blob: 626846bac32f8a5f2c01d4e3f39db9332642a25c [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd2/recovery.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
4 * 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;
13 * part of the ext2fs journaling system.
14 */
15
16#ifndef __KERNEL__
17#include "jfs_user.h"
18#else
19#include <linux/time.h>
20#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070021#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070022#include <linux/errno.h>
Girish Shilamkar818d2762008-01-28 23:58:27 -050023#include <linux/crc32.h>
Jan Kara79feb522012-03-13 22:22:54 -040024#include <linux/blkdev.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070025#endif
26
27/*
28 * Maintain information about the progress of the recovery job, so that
29 * the different passes can carry information between them.
30 */
31struct recovery_info
32{
33 tid_t start_transaction;
34 tid_t end_transaction;
35
36 int nr_replays;
37 int nr_revokes;
38 int nr_revoke_hits;
39};
40
41enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
42static int do_one_pass(journal_t *journal,
43 struct recovery_info *info, enum passtype pass);
44static int scan_revoke_records(journal_t *, struct buffer_head *,
45 tid_t, struct recovery_info *);
46
47#ifdef __KERNEL__
48
49/* Release readahead buffers after use */
50static void journal_brelse_array(struct buffer_head *b[], int n)
51{
52 while (--n >= 0)
53 brelse (b[n]);
54}
55
56
57/*
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.
63 *
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
66 * fixed value.
67 */
68
69#define MAXBUF 8
70static int do_readahead(journal_t *journal, unsigned int start)
71{
72 int err;
73 unsigned int max, nbufs, next;
Mingming Cao18eba7a2006-10-11 01:21:13 -070074 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -070075 struct buffer_head *bh;
76
77 struct buffer_head * bufs[MAXBUF];
78
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;
83
84 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
85 * a time to the block device IO layer. */
86
87 nbufs = 0;
88
89 for (next = start; next < max; next++) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -070090 err = jbd2_journal_bmap(journal, next, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -070091
92 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -040093 printk(KERN_ERR "JBD2: bad block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -070094 next);
95 goto failed;
96 }
97
98 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
99 if (!bh) {
100 err = -ENOMEM;
101 goto failed;
102 }
103
104 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
105 bufs[nbufs++] = bh;
106 if (nbufs == MAXBUF) {
107 ll_rw_block(READ, nbufs, bufs);
108 journal_brelse_array(bufs, nbufs);
109 nbufs = 0;
110 }
111 } else
112 brelse(bh);
113 }
114
115 if (nbufs)
116 ll_rw_block(READ, nbufs, bufs);
117 err = 0;
118
119failed:
120 if (nbufs)
121 journal_brelse_array(bufs, nbufs);
122 return err;
123}
124
125#endif /* __KERNEL__ */
126
127
128/*
129 * Read a block from the journal
130 */
131
132static int jread(struct buffer_head **bhp, journal_t *journal,
133 unsigned int offset)
134{
135 int err;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700136 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700137 struct buffer_head *bh;
138
139 *bhp = NULL;
140
141 if (offset >= journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400142 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700143 return -EIO;
144 }
145
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700146 err = jbd2_journal_bmap(journal, offset, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700147
148 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400149 printk(KERN_ERR "JBD2: bad block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700150 offset);
151 return err;
152 }
153
154 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
155 if (!bh)
156 return -ENOMEM;
157
158 if (!buffer_uptodate(bh)) {
159 /* If this is a brand new buffer, start readahead.
160 Otherwise, we assume we are already reading it. */
161 if (!buffer_req(bh))
162 do_readahead(journal, offset);
163 wait_on_buffer(bh);
164 }
165
166 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400167 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700168 offset);
169 brelse(bh);
170 return -EIO;
171 }
172
173 *bhp = bh;
174 return 0;
175}
176
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400177static int jbd2_descr_block_csum_verify(journal_t *j,
178 void *buf)
179{
180 struct jbd2_journal_block_tail *tail;
181 __u32 provided, calculated;
182
183 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
184 return 1;
185
186 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
187 sizeof(struct jbd2_journal_block_tail));
188 provided = tail->t_checksum;
189 tail->t_checksum = 0;
190 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
191 tail->t_checksum = provided;
192
193 provided = be32_to_cpu(provided);
194 return provided == calculated;
195}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700196
197/*
198 * Count the number of in-use tags in a journal descriptor block.
