]>
Commit | Line | Data |
---|---|---|
3b5386dc | 1 | /* |
13af4b93 | 2 | * linux/fs/jbd2/recovery.c |
e5ea6b14 | 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__ | |
0e8a9560 | 17 | #include "jfs_user.h" |
3b5386dc | 18 | #else |
3de085dd | 19 | #include <linux/time.h> |
3b5386dc | 20 | #include <linux/fs.h> |
13af4b93 | 21 | #include <linux/jbd2.h> |
3b5386dc | 22 | #include <linux/errno.h> |
13af4b93 DW |
23 | #include <linux/crc32.h> |
24 | #include <linux/blkdev.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) { | |
13af4b93 | 93 | printk(KERN_ERR "JBD2: 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 | ||
c8ca2397 TT |
127 | static inline __u32 get_be32(__be32 *p) |
128 | { | |
129 | unsigned char *cp = (unsigned char *) p; | |
130 | __u32 ret; | |
131 | ||
132 | ret = *cp++; | |
133 | ret = (ret << 8) + *cp++; | |
134 | ret = (ret << 8) + *cp++; | |
135 | ret = (ret << 8) + *cp++; | |
136 | return ret; | |
137 | } | |
138 | ||
139 | static inline __u16 get_be16(__be16 *p) | |
140 | { | |
141 | unsigned char *cp = (unsigned char *) p; | |
142 | __u16 ret; | |
143 | ||
144 | ret = *cp++; | |
145 | ret = (ret << 8) + *cp++; | |
146 | return ret; | |
147 | } | |
3b5386dc TT |
148 | |
149 | /* | |
150 | * Read a block from the journal | |
151 | */ | |
152 | ||
e5ea6b14 | 153 | static int jread(struct buffer_head **bhp, journal_t *journal, |
3b5386dc TT |
154 | unsigned int offset) |
155 | { | |
8cf93332 | 156 | int err; |
3b693d0b | 157 | unsigned long long blocknr; |
3b5386dc TT |
158 | struct buffer_head *bh; |
159 | ||
160 | *bhp = NULL; | |
161 | ||
e5ea6b14 | 162 | if (offset >= journal->j_maxlen) { |
13af4b93 | 163 | printk(KERN_ERR "JBD2: corrupted journal superblock\n"); |
b4f02c9f | 164 | return -EFSCORRUPTED; |
e5ea6b14 TT |
165 | } |
166 | ||
8cf93332 TT |
167 | err = journal_bmap(journal, offset, &blocknr); |
168 | ||
169 | if (err) { | |
13af4b93 | 170 | printk(KERN_ERR "JBD2: bad block at offset %u\n", |
3b5386dc | 171 | offset); |
8cf93332 | 172 | return err; |
3b5386dc | 173 | } |
8cf93332 | 174 | |
3de085dd | 175 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); |
3b5386dc TT |
176 | if (!bh) |
177 | return -ENOMEM; | |
8cf93332 | 178 | |
3b5386dc TT |
179 | if (!buffer_uptodate(bh)) { |
180 | /* If this is a brand new buffer, start readahead. | |
181 | Otherwise, we assume we are already reading it. */ | |
182 | if (!buffer_req(bh)) | |
183 | do_readahead(journal, offset); | |
184 | wait_on_buffer(bh); | |
185 | } | |
8cf93332 | 186 | |
3b5386dc | 187 | if (!buffer_uptodate(bh)) { |
13af4b93 | 188 | printk(KERN_ERR "JBD2: Failed to read block at offset %u\n", |
3b5386dc TT |
189 | offset); |
190 | brelse(bh); | |
191 | return -EIO; | |
192 | } | |
8cf93332 | 193 | |
3b5386dc TT |
194 | *bhp = bh; |
195 | return 0; | |
196 | } | |
197 | ||
5c842851 DW |
198 | static int jbd2_descr_block_csum_verify(journal_t *j, |
199 | void *buf) | |
200 | { | |
201 | struct journal_block_tail *tail; | |
13af4b93 DW |
202 | __u32 provided; |
203 | __u32 calculated; | |
5c842851 | 204 | |
38d5adf3 | 205 | if (!journal_has_csum_v2or3(j)) |
5c842851 DW |
206 | return 1; |
207 | ||
478360f5 | 208 | tail = (struct journal_block_tail *)((char *)buf + j->j_blocksize - |
5c842851 DW |
209 | sizeof(struct journal_block_tail)); |
210 | provided = tail->t_checksum; | |
211 | tail->t_checksum = 0; | |
13af4b93 | 212 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
5c842851 DW |
213 | tail->t_checksum = provided; |
214 | ||
13af4b93 | 215 | return provided == ext2fs_cpu_to_be32(calculated); |
5c842851 | 216 | } |
3b5386dc TT |
217 | |
218 | /* | |
219 | * Count the number of in-use tags in a journal descriptor block. | |
