blob: 5b43e96788e6553ec0a716baa45e1a1ac509e838 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd/recovery.c
Mingming Caoae6ddcc2006-09-27 01:49:27 -07003 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 * 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;
Mingming Caoae6ddcc2006-09-27 01:49:27 -070013 * part of the ext2fs journaling system.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 */
15
16#ifndef __KERNEL__
17#include "jfs_user.h"
18#else
19#include <linux/time.h>
20#include <linux/fs.h>
21#include <linux/jbd.h>
22#include <linux/errno.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#endif
24
25/*
26 * Maintain information about the progress of the recovery job, so that
Mingming Caoae6ddcc2006-09-27 01:49:27 -070027 * the different passes can carry information between them.
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -070029struct recovery_info
Linus Torvalds1da177e2005-04-16 15:20:36 -070030{
31 tid_t start_transaction;
32 tid_t end_transaction;
33
34 int nr_replays;
35 int nr_revokes;
36 int nr_revoke_hits;
37};
38
39enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
40static int do_one_pass(journal_t *journal,
41 struct recovery_info *info, enum passtype pass);
42static int scan_revoke_records(journal_t *, struct buffer_head *,
43 tid_t, struct recovery_info *);
44
45#ifdef __KERNEL__
46
47/* Release readahead buffers after use */
Dave Kleikampf71b2f12006-09-29 01:58:39 -070048static void journal_brelse_array(struct buffer_head *b[], int n)
Linus Torvalds1da177e2005-04-16 15:20:36 -070049{
50 while (--n >= 0)
51 brelse (b[n]);
52}
53
54
55/*
56 * When reading from the journal, we are going through the block device
57 * layer directly and so there is no readahead being done for us. We
58 * need to implement any readahead ourselves if we want it to happen at
59 * all. Recovery is basically one long sequential read, so make sure we
60 * do the IO in reasonably large chunks.
61 *
62 * This is not so critical that we need to be enormously clever about
63 * the readahead size, though. 128K is a purely arbitrary, good-enough
64 * fixed value.
65 */
66
67#define MAXBUF 8
68static int do_readahead(journal_t *journal, unsigned int start)
69{
70 int err;
71 unsigned int max, nbufs, next;
Jan Kara9c28cbc2009-08-03 19:21:00 +020072 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 struct buffer_head *bh;
74
75 struct buffer_head * bufs[MAXBUF];
76
77 /* Do up to 128K of readahead */
78 max = start + (128 * 1024 / journal->j_blocksize);
79 if (max > journal->j_maxlen)
80 max = journal->j_maxlen;
81
82 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
83 * a time to the block device IO layer. */
84
85 nbufs = 0;
86
87 for (next = start; next < max; next++) {
88 err = journal_bmap(journal, next, &blocknr);
89
90 if (err) {
91 printk (KERN_ERR "JBD: bad block at offset %u\n",
92 next);
93 goto failed;
94 }
95
96 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
97 if (!bh) {
98 err = -ENOMEM;
99 goto failed;
100 }
101
102 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
103 bufs[nbufs++] = bh;
104 if (nbufs == MAXBUF) {
105 ll_rw_block(READ, nbufs, bufs);
106 journal_brelse_array(bufs, nbufs);
107 nbufs = 0;
108 }
109 } else
110 brelse(bh);
111 }
112
113 if (nbufs)
114 ll_rw_block(READ, nbufs, bufs);
115 err = 0;
116
117failed:
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700118 if (nbufs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 journal_brelse_array(bufs, nbufs);
120 return err;
121}
122
123#endif /* __KERNEL__ */
124
125
126/*
127 * Read a block from the journal
128 */
129
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700130static int jread(struct buffer_head **bhp, journal_t *journal,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 unsigned int offset)
132{
133 int err;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200134 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 struct buffer_head *bh;
136
137 *bhp = NULL;
138
139 if (offset >= journal->j_maxlen) {
140 printk(KERN_ERR "JBD: corrupted journal superblock\n");
141 return -EIO;
142 }
143
144 err = journal_bmap(journal, offset, &blocknr);
145
146 if (err) {
147 printk (KERN_ERR "JBD: bad block at offset %u\n",
148 offset);
149 return err;
150 }
151
152 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
153 if (!bh)
154 return -ENOMEM;
155
156 if (!buffer_uptodate(bh)) {
157 /* If this is a brand new buffer, start readahead.
