]>
Commit | Line | Data |
---|---|---|
3b5386dc | 1 | /* |
e5ea6b14 TT |
2 | * linux/fs/jbd/recovery.c |
3 | * | |
3b5386dc TT |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 |
5 | * | |
0e8a9560 | 6 | * Copyright 1999-2000 Red Hat Software --- All Rights Reserved |
3b5386dc TT |
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; | |
e5ea6b14 | 13 | * part of the ext2fs journaling system. |
3b5386dc TT |
14 | */ |
15 | ||
16 | #ifndef __KERNEL__ | |
d1154eb4 | 17 | #include "config.h" |
0e8a9560 | 18 | #include "jfs_user.h" |
3b5386dc | 19 | #else |
3de085dd | 20 | #include <linux/time.h> |
3b5386dc | 21 | #include <linux/fs.h> |
8cf93332 | 22 | #include <linux/jbd.h> |
3b5386dc | 23 | #include <linux/errno.h> |
8cf93332 | 24 | #include <linux/slab.h> |
0e8a9560 TT |
25 | #endif |
26 | ||
27 | /* | |
28 | * Maintain information about the progress of the recovery job, so that | |
e5ea6b14 | 29 | * the different passes can carry information between them. |
0e8a9560 | 30 | */ |
e5ea6b14 | 31 | struct recovery_info |
0e8a9560 | 32 | { |
e5ea6b14 | 33 | tid_t start_transaction; |
0e8a9560 | 34 | tid_t end_transaction; |
e5ea6b14 | 35 | |
0e8a9560 TT |
36 | int nr_replays; |
37 | int nr_revokes; | |
38 | int nr_revoke_hits; | |
39 | }; | |
3b5386dc | 40 | |
0e8a9560 | 41 | enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY}; |
8cf93332 TT |
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 *); | |
0e8a9560 TT |
46 | |
47 | #ifdef __KERNEL__ | |
3b5386dc TT |
48 | |
49 | /* Release readahead buffers after use */ | |
e5ea6b14 | 50 | static void journal_brelse_array(struct buffer_head *b[], int n) |
3b5386dc TT |
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 | ||
1f735038 | 69 | #define MAXBUF 8 |
3b5386dc TT |
70 | static int do_readahead(journal_t *journal, unsigned int start) |
71 | { | |
72 | int err; | |
8cf93332 | 73 | unsigned int max, nbufs, next; |
3b693d0b | 74 | unsigned long long blocknr; |
3b5386dc | 75 | struct buffer_head *bh; |
e5ea6b14 | 76 | |
3b5386dc | 77 | struct buffer_head * bufs[MAXBUF]; |
e5ea6b14 | 78 | |
3b5386dc TT |
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. */ | |
e5ea6b14 | 86 | |
3b5386dc | 87 | nbufs = 0; |
e5ea6b14 | 88 | |
3b5386dc | 89 | for (next = start; next < max; next++) { |
8cf93332 TT |
90 | err = journal_bmap(journal, next, &blocknr); |
91 | ||
92 | if (err) { | |
93 | printk (KERN_ERR "JBD: bad block at offset %u\n", | |
3b5386dc | 94 | next); |
3b5386dc TT |
95 | goto failed; |
96 | } | |
8cf93332 | 97 | |
3de085dd | 98 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); |
3b5386dc | 99 | if (!bh) { |
3b5386dc TT |
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); | |
8cf93332 | 108 | journal_brelse_array(bufs, nbufs); |
3b5386dc TT |
109 | nbufs = 0; |
110 | } | |
111 | } else | |
112 | brelse(bh); | |
113 | } | |
114 | ||
115 | if (nbufs) | |
116 | ll_rw_block(READ, nbufs, bufs); | |
117 | err = 0; | |
8cf93332 | 118 | |
e5ea6b14 TT |
119 | failed: |
120 | if (nbufs) | |
8cf93332 | 121 | journal_brelse_array(bufs, nbufs); |
3b5386dc TT |
122 | return err; |
123 | } | |
0e8a9560 TT |
124 | |
125 | #endif /* __KERNEL__ */ | |
126 | ||
3b5386dc TT |
127 | |
128 | /* | |
129 | * Read a block from the journal | |
130 | */ | |
