xref: /openbmc/linux/fs/jbd2/recovery.c (revision 4d75f5c664195b970e1cd2fd25b65b5eff257a0a)
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 #endif
23 
24 /*
25  * Maintain information about the progress of the recovery job, so that
26  * the different passes can carry information between them.
27  */
28 struct recovery_info
29 {
30 	tid_t		start_transaction;
31 	tid_t		end_transaction;
32 	unsigned long	head_block;
33 
34 	int		nr_replays;
35 	int		nr_revokes;
36 	int		nr_revoke_hits;
37 };
38 
39 static int do_one_pass(journal_t *journal,
40 				struct recovery_info *info, enum passtype pass);
41 static int scan_revoke_records(journal_t *, struct buffer_head *,
42 				tid_t, struct recovery_info *);
43 
44 #ifdef __KERNEL__
45 
46 /* Release readahead buffers after use */
journal_brelse_array(struct buffer_head * b[],int n)47 static void journal_brelse_array(struct buffer_head *b[], int n)
48 {
49 	while (--n >= 0)
50 		brelse (b[n]);
51 }
52 
53 
54 /*
55  * When reading from the journal, we are going through the block device
56  * layer directly and so there is no readahead being done for us.  We
57  * need to implement any readahead ourselves if we want it to happen at
58  * all.  Recovery is basically one long sequential read, so make sure we
59  * do the IO in reasonably large chunks.
60  *
61  * This is not so critical that we need to be enormously clever about
62  * the readahead size, though.  128K is a purely arbitrary, good-enough
63  * fixed value.
64  */
65 
66 #define MAXBUF 8
do_readahead(journal_t * journal,unsigned int start)67 static int do_readahead(journal_t *journal, unsigned int start)
68 {
69 	int err;
70 	unsigned int max, nbufs, next;
71 	unsigned long long blocknr;
72 	struct buffer_head *bh;
73 
74 	struct buffer_head * bufs[MAXBUF];
75 
76 	/* Do up to 128K of readahead */
77 	max = start + (128 * 1024 / journal->j_blocksize);
78 	if (max > journal->j_total_len)
79 		max = journal->j_total_len;
80 
81 	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
82 	 * a time to the block device IO layer. */
83 
84 	nbufs = 0;
85 
86 	for (next = start; next < max; next++) {
87 		err = jbd2_journal_bmap(journal, next, &blocknr);
88 
89 		if (err) {
90 			printk(KERN_ERR "JBD2: bad block at offset %u\n",
91 				next);
92 			goto failed;
93 		}
94 
95 		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
96 		if (!bh) {
97 			err = -ENOMEM;
98 			goto failed;
99 		}
100 
101 		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
102 			bufs[nbufs++] = bh;
103 			if (nbufs == MAXBUF) {
104 				bh_readahead_batch(nbufs, bufs, 0);
105 				journal_brelse_array(bufs, nbufs);
106 				nbufs = 0;
107 			}
108 		} else
109 			brelse(bh);
110 	}
111 
112 	if (nbufs)
113 		bh_readahead_batch(nbufs, bufs, 0);
114 	err = 0;
115 
116 failed:
117 	if (nbufs)
118 		journal_brelse_array(bufs, nbufs);
119 	return err;
120 }
121 
122 #endif /* __KERNEL__ */
123 
124 
125 /*
126  * Read a block from the journal
127  */
128 
jread(struct buffer_head ** bhp,journal_t * journal,unsigned int offset)129 static int jread(struct buffer_head **bhp, journal_t *journal,
130 		 unsigned int offset)
131 {
132 	int err;
133 	unsigned long long blocknr;
134 	struct buffer_head *bh;
135 
136 	*bhp = NULL;
137 
138 	if (offset >= journal->j_total_len) {
139 		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
140 		return -EFSCORRUPTED;
141 	}
142 
143 	err = jbd2_journal_bmap(journal, offset, &blocknr);
144 
145 	if (err) {
146 		printk(KERN_ERR "JBD2: bad block at offset %u\n",
147 			offset);
148 		return err;
149 	}
150 
151 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
152 	if (!bh)
153 		return -ENOMEM;
154 
155 	if (!buffer_uptodate(bh)) {
156 		/*
157 		 * If this is a brand new buffer, start readahead.
