blob: df4a20e7d854e51add551ea2990b675c64a0d9a9 [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 */
292 if (journal->j_flags & JBD2_BARRIER)
293 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700294 return err;
295}
296
297/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700298 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
Dave Kleikamp470decc2006-10-11 01:20:57 -0700299 * @journal: journal to startup
300 *
301 * Locate any valid recovery information from the journal and set up the
302 * journal structures in memory to ignore it (presumably because the
303 * caller has evidence that it is out of date).
304 * This function does'nt appear to be exorted..
305 *
306 * We perform one pass over the journal to allow us to tell the user how
307 * much recovery information is being erased, and to let us initialise
308 * the journal transaction sequence numbers to the next unused ID.
309 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700310int jbd2_journal_skip_recovery(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700311{
312 int err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313
314 struct recovery_info info;
315
316 memset (&info, 0, sizeof(info));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700317
318 err = do_one_pass(journal, &info, PASS_SCAN);
319
320 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400321 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700322 ++journal->j_transaction_sequence;
323 } else {
Jose R. Santose23291b2007-07-18 08:57:06 -0400324#ifdef CONFIG_JBD2_DEBUG
Andi Kleen5a0790c2010-06-14 13:28:03 -0400325 int dropped = info.end_transaction -
326 be32_to_cpu(journal->j_superblock->s_sequence);
Mingming Caob38bd332007-07-19 01:48:35 -0700327 jbd_debug(1,
Eryu Guanf2a44522011-11-01 19:09:18 -0400328 "JBD2: ignoring %d transaction%s from the journal.\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700329 dropped, (dropped == 1) ? "" : "s");
Theodore Ts'o9a4f6272010-12-18 13:36:33 -0500330#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700331 journal->j_transaction_sequence = ++info.end_transaction;
332 }
333
334 journal->j_tail = 0;
335 return err;
336}
337
Mingming Cao18eba7a2006-10-11 01:21:13 -0700338static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
Zach Brownb517bea2006-10-11 01:21:08 -0700339{
Mingming Cao18eba7a2006-10-11 01:21:13 -0700340 unsigned long long block = be32_to_cpu(tag->t_blocknr);
Mingming Caocd02ff02007-10-16 18:38:25 -0400341 if (tag_bytes > JBD2_TAG_SIZE32)
Zach Brownb517bea2006-10-11 01:21:08 -0700342 block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
343 return block;
344}
345
Girish Shilamkar818d2762008-01-28 23:58:27 -0500346/*
347 * calc_chksums calculates the checksums for the blocks described in the
348 * descriptor block.
349 */
350static int calc_chksums(journal_t *journal, struct buffer_head *bh,
351 unsigned long *next_log_block, __u32 *crc32_sum)
352{
353 int i, num_blks, err;
354 unsigned long io_block;
355 struct buffer_head *obh;
356
357 num_blks = count_tags(journal, bh);
358 /* Calculate checksum of the descriptor block. */
359 *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
360
361 for (i = 0; i < num_blks; i++) {
362 io_block = (*next_log_block)++;
363 wrap(journal, *next_log_block);
364 err = jread(&obh, journal, io_block);
365 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400366 printk(KERN_ERR "JBD2: IO error %d recovering block "
Girish Shilamkar818d2762008-01-28 23:58:27 -0500367 "%lu in log\n", err, io_block);
368 return 1;
369 } else {
370 *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
371 obh->b_size);
372 }
Theodore Ts'o8ea76902008-05-26 10:28:09 -0400373 put_bh(obh);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500374 }
375 return 0;
376}
377
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400378static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
379{
380 struct commit_header *h;
381 __u32 provided, calculated;
382
383 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
384 return 1;
385
386 h = buf;
387 provided = h->h_chksum[0];
388 h->h_chksum[0] = 0;
389 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
390 h->h_chksum[0] = provided;
391
392 provided = be32_to_cpu(provided);
393 return provided == calculated;
394}
395
Dave Kleikamp470decc2006-10-11 01:20:57 -0700396static int do_one_pass(journal_t *journal,
397 struct recovery_info *info, enum passtype pass)
398{
399 unsigned int first_commit_ID, next_commit_ID;
400 unsigned long next_log_block;
401 int err, success = 0;
402 journal_superblock_t * sb;
403 journal_header_t * tmp;
404 struct buffer_head * bh;
405 unsigned int sequence;
406 int blocktype;
Zach Brownb517bea2006-10-11 01:21:08 -0700407 int tag_bytes = journal_tag_bytes(journal);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500408 __u32 crc32_sum = ~0; /* Transactional Checksums */
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400409 int descr_csum_size = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700410
Dave Kleikamp470decc2006-10-11 01:20:57 -0700411 /*
412 * First thing is to establish what we expect to find in the log
413 * (in terms of transaction IDs), and where (in terms of log
414 * block offsets): query the superblock.
