Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
58862699 | 3 | * linux/fs/jbd2/recovery.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 | |
6 | * | |
7 | * Copyright 1999-2000 Red Hat Software --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * 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> | |
f7f4bccb | 18 | #include <linux/jbd2.h> |
470decc6 | 19 | #include <linux/errno.h> |
818d276c | 20 | #include <linux/crc32.h> |
79feb521 | 21 | #include <linux/blkdev.h> |
470decc6 DK |
22 | #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 | */ | |
28 | struct 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 | ||
470decc6 DK |
38 | static int do_one_pass(journal_t *journal, |
39 | struct recovery_info *info, enum passtype pass); | |
40 | static int scan_revoke_records(journal_t *, struct buffer_head *, | |
41 | tid_t, struct recovery_info *); | |
42 | ||
43 | #ifdef __KERNEL__ | |
44 | ||
45 | /* Release readahead buffers after use */ | |
46 | static void journal_brelse_array(struct buffer_head *b[], int n) | |
47 | { | |
48 | while (--n >= 0) | |
49 | brelse (b[n]); | |
50 | } | |
51 | ||
52 | ||
53 | /* | |
54 | * When reading from the journal, we are going through the block device | |
55 | * layer directly and so there is no readahead being done for us. We | |
56 | * need to implement any readahead ourselves if we want it to happen at | |
57 | * all. Recovery is basically one long sequential read, so make sure we | |
58 | * do the IO in reasonably large chunks. | |
59 | * | |
60 | * This is not so critical that we need to be enormously clever about | |
61 | * the readahead size, though. 128K is a purely arbitrary, good-enough | |
62 | * fixed value. | |
63 | */ | |
64 | ||
65 | #define MAXBUF 8 | |
66 | static int do_readahead(journal_t *journal, unsigned int start) | |
67 | { | |
68 | int err; | |
69 | unsigned int max, nbufs, next; | |
18eba7aa | 70 | unsigned long long blocknr; |
470decc6 DK |
71 | struct buffer_head *bh; |
72 | ||
73 | struct buffer_head * bufs[MAXBUF]; | |
74 | ||
75 | /* Do up to 128K of readahead */ | |
76 | max = start + (128 * 1024 / journal->j_blocksize); | |
ede7dc7f HS |
77 | if (max > journal->j_total_len) |
78 | max = journal->j_total_len; | |
470decc6 DK |
79 | |
80 | /* Do the readahead itself. We'll submit MAXBUF buffer_heads at | |
81 | * a time to the block device IO layer. */ | |
82 | ||
83 | nbufs = 0; | |
84 | ||
85 | for (next = start; next < max; next++) { | |
f7f4bccb | 86 | err = jbd2_journal_bmap(journal, next, &blocknr); |
470decc6 DK |
87 | |
88 | if (err) { | |
f2a44523 | 89 | printk(KERN_ERR "JBD2: bad block at offset %u\n", |
470decc6 DK |
90 | next); |
91 | goto failed; | |
92 | } | |
93 | ||
94 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
95 | if (!bh) { | |
96 | err = -ENOMEM; | |
97 | goto failed; | |
98 | } | |
99 | ||
100 | if (!buffer_uptodate(bh) && !buffer_locked(bh)) { | |
101 | bufs[nbufs++] = bh; | |
102 | if (nbufs == MAXBUF) { | |
8c004d1f | 103 | bh_readahead_batch(nbufs, bufs, 0); |
470decc6 DK |
104 | journal_brelse_array(bufs, nbufs); |
105 | nbufs = 0; | |
106 | } | |
107 | } else | |
108 | brelse(bh); | |
109 | } | |
110 | ||
111 | if (nbufs) | |
8c004d1f | 112 | bh_readahead_batch(nbufs, bufs, 0); |
470decc6 DK |
113 | err = 0; |
114 | ||
115 | failed: | |
116 | if (nbufs) | |
117 | journal_brelse_array(bufs, nbufs); | |
118 | return err; | |
119 | } | |
120 | ||
121 | #endif /* __KERNEL__ */ | |
122 | ||
123 | ||
124 | /* | |
125 | * Read a block from the journal | |
126 | */ | |
127 | ||
128 | static int jread(struct buffer_head **bhp, journal_t *journal, | |
129 | unsigned int offset) | |
130 | { | |
131 | int err; | |
18eba7aa | 132 | unsigned long long blocknr; |
470decc6 DK |
133 | struct buffer_head *bh; |
134 | ||
135 | *bhp = NULL; | |
136 | ||
ede7dc7f | 137 | if (offset >= journal->j_total_len) { |
f2a44523 | 138 | printk(KERN_ERR "JBD2: corrupted journal superblock\n"); |
6a797d27 | 139 | return -EFSCORRUPTED; |
470decc6 DK |
140 | } |
141 | ||
f7f4bccb | 142 | err = jbd2_journal_bmap(journal, offset, &blocknr); |
470decc6 DK |
