]>
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 | ||
37 | /* | |
38 | * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions. | |
39 | * @NAME_LESS: name corresponding to the first argument is less than second | |
40 | * @NAME_MATCHES: names match | |
41 | * @NAME_GREATER: name corresponding to the second argument is greater than | |
42 | * first | |
43 | * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media | |
44 | * | |
45 | * These constants were introduce to improve readability. | |
46 | */ | |
47 | enum { | |
48 | NAME_LESS = 0, | |
49 | NAME_MATCHES = 1, | |
50 | NAME_GREATER = 2, | |
51 | NOT_ON_MEDIA = 3, | |
52 | }; | |
53 | ||
54 | /** | |
55 | * insert_old_idx - record an index node obsoleted since the last commit start. | |
56 | * @c: UBIFS file-system description object | |
57 | * @lnum: LEB number of obsoleted index node | |
58 | * @offs: offset of obsoleted index node | |
59 | * | |
60 | * Returns %0 on success, and a negative error code on failure. | |
61 | * | |
62 | * For recovery, there must always be a complete intact version of the index on | |
63 | * flash at all times. That is called the "old index". It is the index as at the | |
64 | * time of the last successful commit. Many of the index nodes in the old index | |
65 | * may be dirty, but they must not be erased until the next successful commit | |
66 | * (at which point that index becomes the old index). | |
67 | * | |
68 | * That means that the garbage collection and the in-the-gaps method of | |
69 | * committing must be able to determine if an index node is in the old index. | |
70 | * Most of the old index nodes can be found by looking up the TNC using the | |
71 | * 'lookup_znode()' function. However, some of the old index nodes may have | |
72 | * been deleted from the current index or may have been changed so much that | |
73 | * they cannot be easily found. In those cases, an entry is added to an RB-tree. | |
74 | * That is what this function does. The RB-tree is ordered by LEB number and | |
75 | * offset because they uniquely identify the old index node. | |
76 | */ | |
77 | static int insert_old_idx(struct ubifs_info *c, int lnum, int offs) | |
78 | { | |
79 | struct ubifs_old_idx *old_idx, *o; | |
80 | struct rb_node **p, *parent = NULL; | |
81 | ||
82 | old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS); | |
83 | if (unlikely(!old_idx)) | |
84 | return -ENOMEM; | |
85 | old_idx->lnum = lnum; | |
86 | old_idx->offs = offs; | |
87 | ||
88 | p = &c->old_idx.rb_node; | |
89 | while (*p) { | |
90 | parent = *p; | |
91 | o = rb_entry(parent, struct ubifs_old_idx, rb); | |
92 | if (lnum < o->lnum) | |
93 | p = &(*p)->rb_left; | |
94 | else if (lnum > o->lnum) | |
95 | p = &(*p)->rb_right; | |
96 | else if (offs < o->offs) | |
97 | p = &(*p)->rb_left; | |
98 | else if (offs > o->offs) | |
99 | p = &(*p)->rb_right; | |
100 | else { | |
235c362b | 101 | ubifs_err(c, "old idx added twice!"); |
1e51764a AB |
102 | kfree(old_idx); |
103 | return 0; | |
104 | } | |
105 | } | |
106 | rb_link_node(&old_idx->rb, parent, p); | |
107 | rb_insert_color(&old_idx->rb, &c->old_idx); | |
108 | return 0; | |
109 | } | |
110 | ||
111 | /** | |
112 | * insert_old_idx_znode - record a znode obsoleted since last commit start. | |
113 | * @c: UBIFS file-system description object | |
114 | * @znode: znode of obsoleted index node | |
115 | * | |
116 | * Returns %0 on success, and a negative error code on failure. | |
117 | */ | |
118 | int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode) | |
119 | { | |
120 | if (znode->parent) { | |
121 | struct ubifs_zbranch *zbr; | |
122 | ||
123 | zbr = &znode->parent->zbranch[znode->iip]; | |
124 | if (zbr->len) | |
125 | return insert_old_idx(c, zbr->lnum, zbr->offs); | |
126 | } else | |
127 | if (c->zroot.len) | |
128 | return insert_old_idx(c, c->zroot.lnum, | |
129 | c->zroot.offs); | |
130 | return 0; | |
131 | } | |
132 | ||
133 | /** | |
134 | * ins_clr_old_idx_znode - record a znode obsoleted since last commit start. | |
135 | * @c: UBIFS file-system description object | |
136 | * @znode: znode of obsoleted index node | |
137 | * | |
138 | * Returns %0 on success, and a negative error code on failure. | |
139 | */ | |
140 | static int ins_clr_old_idx_znode(struct ubifs_info *c, | |
141 | struct ubifs_znode *znode) | |
142 | { | |
143 | int err; | |
144 | ||
145 | if (znode->parent) { | |
146 | struct ubifs_zbranch *zbr; | |
147 | ||
148 | zbr = &znode->parent->zbranch[znode->iip]; | |
149 | if (zbr->len) { | |
150 | err = insert_old_idx(c, zbr->lnum, zbr->offs); | |
151 | if (err) | |
152 | return err; | |
153 | zbr->lnum = 0; | |
154 | zbr->offs = 0; | |
155 | zbr->len = 0; | |
156 | } | |
157 | } else | |
158 | if (c->zroot.len) { | |
159 | err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs); | |
160 | if (err) | |
161 | return err; | |
162 | c->zroot.lnum = 0; | |
163 | c->zroot.offs = 0; | |
164 | c->zroot.len = 0; | |
165 | } | |
166 | return 0; | |
167 | } | |
168 | ||
169 | /** | |
170 | * destroy_old_idx - destroy the old_idx RB-tree. | |
171 | * @c: UBIFS file-system description object | |
172 | * | |
173 | * During start commit, the old_idx RB-tree is used to avoid overwriting index | |
174 | * nodes that were in the index last commit but have since been deleted. This | |
175 | * is necessary for recovery i.e. the old index must be kept intact until the | |
176 | * new index is successfully written. The old-idx RB-tree is used for the | |
177 | * in-the-gaps method of writing index nodes and is destroyed every commit. | |
178 | */ | |
179 | void destroy_old_idx(struct ubifs_info *c) | |
180 | { | |
bb25e49f CS |
181 | struct ubifs_old_idx *old_idx, *n; |
182 | ||
183 | rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb) | |
1e51764a | 184 | kfree(old_idx); |
bb25e49f | 185 | |
1e51764a AB |
186 | c->old_idx = RB_ROOT; |
187 | } | |
188 | ||
189 | /** | |
190 | * copy_znode - copy a dirty znode. | |
191 | * @c: UBIFS file-system description object | |
192 | * @znode: znode to copy | |
193 | * | |
194 | * A dirty znode being committed may not be changed, so it is copied. | |
195 | */ | |
196 | static struct ubifs_znode *copy_znode(struct ubifs_info *c, | |
197 | struct ubifs_znode *znode) | |
198 | { | |
199 | struct ubifs_znode *zn; | |
200 | ||
201 | zn = kmalloc(c->max_znode_sz, GFP_NOFS); | |
202 | if (unlikely(!zn)) | |
203 | return ERR_PTR(-ENOMEM); | |
204 | ||
205 | memcpy(zn, znode, c->max_znode_sz); | |
206 | zn->cnext = NULL; | |
207 | __set_bit(DIRTY_ZNODE, &zn->flags); | |
208 | __clear_bit(COW_ZNODE, &zn->flags); | |
209 | ||
f42eed7c | 210 | ubifs_assert(!ubifs_zn_obsolete(znode)); |
1e51764a AB |
211 | __set_bit(OBSOLETE_ZNODE, &znode->flags); |
212 | ||
213 | if (znode->level != 0) { | |
214 | int i; | |
215 | const int n = zn->child_cnt; | |
216 | ||
217 | /* The children now have new parent */ | |
218 | for (i = 0; i < n; i++) { | |
219 | struct ubifs_zbranch *zbr = &zn->zbranch[i]; | |
220 | ||
221 | if (zbr->znode) | |
222 | zbr->znode->parent = zn; | |
223 | } | |
224 | } | |
225 | ||
226 | atomic_long_inc(&c->dirty_zn_cnt); | |
227 | return zn; | |
228 | } | |
229 | ||
230 | /** | |
231 | * add_idx_dirt - add dirt due to a dirty znode. | |
232 | * @c: UBIFS file-system description object | |
233 | * @lnum: LEB number of index node | |
234 | * @dirt: size of index node | |
235 | * | |
236 | * This function updates lprops dirty space and the new size of the index. | |
237 | */ | |
238 | static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt) | |
239 | { | |
240 | c->calc_idx_sz -= ALIGN(dirt, 8); | |
241 | return ubifs_add_dirt(c, lnum, dirt); | |
242 | } | |
243 | ||
244 | /** | |
245 | * dirty_cow_znode - ensure a znode is not being committed. | |
246 | * @c: UBIFS file-system description object | |
247 | * @zbr: branch of znode to check | |
248 | * | |
249 | * Returns dirtied znode on success or negative error code on failure. | |
250 | */ | |
251 | static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c, | |
252 | struct ubifs_zbranch *zbr) | |
253 | { | |
254 | struct ubifs_znode *znode = zbr->znode; | |
255 | struct ubifs_znode *zn; | |
256 | int err; | |
257 | ||
f42eed7c | 258 | if (!ubifs_zn_cow(znode)) { |
1e51764a AB |
259 | /* znode is not being committed */ |
260 | if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) { | |
261 | atomic_long_inc(&c->dirty_zn_cnt); | |
262 | atomic_long_dec(&c->clean_zn_cnt); | |
263 | atomic_long_dec(&ubifs_clean_zn_cnt); | |
264 | err = add_idx_dirt(c, zbr->lnum, zbr->len); | |
265 | if (unlikely(err)) | |
266 | return ERR_PTR(err); | |
267 | } | |
268 | return znode; | |
269 | } | |
270 | ||
271 | zn = copy_znode(c, znode); | |
8d47aef4 | 272 | if (IS_ERR(zn)) |
1e51764a AB |
273 | return zn; |
274 | ||
275 | if (zbr->len) { | |
276 | err = insert_old_idx(c, zbr->lnum, zbr->offs); | |
277 | if (unlikely(err)) | |
278 | return ERR_PTR(err); | |
279 | err = add_idx_dirt(c, zbr->lnum, zbr->len); | |
280 | } else | |
281 | err = 0; | |
282 | ||
283 | zbr->znode = zn; | |
284 | zbr->lnum = 0; | |
285 | zbr->offs = 0; | |
286 | zbr->len = 0; | |
287 | ||
288 | if (unlikely(err)) | |
289 | return ERR_PTR(err); | |
290 | return zn; | |
291 | } | |
292 | ||
293 | /** | |
294 | * lnc_add - add a leaf node to the leaf node cache. | |
295 | * @c: UBIFS file-system description object | |
296 | * @zbr: zbranch of leaf node | |
297 | * @node: leaf node | |
298 | * | |
299 | * Leaf nodes are non-index nodes directory entry nodes or data nodes. The | |
300 | * purpose of the leaf node cache is to save re-reading the same leaf node over | |
301 | * and over again. Most things are cached by VFS, however the file system must | |
302 | * cache directory entries for readdir and for resolving hash collisions. The | |
303 | * present implementation of the leaf node cache is extremely simple, and | |
304 | * allows for error returns that are not used but that may be needed if a more | |
305 | * complex implementation is created. | |
306 | * | |
307 | * Note, this function does not add the @node object to LNC directly, but | |
308 | * allocates a copy of the object and adds the copy to LNC. The reason for this | |
309 | * is that @node has been allocated outside of the TNC subsystem and will be | |
310 | * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC | |
311 | * may be changed at any time, e.g. freed by the shrinker. | |
312 | */ | |
313 | static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
314 | const void *node) | |
315 | { | |
316 | int err; | |
317 | void *lnc_node; | |
318 | const struct ubifs_dent_node *dent = node; | |
319 | ||
320 | ubifs_assert(!zbr->leaf); | |
321 | ubifs_assert(zbr->len != 0); | |
322 | ubifs_assert(is_hash_key(c, &zbr->key)); | |
323 | ||
324 | err = ubifs_validate_entry(c, dent); | |
325 | if (err) { | |
7c46d0ae | 326 | dump_stack(); |
edf6be24 | 327 | ubifs_dump_node(c, dent); |
1e51764a AB |
328 | return err; |
329 | } | |
330 | ||
eaecf43a | 331 | lnc_node = kmemdup(node, zbr->len, GFP_NOFS); |
1e51764a AB |
332 | if (!lnc_node) |
333 | /* We don't have to have the cache, so no error */ | |
334 | return 0; | |
335 | ||
1e51764a AB |
336 | zbr->leaf = lnc_node; |
337 | return 0; | |
338 | } | |
339 | ||
340 | /** | |
341 | * lnc_add_directly - add a leaf node to the leaf-node-cache. | |
342 | * @c: UBIFS file-system description object | |
343 | * @zbr: zbranch of leaf node | |
344 | * @node: leaf node | |
345 | * | |
346 | * This function is similar to 'lnc_add()', but it does not create a copy of | |
347 | * @node but inserts @node to TNC directly. | |
348 | */ | |
349 | static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
350 | void *node) | |
351 | { | |
352 | int err; | |
353 | ||
354 | ubifs_assert(!zbr->leaf); | |
355 | ubifs_assert(zbr->len != 0); | |
356 | ||
357 | err = ubifs_validate_entry(c, node); | |
358 | if (err) { | |
7c46d0ae | 359 | dump_stack(); |
edf6be24 | 360 | ubifs_dump_node(c, node); |
1e51764a AB |
361 | return err; |
362 | } | |
363 | ||
364 | zbr->leaf = node; | |
365 | return 0; | |
366 | } | |
367 | ||
368 | /** | |
369 | * lnc_free - remove a leaf node from the leaf node cache. | |
370 | * @zbr: zbranch of leaf node | |
371 | * @node: leaf node | |
372 | */ | |
373 | static void lnc_free(struct ubifs_zbranch *zbr) | |
374 | { | |
375 | if (!zbr->leaf) | |
376 | return; | |
377 | kfree(zbr->leaf); | |
378 | zbr->leaf = NULL; | |
379 | } | |
380 | ||
381 | /** | |
382 | * tnc_read_node_nm - read a "hashed" leaf node. | |
383 | * @c: UBIFS file-system description object | |
384 | * @zbr: key and position of the node | |
385 | * @node: node is returned here | |
386 | * | |
387 | * This function reads a "hashed" node defined by @zbr from the leaf node cache | |
388 | * (in it is there) or from the hash media, in which case the node is also | |
389 | * added to LNC. Returns zero in case of success or a negative negative error | |
390 | * code in case of failure. | |
391 | */ | |
392 | static int tnc_read_node_nm(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
393 | void *node) | |
394 | { | |
395 | int err; | |
396 | ||
397 | ubifs_assert(is_hash_key(c, &zbr->key)); | |
398 | ||
399 | if (zbr->leaf) { | |
400 | /* Read from the leaf node cache */ | |
401 | ubifs_assert(zbr->len != 0); | |
402 | memcpy(node, zbr->leaf, zbr->len); | |
403 | return 0; | |
404 | } | |
405 | ||
406 | err = ubifs_tnc_read_node(c, zbr, node); | |
407 | if (err) | |
408 | return err; | |
409 | ||
410 | /* Add the node to the leaf node cache */ | |
411 | err = lnc_add(c, zbr, node); | |
412 | return err; | |
413 | } | |
414 | ||
415 | /** | |
416 | * try_read_node - read a node if it is a node. | |
417 | * @c: UBIFS file-system description object | |
418 | * @buf: buffer to read to | |
419 | * @type: node type | |
420 | * @len: node length (not aligned) | |
421 | * @lnum: LEB number of node to read | |
422 | * @offs: offset of node to read | |
423 | * | |
424 | * This function tries to read a node of known type and length, checks it and | |
425 | * stores it in @buf. This function returns %1 if a node is present and %0 if | |
426 | * a node is not present. A negative error code is returned for I/O errors. | |
427 | * This function performs that same function as ubifs_read_node except that | |
428 | * it does not require that there is actually a node present and instead | |
429 | * the return code indicates if a node was read. | |
6f7ab6d4 AB |
430 | * |
431 | * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc | |
432 | * is true (it is controlled by corresponding mount option). However, if | |
18d1d7fb AB |
433 | * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to |
