blob: a4967b27ffb636212ddd21234b1b2fc211a6038b [file] [log] [blame]
Theodore Ts'of5166762017-12-17 22:00:59 -05001// SPDX-License-Identifier: GPL-2.0+
Dave Kleikamp470decc2006-10-11 01:20:57 -07002/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02003 * linux/fs/jbd2/recovery.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07004 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 *
7 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
8 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07009 * Journal recovery routines for the generic filesystem journaling code;
10 * part of the ext2fs journaling system.
11 */
12
13#ifndef __KERNEL__
14#include "jfs_user.h"
15#else
16#include <linux/time.h>
17#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070018#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070019#include <linux/errno.h>
Girish Shilamkar818d2762008-01-28 23:58:27 -050020#include <linux/crc32.h>
Jan Kara79feb522012-03-13 22:22:54 -040021#include <linux/blkdev.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070022#endif
23
24/*
25 * Maintain information about the progress of the recovery job, so that
26 * the different passes can carry information between them.
27 */
28struct recovery_info
29{
30 tid_t start_transaction;
31 tid_t end_transaction;
32
33 int nr_replays;
34 int nr_revokes;
35 int nr_revoke_hits;
36};
37
38enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
39static int do_one_pass(journal_t *journal,
40 struct recovery_info *info, enum passtype pass);
41static int scan_revoke_records(journal_t *, struct buffer_head *,
42 tid_t, struct recovery_info *);
43
44#ifdef __KERNEL__
45
46/* Release readahead buffers after use */
47static void journal_brelse_array(struct buffer_head *b[], int n)
48{
49 while (--n >= 0)
50 brelse (b[n]);
51}
52
53
54/*
55 * When reading from the journal, we are going through the block device
56 * layer directly and so there is no readahead being done for us. We
57 * need to implement any readahead ourselves if we want it to happen at
58 * all. Recovery is basically one long sequential read, so make sure we
59 * do the IO in reasonably large chunks.
60 *
61 * This is not so critical that we need to be enormously clever about
62 * the readahead size, though. 128K is a purely arbitrary, good-enough
63 * fixed value.
64 */
65
66#define MAXBUF 8
67static int do_readahead(journal_t *journal, unsigned int start)
68{
69 int err;
70 unsigned int max, nbufs, next;
Mingming Cao18eba7a2006-10-11 01:21:13 -070071 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -070072 struct buffer_head *bh;
73
74 struct buffer_head * bufs[MAXBUF];
75
76 /* Do up to 128K of readahead */
77 max = start + (128 * 1024 / journal->j_blocksize);
78 if (max > journal->j_maxlen)
79 max = journal->j_maxlen;
80
81 /* Do the readahead itself. We'll submit MAXBUF buffer_heads at
82 * a time to the block device IO layer. */
83
84 nbufs = 0;
85
86 for (next = start; next < max; next++) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -070087 err = jbd2_journal_bmap(journal, next, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -070088
89 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -040090 printk(KERN_ERR "JBD2: bad block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -070091 next);
92 goto failed;
93 }
94
95 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
96 if (!bh) {
97 err = -ENOMEM;
98 goto failed;
99 }
100
101 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
102 bufs[nbufs++] = bh;
103 if (nbufs == MAXBUF) {
Mike Christiedfec8a12016-06-05 14:31:44 -0500104 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700105 journal_brelse_array(bufs, nbufs);
106 nbufs = 0;
107 }
108 } else
109 brelse(bh);
110 }
111
112 if (nbufs)
Mike Christiedfec8a12016-06-05 14:31:44 -0500113 ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700114 err = 0;
115
116failed:
117 if (nbufs)
118 journal_brelse_array(bufs, nbufs);
119 return err;
120}
121
122#endif /* __KERNEL__ */
123
124
125/*
126 * Read a block from the journal
127 */
128
129static int jread(struct buffer_head **bhp, journal_t *journal,
130 unsigned int offset)
131{
132 int err;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700133 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700134 struct buffer_head *bh;
135
136 *bhp = NULL;
137
138 if (offset >= journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400139 printk(KERN_ERR "JBD2: corrupted journal superblock\n");
Darrick J. Wong6a797d22015-10-17 16:16:04 -0400140 return -EFSCORRUPTED;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700141 }
142
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700143 err = jbd2_journal_bmap(journal, offset, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700144
145 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400146 printk(KERN_ERR "JBD2: bad block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700147 offset);
148 return err;
149 }
150
151 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
152 if (!bh)
153 return -ENOMEM;
154
155 if (!buffer_uptodate(bh)) {
156 /* If this is a brand new buffer, start readahead.