131 | ||
e5ea6b14 | 132 | static int jread(struct buffer_head **bhp, journal_t *journal, |
3b5386dc TT |
133 | unsigned int offset) |
134 | { | |
8cf93332 | 135 | int err; |
3b693d0b | 136 | unsigned long long blocknr; |
3b5386dc TT |
137 | struct buffer_head *bh; |
138 | ||
139 | *bhp = NULL; | |
140 | ||
e5ea6b14 TT |
141 | if (offset >= journal->j_maxlen) { |
142 | printk(KERN_ERR "JBD: corrupted journal superblock\n"); | |
143 | return -EIO; | |
144 | } | |
145 | ||
8cf93332 TT |
146 | err = journal_bmap(journal, offset, &blocknr); |
147 | ||
148 | if (err) { | |
149 | printk (KERN_ERR "JBD: bad block at offset %u\n", | |
3b5386dc | 150 | offset); |
8cf93332 | 151 | return err; |
3b5386dc | 152 | } |
8cf93332 | 153 | |
3de085dd | 154 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); |
3b5386dc TT |
155 | if (!bh) |
156 | return -ENOMEM; | |
8cf93332 | 157 | |
3b5386dc TT |
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 | } | |
8cf93332 | 165 | |
3b5386dc | 166 | if (!buffer_uptodate(bh)) { |
8cf93332 | 167 | printk (KERN_ERR "JBD: Failed to read block at offset %u\n", |
3b5386dc TT |
168 | offset); |
169 | brelse(bh); | |
170 | return -EIO; | |
171 | } | |
8cf93332 | 172 | |
3b5386dc TT |
173 | *bhp = bh; |
174 | return 0; | |
175 | } | |
176 | ||
177 | ||
178 | /* | |
179 | * Count the number of in-use tags in a journal descriptor block. | |
180 | */ | |
181 | ||
185c4aea | 182 | static int count_tags(journal_t *journal, struct buffer_head *bh) |
3b5386dc TT |
183 | { |
184 | char * tagp; | |
185 | journal_block_tag_t * tag; | |
185c4aea TT |
186 | int nr = 0, size = journal->j_blocksize; |
187 | int tag_bytes = journal_tag_bytes(journal); | |
8cf93332 | 188 | |
3b5386dc | 189 | tagp = &bh->b_data[sizeof(journal_header_t)]; |
8cf93332 | 190 | |
185c4aea | 191 | while ((tagp - bh->b_data + tag_bytes) <= size) { |
3b5386dc | 192 | tag = (journal_block_tag_t *) tagp; |
8cf93332 | 193 | |
3b5386dc | 194 | nr++; |
185c4aea | 195 | tagp += tag_bytes; |
2556373a | 196 | if (!(tag->t_flags & cpu_to_be16(JFS_FLAG_SAME_UUID))) |
3b5386dc TT |
197 | tagp += 16; |
198 | ||
2556373a | 199 | if (tag->t_flags & cpu_to_be16(JFS_FLAG_LAST_TAG)) |
3b5386dc TT |
200 | break; |
201 | } | |
8cf93332 | 202 | |
3b5386dc TT |
203 | return nr; |
204 | } | |
205 | ||
206 | ||
207 | /* Make sure we wrap around the log correctly! */ | |
208 | #define wrap(journal, var) \ | |
209 | do { \ | |
210 | if (var >= (journal)->j_last) \ | |
211 | var -= ((journal)->j_last - (journal)->j_first); \ | |
212 | } while (0) | |
213 | ||
3de085dd | 214 | /** |
e5ea6b14 | 215 | * journal_recover - recovers a on-disk journal |
3de085dd | 216 | * @journal: the journal to recover |
3b5386dc TT |
217 | * |
218 | * The primary function for recovering the log contents when mounting a | |
e5ea6b14 TT |
219 | * journaled device. |
220 | * | |
0e8a9560 TT |
221 | * Recovery is done in three passes. In the first pass, we look for the |
222 | * end of the log. In the second, we assemble the list of revoke | |
223 | * blocks. In the third and final pass, we replay any un-revoked blocks | |
e5ea6b14 | 224 | * in the log. |
3b5386dc | 225 | */ |
3b5386dc TT |
226 | int journal_recover(journal_t *journal) |
227 | { | |
0e8a9560 TT |
228 | int err; |
