4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
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.
12 * Journal recovery routines for the generic filesystem journaling code;
13 * part of the ext2fs journaling system.
19 #include <linux/sched.h>
21 #include <linux/jbd.h>
22 #include <linux/errno.h>
23 #include <linux/slab.h>
24 #include <linux/locks.h>
28 * Maintain information about the progress of the recovery job, so that
29 * the different passes can carry information between them.
33 tid_t start_transaction;
34 tid_t end_transaction;
41 enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
42 static int do_one_pass(journal_t *journal,
43 struct recovery_info *info, enum passtype pass);
44 static int scan_revoke_records(journal_t *, struct buffer_head *,
45 tid_t, struct recovery_info *);
49 /* Release readahead buffers after use */
50 void journal_brelse_array(struct buffer_head *b[], int n)
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.
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
70 static int do_readahead(journal_t *journal, unsigned int start)
73 unsigned int max, nbufs, next;
74 unsigned long blocknr;
75 struct buffer_head *bh;
77 struct buffer_head * bufs[MAXBUF];
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;
84 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
85 * a time to the block device IO layer. */
89 for (next = start; next < max; next++) {
90 err = journal_bmap(journal, next, &blocknr);
93 printk (KERN_ERR "JBD: bad block at offset %u\n",
98 bh = getblk(journal->j_dev, blocknr, journal->j_blocksize);
104 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
106 if (nbufs == MAXBUF) {
107 ll_rw_block(READ, nbufs, bufs);
108 journal_brelse_array(bufs, nbufs);
116 ll_rw_block(READ, nbufs, bufs);
121 journal_brelse_array(bufs, nbufs);
125 #endif /* __KERNEL__ */
129 * Read a block from the journal
132 static int jread(struct buffer_head **bhp, journal_t *journal,
136 unsigned long blocknr;
137 struct buffer_head *bh;
141 if (offset >= journal->j_maxlen) {
142 printk(KERN_ERR "JBD: corrupted journal superblock\n");
146 err = journal_bmap(journal, offset, &blocknr);
149 printk (KERN_ERR "JBD: bad block at offset %u\n",
154 bh = getblk(journal->j_dev, blocknr, journal->j_blocksize);
158 if (!buffer_uptodate(bh)) {
159 /* If this is a brand new buffer, start readahead.
160 Otherwise, we assume we are already reading it. */
162 do_readahead(journal, offset);
166 if (!buffer_uptodate(bh)) {
167 printk (KERN_ERR "JBD: Failed to read block at offset %u\n",
179 * Count the number of in-use tags in a journal descriptor block.
182 static int count_tags(struct buffer_head *bh, int size)
185 journal_block_tag_t * tag;
188 tagp = &bh->b_data[sizeof(journal_header_t)];
190 while ((tagp - bh->b_data + sizeof(journal_block_tag_t)) <= size) {
191 tag = (journal_block_tag_t *) tagp;
194 tagp += sizeof(journal_block_tag_t);
195 if (!(tag->t_flags & htonl(JFS_FLAG_SAME_UUID)))
198 if (tag->t_flags & htonl(JFS_FLAG_LAST_TAG))
206 /* Make sure we wrap around the log correctly! */
207 #define wrap(journal, var) \
209 if (var >= (journal)->j_last) \
210 var -= ((journal)->j_last - (journal)->j_first); \
214 * int journal_recover(journal_t *journal) - recovers a on-disk journal
215 * @journal: the journal to recover
217 * The primary function for recovering the log contents when mounting a
220 int journal_recover(journal_t *journal)
223 * Recovery is done in three passes. In the first pass, we look for the
224 * end of the log. In the second, we assemble the list of revoke
225 * blocks. In the third and final pass, we replay any un-revoked blocks
230 journal_superblock_t * sb;
232 struct recovery_info info;
234 memset(&info, 0, sizeof(info));
235 sb = journal->j_superblock;
238 * The journal superblock's s_start field (the current log head)
239 * is always zero if, and only if, the journal was cleanly
244 jbd_debug(1, "No recovery required, last transaction %d\n",
245 ntohl(sb->s_sequence));
246 journal->j_transaction_sequence = ntohl(sb->s_sequence) + 1;
251 err = do_one_pass(journal, &info, PASS_SCAN);
253 err = do_one_pass(journal, &info, PASS_REVOKE);
255 err = do_one_pass(journal, &info, PASS_REPLAY);
257 jbd_debug(0, "JBD: recovery, exit status %d, "
258 "recovered transactions %u to %u\n",
259 err, info.start_transaction, info.end_transaction);
260 jbd_debug(0, "JBD: Replayed %d and revoked %d/%d blocks\n",
261 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
263 /* Restart the log at the next transaction ID, thus invalidating
264 * any existing commit records in the log. */
265 journal->j_transaction_sequence = ++info.end_transaction;
267 journal_clear_revoke(journal);
268 fsync_no_super(journal->j_fs_dev);
273 * int journal_skip_recovery() - Start journal and wipe exiting records
274 * @journal: journal to startup
276 * Locate any valid recovery information from the journal and set up the
277 * journal structures in memory to ignore it (presumably because the
278 * caller has evidence that it is out of date).