199 */
200
Zach Brownb517bea2006-10-11 01:21:08 -0700201static int count_tags(journal_t *journal, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700202{
203 char * tagp;
204 journal_block_tag_t * tag;
Zach Brownb517bea2006-10-11 01:21:08 -0700205 int nr = 0, size = journal->j_blocksize;
206 int tag_bytes = journal_tag_bytes(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700207
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400208 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
209 size -= sizeof(struct jbd2_journal_block_tail);
210
Dave Kleikamp470decc2006-10-11 01:20:57 -0700211 tagp = &bh->b_data[sizeof(journal_header_t)];
212
Zach Brownb517bea2006-10-11 01:21:08 -0700213 while ((tagp - bh->b_data + tag_bytes) <= size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700214 tag = (journal_block_tag_t *) tagp;
215
216 nr++;
Zach Brownb517bea2006-10-11 01:21:08 -0700217 tagp += tag_bytes;
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400218 if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700219 tagp += 16;
220
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400221 if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700222 break;
223 }
224
225 return nr;
226}
227
228
229/* Make sure we wrap around the log correctly! */
230#define wrap(journal, var) \
231do { \
232 if (var >= (journal)->j_last) \
233 var -= ((journal)->j_last - (journal)->j_first); \
234} while (0)
235
236/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700237 * jbd2_journal_recover - recovers a on-disk journal
Dave Kleikamp470decc2006-10-11 01:20:57 -0700238 * @journal: the journal to recover
239 *
240 * The primary function for recovering the log contents when mounting a
241 * journaled device.
242 *
243 * Recovery is done in three passes. In the first pass, we look for the
244 * end of the log. In the second, we assemble the list of revoke
245 * blocks. In the third and final pass, we replay any un-revoked blocks
246 * in the log.
247 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700248int jbd2_journal_recover(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700249{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400250 int err, err2;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700251 journal_superblock_t * sb;
252
253 struct recovery_info info;
254
255 memset(&info, 0, sizeof(info));
256 sb = journal->j_superblock;
257
258 /*
259 * The journal superblock's s_start field (the current log head)
260 * is always zero if, and only if, the journal was cleanly
261 * unmounted.
262 */
263
264 if (!sb->s_start) {
265 jbd_debug(1, "No recovery required, last transaction %d\n",
266 be32_to_cpu(sb->s_sequence));
267 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
268 return 0;
269 }
270
271 err = do_one_pass(journal, &info, PASS_SCAN);
272 if (!err)
273 err = do_one_pass(journal, &info, PASS_REVOKE);
274 if (!err)
275 err = do_one_pass(journal, &info, PASS_REPLAY);
276
Eryu Guanf2a44522011-11-01 19:09:18 -0400277 jbd_debug(1, "JBD2: recovery, exit status %d, "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700278 "recovered transactions %u to %u\n",
279 err, info.start_transaction, info.end_transaction);
Eryu Guanf2a44522011-11-01 19:09:18 -0400280 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700281 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
282
283 /* Restart the log at the next transaction ID, thus invalidating
284 * any existing commit records in the log. */
285 journal->j_transaction_sequence = ++info.end_transaction;
286
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700287 jbd2_journal_clear_revoke(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400288 err2 = sync_blockdev(journal->j_fs_dev);
289 if (!err)
290 err = err2;
Jan Kara79feb522012-03-13 22:22:54 -0400291 /* Make sure all replayed data is on permanent storage */
Theodore Ts'o316e4cf2012-08-17 09:56:17 -0400292 if (journal->j_flags & JBD2_BARRIER) {
293 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
294 if (!err)
295 err = err2;
296 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700297 return err;
298}
299
300/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700301 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
Dave Kleikamp470decc2006-10-11 01:20:57 -0700302 * @journal: journal to startup
303 *
304 * Locate any valid recovery information from the journal and set up the
305 * journal structures in memory to ignore it (presumably because the
306 * caller has evidence that it is out of date).
