Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/recovery.c |
470decc6 DK |
3 | * |
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> | |
f7f4bccb | 21 | #include <linux/jbd2.h> |
470decc6 | 22 | #include <linux/errno.h> |
818d276c | 23 | #include <linux/crc32.h> |
79feb521 | 24 | #include <linux/blkdev.h> |
470decc6 DK |
25 | #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 | */ | |
31 | struct 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 | ||
41 | enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY}; | |
42 | static int do_one_pass(journal_t *journal, | |
43 | struct recovery_info *info, enum passtype pass); | |
44 | static 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 */ | |
50 | static 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 | |
70 | static int do_readahead(journal_t *journal, unsigned int start) | |
71 | { | |
72 | int err; | |
73 | unsigned int max, nbufs, next; | |
18eba7aa | 74 | unsigned long long blocknr; |
470decc6 DK |
75 | 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++) { | |
f7f4bccb | 90 | err = jbd2_journal_bmap(journal, next, &blocknr); |
470decc6 DK |
91 | |
92 | if (err) { | |
f2a44523 | 93 | printk(KERN_ERR "JBD2: bad block at offset %u\n", |
470decc6 DK |
94 | 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 | ||
119 | failed: | |
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 | ||
132 | static int jread(struct buffer_head **bhp, journal_t *journal, | |
133 | unsigned int offset) | |
134 | { | |
135 | int err; | |
18eba7aa | 136 | unsigned long long blocknr; |
470decc6 DK |
137 | struct buffer_head *bh; |
138 | ||
139 | *bhp = NULL; | |
140 | ||
141 | if (offset >= journal->j_maxlen) { | |
f2a44523 | 142 | printk(KERN_ERR "JBD2: corrupted journal superblock\n"); |
6a797d27 | 143 | return -EFSCORRUPTED; |
470decc6 DK |
144 | } |
145 | ||
f7f4bccb | 146 | err = jbd2_journal_bmap(journal, offset, &blocknr); |
470decc6 DK |
147 | |
148 | if (err) { | |
f2a44523 | 149 | printk(KERN_ERR "JBD2: bad block at offset %u\n", |
470decc6 DK |
150 | 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)) { | |
f2a44523 | 167 | printk(KERN_ERR "JBD2: Failed to read block at offset %u\n", |
470decc6 DK |
168 | offset); |
169 | brelse(bh); | |
170 | return -EIO; | |
171 | } | |
172 | ||
173 | *bhp = bh; | |
174 | return 0; | |
175 | } | |
176 | ||
3caa487f DW |
177 | static int jbd2_descr_block_csum_verify(journal_t *j, |
178 | void *buf) | |
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 | ||
187 | tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize - | |
188 | sizeof(struct jbd2_journal_block_tail)); | |
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; | |
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) { |
470decc6 DK |
214 | tag = (journal_block_tag_t *) tagp; |
215 | ||
216 | nr++; | |
b517bea1 | 217 | tagp += tag_bytes; |
8f888ef8 | 218 | if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID))) |
470decc6 DK |
219 | tagp += 16; |
220 | ||
8f888ef8 | 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 { \ | |
232 | if (var >= (journal)->j_last) \ | |
233 | var -= ((journal)->j_last - (journal)->j_first); \ | |
234 | } while (0) | |
235 | ||
236 | /** | |
f7f4bccb | 237 | * jbd2_journal_recover - recovers a on-disk journal |
470decc6 DK |
238 | * @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 | */ | |
f7f4bccb | 248 | int jbd2_journal_recover(journal_t *journal) |
470decc6 | 249 | { |
44519faf | 250 | int err, err2; |
470decc6 DK |
251 | 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 | ||