434 | * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is | |
435 | * because during mounting or re-mounting from R/O mode to R/W mode we may read | |
436 | * journal nodes (when replying the journal or doing the recovery) and the | |
437 | * journal nodes may potentially be corrupted, so checking is required. | |
1e51764a AB |
438 | */ |
439 | static int try_read_node(const struct ubifs_info *c, void *buf, int type, | |
440 | int len, int lnum, int offs) | |
441 | { | |
442 | int err, node_len; | |
443 | struct ubifs_ch *ch = buf; | |
444 | uint32_t crc, node_crc; | |
445 | ||
446 | dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); | |
447 | ||
d304820a | 448 | err = ubifs_leb_read(c, lnum, buf, offs, len, 1); |
1e51764a | 449 | if (err) { |
235c362b | 450 | ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d", |
1e51764a AB |
451 | type, lnum, offs, err); |
452 | return err; | |
453 | } | |
454 | ||
455 | if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) | |
456 | return 0; | |
457 | ||
458 | if (ch->node_type != type) | |
459 | return 0; | |
460 | ||
461 | node_len = le32_to_cpu(ch->len); | |
462 | if (node_len != len) | |
463 | return 0; | |
464 | ||
18d1d7fb AB |
465 | if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting && |
466 | !c->remounting_rw) | |
6f7ab6d4 | 467 | return 1; |
2953e73f | 468 | |
1e51764a AB |
469 | crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8); |
470 | node_crc = le32_to_cpu(ch->crc); | |
471 | if (crc != node_crc) | |
472 | return 0; | |
473 | ||
474 | return 1; | |
475 | } | |
476 | ||
477 | /** | |
478 | * fallible_read_node - try to read a leaf node. | |
479 | * @c: UBIFS file-system description object | |
480 | * @key: key of node to read | |
481 | * @zbr: position of node | |
482 | * @node: node returned | |
483 | * | |
484 | * This function tries to read a node and returns %1 if the node is read, %0 | |
485 | * if the node is not present, and a negative error code in the case of error. | |
486 | */ | |
487 | static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key, | |
488 | struct ubifs_zbranch *zbr, void *node) | |
489 | { | |
490 | int ret; | |
491 | ||
515315a1 | 492 | dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs); |
1e51764a AB |
493 | |
494 | ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum, | |
495 | zbr->offs); | |
496 | if (ret == 1) { | |
497 | union ubifs_key node_key; | |
498 | struct ubifs_dent_node *dent = node; | |
499 | ||
500 | /* All nodes have key in the same place */ | |
501 | key_read(c, &dent->key, &node_key); | |
502 | if (keys_cmp(c, key, &node_key) != 0) | |
503 | ret = 0; | |
504 | } | |
601c0bc4 | 505 | if (ret == 0 && c->replaying) |
515315a1 AB |
506 | dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ", |
507 | zbr->lnum, zbr->offs, zbr->len); | |
1e51764a AB |
508 | return ret; |
509 | } | |
510 | ||
511 | /** | |
512 | * matches_name - determine if a direntry or xattr entry matches a given name. | |
513 | * @c: UBIFS file-system description object | |
514 | * @zbr: zbranch of dent | |
515 | * @nm: name to match | |
516 | * | |
517 | * This function checks if xentry/direntry referred by zbranch @zbr matches name | |
518 | * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by | |
519 | * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case | |
520 | * of failure, a negative error code is returned. | |
521 | */ | |
522 | static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr, | |
523 | const struct qstr *nm) | |
524 | { | |
525 | struct ubifs_dent_node *dent; | |
526 | int nlen, err; | |
527 | ||
528 | /* If possible, match against the dent in the leaf node cache */ | |
529 | if (!zbr->leaf) { | |
530 | dent = kmalloc(zbr->len, GFP_NOFS); | |
531 | if (!dent) | |
532 | return -ENOMEM; | |
533 | ||
534 | err = ubifs_tnc_read_node(c, zbr, dent); | |
535 | if (err) | |
536 | goto out_free; | |
537 | ||
538 | /* Add the node to the leaf node cache */ | |
539 | err = lnc_add_directly(c, zbr, dent); | |
540 | if (err) | |
541 | goto out_free; | |
542 | } else | |
543 | dent = zbr->leaf; | |
544 | ||
545 | nlen = le16_to_cpu(dent->nlen); | |
546 | err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len)); | |
547 | if (err == 0) { | |
548 | if (nlen == nm->len) | |
549 | return NAME_MATCHES; | |
550 | else if (nlen < nm->len) | |
551 | return NAME_LESS; | |
552 | else | |
553 | return NAME_GREATER; | |
554 | } else if (err < 0) | |
555 | return NAME_LESS; | |
556 | else | |
557 | return NAME_GREATER; | |
558 | ||
559 | out_free: | |
560 | kfree(dent); | |
561 | return err; | |
562 | } | |
563 | ||
564 | /** | |
565 | * get_znode - get a TNC znode that may not be loaded yet. | |
566 | * @c: UBIFS file-system description object | |
567 | * @znode: parent znode | |
568 | * @n: znode branch slot number | |
569 | * | |
570 | * This function returns the znode or a negative error code. | |
571 | */ | |
572 | static struct ubifs_znode *get_znode(struct ubifs_info *c, | |
573 | struct ubifs_znode *znode, int n) | |
574 | { | |
575 | struct ubifs_zbranch *zbr; | |
576 | ||
577 | zbr = &znode->zbranch[n]; | |
578 | if (zbr->znode) | |
579 | znode = zbr->znode; | |
580 | else | |
581 | znode = ubifs_load_znode(c, zbr, znode, n); | |
582 | return znode; | |
583 | } | |
584 | ||
585 | /** | |
586 | * tnc_next - find next TNC entry. | |
587 | * @c: UBIFS file-system description object | |
588 | * @zn: znode is passed and returned here | |
589 | * @n: znode branch slot number is passed and returned here | |
590 | * | |
591 | * This function returns %0 if the next TNC entry is found, %-ENOENT if there is | |
592 | * no next entry, or a negative error code otherwise. | |
593 | */ | |
594 | static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n) | |
595 | { | |
596 | struct ubifs_znode *znode = *zn; | |
597 | int nn = *n; | |
598 | ||
599 | nn += 1; | |
600 | if (nn < znode->child_cnt) { | |
601 | *n = nn; | |
602 | return 0; | |
603 | } | |
604 | while (1) { | |
605 | struct ubifs_znode *zp; | |
606 | ||
607 | zp = znode->parent; | |
608 | if (!zp) | |
609 | return -ENOENT; | |
610 | nn = znode->iip + 1; | |
611 | znode = zp; | |
612 | if (nn < znode->child_cnt) { | |
613 | znode = get_znode(c, znode, nn); | |
614 | if (IS_ERR(znode)) | |
615 | return PTR_ERR(znode); | |
616 | while (znode->level != 0) { | |
617 | znode = get_znode(c, znode, 0); | |
618 | if (IS_ERR(znode)) | |
619 | return PTR_ERR(znode); | |
620 | } | |
621 | nn = 0; | |
622 | break; | |
623 | } | |
624 | } | |
625 | *zn = znode; | |
626 | *n = nn; | |
627 | return 0; | |
628 | } | |
629 | ||
630 | /** | |
631 | * tnc_prev - find previous TNC entry. | |
632 | * @c: UBIFS file-system description object | |
633 | * @zn: znode is returned here | |
634 | * @n: znode branch slot number is passed and returned here | |
635 | * | |
636 | * This function returns %0 if the previous TNC entry is found, %-ENOENT if | |
637 | * there is no next entry, or a negative error code otherwise. | |
638 | */ | |
639 | static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n) | |
640 | { | |
641 | struct ubifs_znode *znode = *zn; | |
642 | int nn = *n; | |
643 | ||
644 | if (nn > 0) { | |
645 | *n = nn - 1; | |
646 | return 0; | |
647 | } | |
648 | while (1) { | |
649 | struct ubifs_znode *zp; | |
650 | ||
651 | zp = znode->parent; | |
652 | if (!zp) | |
653 | return -ENOENT; | |
654 | nn = znode->iip - 1; | |
655 | znode = zp; | |
656 | if (nn >= 0) { | |
657 | znode = get_znode(c, znode, nn); | |
658 | if (IS_ERR(znode)) | |
659 | return PTR_ERR(znode); | |
660 | while (znode->level != 0) { | |
661 | nn = znode->child_cnt - 1; | |
662 | znode = get_znode(c, znode, nn); | |
663 | if (IS_ERR(znode)) | |
664 | return PTR_ERR(znode); | |
665 | } | |
666 | nn = znode->child_cnt - 1; | |
667 | break; | |
668 | } | |
669 | } | |
670 | *zn = znode; | |
671 | *n = nn; | |
672 | return 0; | |
673 | } | |
674 | ||
675 | /** | |
676 | * resolve_collision - resolve a collision. | |
677 | * @c: UBIFS file-system description object | |
678 | * @key: key of a directory or extended attribute entry | |
679 | * @zn: znode is returned here | |
680 | * @n: zbranch number is passed and returned here | |
681 | * @nm: name of the entry | |
682 | * | |
683 | * This function is called for "hashed" keys to make sure that the found key | |
684 | * really corresponds to the looked up node (directory or extended attribute | |
685 | * entry). It returns %1 and sets @zn and @n if the collision is resolved. | |
686 | * %0 is returned if @nm is not found and @zn and @n are set to the previous | |
687 | * entry, i.e. to the entry after which @nm could follow if it were in TNC. | |
688 | * This means that @n may be set to %-1 if the leftmost key in @zn is the | |
689 | * previous one. A negative error code is returned on failures. | |
690 | */ | |
691 | static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key, | |
692 | struct ubifs_znode **zn, int *n, | |
693 | const struct qstr *nm) | |
694 | { | |
695 | int err; | |
696 | ||
697 | err = matches_name(c, &(*zn)->zbranch[*n], nm); | |
698 | if (unlikely(err < 0)) | |
699 | return err; | |
700 | if (err == NAME_MATCHES) | |
701 | return 1; | |
702 | ||
703 | if (err == NAME_GREATER) { | |
704 | /* Look left */ | |
705 | while (1) { | |
706 | err = tnc_prev(c, zn, n); | |
707 | if (err == -ENOENT) { | |
708 | ubifs_assert(*n == 0); | |
709 | *n = -1; | |
710 | return 0; | |
711 | } | |
712 | if (err < 0) | |
713 | return err; | |
714 | if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { | |
715 | /* | |
716 | * We have found the branch after which we would | |
717 | * like to insert, but inserting in this znode | |
718 | * may still be wrong. Consider the following 3 | |
719 | * znodes, in the case where we are resolving a | |
720 | * collision with Key2. | |
721 | * | |
722 | * znode zp | |
723 | * ---------------------- | |
724 | * level 1 | Key0 | Key1 | | |
725 | * ----------------------- | |
726 | * | | | |
727 | * znode za | | znode zb | |
728 | * ------------ ------------ | |
729 | * level 0 | Key0 | | Key2 | | |
730 | * ------------ ------------ | |
731 | * | |
732 | * The lookup finds Key2 in znode zb. Lets say | |
733 | * there is no match and the name is greater so | |
734 | * we look left. When we find Key0, we end up | |
735 | * here. If we return now, we will insert into | |
736 | * znode za at slot n = 1. But that is invalid | |
737 | * according to the parent's keys. Key2 must | |
738 | * be inserted into znode zb. | |
739 | * | |
740 | * Note, this problem is not relevant for the | |
741 | * case when we go right, because | |
742 | * 'tnc_insert()' would correct the parent key. | |
743 | */ | |
744 | if (*n == (*zn)->child_cnt - 1) { | |
745 | err = tnc_next(c, zn, n); | |
746 | if (err) { | |
747 | /* Should be impossible */ | |
748 | ubifs_assert(0); | |
749 | if (err == -ENOENT) | |
750 | err = -EINVAL; | |
751 | return err; | |
752 | } | |
753 | ubifs_assert(*n == 0); | |
754 | *n = -1; | |
755 | } | |
756 | return 0; | |
757 | } | |
758 | err = matches_name(c, &(*zn)->zbranch[*n], nm); | |
759 | if (err < 0) | |
760 | return err; | |
761 | if (err == NAME_LESS) | |
762 | return 0; | |
763 | if (err == NAME_MATCHES) | |
764 | return 1; | |
765 | ubifs_assert(err == NAME_GREATER); | |
766 | } | |
767 | } else { | |
768 | int nn = *n; | |
769 | struct ubifs_znode *znode = *zn; | |
770 | ||
771 | /* Look right */ | |
772 | while (1) { | |
773 | err = tnc_next(c, &znode, &nn); | |
774 | if (err == -ENOENT) | |
775 | return 0; | |
776 | if (err < 0) | |
777 | return err; | |
778 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
779 | return 0; | |
780 | err = matches_name(c, &znode->zbranch[nn], nm); | |
781 | if (err < 0) | |
782 | return err; | |
783 | if (err == NAME_GREATER) | |
784 | return 0; | |
785 | *zn = znode; | |
786 | *n = nn; | |
787 | if (err == NAME_MATCHES) | |
788 | return 1; | |
789 | ubifs_assert(err == NAME_LESS); | |
790 | } | |
791 | } | |
792 | } | |
793 | ||
794 | /** | |
795 | * fallible_matches_name - determine if a dent matches a given name. | |
796 | * @c: UBIFS file-system description object | |
797 | * @zbr: zbranch of dent | |
798 | * @nm: name to match | |
799 | * | |
800 | * This is a "fallible" version of 'matches_name()' function which does not | |
801 | * panic if the direntry/xentry referred by @zbr does not exist on the media. | |
802 | * | |
803 | * This function checks if xentry/direntry referred by zbranch @zbr matches name | |
804 | * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr | |
805 | * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA | |
806 | * if xentry/direntry referred by @zbr does not exist on the media. A negative | |
807 | * error code is returned in case of failure. | |
808 | */ | |
809 | static int fallible_matches_name(struct ubifs_info *c, | |
810 | struct ubifs_zbranch *zbr, | |
811 | const struct qstr *nm) | |
812 | { | |
813 | struct ubifs_dent_node *dent; | |
814 | int nlen, err; | |
815 | ||
816 | /* If possible, match against the dent in the leaf node cache */ | |
817 | if (!zbr->leaf) { | |
818 | dent = kmalloc(zbr->len, GFP_NOFS); | |
819 | if (!dent) | |
820 | return -ENOMEM; | |
821 | ||
822 | err = fallible_read_node(c, &zbr->key, zbr, dent); | |
823 | if (err < 0) | |
824 | goto out_free; | |
825 | if (err == 0) { | |
826 | /* The node was not present */ | |
827 | err = NOT_ON_MEDIA; | |
828 | goto out_free; | |
829 | } | |
830 | ubifs_assert(err == 1); | |
831 | ||
832 | err = lnc_add_directly(c, zbr, dent); | |
833 | if (err) | |
834 | goto out_free; | |
835 | } else | |
836 | dent = zbr->leaf; | |
837 | ||
838 | nlen = le16_to_cpu(dent->nlen); | |
839 | err = memcmp(dent->name, nm->name, min_t(int, nlen, nm->len)); | |
840 | if (err == 0) { | |
841 | if (nlen == nm->len) | |
842 | return NAME_MATCHES; | |
843 | else if (nlen < nm->len) | |
844 | return NAME_LESS; | |
845 | else | |
846 | return NAME_GREATER; | |
847 | } else if (err < 0) | |
848 | return NAME_LESS; | |
849 | else | |
850 | return NAME_GREATER; | |
851 | ||
852 | out_free: | |
853 | kfree(dent); | |
854 | return err; | |
855 | } | |
856 | ||
857 | /** | |
858 | * fallible_resolve_collision - resolve a collision even if nodes are missing. | |
859 | * @c: UBIFS file-system description object | |
860 | * @key: key | |
861 | * @zn: znode is returned here | |
862 | * @n: branch number is passed and returned here | |
863 | * @nm: name of directory entry | |
864 | * @adding: indicates caller is adding a key to the TNC | |
865 | * | |
866 | * This is a "fallible" version of the 'resolve_collision()' function which | |
867 | * does not panic if one of the nodes referred to by TNC does not exist on the | |