307 * This function does'nt appear to be exorted..
308 *
309 * We perform one pass over the journal to allow us to tell the user how
310 * much recovery information is being erased, and to let us initialise
311 * the journal transaction sequence numbers to the next unused ID.
312 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700313int jbd2_journal_skip_recovery(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700314{
315 int err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700316
317 struct recovery_info info;
318
319 memset (&info, 0, sizeof(info));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700320
321 err = do_one_pass(journal, &info, PASS_SCAN);
322
323 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400324 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700325 ++journal->j_transaction_sequence;
326 } else {
Jose R. Santose23291b2007-07-18 08:57:06 -0400327#ifdef CONFIG_JBD2_DEBUG
Andi Kleen5a0790c2010-06-14 13:28:03 -0400328 int dropped = info.end_transaction -
329 be32_to_cpu(journal->j_superblock->s_sequence);
Mingming Caob38bd332007-07-19 01:48:35 -0700330 jbd_debug(1,
Eryu Guanf2a44522011-11-01 19:09:18 -0400331 "JBD2: ignoring %d transaction%s from the journal.\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700332 dropped, (dropped == 1) ? "" : "s");
Theodore Ts'o9a4f6272010-12-18 13:36:33 -0500333#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700334 journal->j_transaction_sequence = ++info.end_transaction;
335 }
336
337 journal->j_tail = 0;
338 return err;
339}
340
Mingming Cao18eba7a2006-10-11 01:21:13 -0700341static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
Zach Brownb517bea2006-10-11 01:21:08 -0700342{
Mingming Cao18eba7a2006-10-11 01:21:13 -0700343 unsigned long long block = be32_to_cpu(tag->t_blocknr);
Mingming Caocd02ff02007-10-16 18:38:25 -0400344 if (tag_bytes > JBD2_TAG_SIZE32)
Zach Brownb517bea2006-10-11 01:21:08 -0700345 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
346 return block;
347}
348
Girish Shilamkar818d2762008-01-28 23:58:27 -0500349/*
350 * calc_chksums calculates the checksums for the blocks described in the
351 * descriptor block.
352 */
353static int calc_chksums(journal_t *journal, struct buffer_head *bh,
354 unsigned long *next_log_block, __u32 *crc32_sum)
355{
356 int i, num_blks, err;
357 unsigned long io_block;
358 struct buffer_head *obh;
359
360 num_blks = count_tags(journal, bh);
361 /* Calculate checksum of the descriptor block. */
362 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
363
364 for (i = 0; i < num_blks; i++) {
365 io_block = (*next_log_block)++;
366 wrap(journal, *next_log_block);
367 err = jread(&obh, journal, io_block);
368 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400369 printk(KERN_ERR "JBD2: IO error %d recovering block "
Girish Shilamkar818d2762008-01-28 23:58:27 -0500370 "%lu in log\n", err, io_block);
371 return 1;
372 } else {
373 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
374 obh->b_size);
375 }
Theodore Ts'o8ea76902008-05-26 10:28:09 -0400376 put_bh(obh);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500377 }
378 return 0;
379}
380
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400381static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
382{
383 struct commit_header *h;
384 __u32 provided, calculated;
385
386 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
