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 */
28 struct 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
39 static int do_one_pass(journal_t *journal,
40 struct recovery_info *info, enum passtype pass);
41 static 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 */
journal_brelse_array(struct buffer_head * b[],int n)47 static 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
do_readahead(journal_t * journal,unsigned int start)67 static 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
116 failed:
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
jread(struct buffer_head ** bhp,journal_t * journal,unsigned int offset)129 static 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
jbd2_descriptor_block_csum_verify(journal_t * j,void * buf)179 static 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
count_tags(journal_t * journal,struct buffer_head * bh)202 static 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) \
232 do { \
233 if (var >= (journal)->j_last) \
234 var -= ((journal)->j_last - (journal)->j_first); \
235 } while (0)
236
fc_do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)237 static 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 */
jbd2_journal_recover(journal_t * journal)286 int 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 /* Make sure all replayed data is on permanent storage */
333 if (journal->j_flags & JBD2_BARRIER) {
334 err2 = blkdev_issue_flush(journal->j_fs_dev);
335 if (!err)
336 err = err2;
337 }
338 return err;
339 }
340
341 /**
342 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
343 * @journal: journal to startup
344 *
345 * Locate any valid recovery information from the journal and set up the
346 * journal structures in memory to ignore it (presumably because the
347 * caller has evidence that it is out of date).
348 * This function doesn't appear to be exported..
349 *
350 * We perform one pass over the journal to allow us to tell the user how
351 * much recovery information is being erased, and to let us initialise
352 * the journal transaction sequence numbers to the next unused ID.
353 */
jbd2_journal_skip_recovery(journal_t * journal)354 int jbd2_journal_skip_recovery(journal_t *journal)
355 {
356 int err;
357
358 struct recovery_info info;
359
360 memset (&info, 0, sizeof(info));
361
362 err = do_one_pass(journal, &info, PASS_SCAN);
363
364 if (err) {
365 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
366 ++journal->j_transaction_sequence;
367 journal->j_head = journal->j_first;
368 } else {
369 #ifdef CONFIG_JBD2_DEBUG
370 int dropped = info.end_transaction -
371 be32_to_cpu(journal->j_superblock->s_sequence);
372 jbd2_debug(1,
373 "JBD2: ignoring %d transaction%s from the journal.\n",
374 dropped, (dropped == 1) ? "" : "s");
375 #endif
376 journal->j_transaction_sequence = ++info.end_transaction;
377 journal->j_head = info.head_block;
378 }
379
380 journal->j_tail = 0;
381 return err;
382 }
383
read_tag_block(journal_t * journal,journal_block_tag_t * tag)384 static inline unsigned long long read_tag_block(journal_t *journal,
385 journal_block_tag_t *tag)
386 {
387 unsigned long long block = be32_to_cpu(tag->t_blocknr);
388 if (jbd2_has_feature_64bit(journal))
389 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
390 return block;
391 }
392
393 /*
394 * calc_chksums calculates the checksums for the blocks described in the
395 * descriptor block.
