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