xref: /linux/fs/jbd2/recovery.c (revision 0bab8db4)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * linux/fs/jbd2/recovery.c
4  *
5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6  *
7  * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
8  *
9  * Journal recovery routines for the generic filesystem journaling code;
10  * part of the ext2fs journaling system.
11  */
12 
13 #ifndef __KERNEL__
14 #include "jfs_user.h"
15 #else
16 #include <linux/time.h>
17 #include <linux/fs.h>
18 #include <linux/jbd2.h>
19 #include <linux/errno.h>
20 #include <linux/crc32.h>
21 #include <linux/blkdev.h>
22 #include <linux/string_choices.h>
23 #endif
24 
25 /*
26  * Maintain information about the progress of the recovery job, so that
27  * the different passes can carry information between them.
28  */
29 struct recovery_info
30 {
31 	tid_t		start_transaction;
32 	tid_t		end_transaction;
33 	unsigned long	head_block;
34 
35 	int		nr_replays;
36 	int		nr_revokes;
37 	int		nr_revoke_hits;
38 };
39 
40 static int do_one_pass(journal_t *journal,
41 				struct recovery_info *info, enum passtype pass);
42 static int scan_revoke_records(journal_t *, struct buffer_head *,
43 				tid_t, struct recovery_info *);
44 
45 #ifdef __KERNEL__
46 
47 /* Release readahead buffers after use */
journal_brelse_array(struct buffer_head * b[],int n)48 static void journal_brelse_array(struct buffer_head *b[], int n)
49 {
50 	while (--n >= 0)
51 		brelse (b[n]);
52 }
53 
54 
55 /*
56  * When reading from the journal, we are going through the block device
57  * layer directly and so there is no readahead being done for us.  We
58  * need to implement any readahead ourselves if we want it to happen at
59  * all.  Recovery is basically one long sequential read, so make sure we
60  * do the IO in reasonably large chunks.
61  *
62  * This is not so critical that we need to be enormously clever about
63  * the readahead size, though.  128K is a purely arbitrary, good-enough
64  * fixed value.
65  */
66 
67 #define MAXBUF 8
do_readahead(journal_t * journal,unsigned int start)68 static int do_readahead(journal_t *journal, unsigned int start)
69 {
70 	int err;
71 	unsigned int max, nbufs, next;
72 	unsigned long long blocknr;
73 	struct buffer_head *bh;
74 
75 	struct buffer_head * bufs[MAXBUF];
76 
77 	/* Do up to 128K of readahead */
78 	max = start + (128 * 1024 / journal->j_blocksize);
79 	if (max > journal->j_total_len)
80 		max = journal->j_total_len;
81 
82 	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
83 	 * a time to the block device IO layer. */
84 
85 	nbufs = 0;
86 
87 	for (next = start; next < max; next++) {
88 		err = jbd2_journal_bmap(journal, next, &blocknr);
89 
90 		if (err) {
91 			printk(KERN_ERR "JBD2: bad block at offset %u\n",
92 				next);
93 			goto failed;
94 		}
95 
96 		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
97 		if (!bh) {
98 			err = -ENOMEM;
99 			goto failed;
100 		}
101 
102 		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
103 			bufs[nbufs++] = bh;
104 			if (nbufs == MAXBUF) {
105 				bh_readahead_batch(nbufs, bufs, 0);
106 				journal_brelse_array(bufs, nbufs);
107 				nbufs = 0;
108 			}
109 		} else
110 			brelse(bh);
111 	}
112 
113 	if (nbufs)
114 		bh_readahead_batch(nbufs, bufs, 0);
115 	err = 0;
116 
117 failed:
118 	if (nbufs)
119 		journal_brelse_array(bufs, nbufs);
120 	return err;
121 }
122 
123 #endif /* __KERNEL__ */
124 
125 
126 /*
127  * Read a block from the journal
128  */
129 
jread(struct buffer_head ** bhp,journal_t * journal,unsigned int offset)130 static int jread(struct buffer_head **bhp, journal_t *journal,
131 		 unsigned int offset)
132 {
133 	int err;
134 	unsigned long long blocknr;
135 	struct buffer_head *bh;
136 
137 	*bhp = NULL;
138 
139 	if (offset >= journal->j_total_len) {
140 		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
141 		return -EFSCORRUPTED;
142 	}
143 
144 	err = jbd2_journal_bmap(journal, offset, &blocknr);
145 
146 	if (err) {
147 		printk(KERN_ERR "JBD2: bad block at offset %u\n",
148 			offset);
149 		return err;
150 	}
151 
152 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
153 	if (!bh)
154 		return -ENOMEM;
155 
156 	if (!buffer_uptodate(bh)) {
157 		/*
158 		 * If this is a brand new buffer, start readahead.
