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/blkdev.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
;
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 int 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
);
267 /* Flush disk caches to get replayed data on the permanent storage */
268 if (journal
->j_flags
& JFS_BARRIER
) {
269 err2
= blkdev_issue_flush(journal
->j_fs_dev
, GFP_KERNEL
, NULL
);
278 * journal_skip_recovery - Start journal and wipe exiting records
279 * @journal: journal to startup
281 * Locate any valid recovery information from the journal and set up the
282 * journal structures in memory to ignore it (presumably because the
283 * caller has evidence that it is out of date).
284 * This function does'nt appear to be exorted..
286 * We perform one pass over the journal to allow us to tell the user how
287 * much recovery information is being erased, and to let us initialise
288 * the journal transaction sequence numbers to the next unused ID.
290 int journal_skip_recovery(journal_t
*journal
)
293 struct recovery_info info
;
295 memset (&info
, 0, sizeof(info
));
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
-
305 be32_to_cpu(journal
->j_superblock
->s_sequence
);
307 "JBD: ignoring %d transaction%s from the journal.\n",
308 dropped
, (dropped
== 1) ? "" : "s");
310 journal
->j_transaction_sequence
= ++info
.end_transaction
;
317 static int do_one_pass(journal_t
*journal
,
318 struct recovery_info
*info
, enum passtype pass
)
320 unsigned int first_commit_ID
, next_commit_ID
;
321 unsigned int next_log_block
;
322 int err
, success
= 0;
323 journal_superblock_t
* sb
;
324 journal_header_t
* tmp
;
325 struct buffer_head
* bh
;
326 unsigned int sequence
;
330 * First thing is to establish what we expect to find in the log
331 * (in terms of transaction IDs), and where (in terms of log
332 * block offsets): query the superblock.
335 sb
= journal
->j_superblock
;
336 next_commit_ID
= be32_to_cpu(sb
->s_sequence
);
337 next_log_block
= be32_to_cpu(sb
->s_start
);
339 first_commit_ID
= next_commit_ID
;
340 if (pass
== PASS_SCAN
)
341 info
->start_transaction
= first_commit_ID
;
343 jbd_debug(1, "Starting recovery pass %d\n", pass
);
346 * Now we walk through the log, transaction by transaction,
347 * making sure that each transaction has a commit block in the
348 * expected place. Each complete transaction gets replayed back
349 * into the main filesystem.
355 journal_block_tag_t
* tag
;
356 struct buffer_head
* obh
;
357 struct buffer_head
* nbh
;
361 /* If we already know where to stop the log traversal,
362 * check right now that we haven't gone past the end of
365 if (pass
!= PASS_SCAN
)
366 if (tid_geq(next_commit_ID
, info
->end_transaction
))
369 jbd_debug(2, "Scanning for sequence ID %u at %u/%u\n",
370 next_commit_ID
, next_log_block
, journal
->j_last
);
372 /* Skip over each chunk of the transaction looking
373 * either the next descriptor block or the final commit
376 jbd_debug(3, "JBD: checking block %u\n", next_log_block
);
377 err
= jread(&bh
, journal
, next_log_block
);
382 wrap(journal
, next_log_block
);
384 /* What kind of buffer is it?