f2a44523 | 277 | jbd_debug(1, "JBD2: recovery, exit status %d, " |
470decc6 DK |
278 | "recovered transactions %u to %u\n", |
279 | err, info.start_transaction, info.end_transaction); | |
f2a44523 | 280 | jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n", |
470decc6 DK |
281 | 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 | ||
f7f4bccb | 287 | jbd2_journal_clear_revoke(journal); |
44519faf HK |
288 | err2 = sync_blockdev(journal->j_fs_dev); |
289 | if (!err) | |
290 | err = err2; | |
79feb521 | 291 | /* Make sure all replayed data is on permanent storage */ |
316e4cfd TT |
292 | if (journal->j_flags & JBD2_BARRIER) { |
293 | err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); | |
294 | if (!err) | |
295 | err = err2; | |
296 | } | |
470decc6 DK |
297 | return err; |
298 | } | |
299 | ||
300 | /** | |
f7f4bccb | 301 | * jbd2_journal_skip_recovery - Start journal and wipe exiting records |
470decc6 DK |
302 | * @journal: journal to startup |
303 | * | |
304 | * Locate any valid recovery information from the journal and set up the | |
305 | * journal structures in memory to ignore it (presumably because the | |
306 | * caller has evidence that it is out of date). | |
307 | * This function does'nt appear to be exorted.. | |
308 | * | |
309 | * We perform one pass over the journal to allow us to tell the user how | |
310 | * much recovery information is being erased, and to let us initialise | |
311 | * the journal transaction sequence numbers to the next unused ID. | |
312 | */ | |
f7f4bccb | 313 | int jbd2_journal_skip_recovery(journal_t *journal) |
470decc6 DK |
314 | { |
315 | int err; | |
470decc6 DK |
316 | |
317 | struct recovery_info info; | |
318 | ||
319 | memset (&info, 0, sizeof(info)); | |
470decc6 DK |
320 | |
321 | err = do_one_pass(journal, &info, PASS_SCAN); | |
322 | ||
323 | if (err) { | |
f2a44523 | 324 | printk(KERN_ERR "JBD2: error %d scanning journal\n", err); |
470decc6 DK |
325 | ++journal->j_transaction_sequence; |
326 | } else { | |
e23291b9 | 327 | #ifdef CONFIG_JBD2_DEBUG |
5a0790c2 AK |
328 | int dropped = info.end_transaction - |
329 | be32_to_cpu(journal->j_superblock->s_sequence); | |
b38bd33a | 330 | jbd_debug(1, |
f2a44523 | 331 | "JBD2: ignoring %d transaction%s from the journal.\n", |
470decc6 | 332 | dropped, (dropped == 1) ? "" : "s"); |
9a4f6271 | 333 | #endif |
470decc6 DK |
334 | journal->j_transaction_sequence = ++info.end_transaction; |
335 | } | |
336 | ||
337 | journal->j_tail = 0; | |
338 | return err; | |
339 | } | |
340 | ||
db9ee220 DW |
341 | static inline unsigned long long read_tag_block(journal_t *journal, |
342 | journal_block_tag_t *tag) | |
b517bea1 | 343 | { |
18eba7aa | 344 | unsigned long long block = be32_to_cpu(tag->t_blocknr); |
56316a0d | 345 | if (jbd2_has_feature_64bit(journal)) |
b517bea1 ZB |
346 | block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32; |
347 | return block; | |
348 | } | |
349 | ||
818d276c GS |
350 | /* |
351 | * calc_chksums calculates the checksums for the blocks described in the | |
352 | * descriptor block. | |
353 | */ | |
354 | static int calc_chksums(journal_t *journal, struct buffer_head *bh, | |
355 | unsigned long *next_log_block, __u32 *crc32_sum) | |
356 | { | |
357 | int i, num_blks, err; | |
358 | unsigned long io_block; | |
359 | struct buffer_head *obh; | |
360 | ||
361 | num_blks = count_tags(journal, bh); | |
362 | /* Calculate checksum of the descriptor block. */ | |
363 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); | |
364 | ||
365 | for (i = 0; i < num_blks; i++) { | |
