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 | ||
162d0644 | 26 | err = ovl_copy_up(dentry); |
2ca3c148 | 27 | if (err) { |
1bd0a3ae | 28 | pr_warn_ratelimited("failed to copy up on encode (%pd2, err=%i)\n", |
2ca3c148 AG |
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) | |
5522c9c7 | 78 | return ovl_numlower(oe); |
2ca3c148 AG |
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(OVL_INDEX, d_inode(dentry))) | |
86 | return 0; | |
87 | ||
88 | /* We can get upper/overlay path from indexed/lower dentry */ | |
5522c9c7 | 89 | return ovl_lowerstack(oe)->layer->idx; |
2ca3c148 AG |
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; | |
5522c9c7 | 103 | struct ovl_entry *oe = OVL_E(dentry); |
2ca3c148 AG |
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 | ||
5522c9c7 | 111 | origin_layer = ovl_lowerstack(oe)->layer->idx; |
2ca3c148 AG |
112 | if (ovl_dentry_test_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(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(parent) < origin_layer) { | |
129 | err = ovl_encode_maybe_copy_up(next); | |
130 | break; | |
131 | } | |
132 | ||
133 | /* If @parent is connected or indexed we are done */ | |
134 | if (ovl_dentry_test_flag(OVL_E_CONNECTED, parent) || | |
135 | ovl_test_flag(OVL_INDEX, d_inode(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(OVL_E_CONNECTED, dentry); | |
147 | ||
148 | return err ?: origin_layer; | |
149 | } | |
150 | ||
b305e844 AG |
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 | |
05e1f118 AG |
166 | * Indexed upper | L (*) |
167 | * Non-upper | L (*) | |
b305e844 AG |
168 | * |
169 | * U = upper file handle | |
170 | * L = lower file handle | |
05e1f118 | 171 | * |
16aac5ad | 172 | * (*) Decoding a connected overlay dir from real lower dentry is not always |
2ca3c148 AG |
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 | |
16aac5ad | 175 | * of a decodable file handle for non-upper dir. |
2ca3c148 AG |
176 | * |
177 | * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error. | |
b305e844 | 178 | */ |
c45beebf | 179 | static int ovl_check_encode_origin(struct inode *inode) |
b305e844 | 180 | { |
c45beebf | 181 | struct ovl_fs *ofs = OVL_FS(inode->i_sb); |
16aac5ad | 182 | bool decodable = ofs->config.nfs_export; |
c45beebf AG |
183 | struct dentry *dentry; |
184 | int err; | |
16aac5ad | 185 | |
004b8d14 MS |
186 | /* No upper layer? */ |
187 | if (!ovl_upper_mnt(ofs)) | |
188 | return 1; | |
189 | ||
16aac5ad | 190 | /* Lower file handle for non-upper non-decodable */ |
c45beebf | 191 | if (!ovl_inode_upper(inode) && !decodable) |
c7242a45 | 192 | return 1; |
05e1f118 | 193 | |
2ca3c148 | 194 | /* Upper file handle for pure upper */ |
c45beebf | 195 | if (!ovl_inode_lower(inode)) |
2ca3c148 | 196 | return 0; |
b305e844 | 197 | |
05e1f118 | 198 | /* |
2ca3c148 AG |
199 | * Root is never indexed, so if there's an upper layer, encode upper for |
200 | * root. | |
05e1f118 | 201 | */ |
c45beebf | 202 | if (inode == d_inode(inode->i_sb->s_root)) |
16aac5ad AG |
203 | return 0; |
204 | ||
205 | /* | |
206 | * Upper decodable file handle for non-indexed upper. | |
207 | */ | |
c45beebf AG |
208 | if (ovl_inode_upper(inode) && decodable && |
209 | !ovl_test_flag(OVL_INDEX, inode)) | |
05e1f118 AG |
210 | return 0; |
211 | ||
2ca3c148 AG |
212 | /* |
213 | * Decoding a merge dir, whose origin's ancestor is under a redirected | |