387 return 1;
388
389 h = buf;
390 provided = h->h_chksum[0];
391 h->h_chksum[0] = 0;
392 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
393 h->h_chksum[0] = provided;
394
395 provided = be32_to_cpu(provided);
396 return provided == calculated;
397}
398
Darrick J. Wongc3900872012-05-27 08:12:12 -0400399static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
400 void *buf, __u32 sequence)
401{
402 __u32 provided, calculated;
403
404 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
405 return 1;
406
407 sequence = cpu_to_be32(sequence);
408 calculated = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence,
409 sizeof(sequence));
410 calculated = jbd2_chksum(j, calculated, buf, j->j_blocksize);
411 provided = be32_to_cpu(tag->t_checksum);
412
413 return provided == cpu_to_be32(calculated);
414}
415
Dave Kleikamp470decc2006-10-11 01:20:57 -0700416static int do_one_pass(journal_t *journal,
417 struct recovery_info *info, enum passtype pass)
418{
419 unsigned int first_commit_ID, next_commit_ID;
420 unsigned long next_log_block;
421 int err, success = 0;
422 journal_superblock_t * sb;
423 journal_header_t * tmp;
424 struct buffer_head * bh;
425 unsigned int sequence;
426 int blocktype;
Zach Brownb517bea2006-10-11 01:21:08 -0700427 int tag_bytes = journal_tag_bytes(journal);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500428 __u32 crc32_sum = ~0; /* Transactional Checksums */
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400429 int descr_csum_size = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700430
Dave Kleikamp470decc2006-10-11 01:20:57 -0700431 /*
432 * First thing is to establish what we expect to find in the log
433 * (in terms of transaction IDs), and where (in terms of log
434 * block offsets): query the superblock.
435 */
436
437 sb = journal->j_superblock;
438 next_commit_ID = be32_to_cpu(sb->s_sequence);
439 next_log_block = be32_to_cpu(sb->s_start);
440
441 first_commit_ID = next_commit_ID;
442 if (pass == PASS_SCAN)
443 info->start_transaction = first_commit_ID;
444
445 jbd_debug(1, "Starting recovery pass %d\n", pass);
446
447 /*
448 * Now we walk through the log, transaction by transaction,
449 * making sure that each transaction has a commit block in the
450 * expected place. Each complete transaction gets replayed back
451 * into the main filesystem.
452 */
453
454 while (1) {
455 int flags;
456 char * tagp;
457 journal_block_tag_t * tag;
458 struct buffer_head * obh;
459 struct buffer_head * nbh;
460
Andi Kleene86e1432008-02-06 01:40:11 -0800461 cond_resched();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700462
463 /* If we already know where to stop the log traversal,
464 * check right now that we haven't gone past the end of
465 * the log. */
466
467 if (pass != PASS_SCAN)
468 if (tid_geq(next_commit_ID, info->end_transaction))
469 break;
470
471 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
472 next_commit_ID, next_log_block, journal->j_last);
473
474 /* Skip over each chunk of the transaction looking
475 * either the next descriptor block or the final commit
476 * record. */
477
Eryu Guanf2a44522011-11-01 19:09:18 -0400478 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700479 err = jread(&bh, journal, next_log_block);
480 if (err)
481 goto failed;
482
483 next_log_block++;
484 wrap(journal, next_log_block);
485
486 /* What kind of buffer is it?