220 | */ | |
221 | ||
185c4aea | 222 | static int count_tags(journal_t *journal, struct buffer_head *bh) |
3b5386dc TT |
223 | { |
224 | char * tagp; | |
225 | journal_block_tag_t * tag; | |
185c4aea TT |
226 | int nr = 0, size = journal->j_blocksize; |
227 | int tag_bytes = journal_tag_bytes(journal); | |
8cf93332 | 228 | |
38d5adf3 | 229 | if (journal_has_csum_v2or3(journal)) |
5c842851 DW |
230 | size -= sizeof(struct journal_block_tail); |
231 | ||
3b5386dc | 232 | tagp = &bh->b_data[sizeof(journal_header_t)]; |
8cf93332 | 233 | |
185c4aea | 234 | while ((tagp - bh->b_data + tag_bytes) <= size) { |
3b5386dc | 235 | tag = (journal_block_tag_t *) tagp; |
8cf93332 | 236 | |
3b5386dc | 237 | nr++; |
185c4aea | 238 | tagp += tag_bytes; |
c8ca2397 | 239 | if (!(get_be16(&tag->t_flags) & JFS_FLAG_SAME_UUID)) |
3b5386dc TT |
240 | tagp += 16; |
241 | ||
c8ca2397 | 242 | if (get_be16(&tag->t_flags) & JFS_FLAG_LAST_TAG) |
3b5386dc TT |
243 | break; |
244 | } | |
8cf93332 | 245 | |
3b5386dc TT |
246 | return nr; |
247 | } | |
248 | ||
249 | ||
250 | /* Make sure we wrap around the log correctly! */ | |
251 | #define wrap(journal, var) \ | |
252 | do { \ | |
253 | if (var >= (journal)->j_last) \ | |
254 | var -= ((journal)->j_last - (journal)->j_first); \ | |
255 | } while (0) | |
256 | ||
3de085dd | 257 | /** |
e5ea6b14 | 258 | * journal_recover - recovers a on-disk journal |
3de085dd | 259 | * @journal: the journal to recover |
3b5386dc TT |
260 | * |
261 | * The primary function for recovering the log contents when mounting a | |
e5ea6b14 TT |
262 | * journaled device. |
263 | * | |
0e8a9560 TT |
264 | * Recovery is done in three passes. In the first pass, we look for the |
265 | * end of the log. In the second, we assemble the list of revoke | |
266 | * blocks. In the third and final pass, we replay any un-revoked blocks | |
e5ea6b14 | 267 | * in the log. |
3b5386dc | 268 | */ |
3b5386dc TT |
269 | int journal_recover(journal_t *journal) |
270 | { | |
13af4b93 | 271 | int err, err2; |
0e8a9560 TT |
272 | journal_superblock_t * sb; |
273 | ||
53ef44c4 | 274 | struct recovery_info info; |
e5ea6b14 | 275 | |
1f735038 | 276 | memset(&info, 0, sizeof(info)); |
0e8a9560 | 277 | sb = journal->j_superblock; |
e5ea6b14 TT |
278 | |
279 | /* | |
0e8a9560 TT |
280 | * The journal superblock's s_start field (the current log head) |
281 | * is always zero if, and only if, the journal was cleanly | |
e5ea6b14 | 282 | * unmounted. |
0e8a9560 TT |
283 | */ |
284 | ||
285 | if (!sb->s_start) { | |
8cf93332 | 286 | jbd_debug(1, "No recovery required, last transaction %d\n", |
13af4b93 DW |
287 | ext2fs_be32_to_cpu(sb->s_sequence)); |
288 | journal->j_transaction_sequence = ext2fs_be32_to_cpu(sb->s_sequence) + 1; | |
0e8a9560 TT |
289 | return 0; |
290 | } | |
e5ea6b14 | 291 | |
0e8a9560 TT |
292 | err = do_one_pass(journal, &info, PASS_SCAN); |
293 | if (!err) | |
294 | err = do_one_pass(journal, &info, PASS_REVOKE); | |
295 | if (!err) | |
296 | err = do_one_pass(journal, &info, PASS_REPLAY); | |
297 | ||
13af4b93 | 298 | jbd_debug(1, "JBD2: recovery, exit status %d, " |
0e8a9560 TT |
299 | "recovered transactions %u to %u\n", |
300 | err, info.start_transaction, info.end_transaction); | |
13af4b93 | 301 | jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n", |
0e8a9560 TT |
302 | info.nr_replays, info.nr_revoke_hits, info.nr_revokes); |
303 | ||
304 | /* Restart the log at the next transaction ID, thus invalidating | |
305 | * any existing commit records in the log. */ | |
306 | journal->j_transaction_sequence = ++info.end_transaction; | |
e5ea6b14 | 307 | |
0e8a9560 | 308 | journal_clear_revoke(journal); |
13af4b93 DW |
