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 contains journal replay code. It runs when the file-system is being | |
25 | * mounted and requires no locking. | |
26 | * | |
27 | * The larger is the journal, the longer it takes to scan it, so the longer it | |
28 | * takes to mount UBIFS. This is why the journal has limited size which may be | |
29 | * changed depending on the system requirements. But a larger journal gives | |
30 | * faster I/O speed because it writes the index less frequently. So this is a | |
31 | * trade-off. Also, the journal is indexed by the in-memory index (TNC), so the | |
32 | * larger is the journal, the more memory its index may consume. | |
33 | */ | |
34 | ||
35 | #include "ubifs.h" | |
36 | ||
1e51764a AB |
37 | /** |
38 | * struct replay_entry - replay tree entry. | |
39 | * @lnum: logical eraseblock number of the node | |
40 | * @offs: node offset | |
41 | * @len: node length | |
074bcb9b | 42 | * @deletion: non-zero if this entry corresponds to a node deletion |
1e51764a | 43 | * @sqnum: node sequence number |
1e51764a AB |
44 | * @rb: links the replay tree |
45 | * @key: node key | |
46 | * @nm: directory entry name | |
47 | * @old_size: truncation old size | |
48 | * @new_size: truncation new size | |
1e51764a AB |
49 | * |
50 | * UBIFS journal replay must compare node sequence numbers, which means it must | |
51 | * build a tree of node information to insert into the TNC. | |
52 | */ | |
53 | struct replay_entry { | |
54 | int lnum; | |
55 | int offs; | |
56 | int len; | |
074bcb9b | 57 | unsigned int deletion:1; |
1e51764a | 58 | unsigned long long sqnum; |
1e51764a AB |
59 | struct rb_node rb; |
60 | union ubifs_key key; | |
61 | union { | |
62 | struct qstr nm; | |
63 | struct { | |
64 | loff_t old_size; | |
65 | loff_t new_size; | |
66 | }; | |
1e51764a AB |
67 | }; |
68 | }; | |
69 | ||
70 | /** | |
71 | * struct bud_entry - entry in the list of buds to replay. | |
72 | * @list: next bud in the list | |
73 | * @bud: bud description object | |
1e51764a | 74 | * @sqnum: reference node sequence number |
af1dd412 AB |
75 | * @free: free bytes in the bud |
76 | * @dirty: dirty bytes in the bud | |
1e51764a AB |
77 | */ |
78 | struct bud_entry { | |
79 | struct list_head list; | |
80 | struct ubifs_bud *bud; | |
1e51764a | 81 | unsigned long long sqnum; |
af1dd412 AB |
82 | int free; |
83 | int dirty; | |
1e51764a AB |
84 | }; |
85 | ||
86 | /** | |
87 | * set_bud_lprops - set free and dirty space used by a bud. | |
88 | * @c: UBIFS file-system description object | |
074bcb9b AB |
89 | * @b: bud entry which describes the bud |
90 | * | |
91 | * This function makes sure the LEB properties of bud @b are set correctly | |
92 | * after the replay. Returns zero in case of success and a negative error code | |
93 | * in case of failure. | |
1e51764a | 94 | */ |
074bcb9b | 95 | static int set_bud_lprops(struct ubifs_info *c, struct bud_entry *b) |
1e51764a AB |
96 | { |
97 | const struct ubifs_lprops *lp; | |
98 | int err = 0, dirty; | |
99 | ||
100 | ubifs_get_lprops(c); | |
101 | ||
074bcb9b | 102 | lp = ubifs_lpt_lookup_dirty(c, b->bud->lnum); |
1e51764a AB |
103 | if (IS_ERR(lp)) { |
104 | err = PTR_ERR(lp); | |
105 | goto out; | |
106 | } | |
107 | ||
108 | dirty = lp->dirty; | |
074bcb9b | 109 | if (b->bud->start == 0 && (lp->free != c->leb_size || lp->dirty != 0)) { |
1e51764a AB |
110 | /* |
111 | * The LEB was added to the journal with a starting offset of | |
112 | * zero which means the LEB must have been empty. The LEB | |
074bcb9b AB |
113 | * property values should be @lp->free == @c->leb_size and |
114 | * @lp->dirty == 0, but that is not the case. The reason is that | |
7a9c3e39 AB |
115 | * the LEB had been garbage collected before it became the bud, |
116 | * and there was not commit inbetween. The garbage collector | |
117 | * resets the free and dirty space without recording it | |
118 | * anywhere except lprops, so if there was no commit then | |
119 | * lprops does not have that information. | |
1e51764a AB |
120 | * |
121 | * We do not need to adjust free space because the scan has told | |
122 | * us the exact value which is recorded in the replay entry as | |
074bcb9b | 123 | * @b->free. |
1e51764a AB |
124 | * |
125 | * However we do need to subtract from the dirty space the | |
126 | * amount of space that the garbage collector reclaimed, which | |
127 | * is the whole LEB minus the amount of space that was free. | |
128 | */ | |
074bcb9b | 129 | dbg_mnt("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum, |
