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 __u32 provided, calculated;
182
183 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
184 return 1;
185
186 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
187 sizeof(struct jbd2_journal_block_tail));
188 provided = tail->t_checksum;
189 tail->t_checksum = 0;
190 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
191 tail->t_checksum = provided;
192
193 provided = be32_to_cpu(provided);
194 return provided == 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 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
294 return err;
295}
296
297/**
298 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
299 * @journal: journal to startup
300 *
301 * Locate any valid recovery information from the journal and set up the
302 * journal structures in memory to ignore it (presumably because the
303 * caller has evidence that it is out of date).
304 * This function does'nt appear to be exorted..
305 *
306 * We perform one pass over the journal to allow us to tell the user how
307 * much recovery information is being erased, and to let us initialise
308 * the journal transaction sequence numbers to the next unused ID.
309 */
310int jbd2_journal_skip_recovery(journal_t *journal)
311{
312 int err;
313
314 struct recovery_info info;
315
316 memset (&info, 0, sizeof(info));
317
318 err = do_one_pass(journal, &info, PASS_SCAN);
319
320 if (err) {
321 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
322 ++journal->j_transaction_sequence;
323 } else {
324#ifdef CONFIG_JBD2_DEBUG
325 int dropped = info.end_transaction -
326 be32_to_cpu(journal->j_superblock->s_sequence);
327 jbd_debug(1,
328 "JBD2: ignoring %d transaction%s from the journal.\n",
329 dropped, (dropped == 1) ? "" : "s");
330#endif
331 journal->j_transaction_sequence = ++info.end_transaction;
332 }
333
334 journal->j_tail = 0;
335 return err;
336}
337
338static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
339{
340 unsigned long long block = be32_to_cpu(tag->t_blocknr);
341 if (tag_bytes > JBD2_TAG_SIZE32)
342 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
343 return block;
344}
345
346/*
347 * calc_chksums calculates the checksums for the blocks described in the
348 * descriptor block.
349 */
350static int calc_chksums(journal_t *journal, struct buffer_head *bh,
351 unsigned long *next_log_block, __u32 *crc32_sum)
352{
353 int i, num_blks, err;
354 unsigned long io_block;
355 struct buffer_head *obh;
356
357 num_blks = count_tags(journal, bh);
358 /* Calculate checksum of the descriptor block. */
359 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
360
361 for (i = 0; i < num_blks; i++) {
362 io_block = (*next_log_block)++;
363 wrap(journal, *next_log_block);
364 err = jread(&obh, journal, io_block);
365 if (err) {
366 printk(KERN_ERR "JBD2: IO error %d recovering block "
367 "%lu in log\n", err, io_block);
368 return 1;
369 } else {
370 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
371 obh->b_size);
372 }
373 put_bh(obh);
374 }
375 return 0;
376}
377
378static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
379{
380 struct commit_header *h;
381 __u32 provided, calculated;
382
383 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
384 return 1;
385
386 h = buf;
387 provided = h->h_chksum[0];
388 h->h_chksum[0] = 0;
389 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
390 h->h_chksum[0] = provided;
391
392 provided = be32_to_cpu(provided);
393 return provided == calculated;
394}
395
396static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
397 void *buf, __u32 sequence)
398{
399 __u32 provided, calculated;
400
401 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
402 return 1;
403
404 sequence = cpu_to_be32(sequence);
405 calculated = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence,
406 sizeof(sequence));
407 calculated = jbd2_chksum(j, calculated, buf, j->j_blocksize);
408 provided = be32_to_cpu(tag->t_checksum);
409
410 return provided == cpu_to_be32(calculated);
411}
412
413static int do_one_pass(journal_t *journal,
414 struct recovery_info *info, enum passtype pass)
415{
416 unsigned int first_commit_ID, next_commit_ID;
417 unsigned long next_log_block;
418 int err, success = 0;
419 journal_superblock_t * sb;
420 journal_header_t * tmp;
421 struct buffer_head * bh;
422 unsigned int sequence;
423 int blocktype;
424 int tag_bytes = journal_tag_bytes(journal);
425 __u32 crc32_sum = ~0; /* Transactional Checksums */
426 int descr_csum_size = 0;
427
428 /*
429 * First thing is to establish what we expect to find in the log
430 * (in terms of transaction IDs), and where (in terms of log
431 * block offsets): query the superblock.
432 */
433
434 sb = journal->j_superblock;
435 next_commit_ID = be32_to_cpu(sb->s_sequence);
436 next_log_block = be32_to_cpu(sb->s_start);
437
438 first_commit_ID = next_commit_ID;
439 if (pass == PASS_SCAN)
440 info->start_transaction = first_commit_ID;
441
442 jbd_debug(1, "Starting recovery pass %d\n", pass);
443
444 /*
445 * Now we walk through the log, transaction by transaction,
446 * making sure that each transaction has a commit block in the
447 * expected place. Each complete transaction gets replayed back
448 * into the main filesystem.