279 * This function does'nt appear to be exorted..
281 int journal_skip_recovery(journal_t *journal)
284 * We perform one pass over the journal to allow us to tell the user how
285 * much recovery information is being erased, and to let us initialise
286 * the journal transaction sequence numbers to the next unused ID.
290 journal_superblock_t * sb;
292 struct recovery_info info;
294 memset (&info, 0, sizeof(info));
295 sb = journal->j_superblock;
297 err = do_one_pass(journal, &info, PASS_SCAN);
300 printk(KERN_ERR "JBD: error %d scanning journal\n", err);
301 ++journal->j_transaction_sequence;
303 #ifdef CONFIG_JBD_DEBUG
304 int dropped = info.end_transaction - ntohl(sb->s_sequence);
308 "JBD: ignoring %d transaction%s from the journal.\n",
309 dropped, (dropped == 1) ? "" : "s");
310 journal->j_transaction_sequence = ++info.end_transaction;
318 static int do_one_pass(journal_t *journal,
319 struct recovery_info *info, enum passtype pass)
322 unsigned int first_commit_ID, next_commit_ID;
323 unsigned long next_log_block;
324 int err, success = 0;
325 journal_superblock_t * sb;
326 journal_header_t * tmp;
327 struct buffer_head * bh;
328 unsigned int sequence;
331 /* Precompute the maximum metadata descriptors in a descriptor block */
332 int MAX_BLOCKS_PER_DESC;
333 MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t))
334 / sizeof(journal_block_tag_t));
337 * First thing is to establish what we expect to find in the log
338 * (in terms of transaction IDs), and where (in terms of log
339 * block offsets): query the superblock.
342 sb = journal->j_superblock;
343 next_commit_ID = ntohl(sb->s_sequence);
344 next_log_block = ntohl(sb->s_start);
346 first_commit_ID = next_commit_ID;
347 if (pass == PASS_SCAN)
348 info->start_transaction = first_commit_ID;
350 jbd_debug(1, "Starting recovery pass %d\n", pass);
353 * Now we walk through the log, transaction by transaction,
354 * making sure that each transaction has a commit block in the
355 * expected place. Each complete transaction gets replayed back
356 * into the main filesystem.
362 journal_block_tag_t * tag;
363 struct buffer_head * obh;
364 struct buffer_head * nbh;
366 /* If we already know where to stop the log traversal,
367 * check right now that we haven't gone past the end of
370 if (pass != PASS_SCAN)
371 if (tid_geq(next_commit_ID, info->end_transaction))
374 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
375 next_commit_ID, next_log_block, journal->j_last);
377 /* Skip over each chunk of the transaction looking
378 * either the next descriptor block or the final commit
381 jbd_debug(3, "JBD: checking block %ld\n", next_log_block);
382 err = jread(&bh, journal, next_log_block);
387 wrap(journal, next_log_block);
389 /* What kind of buffer is it?
