Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * linux/fs/jbd2/recovery.c
4 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 *
7 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
8 *
9 * Journal recovery routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system.
11 */
12
13#ifndef __KERNEL__
14#include "jfs_user.h"
15#else
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/jbd2.h>
19#include <linux/errno.h>
20#include <linux/crc32.h>
21#include <linux/blkdev.h>
22#endif
23
24/*
25 * Maintain information about the progress of the recovery job, so that
26 * the different passes can carry information between them.
27 */
28struct recovery_info
29{
30 tid_t start_transaction;
31 tid_t end_transaction;
32
33 int nr_replays;
34 int nr_revokes;
35 int nr_revoke_hits;
36};
37
38static int do_one_pass(journal_t *journal,
39 struct recovery_info *info, enum passtype pass);
40static int scan_revoke_records(journal_t *, struct buffer_head *,
41 tid_t, struct recovery_info *);
42
43#ifdef __KERNEL__
44
45/* Release readahead buffers after use */
46static void journal_brelse_array(struct buffer_head *b[], int n)
47{
48 while (--n >= 0)
49 brelse (b[n]);
50}
51
52
53/*
54 * When reading from the journal, we are going through the block device
55 * layer directly and so there is no readahead being done for us. We
56 * need to implement any readahead ourselves if we want it to happen at
57 * all. Recovery is basically one long sequential read, so make sure we
58 * do the IO in reasonably large chunks.
59 *
60 * This is not so critical that we need to be enormously clever about
61 * the readahead size, though. 128K is a purely arbitrary, good-enough
62 * fixed value.
63 */
64
65#define MAXBUF 8
66static int do_readahead(journal_t *journal, unsigned int start)
67{
68 int err;
69 unsigned int max, nbufs, next;
70 unsigned long long blocknr;
71 struct buffer_head *bh;
72
73 struct buffer_head * bufs[MAXBUF];
74
75 /* Do up to 128K of readahead */
76 max = start + (128 * 1024 / journal->j_blocksize);
77 if (max > journal->j_total_len)
78 max = journal->j_total_len;
79
80 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
81 * a time to the block device IO layer. */
82
83 nbufs = 0;
84
85 for (next = start; next < max; next++) {
86 err = jbd2_journal_bmap(journal, next, &blocknr);
87
88 if (err) {
89 printk(KERN_ERR "JBD2: bad block at offset %u\n",
90 next);
91 goto failed;
92 }
93
94 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
95 if (!bh) {
96 err = -ENOMEM;
97 goto failed;
98 }
99
100 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
101 bufs[nbufs++] = bh;
102 if (nbufs == MAXBUF) {
103 bh_readahead_batch(nbufs, bufs, 0);
104 journal_brelse_array(bufs, nbufs);
105 nbufs = 0;
106 }
107 } else
108 brelse(bh);
109 }
110
111 if (nbufs)
112 bh_readahead_batch(nbufs, bufs, 0);
113 err = 0;
114
115failed:
116 if (nbufs)
117 journal_brelse_array(bufs, nbufs);
118 return err;
119}
120
121#endif /* __KERNEL__ */
122
123
124/*
125 * Read a block from the journal
126 */
127
128static int jread(struct buffer_head **bhp, journal_t *journal,
129 unsigned int offset)
130{
131 int err;
132 unsigned long long blocknr;
133 struct buffer_head *bh;
134
135 *bhp = NULL;
136
137 if (offset >= journal->j_total_len) {
138 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
139 return -EFSCORRUPTED;
140 }
141
142 err = jbd2_journal_bmap(journal, offset, &blocknr);
143
144 if (err) {
145 printk(KERN_ERR "JBD2: bad block at offset %u\n",
146 offset);
147 return err;
148 }
149
150 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
151 if (!bh)
152 return -ENOMEM;
153
154 if (!buffer_uptodate(bh)) {
155 /*
156 * If this is a brand new buffer, start readahead.
157 * Otherwise, we assume we are already reading it.
158 */
159 bool need_readahead = !buffer_req(bh);
160
161 bh_read_nowait(bh, 0);
162 if (need_readahead)
163 do_readahead(journal, offset);
164 wait_on_buffer(bh);
165 }
166
167 if (!buffer_uptodate(bh)) {
168 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
169 offset);
170 brelse(bh);
171 return -EIO;
172 }
173
174 *bhp = bh;
175 return 0;
176}
177
178static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
179{
180 struct jbd2_journal_block_tail *tail;
181 __be32 provided;
182 __u32 calculated;
183
184 if (!jbd2_journal_has_csum_v2or3(j))
185 return 1;
186
187 tail = (struct jbd2_journal_block_tail *)((char *)buf +
188 j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
189 provided = tail->t_checksum;
190 tail->t_checksum = 0;
191 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
192 tail->t_checksum = provided;
193
194 return provided == cpu_to_be32(calculated);
195}
196
197/*
198 * Count the number of in-use tags in a journal descriptor block.
199 */
200
201static int count_tags(journal_t *journal, struct buffer_head *bh)
202{
203 char * tagp;
204 journal_block_tag_t tag;
205 int nr = 0, size = journal->j_blocksize;
206 int tag_bytes = journal_tag_bytes(journal);
207
208 if (jbd2_journal_has_csum_v2or3(journal))
209 size -= sizeof(struct jbd2_journal_block_tail);
210
211 tagp = &bh->b_data[sizeof(journal_header_t)];
212
213 while ((tagp - bh->b_data + tag_bytes) <= size) {
214 memcpy(&tag, tagp, sizeof(tag));
215
216 nr++;
217 tagp += tag_bytes;
218 if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
219 tagp += 16;
220
221 if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
222 break;
223 }
224
225 return nr;
226}
227
228
229/* Make sure we wrap around the log correctly! */
230#define wrap(journal, var) \
231do { \
232 unsigned long _wrap_last = \
233 jbd2_has_feature_fast_commit(journal) ? \
234 (journal)->j_fc_last : (journal)->j_last; \
235 \
236 if (var >= _wrap_last) \
237 var -= (_wrap_last - (journal)->j_first); \
238} while (0)
239
240static int fc_do_one_pass(journal_t *journal,
241 struct recovery_info *info, enum passtype pass)
242{
243 unsigned int expected_commit_id = info->end_transaction;
244 unsigned long next_fc_block;
245 struct buffer_head *bh;
246 int err = 0;
247
248 next_fc_block = journal->j_fc_first;
249 if (!journal->j_fc_replay_callback)
250 return 0;
251
252 while (next_fc_block <= journal->j_fc_last) {
253 jbd2_debug(3, "Fast commit replay: next block %ld\n",
254 next_fc_block);
255 err = jread(&bh, journal, next_fc_block);
256 if (err) {
257 jbd2_debug(3, "Fast commit replay: read error\n");
258 break;
259 }
260
261 err = journal->j_fc_replay_callback(journal, bh, pass,
262 next_fc_block - journal->j_fc_first,
263 expected_commit_id);
264 brelse(bh);
265 next_fc_block++;
266 if (err < 0 || err == JBD2_FC_REPLAY_STOP)
267 break;
268 err = 0;
269 }
270
271 if (err)
272 jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
273
274 return err;
275}
276
277/**
278 * jbd2_journal_recover - recovers a on-disk journal
279 * @journal: the journal to recover
280 *
281 * The primary function for recovering the log contents when mounting a
282 * journaled device.