449 */
450
451 while (1) {
452 int flags;
453 char * tagp;
454 journal_block_tag_t * tag;
455 struct buffer_head * obh;
456 struct buffer_head * nbh;
457
458 cond_resched();
459
460 /* If we already know where to stop the log traversal,
461 * check right now that we haven't gone past the end of
462 * the log. */
463
464 if (pass != PASS_SCAN)
465 if (tid_geq(next_commit_ID, info->end_transaction))
466 break;
467
468 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
469 next_commit_ID, next_log_block, journal->j_last);
470
471 /* Skip over each chunk of the transaction looking
472 * either the next descriptor block or the final commit
473 * record. */
474
475 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
476 err = jread(&bh, journal, next_log_block);
477 if (err)
478 goto failed;
479
480 next_log_block++;
481 wrap(journal, next_log_block);
482
483 /* What kind of buffer is it?
484 *
485 * If it is a descriptor block, check that it has the
486 * expected sequence number. Otherwise, we're all done
487 * here. */
488
489 tmp = (journal_header_t *)bh->b_data;
490
491 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
492 brelse(bh);
493 break;
494 }
495
496 blocktype = be32_to_cpu(tmp->h_blocktype);
497 sequence = be32_to_cpu(tmp->h_sequence);
498 jbd_debug(3, "Found magic %d, sequence %d\n",
499 blocktype, sequence);
500
501 if (sequence != next_commit_ID) {
502 brelse(bh);
503 break;
504 }
505
506 /* OK, we have a valid descriptor block which matches
507 * all of the sequence number checks. What are we going
508 * to do with it? That depends on the pass... */
509
510 switch(blocktype) {
511 case JBD2_DESCRIPTOR_BLOCK:
512 /* Verify checksum first */
513 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
514 JBD2_FEATURE_INCOMPAT_CSUM_V2))
515 descr_csum_size =
516 sizeof(struct jbd2_journal_block_tail);
517 if (descr_csum_size > 0 &&
518 !jbd2_descr_block_csum_verify(journal,
519 bh->b_data)) {
520 err = -EIO;
521 goto failed;
522 }
523
524 /* If it is a valid descriptor block, replay it
525 * in pass REPLAY; if journal_checksums enabled, then
526 * calculate checksums in PASS_SCAN, otherwise,
527 * just skip over the blocks it describes. */
528 if (pass != PASS_REPLAY) {
529 if (pass == PASS_SCAN &&
530 JBD2_HAS_COMPAT_FEATURE(journal,
531 JBD2_FEATURE_COMPAT_CHECKSUM) &&
532 !info->end_transaction) {
533 if (calc_chksums(journal, bh,
534 &next_log_block,
535 &crc32_sum)) {
536 put_bh(bh);
537 break;
538 }
539 put_bh(bh);
540 continue;
541 }
542 next_log_block += count_tags(journal, bh);
543 wrap(journal, next_log_block);
544 put_bh(bh);
545 continue;
546 }
547
548 /* A descriptor block: we can now write all of
549 * the data blocks. Yay, useful work is finally
550 * getting done here! */
551
552 tagp = &bh->b_data[sizeof(journal_header_t)];
553 while ((tagp - bh->b_data + tag_bytes)
554 <= journal->j_blocksize - descr_csum_size) {
555 unsigned long io_block;
556
557 tag = (journal_block_tag_t *) tagp;
558 flags = be16_to_cpu(tag->t_flags);
559
560 io_block = next_log_block++;
561 wrap(journal, next_log_block);
562 err = jread(&obh, journal, io_block);
563 if (err) {
564 /* Recover what we can, but
565 * report failure at the end. */
566 success = err;
567 printk(KERN_ERR
568 "JBD2: IO error %d recovering "
569 "block %ld in log\n",
570 err, io_block);
571 } else {
572 unsigned long long blocknr;
573
574 J_ASSERT(obh != NULL);
575 blocknr = read_tag_block(tag_bytes,
576 tag);
577
578 /* If the block has been
579 * revoked, then we're all done
580 * here. */
581 if (jbd2_journal_test_revoke
582 (journal, blocknr,
583 next_commit_ID)) {
584 brelse(obh);
585 ++info->nr_revoke_hits;
586 goto skip_write;
587 }
588
589 /* Look for block corruption */
590 if (!jbd2_block_tag_csum_verify(
591 journal, tag, obh->b_data,
592 be32_to_cpu(tmp->h_sequence))) {
593 brelse(obh);
594 success = -EIO;
595 printk(KERN_ERR "JBD: Invalid "
596 "checksum recovering "
597 "block %llu in log\n",
598 blocknr);
599 continue;
600 }
601
602 /* Find a buffer for the new
603 * data being restored */
604 nbh = __getblk(journal->j_fs_dev,
605 blocknr,
606 journal->j_blocksize);
607 if (nbh == NULL) {
608 printk(KERN_ERR
609 "JBD2: Out of memory "
610 "during recovery.\n");
611 err = -ENOMEM;
612 brelse(bh);
613 brelse(obh);
614 goto failed;
615 }
616
617 lock_buffer(nbh);
618 memcpy(nbh->b_data, obh->b_data,
619 journal->j_blocksize);
620 if (flags & JBD2_FLAG_ESCAPE) {
621 *((__be32 *)nbh->b_data) =
622 cpu_to_be32(JBD2_MAGIC_NUMBER);
623 }
624
625 BUFFER_TRACE(nbh, "marking dirty");
626 set_buffer_uptodate(nbh);
627 mark_buffer_dirty(nbh);
628 BUFFER_TRACE(nbh, "marking uptodate");
629 ++info->nr_replays;
630 /* ll_rw_block(WRITE, 1, &nbh); */
631 unlock_buffer(nbh);
632 brelse(obh);
633 brelse(nbh);
634 }
635
636 skip_write:
637 tagp += tag_bytes;
638 if (!(flags & JBD2_FLAG_SAME_UUID))
639 tagp += 16;
640
641 if (flags & JBD2_FLAG_LAST_TAG)
642 break;
643 }
644
645 brelse(bh);
646 continue;
647
648 case JBD2_COMMIT_BLOCK:
649 /* How to differentiate between interrupted commit
650 * and journal corruption ?
