Commit | Line | Data |
---|---|---|
d2912cb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
8ed5eec9 AG |
2 | /* |
3 | * Overlayfs NFS export support. | |
4 | * | |
5 | * Amir Goldstein <amir73il@gmail.com> | |
6 | * | |
7 | * Copyright (C) 2017-2018 CTERA Networks. All Rights Reserved. | |
8ed5eec9 AG |
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 | ||
2ca3c148 AG |
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_want_write(dentry); | |
27 | if (!err) { | |
28 | err = ovl_copy_up(dentry); | |
29 | ovl_drop_write(dentry); | |
30 | } | |
31 | ||
32 | if (err) { | |
1bd0a3ae | 33 | pr_warn_ratelimited("failed to copy up on encode (%pd2, err=%i)\n", |
2ca3c148 AG |
34 | dentry, err); |
35 | } | |
36 | ||
37 | return err; | |
38 | } | |
39 | ||
40 | /* | |
41 | * Before encoding a non-upper directory file handle from real layer N, we need | |
42 | * to check if it will be possible to reconnect an overlay dentry from the real | |
43 | * lower decoded dentry. This is done by following the overlay ancestry up to a | |
44 | * "layer N connected" ancestor and verifying that all parents along the way are | |
45 | * "layer N connectable". If an ancestor that is NOT "layer N connectable" is | |
46 | * found, we need to copy up an ancestor, which is "layer N connectable", thus | |
47 | * making that ancestor "layer N connected". For example: | |
48 | * | |
49 | * layer 1: /a | |
50 | * layer 2: /a/b/c | |
51 | * | |
52 | * The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is | |
53 | * copied up and renamed, upper dir /a will be indexed by lower dir /a from | |
54 | * layer 1. The dir /a from layer 2 will never be indexed, so the algorithm (*) | |
55 | * in ovl_lookup_real_ancestor() will not be able to lookup a connected overlay | |
56 | * dentry from the connected lower dentry /a/b/c. | |
57 | * | |
58 | * To avoid this problem on decode time, we need to copy up an ancestor of | |
59 | * /a/b/c, which is "layer 2 connectable", on encode time. That ancestor is | |
60 | * /a/b. After copy up (and index) of /a/b, it will become "layer 2 connected" | |
61 | * and when the time comes to decode the file handle from lower dentry /a/b/c, | |
62 | * ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding | |
63 | * a connected overlay dentry will be accomplished. | |
64 | * | |
65 | * (*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup an | |
66 | * entry /a in the lower layers above layer N and find the indexed dir /a from | |
67 | * layer 1. If that improvement is made, then the check for "layer N connected" | |
68 | * will need to verify there are no redirects in lower layers above N. In the | |
69 | * example above, /a will be "layer 2 connectable". However, if layer 2 dir /a | |
70 | * is a target of a layer 1 redirect, then /a will NOT be "layer 2 connectable": | |
71 | * | |
72 | * layer 1: /A (redirect = /a) | |
73 | * layer 2: /a/b/c | |
74 | */ | |
75 | ||
76 | /* Return the lowest layer for encoding a connectable file handle */ | |
77 | static int ovl_connectable_layer(struct dentry *dentry) | |
78 | { | |
79 | struct ovl_entry *oe = OVL_E(dentry); | |
80 | ||
81 | /* We can get overlay root from root of any layer */ | |
82 | if (dentry == dentry->d_sb->s_root) | |
83 | return oe->numlower; | |
84 | ||
85 | /* | |
86 | * If it's an unindexed merge dir, then it's not connectable with any | |
87 | * lower layer | |
88 | */ | |
89 | if (ovl_dentry_upper(dentry) && | |
90 | !ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
91 | return 0; | |
92 | ||
93 | /* We can get upper/overlay path from indexed/lower dentry */ | |
94 | return oe->lowerstack[0].layer->idx; | |
95 | } | |
96 | ||
97 | /* | |
98 | * @dentry is "connected" if all ancestors up to root or a "connected" ancestor | |
99 | * have the same uppermost lower layer as the origin's layer. We may need to | |
100 | * copy up a "connectable" ancestor to make it "connected". A "connected" dentry | |
101 | * cannot become non "connected", so cache positive result in dentry flags. | |
