]>
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 | ||
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 | 141 | if (offset >= journal->j_maxlen) { |
13af4b93 | 142 | printk(KERN_ERR "JBD2: corrupted journal superblock\n"); |
b4f02c9f | 143 | return -EFSCORRUPTED; |
e5ea6b14 TT |
144 | } |
145 | ||
8cf93332 TT |
146 | err = journal_bmap(journal, offset, &blocknr); |
147 | ||
148 | if (err) { | |
13af4b93 | 149 | printk(KERN_ERR "JBD2: 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)) { |
13af4b93 | 167 | printk(KERN_ERR "JBD2: 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 | ||
5c842851 DW |
177 | static int jbd2_descr_block_csum_verify(journal_t *j, |
178 | void *buf) | |
179 | { | |
180 | struct journal_block_tail *tail; | |
13af4b93 DW |
181 | __u32 provided; |
182 | __u32 calculated; | |
5c842851 | 183 | |
38d5adf3 | 184 | if (!journal_has_csum_v2or3(j)) |
5c842851 DW |
185 | return 1; |
186 | ||
13af4b93 | 187 | tail = (struct journal_block_tail *)(buf + j->j_blocksize - |
5c842851 DW |
188 | sizeof(struct journal_block_tail)); |
189 | provided = tail->t_checksum; | |
190 | tail->t_checksum = 0; | |
13af4b93 | 191 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
5c842851 DW |
192 | tail->t_checksum = provided; |
193 | ||
13af4b93 | 194 | return provided == ext2fs_cpu_to_be32(calculated); |
5c842851 | 195 | } |
3b5386dc TT |
196 | |
197 | /* | |
198 | * Count the number of in-use tags in a journal descriptor block. | |
199 | */ | |
200 | ||
185c4aea | 201 | static int count_tags(journal_t *journal, struct buffer_head *bh) |
3b5386dc TT |
202 | { |
203 | char * tagp; | |
204 | journal_block_tag_t * tag; | |
185c4aea TT |
205 | int nr = 0, size = journal->j_blocksize; |
206 | int tag_bytes = journal_tag_bytes(journal); | |
8cf93332 | 207 | |
38d5adf3 | 208 | if (journal_has_csum_v2or3(journal)) |
5c842851 DW |
209 | size -= sizeof(struct journal_block_tail); |
210 | ||
3b5386dc | 211 | tagp = &bh->b_data[sizeof(journal_header_t)]; |
8cf93332 | 212 | |
185c4aea | 213 | while ((tagp - bh->b_data + tag_bytes) <= size) { |
3b5386dc | 214 | tag = (journal_block_tag_t *) tagp; |
8cf93332 | 215 | |
3b5386dc | 216 | nr++; |
185c4aea | 217 | tagp += tag_bytes; |
13af4b93 | 218 | if (!(tag->t_flags & ext2fs_cpu_to_be16(JFS_FLAG_SAME_UUID))) |
3b5386dc TT |
219 | tagp += 16; |
220 | ||
13af4b93 | 221 | if (tag->t_flags & ext2fs_cpu_to_be16(JFS_FLAG_LAST_TAG)) |
3b5386dc TT |
222 | break; |
223 | } | |
8cf93332 | 224 | |
3b5386dc TT |
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 | ||
3de085dd | 236 | /** |
e5ea6b14 | 237 | * journal_recover - recovers a on-disk journal |
3de085dd | 238 | * @journal: the journal to recover |
3b5386dc TT |
239 | * |
240 | * The primary function for recovering the log contents when mounting a | |
e5ea6b14 TT |
241 | * journaled device. |
242 | * | |
0e8a9560 TT |
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 | |
e5ea6b14 | 246 | * in the log. |
3b5386dc | 247 | */ |
3b5386dc TT |
248 | int journal_recover(journal_t *journal) |
249 | { | |
13af4b93 | 250 | int err, err2; |
0e8a9560 TT |
251 | journal_superblock_t * sb; |
252 | ||
53ef44c4 | 253 | struct recovery_info info; |
