Loading...
1/*
2 * linux/fs/jbd2/recovery.c
3 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
5 *
6 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Journal recovery routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
14 */
15
16#ifndef __KERNEL__
17#include "jfs_user.h"
18#else
19#include <linux/time.h>
20#include <linux/fs.h>
21#include <linux/jbd2.h>
22#include <linux/errno.h>
23#include <linux/crc32.h>
24#include <linux/blkdev.h>
25#endif
26
27/*
28 * Maintain information about the progress of the recovery job, so that
29 * the different passes can carry information between them.
30 */
31struct recovery_info
32{
33 tid_t start_transaction;
34 tid_t end_transaction;
35
36 int nr_replays;
37 int nr_revokes;
38 int nr_revoke_hits;
39};
40
41enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
42static int do_one_pass(journal_t *journal,
43 struct recovery_info *info, enum passtype pass);
44static int scan_revoke_records(journal_t *, struct buffer_head *,
45 tid_t, struct recovery_info *);
46
47#ifdef __KERNEL__
48
49/* Release readahead buffers after use */
50static void journal_brelse_array(struct buffer_head *b[], int n)
51{
52 while (--n >= 0)
53 brelse (b[n]);
54}
55
56
57/*
58 * When reading from the journal, we are going through the block device
59 * layer directly and so there is no readahead being done for us. We
60 * need to implement any readahead ourselves if we want it to happen at
61 * all. Recovery is basically one long sequential read, so make sure we
62 * do the IO in reasonably large chunks.
63 *
64 * This is not so critical that we need to be enormously clever about
65 * the readahead size, though. 128K is a purely arbitrary, good-enough
66 * fixed value.
67 */
68
69#define MAXBUF 8
70static int do_readahead(journal_t *journal, unsigned int start)
71{
72 int err;
73 unsigned int max, nbufs, next;
74 unsigned long long blocknr;
75 struct buffer_head *bh;
76
77 struct buffer_head * bufs[MAXBUF];
78
79 /* Do up to 128K of readahead */
80 max = start + (128 * 1024 / journal->j_blocksize);
81 if (max > journal->j_maxlen)
82 max = journal->j_maxlen;
83
84 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
85 * a time to the block device IO layer. */
86
87 nbufs = 0;
88
89 for (next = start; next < max; next++) {
90 err = jbd2_journal_bmap(journal, next, &blocknr);
91
92 if (err) {
93 printk(KERN_ERR "JBD2: bad block at offset %u\n",
94 next);
95 goto failed;
96 }
97
98 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
99 if (!bh) {
100 err = -ENOMEM;
101 goto failed;
102 }
103
104 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
105 bufs[nbufs++] = bh;
106 if (nbufs == MAXBUF) {
107 ll_rw_block(READ, nbufs, bufs);
108 journal_brelse_array(bufs, nbufs);
109 nbufs = 0;
110 }
111 } else
112 brelse(bh);
113 }
114
115 if (nbufs)
116 ll_rw_block(READ, nbufs, bufs);
117 err = 0;
118
119failed:
120 if (nbufs)
121 journal_brelse_array(bufs, nbufs);
122 return err;
123}
124
125#endif /* __KERNEL__ */
126
127
128/*
129 * Read a block from the journal
130 */
131
132static int jread(struct buffer_head **bhp, journal_t *journal,
133 unsigned int offset)
134{
135 int err;
136 unsigned long long blocknr;
137 struct buffer_head *bh;
138
139 *bhp = NULL;
140
141 if (offset >= journal->j_maxlen) {
142 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
143 return -EIO;
144 }
145
146 err = jbd2_journal_bmap(journal, offset, &blocknr);
147
148 if (err) {
149 printk(KERN_ERR "JBD2: bad block at offset %u\n",
150 offset);
151 return err;
152 }
153
154 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
155 if (!bh)
156 return -ENOMEM;
157
158 if (!buffer_uptodate(bh)) {
159 /* If this is a brand new buffer, start readahead.