309 | err2 = sync_blockdev(journal->j_fs_dev); |
310 | if (!err) | |
311 | err = err2; | |
312 | /* Make sure all replayed data is on permanent storage */ | |
313 | if (journal->j_flags & JFS_BARRIER) { | |
314 | err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); | |
315 | if (!err) | |
316 | err = err2; | |
317 | } | |
0e8a9560 TT |
318 | return err; |
319 | } | |
320 | ||
3de085dd | 321 | /** |
e5ea6b14 | 322 | * journal_skip_recovery - Start journal and wipe exiting records |
3de085dd | 323 | * @journal: journal to startup |
1f735038 TT |
324 | * |
325 | * Locate any valid recovery information from the journal and set up the | |
326 | * journal structures in memory to ignore it (presumably because the | |
e5ea6b14 | 327 | * caller has evidence that it is out of date). |
055866d8 | 328 | * This function doesn't appear to be exported.. |
1f735038 TT |
329 | * |
330 | * We perform one pass over the journal to allow us to tell the user how | |
331 | * much recovery information is being erased, and to let us initialise | |
e5ea6b14 | 332 | * the journal transaction sequence numbers to the next unused ID. |
1f735038 | 333 | */ |
1f735038 TT |
334 | int journal_skip_recovery(journal_t *journal) |
335 | { | |
336 | int err; | |
13af4b93 | 337 | |
1f735038 | 338 | struct recovery_info info; |
e5ea6b14 | 339 | |
8cf93332 | 340 | memset (&info, 0, sizeof(info)); |
e5ea6b14 | 341 | |
1f735038 TT |
342 | err = do_one_pass(journal, &info, PASS_SCAN); |
343 | ||
344 | if (err) { | |
13af4b93 | 345 | printk(KERN_ERR "JBD2: error %d scanning journal\n", err); |
1f735038 TT |
346 | ++journal->j_transaction_sequence; |
347 | } else { | |
13af4b93 | 348 | #ifdef CONFIG_JFS_DEBUG |
97f168b6 | 349 | int dropped = info.end_transaction - |
13af4b93 | 350 | ext2fs_be32_to_cpu(journal->j_superblock->s_sequence); |
e5ea6b14 | 351 | jbd_debug(1, |
13af4b93 | 352 | "JBD2: ignoring %d transaction%s from the journal.\n", |
1f735038 | 353 | dropped, (dropped == 1) ? "" : "s"); |
13af4b93 | 354 | #endif |
1f735038 TT |
355 | journal->j_transaction_sequence = ++info.end_transaction; |
356 | } | |
357 | ||
358 | journal->j_tail = 0; | |
1f735038 TT |
359 | return err; |
360 | } | |
361 | ||
38d5adf3 DW |
362 | static inline unsigned long long read_tag_block(journal_t *journal, |
363 | journal_block_tag_t *tag) | |
185c4aea | 364 | { |
8d7a6392 | 365 | unsigned long long block = get_be32(&tag->t_blocknr); |
86f3b6cf | 366 | if (jfs_has_feature_64bit(journal)) |
8d7a6392 | 367 | block |= (u64)get_be32(&tag->t_blocknr_high) << 32; |
185c4aea TT |
368 | return block; |
369 | } | |
370 | ||
371 | /* | |
372 | * calc_chksums calculates the checksums for the blocks described in the | |
373 | * descriptor block. | |
374 | */ | |
375 | static int calc_chksums(journal_t *journal, struct buffer_head *bh, | |
13af4b93 | 376 | unsigned long *next_log_block, __u32 *crc32_sum) |
185c4aea TT |
377 | { |
378 | int i, num_blks, err; | |
13af4b93 | 379 | unsigned long io_block; |
185c4aea TT |
380 | struct buffer_head *obh; |
381 | ||
382 | num_blks = count_tags(journal, bh); | |
383 | /* Calculate checksum of the descriptor block. */ | |
13af4b93 | 384 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); |
185c4aea TT |
385 | |
386 | for (i = 0; i < num_blks; i++) { | |
387 | io_block = (*next_log_block)++; | |
388 | wrap(journal, *next_log_block); | |
389 | err = jread(&obh, journal, io_block); | |
390 | if (err) { | |
13af4b93 DW |
391 | printk(KERN_ERR "JBD2: IO error %d recovering block " |
392 | "%lu in log\n", err, io_block); | |
185c4aea TT |
393 | return 1; |
394 | } else { | |
13af4b93 DW |
395 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, |
396 | obh->b_size); | |
185c4aea | 397 | } |