158 		 * Otherwise, we assume we are already reading it.
159 		 */
160 		bool need_readahead = !buffer_req(bh);
161 
162 		bh_read_nowait(bh, 0);
163 		if (need_readahead)
164 			do_readahead(journal, offset);
165 		wait_on_buffer(bh);
166 	}
167 
168 	if (!buffer_uptodate(bh)) {
169 		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
170 			offset);
171 		brelse(bh);
172 		return -EIO;
173 	}
174 
175 	*bhp = bh;
176 	return 0;
177 }
178 
jbd2_descriptor_block_csum_verify(journal_t * j,void * buf)179 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
180 {
181 	struct jbd2_journal_block_tail *tail;
182 	__be32 provided;
183 	__u32 calculated;
184 
185 	if (!jbd2_journal_has_csum_v2or3(j))
186 		return 1;
187 
188 	tail = (struct jbd2_journal_block_tail *)((char *)buf +
189 		j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
190 	provided = tail->t_checksum;
191 	tail->t_checksum = 0;
192 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
193 	tail->t_checksum = provided;
194 
195 	return provided == cpu_to_be32(calculated);
196 }
197 
198 /*
199  * Count the number of in-use tags in a journal descriptor block.
200  */
201 
count_tags(journal_t * journal,struct buffer_head * bh)202 static int count_tags(journal_t *journal, struct buffer_head *bh)
203 {
204 	char *			tagp;
205 	journal_block_tag_t	tag;
206 	int			nr = 0, size = journal->j_blocksize;
207 	int			tag_bytes = journal_tag_bytes(journal);
208 
209 	if (jbd2_journal_has_csum_v2or3(journal))
210 		size -= sizeof(struct jbd2_journal_block_tail);
211 
212 	tagp = &bh->b_data[sizeof(journal_header_t)];
213 
214 	while ((tagp - bh->b_data + tag_bytes) <= size) {
215 		memcpy(&tag, tagp, sizeof(tag));
216 
217 		nr++;
218 		tagp += tag_bytes;
219 		if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
220 			tagp += 16;
221 
222 		if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
223 			break;
224 	}
225 
226 	return nr;
227 }
228 
229 
230 /* Make sure we wrap around the log correctly! */
231 #define wrap(journal, var)						\
232 do {									\
233 	if (var >= (journal)->j_last)					\
234 		var -= ((journal)->j_last - (journal)->j_first);	\
235 } while (0)
236 
fc_do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)237 static int fc_do_one_pass(journal_t *journal,
238 			  struct recovery_info *info, enum passtype pass)
239 {
240 	unsigned int expected_commit_id = info->end_transaction;
241 	unsigned long next_fc_block;
242 	struct buffer_head *bh;
243 	int err = 0;
244 
245 	next_fc_block = journal->j_fc_first;
246 	if (!journal->j_fc_replay_callback)
247 		return 0;
248 
249 	while (next_fc_block <= journal->j_fc_last) {
250 		jbd2_debug(3, "Fast commit replay: next block %ld\n",
251 			  next_fc_block);
252 		err = jread(&bh, journal, next_fc_block);
253 		if (err) {
254 			jbd2_debug(3, "Fast commit replay: read error\n");
255 			break;
256 		}
257 
258 		err = journal->j_fc_replay_callback(journal, bh, pass,
259 					next_fc_block - journal->j_fc_first,
260 					expected_commit_id);
261 		brelse(bh);
262 		next_fc_block++;
263 		if (err < 0 || err == JBD2_FC_REPLAY_STOP)
264 			break;
265 		err = 0;
266 	}
267 
268 	if (err)
269 		jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
270 
271 	return err;
272 }
273 
274 /**
275  * jbd2_journal_recover - recovers a on-disk journal
276  * @journal: the journal to recover
277  *
278  * The primary function for recovering the log contents when mounting a
279  * journaled device.