102 | * | |
103 | * Return the connected origin layer or < 0 on error. | |
104 | */ | |
105 | static int ovl_connect_layer(struct dentry *dentry) | |
106 | { | |
107 | struct dentry *next, *parent = NULL; | |
108 | int origin_layer; | |
109 | int err = 0; | |
110 | ||
111 | if (WARN_ON(dentry == dentry->d_sb->s_root) || | |
112 | WARN_ON(!ovl_dentry_lower(dentry))) | |
113 | return -EIO; | |
114 | ||
115 | origin_layer = OVL_E(dentry)->lowerstack[0].layer->idx; | |
116 | if (ovl_dentry_test_flag(OVL_E_CONNECTED, dentry)) | |
117 | return origin_layer; | |
118 | ||
119 | /* Find the topmost origin layer connectable ancestor of @dentry */ | |
120 | next = dget(dentry); | |
121 | for (;;) { | |
122 | parent = dget_parent(next); | |
123 | if (WARN_ON(parent == next)) { | |
124 | err = -EIO; | |
125 | break; | |
126 | } | |
127 | ||
128 | /* | |
129 | * If @parent is not origin layer connectable, then copy up | |
130 | * @next which is origin layer connectable and we are done. | |
131 | */ | |
132 | if (ovl_connectable_layer(parent) < origin_layer) { | |
133 | err = ovl_encode_maybe_copy_up(next); | |
134 | break; | |
135 | } | |
136 | ||
137 | /* If @parent is connected or indexed we are done */ | |
138 | if (ovl_dentry_test_flag(OVL_E_CONNECTED, parent) || | |
139 | ovl_test_flag(OVL_INDEX, d_inode(parent))) | |
140 | break; | |
141 | ||
142 | dput(next); | |
143 | next = parent; | |
144 | } | |
145 | ||
146 | dput(parent); | |
147 | dput(next); | |
148 | ||
149 | if (!err) | |
150 | ovl_dentry_set_flag(OVL_E_CONNECTED, dentry); | |
151 | ||
152 | return err ?: origin_layer; | |
153 | } | |
154 | ||
b305e844 AG |
155 | /* |
156 | * We only need to encode origin if there is a chance that the same object was | |
157 | * encoded pre copy up and then we need to stay consistent with the same | |
158 | * encoding also after copy up. If non-pure upper is not indexed, then it was | |
159 | * copied up before NFS export was enabled. In that case we don't need to worry | |
160 | * about staying consistent with pre copy up encoding and we encode an upper | |
161 | * file handle. Overlay root dentry is a private case of non-indexed upper. | |
162 | * | |
163 | * The following table summarizes the different file handle encodings used for | |
164 | * different overlay object types: | |
165 | * | |
166 | * Object type | Encoding | |
167 | * -------------------------------- | |
168 | * Pure upper | U | |
169 | * Non-indexed upper | U | |
05e1f118 AG |
170 | * Indexed upper | L (*) |
171 | * Non-upper | L (*) | |
b305e844 AG |
172 | * |
173 | * U = upper file handle | |
174 | * L = lower file handle | |
05e1f118 AG |
175 | * |
176 | * (*) Connecting an overlay dir from real lower dentry is not always | |
2ca3c148 AG |
177 | * possible when there are redirects in lower layers and non-indexed merge dirs. |
178 | * To mitigate those case, we may copy up the lower dir ancestor before encode | |
179 | * a lower dir file handle. | |
180 | * | |
181 | * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error. | |
b305e844 | 182 | */ |
2ca3c148 | 183 | static int ovl_check_encode_origin(struct dentry *dentry) |
b305e844 | 184 | { |
05e1f118 AG |
185 | struct ovl_fs *ofs = dentry->d_sb->s_fs_info; |
186 | ||
2ca3c148 | 187 | /* Upper file handle for pure upper */ |
b305e844 | 188 | if (!ovl_dentry_lower(dentry)) |
2ca3c148 | 189 | return 0; |
b305e844 | 190 | |
05e1f118 | 191 | /* |
2ca3c148 | 192 | * Upper file handle for non-indexed upper. |
05e1f118 | 193 | * |
2ca3c148 AG |
194 | * Root is never indexed, so if there's an upper layer, encode upper for |
195 | * root. | |
05e1f118 | 196 | */ |
b305e844 AG |
197 | if (ovl_dentry_upper(dentry) && |
198 | !ovl_test_flag(OVL_INDEX, d_inode(dentry))) | |
05e1f118 AG |
199 | return 0; |
200 | ||
2ca3c148 AG |
201 | /* |
202 | * Decoding a merge dir, whose origin's ancestor is under a redirected | |
203 | * lower dir or under a non-indexed upper is not always possible. | |