487 *
488 * If it is a descriptor block, check that it has the
489 * expected sequence number. Otherwise, we're all done
490 * here. */
491
492 tmp = (journal_header_t *)bh->b_data;
493
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700494 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700495 brelse(bh);
496 break;
497 }
498
499 blocktype = be32_to_cpu(tmp->h_blocktype);
500 sequence = be32_to_cpu(tmp->h_sequence);
501 jbd_debug(3, "Found magic %d, sequence %d\n",
502 blocktype, sequence);
503
504 if (sequence != next_commit_ID) {
505 brelse(bh);
506 break;
507 }
508
509 /* OK, we have a valid descriptor block which matches
510 * all of the sequence number checks. What are we going
511 * to do with it? That depends on the pass... */
512
513 switch(blocktype) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700514 case JBD2_DESCRIPTOR_BLOCK:
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400515 /* Verify checksum first */
516 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
517 JBD2_FEATURE_INCOMPAT_CSUM_V2))
518 descr_csum_size =
519 sizeof(struct jbd2_journal_block_tail);
520 if (descr_csum_size > 0 &&
521 !jbd2_descr_block_csum_verify(journal,
522 bh->b_data)) {
523 err = -EIO;
524 goto failed;
525 }
526
Dave Kleikamp470decc2006-10-11 01:20:57 -0700527 /* If it is a valid descriptor block, replay it
Girish Shilamkar818d2762008-01-28 23:58:27 -0500528 * in pass REPLAY; if journal_checksums enabled, then
529 * calculate checksums in PASS_SCAN, otherwise,
530 * just skip over the blocks it describes. */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531 if (pass != PASS_REPLAY) {
Girish Shilamkar818d2762008-01-28 23:58:27 -0500532 if (pass == PASS_SCAN &&
533 JBD2_HAS_COMPAT_FEATURE(journal,
534 JBD2_FEATURE_COMPAT_CHECKSUM) &&
535 !info->end_transaction) {
536 if (calc_chksums(journal, bh,
537 &next_log_block,
538 &crc32_sum)) {
539 put_bh(bh);
540 break;
541 }
542 put_bh(bh);
543 continue;
544 }
Zach Brownb517bea2006-10-11 01:21:08 -0700545 next_log_block += count_tags(journal, bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546 wrap(journal, next_log_block);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500547 put_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548 continue;
549 }
550
551 /* A descriptor block: we can now write all of
552 * the data blocks. Yay, useful work is finally
553 * getting done here! */
554
555 tagp = &bh->b_data[sizeof(journal_header_t)];
Zach Brownb517bea2006-10-11 01:21:08 -0700556 while ((tagp - bh->b_data + tag_bytes)
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400557 <= journal->j_blocksize - descr_csum_size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558 unsigned long io_block;
559
560 tag = (journal_block_tag_t *) tagp;
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400561 flags = be16_to_cpu(tag->t_flags);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562
563 io_block = next_log_block++;
564 wrap(journal, next_log_block);
565 err = jread(&obh, journal, io_block);
566 if (err) {
567 /* Recover what we can, but
568 * report failure at the end. */
569 success = err;
Eryu Guanf2a44522011-11-01 19:09:18 -0400570 printk(KERN_ERR
571 "JBD2: IO error %d recovering "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700572 "block %ld in log\n",
573 err, io_block);
574 } else {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700575 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576
577 J_ASSERT(obh != NULL);
Zach Brownb517bea2006-10-11 01:21:08 -0700578 blocknr = read_tag_block(tag_bytes,
579 tag);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700580
581 /* If the block has been
582 * revoked, then we're all done
583 * here. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700584 if (jbd2_journal_test_revoke
Dave Kleikamp470decc2006-10-11 01:20:57 -0700585 (journal, blocknr,
586 next_commit_ID)) {
587 brelse(obh);
588 ++info->nr_revoke_hits;
589 goto skip_write;
590 }
591
Darrick J. Wongc3900872012-05-27 08:12:12 -0400592 /* Look for block corruption */
593 if (!jbd2_block_tag_csum_verify(
594 journal, tag, obh->b_data,