159 		 * Otherwise, we assume we are already reading it.
160 		 */
161 		bool need_readahead = !buffer_req(bh);
162 
163 		bh_read_nowait(bh, 0);
164 		if (need_readahead)
165 			do_readahead(journal, offset);
166 		wait_on_buffer(bh);
167 	}
168 
169 	if (!buffer_uptodate(bh)) {
170 		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
171 			offset);
172 		brelse(bh);
173 		return -EIO;
174 	}
175 
176 	*bhp = bh;
177 	return 0;
178 }
179 
jbd2_descriptor_block_csum_verify(journal_t * j,void * buf)180 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
181 {
182 	struct jbd2_journal_block_tail *tail;
183 	__be32 provided;
184 	__u32 calculated;
185 
186 	if (!jbd2_journal_has_csum_v2or3(j))
187 		return 1;
188 
189 	tail = (struct jbd2_journal_block_tail *)((char *)buf +
190 		j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
191 	provided = tail->t_checksum;
192 	tail->t_checksum = 0;
193 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
194 	tail->t_checksum = provided;
195 
196 	return provided == cpu_to_be32(calculated);
197 }
198 
199 /*
200  * Count the number of in-use tags in a journal descriptor block.
201  */
202 
count_tags(journal_t * journal,struct buffer_head * bh)203 static int count_tags(journal_t *journal, struct buffer_head *bh)
204 {
205 	char *			tagp;
206 	journal_block_tag_t	tag;
207 	int			nr = 0, size = journal->j_blocksize;
208 	int			tag_bytes = journal_tag_bytes(journal);
209 
210 	if (jbd2_journal_has_csum_v2or3(journal))
211 		size -= sizeof(struct jbd2_journal_block_tail);
212 
213 	tagp = &bh->b_data[sizeof(journal_header_t)];
214 
215 	while ((tagp - bh->b_data + tag_bytes) <= size) {
216 		memcpy(&tag, tagp, sizeof(tag));
217 
218 		nr++;
219 		tagp += tag_bytes;
220 		if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
221 			tagp += 16;
222 
223 		if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
224 			break;
225 	}
226 
227 	return nr;
228 }
229 
230 
231 /* Make sure we wrap around the log correctly! */
232 #define wrap(journal, var)						\
233 do {									\
234 	if (var >= (journal)->j_last)					\
235 		var -= ((journal)->j_last - (journal)->j_first);	\
236 } while (0)
237 
fc_do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)238 static int fc_do_one_pass(journal_t *journal,
239 			  struct recovery_info *info, enum passtype pass)
240 {
241 	unsigned int expected_commit_id = info->end_transaction;
242 	unsigned long next_fc_block;
243 	struct buffer_head *bh;
244 	int err = 0;
245 
246 	next_fc_block = journal->j_fc_first;
247 	if (!journal->j_fc_replay_callback)
248 		return 0;
249 
250 	while (next_fc_block <= journal->j_fc_last) {
251 		jbd2_debug(3, "Fast commit replay: next block %ld\n",
252 			  next_fc_block);
253 		err = jread(&bh, journal, next_fc_block);
254 		if (err) {
255 			jbd2_debug(3, "Fast commit replay: read error\n");
256 			break;
257 		}
258 
259 		err = journal->j_fc_replay_callback(journal, bh, pass,
260 					next_fc_block - journal->j_fc_first,
261 					expected_commit_id);
262 		brelse(bh);
263 		next_fc_block++;
264 		if (err < 0 || err == JBD2_FC_REPLAY_STOP)
265 			break;
266 		err = 0;
267 	}
268 
269 	if (err)
270 		jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
271 
272 	return err;
273 }
274 
275 /**
276  * jbd2_journal_recover - recovers a on-disk journal
277  * @journal: the journal to recover
278  *
279  * The primary function for recovering the log contents when mounting a
280  * journaled device.