13af4b93 | 398 | put_bh(obh); |
185c4aea TT |
399 | } |
400 | return 0; | |
401 | } | |
402 | ||
721e065c DW |
403 | static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) |
404 | { | |
405 | struct commit_header *h; | |
13af4b93 DW |
406 | __u32 provided; |
407 | __u32 calculated; | |
721e065c | 408 | |
38d5adf3 | 409 | if (!journal_has_csum_v2or3(j)) |
721e065c DW |
410 | return 1; |
411 | ||
412 | h = buf; | |
413 | provided = h->h_chksum[0]; | |
414 | h->h_chksum[0] = 0; | |
13af4b93 | 415 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
721e065c DW |
416 | h->h_chksum[0] = provided; |
417 | ||
13af4b93 | 418 | return provided == ext2fs_cpu_to_be32(calculated); |
721e065c DW |
419 | } |
420 | ||
d9200380 DW |
421 | static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, |
422 | void *buf, __u32 sequence) | |
423 | { | |
38d5adf3 | 424 | journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag; |
13af4b93 DW |
425 | __u32 csum32; |
426 | __u32 seq; | |
d9200380 | 427 | |
38d5adf3 | 428 | if (!journal_has_csum_v2or3(j)) |
d9200380 DW |
429 | return 1; |
430 | ||
13af4b93 DW |
431 | seq = ext2fs_cpu_to_be32(sequence); |
432 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq)); | |
433 | csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize); | |
d9200380 | 434 | |
86f3b6cf | 435 | if (jfs_has_feature_csum3(j)) |
c8ca2397 | 436 | return get_be32(&tag3->t_checksum) == csum32; |
38d5adf3 | 437 | |
c8ca2397 | 438 | return get_be16(&tag->t_checksum) == (csum32 & 0xFFFF); |
d9200380 DW |
439 | } |
440 | ||
8cf93332 TT |
441 | static int do_one_pass(journal_t *journal, |
442 | struct recovery_info *info, enum passtype pass) | |
0e8a9560 | 443 | { |
3b5386dc | 444 | unsigned int first_commit_ID, next_commit_ID; |
13af4b93 | 445 | unsigned long next_log_block; |
3b5386dc | 446 | int err, success = 0; |
0e8a9560 | 447 | journal_superblock_t * sb; |
e5ea6b14 | 448 | journal_header_t * tmp; |
3b5386dc | 449 | struct buffer_head * bh; |
0e8a9560 TT |
450 | unsigned int sequence; |
451 | int blocktype; | |
185c4aea TT |
452 | int tag_bytes = journal_tag_bytes(journal); |
453 | __u32 crc32_sum = ~0; /* Transactional Checksums */ | |
5c842851 | 454 | int descr_csum_size = 0; |
bf93a968 | 455 | int block_error = 0; |
e5ea6b14 | 456 | |
e5ea6b14 | 457 | /* |
3b5386dc TT |
458 | * First thing is to establish what we expect to find in the log |
459 | * (in terms of transaction IDs), and where (in terms of log | |
e5ea6b14 | 460 | * block offsets): query the superblock. |
3b5386dc TT |
461 | */ |
462 | ||
0e8a9560 | 463 | sb = journal->j_superblock; |
13af4b93 DW |
464 | next_commit_ID = ext2fs_be32_to_cpu(sb->s_sequence); |
465 | next_log_block = ext2fs_be32_to_cpu(sb->s_start); | |
3b5386dc TT |
466 | |
467 | first_commit_ID = next_commit_ID; | |
0e8a9560 TT |
468 | if (pass == PASS_SCAN) |
469 | info->start_transaction = first_commit_ID; | |
8cf93332 TT |
470 | |
471 | jbd_debug(1, "Starting recovery pass %d\n", pass); | |
472 | ||
3b5386dc TT |
473 | /* |
474 | * Now we walk through the log, transaction by transaction, | |
475 | * making sure that each transaction has a commit block in the | |
476 | * expected place. Each complete transaction gets replayed back | |
e5ea6b14 | 477 | * into the main filesystem. |
3b5386dc TT |
478 | */ |
479 | ||
0e8a9560 TT |
480 | while (1) { |
481 | int flags; | |
482 | char * tagp; | |
483 | journal_block_tag_t * tag; | |
484 | struct buffer_head * obh; | |
485 | struct buffer_head * nbh; | |
e5ea6b14 TT |
486 | |
487 | cond_resched(); | |
488 | ||
0e8a9560 TT |
489 | /* If we already know where to stop the log traversal, |
490 | * check right now that we haven't gone past the end of | |