1e51764a | 130 | lp->free, lp->dirty); |
074bcb9b | 131 | dbg_gc("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum, |
1e51764a AB |
132 | lp->free, lp->dirty); |
133 | dirty -= c->leb_size - lp->free; | |
134 | /* | |
135 | * If the replay order was perfect the dirty space would now be | |
7d4e9ccb | 136 | * zero. The order is not perfect because the journal heads |
6edbfafd | 137 | * race with each other. This is not a problem but is does mean |
1e51764a AB |
138 | * that the dirty space may temporarily exceed c->leb_size |
139 | * during the replay. | |
140 | */ | |
141 | if (dirty != 0) | |
142 | dbg_msg("LEB %d lp: %d free %d dirty " | |
074bcb9b AB |
143 | "replay: %d free %d dirty", b->bud->lnum, |
144 | lp->free, lp->dirty, b->free, b->dirty); | |
1e51764a | 145 | } |
074bcb9b | 146 | lp = ubifs_change_lp(c, lp, b->free, dirty + b->dirty, |
1e51764a AB |
147 | lp->flags | LPROPS_TAKEN, 0); |
148 | if (IS_ERR(lp)) { | |
149 | err = PTR_ERR(lp); | |
150 | goto out; | |
151 | } | |
52c6e6f9 AB |
152 | |
153 | /* Make sure the journal head points to the latest bud */ | |
074bcb9b AB |
154 | err = ubifs_wbuf_seek_nolock(&c->jheads[b->bud->jhead].wbuf, |
155 | b->bud->lnum, c->leb_size - b->free, | |
156 | UBI_SHORTTERM); | |
52c6e6f9 | 157 | |
1e51764a AB |
158 | out: |
159 | ubifs_release_lprops(c); | |
160 | return err; | |
161 | } | |
162 | ||
074bcb9b AB |
163 | /** |
164 | * set_buds_lprops - set free and dirty space for all replayed buds. | |
165 | * @c: UBIFS file-system description object | |
166 | * | |
167 | * This function sets LEB properties for all replayed buds. Returns zero in | |
168 | * case of success and a negative error code in case of failure. | |
169 | */ | |
170 | static int set_buds_lprops(struct ubifs_info *c) | |
171 | { | |
172 | struct bud_entry *b; | |
173 | int err; | |
174 | ||
175 | list_for_each_entry(b, &c->replay_buds, list) { | |
176 | err = set_bud_lprops(c, b); | |
177 | if (err) | |
178 | return err; | |
179 | } | |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
1e51764a AB |
184 | /** |
185 | * trun_remove_range - apply a replay entry for a truncation to the TNC. | |
186 | * @c: UBIFS file-system description object | |
187 | * @r: replay entry of truncation | |
188 | */ | |
189 | static int trun_remove_range(struct ubifs_info *c, struct replay_entry *r) | |
190 | { | |
191 | unsigned min_blk, max_blk; | |
192 | union ubifs_key min_key, max_key; | |
193 | ino_t ino; | |
194 | ||
195 | min_blk = r->new_size / UBIFS_BLOCK_SIZE; | |
196 | if (r->new_size & (UBIFS_BLOCK_SIZE - 1)) | |
197 | min_blk += 1; | |
198 | ||
199 | max_blk = r->old_size / UBIFS_BLOCK_SIZE; | |
200 | if ((r->old_size & (UBIFS_BLOCK_SIZE - 1)) == 0) | |
201 | max_blk -= 1; | |
202 | ||
203 | ino = key_inum(c, &r->key); | |
204 | ||
205 | data_key_init(c, &min_key, ino, min_blk); | |
206 | data_key_init(c, &max_key, ino, max_blk); | |
207 | ||
208 | return ubifs_tnc_remove_range(c, &min_key, &max_key); | |
209 | } | |
210 | ||
211 | /** | |
212 | * apply_replay_entry - apply a replay entry to the TNC. | |
213 | * @c: UBIFS file-system description object | |
214 | * @r: replay entry to apply | |
215 | * | |
216 | * Apply a replay entry to the TNC. | |
217 | */ | |
218 | static int apply_replay_entry(struct ubifs_info *c, struct replay_entry *r) | |
219 | { | |
074bcb9b | 220 | int err; |
1e51764a | 221 | |
074bcb9b AB |
222 | dbg_mnt("LEB %d:%d len %d deletion %d sqnum %llu %s", r->lnum, |
223 | r->offs, r->len, r->deletion, r->sqnum, DBGKEY(&r->key)); | |
1e51764a AB |
224 | |
225 | /* Set c->replay_sqnum to help deal with dangling branches. */ | |
226 | c->replay_sqnum = r->sqnum; | |
227 | ||
074bcb9b AB |
228 | if (is_hash_key(c, &r->key)) { |
229 | if (r->deletion) | |
1e51764a AB |
230 | err = ubifs_tnc_remove_nm(c, &r->key, &r->nm); |
231 | else | |
232 | err = ubifs_tnc_add_nm(c, &r->key, r->lnum, r->offs, | |
233 | r->len, &r->nm); | |
234 | } else { | |
074bcb9b | 235 | if (r->deletion) |
1e51764a AB |
236 | switch (key_type(c, &r->key)) { |
237 | case UBIFS_INO_KEY: | |
238 | { | |
239 | ino_t inum = key_inum(c, &r->key); | |
240 | ||
241 | err = ubifs_tnc_remove_ino(c, inum); | |
242 | break; | |
243 | } | |
244 | case UBIFS_TRUN_KEY: | |
245 | err = trun_remove_range(c, r); | |
246 | break; | |
247 | default: | |
248 | err = ubifs_tnc_remove(c, &r->key); | |
249 | break; | |
250 | } | |
251 | else | |
252 | err = ubifs_tnc_add(c, &r->key, r->lnum, r->offs, | |
253 | r->len); | |
254 | if (err) | |
255 | return err; | |
256 | ||
257 | if (c->need_recovery) | |