204 | * ovl_connect_layer() will try to make origin's layer "connected" by | |
205 | * copying up a "connectable" ancestor. | |
206 | */ | |
08f4c7c8 | 207 | if (d_is_dir(dentry) && ovl_upper_mnt(ofs)) |
2ca3c148 | 208 | return ovl_connect_layer(dentry); |
05e1f118 | 209 | |
2ca3c148 AG |
210 | /* Lower file handle for indexed and non-upper dir/non-dir */ |
211 | return 1; | |
05e1f118 AG |
212 | } |
213 | ||
1cdb0cb6 PT |
214 | static int ovl_dentry_to_fid(struct ovl_fs *ofs, struct dentry *dentry, |
215 | u32 *fid, int buflen) | |
8ed5eec9 | 216 | { |
8ed5eec9 | 217 | struct ovl_fh *fh = NULL; |
2ca3c148 | 218 | int err, enc_lower; |
cbe7fba8 | 219 | int len; |
8ed5eec9 | 220 | |
05e1f118 | 221 | /* |
2ca3c148 AG |
222 | * Check if we should encode a lower or upper file handle and maybe |
223 | * copy up an ancestor to make lower file handle connectable. | |
05e1f118 | 224 | */ |
2ca3c148 AG |
225 | err = enc_lower = ovl_check_encode_origin(dentry); |
226 | if (enc_lower < 0) | |
227 | goto fail; | |
8ed5eec9 | 228 | |
2ca3c148 | 229 | /* Encode an upper or lower file handle */ |
1cdb0cb6 | 230 | fh = ovl_encode_real_fh(ofs, enc_lower ? ovl_dentry_lower(dentry) : |
5b2cccd3 | 231 | ovl_dentry_upper(dentry), !enc_lower); |
9b6faee0 | 232 | if (IS_ERR(fh)) |
97f024b9 | 233 | return PTR_ERR(fh); |
8ed5eec9 | 234 | |
cbe7fba8 | 235 | len = OVL_FH_LEN(fh); |
144da23b LD |
236 | if (len <= buflen) |
237 | memcpy(fid, fh, len); | |
cbe7fba8 | 238 | err = len; |
8ed5eec9 AG |
239 | |
240 | out: | |
241 | kfree(fh); | |
242 | return err; | |
243 | ||
244 | fail: | |
144da23b LD |
245 | pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n", |
246 | dentry, err); | |
8ed5eec9 AG |
247 | goto out; |
248 | } | |
249 | ||
5b2cccd3 AG |
250 | static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len, |
251 | struct inode *parent) | |
8ed5eec9 | 252 | { |
1cdb0cb6 | 253 | struct ovl_fs *ofs = OVL_FS(inode->i_sb); |
8ed5eec9 | 254 | struct dentry *dentry; |
144da23b | 255 | int bytes, buflen = *max_len << 2; |
8ed5eec9 AG |
256 | |
257 | /* TODO: encode connectable file handles */ | |
258 | if (parent) | |
259 | return FILEID_INVALID; | |
260 | ||
261 | dentry = d_find_any_alias(inode); | |
dd524b7f | 262 | if (!dentry) |
8ed5eec9 AG |
263 | return FILEID_INVALID; |
264 | ||
1cdb0cb6 | 265 | bytes = ovl_dentry_to_fid(ofs, dentry, fid, buflen); |
8ed5eec9 | 266 | dput(dentry); |
cbe7fba8 AG |
267 | if (bytes <= 0) |
268 | return FILEID_INVALID; | |
269 | ||
270 | *max_len = bytes >> 2; | |
144da23b LD |
271 | if (bytes > buflen) |
272 | return FILEID_INVALID; | |
cbe7fba8 AG |
273 | |
274 | return OVL_FILEID_V1; | |
8ed5eec9 AG |
275 | } |
276 | ||
8556a420 | 277 | /* |
f71bd9cf | 278 | * Find or instantiate an overlay dentry from real dentries and index. |
8556a420 AG |
279 | */ |
280 | static struct dentry *ovl_obtain_alias(struct super_block *sb, | |
f71bd9cf AG |
281 | struct dentry *upper_alias, |
282 | struct ovl_path *lowerpath, | |
283 | struct dentry *index) | |
8556a420 | 284 | { |
f941866f | 285 | struct dentry *lower = lowerpath ? lowerpath->dentry : NULL; |
f71bd9cf | 286 | struct dentry *upper = upper_alias ?: index; |
8556a420 | 287 | struct dentry *dentry; |
f941866f | 288 | struct inode *inode; |
8556a420 | 289 | struct ovl_entry *oe; |
ac6a52eb VG |
290 | struct ovl_inode_params oip = { |
291 | .lowerpath = lowerpath, | |
292 | .index = index, | |
293 | .numlower = !!lower | |
294 | }; | |
8556a420 | 295 | |
f71bd9cf AG |
296 | /* We get overlay directory dentries with ovl_lookup_real() */ |
297 | if (d_is_dir(upper ?: lower)) | |
8556a420 AG |
298 | return ERR_PTR(-EIO); |
299 | ||
ac6a52eb VG |
300 | oip.upperdentry = dget(upper); |
301 | inode = ovl_get_inode(sb, &oip); | |
8556a420 AG |
302 | if (IS_ERR(inode)) { |
303 | dput(upper); | |
304 | return ERR_CAST(inode); | |
305 | } | |
306 | ||
9d3dfea3 VG |
307 | if (upper) |
308 | ovl_set_flag(OVL_UPPERDATA, inode); | |
309 | ||
8556a420 | 310 | dentry = d_find_any_alias(inode); |
504f3841 AV |
