2 * linux/fs/jbd/recovery.c
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/time.h>
21 #include <linux/jbd.h>
22 #include <linux/errno.h>
23 #include <linux/slab.h>
27 * Maintain information about the progress of the recovery job, so that
28 * the different passes can carry information between them.
32 tid_t start_transaction
;
33 tid_t end_transaction
;
40 enum passtype
{PASS_SCAN
, PASS_REVOKE
, PASS_REPLAY
};
41 static int do_one_pass(journal_t
*journal
,
42 struct recovery_info
*info
, enum passtype pass
);
43 static int scan_revoke_records(journal_t
*, struct buffer_head
*,
44 tid_t
, struct recovery_info
*);
48 /* Release readahead buffers after use */
49 static void journal_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
69 static int do_readahead(journal_t
*journal
, unsigned int start
)
72 unsigned int max
, nbufs
, next
;
73 unsigned long blocknr
;
74 struct buffer_head
*bh
;
76 struct buffer_head
* bufs
[MAXBUF
];
78 /* Do up to 128K of readahead */
79 max
= start
+ (128 * 1024 / journal
->j_blocksize
);
80 if (max
> journal
->j_maxlen
)
81 max
= journal
->j_maxlen
;
83 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
84 * a time to the block device IO layer. */
88 for (next
= start
; next
< max
; next
++) {
89 err
= journal_bmap(journal
, next
, &blocknr
);
92 printk (KERN_ERR
"JBD: bad block at offset %u\n",
97 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
103 if (!buffer_uptodate(bh
) && !buffer_locked(bh
)) {
105 if (nbufs
== MAXBUF
) {
106 ll_rw_block(READ
, nbufs
, bufs
);
107 journal_brelse_array(bufs
, nbufs
);
115 ll_rw_block(READ
, nbufs
, bufs
);
120 journal_brelse_array(bufs
, nbufs
);
124 #endif /* __KERNEL__ */
128 * Read a block from the journal
131 static int jread(struct buffer_head
**bhp
, journal_t
*journal
,
135 unsigned long blocknr
;
136 struct buffer_head
*bh
;
140 if (offset
>= journal
->j_maxlen
) {
141 printk(KERN_ERR
"JBD: corrupted journal superblock\n");
145 err
= journal_bmap(journal
, offset
, &blocknr
);
148 printk (KERN_ERR
"JBD: bad block at offset %u\n",
153 bh
= __getblk(journal
->j_dev
, blocknr
, journal
->j_blocksize
);
157 if (!buffer_uptodate(bh
)) {
158 /* If this is a brand new buffer, start readahead.
159 Otherwise, we assume we are already reading it. */
161 do_readahead(journal
, offset
);
165 if (!buffer_uptodate(bh
)) {
166 printk (KERN_ERR
"JBD: Failed to read block at offset %u\n",
178 * Count the number of in-use tags in a journal descriptor block.
181 static int count_tags(struct buffer_head
*bh
, int size
)
184 journal_block_tag_t
* tag
;
187 tagp
= &bh
->b_data
[sizeof(journal_header_t
)];
189 while ((tagp
- bh
->b_data
+ sizeof(journal_block_tag_t
)) <= size
) {
190 tag
= (journal_block_tag_t
*) tagp
;
193 tagp
+= sizeof(journal_block_tag_t
);
194 if (!(tag
->t_flags
& cpu_to_be32(JFS_FLAG_SAME_UUID
)))
197 if (tag
->t_flags
& cpu_to_be32(JFS_FLAG_LAST_TAG
))
205 /* Make sure we wrap around the log correctly! */
206 #define wrap(journal, var) \
208 if (var >= (journal)->j_last) \
209 var -= ((journal)->j_last - (journal)->j_first); \
213 * journal_recover - recovers a on-disk journal
214 * @journal: the journal to recover
216 * The primary function for recovering the log contents when mounting a
219 * Recovery is done in three passes. In the first pass, we look for the
220 * end of the log. In the second, we assemble the list of revoke
221 * blocks. In the third and final pass, we replay any un-revoked blocks
224 int journal_recover(journal_t
*journal
)
227 journal_superblock_t
* sb
;
229 struct recovery_info info
;
231 memset(&info
, 0, sizeof(info
));
232 sb
= journal
->j_superblock
;
235 * The journal superblock's s_start field (the current log head)
236 * is always zero if, and only if, the journal was cleanly
241 jbd_debug(1, "No recovery required, last transaction %d\n",
242 be32_to_cpu(sb
->s_sequence
));
243 journal
->j_transaction_sequence
= be32_to_cpu(sb
->s_sequence
) + 1;
247 err
= do_one_pass(journal
, &info
, PASS_SCAN
);
249 err
= do_one_pass(journal
, &info
, PASS_REVOKE
);
251 err
= do_one_pass(journal
, &info
, PASS_REPLAY
);
253 jbd_debug(1, "JBD: recovery, exit status %d, "
254 "recovered transactions %u to %u\n",
255 err
, info
.start_transaction
, info
.end_transaction
);
256 jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n",
257 info
.nr_replays
, info
.nr_revoke_hits
, info
.nr_revokes
);
259 /* Restart the log at the next transaction ID, thus invalidating
260 * any existing commit records in the log. */
261 journal
->j_transaction_sequence
= ++info
.end_transaction
;
263 journal_clear_revoke(journal
);
264 err2
= sync_blockdev(journal
->j_fs_dev
);
272 * journal_skip_recovery - Start journal and wipe exiting records
273 * @journal: journal to startup
275 * Locate any valid recovery information from the journal and set up the
276 * journal structures in memory to ignore it (presumably because the
277 * caller has evidence that it is out of date).