e5ea6b14 | 254 | |
1f735038 | 255 | memset(&info, 0, sizeof(info)); |
0e8a9560 | 256 | sb = journal->j_superblock; |
e5ea6b14 TT |
257 | |
258 | /* | |
0e8a9560 TT |
259 | * The journal superblock's s_start field (the current log head) |
260 | * is always zero if, and only if, the journal was cleanly | |
e5ea6b14 | 261 | * unmounted. |
0e8a9560 TT |
262 | */ |
263 | ||
264 | if (!sb->s_start) { | |
8cf93332 | 265 | jbd_debug(1, "No recovery required, last transaction %d\n", |
13af4b93 DW |
266 | ext2fs_be32_to_cpu(sb->s_sequence)); |
267 | journal->j_transaction_sequence = ext2fs_be32_to_cpu(sb->s_sequence) + 1; | |
0e8a9560 TT |
268 | return 0; |
269 | } | |
e5ea6b14 | 270 | |
0e8a9560 TT |
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 | ||
13af4b93 | 277 | jbd_debug(1, "JBD2: recovery, exit status %d, " |
0e8a9560 TT |
278 | "recovered transactions %u to %u\n", |
279 | err, info.start_transaction, info.end_transaction); | |
13af4b93 | 280 | jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n", |
0e8a9560 TT |
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; | |
e5ea6b14 | 286 | |
0e8a9560 | 287 | journal_clear_revoke(journal); |
13af4b93 DW |
288 | err2 = sync_blockdev(journal->j_fs_dev); |
289 | if (!err) | |
290 | err = err2; | |
291 | /* Make sure all replayed data is on permanent storage */ | |
292 | if (journal->j_flags & JFS_BARRIER) { | |
293 | err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL); | |
294 | if (!err) | |
295 | err = err2; | |
296 | } | |
0e8a9560 TT |
297 | return err; |
298 | } | |
299 | ||
3de085dd | 300 | /** |
e5ea6b14 | 301 | * journal_skip_recovery - Start journal and wipe exiting records |
3de085dd | 302 | * @journal: journal to startup |
1f735038 TT |
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 | |
e5ea6b14 | 306 | * caller has evidence that it is out of date). |
3de085dd | 307 | * This function does'nt appear to be exorted.. |
1f735038 TT |
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 | |
e5ea6b14 | 311 | * the journal transaction sequence numbers to the next unused ID. |
1f735038 | 312 | */ |
1f735038 TT |
313 | int journal_skip_recovery(journal_t *journal) |
314 | { | |
315 | int err; | |
13af4b93 | 316 | |
1f735038 | 317 | struct recovery_info info; |
e5ea6b14 | 318 | |
8cf93332 | 319 | memset (&info, 0, sizeof(info)); |
e5ea6b14 | 320 | |
1f735038 TT |
321 | err = do_one_pass(journal, &info, PASS_SCAN); |
322 | ||
323 | if (err) { | |
13af4b93 | 324 | printk(KERN_ERR "JBD2: error %d scanning journal\n", err); |
1f735038 TT |
325 | ++journal->j_transaction_sequence; |
326 | } else { | |
13af4b93 | 327 | #ifdef CONFIG_JFS_DEBUG |
97f168b6 | 328 | int dropped = info.end_transaction - |
13af4b93 | 329 | ext2fs_be32_to_cpu(journal->j_superblock->s_sequence); |
e5ea6b14 | 330 | jbd_debug(1, |
13af4b93 | 331 | "JBD2: ignoring %d transaction%s from the journal.\n", |
1f735038 | 332 | dropped, (dropped == 1) ? "" : "s"); |
13af4b93 | 333 | #endif |
1f735038 TT |
334 | journal->j_transaction_sequence = ++info.end_transaction; |
335 | } | |
336 | ||
337 | journal->j_tail = 0; | |
1f735038 TT |
338 | return err; |
339 | } | |
340 | ||
38d5adf3 DW |