214 | * lower dir or under a non-indexed upper is not always possible. | |
215 | * ovl_connect_layer() will try to make origin's layer "connected" by | |
216 | * copying up a "connectable" ancestor. | |
217 | */ | |
c45beebf AG |
218 | if (!decodable || !S_ISDIR(inode->i_mode)) |
219 | return 1; | |
220 | ||
221 | dentry = d_find_any_alias(inode); | |
222 | if (!dentry) | |
223 | return -ENOENT; | |
224 | ||
225 | err = ovl_connect_layer(dentry); | |
226 | dput(dentry); | |
227 | if (err < 0) | |
228 | return err; | |
05e1f118 | 229 | |
2ca3c148 AG |
230 | /* Lower file handle for indexed and non-upper dir/non-dir */ |
231 | return 1; | |
05e1f118 AG |
232 | } |
233 | ||
c45beebf | 234 | static int ovl_dentry_to_fid(struct ovl_fs *ofs, struct inode *inode, |
1cdb0cb6 | 235 | u32 *fid, int buflen) |
8ed5eec9 | 236 | { |
8ed5eec9 | 237 | struct ovl_fh *fh = NULL; |
2ca3c148 | 238 | int err, enc_lower; |
cbe7fba8 | 239 | int len; |
8ed5eec9 | 240 | |
05e1f118 | 241 | /* |
2ca3c148 AG |
242 | * Check if we should encode a lower or upper file handle and maybe |
243 | * copy up an ancestor to make lower file handle connectable. | |
05e1f118 | 244 | */ |
c45beebf | 245 | err = enc_lower = ovl_check_encode_origin(inode); |
2ca3c148 AG |
246 | if (enc_lower < 0) |
247 | goto fail; | |
8ed5eec9 | 248 | |
2ca3c148 | 249 | /* Encode an upper or lower file handle */ |
07aeefae AG |
250 | fh = ovl_encode_real_fh(ofs, enc_lower ? ovl_inode_lower(inode) : |
251 | ovl_inode_upper(inode), !enc_lower); | |
9b6faee0 | 252 | if (IS_ERR(fh)) |
97f024b9 | 253 | return PTR_ERR(fh); |
8ed5eec9 | 254 | |
cbe7fba8 | 255 | len = OVL_FH_LEN(fh); |
144da23b LD |
256 | if (len <= buflen) |
257 | memcpy(fid, fh, len); | |
cbe7fba8 | 258 | err = len; |
8ed5eec9 AG |
259 | |
260 | out: | |
261 | kfree(fh); | |
262 | return err; | |
263 | ||
264 | fail: | |
c45beebf AG |
265 | pr_warn_ratelimited("failed to encode file handle (ino=%lu, err=%i)\n", |
266 | inode->i_ino, err); | |
8ed5eec9 AG |
267 | goto out; |
268 | } | |
269 | ||
5b2cccd3 AG |
270 | static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len, |
271 | struct inode *parent) | |
8ed5eec9 | 272 | { |
1cdb0cb6 | 273 | struct ovl_fs *ofs = OVL_FS(inode->i_sb); |
144da23b | 274 | int bytes, buflen = *max_len << 2; |
8ed5eec9 AG |
275 | |
276 | /* TODO: encode connectable file handles */ | |
277 | if (parent) | |
278 | return FILEID_INVALID; | |
279 | ||
c45beebf | 280 | bytes = ovl_dentry_to_fid(ofs, inode, fid, buflen); |
cbe7fba8 AG |
281 | if (bytes <= 0) |
282 | return FILEID_INVALID; | |
283 | ||
284 | *max_len = bytes >> 2; | |
144da23b LD |
285 | if (bytes > buflen) |
286 | return FILEID_INVALID; | |
cbe7fba8 AG |
287 | |
288 | return OVL_FILEID_V1; | |
8ed5eec9 AG |
289 | } |
290 | ||
8556a420 | 291 | /* |
f71bd9cf | 292 | * Find or instantiate an overlay dentry from real dentries and index. |
8556a420 AG |
293 | */ |
294 | static struct dentry *ovl_obtain_alias(struct super_block *sb, | |
f71bd9cf AG |
295 | struct dentry *upper_alias, |
296 | struct ovl_path *lowerpath, | |
297 | struct dentry *index) | |
8556a420 | 298 | { |
f941866f | 299 | struct dentry *lower = lowerpath ? lowerpath->dentry : NULL; |
f71bd9cf | 300 | struct dentry *upper = upper_alias ?: index; |
0af950f5 | 301 | struct inode *inode = NULL; |
8556a420 | 302 | struct ovl_entry *oe; |
ac6a52eb | 303 | struct ovl_inode_params oip = { |