281  *
282  * Recovery is done in three passes.  In the first pass, we look for the
283  * end of the log.  In the second, we assemble the list of revoke
284  * blocks.  In the third and final pass, we replay any un-revoked blocks
285  * in the log.
286  */
jbd2_journal_recover(journal_t * journal)287 int jbd2_journal_recover(journal_t *journal)
288 {
289 	int			err, err2;
290 	journal_superblock_t *	sb;
291 
292 	struct recovery_info	info;
293 
294 	memset(&info, 0, sizeof(info));
295 	sb = journal->j_superblock;
296 
297 	/*
298 	 * The journal superblock's s_start field (the current log head)
299 	 * is always zero if, and only if, the journal was cleanly
300 	 * unmounted.
301 	 */
302 	if (!sb->s_start) {
303 		jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
304 			  be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
305 		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
306 		journal->j_head = be32_to_cpu(sb->s_head);
307 		return 0;
308 	}
309 
310 	err = do_one_pass(journal, &info, PASS_SCAN);
311 	if (!err)
312 		err = do_one_pass(journal, &info, PASS_REVOKE);
313 	if (!err)
314 		err = do_one_pass(journal, &info, PASS_REPLAY);
315 
316 	jbd2_debug(1, "JBD2: recovery, exit status %d, "
317 		  "recovered transactions %u to %u\n",
318 		  err, info.start_transaction, info.end_transaction);
319 	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
320 		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
321 
322 	/* Restart the log at the next transaction ID, thus invalidating
323 	 * any existing commit records in the log. */
324 	journal->j_transaction_sequence = ++info.end_transaction;
325 	journal->j_head = info.head_block;
326 	jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
327 		  journal->j_transaction_sequence, journal->j_head);
328 
329 	jbd2_journal_clear_revoke(journal);
330 	err2 = sync_blockdev(journal->j_fs_dev);
331 	if (!err)
332 		err = err2;
333 	err2 = jbd2_check_fs_dev_write_error(journal);
334 	if (!err)
335 		err = err2;
336 	/* Make sure all replayed data is on permanent storage */
337 	if (journal->j_flags & JBD2_BARRIER) {
338 		err2 = blkdev_issue_flush(journal->j_fs_dev);
339 		if (!err)
340 			err = err2;
341 	}
342 	return err;
343 }
344 
345 /**
346  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
347  * @journal: journal to startup
348  *
349  * Locate any valid recovery information from the journal and set up the
350  * journal structures in memory to ignore it (presumably because the
351  * caller has evidence that it is out of date).
352  * This function doesn't appear to be exported..
353  *
354  * We perform one pass over the journal to allow us to tell the user how
355  * much recovery information is being erased, and to let us initialise
356  * the journal transaction sequence numbers to the next unused ID.
357  */
jbd2_journal_skip_recovery(journal_t * journal)358 int jbd2_journal_skip_recovery(journal_t *journal)
359 {
360 	int			err;
361 
362 	struct recovery_info	info;
363 
364 	memset (&info, 0, sizeof(info));
365 
366 	err = do_one_pass(journal, &info, PASS_SCAN);
367 
368 	if (err) {
369 		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
370 		++journal->j_transaction_sequence;
371 		journal->j_head = journal->j_first;
372 	} else {
373 #ifdef CONFIG_JBD2_DEBUG
374 		int dropped = info.end_transaction -
375 			be32_to_cpu(journal->j_superblock->s_sequence);
376 		jbd2_debug(1,
377 			  "JBD2: ignoring %d transaction%s from the journal.\n",
378 			  dropped, str_plural(dropped));
379 #endif
380 		journal->j_transaction_sequence = ++info.end_transaction;
381 		journal->j_head = info.head_block;
382 	}
383 
384 	journal->j_tail = 0;
385 	return err;
386 }
387 
read_tag_block(journal_t * journal,journal_block_tag_t * tag)388 static inline unsigned long long read_tag_block(journal_t *journal,
389 						journal_block_tag_t *tag)
390 {
391 	unsigned long long block = be32_to_cpu(tag->t_blocknr);
392 	if (jbd2_has_feature_64bit(journal))
393 		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
394 	return block;
395 }
396 
397 /*
398  * calc_chksums calculates the checksums for the blocks described in the
399  * descriptor block.