229 | journal_superblock_t * sb; | |
230 | ||
53ef44c4 | 231 | struct recovery_info info; |
e5ea6b14 | 232 | |
1f735038 | 233 | memset(&info, 0, sizeof(info)); |
0e8a9560 | 234 | sb = journal->j_superblock; |
e5ea6b14 TT |
235 | |
236 | /* | |
0e8a9560 TT |
237 | * The journal superblock's s_start field (the current log head) |
238 | * is always zero if, and only if, the journal was cleanly | |
e5ea6b14 | 239 | * unmounted. |
0e8a9560 TT |
240 | */ |
241 | ||
242 | if (!sb->s_start) { | |
8cf93332 | 243 | jbd_debug(1, "No recovery required, last transaction %d\n", |
e5ea6b14 TT |
244 | be32_to_cpu(sb->s_sequence)); |
245 | journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1; | |
0e8a9560 TT |
246 | return 0; |
247 | } | |
e5ea6b14 | 248 | |
0e8a9560 TT |
249 | err = do_one_pass(journal, &info, PASS_SCAN); |
250 | if (!err) | |
251 | err = do_one_pass(journal, &info, PASS_REVOKE); | |
252 | if (!err) | |
253 | err = do_one_pass(journal, &info, PASS_REPLAY); | |
254 | ||
e5ea6b14 | 255 | jbd_debug(1, "JBD: recovery, exit status %d, " |
0e8a9560 TT |
256 | "recovered transactions %u to %u\n", |
257 | err, info.start_transaction, info.end_transaction); | |
e5ea6b14 | 258 | jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n", |
0e8a9560 TT |
259 | info.nr_replays, info.nr_revoke_hits, info.nr_revokes); |
260 | ||
261 | /* Restart the log at the next transaction ID, thus invalidating | |
262 | * any existing commit records in the log. */ | |
263 | journal->j_transaction_sequence = ++info.end_transaction; | |
e5ea6b14 | 264 | |
0e8a9560 | 265 | journal_clear_revoke(journal); |
3de085dd | 266 | sync_blockdev(journal->j_fs_dev); |
0e8a9560 TT |
267 | return err; |
268 | } | |
269 | ||
3de085dd | 270 | /** |
e5ea6b14 | 271 | * journal_skip_recovery - Start journal and wipe exiting records |
3de085dd | 272 | * @journal: journal to startup |
1f735038 TT |
273 | * |
274 | * Locate any valid recovery information from the journal and set up the | |
275 | * journal structures in memory to ignore it (presumably because the | |
e5ea6b14 | 276 | * caller has evidence that it is out of date). |
3de085dd | 277 | * This function does'nt appear to be exorted.. |
1f735038 TT |
278 | * |
279 | * We perform one pass over the journal to allow us to tell the user how | |
280 | * much recovery information is being erased, and to let us initialise | |
e5ea6b14 | 281 | * the journal transaction sequence numbers to the next unused ID. |
1f735038 | 282 | */ |
1f735038 TT |
283 | int journal_skip_recovery(journal_t *journal) |
284 | { | |
285 | int err; | |
286 | journal_superblock_t * sb; | |
287 | ||
288 | struct recovery_info info; | |
e5ea6b14 | 289 | |
8cf93332 | 290 | memset (&info, 0, sizeof(info)); |
1f735038 | 291 | sb = journal->j_superblock; |
e5ea6b14 | 292 | |
1f735038 TT |
293 | err = do_one_pass(journal, &info, PASS_SCAN); |
294 | ||
295 | if (err) { | |
8cf93332 | 296 | printk(KERN_ERR "JBD: error %d scanning journal\n", err); |
1f735038 TT |
297 | ++journal->j_transaction_sequence; |
298 | } else { | |
8cf93332 | 299 | #ifdef CONFIG_JBD_DEBUG |
e5ea6b14 | 300 | int dropped = info.end_transaction - be32_to_cpu(sb->s_sequence); |
8cf93332 | 301 | #endif |
e5ea6b14 | 302 | jbd_debug(1, |