651 *
652 * {nth transaction}
653 * Checksum Verification Failed
654 * |
655 * ____________________
656 * | |
657 * async_commit sync_commit
658 * | |
659 * | GO TO NEXT "Journal Corruption"
660 * | TRANSACTION
661 * |
662 * {(n+1)th transanction}
663 * |
664 * _______|______________
665 * | |
666 * Commit block found Commit block not found
667 * | |
668 * "Journal Corruption" |
669 * _____________|_________
670 * | |
671 * nth trans corrupt OR nth trans
672 * and (n+1)th interrupted interrupted
673 * before commit block
674 * could reach the disk.
675 * (Cannot find the difference in above
676 * mentioned conditions. Hence assume
677 * "Interrupted Commit".)
678 */
679
680 /* Found an expected commit block: if checksums
681 * are present verify them in PASS_SCAN; else not
682 * much to do other than move on to the next sequence
683 * number. */
684 if (pass == PASS_SCAN &&
685 JBD2_HAS_COMPAT_FEATURE(journal,
686 JBD2_FEATURE_COMPAT_CHECKSUM)) {
687 int chksum_err, chksum_seen;
688 struct commit_header *cbh =
689 (struct commit_header *)bh->b_data;
690 unsigned found_chksum =
691 be32_to_cpu(cbh->h_chksum[0]);
692
693 chksum_err = chksum_seen = 0;
694
695 if (info->end_transaction) {
696 journal->j_failed_commit =
697 info->end_transaction;
698 brelse(bh);
699 break;
700 }
701
702 if (crc32_sum == found_chksum &&
703 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
704 cbh->h_chksum_size ==
705 JBD2_CRC32_CHKSUM_SIZE)
706 chksum_seen = 1;
707 else if (!(cbh->h_chksum_type == 0 &&
708 cbh->h_chksum_size == 0 &&
709 found_chksum == 0 &&
710 !chksum_seen))
711 /*
712 * If fs is mounted using an old kernel and then
713 * kernel with journal_chksum is used then we
714 * get a situation where the journal flag has
715 * checksum flag set but checksums are not
716 * present i.e chksum = 0, in the individual
717 * commit blocks.
718 * Hence to avoid checksum failures, in this
719 * situation, this extra check is added.
720 */
721 chksum_err = 1;
722
723 if (chksum_err) {
724 info->end_transaction = next_commit_ID;
725
726 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
727 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
728 journal->j_failed_commit =
729 next_commit_ID;
730 brelse(bh);
731 break;
732 }
733 }
734 crc32_sum = ~0;
735 }
736 if (pass == PASS_SCAN &&
737 !jbd2_commit_block_csum_verify(journal,
738 bh->b_data)) {
739 info->end_transaction = next_commit_ID;
740
741 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
742 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
743 journal->j_failed_commit =
744 next_commit_ID;
745 brelse(bh);
746 break;
747 }
748 }
749 brelse(bh);
750 next_commit_ID++;
751 continue;
752
753 case JBD2_REVOKE_BLOCK:
754 /* If we aren't in the REVOKE pass, then we can
755 * just skip over this block. */
756 if (pass != PASS_REVOKE) {
757 brelse(bh);
758 continue;
759 }
760
761 err = scan_revoke_records(journal, bh,
762 next_commit_ID, info);
763 brelse(bh);
764 if (err)
765 goto failed;
766 continue;
767
768 default:
769 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
770 blocktype);
771 brelse(bh);
772 goto done;
773 }
774 }
775
776 done:
777 /*
778 * We broke out of the log scan loop: either we came to the
779 * known end of the log or we found an unexpected block in the
780 * log. If the latter happened, then we know that the "current"
781 * transaction marks the end of the valid log.
