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"); |
470decc6 DK |
143 | return -EIO; |
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; | |
181 | __u32 provided, calculated; | |
182 | ||
183 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
184 | return 1; | |
185 | ||
186 | tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize - | |
187 | sizeof(struct jbd2_journal_block_tail)); | |
188 | provided = tail->t_checksum; | |
189 | tail->t_checksum = 0; | |
190 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
191 | tail->t_checksum = provided; | |
192 | ||
193 | provided = be32_to_cpu(provided); | |
194 | return provided == calculated; | |
195 | } | |
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 | |
3caa487f DW |
208 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) |
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 | ||
18eba7aa | 341 | static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag) |
b517bea1 | 342 | { |
18eba7aa | 343 | unsigned long long block = be32_to_cpu(tag->t_blocknr); |
cd02ff0b | 344 | if (tag_bytes > JBD2_TAG_SIZE32) |
b517bea1 ZB |
345 | block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32; |
346 | return block; | |
347 | } | |
348 | ||
818d276c GS |
349 | /* |
350 | * calc_chksums calculates the checksums for the blocks described in the | |
351 | * descriptor block. | |
352 | */ | |
353 | static int calc_chksums(journal_t *journal, struct buffer_head *bh, | |
354 | unsigned long *next_log_block, __u32 *crc32_sum) | |
355 | { | |
356 | int i, num_blks, err; | |
357 | unsigned long io_block; | |
358 | struct buffer_head *obh; | |
359 | ||
360 | num_blks = count_tags(journal, bh); | |
361 | /* Calculate checksum of the descriptor block. */ | |
362 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); | |
363 | ||
364 | for (i = 0; i < num_blks; i++) { | |
365 | io_block = (*next_log_block)++; | |
366 | wrap(journal, *next_log_block); | |
367 | err = jread(&obh, journal, io_block); | |
368 | if (err) { | |
f2a44523 | 369 | printk(KERN_ERR "JBD2: IO error %d recovering block " |
818d276c GS |
370 | "%lu in log\n", err, io_block); |
371 | return 1; | |
372 | } else { | |
373 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, | |
374 | obh->b_size); | |
375 | } | |
8ea76900 | 376 | put_bh(obh); |
818d276c GS |
377 | } |
378 | return 0; | |
379 | } | |
380 | ||
1f56c589 DW |
381 | static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) |
382 | { | |
383 | struct commit_header *h; | |
384 | __u32 provided, calculated; | |
385 | ||
386 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
387 | return 1; | |
388 | ||
389 | h = buf; | |
390 | provided = h->h_chksum[0]; | |
391 | h->h_chksum[0] = 0; | |
392 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
393 | h->h_chksum[0] = provided; | |
394 | ||
395 | provided = be32_to_cpu(provided); | |
396 | return provided == calculated; | |
397 | } | |
398 | ||
c3900875 DW |
399 | static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, |
400 | void *buf, __u32 sequence) | |
401 | { | |
eee06c56 | 402 | __u32 csum32; |
c3900875 DW |
403 | |
404 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
405 | return 1; | |
406 | ||
407 | sequence = cpu_to_be32(sequence); | |
eee06c56 DW |
408 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence, |
409 | sizeof(sequence)); | |
410 | csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize); | |
c3900875 | 411 | |
eee06c56 | 412 | return tag->t_checksum == cpu_to_be16(csum32); |
c3900875 DW |
413 | } |
414 | ||
470decc6 DK |
415 | static int do_one_pass(journal_t *journal, |
416 | struct recovery_info *info, enum passtype pass) | |
417 | { | |
418 | unsigned int first_commit_ID, next_commit_ID; | |
419 | unsigned long next_log_block; | |