157 Otherwise, we assume we are already reading it. */
158 if (!buffer_req(bh))
159 do_readahead(journal, offset);
160 wait_on_buffer(bh);
161 }
162
163 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400164 printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700165 offset);
166 brelse(bh);
167 return -EIO;
168 }
169
170 *bhp = bh;
171 return 0;
172}
173
Jan Kara1101cd42016-02-22 23:19:09 -0500174static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400175{
176 struct jbd2_journal_block_tail *tail;
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400177 __be32 provided;
178 __u32 calculated;
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400179
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400180 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400181 return 1;
182
183 tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
184 sizeof(struct jbd2_journal_block_tail));
185 provided = tail->t_checksum;
186 tail->t_checksum = 0;
187 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
188 tail->t_checksum = provided;
189
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400190 return provided == cpu_to_be32(calculated);
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400191}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700192
193/*
194 * Count the number of in-use tags in a journal descriptor block.
195 */
196
Zach Brownb517bea2006-10-11 01:21:08 -0700197static int count_tags(journal_t *journal, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700198{
199 char * tagp;
200 journal_block_tag_t * tag;
Zach Brownb517bea2006-10-11 01:21:08 -0700201 int nr = 0, size = journal->j_blocksize;
202 int tag_bytes = journal_tag_bytes(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700203
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400204 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400205 size -= sizeof(struct jbd2_journal_block_tail);
206
Dave Kleikamp470decc2006-10-11 01:20:57 -0700207 tagp = &bh->b_data[sizeof(journal_header_t)];
208
Zach Brownb517bea2006-10-11 01:21:08 -0700209 while ((tagp - bh->b_data + tag_bytes) <= size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700210 tag = (journal_block_tag_t *) tagp;
211
212 nr++;
Zach Brownb517bea2006-10-11 01:21:08 -0700213 tagp += tag_bytes;
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400214 if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700215 tagp += 16;
216
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400217 if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700218 break;
219 }
220
221 return nr;
222}
223
224
225/* Make sure we wrap around the log correctly! */
226#define wrap(journal, var) \
227do { \
228 if (var >= (journal)->j_last) \
229 var -= ((journal)->j_last - (journal)->j_first); \
230} while (0)
231
232/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700233 * jbd2_journal_recover - recovers a on-disk journal
Dave Kleikamp470decc2006-10-11 01:20:57 -0700234 * @journal: the journal to recover
235 *
236 * The primary function for recovering the log contents when mounting a
237 * journaled device.
238 *
239 * Recovery is done in three passes. In the first pass, we look for the
240 * end of the log. In the second, we assemble the list of revoke
241 * blocks. In the third and final pass, we replay any un-revoked blocks
242 * in the log.
243 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700244int jbd2_journal_recover(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700245{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400246 int err, err2;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700247 journal_superblock_t * sb;
248
249 struct recovery_info info;
250
251 memset(&info, 0, sizeof(info));
252 sb = journal->j_superblock;
253
254 /*
255 * The journal superblock's s_start field (the current log head)
256 * is always zero if, and only if, the journal was cleanly
257 * unmounted.