341 | static inline unsigned long long read_tag_block(journal_t *journal, |
342 | journal_block_tag_t *tag) | |
185c4aea | 343 | { |
13af4b93 | 344 | unsigned long long block = ext2fs_be32_to_cpu(tag->t_blocknr); |
86f3b6cf | 345 | if (jfs_has_feature_64bit(journal)) |
13af4b93 | 346 | block |= (u64)ext2fs_be32_to_cpu(tag->t_blocknr_high) << 32; |
185c4aea TT |
347 | return block; |
348 | } | |
349 | ||
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, | |
13af4b93 | 355 | unsigned long *next_log_block, __u32 *crc32_sum) |
185c4aea TT |
356 | { |
357 | int i, num_blks, err; | |
13af4b93 | 358 | unsigned long io_block; |
185c4aea TT |
359 | struct buffer_head *obh; |
360 | ||
361 | num_blks = count_tags(journal, bh); | |
362 | /* Calculate checksum of the descriptor block. */ | |
13af4b93 | 363 | *crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size); |
185c4aea TT |
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) { | |
13af4b93 DW |
370 | printk(KERN_ERR "JBD2: IO error %d recovering block " |
371 | "%lu in log\n", err, io_block); | |
185c4aea TT |
372 | return 1; |
373 | } else { | |
13af4b93 DW |
374 | *crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data, |
375 | obh->b_size); | |
185c4aea | 376 | } |
13af4b93 | 377 | put_bh(obh); |
185c4aea TT |
378 | } |
379 | return 0; | |
380 | } | |
381 | ||
721e065c DW |
382 | static int jbd2_commit_block_csum_verify(journal_t *j, void *buf) |
383 | { | |
384 | struct commit_header *h; | |
13af4b93 DW |
385 | __u32 provided; |
386 | __u32 calculated; | |
721e065c | 387 | |
38d5adf3 | 388 | if (!journal_has_csum_v2or3(j)) |
721e065c DW |
389 | return 1; |
390 | ||
391 | h = buf; | |
392 | provided = h->h_chksum[0]; | |
393 | h->h_chksum[0] = 0; | |
13af4b93 | 394 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
721e065c DW |
395 | h->h_chksum[0] = provided; |
396 | ||
13af4b93 | 397 | return provided == ext2fs_cpu_to_be32(calculated); |
721e065c DW |
398 | } |
399 | ||
d9200380 DW |
400 | static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag, |
401 | void *buf, __u32 sequence) | |
402 | { | |
38d5adf3 | 403 | journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag; |
13af4b93 DW |
404 | __u32 csum32; |
405 | __u32 seq; | |
d9200380 | 406 | |
38d5adf3 | 407 | if (!journal_has_csum_v2or3(j)) |
d9200380 DW |
408 | return 1; |
409 | ||
13af4b93 DW |
410 | seq = ext2fs_cpu_to_be32(sequence); |
411 | csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq)); | |
412 | csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize); | |
d9200380 | 413 | |
86f3b6cf | 414 | if (jfs_has_feature_csum3(j)) |
38d5adf3 DW |
415 | return tag3->t_checksum == ext2fs_cpu_to_be32(csum32); |
416 | ||
13af4b93 | 417 | return tag->t_checksum == ext2fs_cpu_to_be16(csum32); |
d9200380 DW |
418 | } |
419 | ||
8cf93332 TT |
420 | static int do_one_pass(journal_t *journal, |
421 | struct recovery_info *info, enum passtype pass) | |
0e8a9560 | 422 | { |
3b5386dc | 423 | unsigned int first_commit_ID, next_commit_ID; |
13af4b93 | 424 | unsigned long next_log_block; |
3b5386dc | 425 | int err, success = 0; |
0e8a9560 | 426 | journal_superblock_t * sb; |
e5ea6b14 | 427 | journal_header_t * tmp; |