280  *
281  * Recovery is done in three passes.  In the first pass, we look for the
282  * end of the log.  In the second, we assemble the list of revoke
283  * blocks.  In the third and final pass, we replay any un-revoked blocks
284  * in the log.
285  */
jbd2_journal_recover(journal_t * journal)286 int jbd2_journal_recover(journal_t *journal)
287 {
288 	int			err, err2;
289 	struct recovery_info	info;
290 	errseq_t		wb_err;
291 	struct address_space	*mapping;
292 
293 	memset(&info, 0, sizeof(info));
294 
295 	/*
296 	 * The journal superblock's s_start field (the current log head)
297 	 * is always zero if, and only if, the journal was cleanly
298 	 * unmounted. We use its in-memory version j_tail here because
299 	 * jbd2_journal_wipe() could have updated it without updating journal
300 	 * superblock.
301 	 */
302 	if (!journal->j_tail) {
303 		journal_superblock_t *sb = journal->j_superblock;
304 
305 		jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
306 			  be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
307 		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
308 		journal->j_head = be32_to_cpu(sb->s_head);
309 		return 0;
310 	}
311 
312 	wb_err = 0;
313 	mapping = journal->j_fs_dev->bd_inode->i_mapping;
314 	errseq_check_and_advance(&mapping->wb_err, &wb_err);
315 	err = do_one_pass(journal, &info, PASS_SCAN);
316 	if (!err)
317 		err = do_one_pass(journal, &info, PASS_REVOKE);
318 	if (!err)
319 		err = do_one_pass(journal, &info, PASS_REPLAY);
320 
321 	jbd2_debug(1, "JBD2: recovery, exit status %d, "
322 		  "recovered transactions %u to %u\n",
323 		  err, info.start_transaction, info.end_transaction);
324 	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
325 		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
326 
327 	/* Restart the log at the next transaction ID, thus invalidating
328 	 * any existing commit records in the log. */
329 	journal->j_transaction_sequence = ++info.end_transaction;
330 	journal->j_head = info.head_block;
331 	jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
332 		  journal->j_transaction_sequence, journal->j_head);
333 
334 	jbd2_journal_clear_revoke(journal);
335 	err2 = sync_blockdev(journal->j_fs_dev);
336 	if (!err)
337 		err = err2;
338 	err2 = errseq_check_and_advance(&mapping->wb_err, &wb_err);
339 	if (!err)
340 		err = err2;
341 	/* Make sure all replayed data is on permanent storage */
342 	if (journal->j_flags & JBD2_BARRIER) {
343 		err2 = blkdev_issue_flush(journal->j_fs_dev);
344 		if (!err)
345 			err = err2;
346 	}
347 	return err;
348 }
349 
350 /**
351  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
352  * @journal: journal to startup
353  *
354  * Locate any valid recovery information from the journal and set up the
355  * journal structures in memory to ignore it (presumably because the
356  * caller has evidence that it is out of date).
357  * This function doesn't appear to be exported..
358  *
359  * We perform one pass over the journal to allow us to tell the user how
360  * much recovery information is being erased, and to let us initialise
361  * the journal transaction sequence numbers to the next unused ID.
362  */
jbd2_journal_skip_recovery(journal_t * journal)363 int jbd2_journal_skip_recovery(journal_t *journal)
364 {
365 	int			err;
366 
367 	struct recovery_info	info;
368 
369 	memset (&info, 0, sizeof(info));
370 
371 	err = do_one_pass(journal, &info, PASS_SCAN);
372 
373 	if (err) {
374 		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
375 		++journal->j_transaction_sequence;
376 		journal->j_head = journal->j_first;
377 	} else {
378 #ifdef CONFIG_JBD2_DEBUG
379 		int dropped = info.end_transaction -
380 			be32_to_cpu(journal->j_superblock->s_sequence);
381 		jbd2_debug(1,
382 			  "JBD2: ignoring %d transaction%s from the journal.\n",
383 			  dropped, (dropped == 1) ? "" : "s");
384 #endif
385 		journal->j_transaction_sequence = ++info.end_transaction;
386 		journal->j_head = info.head_block;
387 	}
388 
389 	journal->j_tail = 0;
390 	return err;
391 }
392 
read_tag_block(journal_t * journal,journal_block_tag_t * tag)393 static inline unsigned long long read_tag_block(journal_t *journal,
394 						journal_block_tag_t *tag)
395 {
396 	unsigned long long block = be32_to_cpu(tag->t_blocknr);
397 	if (jbd2_has_feature_64bit(journal))
398 		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
399 	return block;
400 }
401 
402 /*
403  * calc_chksums calculates the checksums for the blocks described in the
404  * descriptor block.
