Commit | Line | Data |
---|---|---|
1e51764a AB |
1 | /* |
2 | * This file is part of UBIFS. | |
3 | * | |
4 | * Copyright (C) 2006-2008 Nokia Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License version 2 as published by | |
8 | * the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., 51 | |
17 | * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
18 | * | |
19 | * Authors: Adrian Hunter | |
20 | * Artem Bityutskiy (Битюцкий Артём) | |
21 | */ | |
22 | ||
23 | /* | |
24 | * This file implements TNC (Tree Node Cache) which caches indexing nodes of | |
25 | * the UBIFS B-tree. | |
26 | * | |
27 | * At the moment the locking rules of the TNC tree are quite simple and | |
28 | * straightforward. We just have a mutex and lock it when we traverse the | |
29 | * tree. If a znode is not in memory, we read it from flash while still having | |
30 | * the mutex locked. | |
31 | */ | |
32 | ||
33 | #include <linux/crc32.h> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
1e51764a AB |
35 | #include "ubifs.h" |
36 | ||
1cb51a15 | 37 | static int try_read_node(const struct ubifs_info *c, void *buf, int type, |
545bc8f6 | 38 | struct ubifs_zbranch *zbr); |
1cb51a15 RW |
39 | static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key, |
40 | struct ubifs_zbranch *zbr, void *node); | |
41 | ||
1e51764a AB |
42 | /* |
43 | * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions. | |
44 | * @NAME_LESS: name corresponding to the first argument is less than second | |
45 | * @NAME_MATCHES: names match | |
46 | * @NAME_GREATER: name corresponding to the second argument is greater than | |
47 | * first | |
48 | * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media | |
49 | * | |
50 | * These constants were introduce to improve readability. | |
51 | */ | |
52 | enum { | |
53 | NAME_LESS = 0, | |
54 | NAME_MATCHES = 1, | |
55 | NAME_GREATER = 2, | |
56 | NOT_ON_MEDIA = 3, | |
57 | }; | |
58 | ||
59 | /** | |
60 | * insert_old_idx - record an index node obsoleted since the last commit start. | |
61 | * @c: UBIFS file-system description object | |
62 | * @lnum: LEB number of obsoleted index node | |
63 | * @offs: offset of obsoleted index node | |
64 | * | |
65 | * Returns %0 on success, and a negative error code on failure. | |
66 | * | |
67 | * For recovery, there must always be a complete intact version of the index on | |
68 | * flash at all times. That is called the "old index". It is the index as at the | |
69 | * time of the last successful commit. Many of the index nodes in the old index | |
70 | * may be dirty, but they must not be erased until the next successful commit | |
71 | * (at which point that index becomes the old index). | |
72 | * | |
73 | * That means that the garbage collection and the in-the-gaps method of | |
74 | * committing must be able to determine if an index node is in the old index. | |
75 | * Most of the old index nodes can be found by looking up the TNC using the | |
76 | * 'lookup_znode()' function. However, some of the old index nodes may have | |
77 | * been deleted from the current index or may have been changed so much that | |
78 | * they cannot be easily found. In those cases, an entry is added to an RB-tree. | |
79 | * That is what this function does. The RB-tree is ordered by LEB number and | |
80 | * offset because they uniquely identify the old index node. | |
81 | */ | |
82 | static int insert_old_idx(struct ubifs_info *c, int lnum, int offs) | |
83 | { | |
84 | struct ubifs_old_idx *old_idx, *o; | |
85 | struct rb_node **p, *parent = NULL; | |
86 | ||
87 | old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS); | |
88 | if (unlikely(!old_idx)) | |
89 | return -ENOMEM; | |
90 | old_idx->lnum = lnum; | |
91 | old_idx->offs = offs; | |
92 | ||
93 | p = &c->old_idx.rb_node; | |
94 | while (*p) { | |
95 | parent = *p; | |
96 | o = rb_entry(parent, struct ubifs_old_idx, rb); | |
97 | if (lnum < o->lnum) | |
98 | p = &(*p)->rb_left; | |
99 | else if (lnum > o->lnum) | |
100 | p = &(*p)->rb_right; | |
101 | else if (offs < o->offs) | |
102 | p = &(*p)->rb_left; | |
103 | else if (offs > o->offs) | |
104 | p = &(*p)->rb_right; | |
105 | else { | |
235c362b | 106 | ubifs_err(c, "old idx added twice!"); |
1e51764a AB |
107 | kfree(old_idx); |
108 | return 0; | |
109 | } | |
110 | } | |
111 | rb_link_node(&old_idx->rb, parent, p); | |
112 | rb_insert_color(&old_idx->rb, &c->old_idx); | |
113 | return 0; | |
114 | } | |
115 | ||
116 | /** | |
117 | * insert_old_idx_znode - record a znode obsoleted since last commit start. | |
118 | * @c: UBIFS file-system description object | |
119 | * @znode: znode of obsoleted index node | |
120 | * | |
121 | * Returns %0 on success, and a negative error code on failure. | |
122 | */ | |
123 | int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode) | |
124 | { | |
125 | if (znode->parent) { | |
126 | struct ubifs_zbranch *zbr; | |
127 | ||
128 | zbr = &znode->parent->zbranch[znode->iip]; | |
129 | if (zbr->len) | |
130 | return insert_old_idx(c, zbr->lnum, zbr->offs); | |
131 | } else | |
132 | if (c->zroot.len) | |
133 | return insert_old_idx(c, c->zroot.lnum, | |
134 | c->zroot.offs); | |
135 | return 0; | |
136 | } | |
137 | ||
138 | /** | |
139 | * ins_clr_old_idx_znode - record a znode obsoleted since last commit start. | |
140 | * @c: UBIFS file-system description object | |
141 | * @znode: znode of obsoleted index node | |
142 | * | |
143 | * Returns %0 on success, and a negative error code on failure. | |
144 | */ | |
145 | static int ins_clr_old_idx_znode(struct ubifs_info *c, | |
146 | struct ubifs_znode *znode) | |
147 | { | |
148 | int err; | |
149 | ||
150 | if (znode->parent) { | |
151 | struct ubifs_zbranch *zbr; | |
152 | ||
153 | zbr = &znode->parent->zbranch[znode->iip]; | |
154 | if (zbr->len) { | |
155 | err = insert_old_idx(c, zbr->lnum, zbr->offs); | |
156 | if (err) | |
157 | return err; | |
158 | zbr->lnum = 0; | |
159 | zbr->offs = 0; | |
160 | zbr->len = 0; | |
161 | } | |
162 | } else | |
163 | if (c->zroot.len) { | |
164 | err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs); | |
165 | if (err) | |
166 | return err; | |
167 | c->zroot.lnum = 0; | |
168 | c->zroot.offs = 0; | |
169 | c->zroot.len = 0; | |
170 | } | |
171 | return 0; | |
172 | } | |
173 | ||
174 | /** | |
175 | * destroy_old_idx - destroy the old_idx RB-tree. | |
176 | * @c: UBIFS file-system description object | |
177 | * | |
178 | * During start commit, the old_idx RB-tree is used to avoid overwriting index | |
179 | * nodes that were in the index last commit but have since been deleted. This | |
180 | * is necessary for recovery i.e. the old index must be kept intact until the | |
181 | * new index is successfully written. The old-idx RB-tree is used for the | |
182 | * in-the-gaps method of writing index nodes and is destroyed every commit. | |
183 | */ | |
184 | void destroy_old_idx(struct ubifs_info *c) | |
185 | { | |
bb25e49f CS |
186 | struct ubifs_old_idx *old_idx, *n; |
187 | ||
188 | rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb) | |
1e51764a | 189 | kfree(old_idx); |
bb25e49f | 190 | |
1e51764a AB |
191 | c->old_idx = RB_ROOT; |
192 | } | |
193 | ||
194 | /** | |
195 | * copy_znode - copy a dirty znode. | |
196 | * @c: UBIFS file-system description object | |
197 | * @znode: znode to copy | |
198 | * | |
199 | * A dirty znode being committed may not be changed, so it is copied. | |
200 | */ | |
201 | static struct ubifs_znode *copy_znode(struct ubifs_info *c, | |
202 | struct ubifs_znode *znode) | |
203 | { | |
204 | struct ubifs_znode *zn; | |
205 | ||
bbc8a004 | 206 | zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS); |
1e51764a AB |
207 | if (unlikely(!zn)) |
208 | return ERR_PTR(-ENOMEM); | |
209 | ||
1e51764a AB |
210 | zn->cnext = NULL; |
211 | __set_bit(DIRTY_ZNODE, &zn->flags); | |
212 | __clear_bit(COW_ZNODE, &zn->flags); | |
213 | ||
6eb61d58 | 214 | ubifs_assert(c, !ubifs_zn_obsolete(znode)); |
1e51764a AB |
215 | __set_bit(OBSOLETE_ZNODE, &znode->flags); |
216 | ||
217 | if (znode->level != 0) { | |
218 | int i; | |
219 | const int n = zn->child_cnt; | |
220 | ||
221 | /* The children now have new parent */ | |
222 | for (i = 0; i < n; i++) { | |
223 | struct ubifs_zbranch *zbr = &zn->zbranch[i]; | |
224 | ||
225 | if (zbr->znode) | |
226 | zbr->znode->parent = zn; | |
227 | } | |
228 | } | |
229 | ||
230 | atomic_long_inc(&c->dirty_zn_cnt); | |
231 | return zn; | |
232 | } | |
233 | ||
234 | /** | |
235 | * add_idx_dirt - add dirt due to a dirty znode. | |
236 | * @c: UBIFS file-system description object | |
237 | * @lnum: LEB number of index node | |
238 | * @dirt: size of index node | |
239 | * | |
240 | * This function updates lprops dirty space and the new size of the index. | |
241 | */ | |
242 | static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt) | |
243 | { | |
244 | c->calc_idx_sz -= ALIGN(dirt, 8); | |
245 | return ubifs_add_dirt(c, lnum, dirt); | |
246 | } | |
247 | ||
248 | /** | |
249 | * dirty_cow_znode - ensure a znode is not being committed. | |
250 | * @c: UBIFS file-system description object | |
251 | * @zbr: branch of znode to check | |
252 | * | |
253 | * Returns dirtied znode on success or negative error code on failure. | |
254 | */ | |
255 | static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c, | |
256 | struct ubifs_zbranch *zbr) | |
257 | { | |
258 | struct ubifs_znode *znode = zbr->znode; | |
259 | struct ubifs_znode *zn; | |
260 | int err; | |
261 | ||
f42eed7c | 262 | if (!ubifs_zn_cow(znode)) { |
1e51764a AB |
263 | /* znode is not being committed */ |
264 | if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) { | |
265 | atomic_long_inc(&c->dirty_zn_cnt); | |
266 | atomic_long_dec(&c->clean_zn_cnt); | |
267 | atomic_long_dec(&ubifs_clean_zn_cnt); | |
268 | err = add_idx_dirt(c, zbr->lnum, zbr->len); | |
269 | if (unlikely(err)) | |
270 | return ERR_PTR(err); | |
271 | } | |
272 | return znode; | |
273 | } | |
274 | ||
275 | zn = copy_znode(c, znode); | |
8d47aef4 | 276 | if (IS_ERR(zn)) |
1e51764a AB |
277 | return zn; |
278 | ||
279 | if (zbr->len) { | |
280 | err = insert_old_idx(c, zbr->lnum, zbr->offs); | |
281 | if (unlikely(err)) | |
282 | return ERR_PTR(err); | |
283 | err = add_idx_dirt(c, zbr->lnum, zbr->len); | |
284 | } else | |
285 | err = 0; | |
286 | ||
287 | zbr->znode = zn; | |
288 | zbr->lnum = 0; | |
289 | zbr->offs = 0; | |
290 | zbr->len = 0; | |
291 | ||
292 | if (unlikely(err)) | |
293 | return ERR_PTR(err); | |
294 | return zn; | |
295 | } | |
296 | ||
297 | /** | |
298 | * lnc_add - add a leaf node to the leaf node cache. | |
299 | * @c: UBIFS file-system description object | |
300 | * @zbr: zbranch of leaf node | |
301 | * @node: leaf node | |
302 | * | |
303 | * Leaf nodes are non-index nodes directory entry nodes or data nodes. The | |
304 | * purpose of the leaf node cache is to save re-reading the same leaf node over | |
305 | * and over again. Most things are cached by VFS, however the file system must | |
306 | * cache directory entries for readdir and for resolving hash collisions. The | |
307 | * present implementation of the leaf node cache is extremely simple, and | |
308 | * allows for error returns that are not used but that may be needed if a more | |
309 | * complex implementation is created. | |
310 | * | |
311 | * Note, this function does not add the @node object to LNC directly, but | |
312 | * allocates a copy of the object and adds the copy to LNC. The reason for this | |
313 | * is that @node has been allocated outside of the TNC subsystem and will be | |
314 | * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC | |
315 | * may be changed at any time, e.g. freed by the shrinker. | |
316 | */ | |
317 | static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
318 | const void *node) | |
319 | { | |
320 | int err; | |
321 | void *lnc_node; | |
322 | const struct ubifs_dent_node *dent = node; | |
323 | ||
6eb61d58 RW |
324 | ubifs_assert(c, !zbr->leaf); |
325 | ubifs_assert(c, zbr->len != 0); | |
326 | ubifs_assert(c, is_hash_key(c, &zbr->key)); | |
1e51764a AB |
327 | |
328 | err = ubifs_validate_entry(c, dent); | |
329 | if (err) { | |
7c46d0ae | 330 | dump_stack(); |
edf6be24 | 331 | ubifs_dump_node(c, dent); |
1e51764a AB |
332 | return err; |
333 | } | |
334 | ||
eaecf43a | 335 | lnc_node = kmemdup(node, zbr->len, GFP_NOFS); |
1e51764a AB |
336 | if (!lnc_node) |
337 | /* We don't have to have the cache, so no error */ | |
338 | return 0; | |
339 | ||
1e51764a AB |
340 | zbr->leaf = lnc_node; |
341 | return 0; | |
342 | } | |
343 | ||
344 | /** | |
345 | * lnc_add_directly - add a leaf node to the leaf-node-cache. | |
346 | * @c: UBIFS file-system description object | |
347 | * @zbr: zbranch of leaf node | |
348 | * @node: leaf node | |
349 | * | |
350 | * This function is similar to 'lnc_add()', but it does not create a copy of | |
351 | * @node but inserts @node to TNC directly. | |
352 | */ | |
353 | static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
354 | void *node) | |
355 | { | |
356 | int err; | |
357 | ||
6eb61d58 RW |
358 | ubifs_assert(c, !zbr->leaf); |
359 | ubifs_assert(c, zbr->len != 0); | |
1e51764a AB |
360 | |
361 | err = ubifs_validate_entry(c, node); | |
362 | if (err) { | |
7c46d0ae | 363 | dump_stack(); |
edf6be24 | 364 | ubifs_dump_node(c, node); |
1e51764a AB |
365 | return err; |
366 | } | |
367 | ||
368 | zbr->leaf = node; | |
369 | return 0; | |
370 | } | |
371 | ||
372 | /** | |
373 | * lnc_free - remove a leaf node from the leaf node cache. | |
374 | * @zbr: zbranch of leaf node | |
375 | * @node: leaf node | |
376 | */ | |
377 | static void lnc_free(struct ubifs_zbranch *zbr) | |
378 | { | |
379 | if (!zbr->leaf) | |
380 | return; | |
381 | kfree(zbr->leaf); | |
382 | zbr->leaf = NULL; | |
383 | } | |
384 | ||
385 | /** | |
b91dc981 | 386 | * tnc_read_hashed_node - read a "hashed" leaf node. |
1e51764a AB |
387 | * @c: UBIFS file-system description object |
388 | * @zbr: key and position of the node | |
389 | * @node: node is returned here | |
390 | * | |
391 | * This function reads a "hashed" node defined by @zbr from the leaf node cache | |
392 | * (in it is there) or from the hash media, in which case the node is also | |
393 | * added to LNC. Returns zero in case of success or a negative negative error | |
394 | * code in case of failure. | |
395 | */ | |
b91dc981 RW |
396 | static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr, |
397 | void *node) | |
1e51764a AB |
398 | { |
399 | int err; | |
400 | ||
6eb61d58 | 401 | ubifs_assert(c, is_hash_key(c, &zbr->key)); |
1e51764a AB |
402 | |
403 | if (zbr->leaf) { | |
404 | /* Read from the leaf node cache */ | |
6eb61d58 | 405 | ubifs_assert(c, zbr->len != 0); |
1e51764a AB |
406 | memcpy(node, zbr->leaf, zbr->len); |
407 | return 0; | |
408 | } | |
409 | ||
1cb51a15 RW |
410 | if (c->replaying) { |
411 | err = fallible_read_node(c, &zbr->key, zbr, node); | |
412 | /* | |
413 | * When the node was not found, return -ENOENT, 0 otherwise. | |
414 | * Negative return codes stay as-is. | |
415 | */ | |
416 | if (err == 0) | |
417 | err = -ENOENT; | |
418 | else if (err == 1) | |
419 | err = 0; | |
420 | } else { | |
421 | err = ubifs_tnc_read_node(c, zbr, node); | |
422 | } | |
1e51764a AB |
423 | if (err) |
424 | return err; | |
425 | ||
426 | /* Add the node to the leaf node cache */ | |
427 | err = lnc_add(c, zbr, node); | |
428 | return err; | |
429 | } | |
430 | ||
431 | /** | |
432 | * try_read_node - read a node if it is a node. | |
433 | * @c: UBIFS file-system description object | |
434 | * @buf: buffer to read to | |
435 | * @type: node type | |
545bc8f6 | 436 | * @zbr: the zbranch describing the node to read |
1e51764a AB |
437 | * |
438 | * This function tries to read a node of known type and length, checks it and | |
439 | * stores it in @buf. This function returns %1 if a node is present and %0 if | |
440 | * a node is not present. A negative error code is returned for I/O errors. | |