158 Otherwise, we assume we are already reading it. */
159 if (!buffer_req(bh))
160 do_readahead(journal, offset);
161 wait_on_buffer(bh);
162 }
163
164 if (!buffer_uptodate(bh)) {
165 printk (KERN_ERR "JBD: Failed to read block at offset %u\n",
166 offset);
167 brelse(bh);
168 return -EIO;
169 }
170
171 *bhp = bh;
172 return 0;
173}
174
175
176/*
177 * Count the number of in-use tags in a journal descriptor block.
178 */
179
180static int count_tags(struct buffer_head *bh, int size)
181{
182 char * tagp;
183 journal_block_tag_t * tag;
184 int nr = 0;
185
186 tagp = &bh->b_data[sizeof(journal_header_t)];
187
188 while ((tagp - bh->b_data + sizeof(journal_block_tag_t)) <= size) {
189 tag = (journal_block_tag_t *) tagp;
190
191 nr++;
192 tagp += sizeof(journal_block_tag_t);
193 if (!(tag->t_flags & cpu_to_be32(JFS_FLAG_SAME_UUID)))
194 tagp += 16;
195
196 if (tag->t_flags & cpu_to_be32(JFS_FLAG_LAST_TAG))
197 break;
198 }
199
200 return nr;
201}
202
203
204/* Make sure we wrap around the log correctly! */
205#define wrap(journal, var) \
206do { \
207 if (var >= (journal)->j_last) \
208 var -= ((journal)->j_last - (journal)->j_first); \
209} while (0)
210
211/**
Randy Dunlap6c8bec62005-11-07 01:01:04 -0800212 * journal_recover - recovers a on-disk journal
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * @journal: the journal to recover
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700214 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 * The primary function for recovering the log contents when mounting a
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700216 * journaled device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 *
218 * Recovery is done in three passes. In the first pass, we look for the
219 * end of the log. In the second, we assemble the list of revoke
220 * blocks. In the third and final pass, we replay any un-revoked blocks
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700221 * in the log.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 */
223int journal_recover(journal_t *journal)
224{
Hidehiro Kawai4afe9782008-10-22 14:15:00 -0700225 int err, err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 journal_superblock_t * sb;
227
228 struct recovery_info info;
229
230 memset(&info, 0, sizeof(info));
231 sb = journal->j_superblock;
232
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700233 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 * The journal superblock's s_start field (the current log head)
235 * is always zero if, and only if, the journal was cleanly
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700236 * unmounted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 */
238
239 if (!sb->s_start) {
240 jbd_debug(1, "No recovery required, last transaction %d\n",
241 be32_to_cpu(sb->s_sequence));
242 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
243 return 0;
244 }
245
246 err = do_one_pass(journal, &info, PASS_SCAN);
247 if (!err)
248 err = do_one_pass(journal, &info, PASS_REVOKE);
249 if (!err)
250 err = do_one_pass(journal, &info, PASS_REPLAY);
251
Jose R. Santos9ad163a2007-10-18 23:39:23 -0700252 jbd_debug(1, "JBD: recovery, exit status %d, "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 "recovered transactions %u to %u\n",
254 err, info.start_transaction, info.end_transaction);
Jose R. Santos9ad163a2007-10-18 23:39:23 -0700255 jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
257
258 /* Restart the log at the next transaction ID, thus invalidating
259 * any existing commit records in the log. */
260 journal->j_transaction_sequence = ++info.end_transaction;
261
262 journal_clear_revoke(journal);
Hidehiro Kawai4afe9782008-10-22 14:15:00 -0700263 err2 = sync_blockdev(journal->j_fs_dev);
264 if (!err)
265 err = err2;
266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return err;
268}
269
270/**
Randy Dunlap6c8bec62005-11-07 01:01:04 -0800271 * journal_skip_recovery - Start journal and wipe exiting records
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 * @journal: journal to startup
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700273 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 * Locate any valid recovery information from the journal and set up the
275 * journal structures in memory to ignore it (presumably because the
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700276 * caller has evidence that it is out of date).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * This function does'nt appear to be exorted..