415 */
416
417 sb = journal->j_superblock;
418 next_commit_ID = be32_to_cpu(sb->s_sequence);
419 next_log_block = be32_to_cpu(sb->s_start);
420
421 first_commit_ID = next_commit_ID;
422 if (pass == PASS_SCAN)
423 info->start_transaction = first_commit_ID;
424
425 jbd_debug(1, "Starting recovery pass %d\n", pass);
426
427 /*
428 * Now we walk through the log, transaction by transaction,
429 * making sure that each transaction has a commit block in the
430 * expected place. Each complete transaction gets replayed back
431 * into the main filesystem.
432 */
433
434 while (1) {
435 int flags;
436 char * tagp;
437 journal_block_tag_t * tag;
438 struct buffer_head * obh;
439 struct buffer_head * nbh;
440
Andi Kleene86e1432008-02-06 01:40:11 -0800441 cond_resched();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700442
443 /* If we already know where to stop the log traversal,
444 * check right now that we haven't gone past the end of
445 * the log. */
446
447 if (pass != PASS_SCAN)
448 if (tid_geq(next_commit_ID, info->end_transaction))
449 break;
450
451 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
452 next_commit_ID, next_log_block, journal->j_last);
453
454 /* Skip over each chunk of the transaction looking
455 * either the next descriptor block or the final commit
456 * record. */
457
Eryu Guanf2a44522011-11-01 19:09:18 -0400458 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459 err = jread(&bh, journal, next_log_block);
460 if (err)
461 goto failed;
462
463 next_log_block++;
464 wrap(journal, next_log_block);
465
466 /* What kind of buffer is it?
467 *
468 * If it is a descriptor block, check that it has the
469 * expected sequence number. Otherwise, we're all done
470 * here. */
471
472 tmp = (journal_header_t *)bh->b_data;
473
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700474 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700475 brelse(bh);
476 break;
477 }
478
479 blocktype = be32_to_cpu(tmp->h_blocktype);
480 sequence = be32_to_cpu(tmp->h_sequence);
481 jbd_debug(3, "Found magic %d, sequence %d\n",
482 blocktype, sequence);
483
484 if (sequence != next_commit_ID) {
485 brelse(bh);
486 break;
487 }
488
489 /* OK, we have a valid descriptor block which matches
490 * all of the sequence number checks. What are we going
491 * to do with it? That depends on the pass... */
492
493 switch(blocktype) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700494 case JBD2_DESCRIPTOR_BLOCK:
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400495 /* Verify checksum first */
496 if (JBD2_HAS_INCOMPAT_FEATURE(journal,
497 JBD2_FEATURE_INCOMPAT_CSUM_V2))
498 descr_csum_size =
499 sizeof(struct jbd2_journal_block_tail);
500 if (descr_csum_size > 0 &&
501 !jbd2_descr_block_csum_verify(journal,
502 bh->b_data)) {
503 err = -EIO;
504 goto failed;
505 }
506
Dave Kleikamp470decc2006-10-11 01:20:57 -0700507 /* If it is a valid descriptor block, replay it
Girish Shilamkar818d2762008-01-28 23:58:27 -0500508 * in pass REPLAY; if journal_checksums enabled, then
509 * calculate checksums in PASS_SCAN, otherwise,
510 * just skip over the blocks it describes. */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700511 if (pass != PASS_REPLAY) {
Girish Shilamkar818d2762008-01-28 23:58:27 -0500512 if (pass == PASS_SCAN &&
513 JBD2_HAS_COMPAT_FEATURE(journal,
514 JBD2_FEATURE_COMPAT_CHECKSUM) &&
515 !info->end_transaction) {
516 if (calc_chksums(journal, bh,
517 &next_log_block,
518 &crc32_sum)) {
519 put_bh(bh);
520 break;
521 }
522 put_bh(bh);
523 continue;
524 }
Zach Brownb517bea2006-10-11 01:21:08 -0700525 next_log_block += count_tags(journal, bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526 wrap(journal, next_log_block);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500527 put_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700528 continue;
529 }
530
531 /* A descriptor block: we can now write all of
532 * the data blocks. Yay, useful work is finally
533 * getting done here! */
534
535 tagp = &bh->b_data[sizeof(journal_header_t)];
Zach Brownb517bea2006-10-11 01:21:08 -0700536 while ((tagp - bh->b_data + tag_bytes)
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400537 <= journal->j_blocksize - descr_csum_size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538 unsigned long io_block;
539
540 tag = (journal_block_tag_t *) tagp;
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400541 flags = be16_to_cpu(tag->t_flags);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700542
543 io_block = next_log_block++;
544 wrap(journal, next_log_block);
545 err = jread(&obh, journal, io_block);
546 if (err) {
547 /* Recover what we can, but
548 * report failure at the end. */