143 | |
144 | if (err) { | |
f2a44523 | 145 | printk(KERN_ERR "JBD2: bad block at offset %u\n", |
470decc6 DK |
146 | offset); |
147 | return err; | |
148 | } | |
149 | ||
150 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
151 | if (!bh) | |
152 | return -ENOMEM; | |
153 | ||
154 | if (!buffer_uptodate(bh)) { | |
8c004d1f ZY |
155 | /* |
156 | * If this is a brand new buffer, start readahead. | |
157 | * Otherwise, we assume we are already reading it. | |
158 | */ | |
159 | bool need_readahead = !buffer_req(bh); | |
160 | ||
161 | bh_read_nowait(bh, 0); | |
162 | if (need_readahead) | |
470decc6 DK |
163 | do_readahead(journal, offset); |
164 | wait_on_buffer(bh); | |
165 | } | |
166 | ||
167 | if (!buffer_uptodate(bh)) { | |
f2a44523 | 168 | printk(KERN_ERR "JBD2: Failed to read block at offset %u\n", |
470decc6 DK |
169 | offset); |
170 | brelse(bh); | |
171 | return -EIO; | |
172 | } | |
173 | ||
174 | *bhp = bh; | |
175 | return 0; | |
176 | } | |
177 | ||
1101cd4d | 178 | static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf) |
3caa487f DW |
179 | { |
180 | struct jbd2_journal_block_tail *tail; | |
18a6ea1e DW |
181 | __be32 provided; |
182 | __u32 calculated; | |
3caa487f | 183 | |
db9ee220 | 184 | if (!jbd2_journal_has_csum_v2or3(j)) |
3caa487f DW |
185 | return 1; |
186 | ||
4009cc7a TT |
187 | tail = (struct jbd2_journal_block_tail *)((char *)buf + |
188 | j->j_blocksize - sizeof(struct jbd2_journal_block_tail)); | |
3caa487f DW |
189 | provided = tail->t_checksum; |
190 | tail->t_checksum = 0; | |
191 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
192 | tail->t_checksum = provided; | |
193 | ||
18a6ea1e | 194 | return provided == cpu_to_be32(calculated); |
3caa487f | 195 | } |
470decc6 DK |
196 | |
197 | /* | |
198 | * Count the number of in-use tags in a journal descriptor block. | |
199 | */ | |
200 | ||
b517bea1 | 201 | static int count_tags(journal_t *journal, struct buffer_head *bh) |
470decc6 DK |
202 | { |
203 | char * tagp; | |
a20d1ceb | 204 | journal_block_tag_t tag; |
b517bea1 ZB |
205 | int nr = 0, size = journal->j_blocksize; |
206 | int tag_bytes = journal_tag_bytes(journal); | |
470decc6 | 207 | |
db9ee220 | 208 | if (jbd2_journal_has_csum_v2or3(journal)) |
3caa487f DW |
209 | size -= sizeof(struct jbd2_journal_block_tail); |
210 | ||
470decc6 DK |
211 | tagp = &bh->b_data[sizeof(journal_header_t)]; |
212 | ||
b517bea1 | 213 | while ((tagp - bh->b_data + tag_bytes) <= size) { |
a20d1ceb | 214 | memcpy(&tag, tagp, sizeof(tag)); |
470decc6 DK |
215 | |
216 | nr++; | |
b517bea1 | 217 | tagp += tag_bytes; |
a20d1ceb | 218 | if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID))) |
470decc6 DK |
219 | tagp += 16; |
220 | ||
a20d1ceb | 221 | if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG)) |
470decc6 DK |
222 | break; |
223 | } | |
224 | ||
225 | return nr; | |
226 | } | |
227 | ||
228 | ||
229 | /* Make sure we wrap around the log correctly! */ | |
230 | #define wrap(journal, var) \ | |
231 | do { \ | |
5b849b5f HS |
232 | unsigned long _wrap_last = \ |
233 | jbd2_has_feature_fast_commit(journal) ? \ | |
234 | (journal)->j_fc_last : (journal)->j_last; \ | |
235 | \ | |
236 | if (var >= _wrap_last) \ | |
237 | var -= (_wrap_last - (journal)->j_first); \ | |
470decc6 DK |
238 | } while (0) |
239 | ||
5b849b5f HS |
240 | static int fc_do_one_pass(journal_t *journal, |
241 | struct recovery_info *info, enum passtype pass) | |
242 | { | |
243 | unsigned int expected_commit_id = info->end_transaction; | |
244 | unsigned long next_fc_block; | |
245 | struct buffer_head *bh; | |
246 | int err = 0; | |
247 | ||
248 | next_fc_block = journal->j_fc_first; | |
249 | if (!journal->j_fc_replay_callback) | |
250 | return 0; | |
251 | ||
252 | while (next_fc_block <= journal->j_fc_last) { | |
cb3b3bf2 | 253 | jbd2_debug(3, "Fast commit replay: next block %ld\n", |
5b849b5f HS |
254 | next_fc_block); |
255 | err = jread(&bh, journal, next_fc_block); | |
256 | if (err) { | |
cb3b3bf2 | 257 | jbd2_debug(3, "Fast commit replay: read error\n"); |
5b849b5f HS |
258 | break; |
259 | } | |
260 | ||
5b849b5f HS |