160 Otherwise, we assume we are already reading it. */
161 if (!buffer_req(bh))
162 do_readahead(journal, offset);
163 wait_on_buffer(bh);
164 }
165
166 if (!buffer_uptodate(bh)) {
167 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
168 offset);
169 brelse(bh);
170 return -EIO;
171 }
172
173 *bhp = bh;
174 return 0;
175}
176
177static int jbd2_descr_block_csum_verify(journal_t *j,
178 void *buf)
179{
180 struct jbd2_journal_block_tail *tail;
181 __be32 provided;
182 __u32 calculated;
183
184 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
185 return 1;
186
187 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
188 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_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
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 tag = (journal_block_tag_t *) tagp;
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 if (var >= (journal)->j_last) \
233 var -= ((journal)->j_last - (journal)->j_first); \
234} while (0)
235
236/**
237 * jbd2_journal_recover - recovers a on-disk journal
238 * @journal: the journal to recover
239 *
240 * The primary function for recovering the log contents when mounting a
241 * journaled device.
242 *
243 * Recovery is done in three passes. In the first pass, we look for the
244 * end of the log. In the second, we assemble the list of revoke
245 * blocks. In the third and final pass, we replay any un-revoked blocks
246 * in the log.
247 */
248int jbd2_journal_recover(journal_t *journal)
249{
250 int err, err2;
251 journal_superblock_t * sb;
252
253 struct recovery_info info;
254
255 memset(&info, 0, sizeof(info));
256 sb = journal->j_superblock;
257
258 /*
259 * The journal superblock's s_start field (the current log head)
260 * is always zero if, and only if, the journal was cleanly
261 * unmounted.
262 */
263
264 if (!sb->s_start) {
265 jbd_debug(1, "No recovery required, last transaction %d\n",
266 be32_to_cpu(sb->s_sequence));
267 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
268 return 0;
269 }
270
271 err = do_one_pass(journal, &info, PASS_SCAN);
272 if (!err)
273 err = do_one_pass(journal, &info, PASS_REVOKE);
274 if (!err)
275 err = do_one_pass(journal, &info, PASS_REPLAY);
276
277 jbd_debug(1, "JBD2: recovery, exit status %d, "
278 "recovered transactions %u to %u\n",
279 err, info.start_transaction, info.end_transaction);
280 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
281 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
282
283 /* Restart the log at the next transaction ID, thus invalidating
284 * any existing commit records in the log. */
285 journal->j_transaction_sequence = ++info.end_transaction;
286
287 jbd2_journal_clear_revoke(journal);
288 err2 = sync_blockdev(journal->j_fs_dev);
289 if (!err)
290 err = err2;
291 /* Make sure all replayed data is on permanent storage */
292 if (journal->j_flags & JBD2_BARRIER) {
293 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
294 if (!err)
295 err = err2;
296 }
297 return err;
298}
299
300/**
301 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
302 * @journal: journal to startup
303 *
304 * Locate any valid recovery information from the journal and set up the
305 * journal structures in memory to ignore it (presumably because the
306 * caller has evidence that it is out of date).
307 * This function does'nt appear to be exorted..
308 *
309 * We perform one pass over the journal to allow us to tell the user how
310 * much recovery information is being erased, and to let us initialise
311 * the journal transaction sequence numbers to the next unused ID.