283 *
284 * Recovery is done in three passes. In the first pass, we look for the
285 * end of the log. In the second, we assemble the list of revoke
286 * blocks. In the third and final pass, we replay any un-revoked blocks
287 * in the log.
288 */
289int jbd2_journal_recover(journal_t *journal)
290{
291 int err, err2;
292 journal_superblock_t * sb;
293
294 struct recovery_info info;
295
296 memset(&info, 0, sizeof(info));
297 sb = journal->j_superblock;
298
299 /*
300 * The journal superblock's s_start field (the current log head)
301 * is always zero if, and only if, the journal was cleanly
302 * unmounted.
303 */
304
305 if (!sb->s_start) {
306 jbd2_debug(1, "No recovery required, last transaction %d\n",
307 be32_to_cpu(sb->s_sequence));
308 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
309 return 0;
310 }
311
312 err = do_one_pass(journal, &info, PASS_SCAN);
313 if (!err)
314 err = do_one_pass(journal, &info, PASS_REVOKE);
315 if (!err)
316 err = do_one_pass(journal, &info, PASS_REPLAY);
317
318 jbd2_debug(1, "JBD2: recovery, exit status %d, "
319 "recovered transactions %u to %u\n",
320 err, info.start_transaction, info.end_transaction);
321 jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
322 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
323
324 /* Restart the log at the next transaction ID, thus invalidating
325 * any existing commit records in the log. */
326 journal->j_transaction_sequence = ++info.end_transaction;
327
328 jbd2_journal_clear_revoke(journal);
329 err2 = sync_blockdev(journal->j_fs_dev);
330 if (!err)
331 err = err2;
332 /* Make sure all replayed data is on permanent storage */
333 if (journal->j_flags & JBD2_BARRIER) {
334 err2 = blkdev_issue_flush(journal->j_fs_dev);
335 if (!err)
336 err = err2;
337 }
338 return err;
339}
340
341/**
342 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
343 * @journal: journal to startup
344 *
345 * Locate any valid recovery information from the journal and set up the
346 * journal structures in memory to ignore it (presumably because the
347 * caller has evidence that it is out of date).
348 * This function doesn't appear to be exported..
349 *
350 * We perform one pass over the journal to allow us to tell the user how
351 * much recovery information is being erased, and to let us initialise
352 * the journal transaction sequence numbers to the next unused ID.
353 */
354int jbd2_journal_skip_recovery(journal_t *journal)
355{
356 int err;
357
358 struct recovery_info info;
359
360 memset (&info, 0, sizeof(info));
361
362 err = do_one_pass(journal, &info, PASS_SCAN);
363
364 if (err) {
365 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
366 ++journal->j_transaction_sequence;
367 } else {
368#ifdef CONFIG_JBD2_DEBUG
369 int dropped = info.end_transaction -
370 be32_to_cpu(journal->j_superblock->s_sequence);
371 jbd2_debug(1,
372 "JBD2: ignoring %d transaction%s from the journal.\n",
373 dropped, (dropped == 1) ? "" : "s");
374#endif
375 journal->j_transaction_sequence = ++info.end_transaction;
376 }
377
378 journal->j_tail = 0;
379 return err;
380}
381
382static inline unsigned long long read_tag_block(journal_t *journal,
383 journal_block_tag_t *tag)
384{
385 unsigned long long block = be32_to_cpu(tag->t_blocknr);
386 if (jbd2_has_feature_64bit(journal))
387 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
388 return block;
389}
390
391/*
392 * calc_chksums calculates the checksums for the blocks described in the
393 * descriptor block.
394 */
395static int calc_chksums(journal_t *journal, struct buffer_head *bh,
396 unsigned long *next_log_block, __u32 *crc32_sum)
397{
398 int i, num_blks, err;
399 unsigned long io_block;
400 struct buffer_head *obh;
401
402 num_blks = count_tags(journal, bh);
403 /* Calculate checksum of the descriptor block. */
404 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
405
406 for (i = 0; i < num_blks; i++) {
407 io_block = (*next_log_block)++;
408 wrap(journal, *next_log_block);
409 err = jread(&obh, journal, io_block);
410 if (err) {
411 printk(KERN_ERR "JBD2: IO error %d recovering block "
412 "%lu in log\n", err, io_block);
413 return 1;
414 } else {
415 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
416 obh->b_size);
417 }
418 put_bh(obh);
419 }
420 return 0;
421}
422
423static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
424{
425 struct commit_header *h;
426 __be32 provided;
427 __u32 calculated;
428
429 if (!jbd2_journal_has_csum_v2or3(j))
430 return 1;
431
432 h = buf;
433 provided = h->h_chksum[0];
434 h->h_chksum[0] = 0;
435 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
436 h->h_chksum[0] = provided;
437
438 return provided == cpu_to_be32(calculated);
439}
440
441static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
442 journal_block_tag3_t *tag3,
443 void *buf, __u32 sequence)
444{
445 __u32 csum32;
446 __be32 seq;
447
448 if (!jbd2_journal_has_csum_v2or3(j))
449 return 1;
450
451 seq = cpu_to_be32(sequence);
452 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
453 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
454
455 if (jbd2_has_feature_csum3(j))
456 return tag3->t_checksum == cpu_to_be32(csum32);
457 else
458 return tag->t_checksum == cpu_to_be16(csum32);
459}
460
461static int do_one_pass(journal_t *journal,
462 struct recovery_info *info, enum passtype pass)
463{
464 unsigned int first_commit_ID, next_commit_ID;
465 unsigned long next_log_block;
466 int err, success = 0;
467 journal_superblock_t * sb;
468 journal_header_t * tmp;
469 struct buffer_head * bh;
470 unsigned int sequence;
471 int blocktype;
472 int tag_bytes = journal_tag_bytes(journal);
473 __u32 crc32_sum = ~0; /* Transactional Checksums */
474 int descr_csum_size = 0;
475 int block_error = 0;
476 bool need_check_commit_time = false;
477 __u64 last_trans_commit_time = 0, commit_time;
478
479 /*
480 * First thing is to establish what we expect to find in the log
481 * (in terms of transaction IDs), and where (in terms of log
482 * block offsets): query the superblock.
483 */
484
485 sb = journal->j_superblock;
486 next_commit_ID = be32_to_cpu(sb->s_sequence);
487 next_log_block = be32_to_cpu(sb->s_start);
488
489 first_commit_ID = next_commit_ID;
490 if (pass == PASS_SCAN)
491 info->start_transaction = first_commit_ID;
492
493 jbd2_debug(1, "Starting recovery pass %d\n", pass);
494
495 /*
496 * Now we walk through the log, transaction by transaction,
497 * making sure that each transaction has a commit block in the
498 * expected place. Each complete transaction gets replayed back
499 * into the main filesystem.