311 | if (dentry) |
312 | goto out_iput; | |
313 | ||
314 | dentry = d_alloc_anon(inode->i_sb); | |
315 | if (unlikely(!dentry)) | |
316 | goto nomem; | |
317 | oe = ovl_alloc_entry(lower ? 1 : 0); | |
318 | if (!oe) | |
319 | goto nomem; | |
320 | ||
321 | if (lower) { | |
322 | oe->lowerstack->dentry = dget(lower); | |
323 | oe->lowerstack->layer = lowerpath->layer; | |
8556a420 | 324 | } |
504f3841 AV |
325 | dentry->d_fsdata = oe; |
326 | if (upper_alias) | |
327 | ovl_dentry_set_upper_alias(dentry); | |
328 | ||
f4288844 MS |
329 | ovl_dentry_update_reval(dentry, upper, |
330 | DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE); | |
8556a420 AG |
331 | |
332 | return d_instantiate_anon(dentry, inode); | |
333 | ||
334 | nomem: | |
8556a420 | 335 | dput(dentry); |
504f3841 AV |
336 | dentry = ERR_PTR(-ENOMEM); |
337 | out_iput: | |
338 | iput(inode); | |
339 | return dentry; | |
8556a420 AG |
340 | } |
341 | ||
cdf5c9d1 | 342 | /* Get the upper or lower dentry in stack whose on layer @idx */ |
98892516 AG |
343 | static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx) |
344 | { | |
345 | struct ovl_entry *oe = dentry->d_fsdata; | |
346 | int i; | |
347 | ||
348 | if (!idx) | |
349 | return ovl_dentry_upper(dentry); | |
350 | ||
351 | for (i = 0; i < oe->numlower; i++) { | |
352 | if (oe->lowerstack[i].layer->idx == idx) | |
353 | return oe->lowerstack[i].dentry; | |
354 | } | |
355 | ||
356 | return NULL; | |
357 | } | |
358 | ||
3985b70a AG |
359 | /* |
360 | * Lookup a child overlay dentry to get a connected overlay dentry whose real | |
361 | * dentry is @real. If @real is on upper layer, we lookup a child overlay | |
362 | * dentry with the same name as the real dentry. Otherwise, we need to consult | |
363 | * index for lookup. | |
364 | */ | |
365 | static struct dentry *ovl_lookup_real_one(struct dentry *connected, | |
366 | struct dentry *real, | |
13464165 | 367 | const struct ovl_layer *layer) |
3985b70a AG |
368 | { |
369 | struct inode *dir = d_inode(connected); | |
370 | struct dentry *this, *parent = NULL; | |
371 | struct name_snapshot name; | |
372 | int err; | |
373 | ||
3985b70a AG |
374 | /* |
375 | * Lookup child overlay dentry by real name. The dir mutex protects us | |
376 | * from racing with overlay rename. If the overlay dentry that is above | |
377 | * real has already been moved to a parent that is not under the | |
378 | * connected overlay dir, we return -ECHILD and restart the lookup of | |
379 | * connected real path from the top. | |
380 | */ | |
381 | inode_lock_nested(dir, I_MUTEX_PARENT); | |
382 | err = -ECHILD; | |
383 | parent = dget_parent(real); | |
98892516 | 384 | if (ovl_dentry_real_at(connected, layer->idx) != parent) |
3985b70a AG |
385 | goto fail; |
386 | ||
387 | /* | |
388 | * We also need to take a snapshot of real dentry name to protect us | |
389 | * from racing with underlying layer rename. In this case, we don't | |
390 | * care about returning ESTALE, only from dereferencing a free name | |
391 | * pointer because we hold no lock on the real dentry. | |
392 | */ | |
393 | take_dentry_name_snapshot(&name, real); | |
ba9ea771 CB |
394 | /* |
395 | * No mnt_userns handling here: it's an internal lookup. Could skip | |
396 | * permission checking altogether, but for now just use non-mnt_userns | |
397 | * transformed ids. | |
398 | */ | |
230c6402 | 399 | this = lookup_one_len(name.name.name, connected, name.name.len); |
580c6104 | 400 | release_dentry_name_snapshot(&name); |
3985b70a AG |
401 | err = PTR_ERR(this); |
402 | if (IS_ERR(this)) { | |
403 | goto fail; | |
404 | } else if (!this || !this->d_inode) { | |
405 | dput(this); | |
406 | err = -ENOENT; | |
407 | goto fail; | |
98892516 | 408 | } else if (ovl_dentry_real_at(this, layer->idx) != real) { |
3985b70a AG |
409 | dput(this); |
410 | err = -ESTALE; | |
411 | goto fail; | |
412 | } | |
413 | ||
414 | out: | |
3985b70a AG |
415 | dput(parent); |
416 | inode_unlock(dir); | |
417 | return this; | |
418 | ||
419 | fail: | |