441 | * This function performs that same function as ubifs_read_node except that | |
442 | * it does not require that there is actually a node present and instead | |
443 | * the return code indicates if a node was read. | |
6f7ab6d4 AB |
444 | * |
445 | * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc | |
446 | * is true (it is controlled by corresponding mount option). However, if | |
18d1d7fb AB |
447 | * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to |
448 | * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is | |
449 | * because during mounting or re-mounting from R/O mode to R/W mode we may read | |
450 | * journal nodes (when replying the journal or doing the recovery) and the | |
451 | * journal nodes may potentially be corrupted, so checking is required. | |
1e51764a AB |
452 | */ |
453 | static int try_read_node(const struct ubifs_info *c, void *buf, int type, | |
545bc8f6 | 454 | struct ubifs_zbranch *zbr) |
1e51764a | 455 | { |
545bc8f6 SH |
456 | int len = zbr->len; |
457 | int lnum = zbr->lnum; | |
458 | int offs = zbr->offs; | |
1e51764a AB |
459 | int err, node_len; |
460 | struct ubifs_ch *ch = buf; | |
461 | uint32_t crc, node_crc; | |
462 | ||
463 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
464 | ||
d304820a | 465 | err = ubifs_leb_read(c, lnum, buf, offs, len, 1); |
1e51764a | 466 | if (err) { |
235c362b | 467 | ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d", |
1e51764a AB |
468 | type, lnum, offs, err); |
469 | return err; | |
470 | } | |
471 | ||
472 | if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) | |
473 | return 0; | |
474 | ||
475 | if (ch->node_type != type) | |
476 | return 0; | |
477 | ||
478 | node_len = le32_to_cpu(ch->len); | |
479 | if (node_len != len) | |
480 | return 0; | |
481 | ||
e9cd7dfd SH |
482 | if (type != UBIFS_DATA_NODE || !c->no_chk_data_crc || c->mounting || |
483 | c->remounting_rw) { | |
484 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8); | |
485 | node_crc = le32_to_cpu(ch->crc); | |
486 | if (crc != node_crc) | |
487 | return 0; | |
488 | } | |
1e51764a | 489 | |
16a26b20 SH |
490 | err = ubifs_node_check_hash(c, buf, zbr->hash); |
491 | if (err) { | |
492 | ubifs_bad_hash(c, buf, zbr->hash, lnum, offs); | |
493 | return 0; | |
494 | } | |
495 | ||
1e51764a AB |
496 | return 1; |
497 | } | |
498 | ||
499 | /** | |
500 | * fallible_read_node - try to read a leaf node. | |
501 | * @c: UBIFS file-system description object | |
502 | * @key: key of node to read | |
503 | * @zbr: position of node | |
504 | * @node: node returned | |
505 | * | |
506 | * This function tries to read a node and returns %1 if the node is read, %0 | |
507 | * if the node is not present, and a negative error code in the case of error. | |
508 | */ | |
509 | static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key, | |
510 | struct ubifs_zbranch *zbr, void *node) | |
511 | { | |
512 | int ret; | |
513 | ||
515315a1 | 514 | dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs); |
1e51764a | 515 | |
545bc8f6 | 516 | ret = try_read_node(c, node, key_type(c, key), zbr); |
1e51764a AB |
517 | if (ret == 1) { |
518 | union ubifs_key node_key; | |
519 | struct ubifs_dent_node *dent = node; | |
520 | ||
521 | /* All nodes have key in the same place */ | |
522 | key_read(c, &dent->key, &node_key); | |
523 | if (keys_cmp(c, key, &node_key) != 0) | |
524 | ret = 0; | |
525 | } | |
601c0bc4 | 526 | if (ret == 0 && c->replaying) |
515315a1 AB |
527 | dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ", |
528 | zbr->lnum, zbr->offs, zbr->len); | |
1e51764a AB |
529 | return ret; |
530 | } | |
531 | ||
532 | /** | |
533 | * matches_name - determine if a direntry or xattr entry matches a given name. | |
534 | * @c: UBIFS file-system description object | |
535 | * @zbr: zbranch of dent | |
536 | * @nm: name to match | |
537 | * | |
538 | * This function checks if xentry/direntry referred by zbranch @zbr matches name | |
539 | * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by | |
540 | * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case | |
541 | * of failure, a negative error code is returned. | |
542 | */ | |
543 | static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
f4f61d2c | 544 | const struct fscrypt_name *nm) |
1e51764a AB |
545 | { |
546 | struct ubifs_dent_node *dent; | |
547 | int nlen, err; | |
548 | ||
549 | /* If possible, match against the dent in the leaf node cache */ | |
550 | if (!zbr->leaf) { | |
551 | dent = kmalloc(zbr->len, GFP_NOFS); | |
552 | if (!dent) | |
553 | return -ENOMEM; | |
554 | ||
555 | err = ubifs_tnc_read_node(c, zbr, dent); | |
556 | if (err) | |
557 | goto out_free; | |
558 | ||
559 | /* Add the node to the leaf node cache */ | |
560 | err = lnc_add_directly(c, zbr, dent); | |
561 | if (err) | |
562 | goto out_free; | |
563 | } else | |
564 | dent = zbr->leaf; | |
565 | ||
566 | nlen = le16_to_cpu(dent->nlen); | |
f4f61d2c | 567 | err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm))); |
1e51764a | 568 | if (err == 0) { |
f4f61d2c | 569 | if (nlen == fname_len(nm)) |
1e51764a | 570 | return NAME_MATCHES; |
f4f61d2c | 571 | else if (nlen < fname_len(nm)) |
1e51764a AB |
572 | return NAME_LESS; |
573 | else | |
574 | return NAME_GREATER; | |
575 | } else if (err < 0) | |
576 | return NAME_LESS; | |
577 | else | |
578 | return NAME_GREATER; | |
579 | ||
580 | out_free: | |
581 | kfree(dent); | |
582 | return err; | |
583 | } | |
584 | ||
585 | /** | |
586 | * get_znode - get a TNC znode that may not be loaded yet. | |
587 | * @c: UBIFS file-system description object | |
588 | * @znode: parent znode | |
589 | * @n: znode branch slot number | |
590 | * | |
591 | * This function returns the znode or a negative error code. | |
592 | */ | |
593 | static struct ubifs_znode *get_znode(struct ubifs_info *c, | |
594 | struct ubifs_znode *znode, int n) | |
595 | { | |
596 | struct ubifs_zbranch *zbr; | |
597 | ||
598 | zbr = &znode->zbranch[n]; | |
599 | if (zbr->znode) | |
600 | znode = zbr->znode; | |
601 | else | |
602 | znode = ubifs_load_znode(c, zbr, znode, n); | |
603 | return znode; | |
604 | } | |
605 | ||
606 | /** | |
607 | * tnc_next - find next TNC entry. | |
608 | * @c: UBIFS file-system description object | |
609 | * @zn: znode is passed and returned here | |
610 | * @n: znode branch slot number is passed and returned here | |
611 | * | |
612 | * This function returns %0 if the next TNC entry is found, %-ENOENT if there is | |
613 | * no next entry, or a negative error code otherwise. | |
614 | */ | |
615 | static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n) | |
616 | { | |
617 | struct ubifs_znode *znode = *zn; | |
618 | int nn = *n; | |
619 | ||
620 | nn += 1; | |
621 | if (nn < znode->child_cnt) { | |
622 | *n = nn; | |
623 | return 0; | |
624 | } | |
625 | while (1) { | |
626 | struct ubifs_znode *zp; | |
627 | ||
628 | zp = znode->parent; | |
629 | if (!zp) | |
630 | return -ENOENT; | |
631 | nn = znode->iip + 1; | |
632 | znode = zp; | |
633 | if (nn < znode->child_cnt) { | |
634 | znode = get_znode(c, znode, nn); | |
635 | if (IS_ERR(znode)) | |
636 | return PTR_ERR(znode); | |
637 | while (znode->level != 0) { | |
638 | znode = get_znode(c, znode, 0); | |
639 | if (IS_ERR(znode)) | |
640 | return PTR_ERR(znode); | |
641 | } | |
642 | nn = 0; | |
643 | break; | |
644 | } | |
645 | } | |
646 | *zn = znode; | |
647 | *n = nn; | |
648 | return 0; | |
649 | } | |
650 | ||
651 | /** | |
652 | * tnc_prev - find previous TNC entry. | |
653 | * @c: UBIFS file-system description object | |
654 | * @zn: znode is returned here | |
655 | * @n: znode branch slot number is passed and returned here | |
656 | * | |
657 | * This function returns %0 if the previous TNC entry is found, %-ENOENT if | |
658 | * there is no next entry, or a negative error code otherwise. | |
659 | */ | |
660 | static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n) | |
661 | { | |
662 | struct ubifs_znode *znode = *zn; | |
663 | int nn = *n; | |
664 | ||
665 | if (nn > 0) { | |
666 | *n = nn - 1; | |
667 | return 0; | |
668 | } | |
669 | while (1) { | |
670 | struct ubifs_znode *zp; | |
671 | ||
672 | zp = znode->parent; | |
673 | if (!zp) | |
674 | return -ENOENT; | |
675 | nn = znode->iip - 1; | |
676 | znode = zp; | |
677 | if (nn >= 0) { | |
678 | znode = get_znode(c, znode, nn); | |
679 | if (IS_ERR(znode)) | |
680 | return PTR_ERR(znode); | |
681 | while (znode->level != 0) { | |
682 | nn = znode->child_cnt - 1; | |
683 | znode = get_znode(c, znode, nn); | |
684 | if (IS_ERR(znode)) | |
685 | return PTR_ERR(znode); | |
686 | } | |
687 | nn = znode->child_cnt - 1; | |
688 | break; | |
689 | } | |
690 | } | |
691 | *zn = znode; | |
692 | *n = nn; | |
693 | return 0; | |
694 | } | |
695 | ||
696 | /** | |
697 | * resolve_collision - resolve a collision. | |
698 | * @c: UBIFS file-system description object | |
699 | * @key: key of a directory or extended attribute entry | |
700 | * @zn: znode is returned here | |
701 | * @n: zbranch number is passed and returned here | |
702 | * @nm: name of the entry | |
703 | * | |
704 | * This function is called for "hashed" keys to make sure that the found key | |
705 | * really corresponds to the looked up node (directory or extended attribute | |
706 | * entry). It returns %1 and sets @zn and @n if the collision is resolved. | |
707 | * %0 is returned if @nm is not found and @zn and @n are set to the previous | |
708 | * entry, i.e. to the entry after which @nm could follow if it were in TNC. | |
709 | * This means that @n may be set to %-1 if the leftmost key in @zn is the | |
710 | * previous one. A negative error code is returned on failures. | |
711 | */ | |
712 | static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key, | |
713 | struct ubifs_znode **zn, int *n, | |
f4f61d2c | 714 | const struct fscrypt_name *nm) |
1e51764a AB |
715 | { |
716 | int err; | |
717 | ||
718 | err = matches_name(c, &(*zn)->zbranch[*n], nm); | |
719 | if (unlikely(err < 0)) | |
720 | return err; | |
721 | if (err == NAME_MATCHES) | |
722 | return 1; | |
723 | ||
724 | if (err == NAME_GREATER) { | |
725 | /* Look left */ | |
726 | while (1) { | |
727 | err = tnc_prev(c, zn, n); | |
728 | if (err == -ENOENT) { | |
6eb61d58 | 729 | ubifs_assert(c, *n == 0); |
1e51764a AB |
730 | *n = -1; |
731 | return 0; | |
732 | } | |
733 | if (err < 0) | |
734 | return err; | |
735 | if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { | |
736 | /* | |
737 | * We have found the branch after which we would | |
738 | * like to insert, but inserting in this znode | |
739 | * may still be wrong. Consider the following 3 | |
740 | * znodes, in the case where we are resolving a | |
741 | * collision with Key2. | |
742 | * | |
743 | * znode zp | |
744 | * ---------------------- | |
745 | * level 1 | Key0 | Key1 | | |
746 | * ----------------------- | |
747 | * | | | |
748 | * znode za | | znode zb | |
749 | * ------------ ------------ | |
750 | * level 0 | Key0 | | Key2 | | |
751 | * ------------ ------------ | |
752 | * | |
753 | * The lookup finds Key2 in znode zb. Lets say | |
754 | * there is no match and the name is greater so | |
755 | * we look left. When we find Key0, we end up | |
756 | * here. If we return now, we will insert into | |
757 | * znode za at slot n = 1. But that is invalid | |
758 | * according to the parent's keys. Key2 must | |
759 | * be inserted into znode zb. | |
760 | * | |
761 | * Note, this problem is not relevant for the | |
762 | * case when we go right, because | |
763 | * 'tnc_insert()' would correct the parent key. | |
764 | */ | |
765 | if (*n == (*zn)->child_cnt - 1) { | |
766 | err = tnc_next(c, zn, n); | |
767 | if (err) { | |
768 | /* Should be impossible */ | |
6eb61d58 | 769 | ubifs_assert(c, 0); |
1e51764a AB |
770 | if (err == -ENOENT) |
771 | err = -EINVAL; | |
772 | return err; | |
773 | } | |
6eb61d58 | 774 | ubifs_assert(c, *n == 0); |
1e51764a AB |
775 | *n = -1; |
776 | } | |
777 | return 0; | |
778 | } | |
779 | err = matches_name(c, &(*zn)->zbranch[*n], nm); | |
780 | if (err < 0) | |
781 | return err; | |
782 | if (err == NAME_LESS) | |
783 | return 0; | |
784 | if (err == NAME_MATCHES) | |
785 | return 1; | |
6eb61d58 | 786 | ubifs_assert(c, err == NAME_GREATER); |
1e51764a AB |
787 | } |
788 | } else { | |
789 | int nn = *n; | |
790 | struct ubifs_znode *znode = *zn; | |
791 | ||
792 | /* Look right */ | |
793 | while (1) { | |
794 | err = tnc_next(c, &znode, &nn); | |
795 | if (err == -ENOENT) | |
796 | return 0; | |
797 | if (err < 0) | |
798 | return err; | |
799 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
800 | return 0; | |
801 | err = matches_name(c, &znode->zbranch[nn], nm); | |
802 | if (err < 0) | |
803 | return err; | |
804 | if (err == NAME_GREATER) | |
805 | return 0; | |
806 | *zn = znode; | |
807 | *n = nn; | |
808 | if (err == NAME_MATCHES) | |
809 | return 1; | |
6eb61d58 | 810 | ubifs_assert(c, err == NAME_LESS); |
1e51764a AB |
811 | } |
812 | } | |
813 | } | |
814 | ||
815 | /** | |
816 | * fallible_matches_name - determine if a dent matches a given name. | |
817 | * @c: UBIFS file-system description object | |
818 | * @zbr: zbranch of dent | |
819 | * @nm: name to match | |
820 | * | |
821 | * This is a "fallible" version of 'matches_name()' function which does not | |
822 | * panic if the direntry/xentry referred by @zbr does not exist on the media. | |
823 | * | |
824 | * This function checks if xentry/direntry referred by zbranch @zbr matches name | |
825 | * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr | |
826 | * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA | |
827 | * if xentry/direntry referred by @zbr does not exist on the media. A negative | |
828 | * error code is returned in case of failure. | |
829 | */ | |
830 | static int fallible_matches_name(struct ubifs_info *c, | |
831 | struct ubifs_zbranch *zbr, | |
f4f61d2c | 832 | const struct fscrypt_name *nm) |
1e51764a AB |
833 | { |
834 | struct ubifs_dent_node *dent; | |
835 | int nlen, err; | |
836 | ||
837 | /* If possible, match against the dent in the leaf node cache */ | |
838 | if (!zbr->leaf) { | |
839 | dent = kmalloc(zbr->len, GFP_NOFS); | |
840 | if (!dent) | |
841 | return -ENOMEM; | |
842 | ||
843 | err = fallible_read_node(c, &zbr->key, zbr, dent); | |
844 | if (err < 0) | |
845 | goto out_free; | |
846 | if (err == 0) { | |
847 | /* The node was not present */ | |
848 | err = NOT_ON_MEDIA; | |
849 | goto out_free; | |
850 | } | |
6eb61d58 | 851 | ubifs_assert(c, err == 1); |
1e51764a AB |
852 | |
853 | err = lnc_add_directly(c, zbr, dent); | |
854 | if (err) | |
855 | goto out_free; | |
856 | } else | |
857 | dent = zbr->leaf; | |
858 | ||
859 | nlen = le16_to_cpu(dent->nlen); | |
f4f61d2c | 860 | err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm))); |
1e51764a | 861 | if (err == 0) { |
f4f61d2c | 862 | if (nlen == fname_len(nm)) |
1e51764a | 863 | return NAME_MATCHES; |
f4f61d2c | 864 | else if (nlen < fname_len(nm)) |
1e51764a AB |
865 | return NAME_LESS; |
866 | else | |
867 | return NAME_GREATER; | |
868 | } else if (err < 0) | |
869 | return NAME_LESS; | |
870 | else | |
871 | return NAME_GREATER; | |
872 | ||
873 | out_free: | |
874 | kfree(dent); | |
875 | return err; | |
876 | } | |
877 | ||
878 | /** | |
879 | * fallible_resolve_collision - resolve a collision even if nodes are missing. | |
880 | * @c: UBIFS file-system description object | |
881 | * @key: key | |
882 | * @zn: znode is returned here | |
883 | * @n: branch number is passed and returned here | |
884 | * @nm: name of directory entry | |
885 | * @adding: indicates caller is adding a key to the TNC | |
886 | * | |
887 | * This is a "fallible" version of the 'resolve_collision()' function which | |
888 | * does not panic if one of the nodes referred to by TNC does not exist on the | |
889 | * media. This may happen when replaying the journal if a deleted node was | |
890 | * Garbage-collected and the commit was not done. A branch that refers to a node | |