491 | * the log. */ | |
e5ea6b14 | 492 | |
0e8a9560 TT |
493 | if (pass != PASS_SCAN) |
494 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
495 | break; | |
8cf93332 | 496 | |
13af4b93 | 497 | jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", |
3b5386dc | 498 | next_commit_ID, next_log_block, journal->j_last); |
3b5386dc | 499 | |
0e8a9560 TT |
500 | /* Skip over each chunk of the transaction looking |
501 | * either the next descriptor block or the final commit | |
502 | * record. */ | |
e5ea6b14 | 503 | |
13af4b93 | 504 | jbd_debug(3, "JBD2: checking block %ld\n", next_log_block); |
0e8a9560 TT |
505 | err = jread(&bh, journal, next_log_block); |
506 | if (err) | |
507 | goto failed; | |
8cf93332 | 508 | |
0e8a9560 TT |
509 | next_log_block++; |
510 | wrap(journal, next_log_block); | |
e5ea6b14 TT |
511 | |
512 | /* What kind of buffer is it? | |
513 | * | |
0e8a9560 TT |
514 | * If it is a descriptor block, check that it has the |
515 | * expected sequence number. Otherwise, we're all done | |
516 | * here. */ | |
3b5386dc | 517 | |
8cf93332 | 518 | tmp = (journal_header_t *)bh->b_data; |
e5ea6b14 | 519 | |
13af4b93 | 520 | if (tmp->h_magic != ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER)) { |
3b5386dc TT |
521 | brelse(bh); |
522 | break; | |
523 | } | |
8cf93332 | 524 | |
13af4b93 DW |
525 | blocktype = ext2fs_be32_to_cpu(tmp->h_blocktype); |
526 | sequence = ext2fs_be32_to_cpu(tmp->h_sequence); | |
e5ea6b14 | 527 | jbd_debug(3, "Found magic %d, sequence %d\n", |
0e8a9560 | 528 | blocktype, sequence); |
e5ea6b14 | 529 | |
0e8a9560 TT |
530 | if (sequence != next_commit_ID) { |
531 | brelse(bh); | |
532 | break; | |
533 | } | |
e5ea6b14 | 534 | |
0e8a9560 TT |
535 | /* OK, we have a valid descriptor block which matches |
536 | * all of the sequence number checks. What are we going | |
537 | * to do with it? That depends on the pass... */ | |
8cf93332 | 538 | |
0e8a9560 TT |
539 | switch(blocktype) { |
540 | case JFS_DESCRIPTOR_BLOCK: | |
5c842851 | 541 | /* Verify checksum first */ |
38d5adf3 | 542 | if (journal_has_csum_v2or3(journal)) |
5c842851 DW |
543 | descr_csum_size = |
544 | sizeof(struct journal_block_tail); | |
545 | if (descr_csum_size > 0 && | |
546 | !jbd2_descr_block_csum_verify(journal, | |
547 | bh->b_data)) { | |
b4f02c9f | 548 | err = -EFSBADCRC; |
c10700f9 | 549 | brelse(bh); |
5c842851 DW |
550 | goto failed; |
551 | } | |
552 | ||
0e8a9560 | 553 | /* If it is a valid descriptor block, replay it |
185c4aea TT |
554 | * in pass REPLAY; if journal_checksums enabled, then |
555 | * calculate checksums in PASS_SCAN, otherwise, | |
556 | * just skip over the blocks it describes. */ | |
0e8a9560 | 557 | if (pass != PASS_REPLAY) { |
185c4aea | 558 | if (pass == PASS_SCAN && |
86f3b6cf | 559 | jfs_has_feature_checksum(journal) && |
185c4aea TT |
560 | !info->end_transaction) { |
561 | if (calc_chksums(journal, bh, | |
562 | &next_log_block, | |
563 | &crc32_sum)) { | |
13af4b93 | 564 | put_bh(bh); |
185c4aea TT |
565 | break; |
566 | } | |
13af4b93 | 567 | put_bh(bh); |
185c4aea TT |
568 | continue; |
569 | } | |
570 | next_log_block += count_tags(journal, bh); | |
0e8a9560 | 571 | wrap(journal, next_log_block); |
13af4b93 | 572 | put_bh(bh); |
0e8a9560 | 573 | continue; |
3b5386dc | 574 | } |
0e8a9560 | 575 | |
3b5386dc TT |
576 | /* A descriptor block: we can now write all of |
577 | * the data blocks. Yay, useful work is finally | |
578 | * getting done here! */ | |
579 | ||
580 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
185c4aea | 581 | while ((tagp - bh->b_data + tag_bytes) |
5c842851 | 582 | <= journal->j_blocksize - descr_csum_size) { |
13af4b93 | 583 | unsigned long io_block; |
