xref: /openbmc/linux/fs/jbd2/recovery.c (revision e16c4c24)
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 	journal_superblock_t *	sb;
290 
291 	struct recovery_info	info;
292 	errseq_t		wb_err;
293 	struct address_space	*mapping;
294 
295 	memset(&info, 0, sizeof(info));
296 	sb = journal->j_superblock;
297 
298 	/*
299 	 * The journal superblock's s_start field (the current log head)
300 	 * is always zero if, and only if, the journal was cleanly
301 	 * unmounted.
302 	 */
303 	if (!sb->s_start) {
304 		jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
305 			  be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
306 		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
307 		journal->j_head = be32_to_cpu(sb->s_head);
308 		return 0;
309 	}
310 
311 	wb_err = 0;
312 	mapping = journal->j_fs_dev->bd_inode->i_mapping;
313 	errseq_check_and_advance(&mapping->wb_err, &wb_err);
314 	err = do_one_pass(journal, &info, PASS_SCAN);
315 	if (!err)
316 		err = do_one_pass(journal, &info, PASS_REVOKE);
317 	if (!err)
318 		err = do_one_pass(journal, &info, PASS_REPLAY);
319 
320 	jbd2_debug(1, "JBD2: recovery, exit status %d, "
321 		  "recovered transactions %u to %u\n",
322 		  err, info.start_transaction, info.end_transaction);
323 	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
324 		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
325 
326 	/* Restart the log at the next transaction ID, thus invalidating
327 	 * any existing commit records in the log. */
328 	journal->j_transaction_sequence = ++info.end_transaction;
329 	journal->j_head = info.head_block;
330 	jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
331 		  journal->j_transaction_sequence, journal->j_head);
332 
333 	jbd2_journal_clear_revoke(journal);
334 	err2 = sync_blockdev(journal->j_fs_dev);
335 	if (!err)
336 		err = err2;
337 	err2 = errseq_check_and_advance(&mapping->wb_err, &wb_err);
338 	if (!err)
339 		err = err2;
340 	/* Make sure all replayed data is on permanent storage */
341 	if (journal->j_flags & JBD2_BARRIER) {
342 		err2 = blkdev_issue_flush(journal->j_fs_dev);
343 		if (!err)
344 			err = err2;
345 	}
346 	return err;
347 }
348 
349 /**
350  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
351  * @journal: journal to startup
352  *
353  * Locate any valid recovery information from the journal and set up the
354  * journal structures in memory to ignore it (presumably because the
355  * caller has evidence that it is out of date).
356  * This function doesn't appear to be exported..
357  *
358  * We perform one pass over the journal to allow us to tell the user how
359  * much recovery information is being erased, and to let us initialise
360  * the journal transaction sequence numbers to the next unused ID.
361  */
jbd2_journal_skip_recovery(journal_t * journal)362 int jbd2_journal_skip_recovery(journal_t *journal)
363 {
364 	int			err;
365 
366 	struct recovery_info	info;
367 
368 	memset (&info, 0, sizeof(info));
369 
370 	err = do_one_pass(journal, &info, PASS_SCAN);
371 
372 	if (err) {
373 		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
374 		++journal->j_transaction_sequence;
375 		journal->j_head = journal->j_first;
376 	} else {
377 #ifdef CONFIG_JBD2_DEBUG
378 		int dropped = info.end_transaction -
379 			be32_to_cpu(journal->j_superblock->s_sequence);
380 		jbd2_debug(1,
381 			  "JBD2: ignoring %d transaction%s from the journal.\n",
382 			  dropped, (dropped == 1) ? "" : "s");
383 #endif
384 		journal->j_transaction_sequence = ++info.end_transaction;
385 		journal->j_head = info.head_block;
386 	}
387 
388 	journal->j_tail = 0;
389 	return err;
390 }
391 
read_tag_block(journal_t * journal,journal_block_tag_t * tag)392 static inline unsigned long long read_tag_block(journal_t *journal,
393 						journal_block_tag_t *tag)
394 {
395 	unsigned long long block = be32_to_cpu(tag->t_blocknr);
396 	if (jbd2_has_feature_64bit(journal))
397 		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
398 	return block;
399 }
400 
401 /*
402  * calc_chksums calculates the checksums for the blocks described in the
403  * descriptor block.