8cf93332 | 303 | "JBD: ignoring %d transaction%s from the journal.\n", |
1f735038 TT |
304 | dropped, (dropped == 1) ? "" : "s"); |
305 | journal->j_transaction_sequence = ++info.end_transaction; | |
306 | } | |
307 | ||
308 | journal->j_tail = 0; | |
1f735038 TT |
309 | return err; |
310 | } | |
311 | ||
185c4aea TT |
312 | static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag) |
313 | { | |
314 | unsigned long long block = be32_to_cpu(tag->t_blocknr); | |
315 | if (tag_bytes > JBD_TAG_SIZE32) | |
316 | block |= (__u64)be32_to_cpu(tag->t_blocknr_high) << 32; | |
317 | return block; | |
318 | } | |
319 | ||
320 | /* | |
321 | * calc_chksums calculates the checksums for the blocks described in the | |
322 | * descriptor block. | |
323 | */ | |
324 | static int calc_chksums(journal_t *journal, struct buffer_head *bh, | |
3b693d0b | 325 | unsigned long long *next_log_block, __u32 *crc32_sum) |
185c4aea TT |
326 | { |
327 | int i, num_blks, err; | |
3b693d0b | 328 | unsigned long long io_block; |
185c4aea TT |
329 | struct buffer_head *obh; |
330 | ||
331 | num_blks = count_tags(journal, bh); | |
332 | /* Calculate checksum of the descriptor block. */ | |
333 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); | |
334 | ||
335 | for (i = 0; i < num_blks; i++) { | |
336 | io_block = (*next_log_block)++; | |
337 | wrap(journal, *next_log_block); | |
338 | err = jread(&obh, journal, io_block); | |
339 | if (err) { | |
340 | printk(KERN_ERR "JBD: IO error %d recovering block " | |
3b693d0b | 341 | "%llu in log\n", err, io_block); |
185c4aea TT |
342 | return 1; |
343 | } else { | |
344 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, | |
345 | obh->b_size); | |
346 | } | |
347 | brelse(obh); | |
348 | } | |
349 | return 0; | |
350 | } | |
351 | ||
8cf93332 TT |
352 | static int do_one_pass(journal_t *journal, |
353 | struct recovery_info *info, enum passtype pass) | |
0e8a9560 | 354 | { |
3b5386dc | 355 | unsigned int first_commit_ID, next_commit_ID; |
3b693d0b | 356 | unsigned long long next_log_block; |
3b5386dc | 357 | int err, success = 0; |
0e8a9560 | 358 | journal_superblock_t * sb; |
e5ea6b14 | 359 | journal_header_t * tmp; |
3b5386dc | 360 | struct buffer_head * bh; |
0e8a9560 TT |
361 | unsigned int sequence; |
362 | int blocktype; | |
185c4aea TT |
363 | int tag_bytes = journal_tag_bytes(journal); |
364 | __u32 crc32_sum = ~0; /* Transactional Checksums */ | |
e5ea6b14 | 365 | |
3b5386dc TT |
366 | /* Precompute the maximum metadata descriptors in a descriptor block */ |
367 | int MAX_BLOCKS_PER_DESC; | |
368 | MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t)) | |
185c4aea | 369 | / tag_bytes); |
3b5386dc | 370 | |
e5ea6b14 | 371 | /* |
3b5386dc TT |
372 | * First thing is to establish what we expect to find in the log |
373 | * (in terms of transaction IDs), and where (in terms of log | |
e5ea6b14 | 374 | * block offsets): query the superblock. |
3b5386dc TT |
375 | */ |
376 | ||
0e8a9560 | 377 | sb = journal->j_superblock; |
e5ea6b14 TT |
378 | next_commit_ID = be32_to_cpu(sb->s_sequence); |
379 | next_log_block = be32_to_cpu(sb->s_start); | |
3b5386dc TT |
380 | |
381 | first_commit_ID = next_commit_ID; | |
0e8a9560 TT |
382 | if (pass == PASS_SCAN) |
383 | info->start_transaction = first_commit_ID; | |
8cf93332 TT |