261 | err = journal->j_fc_replay_callback(journal, bh, pass, |
262 | next_fc_block - journal->j_fc_first, | |
263 | expected_commit_id); | |
dfff66f3 | 264 | brelse(bh); |
5b849b5f HS |
265 | next_fc_block++; |
266 | if (err < 0 || err == JBD2_FC_REPLAY_STOP) | |
267 | break; | |
268 | err = 0; | |
269 | } | |
270 | ||
271 | if (err) | |
cb3b3bf2 | 272 | jbd2_debug(3, "Fast commit replay failed, err = %d\n", err); |
5b849b5f HS |
273 | |
274 | return err; | |
275 | } | |
276 | ||
470decc6 | 277 | /** |
f7f4bccb | 278 | * jbd2_journal_recover - recovers a on-disk journal |
470decc6 DK |
279 | * @journal: the journal to recover |
280 | * | |
281 | * The primary function for recovering the log contents when mounting a | |
282 | * journaled device. | |
283 | * | |
284 | * Recovery is done in three passes. In the first pass, we look for the | |
285 | * end of the log. In the second, we assemble the list of revoke | |
286 | * blocks. In the third and final pass, we replay any un-revoked blocks | |
287 | * in the log. | |
288 | */ | |
f7f4bccb | 289 | int jbd2_journal_recover(journal_t *journal) |
470decc6 | 290 | { |
44519faf | 291 | int err, err2; |
470decc6 DK |
292 | journal_superblock_t * sb; |
293 | ||
294 | struct recovery_info info; | |
295 | ||
296 | memset(&info, 0, sizeof(info)); | |
297 | sb = journal->j_superblock; | |
298 | ||
299 | /* | |
300 | * The journal superblock's s_start field (the current log head) | |
301 | * is always zero if, and only if, the journal was cleanly | |
302 | * unmounted. | |
303 | */ | |
304 | ||
305 | if (!sb->s_start) { | |
cb3b3bf2 | 306 | jbd2_debug(1, "No recovery required, last transaction %d\n", |
470decc6 DK |
307 | be32_to_cpu(sb->s_sequence)); |
308 | journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1; | |
309 | return 0; | |
310 | } | |
311 | ||
312 | err = do_one_pass(journal, &info, PASS_SCAN); | |
313 | if (!err) | |
314 | err = do_one_pass(journal, &info, PASS_REVOKE); | |
315 | if (!err) | |
316 | err = do_one_pass(journal, &info, PASS_REPLAY); | |
317 | ||
cb3b3bf2 | 318 | jbd2_debug(1, "JBD2: recovery, exit status %d, " |
470decc6 DK |
319 | "recovered transactions %u to %u\n", |
320 | err, info.start_transaction, info.end_transaction); | |
cb3b3bf2 | 321 | jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n", |
470decc6 DK |
322 | info.nr_replays, info.nr_revoke_hits, info.nr_revokes); |
323 | ||
324 | /* Restart the log at the next transaction ID, thus invalidating | |
325 | * any existing commit records in the log. */ | |
326 | journal->j_transaction_sequence = ++info.end_transaction; | |
327 | ||
f7f4bccb | 328 | jbd2_journal_clear_revoke(journal); |
44519faf HK |
329 | err2 = sync_blockdev(journal->j_fs_dev); |
330 | if (!err) | |
331 | err = err2; | |
79feb521 | 332 | /* Make sure all replayed data is on permanent storage */ |
316e4cfd | 333 | if (journal->j_flags & JBD2_BARRIER) { |
c6bf3f0e | 334 | err2 = blkdev_issue_flush(journal->j_fs_dev); |
316e4cfd TT |
335 | if (!err) |
336 | err = err2; | |
337 | } | |
470decc6 DK |
338 | return err; |
339 | } | |
340 | ||
341 | /** | |
f7f4bccb | 342 | * jbd2_journal_skip_recovery - Start journal and wipe exiting records |
470decc6 DK |
343 | * @journal: journal to startup |
344 | * | |
345 | * Locate any valid recovery information from the journal and set up the | |
346 | * journal structures in memory to ignore it (presumably because the | |
347 | * caller has evidence that it is out of date). | |
bd7ced98 | 348 | * This function doesn't appear to be exported.. |
470decc6 DK |
349 | * |
350 | * We perform one pass over the journal to allow us to tell the user how | |
351 | * much recovery information is being erased, and to let us initialise | |
352 | * the journal transaction sequence numbers to the next unused ID. | |
353 | */ | |
f7f4bccb | 354 | int jbd2_journal_skip_recovery(journal_t *journal) |
470decc6 DK |
355 | { |
356 | int err; | |
470decc6 DK |
357 | |
358 | struct recovery_info info; | |
359 | ||
360 | memset (&info, 0, sizeof(info)); | |
470decc6 DK |
361 | |
362 | err = do_one_pass(journal, &info, PASS_SCAN); | |