3b5386dc | 428 | struct buffer_head * bh; |
0e8a9560 TT |
429 | unsigned int sequence; |
430 | int blocktype; | |
185c4aea TT |
431 | int tag_bytes = journal_tag_bytes(journal); |
432 | __u32 crc32_sum = ~0; /* Transactional Checksums */ | |
5c842851 | 433 | int descr_csum_size = 0; |
bf93a968 | 434 | int block_error = 0; |
e5ea6b14 | 435 | |
e5ea6b14 | 436 | /* |
3b5386dc TT |
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 | |
e5ea6b14 | 439 | * block offsets): query the superblock. |
3b5386dc TT |
440 | */ |
441 | ||
0e8a9560 | 442 | sb = journal->j_superblock; |
13af4b93 DW |
443 | next_commit_ID = ext2fs_be32_to_cpu(sb->s_sequence); |
444 | next_log_block = ext2fs_be32_to_cpu(sb->s_start); | |
3b5386dc TT |
445 | |
446 | first_commit_ID = next_commit_ID; | |
0e8a9560 TT |
447 | if (pass == PASS_SCAN) |
448 | info->start_transaction = first_commit_ID; | |
8cf93332 TT |
449 | |
450 | jbd_debug(1, "Starting recovery pass %d\n", pass); | |
451 | ||
3b5386dc TT |
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 | |
e5ea6b14 | 456 | * into the main filesystem. |
3b5386dc TT |
457 | */ |
458 | ||
0e8a9560 TT |
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; | |
e5ea6b14 TT |
465 | |
466 | cond_resched(); | |
467 | ||
0e8a9560 TT |
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. */ | |
e5ea6b14 | 471 | |
0e8a9560 TT |
472 | if (pass != PASS_SCAN) |
473 | if (tid_geq(next_commit_ID, info->end_transaction)) | |
474 | break; | |
8cf93332 | 475 | |
13af4b93 | 476 | jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n", |
3b5386dc | 477 | next_commit_ID, next_log_block, journal->j_last); |
3b5386dc | 478 | |
0e8a9560 TT |
479 | /* Skip over each chunk of the transaction looking |
480 | * either the next descriptor block or the final commit | |
481 | * record. */ | |
e5ea6b14 | 482 | |
13af4b93 | 483 | jbd_debug(3, "JBD2: checking block %ld\n", next_log_block); |
0e8a9560 TT |
484 | err = jread(&bh, journal, next_log_block); |
485 | if (err) | |
486 | goto failed; | |
8cf93332 | 487 | |
0e8a9560 TT |
488 | next_log_block++; |
489 | wrap(journal, next_log_block); | |
e5ea6b14 TT |
490 | |
491 | /* What kind of buffer is it? | |
492 | * | |
0e8a9560 TT |
493 | * If it is a descriptor block, check that it has the |
494 | * expected sequence number. Otherwise, we're all done | |
495 | * here. */ | |
3b5386dc | 496 | |
8cf93332 | 497 | tmp = (journal_header_t *)bh->b_data; |
e5ea6b14 | 498 | |
13af4b93 | 499 | if (tmp->h_magic != ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER)) { |
3b5386dc TT |
500 | brelse(bh); |
501 | break; | |
502 | } | |
8cf93332 | 503 | |
13af4b93 DW |
504 | blocktype = ext2fs_be32_to_cpu(tmp->h_blocktype); |
505 | sequence = ext2fs_be32_to_cpu(tmp->h_sequence); | |
e5ea6b14 | 506 | jbd_debug(3, "Found magic %d, sequence %d\n", |
0e8a9560 | 507 | blocktype, sequence); |
e5ea6b14 | 508 | |
0e8a9560 TT |
509 | if (sequence != next_commit_ID) { |
510 | brelse(bh); | |
511 | break; | |
512 | } | |
e5ea6b14 | 513 | |
0e8a9560 TT |
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... */ | |
8cf93332 | 517 | |
0e8a9560 TT |
518 | switch(blocktype) { |
519 | case JFS_DESCRIPTOR_BLOCK: | |