891 | * that is not present is called a dangling branch. The following are the return | |
892 | * codes for this function: | |
893 | * o if @nm was found, %1 is returned and @zn and @n are set to the found | |
894 | * branch; | |
895 | * o if we are @adding and @nm was not found, %0 is returned; | |
896 | * o if we are not @adding and @nm was not found, but a dangling branch was | |
897 | * found, then %1 is returned and @zn and @n are set to the dangling branch; | |
898 | * o a negative error code is returned in case of failure. | |
899 | */ | |
900 | static int fallible_resolve_collision(struct ubifs_info *c, | |
901 | const union ubifs_key *key, | |
902 | struct ubifs_znode **zn, int *n, | |
f4f61d2c RW |
903 | const struct fscrypt_name *nm, |
904 | int adding) | |
1e51764a AB |
905 | { |
906 | struct ubifs_znode *o_znode = NULL, *znode = *zn; | |
907 | int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n; | |
908 | ||
909 | cmp = fallible_matches_name(c, &znode->zbranch[nn], nm); | |
910 | if (unlikely(cmp < 0)) | |
911 | return cmp; | |
912 | if (cmp == NAME_MATCHES) | |
913 | return 1; | |
914 | if (cmp == NOT_ON_MEDIA) { | |
915 | o_znode = znode; | |
916 | o_n = nn; | |
917 | /* | |
918 | * We are unlucky and hit a dangling branch straight away. | |
919 | * Now we do not really know where to go to find the needed | |
920 | * branch - to the left or to the right. Well, let's try left. | |
921 | */ | |
922 | unsure = 1; | |
923 | } else if (!adding) | |
924 | unsure = 1; /* Remove a dangling branch wherever it is */ | |
925 | ||
926 | if (cmp == NAME_GREATER || unsure) { | |
927 | /* Look left */ | |
928 | while (1) { | |
929 | err = tnc_prev(c, zn, n); | |
930 | if (err == -ENOENT) { | |
6eb61d58 | 931 | ubifs_assert(c, *n == 0); |
1e51764a AB |
932 | *n = -1; |
933 | break; | |
934 | } | |
935 | if (err < 0) | |
936 | return err; | |
937 | if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { | |
938 | /* See comments in 'resolve_collision()' */ | |
939 | if (*n == (*zn)->child_cnt - 1) { | |
940 | err = tnc_next(c, zn, n); | |
941 | if (err) { | |
942 | /* Should be impossible */ | |
6eb61d58 | 943 | ubifs_assert(c, 0); |
1e51764a AB |
944 | if (err == -ENOENT) |
945 | err = -EINVAL; | |
946 | return err; | |
947 | } | |
6eb61d58 | 948 | ubifs_assert(c, *n == 0); |
1e51764a AB |
949 | *n = -1; |
950 | } | |
951 | break; | |
952 | } | |
953 | err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm); | |
954 | if (err < 0) | |
955 | return err; | |
956 | if (err == NAME_MATCHES) | |
957 | return 1; | |
958 | if (err == NOT_ON_MEDIA) { | |
959 | o_znode = *zn; | |
960 | o_n = *n; | |
961 | continue; | |
962 | } | |
963 | if (!adding) | |
964 | continue; | |
965 | if (err == NAME_LESS) | |
966 | break; | |
967 | else | |
968 | unsure = 0; | |
969 | } | |
970 | } | |
971 | ||
972 | if (cmp == NAME_LESS || unsure) { | |
973 | /* Look right */ | |
974 | *zn = znode; | |
975 | *n = nn; | |
976 | while (1) { | |
977 | err = tnc_next(c, &znode, &nn); | |
978 | if (err == -ENOENT) | |
979 | break; | |
980 | if (err < 0) | |
981 | return err; | |
982 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
983 | break; | |
984 | err = fallible_matches_name(c, &znode->zbranch[nn], nm); | |
985 | if (err < 0) | |
986 | return err; | |
987 | if (err == NAME_GREATER) | |
988 | break; | |
989 | *zn = znode; | |
990 | *n = nn; | |
991 | if (err == NAME_MATCHES) | |
992 | return 1; | |
993 | if (err == NOT_ON_MEDIA) { | |
994 | o_znode = znode; | |
995 | o_n = nn; | |
996 | } | |
997 | } | |
998 | } | |
999 | ||
1000 | /* Never match a dangling branch when adding */ | |
1001 | if (adding || !o_znode) | |
1002 | return 0; | |
1003 | ||
515315a1 | 1004 | dbg_mntk(key, "dangling match LEB %d:%d len %d key ", |
1e51764a | 1005 | o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs, |
515315a1 | 1006 | o_znode->zbranch[o_n].len); |
1e51764a AB |
1007 | *zn = o_znode; |
1008 | *n = o_n; | |
1009 | return 1; | |
1010 | } | |
1011 | ||
1012 | /** | |
1013 | * matches_position - determine if a zbranch matches a given position. | |
1014 | * @zbr: zbranch of dent | |
1015 | * @lnum: LEB number of dent to match | |
1016 | * @offs: offset of dent to match | |
1017 | * | |
1018 | * This function returns %1 if @lnum:@offs matches, and %0 otherwise. | |
1019 | */ | |
1020 | static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs) | |
1021 | { | |
1022 | if (zbr->lnum == lnum && zbr->offs == offs) | |
1023 | return 1; | |
1024 | else | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | /** | |
1029 | * resolve_collision_directly - resolve a collision directly. | |
1030 | * @c: UBIFS file-system description object | |
1031 | * @key: key of directory entry | |
1032 | * @zn: znode is passed and returned here | |
1033 | * @n: zbranch number is passed and returned here | |
1034 | * @lnum: LEB number of dent node to match | |
1035 | * @offs: offset of dent node to match | |
1036 | * | |
1037 | * This function is used for "hashed" keys to make sure the found directory or | |
1038 | * extended attribute entry node is what was looked for. It is used when the | |
1039 | * flash address of the right node is known (@lnum:@offs) which makes it much | |
1040 | * easier to resolve collisions (no need to read entries and match full | |
1041 | * names). This function returns %1 and sets @zn and @n if the collision is | |
1042 | * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the | |
1043 | * previous directory entry. Otherwise a negative error code is returned. | |
1044 | */ | |
1045 | static int resolve_collision_directly(struct ubifs_info *c, | |
1046 | const union ubifs_key *key, | |
1047 | struct ubifs_znode **zn, int *n, | |
1048 | int lnum, int offs) | |
1049 | { | |
1050 | struct ubifs_znode *znode; | |
1051 | int nn, err; | |
1052 | ||
1053 | znode = *zn; | |
1054 | nn = *n; | |
1055 | if (matches_position(&znode->zbranch[nn], lnum, offs)) | |
1056 | return 1; | |
1057 | ||
1058 | /* Look left */ | |
1059 | while (1) { | |
1060 | err = tnc_prev(c, &znode, &nn); | |
1061 | if (err == -ENOENT) | |
1062 | break; | |
1063 | if (err < 0) | |
1064 | return err; | |
1065 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
1066 | break; | |
1067 | if (matches_position(&znode->zbranch[nn], lnum, offs)) { | |
1068 | *zn = znode; | |
1069 | *n = nn; | |
1070 | return 1; | |
1071 | } | |
1072 | } | |
1073 | ||
1074 | /* Look right */ | |
1075 | znode = *zn; | |
1076 | nn = *n; | |
1077 | while (1) { | |
1078 | err = tnc_next(c, &znode, &nn); | |
1079 | if (err == -ENOENT) | |
1080 | return 0; | |
1081 | if (err < 0) | |
1082 | return err; | |
1083 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
1084 | return 0; | |
1085 | *zn = znode; | |
1086 | *n = nn; | |
1087 | if (matches_position(&znode->zbranch[nn], lnum, offs)) | |
1088 | return 1; | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | /** | |
1093 | * dirty_cow_bottom_up - dirty a znode and its ancestors. | |
1094 | * @c: UBIFS file-system description object | |
1095 | * @znode: znode to dirty | |
1096 | * | |
1097 | * If we do not have a unique key that resides in a znode, then we cannot | |
1098 | * dirty that znode from the top down (i.e. by using lookup_level0_dirty) | |
1099 | * This function records the path back to the last dirty ancestor, and then | |
1100 | * dirties the znodes on that path. | |
1101 | */ | |
1102 | static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c, | |
1103 | struct ubifs_znode *znode) | |
1104 | { | |
1105 | struct ubifs_znode *zp; | |
1106 | int *path = c->bottom_up_buf, p = 0; | |
1107 | ||
6eb61d58 RW |
1108 | ubifs_assert(c, c->zroot.znode); |
1109 | ubifs_assert(c, znode); | |
1e51764a AB |
1110 | if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) { |
1111 | kfree(c->bottom_up_buf); | |
6da2ec56 KC |
1112 | c->bottom_up_buf = kmalloc_array(c->zroot.znode->level, |
1113 | sizeof(int), | |
1114 | GFP_NOFS); | |
1e51764a AB |
1115 | if (!c->bottom_up_buf) |
1116 | return ERR_PTR(-ENOMEM); | |
1117 | path = c->bottom_up_buf; | |
1118 | } | |
1119 | if (c->zroot.znode->level) { | |
1120 | /* Go up until parent is dirty */ | |
1121 | while (1) { | |
1122 | int n; | |
1123 | ||
1124 | zp = znode->parent; | |
1125 | if (!zp) | |
1126 | break; | |
1127 | n = znode->iip; | |
6eb61d58 | 1128 | ubifs_assert(c, p < c->zroot.znode->level); |
1e51764a AB |
1129 | path[p++] = n; |
1130 | if (!zp->cnext && ubifs_zn_dirty(znode)) | |
1131 | break; | |
1132 | znode = zp; | |
1133 | } | |
1134 | } | |
1135 | ||
1136 | /* Come back down, dirtying as we go */ | |
1137 | while (1) { | |
1138 | struct ubifs_zbranch *zbr; | |
1139 | ||
1140 | zp = znode->parent; | |
1141 | if (zp) { | |
6eb61d58 RW |
1142 | ubifs_assert(c, path[p - 1] >= 0); |
1143 | ubifs_assert(c, path[p - 1] < zp->child_cnt); | |
1e51764a AB |
1144 | zbr = &zp->zbranch[path[--p]]; |
1145 | znode = dirty_cow_znode(c, zbr); | |
1146 | } else { | |
6eb61d58 | 1147 | ubifs_assert(c, znode == c->zroot.znode); |
1e51764a AB |
1148 | znode = dirty_cow_znode(c, &c->zroot); |
1149 | } | |
8d47aef4 | 1150 | if (IS_ERR(znode) || !p) |
1e51764a | 1151 | break; |
6eb61d58 RW |
1152 | ubifs_assert(c, path[p - 1] >= 0); |
1153 | ubifs_assert(c, path[p - 1] < znode->child_cnt); | |
1e51764a AB |
1154 | znode = znode->zbranch[path[p - 1]].znode; |
1155 | } | |
1156 | ||
1157 | return znode; | |
1158 | } | |
1159 | ||
1160 | /** | |
1161 | * ubifs_lookup_level0 - search for zero-level znode. | |
1162 | * @c: UBIFS file-system description object | |
1163 | * @key: key to lookup | |
1164 | * @zn: znode is returned here | |
1165 | * @n: znode branch slot number is returned here | |
1166 | * | |
1167 | * This function looks up the TNC tree and search for zero-level znode which | |
1168 | * refers key @key. The found zero-level znode is returned in @zn. There are 3 | |
1169 | * cases: | |
1170 | * o exact match, i.e. the found zero-level znode contains key @key, then %1 | |
1171 | * is returned and slot number of the matched branch is stored in @n; | |
1172 | * o not exact match, which means that zero-level znode does not contain | |
e3c3efc2 AB |
1173 | * @key, then %0 is returned and slot number of the closest branch is stored |
1174 | * in @n; | |
1e51764a AB |
1175 | * o @key is so small that it is even less than the lowest key of the |
1176 | * leftmost zero-level node, then %0 is returned and %0 is stored in @n. | |
1177 | * | |
1178 | * Note, when the TNC tree is traversed, some znodes may be absent, then this | |
1179 | * function reads corresponding indexing nodes and inserts them to TNC. In | |
1180 | * case of failure, a negative error code is returned. | |
1181 | */ | |
1182 | int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key, | |
1183 | struct ubifs_znode **zn, int *n) | |
1184 | { | |
1185 | int err, exact; | |
1186 | struct ubifs_znode *znode; | |
6cff5732 | 1187 | time64_t time = ktime_get_seconds(); |
1e51764a | 1188 | |
515315a1 | 1189 | dbg_tnck(key, "search key "); |
6eb61d58 | 1190 | ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY); |
1e51764a AB |
1191 | |
1192 | znode = c->zroot.znode; | |
1193 | if (unlikely(!znode)) { | |
1194 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
1195 | if (IS_ERR(znode)) | |
1196 | return PTR_ERR(znode); | |
1197 | } | |
1198 | ||
1199 | znode->time = time; | |
1200 | ||
1201 | while (1) { | |
1202 | struct ubifs_zbranch *zbr; | |
1203 | ||
1204 | exact = ubifs_search_zbranch(c, znode, key, n); | |
1205 | ||
1206 | if (znode->level == 0) | |
1207 | break; | |
1208 | ||
1209 | if (*n < 0) | |
1210 | *n = 0; | |
1211 | zbr = &znode->zbranch[*n]; | |
1212 | ||
1213 | if (zbr->znode) { | |
1214 | znode->time = time; | |
1215 | znode = zbr->znode; | |
1216 | continue; | |
1217 | } | |
1218 | ||
1219 | /* znode is not in TNC cache, load it from the media */ | |
1220 | znode = ubifs_load_znode(c, zbr, znode, *n); | |
1221 | if (IS_ERR(znode)) | |
1222 | return PTR_ERR(znode); | |
1223 | } | |
1224 | ||
1225 | *zn = znode; | |
1226 | if (exact || !is_hash_key(c, key) || *n != -1) { | |
1227 | dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); | |
1228 | return exact; | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * Here is a tricky place. We have not found the key and this is a | |
1233 | * "hashed" key, which may collide. The rest of the code deals with | |
1234 | * situations like this: | |
1235 | * | |
1236 | * | 3 | 5 | | |
1237 | * / \ | |
1238 | * | 3 | 5 | | 6 | 7 | (x) | |
1239 | * | |
1240 | * Or more a complex example: | |
1241 | * | |
1242 | * | 1 | 5 | | |
1243 | * / \ | |
1244 | * | 1 | 3 | | 5 | 8 | | |
1245 | * \ / | |
1246 | * | 5 | 5 | | 6 | 7 | (x) | |
1247 | * | |
1248 | * In the examples, if we are looking for key "5", we may reach nodes | |
1249 | * marked with "(x)". In this case what we have do is to look at the | |
1250 | * left and see if there is "5" key there. If there is, we have to | |
1251 | * return it. | |
1252 | * | |
1253 | * Note, this whole situation is possible because we allow to have | |
1254 | * elements which are equivalent to the next key in the parent in the | |
1255 | * children of current znode. For example, this happens if we split a | |
1256 | * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something | |
1257 | * like this: | |
1258 | * | 3 | 5 | | |
1259 | * / \ | |
1260 | * | 3 | 5 | | 5 | 6 | 7 | | |
1261 | * ^ | |
1262 | * And this becomes what is at the first "picture" after key "5" marked | |
1263 | * with "^" is removed. What could be done is we could prohibit | |
1264 | * splitting in the middle of the colliding sequence. Also, when | |
1265 | * removing the leftmost key, we would have to correct the key of the | |
1266 | * parent node, which would introduce additional complications. Namely, | |
7d4e9ccb | 1267 | * if we changed the leftmost key of the parent znode, the garbage |
1e51764a AB |
1268 | * collector would be unable to find it (GC is doing this when GC'ing |
1269 | * indexing LEBs). Although we already have an additional RB-tree where | |
1270 | * we save such changed znodes (see 'ins_clr_old_idx_znode()') until | |
1271 | * after the commit. But anyway, this does not look easy to implement | |
1272 | * so we did not try this. | |
1273 | */ | |
1274 | err = tnc_prev(c, &znode, n); | |
1275 | if (err == -ENOENT) { | |
1276 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1277 | *n = -1; | |
1278 | return 0; | |
1279 | } | |
1280 | if (unlikely(err < 0)) | |
1281 | return err; | |
1282 | if (keys_cmp(c, key, &znode->zbranch[*n].key)) { | |
1283 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1284 | *n = -1; | |
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); | |
1289 | *zn = znode; | |
1290 | return 1; | |
1291 | } | |
1292 | ||
1293 | /** | |
1294 | * lookup_level0_dirty - search for zero-level znode dirtying. | |
1295 | * @c: UBIFS file-system description object | |
1296 | * @key: key to lookup | |
1297 | * @zn: znode is returned here | |
1298 | * @n: znode branch slot number is returned here | |
1299 | * | |
1300 | * This function looks up the TNC tree and search for zero-level znode which | |
1301 | * refers key @key. The found zero-level znode is returned in @zn. There are 3 | |
1302 | * cases: | |
1303 | * o exact match, i.e. the found zero-level znode contains key @key, then %1 | |
1304 | * is returned and slot number of the matched branch is stored in @n; | |
1305 | * o not exact match, which means that zero-level znode does not contain @key | |
1306 | * then %0 is returned and slot number of the closed branch is stored in | |
1307 | * @n; | |
1308 | * o @key is so small that it is even less than the lowest key of the | |
1309 | * leftmost zero-level node, then %0 is returned and %-1 is stored in @n. | |
1310 | * | |
1311 | * Additionally all znodes in the path from the root to the located zero-level | |
1312 | * znode are marked as dirty. | |
1313 | * | |