500 */
501
502 while (1) {
503 int flags;
504 char * tagp;
505 journal_block_tag_t tag;
506 struct buffer_head * obh;
507 struct buffer_head * nbh;
508
509 cond_resched();
510
511 /* If we already know where to stop the log traversal,
512 * check right now that we haven't gone past the end of
513 * the log. */
514
515 if (pass != PASS_SCAN)
516 if (tid_geq(next_commit_ID, info->end_transaction))
517 break;
518
519 jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
520 next_commit_ID, next_log_block,
521 jbd2_has_feature_fast_commit(journal) ?
522 journal->j_fc_last : journal->j_last);
523
524 /* Skip over each chunk of the transaction looking
525 * either the next descriptor block or the final commit
526 * record. */
527
528 jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
529 err = jread(&bh, journal, next_log_block);
530 if (err)
531 goto failed;
532
533 next_log_block++;
534 wrap(journal, next_log_block);
535
536 /* What kind of buffer is it?
537 *
538 * If it is a descriptor block, check that it has the
539 * expected sequence number. Otherwise, we're all done
540 * here. */
541
542 tmp = (journal_header_t *)bh->b_data;
543
544 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
545 brelse(bh);
546 break;
547 }
548
549 blocktype = be32_to_cpu(tmp->h_blocktype);
550 sequence = be32_to_cpu(tmp->h_sequence);
551 jbd2_debug(3, "Found magic %d, sequence %d\n",
552 blocktype, sequence);
553
554 if (sequence != next_commit_ID) {
555 brelse(bh);
556 break;
557 }
558
559 /* OK, we have a valid descriptor block which matches
560 * all of the sequence number checks. What are we going
561 * to do with it? That depends on the pass... */
562
563 switch(blocktype) {
564 case JBD2_DESCRIPTOR_BLOCK:
565 /* Verify checksum first */
566 if (jbd2_journal_has_csum_v2or3(journal))
567 descr_csum_size =
568 sizeof(struct jbd2_journal_block_tail);
569 if (descr_csum_size > 0 &&
570 !jbd2_descriptor_block_csum_verify(journal,
571 bh->b_data)) {
572 /*
573 * PASS_SCAN can see stale blocks due to lazy
574 * journal init. Don't error out on those yet.
575 */
576 if (pass != PASS_SCAN) {
577 pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
578 next_log_block);
579 err = -EFSBADCRC;
580 brelse(bh);
581 goto failed;
582 }
583 need_check_commit_time = true;
584 jbd2_debug(1,
585 "invalid descriptor block found in %lu\n",
586 next_log_block);
587 }
588
589 /* If it is a valid descriptor block, replay it
590 * in pass REPLAY; if journal_checksums enabled, then
591 * calculate checksums in PASS_SCAN, otherwise,
592 * just skip over the blocks it describes. */
593 if (pass != PASS_REPLAY) {
594 if (pass == PASS_SCAN &&
595 jbd2_has_feature_checksum(journal) &&
596 !need_check_commit_time &&
597 !info->end_transaction) {
598 if (calc_chksums(journal, bh,
599 &next_log_block,
600 &crc32_sum)) {
601 put_bh(bh);
602 break;
603 }
604 put_bh(bh);
605 continue;
606 }
607 next_log_block += count_tags(journal, bh);
608 wrap(journal, next_log_block);
609 put_bh(bh);
610 continue;
611 }
612
613 /* A descriptor block: we can now write all of
614 * the data blocks. Yay, useful work is finally
615 * getting done here! */
616
617 tagp = &bh->b_data[sizeof(journal_header_t)];
618 while ((tagp - bh->b_data + tag_bytes)
619 <= journal->j_blocksize - descr_csum_size) {
620 unsigned long io_block;
621
622 memcpy(&tag, tagp, sizeof(tag));
623 flags = be16_to_cpu(tag.t_flags);
624
625 io_block = next_log_block++;
626 wrap(journal, next_log_block);
627 err = jread(&obh, journal, io_block);
628 if (err) {
629 /* Recover what we can, but
630 * report failure at the end. */
631 success = err;
632 printk(KERN_ERR
633 "JBD2: IO error %d recovering "
634 "block %ld in log\n",
635 err, io_block);
636 } else {
637 unsigned long long blocknr;
638
639 J_ASSERT(obh != NULL);
640 blocknr = read_tag_block(journal,
641 &tag);
642
643 /* If the block has been
644 * revoked, then we're all done
645 * here. */
646 if (jbd2_journal_test_revoke
647 (journal, blocknr,
648 next_commit_ID)) {
649 brelse(obh);
650 ++info->nr_revoke_hits;
651 goto skip_write;
652 }
653
654 /* Look for block corruption */
655 if (!jbd2_block_tag_csum_verify(
656 journal, &tag, (journal_block_tag3_t *)tagp,
657 obh->b_data, be32_to_cpu(tmp->h_sequence))) {
658 brelse(obh);
659 success = -EFSBADCRC;
660 printk(KERN_ERR "JBD2: Invalid "
661 "checksum recovering "
662 "data block %llu in "
663 "log\n", blocknr);
664 block_error = 1;
665 goto skip_write;
666 }
667
668 /* Find a buffer for the new
669 * data being restored */
670 nbh = __getblk(journal->j_fs_dev,
671 blocknr,
672 journal->j_blocksize);
673 if (nbh == NULL) {
674 printk(KERN_ERR
675 "JBD2: Out of memory "
676 "during recovery.\n");
677 err = -ENOMEM;
678 brelse(bh);
679 brelse(obh);
680 goto failed;
681 }
682
683 lock_buffer(nbh);
684 memcpy(nbh->b_data, obh->b_data,
685 journal->j_blocksize);
686 if (flags & JBD2_FLAG_ESCAPE) {
687 *((__be32 *)nbh->b_data) =
688 cpu_to_be32(JBD2_MAGIC_NUMBER);
689 }
690
691 BUFFER_TRACE(nbh, "marking dirty");
692 set_buffer_uptodate(nbh);
693 mark_buffer_dirty(nbh);
694 BUFFER_TRACE(nbh, "marking uptodate");
695 ++info->nr_replays;
696 unlock_buffer(nbh);
697 brelse(obh);
698 brelse(nbh);
699 }
700
701 skip_write:
702 tagp += tag_bytes;
703 if (!(flags & JBD2_FLAG_SAME_UUID))
704 tagp += 16;
705
706 if (flags & JBD2_FLAG_LAST_TAG)
707 break;
708 }
709
710 brelse(bh);
711 continue;
712
713 case JBD2_COMMIT_BLOCK:
714 /* How to differentiate between interrupted commit
715 * and journal corruption ?