278 * This function does'nt appear to be exorted..
280 * We perform one pass over the journal to allow us to tell the user how
281 * much recovery information is being erased, and to let us initialise
282 * the journal transaction sequence numbers to the next unused ID.
284 int journal_skip_recovery(journal_t
*journal
)
287 journal_superblock_t
* sb
;
289 struct recovery_info info
;
291 memset (&info
, 0, sizeof(info
));
292 sb
= journal
->j_superblock
;
294 err
= do_one_pass(journal
, &info
, PASS_SCAN
);
297 printk(KERN_ERR
"JBD: error %d scanning journal\n", err
);
298 ++journal
->j_transaction_sequence
;
300 #ifdef CONFIG_JBD_DEBUG
301 int dropped
= info
.end_transaction
- be32_to_cpu(sb
->s_sequence
);
304 "JBD: ignoring %d transaction%s from the journal.\n",
305 dropped
, (dropped
== 1) ? "" : "s");
306 journal
->j_transaction_sequence
= ++info
.end_transaction
;
313 static int do_one_pass(journal_t
*journal
,
314 struct recovery_info
*info
, enum passtype pass
)
316 unsigned int first_commit_ID
, next_commit_ID
;
317 unsigned long next_log_block
;
318 int err
, success
= 0;
319 journal_superblock_t
* sb
;
320 journal_header_t
* tmp
;
321 struct buffer_head
* bh
;
322 unsigned int sequence
;
325 /* Precompute the maximum metadata descriptors in a descriptor block */
326 int MAX_BLOCKS_PER_DESC
;
327 MAX_BLOCKS_PER_DESC
= ((journal
->j_blocksize
-sizeof(journal_header_t
))
328 / sizeof(journal_block_tag_t
));
331 * First thing is to establish what we expect to find in the log
332 * (in terms of transaction IDs), and where (in terms of log
333 * block offsets): query the superblock.
336 sb
= journal
->j_superblock
;
337 next_commit_ID
= be32_to_cpu(sb
->s_sequence
);
338 next_log_block
= be32_to_cpu(sb
->s_start
);
340 first_commit_ID
= next_commit_ID
;
341 if (pass
== PASS_SCAN
)
342 info
->start_transaction
= first_commit_ID
;
344 jbd_debug(1, "Starting recovery pass %d\n", pass
);
347 * Now we walk through the log, transaction by transaction,
348 * making sure that each transaction has a commit block in the
349 * expected place. Each complete transaction gets replayed back
350 * into the main filesystem.
356 journal_block_tag_t
* tag
;
357 struct buffer_head
* obh
;
358 struct buffer_head
* nbh
;
362 /* If we already know where to stop the log traversal,
363 * check right now that we haven't gone past the end of
366 if (pass
!= PASS_SCAN
)
367 if (tid_geq(next_commit_ID
, info
->end_transaction
))
370 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
371 next_commit_ID
, next_log_block
, journal
->j_last
);
373 /* Skip over each chunk of the transaction looking
374 * either the next descriptor block or the final commit
377 jbd_debug(3, "JBD: checking block %ld\n", next_log_block
);
378 err
= jread(&bh
, journal
, next_log_block
);
383 wrap(journal
, next_log_block
);
385 /* What kind of buffer is it?