386 * If it is a descriptor block, check that it has the
387 * expected sequence number. Otherwise, we're all done
390 tmp
= (journal_header_t
*)bh
->b_data
;
392 if (tmp
->h_magic
!= cpu_to_be32(JFS_MAGIC_NUMBER
)) {
397 blocktype
= be32_to_cpu(tmp
->h_blocktype
);
398 sequence
= be32_to_cpu(tmp
->h_sequence
);
399 jbd_debug(3, "Found magic %d, sequence %d\n",
400 blocktype
, sequence
);
402 if (sequence
!= next_commit_ID
) {
407 /* OK, we have a valid descriptor block which matches
408 * all of the sequence number checks. What are we going
409 * to do with it? That depends on the pass... */
412 case JFS_DESCRIPTOR_BLOCK
:
413 /* If it is a valid descriptor block, replay it
414 * in pass REPLAY; otherwise, just skip over the
415 * blocks it describes. */
416 if (pass
!= PASS_REPLAY
) {
418 count_tags(bh
, journal
->j_blocksize
);
419 wrap(journal
, next_log_block
);
424 /* A descriptor block: we can now write all of
425 * the data blocks. Yay, useful work is finally
426 * getting done here! */
428 tagp
= &bh
->b_data
[sizeof(journal_header_t
)];
429 while ((tagp
- bh
->b_data
+sizeof(journal_block_tag_t
))
430 <= journal
->j_blocksize
) {
431 unsigned int io_block
;
433 tag
= (journal_block_tag_t
*) tagp
;
434 flags
= be32_to_cpu(tag
->t_flags
);
436 io_block
= next_log_block
++;
437 wrap(journal
, next_log_block
);
438 err
= jread(&obh
, journal
, io_block
);
440 /* Recover what we can, but
441 * report failure at the end. */
444 "JBD: IO error %d recovering "
448 unsigned int blocknr
;
450 J_ASSERT(obh
!= NULL
);
451 blocknr
= be32_to_cpu(tag
->t_blocknr
);
453 /* If the block has been
454 * revoked, then we're all done
456 if (journal_test_revoke
460 ++info
->nr_revoke_hits
;
464 /* Find a buffer for the new
465 * data being restored */
466 nbh
= __getblk(journal
->j_fs_dev
,
468 journal
->j_blocksize
);
471 "JBD: Out of memory "
472 "during recovery.\n");
480 memcpy(nbh
->b_data
, obh
->b_data
,
481 journal
->j_blocksize
);
482 if (flags
& JFS_FLAG_ESCAPE
) {
483 *((__be32
*)nbh
->b_data
) =
484 cpu_to_be32(JFS_MAGIC_NUMBER
);
487 BUFFER_TRACE(nbh
, "marking dirty");
488 set_buffer_uptodate(nbh
);
489 mark_buffer_dirty(nbh
);
490 BUFFER_TRACE(nbh
, "marking uptodate");
492 /* ll_rw_block(WRITE, 1, &nbh); */
499 tagp
+= sizeof(journal_block_tag_t
);
500 if (!(flags
& JFS_FLAG_SAME_UUID
))
503 if (flags
& JFS_FLAG_LAST_TAG
)
510 case JFS_COMMIT_BLOCK
:
511 /* Found an expected commit block: not much to
512 * do other than move on to the next sequence
518 case JFS_REVOKE_BLOCK
:
519 /* If we aren't in the REVOKE pass, then we can
520 * just skip over this block. */
521 if (pass
!= PASS_REVOKE
) {
526 err
= scan_revoke_records(journal
, bh
,
527 next_commit_ID
, info
);
534 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
543 * We broke out of the log scan loop: either we came to the
544 * known end of the log or we found an unexpected block in the
545 * log. If the latter happened, then we know that the "current"
546 * transaction marks the end of the valid log.
549 if (pass
== PASS_SCAN
)
550 info
->end_transaction
= next_commit_ID
;
552 /* It's really bad news if different passes end up at
553 * different places (but possible due to IO errors). */
554 if (info
->end_transaction
!= next_commit_ID
) {
555 printk (KERN_ERR
"JBD: recovery pass %d ended at "
556 "transaction %u, expected %u\n",
557 pass
, next_commit_ID
, info
->end_transaction
);
570 /* Scan a revoke record, marking all blocks mentioned as revoked. */
572 static int scan_revoke_records(journal_t
*journal
, struct buffer_head
*bh
,
573 tid_t sequence
, struct recovery_info
*info
)
575 journal_revoke_header_t
*header
;
578 header
= (journal_revoke_header_t
*) bh
->b_data
;
579 offset
= sizeof(journal_revoke_header_t
);
580 max
= be32_to_cpu(header
->r_count
);
582 while (offset
< max
) {
583 unsigned int blocknr
;
586 blocknr
= be32_to_cpu(* ((__be32
*) (bh
->b_data
+offset
)));
588 err
= journal_set_revoke(journal
, blocknr
, sequence
);
This page took 0.041504 seconds and 5 git commands to generate.