716 *
717 * {nth transaction}
718 * Checksum Verification Failed
719 * |
720 * ____________________
721 * | |
722 * async_commit sync_commit
723 * | |
724 * | GO TO NEXT "Journal Corruption"
725 * | TRANSACTION
726 * |
727 * {(n+1)th transanction}
728 * |
729 * _______|______________
730 * | |
731 * Commit block found Commit block not found
732 * | |
733 * "Journal Corruption" |
734 * _____________|_________
735 * | |
736 * nth trans corrupt OR nth trans
737 * and (n+1)th interrupted interrupted
738 * before commit block
739 * could reach the disk.
740 * (Cannot find the difference in above
741 * mentioned conditions. Hence assume
742 * "Interrupted Commit".)
743 */
744 commit_time = be64_to_cpu(
745 ((struct commit_header *)bh->b_data)->h_commit_sec);
746 /*
747 * If need_check_commit_time is set, it means we are in
748 * PASS_SCAN and csum verify failed before. If
749 * commit_time is increasing, it's the same journal,
750 * otherwise it is stale journal block, just end this
751 * recovery.
752 */
753 if (need_check_commit_time) {
754 if (commit_time >= last_trans_commit_time) {
755 pr_err("JBD2: Invalid checksum found in transaction %u\n",
756 next_commit_ID);
757 err = -EFSBADCRC;
758 brelse(bh);
759 goto failed;
760 }
761 ignore_crc_mismatch:
762 /*
763 * It likely does not belong to same journal,
764 * just end this recovery with success.
765 */
766 jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
767 next_commit_ID);
768 brelse(bh);
769 goto done;
770 }
771
772 /*
773 * Found an expected commit block: if checksums
774 * are present, verify them in PASS_SCAN; else not
775 * much to do other than move on to the next sequence
776 * number.
777 */
778 if (pass == PASS_SCAN &&
779 jbd2_has_feature_checksum(journal)) {
780 struct commit_header *cbh =
781 (struct commit_header *)bh->b_data;
782 unsigned found_chksum =
783 be32_to_cpu(cbh->h_chksum[0]);
784
785 if (info->end_transaction) {
786 journal->j_failed_commit =
787 info->end_transaction;
788 brelse(bh);
789 break;
790 }
791
792 /* Neither checksum match nor unused? */
793 if (!((crc32_sum == found_chksum &&
794 cbh->h_chksum_type ==
795 JBD2_CRC32_CHKSUM &&
796 cbh->h_chksum_size ==
797 JBD2_CRC32_CHKSUM_SIZE) ||
798 (cbh->h_chksum_type == 0 &&
799 cbh->h_chksum_size == 0 &&
800 found_chksum == 0)))
801 goto chksum_error;
802
803 crc32_sum = ~0;
804 }
805 if (pass == PASS_SCAN &&
806 !jbd2_commit_block_csum_verify(journal,
807 bh->b_data)) {
808 chksum_error:
809 if (commit_time < last_trans_commit_time)
810 goto ignore_crc_mismatch;
811 info->end_transaction = next_commit_ID;
812
813 if (!jbd2_has_feature_async_commit(journal)) {
814 journal->j_failed_commit =
815 next_commit_ID;
816 brelse(bh);
817 break;
818 }
819 }
820 if (pass == PASS_SCAN)
821 last_trans_commit_time = commit_time;
822 brelse(bh);
823 next_commit_ID++;
824 continue;
825
826 case JBD2_REVOKE_BLOCK:
827 /*
828 * Check revoke block crc in pass_scan, if csum verify
829 * failed, check commit block time later.
830 */
831 if (pass == PASS_SCAN &&
832 !jbd2_descriptor_block_csum_verify(journal,
833 bh->b_data)) {
834 jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
835 next_log_block);
836 need_check_commit_time = true;
837 }
838 /* If we aren't in the REVOKE pass, then we can
839 * just skip over this block. */
840 if (pass != PASS_REVOKE) {
841 brelse(bh);
842 continue;
843 }
844
845 err = scan_revoke_records(journal, bh,
846 next_commit_ID, info);
847 brelse(bh);
848 if (err)
849 goto failed;
850 continue;
851
852 default:
853 jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
854 blocktype);
855 brelse(bh);
856 goto done;
857 }
858 }
859
860 done:
861 /*
862 * We broke out of the log scan loop: either we came to the
863 * known end of the log or we found an unexpected block in the
864 * log. If the latter happened, then we know that the "current"
865 * transaction marks the end of the valid log.
866 */
867
868 if (pass == PASS_SCAN) {
869 if (!info->end_transaction)
870 info->end_transaction = next_commit_ID;
871 } else {
872 /* It's really bad news if different passes end up at
873 * different places (but possible due to IO errors). */
874 if (info->end_transaction != next_commit_ID) {
875 printk(KERN_ERR "JBD2: recovery pass %d ended at "
876 "transaction %u, expected %u\n",
877 pass, next_commit_ID, info->end_transaction);
878 if (!success)
879 success = -EIO;
880 }
881 }
882
883 if (jbd2_has_feature_fast_commit(journal) && pass != PASS_REVOKE) {
884 err = fc_do_one_pass(journal, info, pass);
885 if (err)
886 success = err;
887 }
888
889 if (block_error && success == 0)
890 success = -EIO;
891 return success;
892
893 failed:
894 return err;
895}
896
897/* Scan a revoke record, marking all blocks mentioned as revoked. */
898
899static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
900 tid_t sequence, struct recovery_info *info)
901{
902 jbd2_journal_revoke_header_t *header;
903 int offset, max;
904 unsigned csum_size = 0;
905 __u32 rcount;
906 int record_len = 4;
907
908 header = (jbd2_journal_revoke_header_t *) bh->b_data;
909 offset = sizeof(jbd2_journal_revoke_header_t);
910 rcount = be32_to_cpu(header->r_count);
911
912 if (jbd2_journal_has_csum_v2or3(journal))
913 csum_size = sizeof(struct jbd2_journal_block_tail);
914 if (rcount > journal->j_blocksize - csum_size)
915 return -EINVAL;
916 max = rcount;
917
918 if (jbd2_has_feature_64bit(journal))
919 record_len = 8;
920
921 while (offset + record_len <= max) {
922 unsigned long long blocknr;
923 int err;
924
925 if (record_len == 4)
926 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
927 else
928 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
929 offset += record_len;
930 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
931 if (err)
932 return err;
933 ++info->nr_revokes;
934 }
935 return 0;
936}
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * linux/fs/jbd2/recovery.c
4 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 *
7 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
8 *
9 * Journal recovery routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system.
11 */
12
13#ifndef __KERNEL__
14#include "jfs_user.h"
15#else
16#include <linux/time.h>
17#include <linux/fs.h>
18#include <linux/jbd2.h>
19#include <linux/errno.h>
20#include <linux/crc32.h>
21#include <linux/blkdev.h>
22#include <linux/string_choices.h>
23#endif
24
25/*
26 * Maintain information about the progress of the recovery job, so that
27 * the different passes can carry information between them.