312 */
313int jbd2_journal_skip_recovery(journal_t *journal)
314{
315 int err;
316
317 struct recovery_info info;
318
319 memset (&info, 0, sizeof(info));
320
321 err = do_one_pass(journal, &info, PASS_SCAN);
322
323 if (err) {
324 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
325 ++journal->j_transaction_sequence;
326 } else {
327#ifdef CONFIG_JBD2_DEBUG
328 int dropped = info.end_transaction -
329 be32_to_cpu(journal->j_superblock->s_sequence);
330 jbd_debug(1,
331 "JBD2: ignoring %d transaction%s from the journal.\n",
332 dropped, (dropped == 1) ? "" : "s");
333#endif
334 journal->j_transaction_sequence = ++info.end_transaction;
335 }
336
337 journal->j_tail = 0;
338 return err;
339}
340
341static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
342{
343 unsigned long long block = be32_to_cpu(tag->t_blocknr);
344 if (tag_bytes > JBD2_TAG_SIZE32)
345 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
346 return block;
347}
348
349/*
350 * calc_chksums calculates the checksums for the blocks described in the
351 * descriptor block.
352 */
353static int calc_chksums(journal_t *journal, struct buffer_head *bh,
354 unsigned long *next_log_block, __u32 *crc32_sum)
355{
356 int i, num_blks, err;
357 unsigned long io_block;
358 struct buffer_head *obh;
359
360 num_blks = count_tags(journal, bh);
361 /* Calculate checksum of the descriptor block. */
362 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
363
364 for (i = 0; i < num_blks; i++) {
365 io_block = (*next_log_block)++;
366 wrap(journal, *next_log_block);
367 err = jread(&obh, journal, io_block);
368 if (err) {
369 printk(KERN_ERR "JBD2: IO error %d recovering block "
370 "%lu in log\n", err, io_block);
371 return 1;
372 } else {
373 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
374 obh->b_size);
375 }
376 put_bh(obh);
377 }
378 return 0;
379}
380
381static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
382{
383 struct commit_header *h;
384 __be32 provided;
385 __u32 calculated;
386
387 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
388 return 1;
389
390 h = buf;
391 provided = h->h_chksum[0];
392 h->h_chksum[0] = 0;
393 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
394 h->h_chksum[0] = provided;
395
396 return provided == cpu_to_be32(calculated);
397}
398
399static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
400 void *buf, __u32 sequence)
401{
402 __u32 csum32;
403 __be32 seq;
404
405 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
406 return 1;
407
408 seq = cpu_to_be32(sequence);
409 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
410 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
411
412 return tag->t_checksum == cpu_to_be16(csum32);
413}
414
415static int do_one_pass(journal_t *journal,
416 struct recovery_info *info, enum passtype pass)
417{
418 unsigned int first_commit_ID, next_commit_ID;
419 unsigned long next_log_block;
420 int err, success = 0;
421 journal_superblock_t * sb;
422 journal_header_t * tmp;
423 struct buffer_head * bh;
424 unsigned int sequence;
425 int blocktype;
426 int tag_bytes = journal_tag_bytes(journal);
427 __u32 crc32_sum = ~0; /* Transactional Checksums */
428 int descr_csum_size = 0;
429
430 /*
431 * First thing is to establish what we expect to find in the log
432 * (in terms of transaction IDs), and where (in terms of log
433 * block offsets): query the superblock.
434 */
435
436 sb = journal->j_superblock;
437 next_commit_ID = be32_to_cpu(sb->s_sequence);
438 next_log_block = be32_to_cpu(sb->s_start);
439
440 first_commit_ID = next_commit_ID;
441 if (pass == PASS_SCAN)
442 info->start_transaction = first_commit_ID;
443
444 jbd_debug(1, "Starting recovery pass %d\n", pass);
445
446 /*
447 * Now we walk through the log, transaction by transaction,
448 * making sure that each transaction has a commit block in the
449 * expected place. Each complete transaction gets replayed back
450 * into the main filesystem.