549 success = err;
Eryu Guanf2a44522011-11-01 19:09:18 -0400550 printk(KERN_ERR
551 "JBD2: IO error %d recovering "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700552 "block %ld in log\n",
553 err, io_block);
554 } else {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700555 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700556
557 J_ASSERT(obh != NULL);
Zach Brownb517bea2006-10-11 01:21:08 -0700558 blocknr = read_tag_block(tag_bytes,
559 tag);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700560
561 /* If the block has been
562 * revoked, then we're all done
563 * here. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700564 if (jbd2_journal_test_revoke
Dave Kleikamp470decc2006-10-11 01:20:57 -0700565 (journal, blocknr,
566 next_commit_ID)) {
567 brelse(obh);
568 ++info->nr_revoke_hits;
569 goto skip_write;
570 }
571
572 /* Find a buffer for the new
573 * data being restored */
574 nbh = __getblk(journal->j_fs_dev,
575 blocknr,
576 journal->j_blocksize);
577 if (nbh == NULL) {
578 printk(KERN_ERR
Eryu Guanf2a44522011-11-01 19:09:18 -0400579 "JBD2: Out of memory "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700580 "during recovery.\n");
581 err = -ENOMEM;
582 brelse(bh);
583 brelse(obh);
584 goto failed;
585 }
586
587 lock_buffer(nbh);
588 memcpy(nbh->b_data, obh->b_data,
589 journal->j_blocksize);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700590 if (flags & JBD2_FLAG_ESCAPE) {
Duane Griffind0025672008-03-19 17:00:54 -0700591 *((__be32 *)nbh->b_data) =
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700592 cpu_to_be32(JBD2_MAGIC_NUMBER);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700593 }
594
595 BUFFER_TRACE(nbh, "marking dirty");
596 set_buffer_uptodate(nbh);
597 mark_buffer_dirty(nbh);
598 BUFFER_TRACE(nbh, "marking uptodate");
599 ++info->nr_replays;
600 /* ll_rw_block(WRITE, 1, &nbh); */
601 unlock_buffer(nbh);
602 brelse(obh);
603 brelse(nbh);
604 }
605
606 skip_write:
Zach Brownb517bea2006-10-11 01:21:08 -0700607 tagp += tag_bytes;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700608 if (!(flags & JBD2_FLAG_SAME_UUID))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700609 tagp += 16;
610
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700611 if (flags & JBD2_FLAG_LAST_TAG)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700612 break;
613 }
614
615 brelse(bh);
616 continue;
617
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700618 case JBD2_COMMIT_BLOCK:
Girish Shilamkar818d2762008-01-28 23:58:27 -0500619 /* How to differentiate between interrupted commit
620 * and journal corruption ?
621 *
622 * {nth transaction}
623 * Checksum Verification Failed
624 * |
625 * ____________________
626 * | |
627 * async_commit sync_commit
628 * | |
629 * | GO TO NEXT "Journal Corruption"
630 * | TRANSACTION
631 * |
632 * {(n+1)th transanction}
633 * |
634 * _______|______________
635 * | |
636 * Commit block found Commit block not found
637 * | |
638 * "Journal Corruption" |
639 * _____________|_________
640 * | |
641 * nth trans corrupt OR nth trans
642 * and (n+1)th interrupted interrupted
643 * before commit block
644 * could reach the disk.
645 * (Cannot find the difference in above
646 * mentioned conditions. Hence assume
647 * "Interrupted Commit".)
648 */
649
650 /* Found an expected commit block: if checksums
651 * are present verify them in PASS_SCAN; else not
652 * much to do other than move on to the next sequence
Dave Kleikamp470decc2006-10-11 01:20:57 -0700653 * number. */
Girish Shilamkar818d2762008-01-28 23:58:27 -0500654 if (pass == PASS_SCAN &&
655 JBD2_HAS_COMPAT_FEATURE(journal,
656 JBD2_FEATURE_COMPAT_CHECKSUM)) {
657 int chksum_err, chksum_seen;
658 struct commit_header *cbh =
659 (struct commit_header *)bh->b_data;
660 unsigned found_chksum =
661 be32_to_cpu(cbh->h_chksum[0]);
662
663 chksum_err = chksum_seen = 0;
664
665 if (info->end_transaction) {
Theodore Ts'o624080e2008-06-06 17:50:40 -0400666 journal->j_failed_commit =
667 info->end_transaction;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500668 brelse(bh);
669 break;
670 }
671
672 if (crc32_sum == found_chksum &&
673 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
674 cbh->h_chksum_size ==
675 JBD2_CRC32_CHKSUM_SIZE)
676 chksum_seen = 1;
677 else if (!(cbh->h_chksum_type == 0 &&
678 cbh->h_chksum_size == 0 &&
679 found_chksum == 0 &&
680 !chksum_seen))
681 /*
682 * If fs is mounted using an old kernel and then
683 * kernel with journal_chksum is used then we
684 * get a situation where the journal flag has
685 * checksum flag set but checksums are not
686 * present i.e chksum = 0, in the individual
687 * commit blocks.