782 */
783
784 if (pass == PASS_SCAN) {
785 if (!info->end_transaction)
786 info->end_transaction = next_commit_ID;
787 } else {
788 /* It's really bad news if different passes end up at
789 * different places (but possible due to IO errors). */
790 if (info->end_transaction != next_commit_ID) {
791 printk(KERN_ERR "JBD2: recovery pass %d ended at "
792 "transaction %u, expected %u\n",
793 pass, next_commit_ID, info->end_transaction);
794 if (!success)
795 success = -EIO;
796 }
797 }
798
799 return success;
800
801 failed:
802 return err;
803}
804
805static int jbd2_revoke_block_csum_verify(journal_t *j,
806 void *buf)
807{
808 struct jbd2_journal_revoke_tail *tail;
809 __u32 provided, calculated;
810
811 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
812 return 1;
813
814 tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize -
815 sizeof(struct jbd2_journal_revoke_tail));
816 provided = tail->r_checksum;
817 tail->r_checksum = 0;
818 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
819 tail->r_checksum = provided;
820
821 provided = be32_to_cpu(provided);
822 return provided == calculated;
823}
824
825/* Scan a revoke record, marking all blocks mentioned as revoked. */
826
827static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
828 tid_t sequence, struct recovery_info *info)
829{
830 jbd2_journal_revoke_header_t *header;
831 int offset, max;
832 int record_len = 4;
833
834 header = (jbd2_journal_revoke_header_t *) bh->b_data;
835 offset = sizeof(jbd2_journal_revoke_header_t);
836 max = be32_to_cpu(header->r_count);
837
838 if (!jbd2_revoke_block_csum_verify(journal, header))
839 return -EINVAL;
840
841 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
842 record_len = 8;
843
844 while (offset + record_len <= max) {
845 unsigned long long blocknr;
846 int err;
847
848 if (record_len == 4)
849 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
850 else
851 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
852 offset += record_len;
853 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
854 if (err)
855 return err;
856 ++info->nr_revokes;
857 }
858 return 0;
859}
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
38enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
39static int do_one_pass(journal_t *journal,
40 struct recovery_info *info, enum passtype pass);
41static int scan_revoke_records(journal_t *, struct buffer_head *,
42 tid_t, struct recovery_info *);
43
44#ifdef __KERNEL__
45
46/* Release readahead buffers after use */
47static void journal_brelse_array(struct buffer_head *b[], int n)
48{
49 while (--n >= 0)
50 brelse (b[n]);
51}
52
53
54/*
55 * When reading from the journal, we are going through the block device
56 * layer directly and so there is no readahead being done for us. We
57 * need to implement any readahead ourselves if we want it to happen at
58 * all. Recovery is basically one long sequential read, so make sure we
59 * do the IO in reasonably large chunks.
60 *
61 * This is not so critical that we need to be enormously clever about
62 * the readahead size, though. 128K is a purely arbitrary, good-enough
63 * fixed value.
64 */
65
66#define MAXBUF 8
67static int do_readahead(journal_t *journal, unsigned int start)
68{
69 int err;
70 unsigned int max, nbufs, next;
71 unsigned long long blocknr;
72 struct buffer_head *bh;
73
74 struct buffer_head * bufs[MAXBUF];
75
76 /* Do up to 128K of readahead */
77 max = start + (128 * 1024 / journal->j_blocksize);
78 if (max > journal->j_maxlen)
79 max = journal->j_maxlen;
80
81 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
82 * a time to the block device IO layer. */
83
84 nbufs = 0;
85
86 for (next = start; next < max; next++) {
87 err = jbd2_journal_bmap(journal, next, &blocknr);
88
89 if (err) {
90 printk(KERN_ERR "JBD2: bad block at offset %u\n",
91 next);
92 goto failed;
93 }
94
95 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
96 if (!bh) {
97 err = -ENOMEM;
98 goto failed;
99 }
100
101 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
102 bufs[nbufs++] = bh;
103 if (nbufs == MAXBUF) {
104 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
105 journal_brelse_array(bufs, nbufs);
106 nbufs = 0;
107 }
108 } else
109 brelse(bh);
110 }
111
112 if (nbufs)
113 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
114 err = 0;
115
116failed:
117 if (nbufs)
118 journal_brelse_array(bufs, nbufs);
119 return err;
120}
121
122#endif /* __KERNEL__ */
123
124
125/*
126 * Read a block from the journal
127 */
128
129static int jread(struct buffer_head **bhp, journal_t *journal,
130 unsigned int offset)
131{
132 int err;
133 unsigned long long blocknr;
134 struct buffer_head *bh;
135
136 *bhp = NULL;
137
138 if (offset >= journal->j_maxlen) {
139 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
140 return -EFSCORRUPTED;
141 }
142
143 err = jbd2_journal_bmap(journal, offset, &blocknr);
144
145 if (err) {
146 printk(KERN_ERR "JBD2: bad block at offset %u\n",
147 offset);
148 return err;
149 }
150
151 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
152 if (!bh)
153 return -ENOMEM;
154
155 if (!buffer_uptodate(bh)) {
156 /* If this is a brand new buffer, start readahead.