074bcb9b | 258 | err = ubifs_recover_size_accum(c, &r->key, r->deletion, |
1e51764a AB |
259 | r->new_size); |
260 | } | |
261 | ||
262 | return err; | |
263 | } | |
264 | ||
265 | /** | |
266 | * destroy_replay_tree - destroy the replay. | |
267 | * @c: UBIFS file-system description object | |
268 | * | |
269 | * Destroy the replay tree. | |
270 | */ | |
271 | static void destroy_replay_tree(struct ubifs_info *c) | |
272 | { | |
273 | struct rb_node *this = c->replay_tree.rb_node; | |
274 | struct replay_entry *r; | |
275 | ||
276 | while (this) { | |
277 | if (this->rb_left) { | |
278 | this = this->rb_left; | |
279 | continue; | |
280 | } else if (this->rb_right) { | |
281 | this = this->rb_right; | |
282 | continue; | |
283 | } | |
284 | r = rb_entry(this, struct replay_entry, rb); | |
285 | this = rb_parent(this); | |
286 | if (this) { | |
287 | if (this->rb_left == &r->rb) | |
288 | this->rb_left = NULL; | |
289 | else | |
290 | this->rb_right = NULL; | |
291 | } | |
292 | if (is_hash_key(c, &r->key)) | |
293 | kfree(r->nm.name); | |
294 | kfree(r); | |
295 | } | |
296 | c->replay_tree = RB_ROOT; | |
297 | } | |
298 | ||
299 | /** | |
300 | * apply_replay_tree - apply the replay tree to the TNC. | |
301 | * @c: UBIFS file-system description object | |
302 | * | |
303 | * Apply the replay tree. | |
304 | * Returns zero in case of success and a negative error code in case of | |
305 | * failure. | |
306 | */ | |
307 | static int apply_replay_tree(struct ubifs_info *c) | |
308 | { | |
309 | struct rb_node *this = rb_first(&c->replay_tree); | |
310 | ||
311 | while (this) { | |
312 | struct replay_entry *r; | |
313 | int err; | |
314 | ||
315 | cond_resched(); | |
316 | ||
317 | r = rb_entry(this, struct replay_entry, rb); | |
318 | err = apply_replay_entry(c, r); | |
319 | if (err) | |
320 | return err; | |
321 | this = rb_next(this); | |
322 | } | |
323 | return 0; | |
324 | } | |
325 | ||
326 | /** | |
327 | * insert_node - insert a node to the replay tree. | |
328 | * @c: UBIFS file-system description object | |
329 | * @lnum: node logical eraseblock number | |
330 | * @offs: node offset | |
331 | * @len: node length | |
332 | * @key: node key | |
333 | * @sqnum: sequence number | |
334 | * @deletion: non-zero if this is a deletion | |
335 | * @used: number of bytes in use in a LEB | |
336 | * @old_size: truncation old size | |
337 | * @new_size: truncation new size | |
338 | * | |
339 | * This function inserts a scanned non-direntry node to the replay tree. The | |
340 | * replay tree is an RB-tree containing @struct replay_entry elements which are | |
341 | * indexed by the sequence number. The replay tree is applied at the very end | |
342 | * of the replay process. Since the tree is sorted in sequence number order, | |
343 | * the older modifications are applied first. This function returns zero in | |
344 | * case of success and a negative error code in case of failure. | |
345 | */ | |
346 | static int insert_node(struct ubifs_info *c, int lnum, int offs, int len, | |
347 | union ubifs_key *key, unsigned long long sqnum, | |
348 | int deletion, int *used, loff_t old_size, | |
349 | loff_t new_size) | |
350 | { | |
351 | struct rb_node **p = &c->replay_tree.rb_node, *parent = NULL; | |
352 | struct replay_entry *r; | |
353 | ||
354 | if (key_inum(c, key) >= c->highest_inum) | |
355 | c->highest_inum = key_inum(c, key); | |
356 | ||
357 | dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key)); | |
358 | while (*p) { | |
359 | parent = *p; | |
360 | r = rb_entry(parent, struct replay_entry, rb); | |
361 | if (sqnum < r->sqnum) { | |
362 | p = &(*p)->rb_left; | |
363 | continue; | |
364 | } else if (sqnum > r->sqnum) { | |
365 | p = &(*p)->rb_right; | |
366 | continue; | |
367 | } | |
368 | ubifs_err("duplicate sqnum in replay"); | |
369 | return -EINVAL; | |
370 | } | |
371 | ||
372 | r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL); | |
373 | if (!r) | |
374 | return -ENOMEM; | |
375 | ||
376 | if (!deletion) | |
377 | *used += ALIGN(len, 8); | |
378 | r->lnum = lnum; | |
379 | r->offs = offs; | |
380 | r->len = len; | |
074bcb9b | 381 | r->deletion = !!deletion; |
1e51764a | 382 | r->sqnum = sqnum; |
074bcb9b | 383 | key_copy(c, key, &r->key); |
1e51764a AB |
384 | r->old_size = old_size; |
385 | r->new_size = new_size; | |
1e51764a AB |
386 | |
387 | rb_link_node(&r->rb, parent, p); | |
388 | rb_insert_color(&r->rb, &c->replay_tree); | |
389 | return 0; | |
390 | } | |
391 | ||
392 | /** | |
393 | * insert_dent - insert a directory entry node into the replay tree. | |