1314 | * Note, when the TNC tree is traversed, some znodes may be absent, then this | |
1315 | * function reads corresponding indexing nodes and inserts them to TNC. In | |
1316 | * case of failure, a negative error code is returned. | |
1317 | */ | |
1318 | static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key, | |
1319 | struct ubifs_znode **zn, int *n) | |
1320 | { | |
1321 | int err, exact; | |
1322 | struct ubifs_znode *znode; | |
6cff5732 | 1323 | time64_t time = ktime_get_seconds(); |
1e51764a | 1324 | |
515315a1 | 1325 | dbg_tnck(key, "search and dirty key "); |
1e51764a AB |
1326 | |
1327 | znode = c->zroot.znode; | |
1328 | if (unlikely(!znode)) { | |
1329 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
1330 | if (IS_ERR(znode)) | |
1331 | return PTR_ERR(znode); | |
1332 | } | |
1333 | ||
1334 | znode = dirty_cow_znode(c, &c->zroot); | |
1335 | if (IS_ERR(znode)) | |
1336 | return PTR_ERR(znode); | |
1337 | ||
1338 | znode->time = time; | |
1339 | ||
1340 | while (1) { | |
1341 | struct ubifs_zbranch *zbr; | |
1342 | ||
1343 | exact = ubifs_search_zbranch(c, znode, key, n); | |
1344 | ||
1345 | if (znode->level == 0) | |
1346 | break; | |
1347 | ||
1348 | if (*n < 0) | |
1349 | *n = 0; | |
1350 | zbr = &znode->zbranch[*n]; | |
1351 | ||
1352 | if (zbr->znode) { | |
1353 | znode->time = time; | |
1354 | znode = dirty_cow_znode(c, zbr); | |
1355 | if (IS_ERR(znode)) | |
1356 | return PTR_ERR(znode); | |
1357 | continue; | |
1358 | } | |
1359 | ||
1360 | /* znode is not in TNC cache, load it from the media */ | |
1361 | znode = ubifs_load_znode(c, zbr, znode, *n); | |
1362 | if (IS_ERR(znode)) | |
1363 | return PTR_ERR(znode); | |
1364 | znode = dirty_cow_znode(c, zbr); | |
1365 | if (IS_ERR(znode)) | |
1366 | return PTR_ERR(znode); | |
1367 | } | |
1368 | ||
1369 | *zn = znode; | |
1370 | if (exact || !is_hash_key(c, key) || *n != -1) { | |
1371 | dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); | |
1372 | return exact; | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * See huge comment at 'lookup_level0_dirty()' what is the rest of the | |
1377 | * code. | |
1378 | */ | |
1379 | err = tnc_prev(c, &znode, n); | |
1380 | if (err == -ENOENT) { | |
1381 | *n = -1; | |
1382 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1383 | return 0; | |
1384 | } | |
1385 | if (unlikely(err < 0)) | |
1386 | return err; | |
1387 | if (keys_cmp(c, key, &znode->zbranch[*n].key)) { | |
1388 | *n = -1; | |
1389 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1390 | return 0; | |
1391 | } | |
1392 | ||
1393 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
1394 | znode = dirty_cow_bottom_up(c, znode); | |
1395 | if (IS_ERR(znode)) | |
1396 | return PTR_ERR(znode); | |
1397 | } | |
1398 | ||
1399 | dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); | |
1400 | *zn = znode; | |
1401 | return 1; | |
1402 | } | |
1403 | ||
1404 | /** | |
601c0bc4 | 1405 | * maybe_leb_gced - determine if a LEB may have been garbage collected. |
1e51764a | 1406 | * @c: UBIFS file-system description object |
601c0bc4 AH |
1407 | * @lnum: LEB number |
1408 | * @gc_seq1: garbage collection sequence number | |
1e51764a | 1409 | * |
601c0bc4 AH |
1410 | * This function determines if @lnum may have been garbage collected since |
1411 | * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise | |
1412 | * %0 is returned. | |
1e51764a | 1413 | */ |
601c0bc4 | 1414 | static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1) |
1e51764a | 1415 | { |
601c0bc4 | 1416 | int gc_seq2, gced_lnum; |
1e51764a | 1417 | |
601c0bc4 AH |
1418 | gced_lnum = c->gced_lnum; |
1419 | smp_rmb(); | |
1420 | gc_seq2 = c->gc_seq; | |
1421 | /* Same seq means no GC */ | |
1422 | if (gc_seq1 == gc_seq2) | |
1423 | return 0; | |
1424 | /* Different by more than 1 means we don't know */ | |
1425 | if (gc_seq1 + 1 != gc_seq2) | |
1426 | return 1; | |
1427 | /* | |
1428 | * We have seen the sequence number has increased by 1. Now we need to | |
1429 | * be sure we read the right LEB number, so read it again. | |
1430 | */ | |
1431 | smp_rmb(); | |
1432 | if (gced_lnum != c->gced_lnum) | |
1433 | return 1; | |
1434 | /* Finally we can check lnum */ | |
1435 | if (gced_lnum == lnum) | |
1436 | return 1; | |
1437 | return 0; | |
1e51764a AB |
1438 | } |
1439 | ||
1440 | /** | |
1441 | * ubifs_tnc_locate - look up a file-system node and return it and its location. | |
1442 | * @c: UBIFS file-system description object | |
1443 | * @key: node key to lookup | |
1444 | * @node: the node is returned here | |
1445 | * @lnum: LEB number is returned here | |
1446 | * @offs: offset is returned here | |
1447 | * | |
e3c3efc2 | 1448 | * This function looks up and reads node with key @key. The caller has to make |
601c0bc4 AH |
1449 | * sure the @node buffer is large enough to fit the node. Returns zero in case |
1450 | * of success, %-ENOENT if the node was not found, and a negative error code in | |
1451 | * case of failure. The node location can be returned in @lnum and @offs. | |
1e51764a AB |
1452 | */ |
1453 | int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, | |
1454 | void *node, int *lnum, int *offs) | |
1455 | { | |
601c0bc4 | 1456 | int found, n, err, safely = 0, gc_seq1; |
1e51764a AB |
1457 | struct ubifs_znode *znode; |
1458 | struct ubifs_zbranch zbr, *zt; | |
1459 | ||
601c0bc4 | 1460 | again: |
1e51764a AB |
1461 | mutex_lock(&c->tnc_mutex); |
1462 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
1463 | if (!found) { | |
1464 | err = -ENOENT; | |
1465 | goto out; | |
1466 | } else if (found < 0) { | |
1467 | err = found; | |
1468 | goto out; | |
1469 | } | |
1470 | zt = &znode->zbranch[n]; | |
601c0bc4 AH |
1471 | if (lnum) { |
1472 | *lnum = zt->lnum; | |
1473 | *offs = zt->offs; | |
1474 | } | |
1e51764a AB |
1475 | if (is_hash_key(c, key)) { |
1476 | /* | |
1477 | * In this case the leaf node cache gets used, so we pass the | |
1478 | * address of the zbranch and keep the mutex locked | |
1479 | */ | |
b91dc981 | 1480 | err = tnc_read_hashed_node(c, zt, node); |
1e51764a AB |
1481 | goto out; |
1482 | } | |
601c0bc4 AH |
1483 | if (safely) { |
1484 | err = ubifs_tnc_read_node(c, zt, node); | |
1485 | goto out; | |
1486 | } | |
1487 | /* Drop the TNC mutex prematurely and race with garbage collection */ | |
1e51764a | 1488 | zbr = znode->zbranch[n]; |
601c0bc4 | 1489 | gc_seq1 = c->gc_seq; |
1e51764a AB |
1490 | mutex_unlock(&c->tnc_mutex); |
1491 | ||
601c0bc4 AH |
1492 | if (ubifs_get_wbuf(c, zbr.lnum)) { |
1493 | /* We do not GC journal heads */ | |
1494 | err = ubifs_tnc_read_node(c, &zbr, node); | |
1495 | return err; | |
1496 | } | |
1e51764a | 1497 | |
601c0bc4 | 1498 | err = fallible_read_node(c, key, &zbr, node); |
6dcfac4f | 1499 | if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) { |
601c0bc4 AH |
1500 | /* |
1501 | * The node may have been GC'ed out from under us so try again | |
1502 | * while keeping the TNC mutex locked. | |
1503 | */ | |
1504 | safely = 1; | |
1505 | goto again; | |
1506 | } | |
1507 | return 0; | |
1e51764a AB |
1508 | |
1509 | out: | |
1510 | mutex_unlock(&c->tnc_mutex); | |
1511 | return err; | |
1512 | } | |
1513 | ||
4793e7c5 AH |
1514 | /** |
1515 | * ubifs_tnc_get_bu_keys - lookup keys for bulk-read. | |
1516 | * @c: UBIFS file-system description object | |
1517 | * @bu: bulk-read parameters and results | |
1518 | * | |
1519 | * Lookup consecutive data node keys for the same inode that reside | |
6c0c42cd AB |
1520 | * consecutively in the same LEB. This function returns zero in case of success |
1521 | * and a negative error code in case of failure. | |
1522 | * | |
1523 | * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function | |
1524 | * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares | |
6f7ab6d4 | 1525 | * maximum possible amount of nodes for bulk-read. |
4793e7c5 AH |
1526 | */ |
1527 | int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu) | |
1528 | { | |
1529 | int n, err = 0, lnum = -1, uninitialized_var(offs); | |
1530 | int uninitialized_var(len); | |
1531 | unsigned int block = key_block(c, &bu->key); | |
1532 | struct ubifs_znode *znode; | |
1533 | ||
1534 | bu->cnt = 0; | |
1535 | bu->blk_cnt = 0; | |
1536 | bu->eof = 0; | |
1537 | ||
1538 | mutex_lock(&c->tnc_mutex); | |
1539 | /* Find first key */ | |
1540 | err = ubifs_lookup_level0(c, &bu->key, &znode, &n); | |
1541 | if (err < 0) | |
1542 | goto out; | |
1543 | if (err) { | |
1544 | /* Key found */ | |
1545 | len = znode->zbranch[n].len; | |
1546 | /* The buffer must be big enough for at least 1 node */ | |
1547 | if (len > bu->buf_len) { | |
1548 | err = -EINVAL; | |
1549 | goto out; | |
1550 | } | |
1551 | /* Add this key */ | |
1552 | bu->zbranch[bu->cnt++] = znode->zbranch[n]; | |
1553 | bu->blk_cnt += 1; | |
1554 | lnum = znode->zbranch[n].lnum; | |
1555 | offs = ALIGN(znode->zbranch[n].offs + len, 8); | |
1556 | } | |
1557 | while (1) { | |
1558 | struct ubifs_zbranch *zbr; | |
1559 | union ubifs_key *key; | |
1560 | unsigned int next_block; | |
1561 | ||
1562 | /* Find next key */ | |
1563 | err = tnc_next(c, &znode, &n); | |
1564 | if (err) | |
1565 | goto out; | |
1566 | zbr = &znode->zbranch[n]; | |
1567 | key = &zbr->key; | |
1568 | /* See if there is another data key for this file */ | |
1569 | if (key_inum(c, key) != key_inum(c, &bu->key) || | |
1570 | key_type(c, key) != UBIFS_DATA_KEY) { | |
1571 | err = -ENOENT; | |
1572 | goto out; | |
1573 | } | |
1574 | if (lnum < 0) { | |
1575 | /* First key found */ | |
1576 | lnum = zbr->lnum; | |
1577 | offs = ALIGN(zbr->offs + zbr->len, 8); | |
1578 | len = zbr->len; | |
1579 | if (len > bu->buf_len) { | |
1580 | err = -EINVAL; | |
1581 | goto out; | |
1582 | } | |
1583 | } else { | |
1584 | /* | |
1585 | * The data nodes must be in consecutive positions in | |
1586 | * the same LEB. | |
1587 | */ | |
1588 | if (zbr->lnum != lnum || zbr->offs != offs) | |
1589 | goto out; | |
1590 | offs += ALIGN(zbr->len, 8); | |
1591 | len = ALIGN(len, 8) + zbr->len; | |
1592 | /* Must not exceed buffer length */ | |
1593 | if (len > bu->buf_len) | |
1594 | goto out; | |
1595 | } | |
1596 | /* Allow for holes */ | |
1597 | next_block = key_block(c, key); | |
1598 | bu->blk_cnt += (next_block - block - 1); | |
1599 | if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) | |
1600 | goto out; | |
1601 | block = next_block; | |
1602 | /* Add this key */ | |
1603 | bu->zbranch[bu->cnt++] = *zbr; | |
1604 | bu->blk_cnt += 1; | |
1605 | /* See if we have room for more */ | |
1606 | if (bu->cnt >= UBIFS_MAX_BULK_READ) | |
1607 | goto out; | |
1608 | if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) | |
1609 | goto out; | |
1610 | } | |
1611 | out: | |
1612 | if (err == -ENOENT) { | |
1613 | bu->eof = 1; | |
1614 | err = 0; | |
1615 | } | |
1616 | bu->gc_seq = c->gc_seq; | |
1617 | mutex_unlock(&c->tnc_mutex); | |
1618 | if (err) | |
1619 | return err; | |
1620 | /* | |
1621 | * An enormous hole could cause bulk-read to encompass too many | |
1622 | * page cache pages, so limit the number here. | |
1623 | */ | |
63c300b6 | 1624 | if (bu->blk_cnt > UBIFS_MAX_BULK_READ) |
4793e7c5 AH |
1625 | bu->blk_cnt = UBIFS_MAX_BULK_READ; |
1626 | /* | |
1627 | * Ensure that bulk-read covers a whole number of page cache | |
1628 | * pages. | |
1629 | */ | |
1630 | if (UBIFS_BLOCKS_PER_PAGE == 1 || | |
1631 | !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1))) | |
1632 | return 0; | |
1633 | if (bu->eof) { | |
1634 | /* At the end of file we can round up */ | |
1635 | bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1; | |
1636 | return 0; | |
1637 | } | |
1638 | /* Exclude data nodes that do not make up a whole page cache page */ | |
1639 | block = key_block(c, &bu->key) + bu->blk_cnt; | |
1640 | block &= ~(UBIFS_BLOCKS_PER_PAGE - 1); | |
1641 | while (bu->cnt) { | |
1642 | if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block) | |
1643 | break; | |
1644 | bu->cnt -= 1; | |
1645 | } | |
1646 | return 0; | |
1647 | } | |
1648 | ||
1649 | /** | |
1650 | * read_wbuf - bulk-read from a LEB with a wbuf. | |
1651 | * @wbuf: wbuf that may overlap the read | |
1652 | * @buf: buffer into which to read | |
1653 | * @len: read length | |
1654 | * @lnum: LEB number from which to read | |
1655 | * @offs: offset from which to read | |
1656 | * | |
1657 | * This functions returns %0 on success or a negative error code on failure. | |
1658 | */ | |
1659 | static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum, | |
1660 | int offs) | |
1661 | { | |
1662 | const struct ubifs_info *c = wbuf->c; | |
1663 | int rlen, overlap; | |
1664 | ||
1665 | dbg_io("LEB %d:%d, length %d", lnum, offs, len); | |
6eb61d58 RW |
1666 | ubifs_assert(c, wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0); |
1667 | ubifs_assert(c, !(offs & 7) && offs < c->leb_size); | |
1668 | ubifs_assert(c, offs + len <= c->leb_size); | |
4793e7c5 AH |
1669 | |
1670 | spin_lock(&wbuf->lock); | |
1671 | overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs); | |
1672 | if (!overlap) { | |
1673 | /* We may safely unlock the write-buffer and read the data */ | |
1674 | spin_unlock(&wbuf->lock); | |
d304820a | 1675 | return ubifs_leb_read(c, lnum, buf, offs, len, 0); |
4793e7c5 AH |
1676 | } |
1677 | ||
1678 | /* Don't read under wbuf */ | |
1679 | rlen = wbuf->offs - offs; | |
1680 | if (rlen < 0) | |
1681 | rlen = 0; | |
1682 | ||
1683 | /* Copy the rest from the write-buffer */ | |
1684 | memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen); | |
1685 | spin_unlock(&wbuf->lock); | |
1686 | ||
1687 | if (rlen > 0) | |
1688 | /* Read everything that goes before write-buffer */ | |
d304820a | 1689 | return ubifs_leb_read(c, lnum, buf, offs, rlen, 0); |
4793e7c5 AH |
1690 | |
1691 | return 0; | |
1692 | } | |
1693 | ||
1694 | /** | |
1695 | * validate_data_node - validate data nodes for bulk-read. | |
1696 | * @c: UBIFS file-system description object | |
1697 | * @buf: buffer containing data node to validate | |
1698 | * @zbr: zbranch of data node to validate | |
1699 | * | |
1700 | * This functions returns %0 on success or a negative error code on failure. | |
1701 | */ | |
1702 | static int validate_data_node(struct ubifs_info *c, void *buf, | |
1703 | struct ubifs_zbranch *zbr) | |
1704 | { | |
1705 | union ubifs_key key1; | |
1706 | struct ubifs_ch *ch = buf; | |
1707 | int err, len; | |
1708 | ||
1709 | if (ch->node_type != UBIFS_DATA_NODE) { | |
235c362b | 1710 | ubifs_err(c, "bad node type (%d but expected %d)", |
4793e7c5 AH |
1711 | ch->node_type, UBIFS_DATA_NODE); |
1712 | goto out_err; | |
1713 | } | |
1714 | ||
2953e73f | 1715 | err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0); |
4793e7c5 | 1716 | if (err) { |
235c362b | 1717 | ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE); |
4793e7c5 AH |
1718 | goto out; |
1719 | } | |
1720 | ||
16a26b20 SH |
1721 | err = ubifs_node_check_hash(c, buf, zbr->hash); |
1722 | if (err) { | |
1723 | ubifs_bad_hash(c, buf, zbr->hash, zbr->lnum, zbr->offs); | |
1724 | return err; | |
1725 | } | |
1726 | ||
4793e7c5 AH |
1727 | len = le32_to_cpu(ch->len); |
1728 | if (len != zbr->len) { | |
235c362b | 1729 | ubifs_err(c, "bad node length %d, expected %d", len, zbr->len); |
4793e7c5 AH |
1730 | goto out_err; |
1731 | } | |
1732 | ||
1733 | /* Make sure the key of the read node is correct */ | |
1734 | key_read(c, buf + UBIFS_KEY_OFFSET, &key1); | |
1735 | if (!keys_eq(c, &zbr->key, &key1)) { | |
235c362b | 1736 | ubifs_err(c, "bad key in node at LEB %d:%d", |
4793e7c5 | 1737 | zbr->lnum, zbr->offs); |
515315a1 AB |
1738 | dbg_tnck(&zbr->key, "looked for key "); |
1739 | dbg_tnck(&key1, "found node's key "); | |
4793e7c5 AH |
1740 | goto out_err; |
1741 | } | |
1742 | ||
1743 | return 0; | |
1744 | ||
1745 | out_err: | |
1746 | err = -EINVAL; | |
1747 | out: | |
235c362b | 1748 | ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs); |
edf6be24 | 1749 | ubifs_dump_node(c, buf); |
7c46d0ae | 1750 | dump_stack(); |
4793e7c5 AH |