688 * Hence to avoid checksum failures, in this
689 * situation, this extra check is added.
690 */
691 chksum_err = 1;
692
693 if (chksum_err) {
694 info->end_transaction = next_commit_ID;
695
Aneesh Kumar K.V4d605172008-02-05 10:56:15 -0500696 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
Girish Shilamkar818d2762008-01-28 23:58:27 -0500697 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
Theodore Ts'o624080e2008-06-06 17:50:40 -0400698 journal->j_failed_commit =
699 next_commit_ID;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500700 brelse(bh);
701 break;
702 }
703 }
704 crc32_sum = ~0;
705 }
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400706 if (pass == PASS_SCAN &&
707 !jbd2_commit_block_csum_verify(journal,
708 bh->b_data)) {
709 info->end_transaction = next_commit_ID;
710
711 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
712 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
713 journal->j_failed_commit =
714 next_commit_ID;
715 brelse(bh);
716 break;
717 }
718 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700719 brelse(bh);
720 next_commit_ID++;
721 continue;
722
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700723 case JBD2_REVOKE_BLOCK:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700724 /* If we aren't in the REVOKE pass, then we can
725 * just skip over this block. */
726 if (pass != PASS_REVOKE) {
727 brelse(bh);
728 continue;
729 }
730
731 err = scan_revoke_records(journal, bh,
732 next_commit_ID, info);
733 brelse(bh);
734 if (err)
735 goto failed;
736 continue;
737
738 default:
739 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
740 blocktype);
741 brelse(bh);
742 goto done;
743 }
744 }
745
746 done:
747 /*
748 * We broke out of the log scan loop: either we came to the
749 * known end of the log or we found an unexpected block in the
750 * log. If the latter happened, then we know that the "current"
751 * transaction marks the end of the valid log.
752 */
753
Girish Shilamkar818d2762008-01-28 23:58:27 -0500754 if (pass == PASS_SCAN) {
755 if (!info->end_transaction)
756 info->end_transaction = next_commit_ID;
757 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700758 /* It's really bad news if different passes end up at
759 * different places (but possible due to IO errors). */
760 if (info->end_transaction != next_commit_ID) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400761 printk(KERN_ERR "JBD2: recovery pass %d ended at "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700762 "transaction %u, expected %u\n",
763 pass, next_commit_ID, info->end_transaction);
764 if (!success)
765 success = -EIO;
766 }
767 }
768
769 return success;
770
771 failed:
772 return err;
773}
774
Darrick J. Wong42a71062012-05-27 08:08:24 -0400775static int jbd2_revoke_block_csum_verify(journal_t *j,
776 void *buf)
777{
778 struct jbd2_journal_revoke_tail *tail;
779 __u32 provided, calculated;
780
781 if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
782 return 1;
783
784 tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize -
785 sizeof(struct jbd2_journal_revoke_tail));
786 provided = tail->r_checksum;
787 tail->r_checksum = 0;
788 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
789 tail->r_checksum = provided;
790
791 provided = be32_to_cpu(provided);
792 return provided == calculated;
793}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700794
795/* Scan a revoke record, marking all blocks mentioned as revoked. */
796
797static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
798 tid_t sequence, struct recovery_info *info)
799{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700800 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700801 int offset, max;
Zach Brownb517bea2006-10-11 01:21:08 -0700802 int record_len = 4;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700803
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700804 header = (jbd2_journal_revoke_header_t *) bh->b_data;
805 offset = sizeof(jbd2_journal_revoke_header_t);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700806 max = be32_to_cpu(header->r_count);
807
Darrick J. Wong42a71062012-05-27 08:08:24 -0400808 if (!jbd2_revoke_block_csum_verify(journal, header))
809 return -EINVAL;
810
Zach Brownb517bea2006-10-11 01:21:08 -0700811 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
812 record_len = 8;
813
814 while (offset + record_len <= max) {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700815 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700816 int err;
817
Zach Brownb517bea2006-10-11 01:21:08 -0700818 if (record_len == 4)
819 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
820 else
821 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
822 offset += record_len;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700823 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700824 if (err)
825 return err;
826 ++info->nr_revokes;
827 }
828 return 0;
829}