5c842851 | 520 | /* Verify checksum first */ |
38d5adf3 | 521 | if (journal_has_csum_v2or3(journal)) |
5c842851 DW |
522 | descr_csum_size = |
523 | sizeof(struct journal_block_tail); | |
524 | if (descr_csum_size > 0 && | |
525 | !jbd2_descr_block_csum_verify(journal, | |
526 | bh->b_data)) { | |
b4f02c9f | 527 | err = -EFSBADCRC; |
c10700f9 | 528 | brelse(bh); |
5c842851 DW |
529 | goto failed; |
530 | } | |
531 | ||
0e8a9560 | 532 | /* If it is a valid descriptor block, replay it |
185c4aea TT |
533 | * in pass REPLAY; if journal_checksums enabled, then |
534 | * calculate checksums in PASS_SCAN, otherwise, | |
535 | * just skip over the blocks it describes. */ | |
0e8a9560 | 536 | if (pass != PASS_REPLAY) { |
185c4aea | 537 | if (pass == PASS_SCAN && |
86f3b6cf | 538 | jfs_has_feature_checksum(journal) && |
185c4aea TT |
539 | !info->end_transaction) { |
540 | if (calc_chksums(journal, bh, | |
541 | &next_log_block, | |
542 | &crc32_sum)) { | |
13af4b93 | 543 | put_bh(bh); |
185c4aea TT |
544 | break; |
545 | } | |
13af4b93 | 546 | put_bh(bh); |
185c4aea TT |
547 | continue; |
548 | } | |
549 | next_log_block += count_tags(journal, bh); | |
0e8a9560 | 550 | wrap(journal, next_log_block); |
13af4b93 | 551 | put_bh(bh); |
0e8a9560 | 552 | continue; |
3b5386dc | 553 | } |
0e8a9560 | 554 | |
3b5386dc TT |
555 | /* A descriptor block: we can now write all of |
556 | * the data blocks. Yay, useful work is finally | |
557 | * getting done here! */ | |
558 | ||
559 | tagp = &bh->b_data[sizeof(journal_header_t)]; | |
185c4aea | 560 | while ((tagp - bh->b_data + tag_bytes) |
5c842851 | 561 | <= journal->j_blocksize - descr_csum_size) { |
13af4b93 | 562 | unsigned long io_block; |
8cf93332 | 563 | |
3b5386dc | 564 | tag = (journal_block_tag_t *) tagp; |
13af4b93 | 565 | flags = ext2fs_be16_to_cpu(tag->t_flags); |
e5ea6b14 | 566 | |
0e8a9560 | 567 | io_block = next_log_block++; |
3b5386dc | 568 | wrap(journal, next_log_block); |
0e8a9560 | 569 | err = jread(&obh, journal, io_block); |
3b5386dc TT |
570 | if (err) { |
571 | /* Recover what we can, but | |
572 | * report failure at the end. */ | |
573 | success = err; | |
13af4b93 DW |
574 | printk(KERN_ERR |
575 | "JBD2: IO error %d recovering " | |
576 | "block %ld in log\n", | |
0e8a9560 | 577 | err, io_block); |
3b5386dc | 578 | } else { |
3b693d0b | 579 | unsigned long long blocknr; |
e5ea6b14 | 580 | |
3b5386dc | 581 | J_ASSERT(obh != NULL); |
38d5adf3 | 582 | blocknr = read_tag_block(journal, |
3b693d0b | 583 | tag); |
0e8a9560 TT |
584 | |
585 | /* If the block has been | |
586 | * revoked, then we're all done | |
587 | * here. */ | |
588 | if (journal_test_revoke | |
e5ea6b14 | 589 | (journal, blocknr, |
0e8a9560 TT |
590 | next_commit_ID)) { |
591 | brelse(obh); | |
592 | ++info->nr_revoke_hits; | |
593 | goto skip_write; | |
594 | } | |
e5ea6b14 | 595 | |
d9200380 DW |
596 | /* Look for block corruption */ |
597 | if (!jbd2_block_tag_csum_verify( | |
598 | journal, tag, obh->b_data, | |
13af4b93 | 599 | ext2fs_be32_to_cpu(tmp->h_sequence))) { |
d9200380 | 600 | brelse(obh); |
b4f02c9f | 601 | success = -EFSBADCRC; |
13af4b93 | 602 | printk(KERN_ERR "JBD2: Invalid " |
d9200380 | 603 | "checksum recovering " |