396 */
calc_chksums(journal_t * journal,struct buffer_head * bh,unsigned long * next_log_block,__u32 * crc32_sum)397 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
398 unsigned long *next_log_block, __u32 *crc32_sum)
399 {
400 int i, num_blks, err;
401 unsigned long io_block;
402 struct buffer_head *obh;
403
404 num_blks = count_tags(journal, bh);
405 /* Calculate checksum of the descriptor block. */
406 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
407
408 for (i = 0; i < num_blks; i++) {
409 io_block = (*next_log_block)++;
410 wrap(journal, *next_log_block);
411 err = jread(&obh, journal, io_block);
412 if (err) {
413 printk(KERN_ERR "JBD2: IO error %d recovering block "
414 "%lu in log\n", err, io_block);
415 return 1;
416 } else {
417 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
418 obh->b_size);
419 }
420 put_bh(obh);
421 }
422 return 0;
423 }
424
jbd2_commit_block_csum_verify(journal_t * j,void * buf)425 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
426 {
427 struct commit_header *h;
428 __be32 provided;
429 __u32 calculated;
430
431 if (!jbd2_journal_has_csum_v2or3(j))
432 return 1;
433
434 h = buf;
435 provided = h->h_chksum[0];
436 h->h_chksum[0] = 0;
437 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
438 h->h_chksum[0] = provided;
439
440 return provided == cpu_to_be32(calculated);
441 }
442
jbd2_block_tag_csum_verify(journal_t * j,journal_block_tag_t * tag,journal_block_tag3_t * tag3,void * buf,__u32 sequence)443 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
444 journal_block_tag3_t *tag3,
445 void *buf, __u32 sequence)
446 {
447 __u32 csum32;
448 __be32 seq;
449
450 if (!jbd2_journal_has_csum_v2or3(j))
451 return 1;
452
453 seq = cpu_to_be32(sequence);
454 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
455 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
456
457 if (jbd2_has_feature_csum3(j))
458 return tag3->t_checksum == cpu_to_be32(csum32);
459 else
460 return tag->t_checksum == cpu_to_be16(csum32);
461 }
462
do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)463 static int do_one_pass(journal_t *journal,
464 struct recovery_info *info, enum passtype pass)
465 {
466 unsigned int first_commit_ID, next_commit_ID;
467 unsigned long next_log_block, head_block;
468 int err, success = 0;
469 journal_superblock_t * sb;
470 journal_header_t * tmp;
471 struct buffer_head * bh;
472 unsigned int sequence;
473 int blocktype;
474 int tag_bytes = journal_tag_bytes(journal);
475 __u32 crc32_sum = ~0; /* Transactional Checksums */
476 int descr_csum_size = 0;
477 int block_error = 0;
478 bool need_check_commit_time = false;
479 __u64 last_trans_commit_time = 0, commit_time;
480
481 /*
482 * First thing is to establish what we expect to find in the log
483 * (in terms of transaction IDs), and where (in terms of log
484 * block offsets): query the superblock.
485 */
486
487 sb = journal->j_superblock;
488 next_commit_ID = be32_to_cpu(sb->s_sequence);
489 next_log_block = be32_to_cpu(sb->s_start);
490 head_block = next_log_block;
491
492 first_commit_ID = next_commit_ID;
493 if (pass == PASS_SCAN)
494 info->start_transaction = first_commit_ID;
495
496 jbd2_debug(1, "Starting recovery pass %d\n", pass);
497
498 /*
499 * Now we walk through the log, transaction by transaction,
500 * making sure that each transaction has a commit block in the
501 * expected place. Each complete transaction gets replayed back
502 * into the main filesystem.
503 */
504
505 while (1) {
506 int flags;
507 char * tagp;
508 journal_block_tag_t tag;
509 struct buffer_head * obh;
510 struct buffer_head * nbh;
511
512 cond_resched();
513
514 /* If we already know where to stop the log traversal,
515 * check right now that we haven't gone past the end of
516 * the log. */
517
518 if (pass != PASS_SCAN)
519 if (tid_geq(next_commit_ID, info->end_transaction))
520 break;
521
522 jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
523 next_commit_ID, next_log_block, journal->j_last);
524
525 /* Skip over each chunk of the transaction looking
526 * either the next descriptor block or the final commit
527 * record. */
528
529 jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
530 err = jread(&bh, journal, next_log_block);
531 if (err)
532 goto failed;
533
534 next_log_block++;
535 wrap(journal, next_log_block);
536
537 /* What kind of buffer is it?