451 */
452
453 while (1) {
454 int flags;
455 char * tagp;
456 journal_block_tag_t * tag;
457 struct buffer_head * obh;
458 struct buffer_head * nbh;
459
460 cond_resched();
461
462 /* If we already know where to stop the log traversal,
463 * check right now that we haven't gone past the end of
464 * the log. */
465
466 if (pass != PASS_SCAN)
467 if (tid_geq(next_commit_ID, info->end_transaction))
468 break;
469
470 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
471 next_commit_ID, next_log_block, journal->j_last);
472
473 /* Skip over each chunk of the transaction looking
474 * either the next descriptor block or the final commit
475 * record. */
476
477 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
478 err = jread(&bh, journal, next_log_block);
479 if (err)
480 goto failed;
481
482 next_log_block++;
483 wrap(journal, next_log_block);
484
485 /* What kind of buffer is it?
486 *
487 * If it is a descriptor block, check that it has the
488 * expected sequence number. Otherwise, we're all done
489 * here. */
490
491 tmp = (journal_header_t *)bh->b_data;
492
493 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
494 brelse(bh);
495 break;
496 }
497
498 blocktype = be32_to_cpu(tmp->h_blocktype);
499 sequence = be32_to_cpu(tmp->h_sequence);
500 jbd_debug(3, "Found magic %d, sequence %d\n",
501 blocktype, sequence);
502
503 if (sequence != next_commit_ID) {
504 brelse(bh);
505 break;
506 }
507
508 /* OK, we have a valid descriptor block which matches
509 * all of the sequence number checks. What are we going
510 * to do with it? That depends on the pass... */
511
512 switch(blocktype) {
513 case JBD2_DESCRIPTOR_BLOCK:
514 /* Verify checksum first */
515 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
516 JBD2_FEATURE_INCOMPAT_CSUM_V2))
517 descr_csum_size =
518 sizeof(struct jbd2_journal_block_tail);
519 if (descr_csum_size > 0 &&
520 !jbd2_descr_block_csum_verify(journal,
521 bh->b_data)) {
522 err = -EIO;
523 goto failed;
524 }
525
526 /* If it is a valid descriptor block, replay it
527 * in pass REPLAY; if journal_checksums enabled, then
528 * calculate checksums in PASS_SCAN, otherwise,
529 * just skip over the blocks it describes. */
530 if (pass != PASS_REPLAY) {
531 if (pass == PASS_SCAN &&
532 JBD2_HAS_COMPAT_FEATURE(journal,
533 JBD2_FEATURE_COMPAT_CHECKSUM) &&
534 !info->end_transaction) {
535 if (calc_chksums(journal, bh,
536 &next_log_block,
537 &crc32_sum)) {
538 put_bh(bh);
539 break;
540 }
541 put_bh(bh);
542 continue;
543 }
544 next_log_block += count_tags(journal, bh);
545 wrap(journal, next_log_block);
546 put_bh(bh);
547 continue;
548 }
549
550 /* A descriptor block: we can now write all of
551 * the data blocks. Yay, useful work is finally
552 * getting done here! */
553
554 tagp = &bh->b_data[sizeof(journal_header_t)];
555 while ((tagp - bh->b_data + tag_bytes)
556 <= journal->j_blocksize - descr_csum_size) {
557 unsigned long io_block;
558
559 tag = (journal_block_tag_t *) tagp;
560 flags = be16_to_cpu(tag->t_flags);
561
562 io_block = next_log_block++;
563 wrap(journal, next_log_block);
564 err = jread(&obh, journal, io_block);
565 if (err) {
566 /* Recover what we can, but
567 * report failure at the end. */
568 success = err;
569 printk(KERN_ERR
570 "JBD2: IO error %d recovering "
571 "block %ld in log\n",
572 err, io_block);
573 } else {
574 unsigned long long blocknr;
575
576 J_ASSERT(obh != NULL);
577 blocknr = read_tag_block(tag_bytes,
578 tag);
579
580 /* If the block has been