420 | int err, success = 0; | |
421 | journal_superblock_t * sb; | |
422 | journal_header_t * tmp; | |
423 | struct buffer_head * bh; | |
424 | unsigned int sequence; | |
425 | int blocktype; | |
b517bea1 | 426 | int tag_bytes = journal_tag_bytes(journal); |
818d276c | 427 | __u32 crc32_sum = ~0; /* Transactional Checksums */ |
3caa487f | 428 | int descr_csum_size = 0; |
470decc6 | 429 | |
470decc6 DK |
430 | /* |
431 | * First thing is to establish what we expect to find in the log | |
432 | * (in terms of transaction IDs), and where (in terms of log | |
433 | * block offsets): query the superblock. | |
434 | */ | |
435 | ||
436 | sb = journal->j_superblock; | |
437 | next_commit_ID = be32_to_cpu(sb->s_sequence); | |
438 | next_log_block = be32_to_cpu(sb->s_start); | |
439 | ||
440 | first_commit_ID = next_commit_ID; | |
441 | if (pass == PASS_SCAN) | |
442 | info->start_transaction = first_commit_ID; | |
443 | ||
444 | jbd_debug(1, "Starting recovery pass %d\n", pass); | |
445 | ||
446 | /* | |
447 | * Now we walk through the log, transaction by transaction, | |
448 | * making sure that each transaction has a commit block in the | |
449 | * expected place. Each complete transaction gets replayed back | |
450 | * into the main filesystem. | |
451 | */ | |
452 | ||
453 | while (1) { | |
454 | int flags; | |
455 | char * tagp; | |
456 | journal_block_tag_t * tag; | |
457 | struct buffer_head * obh; | |
458 | struct buffer_head * nbh; | |
459 | ||
e86e1438 | 460 | cond_resched(); |
470decc6 DK |
461 | |
462 | /* If we already know where to stop the log traversal, | |
463 | * check right now that we haven't gone past the end of | |
464 | * the log. */ | |
465 | ||
466 | if (pass != PASS_SCAN) | |
467 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
468 | break; | |
469 | ||
470 | jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", | |
471 | next_commit_ID, next_log_block, journal->j_last); | |
472 | ||
473 | /* Skip over each chunk of the transaction looking | |
474 | * either the next descriptor block or the final commit | |
475 | * record. */ | |
476 | ||
f2a44523 | 477 | jbd_debug(3, "JBD2: checking block %ld\n", next_log_block); |
470decc6 DK |
478 | err = jread(&bh, journal, next_log_block); |
479 | if (err) | |
480 | goto failed; | |
481 | ||
482 | next_log_block++; | |
483 | wrap(journal, next_log_block); | |
484 | ||
485 | /* What kind of buffer is it? | |
486 | * | |
487 | * If it is a descriptor block, check that it has the | |
488 | * expected sequence number. Otherwise, we're all done | |
489 | * here. */ | |
490 | ||
491 | tmp = (journal_header_t *)bh->b_data; | |
492 | ||
f7f4bccb | 493 | if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
494 | brelse(bh); |
495 | break; | |
496 | } | |
497 | ||
498 | blocktype = be32_to_cpu(tmp->h_blocktype); | |
499 | sequence = be32_to_cpu(tmp->h_sequence); | |
500 | jbd_debug(3, "Found magic %d, sequence %d\n", | |
501 | blocktype, sequence); | |
502 | ||
503 | if (sequence != next_commit_ID) { | |
504 | brelse(bh); | |
505 | break; | |
506 | } | |
507 | ||
508 | /* OK, we have a valid descriptor block which matches | |
509 | * all of the sequence number checks. What are we going | |
510 | * to do with it? That depends on the pass... */ | |
511 | ||
512 | switch(blocktype) { | |
f7f4bccb | 513 | case JBD2_DESCRIPTOR_BLOCK: |
3caa487f DW |
514 | /* Verify checksum first */ |
515 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, | |
516 | JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
517 | descr_csum_size = | |
518 | sizeof(struct jbd2_journal_block_tail); | |
519 | if (descr_csum_size > 0 && | |
520 | !jbd2_descr_block_csum_verify(journal, | |
521 | bh->b_data)) { | |
522 | err = -EIO; | |
523 | goto failed; | |
524 | } | |