258 */
259
260 if (!sb->s_start) {
261 jbd_debug(1, "No recovery required, last transaction %d\n",
262 be32_to_cpu(sb->s_sequence));
263 journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
264 return 0;
265 }
266
267 err = do_one_pass(journal, &info, PASS_SCAN);
268 if (!err)
269 err = do_one_pass(journal, &info, PASS_REVOKE);
270 if (!err)
271 err = do_one_pass(journal, &info, PASS_REPLAY);
272
Eryu Guanf2a44522011-11-01 19:09:18 -0400273 jbd_debug(1, "JBD2: recovery, exit status %d, "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700274 "recovered transactions %u to %u\n",
275 err, info.start_transaction, info.end_transaction);
Eryu Guanf2a44522011-11-01 19:09:18 -0400276 jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700277 info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
278
279 /* Restart the log at the next transaction ID, thus invalidating
280 * any existing commit records in the log. */
281 journal->j_transaction_sequence = ++info.end_transaction;
282
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700283 jbd2_journal_clear_revoke(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400284 err2 = sync_blockdev(journal->j_fs_dev);
285 if (!err)
286 err = err2;
Jan Kara79feb522012-03-13 22:22:54 -0400287 /* Make sure all replayed data is on permanent storage */
Theodore Ts'o316e4cf2012-08-17 09:56:17 -0400288 if (journal->j_flags & JBD2_BARRIER) {
289 err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
290 if (!err)
291 err = err2;
292 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700293 return err;
294}
295
296/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700297 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
Dave Kleikamp470decc2006-10-11 01:20:57 -0700298 * @journal: journal to startup
299 *
300 * Locate any valid recovery information from the journal and set up the
301 * journal structures in memory to ignore it (presumably because the
302 * caller has evidence that it is out of date).
Masanari Iidabd7ced92016-02-02 22:31:06 +0900303 * This function doesn't appear to be exported..
Dave Kleikamp470decc2006-10-11 01:20:57 -0700304 *
305 * We perform one pass over the journal to allow us to tell the user how
306 * much recovery information is being erased, and to let us initialise
307 * the journal transaction sequence numbers to the next unused ID.
308 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700309int jbd2_journal_skip_recovery(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700310{
311 int err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700312
313 struct recovery_info info;
314
315 memset (&info, 0, sizeof(info));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700316
317 err = do_one_pass(journal, &info, PASS_SCAN);
318
319 if (err) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400320 printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700321 ++journal->j_transaction_sequence;
322 } else {
Jose R. Santose23291b2007-07-18 08:57:06 -0400323#ifdef CONFIG_JBD2_DEBUG
Andi Kleen5a0790c2010-06-14 13:28:03 -0400324 int dropped = info.end_transaction -
325 be32_to_cpu(journal->j_superblock->s_sequence);
Mingming Caob38bd332007-07-19 01:48:35 -0700326 jbd_debug(1,
Eryu Guanf2a44522011-11-01 19:09:18 -0400327 "JBD2: ignoring %d transaction%s from the journal.\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700328 dropped, (dropped == 1) ? "" : "s");
Theodore Ts'o9a4f6272010-12-18 13:36:33 -0500329#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700330 journal->j_transaction_sequence = ++info.end_transaction;
331 }
332
333 journal->j_tail = 0;
334 return err;
335}
336
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400337static inline unsigned long long read_tag_block(journal_t *journal,
338 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);
Darrick J. Wong56316a02015-10-17 16:18:45 -0400341 if (jbd2_has_feature_64bit(journal))
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;
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400381 __be32 provided;
382 __u32 calculated;
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400383
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400384 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400385 return 1;
386
387 h = buf;
388 provided = h->h_chksum[0];
389 h->h_chksum[0] = 0;
390 calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
391 h->h_chksum[0] = provided;
392
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400393 return provided == cpu_to_be32(calculated);
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400394}
395
Darrick J. Wongc3900872012-05-27 08:12:12 -0400396static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
397 void *buf, __u32 sequence)
398{
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400399 journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
Darrick J. Wongeee06c52013-05-28 07:31:59 -0400400 __u32 csum32;
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400401 __be32 seq;
Darrick J. Wongc3900872012-05-27 08:12:12 -0400402
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400403 if (!jbd2_journal_has_csum_v2or3(j))
Darrick J. Wongc3900872012-05-27 08:12:12 -0400404 return 1;
405
Darrick J. Wong18a6ea12013-08-28 14:59:58 -0400406 seq = cpu_to_be32(sequence);
407 csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
Darrick J. Wongeee06c52013-05-28 07:31:59 -0400408 csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
Darrick J. Wongc3900872012-05-27 08:12:12 -0400409
Darrick J. Wong56316a02015-10-17 16:18:45 -0400410 if (jbd2_has_feature_csum3(j))
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400411 return tag3->t_checksum == cpu_to_be32(csum32);
412 else
413 return tag->t_checksum == cpu_to_be16(csum32);
Darrick J. Wongc3900872012-05-27 08:12:12 -0400414}
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;
Darrick J. Wong022eaa72014-08-27 18:40:05 -0400430 int block_error = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700431
Dave Kleikamp470decc2006-10-11 01:20:57 -0700432 /*
433 * First thing is to establish what we expect to find in the log
434 * (in terms of transaction IDs), and where (in terms of log
435 * block offsets): query the superblock.