8cf93332 | 584 | |
3b5386dc | 585 | tag = (journal_block_tag_t *) tagp; |
c8ca2397 | 586 | flags = get_be16(&tag->t_flags); |
e5ea6b14 | 587 | |
0e8a9560 | 588 | io_block = next_log_block++; |
3b5386dc | 589 | wrap(journal, next_log_block); |
0e8a9560 | 590 | err = jread(&obh, journal, io_block); |
3b5386dc TT |
591 | if (err) { |
592 | /* Recover what we can, but | |
593 | * report failure at the end. */ | |
594 | success = err; | |
13af4b93 DW |
595 | printk(KERN_ERR |
596 | "JBD2: IO error %d recovering " | |
597 | "block %ld in log\n", | |
0e8a9560 | 598 | err, io_block); |
3b5386dc | 599 | } else { |
3b693d0b | 600 | unsigned long long blocknr; |
e5ea6b14 | 601 | |
3b5386dc | 602 | J_ASSERT(obh != NULL); |
38d5adf3 | 603 | blocknr = read_tag_block(journal, |
3b693d0b | 604 | tag); |
0e8a9560 TT |
605 | |
606 | /* If the block has been | |
607 | * revoked, then we're all done | |
608 | * here. */ | |
609 | if (journal_test_revoke | |
e5ea6b14 | 610 | (journal, blocknr, |
0e8a9560 TT |
611 | next_commit_ID)) { |
612 | brelse(obh); | |
613 | ++info->nr_revoke_hits; | |
614 | goto skip_write; | |
615 | } | |
e5ea6b14 | 616 | |
d9200380 DW |
617 | /* Look for block corruption */ |
618 | if (!jbd2_block_tag_csum_verify( | |
619 | journal, tag, obh->b_data, | |
13af4b93 | 620 | ext2fs_be32_to_cpu(tmp->h_sequence))) { |
d9200380 | 621 | brelse(obh); |
b4f02c9f | 622 | success = -EFSBADCRC; |
13af4b93 | 623 | printk(KERN_ERR "JBD2: Invalid " |
d9200380 | 624 | "checksum recovering " |
13af4b93 | 625 | "block %llu in log\n", |
d9200380 | 626 | blocknr); |
bf93a968 DW |
627 | block_error = 1; |
628 | goto skip_write; | |
d9200380 DW |
629 | } |
630 | ||
0e8a9560 TT |
631 | /* Find a buffer for the new |
632 | * data being restored */ | |
e5ea6b14 TT |
633 | nbh = __getblk(journal->j_fs_dev, |
634 | blocknr, | |
635 | journal->j_blocksize); | |
3b5386dc | 636 | if (nbh == NULL) { |
e5ea6b14 | 637 | printk(KERN_ERR |
13af4b93 | 638 | "JBD2: Out of memory " |
3b5386dc TT |
639 | "during recovery.\n"); |
640 | err = -ENOMEM; | |
641 | brelse(bh); | |
642 | brelse(obh); | |
643 | goto failed; | |
644 | } | |
645 | ||
3de085dd | 646 | lock_buffer(nbh); |
8cf93332 TT |
647 | memcpy(nbh->b_data, obh->b_data, |
648 | journal->j_blocksize); | |
3b5386dc | 649 | if (flags & JFS_FLAG_ESCAPE) { |
478360f5 TT |
650 | __u32 magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER); |
651 | memcpy(nbh->b_data, &magic, | |
652 | sizeof(magic)); | |
3b5386dc | 653 | } |
8cf93332 TT |
654 | |
655 | BUFFER_TRACE(nbh, "marking dirty"); | |
3de085dd | 656 | set_buffer_uptodate(nbh); |
8cf93332 TT |
657 | mark_buffer_dirty(nbh); |
658 | BUFFER_TRACE(nbh, "marking uptodate"); | |
0e8a9560 | 659 | ++info->nr_replays; |
53ef44c4 | 660 | /* ll_rw_block(WRITE, 1, &nbh); */ |
3de085dd | 661 | unlock_buffer(nbh); |
3b5386dc TT |
662 | brelse(obh); |
663 | brelse(nbh); | |
664 | } | |
e5ea6b14 | 665 | |
0e8a9560 | 666 | skip_write: |
185c4aea | 667 | tagp += tag_bytes; |
3b5386dc TT |
668 | if (!(flags & JFS_FLAG_SAME_UUID)) |
669 | tagp += 16; | |
670 | ||
671 | if (flags & JFS_FLAG_LAST_TAG) | |
672 | break; | |
0e8a9560 | 673 | } |
e5ea6b14 | 674 | |
0e8a9560 TT |
675 | brelse(bh); |
676 | continue; | |
8cf93332 | 677 | |
0e8a9560 | 678 | case JFS_COMMIT_BLOCK: |
185c4aea TT |
679 | /* How to differentiate between interrupted commit |
680 | * and journal corruption ? | |
681 | * | |
682 | * {nth transaction} | |
683 | * Checksum Verification Failed | |
684 | * | | |
685 | * ____________________ | |
686 | * | | | |
687 | * async_commit sync_commit | |
688 | * | | | |
689 | * | GO TO NEXT "Journal Corruption" | |
690 | * | TRANSACTION | |
691 | * | | |