366 | io_block = (*next_log_block)++; | |
367 | wrap(journal, *next_log_block); | |
368 | err = jread(&obh, journal, io_block); | |
369 | if (err) { | |
f2a44523 | 370 | printk(KERN_ERR "JBD2: IO error %d recovering block " |
818d276c GS |
371 | "%lu in log\n", err, io_block); |
372 | return 1; | |
373 | } else { | |
374 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, | |
375 | obh->b_size); | |
376 | } | |
8ea76900 | 377 | put_bh(obh); |
818d276c GS |
378 | } |
379 | return 0; | |
380 | } | |
381 | ||
1f56c589 DW |
382 | static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) |
383 | { | |
384 | struct commit_header *h; | |
18a6ea1e DW |
385 | __be32 provided; |
386 | __u32 calculated; | |
1f56c589 | 387 | |
db9ee220 | 388 | if (!jbd2_journal_has_csum_v2or3(j)) |
1f56c589 DW |
389 | return 1; |
390 | ||
391 | h = buf; | |
392 | provided = h->h_chksum[0]; | |
393 | h->h_chksum[0] = 0; | |
394 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
395 | h->h_chksum[0] = provided; | |
396 | ||
18a6ea1e | 397 | return provided == cpu_to_be32(calculated); |
1f56c589 DW |
398 | } |
399 | ||
c3900875 DW |
400 | static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, |
401 | void *buf, __u32 sequence) | |
402 | { | |
db9ee220 | 403 | journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag; |
eee06c56 | 404 | __u32 csum32; |
18a6ea1e | 405 | __be32 seq; |
c3900875 | 406 | |
db9ee220 | 407 | if (!jbd2_journal_has_csum_v2or3(j)) |
c3900875 DW |
408 | return 1; |
409 | ||
18a6ea1e DW |
410 | seq = cpu_to_be32(sequence); |
411 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq)); | |
eee06c56 | 412 | csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize); |
c3900875 | 413 | |
56316a0d | 414 | if (jbd2_has_feature_csum3(j)) |
db9ee220 DW |
415 | return tag3->t_checksum == cpu_to_be32(csum32); |
416 | else | |
417 | return tag->t_checksum == cpu_to_be16(csum32); | |
c3900875 DW |
418 | } |
419 | ||
470decc6 DK |
420 | static int do_one_pass(journal_t *journal, |
421 | struct recovery_info *info, enum passtype pass) | |
422 | { | |
423 | unsigned int first_commit_ID, next_commit_ID; | |
424 | unsigned long next_log_block; | |
425 | int err, success = 0; | |
426 | journal_superblock_t * sb; | |
427 | journal_header_t * tmp; | |
428 | struct buffer_head * bh; | |
429 | unsigned int sequence; | |
430 | int blocktype; | |
b517bea1 | 431 | int tag_bytes = journal_tag_bytes(journal); |
818d276c | 432 | __u32 crc32_sum = ~0; /* Transactional Checksums */ |
3caa487f | 433 | int descr_csum_size = 0; |
022eaa75 | 434 | int block_error = 0; |
470decc6 | 435 | |
470decc6 DK |
436 | /* |
437 | * First thing is to establish what we expect to find in the log | |
438 | * (in terms of transaction IDs), and where (in terms of log | |
439 | * block offsets): query the superblock. | |
440 | */ | |
441 | ||
442 | sb = journal->j_superblock; | |
443 | next_commit_ID = be32_to_cpu(sb->s_sequence); | |
444 | next_log_block = be32_to_cpu(sb->s_start); | |
445 | ||
446 | first_commit_ID = next_commit_ID; | |
447 | if (pass == PASS_SCAN) | |
448 | info->start_transaction = first_commit_ID; | |
449 | ||
450 | jbd_debug(1, "Starting recovery pass %d\n", pass); | |
451 | ||
452 | /* | |
453 | * Now we walk through the log, transaction by transaction, | |
454 | * making sure that each transaction has a commit block in the | |
455 | * expected place. Each complete transaction gets replayed back | |
456 | * into the main filesystem. | |
457 | */ | |
458 | ||
459 | while (1) { | |
460 | int flags; | |