868 | * media. This may happen when replaying the journal if a deleted node was | |
869 | * Garbage-collected and the commit was not done. A branch that refers to a node | |
870 | * that is not present is called a dangling branch. The following are the return | |
871 | * codes for this function: | |
872 | * o if @nm was found, %1 is returned and @zn and @n are set to the found | |
873 | * branch; | |
874 | * o if we are @adding and @nm was not found, %0 is returned; | |
875 | * o if we are not @adding and @nm was not found, but a dangling branch was | |
876 | * found, then %1 is returned and @zn and @n are set to the dangling branch; | |
877 | * o a negative error code is returned in case of failure. | |
878 | */ | |
879 | static int fallible_resolve_collision(struct ubifs_info *c, | |
880 | const union ubifs_key *key, | |
881 | struct ubifs_znode **zn, int *n, | |
882 | const struct qstr *nm, int adding) | |
883 | { | |
884 | struct ubifs_znode *o_znode = NULL, *znode = *zn; | |
885 | int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n; | |
886 | ||
887 | cmp = fallible_matches_name(c, &znode->zbranch[nn], nm); | |
888 | if (unlikely(cmp < 0)) | |
889 | return cmp; | |
890 | if (cmp == NAME_MATCHES) | |
891 | return 1; | |
892 | if (cmp == NOT_ON_MEDIA) { | |
893 | o_znode = znode; | |
894 | o_n = nn; | |
895 | /* | |
896 | * We are unlucky and hit a dangling branch straight away. | |
897 | * Now we do not really know where to go to find the needed | |
898 | * branch - to the left or to the right. Well, let's try left. | |
899 | */ | |
900 | unsure = 1; | |
901 | } else if (!adding) | |
902 | unsure = 1; /* Remove a dangling branch wherever it is */ | |
903 | ||
904 | if (cmp == NAME_GREATER || unsure) { | |
905 | /* Look left */ | |
906 | while (1) { | |
907 | err = tnc_prev(c, zn, n); | |
908 | if (err == -ENOENT) { | |
909 | ubifs_assert(*n == 0); | |
910 | *n = -1; | |
911 | break; | |
912 | } | |
913 | if (err < 0) | |
914 | return err; | |
915 | if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { | |
916 | /* See comments in 'resolve_collision()' */ | |
917 | if (*n == (*zn)->child_cnt - 1) { | |
918 | err = tnc_next(c, zn, n); | |
919 | if (err) { | |
920 | /* Should be impossible */ | |
921 | ubifs_assert(0); | |
922 | if (err == -ENOENT) | |
923 | err = -EINVAL; | |
924 | return err; | |
925 | } | |
926 | ubifs_assert(*n == 0); | |
927 | *n = -1; | |
928 | } | |
929 | break; | |
930 | } | |
931 | err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm); | |
932 | if (err < 0) | |
933 | return err; | |
934 | if (err == NAME_MATCHES) | |
935 | return 1; | |
936 | if (err == NOT_ON_MEDIA) { | |
937 | o_znode = *zn; | |
938 | o_n = *n; | |
939 | continue; | |
940 | } | |
941 | if (!adding) | |
942 | continue; | |
943 | if (err == NAME_LESS) | |
944 | break; | |
945 | else | |
946 | unsure = 0; | |
947 | } | |
948 | } | |
949 | ||
950 | if (cmp == NAME_LESS || unsure) { | |
951 | /* Look right */ | |
952 | *zn = znode; | |
953 | *n = nn; | |
954 | while (1) { | |
955 | err = tnc_next(c, &znode, &nn); | |
956 | if (err == -ENOENT) | |
957 | break; | |
958 | if (err < 0) | |
959 | return err; | |
960 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
961 | break; | |
962 | err = fallible_matches_name(c, &znode->zbranch[nn], nm); | |
963 | if (err < 0) | |
964 | return err; | |
965 | if (err == NAME_GREATER) | |
966 | break; | |
967 | *zn = znode; | |
968 | *n = nn; | |
969 | if (err == NAME_MATCHES) | |
970 | return 1; | |
971 | if (err == NOT_ON_MEDIA) { | |
972 | o_znode = znode; | |
973 | o_n = nn; | |
974 | } | |
975 | } | |
976 | } | |
977 | ||
978 | /* Never match a dangling branch when adding */ | |
979 | if (adding || !o_znode) | |
980 | return 0; | |
981 | ||
515315a1 | 982 | dbg_mntk(key, "dangling match LEB %d:%d len %d key ", |
1e51764a | 983 | o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs, |
515315a1 | 984 | o_znode->zbranch[o_n].len); |
1e51764a AB |
985 | *zn = o_znode; |
986 | *n = o_n; | |
987 | return 1; | |
988 | } | |
989 | ||
990 | /** | |
991 | * matches_position - determine if a zbranch matches a given position. | |
992 | * @zbr: zbranch of dent | |
993 | * @lnum: LEB number of dent to match | |
994 | * @offs: offset of dent to match | |
995 | * | |
996 | * This function returns %1 if @lnum:@offs matches, and %0 otherwise. | |
997 | */ | |
998 | static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs) | |
999 | { | |
1000 | if (zbr->lnum == lnum && zbr->offs == offs) | |
1001 | return 1; | |
1002 | else | |
1003 | return 0; | |
1004 | } | |
1005 | ||
1006 | /** | |
1007 | * resolve_collision_directly - resolve a collision directly. | |
1008 | * @c: UBIFS file-system description object | |
1009 | * @key: key of directory entry | |
1010 | * @zn: znode is passed and returned here | |
1011 | * @n: zbranch number is passed and returned here | |
1012 | * @lnum: LEB number of dent node to match | |
1013 | * @offs: offset of dent node to match | |
1014 | * | |
1015 | * This function is used for "hashed" keys to make sure the found directory or | |
1016 | * extended attribute entry node is what was looked for. It is used when the | |
1017 | * flash address of the right node is known (@lnum:@offs) which makes it much | |
1018 | * easier to resolve collisions (no need to read entries and match full | |
1019 | * names). This function returns %1 and sets @zn and @n if the collision is | |
1020 | * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the | |
1021 | * previous directory entry. Otherwise a negative error code is returned. | |
1022 | */ | |
1023 | static int resolve_collision_directly(struct ubifs_info *c, | |
1024 | const union ubifs_key *key, | |
1025 | struct ubifs_znode **zn, int *n, | |
1026 | int lnum, int offs) | |
1027 | { | |
1028 | struct ubifs_znode *znode; | |
1029 | int nn, err; | |
1030 | ||
1031 | znode = *zn; | |
1032 | nn = *n; | |
1033 | if (matches_position(&znode->zbranch[nn], lnum, offs)) | |
1034 | return 1; | |
1035 | ||
1036 | /* Look left */ | |
1037 | while (1) { | |
1038 | err = tnc_prev(c, &znode, &nn); | |
1039 | if (err == -ENOENT) | |
1040 | break; | |
1041 | if (err < 0) | |
1042 | return err; | |
1043 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
1044 | break; | |
1045 | if (matches_position(&znode->zbranch[nn], lnum, offs)) { | |
1046 | *zn = znode; | |
1047 | *n = nn; | |
1048 | return 1; | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | /* Look right */ | |
1053 | znode = *zn; | |
1054 | nn = *n; | |
1055 | while (1) { | |
1056 | err = tnc_next(c, &znode, &nn); | |
1057 | if (err == -ENOENT) | |
1058 | return 0; | |
1059 | if (err < 0) | |
1060 | return err; | |
1061 | if (keys_cmp(c, &znode->zbranch[nn].key, key)) | |
1062 | return 0; | |
1063 | *zn = znode; | |
1064 | *n = nn; | |
1065 | if (matches_position(&znode->zbranch[nn], lnum, offs)) | |
1066 | return 1; | |
1067 | } | |
1068 | } | |
1069 | ||
1070 | /** | |
1071 | * dirty_cow_bottom_up - dirty a znode and its ancestors. | |
1072 | * @c: UBIFS file-system description object | |
1073 | * @znode: znode to dirty | |
1074 | * | |
1075 | * If we do not have a unique key that resides in a znode, then we cannot | |
1076 | * dirty that znode from the top down (i.e. by using lookup_level0_dirty) | |
1077 | * This function records the path back to the last dirty ancestor, and then | |
1078 | * dirties the znodes on that path. | |
1079 | */ | |
1080 | static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c, | |
1081 | struct ubifs_znode *znode) | |
1082 | { | |
1083 | struct ubifs_znode *zp; | |
1084 | int *path = c->bottom_up_buf, p = 0; | |
1085 | ||
1086 | ubifs_assert(c->zroot.znode); | |
1087 | ubifs_assert(znode); | |
1088 | if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) { | |
1089 | kfree(c->bottom_up_buf); | |
1090 | c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int), | |
1091 | GFP_NOFS); | |
1092 | if (!c->bottom_up_buf) | |
1093 | return ERR_PTR(-ENOMEM); | |
1094 | path = c->bottom_up_buf; | |
1095 | } | |
1096 | if (c->zroot.znode->level) { | |
1097 | /* Go up until parent is dirty */ | |
1098 | while (1) { | |
1099 | int n; | |
1100 | ||
1101 | zp = znode->parent; | |
1102 | if (!zp) | |
1103 | break; | |
1104 | n = znode->iip; | |
1105 | ubifs_assert(p < c->zroot.znode->level); | |
1106 | path[p++] = n; | |
1107 | if (!zp->cnext && ubifs_zn_dirty(znode)) | |
1108 | break; | |
1109 | znode = zp; | |
1110 | } | |
1111 | } | |
1112 | ||
1113 | /* Come back down, dirtying as we go */ | |
1114 | while (1) { | |
1115 | struct ubifs_zbranch *zbr; | |
1116 | ||
1117 | zp = znode->parent; | |
1118 | if (zp) { | |
1119 | ubifs_assert(path[p - 1] >= 0); | |
1120 | ubifs_assert(path[p - 1] < zp->child_cnt); | |
1121 | zbr = &zp->zbranch[path[--p]]; | |
1122 | znode = dirty_cow_znode(c, zbr); | |
1123 | } else { | |
1124 | ubifs_assert(znode == c->zroot.znode); | |
1125 | znode = dirty_cow_znode(c, &c->zroot); | |
1126 | } | |
8d47aef4 | 1127 | if (IS_ERR(znode) || !p) |
1e51764a AB |
1128 | break; |
1129 | ubifs_assert(path[p - 1] >= 0); | |
1130 | ubifs_assert(path[p - 1] < znode->child_cnt); | |
1131 | znode = znode->zbranch[path[p - 1]].znode; | |
1132 | } | |
1133 | ||
1134 | return znode; | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * ubifs_lookup_level0 - search for zero-level znode. | |
1139 | * @c: UBIFS file-system description object | |
1140 | * @key: key to lookup | |
1141 | * @zn: znode is returned here | |
1142 | * @n: znode branch slot number is returned here | |
1143 | * | |
1144 | * This function looks up the TNC tree and search for zero-level znode which | |
1145 | * refers key @key. The found zero-level znode is returned in @zn. There are 3 | |
1146 | * cases: | |
1147 | * o exact match, i.e. the found zero-level znode contains key @key, then %1 | |
1148 | * is returned and slot number of the matched branch is stored in @n; | |
1149 | * o not exact match, which means that zero-level znode does not contain | |
e3c3efc2 AB |
1150 | * @key, then %0 is returned and slot number of the closest branch is stored |
1151 | * in @n; | |
1e51764a AB |
1152 | * o @key is so small that it is even less than the lowest key of the |
1153 | * leftmost zero-level node, then %0 is returned and %0 is stored in @n. | |
1154 | * | |
1155 | * Note, when the TNC tree is traversed, some znodes may be absent, then this | |
1156 | * function reads corresponding indexing nodes and inserts them to TNC. In | |
1157 | * case of failure, a negative error code is returned. | |
1158 | */ | |
1159 | int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key, | |
1160 | struct ubifs_znode **zn, int *n) | |
1161 | { | |
1162 | int err, exact; | |
1163 | struct ubifs_znode *znode; | |
1164 | unsigned long time = get_seconds(); | |
1165 | ||
515315a1 | 1166 | dbg_tnck(key, "search key "); |
ba2f48f7 | 1167 | ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY); |
1e51764a AB |
1168 | |
1169 | znode = c->zroot.znode; | |
1170 | if (unlikely(!znode)) { | |
1171 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
1172 | if (IS_ERR(znode)) | |
1173 | return PTR_ERR(znode); | |
1174 | } | |
1175 | ||
1176 | znode->time = time; | |
1177 | ||
1178 | while (1) { | |
1179 | struct ubifs_zbranch *zbr; | |
1180 | ||
1181 | exact = ubifs_search_zbranch(c, znode, key, n); | |
1182 | ||
1183 | if (znode->level == 0) | |
1184 | break; | |
1185 | ||
1186 | if (*n < 0) | |
1187 | *n = 0; | |
1188 | zbr = &znode->zbranch[*n]; | |
1189 | ||
1190 | if (zbr->znode) { | |
1191 | znode->time = time; | |
1192 | znode = zbr->znode; | |
1193 | continue; | |
1194 | } | |
1195 | ||
1196 | /* znode is not in TNC cache, load it from the media */ | |
1197 | znode = ubifs_load_znode(c, zbr, znode, *n); | |
1198 | if (IS_ERR(znode)) | |
1199 | return PTR_ERR(znode); | |
1200 | } | |
1201 | ||
1202 | *zn = znode; | |
1203 | if (exact || !is_hash_key(c, key) || *n != -1) { | |
1204 | dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); | |
1205 | return exact; | |
1206 | } | |
1207 | ||
1208 | /* | |
1209 | * Here is a tricky place. We have not found the key and this is a | |
1210 | * "hashed" key, which may collide. The rest of the code deals with | |
1211 | * situations like this: | |
1212 | * | |
1213 | * | 3 | 5 | | |
1214 | * / \ | |
1215 | * | 3 | 5 | | 6 | 7 | (x) | |
1216 | * | |
1217 | * Or more a complex example: | |
1218 | * | |
1219 | * | 1 | 5 | | |
1220 | * / \ | |
1221 | * | 1 | 3 | | 5 | 8 | | |
1222 | * \ / | |
1223 | * | 5 | 5 | | 6 | 7 | (x) | |
1224 | * | |
1225 | * In the examples, if we are looking for key "5", we may reach nodes | |
1226 | * marked with "(x)". In this case what we have do is to look at the | |
1227 | * left and see if there is "5" key there. If there is, we have to | |
1228 | * return it. | |
1229 | * | |
1230 | * Note, this whole situation is possible because we allow to have | |
1231 | * elements which are equivalent to the next key in the parent in the | |
1232 | * children of current znode. For example, this happens if we split a | |
1233 | * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something | |
1234 | * like this: | |
1235 | * | 3 | 5 | | |
1236 | * / \ | |
1237 | * | 3 | 5 | | 5 | 6 | 7 | | |
1238 | * ^ | |
1239 | * And this becomes what is at the first "picture" after key "5" marked | |
1240 | * with "^" is removed. What could be done is we could prohibit | |
1241 | * splitting in the middle of the colliding sequence. Also, when | |
1242 | * removing the leftmost key, we would have to correct the key of the | |
1243 | * parent node, which would introduce additional complications. Namely, | |
7d4e9ccb | 1244 | * if we changed the leftmost key of the parent znode, the garbage |
1e51764a AB |
1245 | * collector would be unable to find it (GC is doing this when GC'ing |
1246 | * indexing LEBs). Although we already have an additional RB-tree where | |
1247 | * we save such changed znodes (see 'ins_clr_old_idx_znode()') until | |
1248 | * after the commit. But anyway, this does not look easy to implement | |
1249 | * so we did not try this. | |
1250 | */ | |
1251 | err = tnc_prev(c, &znode, n); | |
1252 | if (err == -ENOENT) { | |
1253 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1254 | *n = -1; | |
1255 | return 0; | |
1256 | } | |
1257 | if (unlikely(err < 0)) | |
1258 | return err; | |
1259 | if (keys_cmp(c, key, &znode->zbranch[*n].key)) { | |
1260 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1261 | *n = -1; | |
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); | |
1266 | *zn = znode; | |
1267 | return 1; | |
1268 | } | |
1269 | ||
1270 | /** | |
1271 | * lookup_level0_dirty - search for zero-level znode dirtying. | |