055866d8 | 692 | * {(n+1)th transaction} |
185c4aea TT |
693 | * | |
694 | * _______|______________ | |
695 | * | | | |
696 | * Commit block found Commit block not found | |
697 | * | | | |
698 | * "Journal Corruption" | | |
699 | * _____________|_________ | |
700 | * | | | |
701 | * nth trans corrupt OR nth trans | |
702 | * and (n+1)th interrupted interrupted | |
703 | * before commit block | |
704 | * could reach the disk. | |
705 | * (Cannot find the difference in above | |
706 | * mentioned conditions. Hence assume | |
707 | * "Interrupted Commit".) | |
708 | */ | |
709 | ||
710 | /* Found an expected commit block: if checksums | |
711 | * are present verify them in PASS_SCAN; else not | |
712 | * much to do other than move on to the next sequence | |
0e8a9560 | 713 | * number. */ |
185c4aea | 714 | if (pass == PASS_SCAN && |
86f3b6cf | 715 | jfs_has_feature_checksum(journal)) { |
185c4aea TT |
716 | int chksum_err, chksum_seen; |
717 | struct commit_header *cbh = | |
718 | (struct commit_header *)bh->b_data; | |
719 | unsigned found_chksum = | |
13af4b93 | 720 | ext2fs_be32_to_cpu(cbh->h_chksum[0]); |
185c4aea TT |
721 | |
722 | chksum_err = chksum_seen = 0; | |
723 | ||
185c4aea TT |
724 | if (info->end_transaction) { |
725 | journal->j_failed_commit = | |
726 | info->end_transaction; | |
727 | brelse(bh); | |
728 | break; | |
729 | } | |
730 | ||
731 | if (crc32_sum == found_chksum && | |
13af4b93 | 732 | cbh->h_chksum_type == JFS_CRC32_CHKSUM && |
185c4aea | 733 | cbh->h_chksum_size == |
13af4b93 | 734 | JFS_CRC32_CHKSUM_SIZE) |
185c4aea TT |
735 | chksum_seen = 1; |
736 | else if (!(cbh->h_chksum_type == 0 && | |
737 | cbh->h_chksum_size == 0 && | |
738 | found_chksum == 0 && | |
739 | !chksum_seen)) | |
740 | /* | |
741 | * If fs is mounted using an old kernel and then | |
742 | * kernel with journal_chksum is used then we | |
743 | * get a situation where the journal flag has | |
744 | * checksum flag set but checksums are not | |
745 | * present i.e chksum = 0, in the individual | |
746 | * commit blocks. | |
747 | * Hence to avoid checksum failures, in this | |
748 | * situation, this extra check is added. | |
749 | */ | |
750 | chksum_err = 1; | |
751 | ||
752 | if (chksum_err) { | |
753 | info->end_transaction = next_commit_ID; | |
13af4b93 | 754 | |
86f3b6cf | 755 | if (!jfs_has_feature_async_commit(journal)){ |
185c4aea TT |
756 | journal->j_failed_commit = |
757 | next_commit_ID; | |
758 | brelse(bh); | |
759 | break; | |
760 | } | |
761 | } | |
762 | crc32_sum = ~0; | |
763 | } | |
721e065c DW |
764 | if (pass == PASS_SCAN && |
765 | !jbd2_commit_block_csum_verify(journal, | |
766 | bh->b_data)) { | |
767 | info->end_transaction = next_commit_ID; | |
768 | ||
86f3b6cf | 769 | if (!jfs_has_feature_async_commit(journal)) { |
721e065c DW |
770 | journal->j_failed_commit = |
771 | next_commit_ID; | |
772 | brelse(bh); | |
773 | break; | |
774 | } | |
775 | } | |
0e8a9560 TT |
776 | brelse(bh); |
777 | next_commit_ID++; | |
778 | continue; | |
3b5386dc | 779 | |
0e8a9560 TT |
780 | case JFS_REVOKE_BLOCK: |
781 | /* If we aren't in the REVOKE pass, then we can | |
782 | * just skip over this block. */ | |
783 | if (pass != PASS_REVOKE) { | |
784 | brelse(bh); | |
785 | continue; | |
786 | } | |
3b5386dc | 787 | |
8cf93332 | 788 | err = scan_revoke_records(journal, bh, |
0e8a9560 | 789 | next_commit_ID, info); |
3b5386dc | 790 | brelse(bh); |
0e8a9560 TT |
791 | if (err) |
792 | goto failed; | |
793 | continue; | |
794 | ||
795 | default: | |
8cf93332 | 796 | jbd_debug(3, "Unrecognised magic %d, end of scan.\n", |
0e8a9560 | 797 | blocktype); |
2aa362f5 | 798 | brelse(bh); |
0e8a9560 TT |
799 | goto done; |
800 | } | |