404  */
calc_chksums(journal_t * journal,struct buffer_head * bh,unsigned long * next_log_block,__u32 * crc32_sum)405 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
406 			unsigned long *next_log_block, __u32 *crc32_sum)
407 {
408 	int i, num_blks, err;
409 	unsigned long io_block;
410 	struct buffer_head *obh;
411 
412 	num_blks = count_tags(journal, bh);
413 	/* Calculate checksum of the descriptor block. */
414 	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
415 
416 	for (i = 0; i < num_blks; i++) {
417 		io_block = (*next_log_block)++;
418 		wrap(journal, *next_log_block);
419 		err = jread(&obh, journal, io_block);
420 		if (err) {
421 			printk(KERN_ERR "JBD2: IO error %d recovering block "
422 				"%lu in log\n", err, io_block);
423 			return 1;
424 		} else {
425 			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
426 				     obh->b_size);
427 		}
428 		put_bh(obh);
429 	}
430 	return 0;
431 }
432 
jbd2_commit_block_csum_verify(journal_t * j,void * buf)433 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
434 {
435 	struct commit_header *h;
436 	__be32 provided;
437 	__u32 calculated;
438 
439 	if (!jbd2_journal_has_csum_v2or3(j))
440 		return 1;
441 
442 	h = buf;
443 	provided = h->h_chksum[0];
444 	h->h_chksum[0] = 0;
445 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
446 	h->h_chksum[0] = provided;
447 
448 	return provided == cpu_to_be32(calculated);
449 }
450 
jbd2_commit_block_csum_verify_partial(journal_t * j,void * buf)451 static bool jbd2_commit_block_csum_verify_partial(journal_t *j, void *buf)
452 {
453 	struct commit_header *h;
454 	__be32 provided;
455 	__u32 calculated;
456 	void *tmpbuf;
457 
458 	tmpbuf = kzalloc(j->j_blocksize, GFP_KERNEL);
459 	if (!tmpbuf)
460 		return false;
461 
462 	memcpy(tmpbuf, buf, sizeof(struct commit_header));
463 	h = tmpbuf;
464 	provided = h->h_chksum[0];
465 	h->h_chksum[0] = 0;
466 	calculated = jbd2_chksum(j, j->j_csum_seed, tmpbuf, j->j_blocksize);
467 	kfree(tmpbuf);
468 
469 	return provided == cpu_to_be32(calculated);
470 }
471 
jbd2_block_tag_csum_verify(journal_t * j,journal_block_tag_t * tag,journal_block_tag3_t * tag3,void * buf,__u32 sequence)472 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
473 				      journal_block_tag3_t *tag3,
474 				      void *buf, __u32 sequence)
475 {
476 	__u32 csum32;
477 	__be32 seq;
478 
479 	if (!jbd2_journal_has_csum_v2or3(j))
480 		return 1;
481 
482 	seq = cpu_to_be32(sequence);
483 	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
484 	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
485 
486 	if (jbd2_has_feature_csum3(j))
487 		return tag3->t_checksum == cpu_to_be32(csum32);
488 	else
489 		return tag->t_checksum == cpu_to_be16(csum32);
490 }
491 
do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)492 static int do_one_pass(journal_t *journal,
493 			struct recovery_info *info, enum passtype pass)
494 {
495 	unsigned int		first_commit_ID, next_commit_ID;
496 	unsigned long		next_log_block, head_block;
497 	int			err, success = 0;
498 	journal_superblock_t *	sb;
499 	journal_header_t *	tmp;
500 	struct buffer_head *	bh;
501 	unsigned int		sequence;
502 	int			blocktype;
503 	int			tag_bytes = journal_tag_bytes(journal);
504 	__u32			crc32_sum = ~0; /* Transactional Checksums */
505 	int			descr_csum_size = 0;
506 	int			block_error = 0;
507 	bool			need_check_commit_time = false;
508 	__u64			last_trans_commit_time = 0, commit_time;
509 
510 	/*
511 	 * First thing is to establish what we expect to find in the log
512 	 * (in terms of transaction IDs), and where (in terms of log
513 	 * block offsets): query the superblock.
514 	 */
515 
516 	sb = journal->j_superblock;
517 	next_commit_ID = be32_to_cpu(sb->s_sequence);
518 	next_log_block = be32_to_cpu(sb->s_start);
519 	head_block = next_log_block;
520 
521 	first_commit_ID = next_commit_ID;
522 	if (pass == PASS_SCAN)
523 		info->start_transaction = first_commit_ID;
524 
525 	jbd2_debug(1, "Starting recovery pass %d\n", pass);
526 
527 	/*
528 	 * Now we walk through the log, transaction by transaction,
529 	 * making sure that each transaction has a commit block in the
530 	 * expected place.  Each complete transaction gets replayed back
531 	 * into the main filesystem.