405  */
calc_chksums(journal_t * journal,struct buffer_head * bh,unsigned long * next_log_block,__u32 * crc32_sum)406 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
407 			unsigned long *next_log_block, __u32 *crc32_sum)
408 {
409 	int i, num_blks, err;
410 	unsigned long io_block;
411 	struct buffer_head *obh;
412 
413 	num_blks = count_tags(journal, bh);
414 	/* Calculate checksum of the descriptor block. */
415 	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
416 
417 	for (i = 0; i < num_blks; i++) {
418 		io_block = (*next_log_block)++;
419 		wrap(journal, *next_log_block);
420 		err = jread(&obh, journal, io_block);
421 		if (err) {
422 			printk(KERN_ERR "JBD2: IO error %d recovering block "
423 				"%lu in log\n", err, io_block);
424 			return 1;
425 		} else {
426 			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
427 				     obh->b_size);
428 		}
429 		put_bh(obh);
430 	}
431 	return 0;
432 }
433 
jbd2_commit_block_csum_verify(journal_t * j,void * buf)434 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
435 {
436 	struct commit_header *h;
437 	__be32 provided;
438 	__u32 calculated;
439 
440 	if (!jbd2_journal_has_csum_v2or3(j))
441 		return 1;
442 
443 	h = buf;
444 	provided = h->h_chksum[0];
445 	h->h_chksum[0] = 0;
446 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
447 	h->h_chksum[0] = provided;
448 
449 	return provided == cpu_to_be32(calculated);
450 }
451 
jbd2_commit_block_csum_verify_partial(journal_t * j,void * buf)452 static bool jbd2_commit_block_csum_verify_partial(journal_t *j, void *buf)
453 {
454 	struct commit_header *h;
455 	__be32 provided;
456 	__u32 calculated;
457 	void *tmpbuf;
458 
459 	tmpbuf = kzalloc(j->j_blocksize, GFP_KERNEL);
460 	if (!tmpbuf)
461 		return false;
462 
463 	memcpy(tmpbuf, buf, sizeof(struct commit_header));
464 	h = tmpbuf;
465 	provided = h->h_chksum[0];
466 	h->h_chksum[0] = 0;
467 	calculated = jbd2_chksum(j, j->j_csum_seed, tmpbuf, j->j_blocksize);
468 	kfree(tmpbuf);
469 
470 	return provided == cpu_to_be32(calculated);
471 }
472 
jbd2_block_tag_csum_verify(journal_t * j,journal_block_tag_t * tag,journal_block_tag3_t * tag3,void * buf,__u32 sequence)473 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
474 				      journal_block_tag3_t *tag3,
475 				      void *buf, __u32 sequence)
476 {
477 	__u32 csum32;
478 	__be32 seq;
479 
480 	if (!jbd2_journal_has_csum_v2or3(j))
481 		return 1;
482 
483 	seq = cpu_to_be32(sequence);
484 	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
485 	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
486 
487 	if (jbd2_has_feature_csum3(j))
488 		return tag3->t_checksum == cpu_to_be32(csum32);
489 	else
490 		return tag->t_checksum == cpu_to_be16(csum32);
491 }
492 
do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)493 static int do_one_pass(journal_t *journal,
494 			struct recovery_info *info, enum passtype pass)
495 {
496 	unsigned int		first_commit_ID, next_commit_ID;
497 	unsigned long		next_log_block, head_block;
498 	int			err, success = 0;
499 	journal_superblock_t *	sb;
500 	journal_header_t *	tmp;
501 	struct buffer_head *	bh;
502 	unsigned int		sequence;
503 	int			blocktype;
504 	int			tag_bytes = journal_tag_bytes(journal);
505 	__u32			crc32_sum = ~0; /* Transactional Checksums */
506 	int			descr_csum_size = 0;
507 	int			block_error = 0;
508 	bool			need_check_commit_time = false;
509 	__u64			last_trans_commit_time = 0, commit_time;
510 
511 	/*
512 	 * First thing is to establish what we expect to find in the log
513 	 * (in terms of transaction IDs), and where (in terms of log
514 	 * block offsets): query the superblock.