436 */
437
438 sb = journal->j_superblock;
439 next_commit_ID = be32_to_cpu(sb->s_sequence);
440 next_log_block = be32_to_cpu(sb->s_start);
441
442 first_commit_ID = next_commit_ID;
443 if (pass == PASS_SCAN)
444 info->start_transaction = first_commit_ID;
445
446 jbd_debug(1, "Starting recovery pass %d\n", pass);
447
448 /*
449 * Now we walk through the log, transaction by transaction,
450 * making sure that each transaction has a commit block in the
451 * expected place. Each complete transaction gets replayed back
452 * into the main filesystem.
453 */
454
455 while (1) {
456 int flags;
457 char * tagp;
458 journal_block_tag_t * tag;
459 struct buffer_head * obh;
460 struct buffer_head * nbh;
461
Andi Kleene86e1432008-02-06 01:40:11 -0800462 cond_resched();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700463
464 /* If we already know where to stop the log traversal,
465 * check right now that we haven't gone past the end of
466 * the log. */
467
468 if (pass != PASS_SCAN)
469 if (tid_geq(next_commit_ID, info->end_transaction))
470 break;
471
472 jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
473 next_commit_ID, next_log_block, journal->j_last);
474
475 /* Skip over each chunk of the transaction looking
476 * either the next descriptor block or the final commit
477 * record. */
478
Eryu Guanf2a44522011-11-01 19:09:18 -0400479 jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700480 err = jread(&bh, journal, next_log_block);
481 if (err)
482 goto failed;
483
484 next_log_block++;
485 wrap(journal, next_log_block);
486
487 /* What kind of buffer is it?
488 *
489 * If it is a descriptor block, check that it has the
490 * expected sequence number. Otherwise, we're all done
491 * here. */
492
493 tmp = (journal_header_t *)bh->b_data;
494
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700495 if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700496 brelse(bh);
497 break;
498 }
499
500 blocktype = be32_to_cpu(tmp->h_blocktype);
501 sequence = be32_to_cpu(tmp->h_sequence);
502 jbd_debug(3, "Found magic %d, sequence %d\n",
503 blocktype, sequence);
504
505 if (sequence != next_commit_ID) {
506 brelse(bh);
507 break;
508 }
509
510 /* OK, we have a valid descriptor block which matches
511 * all of the sequence number checks. What are we going
512 * to do with it? That depends on the pass... */
513
514 switch(blocktype) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700515 case JBD2_DESCRIPTOR_BLOCK:
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400516 /* Verify checksum first */
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400517 if (jbd2_journal_has_csum_v2or3(journal))
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400518 descr_csum_size =
519 sizeof(struct jbd2_journal_block_tail);
520 if (descr_csum_size > 0 &&
Jan Kara1101cd42016-02-22 23:19:09 -0500521 !jbd2_descriptor_block_csum_verify(journal,
522 bh->b_data)) {
Darrick J. Wongb6924222015-01-19 15:59:58 -0500523 printk(KERN_ERR "JBD2: Invalid checksum "
524 "recovering block %lu in log\n",
525 next_log_block);
Darrick J. Wong6a797d22015-10-17 16:16:04 -0400526 err = -EFSBADCRC;
Darrick J. Wong064d8382014-09-16 14:43:09 -0400527 brelse(bh);
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400528 goto failed;
529 }
530
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531 /* If it is a valid descriptor block, replay it
Girish Shilamkar818d2762008-01-28 23:58:27 -0500532 * in pass REPLAY; if journal_checksums enabled, then
533 * calculate checksums in PASS_SCAN, otherwise,
534 * just skip over the blocks it describes. */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700535 if (pass != PASS_REPLAY) {
Girish Shilamkar818d2762008-01-28 23:58:27 -0500536 if (pass == PASS_SCAN &&
Darrick J. Wong56316a02015-10-17 16:18:45 -0400537 jbd2_has_feature_checksum(journal) &&
Girish Shilamkar818d2762008-01-28 23:58:27 -0500538 !info->end_transaction) {
539 if (calc_chksums(journal, bh,
540 &next_log_block,
541 &crc32_sum)) {
542 put_bh(bh);
543 break;
544 }
545 put_bh(bh);
546 continue;
547 }
Zach Brownb517bea2006-10-11 01:21:08 -0700548 next_log_block += count_tags(journal, bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700549 wrap(journal, next_log_block);