394 | * @c: UBIFS file-system description object | |
395 | * @lnum: node logical eraseblock number | |
396 | * @offs: node offset | |
397 | * @len: node length | |
398 | * @key: node key | |
399 | * @name: directory entry name | |
400 | * @nlen: directory entry name length | |
401 | * @sqnum: sequence number | |
402 | * @deletion: non-zero if this is a deletion | |
403 | * @used: number of bytes in use in a LEB | |
404 | * | |
405 | * This function inserts a scanned directory entry node to the replay tree. | |
406 | * Returns zero in case of success and a negative error code in case of | |
407 | * failure. | |
408 | * | |
409 | * This function is also used for extended attribute entries because they are | |
410 | * implemented as directory entry nodes. | |
411 | */ | |
412 | static int insert_dent(struct ubifs_info *c, int lnum, int offs, int len, | |
413 | union ubifs_key *key, const char *name, int nlen, | |
414 | unsigned long long sqnum, int deletion, int *used) | |
415 | { | |
416 | struct rb_node **p = &c->replay_tree.rb_node, *parent = NULL; | |
417 | struct replay_entry *r; | |
418 | char *nbuf; | |
419 | ||
420 | if (key_inum(c, key) >= c->highest_inum) | |
421 | c->highest_inum = key_inum(c, key); | |
422 | ||
423 | dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key)); | |
424 | while (*p) { | |
425 | parent = *p; | |
426 | r = rb_entry(parent, struct replay_entry, rb); | |
427 | if (sqnum < r->sqnum) { | |
428 | p = &(*p)->rb_left; | |
429 | continue; | |
430 | } | |
431 | if (sqnum > r->sqnum) { | |
432 | p = &(*p)->rb_right; | |
433 | continue; | |
434 | } | |
435 | ubifs_err("duplicate sqnum in replay"); | |
436 | return -EINVAL; | |
437 | } | |
438 | ||
439 | r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL); | |
440 | if (!r) | |
441 | return -ENOMEM; | |
442 | nbuf = kmalloc(nlen + 1, GFP_KERNEL); | |
443 | if (!nbuf) { | |
444 | kfree(r); | |
445 | return -ENOMEM; | |
446 | } | |
447 | ||
448 | if (!deletion) | |
449 | *used += ALIGN(len, 8); | |
450 | r->lnum = lnum; | |
451 | r->offs = offs; | |
452 | r->len = len; | |
074bcb9b | 453 | r->deletion = !!deletion; |
1e51764a | 454 | r->sqnum = sqnum; |
074bcb9b | 455 | key_copy(c, key, &r->key); |
1e51764a AB |
456 | r->nm.len = nlen; |
457 | memcpy(nbuf, name, nlen); | |
458 | nbuf[nlen] = '\0'; | |
459 | r->nm.name = nbuf; | |
1e51764a AB |
460 | |
461 | ubifs_assert(!*p); | |
462 | rb_link_node(&r->rb, parent, p); | |
463 | rb_insert_color(&r->rb, &c->replay_tree); | |
464 | return 0; | |
465 | } | |
466 | ||
467 | /** | |
468 | * ubifs_validate_entry - validate directory or extended attribute entry node. | |
469 | * @c: UBIFS file-system description object | |
470 | * @dent: the node to validate | |
471 | * | |
472 | * This function validates directory or extended attribute entry node @dent. | |
473 | * Returns zero if the node is all right and a %-EINVAL if not. | |
474 | */ | |
475 | int ubifs_validate_entry(struct ubifs_info *c, | |
476 | const struct ubifs_dent_node *dent) | |
477 | { | |
478 | int key_type = key_type_flash(c, dent->key); | |
479 | int nlen = le16_to_cpu(dent->nlen); | |
480 | ||
481 | if (le32_to_cpu(dent->ch.len) != nlen + UBIFS_DENT_NODE_SZ + 1 || | |
482 | dent->type >= UBIFS_ITYPES_CNT || | |
483 | nlen > UBIFS_MAX_NLEN || dent->name[nlen] != 0 || | |
484 | strnlen(dent->name, nlen) != nlen || | |
485 | le64_to_cpu(dent->inum) > MAX_INUM) { | |
486 | ubifs_err("bad %s node", key_type == UBIFS_DENT_KEY ? | |
487 | "directory entry" : "extended attribute entry"); | |
488 | return -EINVAL; | |
489 | } | |
490 | ||
491 | if (key_type != UBIFS_DENT_KEY && key_type != UBIFS_XENT_KEY) { | |
492 | ubifs_err("bad key type %d", key_type); | |
493 | return -EINVAL; | |
494 | } | |
495 | ||
496 | return 0; | |
497 | } | |
498 | ||
499 | /** | |
500 | * replay_bud - replay a bud logical eraseblock. | |
501 | * @c: UBIFS file-system description object | |
502 | * @lnum: bud logical eraseblock number to replay | |
503 | * @offs: bud start offset | |
504 | * @jhead: journal head to which this bud belongs | |
505 | * @free: amount of free space in the bud is returned here | |
506 | * @dirty: amount of dirty space from padding and deletion nodes is returned | |
507 | * here | |
508 | * | |
509 | * This function returns zero in case of success and a negative error code in | |
510 | * case of failure. | |
511 | */ | |
512 | static int replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead, | |
513 | int *free, int *dirty) | |
514 | { | |
515 | int err = 0, used = 0; | |
516 | struct ubifs_scan_leb *sleb; | |