525 | ||
470decc6 | 526 | /* If it is a valid descriptor block, replay it |
818d276c GS |
527 | * in pass REPLAY; if journal_checksums enabled, then |
528 | * calculate checksums in PASS_SCAN, otherwise, | |
529 | * just skip over the blocks it describes. */ | |
470decc6 | 530 | if (pass != PASS_REPLAY) { |
818d276c GS |
531 | if (pass == PASS_SCAN && |
532 | JBD2_HAS_COMPAT_FEATURE(journal, | |
533 | JBD2_FEATURE_COMPAT_CHECKSUM) && | |
534 | !info->end_transaction) { | |
535 | if (calc_chksums(journal, bh, | |
536 | &next_log_block, | |
537 | &crc32_sum)) { | |
538 | put_bh(bh); | |
539 | break; | |
540 | } | |
541 | put_bh(bh); | |
542 | continue; | |
543 | } | |
b517bea1 | 544 | next_log_block += count_tags(journal, bh); |
470decc6 | 545 | wrap(journal, next_log_block); |
818d276c | 546 | put_bh(bh); |
470decc6 DK |
547 | continue; |
548 | } | |
549 | ||
550 | /* A descriptor block: we can now write all of | |
551 | * the data blocks. Yay, useful work is finally | |
552 | * getting done here! */ | |
553 | ||
554 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
b517bea1 | 555 | while ((tagp - bh->b_data + tag_bytes) |
3caa487f | 556 | <= journal->j_blocksize - descr_csum_size) { |
470decc6 DK |
557 | unsigned long io_block; |
558 | ||
559 | tag = (journal_block_tag_t *) tagp; | |
8f888ef8 | 560 | flags = be16_to_cpu(tag->t_flags); |
470decc6 DK |
561 | |
562 | io_block = next_log_block++; | |
563 | wrap(journal, next_log_block); | |
564 | err = jread(&obh, journal, io_block); | |
565 | if (err) { | |
566 | /* Recover what we can, but | |
567 | * report failure at the end. */ | |
568 | success = err; | |
f2a44523 EG |
569 | printk(KERN_ERR |
570 | "JBD2: IO error %d recovering " | |
470decc6 DK |
571 | "block %ld in log\n", |
572 | err, io_block); | |
573 | } else { | |
18eba7aa | 574 | unsigned long long blocknr; |
470decc6 DK |
575 | |
576 | J_ASSERT(obh != NULL); | |
b517bea1 ZB |
577 | blocknr = read_tag_block(tag_bytes, |
578 | tag); | |
470decc6 DK |
579 | |
580 | /* If the block has been | |
581 | * revoked, then we're all done | |
582 | * here. */ | |
f7f4bccb | 583 | if (jbd2_journal_test_revoke |
470decc6 DK |
584 | (journal, blocknr, |
585 | next_commit_ID)) { | |
586 | brelse(obh); | |
587 | ++info->nr_revoke_hits; | |
588 | goto skip_write; | |
589 | } | |
590 | ||
c3900875 DW |
591 | /* Look for block corruption */ |
592 | if (!jbd2_block_tag_csum_verify( | |
593 | journal, tag, obh->b_data, | |
594 | be32_to_cpu(tmp->h_sequence))) { | |
595 | brelse(obh); | |
596 | success = -EIO; | |
597 | printk(KERN_ERR "JBD: Invalid " | |
598 | "checksum recovering " | |
599 | "block %llu in log\n", | |
600 | blocknr); | |
601 | continue; | |
602 | } | |
603 | ||
470decc6 DK |
604 | /* Find a buffer for the new |
605 | * data being restored */ | |
606 | nbh = __getblk(journal->j_fs_dev, | |
607 | blocknr, | |
608 | journal->j_blocksize); | |
609 | if (nbh == NULL) { | |
610 | printk(KERN_ERR | |
f2a44523 | 611 | "JBD2: Out of memory " |
470decc6 DK |
612 | "during recovery.\n"); |
613 | err = -ENOMEM; | |
614 | brelse(bh); | |
615 | brelse(obh); | |
616 | goto failed; | |
617 | } | |
618 | ||
619 | lock_buffer(nbh); | |
620 | memcpy(nbh->b_data, obh->b_data, | |
621 | journal->j_blocksize); | |
f7f4bccb | 622 | if (flags & JBD2_FLAG_ESCAPE) { |
d0025676 | 623 | *((__be32 *)nbh->b_data) = |
f7f4bccb | 624 | cpu_to_be32(JBD2_MAGIC_NUMBER); |
470decc6 DK |
625 | } |
626 | ||
627 | BUFFER_TRACE(nbh, "marking dirty"); | |
628 | set_buffer_uptodate(nbh); | |
629 | mark_buffer_dirty(nbh); | |
630 | BUFFER_TRACE(nbh, "marking uptodate"); | |
631 | ++info->nr_replays; | |
632 | /* ll_rw_block(WRITE, 1, &nbh); */ | |
633 | unlock_buffer(nbh); | |
634 | brelse(obh); | |
635 | brelse(nbh); | |
636 | } | |
637 | ||