Girish Shilamkar818d2762008-01-28 23:58:27 -0500550 put_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700551 continue;
552 }
553
554 /* A descriptor block: we can now write all of
555 * the data blocks. Yay, useful work is finally
556 * getting done here! */
557
558 tagp = &bh->b_data[sizeof(journal_header_t)];
Zach Brownb517bea2006-10-11 01:21:08 -0700559 while ((tagp - bh->b_data + tag_bytes)
Darrick J. Wong3caa4872012-05-27 08:10:22 -0400560 <= journal->j_blocksize - descr_csum_size) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700561 unsigned long io_block;
562
563 tag = (journal_block_tag_t *) tagp;
Darrick J. Wong8f888ef2012-05-22 22:43:41 -0400564 flags = be16_to_cpu(tag->t_flags);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700565
566 io_block = next_log_block++;
567 wrap(journal, next_log_block);
568 err = jread(&obh, journal, io_block);
569 if (err) {
570 /* Recover what we can, but
571 * report failure at the end. */
572 success = err;
Eryu Guanf2a44522011-11-01 19:09:18 -0400573 printk(KERN_ERR
574 "JBD2: IO error %d recovering "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700575 "block %ld in log\n",
576 err, io_block);
577 } else {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700578 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700579
580 J_ASSERT(obh != NULL);
Darrick J. Wongdb9ee222014-08-27 18:40:07 -0400581 blocknr = read_tag_block(journal,
Zach Brownb517bea2006-10-11 01:21:08 -0700582 tag);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700583
584 /* If the block has been
585 * revoked, then we're all done
586 * here. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700587 if (jbd2_journal_test_revoke
Dave Kleikamp470decc2006-10-11 01:20:57 -0700588 (journal, blocknr,
589 next_commit_ID)) {
590 brelse(obh);
591 ++info->nr_revoke_hits;
592 goto skip_write;
593 }
594
Darrick J. Wongc3900872012-05-27 08:12:12 -0400595 /* Look for block corruption */
596 if (!jbd2_block_tag_csum_verify(
597 journal, tag, obh->b_data,
598 be32_to_cpu(tmp->h_sequence))) {
599 brelse(obh);
Darrick J. Wong6a797d22015-10-17 16:16:04 -0400600 success = -EFSBADCRC;
Dmitry Monakhova67c8482013-12-08 21:14:59 -0500601 printk(KERN_ERR "JBD2: Invalid "
Darrick J. Wongc3900872012-05-27 08:12:12 -0400602 "checksum recovering "
Theodore Ts'oed65b002018-02-18 21:33:13 -0500603 "data block %llu in "
604 "log\n", blocknr);
Darrick J. Wong022eaa72014-08-27 18:40:05 -0400605 block_error = 1;
606 goto skip_write;
Darrick J. Wongc3900872012-05-27 08:12:12 -0400607 }
608
Dave Kleikamp470decc2006-10-11 01:20:57 -0700609 /* Find a buffer for the new
610 * data being restored */
611 nbh = __getblk(journal->j_fs_dev,
612 blocknr,
613 journal->j_blocksize);
614 if (nbh == NULL) {
615 printk(KERN_ERR
Eryu Guanf2a44522011-11-01 19:09:18 -0400616 "JBD2: Out of memory "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700617 "during recovery.\n");
618 err = -ENOMEM;
619 brelse(bh);
620 brelse(obh);
621 goto failed;
622 }
623
624 lock_buffer(nbh);
625 memcpy(nbh->b_data, obh->b_data,
626 journal->j_blocksize);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700627 if (flags & JBD2_FLAG_ESCAPE) {
Duane Griffind0025672008-03-19 17:00:54 -0700628 *((__be32 *)nbh->b_data) =
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700629 cpu_to_be32(JBD2_MAGIC_NUMBER);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700630 }
631
632 BUFFER_TRACE(nbh, "marking dirty");
633 set_buffer_uptodate(nbh);
634 mark_buffer_dirty(nbh);
635 BUFFER_TRACE(nbh, "marking uptodate");
636 ++info->nr_replays;
637 /* ll_rw_block(WRITE, 1, &nbh); */
638 unlock_buffer(nbh);
639 brelse(obh);
640 brelse(nbh);
641 }
642
643 skip_write:
Zach Brownb517bea2006-10-11 01:21:08 -0700644 tagp += tag_bytes;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700645 if (!(flags & JBD2_FLAG_SAME_UUID))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646 tagp += 16;
647
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700648 if (flags & JBD2_FLAG_LAST_TAG)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700649 break;
650 }
651
652 brelse(bh);
653 continue;
654
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700655 case JBD2_COMMIT_BLOCK:
Girish Shilamkar818d2762008-01-28 23:58:27 -0500656 /* How to differentiate between interrupted commit
657 * and journal corruption ?