1751 | return err; |
1752 | } | |
1753 | ||
1754 | /** | |
1755 | * ubifs_tnc_bulk_read - read a number of data nodes in one go. | |
1756 | * @c: UBIFS file-system description object | |
1757 | * @bu: bulk-read parameters and results | |
1758 | * | |
1759 | * This functions reads and validates the data nodes that were identified by the | |
1760 | * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success, | |
1761 | * -EAGAIN to indicate a race with GC, or another negative error code on | |
1762 | * failure. | |
1763 | */ | |
1764 | int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu) | |
1765 | { | |
1766 | int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i; | |
1767 | struct ubifs_wbuf *wbuf; | |
1768 | void *buf; | |
1769 | ||
1770 | len = bu->zbranch[bu->cnt - 1].offs; | |
1771 | len += bu->zbranch[bu->cnt - 1].len - offs; | |
1772 | if (len > bu->buf_len) { | |
235c362b | 1773 | ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len); |
4793e7c5 AH |
1774 | return -EINVAL; |
1775 | } | |
1776 | ||
1777 | /* Do the read */ | |
1778 | wbuf = ubifs_get_wbuf(c, lnum); | |
1779 | if (wbuf) | |
1780 | err = read_wbuf(wbuf, bu->buf, len, lnum, offs); | |
1781 | else | |
d304820a | 1782 | err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0); |
4793e7c5 AH |
1783 | |
1784 | /* Check for a race with GC */ | |
1785 | if (maybe_leb_gced(c, lnum, bu->gc_seq)) | |
1786 | return -EAGAIN; | |
1787 | ||
1788 | if (err && err != -EBADMSG) { | |
235c362b | 1789 | ubifs_err(c, "failed to read from LEB %d:%d, error %d", |
4793e7c5 | 1790 | lnum, offs, err); |
7c46d0ae | 1791 | dump_stack(); |
515315a1 | 1792 | dbg_tnck(&bu->key, "key "); |
4793e7c5 AH |
1793 | return err; |
1794 | } | |
1795 | ||
1796 | /* Validate the nodes read */ | |
1797 | buf = bu->buf; | |
1798 | for (i = 0; i < bu->cnt; i++) { | |
1799 | err = validate_data_node(c, buf, &bu->zbranch[i]); | |
1800 | if (err) | |
1801 | return err; | |
1802 | buf = buf + ALIGN(bu->zbranch[i].len, 8); | |
1803 | } | |
1804 | ||
1805 | return 0; | |
1806 | } | |
1807 | ||
1e51764a AB |
1808 | /** |
1809 | * do_lookup_nm- look up a "hashed" node. | |
1810 | * @c: UBIFS file-system description object | |
1811 | * @key: node key to lookup | |
1812 | * @node: the node is returned here | |
1813 | * @nm: node name | |
1814 | * | |
528e3d17 | 1815 | * This function looks up and reads a node which contains name hash in the key. |
1e51764a AB |
1816 | * Since the hash may have collisions, there may be many nodes with the same |
1817 | * key, so we have to sequentially look to all of them until the needed one is | |
1818 | * found. This function returns zero in case of success, %-ENOENT if the node | |
1819 | * was not found, and a negative error code in case of failure. | |
1820 | */ | |
1821 | static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, | |
f4f61d2c | 1822 | void *node, const struct fscrypt_name *nm) |
1e51764a AB |
1823 | { |
1824 | int found, n, err; | |
1825 | struct ubifs_znode *znode; | |
1e51764a | 1826 | |
35ee314c | 1827 | dbg_tnck(key, "key "); |
1e51764a AB |
1828 | mutex_lock(&c->tnc_mutex); |
1829 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
1830 | if (!found) { | |
1831 | err = -ENOENT; | |
1832 | goto out_unlock; | |
1833 | } else if (found < 0) { | |
1834 | err = found; | |
1835 | goto out_unlock; | |
1836 | } | |
1837 | ||
6eb61d58 | 1838 | ubifs_assert(c, n >= 0); |
1e51764a AB |
1839 | |
1840 | err = resolve_collision(c, key, &znode, &n, nm); | |
1841 | dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); | |
1842 | if (unlikely(err < 0)) | |
1843 | goto out_unlock; | |
1844 | if (err == 0) { | |
1845 | err = -ENOENT; | |
1846 | goto out_unlock; | |
1847 | } | |
1848 | ||
b91dc981 | 1849 | err = tnc_read_hashed_node(c, &znode->zbranch[n], node); |
1e51764a AB |
1850 | |
1851 | out_unlock: | |
1852 | mutex_unlock(&c->tnc_mutex); | |
1853 | return err; | |
1854 | } | |
1855 | ||
1856 | /** | |
1857 | * ubifs_tnc_lookup_nm - look up a "hashed" node. | |
1858 | * @c: UBIFS file-system description object | |
1859 | * @key: node key to lookup | |
1860 | * @node: the node is returned here | |
1861 | * @nm: node name | |
1862 | * | |
528e3d17 | 1863 | * This function looks up and reads a node which contains name hash in the key. |
1e51764a AB |
1864 | * Since the hash may have collisions, there may be many nodes with the same |
1865 | * key, so we have to sequentially look to all of them until the needed one is | |
1866 | * found. This function returns zero in case of success, %-ENOENT if the node | |
1867 | * was not found, and a negative error code in case of failure. | |
1868 | */ | |
1869 | int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, | |
f4f61d2c | 1870 | void *node, const struct fscrypt_name *nm) |
1e51764a AB |
1871 | { |
1872 | int err, len; | |
1873 | const struct ubifs_dent_node *dent = node; | |
1874 | ||
1875 | /* | |
1876 | * We assume that in most of the cases there are no name collisions and | |
1877 | * 'ubifs_tnc_lookup()' returns us the right direntry. | |
1878 | */ | |
1879 | err = ubifs_tnc_lookup(c, key, node); | |
1880 | if (err) | |
1881 | return err; | |
1882 | ||
1883 | len = le16_to_cpu(dent->nlen); | |
f4f61d2c | 1884 | if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len)) |
1e51764a AB |
1885 | return 0; |
1886 | ||
1887 | /* | |
1888 | * Unluckily, there are hash collisions and we have to iterate over | |
1889 | * them look at each direntry with colliding name hash sequentially. | |
1890 | */ | |
528e3d17 | 1891 | |
1e51764a AB |
1892 | return do_lookup_nm(c, key, node, nm); |
1893 | } | |
1894 | ||
781f675e RW |
1895 | static int search_dh_cookie(struct ubifs_info *c, const union ubifs_key *key, |
1896 | struct ubifs_dent_node *dent, uint32_t cookie, | |
1897 | struct ubifs_znode **zn, int *n) | |
528e3d17 | 1898 | { |
781f675e RW |
1899 | int err; |
1900 | struct ubifs_znode *znode = *zn; | |
528e3d17 | 1901 | struct ubifs_zbranch *zbr; |
781f675e | 1902 | union ubifs_key *dkey; |
528e3d17 RW |
1903 | |
1904 | for (;;) { | |
781f675e | 1905 | zbr = &znode->zbranch[*n]; |
528e3d17 RW |
1906 | dkey = &zbr->key; |
1907 | ||
1908 | if (key_inum(c, dkey) != key_inum(c, key) || | |
781f675e | 1909 | key_type(c, dkey) != key_type(c, key)) { |
c877154d | 1910 | return -ENOENT; |
528e3d17 RW |
1911 | } |
1912 | ||
1913 | err = tnc_read_hashed_node(c, zbr, dent); | |
1914 | if (err) | |
c877154d | 1915 | return err; |
528e3d17 RW |
1916 | |
1917 | if (key_hash(c, key) == key_hash(c, dkey) && | |
781f675e RW |
1918 | le32_to_cpu(dent->cookie) == cookie) { |
1919 | *zn = znode; | |
c877154d | 1920 | return 0; |
781f675e | 1921 | } |
781f675e | 1922 | |
c877154d GU |
1923 | err = tnc_next(c, &znode, n); |
1924 | if (err) | |
1925 | return err; | |
1926 | } | |
781f675e RW |
1927 | } |
1928 | ||
1929 | static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key, | |
1930 | struct ubifs_dent_node *dent, uint32_t cookie) | |
1931 | { | |
1932 | int n, err; | |
1933 | struct ubifs_znode *znode; | |
1934 | union ubifs_key start_key; | |
1935 | ||
6eb61d58 | 1936 | ubifs_assert(c, is_hash_key(c, key)); |
781f675e RW |
1937 | |
1938 | lowest_dent_key(c, &start_key, key_inum(c, key)); | |
1939 | ||
1940 | mutex_lock(&c->tnc_mutex); | |
1941 | err = ubifs_lookup_level0(c, &start_key, &znode, &n); | |
1942 | if (unlikely(err < 0)) | |
1943 | goto out_unlock; | |
1944 | ||
1945 | err = search_dh_cookie(c, key, dent, cookie, &znode, &n); | |
1946 | ||
528e3d17 RW |
1947 | out_unlock: |
1948 | mutex_unlock(&c->tnc_mutex); | |
1949 | return err; | |
1950 | } | |
1951 | ||
1952 | /** | |
1953 | * ubifs_tnc_lookup_dh - look up a "double hashed" node. | |
1954 | * @c: UBIFS file-system description object | |
1955 | * @key: node key to lookup | |
1956 | * @node: the node is returned here | |
1957 | * @cookie: node cookie for collision resolution | |
1958 | * | |
1959 | * This function looks up and reads a node which contains name hash in the key. | |
1960 | * Since the hash may have collisions, there may be many nodes with the same | |
1961 | * key, so we have to sequentially look to all of them until the needed one | |
1962 | * with the same cookie value is found. | |
1963 | * This function returns zero in case of success, %-ENOENT if the node | |
1964 | * was not found, and a negative error code in case of failure. | |
1965 | */ | |
1966 | int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key, | |
1967 | void *node, uint32_t cookie) | |
1968 | { | |
1969 | int err; | |
1970 | const struct ubifs_dent_node *dent = node; | |
1971 | ||
d63d61c1 RW |
1972 | if (!c->double_hash) |
1973 | return -EOPNOTSUPP; | |
1974 | ||
528e3d17 RW |
1975 | /* |
1976 | * We assume that in most of the cases there are no name collisions and | |
1977 | * 'ubifs_tnc_lookup()' returns us the right direntry. | |
1978 | */ | |
1979 | err = ubifs_tnc_lookup(c, key, node); | |
1980 | if (err) | |
1981 | return err; | |
1982 | ||
1983 | if (le32_to_cpu(dent->cookie) == cookie) | |
1984 | return 0; | |
1985 | ||
1986 | /* | |
1987 | * Unluckily, there are hash collisions and we have to iterate over | |
1988 | * them look at each direntry with colliding name hash sequentially. | |
1989 | */ | |
1990 | return do_lookup_dh(c, key, node, cookie); | |
1991 | } | |
1992 | ||
1e51764a AB |
1993 | /** |
1994 | * correct_parent_keys - correct parent znodes' keys. | |
1995 | * @c: UBIFS file-system description object | |
1996 | * @znode: znode to correct parent znodes for | |
1997 | * | |
1998 | * This is a helper function for 'tnc_insert()'. When the key of the leftmost | |
1999 | * zbranch changes, keys of parent znodes have to be corrected. This helper | |
2000 | * function is called in such situations and corrects the keys if needed. | |
2001 | */ | |
2002 | static void correct_parent_keys(const struct ubifs_info *c, | |
2003 | struct ubifs_znode *znode) | |
2004 | { | |
2005 | union ubifs_key *key, *key1; | |
2006 | ||
6eb61d58 RW |
2007 | ubifs_assert(c, znode->parent); |
2008 | ubifs_assert(c, znode->iip == 0); | |
1e51764a AB |
2009 | |
2010 | key = &znode->zbranch[0].key; | |
2011 | key1 = &znode->parent->zbranch[0].key; | |
2012 | ||
2013 | while (keys_cmp(c, key, key1) < 0) { | |
2014 | key_copy(c, key, key1); | |
2015 | znode = znode->parent; | |
2016 | znode->alt = 1; | |
2017 | if (!znode->parent || znode->iip) | |
2018 | break; | |
2019 | key1 = &znode->parent->zbranch[0].key; | |
2020 | } | |
2021 | } | |
2022 | ||
2023 | /** | |
2024 | * insert_zbranch - insert a zbranch into a znode. | |
6eb61d58 | 2025 | * @c: UBIFS file-system description object |
1e51764a AB |
2026 | * @znode: znode into which to insert |
2027 | * @zbr: zbranch to insert | |
2028 | * @n: slot number to insert to | |
2029 | * | |
2030 | * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in | |
2031 | * znode's array of zbranches and keeps zbranches consolidated, so when a new | |
2032 | * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th | |
2033 | * slot, zbranches starting from @n have to be moved right. | |
2034 | */ | |
6eb61d58 | 2035 | static void insert_zbranch(struct ubifs_info *c, struct ubifs_znode *znode, |
1e51764a AB |
2036 | const struct ubifs_zbranch *zbr, int n) |
2037 | { | |
2038 | int i; | |
2039 | ||
6eb61d58 | 2040 | ubifs_assert(c, ubifs_zn_dirty(znode)); |
1e51764a AB |
2041 | |
2042 | if (znode->level) { | |
2043 | for (i = znode->child_cnt; i > n; i--) { | |
2044 | znode->zbranch[i] = znode->zbranch[i - 1]; | |
2045 | if (znode->zbranch[i].znode) | |
2046 | znode->zbranch[i].znode->iip = i; | |
2047 | } | |
2048 | if (zbr->znode) | |
2049 | zbr->znode->iip = n; | |
2050 | } else | |
2051 | for (i = znode->child_cnt; i > n; i--) | |
2052 | znode->zbranch[i] = znode->zbranch[i - 1]; | |
2053 | ||
2054 | znode->zbranch[n] = *zbr; | |
2055 | znode->child_cnt += 1; | |
2056 | ||
2057 | /* | |
2058 | * After inserting at slot zero, the lower bound of the key range of | |
2059 | * this znode may have changed. If this znode is subsequently split | |
2060 | * then the upper bound of the key range may change, and furthermore | |
2061 | * it could change to be lower than the original lower bound. If that | |
2062 | * happens, then it will no longer be possible to find this znode in the | |
2063 | * TNC using the key from the index node on flash. That is bad because | |
2064 | * if it is not found, we will assume it is obsolete and may overwrite | |
2065 | * it. Then if there is an unclean unmount, we will start using the | |
2066 | * old index which will be broken. | |
2067 | * | |
2068 | * So we first mark znodes that have insertions at slot zero, and then | |
2069 | * if they are split we add their lnum/offs to the old_idx tree. | |
2070 | */ | |
2071 | if (n == 0) | |
2072 | znode->alt = 1; | |
2073 | } | |
2074 | ||
2075 | /** | |
2076 | * tnc_insert - insert a node into TNC. | |
2077 | * @c: UBIFS file-system description object | |
2078 | * @znode: znode to insert into | |
2079 | * @zbr: branch to insert | |
2080 | * @n: slot number to insert new zbranch to | |
2081 | * | |
2082 | * This function inserts a new node described by @zbr into znode @znode. If | |
2083 | * znode does not have a free slot for new zbranch, it is split. Parent znodes | |
2084 | * are splat as well if needed. Returns zero in case of success or a negative | |
2085 | * error code in case of failure. | |
2086 | */ | |
2087 | static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode, | |
2088 | struct ubifs_zbranch *zbr, int n) | |
2089 | { | |
2090 | struct ubifs_znode *zn, *zi, *zp; | |
2091 | int i, keep, move, appending = 0; | |
2242c689 | 2092 | union ubifs_key *key = &zbr->key, *key1; |
1e51764a | 2093 | |
6eb61d58 | 2094 | ubifs_assert(c, n >= 0 && n <= c->fanout); |
1e51764a AB |
2095 | |
2096 | /* Implement naive insert for now */ | |
2097 | again: | |
2098 | zp = znode->parent; | |
2099 | if (znode->child_cnt < c->fanout) { | |
6eb61d58 | 2100 | ubifs_assert(c, n != c->fanout); |
515315a1 | 2101 | dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level); |
1e51764a | 2102 | |
6eb61d58 | 2103 | insert_zbranch(c, znode, zbr, n); |
1e51764a AB |
2104 | |
2105 | /* Ensure parent's key is correct */ | |
2106 | if (n == 0 && zp && znode->iip == 0) | |
2107 | correct_parent_keys(c, znode); | |
2108 | ||
2109 | return 0; | |
2110 | } | |
2111 | ||
2112 | /* | |
2113 | * Unfortunately, @znode does not have more empty slots and we have to | |
2114 | * split it. | |
2115 | */ | |
515315a1 | 2116 | dbg_tnck(key, "splitting level %d, key ", znode->level); |
1e51764a AB |
2117 | |
2118 | if (znode->alt) | |
2119 | /* | |
2120 | * We can no longer be sure of finding this znode by key, so we | |
2121 | * record it in the old_idx tree. | |
2122 | */ | |
2123 | ins_clr_old_idx_znode(c, znode); | |
2124 | ||
2125 | zn = kzalloc(c->max_znode_sz, GFP_NOFS); | |
2126 | if (!zn) | |
2127 | return -ENOMEM; | |
2128 | zn->parent = zp; | |
2129 | zn->level = znode->level; | |
2130 | ||
2131 | /* Decide where to split */ | |
2242c689 AH |
2132 | if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) { |
2133 | /* Try not to split consecutive data keys */ | |
2134 | if (n == c->fanout) { | |
2135 | key1 = &znode->zbranch[n - 1].key; | |
2136 | if (key_inum(c, key1) == key_inum(c, key) && | |
2137 | key_type(c, key1) == UBIFS_DATA_KEY) | |
2138 | appending = 1; | |
2139 | } else | |
2140 | goto check_split; | |
2141 | } else if (appending && n != c->fanout) { | |
2142 | /* Try not to split consecutive data keys */ | |
2143 | appending = 0; | |
2144 | check_split: | |
2145 | if (n >= (c->fanout + 1) / 2) { | |
2146 | key1 = &znode->zbranch[0].key; | |
2147 | if (key_inum(c, key1) == key_inum(c, key) && | |
2148 | key_type(c, key1) == UBIFS_DATA_KEY) { | |
2149 | key1 = &znode->zbranch[n].key; | |
2150 | if (key_inum(c, key1) != key_inum(c, key) || | |
2151 | key_type(c, key1) != UBIFS_DATA_KEY) { | |
2152 | keep = n; | |
2153 | move = c->fanout - keep; | |