ac6a52eb | 304 | .index = index, |
ac6a52eb | 305 | }; |
8556a420 | 306 | |
f71bd9cf AG |
307 | /* We get overlay directory dentries with ovl_lookup_real() */ |
308 | if (d_is_dir(upper ?: lower)) | |
8556a420 AG |
309 | return ERR_PTR(-EIO); |
310 | ||
0af950f5 AG |
311 | oe = ovl_alloc_entry(!!lower); |
312 | if (!oe) | |
313 | return ERR_PTR(-ENOMEM); | |
314 | ||
ac6a52eb | 315 | oip.upperdentry = dget(upper); |
0af950f5 AG |
316 | if (lower) { |
317 | ovl_lowerstack(oe)->dentry = dget(lower); | |
318 | ovl_lowerstack(oe)->layer = lowerpath->layer; | |
319 | } | |
320 | oip.oe = oe; | |
ac6a52eb | 321 | inode = ovl_get_inode(sb, &oip); |
8556a420 | 322 | if (IS_ERR(inode)) { |
0af950f5 | 323 | ovl_free_entry(oe); |
8556a420 AG |
324 | dput(upper); |
325 | return ERR_CAST(inode); | |
326 | } | |
327 | ||
9d3dfea3 VG |
328 | if (upper) |
329 | ovl_set_flag(OVL_UPPERDATA, inode); | |
330 | ||
484d4fbf | 331 | return d_obtain_alias(inode); |
8556a420 AG |
332 | } |
333 | ||
cdf5c9d1 | 334 | /* Get the upper or lower dentry in stack whose on layer @idx */ |
98892516 AG |
335 | static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx) |
336 | { | |
a6ff2bc0 | 337 | struct ovl_entry *oe = OVL_E(dentry); |
5522c9c7 | 338 | struct ovl_path *lowerstack = ovl_lowerstack(oe); |
98892516 AG |
339 | int i; |
340 | ||
341 | if (!idx) | |
342 | return ovl_dentry_upper(dentry); | |
343 | ||
5522c9c7 AG |
344 | for (i = 0; i < ovl_numlower(oe); i++) { |
345 | if (lowerstack[i].layer->idx == idx) | |
346 | return lowerstack[i].dentry; | |
98892516 AG |
347 | } |
348 | ||
349 | return NULL; | |
350 | } | |
351 | ||
3985b70a AG |
352 | /* |
353 | * Lookup a child overlay dentry to get a connected overlay dentry whose real | |
354 | * dentry is @real. If @real is on upper layer, we lookup a child overlay | |
355 | * dentry with the same name as the real dentry. Otherwise, we need to consult | |
356 | * index for lookup. | |
357 | */ | |
358 | static struct dentry *ovl_lookup_real_one(struct dentry *connected, | |
359 | struct dentry *real, | |
13464165 | 360 | const struct ovl_layer *layer) |
3985b70a AG |
361 | { |
362 | struct inode *dir = d_inode(connected); | |
363 | struct dentry *this, *parent = NULL; | |
364 | struct name_snapshot name; | |
365 | int err; | |
366 | ||
3985b70a AG |
367 | /* |
368 | * Lookup child overlay dentry by real name. The dir mutex protects us | |
369 | * from racing with overlay rename. If the overlay dentry that is above | |
370 | * real has already been moved to a parent that is not under the | |
371 | * connected overlay dir, we return -ECHILD and restart the lookup of | |
372 | * connected real path from the top. | |
373 | */ | |
374 | inode_lock_nested(dir, I_MUTEX_PARENT); | |
375 | err = -ECHILD; | |
376 | parent = dget_parent(real); | |
98892516 | 377 | if (ovl_dentry_real_at(connected, layer->idx) != parent) |
3985b70a AG |
378 | goto fail; |
379 | ||
380 | /* | |
381 | * We also need to take a snapshot of real dentry name to protect us | |
382 | * from racing with underlying layer rename. In this case, we don't | |
383 | * care about returning ESTALE, only from dereferencing a free name | |
384 | * pointer because we hold no lock on the real dentry. | |
385 | */ | |
386 | take_dentry_name_snapshot(&name, real); | |
ba9ea771 | 387 | /* |
fa6fe07d | 388 | * No idmap handling here: it's an internal lookup. |