384 | |
385 | jbd_debug(1, "Starting recovery pass %d\n", pass); | |
386 | ||
3b5386dc TT |
387 | /* |
388 | * Now we walk through the log, transaction by transaction, | |
389 | * making sure that each transaction has a commit block in the | |
390 | * expected place. Each complete transaction gets replayed back | |
e5ea6b14 | 391 | * into the main filesystem. |
3b5386dc TT |
392 | */ |
393 | ||
0e8a9560 TT |
394 | while (1) { |
395 | int flags; | |
396 | char * tagp; | |
397 | journal_block_tag_t * tag; | |
398 | struct buffer_head * obh; | |
399 | struct buffer_head * nbh; | |
e5ea6b14 TT |
400 | |
401 | cond_resched(); | |
402 | ||
0e8a9560 TT |
403 | /* If we already know where to stop the log traversal, |
404 | * check right now that we haven't gone past the end of | |
405 | * the log. */ | |
e5ea6b14 | 406 | |
0e8a9560 TT |
407 | if (pass != PASS_SCAN) |
408 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
409 | break; | |
8cf93332 TT |
410 | |
411 | jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", | |
3b5386dc | 412 | next_commit_ID, next_log_block, journal->j_last); |
3b5386dc | 413 | |
0e8a9560 TT |
414 | /* Skip over each chunk of the transaction looking |
415 | * either the next descriptor block or the final commit | |
416 | * record. */ | |
e5ea6b14 | 417 | |
8cf93332 | 418 | jbd_debug(3, "JBD: checking block %ld\n", next_log_block); |
0e8a9560 TT |
419 | err = jread(&bh, journal, next_log_block); |
420 | if (err) | |
421 | goto failed; | |
8cf93332 | 422 | |
0e8a9560 TT |
423 | next_log_block++; |
424 | wrap(journal, next_log_block); | |
e5ea6b14 TT |
425 | |
426 | /* What kind of buffer is it? | |
427 | * | |
0e8a9560 TT |
428 | * If it is a descriptor block, check that it has the |
429 | * expected sequence number. Otherwise, we're all done | |
430 | * here. */ | |
3b5386dc | 431 | |
8cf93332 | 432 | tmp = (journal_header_t *)bh->b_data; |
e5ea6b14 TT |
433 | |
434 | if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) { | |
3b5386dc TT |
435 | brelse(bh); |
436 | break; | |
437 | } | |
8cf93332 | 438 | |
e5ea6b14 TT |
439 | blocktype = be32_to_cpu(tmp->h_blocktype); |
440 | sequence = be32_to_cpu(tmp->h_sequence); | |
441 | jbd_debug(3, "Found magic %d, sequence %d\n", | |
0e8a9560 | 442 | blocktype, sequence); |
e5ea6b14 | 443 | |
0e8a9560 TT |
444 | if (sequence != next_commit_ID) { |
445 | brelse(bh); | |
446 | break; | |
447 | } | |
e5ea6b14 | 448 | |
0e8a9560 TT |
449 | /* OK, we have a valid descriptor block which matches |
450 | * all of the sequence number checks. What are we going | |
451 | * to do with it? That depends on the pass... */ | |
8cf93332 | 452 | |
0e8a9560 TT |
453 | switch(blocktype) { |
454 | case JFS_DESCRIPTOR_BLOCK: | |
455 | /* If it is a valid descriptor block, replay it | |
185c4aea TT |
456 | * in pass REPLAY; if journal_checksums enabled, then |
457 | * calculate checksums in PASS_SCAN, otherwise, | |
458 | * just skip over the blocks it describes. */ | |
0e8a9560 | 459 | if (pass != PASS_REPLAY) { |
185c4aea TT |
460 | if (pass == PASS_SCAN && |
461 | JFS_HAS_COMPAT_FEATURE(journal, | |
462 | JFS_FEATURE_COMPAT_CHECKSUM) && | |
463 | !info->end_transaction) { | |
464 | if (calc_chksums(journal, bh, | |
465 | &next_log_block, | |
466 | &crc32_sum)) { | |