157 Otherwise, we assume we are already reading it. */
158 if (!buffer_req(bh))
159 do_readahead(journal, offset);
160 wait_on_buffer(bh);
161 }
162
163 if (!buffer_uptodate(bh)) {
164 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
165 offset);
166 brelse(bh);
167 return -EIO;
168 }
169
170 *bhp = bh;
171 return 0;
172}
173
174static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
175{
176 struct jbd2_journal_block_tail *tail;
177 __be32 provided;
178 __u32 calculated;
179
180 if (!jbd2_journal_has_csum_v2or3(j))
181 return 1;
182
183 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
184 sizeof(struct jbd2_journal_block_tail));
185 provided = tail->t_checksum;
186 tail->t_checksum = 0;
187 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
188 tail->t_checksum = provided;
189
190 return provided == cpu_to_be32(calculated);
191}
192
193/*
194 * Count the number of in-use tags in a journal descriptor block.
195 */
196
197static int count_tags(journal_t *journal, struct buffer_head *bh)
198{
199 char * tagp;
200 journal_block_tag_t * tag;
201 int nr = 0, size = journal->j_blocksize;
202 int tag_bytes = journal_tag_bytes(journal);
203
204 if (jbd2_journal_has_csum_v2or3(journal))
205 size -= sizeof(struct jbd2_journal_block_tail);
206
207 tagp = &bh->b_data[sizeof(journal_header_t)];
208
209 while ((tagp - bh->b_data + tag_bytes) <= size) {
210 tag = (journal_block_tag_t *) tagp;
211
212 nr++;
213 tagp += tag_bytes;
214 if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
215 tagp += 16;
216
217 if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
218 break;
219 }
220
221 return nr;
222}
223
224
225/* Make sure we wrap around the log correctly! */
226#define wrap(journal, var) \
227do { \
228 if (var >= (journal)->j_last) \
229 var -= ((journal)->j_last - (journal)->j_first); \
230} while (0)
231
232/**
233 * jbd2_journal_recover - recovers a on-disk journal
234 * @journal: the journal to recover
235 *
236 * The primary function for recovering the log contents when mounting a
237 * journaled device.
238 *
239 * Recovery is done in three passes. In the first pass, we look for the
240 * end of the log. In the second, we assemble the list of revoke
241 * blocks. In the third and final pass, we replay any un-revoked blocks
242 * in the log.
243 */
244int jbd2_journal_recover(journal_t *journal)
245{
246 int err, err2;
247 journal_superblock_t * sb;
248
249 struct recovery_info info;
250
251 memset(&info, 0, sizeof(info));
252 sb = journal->j_superblock;
253
254 /*
255 * The journal superblock's s_start field (the current log head)
256 * is always zero if, and only if, the journal was cleanly
257 * unmounted.
258 */
259
260 if (!sb->s_start) {
261 jbd_debug(1, "No recovery required, last transaction %d\n",
262 be32_to_cpu(sb->s_sequence));
263 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
264 return 0;
265 }
266
267 err = do_one_pass(journal, &info, PASS_SCAN);
268 if (!err)
269 err = do_one_pass(journal, &info, PASS_REVOKE);
270 if (!err)
271 err = do_one_pass(journal, &info, PASS_REPLAY);
272
273 jbd_debug(1, "JBD2: recovery, exit status %d, "
274 "recovered transactions %u to %u\n",
275 err, info.start_transaction, info.end_transaction);
276 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
277 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
278
279 /* Restart the log at the next transaction ID, thus invalidating
280 * any existing commit records in the log. */
281 journal->j_transaction_sequence = ++info.end_transaction;
282
283 jbd2_journal_clear_revoke(journal);
284 err2 = sync_blockdev(journal->j_fs_dev);
285 if (!err)
286 err = err2;
287 /* Make sure all replayed data is on permanent storage */
288 if (journal->j_flags & JBD2_BARRIER) {
289 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
290 if (!err)
291 err = err2;
292 }
293 return err;
294}
295
296/**
297 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
298 * @journal: journal to startup
299 *
300 * Locate any valid recovery information from the journal and set up the
301 * journal structures in memory to ignore it (presumably because the
302 * caller has evidence that it is out of date).
303 * This function doesn't appear to be exported..
304 *
305 * We perform one pass over the journal to allow us to tell the user how
306 * much recovery information is being erased, and to let us initialise
307 * the journal transaction sequence numbers to the next unused ID.
308 */
309int jbd2_journal_skip_recovery(journal_t *journal)
310{
311 int err;
312
313 struct recovery_info info;
314
315 memset (&info, 0, sizeof(info));
316
317 err = do_one_pass(journal, &info, PASS_SCAN);
318
319 if (err) {
320 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
321 ++journal->j_transaction_sequence;
322 } else {
323#ifdef CONFIG_JBD2_DEBUG
324 int dropped = info.end_transaction -
325 be32_to_cpu(journal->j_superblock->s_sequence);
326 jbd_debug(1,
327 "JBD2: ignoring %d transaction%s from the journal.\n",
328 dropped, (dropped == 1) ? "" : "s");
329#endif
330 journal->j_transaction_sequence = ++info.end_transaction;
331 }
332
333 journal->j_tail = 0;
334 return err;
335}
336
337static inline unsigned long long read_tag_block(journal_t *journal,
338 journal_block_tag_t *tag)
339{
340 unsigned long long block = be32_to_cpu(tag->t_blocknr);
341 if (jbd2_has_feature_64bit(journal))
342 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
343 return block;
344}
345
346/*
347 * calc_chksums calculates the checksums for the blocks described in the
348 * descriptor block.