658 *
659 * {nth transaction}
660 * Checksum Verification Failed
661 * |
662 * ____________________
663 * | |
664 * async_commit sync_commit
665 * | |
666 * | GO TO NEXT "Journal Corruption"
667 * | TRANSACTION
668 * |
669 * {(n+1)th transanction}
670 * |
671 * _______|______________
672 * | |
673 * Commit block found Commit block not found
674 * | |
675 * "Journal Corruption" |
676 * _____________|_________
677 * | |
678 * nth trans corrupt OR nth trans
679 * and (n+1)th interrupted interrupted
680 * before commit block
681 * could reach the disk.
682 * (Cannot find the difference in above
683 * mentioned conditions. Hence assume
684 * "Interrupted Commit".)
685 */
686
687 /* Found an expected commit block: if checksums
688 * are present verify them in PASS_SCAN; else not
689 * much to do other than move on to the next sequence
Dave Kleikamp470decc2006-10-11 01:20:57 -0700690 * number. */
Girish Shilamkar818d2762008-01-28 23:58:27 -0500691 if (pass == PASS_SCAN &&
Darrick J. Wong56316a02015-10-17 16:18:45 -0400692 jbd2_has_feature_checksum(journal)) {
Girish Shilamkar818d2762008-01-28 23:58:27 -0500693 int chksum_err, chksum_seen;
694 struct commit_header *cbh =
695 (struct commit_header *)bh->b_data;
696 unsigned found_chksum =
697 be32_to_cpu(cbh->h_chksum[0]);
698
699 chksum_err = chksum_seen = 0;
700
701 if (info->end_transaction) {
Theodore Ts'o624080e2008-06-06 17:50:40 -0400702 journal->j_failed_commit =
703 info->end_transaction;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500704 brelse(bh);
705 break;
706 }
707
708 if (crc32_sum == found_chksum &&
709 cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
710 cbh->h_chksum_size ==
711 JBD2_CRC32_CHKSUM_SIZE)
712 chksum_seen = 1;
713 else if (!(cbh->h_chksum_type == 0 &&
714 cbh->h_chksum_size == 0 &&
715 found_chksum == 0 &&
716 !chksum_seen))
717 /*
718 * If fs is mounted using an old kernel and then
719 * kernel with journal_chksum is used then we
720 * get a situation where the journal flag has
721 * checksum flag set but checksums are not
722 * present i.e chksum = 0, in the individual
723 * commit blocks.
724 * Hence to avoid checksum failures, in this
725 * situation, this extra check is added.