28 */
29struct recovery_info
30{
31 tid_t start_transaction;
32 tid_t end_transaction;
33 unsigned long head_block;
34
35 int nr_replays;
36 int nr_revokes;
37 int nr_revoke_hits;
38};
39
40static int do_one_pass(journal_t *journal,
41 struct recovery_info *info, enum passtype pass);
42static int scan_revoke_records(journal_t *, struct buffer_head *,
43 tid_t, struct recovery_info *);
44
45#ifdef __KERNEL__
46
47/* Release readahead buffers after use */
48static void journal_brelse_array(struct buffer_head *b[], int n)
49{
50 while (--n >= 0)
51 brelse (b[n]);
52}
53
54
55/*
56 * When reading from the journal, we are going through the block device
57 * layer directly and so there is no readahead being done for us. We
58 * need to implement any readahead ourselves if we want it to happen at
59 * all. Recovery is basically one long sequential read, so make sure we
60 * do the IO in reasonably large chunks.
61 *
62 * This is not so critical that we need to be enormously clever about
63 * the readahead size, though. 128K is a purely arbitrary, good-enough
64 * fixed value.
65 */
66
67#define MAXBUF 8
68static int do_readahead(journal_t *journal, unsigned int start)
69{
70 int err;
71 unsigned int max, nbufs, next;
72 unsigned long long blocknr;
73 struct buffer_head *bh;
74
75 struct buffer_head * bufs[MAXBUF];
76
77 /* Do up to 128K of readahead */
78 max = start + (128 * 1024 / journal->j_blocksize);
79 if (max > journal->j_total_len)
80 max = journal->j_total_len;
81
82 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
83 * a time to the block device IO layer. */
84
85 nbufs = 0;
86
87 for (next = start; next < max; next++) {
88 err = jbd2_journal_bmap(journal, next, &blocknr);
89
90 if (err) {
91 printk(KERN_ERR "JBD2: bad block at offset %u\n",
92 next);
93 goto failed;
94 }
95
96 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
97 if (!bh) {
98 err = -ENOMEM;
99 goto failed;
100 }
101
102 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
103 bufs[nbufs++] = bh;
104 if (nbufs == MAXBUF) {
105 bh_readahead_batch(nbufs, bufs, 0);
106 journal_brelse_array(bufs, nbufs);
107 nbufs = 0;
108 }
109 } else
110 brelse(bh);
111 }
112
113 if (nbufs)
114 bh_readahead_batch(nbufs, bufs, 0);
115 err = 0;
116
117failed:
118 if (nbufs)
119 journal_brelse_array(bufs, nbufs);
120 return err;
121}
122
123#endif /* __KERNEL__ */
124
125
126/*
127 * Read a block from the journal
128 */
129
130static int jread(struct buffer_head **bhp, journal_t *journal,
131 unsigned int offset)
132{
133 int err;
134 unsigned long long blocknr;
135 struct buffer_head *bh;
136
137 *bhp = NULL;
138
139 if (offset >= journal->j_total_len) {
140 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
141 return -EFSCORRUPTED;
142 }
143
144 err = jbd2_journal_bmap(journal, offset, &blocknr);
145
146 if (err) {
147 printk(KERN_ERR "JBD2: bad block at offset %u\n",
148 offset);
149 return err;
150 }
151
152 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
153 if (!bh)
154 return -ENOMEM;
155
156 if (!buffer_uptodate(bh)) {
157 /*
158 * If this is a brand new buffer, start readahead.
159 * Otherwise, we assume we are already reading it.
160 */
161 bool need_readahead = !buffer_req(bh);
162
163 bh_read_nowait(bh, 0);
164 if (need_readahead)
165 do_readahead(journal, offset);
166 wait_on_buffer(bh);
167 }
168
169 if (!buffer_uptodate(bh)) {
170 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
171 offset);
172 brelse(bh);
173 return -EIO;
174 }
175
176 *bhp = bh;
177 return 0;
178}
179
180static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
181{
182 struct jbd2_journal_block_tail *tail;
183 __be32 provided;
184 __u32 calculated;
185
186 if (!jbd2_journal_has_csum_v2or3(j))
187 return 1;
188
189 tail = (struct jbd2_journal_block_tail *)((char *)buf +
190 j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
191 provided = tail->t_checksum;
192 tail->t_checksum = 0;
193 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
194 tail->t_checksum = provided;
195
196 return provided == cpu_to_be32(calculated);
197}
198
199/*
200 * Count the number of in-use tags in a journal descriptor block.
201 */
202
203static int count_tags(journal_t *journal, struct buffer_head *bh)
204{
205 char * tagp;
206 journal_block_tag_t tag;
207 int nr = 0, size = journal->j_blocksize;
208 int tag_bytes = journal_tag_bytes(journal);
209
210 if (jbd2_journal_has_csum_v2or3(journal))
211 size -= sizeof(struct jbd2_journal_block_tail);
212
213 tagp = &bh->b_data[sizeof(journal_header_t)];
214
215 while ((tagp - bh->b_data + tag_bytes) <= size) {
216 memcpy(&tag, tagp, sizeof(tag));
217
218 nr++;
219 tagp += tag_bytes;
220 if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
221 tagp += 16;
222
223 if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
224 break;
225 }
226
227 return nr;
228}
229
230
231/* Make sure we wrap around the log correctly! */
232#define wrap(journal, var) \
233do { \
234 if (var >= (journal)->j_last) \
235 var -= ((journal)->j_last - (journal)->j_first); \
236} while (0)
237
238static int fc_do_one_pass(journal_t *journal,
239 struct recovery_info *info, enum passtype pass)
240{
241 unsigned int expected_commit_id = info->end_transaction;
242 unsigned long next_fc_block;
243 struct buffer_head *bh;
244 int err = 0;
245
246 next_fc_block = journal->j_fc_first;
247 if (!journal->j_fc_replay_callback)
248 return 0;
249
250 while (next_fc_block <= journal->j_fc_last) {
251 jbd2_debug(3, "Fast commit replay: next block %ld\n",
252 next_fc_block);
253 err = jread(&bh, journal, next_fc_block);
254 if (err) {
255 jbd2_debug(3, "Fast commit replay: read error\n");
256 break;
257 }
258
259 err = journal->j_fc_replay_callback(journal, bh, pass,
260 next_fc_block - journal->j_fc_first,
261 expected_commit_id);
262 brelse(bh);
263 next_fc_block++;
264 if (err < 0 || err == JBD2_FC_REPLAY_STOP)
265 break;
266 err = 0;
267 }
268
269 if (err)
270 jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
271
272 return err;
273}
274
275/**
276 * jbd2_journal_recover - recovers a on-disk journal
277 * @journal: the journal to recover
278 *
279 * The primary function for recovering the log contents when mounting a
280 * journaled device.
281 *
282 * Recovery is done in three passes. In the first pass, we look for the
283 * end of the log. In the second, we assemble the list of revoke
284 * blocks. In the third and final pass, we replay any un-revoked blocks
285 * in the log.