517 | struct ubifs_scan_node *snod; | |
518 | struct ubifs_bud *bud; | |
519 | ||
c839e297 | 520 | dbg_mnt("replay bud LEB %d, head %d, offs %d", lnum, jhead, offs); |
1e51764a AB |
521 | if (c->need_recovery) |
522 | sleb = ubifs_recover_leb(c, lnum, offs, c->sbuf, jhead != GCHD); | |
523 | else | |
348709ba | 524 | sleb = ubifs_scan(c, lnum, offs, c->sbuf, 0); |
1e51764a AB |
525 | if (IS_ERR(sleb)) |
526 | return PTR_ERR(sleb); | |
527 | ||
528 | /* | |
529 | * The bud does not have to start from offset zero - the beginning of | |
530 | * the 'lnum' LEB may contain previously committed data. One of the | |
531 | * things we have to do in replay is to correctly update lprops with | |
532 | * newer information about this LEB. | |
533 | * | |
534 | * At this point lprops thinks that this LEB has 'c->leb_size - offs' | |
535 | * bytes of free space because it only contain information about | |
536 | * committed data. | |
537 | * | |
538 | * But we know that real amount of free space is 'c->leb_size - | |
539 | * sleb->endpt', and the space in the 'lnum' LEB between 'offs' and | |
540 | * 'sleb->endpt' is used by bud data. We have to correctly calculate | |
541 | * how much of these data are dirty and update lprops with this | |
542 | * information. | |
543 | * | |
544 | * The dirt in that LEB region is comprised of padding nodes, deletion | |
545 | * nodes, truncation nodes and nodes which are obsoleted by subsequent | |
546 | * nodes in this LEB. So instead of calculating clean space, we | |
547 | * calculate used space ('used' variable). | |
548 | */ | |
549 | ||
550 | list_for_each_entry(snod, &sleb->nodes, list) { | |
551 | int deletion = 0; | |
552 | ||
553 | cond_resched(); | |
554 | ||
555 | if (snod->sqnum >= SQNUM_WATERMARK) { | |
556 | ubifs_err("file system's life ended"); | |
557 | goto out_dump; | |
558 | } | |
559 | ||
560 | if (snod->sqnum > c->max_sqnum) | |
561 | c->max_sqnum = snod->sqnum; | |
562 | ||
563 | switch (snod->type) { | |
564 | case UBIFS_INO_NODE: | |
565 | { | |
566 | struct ubifs_ino_node *ino = snod->node; | |
567 | loff_t new_size = le64_to_cpu(ino->size); | |
568 | ||
569 | if (le32_to_cpu(ino->nlink) == 0) | |
570 | deletion = 1; | |
571 | err = insert_node(c, lnum, snod->offs, snod->len, | |
572 | &snod->key, snod->sqnum, deletion, | |
573 | &used, 0, new_size); | |
574 | break; | |
575 | } | |
576 | case UBIFS_DATA_NODE: | |
577 | { | |
578 | struct ubifs_data_node *dn = snod->node; | |
579 | loff_t new_size = le32_to_cpu(dn->size) + | |
580 | key_block(c, &snod->key) * | |
581 | UBIFS_BLOCK_SIZE; | |
582 | ||
583 | err = insert_node(c, lnum, snod->offs, snod->len, | |
584 | &snod->key, snod->sqnum, deletion, | |
585 | &used, 0, new_size); | |
586 | break; | |
587 | } | |
588 | case UBIFS_DENT_NODE: | |
589 | case UBIFS_XENT_NODE: | |
590 | { | |
591 | struct ubifs_dent_node *dent = snod->node; | |
592 | ||
593 | err = ubifs_validate_entry(c, dent); | |
594 | if (err) | |
595 | goto out_dump; | |
596 | ||
597 | err = insert_dent(c, lnum, snod->offs, snod->len, | |
598 | &snod->key, dent->name, | |
599 | le16_to_cpu(dent->nlen), snod->sqnum, | |
600 | !le64_to_cpu(dent->inum), &used); | |
601 | break; | |
602 | } | |
603 | case UBIFS_TRUN_NODE: | |
604 | { | |
605 | struct ubifs_trun_node *trun = snod->node; | |
606 | loff_t old_size = le64_to_cpu(trun->old_size); | |
607 | loff_t new_size = le64_to_cpu(trun->new_size); | |
608 | union ubifs_key key; | |
609 | ||
610 | /* Validate truncation node */ | |
611 | if (old_size < 0 || old_size > c->max_inode_sz || | |
612 | new_size < 0 || new_size > c->max_inode_sz || | |
613 | old_size <= new_size) { | |
614 | ubifs_err("bad truncation node"); | |
615 | goto out_dump; | |
616 | } | |
617 | ||
618 | /* | |
619 | * Create a fake truncation key just to use the same | |
620 | * functions which expect nodes to have keys. | |
621 | */ | |
622 | trun_key_init(c, &key, le32_to_cpu(trun->inum)); | |
623 | err = insert_node(c, lnum, snod->offs, snod->len, | |
624 | &key, snod->sqnum, 1, &used, | |
625 | old_size, new_size); | |
626 | break; | |
627 | } | |
628 | default: | |
629 | ubifs_err("unexpected node type %d in bud LEB %d:%d", | |
630 | snod->type, lnum, snod->offs); | |
631 | err = -EINVAL; | |
632 | goto out_dump; | |
633 | } | |
634 | if (err) | |
635 | goto out; | |
636 | } | |
637 | ||
638 | bud = ubifs_search_bud(c, lnum); | |
639 | if (!bud) | |
640 | BUG(); | |
641 | ||
642 | ubifs_assert(sleb->endpt - offs >= used); | |
643 | ubifs_assert(sleb->endpt % c->min_io_size == 0); | |
644 | ||
1e51764a AB |
645 | *dirty = sleb->endpt - offs - used; |