1272 | * @c: UBIFS file-system description object | |
1273 | * @key: key to lookup | |
1274 | * @zn: znode is returned here | |
1275 | * @n: znode branch slot number is returned here | |
1276 | * | |
1277 | * This function looks up the TNC tree and search for zero-level znode which | |
1278 | * refers key @key. The found zero-level znode is returned in @zn. There are 3 | |
1279 | * cases: | |
1280 | * o exact match, i.e. the found zero-level znode contains key @key, then %1 | |
1281 | * is returned and slot number of the matched branch is stored in @n; | |
1282 | * o not exact match, which means that zero-level znode does not contain @key | |
1283 | * then %0 is returned and slot number of the closed branch is stored in | |
1284 | * @n; | |
1285 | * o @key is so small that it is even less than the lowest key of the | |
1286 | * leftmost zero-level node, then %0 is returned and %-1 is stored in @n. | |
1287 | * | |
1288 | * Additionally all znodes in the path from the root to the located zero-level | |
1289 | * znode are marked as dirty. | |
1290 | * | |
1291 | * Note, when the TNC tree is traversed, some znodes may be absent, then this | |
1292 | * function reads corresponding indexing nodes and inserts them to TNC. In | |
1293 | * case of failure, a negative error code is returned. | |
1294 | */ | |
1295 | static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key, | |
1296 | struct ubifs_znode **zn, int *n) | |
1297 | { | |
1298 | int err, exact; | |
1299 | struct ubifs_znode *znode; | |
1300 | unsigned long time = get_seconds(); | |
1301 | ||
515315a1 | 1302 | dbg_tnck(key, "search and dirty key "); |
1e51764a AB |
1303 | |
1304 | znode = c->zroot.znode; | |
1305 | if (unlikely(!znode)) { | |
1306 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
1307 | if (IS_ERR(znode)) | |
1308 | return PTR_ERR(znode); | |
1309 | } | |
1310 | ||
1311 | znode = dirty_cow_znode(c, &c->zroot); | |
1312 | if (IS_ERR(znode)) | |
1313 | return PTR_ERR(znode); | |
1314 | ||
1315 | znode->time = time; | |
1316 | ||
1317 | while (1) { | |
1318 | struct ubifs_zbranch *zbr; | |
1319 | ||
1320 | exact = ubifs_search_zbranch(c, znode, key, n); | |
1321 | ||
1322 | if (znode->level == 0) | |
1323 | break; | |
1324 | ||
1325 | if (*n < 0) | |
1326 | *n = 0; | |
1327 | zbr = &znode->zbranch[*n]; | |
1328 | ||
1329 | if (zbr->znode) { | |
1330 | znode->time = time; | |
1331 | znode = dirty_cow_znode(c, zbr); | |
1332 | if (IS_ERR(znode)) | |
1333 | return PTR_ERR(znode); | |
1334 | continue; | |
1335 | } | |
1336 | ||
1337 | /* znode is not in TNC cache, load it from the media */ | |
1338 | znode = ubifs_load_znode(c, zbr, znode, *n); | |
1339 | if (IS_ERR(znode)) | |
1340 | return PTR_ERR(znode); | |
1341 | znode = dirty_cow_znode(c, zbr); | |
1342 | if (IS_ERR(znode)) | |
1343 | return PTR_ERR(znode); | |
1344 | } | |
1345 | ||
1346 | *zn = znode; | |
1347 | if (exact || !is_hash_key(c, key) || *n != -1) { | |
1348 | dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); | |
1349 | return exact; | |
1350 | } | |
1351 | ||
1352 | /* | |
1353 | * See huge comment at 'lookup_level0_dirty()' what is the rest of the | |
1354 | * code. | |
1355 | */ | |
1356 | err = tnc_prev(c, &znode, n); | |
1357 | if (err == -ENOENT) { | |
1358 | *n = -1; | |
1359 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1360 | return 0; | |
1361 | } | |
1362 | if (unlikely(err < 0)) | |
1363 | return err; | |
1364 | if (keys_cmp(c, key, &znode->zbranch[*n].key)) { | |
1365 | *n = -1; | |
1366 | dbg_tnc("found 0, lvl %d, n -1", znode->level); | |
1367 | return 0; | |
1368 | } | |
1369 | ||
1370 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
1371 | znode = dirty_cow_bottom_up(c, znode); | |
1372 | if (IS_ERR(znode)) | |
1373 | return PTR_ERR(znode); | |
1374 | } | |
1375 | ||
1376 | dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); | |
1377 | *zn = znode; | |
1378 | return 1; | |
1379 | } | |
1380 | ||
1381 | /** | |
601c0bc4 | 1382 | * maybe_leb_gced - determine if a LEB may have been garbage collected. |
1e51764a | 1383 | * @c: UBIFS file-system description object |
601c0bc4 AH |
1384 | * @lnum: LEB number |
1385 | * @gc_seq1: garbage collection sequence number | |
1e51764a | 1386 | * |
601c0bc4 AH |
1387 | * This function determines if @lnum may have been garbage collected since |
1388 | * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise | |
1389 | * %0 is returned. | |
1e51764a | 1390 | */ |
601c0bc4 | 1391 | static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1) |
1e51764a | 1392 | { |
601c0bc4 | 1393 | int gc_seq2, gced_lnum; |
1e51764a | 1394 | |
601c0bc4 AH |
1395 | gced_lnum = c->gced_lnum; |
1396 | smp_rmb(); | |
1397 | gc_seq2 = c->gc_seq; | |
1398 | /* Same seq means no GC */ | |
1399 | if (gc_seq1 == gc_seq2) | |
1400 | return 0; | |
1401 | /* Different by more than 1 means we don't know */ | |
1402 | if (gc_seq1 + 1 != gc_seq2) | |
1403 | return 1; | |
1404 | /* | |
1405 | * We have seen the sequence number has increased by 1. Now we need to | |
1406 | * be sure we read the right LEB number, so read it again. | |
1407 | */ | |
1408 | smp_rmb(); | |
1409 | if (gced_lnum != c->gced_lnum) | |
1410 | return 1; | |
1411 | /* Finally we can check lnum */ | |
1412 | if (gced_lnum == lnum) | |
1413 | return 1; | |
1414 | return 0; | |
1e51764a AB |
1415 | } |
1416 | ||
1417 | /** | |
1418 | * ubifs_tnc_locate - look up a file-system node and return it and its location. | |
1419 | * @c: UBIFS file-system description object | |
1420 | * @key: node key to lookup | |
1421 | * @node: the node is returned here | |
1422 | * @lnum: LEB number is returned here | |
1423 | * @offs: offset is returned here | |
1424 | * | |
e3c3efc2 | 1425 | * This function looks up and reads node with key @key. The caller has to make |
601c0bc4 AH |
1426 | * sure the @node buffer is large enough to fit the node. Returns zero in case |
1427 | * of success, %-ENOENT if the node was not found, and a negative error code in | |
1428 | * case of failure. The node location can be returned in @lnum and @offs. | |
1e51764a AB |
1429 | */ |
1430 | int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, | |
1431 | void *node, int *lnum, int *offs) | |
1432 | { | |
601c0bc4 | 1433 | int found, n, err, safely = 0, gc_seq1; |
1e51764a AB |
1434 | struct ubifs_znode *znode; |
1435 | struct ubifs_zbranch zbr, *zt; | |
1436 | ||
601c0bc4 | 1437 | again: |
1e51764a AB |
1438 | mutex_lock(&c->tnc_mutex); |
1439 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
1440 | if (!found) { | |
1441 | err = -ENOENT; | |
1442 | goto out; | |
1443 | } else if (found < 0) { | |
1444 | err = found; | |
1445 | goto out; | |
1446 | } | |
1447 | zt = &znode->zbranch[n]; | |
601c0bc4 AH |
1448 | if (lnum) { |
1449 | *lnum = zt->lnum; | |
1450 | *offs = zt->offs; | |
1451 | } | |
1e51764a AB |
1452 | if (is_hash_key(c, key)) { |
1453 | /* | |
1454 | * In this case the leaf node cache gets used, so we pass the | |
1455 | * address of the zbranch and keep the mutex locked | |
1456 | */ | |
1e51764a AB |
1457 | err = tnc_read_node_nm(c, zt, node); |
1458 | goto out; | |
1459 | } | |
601c0bc4 AH |
1460 | if (safely) { |
1461 | err = ubifs_tnc_read_node(c, zt, node); | |
1462 | goto out; | |
1463 | } | |
1464 | /* Drop the TNC mutex prematurely and race with garbage collection */ | |
1e51764a | 1465 | zbr = znode->zbranch[n]; |
601c0bc4 | 1466 | gc_seq1 = c->gc_seq; |
1e51764a AB |
1467 | mutex_unlock(&c->tnc_mutex); |
1468 | ||
601c0bc4 AH |
1469 | if (ubifs_get_wbuf(c, zbr.lnum)) { |
1470 | /* We do not GC journal heads */ | |
1471 | err = ubifs_tnc_read_node(c, &zbr, node); | |
1472 | return err; | |
1473 | } | |
1e51764a | 1474 | |
601c0bc4 | 1475 | err = fallible_read_node(c, key, &zbr, node); |
6dcfac4f | 1476 | if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) { |
601c0bc4 AH |
1477 | /* |
1478 | * The node may have been GC'ed out from under us so try again | |
1479 | * while keeping the TNC mutex locked. | |
1480 | */ | |
1481 | safely = 1; | |
1482 | goto again; | |
1483 | } | |
1484 | return 0; | |
1e51764a AB |
1485 | |
1486 | out: | |
1487 | mutex_unlock(&c->tnc_mutex); | |
1488 | return err; | |
1489 | } | |
1490 | ||
4793e7c5 AH |
1491 | /** |
1492 | * ubifs_tnc_get_bu_keys - lookup keys for bulk-read. | |
1493 | * @c: UBIFS file-system description object | |
1494 | * @bu: bulk-read parameters and results | |
1495 | * | |
1496 | * Lookup consecutive data node keys for the same inode that reside | |
6c0c42cd AB |
1497 | * consecutively in the same LEB. This function returns zero in case of success |
1498 | * and a negative error code in case of failure. | |
1499 | * | |
1500 | * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function | |
1501 | * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares | |
6f7ab6d4 | 1502 | * maximum possible amount of nodes for bulk-read. |
4793e7c5 AH |
1503 | */ |
1504 | int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu) | |
1505 | { | |
1506 | int n, err = 0, lnum = -1, uninitialized_var(offs); | |
1507 | int uninitialized_var(len); | |
1508 | unsigned int block = key_block(c, &bu->key); | |
1509 | struct ubifs_znode *znode; | |
1510 | ||
1511 | bu->cnt = 0; | |
1512 | bu->blk_cnt = 0; | |
1513 | bu->eof = 0; | |
1514 | ||
1515 | mutex_lock(&c->tnc_mutex); | |
1516 | /* Find first key */ | |
1517 | err = ubifs_lookup_level0(c, &bu->key, &znode, &n); | |
1518 | if (err < 0) | |
1519 | goto out; | |
1520 | if (err) { | |
1521 | /* Key found */ | |
1522 | len = znode->zbranch[n].len; | |
1523 | /* The buffer must be big enough for at least 1 node */ | |
1524 | if (len > bu->buf_len) { | |
1525 | err = -EINVAL; | |
1526 | goto out; | |
1527 | } | |
1528 | /* Add this key */ | |
1529 | bu->zbranch[bu->cnt++] = znode->zbranch[n]; | |
1530 | bu->blk_cnt += 1; | |
1531 | lnum = znode->zbranch[n].lnum; | |
1532 | offs = ALIGN(znode->zbranch[n].offs + len, 8); | |
1533 | } | |
1534 | while (1) { | |
1535 | struct ubifs_zbranch *zbr; | |
1536 | union ubifs_key *key; | |
1537 | unsigned int next_block; | |
1538 | ||
1539 | /* Find next key */ | |
1540 | err = tnc_next(c, &znode, &n); | |
1541 | if (err) | |
1542 | goto out; | |
1543 | zbr = &znode->zbranch[n]; | |
1544 | key = &zbr->key; | |
1545 | /* See if there is another data key for this file */ | |
1546 | if (key_inum(c, key) != key_inum(c, &bu->key) || | |
1547 | key_type(c, key) != UBIFS_DATA_KEY) { | |
1548 | err = -ENOENT; | |
1549 | goto out; | |
1550 | } | |
1551 | if (lnum < 0) { | |
1552 | /* First key found */ | |
1553 | lnum = zbr->lnum; | |
1554 | offs = ALIGN(zbr->offs + zbr->len, 8); | |
1555 | len = zbr->len; | |
1556 | if (len > bu->buf_len) { | |
1557 | err = -EINVAL; | |
1558 | goto out; | |
1559 | } | |
1560 | } else { | |
1561 | /* | |
1562 | * The data nodes must be in consecutive positions in | |
1563 | * the same LEB. | |
1564 | */ | |
1565 | if (zbr->lnum != lnum || zbr->offs != offs) | |
1566 | goto out; | |
1567 | offs += ALIGN(zbr->len, 8); | |
1568 | len = ALIGN(len, 8) + zbr->len; | |
1569 | /* Must not exceed buffer length */ | |
1570 | if (len > bu->buf_len) | |
1571 | goto out; | |
1572 | } | |
1573 | /* Allow for holes */ | |
1574 | next_block = key_block(c, key); | |
1575 | bu->blk_cnt += (next_block - block - 1); | |
1576 | if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) | |
1577 | goto out; | |
1578 | block = next_block; | |
1579 | /* Add this key */ | |
1580 | bu->zbranch[bu->cnt++] = *zbr; | |
1581 | bu->blk_cnt += 1; | |
1582 | /* See if we have room for more */ | |
1583 | if (bu->cnt >= UBIFS_MAX_BULK_READ) | |
1584 | goto out; | |
1585 | if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) | |
1586 | goto out; | |
1587 | } | |
1588 | out: | |
1589 | if (err == -ENOENT) { | |
1590 | bu->eof = 1; | |
1591 | err = 0; | |
1592 | } | |
1593 | bu->gc_seq = c->gc_seq; | |
1594 | mutex_unlock(&c->tnc_mutex); | |
1595 | if (err) | |
1596 | return err; | |
1597 | /* | |
1598 | * An enormous hole could cause bulk-read to encompass too many | |
1599 | * page cache pages, so limit the number here. | |
1600 | */ | |
63c300b6 | 1601 | if (bu->blk_cnt > UBIFS_MAX_BULK_READ) |
4793e7c5 AH |
1602 | bu->blk_cnt = UBIFS_MAX_BULK_READ; |
1603 | /* | |
1604 | * Ensure that bulk-read covers a whole number of page cache | |
1605 | * pages. | |
1606 | */ | |
1607 | if (UBIFS_BLOCKS_PER_PAGE == 1 || | |
1608 | !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1))) | |
1609 | return 0; | |
1610 | if (bu->eof) { | |
1611 | /* At the end of file we can round up */ | |
1612 | bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1; | |
1613 | return 0; | |
1614 | } | |
1615 | /* Exclude data nodes that do not make up a whole page cache page */ | |
1616 | block = key_block(c, &bu->key) + bu->blk_cnt; | |
1617 | block &= ~(UBIFS_BLOCKS_PER_PAGE - 1); | |
1618 | while (bu->cnt) { | |
1619 | if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block) | |
1620 | break; | |
1621 | bu->cnt -= 1; | |
1622 | } | |
1623 | return 0; | |
1624 | } | |
1625 | ||
1626 | /** | |
1627 | * read_wbuf - bulk-read from a LEB with a wbuf. | |
1628 | * @wbuf: wbuf that may overlap the read | |
1629 | * @buf: buffer into which to read | |
1630 | * @len: read length | |
1631 | * @lnum: LEB number from which to read | |
1632 | * @offs: offset from which to read | |
1633 | * | |
1634 | * This functions returns %0 on success or a negative error code on failure. | |
1635 | */ | |
1636 | static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum, | |
1637 | int offs) | |
1638 | { | |
1639 | const struct ubifs_info *c = wbuf->c; | |
1640 | int rlen, overlap; | |
1641 | ||
1642 | dbg_io("LEB %d:%d, length %d", lnum, offs, len); | |
1643 | ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0); | |
1644 | ubifs_assert(!(offs & 7) && offs < c->leb_size); | |
1645 | ubifs_assert(offs + len <= c->leb_size); | |
1646 | ||
1647 | spin_lock(&wbuf->lock); | |
1648 | overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs); | |
1649 | if (!overlap) { | |
1650 | /* We may safely unlock the write-buffer and read the data */ | |
1651 | spin_unlock(&wbuf->lock); | |
d304820a | 1652 | return ubifs_leb_read(c, lnum, buf, offs, len, 0); |
4793e7c5 AH |
1653 | } |
1654 | ||
1655 | /* Don't read under wbuf */ | |
1656 | rlen = wbuf->offs - offs; | |
1657 | if (rlen < 0) | |
1658 | rlen = 0; | |
1659 | ||
1660 | /* Copy the rest from the write-buffer */ | |
1661 | memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen); | |
1662 | spin_unlock(&wbuf->lock); | |
1663 | ||
1664 | if (rlen > 0) | |
1665 | /* Read everything that goes before write-buffer */ | |