515 	 */
516 
517 	sb = journal->j_superblock;
518 	next_commit_ID = be32_to_cpu(sb->s_sequence);
519 	next_log_block = be32_to_cpu(sb->s_start);
520 	head_block = next_log_block;
521 
522 	first_commit_ID = next_commit_ID;
523 	if (pass == PASS_SCAN)
524 		info->start_transaction = first_commit_ID;
525 
526 	jbd2_debug(1, "Starting recovery pass %d\n", pass);
527 
528 	/*
529 	 * Now we walk through the log, transaction by transaction,
530 	 * making sure that each transaction has a commit block in the
531 	 * expected place.  Each complete transaction gets replayed back
532 	 * into the main filesystem.
533 	 */
534 
535 	while (1) {
536 		int			flags;
537 		char *			tagp;
538 		journal_block_tag_t	tag;
539 		struct buffer_head *	obh;
540 		struct buffer_head *	nbh;
541 
542 		cond_resched();
543 
544 		/* If we already know where to stop the log traversal,
545 		 * check right now that we haven't gone past the end of
546 		 * the log. */
547 
548 		if (pass != PASS_SCAN)
549 			if (tid_geq(next_commit_ID, info->end_transaction))
550 				break;
551 
552 		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
553 			  next_commit_ID, next_log_block, journal->j_last);
554 
555 		/* Skip over each chunk of the transaction looking
556 		 * either the next descriptor block or the final commit
557 		 * record. */
558 
559 		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
560 		err = jread(&bh, journal, next_log_block);
561 		if (err)
562 			goto failed;
563 
564 		next_log_block++;
565 		wrap(journal, next_log_block);
566 
567 		/* What kind of buffer is it?
568 		 *
569 		 * If it is a descriptor block, check that it has the
570 		 * expected sequence number.  Otherwise, we're all done
571 		 * here. */
572 
573 		tmp = (journal_header_t *)bh->b_data;
574 
575 		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
576 			brelse(bh);
577 			break;
578 		}
579 
580 		blocktype = be32_to_cpu(tmp->h_blocktype);
581 		sequence = be32_to_cpu(tmp->h_sequence);
582 		jbd2_debug(3, "Found magic %d, sequence %d\n",
583 			  blocktype, sequence);
584 
585 		if (sequence != next_commit_ID) {
586 			brelse(bh);
587 			break;
588 		}
589 
590 		/* OK, we have a valid descriptor block which matches
591 		 * all of the sequence number checks.  What are we going
592 		 * to do with it?  That depends on the pass... */
593 
594 		switch(blocktype) {
595 		case JBD2_DESCRIPTOR_BLOCK:
596 			/* Verify checksum first */
597 			if (jbd2_journal_has_csum_v2or3(journal))
598 				descr_csum_size =
599 					sizeof(struct jbd2_journal_block_tail);
600 			if (descr_csum_size > 0 &&
601 			    !jbd2_descriptor_block_csum_verify(journal,
602 							       bh->b_data)) {
603 				/*
604 				 * PASS_SCAN can see stale blocks due to lazy
605 				 * journal init. Don't error out on those yet.
