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/jfs.h>
22 #include <linux/errno.h>
23 #include <linux/malloc.h>
24 #include <linux/locks.h>
25 #include <linux/buffer.h>
29 * Maintain information about the progress of the recovery job, so that
30 * the different passes can carry information between them.
34 tid_t start_transaction;
35 tid_t end_transaction;
42 enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
43 static int do_one_pass(journal_t *, struct recovery_info *, enum passtype);
44 static int scan_revoke_records(journal_t *, struct buffer_head *, tid_t, struct recovery_info *);
48 /* Release readahead buffers after use */
49 static void brelse_array(struct buffer_head *b[], int n)
57 * When reading from the journal, we are going through the block device
58 * layer directly and so there is no readahead being done for us. We
59 * need to implement any readahead ourselves if we want it to happen at
60 * all. Recovery is basically one long sequential read, so make sure we
61 * do the IO in reasonably large chunks.
63 * This is not so critical that we need to be enormously clever about
64 * the readahead size, though. 128K is a purely arbitrary, good-enough
68 static int do_readahead(journal_t *journal, unsigned int start)
71 unsigned int max, nbufs, next, blocknr;
72 struct buffer_head *bh;
75 struct buffer_head * bufs[MAXBUF];
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;
82 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
83 * a time to the block device IO layer. */
87 for (next = start; next < max; next++) {
90 blocknr = bmap(journal->j_inode, next);
92 printk (KERN_ERR "JFS: 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 brelse_array(bufs, nbufs);
116 ll_rw_block(READ, nbufs, bufs);
121 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,
135 unsigned int blocknr;
136 struct buffer_head *bh;
140 J_ASSERT (offset < journal->j_maxlen);
143 if (journal->j_inode)
144 blocknr = bmap(journal->j_inode, offset);
147 printk (KERN_ERR "JFS: bad block at offset %u\n",
152 bh = getblk(journal->j_dev, blocknr, journal->j_blocksize);
156 if (!buffer_uptodate(bh)) {
157 /* If this is a brand new buffer, start readahead.
158 Otherwise, we assume we are already reading it. */
160 do_readahead(journal, offset);
164 if (!buffer_uptodate(bh)) {
165 printk (KERN_ERR "JFS: Failed to read block at offset %u\n",
177 * Count the number of in-use tags in a journal descriptor block.
180 int count_tags(struct buffer_head *bh, int size)
183 journal_block_tag_t * tag;
186 tagp = &bh->b_data[sizeof(journal_header_t)];
188 while ((tagp - bh->b_data + sizeof(journal_block_tag_t)) <= size) {
189 tag = (journal_block_tag_t *) tagp;
192 tagp += sizeof(journal_block_tag_t);
193 if (!(tag->t_flags & htonl(JFS_FLAG_SAME_UUID)))
196 if (tag->t_flags & htonl(JFS_FLAG_LAST_TAG))
204 /* Make sure we wrap around the log correctly! */
205 #define wrap(journal, var) \
207 if (var >= (journal)->j_last) \
208 var -= ((journal)->j_last - (journal)->j_first); \
214 * The primary function for recovering the log contents when mounting a
217 * Recovery is done in three passes. In the first pass, we look for the
218 * end of the log. In the second, we assemble the list of revoke
219 * blocks. In the third and final pass, we replay any un-revoked blocks
223 int journal_recover(journal_t *journal)
226 journal_superblock_t * sb;
228 struct recovery_info info = {};
230 sb = journal->j_superblock;
233 * The journal superblock's s_start field (the current log head)
234 * is always zero if, and only if, the journal was cleanly
239 jfs_debug(1, "No recovery required, last transaction %d\n",
240 ntohl(sb->s_sequence));
241 journal->j_transaction_sequence = ntohl(sb->s_sequence) + 1;
246 err = do_one_pass(journal, &info, PASS_SCAN);
248 err = do_one_pass(journal, &info, PASS_REVOKE);
250 err = do_one_pass(journal, &info, PASS_REPLAY);
252 jfs_debug(0, "JFS: recovery, exit status %d, "
253 "recovered transactions %u to %u\n",
254 err, info.start_transaction, info.end_transaction);
255 jfs_debug(0, "JFS: Replayed %d and revoked %d/%d blocks\n",
256 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
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;
262 journal_clear_revoke(journal);
263 fsync_dev(journal->j_dev);
267 static int do_one_pass(journal_t *journal, struct recovery_info *info,
271 unsigned int first_commit_ID, next_commit_ID;
272 unsigned long next_log_block;
273 int err, success = 0;
274 journal_superblock_t * sb;
275 journal_header_t * tmp;
276 struct buffer_head * bh;
277 unsigned int sequence;
280 /* Precompute the maximum metadata descriptors in a descriptor block */
281 int MAX_BLOCKS_PER_DESC;
282 MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t))
283 / sizeof(journal_block_tag_t));
286 * First thing is to establish what we expect to find in the log
287 * (in terms of transaction IDs), and where (in terms of log
288 * block offsets): query the superblock.