538 *
539 * If it is a descriptor block, check that it has the
540 * expected sequence number. Otherwise, we're all done
541 * here. */
542
543 tmp = (journal_header_t *)bh->b_data;
544
545 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
546 brelse(bh);
547 break;
548 }
549
550 blocktype = be32_to_cpu(tmp->h_blocktype);
551 sequence = be32_to_cpu(tmp->h_sequence);
552 jbd2_debug(3, "Found magic %d, sequence %d\n",
553 blocktype, sequence);
554
555 if (sequence != next_commit_ID) {
556 brelse(bh);
557 break;
558 }
559
560 /* OK, we have a valid descriptor block which matches
561 * all of the sequence number checks. What are we going
562 * to do with it? That depends on the pass... */
563
564 switch(blocktype) {
565 case JBD2_DESCRIPTOR_BLOCK:
566 /* Verify checksum first */
567 if (jbd2_journal_has_csum_v2or3(journal))
568 descr_csum_size =
569 sizeof(struct jbd2_journal_block_tail);
570 if (descr_csum_size > 0 &&
571 !jbd2_descriptor_block_csum_verify(journal,
572 bh->b_data)) {
573 /*
574 * PASS_SCAN can see stale blocks due to lazy
575 * journal init. Don't error out on those yet.
576 */
577 if (pass != PASS_SCAN) {
578 pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
579 next_log_block);
580 err = -EFSBADCRC;
581 brelse(bh);
582 goto failed;
583 }
584 need_check_commit_time = true;
585 jbd2_debug(1,
586 "invalid descriptor block found in %lu\n",
587 next_log_block);
588 }
589
590 /* If it is a valid descriptor block, replay it
591 * in pass REPLAY; if journal_checksums enabled, then
592 * calculate checksums in PASS_SCAN, otherwise,
593 * just skip over the blocks it describes. */
594 if (pass != PASS_REPLAY) {
595 if (pass == PASS_SCAN &&
596 jbd2_has_feature_checksum(journal) &&
597 !need_check_commit_time &&
598 !info->end_transaction) {
599 if (calc_chksums(journal, bh,
600 &next_log_block,
601 &crc32_sum)) {
602 put_bh(bh);
603 break;
604 }
605 put_bh(bh);
606 continue;
607 }
608 next_log_block += count_tags(journal, bh);
609 wrap(journal, next_log_block);
610 put_bh(bh);
611 continue;
612 }
613
614 /* A descriptor block: we can now write all of
615 * the data blocks. Yay, useful work is finally
616 * getting done here! */
617
618 tagp = &bh->b_data[sizeof(journal_header_t)];
619 while ((tagp - bh->b_data + tag_bytes)
620 <= journal->j_blocksize - descr_csum_size) {
621 unsigned long io_block;
622
623 memcpy(&tag, tagp, sizeof(tag));
624 flags = be16_to_cpu(tag.t_flags);
625
626 io_block = next_log_block++;
627 wrap(journal, next_log_block);
628 err = jread(&obh, journal, io_block);
629 if (err) {
630 /* Recover what we can, but
631 * report failure at the end. */
632 success = err;
633 printk(KERN_ERR
634 "JBD2: IO error %d recovering "
635 "block %ld in log\n",
636 err, io_block);
637 } else {
638 unsigned long long blocknr;
639
640 J_ASSERT(obh != NULL);
641 blocknr = read_tag_block(journal,
642 &tag);
643
644 /* If the block has been
645 * revoked, then we're all done
646 * here. */
647 if (jbd2_journal_test_revoke
648 (journal, blocknr,
649 next_commit_ID)) {
650 brelse(obh);
651 ++info->nr_revoke_hits;
652 goto skip_write;
653 }
654
655 /* Look for block corruption */
656 if (!jbd2_block_tag_csum_verify(
657 journal, &tag, (journal_block_tag3_t *)tagp,
658 obh->b_data, be32_to_cpu(tmp->h_sequence))) {
659 brelse(obh);
660 success = -EFSBADCRC;
661 printk(KERN_ERR "JBD2: Invalid "
662 "checksum recovering "
663 "data block %llu in "
664 "log\n", blocknr);
665 block_error = 1;
666 goto skip_write;
667 }
668
669 /* Find a buffer for the new
670 * data being restored */
671 nbh = __getblk(journal->j_fs_dev,
672 blocknr,
673 journal->j_blocksize);
674 if (nbh == NULL) {
675 printk(KERN_ERR
676 "JBD2: Out of memory "
677 "during recovery.\n");
678 err = -ENOMEM;
679 brelse(bh);
680 brelse(obh);
681 goto failed;
682 }