13af4b93 | 604 | "block %llu in log\n", |
d9200380 | 605 | blocknr); |
bf93a968 DW |
606 | block_error = 1; |
607 | goto skip_write; | |
d9200380 DW |
608 | } |
609 | ||
0e8a9560 TT |
610 | /* Find a buffer for the new |
611 | * data being restored */ | |
e5ea6b14 TT |
612 | nbh = __getblk(journal->j_fs_dev, |
613 | blocknr, | |
614 | journal->j_blocksize); | |
3b5386dc | 615 | if (nbh == NULL) { |
e5ea6b14 | 616 | printk(KERN_ERR |
13af4b93 | 617 | "JBD2: Out of memory " |
3b5386dc TT |
618 | "during recovery.\n"); |
619 | err = -ENOMEM; | |
620 | brelse(bh); | |
621 | brelse(obh); | |
622 | goto failed; | |
623 | } | |
624 | ||
3de085dd | 625 | lock_buffer(nbh); |
8cf93332 TT |
626 | memcpy(nbh->b_data, obh->b_data, |
627 | journal->j_blocksize); | |
3b5386dc | 628 | if (flags & JFS_FLAG_ESCAPE) { |
13af4b93 DW |
629 | *((__u32 *)nbh->b_data) = |
630 | ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER); | |
3b5386dc | 631 | } |
8cf93332 TT |
632 | |
633 | BUFFER_TRACE(nbh, "marking dirty"); | |
3de085dd | 634 | set_buffer_uptodate(nbh); |
8cf93332 TT |
635 | mark_buffer_dirty(nbh); |
636 | BUFFER_TRACE(nbh, "marking uptodate"); | |
0e8a9560 | 637 | ++info->nr_replays; |
53ef44c4 | 638 | /* ll_rw_block(WRITE, 1, &nbh); */ |
3de085dd | 639 | unlock_buffer(nbh); |
3b5386dc TT |
640 | brelse(obh); |
641 | brelse(nbh); | |
642 | } | |
e5ea6b14 | 643 | |
0e8a9560 | 644 | skip_write: |
185c4aea | 645 | tagp += tag_bytes; |
3b5386dc TT |
646 | if (!(flags & JFS_FLAG_SAME_UUID)) |
647 | tagp += 16; | |
648 | ||
649 | if (flags & JFS_FLAG_LAST_TAG) | |
650 | break; | |
0e8a9560 | 651 | } |
e5ea6b14 | 652 | |
0e8a9560 TT |
653 | brelse(bh); |
654 | continue; | |
8cf93332 | 655 | |
0e8a9560 | 656 | case JFS_COMMIT_BLOCK: |
185c4aea TT |
657 | /* How to differentiate between interrupted commit |
658 | * and journal corruption ? | |
659 | * | |
660 | * {nth transaction} | |
661 | * Checksum Verification Failed | |
662 | * | | |
663 | * ____________________ | |
664 | * | | | |
665 | * async_commit sync_commit | |
666 | * | | | |
667 | * | GO TO NEXT "Journal Corruption" | |
668 | * | TRANSACTION | |
669 | * | | |
670 | * {(n+1)th transanction} | |
671 | * | | |
672 | * _______|______________ | |
673 | * | | | |
674 | * Commit block found Commit block not found | |
675 | * | | | |
676 | * "Journal Corruption" | | |
677 | * _____________|_________ | |
678 | * | | | |
679 | * nth trans corrupt OR nth trans | |
680 | * and (n+1)th interrupted interrupted | |
681 | * before commit block | |
682 | * could reach the disk. | |
683 | * (Cannot find the difference in above | |
684 | * mentioned conditions. Hence assume | |
685 | * "Interrupted Commit".) | |
686 | */ | |
687 | ||
688 | /* Found an expected commit block: if checksums | |
689 | * are present verify them in PASS_SCAN; else not | |
690 | * much to do other than move on to the next sequence | |
0e8a9560 | 691 | * number. */ |
185c4aea | 692 | if (pass == PASS_SCAN && |
86f3b6cf | 693 | jfs_has_feature_checksum(journal)) { |
185c4aea TT |
694 | int chksum_err, chksum_seen; |
695 | struct commit_header *cbh = | |
696 | (struct commit_header *)bh->b_data; | |
697 | unsigned found_chksum = | |