d304820a | 1666 | return ubifs_leb_read(c, lnum, buf, offs, rlen, 0); |
4793e7c5 AH |
1667 | |
1668 | return 0; | |
1669 | } | |
1670 | ||
1671 | /** | |
1672 | * validate_data_node - validate data nodes for bulk-read. | |
1673 | * @c: UBIFS file-system description object | |
1674 | * @buf: buffer containing data node to validate | |
1675 | * @zbr: zbranch of data node to validate | |
1676 | * | |
1677 | * This functions returns %0 on success or a negative error code on failure. | |
1678 | */ | |
1679 | static int validate_data_node(struct ubifs_info *c, void *buf, | |
1680 | struct ubifs_zbranch *zbr) | |
1681 | { | |
1682 | union ubifs_key key1; | |
1683 | struct ubifs_ch *ch = buf; | |
1684 | int err, len; | |
1685 | ||
1686 | if (ch->node_type != UBIFS_DATA_NODE) { | |
235c362b | 1687 | ubifs_err(c, "bad node type (%d but expected %d)", |
4793e7c5 AH |
1688 | ch->node_type, UBIFS_DATA_NODE); |
1689 | goto out_err; | |
1690 | } | |
1691 | ||
2953e73f | 1692 | err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0); |
4793e7c5 | 1693 | if (err) { |
235c362b | 1694 | ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE); |
4793e7c5 AH |
1695 | goto out; |
1696 | } | |
1697 | ||
1698 | len = le32_to_cpu(ch->len); | |
1699 | if (len != zbr->len) { | |
235c362b | 1700 | ubifs_err(c, "bad node length %d, expected %d", len, zbr->len); |
4793e7c5 AH |
1701 | goto out_err; |
1702 | } | |
1703 | ||
1704 | /* Make sure the key of the read node is correct */ | |
1705 | key_read(c, buf + UBIFS_KEY_OFFSET, &key1); | |
1706 | if (!keys_eq(c, &zbr->key, &key1)) { | |
235c362b | 1707 | ubifs_err(c, "bad key in node at LEB %d:%d", |
4793e7c5 | 1708 | zbr->lnum, zbr->offs); |
515315a1 AB |
1709 | dbg_tnck(&zbr->key, "looked for key "); |
1710 | dbg_tnck(&key1, "found node's key "); | |
4793e7c5 AH |
1711 | goto out_err; |
1712 | } | |
1713 | ||
1714 | return 0; | |
1715 | ||
1716 | out_err: | |
1717 | err = -EINVAL; | |
1718 | out: | |
235c362b | 1719 | ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs); |
edf6be24 | 1720 | ubifs_dump_node(c, buf); |
7c46d0ae | 1721 | dump_stack(); |
4793e7c5 AH |
1722 | return err; |
1723 | } | |
1724 | ||
1725 | /** | |
1726 | * ubifs_tnc_bulk_read - read a number of data nodes in one go. | |
1727 | * @c: UBIFS file-system description object | |
1728 | * @bu: bulk-read parameters and results | |
1729 | * | |
1730 | * This functions reads and validates the data nodes that were identified by the | |
1731 | * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success, | |
1732 | * -EAGAIN to indicate a race with GC, or another negative error code on | |
1733 | * failure. | |
1734 | */ | |
1735 | int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu) | |
1736 | { | |
1737 | int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i; | |
1738 | struct ubifs_wbuf *wbuf; | |
1739 | void *buf; | |
1740 | ||
1741 | len = bu->zbranch[bu->cnt - 1].offs; | |
1742 | len += bu->zbranch[bu->cnt - 1].len - offs; | |
1743 | if (len > bu->buf_len) { | |
235c362b | 1744 | ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len); |
4793e7c5 AH |
1745 | return -EINVAL; |
1746 | } | |
1747 | ||
1748 | /* Do the read */ | |
1749 | wbuf = ubifs_get_wbuf(c, lnum); | |
1750 | if (wbuf) | |
1751 | err = read_wbuf(wbuf, bu->buf, len, lnum, offs); | |
1752 | else | |
d304820a | 1753 | err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0); |
4793e7c5 AH |
1754 | |
1755 | /* Check for a race with GC */ | |
1756 | if (maybe_leb_gced(c, lnum, bu->gc_seq)) | |
1757 | return -EAGAIN; | |
1758 | ||
1759 | if (err && err != -EBADMSG) { | |
235c362b | 1760 | ubifs_err(c, "failed to read from LEB %d:%d, error %d", |
4793e7c5 | 1761 | lnum, offs, err); |
7c46d0ae | 1762 | dump_stack(); |
515315a1 | 1763 | dbg_tnck(&bu->key, "key "); |
4793e7c5 AH |
1764 | return err; |
1765 | } | |
1766 | ||
1767 | /* Validate the nodes read */ | |
1768 | buf = bu->buf; | |
1769 | for (i = 0; i < bu->cnt; i++) { | |
1770 | err = validate_data_node(c, buf, &bu->zbranch[i]); | |
1771 | if (err) | |
1772 | return err; | |
1773 | buf = buf + ALIGN(bu->zbranch[i].len, 8); | |
1774 | } | |
1775 | ||
1776 | return 0; | |
1777 | } | |
1778 | ||
1e51764a AB |
1779 | /** |
1780 | * do_lookup_nm- look up a "hashed" node. | |
1781 | * @c: UBIFS file-system description object | |
1782 | * @key: node key to lookup | |
1783 | * @node: the node is returned here | |
1784 | * @nm: node name | |
1785 | * | |
1786 | * This function look up and reads a node which contains name hash in the key. | |
1787 | * Since the hash may have collisions, there may be many nodes with the same | |
1788 | * key, so we have to sequentially look to all of them until the needed one is | |
1789 | * found. This function returns zero in case of success, %-ENOENT if the node | |
1790 | * was not found, and a negative error code in case of failure. | |
1791 | */ | |
1792 | static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, | |
1793 | void *node, const struct qstr *nm) | |
1794 | { | |
1795 | int found, n, err; | |
1796 | struct ubifs_znode *znode; | |
1e51764a | 1797 | |
515315a1 | 1798 | dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name); |
1e51764a AB |
1799 | mutex_lock(&c->tnc_mutex); |
1800 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
1801 | if (!found) { | |
1802 | err = -ENOENT; | |
1803 | goto out_unlock; | |
1804 | } else if (found < 0) { | |
1805 | err = found; | |
1806 | goto out_unlock; | |
1807 | } | |
1808 | ||
1809 | ubifs_assert(n >= 0); | |
1810 | ||
1811 | err = resolve_collision(c, key, &znode, &n, nm); | |
1812 | dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); | |
1813 | if (unlikely(err < 0)) | |
1814 | goto out_unlock; | |
1815 | if (err == 0) { | |
1816 | err = -ENOENT; | |
1817 | goto out_unlock; | |
1818 | } | |
1819 | ||
761e29f3 | 1820 | err = tnc_read_node_nm(c, &znode->zbranch[n], node); |
1e51764a AB |
1821 | |
1822 | out_unlock: | |
1823 | mutex_unlock(&c->tnc_mutex); | |
1824 | return err; | |
1825 | } | |
1826 | ||
1827 | /** | |
1828 | * ubifs_tnc_lookup_nm - look up a "hashed" node. | |
1829 | * @c: UBIFS file-system description object | |
1830 | * @key: node key to lookup | |
1831 | * @node: the node is returned here | |
1832 | * @nm: node name | |
1833 | * | |
1834 | * This function look up and reads a node which contains name hash in the key. | |
1835 | * Since the hash may have collisions, there may be many nodes with the same | |
1836 | * key, so we have to sequentially look to all of them until the needed one is | |
1837 | * found. This function returns zero in case of success, %-ENOENT if the node | |
1838 | * was not found, and a negative error code in case of failure. | |
1839 | */ | |
1840 | int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, | |
1841 | void *node, const struct qstr *nm) | |
1842 | { | |
1843 | int err, len; | |
1844 | const struct ubifs_dent_node *dent = node; | |
1845 | ||
1846 | /* | |
1847 | * We assume that in most of the cases there are no name collisions and | |
1848 | * 'ubifs_tnc_lookup()' returns us the right direntry. | |
1849 | */ | |
1850 | err = ubifs_tnc_lookup(c, key, node); | |
1851 | if (err) | |
1852 | return err; | |
1853 | ||
1854 | len = le16_to_cpu(dent->nlen); | |
1855 | if (nm->len == len && !memcmp(dent->name, nm->name, len)) | |
1856 | return 0; | |
1857 | ||
1858 | /* | |
1859 | * Unluckily, there are hash collisions and we have to iterate over | |
1860 | * them look at each direntry with colliding name hash sequentially. | |
1861 | */ | |
1862 | return do_lookup_nm(c, key, node, nm); | |
1863 | } | |
1864 | ||
1865 | /** | |
1866 | * correct_parent_keys - correct parent znodes' keys. | |
1867 | * @c: UBIFS file-system description object | |
1868 | * @znode: znode to correct parent znodes for | |
1869 | * | |
1870 | * This is a helper function for 'tnc_insert()'. When the key of the leftmost | |
1871 | * zbranch changes, keys of parent znodes have to be corrected. This helper | |
1872 | * function is called in such situations and corrects the keys if needed. | |
1873 | */ | |
1874 | static void correct_parent_keys(const struct ubifs_info *c, | |
1875 | struct ubifs_znode *znode) | |
1876 | { | |
1877 | union ubifs_key *key, *key1; | |
1878 | ||
1879 | ubifs_assert(znode->parent); | |
1880 | ubifs_assert(znode->iip == 0); | |
1881 | ||
1882 | key = &znode->zbranch[0].key; | |
1883 | key1 = &znode->parent->zbranch[0].key; | |
1884 | ||
1885 | while (keys_cmp(c, key, key1) < 0) { | |
1886 | key_copy(c, key, key1); | |
1887 | znode = znode->parent; | |
1888 | znode->alt = 1; | |
1889 | if (!znode->parent || znode->iip) | |
1890 | break; | |
1891 | key1 = &znode->parent->zbranch[0].key; | |
1892 | } | |
1893 | } | |
1894 | ||
1895 | /** | |
1896 | * insert_zbranch - insert a zbranch into a znode. | |
1897 | * @znode: znode into which to insert | |
1898 | * @zbr: zbranch to insert | |
1899 | * @n: slot number to insert to | |
1900 | * | |
1901 | * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in | |
1902 | * znode's array of zbranches and keeps zbranches consolidated, so when a new | |
1903 | * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th | |
1904 | * slot, zbranches starting from @n have to be moved right. | |
1905 | */ | |
1906 | static void insert_zbranch(struct ubifs_znode *znode, | |
1907 | const struct ubifs_zbranch *zbr, int n) | |
1908 | { | |
1909 | int i; | |
1910 | ||
1911 | ubifs_assert(ubifs_zn_dirty(znode)); | |
1912 | ||
1913 | if (znode->level) { | |
1914 | for (i = znode->child_cnt; i > n; i--) { | |
1915 | znode->zbranch[i] = znode->zbranch[i - 1]; | |
1916 | if (znode->zbranch[i].znode) | |
1917 | znode->zbranch[i].znode->iip = i; | |
1918 | } | |
1919 | if (zbr->znode) | |
1920 | zbr->znode->iip = n; | |
1921 | } else | |
1922 | for (i = znode->child_cnt; i > n; i--) | |
1923 | znode->zbranch[i] = znode->zbranch[i - 1]; | |
1924 | ||
1925 | znode->zbranch[n] = *zbr; | |
1926 | znode->child_cnt += 1; | |
1927 | ||
1928 | /* | |
1929 | * After inserting at slot zero, the lower bound of the key range of | |
1930 | * this znode may have changed. If this znode is subsequently split | |
1931 | * then the upper bound of the key range may change, and furthermore | |
1932 | * it could change to be lower than the original lower bound. If that | |
1933 | * happens, then it will no longer be possible to find this znode in the | |
1934 | * TNC using the key from the index node on flash. That is bad because | |
1935 | * if it is not found, we will assume it is obsolete and may overwrite | |
1936 | * it. Then if there is an unclean unmount, we will start using the | |
1937 | * old index which will be broken. | |
1938 | * | |
1939 | * So we first mark znodes that have insertions at slot zero, and then | |
1940 | * if they are split we add their lnum/offs to the old_idx tree. | |
1941 | */ | |
1942 | if (n == 0) | |
1943 | znode->alt = 1; | |
1944 | } | |
1945 | ||
1946 | /** | |
1947 | * tnc_insert - insert a node into TNC. | |
1948 | * @c: UBIFS file-system description object | |
1949 | * @znode: znode to insert into | |
1950 | * @zbr: branch to insert | |
1951 | * @n: slot number to insert new zbranch to | |
1952 | * | |
1953 | * This function inserts a new node described by @zbr into znode @znode. If | |
1954 | * znode does not have a free slot for new zbranch, it is split. Parent znodes | |
1955 | * are splat as well if needed. Returns zero in case of success or a negative | |
1956 | * error code in case of failure. | |
1957 | */ | |
1958 | static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode, | |
1959 | struct ubifs_zbranch *zbr, int n) | |
1960 | { | |
1961 | struct ubifs_znode *zn, *zi, *zp; | |
1962 | int i, keep, move, appending = 0; | |
2242c689 | 1963 | union ubifs_key *key = &zbr->key, *key1; |
1e51764a AB |
1964 | |
1965 | ubifs_assert(n >= 0 && n <= c->fanout); | |
1966 | ||
1967 | /* Implement naive insert for now */ | |
1968 | again: | |
1969 | zp = znode->parent; | |
1970 | if (znode->child_cnt < c->fanout) { | |
1971 | ubifs_assert(n != c->fanout); | |
515315a1 | 1972 | dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level); |
1e51764a AB |
1973 | |
1974 | insert_zbranch(znode, zbr, n); | |
1975 | ||
1976 | /* Ensure parent's key is correct */ | |
1977 | if (n == 0 && zp && znode->iip == 0) | |
1978 | correct_parent_keys(c, znode); | |
1979 | ||
1980 | return 0; | |
1981 | } | |
1982 | ||
1983 | /* | |
1984 | * Unfortunately, @znode does not have more empty slots and we have to | |
1985 | * split it. | |
1986 | */ | |
515315a1 | 1987 | dbg_tnck(key, "splitting level %d, key ", znode->level); |
1e51764a AB |
1988 | |
1989 | if (znode->alt) | |
1990 | /* | |
1991 | * We can no longer be sure of finding this znode by key, so we | |
1992 | * record it in the old_idx tree. | |
1993 | */ | |
1994 | ins_clr_old_idx_znode(c, znode); | |
1995 | ||
1996 | zn = kzalloc(c->max_znode_sz, GFP_NOFS); | |
1997 | if (!zn) | |
1998 | return -ENOMEM; | |
1999 | zn->parent = zp; | |
2000 | zn->level = znode->level; | |
2001 | ||
2002 | /* Decide where to split */ | |
2242c689 AH |
2003 | if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) { |
2004 | /* Try not to split consecutive data keys */ | |
2005 | if (n == c->fanout) { | |
2006 | key1 = &znode->zbranch[n - 1].key; | |
2007 | if (key_inum(c, key1) == key_inum(c, key) && | |
2008 | key_type(c, key1) == UBIFS_DATA_KEY) | |
2009 | appending = 1; | |
2010 | } else | |
2011 | goto check_split; | |
2012 | } else if (appending && n != c->fanout) { | |
2013 | /* Try not to split consecutive data keys */ | |
2014 | appending = 0; | |
2015 | check_split: | |
2016 | if (n >= (c->fanout + 1) / 2) { | |
2017 | key1 = &znode->zbranch[0].key; | |
2018 | if (key_inum(c, key1) == key_inum(c, key) && | |
2019 | key_type(c, key1) == UBIFS_DATA_KEY) { | |
2020 | key1 = &znode->zbranch[n].key; | |
2021 | if (key_inum(c, key1) != key_inum(c, key) || | |
2022 | key_type(c, key1) != UBIFS_DATA_KEY) { | |
2023 | keep = n; | |
2024 | move = c->fanout - keep; | |
2025 | zi = znode; | |
2026 | goto do_split; | |
2027 | } | |
2028 | } | |
2029 | } | |
1e51764a AB |
2030 | } |
2031 | ||
2032 | if (appending) { | |
2033 | keep = c->fanout; | |
2034 | move = 0; | |
2035 | } else { | |
2036 | keep = (c->fanout + 1) / 2; | |
2037 | move = c->fanout - keep; | |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * Although we don't at present, we could look at the neighbors and see | |
2042 | * if we can move some zbranches there. | |
2043 | */ | |
2044 | ||
2045 | if (n < keep) { | |
2046 | /* Insert into existing znode */ | |
2047 | zi = znode; | |
2048 | move += 1; | |
2049 | keep -= 1; | |
2050 | } else { | |
2051 | /* Insert into new znode */ | |