595 be32_to_cpu(tmp->h_sequence))) {
596 brelse(obh);
597 success = -EIO;
598 printk(KERN_ERR "JBD: Invalid "
599 "checksum recovering "
600 "block %llu in log\n",
601 blocknr);
602 continue;
603 }
604
Dave Kleikamp470decc2006-10-11 01:20:57 -0700605 /* Find a buffer for the new
606 * data being restored */
607 nbh = __getblk(journal->j_fs_dev,
608 blocknr,
609 journal->j_blocksize);
610 if (nbh == NULL) {
611 printk(KERN_ERR
Eryu Guanf2a44522011-11-01 19:09:18 -0400612 "JBD2: Out of memory "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700613 "during recovery.\n");
614 err = -ENOMEM;
615 brelse(bh);
616 brelse(obh);
617 goto failed;
618 }
619
620 lock_buffer(nbh);
621 memcpy(nbh->b_data, obh->b_data,
622 journal->j_blocksize);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700623 if (flags & JBD2_FLAG_ESCAPE) {
Duane Griffind0025672008-03-19 17:00:54 -0700624 *((__be32 *)nbh->b_data) =
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700625 cpu_to_be32(JBD2_MAGIC_NUMBER);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700626 }
627
628 BUFFER_TRACE(nbh, "marking dirty");
629 set_buffer_uptodate(nbh);
630 mark_buffer_dirty(nbh);
631 BUFFER_TRACE(nbh, "marking uptodate");
632 ++info->nr_replays;
633 /* ll_rw_block(WRITE, 1, &nbh); */
634 unlock_buffer(nbh);
635 brelse(obh);
636 brelse(nbh);
637 }
638
639 skip_write:
Zach Brownb517bea2006-10-11 01:21:08 -0700640 tagp += tag_bytes;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700641 if (!(flags & JBD2_FLAG_SAME_UUID))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700642 tagp += 16;
643
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700644 if (flags & JBD2_FLAG_LAST_TAG)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700645 break;
646 }
647
648 brelse(bh);
649 continue;
650
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700651 case JBD2_COMMIT_BLOCK:
Girish Shilamkar818d2762008-01-28 23:58:27 -0500652 /* How to differentiate between interrupted commit
653 * and journal corruption ?
654 *
655 * {nth transaction}
656 * Checksum Verification Failed
657 * |
658 * ____________________
659 * | |
660 * async_commit sync_commit
661 * | |
662 * | GO TO NEXT "Journal Corruption"
663 * | TRANSACTION
664 * |
665 * {(n+1)th transanction}
666 * |
667 * _______|______________
668 * | |
669 * Commit block found Commit block not found
670 * | |
671 * "Journal Corruption" |
672 * _____________|_________
673 * | |
674 * nth trans corrupt OR nth trans
675 * and (n+1)th interrupted interrupted
676 * before commit block
677 * could reach the disk.
678 * (Cannot find the difference in above
679 * mentioned conditions. Hence assume
680 * "Interrupted Commit".)
681 */
682
683 /* Found an expected commit block: if checksums
684 * are present verify them in PASS_SCAN; else not
685 * much to do other than move on to the next sequence
Dave Kleikamp470decc2006-10-11 01:20:57 -0700686 * number. */
Girish Shilamkar818d2762008-01-28 23:58:27 -0500687 if (pass == PASS_SCAN &&
688 JBD2_HAS_COMPAT_FEATURE(journal,
689 JBD2_FEATURE_COMPAT_CHECKSUM)) {
690 int chksum_err, chksum_seen;
691 struct commit_header *cbh =
692 (struct commit_header *)bh->b_data;
693 unsigned found_chksum =
694 be32_to_cpu(cbh->h_chksum[0]);
695
696 chksum_err = chksum_seen = 0;
697
698 if (info->end_transaction) {
Theodore Ts'o624080e2008-06-06 17:50:40 -0400699 journal->j_failed_commit =
700 info->end_transaction;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500701 brelse(bh);
702 break;
703 }
704
705 if (crc32_sum == found_chksum &&
706 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
707 cbh->h_chksum_size ==
708 JBD2_CRC32_CHKSUM_SIZE)
709 chksum_seen = 1;
710 else if (!(cbh->h_chksum_type == 0 &&
711 cbh->h_chksum_size == 0 &&
712 found_chksum == 0 &&
713 !chksum_seen))
714 /*
715 * If fs is mounted using an old kernel and then
716 * kernel with journal_chksum is used then we
717 * get a situation where the journal flag has
718 * checksum flag set but checksums are not
719 * present i.e chksum = 0, in the individual
720 * commit blocks.
721 * Hence to avoid checksum failures, in this
722 * situation, this extra check is added.