2154 | zi = znode; | |
2155 | goto do_split; | |
2156 | } | |
2157 | } | |
2158 | } | |
1e51764a AB |
2159 | } |
2160 | ||
2161 | if (appending) { | |
2162 | keep = c->fanout; | |
2163 | move = 0; | |
2164 | } else { | |
2165 | keep = (c->fanout + 1) / 2; | |
2166 | move = c->fanout - keep; | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * Although we don't at present, we could look at the neighbors and see | |
2171 | * if we can move some zbranches there. | |
2172 | */ | |
2173 | ||
2174 | if (n < keep) { | |
2175 | /* Insert into existing znode */ | |
2176 | zi = znode; | |
2177 | move += 1; | |
2178 | keep -= 1; | |
2179 | } else { | |
2180 | /* Insert into new znode */ | |
2181 | zi = zn; | |
2182 | n -= keep; | |
2183 | /* Re-parent */ | |
2184 | if (zn->level != 0) | |
2185 | zbr->znode->parent = zn; | |
2186 | } | |
2187 | ||
2242c689 AH |
2188 | do_split: |
2189 | ||
1e51764a AB |
2190 | __set_bit(DIRTY_ZNODE, &zn->flags); |
2191 | atomic_long_inc(&c->dirty_zn_cnt); | |
2192 | ||
2193 | zn->child_cnt = move; | |
2194 | znode->child_cnt = keep; | |
2195 | ||
2196 | dbg_tnc("moving %d, keeping %d", move, keep); | |
2197 | ||
2198 | /* Move zbranch */ | |
2199 | for (i = 0; i < move; i++) { | |
2200 | zn->zbranch[i] = znode->zbranch[keep + i]; | |
2201 | /* Re-parent */ | |
2202 | if (zn->level != 0) | |
2203 | if (zn->zbranch[i].znode) { | |
2204 | zn->zbranch[i].znode->parent = zn; | |
2205 | zn->zbranch[i].znode->iip = i; | |
2206 | } | |
2207 | } | |
2208 | ||
2209 | /* Insert new key and branch */ | |
515315a1 | 2210 | dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level); |
1e51764a | 2211 | |
6eb61d58 | 2212 | insert_zbranch(c, zi, zbr, n); |
1e51764a AB |
2213 | |
2214 | /* Insert new znode (produced by spitting) into the parent */ | |
2215 | if (zp) { | |
2242c689 AH |
2216 | if (n == 0 && zi == znode && znode->iip == 0) |
2217 | correct_parent_keys(c, znode); | |
2218 | ||
1e51764a AB |
2219 | /* Locate insertion point */ |
2220 | n = znode->iip + 1; | |
1e51764a AB |
2221 | |
2222 | /* Tail recursion */ | |
2223 | zbr->key = zn->zbranch[0].key; | |
2224 | zbr->znode = zn; | |
2225 | zbr->lnum = 0; | |
2226 | zbr->offs = 0; | |
2227 | zbr->len = 0; | |
2228 | znode = zp; | |
2229 | ||
2230 | goto again; | |
2231 | } | |
2232 | ||
2233 | /* We have to split root znode */ | |
2234 | dbg_tnc("creating new zroot at level %d", znode->level + 1); | |
2235 | ||
2236 | zi = kzalloc(c->max_znode_sz, GFP_NOFS); | |
2237 | if (!zi) | |
2238 | return -ENOMEM; | |
2239 | ||
2240 | zi->child_cnt = 2; | |
2241 | zi->level = znode->level + 1; | |
2242 | ||
2243 | __set_bit(DIRTY_ZNODE, &zi->flags); | |
2244 | atomic_long_inc(&c->dirty_zn_cnt); | |
2245 | ||
2246 | zi->zbranch[0].key = znode->zbranch[0].key; | |
2247 | zi->zbranch[0].znode = znode; | |
2248 | zi->zbranch[0].lnum = c->zroot.lnum; | |
2249 | zi->zbranch[0].offs = c->zroot.offs; | |
2250 | zi->zbranch[0].len = c->zroot.len; | |
2251 | zi->zbranch[1].key = zn->zbranch[0].key; | |
2252 | zi->zbranch[1].znode = zn; | |
2253 | ||
2254 | c->zroot.lnum = 0; | |
2255 | c->zroot.offs = 0; | |
2256 | c->zroot.len = 0; | |
2257 | c->zroot.znode = zi; | |
2258 | ||
2259 | zn->parent = zi; | |
2260 | zn->iip = 1; | |
2261 | znode->parent = zi; | |
2262 | znode->iip = 0; | |
2263 | ||
2264 | return 0; | |
2265 | } | |
2266 | ||
2267 | /** | |
2268 | * ubifs_tnc_add - add a node to TNC. | |
2269 | * @c: UBIFS file-system description object | |
2270 | * @key: key to add | |
2271 | * @lnum: LEB number of node | |
2272 | * @offs: node offset | |
2273 | * @len: node length | |
823838a4 | 2274 | * @hash: The hash over the node |
1e51764a AB |
2275 | * |
2276 | * This function adds a node with key @key to TNC. The node may be new or it may | |
2277 | * obsolete some existing one. Returns %0 on success or negative error code on | |
2278 | * failure. | |
2279 | */ | |
2280 | int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum, | |
823838a4 | 2281 | int offs, int len, const u8 *hash) |
1e51764a AB |
2282 | { |
2283 | int found, n, err = 0; | |
2284 | struct ubifs_znode *znode; | |
2285 | ||
2286 | mutex_lock(&c->tnc_mutex); | |
515315a1 | 2287 | dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len); |
1e51764a AB |
2288 | found = lookup_level0_dirty(c, key, &znode, &n); |
2289 | if (!found) { | |
2290 | struct ubifs_zbranch zbr; | |
2291 | ||
2292 | zbr.znode = NULL; | |
2293 | zbr.lnum = lnum; | |
2294 | zbr.offs = offs; | |
2295 | zbr.len = len; | |
823838a4 | 2296 | ubifs_copy_hash(c, hash, zbr.hash); |
1e51764a AB |
2297 | key_copy(c, key, &zbr.key); |
2298 | err = tnc_insert(c, znode, &zbr, n + 1); | |
2299 | } else if (found == 1) { | |
2300 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2301 | ||
2302 | lnc_free(zbr); | |
2303 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2304 | zbr->lnum = lnum; | |
2305 | zbr->offs = offs; | |
2306 | zbr->len = len; | |
823838a4 | 2307 | ubifs_copy_hash(c, hash, zbr->hash); |
1e51764a AB |
2308 | } else |
2309 | err = found; | |
2310 | if (!err) | |
2311 | err = dbg_check_tnc(c, 0); | |
2312 | mutex_unlock(&c->tnc_mutex); | |
2313 | ||
2314 | return err; | |
2315 | } | |
2316 | ||
2317 | /** | |
2318 | * ubifs_tnc_replace - replace a node in the TNC only if the old node is found. | |
2319 | * @c: UBIFS file-system description object | |
2320 | * @key: key to add | |
2321 | * @old_lnum: LEB number of old node | |
2322 | * @old_offs: old node offset | |
2323 | * @lnum: LEB number of node | |
2324 | * @offs: node offset | |
2325 | * @len: node length | |
2326 | * | |
2327 | * This function replaces a node with key @key in the TNC only if the old node | |
2328 | * is found. This function is called by garbage collection when node are moved. | |
2329 | * Returns %0 on success or negative error code on failure. | |
2330 | */ | |
2331 | int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key, | |
2332 | int old_lnum, int old_offs, int lnum, int offs, int len) | |
2333 | { | |
2334 | int found, n, err = 0; | |
2335 | struct ubifs_znode *znode; | |
2336 | ||
2337 | mutex_lock(&c->tnc_mutex); | |
515315a1 AB |
2338 | dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum, |
2339 | old_offs, lnum, offs, len); | |
1e51764a AB |
2340 | found = lookup_level0_dirty(c, key, &znode, &n); |
2341 | if (found < 0) { | |
2342 | err = found; | |
2343 | goto out_unlock; | |
2344 | } | |
2345 | ||
2346 | if (found == 1) { | |
2347 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2348 | ||
2349 | found = 0; | |
2350 | if (zbr->lnum == old_lnum && zbr->offs == old_offs) { | |
2351 | lnc_free(zbr); | |
2352 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2353 | if (err) | |
2354 | goto out_unlock; | |
2355 | zbr->lnum = lnum; | |
2356 | zbr->offs = offs; | |
2357 | zbr->len = len; | |
2358 | found = 1; | |
2359 | } else if (is_hash_key(c, key)) { | |
2360 | found = resolve_collision_directly(c, key, &znode, &n, | |
2361 | old_lnum, old_offs); | |
2362 | dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d", | |
2363 | found, znode, n, old_lnum, old_offs); | |
2364 | if (found < 0) { | |
2365 | err = found; | |
2366 | goto out_unlock; | |
2367 | } | |
2368 | ||
2369 | if (found) { | |
2370 | /* Ensure the znode is dirtied */ | |
2371 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2372 | znode = dirty_cow_bottom_up(c, znode); |
2373 | if (IS_ERR(znode)) { | |
2374 | err = PTR_ERR(znode); | |
2375 | goto out_unlock; | |
2376 | } | |
1e51764a AB |
2377 | } |
2378 | zbr = &znode->zbranch[n]; | |
2379 | lnc_free(zbr); | |
2380 | err = ubifs_add_dirt(c, zbr->lnum, | |
2381 | zbr->len); | |
2382 | if (err) | |
2383 | goto out_unlock; | |
2384 | zbr->lnum = lnum; | |
2385 | zbr->offs = offs; | |
2386 | zbr->len = len; | |
2387 | } | |
2388 | } | |
2389 | } | |
2390 | ||
2391 | if (!found) | |
2392 | err = ubifs_add_dirt(c, lnum, len); | |
2393 | ||
2394 | if (!err) | |
2395 | err = dbg_check_tnc(c, 0); | |
2396 | ||
2397 | out_unlock: | |
2398 | mutex_unlock(&c->tnc_mutex); | |
2399 | return err; | |
2400 | } | |
2401 | ||
2402 | /** | |
2403 | * ubifs_tnc_add_nm - add a "hashed" node to TNC. | |
2404 | * @c: UBIFS file-system description object | |
2405 | * @key: key to add | |
2406 | * @lnum: LEB number of node | |
2407 | * @offs: node offset | |
2408 | * @len: node length | |
823838a4 | 2409 | * @hash: The hash over the node |
1e51764a AB |
2410 | * @nm: node name |
2411 | * | |
2412 | * This is the same as 'ubifs_tnc_add()' but it should be used with keys which | |
2413 | * may have collisions, like directory entry keys. | |
2414 | */ | |
2415 | int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key, | |
823838a4 | 2416 | int lnum, int offs, int len, const u8 *hash, |
f4f61d2c | 2417 | const struct fscrypt_name *nm) |
1e51764a AB |
2418 | { |
2419 | int found, n, err = 0; | |
2420 | struct ubifs_znode *znode; | |
2421 | ||
2422 | mutex_lock(&c->tnc_mutex); | |
35ee314c | 2423 | dbg_tnck(key, "LEB %d:%d, key ", lnum, offs); |
1e51764a AB |
2424 | found = lookup_level0_dirty(c, key, &znode, &n); |
2425 | if (found < 0) { | |
2426 | err = found; | |
2427 | goto out_unlock; | |
2428 | } | |
2429 | ||
2430 | if (found == 1) { | |
2431 | if (c->replaying) | |
2432 | found = fallible_resolve_collision(c, key, &znode, &n, | |
2433 | nm, 1); | |
2434 | else | |
2435 | found = resolve_collision(c, key, &znode, &n, nm); | |
2436 | dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n); | |
2437 | if (found < 0) { | |
2438 | err = found; | |
2439 | goto out_unlock; | |
2440 | } | |
2441 | ||
2442 | /* Ensure the znode is dirtied */ | |
2443 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2444 | znode = dirty_cow_bottom_up(c, znode); |
2445 | if (IS_ERR(znode)) { | |
2446 | err = PTR_ERR(znode); | |
2447 | goto out_unlock; | |
2448 | } | |
1e51764a AB |
2449 | } |
2450 | ||
2451 | if (found == 1) { | |
2452 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2453 | ||
2454 | lnc_free(zbr); | |
2455 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2456 | zbr->lnum = lnum; | |
2457 | zbr->offs = offs; | |
2458 | zbr->len = len; | |
823838a4 | 2459 | ubifs_copy_hash(c, hash, zbr->hash); |
1e51764a AB |
2460 | goto out_unlock; |
2461 | } | |
2462 | } | |
2463 | ||
2464 | if (!found) { | |
2465 | struct ubifs_zbranch zbr; | |
2466 | ||
2467 | zbr.znode = NULL; | |
2468 | zbr.lnum = lnum; | |
2469 | zbr.offs = offs; | |
2470 | zbr.len = len; | |
823838a4 | 2471 | ubifs_copy_hash(c, hash, zbr.hash); |
1e51764a AB |
2472 | key_copy(c, key, &zbr.key); |
2473 | err = tnc_insert(c, znode, &zbr, n + 1); | |
2474 | if (err) | |
2475 | goto out_unlock; | |
2476 | if (c->replaying) { | |
2477 | /* | |
2478 | * We did not find it in the index so there may be a | |
2479 | * dangling branch still in the index. So we remove it | |
2480 | * by passing 'ubifs_tnc_remove_nm()' the same key but | |
2481 | * an unmatchable name. | |
2482 | */ | |
f4f61d2c | 2483 | struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } }; |
1e51764a AB |
2484 | |
2485 | err = dbg_check_tnc(c, 0); | |
2486 | mutex_unlock(&c->tnc_mutex); | |
2487 | if (err) | |
2488 | return err; | |
2489 | return ubifs_tnc_remove_nm(c, key, &noname); | |
2490 | } | |
2491 | } | |
2492 | ||
2493 | out_unlock: | |
2494 | if (!err) | |
2495 | err = dbg_check_tnc(c, 0); | |
2496 | mutex_unlock(&c->tnc_mutex); | |
2497 | return err; | |
2498 | } | |
2499 | ||
2500 | /** | |
2501 | * tnc_delete - delete a znode form TNC. | |
2502 | * @c: UBIFS file-system description object | |
2503 | * @znode: znode to delete from | |
2504 | * @n: zbranch slot number to delete | |
2505 | * | |
2506 | * This function deletes a leaf node from @n-th slot of @znode. Returns zero in | |
2507 | * case of success and a negative error code in case of failure. | |
2508 | */ | |
2509 | static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n) | |
2510 | { | |
2511 | struct ubifs_zbranch *zbr; | |
2512 | struct ubifs_znode *zp; | |
2513 | int i, err; | |
2514 | ||
2515 | /* Delete without merge for now */ | |
6eb61d58 RW |
2516 | ubifs_assert(c, znode->level == 0); |
2517 | ubifs_assert(c, n >= 0 && n < c->fanout); | |
515315a1 | 2518 | dbg_tnck(&znode->zbranch[n].key, "deleting key "); |
1e51764a AB |
2519 | |
2520 | zbr = &znode->zbranch[n]; | |
2521 | lnc_free(zbr); | |
2522 | ||
2523 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2524 | if (err) { | |
edf6be24 | 2525 | ubifs_dump_znode(c, znode); |
1e51764a AB |
2526 | return err; |
2527 | } | |
2528 | ||
2529 | /* We do not "gap" zbranch slots */ | |
2530 | for (i = n; i < znode->child_cnt - 1; i++) | |
2531 | znode->zbranch[i] = znode->zbranch[i + 1]; | |
2532 | znode->child_cnt -= 1; | |
2533 | ||
2534 | if (znode->child_cnt > 0) | |
2535 | return 0; | |
2536 | ||
2537 | /* | |
2538 | * This was the last zbranch, we have to delete this znode from the | |
2539 | * parent. | |
2540 | */ | |
2541 | ||
2542 | do { | |
6eb61d58 RW |
2543 | ubifs_assert(c, !ubifs_zn_obsolete(znode)); |
2544 | ubifs_assert(c, ubifs_zn_dirty(znode)); | |
1e51764a AB |
2545 | |
2546 | zp = znode->parent; | |
2547 | n = znode->iip; | |
2548 | ||
2549 | atomic_long_dec(&c->dirty_zn_cnt); | |
2550 | ||
2551 | err = insert_old_idx_znode(c, znode); | |
2552 | if (err) | |
2553 | return err; | |
2554 | ||
2555 | if (znode->cnext) { | |
2556 | __set_bit(OBSOLETE_ZNODE, &znode->flags); | |
2557 | atomic_long_inc(&c->clean_zn_cnt); | |
2558 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
2559 | } else | |
2560 | kfree(znode); | |
2561 | znode = zp; | |
2562 | } while (znode->child_cnt == 1); /* while removing last child */ | |
2563 | ||
2564 | /* Remove from znode, entry n - 1 */ | |
2565 | znode->child_cnt -= 1; | |
6eb61d58 | 2566 | ubifs_assert(c, znode->level != 0); |
1e51764a AB |
2567 | for (i = n; i < znode->child_cnt; i++) { |
2568 | znode->zbranch[i] = znode->zbranch[i + 1]; | |
2569 | if (znode->zbranch[i].znode) | |
2570 | znode->zbranch[i].znode->iip = i; | |
2571 | } | |
2572 | ||
2573 | /* | |
2574 | * If this is the root and it has only 1 child then | |
2575 | * collapse the tree. | |
2576 | */ | |
2577 | if (!znode->parent) { | |
2578 | while (znode->child_cnt == 1 && znode->level != 0) { | |
2579 | zp = znode; | |
2580 | zbr = &znode->zbranch[0]; | |
2581 | znode = get_znode(c, znode, 0); | |
2582 | if (IS_ERR(znode)) | |
2583 | return PTR_ERR(znode); | |
2584 | znode = dirty_cow_znode(c, zbr); | |
2585 | if (IS_ERR(znode)) | |
2586 | return PTR_ERR(znode); | |
2587 | znode->parent = NULL; | |
2588 | znode->iip = 0; | |
2589 | if (c->zroot.len) { | |
2590 | err = insert_old_idx(c, c->zroot.lnum, | |
2591 | c->zroot.offs); | |
2592 | if (err) | |
2593 | return err; | |
2594 | } | |
2595 | c->zroot.lnum = zbr->lnum; | |
2596 | c->zroot.offs = zbr->offs; | |
2597 | c->zroot.len = zbr->len; | |
2598 | c->zroot.znode = znode; | |
6eb61d58 RW |
2599 | ubifs_assert(c, !ubifs_zn_obsolete(zp)); |
2600 | ubifs_assert(c, ubifs_zn_dirty(zp)); | |
1e51764a AB |
2601 | atomic_long_dec(&c->dirty_zn_cnt); |
2602 | ||
2603 | if (zp->cnext) { | |
2604 | __set_bit(OBSOLETE_ZNODE, &zp->flags); | |
2605 | atomic_long_inc(&c->clean_zn_cnt); | |
2606 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
2607 | } else | |
2608 | kfree(zp); | |
2609 | } | |
2610 | } | |
2611 | ||
2612 | return 0; | |
2613 | } | |
2614 | ||
2615 | /** | |
2616 | * ubifs_tnc_remove - remove an index entry of a node. | |
2617 | * @c: UBIFS file-system description object | |
2618 | * @key: key of node | |
2619 | * | |
2620 | * Returns %0 on success or negative error code on failure. | |
2621 | */ | |
2622 | int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key) | |
2623 | { | |
2624 | int found, n, err = 0; | |
2625 | struct ubifs_znode *znode; | |
2626 | ||
2627 | mutex_lock(&c->tnc_mutex); | |
515315a1 | 2628 | dbg_tnck(key, "key "); |
1e51764a AB |
2629 | found = lookup_level0_dirty(c, key, &znode, &n); |
2630 | if (found < 0) { | |
2631 | err = found; | |