581 * revoked, then we're all done
582 * here. */
583 if (jbd2_journal_test_revoke
584 (journal, blocknr,
585 next_commit_ID)) {
586 brelse(obh);
587 ++info->nr_revoke_hits;
588 goto skip_write;
589 }
590
591 /* Look for block corruption */
592 if (!jbd2_block_tag_csum_verify(
593 journal, tag, obh->b_data,
594 be32_to_cpu(tmp->h_sequence))) {
595 brelse(obh);
596 success = -EIO;
597 printk(KERN_ERR "JBD2: Invalid "
598 "checksum recovering "
599 "block %llu in log\n",
600 blocknr);
601 continue;
602 }
603
604 /* Find a buffer for the new
605 * data being restored */
606 nbh = __getblk(journal->j_fs_dev,
607 blocknr,
608 journal->j_blocksize);
609 if (nbh == NULL) {
610 printk(KERN_ERR
611 "JBD2: Out of memory "
612 "during recovery.\n");
613 err = -ENOMEM;
614 brelse(bh);
615 brelse(obh);
616 goto failed;
617 }
618
619 lock_buffer(nbh);
620 memcpy(nbh->b_data, obh->b_data,
621 journal->j_blocksize);
622 if (flags & JBD2_FLAG_ESCAPE) {
623 *((__be32 *)nbh->b_data) =
624 cpu_to_be32(JBD2_MAGIC_NUMBER);
625 }
626
627 BUFFER_TRACE(nbh, "marking dirty");
628 set_buffer_uptodate(nbh);
629 mark_buffer_dirty(nbh);
630 BUFFER_TRACE(nbh, "marking uptodate");
631 ++info->nr_replays;
632 /* ll_rw_block(WRITE, 1, &nbh); */
633 unlock_buffer(nbh);
634 brelse(obh);
635 brelse(nbh);
636 }
637
638 skip_write:
639 tagp += tag_bytes;
640 if (!(flags & JBD2_FLAG_SAME_UUID))
641 tagp += 16;
642
643 if (flags & JBD2_FLAG_LAST_TAG)
644 break;
645 }
646
647 brelse(bh);
648 continue;
649
650 case JBD2_COMMIT_BLOCK:
651 /* How to differentiate between interrupted commit
652 * and journal corruption ?
653 *
654 * {nth transaction}
655 * Checksum Verification Failed
656 * |
657 * ____________________
658 * | |
659 * async_commit sync_commit
660 * | |
661 * | GO TO NEXT "Journal Corruption"
662 * | TRANSACTION
663 * |
664 * {(n+1)th transanction}
665 * |
666 * _______|______________
667 * | |
668 * Commit block found Commit block not found
669 * | |
670 * "Journal Corruption" |
671 * _____________|_________
672 * | |
673 * nth trans corrupt OR nth trans
674 * and (n+1)th interrupted interrupted
675 * before commit block
676 * could reach the disk.
677 * (Cannot find the difference in above
678 * mentioned conditions. Hence assume
679 * "Interrupted Commit".)
680 */
681
682 /* Found an expected commit block: if checksums
683 * are present verify them in PASS_SCAN; else not
684 * much to do other than move on to the next sequence
685 * number. */
686 if (pass == PASS_SCAN &&
687 JBD2_HAS_COMPAT_FEATURE(journal,
688 JBD2_FEATURE_COMPAT_CHECKSUM)) {
689 int chksum_err, chksum_seen;
690 struct commit_header *cbh =
691 (struct commit_header *)bh->b_data;
692 unsigned found_chksum =
693 be32_to_cpu(cbh->h_chksum[0]);
694
695 chksum_err = chksum_seen = 0;
696
697 if (info->end_transaction) {
698 journal->j_failed_commit =
699 info->end_transaction;
700 brelse(bh);
701 break;
702 }
703
704 if (crc32_sum == found_chksum &&
705 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
706 cbh->h_chksum_size ==
707 JBD2_CRC32_CHKSUM_SIZE)
708 chksum_seen = 1;
709 else if (!(cbh->h_chksum_type == 0 &&
710 cbh->h_chksum_size == 0 &&
711 found_chksum == 0 &&
712 !chksum_seen))
713 /*
714 * If fs is mounted using an old kernel and then
715 * kernel with journal_chksum is used then we
716 * get a situation where the journal flag has
717 * checksum flag set but checksums are not
718 * present i.e chksum = 0, in the individual
719 * commit blocks.
