Loading...
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#include <linux/list_sort.h>
37
38/**
39 * struct replay_entry - replay list entry.
40 * @lnum: logical eraseblock number of the node
41 * @offs: node offset
42 * @len: node length
43 * @deletion: non-zero if this entry corresponds to a node deletion
44 * @sqnum: node sequence number
45 * @list: links the replay list
46 * @key: node key
47 * @nm: directory entry name
48 * @old_size: truncation old size
49 * @new_size: truncation new size
50 *
51 * The replay process first scans all buds and builds the replay list, then
52 * sorts the replay list in nodes sequence number order, and then inserts all
53 * the replay entries to the TNC.
54 */
55struct replay_entry {
56 int lnum;
57 int offs;
58 int len;
59 unsigned int deletion:1;
60 unsigned long long sqnum;
61 struct list_head list;
62 union ubifs_key key;
63 union {
64 struct qstr nm;
65 struct {
66 loff_t old_size;
67 loff_t new_size;
68 };
69 };
70};
71
72/**
73 * struct bud_entry - entry in the list of buds to replay.
74 * @list: next bud in the list
75 * @bud: bud description object
76 * @sqnum: reference node sequence number
77 * @free: free bytes in the bud
78 * @dirty: dirty bytes in the bud
79 */
80struct bud_entry {
81 struct list_head list;
82 struct ubifs_bud *bud;
83 unsigned long long sqnum;
84 int free;
85 int dirty;
86};
87
88/**
89 * set_bud_lprops - set free and dirty space used by a bud.
90 * @c: UBIFS file-system description object
91 * @b: bud entry which describes the bud
92 *
93 * This function makes sure the LEB properties of bud @b are set correctly
94 * after the replay. Returns zero in case of success and a negative error code
95 * in case of failure.
96 */
97static int set_bud_lprops(struct ubifs_info *c, struct bud_entry *b)
98{
99 const struct ubifs_lprops *lp;
100 int err = 0, dirty;
101
102 ubifs_get_lprops(c);
103
104 lp = ubifs_lpt_lookup_dirty(c, b->bud->lnum);
105 if (IS_ERR(lp)) {
106 err = PTR_ERR(lp);
107 goto out;
108 }
109
110 dirty = lp->dirty;
111 if (b->bud->start == 0 && (lp->free != c->leb_size || lp->dirty != 0)) {
112 /*
113 * The LEB was added to the journal with a starting offset of
114 * zero which means the LEB must have been empty. The LEB
115 * property values should be @lp->free == @c->leb_size and
116 * @lp->dirty == 0, but that is not the case. The reason is that
117 * the LEB had been garbage collected before it became the bud,
118 * and there was not commit inbetween. The garbage collector
119 * resets the free and dirty space without recording it
120 * anywhere except lprops, so if there was no commit then
121 * lprops does not have that information.
122 *
123 * We do not need to adjust free space because the scan has told
124 * us the exact value which is recorded in the replay entry as
125 * @b->free.
126 *
127 * However we do need to subtract from the dirty space the
128 * amount of space that the garbage collector reclaimed, which
129 * is the whole LEB minus the amount of space that was free.
130 */
131 dbg_mnt("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
132 lp->free, lp->dirty);
133 dbg_gc("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
134 lp->free, lp->dirty);
135 dirty -= c->leb_size - lp->free;
136 /*
137 * If the replay order was perfect the dirty space would now be
138 * zero. The order is not perfect because the journal heads
139 * race with each other. This is not a problem but is does mean
140 * that the dirty space may temporarily exceed c->leb_size
141 * during the replay.
142 */
143 if (dirty != 0)
144 dbg_msg("LEB %d lp: %d free %d dirty "
145 "replay: %d free %d dirty", b->bud->lnum,
146 lp->free, lp->dirty, b->free, b->dirty);
147 }
148 lp = ubifs_change_lp(c, lp, b->free, dirty + b->dirty,
149 lp->flags | LPROPS_TAKEN, 0);
150 if (IS_ERR(lp)) {
151 err = PTR_ERR(lp);
152 goto out;
153 }
154
155 /* Make sure the journal head points to the latest bud */
156 err = ubifs_wbuf_seek_nolock(&c->jheads[b->bud->jhead].wbuf,
157 b->bud->lnum, c->leb_size - b->free,
158 UBI_SHORTTERM);
159
160out:
161 ubifs_release_lprops(c);
162 return err;
163}
164
165/**
166 * set_buds_lprops - set free and dirty space for all replayed buds.
167 * @c: UBIFS file-system description object
168 *
169 * This function sets LEB properties for all replayed buds. Returns zero in
170 * case of success and a negative error code in case of failure.
171 */
172static int set_buds_lprops(struct ubifs_info *c)
173{
174 struct bud_entry *b;
175 int err;
176
177 list_for_each_entry(b, &c->replay_buds, list) {
178 err = set_bud_lprops(c, b);
179 if (err)
180 return err;
181 }
182
183 return 0;
184}
185
186/**
187 * trun_remove_range - apply a replay entry for a truncation to the TNC.
188 * @c: UBIFS file-system description object
189 * @r: replay entry of truncation
190 */
191static int trun_remove_range(struct ubifs_info *c, struct replay_entry *r)
192{
193 unsigned min_blk, max_blk;
194 union ubifs_key min_key, max_key;
195 ino_t ino;
196
197 min_blk = r->new_size / UBIFS_BLOCK_SIZE;
198 if (r->new_size & (UBIFS_BLOCK_SIZE - 1))
199 min_blk += 1;
200
201 max_blk = r->old_size / UBIFS_BLOCK_SIZE;
202 if ((r->old_size & (UBIFS_BLOCK_SIZE - 1)) == 0)
203 max_blk -= 1;
204
205 ino = key_inum(c, &r->key);
206
207 data_key_init(c, &min_key, ino, min_blk);
208 data_key_init(c, &max_key, ino, max_blk);
209
210 return ubifs_tnc_remove_range(c, &min_key, &max_key);
211}
212
213/**
214 * apply_replay_entry - apply a replay entry to the TNC.
215 * @c: UBIFS file-system description object
216 * @r: replay entry to apply
217 *
218 * Apply a replay entry to the TNC.
219 */
220static int apply_replay_entry(struct ubifs_info *c, struct replay_entry *r)
221{
222 int err;
223
224 dbg_mnt("LEB %d:%d len %d deletion %d sqnum %llu %s", r->lnum,
225 r->offs, r->len, r->deletion, r->sqnum, DBGKEY(&r->key));
226
227 /* Set c->replay_sqnum to help deal with dangling branches. */
228 c->replay_sqnum = r->sqnum;
229
230 if (is_hash_key(c, &r->key)) {
231 if (r->deletion)
232 err = ubifs_tnc_remove_nm(c, &r->key, &r->nm);
233 else
234 err = ubifs_tnc_add_nm(c, &r->key, r->lnum, r->offs,
235 r->len, &r->nm);
236 } else {
237 if (r->deletion)
238 switch (key_type(c, &r->key)) {
239 case UBIFS_INO_KEY:
240 {
241 ino_t inum = key_inum(c, &r->key);
242
243 err = ubifs_tnc_remove_ino(c, inum);
244 break;
245 }
246 case UBIFS_TRUN_KEY:
247 err = trun_remove_range(c, r);
248 break;
249 default:
250 err = ubifs_tnc_remove(c, &r->key);
251 break;
252 }
253 else
254 err = ubifs_tnc_add(c, &r->key, r->lnum, r->offs,
255 r->len);
256 if (err)
257 return err;
258
259 if (c->need_recovery)
260 err = ubifs_recover_size_accum(c, &r->key, r->deletion,
261 r->new_size);
262 }
263
264 return err;
265}
266
267/**
268 * replay_entries_cmp - compare 2 replay entries.
269 * @priv: UBIFS file-system description object
270 * @a: first replay entry
271 * @a: second replay entry
272 *
273 * This is a comparios function for 'list_sort()' which compares 2 replay
274 * entries @a and @b by comparing their sequence numer. Returns %1 if @a has
275 * greater sequence number and %-1 otherwise.
276 */
277static int replay_entries_cmp(void *priv, struct list_head *a,
278 struct list_head *b)
279{
280 struct replay_entry *ra, *rb;
281
282 cond_resched();
283 if (a == b)
284 return 0;
285
286 ra = list_entry(a, struct replay_entry, list);
287 rb = list_entry(b, struct replay_entry, list);
288 ubifs_assert(ra->sqnum != rb->sqnum);
289 if (ra->sqnum > rb->sqnum)
290 return 1;
291 return -1;
292}
293
294/**
295 * apply_replay_list - apply the replay list to the TNC.
296 * @c: UBIFS file-system description object
297 *
298 * Apply all entries in the replay list to the TNC. Returns zero in case of
299 * success and a negative error code in case of failure.