638 | skip_write: | |
b517bea1 | 639 | tagp += tag_bytes; |
f7f4bccb | 640 | if (!(flags & JBD2_FLAG_SAME_UUID)) |
470decc6 DK |
641 | tagp += 16; |
642 | ||
f7f4bccb | 643 | if (flags & JBD2_FLAG_LAST_TAG) |
470decc6 DK |
644 | break; |
645 | } | |
646 | ||
647 | brelse(bh); | |
648 | continue; | |
649 | ||
f7f4bccb | 650 | case JBD2_COMMIT_BLOCK: |
818d276c GS |
651 | /* How to differentiate between interrupted commit |
652 | * and journal corruption ? | |
653 | * | |
654 | * {nth transaction} | |
655 | * Checksum Verification Failed | |
656 | * | | |
657 | * ____________________ | |
658 | * | | | |
659 | * async_commit sync_commit | |
660 | * | | | |
661 | * | GO TO NEXT "Journal Corruption" | |
662 | * | TRANSACTION | |
663 | * | | |
664 | * {(n+1)th transanction} | |
665 | * | | |
666 | * _______|______________ | |
667 | * | | | |
668 | * Commit block found Commit block not found | |
669 | * | | | |
670 | * "Journal Corruption" | | |
671 | * _____________|_________ | |
672 | * | | | |
673 | * nth trans corrupt OR nth trans | |
674 | * and (n+1)th interrupted interrupted | |
675 | * before commit block | |
676 | * could reach the disk. | |
677 | * (Cannot find the difference in above | |
678 | * mentioned conditions. Hence assume | |
679 | * "Interrupted Commit".) | |
680 | */ | |
681 | ||
682 | /* Found an expected commit block: if checksums | |
683 | * are present verify them in PASS_SCAN; else not | |
684 | * much to do other than move on to the next sequence | |
470decc6 | 685 | * number. */ |
818d276c GS |
686 | if (pass == PASS_SCAN && |
687 | JBD2_HAS_COMPAT_FEATURE(journal, | |
688 | JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
689 | int chksum_err, chksum_seen; | |
690 | struct commit_header *cbh = | |
691 | (struct commit_header *)bh->b_data; | |
692 | unsigned found_chksum = | |
693 | be32_to_cpu(cbh->h_chksum[0]); | |
694 | ||
695 | chksum_err = chksum_seen = 0; | |
696 | ||
697 | if (info->end_transaction) { | |
624080ed TT |
698 | journal->j_failed_commit = |
699 | info->end_transaction; | |
818d276c GS |
700 | brelse(bh); |
701 | break; | |
702 | } | |
703 | ||
704 | if (crc32_sum == found_chksum && | |
705 | cbh->h_chksum_type == JBD2_CRC32_CHKSUM && | |
706 | cbh->h_chksum_size == | |
707 | JBD2_CRC32_CHKSUM_SIZE) | |
708 | chksum_seen = 1; | |
709 | else if (!(cbh->h_chksum_type == 0 && | |
710 | cbh->h_chksum_size == 0 && | |
711 | found_chksum == 0 && | |
712 | !chksum_seen)) | |
713 | /* | |
714 | * If fs is mounted using an old kernel and then | |
715 | * kernel with journal_chksum is used then we | |
716 | * get a situation where the journal flag has | |
717 | * checksum flag set but checksums are not | |
718 | * present i.e chksum = 0, in the individual | |
719 | * commit blocks. | |
720 | * Hence to avoid checksum failures, in this | |
721 | * situation, this extra check is added. | |
722 | */ | |
723 | chksum_err = 1; | |
724 | ||
725 | if (chksum_err) { | |
726 | info->end_transaction = next_commit_ID; | |
727 | ||
4d605179 | 728 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, |
818d276c | 729 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){ |
624080ed TT |
730 | journal->j_failed_commit = |
731 | next_commit_ID; | |
818d276c GS |
732 | brelse(bh); |
733 | break; | |
734 | } | |
735 | } | |
736 | crc32_sum = ~0; | |
737 | } | |
1f56c589 DW |
738 | if (pass == PASS_SCAN && |
739 | !jbd2_commit_block_csum_verify(journal, | |
740 | bh->b_data)) { | |
741 | info->end_transaction = next_commit_ID; | |
742 | ||
743 | if (!JBD2_HAS_INCOMPAT_FEATURE(journal, | |
744 | JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { | |
745 | journal->j_failed_commit = | |
746 | next_commit_ID; | |
747 | brelse(bh); | |
748 | break; | |
749 | } | |
750 | } | |
470decc6 DK |