363 | ||
364 | if (err) { | |
f2a44523 | 365 | printk(KERN_ERR "JBD2: error %d scanning journal\n", err); |
470decc6 DK |
366 | ++journal->j_transaction_sequence; |
367 | } else { | |
e23291b9 | 368 | #ifdef CONFIG_JBD2_DEBUG |
5a0790c2 AK |
369 | int dropped = info.end_transaction - |
370 | be32_to_cpu(journal->j_superblock->s_sequence); | |
cb3b3bf2 | 371 | jbd2_debug(1, |
f2a44523 | 372 | "JBD2: ignoring %d transaction%s from the journal.\n", |
470decc6 | 373 | dropped, (dropped == 1) ? "" : "s"); |
9a4f6271 | 374 | #endif |
470decc6 DK |
375 | journal->j_transaction_sequence = ++info.end_transaction; |
376 | } | |
377 | ||
378 | journal->j_tail = 0; | |
379 | return err; | |
380 | } | |
381 | ||
db9ee220 DW |
382 | static inline unsigned long long read_tag_block(journal_t *journal, |
383 | journal_block_tag_t *tag) | |
b517bea1 | 384 | { |
18eba7aa | 385 | unsigned long long block = be32_to_cpu(tag->t_blocknr); |
56316a0d | 386 | if (jbd2_has_feature_64bit(journal)) |
b517bea1 ZB |
387 | block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32; |
388 | return block; | |
389 | } | |
390 | ||
818d276c GS |
391 | /* |
392 | * calc_chksums calculates the checksums for the blocks described in the | |
393 | * descriptor block. | |
394 | */ | |
395 | static int calc_chksums(journal_t *journal, struct buffer_head *bh, | |
396 | unsigned long *next_log_block, __u32 *crc32_sum) | |
397 | { | |
398 | int i, num_blks, err; | |
399 | unsigned long io_block; | |
400 | struct buffer_head *obh; | |
401 | ||
402 | num_blks = count_tags(journal, bh); | |
403 | /* Calculate checksum of the descriptor block. */ | |
404 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); | |
405 | ||
406 | for (i = 0; i < num_blks; i++) { | |
407 | io_block = (*next_log_block)++; | |
408 | wrap(journal, *next_log_block); | |
409 | err = jread(&obh, journal, io_block); | |
410 | if (err) { | |
f2a44523 | 411 | printk(KERN_ERR "JBD2: IO error %d recovering block " |
818d276c GS |
412 | "%lu in log\n", err, io_block); |
413 | return 1; | |
414 | } else { | |
415 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, | |
416 | obh->b_size); | |
417 | } | |
8ea76900 | 418 | put_bh(obh); |
818d276c GS |
419 | } |
420 | return 0; | |
421 | } | |
422 | ||
1f56c589 DW |
423 | static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) |
424 | { | |
425 | struct commit_header *h; | |
18a6ea1e DW |
426 | __be32 provided; |
427 | __u32 calculated; | |
1f56c589 | 428 | |
db9ee220 | 429 | if (!jbd2_journal_has_csum_v2or3(j)) |
1f56c589 DW |
430 | return 1; |
431 | ||
432 | h = buf; | |
433 | provided = h->h_chksum[0]; | |
434 | h->h_chksum[0] = 0; | |
435 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
436 | h->h_chksum[0] = provided; | |
437 | ||
18a6ea1e | 438 | return provided == cpu_to_be32(calculated); |
1f56c589 DW |
439 | } |
440 | ||
c3900875 | 441 | static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, |
a20d1ceb | 442 | journal_block_tag3_t *tag3, |
c3900875 DW |
443 | void *buf, __u32 sequence) |
444 | { | |
eee06c56 | 445 | __u32 csum32; |
18a6ea1e | 446 | __be32 seq; |
c3900875 | 447 | |
db9ee220 | 448 | if (!jbd2_journal_has_csum_v2or3(j)) |
c3900875 DW |
449 | return 1; |
450 | ||
18a6ea1e DW |
451 | seq = cpu_to_be32(sequence); |
452 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq)); | |
eee06c56 | 453 | csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize); |
c3900875 | 454 | |
56316a0d | 455 | if (jbd2_has_feature_csum3(j)) |
db9ee220 DW |
456 | return tag3->t_checksum == cpu_to_be32(csum32); |
457 | else | |
458 | return tag->t_checksum == cpu_to_be16(csum32); | |
c3900875 DW |
459 | } |
460 | ||
470decc6 DK |
461 | static int do_one_pass(journal_t *journal, |
462 | struct recovery_info *info, enum passtype pass) | |
463 | { | |
464 | unsigned int first_commit_ID, next_commit_ID; | |
465 | unsigned long next_log_block; | |
466 | int err, success = 0; | |
467 | journal_superblock_t * sb; | |
468 | journal_header_t * tmp; | |
469 | struct buffer_head * bh; | |
470 | unsigned int sequence; | |
471 | int blocktype; | |