300 */
301static int apply_replay_list(struct ubifs_info *c)
302{
303 struct replay_entry *r;
304 int err;
305
306 list_sort(c, &c->replay_list, &replay_entries_cmp);
307
308 list_for_each_entry(r, &c->replay_list, list) {
309 cond_resched();
310
311 err = apply_replay_entry(c, r);
312 if (err)
313 return err;
314 }
315
316 return 0;
317}
318
319/**
320 * destroy_replay_list - destroy the replay.
321 * @c: UBIFS file-system description object
322 *
323 * Destroy the replay list.
324 */
325static void destroy_replay_list(struct ubifs_info *c)
326{
327 struct replay_entry *r, *tmp;
328
329 list_for_each_entry_safe(r, tmp, &c->replay_list, list) {
330 if (is_hash_key(c, &r->key))
331 kfree(r->nm.name);
332 list_del(&r->list);
333 kfree(r);
334 }
335}
336
337/**
338 * insert_node - insert a node to the replay list
339 * @c: UBIFS file-system description object
340 * @lnum: node logical eraseblock number
341 * @offs: node offset
342 * @len: node length
343 * @key: node key
344 * @sqnum: sequence number
345 * @deletion: non-zero if this is a deletion
346 * @used: number of bytes in use in a LEB
347 * @old_size: truncation old size
348 * @new_size: truncation new size
349 *
350 * This function inserts a scanned non-direntry node to the replay list. The
351 * replay list contains @struct replay_entry elements, and we sort this list in
352 * sequence number order before applying it. The replay list is applied at the
353 * very end of the replay process. Since the list is sorted in sequence number
354 * order, the older modifications are applied first. This function returns zero
355 * in case of success and a negative error code in case of failure.
356 */
357static int insert_node(struct ubifs_info *c, int lnum, int offs, int len,
358 union ubifs_key *key, unsigned long long sqnum,
359 int deletion, int *used, loff_t old_size,
360 loff_t new_size)
361{
362 struct replay_entry *r;
363
364 dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key));
365
366 if (key_inum(c, key) >= c->highest_inum)
367 c->highest_inum = key_inum(c, key);
368
369 r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
370 if (!r)
371 return -ENOMEM;
372
373 if (!deletion)
374 *used += ALIGN(len, 8);
375 r->lnum = lnum;
376 r->offs = offs;
377 r->len = len;
378 r->deletion = !!deletion;
379 r->sqnum = sqnum;
380 key_copy(c, key, &r->key);
381 r->old_size = old_size;
382 r->new_size = new_size;
383
384 list_add_tail(&r->list, &c->replay_list);
385 return 0;
386}
387
388/**
389 * insert_dent - insert a directory entry node into the replay list.
390 * @c: UBIFS file-system description object
391 * @lnum: node logical eraseblock number
392 * @offs: node offset
393 * @len: node length
394 * @key: node key
395 * @name: directory entry name
396 * @nlen: directory entry name length
397 * @sqnum: sequence number
398 * @deletion: non-zero if this is a deletion
399 * @used: number of bytes in use in a LEB
400 *
401 * This function inserts a scanned directory entry node or an extended
402 * attribute entry to the replay list. Returns zero in case of success and a
403 * negative error code in case of failure.
404 */
405static int insert_dent(struct ubifs_info *c, int lnum, int offs, int len,
406 union ubifs_key *key, const char *name, int nlen,
407 unsigned long long sqnum, int deletion, int *used)
408{
409 struct replay_entry *r;
410 char *nbuf;
411
412 dbg_mnt("add LEB %d:%d, key %s", lnum, offs, DBGKEY(key));
413 if (key_inum(c, key) >= c->highest_inum)
414 c->highest_inum = key_inum(c, key);
415
416 r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
417 if (!r)
418 return -ENOMEM;
419
420 nbuf = kmalloc(nlen + 1, GFP_KERNEL);
421 if (!nbuf) {
422 kfree(r);
423 return -ENOMEM;
424 }
425
426 if (!deletion)
427 *used += ALIGN(len, 8);
428 r->lnum = lnum;
429 r->offs = offs;
430 r->len = len;
431 r->deletion = !!deletion;
432 r->sqnum = sqnum;
433 key_copy(c, key, &r->key);
434 r->nm.len = nlen;
435 memcpy(nbuf, name, nlen);
436 nbuf[nlen] = '\0';
437 r->nm.name = nbuf;
438
439 list_add_tail(&r->list, &c->replay_list);
440 return 0;
441}
442
443/**
444 * ubifs_validate_entry - validate directory or extended attribute entry node.
445 * @c: UBIFS file-system description object
446 * @dent: the node to validate
447 *
448 * This function validates directory or extended attribute entry node @dent.
449 * Returns zero if the node is all right and a %-EINVAL if not.
450 */
451int ubifs_validate_entry(struct ubifs_info *c,
452 const struct ubifs_dent_node *dent)
453{
454 int key_type = key_type_flash(c, dent->key);
455 int nlen = le16_to_cpu(dent->nlen);
456
457 if (le32_to_cpu(dent->ch.len) != nlen + UBIFS_DENT_NODE_SZ + 1 ||
458 dent->type >= UBIFS_ITYPES_CNT ||
459 nlen > UBIFS_MAX_NLEN || dent->name[nlen] != 0 ||
460 strnlen(dent->name, nlen) != nlen ||
461 le64_to_cpu(dent->inum) > MAX_INUM) {
462 ubifs_err("bad %s node", key_type == UBIFS_DENT_KEY ?
463 "directory entry" : "extended attribute entry");
464 return -EINVAL;
465 }
466
467 if (key_type != UBIFS_DENT_KEY && key_type != UBIFS_XENT_KEY) {
468 ubifs_err("bad key type %d", key_type);
469 return -EINVAL;
470 }
471
472 return 0;
473}
474
475/**
476 * is_last_bud - check if the bud is the last in the journal head.
477 * @c: UBIFS file-system description object
478 * @bud: bud description object
479 *
480 * This function checks if bud @bud is the last bud in its journal head. This
481 * information is then used by 'replay_bud()' to decide whether the bud can
482 * have corruptions or not. Indeed, only last buds can be corrupted by power
483 * cuts. Returns %1 if this is the last bud, and %0 if not.
484 */
485static int is_last_bud(struct ubifs_info *c, struct ubifs_bud *bud)
486{
487 struct ubifs_jhead *jh = &c->jheads[bud->jhead];
488 struct ubifs_bud *next;
489 uint32_t data;
490 int err;
491
492 if (list_is_last(&bud->list, &jh->buds_list))
493 return 1;
494
495 /*
496 * The following is a quirk to make sure we work correctly with UBIFS
497 * images used with older UBIFS.
498 *
499 * Normally, the last bud will be the last in the journal head's list
500 * of bud. However, there is one exception if the UBIFS image belongs
501 * to older UBIFS. This is fairly unlikely: one would need to use old
502 * UBIFS, then have a power cut exactly at the right point, and then
503 * try to mount this image with new UBIFS.
504 *
505 * The exception is: it is possible to have 2 buds A and B, A goes
506 * before B, and B is the last, bud B is contains no data, and bud A is
507 * corrupted at the end. The reason is that in older versions when the
508 * journal code switched the next bud (from A to B), it first added a
509 * log reference node for the new bud (B), and only after this it
510 * synchronized the write-buffer of current bud (A). But later this was
511 * changed and UBIFS started to always synchronize the write-buffer of
512 * the bud (A) before writing the log reference for the new bud (B).
513 *
514 * But because older UBIFS always synchronized A's write-buffer before
515 * writing to B, we can recognize this exceptional situation but
516 * checking the contents of bud B - if it is empty, then A can be
517 * treated as the last and we can recover it.
518 *
519 * TODO: remove this piece of code in a couple of years (today it is
520 * 16.05.2011).
521 */
522 next = list_entry(bud->list.next, struct ubifs_bud, list);
523 if (!list_is_last(&next->list, &jh->buds_list))
524 return 0;
525
526 err = ubifs_leb_read(c, next->lnum, (char *)&data, next->start, 4, 1);
527 if (err)
528 return 0;
529
530 return data == 0xFFFFFFFF;
531}
532
533/**
534 * replay_bud - replay a bud logical eraseblock.
535 * @c: UBIFS file-system description object
536 * @b: bud entry which describes the bud
537 *
538 * This function replays bud @bud, recovers it if needed, and adds all nodes
539 * from this bud to the replay list. Returns zero in case of success and a
540 * negative error code in case of failure.
541 */
542static int replay_bud(struct ubifs_info *c, struct bud_entry *b)
543{
544 int is_last = is_last_bud(c, b->bud);
545 int err = 0, used = 0, lnum = b->bud->lnum, offs = b->bud->start;
546 struct ubifs_scan_leb *sleb;
547 struct ubifs_scan_node *snod;
548
549 dbg_mnt("replay bud LEB %d, head %d, offs %d, is_last %d",
550 lnum, b->bud->jhead, offs, is_last);
551
552 if (c->need_recovery && is_last)
553 /*
554 * Recover only last LEBs in the journal heads, because power
555 * cuts may cause corruptions only in these LEBs, because only
556 * these LEBs could possibly be written to at the power cut
557 * time.