278 *
279 * We perform one pass over the journal to allow us to tell the user how
280 * much recovery information is being erased, and to let us initialise
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700281 * the journal transaction sequence numbers to the next unused ID.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 */
283int journal_skip_recovery(journal_t *journal)
284{
285 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 struct recovery_info info;
287
288 memset (&info, 0, sizeof(info));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
290 err = do_one_pass(journal, &info, PASS_SCAN);
291
292 if (err) {
293 printk(KERN_ERR "JBD: error %d scanning journal\n", err);
294 ++journal->j_transaction_sequence;
295 } else {
296#ifdef CONFIG_JBD_DEBUG
Andi Kleen0411ba72010-06-10 13:10:53 +0200297 int dropped = info.end_transaction -
298 be32_to_cpu(journal->j_superblock->s_sequence);
Jose R. Santos9ad163a2007-10-18 23:39:23 -0700299 jbd_debug(1,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 "JBD: ignoring %d transaction%s from the journal.\n",
301 dropped, (dropped == 1) ? "" : "s");
Namhyung Kimc5639be2010-10-04 17:26:53 +0900302#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 journal->j_transaction_sequence = ++info.end_transaction;
304 }
305
306 journal->j_tail = 0;
307 return err;
308}
309
310static int do_one_pass(journal_t *journal,
311 struct recovery_info *info, enum passtype pass)
312{
313 unsigned int first_commit_ID, next_commit_ID;
Jan Kara9c28cbc2009-08-03 19:21:00 +0200314 unsigned int next_log_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 int err, success = 0;
316 journal_superblock_t * sb;
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700317 journal_header_t * tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 struct buffer_head * bh;
319 unsigned int sequence;
320 int blocktype;
321
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700322 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * First thing is to establish what we expect to find in the log
324 * (in terms of transaction IDs), and where (in terms of log
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700325 * block offsets): query the superblock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 */
327
328 sb = journal->j_superblock;
329 next_commit_ID = be32_to_cpu(sb->s_sequence);
330 next_log_block = be32_to_cpu(sb->s_start);
331
332 first_commit_ID = next_commit_ID;
333 if (pass == PASS_SCAN)
334 info->start_transaction = first_commit_ID;
335
336 jbd_debug(1, "Starting recovery pass %d\n", pass);
337
338 /*
339 * Now we walk through the log, transaction by transaction,
340 * making sure that each transaction has a commit block in the
341 * expected place. Each complete transaction gets replayed back
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700342 * into the main filesystem.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 */
344
345 while (1) {
346 int flags;
347 char * tagp;
348 journal_block_tag_t * tag;
349 struct buffer_head * obh;
350 struct buffer_head * nbh;
351
Andi Kleene86e1432008-02-06 01:40:11 -0800352 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 /* If we already know where to stop the log traversal,
355 * check right now that we haven't gone past the end of
356 * the log. */
357
358 if (pass != PASS_SCAN)
359 if (tid_geq(next_commit_ID, info->end_transaction))
360 break;
361
Jan Kara9c28cbc2009-08-03 19:21:00 +0200362 jbd_debug(2, "Scanning for sequence ID %u at %u/%u\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 next_commit_ID, next_log_block, journal->j_last);
364
365 /* Skip over each chunk of the transaction looking
366 * either the next descriptor block or the final commit
367 * record. */
368
Jan Kara9c28cbc2009-08-03 19:21:00 +0200369 jbd_debug(3, "JBD: checking block %u\n", next_log_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 err = jread(&bh, journal, next_log_block);
371 if (err)
372 goto failed;
373
374 next_log_block++;
375 wrap(journal, next_log_block);
376
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700377 /* What kind of buffer is it?
378 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 * If it is a descriptor block, check that it has the
380 * expected sequence number. Otherwise, we're all done
381 * here. */
382
383 tmp = (journal_header_t *)bh->b_data;
384
385 if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) {
386 brelse(bh);
387 break;
388 }
389
390 blocktype = be32_to_cpu(tmp->h_blocktype);
391 sequence = be32_to_cpu(tmp->h_sequence);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700392 jbd_debug(3, "Found magic %d, sequence %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 blocktype, sequence);
394
395 if (sequence != next_commit_ID) {
396 brelse(bh);
397 break;
398 }
399
400 /* OK, we have a valid descriptor block which matches
401 * all of the sequence number checks. What are we going
402 * to do with it? That depends on the pass... */
403
404 switch(blocktype) {
405 case JFS_DESCRIPTOR_BLOCK:
406 /* If it is a valid descriptor block, replay it
407 * in pass REPLAY; otherwise, just skip over the
408 * blocks it describes. */
409 if (pass != PASS_REPLAY) {
410 next_log_block +=