683
684 lock_buffer(nbh);
685 memcpy(nbh->b_data, obh->b_data,
686 journal->j_blocksize);
687 if (flags & JBD2_FLAG_ESCAPE) {
688 *((__be32 *)nbh->b_data) =
689 cpu_to_be32(JBD2_MAGIC_NUMBER);
690 }
691
692 BUFFER_TRACE(nbh, "marking dirty");
693 set_buffer_uptodate(nbh);
694 mark_buffer_dirty(nbh);
695 BUFFER_TRACE(nbh, "marking uptodate");
696 ++info->nr_replays;
697 unlock_buffer(nbh);
698 brelse(obh);
699 brelse(nbh);
700 }
701
702 skip_write:
703 tagp += tag_bytes;
704 if (!(flags & JBD2_FLAG_SAME_UUID))
705 tagp += 16;
706
707 if (flags & JBD2_FLAG_LAST_TAG)
708 break;
709 }
710
711 brelse(bh);
712 continue;
713
714 case JBD2_COMMIT_BLOCK:
715 /* How to differentiate between interrupted commit
716 * and journal corruption ?
717 *
718 * {nth transaction}
719 * Checksum Verification Failed
720 * |
721 * ____________________
722 * | |
723 * async_commit sync_commit
724 * | |
725 * | GO TO NEXT "Journal Corruption"
726 * | TRANSACTION
727 * |
728 * {(n+1)th transanction}
729 * |
730 * _______|______________
731 * | |
732 * Commit block found Commit block not found
733 * | |
734 * "Journal Corruption" |
735 * _____________|_________
736 * | |
737 * nth trans corrupt OR nth trans
738 * and (n+1)th interrupted interrupted
739 * before commit block
740 * could reach the disk.
741 * (Cannot find the difference in above
742 * mentioned conditions. Hence assume
743 * "Interrupted Commit".)
744 */
745 commit_time = be64_to_cpu(
746 ((struct commit_header *)bh->b_data)->h_commit_sec);
747 /*
748 * If need_check_commit_time is set, it means we are in
749 * PASS_SCAN and csum verify failed before. If
750 * commit_time is increasing, it's the same journal,
751 * otherwise it is stale journal block, just end this
752 * recovery.
753 */
754 if (need_check_commit_time) {
755 if (commit_time >= last_trans_commit_time) {
756 pr_err("JBD2: Invalid checksum found in transaction %u\n",
757 next_commit_ID);
758 err = -EFSBADCRC;
759 brelse(bh);
760 goto failed;
761 }
762 ignore_crc_mismatch:
763 /*
764 * It likely does not belong to same journal,
765 * just end this recovery with success.
766 */
767 jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
768 next_commit_ID);
769 brelse(bh);
770 goto done;
771 }
772
773 /*
774 * Found an expected commit block: if checksums
775 * are present, verify them in PASS_SCAN; else not
776 * much to do other than move on to the next sequence
777 * number.
778 */
779 if (pass == PASS_SCAN &&
780 jbd2_has_feature_checksum(journal)) {
781 struct commit_header *cbh =
782 (struct commit_header *)bh->b_data;
783 unsigned found_chksum =
784 be32_to_cpu(cbh->h_chksum[0]);
785
786 if (info->end_transaction) {
787 journal->j_failed_commit =
788 info->end_transaction;
789 brelse(bh);
790 break;
791 }
792
793 /* Neither checksum match nor unused? */
794 if (!((crc32_sum == found_chksum &&
795 cbh->h_chksum_type ==
796 JBD2_CRC32_CHKSUM &&
797 cbh->h_chksum_size ==
798 JBD2_CRC32_CHKSUM_SIZE) ||
799 (cbh->h_chksum_type == 0 &&
800 cbh->h_chksum_size == 0 &&
801 found_chksum == 0)))
802 goto chksum_error;
803
804 crc32_sum = ~0;
805 }
806 if (pass == PASS_SCAN &&
807 !jbd2_commit_block_csum_verify(journal,
808 bh->b_data)) {
809 chksum_error:
810 if (commit_time < last_trans_commit_time)
811 goto ignore_crc_mismatch;
812 info->end_transaction = next_commit_ID;
813 info->head_block = head_block;
814
815 if (!jbd2_has_feature_async_commit(journal)) {
816 journal->j_failed_commit =
817 next_commit_ID;
818 brelse(bh);
819 break;
820 }
821 }
822 if (pass == PASS_SCAN) {
823 last_trans_commit_time = commit_time;
824 head_block = next_log_block;
825 }
826 brelse(bh);
827 next_commit_ID++;
828 continue;
829
830 case JBD2_REVOKE_BLOCK:
831 /*
832 * Check revoke block crc in pass_scan, if csum verify
833 * failed, check commit block time later.