13af4b93 | 698 | ext2fs_be32_to_cpu(cbh->h_chksum[0]); |
185c4aea TT |
699 | |
700 | chksum_err = chksum_seen = 0; | |
701 | ||
185c4aea TT |
702 | if (info->end_transaction) { |
703 | journal->j_failed_commit = | |
704 | info->end_transaction; | |
705 | brelse(bh); | |
706 | break; | |
707 | } | |
708 | ||
709 | if (crc32_sum == found_chksum && | |
13af4b93 | 710 | cbh->h_chksum_type == JFS_CRC32_CHKSUM && |
185c4aea | 711 | cbh->h_chksum_size == |
13af4b93 | 712 | JFS_CRC32_CHKSUM_SIZE) |
185c4aea TT |
713 | chksum_seen = 1; |
714 | else if (!(cbh->h_chksum_type == 0 && | |
715 | cbh->h_chksum_size == 0 && | |
716 | found_chksum == 0 && | |
717 | !chksum_seen)) | |
718 | /* | |
719 | * If fs is mounted using an old kernel and then | |
720 | * kernel with journal_chksum is used then we | |
721 | * get a situation where the journal flag has | |
722 | * checksum flag set but checksums are not | |
723 | * present i.e chksum = 0, in the individual | |
724 | * commit blocks. | |
725 | * Hence to avoid checksum failures, in this | |
726 | * situation, this extra check is added. | |
727 | */ | |
728 | chksum_err = 1; | |
729 | ||
730 | if (chksum_err) { | |
731 | info->end_transaction = next_commit_ID; | |
13af4b93 | 732 | |
86f3b6cf | 733 | if (!jfs_has_feature_async_commit(journal)){ |
185c4aea TT |
734 | journal->j_failed_commit = |
735 | next_commit_ID; | |
736 | brelse(bh); | |
737 | break; | |
738 | } | |
739 | } | |
740 | crc32_sum = ~0; | |
741 | } | |
721e065c DW |
742 | if (pass == PASS_SCAN && |
743 | !jbd2_commit_block_csum_verify(journal, | |
744 | bh->b_data)) { | |
745 | info->end_transaction = next_commit_ID; | |
746 | ||
86f3b6cf | 747 | if (!jfs_has_feature_async_commit(journal)) { |
721e065c DW |
748 | journal->j_failed_commit = |
749 | next_commit_ID; | |
750 | brelse(bh); | |
751 | break; | |
752 | } | |
753 | } | |
0e8a9560 TT |
754 | brelse(bh); |
755 | next_commit_ID++; | |
756 | continue; | |
3b5386dc | 757 | |
0e8a9560 TT |
758 | case JFS_REVOKE_BLOCK: |
759 | /* If we aren't in the REVOKE pass, then we can | |
760 | * just skip over this block. */ | |
761 | if (pass != PASS_REVOKE) { | |
762 | brelse(bh); | |
763 | continue; | |
764 | } | |
3b5386dc | 765 | |
8cf93332 | 766 | err = scan_revoke_records(journal, bh, |
0e8a9560 | 767 | next_commit_ID, info); |
3b5386dc | 768 | brelse(bh); |
0e8a9560 TT |
769 | if (err) |
770 | goto failed; | |
771 | continue; | |
772 | ||
773 | default: | |
8cf93332 | 774 | jbd_debug(3, "Unrecognised magic %d, end of scan.\n", |
0e8a9560 | 775 | blocktype); |
2aa362f5 | 776 | brelse(bh); |
0e8a9560 TT |
777 | goto done; |
778 | } | |
3b5386dc | 779 | } |
3b5386dc | 780 | |
0e8a9560 | 781 | done: |
e5ea6b14 | 782 | /* |
0e8a9560 TT |
783 | * We broke out of the log scan loop: either we came to the |
784 | * known end of the log or we found an unexpected block in the | |
785 | * log. If the latter happened, then we know that the "current" | |
786 | * transaction marks the end of the valid log. | |
787 | */ | |
e5ea6b14 | 788 | |
185c4aea TT |
789 | if (pass == PASS_SCAN) { |
790 | if (!info->end_transaction) | |
791 | info->end_transaction = next_commit_ID; | |
792 | } else { | |