400  */
calc_chksums(journal_t * journal,struct buffer_head * bh,unsigned long * next_log_block,__u32 * crc32_sum)401 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
402 			unsigned long *next_log_block, __u32 *crc32_sum)
403 {
404 	int i, num_blks, err;
405 	unsigned long io_block;
406 	struct buffer_head *obh;
407 
408 	num_blks = count_tags(journal, bh);
409 	/* Calculate checksum of the descriptor block. */
410 	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
411 
412 	for (i = 0; i < num_blks; i++) {
413 		io_block = (*next_log_block)++;
414 		wrap(journal, *next_log_block);
415 		err = jread(&obh, journal, io_block);
416 		if (err) {
417 			printk(KERN_ERR "JBD2: IO error %d recovering block "
418 				"%lu in log\n", err, io_block);
419 			return 1;
420 		} else {
421 			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
422 				     obh->b_size);
423 		}
424 		put_bh(obh);
425 	}
426 	return 0;
427 }
428 
jbd2_commit_block_csum_verify(journal_t * j,void * buf)429 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
430 {
431 	struct commit_header *h;
432 	__be32 provided;
433 	__u32 calculated;
434 
435 	if (!jbd2_journal_has_csum_v2or3(j))
436 		return 1;
437 
438 	h = buf;
439 	provided = h->h_chksum[0];
440 	h->h_chksum[0] = 0;
441 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
442 	h->h_chksum[0] = provided;
443 
444 	return provided == cpu_to_be32(calculated);
445 }
446 
jbd2_commit_block_csum_verify_partial(journal_t * j,void * buf)447 static bool jbd2_commit_block_csum_verify_partial(journal_t *j, void *buf)
448 {
449 	struct commit_header *h;
450 	__be32 provided;
451 	__u32 calculated;
452 	void *tmpbuf;
453 
454 	tmpbuf = kzalloc(j->j_blocksize, GFP_KERNEL);
455 	if (!tmpbuf)
456 		return false;
457 
458 	memcpy(tmpbuf, buf, sizeof(struct commit_header));
459 	h = tmpbuf;
460 	provided = h->h_chksum[0];
461 	h->h_chksum[0] = 0;
462 	calculated = jbd2_chksum(j, j->j_csum_seed, tmpbuf, j->j_blocksize);
463 	kfree(tmpbuf);
464 
465 	return provided == cpu_to_be32(calculated);
466 }
467 
jbd2_block_tag_csum_verify(journal_t * j,journal_block_tag_t * tag,journal_block_tag3_t * tag3,void * buf,__u32 sequence)468 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
469 				      journal_block_tag3_t *tag3,
470 				      void *buf, __u32 sequence)
471 {
472 	__u32 csum32;
473 	__be32 seq;
474 
475 	if (!jbd2_journal_has_csum_v2or3(j))
476 		return 1;
477 
478 	seq = cpu_to_be32(sequence);
479 	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
480 	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
481 
482 	if (jbd2_has_feature_csum3(j))
483 		return tag3->t_checksum == cpu_to_be32(csum32);
484 	else
485 		return tag->t_checksum == cpu_to_be16(csum32);
486 }
487 
do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)488 static int do_one_pass(journal_t *journal,
489 			struct recovery_info *info, enum passtype pass)
490 {
491 	unsigned int		first_commit_ID, next_commit_ID;
492 	unsigned long		next_log_block, head_block;
493 	int			err, success = 0;
494 	journal_superblock_t *	sb;
495 	journal_header_t *	tmp;
496 	struct buffer_head *	bh;
497 	unsigned int		sequence;
498 	int			blocktype;
499 	int			tag_bytes = journal_tag_bytes(journal);
500 	__u32			crc32_sum = ~0; /* Transactional Checksums */
501 	int			descr_csum_size = 0;
502 	int			block_error = 0;
503 	bool			need_check_commit_time = false;
504 	__u64			last_trans_commit_time = 0, commit_time;
505 
506 	/*
507 	 * First thing is to establish what we expect to find in the log
508 	 * (in terms of transaction IDs), and where (in terms of log
509 	 * block offsets): query the superblock.