286 */
287int jbd2_journal_recover(journal_t *journal)
288{
289 int err, err2;
290 journal_superblock_t * sb;
291
292 struct recovery_info info;
293
294 memset(&info, 0, sizeof(info));
295 sb = journal->j_superblock;
296
297 /*
298 * The journal superblock's s_start field (the current log head)
299 * is always zero if, and only if, the journal was cleanly
300 * unmounted.
301 */
302 if (!sb->s_start) {
303 jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
304 be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
305 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
306 journal->j_head = be32_to_cpu(sb->s_head);
307 return 0;
308 }
309
310 err = do_one_pass(journal, &info, PASS_SCAN);
311 if (!err)
312 err = do_one_pass(journal, &info, PASS_REVOKE);
313 if (!err)
314 err = do_one_pass(journal, &info, PASS_REPLAY);
315
316 jbd2_debug(1, "JBD2: recovery, exit status %d, "
317 "recovered transactions %u to %u\n",
318 err, info.start_transaction, info.end_transaction);
319 jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
320 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
321
322 /* Restart the log at the next transaction ID, thus invalidating
323 * any existing commit records in the log. */
324 journal->j_transaction_sequence = ++info.end_transaction;
325 journal->j_head = info.head_block;
326 jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
327 journal->j_transaction_sequence, journal->j_head);
328
329 jbd2_journal_clear_revoke(journal);
330 err2 = sync_blockdev(journal->j_fs_dev);
331 if (!err)
332 err = err2;
333 err2 = jbd2_check_fs_dev_write_error(journal);
334 if (!err)
335 err = err2;
336 /* Make sure all replayed data is on permanent storage */
337 if (journal->j_flags & JBD2_BARRIER) {
338 err2 = blkdev_issue_flush(journal->j_fs_dev);
339 if (!err)
340 err = err2;
341 }
342 return err;
343}
344
345/**
346 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
347 * @journal: journal to startup
348 *
349 * Locate any valid recovery information from the journal and set up the
350 * journal structures in memory to ignore it (presumably because the
351 * caller has evidence that it is out of date).
352 * This function doesn't appear to be exported..
353 *
354 * We perform one pass over the journal to allow us to tell the user how
355 * much recovery information is being erased, and to let us initialise
356 * the journal transaction sequence numbers to the next unused ID.
357 */
358int jbd2_journal_skip_recovery(journal_t *journal)
359{
360 int err;
361
362 struct recovery_info info;
363
364 memset (&info, 0, sizeof(info));
365
366 err = do_one_pass(journal, &info, PASS_SCAN);
367
368 if (err) {
369 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
370 ++journal->j_transaction_sequence;
371 journal->j_head = journal->j_first;
372 } else {
373#ifdef CONFIG_JBD2_DEBUG
374 int dropped = info.end_transaction -
375 be32_to_cpu(journal->j_superblock->s_sequence);
376 jbd2_debug(1,
377 "JBD2: ignoring %d transaction%s from the journal.\n",
378 dropped, str_plural(dropped));
379#endif
380 journal->j_transaction_sequence = ++info.end_transaction;
381 journal->j_head = info.head_block;
382 }
383
384 journal->j_tail = 0;
385 return err;
386}
387
388static inline unsigned long long read_tag_block(journal_t *journal,
389 journal_block_tag_t *tag)
390{
391 unsigned long long block = be32_to_cpu(tag->t_blocknr);
392 if (jbd2_has_feature_64bit(journal))
393 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
394 return block;
395}
396
397/*
398 * calc_chksums calculates the checksums for the blocks described in the
399 * descriptor block.
400 */
401static int calc_chksums(journal_t *journal, struct buffer_head *bh,
402 unsigned long *next_log_block, __u32 *crc32_sum)
403{
404 int i, num_blks, err;
405 unsigned long io_block;
406 struct buffer_head *obh;
407
408 num_blks = count_tags(journal, bh);
409 /* Calculate checksum of the descriptor block. */
410 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
411
412 for (i = 0; i < num_blks; i++) {
413 io_block = (*next_log_block)++;
414 wrap(journal, *next_log_block);
415 err = jread(&obh, journal, io_block);
416 if (err) {
417 printk(KERN_ERR "JBD2: IO error %d recovering block "
418 "%lu in log\n", err, io_block);
419 return 1;
420 } else {
421 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
422 obh->b_size);
423 }
424 put_bh(obh);
425 }
426 return 0;
427}
428
429static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
430{
431 struct commit_header *h;
432 __be32 provided;
433 __u32 calculated;
434
435 if (!jbd2_journal_has_csum_v2or3(j))
436 return 1;
437
438 h = buf;
439 provided = h->h_chksum[0];
440 h->h_chksum[0] = 0;
441 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
442 h->h_chksum[0] = provided;
443
444 return provided == cpu_to_be32(calculated);
445}
446
447static bool jbd2_commit_block_csum_verify_partial(journal_t *j, void *buf)
448{
449 struct commit_header *h;
450 __be32 provided;
451 __u32 calculated;
452 void *tmpbuf;
453
454 tmpbuf = kzalloc(j->j_blocksize, GFP_KERNEL);
455 if (!tmpbuf)
456 return false;
457
458 memcpy(tmpbuf, buf, sizeof(struct commit_header));
459 h = tmpbuf;
460 provided = h->h_chksum[0];
461 h->h_chksum[0] = 0;
462 calculated = jbd2_chksum(j, j->j_csum_seed, tmpbuf, j->j_blocksize);
463 kfree(tmpbuf);
464
465 return provided == cpu_to_be32(calculated);
466}
467
468static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
469 journal_block_tag3_t *tag3,
470 void *buf, __u32 sequence)
471{
472 __u32 csum32;
473 __be32 seq;
474
475 if (!jbd2_journal_has_csum_v2or3(j))
476 return 1;
477
478 seq = cpu_to_be32(sequence);
479 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
480 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
481
482 if (jbd2_has_feature_csum3(j))
483 return tag3->t_checksum == cpu_to_be32(csum32);
484 else
485 return tag->t_checksum == cpu_to_be16(csum32);
486}
487
488static __always_inline int jbd2_do_replay(journal_t *journal,
489 struct recovery_info *info,
490 struct buffer_head *bh,
491 unsigned long *next_log_block,
492 unsigned int next_commit_ID)
493{
494 char *tagp;
495 int flags;
496 int ret = 0;
497 int tag_bytes = journal_tag_bytes(journal);
498 int descr_csum_size = 0;
499 unsigned long io_block;