b517bea1 | 472 | int tag_bytes = journal_tag_bytes(journal); |
818d276c | 473 | __u32 crc32_sum = ~0; /* Transactional Checksums */ |
3caa487f | 474 | int descr_csum_size = 0; |
022eaa75 | 475 | int block_error = 0; |
fc750a3b | 476 | bool need_check_commit_time = false; |
477 | __u64 last_trans_commit_time = 0, commit_time; | |
470decc6 | 478 | |
470decc6 DK |
479 | /* |
480 | * First thing is to establish what we expect to find in the log | |
481 | * (in terms of transaction IDs), and where (in terms of log | |
482 | * block offsets): query the superblock. | |
483 | */ | |
484 | ||
485 | sb = journal->j_superblock; | |
486 | next_commit_ID = be32_to_cpu(sb->s_sequence); | |
487 | next_log_block = be32_to_cpu(sb->s_start); | |
488 | ||
489 | first_commit_ID = next_commit_ID; | |
490 | if (pass == PASS_SCAN) | |
491 | info->start_transaction = first_commit_ID; | |
492 | ||
cb3b3bf2 | 493 | jbd2_debug(1, "Starting recovery pass %d\n", pass); |
470decc6 DK |
494 | |
495 | /* | |
496 | * Now we walk through the log, transaction by transaction, | |
497 | * making sure that each transaction has a commit block in the | |
498 | * expected place. Each complete transaction gets replayed back | |
499 | * into the main filesystem. | |
500 | */ | |
501 | ||
502 | while (1) { | |
503 | int flags; | |
504 | char * tagp; | |
a20d1ceb | 505 | journal_block_tag_t tag; |
470decc6 DK |
506 | struct buffer_head * obh; |
507 | struct buffer_head * nbh; | |
508 | ||
e86e1438 | 509 | cond_resched(); |
470decc6 DK |
510 | |
511 | /* If we already know where to stop the log traversal, | |
512 | * check right now that we haven't gone past the end of | |
513 | * the log. */ | |
514 | ||
515 | if (pass != PASS_SCAN) | |
516 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
517 | break; | |
518 | ||
cb3b3bf2 | 519 | jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", |
5b849b5f HS |
520 | next_commit_ID, next_log_block, |
521 | jbd2_has_feature_fast_commit(journal) ? | |
522 | journal->j_fc_last : journal->j_last); | |
470decc6 DK |
523 | |
524 | /* Skip over each chunk of the transaction looking | |
525 | * either the next descriptor block or the final commit | |
526 | * record. */ | |
527 | ||
cb3b3bf2 | 528 | jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block); |
470decc6 DK |
529 | err = jread(&bh, journal, next_log_block); |
530 | if (err) | |
531 | goto failed; | |
532 | ||
533 | next_log_block++; | |
534 | wrap(journal, next_log_block); | |
535 | ||
536 | /* What kind of buffer is it? | |
537 | * | |
538 | * If it is a descriptor block, check that it has the | |
539 | * expected sequence number. Otherwise, we're all done | |
540 | * here. */ | |
541 | ||
542 | tmp = (journal_header_t *)bh->b_data; | |
543 | ||
f7f4bccb | 544 | if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
545 | brelse(bh); |
546 | break; | |
547 | } | |
548 | ||
549 | blocktype = be32_to_cpu(tmp->h_blocktype); | |
550 | sequence = be32_to_cpu(tmp->h_sequence); | |
cb3b3bf2 | 551 | jbd2_debug(3, "Found magic %d, sequence %d\n", |
470decc6 DK |
552 | blocktype, sequence); |
553 | ||
554 | if (sequence != next_commit_ID) { | |
555 | brelse(bh); | |
556 | break; | |
557 | } | |
558 | ||
559 | /* OK, we have a valid descriptor block which matches | |
560 | * all of the sequence number checks. What are we going | |
561 | * to do with it? That depends on the pass... */ | |
562 | ||
563 | switch(blocktype) { | |
f7f4bccb | 564 | case JBD2_DESCRIPTOR_BLOCK: |
3caa487f | 565 | /* Verify checksum first */ |
db9ee220 | 566 | if (jbd2_journal_has_csum_v2or3(journal)) |
3caa487f DW |
567 | descr_csum_size = |
568 | sizeof(struct jbd2_journal_block_tail); | |
569 | if (descr_csum_size > 0 && | |
1101cd4d JK |
570 | !jbd2_descriptor_block_csum_verify(journal, |
571 | bh->b_data)) { | |
fc750a3b | 572 | /* |
573 | * PASS_SCAN can see stale blocks due to lazy | |
574 | * journal init. Don't error out on those yet. | |
575 | */ | |
576 | if (pass != PASS_SCAN) { | |
577 | pr_err("JBD2: Invalid checksum recovering block %lu in log\n", | |
578 | next_log_block); | |
579 | err = -EFSBADCRC; | |
580 | brelse(bh); | |