510 	 */
511 
512 	sb = journal->j_superblock;
513 	next_commit_ID = be32_to_cpu(sb->s_sequence);
514 	next_log_block = be32_to_cpu(sb->s_start);
515 	head_block = next_log_block;
516 
517 	first_commit_ID = next_commit_ID;
518 	if (pass == PASS_SCAN)
519 		info->start_transaction = first_commit_ID;
520 
521 	jbd2_debug(1, "Starting recovery pass %d\n", pass);
522 
523 	/*
524 	 * Now we walk through the log, transaction by transaction,
525 	 * making sure that each transaction has a commit block in the
526 	 * expected place.  Each complete transaction gets replayed back
527 	 * into the main filesystem.
528 	 */
529 
530 	while (1) {
531 		int			flags;
532 		char *			tagp;
533 		journal_block_tag_t	tag;
534 		struct buffer_head *	obh;
535 		struct buffer_head *	nbh;
536 
537 		cond_resched();
538 
539 		/* If we already know where to stop the log traversal,
540 		 * check right now that we haven't gone past the end of
541 		 * the log. */
542 
543 		if (pass != PASS_SCAN)
544 			if (tid_geq(next_commit_ID, info->end_transaction))
545 				break;
546 
547 		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
548 			  next_commit_ID, next_log_block, journal->j_last);
549 
550 		/* Skip over each chunk of the transaction looking
551 		 * either the next descriptor block or the final commit
552 		 * record. */
553 
554 		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
555 		err = jread(&bh, journal, next_log_block);
556 		if (err)
557 			goto failed;
558 
559 		next_log_block++;
560 		wrap(journal, next_log_block);
561 
562 		/* What kind of buffer is it?
563 		 *
564 		 * If it is a descriptor block, check that it has the
565 		 * expected sequence number.  Otherwise, we're all done
566 		 * here. */
567 
568 		tmp = (journal_header_t *)bh->b_data;
569 
570 		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
571 			brelse(bh);
572 			break;
573 		}
574 
575 		blocktype = be32_to_cpu(tmp->h_blocktype);
576 		sequence = be32_to_cpu(tmp->h_sequence);
577 		jbd2_debug(3, "Found magic %d, sequence %d\n",
578 			  blocktype, sequence);
579 
580 		if (sequence != next_commit_ID) {
581 			brelse(bh);
582 			break;
583 		}
584 
585 		/* OK, we have a valid descriptor block which matches
586 		 * all of the sequence number checks.  What are we going
587 		 * to do with it?  That depends on the pass... */
588 
589 		switch(blocktype) {
590 		case JBD2_DESCRIPTOR_BLOCK:
591 			/* Verify checksum first */
592 			if (jbd2_journal_has_csum_v2or3(journal))
593 				descr_csum_size =
594 					sizeof(struct jbd2_journal_block_tail);
595 			if (descr_csum_size > 0 &&
596 			    !jbd2_descriptor_block_csum_verify(journal,
597 							       bh->b_data)) {
598 				/*
599 				 * PASS_SCAN can see stale blocks due to lazy
600 				 * journal init. Don't error out on those yet.