467 | brelse(bh); | |
468 | break; | |
469 | } | |
470 | brelse(bh); | |
471 | continue; | |
472 | } | |
473 | next_log_block += count_tags(journal, bh); | |
0e8a9560 | 474 | wrap(journal, next_log_block); |
3b5386dc | 475 | brelse(bh); |
0e8a9560 | 476 | continue; |
3b5386dc | 477 | } |
0e8a9560 | 478 | |
3b5386dc TT |
479 | /* A descriptor block: we can now write all of |
480 | * the data blocks. Yay, useful work is finally | |
481 | * getting done here! */ | |
482 | ||
483 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
185c4aea | 484 | while ((tagp - bh->b_data + tag_bytes) |
3b5386dc | 485 | <= journal->j_blocksize) { |
3b693d0b | 486 | unsigned long long io_block; |
8cf93332 | 487 | |
3b5386dc | 488 | tag = (journal_block_tag_t *) tagp; |
2556373a | 489 | flags = be16_to_cpu(tag->t_flags); |
e5ea6b14 | 490 | |
0e8a9560 | 491 | io_block = next_log_block++; |
3b5386dc | 492 | wrap(journal, next_log_block); |
0e8a9560 | 493 | err = jread(&obh, journal, io_block); |
3b5386dc TT |
494 | if (err) { |
495 | /* Recover what we can, but | |
496 | * report failure at the end. */ | |
497 | success = err; | |
e5ea6b14 | 498 | printk (KERN_ERR |
8cf93332 | 499 | "JBD: IO error %d recovering " |
3b693d0b | 500 | "block %llu in log\n", |
0e8a9560 | 501 | err, io_block); |
3b5386dc | 502 | } else { |
3b693d0b | 503 | unsigned long long blocknr; |
e5ea6b14 | 504 | |
3b5386dc | 505 | J_ASSERT(obh != NULL); |
3b693d0b TT |
506 | blocknr = read_tag_block(tag_bytes, |
507 | tag); | |
0e8a9560 TT |
508 | |
509 | /* If the block has been | |
510 | * revoked, then we're all done | |
511 | * here. */ | |
512 | if (journal_test_revoke | |
e5ea6b14 | 513 | (journal, blocknr, |
0e8a9560 TT |
514 | next_commit_ID)) { |
515 | brelse(obh); | |
516 | ++info->nr_revoke_hits; | |
517 | goto skip_write; | |
518 | } | |
e5ea6b14 | 519 | |
0e8a9560 TT |
520 | /* Find a buffer for the new |
521 | * data being restored */ | |
e5ea6b14 TT |
522 | nbh = __getblk(journal->j_fs_dev, |
523 | blocknr, | |
524 | journal->j_blocksize); | |
3b5386dc | 525 | if (nbh == NULL) { |
e5ea6b14 | 526 | printk(KERN_ERR |
8cf93332 | 527 | "JBD: Out of memory " |
3b5386dc TT |
528 | "during recovery.\n"); |
529 | err = -ENOMEM; | |
530 | brelse(bh); | |
531 | brelse(obh); | |
532 | goto failed; | |
533 | } | |
534 | ||
3de085dd | 535 | lock_buffer(nbh); |
8cf93332 TT |
536 | memcpy(nbh->b_data, obh->b_data, |
537 | journal->j_blocksize); | |
3b5386dc | 538 | if (flags & JFS_FLAG_ESCAPE) { |
c816ecb2 ES |
539 | journal_header_t *header; |
540 | ||
541 | header = (journal_header_t *) &nbh->b_data[0]; | |
542 | header->h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
3b5386dc | 543 | } |
8cf93332 TT |
544 | |
545 | BUFFER_TRACE(nbh, "marking dirty"); | |
3de085dd | 546 | set_buffer_uptodate(nbh); |
8cf93332 TT |
547 | mark_buffer_dirty(nbh); |
548 | BUFFER_TRACE(nbh, "marking uptodate"); | |
0e8a9560 | 549 | ++info->nr_replays; |
53ef44c4 | 550 | /* ll_rw_block(WRITE, 1, &nbh); */ |
3de085dd | 551 | unlock_buffer(nbh); |
3b5386dc TT |
552 | brelse(obh); |
553 | brelse(nbh); | |
554 | } | |
e5ea6b14 | 555 | |
0e8a9560 | 556 | skip_write: |
185c4aea | 557 | tagp += tag_bytes; |
3b5386dc TT |
558 | if (!(flags & JFS_FLAG_SAME_UUID)) |
559 | tagp += 16; | |