646 | *free = c->leb_size - sleb->endpt; | |
c839e297 | 647 | dbg_mnt("bud LEB %d replied: dirty %d, free %d", lnum, *dirty, *free); |
1e51764a AB |
648 | |
649 | out: | |
650 | ubifs_scan_destroy(sleb); | |
651 | return err; | |
652 | ||
653 | out_dump: | |
654 | ubifs_err("bad node is at LEB %d:%d", lnum, snod->offs); | |
655 | dbg_dump_node(c, snod->node); | |
656 | ubifs_scan_destroy(sleb); | |
657 | return -EINVAL; | |
658 | } | |
659 | ||
1e51764a AB |
660 | /** |
661 | * replay_buds - replay all buds. | |
662 | * @c: UBIFS file-system description object | |
663 | * | |
664 | * This function returns zero in case of success and a negative error code in | |
665 | * case of failure. | |
666 | */ | |
667 | static int replay_buds(struct ubifs_info *c) | |
668 | { | |
669 | struct bud_entry *b; | |
074bcb9b | 670 | int err; |
7703f09d | 671 | unsigned long long prev_sqnum = 0; |
1e51764a AB |
672 | |
673 | list_for_each_entry(b, &c->replay_buds, list) { | |
674 | err = replay_bud(c, b->bud->lnum, b->bud->start, b->bud->jhead, | |
074bcb9b | 675 | &b->free, &b->dirty); |
1e51764a AB |
676 | if (err) |
677 | return err; | |
7703f09d AB |
678 | |
679 | ubifs_assert(b->sqnum > prev_sqnum); | |
680 | prev_sqnum = b->sqnum; | |
1e51764a AB |
681 | } |
682 | ||
683 | return 0; | |
684 | } | |
685 | ||
686 | /** | |
687 | * destroy_bud_list - destroy the list of buds to replay. | |
688 | * @c: UBIFS file-system description object | |
689 | */ | |
690 | static void destroy_bud_list(struct ubifs_info *c) | |
691 | { | |
692 | struct bud_entry *b; | |
693 | ||
694 | while (!list_empty(&c->replay_buds)) { | |
695 | b = list_entry(c->replay_buds.next, struct bud_entry, list); | |
696 | list_del(&b->list); | |
697 | kfree(b); | |
698 | } | |
699 | } | |
700 | ||
701 | /** | |
702 | * add_replay_bud - add a bud to the list of buds to replay. | |
703 | * @c: UBIFS file-system description object | |
704 | * @lnum: bud logical eraseblock number to replay | |
705 | * @offs: bud start offset | |
706 | * @jhead: journal head to which this bud belongs | |
707 | * @sqnum: reference node sequence number | |
708 | * | |
709 | * This function returns zero in case of success and a negative error code in | |
710 | * case of failure. | |
711 | */ | |
712 | static int add_replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead, | |
713 | unsigned long long sqnum) | |
714 | { | |
715 | struct ubifs_bud *bud; | |
716 | struct bud_entry *b; | |
717 | ||
718 | dbg_mnt("add replay bud LEB %d:%d, head %d", lnum, offs, jhead); | |
719 | ||
720 | bud = kmalloc(sizeof(struct ubifs_bud), GFP_KERNEL); | |
721 | if (!bud) | |
722 | return -ENOMEM; | |
723 | ||
724 | b = kmalloc(sizeof(struct bud_entry), GFP_KERNEL); | |
725 | if (!b) { | |
726 | kfree(bud); | |
727 | return -ENOMEM; | |
728 | } | |
729 | ||
730 | bud->lnum = lnum; | |
731 | bud->start = offs; | |
732 | bud->jhead = jhead; | |
733 | ubifs_add_bud(c, bud); | |
734 | ||
735 | b->bud = bud; | |
736 | b->sqnum = sqnum; | |
737 | list_add_tail(&b->list, &c->replay_buds); | |
738 | ||
739 | return 0; | |
740 | } | |
741 | ||
742 | /** | |
743 | * validate_ref - validate a reference node. | |
744 | * @c: UBIFS file-system description object | |
745 | * @ref: the reference node to validate | |
746 | * @ref_lnum: LEB number of the reference node | |
747 | * @ref_offs: reference node offset | |
748 | * | |
749 | * This function returns %1 if a bud reference already exists for the LEB. %0 is | |
750 | * returned if the reference node is new, otherwise %-EINVAL is returned if | |
751 | * validation failed. | |
752 | */ | |
753 | static int validate_ref(struct ubifs_info *c, const struct ubifs_ref_node *ref) | |
754 | { | |
755 | struct ubifs_bud *bud; | |
756 | int lnum = le32_to_cpu(ref->lnum); | |
757 | unsigned int offs = le32_to_cpu(ref->offs); | |
758 | unsigned int jhead = le32_to_cpu(ref->jhead); | |
759 | ||
760 | /* | |
761 | * ref->offs may point to the end of LEB when the journal head points | |
762 | * to the end of LEB and we write reference node for it during commit. | |
763 | * So this is why we require 'offs > c->leb_size'. | |
764 | */ | |
765 | if (jhead >= c->jhead_cnt || lnum >= c->leb_cnt || | |
766 | lnum < c->main_first || offs > c->leb_size || | |
767 | offs & (c->min_io_size - 1)) | |
768 | return -EINVAL; | |
769 | ||
770 | /* Make sure we have not already looked at this bud */ | |
771 | bud = ubifs_search_bud(c, lnum); | |
772 | if (bud) { | |
773 | if (bud->jhead == jhead && bud->start <= offs) | |
774 | return 1; | |