1bd0a3ae | 420 | pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n", |
3985b70a AG |
421 | real, layer->idx, connected, err); |
422 | this = ERR_PTR(err); | |
423 | goto out; | |
424 | } | |
425 | ||
06170154 AG |
426 | static struct dentry *ovl_lookup_real(struct super_block *sb, |
427 | struct dentry *real, | |
13464165 | 428 | const struct ovl_layer *layer); |
06170154 | 429 | |
4b91c30a AG |
430 | /* |
431 | * Lookup an indexed or hashed overlay dentry by real inode. | |
432 | */ | |
433 | static struct dentry *ovl_lookup_real_inode(struct super_block *sb, | |
434 | struct dentry *real, | |
13464165 | 435 | const struct ovl_layer *layer) |
4b91c30a | 436 | { |
06170154 | 437 | struct ovl_fs *ofs = sb->s_fs_info; |
06170154 | 438 | struct dentry *index = NULL; |
4b91c30a AG |
439 | struct dentry *this = NULL; |
440 | struct inode *inode; | |
441 | ||
06170154 AG |
442 | /* |
443 | * Decoding upper dir from index is expensive, so first try to lookup | |
444 | * overlay dentry in inode/dcache. | |
445 | */ | |
4b91c30a AG |
446 | inode = ovl_lookup_inode(sb, real, !layer->idx); |
447 | if (IS_ERR(inode)) | |
448 | return ERR_CAST(inode); | |
449 | if (inode) { | |
450 | this = d_find_any_alias(inode); | |
451 | iput(inode); | |
452 | } | |
453 | ||
06170154 AG |
454 | /* |
455 | * For decoded lower dir file handle, lookup index by origin to check | |
456 | * if lower dir was copied up and and/or removed. | |
457 | */ | |
458 | if (!this && layer->idx && ofs->indexdir && !WARN_ON(!d_is_dir(real))) { | |
459 | index = ovl_lookup_index(ofs, NULL, real, false); | |
460 | if (IS_ERR(index)) | |
461 | return index; | |
462 | } | |
463 | ||
464 | /* Get connected upper overlay dir from index */ | |
465 | if (index) { | |
8ea28765 | 466 | struct dentry *upper = ovl_index_upper(ofs, index, true); |
06170154 AG |
467 | |
468 | dput(index); | |
469 | if (IS_ERR_OR_NULL(upper)) | |
470 | return upper; | |
471 | ||
472 | /* | |
473 | * ovl_lookup_real() in lower layer may call recursively once to | |
474 | * ovl_lookup_real() in upper layer. The first level call walks | |
475 | * back lower parents to the topmost indexed parent. The second | |
476 | * recursive call walks back from indexed upper to the topmost | |
477 | * connected/hashed upper parent (or up to root). | |
478 | */ | |
94375f9d | 479 | this = ovl_lookup_real(sb, upper, &ofs->layers[0]); |
06170154 AG |
480 | dput(upper); |
481 | } | |
482 | ||
7168179f AG |
483 | if (IS_ERR_OR_NULL(this)) |
484 | return this; | |
4b91c30a | 485 | |
124c2de2 | 486 | if (ovl_dentry_real_at(this, layer->idx) != real) { |
4b91c30a AG |
487 | dput(this); |
488 | this = ERR_PTR(-EIO); | |
489 | } | |
490 | ||
491 | return this; | |
492 | } | |
493 | ||
494 | /* | |
495 | * Lookup an indexed or hashed overlay dentry, whose real dentry is an | |
496 | * ancestor of @real. | |
497 | */ | |
498 | static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb, | |
499 | struct dentry *real, | |
13464165 | 500 | const struct ovl_layer *layer) |
4b91c30a AG |
501 | { |
502 | struct dentry *next, *parent = NULL; | |
503 | struct dentry *ancestor = ERR_PTR(-EIO); | |
504 | ||
505 | if (real == layer->mnt->mnt_root) | |
506 | return dget(sb->s_root); | |
507 | ||
508 | /* Find the topmost indexed or hashed ancestor */ | |
509 | next = dget(real); | |
510 | for (;;) { | |
511 | parent = dget_parent(next); | |
512 | ||
513 | /* | |
514 | * Lookup a matching overlay dentry in inode/dentry | |
515 | * cache or in index by real inode. | |
516 | */ | |
517 | ancestor = ovl_lookup_real_inode(sb, next, layer); | |
518 | if (ancestor) | |
519 | break; | |
520 | ||
521 | if (parent == layer->mnt->mnt_root) { | |
522 | ancestor = dget(sb->s_root); | |
523 | break; | |
524 | } | |
525 | ||
526 | /* | |
527 | * If @real has been moved out of the layer root directory, | |
528 | * we will eventully hit the real fs root. This cannot happen | |
529 | * by legit overlay rename, so we return error in that case. | |
530 | */ | |