461 | char * tagp; | |
462 | journal_block_tag_t * tag; | |
463 | struct buffer_head * obh; | |
464 | struct buffer_head * nbh; | |
465 | ||
e86e1438 | 466 | cond_resched(); |
470decc6 DK |
467 | |
468 | /* If we already know where to stop the log traversal, | |
469 | * check right now that we haven't gone past the end of | |
470 | * the log. */ | |
471 | ||
472 | if (pass != PASS_SCAN) | |
473 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
474 | break; | |
475 | ||
476 | jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", | |
477 | next_commit_ID, next_log_block, journal->j_last); | |
478 | ||
479 | /* Skip over each chunk of the transaction looking | |
480 | * either the next descriptor block or the final commit | |
481 | * record. */ | |
482 | ||
f2a44523 | 483 | jbd_debug(3, "JBD2: checking block %ld\n", next_log_block); |
470decc6 DK |
484 | err = jread(&bh, journal, next_log_block); |
485 | if (err) | |
486 | goto failed; | |
487 | ||
488 | next_log_block++; | |
489 | wrap(journal, next_log_block); | |
490 | ||
491 | /* What kind of buffer is it? | |
492 | * | |
493 | * If it is a descriptor block, check that it has the | |
494 | * expected sequence number. Otherwise, we're all done | |
495 | * here. */ | |
496 | ||
497 | tmp = (journal_header_t *)bh->b_data; | |
498 | ||
f7f4bccb | 499 | if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
500 | brelse(bh); |
501 | break; | |
502 | } | |
503 | ||
504 | blocktype = be32_to_cpu(tmp->h_blocktype); | |
505 | sequence = be32_to_cpu(tmp->h_sequence); | |
506 | jbd_debug(3, "Found magic %d, sequence %d\n", | |
507 | blocktype, sequence); | |
508 | ||
509 | if (sequence != next_commit_ID) { | |
510 | brelse(bh); | |
511 | break; | |
512 | } | |
513 | ||
514 | /* OK, we have a valid descriptor block which matches | |
515 | * all of the sequence number checks. What are we going | |
516 | * to do with it? That depends on the pass... */ | |
517 | ||
518 | switch(blocktype) { | |
f7f4bccb | 519 | case JBD2_DESCRIPTOR_BLOCK: |
3caa487f | 520 | /* Verify checksum first */ |
db9ee220 | 521 | if (jbd2_journal_has_csum_v2or3(journal)) |
3caa487f DW |
522 | descr_csum_size = |
523 | sizeof(struct jbd2_journal_block_tail); | |
524 | if (descr_csum_size > 0 && | |
525 | !jbd2_descr_block_csum_verify(journal, | |
526 | bh->b_data)) { | |
b6924225 DW |
527 | printk(KERN_ERR "JBD2: Invalid checksum " |
528 | "recovering block %lu in log\n", | |
529 | next_log_block); | |
6a797d27 | 530 | err = -EFSBADCRC; |
064d8389 | 531 | brelse(bh); |
3caa487f DW |
532 | goto failed; |
533 | } | |
534 | ||
470decc6 | 535 | /* If it is a valid descriptor block, replay it |
818d276c GS |
536 | * in pass REPLAY; if journal_checksums enabled, then |
537 | * calculate checksums in PASS_SCAN, otherwise, | |
538 | * just skip over the blocks it describes. */ | |
470decc6 | 539 | if (pass != PASS_REPLAY) { |
818d276c | 540 | if (pass == PASS_SCAN && |
56316a0d | 541 | jbd2_has_feature_checksum(journal) && |
818d276c GS |
542 | !info->end_transaction) { |
543 | if (calc_chksums(journal, bh, | |
544 | &next_log_block, | |
545 | &crc32_sum)) { | |
546 | put_bh(bh); | |
547 | break; | |
548 | } | |
549 | put_bh(bh); | |
550 | continue; | |
551 | } | |
b517bea1 | 552 | next_log_block += count_tags(journal, bh); |
470decc6 | 553 | wrap(journal, next_log_block); |
818d276c | 554 | put_bh(bh); |
470decc6 DK |
555 | continue; |
556 | } | |
557 | ||
558 | /* A descriptor block: we can now write all of | |
559 | * the data blocks. Yay, useful work is finally | |
560 | * getting done here! */ | |