500 journal_block_tag_t tag;
501 struct buffer_head *obh;
502 struct buffer_head *nbh;
503
504 if (jbd2_journal_has_csum_v2or3(journal))
505 descr_csum_size = sizeof(struct jbd2_journal_block_tail);
506
507 tagp = &bh->b_data[sizeof(journal_header_t)];
508 while (tagp - bh->b_data + tag_bytes <=
509 journal->j_blocksize - descr_csum_size) {
510 int err;
511
512 memcpy(&tag, tagp, sizeof(tag));
513 flags = be16_to_cpu(tag.t_flags);
514
515 io_block = (*next_log_block)++;
516 wrap(journal, *next_log_block);
517 err = jread(&obh, journal, io_block);
518 if (err) {
519 /* Recover what we can, but report failure at the end. */
520 ret = err;
521 pr_err("JBD2: IO error %d recovering block %lu in log\n",
522 err, io_block);
523 } else {
524 unsigned long long blocknr;
525
526 J_ASSERT(obh != NULL);
527 blocknr = read_tag_block(journal, &tag);
528
529 /* If the block has been revoked, then we're all done here. */
530 if (jbd2_journal_test_revoke(journal, blocknr,
531 next_commit_ID)) {
532 brelse(obh);
533 ++info->nr_revoke_hits;
534 goto skip_write;
535 }
536
537 /* Look for block corruption */
538 if (!jbd2_block_tag_csum_verify(journal, &tag,
539 (journal_block_tag3_t *)tagp,
540 obh->b_data, next_commit_ID)) {
541 brelse(obh);
542 ret = -EFSBADCRC;
543 pr_err("JBD2: Invalid checksum recovering data block %llu in journal block %lu\n",
544 blocknr, io_block);
545 goto skip_write;
546 }
547
548 /* Find a buffer for the new data being restored */
549 nbh = __getblk(journal->j_fs_dev, blocknr,
550 journal->j_blocksize);
551 if (nbh == NULL) {
552 pr_err("JBD2: Out of memory during recovery.\n");
553 brelse(obh);
554 return -ENOMEM;
555 }
556
557 lock_buffer(nbh);
558 memcpy(nbh->b_data, obh->b_data, journal->j_blocksize);
559 if (flags & JBD2_FLAG_ESCAPE) {
560 *((__be32 *)nbh->b_data) =
561 cpu_to_be32(JBD2_MAGIC_NUMBER);
562 }
563
564 BUFFER_TRACE(nbh, "marking dirty");
565 set_buffer_uptodate(nbh);
566 mark_buffer_dirty(nbh);
567 BUFFER_TRACE(nbh, "marking uptodate");
568 ++info->nr_replays;
569 unlock_buffer(nbh);
570 brelse(obh);
571 brelse(nbh);
572 }
573
574skip_write:
575 tagp += tag_bytes;
576 if (!(flags & JBD2_FLAG_SAME_UUID))
577 tagp += 16;
578
579 if (flags & JBD2_FLAG_LAST_TAG)
580 break;
581 }
582
583 return ret;
584}
585
586static int do_one_pass(journal_t *journal,
587 struct recovery_info *info, enum passtype pass)
588{
589 unsigned int first_commit_ID, next_commit_ID;
590 unsigned long next_log_block, head_block;
591 int err, success = 0;
592 journal_superblock_t * sb;
593 journal_header_t * tmp;
594 struct buffer_head *bh = NULL;
595 unsigned int sequence;
596 int blocktype;
597 __u32 crc32_sum = ~0; /* Transactional Checksums */
598 bool need_check_commit_time = false;
599 __u64 last_trans_commit_time = 0, commit_time;
600
601 /*
602 * First thing is to establish what we expect to find in the log
603 * (in terms of transaction IDs), and where (in terms of log
604 * block offsets): query the superblock.
605 */
606
607 sb = journal->j_superblock;
608 next_commit_ID = be32_to_cpu(sb->s_sequence);
609 next_log_block = be32_to_cpu(sb->s_start);
610 head_block = next_log_block;
611
612 first_commit_ID = next_commit_ID;
613 if (pass == PASS_SCAN)
614 info->start_transaction = first_commit_ID;
615
616 jbd2_debug(1, "Starting recovery pass %d\n", pass);
617
618 /*
619 * Now we walk through the log, transaction by transaction,
620 * making sure that each transaction has a commit block in the
621 * expected place. Each complete transaction gets replayed back
622 * into the main filesystem.
623 */
624
625 while (1) {
626 cond_resched();
627
628 /* If we already know where to stop the log traversal,
629 * check right now that we haven't gone past the end of
630 * the log. */
631
632 if (pass != PASS_SCAN)
633 if (tid_geq(next_commit_ID, info->end_transaction))
634 break;
635
636 jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
637 next_commit_ID, next_log_block, journal->j_last);
638
639 /* Skip over each chunk of the transaction looking
640 * either the next descriptor block or the final commit
641 * record. */
642
643 jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
644 brelse(bh);
645 bh = NULL;
646 err = jread(&bh, journal, next_log_block);
647 if (err)
648 goto failed;
649
650 next_log_block++;
651 wrap(journal, next_log_block);
652
653 /* What kind of buffer is it?
654 *
655 * If it is a descriptor block, check that it has the
656 * expected sequence number. Otherwise, we're all done
657 * here. */
658
659 tmp = (journal_header_t *)bh->b_data;
660
661 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER))
662 break;
663
664 blocktype = be32_to_cpu(tmp->h_blocktype);
665 sequence = be32_to_cpu(tmp->h_sequence);
666 jbd2_debug(3, "Found magic %d, sequence %d\n",
667 blocktype, sequence);
668
669 if (sequence != next_commit_ID)
670 break;
671
672 /* OK, we have a valid descriptor block which matches
673 * all of the sequence number checks. What are we going
674 * to do with it? That depends on the pass... */
675
676 switch(blocktype) {
677 case JBD2_DESCRIPTOR_BLOCK:
678 /* Verify checksum first */
679 if (!jbd2_descriptor_block_csum_verify(journal,
680 bh->b_data)) {
681 /*
682 * PASS_SCAN can see stale blocks due to lazy
683 * journal init. Don't error out on those yet.
684 */
685 if (pass != PASS_SCAN) {
686 pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
687 next_log_block);
688 err = -EFSBADCRC;
689 goto failed;
690 }
691 need_check_commit_time = true;
692 jbd2_debug(1,
693 "invalid descriptor block found in %lu\n",
694 next_log_block);
695 }
696
697 /* If it is a valid descriptor block, replay it
698 * in pass REPLAY; if journal_checksums enabled, then
699 * calculate checksums in PASS_SCAN, otherwise,
700 * just skip over the blocks it describes. */
701 if (pass != PASS_REPLAY) {
702 if (pass == PASS_SCAN &&
703 jbd2_has_feature_checksum(journal) &&
704 !info->end_transaction) {
705 if (calc_chksums(journal, bh,
706 &next_log_block,
707 &crc32_sum))
708 break;
709 continue;
710 }
711 next_log_block += count_tags(journal, bh);
712 wrap(journal, next_log_block);
713 continue;
714 }
715
716 /*
717 * A descriptor block: we can now write all of the
718 * data blocks. Yay, useful work is finally getting
719 * done here!