531 | if (parent == next) { | |
532 | ancestor = ERR_PTR(-EXDEV); | |
533 | break; | |
534 | } | |
535 | ||
536 | dput(next); | |
537 | next = parent; | |
538 | } | |
539 | ||
540 | dput(parent); | |
541 | dput(next); | |
542 | ||
543 | return ancestor; | |
544 | } | |
545 | ||
3985b70a AG |
546 | /* |
547 | * Lookup a connected overlay dentry whose real dentry is @real. | |
548 | * If @real is on upper layer, we lookup a child overlay dentry with the same | |
549 | * path the real dentry. Otherwise, we need to consult index for lookup. | |
550 | */ | |
551 | static struct dentry *ovl_lookup_real(struct super_block *sb, | |
552 | struct dentry *real, | |
13464165 | 553 | const struct ovl_layer *layer) |
3985b70a AG |
554 | { |
555 | struct dentry *connected; | |
556 | int err = 0; | |
557 | ||
4b91c30a AG |
558 | connected = ovl_lookup_real_ancestor(sb, real, layer); |
559 | if (IS_ERR(connected)) | |
560 | return connected; | |
3985b70a | 561 | |
3985b70a AG |
562 | while (!err) { |
563 | struct dentry *next, *this; | |
564 | struct dentry *parent = NULL; | |
98892516 AG |
565 | struct dentry *real_connected = ovl_dentry_real_at(connected, |
566 | layer->idx); | |
3985b70a AG |
567 | |
568 | if (real_connected == real) | |
569 | break; | |
570 | ||
571 | /* Find the topmost dentry not yet connected */ | |
572 | next = dget(real); | |
573 | for (;;) { | |
574 | parent = dget_parent(next); | |
575 | ||
576 | if (parent == real_connected) | |
577 | break; | |
578 | ||
579 | /* | |
580 | * If real has been moved out of 'real_connected', | |
581 | * we will not find 'real_connected' and hit the layer | |
582 | * root. In that case, we need to restart connecting. | |
583 | * This game can go on forever in the worst case. We | |
584 | * may want to consider taking s_vfs_rename_mutex if | |
585 | * this happens more than once. | |
586 | */ | |
587 | if (parent == layer->mnt->mnt_root) { | |
588 | dput(connected); | |
589 | connected = dget(sb->s_root); | |
590 | break; | |
591 | } | |
592 | ||
593 | /* | |
594 | * If real file has been moved out of the layer root | |
595 | * directory, we will eventully hit the real fs root. | |
596 | * This cannot happen by legit overlay rename, so we | |
597 | * return error in that case. | |
598 | */ | |
599 | if (parent == next) { | |
600 | err = -EXDEV; | |
601 | break; | |
602 | } | |
603 | ||
604 | dput(next); | |
605 | next = parent; | |
606 | } | |
607 | ||
608 | if (!err) { | |
609 | this = ovl_lookup_real_one(connected, next, layer); | |
610 | if (IS_ERR(this)) | |
611 | err = PTR_ERR(this); | |
612 | ||
613 | /* | |
614 | * Lookup of child in overlay can fail when racing with | |
615 | * overlay rename of child away from 'connected' parent. | |
616 | * In this case, we need to restart the lookup from the | |
617 | * top, because we cannot trust that 'real_connected' is | |
4b91c30a AG |
618 | * still an ancestor of 'real'. There is a good chance |
619 | * that the renamed overlay ancestor is now in cache, so | |
620 | * ovl_lookup_real_ancestor() will find it and we can | |
621 | * continue to connect exactly from where lookup failed. | |
3985b70a AG |
622 | */ |
623 | if (err == -ECHILD) { | |
4b91c30a AG |
624 | this = ovl_lookup_real_ancestor(sb, real, |
625 | layer); | |
b5095f24 | 626 | err = PTR_ERR_OR_ZERO(this); |
3985b70a AG |
627 | } |
628 | if (!err) { | |
629 | dput(connected); | |
630 | connected = this; | |
631 | } | |
632 | } | |
633 | ||
634 | dput(parent); | |
635 | dput(next); | |
636 | } | |
637 | ||
638 | if (err) | |
639 | goto fail; | |
640 | ||
641 | return connected; | |
642 | ||
643 | fail: | |
1bd0a3ae | 644 | pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n", |
3985b70a AG |
645 | real, layer->idx, connected, err); |
646 | dput(connected); | |
647 | return ERR_PTR(err); | |
648 | } | |
649 | ||
650 | /* | |
f71bd9cf | 651 | * Get an overlay dentry from upper/lower real dentries and index. |
3985b70a AG |
652 | */ |
653 | static struct dentry *ovl_get_dentry(struct super_block *sb, | |