ba9ea771 | 389 | */ |
fa6fe07d | 390 | this = lookup_noperm(&name.name, connected); |
580c6104 | 391 | release_dentry_name_snapshot(&name); |
3985b70a AG |
392 | err = PTR_ERR(this); |
393 | if (IS_ERR(this)) { | |
394 | goto fail; | |
395 | } else if (!this || !this->d_inode) { | |
396 | dput(this); | |
397 | err = -ENOENT; | |
398 | goto fail; | |
98892516 | 399 | } else if (ovl_dentry_real_at(this, layer->idx) != real) { |
3985b70a AG |
400 | dput(this); |
401 | err = -ESTALE; | |
402 | goto fail; | |
403 | } | |
404 | ||
405 | out: | |
3985b70a AG |
406 | dput(parent); |
407 | inode_unlock(dir); | |
408 | return this; | |
409 | ||
410 | fail: | |
1bd0a3ae | 411 | pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n", |
3985b70a AG |
412 | real, layer->idx, connected, err); |
413 | this = ERR_PTR(err); | |
414 | goto out; | |
415 | } | |
416 | ||
06170154 AG |
417 | static struct dentry *ovl_lookup_real(struct super_block *sb, |
418 | struct dentry *real, | |
13464165 | 419 | const struct ovl_layer *layer); |
06170154 | 420 | |
4b91c30a AG |
421 | /* |
422 | * Lookup an indexed or hashed overlay dentry by real inode. | |
423 | */ | |
424 | static struct dentry *ovl_lookup_real_inode(struct super_block *sb, | |
425 | struct dentry *real, | |
13464165 | 426 | const struct ovl_layer *layer) |
4b91c30a | 427 | { |
f01d0889 | 428 | struct ovl_fs *ofs = OVL_FS(sb); |
06170154 | 429 | struct dentry *index = NULL; |
4b91c30a AG |
430 | struct dentry *this = NULL; |
431 | struct inode *inode; | |
432 | ||
06170154 AG |
433 | /* |
434 | * Decoding upper dir from index is expensive, so first try to lookup | |
435 | * overlay dentry in inode/dcache. | |
436 | */ | |
4b91c30a AG |
437 | inode = ovl_lookup_inode(sb, real, !layer->idx); |
438 | if (IS_ERR(inode)) | |
439 | return ERR_CAST(inode); | |
440 | if (inode) { | |
441 | this = d_find_any_alias(inode); | |
442 | iput(inode); | |
443 | } | |
444 | ||
06170154 AG |
445 | /* |
446 | * For decoded lower dir file handle, lookup index by origin to check | |
447 | * if lower dir was copied up and and/or removed. | |
448 | */ | |
02d70090 | 449 | if (!this && layer->idx && ovl_indexdir(sb) && !WARN_ON(!d_is_dir(real))) { |
06170154 AG |
450 | index = ovl_lookup_index(ofs, NULL, real, false); |
451 | if (IS_ERR(index)) | |
452 | return index; | |
453 | } | |
454 | ||
455 | /* Get connected upper overlay dir from index */ | |
456 | if (index) { | |
8ea28765 | 457 | struct dentry *upper = ovl_index_upper(ofs, index, true); |
06170154 AG |
458 | |
459 | dput(index); | |
460 | if (IS_ERR_OR_NULL(upper)) | |
461 | return upper; | |
462 | ||
463 | /* | |
464 | * ovl_lookup_real() in lower layer may call recursively once to | |
465 | * ovl_lookup_real() in upper layer. The first level call walks | |
466 | * back lower parents to the topmost indexed parent. The second | |
467 | * recursive call walks back from indexed upper to the topmost | |
468 | * connected/hashed upper parent (or up to root). | |
469 | */ | |
94375f9d | 470 | this = ovl_lookup_real(sb, upper, &ofs->layers[0]); |
06170154 AG |
471 | dput(upper); |
472 | } | |
473 | ||
7168179f AG |
474 | if (IS_ERR_OR_NULL(this)) |
475 | return this; | |
4b91c30a | 476 | |
124c2de2 | 477 | if (ovl_dentry_real_at(this, layer->idx) != real) { |
4b91c30a AG |
478 | dput(this); |
479 | this = ERR_PTR(-EIO); | |
480 | } | |
481 | ||
482 | return this; | |