558 */
559 sleb = ubifs_recover_leb(c, lnum, offs, c->sbuf, b->bud->jhead);
560 else
561 sleb = ubifs_scan(c, lnum, offs, c->sbuf, 0);
562 if (IS_ERR(sleb))
563 return PTR_ERR(sleb);
564
565 /*
566 * The bud does not have to start from offset zero - the beginning of
567 * the 'lnum' LEB may contain previously committed data. One of the
568 * things we have to do in replay is to correctly update lprops with
569 * newer information about this LEB.
570 *
571 * At this point lprops thinks that this LEB has 'c->leb_size - offs'
572 * bytes of free space because it only contain information about
573 * committed data.
574 *
575 * But we know that real amount of free space is 'c->leb_size -
576 * sleb->endpt', and the space in the 'lnum' LEB between 'offs' and
577 * 'sleb->endpt' is used by bud data. We have to correctly calculate
578 * how much of these data are dirty and update lprops with this
579 * information.
580 *
581 * The dirt in that LEB region is comprised of padding nodes, deletion
582 * nodes, truncation nodes and nodes which are obsoleted by subsequent
583 * nodes in this LEB. So instead of calculating clean space, we
584 * calculate used space ('used' variable).
585 */
586
587 list_for_each_entry(snod, &sleb->nodes, list) {
588 int deletion = 0;
589
590 cond_resched();
591
592 if (snod->sqnum >= SQNUM_WATERMARK) {
593 ubifs_err("file system's life ended");
594 goto out_dump;
595 }
596
597 if (snod->sqnum > c->max_sqnum)
598 c->max_sqnum = snod->sqnum;
599
600 switch (snod->type) {
601 case UBIFS_INO_NODE:
602 {
603 struct ubifs_ino_node *ino = snod->node;
604 loff_t new_size = le64_to_cpu(ino->size);
605
606 if (le32_to_cpu(ino->nlink) == 0)
607 deletion = 1;
608 err = insert_node(c, lnum, snod->offs, snod->len,
609 &snod->key, snod->sqnum, deletion,
610 &used, 0, new_size);
611 break;
612 }
613 case UBIFS_DATA_NODE:
614 {
615 struct ubifs_data_node *dn = snod->node;
616 loff_t new_size = le32_to_cpu(dn->size) +
617 key_block(c, &snod->key) *
618 UBIFS_BLOCK_SIZE;
619
620 err = insert_node(c, lnum, snod->offs, snod->len,
621 &snod->key, snod->sqnum, deletion,
622 &used, 0, new_size);
623 break;
624 }
625 case UBIFS_DENT_NODE:
626 case UBIFS_XENT_NODE:
627 {
628 struct ubifs_dent_node *dent = snod->node;
629
630 err = ubifs_validate_entry(c, dent);
631 if (err)
632 goto out_dump;
633
634 err = insert_dent(c, lnum, snod->offs, snod->len,
635 &snod->key, dent->name,
636 le16_to_cpu(dent->nlen), snod->sqnum,
637 !le64_to_cpu(dent->inum), &used);
638 break;
639 }
640 case UBIFS_TRUN_NODE:
641 {
642 struct ubifs_trun_node *trun = snod->node;
643 loff_t old_size = le64_to_cpu(trun->old_size);
644 loff_t new_size = le64_to_cpu(trun->new_size);
645 union ubifs_key key;
646
647 /* Validate truncation node */
648 if (old_size < 0 || old_size > c->max_inode_sz ||
649 new_size < 0 || new_size > c->max_inode_sz ||
650 old_size <= new_size) {
651 ubifs_err("bad truncation node");
652 goto out_dump;
653 }
654
655 /*
656 * Create a fake truncation key just to use the same
657 * functions which expect nodes to have keys.
658 */
659 trun_key_init(c, &key, le32_to_cpu(trun->inum));
660 err = insert_node(c, lnum, snod->offs, snod->len,
661 &key, snod->sqnum, 1, &used,
662 old_size, new_size);
663 break;
664 }
665 default:
666 ubifs_err("unexpected node type %d in bud LEB %d:%d",
667 snod->type, lnum, snod->offs);
668 err = -EINVAL;
669 goto out_dump;
670 }
671 if (err)
672 goto out;
673 }
674
675 ubifs_assert(ubifs_search_bud(c, lnum));
676 ubifs_assert(sleb->endpt - offs >= used);
677 ubifs_assert(sleb->endpt % c->min_io_size == 0);
678
679 b->dirty = sleb->endpt - offs - used;
680 b->free = c->leb_size - sleb->endpt;
681 dbg_mnt("bud LEB %d replied: dirty %d, free %d", lnum, b->dirty, b->free);
682
683out:
684 ubifs_scan_destroy(sleb);
685 return err;
686
687out_dump:
688 ubifs_err("bad node is at LEB %d:%d", lnum, snod->offs);
689 dbg_dump_node(c, snod->node);
690 ubifs_scan_destroy(sleb);
691 return -EINVAL;
692}
693
694/**
695 * replay_buds - replay all buds.
696 * @c: UBIFS file-system description object
697 *
698 * This function returns zero in case of success and a negative error code in
699 * case of failure.
700 */
701static int replay_buds(struct ubifs_info *c)
702{
703 struct bud_entry *b;
704 int err;
705 unsigned long long prev_sqnum = 0;
706
707 list_for_each_entry(b, &c->replay_buds, list) {
708 err = replay_bud(c, b);
709 if (err)
710 return err;
711
712 ubifs_assert(b->sqnum > prev_sqnum);
713 prev_sqnum = b->sqnum;
714 }
715
716 return 0;
717}
718
719/**
720 * destroy_bud_list - destroy the list of buds to replay.
721 * @c: UBIFS file-system description object
722 */
723static void destroy_bud_list(struct ubifs_info *c)
724{
725 struct bud_entry *b;
726
727 while (!list_empty(&c->replay_buds)) {
728 b = list_entry(c->replay_buds.next, struct bud_entry, list);
729 list_del(&b->list);
730 kfree(b);
731 }
732}
733
734/**
735 * add_replay_bud - add a bud to the list of buds to replay.
736 * @c: UBIFS file-system description object
737 * @lnum: bud logical eraseblock number to replay
738 * @offs: bud start offset
739 * @jhead: journal head to which this bud belongs
740 * @sqnum: reference node sequence number
741 *
742 * This function returns zero in case of success and a negative error code in
743 * case of failure.
744 */
745static int add_replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead,
746 unsigned long long sqnum)
747{
748 struct ubifs_bud *bud;
749 struct bud_entry *b;
750
751 dbg_mnt("add replay bud LEB %d:%d, head %d", lnum, offs, jhead);
752
753 bud = kmalloc(sizeof(struct ubifs_bud), GFP_KERNEL);
754 if (!bud)
755 return -ENOMEM;
756
757 b = kmalloc(sizeof(struct bud_entry), GFP_KERNEL);
758 if (!b) {
759 kfree(bud);
760 return -ENOMEM;
761 }
762
763 bud->lnum = lnum;
764 bud->start = offs;
765 bud->jhead = jhead;
766 ubifs_add_bud(c, bud);
767
768 b->bud = bud;
769 b->sqnum = sqnum;
770 list_add_tail(&b->list, &c->replay_buds);
771
772 return 0;
773}
774
775/**
776 * validate_ref - validate a reference node.
777 * @c: UBIFS file-system description object
778 * @ref: the reference node to validate
779 * @ref_lnum: LEB number of the reference node
780 * @ref_offs: reference node offset
781 *
782 * This function returns %1 if a bud reference already exists for the LEB. %0 is
783 * returned if the reference node is new, otherwise %-EINVAL is returned if
784 * validation failed.
785 */
786static int validate_ref(struct ubifs_info *c, const struct ubifs_ref_node *ref)
787{
788 struct ubifs_bud *bud;
789 int lnum = le32_to_cpu(ref->lnum);
790 unsigned int offs = le32_to_cpu(ref->offs);
791 unsigned int jhead = le32_to_cpu(ref->jhead);
792
793 /*
794 * ref->offs may point to the end of LEB when the journal head points
795 * to the end of LEB and we write reference node for it during commit.
796 * So this is why we require 'offs > c->leb_size'.
797 */
798 if (jhead >= c->jhead_cnt || lnum >= c->leb_cnt ||
799 lnum < c->main_first || offs > c->leb_size ||
800 offs & (c->min_io_size - 1))
801 return -EINVAL;
802
803 /* Make sure we have not already looked at this bud */
804 bud = ubifs_search_bud(c, lnum);
805 if (bud) {
806 if (bud->jhead == jhead && bud->start <= offs)
807 return 1;
808 ubifs_err("bud at LEB %d:%d was already referred", lnum, offs);
809 return -EINVAL;
810 }
811
812 return 0;
813}
814
815/**
816 * replay_log_leb - replay a log logical eraseblock.