606 				 */
607 				if (pass != PASS_SCAN) {
608 					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
609 					       next_log_block);
610 					err = -EFSBADCRC;
611 					brelse(bh);
612 					goto failed;
613 				}
614 				need_check_commit_time = true;
615 				jbd2_debug(1,
616 					"invalid descriptor block found in %lu\n",
617 					next_log_block);
618 			}
619 
620 			/* If it is a valid descriptor block, replay it
621 			 * in pass REPLAY; if journal_checksums enabled, then
622 			 * calculate checksums in PASS_SCAN, otherwise,
623 			 * just skip over the blocks it describes. */
624 			if (pass != PASS_REPLAY) {
625 				if (pass == PASS_SCAN &&
626 				    jbd2_has_feature_checksum(journal) &&
627 				    !need_check_commit_time &&
628 				    !info->end_transaction) {
629 					if (calc_chksums(journal, bh,
630 							&next_log_block,
631 							&crc32_sum)) {
632 						put_bh(bh);
633 						break;
634 					}
635 					put_bh(bh);
636 					continue;
637 				}
638 				next_log_block += count_tags(journal, bh);
639 				wrap(journal, next_log_block);
640 				put_bh(bh);
641 				continue;
642 			}
643 
644 			/* A descriptor block: we can now write all of
645 			 * the data blocks.  Yay, useful work is finally
646 			 * getting done here! */
647 
648 			tagp = &bh->b_data[sizeof(journal_header_t)];
649 			while ((tagp - bh->b_data + tag_bytes)
650 			       <= journal->j_blocksize - descr_csum_size) {
651 				unsigned long io_block;
652 
653 				memcpy(&tag, tagp, sizeof(tag));
654 				flags = be16_to_cpu(tag.t_flags);
655 
656 				io_block = next_log_block++;
657 				wrap(journal, next_log_block);
658 				err = jread(&obh, journal, io_block);
659 				if (err) {
660 					/* Recover what we can, but
661 					 * report failure at the end. */
662 					success = err;
663 					printk(KERN_ERR
664 						"JBD2: IO error %d recovering "
665 						"block %ld in log\n",
666 						err, io_block);
667 				} else {
668 					unsigned long long blocknr;
669 
670 					J_ASSERT(obh != NULL);
671 					blocknr = read_tag_block(journal,
672 								 &tag);
673 
674 					/* If the block has been
675 					 * revoked, then we're all done
676 					 * here. */
677 					if (jbd2_journal_test_revoke
678 					    (journal, blocknr,
679 					     next_commit_ID)) {
680 						brelse(obh);
681 						++info->nr_revoke_hits;
682 						goto skip_write;
683 					}
684 
685 					/* Look for block corruption */
686 					if (!jbd2_block_tag_csum_verify(
687 			journal, &tag, (journal_block_tag3_t *)tagp,
688 			obh->b_data, be32_to_cpu(tmp->h_sequence))) {
689 						brelse(obh);
690 						success = -EFSBADCRC;
691 						printk(KERN_ERR "JBD2: Invalid "
692 						       "checksum recovering "
693 						       "data block %llu in "
694 						       "log\n", blocknr);
695 						block_error = 1;
696 						goto skip_write;
697 					}
698 
699 					/* Find a buffer for the new
700 					 * data being restored */
701 					nbh = __getblk(journal->j_fs_dev,
702 							blocknr,
703 							journal->j_blocksize);
704 					if (nbh == NULL) {
705 						printk(KERN_ERR
706 						       "JBD2: Out of memory "
707 						       "during recovery.\n");
708 						err = -ENOMEM;
709 						brelse(bh);
710 						brelse(obh);
711 						goto failed;
712 					}
713 
714 					lock_buffer(nbh);
715 					memcpy(nbh->b_data, obh->b_data,
716 							journal->j_blocksize);
717 					if (flags & JBD2_FLAG_ESCAPE) {
718 						*((__be32 *)nbh->b_data) =
719 						cpu_to_be32(JBD2_MAGIC_NUMBER);
720 					}
721 
722 					BUFFER_TRACE(nbh, "marking dirty");
723 					set_buffer_uptodate(nbh);
724 					mark_buffer_dirty(nbh);
725 					BUFFER_TRACE(nbh, "marking uptodate");
726 					++info->nr_replays;
727 					unlock_buffer(nbh);
728 					brelse(obh);
729 					brelse(nbh);
730 				}
731 
732 			skip_write:
733 				tagp += tag_bytes;
734 				if (!(flags & JBD2_FLAG_SAME_UUID))
735 					tagp += 16;
736 
737 				if (flags & JBD2_FLAG_LAST_TAG)
738 					break;
739 			}
740 
741 			brelse(bh);
742 			continue;
743 
744 		case JBD2_COMMIT_BLOCK:
745 			/*     How to differentiate between interrupted commit
746 			 *               and journal corruption ?