483 | } | |
484 | ||
485 | /* | |
486 | * Lookup an indexed or hashed overlay dentry, whose real dentry is an | |
487 | * ancestor of @real. | |
488 | */ | |
489 | static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb, | |
490 | struct dentry *real, | |
13464165 | 491 | const struct ovl_layer *layer) |
4b91c30a AG |
492 | { |
493 | struct dentry *next, *parent = NULL; | |
494 | struct dentry *ancestor = ERR_PTR(-EIO); | |
495 | ||
496 | if (real == layer->mnt->mnt_root) | |
497 | return dget(sb->s_root); | |
498 | ||
499 | /* Find the topmost indexed or hashed ancestor */ | |
500 | next = dget(real); | |
501 | for (;;) { | |
502 | parent = dget_parent(next); | |
503 | ||
504 | /* | |
505 | * Lookup a matching overlay dentry in inode/dentry | |
506 | * cache or in index by real inode. | |
507 | */ | |
508 | ancestor = ovl_lookup_real_inode(sb, next, layer); | |
509 | if (ancestor) | |
510 | break; | |
511 | ||
512 | if (parent == layer->mnt->mnt_root) { | |
513 | ancestor = dget(sb->s_root); | |
514 | break; | |
515 | } | |
516 | ||
517 | /* | |
518 | * If @real has been moved out of the layer root directory, | |
519 | * we will eventully hit the real fs root. This cannot happen | |
520 | * by legit overlay rename, so we return error in that case. | |
521 | */ | |
522 | if (parent == next) { | |
523 | ancestor = ERR_PTR(-EXDEV); | |
524 | break; | |
525 | } | |
526 | ||
527 | dput(next); | |
528 | next = parent; | |
529 | } | |
530 | ||
531 | dput(parent); | |
532 | dput(next); | |
533 | ||
534 | return ancestor; | |
535 | } | |
536 | ||
3985b70a AG |
537 | /* |
538 | * Lookup a connected overlay dentry whose real dentry is @real. | |
539 | * If @real is on upper layer, we lookup a child overlay dentry with the same | |
540 | * path the real dentry. Otherwise, we need to consult index for lookup. | |
541 | */ | |
542 | static struct dentry *ovl_lookup_real(struct super_block *sb, | |
543 | struct dentry *real, | |
13464165 | 544 | const struct ovl_layer *layer) |
3985b70a AG |
545 | { |
546 | struct dentry *connected; | |
547 | int err = 0; | |
548 | ||
4b91c30a AG |
549 | connected = ovl_lookup_real_ancestor(sb, real, layer); |
550 | if (IS_ERR(connected)) | |
551 | return connected; | |
3985b70a | 552 | |
3985b70a AG |
553 | while (!err) { |
554 | struct dentry *next, *this; | |
555 | struct dentry *parent = NULL; | |
98892516 AG |
556 | struct dentry *real_connected = ovl_dentry_real_at(connected, |
557 | layer->idx); | |
3985b70a AG |
558 | |
559 | if (real_connected == real) | |
560 | break; | |
561 | ||
562 | /* Find the topmost dentry not yet connected */ | |
563 | next = dget(real); | |
564 | for (;;) { | |
565 | parent = dget_parent(next); | |
566 | ||
567 | if (parent == real_connected) | |
568 | break; | |
569 | ||
570 | /* | |
571 | * If real has been moved out of 'real_connected', | |
572 | * we will not find 'real_connected' and hit the layer | |
573 | * root. In that case, we need to restart connecting. | |
574 | * This game can go on forever in the worst case. We | |
575 | * may want to consider taking s_vfs_rename_mutex if | |
576 | * this happens more than once. | |
577 | */ | |
578 | if (parent == layer->mnt->mnt_root) { | |
579 | dput(connected); | |
580 | connected = dget(sb->s_root); | |
581 | break; | |
582 | } | |
583 | ||
584 | /* | |
585 | * If real file has been moved out of the layer root | |