723 */
724 chksum_err = 1;
725
726 if (chksum_err) {
727 info->end_transaction = next_commit_ID;
728
Aneesh Kumar K.V4d605172008-02-05 10:56:15 -0500729 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
Girish Shilamkar818d2762008-01-28 23:58:27 -0500730 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
Theodore Ts'o624080e2008-06-06 17:50:40 -0400731 journal->j_failed_commit =
732 next_commit_ID;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500733 brelse(bh);
734 break;
735 }
736 }
737 crc32_sum = ~0;
738 }
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400739 if (pass == PASS_SCAN &&
740 !jbd2_commit_block_csum_verify(journal,
741 bh->b_data)) {
742 info->end_transaction = next_commit_ID;
743
744 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
745 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
746 journal->j_failed_commit =
747 next_commit_ID;
748 brelse(bh);
749 break;
750 }
751 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700752 brelse(bh);
753 next_commit_ID++;
754 continue;
755
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700756 case JBD2_REVOKE_BLOCK:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700757 /* If we aren't in the REVOKE pass, then we can
758 * just skip over this block. */
759 if (pass != PASS_REVOKE) {
760 brelse(bh);
761 continue;
762 }
763
764 err = scan_revoke_records(journal, bh,
765 next_commit_ID, info);
766 brelse(bh);
767 if (err)
768 goto failed;
769 continue;
770
771 default:
772 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
773 blocktype);
774 brelse(bh);
775 goto done;
776 }
777 }
778
779 done:
780 /*
781 * We broke out of the log scan loop: either we came to the
782 * known end of the log or we found an unexpected block in the
783 * log. If the latter happened, then we know that the "current"
784 * transaction marks the end of the valid log.
785 */
786
Girish Shilamkar818d2762008-01-28 23:58:27 -0500787 if (pass == PASS_SCAN) {
788 if (!info->end_transaction)
789 info->end_transaction = next_commit_ID;
790 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700791 /* It's really bad news if different passes end up at
792 * different places (but possible due to IO errors). */
793 if (info->end_transaction != next_commit_ID) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400794 printk(KERN_ERR "JBD2: recovery pass %d ended at "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700795 "transaction %u, expected %u\n",
796 pass, next_commit_ID, info->end_transaction);
797 if (!success)
798 success = -EIO;
799 }
800 }
801
802 return success;
803
804 failed:
805 return err;
806}
807
Darrick J. Wong42a71062012-05-27 08:08:24 -0400808static int jbd2_revoke_block_csum_verify(journal_t *j,
809 void *buf)
810{
811 struct jbd2_journal_revoke_tail *tail;
812 __u32 provided, calculated;
813
814 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
815 return 1;
816
817 tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize -
818 sizeof(struct jbd2_journal_revoke_tail));
819 provided = tail->r_checksum;
820 tail->r_checksum = 0;
821 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
822 tail->r_checksum = provided;
823
824 provided = be32_to_cpu(provided);
825 return provided == calculated;
826}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700827
828/* Scan a revoke record, marking all blocks mentioned as revoked. */
829
830static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
831 tid_t sequence, struct recovery_info *info)
832{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700833 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700834 int offset, max;
Zach Brownb517bea2006-10-11 01:21:08 -0700835 int record_len = 4;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700836
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700837 header = (jbd2_journal_revoke_header_t *) bh->b_data;
838 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700839 max = be32_to_cpu(header->r_count);
840
Darrick J. Wong42a71062012-05-27 08:08:24 -0400841 if (!jbd2_revoke_block_csum_verify(journal, header))
842 return -EINVAL;
843
Zach Brownb517bea2006-10-11 01:21:08 -0700844 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
845 record_len = 8;
846
847 while (offset + record_len <= max) {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700848 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700849 int err;
850
Zach Brownb517bea2006-10-11 01:21:08 -0700851 if (record_len == 4)
852 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
853 else
854 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
855 offset += record_len;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700856 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700857 if (err)
858 return err;
859 ++info->nr_revokes;
860 }
861 return 0;
862}