411 count_tags(bh, journal->j_blocksize);
412 wrap(journal, next_log_block);
413 brelse(bh);
414 continue;
415 }
416
417 /* A descriptor block: we can now write all of
418 * the data blocks. Yay, useful work is finally
419 * getting done here! */
420
421 tagp = &bh->b_data[sizeof(journal_header_t)];
422 while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
423 <= journal->j_blocksize) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200424 unsigned int io_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 tag = (journal_block_tag_t *) tagp;
427 flags = be32_to_cpu(tag->t_flags);
428
429 io_block = next_log_block++;
430 wrap(journal, next_log_block);
431 err = jread(&obh, journal, io_block);
432 if (err) {
433 /* Recover what we can, but
434 * report failure at the end. */
435 success = err;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700436 printk (KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 "JBD: IO error %d recovering "
Jan Kara9c28cbc2009-08-03 19:21:00 +0200438 "block %u in log\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 err, io_block);
440 } else {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200441 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
443 J_ASSERT(obh != NULL);
444 blocknr = be32_to_cpu(tag->t_blocknr);
445
446 /* If the block has been
447 * revoked, then we're all done
448 * here. */
449 if (journal_test_revoke
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700450 (journal, blocknr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 next_commit_ID)) {
452 brelse(obh);
453 ++info->nr_revoke_hits;
454 goto skip_write;
455 }
456
457 /* Find a buffer for the new
458 * data being restored */
459 nbh = __getblk(journal->j_fs_dev,
460 blocknr,
461 journal->j_blocksize);
462 if (nbh == NULL) {
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700463 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 "JBD: Out of memory "
465 "during recovery.\n");
466 err = -ENOMEM;
467 brelse(bh);
468 brelse(obh);
469 goto failed;
470 }
471
472 lock_buffer(nbh);
473 memcpy(nbh->b_data, obh->b_data,
474 journal->j_blocksize);
475 if (flags & JFS_FLAG_ESCAPE) {
Duane Griffin439aeec2008-03-19 17:00:53 -0700476 *((__be32 *)nbh->b_data) =
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 cpu_to_be32(JFS_MAGIC_NUMBER);
478 }
479
480 BUFFER_TRACE(nbh, "marking dirty");
481 set_buffer_uptodate(nbh);
482 mark_buffer_dirty(nbh);
483 BUFFER_TRACE(nbh, "marking uptodate");
484 ++info->nr_replays;
485 /* ll_rw_block(WRITE, 1, &nbh); */
486 unlock_buffer(nbh);
487 brelse(obh);
488 brelse(nbh);
489 }
490
491 skip_write:
492 tagp += sizeof(journal_block_tag_t);
493 if (!(flags & JFS_FLAG_SAME_UUID))
494 tagp += 16;
495
496 if (flags & JFS_FLAG_LAST_TAG)
497 break;
498 }
499
500 brelse(bh);
501 continue;
502
503 case JFS_COMMIT_BLOCK:
504 /* Found an expected commit block: not much to
505 * do other than move on to the next sequence
506 * number. */
507 brelse(bh);
508 next_commit_ID++;
509 continue;
510
511 case JFS_REVOKE_BLOCK:
512 /* If we aren't in the REVOKE pass, then we can
513 * just skip over this block. */
514 if (pass != PASS_REVOKE) {
515 brelse(bh);
516 continue;
517 }
518
519 err = scan_revoke_records(journal, bh,
520 next_commit_ID, info);
521 brelse(bh);
522 if (err)
523 goto failed;
524 continue;
525
526 default:
527 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
528 blocktype);
Theodore Ts'oe8f1c622006-06-25 05:47:51 -0700529 brelse(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 goto done;
531 }
532 }
533
534 done:
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700535 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 * We broke out of the log scan loop: either we came to the
537 * known end of the log or we found an unexpected block in the
538 * log. If the latter happened, then we know that the "current"
539 * transaction marks the end of the valid log.
540 */
541
542 if (pass == PASS_SCAN)
543 info->end_transaction = next_commit_ID;
544 else {
545 /* It's really bad news if different passes end up at
546 * different places (but possible due to IO errors). */
547 if (info->end_transaction != next_commit_ID) {
548 printk (KERN_ERR "JBD: recovery pass %d ended at "
549 "transaction %u, expected %u\n",
550 pass, next_commit_ID, info->end_transaction);
551 if (!success)
552 success = -EIO;
553 }
554 }
555
556 return success;
557
558 failed:
559 return err;
560}
561
562
563/* Scan a revoke record, marking all blocks mentioned as revoked. */
564
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700565static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 tid_t sequence, struct recovery_info *info)
567{
568 journal_revoke_header_t *header;
569 int offset, max;
570
571 header = (journal_revoke_header_t *) bh->b_data;
572 offset = sizeof(journal_revoke_header_t);
573 max = be32_to_cpu(header->r_count);
574
575 while (offset < max) {
Jan Kara9c28cbc2009-08-03 19:21:00 +0200576 unsigned int blocknr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 int err;
578
579 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
580 offset += 4;
581 err = journal_set_revoke(journal, blocknr, sequence);
582 if (err)
583 return err;
584 ++info->nr_revokes;
585 }
586 return 0;
587}