586 | * directory, we will eventully hit the real fs root. | |
587 | * This cannot happen by legit overlay rename, so we | |
588 | * return error in that case. | |
589 | */ | |
590 | if (parent == next) { | |
591 | err = -EXDEV; | |
592 | break; | |
593 | } | |
594 | ||
595 | dput(next); | |
596 | next = parent; | |
597 | } | |
598 | ||
599 | if (!err) { | |
600 | this = ovl_lookup_real_one(connected, next, layer); | |
601 | if (IS_ERR(this)) | |
602 | err = PTR_ERR(this); | |
603 | ||
604 | /* | |
605 | * Lookup of child in overlay can fail when racing with | |
606 | * overlay rename of child away from 'connected' parent. | |
607 | * In this case, we need to restart the lookup from the | |
608 | * top, because we cannot trust that 'real_connected' is | |
4b91c30a AG |
609 | * still an ancestor of 'real'. There is a good chance |
610 | * that the renamed overlay ancestor is now in cache, so | |
611 | * ovl_lookup_real_ancestor() will find it and we can | |
612 | * continue to connect exactly from where lookup failed. | |
3985b70a AG |
613 | */ |
614 | if (err == -ECHILD) { | |
4b91c30a AG |
615 | this = ovl_lookup_real_ancestor(sb, real, |
616 | layer); | |
b5095f24 | 617 | err = PTR_ERR_OR_ZERO(this); |
3985b70a AG |
618 | } |
619 | if (!err) { | |
620 | dput(connected); | |
621 | connected = this; | |
622 | } | |
623 | } | |
624 | ||
625 | dput(parent); | |
626 | dput(next); | |
627 | } | |
628 | ||
629 | if (err) | |
630 | goto fail; | |
631 | ||
632 | return connected; | |
633 | ||
634 | fail: | |
1bd0a3ae | 635 | pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n", |
3985b70a AG |
636 | real, layer->idx, connected, err); |
637 | dput(connected); | |
638 | return ERR_PTR(err); | |
639 | } | |
640 | ||
641 | /* | |
f71bd9cf | 642 | * Get an overlay dentry from upper/lower real dentries and index. |
3985b70a AG |
643 | */ |
644 | static struct dentry *ovl_get_dentry(struct super_block *sb, | |
645 | struct dentry *upper, | |
f71bd9cf AG |
646 | struct ovl_path *lowerpath, |
647 | struct dentry *index) | |
3985b70a | 648 | { |
f01d0889 | 649 | struct ovl_fs *ofs = OVL_FS(sb); |
13464165 | 650 | const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer; |
f71bd9cf | 651 | struct dentry *real = upper ?: (index ?: lowerpath->dentry); |
3985b70a | 652 | |
f941866f | 653 | /* |
f71bd9cf AG |
654 | * Obtain a disconnected overlay dentry from a non-dir real dentry |
655 | * and index. | |
f941866f | 656 | */ |
f71bd9cf AG |
657 | if (!d_is_dir(real)) |
658 | return ovl_obtain_alias(sb, upper, lowerpath, index); | |
f941866f | 659 | |
3985b70a | 660 | /* Removed empty directory? */ |
98892516 | 661 | if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(real)) |
3985b70a AG |
662 | return ERR_PTR(-ENOENT); |
663 | ||
664 | /* | |
98892516 AG |
665 | * If real dentry is connected and hashed, get a connected overlay |
666 | * dentry whose real dentry is @real. | |
3985b70a | 667 | */ |
98892516 | 668 | return ovl_lookup_real(sb, real, layer); |
3985b70a AG |
669 | } |
670 | ||
8556a420 AG |
671 | static struct dentry *ovl_upper_fh_to_d(struct super_block *sb, |
672 | struct ovl_fh *fh) | |
673 | { | |
f01d0889 | 674 | struct ovl_fs *ofs = OVL_FS(sb); |
8556a420 AG |
675 | struct dentry *dentry; |
676 | struct dentry *upper; | |
677 | ||
08f4c7c8 | 678 | if (!ovl_upper_mnt(ofs)) |
8556a420 AG |
679 | return ERR_PTR(-EACCES); |
680 | ||