817 * @c: UBIFS file-system description object
818 * @lnum: log logical eraseblock to replay
819 * @offs: offset to start replaying from
820 * @sbuf: scan buffer
821 *
822 * This function replays a log LEB and returns zero in case of success, %1 if
823 * this is the last LEB in the log, and a negative error code in case of
824 * failure.
825 */
826static int replay_log_leb(struct ubifs_info *c, int lnum, int offs, void *sbuf)
827{
828 int err;
829 struct ubifs_scan_leb *sleb;
830 struct ubifs_scan_node *snod;
831 const struct ubifs_cs_node *node;
832
833 dbg_mnt("replay log LEB %d:%d", lnum, offs);
834 sleb = ubifs_scan(c, lnum, offs, sbuf, c->need_recovery);
835 if (IS_ERR(sleb)) {
836 if (PTR_ERR(sleb) != -EUCLEAN || !c->need_recovery)
837 return PTR_ERR(sleb);
838 /*
839 * Note, the below function will recover this log LEB only if
840 * it is the last, because unclean reboots can possibly corrupt
841 * only the tail of the log.
842 */
843 sleb = ubifs_recover_log_leb(c, lnum, offs, sbuf);
844 if (IS_ERR(sleb))
845 return PTR_ERR(sleb);
846 }
847
848 if (sleb->nodes_cnt == 0) {
849 err = 1;
850 goto out;
851 }
852
853 node = sleb->buf;
854 snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);
855 if (c->cs_sqnum == 0) {
856 /*
857 * This is the first log LEB we are looking at, make sure that
858 * the first node is a commit start node. Also record its
859 * sequence number so that UBIFS can determine where the log
860 * ends, because all nodes which were have higher sequence
861 * numbers.
862 */
863 if (snod->type != UBIFS_CS_NODE) {
864 dbg_err("first log node at LEB %d:%d is not CS node",
865 lnum, offs);
866 goto out_dump;
867 }
868 if (le64_to_cpu(node->cmt_no) != c->cmt_no) {
869 dbg_err("first CS node at LEB %d:%d has wrong "
870 "commit number %llu expected %llu",
871 lnum, offs,
872 (unsigned long long)le64_to_cpu(node->cmt_no),
873 c->cmt_no);
874 goto out_dump;
875 }
876
877 c->cs_sqnum = le64_to_cpu(node->ch.sqnum);
878 dbg_mnt("commit start sqnum %llu", c->cs_sqnum);
879 }
880
881 if (snod->sqnum < c->cs_sqnum) {
882 /*
883 * This means that we reached end of log and now
884 * look to the older log data, which was already
885 * committed but the eraseblock was not erased (UBIFS
886 * only un-maps it). So this basically means we have to
887 * exit with "end of log" code.
888 */
889 err = 1;
890 goto out;
891 }
892
893 /* Make sure the first node sits at offset zero of the LEB */
894 if (snod->offs != 0) {
895 dbg_err("first node is not at zero offset");
896 goto out_dump;
897 }
898
899 list_for_each_entry(snod, &sleb->nodes, list) {
900 cond_resched();
901
902 if (snod->sqnum >= SQNUM_WATERMARK) {
903 ubifs_err("file system's life ended");
904 goto out_dump;
905 }
906
907 if (snod->sqnum < c->cs_sqnum) {
908 dbg_err("bad sqnum %llu, commit sqnum %llu",
909 snod->sqnum, c->cs_sqnum);
910 goto out_dump;
911 }
912
913 if (snod->sqnum > c->max_sqnum)
914 c->max_sqnum = snod->sqnum;
915
916 switch (snod->type) {
917 case UBIFS_REF_NODE: {
918 const struct ubifs_ref_node *ref = snod->node;
919
920 err = validate_ref(c, ref);
921 if (err == 1)
922 break; /* Already have this bud */
923 if (err)
924 goto out_dump;
925
926 err = add_replay_bud(c, le32_to_cpu(ref->lnum),
927 le32_to_cpu(ref->offs),
928 le32_to_cpu(ref->jhead),
929 snod->sqnum);
930 if (err)
931 goto out;
932
933 break;
934 }
935 case UBIFS_CS_NODE:
936 /* Make sure it sits at the beginning of LEB */
937 if (snod->offs != 0) {
938 ubifs_err("unexpected node in log");
939 goto out_dump;
940 }
941 break;
942 default:
943 ubifs_err("unexpected node in log");
944 goto out_dump;
945 }
946 }
947
948 if (sleb->endpt || c->lhead_offs >= c->leb_size) {
949 c->lhead_lnum = lnum;
950 c->lhead_offs = sleb->endpt;
951 }
952
953 err = !sleb->endpt;
954out:
955 ubifs_scan_destroy(sleb);
956 return err;
957
958out_dump:
959 ubifs_err("log error detected while replaying the log at LEB %d:%d",
960 lnum, offs + snod->offs);
961 dbg_dump_node(c, snod->node);
962 ubifs_scan_destroy(sleb);
963 return -EINVAL;
964}
965
966/**
967 * take_ihead - update the status of the index head in lprops to 'taken'.
968 * @c: UBIFS file-system description object
969 *
970 * This function returns the amount of free space in the index head LEB or a
971 * negative error code.
972 */
973static int take_ihead(struct ubifs_info *c)
974{
975 const struct ubifs_lprops *lp;
976 int err, free;
977
978 ubifs_get_lprops(c);
979
980 lp = ubifs_lpt_lookup_dirty(c, c->ihead_lnum);
981 if (IS_ERR(lp)) {
982 err = PTR_ERR(lp);
983 goto out;
984 }
985
986 free = lp->free;
987
988 lp = ubifs_change_lp(c, lp, LPROPS_NC, LPROPS_NC,
989 lp->flags | LPROPS_TAKEN, 0);
990 if (IS_ERR(lp)) {
991 err = PTR_ERR(lp);
992 goto out;
993 }
994
995 err = free;
996out:
997 ubifs_release_lprops(c);
998 return err;
999}
1000
1001/**
1002 * ubifs_replay_journal - replay journal.
1003 * @c: UBIFS file-system description object
1004 *
1005 * This function scans the journal, replays and cleans it up. It makes sure all
1006 * memory data structures related to uncommitted journal are built (dirty TNC
1007 * tree, tree of buds, modified lprops, etc).
1008 */
1009int ubifs_replay_journal(struct ubifs_info *c)
1010{
1011 int err, i, lnum, offs, free;
1012
1013 BUILD_BUG_ON(UBIFS_TRUN_KEY > 5);
1014
1015 /* Update the status of the index head in lprops to 'taken' */
1016 free = take_ihead(c);
1017 if (free < 0)
1018 return free; /* Error code */
1019
1020 if (c->ihead_offs != c->leb_size - free) {
1021 ubifs_err("bad index head LEB %d:%d", c->ihead_lnum,
1022 c->ihead_offs);
1023 return -EINVAL;
1024 }
1025
1026 dbg_mnt("start replaying the journal");
1027 c->replaying = 1;
1028 lnum = c->ltail_lnum = c->lhead_lnum;
1029 offs = c->lhead_offs;
1030
1031 for (i = 0; i < c->log_lebs; i++, lnum++) {
1032 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs) {
1033 /*
1034 * The log is logically circular, we reached the last
1035 * LEB, switch to the first one.
1036 */
1037 lnum = UBIFS_LOG_LNUM;
1038 offs = 0;
1039 }
1040 err = replay_log_leb(c, lnum, offs, c->sbuf);
1041 if (err == 1)
1042 /* We hit the end of the log */
1043 break;
1044 if (err)
1045 goto out;
1046 offs = 0;
1047 }
1048
1049 err = replay_buds(c);
1050 if (err)
1051 goto out;
1052
1053 err = apply_replay_list(c);
1054 if (err)
1055 goto out;
1056
1057 err = set_buds_lprops(c);
1058 if (err)
1059 goto out;
1060
1061 /*
1062 * UBIFS budgeting calculations use @c->bi.uncommitted_idx variable
1063 * to roughly estimate index growth. Things like @c->bi.min_idx_lebs
1064 * depend on it. This means we have to initialize it to make sure
1065 * budgeting works properly.
1066 */
1067 c->bi.uncommitted_idx = atomic_long_read(&c->dirty_zn_cnt);
1068 c->bi.uncommitted_idx *= c->max_idx_node_sz;
1069
1070 ubifs_assert(c->bud_bytes <= c->max_bud_bytes || c->need_recovery);
1071 dbg_mnt("finished, log head LEB %d:%d, max_sqnum %llu, "
1072 "highest_inum %lu", c->lhead_lnum, c->lhead_offs, c->max_sqnum,
1073 (unsigned long)c->highest_inum);
1074out:
1075 destroy_replay_list(c);
1076 destroy_bud_list(c);
1077 c->replaying = 0;
1078 return err;
1079}
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#include <linux/list_sort.h>
37
38/**
39 * struct replay_entry - replay list entry.