2052 | zi = zn; | |
2053 | n -= keep; | |
2054 | /* Re-parent */ | |
2055 | if (zn->level != 0) | |
2056 | zbr->znode->parent = zn; | |
2057 | } | |
2058 | ||
2242c689 AH |
2059 | do_split: |
2060 | ||
1e51764a AB |
2061 | __set_bit(DIRTY_ZNODE, &zn->flags); |
2062 | atomic_long_inc(&c->dirty_zn_cnt); | |
2063 | ||
2064 | zn->child_cnt = move; | |
2065 | znode->child_cnt = keep; | |
2066 | ||
2067 | dbg_tnc("moving %d, keeping %d", move, keep); | |
2068 | ||
2069 | /* Move zbranch */ | |
2070 | for (i = 0; i < move; i++) { | |
2071 | zn->zbranch[i] = znode->zbranch[keep + i]; | |
2072 | /* Re-parent */ | |
2073 | if (zn->level != 0) | |
2074 | if (zn->zbranch[i].znode) { | |
2075 | zn->zbranch[i].znode->parent = zn; | |
2076 | zn->zbranch[i].znode->iip = i; | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | /* Insert new key and branch */ | |
515315a1 | 2081 | dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level); |
1e51764a AB |
2082 | |
2083 | insert_zbranch(zi, zbr, n); | |
2084 | ||
2085 | /* Insert new znode (produced by spitting) into the parent */ | |
2086 | if (zp) { | |
2242c689 AH |
2087 | if (n == 0 && zi == znode && znode->iip == 0) |
2088 | correct_parent_keys(c, znode); | |
2089 | ||
1e51764a AB |
2090 | /* Locate insertion point */ |
2091 | n = znode->iip + 1; | |
1e51764a AB |
2092 | |
2093 | /* Tail recursion */ | |
2094 | zbr->key = zn->zbranch[0].key; | |
2095 | zbr->znode = zn; | |
2096 | zbr->lnum = 0; | |
2097 | zbr->offs = 0; | |
2098 | zbr->len = 0; | |
2099 | znode = zp; | |
2100 | ||
2101 | goto again; | |
2102 | } | |
2103 | ||
2104 | /* We have to split root znode */ | |
2105 | dbg_tnc("creating new zroot at level %d", znode->level + 1); | |
2106 | ||
2107 | zi = kzalloc(c->max_znode_sz, GFP_NOFS); | |
2108 | if (!zi) | |
2109 | return -ENOMEM; | |
2110 | ||
2111 | zi->child_cnt = 2; | |
2112 | zi->level = znode->level + 1; | |
2113 | ||
2114 | __set_bit(DIRTY_ZNODE, &zi->flags); | |
2115 | atomic_long_inc(&c->dirty_zn_cnt); | |
2116 | ||
2117 | zi->zbranch[0].key = znode->zbranch[0].key; | |
2118 | zi->zbranch[0].znode = znode; | |
2119 | zi->zbranch[0].lnum = c->zroot.lnum; | |
2120 | zi->zbranch[0].offs = c->zroot.offs; | |
2121 | zi->zbranch[0].len = c->zroot.len; | |
2122 | zi->zbranch[1].key = zn->zbranch[0].key; | |
2123 | zi->zbranch[1].znode = zn; | |
2124 | ||
2125 | c->zroot.lnum = 0; | |
2126 | c->zroot.offs = 0; | |
2127 | c->zroot.len = 0; | |
2128 | c->zroot.znode = zi; | |
2129 | ||
2130 | zn->parent = zi; | |
2131 | zn->iip = 1; | |
2132 | znode->parent = zi; | |
2133 | znode->iip = 0; | |
2134 | ||
2135 | return 0; | |
2136 | } | |
2137 | ||
2138 | /** | |
2139 | * ubifs_tnc_add - add a node to TNC. | |
2140 | * @c: UBIFS file-system description object | |
2141 | * @key: key to add | |
2142 | * @lnum: LEB number of node | |
2143 | * @offs: node offset | |
2144 | * @len: node length | |
2145 | * | |
2146 | * This function adds a node with key @key to TNC. The node may be new or it may | |
2147 | * obsolete some existing one. Returns %0 on success or negative error code on | |
2148 | * failure. | |
2149 | */ | |
2150 | int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum, | |
2151 | int offs, int len) | |
2152 | { | |
2153 | int found, n, err = 0; | |
2154 | struct ubifs_znode *znode; | |
2155 | ||
2156 | mutex_lock(&c->tnc_mutex); | |
515315a1 | 2157 | dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len); |
1e51764a AB |
2158 | found = lookup_level0_dirty(c, key, &znode, &n); |
2159 | if (!found) { | |
2160 | struct ubifs_zbranch zbr; | |
2161 | ||
2162 | zbr.znode = NULL; | |
2163 | zbr.lnum = lnum; | |
2164 | zbr.offs = offs; | |
2165 | zbr.len = len; | |
2166 | key_copy(c, key, &zbr.key); | |
2167 | err = tnc_insert(c, znode, &zbr, n + 1); | |
2168 | } else if (found == 1) { | |
2169 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2170 | ||
2171 | lnc_free(zbr); | |
2172 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2173 | zbr->lnum = lnum; | |
2174 | zbr->offs = offs; | |
2175 | zbr->len = len; | |
2176 | } else | |
2177 | err = found; | |
2178 | if (!err) | |
2179 | err = dbg_check_tnc(c, 0); | |
2180 | mutex_unlock(&c->tnc_mutex); | |
2181 | ||
2182 | return err; | |
2183 | } | |
2184 | ||
2185 | /** | |
2186 | * ubifs_tnc_replace - replace a node in the TNC only if the old node is found. | |
2187 | * @c: UBIFS file-system description object | |
2188 | * @key: key to add | |
2189 | * @old_lnum: LEB number of old node | |
2190 | * @old_offs: old node offset | |
2191 | * @lnum: LEB number of node | |
2192 | * @offs: node offset | |
2193 | * @len: node length | |
2194 | * | |
2195 | * This function replaces a node with key @key in the TNC only if the old node | |
2196 | * is found. This function is called by garbage collection when node are moved. | |
2197 | * Returns %0 on success or negative error code on failure. | |
2198 | */ | |
2199 | int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key, | |
2200 | int old_lnum, int old_offs, int lnum, int offs, int len) | |
2201 | { | |
2202 | int found, n, err = 0; | |
2203 | struct ubifs_znode *znode; | |
2204 | ||
2205 | mutex_lock(&c->tnc_mutex); | |
515315a1 AB |
2206 | dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum, |
2207 | old_offs, lnum, offs, len); | |
1e51764a AB |
2208 | found = lookup_level0_dirty(c, key, &znode, &n); |
2209 | if (found < 0) { | |
2210 | err = found; | |
2211 | goto out_unlock; | |
2212 | } | |
2213 | ||
2214 | if (found == 1) { | |
2215 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2216 | ||
2217 | found = 0; | |
2218 | if (zbr->lnum == old_lnum && zbr->offs == old_offs) { | |
2219 | lnc_free(zbr); | |
2220 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2221 | if (err) | |
2222 | goto out_unlock; | |
2223 | zbr->lnum = lnum; | |
2224 | zbr->offs = offs; | |
2225 | zbr->len = len; | |
2226 | found = 1; | |
2227 | } else if (is_hash_key(c, key)) { | |
2228 | found = resolve_collision_directly(c, key, &znode, &n, | |
2229 | old_lnum, old_offs); | |
2230 | dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d", | |
2231 | found, znode, n, old_lnum, old_offs); | |
2232 | if (found < 0) { | |
2233 | err = found; | |
2234 | goto out_unlock; | |
2235 | } | |
2236 | ||
2237 | if (found) { | |
2238 | /* Ensure the znode is dirtied */ | |
2239 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2240 | znode = dirty_cow_bottom_up(c, znode); |
2241 | if (IS_ERR(znode)) { | |
2242 | err = PTR_ERR(znode); | |
2243 | goto out_unlock; | |
2244 | } | |
1e51764a AB |
2245 | } |
2246 | zbr = &znode->zbranch[n]; | |
2247 | lnc_free(zbr); | |
2248 | err = ubifs_add_dirt(c, zbr->lnum, | |
2249 | zbr->len); | |
2250 | if (err) | |
2251 | goto out_unlock; | |
2252 | zbr->lnum = lnum; | |
2253 | zbr->offs = offs; | |
2254 | zbr->len = len; | |
2255 | } | |
2256 | } | |
2257 | } | |
2258 | ||
2259 | if (!found) | |
2260 | err = ubifs_add_dirt(c, lnum, len); | |
2261 | ||
2262 | if (!err) | |
2263 | err = dbg_check_tnc(c, 0); | |
2264 | ||
2265 | out_unlock: | |
2266 | mutex_unlock(&c->tnc_mutex); | |
2267 | return err; | |
2268 | } | |
2269 | ||
2270 | /** | |
2271 | * ubifs_tnc_add_nm - add a "hashed" node to TNC. | |
2272 | * @c: UBIFS file-system description object | |
2273 | * @key: key to add | |
2274 | * @lnum: LEB number of node | |
2275 | * @offs: node offset | |
2276 | * @len: node length | |
2277 | * @nm: node name | |
2278 | * | |
2279 | * This is the same as 'ubifs_tnc_add()' but it should be used with keys which | |
2280 | * may have collisions, like directory entry keys. | |
2281 | */ | |
2282 | int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key, | |
2283 | int lnum, int offs, int len, const struct qstr *nm) | |
2284 | { | |
2285 | int found, n, err = 0; | |
2286 | struct ubifs_znode *znode; | |
2287 | ||
2288 | mutex_lock(&c->tnc_mutex); | |
515315a1 AB |
2289 | dbg_tnck(key, "LEB %d:%d, name '%.*s', key ", |
2290 | lnum, offs, nm->len, nm->name); | |
1e51764a AB |
2291 | found = lookup_level0_dirty(c, key, &znode, &n); |
2292 | if (found < 0) { | |
2293 | err = found; | |
2294 | goto out_unlock; | |
2295 | } | |
2296 | ||
2297 | if (found == 1) { | |
2298 | if (c->replaying) | |
2299 | found = fallible_resolve_collision(c, key, &znode, &n, | |
2300 | nm, 1); | |
2301 | else | |
2302 | found = resolve_collision(c, key, &znode, &n, nm); | |
2303 | dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n); | |
2304 | if (found < 0) { | |
2305 | err = found; | |
2306 | goto out_unlock; | |
2307 | } | |
2308 | ||
2309 | /* Ensure the znode is dirtied */ | |
2310 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2311 | znode = dirty_cow_bottom_up(c, znode); |
2312 | if (IS_ERR(znode)) { | |
2313 | err = PTR_ERR(znode); | |
2314 | goto out_unlock; | |
2315 | } | |
1e51764a AB |
2316 | } |
2317 | ||
2318 | if (found == 1) { | |
2319 | struct ubifs_zbranch *zbr = &znode->zbranch[n]; | |
2320 | ||
2321 | lnc_free(zbr); | |
2322 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2323 | zbr->lnum = lnum; | |
2324 | zbr->offs = offs; | |
2325 | zbr->len = len; | |
2326 | goto out_unlock; | |
2327 | } | |
2328 | } | |
2329 | ||
2330 | if (!found) { | |
2331 | struct ubifs_zbranch zbr; | |
2332 | ||
2333 | zbr.znode = NULL; | |
2334 | zbr.lnum = lnum; | |
2335 | zbr.offs = offs; | |
2336 | zbr.len = len; | |
2337 | key_copy(c, key, &zbr.key); | |
2338 | err = tnc_insert(c, znode, &zbr, n + 1); | |
2339 | if (err) | |
2340 | goto out_unlock; | |
2341 | if (c->replaying) { | |
2342 | /* | |
2343 | * We did not find it in the index so there may be a | |
2344 | * dangling branch still in the index. So we remove it | |
2345 | * by passing 'ubifs_tnc_remove_nm()' the same key but | |
2346 | * an unmatchable name. | |
2347 | */ | |
26fe5750 | 2348 | struct qstr noname = { .name = "" }; |
1e51764a AB |
2349 | |
2350 | err = dbg_check_tnc(c, 0); | |
2351 | mutex_unlock(&c->tnc_mutex); | |
2352 | if (err) | |
2353 | return err; | |
2354 | return ubifs_tnc_remove_nm(c, key, &noname); | |
2355 | } | |
2356 | } | |
2357 | ||
2358 | out_unlock: | |
2359 | if (!err) | |
2360 | err = dbg_check_tnc(c, 0); | |
2361 | mutex_unlock(&c->tnc_mutex); | |
2362 | return err; | |
2363 | } | |
2364 | ||
2365 | /** | |
2366 | * tnc_delete - delete a znode form TNC. | |
2367 | * @c: UBIFS file-system description object | |
2368 | * @znode: znode to delete from | |
2369 | * @n: zbranch slot number to delete | |
2370 | * | |
2371 | * This function deletes a leaf node from @n-th slot of @znode. Returns zero in | |
2372 | * case of success and a negative error code in case of failure. | |
2373 | */ | |
2374 | static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n) | |
2375 | { | |
2376 | struct ubifs_zbranch *zbr; | |
2377 | struct ubifs_znode *zp; | |
2378 | int i, err; | |
2379 | ||
2380 | /* Delete without merge for now */ | |
2381 | ubifs_assert(znode->level == 0); | |
2382 | ubifs_assert(n >= 0 && n < c->fanout); | |
515315a1 | 2383 | dbg_tnck(&znode->zbranch[n].key, "deleting key "); |
1e51764a AB |
2384 | |
2385 | zbr = &znode->zbranch[n]; | |
2386 | lnc_free(zbr); | |
2387 | ||
2388 | err = ubifs_add_dirt(c, zbr->lnum, zbr->len); | |
2389 | if (err) { | |
edf6be24 | 2390 | ubifs_dump_znode(c, znode); |
1e51764a AB |
2391 | return err; |
2392 | } | |
2393 | ||
2394 | /* We do not "gap" zbranch slots */ | |
2395 | for (i = n; i < znode->child_cnt - 1; i++) | |
2396 | znode->zbranch[i] = znode->zbranch[i + 1]; | |
2397 | znode->child_cnt -= 1; | |
2398 | ||
2399 | if (znode->child_cnt > 0) | |
2400 | return 0; | |
2401 | ||
2402 | /* | |
2403 | * This was the last zbranch, we have to delete this znode from the | |
2404 | * parent. | |
2405 | */ | |
2406 | ||
2407 | do { | |
f42eed7c | 2408 | ubifs_assert(!ubifs_zn_obsolete(znode)); |
1e51764a AB |
2409 | ubifs_assert(ubifs_zn_dirty(znode)); |
2410 | ||
2411 | zp = znode->parent; | |
2412 | n = znode->iip; | |
2413 | ||
2414 | atomic_long_dec(&c->dirty_zn_cnt); | |
2415 | ||
2416 | err = insert_old_idx_znode(c, znode); | |
2417 | if (err) | |
2418 | return err; | |
2419 | ||
2420 | if (znode->cnext) { | |
2421 | __set_bit(OBSOLETE_ZNODE, &znode->flags); | |
2422 | atomic_long_inc(&c->clean_zn_cnt); | |
2423 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
2424 | } else | |
2425 | kfree(znode); | |
2426 | znode = zp; | |
2427 | } while (znode->child_cnt == 1); /* while removing last child */ | |
2428 | ||
2429 | /* Remove from znode, entry n - 1 */ | |
2430 | znode->child_cnt -= 1; | |
2431 | ubifs_assert(znode->level != 0); | |
2432 | for (i = n; i < znode->child_cnt; i++) { | |
2433 | znode->zbranch[i] = znode->zbranch[i + 1]; | |
2434 | if (znode->zbranch[i].znode) | |
2435 | znode->zbranch[i].znode->iip = i; | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * If this is the root and it has only 1 child then | |
2440 | * collapse the tree. | |
2441 | */ | |
2442 | if (!znode->parent) { | |
2443 | while (znode->child_cnt == 1 && znode->level != 0) { | |
2444 | zp = znode; | |
2445 | zbr = &znode->zbranch[0]; | |
2446 | znode = get_znode(c, znode, 0); | |
2447 | if (IS_ERR(znode)) | |
2448 | return PTR_ERR(znode); | |
2449 | znode = dirty_cow_znode(c, zbr); | |
2450 | if (IS_ERR(znode)) | |
2451 | return PTR_ERR(znode); | |
2452 | znode->parent = NULL; | |
2453 | znode->iip = 0; | |
2454 | if (c->zroot.len) { | |
2455 | err = insert_old_idx(c, c->zroot.lnum, | |
2456 | c->zroot.offs); | |
2457 | if (err) | |
2458 | return err; | |
2459 | } | |
2460 | c->zroot.lnum = zbr->lnum; | |
2461 | c->zroot.offs = zbr->offs; | |
2462 | c->zroot.len = zbr->len; | |
2463 | c->zroot.znode = znode; | |
f42eed7c AB |
2464 | ubifs_assert(!ubifs_zn_obsolete(zp)); |
2465 | ubifs_assert(ubifs_zn_dirty(zp)); | |
1e51764a AB |
2466 | atomic_long_dec(&c->dirty_zn_cnt); |
2467 | ||
2468 | if (zp->cnext) { | |
2469 | __set_bit(OBSOLETE_ZNODE, &zp->flags); | |
2470 | atomic_long_inc(&c->clean_zn_cnt); | |
2471 | atomic_long_inc(&ubifs_clean_zn_cnt); | |
2472 | } else | |
2473 | kfree(zp); | |
2474 | } | |
2475 | } | |
2476 | ||
2477 | return 0; | |
2478 | } | |
2479 | ||
2480 | /** | |
2481 | * ubifs_tnc_remove - remove an index entry of a node. | |
2482 | * @c: UBIFS file-system description object | |
2483 | * @key: key of node | |
2484 | * | |
2485 | * Returns %0 on success or negative error code on failure. | |
2486 | */ | |
2487 | int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key) | |
2488 | { | |
2489 | int found, n, err = 0; | |
2490 | struct ubifs_znode *znode; | |
2491 | ||
2492 | mutex_lock(&c->tnc_mutex); | |
515315a1 | 2493 | dbg_tnck(key, "key "); |
1e51764a AB |
2494 | found = lookup_level0_dirty(c, key, &znode, &n); |
2495 | if (found < 0) { | |