0e8a9560 TT |
793 | /* It's really bad news if different passes end up at |
794 | * different places (but possible due to IO errors). */ | |
795 | if (info->end_transaction != next_commit_ID) { | |
13af4b93 | 796 | printk(KERN_ERR "JBD2: recovery pass %d ended at " |
0e8a9560 TT |
797 | "transaction %u, expected %u\n", |
798 | pass, next_commit_ID, info->end_transaction); | |
799 | if (!success) | |
800 | success = -EIO; | |
801 | } | |
802 | } | |
bf93a968 DW |
803 | if (block_error && success == 0) |
804 | success = -EIO; | |
0e8a9560 | 805 | return success; |
3b5386dc | 806 | |
0e8a9560 | 807 | failed: |
3b5386dc TT |
808 | return err; |
809 | } | |
0e8a9560 | 810 | |
d35ed158 DW |
811 | static int jbd2_revoke_block_csum_verify(journal_t *j, |
812 | void *buf) | |
813 | { | |
814 | struct journal_revoke_tail *tail; | |
13af4b93 DW |
815 | __u32 provided; |
816 | __u32 calculated; | |
d35ed158 | 817 | |
38d5adf3 | 818 | if (!journal_has_csum_v2or3(j)) |
d35ed158 DW |
819 | return 1; |
820 | ||
13af4b93 | 821 | tail = (struct journal_revoke_tail *)(buf + j->j_blocksize - |
d35ed158 DW |
822 | sizeof(struct journal_revoke_tail)); |
823 | provided = tail->r_checksum; | |
824 | tail->r_checksum = 0; | |
13af4b93 | 825 | calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize); |
d35ed158 DW |
826 | tail->r_checksum = provided; |
827 | ||
13af4b93 | 828 | return provided == ext2fs_cpu_to_be32(calculated); |
d35ed158 | 829 | } |
0e8a9560 TT |
830 | |
831 | /* Scan a revoke record, marking all blocks mentioned as revoked. */ | |
832 | ||
e5ea6b14 | 833 | static int scan_revoke_records(journal_t *journal, struct buffer_head *bh, |
0e8a9560 TT |
834 | tid_t sequence, struct recovery_info *info) |
835 | { | |
836 | journal_revoke_header_t *header; | |
837 | int offset, max; | |
04c66cb2 DW |
838 | int csum_size = 0; |
839 | __u32 rcount; | |
79a4dddb | 840 | int record_len = 4; |
8cf93332 | 841 | |
0e8a9560 TT |
842 | header = (journal_revoke_header_t *) bh->b_data; |
843 | offset = sizeof(journal_revoke_header_t); | |
04c66cb2 | 844 | rcount = ext2fs_be32_to_cpu(header->r_count); |
e5ea6b14 | 845 | |
d35ed158 | 846 | if (!jbd2_revoke_block_csum_verify(journal, header)) |
b4f02c9f | 847 | return -EFSBADCRC; |
d35ed158 | 848 | |
04c66cb2 DW |
849 | if (journal_has_csum_v2or3(journal)) |
850 | csum_size = sizeof(struct journal_revoke_tail); | |
851 | if (rcount > journal->j_blocksize - csum_size) | |
852 | return -EINVAL; | |
853 | max = rcount; | |
854 | ||
86f3b6cf | 855 | if (jfs_has_feature_64bit(journal)) |
79a4dddb DW |
856 | record_len = 8; |
857 | ||
13af4b93 | 858 | while (offset + record_len <= max) { |
3b693d0b | 859 | unsigned long long blocknr; |
0e8a9560 | 860 | int err; |
e5ea6b14 | 861 | |
13af4b93 DW |
862 | if (record_len == 4) |
863 | blocknr = ext2fs_be32_to_cpu(* ((__u32 *) (bh->b_data+offset))); | |
864 | else | |
865 | blocknr = ext2fs_be64_to_cpu(* ((__u64 *) (bh->b_data+offset))); | |
79a4dddb | 866 | offset += record_len; |
0e8a9560 TT |
867 | err = journal_set_revoke(journal, blocknr, sequence); |
868 | if (err) | |
869 | return err; | |
870 | ++info->nr_revokes; | |
871 | } | |
872 | return 0; | |
873 | } |