581 | goto failed; | |
582 | } | |
583 | need_check_commit_time = true; | |
cb3b3bf2 | 584 | jbd2_debug(1, |
fc750a3b | 585 | "invalid descriptor block found in %lu\n", |
586 | next_log_block); | |
3caa487f DW |
587 | } |
588 | ||
470decc6 | 589 | /* If it is a valid descriptor block, replay it |
818d276c GS |
590 | * in pass REPLAY; if journal_checksums enabled, then |
591 | * calculate checksums in PASS_SCAN, otherwise, | |
592 | * just skip over the blocks it describes. */ | |
470decc6 | 593 | if (pass != PASS_REPLAY) { |
818d276c | 594 | if (pass == PASS_SCAN && |
56316a0d | 595 | jbd2_has_feature_checksum(journal) && |
fc750a3b | 596 | !need_check_commit_time && |
818d276c GS |
597 | !info->end_transaction) { |
598 | if (calc_chksums(journal, bh, | |
599 | &next_log_block, | |
600 | &crc32_sum)) { | |
601 | put_bh(bh); | |
602 | break; | |
603 | } | |
604 | put_bh(bh); | |
605 | continue; | |
606 | } | |
b517bea1 | 607 | next_log_block += count_tags(journal, bh); |
470decc6 | 608 | wrap(journal, next_log_block); |
818d276c | 609 | put_bh(bh); |
470decc6 DK |
610 | continue; |
611 | } | |
612 | ||
613 | /* A descriptor block: we can now write all of | |
614 | * the data blocks. Yay, useful work is finally | |
615 | * getting done here! */ | |
616 | ||
617 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
b517bea1 | 618 | while ((tagp - bh->b_data + tag_bytes) |
3caa487f | 619 | <= journal->j_blocksize - descr_csum_size) { |
470decc6 DK |
620 | unsigned long io_block; |
621 | ||
a20d1ceb TT |
622 | memcpy(&tag, tagp, sizeof(tag)); |
623 | flags = be16_to_cpu(tag.t_flags); | |
470decc6 DK |
624 | |
625 | io_block = next_log_block++; | |
626 | wrap(journal, next_log_block); | |
627 | err = jread(&obh, journal, io_block); | |
628 | if (err) { | |
629 | /* Recover what we can, but | |
630 | * report failure at the end. */ | |
631 | success = err; | |
f2a44523 EG |
632 | printk(KERN_ERR |
633 | "JBD2: IO error %d recovering " | |
470decc6 DK |
634 | "block %ld in log\n", |
635 | err, io_block); | |
636 | } else { | |
18eba7aa | 637 | unsigned long long blocknr; |
470decc6 DK |
638 | |
639 | J_ASSERT(obh != NULL); | |
db9ee220 | 640 | blocknr = read_tag_block(journal, |
a20d1ceb | 641 | &tag); |
470decc6 DK |
642 | |
643 | /* If the block has been | |
644 | * revoked, then we're all done | |
645 | * here. */ | |
f7f4bccb | 646 | if (jbd2_journal_test_revoke |
470decc6 DK |
647 | (journal, blocknr, |
648 | next_commit_ID)) { | |
649 | brelse(obh); | |
650 | ++info->nr_revoke_hits; | |
651 | goto skip_write; | |
652 | } | |
653 | ||
c3900875 DW |
654 | /* Look for block corruption */ |
655 | if (!jbd2_block_tag_csum_verify( | |
a20d1ceb TT |
656 | journal, &tag, (journal_block_tag3_t *)tagp, |
657 | obh->b_data, be32_to_cpu(tmp->h_sequence))) { | |
c3900875 | 658 | brelse(obh); |
6a797d27 | 659 | success = -EFSBADCRC; |
a67c848a | 660 | printk(KERN_ERR "JBD2: Invalid " |
c3900875 | 661 | "checksum recovering " |
ed65b00f TT |
662 | "data block %llu in " |
663 | "log\n", blocknr); | |
022eaa75 DW |
664 | block_error = 1; |
665 | goto skip_write; | |
c3900875 DW |
666 | } |
667 | ||
470decc6 DK |
668 | /* Find a buffer for the new |
669 | * data being restored */ | |
670 | nbh = __getblk(journal->j_fs_dev, | |
671 | blocknr, | |
672 | journal->j_blocksize); | |
673 | if (nbh == NULL) { | |
674 | printk(KERN_ERR | |
f2a44523 | 675 | "JBD2: Out of memory " |
470decc6 DK |
676 | "during recovery.\n"); |
677 | err = -ENOMEM; | |
678 | brelse(bh); | |
679 | brelse(obh); | |
680 | goto failed; | |
681 | } | |
682 | ||
683 | lock_buffer(nbh); | |
684 | memcpy(nbh->b_data, obh->b_data, | |
685 | journal->j_blocksize); | |
f7f4bccb | 686 | if (flags & JBD2_FLAG_ESCAPE) { |
d0025676 | 687 | *((__be32 *)nbh->b_data) = |
f7f4bccb | 688 | cpu_to_be32(JBD2_MAGIC_NUMBER); |
470decc6 DK |
689 | } |
690 | ||
691 | BUFFER_TRACE(nbh, "marking dirty"); | |
692 | set_buffer_uptodate(nbh); | |
693 | mark_buffer_dirty(nbh); | |
694 | BUFFER_TRACE(nbh, "marking uptodate"); | |
695 | ++info->nr_replays; | |
470decc6 DK |
696 | unlock_buffer(nbh); |
697 | brelse(obh); | |