2496 | err = found; | |
2497 | goto out_unlock; | |
2498 | } | |
2499 | if (found == 1) | |
2500 | err = tnc_delete(c, znode, n); | |
2501 | if (!err) | |
2502 | err = dbg_check_tnc(c, 0); | |
2503 | ||
2504 | out_unlock: | |
2505 | mutex_unlock(&c->tnc_mutex); | |
2506 | return err; | |
2507 | } | |
2508 | ||
2509 | /** | |
2510 | * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node. | |
2511 | * @c: UBIFS file-system description object | |
2512 | * @key: key of node | |
2513 | * @nm: directory entry name | |
2514 | * | |
2515 | * Returns %0 on success or negative error code on failure. | |
2516 | */ | |
2517 | int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key, | |
2518 | const struct qstr *nm) | |
2519 | { | |
2520 | int n, err; | |
2521 | struct ubifs_znode *znode; | |
2522 | ||
2523 | mutex_lock(&c->tnc_mutex); | |
515315a1 | 2524 | dbg_tnck(key, "%.*s, key ", nm->len, nm->name); |
1e51764a AB |
2525 | err = lookup_level0_dirty(c, key, &znode, &n); |
2526 | if (err < 0) | |
2527 | goto out_unlock; | |
2528 | ||
2529 | if (err) { | |
2530 | if (c->replaying) | |
2531 | err = fallible_resolve_collision(c, key, &znode, &n, | |
2532 | nm, 0); | |
2533 | else | |
2534 | err = resolve_collision(c, key, &znode, &n, nm); | |
2535 | dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); | |
2536 | if (err < 0) | |
2537 | goto out_unlock; | |
2538 | if (err) { | |
2539 | /* Ensure the znode is dirtied */ | |
2540 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
c4361570 AB |
2541 | znode = dirty_cow_bottom_up(c, znode); |
2542 | if (IS_ERR(znode)) { | |
2543 | err = PTR_ERR(znode); | |
2544 | goto out_unlock; | |
2545 | } | |
1e51764a AB |
2546 | } |
2547 | err = tnc_delete(c, znode, n); | |
2548 | } | |
2549 | } | |
2550 | ||
2551 | out_unlock: | |
2552 | if (!err) | |
2553 | err = dbg_check_tnc(c, 0); | |
2554 | mutex_unlock(&c->tnc_mutex); | |
2555 | return err; | |
2556 | } | |
2557 | ||
2558 | /** | |
2559 | * key_in_range - determine if a key falls within a range of keys. | |
2560 | * @c: UBIFS file-system description object | |
2561 | * @key: key to check | |
2562 | * @from_key: lowest key in range | |
2563 | * @to_key: highest key in range | |
2564 | * | |
2565 | * This function returns %1 if the key is in range and %0 otherwise. | |
2566 | */ | |
2567 | static int key_in_range(struct ubifs_info *c, union ubifs_key *key, | |
2568 | union ubifs_key *from_key, union ubifs_key *to_key) | |
2569 | { | |
2570 | if (keys_cmp(c, key, from_key) < 0) | |
2571 | return 0; | |
2572 | if (keys_cmp(c, key, to_key) > 0) | |
2573 | return 0; | |
2574 | return 1; | |
2575 | } | |
2576 | ||
2577 | /** | |
2578 | * ubifs_tnc_remove_range - remove index entries in range. | |
2579 | * @c: UBIFS file-system description object | |
2580 | * @from_key: lowest key to remove | |
2581 | * @to_key: highest key to remove | |
2582 | * | |
2583 | * This function removes index entries starting at @from_key and ending at | |
2584 | * @to_key. This function returns zero in case of success and a negative error | |
2585 | * code in case of failure. | |
2586 | */ | |
2587 | int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key, | |
2588 | union ubifs_key *to_key) | |
2589 | { | |
2590 | int i, n, k, err = 0; | |
2591 | struct ubifs_znode *znode; | |
2592 | union ubifs_key *key; | |
2593 | ||
2594 | mutex_lock(&c->tnc_mutex); | |
2595 | while (1) { | |
2596 | /* Find first level 0 znode that contains keys to remove */ | |
2597 | err = ubifs_lookup_level0(c, from_key, &znode, &n); | |
2598 | if (err < 0) | |
2599 | goto out_unlock; | |
2600 | ||
2601 | if (err) | |
2602 | key = from_key; | |
2603 | else { | |
2604 | err = tnc_next(c, &znode, &n); | |
2605 | if (err == -ENOENT) { | |
2606 | err = 0; | |
2607 | goto out_unlock; | |
2608 | } | |
2609 | if (err < 0) | |
2610 | goto out_unlock; | |
2611 | key = &znode->zbranch[n].key; | |
2612 | if (!key_in_range(c, key, from_key, to_key)) { | |
2613 | err = 0; | |
2614 | goto out_unlock; | |
2615 | } | |
2616 | } | |
2617 | ||
2618 | /* Ensure the znode is dirtied */ | |
2619 | if (znode->cnext || !ubifs_zn_dirty(znode)) { | |
f92b9826 AB |
2620 | znode = dirty_cow_bottom_up(c, znode); |
2621 | if (IS_ERR(znode)) { | |
2622 | err = PTR_ERR(znode); | |
2623 | goto out_unlock; | |
2624 | } | |
1e51764a AB |
2625 | } |
2626 | ||
2627 | /* Remove all keys in range except the first */ | |
2628 | for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) { | |
2629 | key = &znode->zbranch[i].key; | |
2630 | if (!key_in_range(c, key, from_key, to_key)) | |
2631 | break; | |
2632 | lnc_free(&znode->zbranch[i]); | |
2633 | err = ubifs_add_dirt(c, znode->zbranch[i].lnum, | |
2634 | znode->zbranch[i].len); | |
2635 | if (err) { | |
edf6be24 | 2636 | ubifs_dump_znode(c, znode); |
1e51764a AB |
2637 | goto out_unlock; |
2638 | } | |
515315a1 | 2639 | dbg_tnck(key, "removing key "); |
1e51764a AB |
2640 | } |
2641 | if (k) { | |
2642 | for (i = n + 1 + k; i < znode->child_cnt; i++) | |
2643 | znode->zbranch[i - k] = znode->zbranch[i]; | |
2644 | znode->child_cnt -= k; | |
2645 | } | |
2646 | ||
2647 | /* Now delete the first */ | |
2648 | err = tnc_delete(c, znode, n); | |
2649 | if (err) | |
2650 | goto out_unlock; | |
2651 | } | |
2652 | ||
2653 | out_unlock: | |
2654 | if (!err) | |
2655 | err = dbg_check_tnc(c, 0); | |
2656 | mutex_unlock(&c->tnc_mutex); | |
2657 | return err; | |
2658 | } | |
2659 | ||
2660 | /** | |
2661 | * ubifs_tnc_remove_ino - remove an inode from TNC. | |
2662 | * @c: UBIFS file-system description object | |
2663 | * @inum: inode number to remove | |
2664 | * | |
2665 | * This function remove inode @inum and all the extended attributes associated | |
2666 | * with the anode from TNC and returns zero in case of success or a negative | |
2667 | * error code in case of failure. | |
2668 | */ | |
2669 | int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum) | |
2670 | { | |
2671 | union ubifs_key key1, key2; | |
2672 | struct ubifs_dent_node *xent, *pxent = NULL; | |
2673 | struct qstr nm = { .name = NULL }; | |
2674 | ||
e84461ad | 2675 | dbg_tnc("ino %lu", (unsigned long)inum); |
1e51764a AB |
2676 | |
2677 | /* | |
2678 | * Walk all extended attribute entries and remove them together with | |
2679 | * corresponding extended attribute inodes. | |
2680 | */ | |
2681 | lowest_xent_key(c, &key1, inum); | |
2682 | while (1) { | |
2683 | ino_t xattr_inum; | |
2684 | int err; | |
2685 | ||
2686 | xent = ubifs_tnc_next_ent(c, &key1, &nm); | |
2687 | if (IS_ERR(xent)) { | |
2688 | err = PTR_ERR(xent); | |
2689 | if (err == -ENOENT) | |
2690 | break; | |
2691 | return err; | |
2692 | } | |
2693 | ||
2694 | xattr_inum = le64_to_cpu(xent->inum); | |
e84461ad AB |
2695 | dbg_tnc("xent '%s', ino %lu", xent->name, |
2696 | (unsigned long)xattr_inum); | |
1e51764a AB |
2697 | |
2698 | nm.name = xent->name; | |
2699 | nm.len = le16_to_cpu(xent->nlen); | |
2700 | err = ubifs_tnc_remove_nm(c, &key1, &nm); | |
2701 | if (err) { | |
2702 | kfree(xent); | |
2703 | return err; | |
2704 | } | |
2705 | ||
2706 | lowest_ino_key(c, &key1, xattr_inum); | |
2707 | highest_ino_key(c, &key2, xattr_inum); | |
2708 | err = ubifs_tnc_remove_range(c, &key1, &key2); | |
2709 | if (err) { | |
2710 | kfree(xent); | |
2711 | return err; | |
2712 | } | |
2713 | ||
2714 | kfree(pxent); | |
2715 | pxent = xent; | |
2716 | key_read(c, &xent->key, &key1); | |
2717 | } | |
2718 | ||
2719 | kfree(pxent); | |
2720 | lowest_ino_key(c, &key1, inum); | |
2721 | highest_ino_key(c, &key2, inum); | |
2722 | ||
2723 | return ubifs_tnc_remove_range(c, &key1, &key2); | |
2724 | } | |
2725 | ||
2726 | /** | |
2727 | * ubifs_tnc_next_ent - walk directory or extended attribute entries. | |
2728 | * @c: UBIFS file-system description object | |
2729 | * @key: key of last entry | |
2730 | * @nm: name of last entry found or %NULL | |
2731 | * | |
2732 | * This function finds and reads the next directory or extended attribute entry | |
2733 | * after the given key (@key) if there is one. @nm is used to resolve | |
2734 | * collisions. | |
2735 | * | |
2736 | * If the name of the current entry is not known and only the key is known, | |
2737 | * @nm->name has to be %NULL. In this case the semantics of this function is a | |
2738 | * little bit different and it returns the entry corresponding to this key, not | |
2739 | * the next one. If the key was not found, the closest "right" entry is | |
2740 | * returned. | |
2741 | * | |
2742 | * If the fist entry has to be found, @key has to contain the lowest possible | |
2743 | * key value for this inode and @name has to be %NULL. | |
2744 | * | |
2745 | * This function returns the found directory or extended attribute entry node | |
2746 | * in case of success, %-ENOENT is returned if no entry was found, and a | |
2747 | * negative error code is returned in case of failure. | |
2748 | */ | |
2749 | struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c, | |
2750 | union ubifs_key *key, | |
2751 | const struct qstr *nm) | |
2752 | { | |
2753 | int n, err, type = key_type(c, key); | |
2754 | struct ubifs_znode *znode; | |
2755 | struct ubifs_dent_node *dent; | |
2756 | struct ubifs_zbranch *zbr; | |
2757 | union ubifs_key *dkey; | |
2758 | ||
515315a1 | 2759 | dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)"); |
1e51764a AB |
2760 | ubifs_assert(is_hash_key(c, key)); |
2761 | ||
2762 | mutex_lock(&c->tnc_mutex); | |
2763 | err = ubifs_lookup_level0(c, key, &znode, &n); | |
2764 | if (unlikely(err < 0)) | |
2765 | goto out_unlock; | |
2766 | ||
2767 | if (nm->name) { | |
2768 | if (err) { | |
2769 | /* Handle collisions */ | |
2770 | err = resolve_collision(c, key, &znode, &n, nm); | |
2771 | dbg_tnc("rc returned %d, znode %p, n %d", | |
2772 | err, znode, n); | |
2773 | if (unlikely(err < 0)) | |
2774 | goto out_unlock; | |
2775 | } | |
2776 | ||
2777 | /* Now find next entry */ | |
2778 | err = tnc_next(c, &znode, &n); | |
2779 | if (unlikely(err)) | |
2780 | goto out_unlock; | |
2781 | } else { | |
2782 | /* | |
2783 | * The full name of the entry was not given, in which case the | |
2784 | * behavior of this function is a little different and it | |
2785 | * returns current entry, not the next one. | |
2786 | */ | |
2787 | if (!err) { | |
2788 | /* | |
2789 | * However, the given key does not exist in the TNC | |
2790 | * tree and @znode/@n variables contain the closest | |
2791 | * "preceding" element. Switch to the next one. | |
2792 | */ | |
2793 | err = tnc_next(c, &znode, &n); | |
2794 | if (err) | |
2795 | goto out_unlock; | |
2796 | } | |
2797 | } | |
2798 | ||
2799 | zbr = &znode->zbranch[n]; | |
2800 | dent = kmalloc(zbr->len, GFP_NOFS); | |
2801 | if (unlikely(!dent)) { | |
2802 | err = -ENOMEM; | |
2803 | goto out_unlock; | |
2804 | } | |
2805 | ||
2806 | /* | |
2807 | * The above 'tnc_next()' call could lead us to the next inode, check | |
2808 | * this. | |
2809 | */ | |
2810 | dkey = &zbr->key; | |
2811 | if (key_inum(c, dkey) != key_inum(c, key) || | |
2812 | key_type(c, dkey) != type) { | |
2813 | err = -ENOENT; | |
2814 | goto out_free; | |
2815 | } | |
2816 | ||
2817 | err = tnc_read_node_nm(c, zbr, dent); | |
2818 | if (unlikely(err)) | |
2819 | goto out_free; | |
2820 | ||
2821 | mutex_unlock(&c->tnc_mutex); | |
2822 | return dent; | |
2823 | ||
2824 | out_free: | |
2825 | kfree(dent); | |
2826 | out_unlock: | |
2827 | mutex_unlock(&c->tnc_mutex); | |
2828 | return ERR_PTR(err); | |
2829 | } | |
2830 | ||
2831 | /** | |
2832 | * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit. | |
2833 | * @c: UBIFS file-system description object | |
2834 | * | |
2835 | * Destroy left-over obsolete znodes from a failed commit. | |
2836 | */ | |
2837 | static void tnc_destroy_cnext(struct ubifs_info *c) | |
2838 | { | |
2839 | struct ubifs_znode *cnext; | |
2840 | ||
2841 | if (!c->cnext) | |
2842 | return; | |
2843 | ubifs_assert(c->cmt_state == COMMIT_BROKEN); | |
2844 | cnext = c->cnext; | |
2845 | do { | |
2846 | struct ubifs_znode *znode = cnext; | |
2847 | ||
2848 | cnext = cnext->cnext; | |
f42eed7c | 2849 | if (ubifs_zn_obsolete(znode)) |
1e51764a AB |
2850 | kfree(znode); |
2851 | } while (cnext && cnext != c->cnext); | |
2852 | } | |
2853 | ||
2854 | /** | |
2855 | * ubifs_tnc_close - close TNC subsystem and free all related resources. | |
2856 | * @c: UBIFS file-system description object | |
2857 | */ | |
2858 | void ubifs_tnc_close(struct ubifs_info *c) | |
2859 | { | |
1e51764a AB |
2860 | tnc_destroy_cnext(c); |
2861 | if (c->zroot.znode) { | |
380347e9 | 2862 | long n, freed; |
83707237 | 2863 | |
83707237 | 2864 | n = atomic_long_read(&c->clean_zn_cnt); |
380347e9 | 2865 | freed = ubifs_destroy_tnc_subtree(c->zroot.znode); |
2866 | ubifs_assert(freed == n); | |
83707237 | 2867 | atomic_long_sub(n, &ubifs_clean_zn_cnt); |
1e51764a AB |
2868 | } |
2869 | kfree(c->gap_lebs); | |
2870 | kfree(c->ilebs); | |
2871 | destroy_old_idx(c); | |
2872 | } | |
2873 | ||
2874 | /** | |
2875 | * left_znode - get the znode to the left. | |
2876 | * @c: UBIFS file-system description object | |
2877 | * @znode: znode | |
2878 | * | |
2879 | * This function returns a pointer to the znode to the left of @znode or NULL if | |
2880 | * there is not one. A negative error code is returned on failure. | |
2881 | */ | |
2882 | static struct ubifs_znode *left_znode(struct ubifs_info *c, | |
2883 | struct ubifs_znode *znode) | |
2884 | { | |
2885 | int level = znode->level; | |
2886 | ||
2887 | while (1) { | |
2888 | int n = znode->iip - 1; | |
2889 | ||
2890 | /* Go up until we can go left */ | |
2891 | znode = znode->parent; | |
2892 | if (!znode) | |
2893 | return NULL; | |
2894 | if (n >= 0) { | |
2895 | /* Now go down the rightmost branch to 'level' */ | |
2896 | znode = get_znode(c, znode, n); | |
2897 | if (IS_ERR(znode)) | |
2898 | return znode; | |
2899 | while (znode->level != level) { | |
2900 | n = znode->child_cnt - 1; | |
2901 | znode = get_znode(c, znode, n); | |
2902 | if (IS_ERR(znode)) | |
2903 | return znode; | |
2904 | } | |
2905 | break; | |
2906 | } | |
2907 | } | |
2908 | return znode; | |
2909 | } | |
2910 | ||
2911 | /** | |
2912 | * right_znode - get the znode to the right. | |
2913 | * @c: UBIFS file-system description object | |
2914 | * @znode: znode | |
2915 | * | |
2916 | * This function returns a pointer to the znode to the right of @znode or NULL | |
2917 | * if there is not one. A negative error code is returned on failure. | |
2918 | */ | |
2919 | static struct ubifs_znode *right_znode(struct ubifs_info *c, | |
2920 | struct ubifs_znode *znode) | |
2921 | { | |
2922 | int level = znode->level; | |
2923 | ||
2924 | while (1) { | |
2925 | int n = znode->iip + 1; | |