561 | ||
562 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
b517bea1 | 563 | while ((tagp - bh->b_data + tag_bytes) |
3caa487f | 564 | <= journal->j_blocksize - descr_csum_size) { |
470decc6 DK |
565 | unsigned long io_block; |
566 | ||
567 | tag = (journal_block_tag_t *) tagp; | |
8f888ef8 | 568 | flags = be16_to_cpu(tag->t_flags); |
470decc6 DK |
569 | |
570 | io_block = next_log_block++; | |
571 | wrap(journal, next_log_block); | |
572 | err = jread(&obh, journal, io_block); | |
573 | if (err) { | |
574 | /* Recover what we can, but | |
575 | * report failure at the end. */ | |
576 | success = err; | |
f2a44523 EG |
577 | printk(KERN_ERR |
578 | "JBD2: IO error %d recovering " | |
470decc6 DK |
579 | "block %ld in log\n", |
580 | err, io_block); | |
581 | } else { | |
18eba7aa | 582 | unsigned long long blocknr; |
470decc6 DK |
583 | |
584 | J_ASSERT(obh != NULL); | |
db9ee220 | 585 | blocknr = read_tag_block(journal, |
b517bea1 | 586 | tag); |
470decc6 DK |
587 | |
588 | /* If the block has been | |
589 | * revoked, then we're all done | |
590 | * here. */ | |
f7f4bccb | 591 | if (jbd2_journal_test_revoke |
470decc6 DK |
592 | (journal, blocknr, |
593 | next_commit_ID)) { | |
594 | brelse(obh); | |
595 | ++info->nr_revoke_hits; | |
596 | goto skip_write; | |
597 | } | |
598 | ||
c3900875 DW |
599 | /* Look for block corruption */ |
600 | if (!jbd2_block_tag_csum_verify( | |
601 | journal, tag, obh->b_data, | |
602 | be32_to_cpu(tmp->h_sequence))) { | |
603 | brelse(obh); | |
6a797d27 | 604 | success = -EFSBADCRC; |
a67c848a | 605 | printk(KERN_ERR "JBD2: Invalid " |
c3900875 DW |
606 | "checksum recovering " |
607 | "block %llu in log\n", | |
608 | blocknr); | |
022eaa75 DW |
609 | block_error = 1; |
610 | goto skip_write; | |
c3900875 DW |
611 | } |
612 | ||
470decc6 DK |
613 | /* Find a buffer for the new |
614 | * data being restored */ | |
615 | nbh = __getblk(journal->j_fs_dev, | |
616 | blocknr, | |
617 | journal->j_blocksize); | |
618 | if (nbh == NULL) { | |
619 | printk(KERN_ERR | |
f2a44523 | 620 | "JBD2: Out of memory " |
470decc6 DK |
621 | "during recovery.\n"); |
622 | err = -ENOMEM; | |
623 | brelse(bh); | |
624 | brelse(obh); | |
625 | goto failed; | |
626 | } | |
627 | ||
628 | lock_buffer(nbh); | |
629 | memcpy(nbh->b_data, obh->b_data, | |
630 | journal->j_blocksize); | |
f7f4bccb | 631 | if (flags & JBD2_FLAG_ESCAPE) { |
d0025676 | 632 | *((__be32 *)nbh->b_data) = |
f7f4bccb | 633 | cpu_to_be32(JBD2_MAGIC_NUMBER); |
470decc6 DK |
634 | } |
635 | ||
636 | BUFFER_TRACE(nbh, "marking dirty"); | |
637 | set_buffer_uptodate(nbh); | |
638 | mark_buffer_dirty(nbh); | |
639 | BUFFER_TRACE(nbh, "marking uptodate"); | |
640 | ++info->nr_replays; | |
641 | /* ll_rw_block(WRITE, 1, &nbh); */ | |
642 | unlock_buffer(nbh); | |
643 | brelse(obh); | |
644 | brelse(nbh); | |
645 | } | |
646 | ||
647 | skip_write: | |
b517bea1 | 648 | tagp += tag_bytes; |
f7f4bccb | 649 | if (!(flags & JBD2_FLAG_SAME_UUID)) |
470decc6 DK |
650 | tagp += 16; |
651 | ||
f7f4bccb | 652 | if (flags & JBD2_FLAG_LAST_TAG) |
470decc6 DK |
653 | break; |
654 | } | |
655 | ||
656 | brelse(bh); | |
657 | continue; | |
658 | ||
f7f4bccb | 659 | case JBD2_COMMIT_BLOCK: |
818d276c GS |
660 | /* How to differentiate between interrupted commit |
661 | * and journal corruption ? | |
662 | * | |
663 | * {nth transaction} | |
664 | * Checksum Verification Failed | |
665 | * | | |
666 | * ____________________ | |
667 | * | | | |
668 | * async_commit sync_commit | |
669 | * | | | |
670 | * | GO TO NEXT "Journal Corruption" | |