834 */
835 if (pass == PASS_SCAN &&
836 !jbd2_descriptor_block_csum_verify(journal,
837 bh->b_data)) {
838 jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
839 next_log_block);
840 need_check_commit_time = true;
841 }
842 /* If we aren't in the REVOKE pass, then we can
843 * just skip over this block. */
844 if (pass != PASS_REVOKE) {
845 brelse(bh);
846 continue;
847 }
848
849 err = scan_revoke_records(journal, bh,
850 next_commit_ID, info);
851 brelse(bh);
852 if (err)
853 goto failed;
854 continue;
855
856 default:
857 jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
858 blocktype);
859 brelse(bh);
860 goto done;
861 }
862 }
863
864 done:
865 /*
866 * We broke out of the log scan loop: either we came to the
867 * known end of the log or we found an unexpected block in the
868 * log. If the latter happened, then we know that the "current"
869 * transaction marks the end of the valid log.
870 */
871
872 if (pass == PASS_SCAN) {
873 if (!info->end_transaction)
874 info->end_transaction = next_commit_ID;
875 if (!info->head_block)
876 info->head_block = head_block;
877 } else {
878 /* It's really bad news if different passes end up at
879 * different places (but possible due to IO errors). */
880 if (info->end_transaction != next_commit_ID) {
881 printk(KERN_ERR "JBD2: recovery pass %d ended at "
882 "transaction %u, expected %u\n",
883 pass, next_commit_ID, info->end_transaction);
884 if (!success)
885 success = -EIO;
886 }
887 }
888
889 if (jbd2_has_feature_fast_commit(journal) && pass != PASS_REVOKE) {
890 err = fc_do_one_pass(journal, info, pass);
891 if (err)
892 success = err;
893 }
894
895 if (block_error && success == 0)
896 success = -EIO;
897 return success;
898
899 failed:
900 return err;
901 }
902
903 /* Scan a revoke record, marking all blocks mentioned as revoked. */
904
scan_revoke_records(journal_t * journal,struct buffer_head * bh,tid_t sequence,struct recovery_info * info)905 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
906 tid_t sequence, struct recovery_info *info)
907 {
908 jbd2_journal_revoke_header_t *header;
909 int offset, max;
910 unsigned csum_size = 0;
911 __u32 rcount;
912 int record_len = 4;
913
914 header = (jbd2_journal_revoke_header_t *) bh->b_data;
915 offset = sizeof(jbd2_journal_revoke_header_t);
916 rcount = be32_to_cpu(header->r_count);
917
918 if (jbd2_journal_has_csum_v2or3(journal))
919 csum_size = sizeof(struct jbd2_journal_block_tail);
920 if (rcount > journal->j_blocksize - csum_size)
921 return -EINVAL;
922 max = rcount;
923
924 if (jbd2_has_feature_64bit(journal))
925 record_len = 8;
926
927 while (offset + record_len <= max) {
928 unsigned long long blocknr;
929 int err;
930
931 if (record_len == 4)
932 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
933 else
934 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
935 offset += record_len;
936 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
937 if (err)
938 return err;
939 ++info->nr_revokes;
940 }
941 return 0;
942 }
943