3b5386dc | 801 | } |
3b5386dc | 802 | |
0e8a9560 | 803 | done: |
e5ea6b14 | 804 | /* |
0e8a9560 TT |
805 | * We broke out of the log scan loop: either we came to the |
806 | * known end of the log or we found an unexpected block in the | |
807 | * log. If the latter happened, then we know that the "current" | |
808 | * transaction marks the end of the valid log. | |
809 | */ | |
e5ea6b14 | 810 | |
185c4aea TT |
811 | if (pass == PASS_SCAN) { |
812 | if (!info->end_transaction) | |
813 | info->end_transaction = next_commit_ID; | |
814 | } else { | |
0e8a9560 TT |
815 | /* It's really bad news if different passes end up at |
816 | * different places (but possible due to IO errors). */ | |
817 | if (info->end_transaction != next_commit_ID) { | |
13af4b93 | 818 | printk(KERN_ERR "JBD2: recovery pass %d ended at " |
0e8a9560 TT |
819 | "transaction %u, expected %u\n", |
820 | pass, next_commit_ID, info->end_transaction); | |
821 | if (!success) | |
822 | success = -EIO; | |
823 | } | |
824 | } | |
bf93a968 DW |
825 | if (block_error && success == 0) |
826 | success = -EIO; | |
0e8a9560 | 827 | return success; |
3b5386dc | 828 | |
0e8a9560 | 829 | failed: |
3b5386dc TT |
830 | return err; |
831 | } | |
0e8a9560 | 832 | |
d35ed158 DW |
833 | static int jbd2_revoke_block_csum_verify(journal_t *j, |
834 | void *buf) | |
835 | { | |
836 | struct journal_revoke_tail *tail; | |
13af4b93 DW |
837 | __u32 provided; |
838 | __u32 calculated; | |
d35ed158 | 839 | |
38d5adf3 | 840 | if (!journal_has_csum_v2or3(j)) |
d35ed158 DW |
841 | return 1; |
842 | ||
478360f5 | 843 | tail = (struct journal_revoke_tail *)((char *)buf + j->j_blocksize - |
d35ed158 DW |
844 | sizeof(struct journal_revoke_tail)); |
845 | provided = tail->r_checksum; | |
846 | tail->r_checksum = 0; | |
13af4b93 | 847 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
d35ed158 DW |
848 | tail->r_checksum = provided; |
849 | ||
13af4b93 | 850 | return provided == ext2fs_cpu_to_be32(calculated); |
d35ed158 | 851 | } |
0e8a9560 TT |
852 | |
853 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ | |
854 | ||
e5ea6b14 | 855 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, |
0e8a9560 TT |
856 | tid_t sequence, struct recovery_info *info) |
857 | { | |
858 | journal_revoke_header_t *header; | |
859 | int offset, max; | |
478360f5 | 860 | unsigned csum_size = 0; |
04c66cb2 | 861 | __u32 rcount; |
79a4dddb | 862 | int record_len = 4; |
8cf93332 | 863 | |
0e8a9560 TT |
864 | header = (journal_revoke_header_t *) bh->b_data; |
865 | offset = sizeof(journal_revoke_header_t); | |
04c66cb2 | 866 | rcount = ext2fs_be32_to_cpu(header->r_count); |
e5ea6b14 | 867 | |
d35ed158 | 868 | if (!jbd2_revoke_block_csum_verify(journal, header)) |
b4f02c9f | 869 | return -EFSBADCRC; |
d35ed158 | 870 | |
04c66cb2 DW |
871 | if (journal_has_csum_v2or3(journal)) |
872 | csum_size = sizeof(struct journal_revoke_tail); | |
873 | if (rcount > journal->j_blocksize - csum_size) | |
874 | return -EINVAL; | |
875 | max = rcount; | |
876 | ||
86f3b6cf | 877 | if (jfs_has_feature_64bit(journal)) |
79a4dddb DW |
878 | record_len = 8; |
879 | ||
13af4b93 | 880 | while (offset + record_len <= max) { |
3b693d0b | 881 | unsigned long long blocknr; |
0e8a9560 | 882 | int err; |
e5ea6b14 | 883 | |
13af4b93 DW |
884 | if (record_len == 4) |
885 | blocknr = ext2fs_be32_to_cpu(* ((__u32 *) (bh->b_data+offset))); | |
886 | else | |
887 | blocknr = ext2fs_be64_to_cpu(* ((__u64 *) (bh->b_data+offset))); | |
79a4dddb | 888 | offset += record_len; |
0e8a9560 TT |
889 | err = journal_set_revoke(journal, blocknr, sequence); |
890 | if (err) | |
891 | return err; | |
892 | ++info->nr_revokes; | |
893 | } | |
894 | return 0; | |
895 | } |