654 | struct dentry *upper, | |
f71bd9cf AG |
655 | struct ovl_path *lowerpath, |
656 | struct dentry *index) | |
3985b70a AG |
657 | { |
658 | struct ovl_fs *ofs = sb->s_fs_info; | |
13464165 | 659 | const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer; |
f71bd9cf | 660 | struct dentry *real = upper ?: (index ?: lowerpath->dentry); |
3985b70a | 661 | |
f941866f | 662 | /* |
f71bd9cf AG |
663 | * Obtain a disconnected overlay dentry from a non-dir real dentry |
664 | * and index. | |
f941866f | 665 | */ |
f71bd9cf AG |
666 | if (!d_is_dir(real)) |
667 | return ovl_obtain_alias(sb, upper, lowerpath, index); | |
f941866f | 668 | |
3985b70a | 669 | /* Removed empty directory? */ |
98892516 | 670 | if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real)) |
3985b70a AG |
671 | return ERR_PTR(-ENOENT); |
672 | ||
673 | /* | |
98892516 AG |
674 | * If real dentry is connected and hashed, get a connected overlay |
675 | * dentry whose real dentry is @real. | |
3985b70a | 676 | */ |
98892516 | 677 | return ovl_lookup_real(sb, real, layer); |
3985b70a AG |
678 | } |
679 | ||
8556a420 AG |
680 | static struct dentry *ovl_upper_fh_to_d(struct super_block *sb, |
681 | struct ovl_fh *fh) | |
682 | { | |
683 | struct ovl_fs *ofs = sb->s_fs_info; | |
684 | struct dentry *dentry; | |
685 | struct dentry *upper; | |
686 | ||
08f4c7c8 | 687 | if (!ovl_upper_mnt(ofs)) |
8556a420 AG |
688 | return ERR_PTR(-EACCES); |
689 | ||
1cdb0cb6 | 690 | upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true); |
8556a420 AG |
691 | if (IS_ERR_OR_NULL(upper)) |
692 | return upper; | |
693 | ||
f71bd9cf | 694 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); |
8556a420 AG |
695 | dput(upper); |
696 | ||
697 | return dentry; | |
698 | } | |
699 | ||
f941866f AG |
700 | static struct dentry *ovl_lower_fh_to_d(struct super_block *sb, |
701 | struct ovl_fh *fh) | |
702 | { | |
703 | struct ovl_fs *ofs = sb->s_fs_info; | |
704 | struct ovl_path origin = { }; | |
705 | struct ovl_path *stack = &origin; | |
706 | struct dentry *dentry = NULL; | |
f71bd9cf | 707 | struct dentry *index = NULL; |
8b58924a | 708 | struct inode *inode; |
f941866f AG |
709 | int err; |
710 | ||
8b58924a AG |
711 | /* First lookup overlay inode in inode cache by origin fh */ |
712 | err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack); | |
713 | if (err) | |
714 | return ERR_PTR(err); | |
715 | ||
716 | if (!d_is_dir(origin.dentry) || | |
717 | !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) { | |
718 | inode = ovl_lookup_inode(sb, origin.dentry, false); | |
719 | err = PTR_ERR(inode); | |
720 | if (IS_ERR(inode)) | |
721 | goto out_err; | |
722 | if (inode) { | |
723 | dentry = d_find_any_alias(inode); | |
724 | iput(inode); | |
725 | if (dentry) | |
726 | goto out; | |
727 | } | |
728 | } | |
729 | ||
730 | /* Then lookup indexed upper/whiteout by origin fh */ | |
f71bd9cf AG |
731 | if (ofs->indexdir) { |
732 | index = ovl_get_index_fh(ofs, fh); | |
733 | err = PTR_ERR(index); | |
9436a1a3 | 734 | if (IS_ERR(index)) { |
9436a1a3 | 735 | index = NULL; |
8b58924a | 736 | goto out_err; |
9436a1a3 | 737 | } |
f71bd9cf AG |
738 | } |
739 | ||
8b58924a | 740 | /* Then try to get a connected upper dir by index */ |
3b0bfc6e | 741 | if (index && d_is_dir(index)) { |
8ea28765 | 742 | struct dentry *upper = ovl_index_upper(ofs, index, true); |
3b0bfc6e AG |
743 | |
744 | err = PTR_ERR(upper); | |
745 | if (IS_ERR_OR_NULL(upper)) | |
746 | goto out_err; | |
747 | ||
748 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); | |
749 | dput(upper); | |
750 | goto out; | |
751 | } | |
752 | ||
155b8a04 AG |
753 | /* Find origin.dentry again with ovl_acceptable() layer check */ |
754 | if (d_is_dir(origin.dentry)) { | |
8b58924a AG |
755 | dput(origin.dentry); |
756 | origin.dentry = NULL; | |
757 | err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack); | |
f71bd9cf AG |
758 | if (err) |
759 | goto out_err; | |
8b58924a AG |
760 | } |