391 * If it is a descriptor block, check that it has the
392 * expected sequence number. Otherwise, we're all done
395 tmp = (journal_header_t *)bh->b_data;
397 if (tmp->h_magic != htonl(JFS_MAGIC_NUMBER)) {
402 blocktype = ntohl(tmp->h_blocktype);
403 sequence = ntohl(tmp->h_sequence);
404 jbd_debug(3, "Found magic %d, sequence %d\n",
405 blocktype, sequence);
407 if (sequence != next_commit_ID) {
412 /* OK, we have a valid descriptor block which matches
413 * all of the sequence number checks. What are we going
414 * to do with it? That depends on the pass... */
417 case JFS_DESCRIPTOR_BLOCK:
418 /* If it is a valid descriptor block, replay it
419 * in pass REPLAY; otherwise, just skip over the
420 * blocks it describes. */
421 if (pass != PASS_REPLAY) {
423 count_tags(bh, journal->j_blocksize);
424 wrap(journal, next_log_block);
429 /* A descriptor block: we can now write all of
430 * the data blocks. Yay, useful work is finally
431 * getting done here! */
433 tagp = &bh->b_data[sizeof(journal_header_t)];
434 while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
435 <= journal->j_blocksize) {
436 unsigned long io_block;
438 tag = (journal_block_tag_t *) tagp;
439 flags = ntohl(tag->t_flags);
441 io_block = next_log_block++;
442 wrap(journal, next_log_block);
443 err = jread(&obh, journal, io_block);
445 /* Recover what we can, but
446 * report failure at the end. */
449 "JBD: IO error %d recovering "
450 "block %ld in log\n",
453 unsigned long blocknr;
455 J_ASSERT(obh != NULL);
456 blocknr = ntohl(tag->t_blocknr);
458 /* If the block has been
459 * revoked, then we're all done
461 if (journal_test_revoke
465 ++info->nr_revoke_hits;
469 /* Find a buffer for the new
470 * data being restored */
471 nbh = getblk(journal->j_fs_dev, blocknr,
472 journal->j_blocksize);
475 "JBD: Out of memory "
476 "during recovery.\n");
484 memcpy(nbh->b_data, obh->b_data,
485 journal->j_blocksize);
486 if (flags & JFS_FLAG_ESCAPE) {
487 *((unsigned int *)bh->b_data) =
488 htonl(JFS_MAGIC_NUMBER);
491 BUFFER_TRACE(nbh, "marking dirty");
492 mark_buffer_dirty(nbh);
493 BUFFER_TRACE(nbh, "marking uptodate");
494 mark_buffer_uptodate(nbh, 1);
497 /* ll_rw_block(WRITE, 1, &nbh); */
503 tagp += sizeof(journal_block_tag_t);
504 if (!(flags & JFS_FLAG_SAME_UUID))
507 if (flags & JFS_FLAG_LAST_TAG)
514 case JFS_COMMIT_BLOCK:
515 /* Found an expected commit block: not much to
516 * do other than move on to the next sequence
522 case JFS_REVOKE_BLOCK:
523 /* If we aren't in the REVOKE pass, then we can
524 * just skip over this block. */
525 if (pass != PASS_REVOKE) {
530 err = scan_revoke_records(journal, bh,
531 next_commit_ID, info);
538 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
547 * We broke out of the log scan loop: either we came to the
548 * known end of the log or we found an unexpected block in the
549 * log. If the latter happened, then we know that the "current"
550 * transaction marks the end of the valid log.
553 if (pass == PASS_SCAN)
554 info->end_transaction = next_commit_ID;
556 /* It's really bad news if different passes end up at
557 * different places (but possible due to IO errors). */
558 if (info->end_transaction != next_commit_ID) {
559 printk (KERN_ERR "JBD: recovery pass %d ended at "
560 "transaction %u, expected %u\n",
561 pass, next_commit_ID, info->end_transaction);
574 /* Scan a revoke record, marking all blocks mentioned as revoked. */
576 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
577 tid_t sequence, struct recovery_info *info)
579 journal_revoke_header_t *header;
582 header = (journal_revoke_header_t *) bh->b_data;
583 offset = sizeof(journal_revoke_header_t);
584 max = ntohl(header->r_count);
586 while (offset < max) {
587 unsigned long blocknr;
590 blocknr = ntohl(* ((unsigned int *) (bh->b_data+offset)));
592 err = journal_set_revoke(journal, blocknr, sequence);