775 | ubifs_err("bud at LEB %d:%d was already referred", lnum, offs); | |
776 | return -EINVAL; | |
777 | } | |
778 | ||
779 | return 0; | |
780 | } | |
781 | ||
782 | /** | |
783 | * replay_log_leb - replay a log logical eraseblock. | |
784 | * @c: UBIFS file-system description object | |
785 | * @lnum: log logical eraseblock to replay | |
786 | * @offs: offset to start replaying from | |
787 | * @sbuf: scan buffer | |
788 | * | |
789 | * This function replays a log LEB and returns zero in case of success, %1 if | |
790 | * this is the last LEB in the log, and a negative error code in case of | |
791 | * failure. | |
792 | */ | |
793 | static int replay_log_leb(struct ubifs_info *c, int lnum, int offs, void *sbuf) | |
794 | { | |
795 | int err; | |
796 | struct ubifs_scan_leb *sleb; | |
797 | struct ubifs_scan_node *snod; | |
798 | const struct ubifs_cs_node *node; | |
799 | ||
800 | dbg_mnt("replay log LEB %d:%d", lnum, offs); | |
348709ba AB |
801 | sleb = ubifs_scan(c, lnum, offs, sbuf, c->need_recovery); |
802 | if (IS_ERR(sleb)) { | |
ed43f2f0 AB |
803 | if (PTR_ERR(sleb) != -EUCLEAN || !c->need_recovery) |
804 | return PTR_ERR(sleb); | |
7d08ae3c AB |
805 | /* |
806 | * Note, the below function will recover this log LEB only if | |
807 | * it is the last, because unclean reboots can possibly corrupt | |
808 | * only the tail of the log. | |
809 | */ | |
ed43f2f0 | 810 | sleb = ubifs_recover_log_leb(c, lnum, offs, sbuf); |
1e51764a AB |
811 | if (IS_ERR(sleb)) |
812 | return PTR_ERR(sleb); | |
813 | } | |
814 | ||
815 | if (sleb->nodes_cnt == 0) { | |
816 | err = 1; | |
817 | goto out; | |
818 | } | |
819 | ||
820 | node = sleb->buf; | |
1e51764a AB |
821 | snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list); |
822 | if (c->cs_sqnum == 0) { | |
823 | /* | |
824 | * This is the first log LEB we are looking at, make sure that | |
825 | * the first node is a commit start node. Also record its | |
826 | * sequence number so that UBIFS can determine where the log | |
827 | * ends, because all nodes which were have higher sequence | |
828 | * numbers. | |
829 | */ | |
830 | if (snod->type != UBIFS_CS_NODE) { | |
831 | dbg_err("first log node at LEB %d:%d is not CS node", | |
832 | lnum, offs); | |
833 | goto out_dump; | |
834 | } | |
835 | if (le64_to_cpu(node->cmt_no) != c->cmt_no) { | |
836 | dbg_err("first CS node at LEB %d:%d has wrong " | |
837 | "commit number %llu expected %llu", | |
838 | lnum, offs, | |
839 | (unsigned long long)le64_to_cpu(node->cmt_no), | |
840 | c->cmt_no); | |
841 | goto out_dump; | |
842 | } | |
843 | ||
844 | c->cs_sqnum = le64_to_cpu(node->ch.sqnum); | |
845 | dbg_mnt("commit start sqnum %llu", c->cs_sqnum); | |
846 | } | |
847 | ||
848 | if (snod->sqnum < c->cs_sqnum) { | |
849 | /* | |
850 | * This means that we reached end of log and now | |
851 | * look to the older log data, which was already | |
852 | * committed but the eraseblock was not erased (UBIFS | |
6edbfafd | 853 | * only un-maps it). So this basically means we have to |
1e51764a AB |
854 | * exit with "end of log" code. |
855 | */ | |
856 | err = 1; | |
857 | goto out; | |
858 | } | |
859 | ||
860 | /* Make sure the first node sits at offset zero of the LEB */ | |
861 | if (snod->offs != 0) { | |
862 | dbg_err("first node is not at zero offset"); | |
863 | goto out_dump; | |
864 | } | |
865 | ||
866 | list_for_each_entry(snod, &sleb->nodes, list) { | |
1e51764a AB |
867 | cond_resched(); |
868 | ||
869 | if (snod->sqnum >= SQNUM_WATERMARK) { | |
870 | ubifs_err("file system's life ended"); | |
871 | goto out_dump; | |
872 | } | |
873 | ||
874 | if (snod->sqnum < c->cs_sqnum) { | |
875 | dbg_err("bad sqnum %llu, commit sqnum %llu", | |
876 | snod->sqnum, c->cs_sqnum); | |
877 | goto out_dump; | |
878 | } | |
879 | ||
880 | if (snod->sqnum > c->max_sqnum) | |
881 | c->max_sqnum = snod->sqnum; | |
882 | ||
883 | switch (snod->type) { | |
884 | case UBIFS_REF_NODE: { | |
885 | const struct ubifs_ref_node *ref = snod->node; | |
886 | ||
887 | err = validate_ref(c, ref); | |
888 | if (err == 1) | |
889 | break; /* Already have this bud */ | |
890 | if (err) | |
891 | goto out_dump; | |
892 | ||
893 | err = add_replay_bud(c, le32_to_cpu(ref->lnum), | |
894 | le32_to_cpu(ref->offs), | |
895 | le32_to_cpu(ref->jhead), | |
896 | snod->sqnum); | |
897 | if (err) | |
898 | goto out; | |
899 | ||
900 | break; | |
901 | } | |
902 | case UBIFS_CS_NODE: | |
903 | /* Make sure it sits at the beginning of LEB */ | |
904 | if (snod->offs != 0) { | |
905 | ubifs_err("unexpected node in log"); | |
906 | goto out_dump; | |