720 */
721 err = jbd2_do_replay(journal, info, bh, &next_log_block,
722 next_commit_ID);
723 if (err) {
724 if (err == -ENOMEM)
725 goto failed;
726 success = err;
727 }
728
729 continue;
730
731 case JBD2_COMMIT_BLOCK:
732 if (pass != PASS_SCAN) {
733 next_commit_ID++;
734 continue;
735 }
736
737 /* How to differentiate between interrupted commit
738 * and journal corruption ?
739 *
740 * {nth transaction}
741 * Checksum Verification Failed
742 * |
743 * ____________________
744 * | |
745 * async_commit sync_commit
746 * | |
747 * | GO TO NEXT "Journal Corruption"
748 * | TRANSACTION
749 * |
750 * {(n+1)th transanction}
751 * |
752 * _______|______________
753 * | |
754 * Commit block found Commit block not found
755 * | |
756 * "Journal Corruption" |
757 * _____________|_________
758 * | |
759 * nth trans corrupt OR nth trans
760 * and (n+1)th interrupted interrupted
761 * before commit block
762 * could reach the disk.
763 * (Cannot find the difference in above
764 * mentioned conditions. Hence assume
765 * "Interrupted Commit".)
766 */
767 commit_time = be64_to_cpu(
768 ((struct commit_header *)bh->b_data)->h_commit_sec);
769 /*
770 * If need_check_commit_time is set, it means we are in
771 * PASS_SCAN and csum verify failed before. If
772 * commit_time is increasing, it's the same journal,
773 * otherwise it is stale journal block, just end this
774 * recovery.
775 */
776 if (need_check_commit_time) {
777 if (commit_time >= last_trans_commit_time) {
778 pr_err("JBD2: Invalid checksum found in transaction %u\n",
779 next_commit_ID);
780 err = -EFSBADCRC;
781 goto failed;
782 }
783 ignore_crc_mismatch:
784 /*
785 * It likely does not belong to same journal,
786 * just end this recovery with success.
787 */
788 jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
789 next_commit_ID);
790 goto done;
791 }
792
793 /*
794 * Found an expected commit block: if checksums
795 * are present, verify them in PASS_SCAN; else not
796 * much to do other than move on to the next sequence
797 * number.
798 */
799 if (jbd2_has_feature_checksum(journal)) {
800 struct commit_header *cbh =
801 (struct commit_header *)bh->b_data;
802 unsigned found_chksum =
803 be32_to_cpu(cbh->h_chksum[0]);
804
805 if (info->end_transaction) {
806 journal->j_failed_commit =
807 info->end_transaction;
808 break;
809 }
810
811 /* Neither checksum match nor unused? */
812 if (!((crc32_sum == found_chksum &&
813 cbh->h_chksum_type ==
814 JBD2_CRC32_CHKSUM &&
815 cbh->h_chksum_size ==
816 JBD2_CRC32_CHKSUM_SIZE) ||
817 (cbh->h_chksum_type == 0 &&
818 cbh->h_chksum_size == 0 &&
819 found_chksum == 0)))
820 goto chksum_error;
821
822 crc32_sum = ~0;
823 goto chksum_ok;
824 }
825
826 if (jbd2_commit_block_csum_verify(journal, bh->b_data))
827 goto chksum_ok;
828
829 if (jbd2_commit_block_csum_verify_partial(journal,
830 bh->b_data)) {
831 pr_notice("JBD2: Find incomplete commit block in transaction %u block %lu\n",
832 next_commit_ID, next_log_block);
833 goto chksum_ok;
834 }
835
836chksum_error:
837 if (commit_time < last_trans_commit_time)
838 goto ignore_crc_mismatch;
839 info->end_transaction = next_commit_ID;
840 info->head_block = head_block;
841
842 if (!jbd2_has_feature_async_commit(journal)) {
843 journal->j_failed_commit = next_commit_ID;
844 break;
845 }
846
847chksum_ok:
848 last_trans_commit_time = commit_time;
849 head_block = next_log_block;
850 next_commit_ID++;
851 continue;
852
853 case JBD2_REVOKE_BLOCK:
854 /*
855 * Check revoke block crc in pass_scan, if csum verify
856 * failed, check commit block time later.
857 */
858 if (pass == PASS_SCAN &&
859 !jbd2_descriptor_block_csum_verify(journal,
860 bh->b_data)) {
861 jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
862 next_log_block);
863 need_check_commit_time = true;
864 }
865
866 /* If we aren't in the REVOKE pass, then we can
867 * just skip over this block. */
868 if (pass != PASS_REVOKE)
869 continue;
870
871 err = scan_revoke_records(journal, bh,
872 next_commit_ID, info);
873 if (err)
874 goto failed;
875 continue;
876
877 default:
878 jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
879 blocktype);
880 goto done;
881 }
882 }
883
884 done:
885 brelse(bh);
886 /*
887 * We broke out of the log scan loop: either we came to the
888 * known end of the log or we found an unexpected block in the
889 * log. If the latter happened, then we know that the "current"
890 * transaction marks the end of the valid log.
891 */
892
893 if (pass == PASS_SCAN) {
894 if (!info->end_transaction)
895 info->end_transaction = next_commit_ID;
896 if (!info->head_block)
897 info->head_block = head_block;
898 } else {
899 /* It's really bad news if different passes end up at
900 * different places (but possible due to IO errors). */
901 if (info->end_transaction != next_commit_ID) {
902 printk(KERN_ERR "JBD2: recovery pass %d ended at "
903 "transaction %u, expected %u\n",
904 pass, next_commit_ID, info->end_transaction);
905 if (!success)
906 success = -EIO;
907 }
908 }
909
910 if (jbd2_has_feature_fast_commit(journal) && pass != PASS_REVOKE) {
911 err = fc_do_one_pass(journal, info, pass);
912 if (err)
913 success = err;
914 }
915
916 return success;
917
918 failed:
919 brelse(bh);
920 return err;
921}
922
923/* Scan a revoke record, marking all blocks mentioned as revoked. */
924
925static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
926 tid_t sequence, struct recovery_info *info)
927{
928 jbd2_journal_revoke_header_t *header;
929 int offset, max;
930 unsigned csum_size = 0;
931 __u32 rcount;
932 int record_len = 4;
933
934 header = (jbd2_journal_revoke_header_t *) bh->b_data;
935 offset = sizeof(jbd2_journal_revoke_header_t);
936 rcount = be32_to_cpu(header->r_count);
937
938 if (jbd2_journal_has_csum_v2or3(journal))
939 csum_size = sizeof(struct jbd2_journal_block_tail);
940 if (rcount > journal->j_blocksize - csum_size)
941 return -EINVAL;
942 max = rcount;
943
944 if (jbd2_has_feature_64bit(journal))
945 record_len = 8;
946
947 while (offset + record_len <= max) {
948 unsigned long long blocknr;
949 int err;
950
951 if (record_len == 4)
952 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
953 else
954 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
955 offset += record_len;
956 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
957 if (err)
958 return err;
959 ++info->nr_revokes;
960 }
961 return 0;
962}