751 | brelse(bh); |
752 | next_commit_ID++; | |
753 | continue; | |
754 | ||
f7f4bccb | 755 | case JBD2_REVOKE_BLOCK: |
470decc6 DK |
756 | /* If we aren't in the REVOKE pass, then we can |
757 | * just skip over this block. */ | |
758 | if (pass != PASS_REVOKE) { | |
759 | brelse(bh); | |
760 | continue; | |
761 | } | |
762 | ||
763 | err = scan_revoke_records(journal, bh, | |
764 | next_commit_ID, info); | |
765 | brelse(bh); | |
766 | if (err) | |
767 | goto failed; | |
768 | continue; | |
769 | ||
770 | default: | |
771 | jbd_debug(3, "Unrecognised magic %d, end of scan.\n", | |
772 | blocktype); | |
773 | brelse(bh); | |
774 | goto done; | |
775 | } | |
776 | } | |
777 | ||
778 | done: | |
779 | /* | |
780 | * We broke out of the log scan loop: either we came to the | |
781 | * known end of the log or we found an unexpected block in the | |
782 | * log. If the latter happened, then we know that the "current" | |
783 | * transaction marks the end of the valid log. | |
784 | */ | |
785 | ||
818d276c GS |
786 | if (pass == PASS_SCAN) { |
787 | if (!info->end_transaction) | |
788 | info->end_transaction = next_commit_ID; | |
789 | } else { | |
470decc6 DK |
790 | /* It's really bad news if different passes end up at |
791 | * different places (but possible due to IO errors). */ | |
792 | if (info->end_transaction != next_commit_ID) { | |
f2a44523 | 793 | printk(KERN_ERR "JBD2: recovery pass %d ended at " |
470decc6 DK |
794 | "transaction %u, expected %u\n", |
795 | pass, next_commit_ID, info->end_transaction); | |
796 | if (!success) | |
797 | success = -EIO; | |
798 | } | |
799 | } | |
800 | ||
801 | return success; | |
802 | ||
803 | failed: | |
804 | return err; | |
805 | } | |
806 | ||
42a7106d DW |
807 | static int jbd2_revoke_block_csum_verify(journal_t *j, |
808 | void *buf) | |
809 | { | |
810 | struct jbd2_journal_revoke_tail *tail; | |
811 | __u32 provided, calculated; | |
812 | ||
813 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
814 | return 1; | |
815 | ||
816 | tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize - | |
817 | sizeof(struct jbd2_journal_revoke_tail)); | |
818 | provided = tail->r_checksum; | |
819 | tail->r_checksum = 0; | |
820 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); | |
821 | tail->r_checksum = provided; | |
822 | ||
823 | provided = be32_to_cpu(provided); | |
824 | return provided == calculated; | |
825 | } | |
470decc6 DK |
826 | |
827 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ | |
828 | ||
829 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, | |
830 | tid_t sequence, struct recovery_info *info) | |
831 | { | |
f7f4bccb | 832 | jbd2_journal_revoke_header_t *header; |
470decc6 | 833 | int offset, max; |
b517bea1 | 834 | int record_len = 4; |
470decc6 | 835 | |
f7f4bccb MC |
836 | header = (jbd2_journal_revoke_header_t *) bh->b_data; |
837 | offset = sizeof(jbd2_journal_revoke_header_t); | |
470decc6 DK |
838 | max = be32_to_cpu(header->r_count); |
839 | ||
42a7106d DW |
840 | if (!jbd2_revoke_block_csum_verify(journal, header)) |
841 | return -EINVAL; | |
842 | ||
b517bea1 ZB |
843 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) |
844 | record_len = 8; | |
845 | ||
846 | while (offset + record_len <= max) { | |
18eba7aa | 847 | unsigned long long blocknr; |
470decc6 DK |
848 | int err; |
849 | ||
b517bea1 ZB |
850 | if (record_len == 4) |
851 | blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset))); | |
852 | else | |
853 | blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset))); | |
854 | offset += record_len; | |
f7f4bccb | 855 | err = jbd2_journal_set_revoke(journal, blocknr, sequence); |
470decc6 DK |
856 | if (err) |
857 | return err; | |
858 | ++info->nr_revokes; | |
859 | } | |
860 | return 0; | |
861 | } |