1cdb0cb6 | 681 | upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true); |
8556a420 AG |
682 | if (IS_ERR_OR_NULL(upper)) |
683 | return upper; | |
684 | ||
f71bd9cf | 685 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); |
8556a420 AG |
686 | dput(upper); |
687 | ||
688 | return dentry; | |
689 | } | |
690 | ||
f941866f AG |
691 | static struct dentry *ovl_lower_fh_to_d(struct super_block *sb, |
692 | struct ovl_fh *fh) | |
693 | { | |
f01d0889 | 694 | struct ovl_fs *ofs = OVL_FS(sb); |
f941866f AG |
695 | struct ovl_path origin = { }; |
696 | struct ovl_path *stack = &origin; | |
697 | struct dentry *dentry = NULL; | |
f71bd9cf | 698 | struct dentry *index = NULL; |
8b58924a | 699 | struct inode *inode; |
f941866f AG |
700 | int err; |
701 | ||
8b58924a AG |
702 | /* First lookup overlay inode in inode cache by origin fh */ |
703 | err = ovl_check_origin_fh(ofs, fh, false, NULL, &stack); | |
704 | if (err) | |
705 | return ERR_PTR(err); | |
706 | ||
707 | if (!d_is_dir(origin.dentry) || | |
708 | !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) { | |
709 | inode = ovl_lookup_inode(sb, origin.dentry, false); | |
710 | err = PTR_ERR(inode); | |
711 | if (IS_ERR(inode)) | |
712 | goto out_err; | |
713 | if (inode) { | |
714 | dentry = d_find_any_alias(inode); | |
715 | iput(inode); | |
716 | if (dentry) | |
717 | goto out; | |
718 | } | |
719 | } | |
720 | ||
721 | /* Then lookup indexed upper/whiteout by origin fh */ | |
02d70090 | 722 | if (ovl_indexdir(sb)) { |
f71bd9cf AG |
723 | index = ovl_get_index_fh(ofs, fh); |
724 | err = PTR_ERR(index); | |
9436a1a3 | 725 | if (IS_ERR(index)) { |
9436a1a3 | 726 | index = NULL; |
8b58924a | 727 | goto out_err; |
9436a1a3 | 728 | } |
f71bd9cf AG |
729 | } |
730 | ||
8b58924a | 731 | /* Then try to get a connected upper dir by index */ |
3b0bfc6e | 732 | if (index && d_is_dir(index)) { |
8ea28765 | 733 | struct dentry *upper = ovl_index_upper(ofs, index, true); |
3b0bfc6e AG |
734 | |
735 | err = PTR_ERR(upper); | |
736 | if (IS_ERR_OR_NULL(upper)) | |
737 | goto out_err; | |
738 | ||
739 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); | |
740 | dput(upper); | |
741 | goto out; | |
742 | } | |
743 | ||
155b8a04 AG |
744 | /* Find origin.dentry again with ovl_acceptable() layer check */ |
745 | if (d_is_dir(origin.dentry)) { | |
8b58924a AG |
746 | dput(origin.dentry); |
747 | origin.dentry = NULL; | |
748 | err = ovl_check_origin_fh(ofs, fh, true, NULL, &stack); | |
f71bd9cf AG |
749 | if (err) |
750 | goto out_err; | |
8b58924a AG |
751 | } |
752 | if (index) { | |
610afc0b | 753 | err = ovl_verify_origin(ofs, index, origin.dentry, false); |
8b58924a | 754 | if (err) |
9436a1a3 | 755 | goto out_err; |
f71bd9cf | 756 | } |
f941866f | 757 | |
155b8a04 | 758 | /* Get a connected non-upper dir or disconnected non-dir */ |
f71bd9cf | 759 | dentry = ovl_get_dentry(sb, NULL, &origin, index); |
f941866f | 760 | |
f71bd9cf AG |
761 | out: |
762 | dput(origin.dentry); | |
763 | dput(index); | |
f941866f | 764 | return dentry; |
f71bd9cf AG |
765 | |
766 | out_err: | |
767 | dentry = ERR_PTR(err); | |
768 | goto out; | |
f941866f AG |
769 | } |
770 | ||
cbe7fba8 AG |
771 | static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type) |
772 | { | |
773 | struct ovl_fh *fh; | |
774 | ||
775 | /* If on-wire inner fid is aligned - nothing to do */ | |
776 | if (fh_type == OVL_FILEID_V1) | |