726 */
727 chksum_err = 1;
728
729 if (chksum_err) {
730 info->end_transaction = next_commit_ID;
731
Darrick J. Wong56316a02015-10-17 16:18:45 -0400732 if (!jbd2_has_feature_async_commit(journal)) {
Theodore Ts'o624080e2008-06-06 17:50:40 -0400733 journal->j_failed_commit =
734 next_commit_ID;
Girish Shilamkar818d2762008-01-28 23:58:27 -0500735 brelse(bh);
736 break;
737 }
738 }
739 crc32_sum = ~0;
740 }
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400741 if (pass == PASS_SCAN &&
742 !jbd2_commit_block_csum_verify(journal,
743 bh->b_data)) {
744 info->end_transaction = next_commit_ID;
745
Darrick J. Wong56316a02015-10-17 16:18:45 -0400746 if (!jbd2_has_feature_async_commit(journal)) {
Darrick J. Wong1f56c582012-05-27 08:10:25 -0400747 journal->j_failed_commit =
748 next_commit_ID;
749 brelse(bh);
750 break;
751 }
752 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700753 brelse(bh);
754 next_commit_ID++;
755 continue;
756
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700757 case JBD2_REVOKE_BLOCK:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700758 /* If we aren't in the REVOKE pass, then we can
759 * just skip over this block. */
760 if (pass != PASS_REVOKE) {
761 brelse(bh);
762 continue;
763 }
764
765 err = scan_revoke_records(journal, bh,
766 next_commit_ID, info);
767 brelse(bh);
768 if (err)
769 goto failed;
770 continue;
771
772 default:
773 jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
774 blocktype);
775 brelse(bh);
776 goto done;
777 }
778 }
779
780 done:
781 /*
782 * We broke out of the log scan loop: either we came to the
783 * known end of the log or we found an unexpected block in the
784 * log. If the latter happened, then we know that the "current"
785 * transaction marks the end of the valid log.
786 */
787
Girish Shilamkar818d2762008-01-28 23:58:27 -0500788 if (pass == PASS_SCAN) {
789 if (!info->end_transaction)
790 info->end_transaction = next_commit_ID;
791 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700792 /* It's really bad news if different passes end up at
793 * different places (but possible due to IO errors). */
794 if (info->end_transaction != next_commit_ID) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400795 printk(KERN_ERR "JBD2: recovery pass %d ended at "
Dave Kleikamp470decc2006-10-11 01:20:57 -0700796 "transaction %u, expected %u\n",
797 pass, next_commit_ID, info->end_transaction);
798 if (!success)
799 success = -EIO;
800 }
801 }
Darrick J. Wong022eaa72014-08-27 18:40:05 -0400802 if (block_error && success == 0)
803 success = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700804 return success;
805
806 failed:
807 return err;
808}
809
Dave Kleikamp470decc2006-10-11 01:20:57 -0700810/* Scan a revoke record, marking all blocks mentioned as revoked. */
811
812static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
813 tid_t sequence, struct recovery_info *info)
814{
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700815 jbd2_journal_revoke_header_t *header;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700816 int offset, max;
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400817 int csum_size = 0;
818 __u32 rcount;
Zach Brownb517bea2006-10-11 01:21:08 -0700819 int record_len = 4;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700820
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700821 header = (jbd2_journal_revoke_header_t *) bh->b_data;
822 offset = sizeof(jbd2_journal_revoke_header_t);
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400823 rcount = be32_to_cpu(header->r_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700824
Jan Kara1101cd42016-02-22 23:19:09 -0500825 if (!jbd2_descriptor_block_csum_verify(journal, header))
Darrick J. Wong6a797d22015-10-17 16:16:04 -0400826 return -EFSBADCRC;
Darrick J. Wong42a71062012-05-27 08:08:24 -0400827
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400828 if (jbd2_journal_has_csum_v2or3(journal))
Jan Kara1101cd42016-02-22 23:19:09 -0500829 csum_size = sizeof(struct jbd2_journal_block_tail);
Darrick J. Wonge531d0b2015-05-14 19:11:50 -0400830 if (rcount > journal->j_blocksize - csum_size)
831 return -EINVAL;
832 max = rcount;
833
Darrick J. Wong56316a02015-10-17 16:18:45 -0400834 if (jbd2_has_feature_64bit(journal))
Zach Brownb517bea2006-10-11 01:21:08 -0700835 record_len = 8;
836
837 while (offset + record_len <= max) {
Mingming Cao18eba7a2006-10-11 01:21:13 -0700838 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700839 int err;
840
Zach Brownb517bea2006-10-11 01:21:08 -0700841 if (record_len == 4)
842 blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
843 else
844 blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
845 offset += record_len;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700846 err = jbd2_journal_set_revoke(journal, blocknr, sequence);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700847 if (err)
848 return err;
849 ++info->nr_revokes;
850 }
851 return 0;
852}