601 				 */
602 				if (pass != PASS_SCAN) {
603 					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
604 					       next_log_block);
605 					err = -EFSBADCRC;
606 					brelse(bh);
607 					goto failed;
608 				}
609 				need_check_commit_time = true;
610 				jbd2_debug(1,
611 					"invalid descriptor block found in %lu\n",
612 					next_log_block);
613 			}
614 
615 			/* If it is a valid descriptor block, replay it
616 			 * in pass REPLAY; if journal_checksums enabled, then
617 			 * calculate checksums in PASS_SCAN, otherwise,
618 			 * just skip over the blocks it describes. */
619 			if (pass != PASS_REPLAY) {
620 				if (pass == PASS_SCAN &&
621 				    jbd2_has_feature_checksum(journal) &&
622 				    !need_check_commit_time &&
623 				    !info->end_transaction) {
624 					if (calc_chksums(journal, bh,
625 							&next_log_block,
626 							&crc32_sum)) {
627 						put_bh(bh);
628 						break;
629 					}
630 					put_bh(bh);
631 					continue;
632 				}
633 				next_log_block += count_tags(journal, bh);
634 				wrap(journal, next_log_block);
635 				put_bh(bh);
636 				continue;
637 			}
638 
639 			/* A descriptor block: we can now write all of
640 			 * the data blocks.  Yay, useful work is finally
641 			 * getting done here! */
642 
643 			tagp = &bh->b_data[sizeof(journal_header_t)];
644 			while ((tagp - bh->b_data + tag_bytes)
645 			       <= journal->j_blocksize - descr_csum_size) {
646 				unsigned long io_block;
647 
648 				memcpy(&tag, tagp, sizeof(tag));
649 				flags = be16_to_cpu(tag.t_flags);
650 
651 				io_block = next_log_block++;
652 				wrap(journal, next_log_block);
653 				err = jread(&obh, journal, io_block);
654 				if (err) {
655 					/* Recover what we can, but
656 					 * report failure at the end. */
657 					success = err;
658 					printk(KERN_ERR
659 						"JBD2: IO error %d recovering "
660 						"block %lu in log\n",
661 						err, io_block);
662 				} else {
663 					unsigned long long blocknr;
664 
665 					J_ASSERT(obh != NULL);
666 					blocknr = read_tag_block(journal,
667 								 &tag);
668 
669 					/* If the block has been
670 					 * revoked, then we're all done
671 					 * here. */
672 					if (jbd2_journal_test_revoke
673 					    (journal, blocknr,
674 					     next_commit_ID)) {
675 						brelse(obh);
676 						++info->nr_revoke_hits;
677 						goto skip_write;
678 					}
679 
680 					/* Look for block corruption */
681 					if (!jbd2_block_tag_csum_verify(
682 			journal, &tag, (journal_block_tag3_t *)tagp,
683 			obh->b_data, be32_to_cpu(tmp->h_sequence))) {
684 						brelse(obh);
685 						success = -EFSBADCRC;
686 						printk(KERN_ERR "JBD2: Invalid "
687 						       "checksum recovering "
688 						       "data block %llu in "
689 						       "journal block %lu\n",
690 						       blocknr, io_block);
691 						block_error = 1;
692 						goto skip_write;
693 					}
694 
695 					/* Find a buffer for the new
696 					 * data being restored */
697 					nbh = __getblk(journal->j_fs_dev,
698 							blocknr,
699 							journal->j_blocksize);
700 					if (nbh == NULL) {
701 						printk(KERN_ERR
702 						       "JBD2: Out of memory "
703 						       "during recovery.\n");
704 						err = -ENOMEM;
705 						brelse(bh);
706 						brelse(obh);
707 						goto failed;
708 					}
709 
710 					lock_buffer(nbh);
711 					memcpy(nbh->b_data, obh->b_data,
712 							journal->j_blocksize);
713 					if (flags & JBD2_FLAG_ESCAPE) {
714 						*((__be32 *)nbh->b_data) =
715 						cpu_to_be32(JBD2_MAGIC_NUMBER);
716 					}
717 
718 					BUFFER_TRACE(nbh, "marking dirty");
719 					set_buffer_uptodate(nbh);
720 					mark_buffer_dirty(nbh);
721 					BUFFER_TRACE(nbh, "marking uptodate");
722 					++info->nr_replays;
723 					unlock_buffer(nbh);
724 					brelse(obh);
725 					brelse(nbh);
726 				}
727 
728 			skip_write:
729 				tagp += tag_bytes;
730 				if (!(flags & JBD2_FLAG_SAME_UUID))
731 					tagp += 16;
732 
733 				if (flags & JBD2_FLAG_LAST_TAG)
734 					break;
735 			}
736 
737 			brelse(bh);
738 			continue;
739 
740 		case JBD2_COMMIT_BLOCK:
741 			/*     How to differentiate between interrupted commit
742 			 *               and journal corruption ?