40 * @lnum: logical eraseblock number of the node
41 * @offs: node offset
42 * @len: node length
43 * @deletion: non-zero if this entry corresponds to a node deletion
44 * @sqnum: node sequence number
45 * @list: links the replay list
46 * @key: node key
47 * @nm: directory entry name
48 * @old_size: truncation old size
49 * @new_size: truncation new size
50 *
51 * The replay process first scans all buds and builds the replay list, then
52 * sorts the replay list in nodes sequence number order, and then inserts all
53 * the replay entries to the TNC.
54 */
55struct replay_entry {
56 int lnum;
57 int offs;
58 int len;
59 unsigned int deletion:1;
60 unsigned long long sqnum;
61 struct list_head list;
62 union ubifs_key key;
63 union {
64 struct qstr nm;
65 struct {
66 loff_t old_size;
67 loff_t new_size;
68 };
69 };
70};
71
72/**
73 * struct bud_entry - entry in the list of buds to replay.
74 * @list: next bud in the list
75 * @bud: bud description object
76 * @sqnum: reference node sequence number
77 * @free: free bytes in the bud
78 * @dirty: dirty bytes in the bud
79 */
80struct bud_entry {
81 struct list_head list;
82 struct ubifs_bud *bud;
83 unsigned long long sqnum;
84 int free;
85 int dirty;
86};
87
88/**
89 * set_bud_lprops - set free and dirty space used by a bud.
90 * @c: UBIFS file-system description object
91 * @b: bud entry which describes the bud
92 *
93 * This function makes sure the LEB properties of bud @b are set correctly
94 * after the replay. Returns zero in case of success and a negative error code
95 * in case of failure.
96 */
97static int set_bud_lprops(struct ubifs_info *c, struct bud_entry *b)
98{
99 const struct ubifs_lprops *lp;
100 int err = 0, dirty;
101
102 ubifs_get_lprops(c);
103
104 lp = ubifs_lpt_lookup_dirty(c, b->bud->lnum);
105 if (IS_ERR(lp)) {
106 err = PTR_ERR(lp);
107 goto out;
108 }
109
110 dirty = lp->dirty;
111 if (b->bud->start == 0 && (lp->free != c->leb_size || lp->dirty != 0)) {
112 /*
113 * The LEB was added to the journal with a starting offset of
114 * zero which means the LEB must have been empty. The LEB
115 * property values should be @lp->free == @c->leb_size and
116 * @lp->dirty == 0, but that is not the case. The reason is that
117 * the LEB had been garbage collected before it became the bud,
118 * and there was not commit inbetween. The garbage collector
119 * resets the free and dirty space without recording it
120 * anywhere except lprops, so if there was no commit then
121 * lprops does not have that information.
122 *
123 * We do not need to adjust free space because the scan has told
124 * us the exact value which is recorded in the replay entry as
125 * @b->free.
126 *
127 * However we do need to subtract from the dirty space the
128 * amount of space that the garbage collector reclaimed, which
129 * is the whole LEB minus the amount of space that was free.
130 */
131 dbg_mnt("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
132 lp->free, lp->dirty);
133 dbg_gc("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
134 lp->free, lp->dirty);
135 dirty -= c->leb_size - lp->free;
136 /*
137 * If the replay order was perfect the dirty space would now be
138 * zero. The order is not perfect because the journal heads
139 * race with each other. This is not a problem but is does mean
140 * that the dirty space may temporarily exceed c->leb_size
141 * during the replay.
142 */
143 if (dirty != 0)
144 dbg_msg("LEB %d lp: %d free %d dirty "
145 "replay: %d free %d dirty", b->bud->lnum,
146 lp->free, lp->dirty, b->free, b->dirty);
147 }
148 lp = ubifs_change_lp(c, lp, b->free, dirty + b->dirty,
149 lp->flags | LPROPS_TAKEN, 0);
150 if (IS_ERR(lp)) {
151 err = PTR_ERR(lp);
152 goto out;
153 }
154
155 /* Make sure the journal head points to the latest bud */
156 err = ubifs_wbuf_seek_nolock(&c->jheads[b->bud->jhead].wbuf,
157 b->bud->lnum, c->leb_size - b->free);
158
159out:
160 ubifs_release_lprops(c);
161 return err;
162}
163
164/**
165 * set_buds_lprops - set free and dirty space for all replayed buds.
166 * @c: UBIFS file-system description object
167 *
168 * This function sets LEB properties for all replayed buds. Returns zero in
169 * case of success and a negative error code in case of failure.
170 */
171static int set_buds_lprops(struct ubifs_info *c)
172{
173 struct bud_entry *b;
174 int err;
175
176 list_for_each_entry(b, &c->replay_buds, list) {
177 err = set_bud_lprops(c, b);
178 if (err)
179 return err;
180 }
181
182 return 0;
183}
184
185/**
186 * trun_remove_range - apply a replay entry for a truncation to the TNC.
187 * @c: UBIFS file-system description object
188 * @r: replay entry of truncation
189 */
190static int trun_remove_range(struct ubifs_info *c, struct replay_entry *r)
191{
192 unsigned min_blk, max_blk;
193 union ubifs_key min_key, max_key;
194 ino_t ino;
195
196 min_blk = r->new_size / UBIFS_BLOCK_SIZE;
197 if (r->new_size & (UBIFS_BLOCK_SIZE - 1))
198 min_blk += 1;
199
200 max_blk = r->old_size / UBIFS_BLOCK_SIZE;
201 if ((r->old_size & (UBIFS_BLOCK_SIZE - 1)) == 0)
202 max_blk -= 1;
203
204 ino = key_inum(c, &r->key);
205
206 data_key_init(c, &min_key, ino, min_blk);
207 data_key_init(c, &max_key, ino, max_blk);
208
209 return ubifs_tnc_remove_range(c, &min_key, &max_key);
210}
211
212/**
213 * apply_replay_entry - apply a replay entry to the TNC.
214 * @c: UBIFS file-system description object
215 * @r: replay entry to apply
216 *
217 * Apply a replay entry to the TNC.
218 */
219static int apply_replay_entry(struct ubifs_info *c, struct replay_entry *r)
220{
221 int err;
222
223 dbg_mntk(&r->key, "LEB %d:%d len %d deletion %d sqnum %llu key ",
224 r->lnum, r->offs, r->len, r->deletion, r->sqnum);
225
226 /* Set c->replay_sqnum to help deal with dangling branches. */
227 c->replay_sqnum = r->sqnum;
228
229 if (is_hash_key(c, &r->key)) {
230 if (r->deletion)
231 err = ubifs_tnc_remove_nm(c, &r->key, &r->nm);
232 else
233 err = ubifs_tnc_add_nm(c, &r->key, r->lnum, r->offs,
234 r->len, &r->nm);
235 } else {
236 if (r->deletion)
237 switch (key_type(c, &r->key)) {
238 case UBIFS_INO_KEY:
239 {
240 ino_t inum = key_inum(c, &r->key);
241
242 err = ubifs_tnc_remove_ino(c, inum);
243 break;
244 }
245 case UBIFS_TRUN_KEY:
246 err = trun_remove_range(c, r);
247 break;
248 default:
249 err = ubifs_tnc_remove(c, &r->key);
250 break;
251 }
252 else
253 err = ubifs_tnc_add(c, &r->key, r->lnum, r->offs,
254 r->len);
255 if (err)
256 return err;
257
258 if (c->need_recovery)
259 err = ubifs_recover_size_accum(c, &r->key, r->deletion,
260 r->new_size);
261 }
262
263 return err;
264}
265
266/**
267 * replay_entries_cmp - compare 2 replay entries.
268 * @priv: UBIFS file-system description object
269 * @a: first replay entry
270 * @a: second replay entry
271 *
272 * This is a comparios function for 'list_sort()' which compares 2 replay
273 * entries @a and @b by comparing their sequence numer. Returns %1 if @a has
274 * greater sequence number and %-1 otherwise.
275 */
276static int replay_entries_cmp(void *priv, struct list_head *a,
277 struct list_head *b)
278{
279 struct replay_entry *ra, *rb;
280
281 cond_resched();
282 if (a == b)
283 return 0;
284
285 ra = list_entry(a, struct replay_entry, list);
286 rb = list_entry(b, struct replay_entry, list);
287 ubifs_assert(ra->sqnum != rb->sqnum);
288 if (ra->sqnum > rb->sqnum)
289 return 1;
290 return -1;
291}
292
293/**
294 * apply_replay_list - apply the replay list to the TNC.
295 * @c: UBIFS file-system description object
296 *
297 * Apply all entries in the replay list to the TNC. Returns zero in case of
298 * success and a negative error code in case of failure.
299 */
300static int apply_replay_list(struct ubifs_info *c)
301{
302 struct replay_entry *r;
303 int err;
304
305 list_sort(c, &c->replay_list, &replay_entries_cmp);
306
307 list_for_each_entry(r, &c->replay_list, list) {
308 cond_resched();
309
310 err = apply_replay_entry(c, r);
311 if (err)
312 return err;
313 }
314
315 return 0;
316}
317
318/**
319 * destroy_replay_list - destroy the replay.