291 sb = journal->j_superblock;
292 next_commit_ID = ntohl(sb->s_sequence);
293 next_log_block = ntohl(sb->s_start);
295 first_commit_ID = next_commit_ID;
296 if (pass == PASS_SCAN)
297 info->start_transaction = first_commit_ID;
299 jfs_debug(1, "Starting recovery pass %d\n", pass);
302 * Now we walk through the log, transaction by transaction,
303 * making sure that each transaction has a commit block in the
304 * expected place. Each complete transaction gets replayed back
305 * into the main filesystem.
311 journal_block_tag_t * tag;
312 struct buffer_head * obh;
313 struct buffer_head * nbh;
315 /* If we already know where to stop the log traversal,
316 * check right now that we haven't gone past the end of
319 if (pass != PASS_SCAN)
320 if (tid_geq(next_commit_ID, info->end_transaction))
323 jfs_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
324 next_commit_ID, next_log_block, journal->j_last);
326 /* Skip over each chunk of the transaction looking
327 * either the next descriptor block or the final commit
330 jfs_debug(3, "JFS: checking block %ld\n", next_log_block);
331 err = jread(&bh, journal, next_log_block);
336 wrap(journal, next_log_block);
338 /* What kind of buffer is it?
340 * If it is a descriptor block, check that it has the
341 * expected sequence number. Otherwise, we're all done
344 tmp = (journal_header_t *) bh->b_data;
346 if (tmp->h_magic != htonl(JFS_MAGIC_NUMBER)) {
351 blocktype = ntohl(tmp->h_blocktype);
352 sequence = ntohl(tmp->h_sequence);
353 jfs_debug(3, "Found magic %d, sequence %d\n",
354 blocktype, sequence);
356 if (sequence != next_commit_ID) {
361 /* OK, we have a valid descriptor block which matches
362 * all of the sequence number checks. What are we going
363 * to do with it? That depends on the pass... */
366 case JFS_DESCRIPTOR_BLOCK:
367 /* If it is a valid descriptor block, replay it
368 * in pass REPLAY; otherwise, just skip over the
369 * blocks it describes. */
370 if (pass != PASS_REPLAY) {
371 next_log_block += count_tags(bh, journal->j_blocksize);
372 wrap(journal, next_log_block);
377 /* A descriptor block: we can now write all of
378 * the data blocks. Yay, useful work is finally
379 * getting done here! */
381 tagp = &bh->b_data[sizeof(journal_header_t)];
382 while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
383 <= journal->j_blocksize) {
384 unsigned long io_block;
386 tag = (journal_block_tag_t *) tagp;
387 flags = ntohl(tag->t_flags);
389 io_block = next_log_block++;
390 wrap(journal, next_log_block);
391 err = jread(&obh, journal, io_block);
393 /* Recover what we can, but
394 * report failure at the end. */
397 "JFS: IO error %d recovering "
398 "block %ld in log\n",
401 unsigned long blocknr;
403 J_ASSERT(obh != NULL);
404 blocknr = ntohl(tag->t_blocknr);
406 /* If the block has been
407 * revoked, then we're all done
409 if (journal_test_revoke
413 ++info->nr_revoke_hits;
417 /* Find a buffer for the new
418 * data being restored */
419 nbh = getblk(journal->j_dev, blocknr,
420 journal->j_blocksize);
423 "JFS: Out of memory "
424 "during recovery.\n");
431 memcpy(nbh->b_data, obh->b_data,
432 journal->j_blocksize);
433 if (flags & JFS_FLAG_ESCAPE) {
434 * ((unsigned int *) bh->b_data) = htonl(JFS_MAGIC_NUMBER);
437 mark_buffer_dirty(nbh, 1);
439 // ll_rw_block(WRITE, 1, &nbh);
445 tagp += sizeof(journal_block_tag_t);
446 if (!(flags & JFS_FLAG_SAME_UUID))
449 if (flags & JFS_FLAG_LAST_TAG)
456 case JFS_COMMIT_BLOCK:
457 /* Found an expected commit block: not much to
458 * do other than move on to the next sequence
464 case JFS_REVOKE_BLOCK:
465 /* If we aren't in the REVOKE pass, then we can
466 * just skip over this block. */
467 if (pass != PASS_REVOKE) {
472 err = scan_revoke_records(journal, bh,
473 next_commit_ID, info);
480 jfs_debug(3, "Unrecognised magic %d, end of scan.\n",
488 * We broke out of the log scan loop: either we came to the
489 * known end of the log or we found an unexpected block in the
490 * log. If the latter happened, then we know that the "current"
491 * transaction marks the end of the valid log.
494 if (pass == PASS_SCAN)
495 info->end_transaction = next_commit_ID;
497 /* It's really bad news if different passes end up at
498 * different places (but possible due to IO errors). */
499 if (info->end_transaction != next_commit_ID) {
500 printk (KERN_ERR "JFS: recovery pass %d ended at "
501 "transaction %u, expected %u\n",
502 pass, next_commit_ID, info->end_transaction);
515 /* Scan a revoke record, marking all blocks mentioned as revoked. */
517 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
518 tid_t sequence, struct recovery_info *info)
520 journal_revoke_header_t *header;
523 header = (journal_revoke_header_t *) bh->b_data;
524 offset = sizeof(journal_revoke_header_t);
525 max = ntohl(header->r_count);
527 while (offset < max) {
528 unsigned long blocknr;
531 blocknr = * ((unsigned int *) bh->b_data+offset);
533 err = journal_set_revoke(journal, blocknr, sequence);