671 | * | TRANSACTION | |
672 | * | | |
673 | * {(n+1)th transanction} | |
674 | * | | |
675 | * _______|______________ | |
676 | * | | | |
677 | * Commit block found Commit block not found | |
678 | * | | | |
679 | * "Journal Corruption" | | |
680 | * _____________|_________ | |
681 | * | | | |
682 | * nth trans corrupt OR nth trans | |
683 | * and (n+1)th interrupted interrupted | |
684 | * before commit block | |
685 | * could reach the disk. | |
686 | * (Cannot find the difference in above | |
687 | * mentioned conditions. Hence assume | |
688 | * "Interrupted Commit".) | |
689 | */ | |
690 | ||
691 | /* Found an expected commit block: if checksums | |
692 | * are present verify them in PASS_SCAN; else not | |
693 | * much to do other than move on to the next sequence | |
470decc6 | 694 | * number. */ |
818d276c | 695 | if (pass == PASS_SCAN && |
56316a0d | 696 | jbd2_has_feature_checksum(journal)) { |
818d276c GS |
697 | int chksum_err, chksum_seen; |
698 | struct commit_header *cbh = | |
699 | (struct commit_header *)bh->b_data; | |
700 | unsigned found_chksum = | |
701 | be32_to_cpu(cbh->h_chksum[0]); | |
702 | ||
703 | chksum_err = chksum_seen = 0; | |
704 | ||
705 | if (info->end_transaction) { | |
624080ed TT |
706 | journal->j_failed_commit = |
707 | info->end_transaction; | |
818d276c GS |
708 | brelse(bh); |
709 | break; | |
710 | } | |
711 | ||
712 | if (crc32_sum == found_chksum && | |
713 | cbh->h_chksum_type == JBD2_CRC32_CHKSUM && | |
714 | cbh->h_chksum_size == | |
715 | JBD2_CRC32_CHKSUM_SIZE) | |
716 | chksum_seen = 1; | |
717 | else if (!(cbh->h_chksum_type == 0 && | |
718 | cbh->h_chksum_size == 0 && | |
719 | found_chksum == 0 && | |
720 | !chksum_seen)) | |
721 | /* | |
722 | * If fs is mounted using an old kernel and then | |
723 | * kernel with journal_chksum is used then we | |
724 | * get a situation where the journal flag has | |
725 | * checksum flag set but checksums are not | |
726 | * present i.e chksum = 0, in the individual | |
727 | * commit blocks. | |
728 | * Hence to avoid checksum failures, in this | |
729 | * situation, this extra check is added. | |
730 | */ | |
731 | chksum_err = 1; | |
732 | ||
733 | if (chksum_err) { | |
734 | info->end_transaction = next_commit_ID; | |
735 | ||
56316a0d | 736 | if (!jbd2_has_feature_async_commit(journal)) { |
624080ed TT |
737 | journal->j_failed_commit = |
738 | next_commit_ID; | |
818d276c GS |
739 | brelse(bh); |
740 | break; | |
741 | } | |
742 | } | |
743 | crc32_sum = ~0; | |
744 | } | |
1f56c589 DW |
745 | if (pass == PASS_SCAN && |
746 | !jbd2_commit_block_csum_verify(journal, | |
747 | bh->b_data)) { | |
748 | info->end_transaction = next_commit_ID; | |
749 | ||
56316a0d | 750 | if (!jbd2_has_feature_async_commit(journal)) { |
1f56c589 DW |
751 | journal->j_failed_commit = |
752 | next_commit_ID; | |
753 | brelse(bh); | |
754 | break; | |
755 | } | |
756 | } | |
470decc6 DK |
757 | brelse(bh); |
758 | next_commit_ID++; | |
759 | continue; | |
760 | ||
f7f4bccb | 761 | case JBD2_REVOKE_BLOCK: |
470decc6 DK |
762 | /* If we aren't in the REVOKE pass, then we can |
763 | * just skip over this block. */ | |
764 | if (pass != PASS_REVOKE) { | |
765 | brelse(bh); | |
766 | continue; | |
767 | } | |
768 | ||
769 | err = scan_revoke_records(journal, bh, | |
770 | next_commit_ID, info); | |
771 | brelse(bh); | |
772 | if (err) | |
773 | goto failed; | |
774 | continue; | |
775 | ||
776 | default: | |
777 | jbd_debug(3, "Unrecognised magic %d, end of scan.\n", | |
778 | blocktype); | |
779 | brelse(bh); | |
780 | goto done; | |