320 * @c: UBIFS file-system description object
321 *
322 * Destroy the replay list.
323 */
324static void destroy_replay_list(struct ubifs_info *c)
325{
326 struct replay_entry *r, *tmp;
327
328 list_for_each_entry_safe(r, tmp, &c->replay_list, list) {
329 if (is_hash_key(c, &r->key))
330 kfree(r->nm.name);
331 list_del(&r->list);
332 kfree(r);
333 }
334}
335
336/**
337 * insert_node - insert a node to the replay list
338 * @c: UBIFS file-system description object
339 * @lnum: node logical eraseblock number
340 * @offs: node offset
341 * @len: node length
342 * @key: node key
343 * @sqnum: sequence number
344 * @deletion: non-zero if this is a deletion
345 * @used: number of bytes in use in a LEB
346 * @old_size: truncation old size
347 * @new_size: truncation new size
348 *
349 * This function inserts a scanned non-direntry node to the replay list. The
350 * replay list contains @struct replay_entry elements, and we sort this list in
351 * sequence number order before applying it. The replay list is applied at the
352 * very end of the replay process. Since the list is sorted in sequence number
353 * order, the older modifications are applied first. This function returns zero
354 * in case of success and a negative error code in case of failure.
355 */
356static int insert_node(struct ubifs_info *c, int lnum, int offs, int len,
357 union ubifs_key *key, unsigned long long sqnum,
358 int deletion, int *used, loff_t old_size,
359 loff_t new_size)
360{
361 struct replay_entry *r;
362
363 dbg_mntk(key, "add LEB %d:%d, key ", lnum, offs);
364
365 if (key_inum(c, key) >= c->highest_inum)
366 c->highest_inum = key_inum(c, key);
367
368 r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
369 if (!r)
370 return -ENOMEM;
371
372 if (!deletion)
373 *used += ALIGN(len, 8);
374 r->lnum = lnum;
375 r->offs = offs;
376 r->len = len;
377 r->deletion = !!deletion;
378 r->sqnum = sqnum;
379 key_copy(c, key, &r->key);
380 r->old_size = old_size;
381 r->new_size = new_size;
382
383 list_add_tail(&r->list, &c->replay_list);
384 return 0;
385}
386
387/**
388 * insert_dent - insert a directory entry node into the replay list.
389 * @c: UBIFS file-system description object
390 * @lnum: node logical eraseblock number
391 * @offs: node offset
392 * @len: node length
393 * @key: node key
394 * @name: directory entry name
395 * @nlen: directory entry name length
396 * @sqnum: sequence number
397 * @deletion: non-zero if this is a deletion
398 * @used: number of bytes in use in a LEB
399 *
400 * This function inserts a scanned directory entry node or an extended
401 * attribute entry to the replay list. Returns zero in case of success and a
402 * negative error code in case of failure.
403 */
404static int insert_dent(struct ubifs_info *c, int lnum, int offs, int len,
405 union ubifs_key *key, const char *name, int nlen,
406 unsigned long long sqnum, int deletion, int *used)
407{
408 struct replay_entry *r;
409 char *nbuf;
410
411 dbg_mntk(key, "add LEB %d:%d, key ", lnum, offs);
412 if (key_inum(c, key) >= c->highest_inum)
413 c->highest_inum = key_inum(c, key);
414
415 r = kzalloc(sizeof(struct replay_entry), GFP_KERNEL);
416 if (!r)
417 return -ENOMEM;
418
419 nbuf = kmalloc(nlen + 1, GFP_KERNEL);
420 if (!nbuf) {
421 kfree(r);
422 return -ENOMEM;
423 }
424
425 if (!deletion)
426 *used += ALIGN(len, 8);
427 r->lnum = lnum;
428 r->offs = offs;
429 r->len = len;
430 r->deletion = !!deletion;
431 r->sqnum = sqnum;
432 key_copy(c, key, &r->key);
433 r->nm.len = nlen;
434 memcpy(nbuf, name, nlen);
435 nbuf[nlen] = '\0';
436 r->nm.name = nbuf;
437
438 list_add_tail(&r->list, &c->replay_list);
439 return 0;
440}
441
442/**
443 * ubifs_validate_entry - validate directory or extended attribute entry node.
444 * @c: UBIFS file-system description object
445 * @dent: the node to validate
446 *
447 * This function validates directory or extended attribute entry node @dent.
448 * Returns zero if the node is all right and a %-EINVAL if not.
449 */
450int ubifs_validate_entry(struct ubifs_info *c,
451 const struct ubifs_dent_node *dent)
452{
453 int key_type = key_type_flash(c, dent->key);
454 int nlen = le16_to_cpu(dent->nlen);
455
456 if (le32_to_cpu(dent->ch.len) != nlen + UBIFS_DENT_NODE_SZ + 1 ||
457 dent->type >= UBIFS_ITYPES_CNT ||
458 nlen > UBIFS_MAX_NLEN || dent->name[nlen] != 0 ||
459 strnlen(dent->name, nlen) != nlen ||
460 le64_to_cpu(dent->inum) > MAX_INUM) {
461 ubifs_err("bad %s node", key_type == UBIFS_DENT_KEY ?
462 "directory entry" : "extended attribute entry");
463 return -EINVAL;
464 }
465
466 if (key_type != UBIFS_DENT_KEY && key_type != UBIFS_XENT_KEY) {
467 ubifs_err("bad key type %d", key_type);
468 return -EINVAL;
469 }
470
471 return 0;
472}
473
474/**
475 * is_last_bud - check if the bud is the last in the journal head.
476 * @c: UBIFS file-system description object
477 * @bud: bud description object
478 *
479 * This function checks if bud @bud is the last bud in its journal head. This
480 * information is then used by 'replay_bud()' to decide whether the bud can
481 * have corruptions or not. Indeed, only last buds can be corrupted by power
482 * cuts. Returns %1 if this is the last bud, and %0 if not.
483 */
484static int is_last_bud(struct ubifs_info *c, struct ubifs_bud *bud)
485{
486 struct ubifs_jhead *jh = &c->jheads[bud->jhead];
487 struct ubifs_bud *next;
488 uint32_t data;
489 int err;
490
491 if (list_is_last(&bud->list, &jh->buds_list))
492 return 1;
493
494 /*
495 * The following is a quirk to make sure we work correctly with UBIFS
496 * images used with older UBIFS.
497 *
498 * Normally, the last bud will be the last in the journal head's list
499 * of bud. However, there is one exception if the UBIFS image belongs
500 * to older UBIFS. This is fairly unlikely: one would need to use old
501 * UBIFS, then have a power cut exactly at the right point, and then
502 * try to mount this image with new UBIFS.
503 *
504 * The exception is: it is possible to have 2 buds A and B, A goes
505 * before B, and B is the last, bud B is contains no data, and bud A is
506 * corrupted at the end. The reason is that in older versions when the
507 * journal code switched the next bud (from A to B), it first added a
508 * log reference node for the new bud (B), and only after this it
509 * synchronized the write-buffer of current bud (A). But later this was
510 * changed and UBIFS started to always synchronize the write-buffer of
511 * the bud (A) before writing the log reference for the new bud (B).
512 *
513 * But because older UBIFS always synchronized A's write-buffer before
514 * writing to B, we can recognize this exceptional situation but
515 * checking the contents of bud B - if it is empty, then A can be
516 * treated as the last and we can recover it.
517 *
518 * TODO: remove this piece of code in a couple of years (today it is
519 * 16.05.2011).
520 */
521 next = list_entry(bud->list.next, struct ubifs_bud, list);
522 if (!list_is_last(&next->list, &jh->buds_list))
523 return 0;
524
525 err = ubifs_leb_read(c, next->lnum, (char *)&data, next->start, 4, 1);
526 if (err)
527 return 0;
528
529 return data == 0xFFFFFFFF;
530}
531
532/**
533 * replay_bud - replay a bud logical eraseblock.
534 * @c: UBIFS file-system description object
535 * @b: bud entry which describes the bud
536 *
537 * This function replays bud @bud, recovers it if needed, and adds all nodes
538 * from this bud to the replay list. Returns zero in case of success and a
539 * negative error code in case of failure.
540 */
541static int replay_bud(struct ubifs_info *c, struct bud_entry *b)
542{
543 int is_last = is_last_bud(c, b->bud);
544 int err = 0, used = 0, lnum = b->bud->lnum, offs = b->bud->start;
545 struct ubifs_scan_leb *sleb;
546 struct ubifs_scan_node *snod;
547
548 dbg_mnt("replay bud LEB %d, head %d, offs %d, is_last %d",
549 lnum, b->bud->jhead, offs, is_last);
550
551 if (c->need_recovery && is_last)
552 /*
553 * Recover only last LEBs in the journal heads, because power
554 * cuts may cause corruptions only in these LEBs, because only
555 * these LEBs could possibly be written to at the power cut
556 * time.