560 | ||
561 | if (flags & JFS_FLAG_LAST_TAG) | |
562 | break; | |
0e8a9560 | 563 | } |
e5ea6b14 | 564 | |
0e8a9560 TT |
565 | brelse(bh); |
566 | continue; | |
8cf93332 | 567 | |
0e8a9560 | 568 | case JFS_COMMIT_BLOCK: |
185c4aea TT |
569 | jbd_debug(3, "Commit block for #%u found\n", |
570 | next_commit_ID); | |
571 | /* How to differentiate between interrupted commit | |
572 | * and journal corruption ? | |
573 | * | |
574 | * {nth transaction} | |
575 | * Checksum Verification Failed | |
576 | * | | |
577 | * ____________________ | |
578 | * | | | |
579 | * async_commit sync_commit | |
580 | * | | | |
581 | * | GO TO NEXT "Journal Corruption" | |
582 | * | TRANSACTION | |
583 | * | | |
584 | * {(n+1)th transanction} | |
585 | * | | |
586 | * _______|______________ | |
587 | * | | | |
588 | * Commit block found Commit block not found | |
589 | * | | | |
590 | * "Journal Corruption" | | |
591 | * _____________|_________ | |
592 | * | | | |
593 | * nth trans corrupt OR nth trans | |
594 | * and (n+1)th interrupted interrupted | |
595 | * before commit block | |
596 | * could reach the disk. | |
597 | * (Cannot find the difference in above | |
598 | * mentioned conditions. Hence assume | |
599 | * "Interrupted Commit".) | |
600 | */ | |
601 | ||
602 | /* Found an expected commit block: if checksums | |
603 | * are present verify them in PASS_SCAN; else not | |
604 | * much to do other than move on to the next sequence | |
0e8a9560 | 605 | * number. */ |
185c4aea TT |
606 | if (pass == PASS_SCAN && |
607 | JFS_HAS_COMPAT_FEATURE(journal, | |
608 | JFS_FEATURE_COMPAT_CHECKSUM)) { | |
609 | int chksum_err, chksum_seen; | |
610 | struct commit_header *cbh = | |
611 | (struct commit_header *)bh->b_data; | |
612 | unsigned found_chksum = | |
613 | be32_to_cpu(cbh->h_chksum[0]); | |
614 | ||
615 | chksum_err = chksum_seen = 0; | |
616 | ||
617 | jbd_debug(3, "Checksums %x %x\n", | |
618 | crc32_sum, found_chksum); | |
619 | if (info->end_transaction) { | |
620 | journal->j_failed_commit = | |
621 | info->end_transaction; | |
622 | brelse(bh); | |
623 | break; | |
624 | } | |
625 | ||
626 | if (crc32_sum == found_chksum && | |
627 | cbh->h_chksum_type == JBD2_CRC32_CHKSUM && | |
628 | cbh->h_chksum_size == | |
629 | JBD2_CRC32_CHKSUM_SIZE) | |
630 | chksum_seen = 1; | |
631 | else if (!(cbh->h_chksum_type == 0 && | |
632 | cbh->h_chksum_size == 0 && | |
633 | found_chksum == 0 && | |
634 | !chksum_seen)) | |
635 | /* | |
636 | * If fs is mounted using an old kernel and then | |
637 | * kernel with journal_chksum is used then we | |
638 | * get a situation where the journal flag has | |
639 | * checksum flag set but checksums are not | |
640 | * present i.e chksum = 0, in the individual | |
641 | * commit blocks. | |
642 | * Hence to avoid checksum failures, in this | |
643 | * situation, this extra check is added. | |
644 | */ | |
645 | chksum_err = 1; | |
646 | ||
647 | if (chksum_err) { | |
648 | info->end_transaction = next_commit_ID; | |
2acad6b4 TT |
649 | jbd_debug(1, "Checksum_err %x %x\n", |
650 | crc32_sum, found_chksum); | |
185c4aea TT |
651 | if (!JFS_HAS_INCOMPAT_FEATURE(journal, |
652 | JFS_FEATURE_INCOMPAT_ASYNC_COMMIT)){ | |
653 | journal->j_failed_commit = | |
654 | next_commit_ID; | |
655 | brelse(bh); | |