747 			 *
748 			 * {nth transaction}
749 			 *        Checksum Verification Failed
750 			 *			 |
751 			 *		 ____________________
752 			 *		|		     |
753 			 * 	async_commit             sync_commit
754 			 *     		|                    |
755 			 *		| GO TO NEXT    "Journal Corruption"
756 			 *		| TRANSACTION
757 			 *		|
758 			 * {(n+1)th transanction}
759 			 *		|
760 			 * 	 _______|______________
761 			 * 	|	 	      |
762 			 * Commit block found	Commit block not found
763 			 *      |		      |
764 			 * "Journal Corruption"       |
765 			 *		 _____________|_________
766 			 *     		|	           	|
767 			 *	nth trans corrupt	OR   nth trans
768 			 *	and (n+1)th interrupted     interrupted
769 			 *	before commit block
770 			 *      could reach the disk.
771 			 *	(Cannot find the difference in above
772 			 *	 mentioned conditions. Hence assume
773 			 *	 "Interrupted Commit".)
774 			 */
775 			commit_time = be64_to_cpu(
776 				((struct commit_header *)bh->b_data)->h_commit_sec);
777 			/*
778 			 * If need_check_commit_time is set, it means we are in
779 			 * PASS_SCAN and csum verify failed before. If
780 			 * commit_time is increasing, it's the same journal,
781 			 * otherwise it is stale journal block, just end this
782 			 * recovery.
783 			 */
784 			if (need_check_commit_time) {
785 				if (commit_time >= last_trans_commit_time) {
786 					pr_err("JBD2: Invalid checksum found in transaction %u\n",
787 					       next_commit_ID);
788 					err = -EFSBADCRC;
789 					brelse(bh);
790 					goto failed;
791 				}
792 			ignore_crc_mismatch:
793 				/*
794 				 * It likely does not belong to same journal,
795 				 * just end this recovery with success.
796 				 */
797 				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
798 					  next_commit_ID);
799 				brelse(bh);
800 				goto done;
801 			}
802 
803 			/*
804 			 * Found an expected commit block: if checksums
805 			 * are present, verify them in PASS_SCAN; else not
806 			 * much to do other than move on to the next sequence
807 			 * number.
808 			 */
809 			if (pass == PASS_SCAN &&
810 			    jbd2_has_feature_checksum(journal)) {
811 				struct commit_header *cbh =
812 					(struct commit_header *)bh->b_data;
813 				unsigned found_chksum =
814 					be32_to_cpu(cbh->h_chksum[0]);
815 
816 				if (info->end_transaction) {
817 					journal->j_failed_commit =
818 						info->end_transaction;
819 					brelse(bh);
820 					break;
821 				}
822 
823 				/* Neither checksum match nor unused? */
824 				if (!((crc32_sum == found_chksum &&
825 				       cbh->h_chksum_type ==
826 						JBD2_CRC32_CHKSUM &&
827 				       cbh->h_chksum_size ==
828 						JBD2_CRC32_CHKSUM_SIZE) ||
829 				      (cbh->h_chksum_type == 0 &&
830 				       cbh->h_chksum_size == 0 &&
831 				       found_chksum == 0)))
832 					goto chksum_error;
833 
834 				crc32_sum = ~0;
835 			}
836 			if (pass == PASS_SCAN &&
837 			    !jbd2_commit_block_csum_verify(journal,
838 							   bh->b_data)) {
839 				if (jbd2_commit_block_csum_verify_partial(
840 								  journal,
841 								  bh->b_data)) {
842 					pr_notice("JBD2: Find incomplete commit block in transaction %u block %lu\n",
843 						  next_commit_ID, next_log_block);
844 					goto chksum_ok;
845 				}
846 			chksum_error:
847 				if (commit_time < last_trans_commit_time)
848 					goto ignore_crc_mismatch;
849 				info->end_transaction = next_commit_ID;
850 				info->head_block = head_block;
851 
852 				if (!jbd2_has_feature_async_commit(journal)) {
853 					journal->j_failed_commit =
854 						next_commit_ID;
855 					brelse(bh);
856 					break;
857 				}
858 			}
859 			if (pass == PASS_SCAN) {
860 			chksum_ok:
861 				last_trans_commit_time = commit_time;
862 				head_block = next_log_block;
863 			}
864 			brelse(bh);
865 			next_commit_ID++;
866 			continue;
867 
868 		case JBD2_REVOKE_BLOCK:
869 			/*
870 			 * Check revoke block crc in pass_scan, if csum verify
871 			 * failed, check commit block time later.