349 */
350static int calc_chksums(journal_t *journal, struct buffer_head *bh,
351 unsigned long *next_log_block, __u32 *crc32_sum)
352{
353 int i, num_blks, err;
354 unsigned long io_block;
355 struct buffer_head *obh;
356
357 num_blks = count_tags(journal, bh);
358 /* Calculate checksum of the descriptor block. */
359 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
360
361 for (i = 0; i < num_blks; i++) {
362 io_block = (*next_log_block)++;
363 wrap(journal, *next_log_block);
364 err = jread(&obh, journal, io_block);
365 if (err) {
366 printk(KERN_ERR "JBD2: IO error %d recovering block "
367 "%lu in log\n", err, io_block);
368 return 1;
369 } else {
370 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
371 obh->b_size);
372 }
373 put_bh(obh);
374 }
375 return 0;
376}
377
378static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
379{
380 struct commit_header *h;
381 __be32 provided;
382 __u32 calculated;
383
384 if (!jbd2_journal_has_csum_v2or3(j))
385 return 1;
386
387 h = buf;
388 provided = h->h_chksum[0];
389 h->h_chksum[0] = 0;
390 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
391 h->h_chksum[0] = provided;
392
393 return provided == cpu_to_be32(calculated);
394}
395
396static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
397 void *buf, __u32 sequence)
398{
399 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
400 __u32 csum32;
401 __be32 seq;
402
403 if (!jbd2_journal_has_csum_v2or3(j))
404 return 1;
405
406 seq = cpu_to_be32(sequence);
407 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
408 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
409
410 if (jbd2_has_feature_csum3(j))
411 return tag3->t_checksum == cpu_to_be32(csum32);
412 else
413 return tag->t_checksum == cpu_to_be16(csum32);
414}
415
416static int do_one_pass(journal_t *journal,
417 struct recovery_info *info, enum passtype pass)
418{
419 unsigned int first_commit_ID, next_commit_ID;
420 unsigned long next_log_block;
421 int err, success = 0;
422 journal_superblock_t * sb;
423 journal_header_t * tmp;
424 struct buffer_head * bh;
425 unsigned int sequence;
426 int blocktype;
427 int tag_bytes = journal_tag_bytes(journal);
428 __u32 crc32_sum = ~0; /* Transactional Checksums */
429 int descr_csum_size = 0;
430 int block_error = 0;
431
432 /*
433 * First thing is to establish what we expect to find in the log
434 * (in terms of transaction IDs), and where (in terms of log
435 * block offsets): query the superblock.
436 */
437
438 sb = journal->j_superblock;
439 next_commit_ID = be32_to_cpu(sb->s_sequence);
440 next_log_block = be32_to_cpu(sb->s_start);
441
442 first_commit_ID = next_commit_ID;
443 if (pass == PASS_SCAN)
444 info->start_transaction = first_commit_ID;
445
446 jbd_debug(1, "Starting recovery pass %d\n", pass);
447
448 /*
449 * Now we walk through the log, transaction by transaction,
450 * making sure that each transaction has a commit block in the
451 * expected place. Each complete transaction gets replayed back
452 * into the main filesystem.
453 */
454
455 while (1) {
456 int flags;
457 char * tagp;
458 journal_block_tag_t * tag;
459 struct buffer_head * obh;
460 struct buffer_head * nbh;
461
462 cond_resched();
463
464 /* If we already know where to stop the log traversal,
465 * check right now that we haven't gone past the end of
466 * the log. */
467
468 if (pass != PASS_SCAN)
469 if (tid_geq(next_commit_ID, info->end_transaction))
470 break;
471
472 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
473 next_commit_ID, next_log_block, journal->j_last);
474
475 /* Skip over each chunk of the transaction looking
476 * either the next descriptor block or the final commit
477 * record. */
478
479 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
480 err = jread(&bh, journal, next_log_block);
481 if (err)
482 goto failed;
483
484 next_log_block++;
485 wrap(journal, next_log_block);
486
487 /* What kind of buffer is it?