781 | } | |
782 | } | |
783 | ||
784 | done: | |
785 | /* | |
786 | * We broke out of the log scan loop: either we came to the | |
787 | * known end of the log or we found an unexpected block in the | |
788 | * log. If the latter happened, then we know that the "current" | |
789 | * transaction marks the end of the valid log. | |
790 | */ | |
791 | ||
818d276c GS |
792 | if (pass == PASS_SCAN) { |
793 | if (!info->end_transaction) | |
794 | info->end_transaction = next_commit_ID; | |
795 | } else { | |
470decc6 DK |
796 | /* It's really bad news if different passes end up at |
797 | * different places (but possible due to IO errors). */ | |
798 | if (info->end_transaction != next_commit_ID) { | |
f2a44523 | 799 | printk(KERN_ERR "JBD2: recovery pass %d ended at " |
470decc6 DK |
800 | "transaction %u, expected %u\n", |
801 | pass, next_commit_ID, info->end_transaction); | |
802 | if (!success) | |
803 | success = -EIO; | |
804 | } | |
805 | } | |
022eaa75 DW |
806 | if (block_error && success == 0) |
807 | success = -EIO; | |
470decc6 DK |
808 | return success; |
809 | ||
810 | failed: | |
811 | return err; | |
812 | } | |
813 | ||
42a7106d DW |
814 | static int jbd2_revoke_block_csum_verify(journal_t *j, |
815 | void *buf) | |
816 | { | |
817 | struct jbd2_journal_revoke_tail *tail; | |
18a6ea1e DW |
818 | __be32 provided; |
819 | __u32 calculated; | |
42a7106d | 820 | |
db9ee220 | 821 | if (!jbd2_journal_has_csum_v2or3(j)) |
42a7106d DW |
822 | return 1; |
823 | ||
824 | tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize - | |
825 | sizeof(struct jbd2_journal_revoke_tail)); | |
826 | provided = tail->r_checksum; | |
827 | tail->r_checksum = 0; | |
828 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
829 | tail->r_checksum = provided; | |
830 | ||
18a6ea1e | 831 | return provided == cpu_to_be32(calculated); |
42a7106d | 832 | } |
470decc6 DK |
833 | |
834 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ | |
835 | ||
836 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, | |
837 | tid_t sequence, struct recovery_info *info) | |
838 | { | |
f7f4bccb | 839 | jbd2_journal_revoke_header_t *header; |
470decc6 | 840 | int offset, max; |
e531d0bc DW |
841 | int csum_size = 0; |
842 | __u32 rcount; | |
b517bea1 | 843 | int record_len = 4; |
470decc6 | 844 | |
f7f4bccb MC |
845 | header = (jbd2_journal_revoke_header_t *) bh->b_data; |
846 | offset = sizeof(jbd2_journal_revoke_header_t); | |
e531d0bc | 847 | rcount = be32_to_cpu(header->r_count); |
470decc6 | 848 | |
42a7106d | 849 | if (!jbd2_revoke_block_csum_verify(journal, header)) |
6a797d27 | 850 | return -EFSBADCRC; |
42a7106d | 851 | |
e531d0bc DW |
852 | if (jbd2_journal_has_csum_v2or3(journal)) |
853 | csum_size = sizeof(struct jbd2_journal_revoke_tail); | |
854 | if (rcount > journal->j_blocksize - csum_size) | |
855 | return -EINVAL; | |
856 | max = rcount; | |
857 | ||
56316a0d | 858 | if (jbd2_has_feature_64bit(journal)) |
b517bea1 ZB |
859 | record_len = 8; |
860 | ||
861 | while (offset + record_len <= max) { | |
18eba7aa | 862 | unsigned long long blocknr; |
470decc6 DK |
863 | int err; |
864 | ||
b517bea1 ZB |
865 | if (record_len == 4) |
866 | blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset))); | |
867 | else | |
868 | blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset))); | |
869 | offset += record_len; | |
f7f4bccb | 870 | err = jbd2_journal_set_revoke(journal, blocknr, sequence); |
470decc6 DK |
871 | if (err) |
872 | return err; | |
873 | ++info->nr_revokes; | |
874 | } | |
875 | return 0; | |
876 | } |