557 */
558 sleb = ubifs_recover_leb(c, lnum, offs, c->sbuf, b->bud->jhead);
559 else
560 sleb = ubifs_scan(c, lnum, offs, c->sbuf, 0);
561 if (IS_ERR(sleb))
562 return PTR_ERR(sleb);
563
564 /*
565 * The bud does not have to start from offset zero - the beginning of
566 * the 'lnum' LEB may contain previously committed data. One of the
567 * things we have to do in replay is to correctly update lprops with
568 * newer information about this LEB.
569 *
570 * At this point lprops thinks that this LEB has 'c->leb_size - offs'
571 * bytes of free space because it only contain information about
572 * committed data.
573 *
574 * But we know that real amount of free space is 'c->leb_size -
575 * sleb->endpt', and the space in the 'lnum' LEB between 'offs' and
576 * 'sleb->endpt' is used by bud data. We have to correctly calculate
577 * how much of these data are dirty and update lprops with this
578 * information.
579 *
580 * The dirt in that LEB region is comprised of padding nodes, deletion
581 * nodes, truncation nodes and nodes which are obsoleted by subsequent
582 * nodes in this LEB. So instead of calculating clean space, we
583 * calculate used space ('used' variable).
584 */
585
586 list_for_each_entry(snod, &sleb->nodes, list) {
587 int deletion = 0;
588
589 cond_resched();
590
591 if (snod->sqnum >= SQNUM_WATERMARK) {
592 ubifs_err("file system's life ended");
593 goto out_dump;
594 }
595
596 if (snod->sqnum > c->max_sqnum)
597 c->max_sqnum = snod->sqnum;
598
599 switch (snod->type) {
600 case UBIFS_INO_NODE:
601 {
602 struct ubifs_ino_node *ino = snod->node;
603 loff_t new_size = le64_to_cpu(ino->size);
604
605 if (le32_to_cpu(ino->nlink) == 0)
606 deletion = 1;
607 err = insert_node(c, lnum, snod->offs, snod->len,
608 &snod->key, snod->sqnum, deletion,
609 &used, 0, new_size);
610 break;
611 }
612 case UBIFS_DATA_NODE:
613 {
614 struct ubifs_data_node *dn = snod->node;
615 loff_t new_size = le32_to_cpu(dn->size) +
616 key_block(c, &snod->key) *
617 UBIFS_BLOCK_SIZE;
618
619 err = insert_node(c, lnum, snod->offs, snod->len,
620 &snod->key, snod->sqnum, deletion,
621 &used, 0, new_size);
622 break;
623 }
624 case UBIFS_DENT_NODE:
625 case UBIFS_XENT_NODE:
626 {
627 struct ubifs_dent_node *dent = snod->node;
628
629 err = ubifs_validate_entry(c, dent);
630 if (err)
631 goto out_dump;
632
633 err = insert_dent(c, lnum, snod->offs, snod->len,
634 &snod->key, dent->name,
635 le16_to_cpu(dent->nlen), snod->sqnum,
636 !le64_to_cpu(dent->inum), &used);
637 break;
638 }
639 case UBIFS_TRUN_NODE:
640 {
641 struct ubifs_trun_node *trun = snod->node;
642 loff_t old_size = le64_to_cpu(trun->old_size);
643 loff_t new_size = le64_to_cpu(trun->new_size);
644 union ubifs_key key;
645
646 /* Validate truncation node */
647 if (old_size < 0 || old_size > c->max_inode_sz ||
648 new_size < 0 || new_size > c->max_inode_sz ||
649 old_size <= new_size) {
650 ubifs_err("bad truncation node");
651 goto out_dump;
652 }
653
654 /*
655 * Create a fake truncation key just to use the same
656 * functions which expect nodes to have keys.
657 */
658 trun_key_init(c, &key, le32_to_cpu(trun->inum));
659 err = insert_node(c, lnum, snod->offs, snod->len,
660 &key, snod->sqnum, 1, &used,
661 old_size, new_size);
662 break;
663 }
664 default:
665 ubifs_err("unexpected node type %d in bud LEB %d:%d",
666 snod->type, lnum, snod->offs);
667 err = -EINVAL;
668 goto out_dump;
669 }
670 if (err)
671 goto out;
672 }
673
674 ubifs_assert(ubifs_search_bud(c, lnum));
675 ubifs_assert(sleb->endpt - offs >= used);
676 ubifs_assert(sleb->endpt % c->min_io_size == 0);
677
678 b->dirty = sleb->endpt - offs - used;
679 b->free = c->leb_size - sleb->endpt;
680 dbg_mnt("bud LEB %d replied: dirty %d, free %d", lnum, b->dirty, b->free);
681
682out:
683 ubifs_scan_destroy(sleb);
684 return err;
685
686out_dump:
687 ubifs_err("bad node is at LEB %d:%d", lnum, snod->offs);
688 ubifs_dump_node(c, snod->node);
689 ubifs_scan_destroy(sleb);
690 return -EINVAL;
691}
692
693/**
694 * replay_buds - replay all buds.
695 * @c: UBIFS file-system description object
696 *
697 * This function returns zero in case of success and a negative error code in
698 * case of failure.
699 */
700static int replay_buds(struct ubifs_info *c)
701{
702 struct bud_entry *b;
703 int err;
704 unsigned long long prev_sqnum = 0;
705
706 list_for_each_entry(b, &c->replay_buds, list) {
707 err = replay_bud(c, b);
708 if (err)
709 return err;
710
711 ubifs_assert(b->sqnum > prev_sqnum);
712 prev_sqnum = b->sqnum;
713 }
714
715 return 0;
716}
717
718/**
719 * destroy_bud_list - destroy the list of buds to replay.
720 * @c: UBIFS file-system description object
721 */
722static void destroy_bud_list(struct ubifs_info *c)
723{
724 struct bud_entry *b;
725
726 while (!list_empty(&c->replay_buds)) {
727 b = list_entry(c->replay_buds.next, struct bud_entry, list);
728 list_del(&b->list);
729 kfree(b);
730 }
731}
732
733/**
734 * add_replay_bud - add a bud to the list of buds to replay.
735 * @c: UBIFS file-system description object
736 * @lnum: bud logical eraseblock number to replay
737 * @offs: bud start offset
738 * @jhead: journal head to which this bud belongs
739 * @sqnum: reference node sequence number
740 *
741 * This function returns zero in case of success and a negative error code in
742 * case of failure.
743 */
744static int add_replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead,
745 unsigned long long sqnum)
746{
747 struct ubifs_bud *bud;
748 struct bud_entry *b;
749
750 dbg_mnt("add replay bud LEB %d:%d, head %d", lnum, offs, jhead);
751
752 bud = kmalloc(sizeof(struct ubifs_bud), GFP_KERNEL);
753 if (!bud)
754 return -ENOMEM;
755
756 b = kmalloc(sizeof(struct bud_entry), GFP_KERNEL);
757 if (!b) {
758 kfree(bud);
759 return -ENOMEM;
760 }
761
762 bud->lnum = lnum;
763 bud->start = offs;
764 bud->jhead = jhead;
765 ubifs_add_bud(c, bud);
766
767 b->bud = bud;
768 b->sqnum = sqnum;
769 list_add_tail(&b->list, &c->replay_buds);
770
771 return 0;
772}
773
774/**
775 * validate_ref - validate a reference node.
776 * @c: UBIFS file-system description object
777 * @ref: the reference node to validate
778 * @ref_lnum: LEB number of the reference node
779 * @ref_offs: reference node offset
780 *
781 * This function returns %1 if a bud reference already exists for the LEB. %0 is
782 * returned if the reference node is new, otherwise %-EINVAL is returned if
783 * validation failed.
784 */
785static int validate_ref(struct ubifs_info *c, const struct ubifs_ref_node *ref)
786{
787 struct ubifs_bud *bud;
788 int lnum = le32_to_cpu(ref->lnum);
789 unsigned int offs = le32_to_cpu(ref->offs);
790 unsigned int jhead = le32_to_cpu(ref->jhead);
791
792 /*
793 * ref->offs may point to the end of LEB when the journal head points
794 * to the end of LEB and we write reference node for it during commit.
795 * So this is why we require 'offs > c->leb_size'.
796 */
797 if (jhead >= c->jhead_cnt || lnum >= c->leb_cnt ||
798 lnum < c->main_first || offs > c->leb_size ||
799 offs & (c->min_io_size - 1))
800 return -EINVAL;
801
802 /* Make sure we have not already looked at this bud */
803 bud = ubifs_search_bud(c, lnum);
804 if (bud) {
805 if (bud->jhead == jhead && bud->start <= offs)
806 return 1;
807 ubifs_err("bud at LEB %d:%d was already referred", lnum, offs);
808 return -EINVAL;
809 }
810
811 return 0;
812}
813
814/**
815 * replay_log_leb - replay a log logical eraseblock.