872 			 */
873 			if (pass == PASS_SCAN &&
874 			    !jbd2_descriptor_block_csum_verify(journal,
875 							       bh->b_data)) {
876 				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
877 					  next_log_block);
878 				need_check_commit_time = true;
879 			}
880 
881 			/* If we aren't in the REVOKE pass, then we can
882 			 * just skip over this block. */
883 			if (pass != PASS_REVOKE) {
884 				brelse(bh);
885 				continue;
886 			}
887 
888 			err = scan_revoke_records(journal, bh,
889 						  next_commit_ID, info);
890 			brelse(bh);
891 			if (err)
892 				goto failed;
893 			continue;
894 
895 		default:
896 			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
897 				  blocktype);
898 			brelse(bh);
899 			goto done;
900 		}
901 	}
902 
903  done:
904 	/*
905 	 * We broke out of the log scan loop: either we came to the
906 	 * known end of the log or we found an unexpected block in the
907 	 * log.  If the latter happened, then we know that the "current"
908 	 * transaction marks the end of the valid log.
909 	 */
910 
911 	if (pass == PASS_SCAN) {
912 		if (!info->end_transaction)
913 			info->end_transaction = next_commit_ID;
914 		if (!info->head_block)
915 			info->head_block = head_block;
916 	} else {
917 		/* It's really bad news if different passes end up at
918 		 * different places (but possible due to IO errors). */
919 		if (info->end_transaction != next_commit_ID) {
920 			printk(KERN_ERR "JBD2: recovery pass %d ended at "
921 				"transaction %u, expected %u\n",
922 				pass, next_commit_ID, info->end_transaction);
923 			if (!success)
924 				success = -EIO;
925 		}
926 	}
927 
928 	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
929 		err = fc_do_one_pass(journal, info, pass);
930 		if (err)
931 			success = err;
932 	}
933 
934 	if (block_error && success == 0)
935 		success = -EIO;
936 	return success;
937 
938  failed:
939 	return err;
940 }
941 
942 /* Scan a revoke record, marking all blocks mentioned as revoked. */
943 
scan_revoke_records(journal_t * journal,struct buffer_head * bh,tid_t sequence,struct recovery_info * info)944 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
945 			       tid_t sequence, struct recovery_info *info)
946 {
947 	jbd2_journal_revoke_header_t *header;
948 	int offset, max;
949 	unsigned csum_size = 0;
950 	__u32 rcount;
951 	int record_len = 4;
952 
953 	header = (jbd2_journal_revoke_header_t *) bh->b_data;
954 	offset = sizeof(jbd2_journal_revoke_header_t);
955 	rcount = be32_to_cpu(header->r_count);
956 
957 	if (jbd2_journal_has_csum_v2or3(journal))
958 		csum_size = sizeof(struct jbd2_journal_block_tail);
959 	if (rcount > journal->j_blocksize - csum_size)
960 		return -EINVAL;
961 	max = rcount;
962 
963 	if (jbd2_has_feature_64bit(journal))
964 		record_len = 8;
965 
966 	while (offset + record_len <= max) {
967 		unsigned long long blocknr;
968 		int err;
969 
970 		if (record_len == 4)
971 			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
972 		else
973 			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
974 		offset += record_len;
975 		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
976 		if (err)
977 			return err;
978 		++info->nr_revokes;
979 	}
980 	return 0;
981 }
982