656 | break; | |
657 | } | |
658 | } | |
659 | crc32_sum = ~0; | |
660 | } | |
0e8a9560 TT |
661 | brelse(bh); |
662 | next_commit_ID++; | |
663 | continue; | |
3b5386dc | 664 | |
0e8a9560 TT |
665 | case JFS_REVOKE_BLOCK: |
666 | /* If we aren't in the REVOKE pass, then we can | |
667 | * just skip over this block. */ | |
668 | if (pass != PASS_REVOKE) { | |
669 | brelse(bh); | |
670 | continue; | |
671 | } | |
3b5386dc | 672 | |
8cf93332 | 673 | err = scan_revoke_records(journal, bh, |
0e8a9560 | 674 | next_commit_ID, info); |
3b5386dc | 675 | brelse(bh); |
0e8a9560 TT |
676 | if (err) |
677 | goto failed; | |
678 | continue; | |
679 | ||
680 | default: | |
8cf93332 | 681 | jbd_debug(3, "Unrecognised magic %d, end of scan.\n", |
0e8a9560 | 682 | blocktype); |
2aa362f5 | 683 | brelse(bh); |
0e8a9560 TT |
684 | goto done; |
685 | } | |
3b5386dc | 686 | } |
3b5386dc | 687 | |
0e8a9560 | 688 | done: |
e5ea6b14 | 689 | /* |
0e8a9560 TT |
690 | * We broke out of the log scan loop: either we came to the |
691 | * known end of the log or we found an unexpected block in the | |
692 | * log. If the latter happened, then we know that the "current" | |
693 | * transaction marks the end of the valid log. | |
694 | */ | |
e5ea6b14 | 695 | |
185c4aea TT |
696 | if (pass == PASS_SCAN) { |
697 | if (!info->end_transaction) | |
698 | info->end_transaction = next_commit_ID; | |
699 | } else { | |
0e8a9560 TT |
700 | /* It's really bad news if different passes end up at |
701 | * different places (but possible due to IO errors). */ | |
702 | if (info->end_transaction != next_commit_ID) { | |
8cf93332 | 703 | printk (KERN_ERR "JBD: recovery pass %d ended at " |
0e8a9560 TT |
704 | "transaction %u, expected %u\n", |
705 | pass, next_commit_ID, info->end_transaction); | |
706 | if (!success) | |
707 | success = -EIO; | |
708 | } | |
709 | } | |
710 | ||
711 | return success; | |
3b5386dc | 712 | |
0e8a9560 | 713 | failed: |
3b5386dc TT |
714 | return err; |
715 | } | |
0e8a9560 TT |
716 | |
717 | ||
718 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ | |
719 | ||
e5ea6b14 | 720 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, |
0e8a9560 TT |
721 | tid_t sequence, struct recovery_info *info) |
722 | { | |
723 | journal_revoke_header_t *header; | |
724 | int offset, max; | |
79a4dddb | 725 | int record_len = 4; |
8cf93332 | 726 | |
0e8a9560 TT |
727 | header = (journal_revoke_header_t *) bh->b_data; |
728 | offset = sizeof(journal_revoke_header_t); | |
e5ea6b14 TT |
729 | max = be32_to_cpu(header->r_count); |
730 | ||
79a4dddb DW |
731 | if (JFS_HAS_INCOMPAT_FEATURE(journal, JFS_FEATURE_INCOMPAT_64BIT)) |
732 | record_len = 8; | |
733 | ||
0e8a9560 | 734 | while (offset < max) { |
3b693d0b | 735 | unsigned long long blocknr; |
0e8a9560 | 736 | int err; |
e5ea6b14 | 737 | |
79a4dddb DW |
738 | if (record_len == 4) |
739 | blocknr = ext2fs_be32_to_cpu(*((__be32 *)(bh->b_data + | |
740 | offset))); | |
741 | else | |
742 | blocknr = ext2fs_be64_to_cpu(*((__be64 *)(bh->b_data + | |
743 | offset))); | |
744 | offset += record_len; | |
0e8a9560 TT |
745 | err = journal_set_revoke(journal, blocknr, sequence); |
746 | if (err) | |
747 | return err; | |
748 | ++info->nr_revokes; | |
749 | } | |
750 | return 0; | |
751 | } |