2926 | ||
2927 | /* Go up until we can go right */ | |
2928 | znode = znode->parent; | |
2929 | if (!znode) | |
2930 | return NULL; | |
2931 | if (n < znode->child_cnt) { | |
2932 | /* Now go down the leftmost branch to 'level' */ | |
2933 | znode = get_znode(c, znode, n); | |
2934 | if (IS_ERR(znode)) | |
2935 | return znode; | |
2936 | while (znode->level != level) { | |
2937 | znode = get_znode(c, znode, 0); | |
2938 | if (IS_ERR(znode)) | |
2939 | return znode; | |
2940 | } | |
2941 | break; | |
2942 | } | |
2943 | } | |
2944 | return znode; | |
2945 | } | |
2946 | ||
2947 | /** | |
2948 | * lookup_znode - find a particular indexing node from TNC. | |
2949 | * @c: UBIFS file-system description object | |
2950 | * @key: index node key to lookup | |
2951 | * @level: index node level | |
2952 | * @lnum: index node LEB number | |
2953 | * @offs: index node offset | |
2954 | * | |
2955 | * This function searches an indexing node by its first key @key and its | |
2956 | * address @lnum:@offs. It looks up the indexing tree by pulling all indexing | |
ba2f48f7 | 2957 | * nodes it traverses to TNC. This function is called for indexing nodes which |
1e51764a AB |
2958 | * were found on the media by scanning, for example when garbage-collecting or |
2959 | * when doing in-the-gaps commit. This means that the indexing node which is | |
2960 | * looked for does not have to have exactly the same leftmost key @key, because | |
2961 | * the leftmost key may have been changed, in which case TNC will contain a | |
2962 | * dirty znode which still refers the same @lnum:@offs. This function is clever | |
2963 | * enough to recognize such indexing nodes. | |
2964 | * | |
2965 | * Note, if a znode was deleted or changed too much, then this function will | |
2966 | * not find it. For situations like this UBIFS has the old index RB-tree | |
2967 | * (indexed by @lnum:@offs). | |
2968 | * | |
2969 | * This function returns a pointer to the znode found or %NULL if it is not | |
2970 | * found. A negative error code is returned on failure. | |
2971 | */ | |
2972 | static struct ubifs_znode *lookup_znode(struct ubifs_info *c, | |
2973 | union ubifs_key *key, int level, | |
2974 | int lnum, int offs) | |
2975 | { | |
2976 | struct ubifs_znode *znode, *zn; | |
2977 | int n, nn; | |
2978 | ||
ba2f48f7 AB |
2979 | ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY); |
2980 | ||
1e51764a AB |
2981 | /* |
2982 | * The arguments have probably been read off flash, so don't assume | |
2983 | * they are valid. | |
2984 | */ | |
2985 | if (level < 0) | |
2986 | return ERR_PTR(-EINVAL); | |
2987 | ||
2988 | /* Get the root znode */ | |
2989 | znode = c->zroot.znode; | |
2990 | if (!znode) { | |
2991 | znode = ubifs_load_znode(c, &c->zroot, NULL, 0); | |
2992 | if (IS_ERR(znode)) | |
2993 | return znode; | |
2994 | } | |
2995 | /* Check if it is the one we are looking for */ | |
2996 | if (c->zroot.lnum == lnum && c->zroot.offs == offs) | |
2997 | return znode; | |
2998 | /* Descend to the parent level i.e. (level + 1) */ | |
2999 | if (level >= znode->level) | |
3000 | return NULL; | |
3001 | while (1) { | |
3002 | ubifs_search_zbranch(c, znode, key, &n); | |
3003 | if (n < 0) { | |
3004 | /* | |
3005 | * We reached a znode where the leftmost key is greater | |
3006 | * than the key we are searching for. This is the same | |
3007 | * situation as the one described in a huge comment at | |
3008 | * the end of the 'ubifs_lookup_level0()' function. And | |
3009 | * for exactly the same reasons we have to try to look | |
3010 | * left before giving up. | |
3011 | */ | |
3012 | znode = left_znode(c, znode); | |
3013 | if (!znode) | |
3014 | return NULL; | |
3015 | if (IS_ERR(znode)) | |
3016 | return znode; | |
3017 | ubifs_search_zbranch(c, znode, key, &n); | |
3018 | ubifs_assert(n >= 0); | |
3019 | } | |
3020 | if (znode->level == level + 1) | |
3021 | break; | |
3022 | znode = get_znode(c, znode, n); | |
3023 | if (IS_ERR(znode)) | |
3024 | return znode; | |
3025 | } | |
3026 | /* Check if the child is the one we are looking for */ | |
3027 | if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs) | |
3028 | return get_znode(c, znode, n); | |
3029 | /* If the key is unique, there is nowhere else to look */ | |
3030 | if (!is_hash_key(c, key)) | |
3031 | return NULL; | |
3032 | /* | |
3033 | * The key is not unique and so may be also in the znodes to either | |
3034 | * side. | |
3035 | */ | |
3036 | zn = znode; | |
3037 | nn = n; | |
3038 | /* Look left */ | |
3039 | while (1) { | |
3040 | /* Move one branch to the left */ | |
3041 | if (n) | |
3042 | n -= 1; | |
3043 | else { | |
3044 | znode = left_znode(c, znode); | |
3045 | if (!znode) | |
3046 | break; | |
3047 | if (IS_ERR(znode)) | |
3048 | return znode; | |
3049 | n = znode->child_cnt - 1; | |
3050 | } | |
3051 | /* Check it */ | |
3052 | if (znode->zbranch[n].lnum == lnum && | |
3053 | znode->zbranch[n].offs == offs) | |
3054 | return get_znode(c, znode, n); | |
3055 | /* Stop if the key is less than the one we are looking for */ | |
3056 | if (keys_cmp(c, &znode->zbranch[n].key, key) < 0) | |
3057 | break; | |
3058 | } | |
3059 | /* Back to the middle */ | |
3060 | znode = zn; | |
3061 | n = nn; | |
3062 | /* Look right */ | |
3063 | while (1) { | |
3064 | /* Move one branch to the right */ | |
3065 | if (++n >= znode->child_cnt) { | |
3066 | znode = right_znode(c, znode); | |
3067 | if (!znode) | |
3068 | break; | |
3069 | if (IS_ERR(znode)) | |
3070 | return znode; | |
3071 | n = 0; | |
3072 | } | |
3073 | /* Check it */ | |
3074 | if (znode->zbranch[n].lnum == lnum && | |
3075 | znode->zbranch[n].offs == offs) | |
3076 | return get_znode(c, znode, n); | |
3077 | /* Stop if the key is greater than the one we are looking for */ | |
3078 | if (keys_cmp(c, &znode->zbranch[n].key, key) > 0) | |
3079 | break; | |
3080 | } | |
3081 | return NULL; | |
3082 | } | |
3083 | ||
3084 | /** | |
3085 | * is_idx_node_in_tnc - determine if an index node is in the TNC. | |
3086 | * @c: UBIFS file-system description object | |
3087 | * @key: key of index node | |
3088 | * @level: index node level | |
3089 | * @lnum: LEB number of index node | |
3090 | * @offs: offset of index node | |
3091 | * | |
3092 | * This function returns %0 if the index node is not referred to in the TNC, %1 | |
3093 | * if the index node is referred to in the TNC and the corresponding znode is | |
3094 | * dirty, %2 if an index node is referred to in the TNC and the corresponding | |
3095 | * znode is clean, and a negative error code in case of failure. | |
3096 | * | |
3097 | * Note, the @key argument has to be the key of the first child. Also note, | |
3098 | * this function relies on the fact that 0:0 is never a valid LEB number and | |
3099 | * offset for a main-area node. | |
3100 | */ | |
3101 | int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level, | |
3102 | int lnum, int offs) | |
3103 | { | |
3104 | struct ubifs_znode *znode; | |
3105 | ||
3106 | znode = lookup_znode(c, key, level, lnum, offs); | |
3107 | if (!znode) | |
3108 | return 0; | |
3109 | if (IS_ERR(znode)) | |
3110 | return PTR_ERR(znode); | |
3111 | ||
3112 | return ubifs_zn_dirty(znode) ? 1 : 2; | |
3113 | } | |
3114 | ||
3115 | /** | |
3116 | * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC. | |
3117 | * @c: UBIFS file-system description object | |
3118 | * @key: node key | |
3119 | * @lnum: node LEB number | |
3120 | * @offs: node offset | |
3121 | * | |
3122 | * This function returns %1 if the node is referred to in the TNC, %0 if it is | |
3123 | * not, and a negative error code in case of failure. | |
3124 | * | |
3125 | * Note, this function relies on the fact that 0:0 is never a valid LEB number | |
3126 | * and offset for a main-area node. | |
3127 | */ | |
3128 | static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, | |
3129 | int lnum, int offs) | |
3130 | { | |
3131 | struct ubifs_zbranch *zbr; | |
3132 | struct ubifs_znode *znode, *zn; | |
3133 | int n, found, err, nn; | |
3134 | const int unique = !is_hash_key(c, key); | |
3135 | ||
3136 | found = ubifs_lookup_level0(c, key, &znode, &n); | |
3137 | if (found < 0) | |
3138 | return found; /* Error code */ | |
3139 | if (!found) | |
3140 | return 0; | |
3141 | zbr = &znode->zbranch[n]; | |
3142 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3143 | return 1; /* Found it */ | |
3144 | if (unique) | |
3145 | return 0; | |
3146 | /* | |
3147 | * Because the key is not unique, we have to look left | |
3148 | * and right as well | |
3149 | */ | |
3150 | zn = znode; | |
3151 | nn = n; | |
3152 | /* Look left */ | |
3153 | while (1) { | |
3154 | err = tnc_prev(c, &znode, &n); | |
3155 | if (err == -ENOENT) | |
3156 | break; | |
3157 | if (err) | |
3158 | return err; | |
3159 | if (keys_cmp(c, key, &znode->zbranch[n].key)) | |
3160 | break; | |
3161 | zbr = &znode->zbranch[n]; | |
3162 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3163 | return 1; /* Found it */ | |
3164 | } | |
3165 | /* Look right */ | |
3166 | znode = zn; | |
3167 | n = nn; | |
3168 | while (1) { | |
3169 | err = tnc_next(c, &znode, &n); | |
3170 | if (err) { | |
3171 | if (err == -ENOENT) | |
3172 | return 0; | |
3173 | return err; | |
3174 | } | |
3175 | if (keys_cmp(c, key, &znode->zbranch[n].key)) | |
3176 | break; | |
3177 | zbr = &znode->zbranch[n]; | |
3178 | if (lnum == zbr->lnum && offs == zbr->offs) | |
3179 | return 1; /* Found it */ | |
3180 | } | |
3181 | return 0; | |
3182 | } | |
3183 | ||
3184 | /** | |
3185 | * ubifs_tnc_has_node - determine whether a node is in the TNC. | |
3186 | * @c: UBIFS file-system description object | |
3187 | * @key: node key | |
3188 | * @level: index node level (if it is an index node) | |
3189 | * @lnum: node LEB number | |
3190 | * @offs: node offset | |
3191 | * @is_idx: non-zero if the node is an index node | |
3192 | * | |
3193 | * This function returns %1 if the node is in the TNC, %0 if it is not, and a | |
3194 | * negative error code in case of failure. For index nodes, @key has to be the | |
3195 | * key of the first child. An index node is considered to be in the TNC only if | |
3196 | * the corresponding znode is clean or has not been loaded. | |
3197 | */ | |
3198 | int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level, | |
3199 | int lnum, int offs, int is_idx) | |
3200 | { | |
3201 | int err; | |
3202 | ||
3203 | mutex_lock(&c->tnc_mutex); | |
3204 | if (is_idx) { | |
3205 | err = is_idx_node_in_tnc(c, key, level, lnum, offs); | |
3206 | if (err < 0) | |
3207 | goto out_unlock; | |
3208 | if (err == 1) | |
3209 | /* The index node was found but it was dirty */ | |
3210 | err = 0; | |
3211 | else if (err == 2) | |
3212 | /* The index node was found and it was clean */ | |
3213 | err = 1; | |
3214 | else | |
3215 | BUG_ON(err != 0); | |
3216 | } else | |
3217 | err = is_leaf_node_in_tnc(c, key, lnum, offs); | |
3218 | ||
3219 | out_unlock: | |
3220 | mutex_unlock(&c->tnc_mutex); | |
3221 | return err; | |
3222 | } | |
3223 | ||
3224 | /** | |
3225 | * ubifs_dirty_idx_node - dirty an index node. | |
3226 | * @c: UBIFS file-system description object | |
3227 | * @key: index node key | |
3228 | * @level: index node level | |
3229 | * @lnum: index node LEB number | |
3230 | * @offs: index node offset | |
3231 | * | |
3232 | * This function loads and dirties an index node so that it can be garbage | |
3233 | * collected. The @key argument has to be the key of the first child. This | |
3234 | * function relies on the fact that 0:0 is never a valid LEB number and offset | |
3235 | * for a main-area node. Returns %0 on success and a negative error code on | |
3236 | * failure. | |
3237 | */ | |
3238 | int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level, | |
3239 | int lnum, int offs) | |
3240 | { | |
3241 | struct ubifs_znode *znode; | |
3242 | int err = 0; | |
3243 | ||
3244 | mutex_lock(&c->tnc_mutex); | |
3245 | znode = lookup_znode(c, key, level, lnum, offs); | |
3246 | if (!znode) | |
3247 | goto out_unlock; | |
3248 | if (IS_ERR(znode)) { | |
3249 | err = PTR_ERR(znode); | |
3250 | goto out_unlock; | |
3251 | } | |
3252 | znode = dirty_cow_bottom_up(c, znode); | |
3253 | if (IS_ERR(znode)) { | |
3254 | err = PTR_ERR(znode); | |
3255 | goto out_unlock; | |
3256 | } | |
3257 | ||
3258 | out_unlock: | |
3259 | mutex_unlock(&c->tnc_mutex); | |
3260 | return err; | |
3261 | } | |
e3c3efc2 | 3262 | |
e3c3efc2 AB |
3263 | /** |
3264 | * dbg_check_inode_size - check if inode size is correct. | |
3265 | * @c: UBIFS file-system description object | |
3266 | * @inum: inode number | |
3267 | * @size: inode size | |
3268 | * | |
3269 | * This function makes sure that the inode size (@size) is correct and it does | |
3270 | * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL | |
3271 | * if it has a data page beyond @size, and other negative error code in case of | |
3272 | * other errors. | |
3273 | */ | |
3274 | int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode, | |
3275 | loff_t size) | |
3276 | { | |
3277 | int err, n; | |
3278 | union ubifs_key from_key, to_key, *key; | |
3279 | struct ubifs_znode *znode; | |
3280 | unsigned int block; | |
3281 | ||
3282 | if (!S_ISREG(inode->i_mode)) | |
3283 | return 0; | |
2b1844a8 | 3284 | if (!dbg_is_chk_gen(c)) |
e3c3efc2 AB |
3285 | return 0; |
3286 | ||
3287 | block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT; | |
3288 | data_key_init(c, &from_key, inode->i_ino, block); | |
3289 | highest_data_key(c, &to_key, inode->i_ino); | |
3290 | ||
3291 | mutex_lock(&c->tnc_mutex); | |
3292 | err = ubifs_lookup_level0(c, &from_key, &znode, &n); | |
3293 | if (err < 0) | |
3294 | goto out_unlock; | |
3295 | ||
3296 | if (err) { | |
e3c3efc2 AB |
3297 | key = &from_key; |
3298 | goto out_dump; | |
3299 | } | |
3300 | ||
3301 | err = tnc_next(c, &znode, &n); | |
3302 | if (err == -ENOENT) { | |
3303 | err = 0; | |
3304 | goto out_unlock; | |
3305 | } | |
3306 | if (err < 0) | |
3307 | goto out_unlock; | |
3308 | ||
3309 | ubifs_assert(err == 0); | |
3310 | key = &znode->zbranch[n].key; | |
3311 | if (!key_in_range(c, key, &from_key, &to_key)) | |
3312 | goto out_unlock; | |
3313 | ||
3314 | out_dump: | |
3315 | block = key_block(c, key); | |
235c362b | 3316 | ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld", |
515315a1 AB |
3317 | (unsigned long)inode->i_ino, size, |
3318 | ((loff_t)block) << UBIFS_BLOCK_SHIFT); | |
4315fb40 | 3319 | mutex_unlock(&c->tnc_mutex); |
edf6be24 | 3320 | ubifs_dump_inode(c, inode); |
7c46d0ae | 3321 | dump_stack(); |
4315fb40 | 3322 | return -EINVAL; |
e3c3efc2 AB |
3323 | |
3324 | out_unlock: | |
3325 | mutex_unlock(&c->tnc_mutex); | |
3326 | return err; | |
3327 | } |