907 | } | |
908 | break; | |
909 | default: | |
910 | ubifs_err("unexpected node in log"); | |
911 | goto out_dump; | |
912 | } | |
913 | } | |
914 | ||
915 | if (sleb->endpt || c->lhead_offs >= c->leb_size) { | |
916 | c->lhead_lnum = lnum; | |
917 | c->lhead_offs = sleb->endpt; | |
918 | } | |
919 | ||
920 | err = !sleb->endpt; | |
921 | out: | |
922 | ubifs_scan_destroy(sleb); | |
923 | return err; | |
924 | ||
925 | out_dump: | |
681947d2 | 926 | ubifs_err("log error detected while replaying the log at LEB %d:%d", |
1e51764a AB |
927 | lnum, offs + snod->offs); |
928 | dbg_dump_node(c, snod->node); | |
929 | ubifs_scan_destroy(sleb); | |
930 | return -EINVAL; | |
931 | } | |
932 | ||
933 | /** | |
934 | * take_ihead - update the status of the index head in lprops to 'taken'. | |
935 | * @c: UBIFS file-system description object | |
936 | * | |
937 | * This function returns the amount of free space in the index head LEB or a | |
938 | * negative error code. | |
939 | */ | |
940 | static int take_ihead(struct ubifs_info *c) | |
941 | { | |
942 | const struct ubifs_lprops *lp; | |
943 | int err, free; | |
944 | ||
945 | ubifs_get_lprops(c); | |
946 | ||
947 | lp = ubifs_lpt_lookup_dirty(c, c->ihead_lnum); | |
948 | if (IS_ERR(lp)) { | |
949 | err = PTR_ERR(lp); | |
950 | goto out; | |
951 | } | |
952 | ||
953 | free = lp->free; | |
954 | ||
955 | lp = ubifs_change_lp(c, lp, LPROPS_NC, LPROPS_NC, | |
956 | lp->flags | LPROPS_TAKEN, 0); | |
957 | if (IS_ERR(lp)) { | |
958 | err = PTR_ERR(lp); | |
959 | goto out; | |
960 | } | |
961 | ||
962 | err = free; | |
963 | out: | |
964 | ubifs_release_lprops(c); | |
965 | return err; | |
966 | } | |
967 | ||
968 | /** | |
969 | * ubifs_replay_journal - replay journal. | |
970 | * @c: UBIFS file-system description object | |
971 | * | |
972 | * This function scans the journal, replays and cleans it up. It makes sure all | |
973 | * memory data structures related to uncommitted journal are built (dirty TNC | |
974 | * tree, tree of buds, modified lprops, etc). | |
975 | */ | |
976 | int ubifs_replay_journal(struct ubifs_info *c) | |
977 | { | |
978 | int err, i, lnum, offs, free; | |
1e51764a AB |
979 | |
980 | BUILD_BUG_ON(UBIFS_TRUN_KEY > 5); | |
981 | ||
982 | /* Update the status of the index head in lprops to 'taken' */ | |
983 | free = take_ihead(c); | |
984 | if (free < 0) | |
985 | return free; /* Error code */ | |
986 | ||
987 | if (c->ihead_offs != c->leb_size - free) { | |
988 | ubifs_err("bad index head LEB %d:%d", c->ihead_lnum, | |
989 | c->ihead_offs); | |
990 | return -EINVAL; | |
991 | } | |
992 | ||
1e51764a | 993 | dbg_mnt("start replaying the journal"); |
1e51764a | 994 | c->replaying = 1; |
1e51764a AB |
995 | lnum = c->ltail_lnum = c->lhead_lnum; |
996 | offs = c->lhead_offs; | |
997 | ||
998 | for (i = 0; i < c->log_lebs; i++, lnum++) { | |
999 | if (lnum >= UBIFS_LOG_LNUM + c->log_lebs) { | |
1000 | /* | |
1001 | * The log is logically circular, we reached the last | |
1002 | * LEB, switch to the first one. | |
1003 | */ | |
1004 | lnum = UBIFS_LOG_LNUM; | |
1005 | offs = 0; | |
1006 | } | |
6599fcbd | 1007 | err = replay_log_leb(c, lnum, offs, c->sbuf); |
1e51764a AB |
1008 | if (err == 1) |
1009 | /* We hit the end of the log */ | |
1010 | break; | |
1011 | if (err) | |
1012 | goto out; | |
1013 | offs = 0; | |
1014 | } | |
1015 | ||
1016 | err = replay_buds(c); | |
1017 | if (err) | |
1018 | goto out; | |
1019 | ||
1020 | err = apply_replay_tree(c); | |
1021 | if (err) | |
1022 | goto out; | |
1023 | ||
074bcb9b AB |
1024 | err = set_buds_lprops(c); |
1025 | if (err) | |
1026 | goto out; | |
1027 | ||
6edbfafd | 1028 | /* |
b137545c AB |
1029 | * UBIFS budgeting calculations use @c->bi.uncommitted_idx variable |
1030 | * to roughly estimate index growth. Things like @c->bi.min_idx_lebs | |
6edbfafd AB |
1031 | * depend on it. This means we have to initialize it to make sure |
1032 | * budgeting works properly. | |
1033 | */ | |
b137545c AB |
1034 | c->bi.uncommitted_idx = atomic_long_read(&c->dirty_zn_cnt); |
1035 | c->bi.uncommitted_idx *= c->max_idx_node_sz; | |
6edbfafd | 1036 | |
1e51764a AB |
1037 | ubifs_assert(c->bud_bytes <= c->max_bud_bytes || c->need_recovery); |
1038 | dbg_mnt("finished, log head LEB %d:%d, max_sqnum %llu, " | |
1039 | "highest_inum %lu", c->lhead_lnum, c->lhead_offs, c->max_sqnum, | |
e84461ad | 1040 | (unsigned long)c->highest_inum); |
1e51764a AB |
1041 | out: |
1042 | destroy_replay_tree(c); | |
1043 | destroy_bud_list(c); | |
1e51764a AB |
1044 | c->replaying = 0; |
1045 | return err; | |
1046 | } |