720 * Hence to avoid checksum failures, in this
721 * situation, this extra check is added.
722 */
723 chksum_err = 1;
724
725 if (chksum_err) {
726 info->end_transaction = next_commit_ID;
727
728 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
729 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
730 journal->j_failed_commit =
731 next_commit_ID;
732 brelse(bh);
733 break;
734 }
735 }
736 crc32_sum = ~0;
737 }
738 if (pass == PASS_SCAN &&
739 !jbd2_commit_block_csum_verify(journal,
740 bh->b_data)) {
741 info->end_transaction = next_commit_ID;
742
743 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
744 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
745 journal->j_failed_commit =
746 next_commit_ID;
747 brelse(bh);
748 break;
749 }
750 }
751 brelse(bh);
752 next_commit_ID++;
753 continue;
754
755 case JBD2_REVOKE_BLOCK:
756 /* If we aren't in the REVOKE pass, then we can
757 * just skip over this block. */
758 if (pass != PASS_REVOKE) {
759 brelse(bh);
760 continue;
761 }
762
763 err = scan_revoke_records(journal, bh,
764 next_commit_ID, info);
765 brelse(bh);
766 if (err)
767 goto failed;
768 continue;
769
770 default:
771 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
772 blocktype);
773 brelse(bh);
774 goto done;
775 }
776 }
777
778 done:
779 /*
780 * We broke out of the log scan loop: either we came to the
781 * known end of the log or we found an unexpected block in the
782 * log. If the latter happened, then we know that the "current"
783 * transaction marks the end of the valid log.
784 */
785
786 if (pass == PASS_SCAN) {
787 if (!info->end_transaction)
788 info->end_transaction = next_commit_ID;
789 } else {
790 /* It's really bad news if different passes end up at
791 * different places (but possible due to IO errors). */
792 if (info->end_transaction != next_commit_ID) {
793 printk(KERN_ERR "JBD2: recovery pass %d ended at "
794 "transaction %u, expected %u\n",
795 pass, next_commit_ID, info->end_transaction);
796 if (!success)
797 success = -EIO;
798 }
799 }
800
801 return success;
802
803 failed:
804 return err;
805}
806
807static int jbd2_revoke_block_csum_verify(journal_t *j,
808 void *buf)
809{
810 struct jbd2_journal_revoke_tail *tail;
811 __be32 provided;
812 __u32 calculated;
813
814 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
815 return 1;
816
817 tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize -
818 sizeof(struct jbd2_journal_revoke_tail));
819 provided = tail->r_checksum;
820 tail->r_checksum = 0;
821 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
822 tail->r_checksum = provided;
823
824 return provided == cpu_to_be32(calculated);
825}
826
827/* Scan a revoke record, marking all blocks mentioned as revoked. */
828
829static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
830 tid_t sequence, struct recovery_info *info)
831{
832 jbd2_journal_revoke_header_t *header;
833 int offset, max;
834 int record_len = 4;
835
836 header = (jbd2_journal_revoke_header_t *) bh->b_data;
837 offset = sizeof(jbd2_journal_revoke_header_t);
838 max = be32_to_cpu(header->r_count);
839
840 if (!jbd2_revoke_block_csum_verify(journal, header))
841 return -EINVAL;
842
843 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
844 record_len = 8;
845
846 while (offset + record_len <= max) {
847 unsigned long long blocknr;
848 int err;
849
850 if (record_len == 4)
851 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
852 else
853 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
854 offset += record_len;
855 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
856 if (err)
857 return err;
858 ++info->nr_revokes;
859 }
860 return 0;
861}
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}