743 			 *
744 			 * {nth transaction}
745 			 *        Checksum Verification Failed
746 			 *			 |
747 			 *		 ____________________
748 			 *		|		     |
749 			 * 	async_commit             sync_commit
750 			 *     		|                    |
751 			 *		| GO TO NEXT    "Journal Corruption"
752 			 *		| TRANSACTION
753 			 *		|
754 			 * {(n+1)th transanction}
755 			 *		|
756 			 * 	 _______|______________
757 			 * 	|	 	      |
758 			 * Commit block found	Commit block not found
759 			 *      |		      |
760 			 * "Journal Corruption"       |
761 			 *		 _____________|_________
762 			 *     		|	           	|
763 			 *	nth trans corrupt	OR   nth trans
764 			 *	and (n+1)th interrupted     interrupted
765 			 *	before commit block
766 			 *      could reach the disk.
767 			 *	(Cannot find the difference in above
768 			 *	 mentioned conditions. Hence assume
769 			 *	 "Interrupted Commit".)
770 			 */
771 			commit_time = be64_to_cpu(
772 				((struct commit_header *)bh->b_data)->h_commit_sec);
773 			/*
774 			 * If need_check_commit_time is set, it means we are in
775 			 * PASS_SCAN and csum verify failed before. If
776 			 * commit_time is increasing, it's the same journal,
777 			 * otherwise it is stale journal block, just end this
778 			 * recovery.
779 			 */
780 			if (need_check_commit_time) {
781 				if (commit_time >= last_trans_commit_time) {
782 					pr_err("JBD2: Invalid checksum found in transaction %u\n",
783 					       next_commit_ID);
784 					err = -EFSBADCRC;
785 					brelse(bh);
786 					goto failed;
787 				}
788 			ignore_crc_mismatch:
789 				/*
790 				 * It likely does not belong to same journal,
791 				 * just end this recovery with success.
792 				 */
793 				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
794 					  next_commit_ID);
795 				brelse(bh);
796 				goto done;
797 			}
798 
799 			/*
800 			 * Found an expected commit block: if checksums
801 			 * are present, verify them in PASS_SCAN; else not
802 			 * much to do other than move on to the next sequence
803 			 * number.
804 			 */
805 			if (pass == PASS_SCAN &&
806 			    jbd2_has_feature_checksum(journal)) {
807 				struct commit_header *cbh =
808 					(struct commit_header *)bh->b_data;
809 				unsigned found_chksum =
810 					be32_to_cpu(cbh->h_chksum[0]);
811 
812 				if (info->end_transaction) {
813 					journal->j_failed_commit =
814 						info->end_transaction;
815 					brelse(bh);
816 					break;
817 				}
818 
819 				/* Neither checksum match nor unused? */
820 				if (!((crc32_sum == found_chksum &&
821 				       cbh->h_chksum_type ==
822 						JBD2_CRC32_CHKSUM &&
823 				       cbh->h_chksum_size ==
824 						JBD2_CRC32_CHKSUM_SIZE) ||
825 				      (cbh->h_chksum_type == 0 &&
826 				       cbh->h_chksum_size == 0 &&
827 				       found_chksum == 0)))
828 					goto chksum_error;
829 
830 				crc32_sum = ~0;
831 			}
832 			if (pass == PASS_SCAN &&
833 			    !jbd2_commit_block_csum_verify(journal,
834 							   bh->b_data)) {
835 				if (jbd2_commit_block_csum_verify_partial(
836 								  journal,
837 								  bh->b_data)) {
838 					pr_notice("JBD2: Find incomplete commit block in transaction %u block %lu\n",
839 						  next_commit_ID, next_log_block);
840 					goto chksum_ok;
841 				}
842 			chksum_error:
843 				if (commit_time < last_trans_commit_time)
844 					goto ignore_crc_mismatch;
845 				info->end_transaction = next_commit_ID;
846 				info->head_block = head_block;
847 
848 				if (!jbd2_has_feature_async_commit(journal)) {
849 					journal->j_failed_commit =
850 						next_commit_ID;
851 					brelse(bh);
852 					break;
853 				}
854 			}
855 			if (pass == PASS_SCAN) {
856 			chksum_ok:
857 				last_trans_commit_time = commit_time;
858 				head_block = next_log_block;
859 			}
860 			brelse(bh);
861 			next_commit_ID++;
862 			continue;
863 
864 		case JBD2_REVOKE_BLOCK:
865 			/*
866 			 * Check revoke block crc in pass_scan, if csum verify
867 			 * failed, check commit block time later.