698 | brelse(nbh); | |
699 | } | |
700 | ||
701 | skip_write: | |
b517bea1 | 702 | tagp += tag_bytes; |
f7f4bccb | 703 | if (!(flags & JBD2_FLAG_SAME_UUID)) |
470decc6 DK |
704 | tagp += 16; |
705 | ||
f7f4bccb | 706 | if (flags & JBD2_FLAG_LAST_TAG) |
470decc6 DK |
707 | break; |
708 | } | |
709 | ||
710 | brelse(bh); | |
711 | continue; | |
712 | ||
f7f4bccb | 713 | case JBD2_COMMIT_BLOCK: |
818d276c GS |
714 | /* How to differentiate between interrupted commit |
715 | * and journal corruption ? | |
716 | * | |
717 | * {nth transaction} | |
718 | * Checksum Verification Failed | |
719 | * | | |
720 | * ____________________ | |
721 | * | | | |
722 | * async_commit sync_commit | |
723 | * | | | |
724 | * | GO TO NEXT "Journal Corruption" | |
725 | * | TRANSACTION | |
726 | * | | |
727 | * {(n+1)th transanction} | |
728 | * | | |
729 | * _______|______________ | |
730 | * | | | |
731 | * Commit block found Commit block not found | |
732 | * | | | |
733 | * "Journal Corruption" | | |
734 | * _____________|_________ | |
735 | * | | | |
736 | * nth trans corrupt OR nth trans | |
737 | * and (n+1)th interrupted interrupted | |
738 | * before commit block | |
739 | * could reach the disk. | |
740 | * (Cannot find the difference in above | |
741 | * mentioned conditions. Hence assume | |
742 | * "Interrupted Commit".) | |
743 | */ | |
fc750a3b | 744 | commit_time = be64_to_cpu( |
745 | ((struct commit_header *)bh->b_data)->h_commit_sec); | |
746 | /* | |
747 | * If need_check_commit_time is set, it means we are in | |
748 | * PASS_SCAN and csum verify failed before. If | |
749 | * commit_time is increasing, it's the same journal, | |
750 | * otherwise it is stale journal block, just end this | |
751 | * recovery. | |
752 | */ | |
753 | if (need_check_commit_time) { | |
754 | if (commit_time >= last_trans_commit_time) { | |
755 | pr_err("JBD2: Invalid checksum found in transaction %u\n", | |
756 | next_commit_ID); | |
757 | err = -EFSBADCRC; | |
758 | brelse(bh); | |
759 | goto failed; | |
760 | } | |
761 | ignore_crc_mismatch: | |
762 | /* | |
763 | * It likely does not belong to same journal, | |
764 | * just end this recovery with success. | |
765 | */ | |
cb3b3bf2 | 766 | jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n", |
fc750a3b | 767 | next_commit_ID); |
fc750a3b | 768 | brelse(bh); |
769 | goto done; | |
770 | } | |
818d276c | 771 | |
fc750a3b | 772 | /* |
773 | * Found an expected commit block: if checksums | |
774 | * are present, verify them in PASS_SCAN; else not | |
818d276c | 775 | * much to do other than move on to the next sequence |
fc750a3b | 776 | * number. |
777 | */ | |
818d276c | 778 | if (pass == PASS_SCAN && |
56316a0d | 779 | jbd2_has_feature_checksum(journal)) { |
818d276c GS |
780 | struct commit_header *cbh = |
781 | (struct commit_header *)bh->b_data; | |
782 | unsigned found_chksum = | |
783 | be32_to_cpu(cbh->h_chksum[0]); | |
784 | ||
818d276c | 785 | if (info->end_transaction) { |
624080ed TT |
786 | journal->j_failed_commit = |
787 | info->end_transaction; | |
818d276c GS |
788 | brelse(bh); |
789 | break; | |
790 | } | |
791 | ||
00a3fff0 SL |
792 | /* Neither checksum match nor unused? */ |
793 | if (!((crc32_sum == found_chksum && | |
794 | cbh->h_chksum_type == | |
795 | JBD2_CRC32_CHKSUM && | |
796 | cbh->h_chksum_size == | |
797 | JBD2_CRC32_CHKSUM_SIZE) || | |
798 | (cbh->h_chksum_type == 0 && | |
799 | cbh->h_chksum_size == 0 && | |
800 | found_chksum == 0))) | |
801 | goto chksum_error; | |
802 | ||
818d276c GS |
803 | crc32_sum = ~0; |
804 | } | |
1f56c589 DW |
805 | if (pass == PASS_SCAN && |
806 | !jbd2_commit_block_csum_verify(journal, | |
807 | bh->b_data)) { | |
00a3fff0 | 808 | chksum_error: |
fc750a3b | 809 | if (commit_time < last_trans_commit_time) |
810 | goto ignore_crc_mismatch; | |
1f56c589 DW |
811 | info->end_transaction = next_commit_ID; |
812 | ||
56316a0d | 813 | if (!jbd2_has_feature_async_commit(journal)) { |
1f56c589 DW |
814 | journal->j_failed_commit = |
815 | next_commit_ID; | |
816 | brelse(bh); | |
817 | break; | |
818 | } | |
819 | } | |
fc750a3b | 820 | if (pass == PASS_SCAN) |