2632 | goto out_unlock; | |
2633 | } | |
2634 | if (found == 1) | |
2635 | err = tnc_delete(c, znode, n); | |
2636 | if (!err) | |
2637 | err = dbg_check_tnc(c, 0); | |
2638 | ||
2639 | out_unlock: | |
2640 | mutex_unlock(&c->tnc_mutex); | |
2641 | return err; | |
2642 | } | |
2643 | ||
2644 | /** | |
2645 | * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node. | |
2646 | * @c: UBIFS file-system description object | |
2647 | * @key: key of node | |
2648 | * @nm: directory entry name | |
2649 | * | |
2650 | * Returns %0 on success or negative error code on failure. | |
2651 | */ | |
2652 | int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key, | |
f4f61d2c | 2653 | const struct fscrypt_name *nm) |
1e51764a AB |
2654 | { |
2655 | int n, err; | |
2656 | struct ubifs_znode *znode; | |
2657 | ||
2658 | mutex_lock(&c->tnc_mutex); | |
35ee314c | 2659 | dbg_tnck(key, "key "); |
1e51764a AB |
2660 | err = lookup_level0_dirty(c, key, &znode, &n); |
2661 | if (err < 0) | |
2662 | goto out_unlock; | |
2663 | ||
2664 | if (err) { | |
2665 | if (c->replaying) | |
2666 | err = fallible_resolve_collision(c, key, &znode, &n, | |
2667 | nm, 0); | |
2668 | else | |
2669 | err = resolve_collision(c, key, &znode, &n, nm); | |
2670 | dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); | |
2671 | if (err < 0) | |
2672 | goto out_unlock; | |
2673 | if (err) { | |
2674 | /* Ensure the znode is dirtied */ | |
2675 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
c4361570 AB |
2676 | znode = dirty_cow_bottom_up(c, znode); |
2677 | if (IS_ERR(znode)) { | |
2678 | err = PTR_ERR(znode); | |
2679 | goto out_unlock; | |
2680 | } | |
1e51764a AB |
2681 | } |
2682 | err = tnc_delete(c, znode, n); | |
2683 | } | |
2684 | } | |
2685 | ||
2686 | out_unlock: | |
2687 | if (!err) | |
2688 | err = dbg_check_tnc(c, 0); | |
2689 | mutex_unlock(&c->tnc_mutex); | |
2690 | return err; | |
2691 | } | |
2692 | ||
781f675e RW |
2693 | /** |
2694 | * ubifs_tnc_remove_dh - remove an index entry for a "double hashed" node. | |
2695 | * @c: UBIFS file-system description object | |
2696 | * @key: key of node | |
2697 | * @cookie: node cookie for collision resolution | |
2698 | * | |
2699 | * Returns %0 on success or negative error code on failure. | |
2700 | */ | |
2701 | int ubifs_tnc_remove_dh(struct ubifs_info *c, const union ubifs_key *key, | |
2702 | uint32_t cookie) | |
2703 | { | |
2704 | int n, err; | |
2705 | struct ubifs_znode *znode; | |
2706 | struct ubifs_dent_node *dent; | |
2707 | struct ubifs_zbranch *zbr; | |
2708 | ||
2709 | if (!c->double_hash) | |
2710 | return -EOPNOTSUPP; | |
2711 | ||
2712 | mutex_lock(&c->tnc_mutex); | |
2713 | err = lookup_level0_dirty(c, key, &znode, &n); | |
2714 | if (err <= 0) | |
2715 | goto out_unlock; | |
2716 | ||
2717 | zbr = &znode->zbranch[n]; | |
2718 | dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS); | |
2719 | if (!dent) { | |
2720 | err = -ENOMEM; | |
2721 | goto out_unlock; | |
2722 | } | |
2723 | ||
2724 | err = tnc_read_hashed_node(c, zbr, dent); | |
2725 | if (err) | |
2726 | goto out_free; | |
2727 | ||
2728 | /* If the cookie does not match, we're facing a hash collision. */ | |
2729 | if (le32_to_cpu(dent->cookie) != cookie) { | |
2730 | union ubifs_key start_key; | |
2731 | ||
2732 | lowest_dent_key(c, &start_key, key_inum(c, key)); | |
2733 | ||
2734 | err = ubifs_lookup_level0(c, &start_key, &znode, &n); | |
2735 | if (unlikely(err < 0)) | |
2736 | goto out_free; | |
2737 | ||
2738 | err = search_dh_cookie(c, key, dent, cookie, &znode, &n); | |
2739 | if (err) | |
2740 | goto out_free; | |
2741 | } | |
2742 | ||
2743 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
2744 | znode = dirty_cow_bottom_up(c, znode); | |
2745 | if (IS_ERR(znode)) { | |
2746 | err = PTR_ERR(znode); | |
2747 | goto out_free; | |
2748 | } | |
2749 | } | |
2750 | err = tnc_delete(c, znode, n); | |
2751 | ||
2752 | out_free: | |
2753 | kfree(dent); | |
2754 | out_unlock: | |
2755 | if (!err) | |
2756 | err = dbg_check_tnc(c, 0); | |
2757 | mutex_unlock(&c->tnc_mutex); | |
2758 | return err; | |
2759 | } | |
2760 | ||
1e51764a AB |
2761 | /** |
2762 | * key_in_range - determine if a key falls within a range of keys. | |
2763 | * @c: UBIFS file-system description object | |
2764 | * @key: key to check | |
2765 | * @from_key: lowest key in range | |
2766 | * @to_key: highest key in range | |
2767 | * | |
2768 | * This function returns %1 if the key is in range and %0 otherwise. | |
2769 | */ | |
2770 | static int key_in_range(struct ubifs_info *c, union ubifs_key *key, | |
2771 | union ubifs_key *from_key, union ubifs_key *to_key) | |
2772 | { | |
2773 | if (keys_cmp(c, key, from_key) < 0) | |
2774 | return 0; | |
2775 | if (keys_cmp(c, key, to_key) > 0) | |
2776 | return 0; | |
2777 | return 1; | |
2778 | } | |
2779 | ||
2780 | /** | |
2781 | * ubifs_tnc_remove_range - remove index entries in range. | |
2782 | * @c: UBIFS file-system description object | |
2783 | * @from_key: lowest key to remove | |
2784 | * @to_key: highest key to remove | |
2785 | * | |
2786 | * This function removes index entries starting at @from_key and ending at | |
2787 | * @to_key. This function returns zero in case of success and a negative error | |
2788 | * code in case of failure. | |
2789 | */ | |
2790 | int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key, | |
2791 | union ubifs_key *to_key) | |
2792 | { | |
2793 | int i, n, k, err = 0; | |
2794 | struct ubifs_znode *znode; | |
2795 | union ubifs_key *key; | |
2796 | ||
2797 | mutex_lock(&c->tnc_mutex); | |
2798 | while (1) { | |
2799 | /* Find first level 0 znode that contains keys to remove */ | |
2800 | err = ubifs_lookup_level0(c, from_key, &znode, &n); | |
2801 | if (err < 0) | |
2802 | goto out_unlock; | |
2803 | ||
2804 | if (err) | |
2805 | key = from_key; | |
2806 | else { | |
2807 | err = tnc_next(c, &znode, &n); | |
2808 | if (err == -ENOENT) { | |
2809 | err = 0; | |
2810 | goto out_unlock; | |
2811 | } | |
2812 | if (err < 0) | |
2813 | goto out_unlock; | |
2814 | key = &znode->zbranch[n].key; | |
2815 | if (!key_in_range(c, key, from_key, to_key)) { | |
2816 | err = 0; | |
2817 | goto out_unlock; | |
2818 | } | |
2819 | } | |
2820 | ||
2821 | /* Ensure the znode is dirtied */ | |
2822 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2823 | znode = dirty_cow_bottom_up(c, znode); |
2824 | if (IS_ERR(znode)) { | |
2825 | err = PTR_ERR(znode); | |
2826 | goto out_unlock; | |
2827 | } | |
1e51764a AB |
2828 | } |
2829 | ||
2830 | /* Remove all keys in range except the first */ | |
2831 | for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) { | |
2832 | key = &znode->zbranch[i].key; | |
2833 | if (!key_in_range(c, key, from_key, to_key)) | |
2834 | break; | |
2835 | lnc_free(&znode->zbranch[i]); | |
2836 | err = ubifs_add_dirt(c, znode->zbranch[i].lnum, | |
2837 | znode->zbranch[i].len); | |
2838 | if (err) { | |
edf6be24 | 2839 | ubifs_dump_znode(c, znode); |
1e51764a AB |
2840 | goto out_unlock; |
2841 | } | |
515315a1 | 2842 | dbg_tnck(key, "removing key "); |
1e51764a AB |
2843 | } |
2844 | if (k) { | |
2845 | for (i = n + 1 + k; i < znode->child_cnt; i++) | |
2846 | znode->zbranch[i - k] = znode->zbranch[i]; | |
2847 | znode->child_cnt -= k; | |
2848 | } | |
2849 | ||
2850 | /* Now delete the first */ | |
2851 | err = tnc_delete(c, znode, n); | |
2852 | if (err) | |
2853 | goto out_unlock; | |
2854 | } | |
2855 | ||
2856 | out_unlock: | |
2857 | if (!err) | |
2858 | err = dbg_check_tnc(c, 0); | |
2859 | mutex_unlock(&c->tnc_mutex); | |
2860 | return err; | |
2861 | } | |
2862 | ||
2863 | /** | |
2864 | * ubifs_tnc_remove_ino - remove an inode from TNC. | |
2865 | * @c: UBIFS file-system description object | |
2866 | * @inum: inode number to remove | |
2867 | * | |
2868 | * This function remove inode @inum and all the extended attributes associated | |
2869 | * with the anode from TNC and returns zero in case of success or a negative | |
2870 | * error code in case of failure. | |
2871 | */ | |
2872 | int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum) | |
2873 | { | |
2874 | union ubifs_key key1, key2; | |
2875 | struct ubifs_dent_node *xent, *pxent = NULL; | |
f4f61d2c | 2876 | struct fscrypt_name nm = {0}; |
1e51764a | 2877 | |
e84461ad | 2878 | dbg_tnc("ino %lu", (unsigned long)inum); |
1e51764a AB |
2879 | |
2880 | /* | |
2881 | * Walk all extended attribute entries and remove them together with | |
2882 | * corresponding extended attribute inodes. | |
2883 | */ | |
2884 | lowest_xent_key(c, &key1, inum); | |
2885 | while (1) { | |
2886 | ino_t xattr_inum; | |
2887 | int err; | |
2888 | ||
2889 | xent = ubifs_tnc_next_ent(c, &key1, &nm); | |
2890 | if (IS_ERR(xent)) { | |
2891 | err = PTR_ERR(xent); | |
2892 | if (err == -ENOENT) | |
2893 | break; | |
2894 | return err; | |
2895 | } | |
2896 | ||
2897 | xattr_inum = le64_to_cpu(xent->inum); | |
e84461ad AB |
2898 | dbg_tnc("xent '%s', ino %lu", xent->name, |
2899 | (unsigned long)xattr_inum); | |
1e51764a | 2900 | |
272eda82 RW |
2901 | ubifs_evict_xattr_inode(c, xattr_inum); |
2902 | ||
f4f61d2c RW |
2903 | fname_name(&nm) = xent->name; |
2904 | fname_len(&nm) = le16_to_cpu(xent->nlen); | |
1e51764a AB |
2905 | err = ubifs_tnc_remove_nm(c, &key1, &nm); |
2906 | if (err) { | |
2907 | kfree(xent); | |
2908 | return err; | |
2909 | } | |
2910 | ||
2911 | lowest_ino_key(c, &key1, xattr_inum); | |
2912 | highest_ino_key(c, &key2, xattr_inum); | |
2913 | err = ubifs_tnc_remove_range(c, &key1, &key2); | |
2914 | if (err) { | |
2915 | kfree(xent); | |
2916 | return err; | |
2917 | } | |
2918 | ||
2919 | kfree(pxent); | |
2920 | pxent = xent; | |
2921 | key_read(c, &xent->key, &key1); | |
2922 | } | |
2923 | ||
2924 | kfree(pxent); | |
2925 | lowest_ino_key(c, &key1, inum); | |
2926 | highest_ino_key(c, &key2, inum); | |
2927 | ||
2928 | return ubifs_tnc_remove_range(c, &key1, &key2); | |
2929 | } | |
2930 | ||
2931 | /** | |
2932 | * ubifs_tnc_next_ent - walk directory or extended attribute entries. | |
2933 | * @c: UBIFS file-system description object | |
2934 | * @key: key of last entry | |
2935 | * @nm: name of last entry found or %NULL | |
2936 | * | |
2937 | * This function finds and reads the next directory or extended attribute entry | |
2938 | * after the given key (@key) if there is one. @nm is used to resolve | |
2939 | * collisions. | |
2940 | * | |
2941 | * If the name of the current entry is not known and only the key is known, | |
2942 | * @nm->name has to be %NULL. In this case the semantics of this function is a | |
2943 | * little bit different and it returns the entry corresponding to this key, not | |
2944 | * the next one. If the key was not found, the closest "right" entry is | |
2945 | * returned. | |
2946 | * | |
2947 | * If the fist entry has to be found, @key has to contain the lowest possible | |
2948 | * key value for this inode and @name has to be %NULL. | |
2949 | * | |
2950 | * This function returns the found directory or extended attribute entry node | |
2951 | * in case of success, %-ENOENT is returned if no entry was found, and a | |
2952 | * negative error code is returned in case of failure. | |
2953 | */ | |
2954 | struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c, | |
2955 | union ubifs_key *key, | |
f4f61d2c | 2956 | const struct fscrypt_name *nm) |
1e51764a AB |
2957 | { |
2958 | int n, err, type = key_type(c, key); | |
2959 | struct ubifs_znode *znode; | |
2960 | struct ubifs_dent_node *dent; | |
2961 | struct ubifs_zbranch *zbr; | |
2962 | union ubifs_key *dkey; | |
2963 | ||
35ee314c | 2964 | dbg_tnck(key, "key "); |
6eb61d58 | 2965 | ubifs_assert(c, is_hash_key(c, key)); |
1e51764a AB |
2966 | |
2967 | mutex_lock(&c->tnc_mutex); | |
2968 | err = ubifs_lookup_level0(c, key, &znode, &n); | |
2969 | if (unlikely(err < 0)) | |
2970 | goto out_unlock; | |
2971 | ||
f4f61d2c | 2972 | if (fname_len(nm) > 0) { |
1e51764a AB |
2973 | if (err) { |
2974 | /* Handle collisions */ | |
1cb51a15 RW |
2975 | if (c->replaying) |
2976 | err = fallible_resolve_collision(c, key, &znode, &n, | |
2977 | nm, 0); | |
2978 | else | |
2979 | err = resolve_collision(c, key, &znode, &n, nm); | |
1e51764a AB |
2980 | dbg_tnc("rc returned %d, znode %p, n %d", |
2981 | err, znode, n); | |
2982 | if (unlikely(err < 0)) | |
2983 | goto out_unlock; | |
2984 | } | |
2985 | ||
2986 | /* Now find next entry */ | |
2987 | err = tnc_next(c, &znode, &n); | |
2988 | if (unlikely(err)) | |
2989 | goto out_unlock; | |
2990 | } else { | |
2991 | /* | |
2992 | * The full name of the entry was not given, in which case the | |
2993 | * behavior of this function is a little different and it | |
2994 | * returns current entry, not the next one. | |
2995 | */ | |
2996 | if (!err) { | |
2997 | /* | |
2998 | * However, the given key does not exist in the TNC | |
2999 | * tree and @znode/@n variables contain the closest | |
3000 | * "preceding" element. Switch to the next one. | |
3001 | */ | |
3002 | err = tnc_next(c, &znode, &n); | |
3003 | if (err) | |
3004 | goto out_unlock; | |
3005 | } | |
3006 | } | |
3007 | ||
3008 | zbr = &znode->zbranch[n]; | |
3009 | dent = kmalloc(zbr->len, GFP_NOFS); | |
3010 | if (unlikely(!dent)) { | |
3011 | err = -ENOMEM; | |
3012 | goto out_unlock; | |
3013 | } | |
3014 | ||
3015 | /* | |
3016 | * The above 'tnc_next()' call could lead us to the next inode, check | |
3017 | * this. | |
3018 | */ | |
3019 | dkey = &zbr->key; | |
3020 | if (key_inum(c, dkey) != key_inum(c, key) || | |
3021 | key_type(c, dkey) != type) { | |
3022 | err = -ENOENT; | |
3023 | goto out_free; | |
3024 | } | |
3025 | ||
b91dc981 | 3026 | err = tnc_read_hashed_node(c, zbr, dent); |
1e51764a AB |
3027 | if (unlikely(err)) |
3028 | goto out_free; | |
3029 | ||
3030 | mutex_unlock(&c->tnc_mutex); | |
3031 | return dent; | |
3032 | ||
3033 | out_free: | |
3034 | kfree(dent); | |
3035 | out_unlock: | |
3036 | mutex_unlock(&c->tnc_mutex); | |
3037 | return ERR_PTR(err); | |
3038 | } | |
3039 | ||
3040 | /** | |
3041 | * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit. | |
3042 | * @c: UBIFS file-system description object | |
3043 | * | |
3044 | * Destroy left-over obsolete znodes from a failed commit. | |
3045 | */ | |
3046 | static void tnc_destroy_cnext(struct ubifs_info *c) | |
3047 | { | |
3048 | struct ubifs_znode *cnext; | |
3049 | ||
3050 | if (!c->cnext) | |
3051 | return; | |
6eb61d58 | 3052 | ubifs_assert(c, c->cmt_state == COMMIT_BROKEN); |
1e51764a AB |
3053 | cnext = c->cnext; |
3054 | do { | |
3055 | struct ubifs_znode *znode = cnext; | |
3056 | ||
3057 | cnext = cnext->cnext; | |
f42eed7c | 3058 | if (ubifs_zn_obsolete(znode)) |
1e51764a AB |
3059 | kfree(znode); |
3060 | } while (cnext && cnext != c->cnext); | |
3061 | } | |
3062 | ||
3063 | /** | |
3064 | * ubifs_tnc_close - close TNC subsystem and free all related resources. | |
3065 | * @c: UBIFS file-system description object | |
3066 | */ | |
3067 | void ubifs_tnc_close(struct ubifs_info *c) | |
3068 | { | |
1e51764a AB |
3069 | tnc_destroy_cnext(c); |
3070 | if (c->zroot.znode) { | |
380347e9 | 3071 | long n, freed; |
83707237 | 3072 | |
83707237 | 3073 | n = atomic_long_read(&c->clean_zn_cnt); |
6eb61d58 RW |
3074 | freed = ubifs_destroy_tnc_subtree(c, c->zroot.znode); |
3075 | ubifs_assert(c, freed == n); | |
83707237 | 3076 | atomic_long_sub(n, &ubifs_clean_zn_cnt); |
1e51764a AB |
3077 | } |
3078 | kfree(c->gap_lebs); | |
3079 | kfree(c->ilebs); | |
3080 | destroy_old_idx(c); | |
3081 | } | |
3082 | ||
3083 | /** | |
3084 | * left_znode - get the znode to the left. | |
3085 | * @c: UBIFS file-system description object | |
3086 | * @znode: znode | |
3087 | * | |
3088 | * This function returns a pointer to the znode to the left of @znode or NULL if | |
3089 | * there is not one. A negative error code is returned on failure. | |
3090 | */ | |
3091 | static struct ubifs_znode *left_znode(struct ubifs_info *c, | |
3092 | struct ubifs_znode *znode) | |
3093 | { | |
3094 | int level = znode->level; | |