761 | if (index) { | |
610afc0b | 762 | err = ovl_verify_origin(ofs, index, origin.dentry, false); |
8b58924a | 763 | if (err) |
9436a1a3 | 764 | goto out_err; |
f71bd9cf | 765 | } |
f941866f | 766 | |
155b8a04 | 767 | /* Get a connected non-upper dir or disconnected non-dir */ |
f71bd9cf | 768 | dentry = ovl_get_dentry(sb, NULL, &origin, index); |
f941866f | 769 | |
f71bd9cf AG |
770 | out: |
771 | dput(origin.dentry); | |
772 | dput(index); | |
f941866f | 773 | return dentry; |
f71bd9cf AG |
774 | |
775 | out_err: | |
776 | dentry = ERR_PTR(err); | |
777 | goto out; | |
f941866f AG |
778 | } |
779 | ||
cbe7fba8 AG |
780 | static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type) |
781 | { | |
782 | struct ovl_fh *fh; | |
783 | ||
784 | /* If on-wire inner fid is aligned - nothing to do */ | |
785 | if (fh_type == OVL_FILEID_V1) | |
786 | return (struct ovl_fh *)fid; | |
787 | ||
788 | if (fh_type != OVL_FILEID_V0) | |
789 | return ERR_PTR(-EINVAL); | |
790 | ||
9aafc1b0 DC |
791 | if (buflen <= OVL_FH_WIRE_OFFSET) |
792 | return ERR_PTR(-EINVAL); | |
793 | ||
cbe7fba8 AG |
794 | fh = kzalloc(buflen, GFP_KERNEL); |
795 | if (!fh) | |
796 | return ERR_PTR(-ENOMEM); | |
797 | ||
798 | /* Copy unaligned inner fh into aligned buffer */ | |
cf8aa9bf | 799 | memcpy(fh->buf, fid, buflen - OVL_FH_WIRE_OFFSET); |
cbe7fba8 AG |
800 | return fh; |
801 | } | |
802 | ||
8556a420 AG |
803 | static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid, |
804 | int fh_len, int fh_type) | |
805 | { | |
806 | struct dentry *dentry = NULL; | |
cbe7fba8 | 807 | struct ovl_fh *fh = NULL; |
8556a420 AG |
808 | int len = fh_len << 2; |
809 | unsigned int flags = 0; | |
810 | int err; | |
811 | ||
cbe7fba8 AG |
812 | fh = ovl_fid_to_fh(fid, len, fh_type); |
813 | err = PTR_ERR(fh); | |
814 | if (IS_ERR(fh)) | |
8556a420 AG |
815 | goto out_err; |
816 | ||
817 | err = ovl_check_fh_len(fh, len); | |
818 | if (err) | |
819 | goto out_err; | |
820 | ||
cbe7fba8 | 821 | flags = fh->fb.flags; |
f941866f AG |
822 | dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ? |
823 | ovl_upper_fh_to_d(sb, fh) : | |
824 | ovl_lower_fh_to_d(sb, fh); | |
8556a420 AG |
825 | err = PTR_ERR(dentry); |
826 | if (IS_ERR(dentry) && err != -ESTALE) | |
827 | goto out_err; | |
828 | ||
cbe7fba8 AG |
829 | out: |
830 | /* We may have needed to re-align OVL_FILEID_V0 */ | |
831 | if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid) | |
832 | kfree(fh); | |
833 | ||
8556a420 AG |
834 | return dentry; |
835 | ||
836 | out_err: | |
1bd0a3ae | 837 | pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n", |
cbe7fba8 AG |
838 | fh_len, fh_type, flags, err); |
839 | dentry = ERR_PTR(err); | |
840 | goto out; | |
8556a420 AG |
841 | } |
842 | ||
3985b70a AG |
843 | static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid, |
844 | int fh_len, int fh_type) | |
845 | { | |
1bd0a3ae | 846 | pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n"); |
3985b70a AG |
847 | return ERR_PTR(-EACCES); |
848 | } | |
849 | ||
850 | static int ovl_get_name(struct dentry *parent, char *name, | |
851 | struct dentry *child) | |
852 | { | |
853 | /* | |
854 | * ovl_fh_to_dentry() returns connected dir overlay dentries and | |
855 | * ovl_fh_to_parent() is not implemented, so we should not get here. | |
856 | */ | |
857 | WARN_ON_ONCE(1); | |
858 | return -EIO; | |
859 | } | |
860 | ||
861 | static struct dentry *ovl_get_parent(struct dentry *dentry) | |
862 | { | |
863 | /* | |
864 | * ovl_fh_to_dentry() returns connected dir overlay dentries, so we | |
865 | * should not get here. | |
866 | */ | |
867 | WARN_ON_ONCE(1); | |
868 | return ERR_PTR(-EIO); | |
869 | } | |
870 | ||
8ed5eec9 | 871 | const struct export_operations ovl_export_operations = { |
5b2cccd3 | 872 | .encode_fh = ovl_encode_fh, |
8556a420 | 873 | .fh_to_dentry = ovl_fh_to_dentry, |
3985b70a AG |
874 | .fh_to_parent = ovl_fh_to_parent, |
875 | .get_name = ovl_get_name, | |
876 | .get_parent = ovl_get_parent, | |
8ed5eec9 | 877 | }; |