821 | last_trans_commit_time = commit_time; | |
470decc6 DK |
822 | brelse(bh); |
823 | next_commit_ID++; | |
824 | continue; | |
825 | ||
f7f4bccb | 826 | case JBD2_REVOKE_BLOCK: |
fc750a3b | 827 | /* |
828 | * Check revoke block crc in pass_scan, if csum verify | |
829 | * failed, check commit block time later. | |
830 | */ | |
831 | if (pass == PASS_SCAN && | |
832 | !jbd2_descriptor_block_csum_verify(journal, | |
833 | bh->b_data)) { | |
cb3b3bf2 | 834 | jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n", |
fc750a3b | 835 | next_log_block); |
836 | need_check_commit_time = true; | |
837 | } | |
470decc6 DK |
838 | /* If we aren't in the REVOKE pass, then we can |
839 | * just skip over this block. */ | |
840 | if (pass != PASS_REVOKE) { | |
841 | brelse(bh); | |
842 | continue; | |
843 | } | |
844 | ||
845 | err = scan_revoke_records(journal, bh, | |
846 | next_commit_ID, info); | |
847 | brelse(bh); | |
848 | if (err) | |
849 | goto failed; | |
850 | continue; | |
851 | ||
852 | default: | |
cb3b3bf2 | 853 | jbd2_debug(3, "Unrecognised magic %d, end of scan.\n", |
470decc6 DK |
854 | blocktype); |
855 | brelse(bh); | |
856 | goto done; | |
857 | } | |
858 | } | |
859 | ||
860 | done: | |
861 | /* | |
862 | * We broke out of the log scan loop: either we came to the | |
863 | * known end of the log or we found an unexpected block in the | |
864 | * log. If the latter happened, then we know that the "current" | |
865 | * transaction marks the end of the valid log. | |
866 | */ | |
867 | ||
818d276c GS |
868 | if (pass == PASS_SCAN) { |
869 | if (!info->end_transaction) | |
870 | info->end_transaction = next_commit_ID; | |
871 | } else { | |
470decc6 DK |
872 | /* It's really bad news if different passes end up at |
873 | * different places (but possible due to IO errors). */ | |
874 | if (info->end_transaction != next_commit_ID) { | |
f2a44523 | 875 | printk(KERN_ERR "JBD2: recovery pass %d ended at " |
470decc6 DK |
876 | "transaction %u, expected %u\n", |
877 | pass, next_commit_ID, info->end_transaction); | |
878 | if (!success) | |
879 | success = -EIO; | |
880 | } | |
881 | } | |
5b849b5f HS |
882 | |
883 | if (jbd2_has_feature_fast_commit(journal) && pass != PASS_REVOKE) { | |
884 | err = fc_do_one_pass(journal, info, pass); | |
885 | if (err) | |
886 | success = err; | |
887 | } | |
888 | ||
022eaa75 DW |
889 | if (block_error && success == 0) |
890 | success = -EIO; | |
470decc6 DK |
891 | return success; |
892 | ||
893 | failed: | |
894 | return err; | |
895 | } | |
896 | ||
470decc6 DK |
897 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ |
898 | ||
899 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, | |
900 | tid_t sequence, struct recovery_info *info) | |
901 | { | |
f7f4bccb | 902 | jbd2_journal_revoke_header_t *header; |
470decc6 | 903 | int offset, max; |
4009cc7a | 904 | unsigned csum_size = 0; |
e531d0bc | 905 | __u32 rcount; |
b517bea1 | 906 | int record_len = 4; |
470decc6 | 907 | |
f7f4bccb MC |
908 | header = (jbd2_journal_revoke_header_t *) bh->b_data; |
909 | offset = sizeof(jbd2_journal_revoke_header_t); | |
e531d0bc | 910 | rcount = be32_to_cpu(header->r_count); |
470decc6 | 911 | |
e531d0bc | 912 | if (jbd2_journal_has_csum_v2or3(journal)) |
1101cd4d | 913 | csum_size = sizeof(struct jbd2_journal_block_tail); |
e531d0bc DW |
914 | if (rcount > journal->j_blocksize - csum_size) |
915 | return -EINVAL; | |
916 | max = rcount; | |
917 | ||
56316a0d | 918 | if (jbd2_has_feature_64bit(journal)) |
b517bea1 ZB |
919 | record_len = 8; |
920 | ||
921 | while (offset + record_len <= max) { | |
18eba7aa | 922 | unsigned long long blocknr; |
470decc6 DK |
923 | int err; |
924 | ||
b517bea1 ZB |
925 | if (record_len == 4) |
926 | blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset))); | |
927 | else | |
928 | blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset))); | |
929 | offset += record_len; | |
f7f4bccb | 930 | err = jbd2_journal_set_revoke(journal, blocknr, sequence); |
470decc6 DK |
931 | if (err) |
932 | return err; | |
933 | ++info->nr_revokes; | |
934 | } | |
935 | return 0; | |
936 | } |