3095 | ||
3096 | while (1) { | |
3097 | int n = znode->iip - 1; | |
3098 | ||
3099 | /* Go up until we can go left */ | |
3100 | znode = znode->parent; | |
3101 | if (!znode) | |
3102 | return NULL; | |
3103 | if (n >= 0) { | |
3104 | /* Now go down the rightmost branch to 'level' */ | |
3105 | znode = get_znode(c, znode, n); | |
3106 | if (IS_ERR(znode)) | |
3107 | return znode; | |
3108 | while (znode->level != level) { | |
3109 | n = znode->child_cnt - 1; | |
3110 | znode = get_znode(c, znode, n); | |
3111 | if (IS_ERR(znode)) | |
3112 | return znode; | |
3113 | } | |
3114 | break; | |
3115 | } | |
3116 | } | |
3117 | return znode; | |
3118 | } | |
3119 | ||
3120 | /** | |
3121 | * right_znode - get the znode to the right. | |
3122 | * @c: UBIFS file-system description object | |
3123 | * @znode: znode | |
3124 | * | |
3125 | * This function returns a pointer to the znode to the right of @znode or NULL | |
3126 | * if there is not one. A negative error code is returned on failure. | |
3127 | */ | |
3128 | static struct ubifs_znode *right_znode(struct ubifs_info *c, | |
3129 | struct ubifs_znode *znode) | |
3130 | { | |
3131 | int level = znode->level; | |
3132 | ||
3133 | while (1) { | |
3134 | int n = znode->iip + 1; | |
3135 | ||
3136 | /* Go up until we can go right */ | |
3137 | znode = znode->parent; | |
3138 | if (!znode) | |
3139 | return NULL; | |
3140 | if (n < znode->child_cnt) { | |
3141 | /* Now go down the leftmost branch to 'level' */ | |
3142 | znode = get_znode(c, znode, n); | |
3143 | if (IS_ERR(znode)) | |
3144 | return znode; | |
3145 | while (znode->level != level) { | |
3146 | znode = get_znode(c, znode, 0); | |
3147 | if (IS_ERR(znode)) | |
3148 | return znode; | |
3149 | } | |
3150 | break; | |
3151 | } | |
3152 | } | |
3153 | return znode; | |
3154 | } | |
3155 | ||
3156 | /** | |
3157 | * lookup_znode - find a particular indexing node from TNC. | |
3158 | * @c: UBIFS file-system description object | |
3159 | * @key: index node key to lookup | |
3160 | * @level: index node level | |
3161 | * @lnum: index node LEB number | |
3162 | * @offs: index node offset | |
3163 | * | |
3164 | * This function searches an indexing node by its first key @key and its | |
3165 | * address @lnum:@offs. It looks up the indexing tree by pulling all indexing | |
ba2f48f7 | 3166 | * nodes it traverses to TNC. This function is called for indexing nodes which |
1e51764a AB |
3167 | * were found on the media by scanning, for example when garbage-collecting or |
3168 | * when doing in-the-gaps commit. This means that the indexing node which is | |
3169 | * looked for does not have to have exactly the same leftmost key @key, because | |
3170 | * the leftmost key may have been changed, in which case TNC will contain a | |
3171 | * dirty znode which still refers the same @lnum:@offs. This function is clever | |
3172 | * enough to recognize such indexing nodes. | |
3173 | * | |
3174 | * Note, if a znode was deleted or changed too much, then this function will | |
3175 | * not find it. For situations like this UBIFS has the old index RB-tree | |
3176 | * (indexed by @lnum:@offs). | |
3177 | * | |
3178 | * This function returns a pointer to the znode found or %NULL if it is not | |
3179 | * found. A negative error code is returned on failure. | |
3180 | */ | |
3181 | static struct ubifs_znode *lookup_znode(struct ubifs_info *c, | |
3182 | union ubifs_key *key, int level, | |
3183 | int lnum, int offs) | |
3184 | { | |
3185 | struct ubifs_znode *znode, *zn; | |
3186 | int n, nn; | |
3187 | ||
6eb61d58 | 3188 | ubifs_assert(c, key_type(c, key) < UBIFS_INVALID_KEY); |
ba2f48f7 | 3189 | |
1e51764a AB |
3190 | /* |
3191 | * The arguments have probably been read off flash, so don't assume | |
3192 | * they are valid. | |
3193 | */ | |
3194 | if (level < 0) | |
3195 | return ERR_PTR(-EINVAL); | |
3196 | ||
3197 | /* Get the root znode */ | |
3198 | znode = c->zroot.znode; | |
3199 | if (!znode) { | |
3200 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
3201 | if (IS_ERR(znode)) | |
3202 | return znode; | |
3203 | } | |
3204 | /* Check if it is the one we are looking for */ | |
3205 | if (c->zroot.lnum == lnum && c->zroot.offs == offs) | |
3206 | return znode; | |
3207 | /* Descend to the parent level i.e. (level + 1) */ | |
3208 | if (level >= znode->level) | |
3209 | return NULL; | |
3210 | while (1) { | |
3211 | ubifs_search_zbranch(c, znode, key, &n); | |
3212 | if (n < 0) { | |
3213 | /* | |
3214 | * We reached a znode where the leftmost key is greater | |
3215 | * than the key we are searching for. This is the same | |
3216 | * situation as the one described in a huge comment at | |
3217 | * the end of the 'ubifs_lookup_level0()' function. And | |
3218 | * for exactly the same reasons we have to try to look | |
3219 | * left before giving up. | |
3220 | */ | |
3221 | znode = left_znode(c, znode); | |
3222 | if (!znode) | |
3223 | return NULL; | |
3224 | if (IS_ERR(znode)) | |
3225 | return znode; | |
3226 | ubifs_search_zbranch(c, znode, key, &n); | |
6eb61d58 | 3227 | ubifs_assert(c, n >= 0); |
1e51764a AB |
3228 | } |
3229 | if (znode->level == level + 1) | |
3230 | break; | |
3231 | znode = get_znode(c, znode, n); | |
3232 | if (IS_ERR(znode)) | |
3233 | return znode; | |
3234 | } | |
3235 | /* Check if the child is the one we are looking for */ | |
3236 | if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs) | |
3237 | return get_znode(c, znode, n); | |
3238 | /* If the key is unique, there is nowhere else to look */ | |
3239 | if (!is_hash_key(c, key)) | |
3240 | return NULL; | |
3241 | /* | |
3242 | * The key is not unique and so may be also in the znodes to either | |
3243 | * side. | |
3244 | */ | |
3245 | zn = znode; | |
3246 | nn = n; | |
3247 | /* Look left */ | |
3248 | while (1) { | |
3249 | /* Move one branch to the left */ | |
3250 | if (n) | |
3251 | n -= 1; | |
3252 | else { | |
3253 | znode = left_znode(c, znode); | |
3254 | if (!znode) | |
3255 | break; | |
3256 | if (IS_ERR(znode)) | |
3257 | return znode; | |
3258 | n = znode->child_cnt - 1; | |
3259 | } | |
3260 | /* Check it */ | |
3261 | if (znode->zbranch[n].lnum == lnum && | |
3262 | znode->zbranch[n].offs == offs) | |
3263 | return get_znode(c, znode, n); | |
3264 | /* Stop if the key is less than the one we are looking for */ | |
3265 | if (keys_cmp(c, &znode->zbranch[n].key, key) < 0) | |
3266 | break; | |
3267 | } | |
3268 | /* Back to the middle */ | |
3269 | znode = zn; | |
3270 | n = nn; | |
3271 | /* Look right */ | |
3272 | while (1) { | |
3273 | /* Move one branch to the right */ | |
3274 | if (++n >= znode->child_cnt) { | |
3275 | znode = right_znode(c, znode); | |
3276 | if (!znode) | |
3277 | break; | |
3278 | if (IS_ERR(znode)) | |
3279 | return znode; | |
3280 | n = 0; | |
3281 | } | |
3282 | /* Check it */ | |
3283 | if (znode->zbranch[n].lnum == lnum && | |
3284 | znode->zbranch[n].offs == offs) | |
3285 | return get_znode(c, znode, n); | |
3286 | /* Stop if the key is greater than the one we are looking for */ | |
3287 | if (keys_cmp(c, &znode->zbranch[n].key, key) > 0) | |
3288 | break; | |
3289 | } | |
3290 | return NULL; | |
3291 | } | |
3292 | ||
3293 | /** | |
3294 | * is_idx_node_in_tnc - determine if an index node is in the TNC. | |
3295 | * @c: UBIFS file-system description object | |
3296 | * @key: key of index node | |
3297 | * @level: index node level | |
3298 | * @lnum: LEB number of index node | |
3299 | * @offs: offset of index node | |
3300 | * | |
3301 | * This function returns %0 if the index node is not referred to in the TNC, %1 | |
3302 | * if the index node is referred to in the TNC and the corresponding znode is | |
3303 | * dirty, %2 if an index node is referred to in the TNC and the corresponding | |
3304 | * znode is clean, and a negative error code in case of failure. | |
3305 | * | |
3306 | * Note, the @key argument has to be the key of the first child. Also note, | |
3307 | * this function relies on the fact that 0:0 is never a valid LEB number and | |
3308 | * offset for a main-area node. | |
3309 | */ | |
3310 | int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level, | |
3311 | int lnum, int offs) | |
3312 | { | |
3313 | struct ubifs_znode *znode; | |
3314 | ||
3315 | znode = lookup_znode(c, key, level, lnum, offs); | |
3316 | if (!znode) | |
3317 | return 0; | |
3318 | if (IS_ERR(znode)) | |
3319 | return PTR_ERR(znode); | |
3320 | ||
3321 | return ubifs_zn_dirty(znode) ? 1 : 2; | |
3322 | } | |
3323 | ||
3324 | /** | |
3325 | * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC. | |
3326 | * @c: UBIFS file-system description object | |
3327 | * @key: node key | |
3328 | * @lnum: node LEB number | |
3329 | * @offs: node offset | |
3330 | * | |
3331 | * This function returns %1 if the node is referred to in the TNC, %0 if it is | |
3332 | * not, and a negative error code in case of failure. | |
3333 | * | |
3334 | * Note, this function relies on the fact that 0:0 is never a valid LEB number | |
3335 | * and offset for a main-area node. | |
3336 | */ | |
3337 | static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, | |
3338 | int lnum, int offs) | |
3339 | { | |
3340 | struct ubifs_zbranch *zbr; | |
3341 | struct ubifs_znode *znode, *zn; | |
3342 | int n, found, err, nn; | |
3343 | const int unique = !is_hash_key(c, key); | |
3344 | ||
3345 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
3346 | if (found < 0) | |
3347 | return found; /* Error code */ | |
3348 | if (!found) | |
3349 | return 0; | |
3350 | zbr = &znode->zbranch[n]; | |
3351 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3352 | return 1; /* Found it */ | |
3353 | if (unique) | |
3354 | return 0; | |
3355 | /* | |
3356 | * Because the key is not unique, we have to look left | |
3357 | * and right as well | |
3358 | */ | |
3359 | zn = znode; | |
3360 | nn = n; | |
3361 | /* Look left */ | |
3362 | while (1) { | |
3363 | err = tnc_prev(c, &znode, &n); | |
3364 | if (err == -ENOENT) | |
3365 | break; | |
3366 | if (err) | |
3367 | return err; | |
3368 | if (keys_cmp(c, key, &znode->zbranch[n].key)) | |
3369 | break; | |
3370 | zbr = &znode->zbranch[n]; | |
3371 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3372 | return 1; /* Found it */ | |
3373 | } | |
3374 | /* Look right */ | |
3375 | znode = zn; | |
3376 | n = nn; | |
3377 | while (1) { | |
3378 | err = tnc_next(c, &znode, &n); | |
3379 | if (err) { | |
3380 | if (err == -ENOENT) | |
3381 | return 0; | |
3382 | return err; | |
3383 | } | |
3384 | if (keys_cmp(c, key, &znode->zbranch[n].key)) | |
3385 | break; | |
3386 | zbr = &znode->zbranch[n]; | |
3387 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3388 | return 1; /* Found it */ | |
3389 | } | |
3390 | return 0; | |
3391 | } | |
3392 | ||
3393 | /** | |
3394 | * ubifs_tnc_has_node - determine whether a node is in the TNC. | |
3395 | * @c: UBIFS file-system description object | |
3396 | * @key: node key | |
3397 | * @level: index node level (if it is an index node) | |
3398 | * @lnum: node LEB number | |
3399 | * @offs: node offset | |
3400 | * @is_idx: non-zero if the node is an index node | |
3401 | * | |
3402 | * This function returns %1 if the node is in the TNC, %0 if it is not, and a | |
3403 | * negative error code in case of failure. For index nodes, @key has to be the | |
3404 | * key of the first child. An index node is considered to be in the TNC only if | |
3405 | * the corresponding znode is clean or has not been loaded. | |
3406 | */ | |
3407 | int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level, | |
3408 | int lnum, int offs, int is_idx) | |
3409 | { | |
3410 | int err; | |
3411 | ||
3412 | mutex_lock(&c->tnc_mutex); | |
3413 | if (is_idx) { | |
3414 | err = is_idx_node_in_tnc(c, key, level, lnum, offs); | |
3415 | if (err < 0) | |
3416 | goto out_unlock; | |
3417 | if (err == 1) | |
3418 | /* The index node was found but it was dirty */ | |
3419 | err = 0; | |
3420 | else if (err == 2) | |
3421 | /* The index node was found and it was clean */ | |
3422 | err = 1; | |
3423 | else | |
3424 | BUG_ON(err != 0); | |
3425 | } else | |
3426 | err = is_leaf_node_in_tnc(c, key, lnum, offs); | |
3427 | ||
3428 | out_unlock: | |
3429 | mutex_unlock(&c->tnc_mutex); | |
3430 | return err; | |
3431 | } | |
3432 | ||
3433 | /** | |
3434 | * ubifs_dirty_idx_node - dirty an index node. | |
3435 | * @c: UBIFS file-system description object | |
3436 | * @key: index node key | |
3437 | * @level: index node level | |
3438 | * @lnum: index node LEB number | |
3439 | * @offs: index node offset | |
3440 | * | |
3441 | * This function loads and dirties an index node so that it can be garbage | |
3442 | * collected. The @key argument has to be the key of the first child. This | |
3443 | * function relies on the fact that 0:0 is never a valid LEB number and offset | |
3444 | * for a main-area node. Returns %0 on success and a negative error code on | |
3445 | * failure. | |
3446 | */ | |
3447 | int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level, | |
3448 | int lnum, int offs) | |
3449 | { | |
3450 | struct ubifs_znode *znode; | |
3451 | int err = 0; | |
3452 | ||
3453 | mutex_lock(&c->tnc_mutex); | |
3454 | znode = lookup_znode(c, key, level, lnum, offs); | |
3455 | if (!znode) | |
3456 | goto out_unlock; | |
3457 | if (IS_ERR(znode)) { | |
3458 | err = PTR_ERR(znode); | |
3459 | goto out_unlock; | |
3460 | } | |
3461 | znode = dirty_cow_bottom_up(c, znode); | |
3462 | if (IS_ERR(znode)) { | |
3463 | err = PTR_ERR(znode); | |
3464 | goto out_unlock; | |
3465 | } | |
3466 | ||
3467 | out_unlock: | |
3468 | mutex_unlock(&c->tnc_mutex); | |
3469 | return err; | |
3470 | } | |
e3c3efc2 | 3471 | |
e3c3efc2 AB |
3472 | /** |
3473 | * dbg_check_inode_size - check if inode size is correct. | |
3474 | * @c: UBIFS file-system description object | |
3475 | * @inum: inode number | |
3476 | * @size: inode size | |
3477 | * | |
3478 | * This function makes sure that the inode size (@size) is correct and it does | |
3479 | * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL | |
3480 | * if it has a data page beyond @size, and other negative error code in case of | |
3481 | * other errors. | |
3482 | */ | |
3483 | int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode, | |
3484 | loff_t size) | |
3485 | { | |
3486 | int err, n; | |
3487 | union ubifs_key from_key, to_key, *key; | |
3488 | struct ubifs_znode *znode; | |
3489 | unsigned int block; | |
3490 | ||
3491 | if (!S_ISREG(inode->i_mode)) | |
3492 | return 0; | |
2b1844a8 | 3493 | if (!dbg_is_chk_gen(c)) |
e3c3efc2 AB |
3494 | return 0; |
3495 | ||
3496 | block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT; | |
3497 | data_key_init(c, &from_key, inode->i_ino, block); | |
3498 | highest_data_key(c, &to_key, inode->i_ino); | |
3499 | ||
3500 | mutex_lock(&c->tnc_mutex); | |
3501 | err = ubifs_lookup_level0(c, &from_key, &znode, &n); | |
3502 | if (err < 0) | |
3503 | goto out_unlock; | |
3504 | ||
3505 | if (err) { | |
e3c3efc2 AB |
3506 | key = &from_key; |
3507 | goto out_dump; | |
3508 | } | |
3509 | ||
3510 | err = tnc_next(c, &znode, &n); | |
3511 | if (err == -ENOENT) { | |
3512 | err = 0; | |
3513 | goto out_unlock; | |
3514 | } | |
3515 | if (err < 0) | |
3516 | goto out_unlock; | |
3517 | ||
6eb61d58 | 3518 | ubifs_assert(c, err == 0); |
e3c3efc2 AB |
3519 | key = &znode->zbranch[n].key; |
3520 | if (!key_in_range(c, key, &from_key, &to_key)) | |
3521 | goto out_unlock; | |
3522 | ||
3523 | out_dump: | |
3524 | block = key_block(c, key); | |
235c362b | 3525 | ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld", |
515315a1 AB |
3526 | (unsigned long)inode->i_ino, size, |
3527 | ((loff_t)block) << UBIFS_BLOCK_SHIFT); | |
4315fb40 | 3528 | mutex_unlock(&c->tnc_mutex); |
edf6be24 | 3529 | ubifs_dump_inode(c, inode); |
7c46d0ae | 3530 | dump_stack(); |
4315fb40 | 3531 | return -EINVAL; |
e3c3efc2 AB |
3532 | |
3533 | out_unlock: | |
3534 | mutex_unlock(&c->tnc_mutex); | |
3535 | return err; | |
3536 | } |