387 * If it is a descriptor block, check that it has the
388 * expected sequence number. Otherwise, we're all done
391 tmp
= (journal_header_t
*)bh
->b_data
;
393 if (tmp
->h_magic
!= cpu_to_be32(JFS_MAGIC_NUMBER
)) {
398 blocktype
= be32_to_cpu(tmp
->h_blocktype
);
399 sequence
= be32_to_cpu(tmp
->h_sequence
);
400 jbd_debug(3, "Found magic %d, sequence %d\n",
401 blocktype
, sequence
);
403 if (sequence
!= next_commit_ID
) {
408 /* OK, we have a valid descriptor block which matches
409 * all of the sequence number checks. What are we going
410 * to do with it? That depends on the pass... */
413 case JFS_DESCRIPTOR_BLOCK
:
414 /* If it is a valid descriptor block, replay it
415 * in pass REPLAY; otherwise, just skip over the
416 * blocks it describes. */
417 if (pass
!= PASS_REPLAY
) {
419 count_tags(bh
, journal
->j_blocksize
);
420 wrap(journal
, next_log_block
);
425 /* A descriptor block: we can now write all of
426 * the data blocks. Yay, useful work is finally
427 * getting done here! */
429 tagp
= &bh
->b_data
[sizeof(journal_header_t
)];
430 while ((tagp
- bh
->b_data
+sizeof(journal_block_tag_t
))
431 <= journal
->j_blocksize
) {
432 unsigned long io_block
;
434 tag
= (journal_block_tag_t
*) tagp
;
435 flags
= be32_to_cpu(tag
->t_flags
);
437 io_block
= next_log_block
++;
438 wrap(journal
, next_log_block
);
439 err
= jread(&obh
, journal
, io_block
);
441 /* Recover what we can, but
442 * report failure at the end. */
445 "JBD: IO error %d recovering "
446 "block %ld in log\n",
449 unsigned long blocknr
;
451 J_ASSERT(obh
!= NULL
);
452 blocknr
= be32_to_cpu(tag
->t_blocknr
);
454 /* If the block has been
455 * revoked, then we're all done
457 if (journal_test_revoke
461 ++info
->nr_revoke_hits
;
465 /* Find a buffer for the new
466 * data being restored */
467 nbh
= __getblk(journal
->j_fs_dev
,
469 journal
->j_blocksize
);
472 "JBD: Out of memory "
473 "during recovery.\n");
481 memcpy(nbh
->b_data
, obh
->b_data
,
482 journal
->j_blocksize
);
483 if (flags
& JFS_FLAG_ESCAPE
) {
484 *((__be32
*)nbh
->b_data
) =
485 cpu_to_be32(JFS_MAGIC_NUMBER
);
488 BUFFER_TRACE(nbh
, "marking dirty");
489 set_buffer_uptodate(nbh
);
490 mark_buffer_dirty(nbh
);
491 BUFFER_TRACE(nbh
, "marking uptodate");
493 /* ll_rw_block(WRITE, 1, &nbh); */
500 tagp
+= sizeof(journal_block_tag_t
);
501 if (!(flags
& JFS_FLAG_SAME_UUID
))
504 if (flags
& JFS_FLAG_LAST_TAG
)
511 case JFS_COMMIT_BLOCK
:
512 /* Found an expected commit block: not much to
513 * do other than move on to the next sequence
519 case JFS_REVOKE_BLOCK
:
520 /* If we aren't in the REVOKE pass, then we can
521 * just skip over this block. */
522 if (pass
!= PASS_REVOKE
) {
527 err
= scan_revoke_records(journal
, bh
,
528 next_commit_ID
, info
);
535 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
544 * We broke out of the log scan loop: either we came to the
545 * known end of the log or we found an unexpected block in the
546 * log. If the latter happened, then we know that the "current"
547 * transaction marks the end of the valid log.
550 if (pass
== PASS_SCAN
)
551 info
->end_transaction
= next_commit_ID
;
553 /* It's really bad news if different passes end up at
554 * different places (but possible due to IO errors). */
555 if (info
->end_transaction
!= next_commit_ID
) {
556 printk (KERN_ERR
"JBD: recovery pass %d ended at "
557 "transaction %u, expected %u\n",
558 pass
, next_commit_ID
, info
->end_transaction
);
571 /* Scan a revoke record, marking all blocks mentioned as revoked. */
573 static int scan_revoke_records(journal_t
*journal
, struct buffer_head
*bh
,
574 tid_t sequence
, struct recovery_info
*info
)
576 journal_revoke_header_t
*header
;
579 header
= (journal_revoke_header_t
*) bh
->b_data
;
580 offset
= sizeof(journal_revoke_header_t
);
581 max
= be32_to_cpu(header
->r_count
);
583 while (offset
< max
) {
584 unsigned long blocknr
;
587 blocknr
= be32_to_cpu(* ((__be32
*) (bh
->b_data
+offset
)));
589 err
= journal_set_revoke(journal
, blocknr
, sequence
);
This page took 0.042779 seconds and 5 git commands to generate.