488 *
489 * If it is a descriptor block, check that it has the
490 * expected sequence number. Otherwise, we're all done
491 * here. */
492
493 tmp = (journal_header_t *)bh->b_data;
494
495 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
496 brelse(bh);
497 break;
498 }
499
500 blocktype = be32_to_cpu(tmp->h_blocktype);
501 sequence = be32_to_cpu(tmp->h_sequence);
502 jbd_debug(3, "Found magic %d, sequence %d\n",
503 blocktype, sequence);
504
505 if (sequence != next_commit_ID) {
506 brelse(bh);
507 break;
508 }
509
510 /* OK, we have a valid descriptor block which matches
511 * all of the sequence number checks. What are we going
512 * to do with it? That depends on the pass... */
513
514 switch(blocktype) {
515 case JBD2_DESCRIPTOR_BLOCK:
516 /* Verify checksum first */
517 if (jbd2_journal_has_csum_v2or3(journal))
518 descr_csum_size =
519 sizeof(struct jbd2_journal_block_tail);
520 if (descr_csum_size > 0 &&
521 !jbd2_descriptor_block_csum_verify(journal,
522 bh->b_data)) {
523 printk(KERN_ERR "JBD2: Invalid checksum "
524 "recovering block %lu in log\n",
525 next_log_block);
526 err = -EFSBADCRC;
527 brelse(bh);
528 goto failed;
529 }
530
531 /* If it is a valid descriptor block, replay it
532 * in pass REPLAY; if journal_checksums enabled, then
533 * calculate checksums in PASS_SCAN, otherwise,
534 * just skip over the blocks it describes. */
535 if (pass != PASS_REPLAY) {
536 if (pass == PASS_SCAN &&
537 jbd2_has_feature_checksum(journal) &&
538 !info->end_transaction) {
539 if (calc_chksums(journal, bh,
540 &next_log_block,
541 &crc32_sum)) {
542 put_bh(bh);
543 break;
544 }
545 put_bh(bh);
546 continue;
547 }
548 next_log_block += count_tags(journal, bh);
549 wrap(journal, next_log_block);
550 put_bh(bh);
551 continue;
552 }
553
554 /* A descriptor block: we can now write all of
555 * the data blocks. Yay, useful work is finally
556 * getting done here! */
557
558 tagp = &bh->b_data[sizeof(journal_header_t)];
559 while ((tagp - bh->b_data + tag_bytes)
560 <= journal->j_blocksize - descr_csum_size) {
561 unsigned long io_block;
562
563 tag = (journal_block_tag_t *) tagp;
564 flags = be16_to_cpu(tag->t_flags);
565
566 io_block = next_log_block++;
567 wrap(journal, next_log_block);
568 err = jread(&obh, journal, io_block);
569 if (err) {
570 /* Recover what we can, but
571 * report failure at the end. */
572 success = err;
573 printk(KERN_ERR
574 "JBD2: IO error %d recovering "
575 "block %ld in log\n",
576 err, io_block);
577 } else {
578 unsigned long long blocknr;
579
580 J_ASSERT(obh != NULL);
581 blocknr = read_tag_block(journal,
582 tag);
583
584 /* If the block has been
585 * revoked, then we're all done
586 * here. */
587 if (jbd2_journal_test_revoke
588 (journal, blocknr,
589 next_commit_ID)) {
590 brelse(obh);
591 ++info->nr_revoke_hits;
592 goto skip_write;
593 }
594
595 /* Look for block corruption */
596 if (!jbd2_block_tag_csum_verify(
597 journal, tag, obh->b_data,
598 be32_to_cpu(tmp->h_sequence))) {
599 brelse(obh);
600 success = -EFSBADCRC;
601 printk(KERN_ERR "JBD2: Invalid "
602 "checksum recovering "
603 "data block %llu in "
604 "log\n", blocknr);
605 block_error = 1;
606 goto skip_write;
607 }
608
609 /* Find a buffer for the new
610 * data being restored */
611 nbh = __getblk(journal->j_fs_dev,
612 blocknr,
613 journal->j_blocksize);
614 if (nbh == NULL) {
615 printk(KERN_ERR
616 "JBD2: Out of memory "
617 "during recovery.\n");
618 err = -ENOMEM;
619 brelse(bh);
620 brelse(obh);
621 goto failed;
622 }
623
624 lock_buffer(nbh);
625 memcpy(nbh->b_data, obh->b_data,
626 journal->j_blocksize);
627 if (flags & JBD2_FLAG_ESCAPE) {
628 *((__be32 *)nbh->b_data) =
629 cpu_to_be32(JBD2_MAGIC_NUMBER);
630 }
631
632 BUFFER_TRACE(nbh, "marking dirty");
633 set_buffer_uptodate(nbh);
634 mark_buffer_dirty(nbh);
635 BUFFER_TRACE(nbh, "marking uptodate");
636 ++info->nr_replays;
637 /* ll_rw_block(WRITE, 1, &nbh); */
638 unlock_buffer(nbh);
639 brelse(obh);
640 brelse(nbh);
641 }
642
643 skip_write:
644 tagp += tag_bytes;
645 if (!(flags & JBD2_FLAG_SAME_UUID))
646 tagp += 16;
647
648 if (flags & JBD2_FLAG_LAST_TAG)
649 break;
650 }
651
652 brelse(bh);
653 continue;
654
655 case JBD2_COMMIT_BLOCK:
656 /* How to differentiate between interrupted commit
657 * and journal corruption ?
658 *
659 * {nth transaction}
660 * Checksum Verification Failed
661 * |
662 * ____________________
663 * | |
664 * async_commit sync_commit
665 * | |
666 * | GO TO NEXT "Journal Corruption"
667 * | TRANSACTION
668 * |
669 * {(n+1)th transanction}
670 * |
671 * _______|______________
672 * | |
673 * Commit block found Commit block not found
674 * | |
675 * "Journal Corruption" |
676 * _____________|_________
677 * | |
678 * nth trans corrupt OR nth trans
679 * and (n+1)th interrupted interrupted
680 * before commit block
681 * could reach the disk.