868 			 */
869 			if (pass == PASS_SCAN &&
870 			    !jbd2_descriptor_block_csum_verify(journal,
871 							       bh->b_data)) {
872 				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
873 					  next_log_block);
874 				need_check_commit_time = true;
875 			}
876 
877 			/* If we aren't in the REVOKE pass, then we can
878 			 * just skip over this block. */
879 			if (pass != PASS_REVOKE) {
880 				brelse(bh);
881 				continue;
882 			}
883 
884 			err = scan_revoke_records(journal, bh,
885 						  next_commit_ID, info);
886 			brelse(bh);
887 			if (err)
888 				goto failed;
889 			continue;
890 
891 		default:
892 			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
893 				  blocktype);
894 			brelse(bh);
895 			goto done;
896 		}
897 	}
898 
899  done:
900 	/*
901 	 * We broke out of the log scan loop: either we came to the
902 	 * known end of the log or we found an unexpected block in the
903 	 * log.  If the latter happened, then we know that the "current"
904 	 * transaction marks the end of the valid log.
905 	 */
906 
907 	if (pass == PASS_SCAN) {
908 		if (!info->end_transaction)
909 			info->end_transaction = next_commit_ID;
910 		if (!info->head_block)
911 			info->head_block = head_block;
912 	} else {
913 		/* It's really bad news if different passes end up at
914 		 * different places (but possible due to IO errors). */
915 		if (info->end_transaction != next_commit_ID) {
916 			printk(KERN_ERR "JBD2: recovery pass %d ended at "
917 				"transaction %u, expected %u\n",
918 				pass, next_commit_ID, info->end_transaction);
919 			if (!success)
920 				success = -EIO;
921 		}
922 	}
923 
924 	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
925 		err = fc_do_one_pass(journal, info, pass);
926 		if (err)
927 			success = err;
928 	}
929 
930 	if (block_error && success == 0)
931 		success = -EIO;
932 	return success;
933 
934  failed:
935 	return err;
936 }
937 
938 /* Scan a revoke record, marking all blocks mentioned as revoked. */
939 
scan_revoke_records(journal_t * journal,struct buffer_head * bh,tid_t sequence,struct recovery_info * info)940 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
941 			       tid_t sequence, struct recovery_info *info)
942 {
943 	jbd2_journal_revoke_header_t *header;
944 	int offset, max;
945 	unsigned csum_size = 0;
946 	__u32 rcount;
947 	int record_len = 4;
948 
949 	header = (jbd2_journal_revoke_header_t *) bh->b_data;
950 	offset = sizeof(jbd2_journal_revoke_header_t);
951 	rcount = be32_to_cpu(header->r_count);
952 
953 	if (jbd2_journal_has_csum_v2or3(journal))
954 		csum_size = sizeof(struct jbd2_journal_block_tail);
955 	if (rcount > journal->j_blocksize - csum_size)
956 		return -EINVAL;
957 	max = rcount;
958 
959 	if (jbd2_has_feature_64bit(journal))
960 		record_len = 8;
961 
962 	while (offset + record_len <= max) {
963 		unsigned long long blocknr;
964 		int err;
965 
966 		if (record_len == 4)
967 			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
968 		else
969 			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
970 		offset += record_len;
971 		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
972 		if (err)
973 			return err;
974 		++info->nr_revokes;
975 	}
976 	return 0;
977 }
978