777 | return (struct ovl_fh *)fid; | |
778 | ||
779 | if (fh_type != OVL_FILEID_V0) | |
780 | return ERR_PTR(-EINVAL); | |
781 | ||
9aafc1b0 DC |
782 | if (buflen <= OVL_FH_WIRE_OFFSET) |
783 | return ERR_PTR(-EINVAL); | |
784 | ||
cbe7fba8 AG |
785 | fh = kzalloc(buflen, GFP_KERNEL); |
786 | if (!fh) | |
787 | return ERR_PTR(-ENOMEM); | |
788 | ||
789 | /* Copy unaligned inner fh into aligned buffer */ | |
cf8aa9bf | 790 | memcpy(fh->buf, fid, buflen - OVL_FH_WIRE_OFFSET); |
cbe7fba8 AG |
791 | return fh; |
792 | } | |
793 | ||
8556a420 AG |
794 | static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid, |
795 | int fh_len, int fh_type) | |
796 | { | |
797 | struct dentry *dentry = NULL; | |
cbe7fba8 | 798 | struct ovl_fh *fh = NULL; |
8556a420 AG |
799 | int len = fh_len << 2; |
800 | unsigned int flags = 0; | |
801 | int err; | |
802 | ||
cbe7fba8 AG |
803 | fh = ovl_fid_to_fh(fid, len, fh_type); |
804 | err = PTR_ERR(fh); | |
805 | if (IS_ERR(fh)) | |
8556a420 AG |
806 | goto out_err; |
807 | ||
808 | err = ovl_check_fh_len(fh, len); | |
809 | if (err) | |
810 | goto out_err; | |
811 | ||
cbe7fba8 | 812 | flags = fh->fb.flags; |
f941866f AG |
813 | dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ? |
814 | ovl_upper_fh_to_d(sb, fh) : | |
815 | ovl_lower_fh_to_d(sb, fh); | |
8556a420 AG |
816 | err = PTR_ERR(dentry); |
817 | if (IS_ERR(dentry) && err != -ESTALE) | |
818 | goto out_err; | |
819 | ||
cbe7fba8 AG |
820 | out: |
821 | /* We may have needed to re-align OVL_FILEID_V0 */ | |
822 | if (!IS_ERR_OR_NULL(fh) && fh != (void *)fid) | |
823 | kfree(fh); | |
824 | ||
8556a420 AG |
825 | return dentry; |
826 | ||
827 | out_err: | |
1bd0a3ae | 828 | pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n", |
cbe7fba8 AG |
829 | fh_len, fh_type, flags, err); |
830 | dentry = ERR_PTR(err); | |
831 | goto out; | |
8556a420 AG |
832 | } |
833 | ||
3985b70a AG |
834 | static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid, |
835 | int fh_len, int fh_type) | |
836 | { | |
1bd0a3ae | 837 | pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n"); |
3985b70a AG |
838 | return ERR_PTR(-EACCES); |
839 | } | |
840 | ||
841 | static int ovl_get_name(struct dentry *parent, char *name, | |
842 | struct dentry *child) | |
843 | { | |
844 | /* | |
845 | * ovl_fh_to_dentry() returns connected dir overlay dentries and | |
846 | * ovl_fh_to_parent() is not implemented, so we should not get here. | |
847 | */ | |
848 | WARN_ON_ONCE(1); | |
849 | return -EIO; | |
850 | } | |
851 | ||
852 | static struct dentry *ovl_get_parent(struct dentry *dentry) | |
853 | { | |
854 | /* | |
855 | * ovl_fh_to_dentry() returns connected dir overlay dentries, so we | |
856 | * should not get here. | |
857 | */ | |
858 | WARN_ON_ONCE(1); | |
859 | return ERR_PTR(-EIO); | |
860 | } | |
861 | ||
8ed5eec9 | 862 | const struct export_operations ovl_export_operations = { |
5b2cccd3 | 863 | .encode_fh = ovl_encode_fh, |
8556a420 | 864 | .fh_to_dentry = ovl_fh_to_dentry, |
3985b70a AG |
865 | .fh_to_parent = ovl_fh_to_parent, |
866 | .get_name = ovl_get_name, | |
867 | .get_parent = ovl_get_parent, | |
8ed5eec9 | 868 | }; |
16aac5ad AG |
869 | |
870 | /* encode_fh() encodes non-decodable file handles with nfs_export=off */ | |
871 | const struct export_operations ovl_export_fid_operations = { | |
872 | .encode_fh = ovl_encode_fh, | |
873 | }; |