816 * @c: UBIFS file-system description object
817 * @lnum: log logical eraseblock to replay
818 * @offs: offset to start replaying from
819 * @sbuf: scan buffer
820 *
821 * This function replays a log LEB and returns zero in case of success, %1 if
822 * this is the last LEB in the log, and a negative error code in case of
823 * failure.
824 */
825static int replay_log_leb(struct ubifs_info *c, int lnum, int offs, void *sbuf)
826{
827 int err;
828 struct ubifs_scan_leb *sleb;
829 struct ubifs_scan_node *snod;
830 const struct ubifs_cs_node *node;
831
832 dbg_mnt("replay log LEB %d:%d", lnum, offs);
833 sleb = ubifs_scan(c, lnum, offs, sbuf, c->need_recovery);
834 if (IS_ERR(sleb)) {
835 if (PTR_ERR(sleb) != -EUCLEAN || !c->need_recovery)
836 return PTR_ERR(sleb);
837 /*
838 * Note, the below function will recover this log LEB only if
839 * it is the last, because unclean reboots can possibly corrupt
840 * only the tail of the log.
841 */
842 sleb = ubifs_recover_log_leb(c, lnum, offs, sbuf);
843 if (IS_ERR(sleb))
844 return PTR_ERR(sleb);
845 }
846
847 if (sleb->nodes_cnt == 0) {
848 err = 1;
849 goto out;
850 }
851
852 node = sleb->buf;
853 snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);
854 if (c->cs_sqnum == 0) {
855 /*
856 * This is the first log LEB we are looking at, make sure that
857 * the first node is a commit start node. Also record its
858 * sequence number so that UBIFS can determine where the log
859 * ends, because all nodes which were have higher sequence
860 * numbers.
861 */
862 if (snod->type != UBIFS_CS_NODE) {
863 ubifs_err("first log node at LEB %d:%d is not CS node",
864 lnum, offs);
865 goto out_dump;
866 }
867 if (le64_to_cpu(node->cmt_no) != c->cmt_no) {
868 ubifs_err("first CS node at LEB %d:%d has wrong "
869 "commit number %llu expected %llu",
870 lnum, offs,
871 (unsigned long long)le64_to_cpu(node->cmt_no),
872 c->cmt_no);
873 goto out_dump;
874 }
875
876 c->cs_sqnum = le64_to_cpu(node->ch.sqnum);
877 dbg_mnt("commit start sqnum %llu", c->cs_sqnum);
878 }
879
880 if (snod->sqnum < c->cs_sqnum) {
881 /*
882 * This means that we reached end of log and now
883 * look to the older log data, which was already
884 * committed but the eraseblock was not erased (UBIFS
885 * only un-maps it). So this basically means we have to
886 * exit with "end of log" code.
887 */
888 err = 1;
889 goto out;
890 }
891
892 /* Make sure the first node sits at offset zero of the LEB */
893 if (snod->offs != 0) {
894 ubifs_err("first node is not at zero offset");
895 goto out_dump;
896 }
897
898 list_for_each_entry(snod, &sleb->nodes, list) {
899 cond_resched();
900
901 if (snod->sqnum >= SQNUM_WATERMARK) {
902 ubifs_err("file system's life ended");
903 goto out_dump;
904 }
905
906 if (snod->sqnum < c->cs_sqnum) {
907 ubifs_err("bad sqnum %llu, commit sqnum %llu",
908 snod->sqnum, c->cs_sqnum);
909 goto out_dump;
910 }
911
912 if (snod->sqnum > c->max_sqnum)
913 c->max_sqnum = snod->sqnum;
914
915 switch (snod->type) {
916 case UBIFS_REF_NODE: {
917 const struct ubifs_ref_node *ref = snod->node;
918
919 err = validate_ref(c, ref);
920 if (err == 1)
921 break; /* Already have this bud */
922 if (err)
923 goto out_dump;
924
925 err = add_replay_bud(c, le32_to_cpu(ref->lnum),
926 le32_to_cpu(ref->offs),
927 le32_to_cpu(ref->jhead),
928 snod->sqnum);
929 if (err)
930 goto out;
931
932 break;
933 }
934 case UBIFS_CS_NODE:
935 /* Make sure it sits at the beginning of LEB */
936 if (snod->offs != 0) {
937 ubifs_err("unexpected node in log");
938 goto out_dump;
939 }
940 break;
941 default:
942 ubifs_err("unexpected node in log");
943 goto out_dump;
944 }
945 }
946
947 if (sleb->endpt || c->lhead_offs >= c->leb_size) {
948 c->lhead_lnum = lnum;
949 c->lhead_offs = sleb->endpt;
950 }
951
952 err = !sleb->endpt;
953out:
954 ubifs_scan_destroy(sleb);
955 return err;
956
957out_dump:
958 ubifs_err("log error detected while replaying the log at LEB %d:%d",
959 lnum, offs + snod->offs);
960 ubifs_dump_node(c, snod->node);
961 ubifs_scan_destroy(sleb);
962 return -EINVAL;
963}
964
965/**
966 * take_ihead - update the status of the index head in lprops to 'taken'.
967 * @c: UBIFS file-system description object
968 *
969 * This function returns the amount of free space in the index head LEB or a
970 * negative error code.
971 */
972static int take_ihead(struct ubifs_info *c)
973{
974 const struct ubifs_lprops *lp;
975 int err, free;
976
977 ubifs_get_lprops(c);
978
979 lp = ubifs_lpt_lookup_dirty(c, c->ihead_lnum);
980 if (IS_ERR(lp)) {
981 err = PTR_ERR(lp);
982 goto out;
983 }
984
985 free = lp->free;
986
987 lp = ubifs_change_lp(c, lp, LPROPS_NC, LPROPS_NC,
988 lp->flags | LPROPS_TAKEN, 0);
989 if (IS_ERR(lp)) {
990 err = PTR_ERR(lp);
991 goto out;
992 }
993
994 err = free;
995out:
996 ubifs_release_lprops(c);
997 return err;
998}
999
1000/**
1001 * ubifs_replay_journal - replay journal.
1002 * @c: UBIFS file-system description object
1003 *
1004 * This function scans the journal, replays and cleans it up. It makes sure all
1005 * memory data structures related to uncommitted journal are built (dirty TNC
1006 * tree, tree of buds, modified lprops, etc).
1007 */
1008int ubifs_replay_journal(struct ubifs_info *c)
1009{
1010 int err, i, lnum, offs, free;
1011
1012 BUILD_BUG_ON(UBIFS_TRUN_KEY > 5);
1013
1014 /* Update the status of the index head in lprops to 'taken' */
1015 free = take_ihead(c);
1016 if (free < 0)
1017 return free; /* Error code */
1018
1019 if (c->ihead_offs != c->leb_size - free) {
1020 ubifs_err("bad index head LEB %d:%d", c->ihead_lnum,
1021 c->ihead_offs);
1022 return -EINVAL;
1023 }
1024
1025 dbg_mnt("start replaying the journal");
1026 c->replaying = 1;
1027 lnum = c->ltail_lnum = c->lhead_lnum;
1028 offs = c->lhead_offs;
1029
1030 for (i = 0; i < c->log_lebs; i++, lnum++) {
1031 if (lnum >= UBIFS_LOG_LNUM + c->log_lebs) {
1032 /*
1033 * The log is logically circular, we reached the last
1034 * LEB, switch to the first one.
1035 */
1036 lnum = UBIFS_LOG_LNUM;
1037 offs = 0;
1038 }
1039 err = replay_log_leb(c, lnum, offs, c->sbuf);
1040 if (err == 1)
1041 /* We hit the end of the log */
1042 break;
1043 if (err)
1044 goto out;
1045 offs = 0;
1046 }
1047
1048 err = replay_buds(c);
1049 if (err)
1050 goto out;
1051
1052 err = apply_replay_list(c);
1053 if (err)
1054 goto out;
1055
1056 err = set_buds_lprops(c);
1057 if (err)
1058 goto out;
1059
1060 /*
1061 * UBIFS budgeting calculations use @c->bi.uncommitted_idx variable
1062 * to roughly estimate index growth. Things like @c->bi.min_idx_lebs
1063 * depend on it. This means we have to initialize it to make sure
1064 * budgeting works properly.
1065 */
1066 c->bi.uncommitted_idx = atomic_long_read(&c->dirty_zn_cnt);
1067 c->bi.uncommitted_idx *= c->max_idx_node_sz;
1068
1069 ubifs_assert(c->bud_bytes <= c->max_bud_bytes || c->need_recovery);
1070 dbg_mnt("finished, log head LEB %d:%d, max_sqnum %llu, "
1071 "highest_inum %lu", c->lhead_lnum, c->lhead_offs, c->max_sqnum,
1072 (unsigned long)c->highest_inum);
1073out:
1074 destroy_replay_list(c);
1075 destroy_bud_list(c);
1076 c->replaying = 0;
1077 return err;
1078}