682 * (Cannot find the difference in above
683 * mentioned conditions. Hence assume
684 * "Interrupted Commit".)
685 */
686
687 /* Found an expected commit block: if checksums
688 * are present verify them in PASS_SCAN; else not
689 * much to do other than move on to the next sequence
690 * number. */
691 if (pass == PASS_SCAN &&
692 jbd2_has_feature_checksum(journal)) {
693 int chksum_err, chksum_seen;
694 struct commit_header *cbh =
695 (struct commit_header *)bh->b_data;
696 unsigned found_chksum =
697 be32_to_cpu(cbh->h_chksum[0]);
698
699 chksum_err = chksum_seen = 0;
700
701 if (info->end_transaction) {
702 journal->j_failed_commit =
703 info->end_transaction;
704 brelse(bh);
705 break;
706 }
707
708 if (crc32_sum == found_chksum &&
709 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
710 cbh->h_chksum_size ==
711 JBD2_CRC32_CHKSUM_SIZE)
712 chksum_seen = 1;
713 else if (!(cbh->h_chksum_type == 0 &&
714 cbh->h_chksum_size == 0 &&
715 found_chksum == 0 &&
716 !chksum_seen))
717 /*
718 * If fs is mounted using an old kernel and then
719 * kernel with journal_chksum is used then we
720 * get a situation where the journal flag has
721 * checksum flag set but checksums are not
722 * present i.e chksum = 0, in the individual
723 * commit blocks.
724 * Hence to avoid checksum failures, in this
725 * situation, this extra check is added.
726 */
727 chksum_err = 1;
728
729 if (chksum_err) {
730 info->end_transaction = next_commit_ID;
731
732 if (!jbd2_has_feature_async_commit(journal)) {
733 journal->j_failed_commit =
734 next_commit_ID;
735 brelse(bh);
736 break;
737 }
738 }
739 crc32_sum = ~0;
740 }
741 if (pass == PASS_SCAN &&
742 !jbd2_commit_block_csum_verify(journal,
743 bh->b_data)) {
744 info->end_transaction = next_commit_ID;
745
746 if (!jbd2_has_feature_async_commit(journal)) {
747 journal->j_failed_commit =
748 next_commit_ID;
749 brelse(bh);
750 break;
751 }
752 }
753 brelse(bh);
754 next_commit_ID++;
755 continue;
756
757 case JBD2_REVOKE_BLOCK:
758 /* If we aren't in the REVOKE pass, then we can
759 * just skip over this block. */
760 if (pass != PASS_REVOKE) {
761 brelse(bh);
762 continue;
763 }
764
765 err = scan_revoke_records(journal, bh,
766 next_commit_ID, info);
767 brelse(bh);
768 if (err)
769 goto failed;
770 continue;
771
772 default:
773 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
774 blocktype);
775 brelse(bh);
776 goto done;
777 }
778 }
779
780 done:
781 /*
782 * We broke out of the log scan loop: either we came to the
783 * known end of the log or we found an unexpected block in the
784 * log. If the latter happened, then we know that the "current"
785 * transaction marks the end of the valid log.
786 */
787
788 if (pass == PASS_SCAN) {
789 if (!info->end_transaction)
790 info->end_transaction = next_commit_ID;
791 } else {
792 /* It's really bad news if different passes end up at
793 * different places (but possible due to IO errors). */
794 if (info->end_transaction != next_commit_ID) {
795 printk(KERN_ERR "JBD2: recovery pass %d ended at "
796 "transaction %u, expected %u\n",
797 pass, next_commit_ID, info->end_transaction);
798 if (!success)
799 success = -EIO;
800 }
801 }
802 if (block_error && success == 0)
803 success = -EIO;
804 return success;
805
806 failed:
807 return err;
808}
809
810/* Scan a revoke record, marking all blocks mentioned as revoked. */
811
812static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
813 tid_t sequence, struct recovery_info *info)
814{
815 jbd2_journal_revoke_header_t *header;
816 int offset, max;
817 int csum_size = 0;
818 __u32 rcount;
819 int record_len = 4;
820
821 header = (jbd2_journal_revoke_header_t *) bh->b_data;
822 offset = sizeof(jbd2_journal_revoke_header_t);
823 rcount = be32_to_cpu(header->r_count);
824
825 if (!jbd2_descriptor_block_csum_verify(journal, header))
826 return -EFSBADCRC;
827
828 if (jbd2_journal_has_csum_v2or3(journal))
829 csum_size = sizeof(struct jbd2_journal_block_tail);
830 if (rcount > journal->j_blocksize - csum_size)
831 return -EINVAL;
832 max = rcount;
833
834 if (jbd2_has_feature_64bit(journal))
835 record_len = 8;
836
837 while (offset + record_len <= max) {
838 unsigned long long blocknr;
839 int err;
840
841 if (record_len == 4)
842 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
843 else
844 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
845 offset += record_len;
846 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
847 if (err)
848 return err;
849 ++info->nr_revokes;
850 }
851 return 0;
852}