532 	 */
533 
534 	while (1) {
535 		int			flags;
536 		char *			tagp;
537 		journal_block_tag_t	tag;
538 		struct buffer_head *	obh;
539 		struct buffer_head *	nbh;
540 
541 		cond_resched();
542 
543 		/* If we already know where to stop the log traversal,
544 		 * check right now that we haven't gone past the end of
545 		 * the log. */
546 
547 		if (pass != PASS_SCAN)
548 			if (tid_geq(next_commit_ID, info->end_transaction))
549 				break;
550 
551 		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
552 			  next_commit_ID, next_log_block, journal->j_last);
553 
554 		/* Skip over each chunk of the transaction looking
555 		 * either the next descriptor block or the final commit
556 		 * record. */
557 
558 		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
559 		err = jread(&bh, journal, next_log_block);
560 		if (err)
561 			goto failed;
562 
563 		next_log_block++;
564 		wrap(journal, next_log_block);
565 
566 		/* What kind of buffer is it?
567 		 *
568 		 * If it is a descriptor block, check that it has the
569 		 * expected sequence number.  Otherwise, we're all done
570 		 * here. */
571 
572 		tmp = (journal_header_t *)bh->b_data;
573 
574 		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
575 			brelse(bh);
576 			break;
577 		}
578 
579 		blocktype = be32_to_cpu(tmp->h_blocktype);
580 		sequence = be32_to_cpu(tmp->h_sequence);
581 		jbd2_debug(3, "Found magic %d, sequence %d\n",
582 			  blocktype, sequence);
583 
584 		if (sequence != next_commit_ID) {
585 			brelse(bh);
586 			break;
587 		}
588 
589 		/* OK, we have a valid descriptor block which matches
590 		 * all of the sequence number checks.  What are we going
591 		 * to do with it?  That depends on the pass... */
592 
593 		switch(blocktype) {
594 		case JBD2_DESCRIPTOR_BLOCK:
595 			/* Verify checksum first */
596 			if (jbd2_journal_has_csum_v2or3(journal))
597 				descr_csum_size =
598 					sizeof(struct jbd2_journal_block_tail);
599 			if (descr_csum_size > 0 &&
600 			    !jbd2_descriptor_block_csum_verify(journal,
601 							       bh->b_data)) {
602 				/*
603 				 * PASS_SCAN can see stale blocks due to lazy
604 				 * journal init. Don't error out on those yet.
605 				 */
606 				if (pass != PASS_SCAN) {
607 					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
608 					       next_log_block);
609 					err = -EFSBADCRC;
610 					brelse(bh);
611 					goto failed;
612 				}
613 				need_check_commit_time = true;
614 				jbd2_debug(1,
615 					"invalid descriptor block found in %lu\n",
616 					next_log_block);
617 			}
618 
619 			/* If it is a valid descriptor block, replay it
620 			 * in pass REPLAY; if journal_checksums enabled, then
621 			 * calculate checksums in PASS_SCAN, otherwise,
622 			 * just skip over the blocks it describes. */
623 			if (pass != PASS_REPLAY) {
624 				if (pass == PASS_SCAN &&
625 				    jbd2_has_feature_checksum(journal) &&
626 				    !need_check_commit_time &&
627 				    !info->end_transaction) {
628 					if (calc_chksums(journal, bh,
629 							&next_log_block,
630 							&crc32_sum)) {
631 						put_bh(bh);
632 						break;
633 					}
634 					put_bh(bh);
635 					continue;
636 				}
637 				next_log_block += count_tags(journal, bh);
638 				wrap(journal, next_log_block);
639 				put_bh(bh);
640 				continue;
641 			}
642 
643 			/* A descriptor block: we can now write all of
644 			 * the data blocks.  Yay, useful work is finally
645 			 * getting done here! */
646 
647 			tagp = &bh->b_data[sizeof(journal_header_t)];
648 			while ((tagp - bh->b_data + tag_bytes)
649 			       <= journal->j_blocksize - descr_csum_size) {
650 				unsigned long io_block;
651 
652 				memcpy(&tag, tagp, sizeof(tag));
653 				flags = be16_to_cpu(tag.t_flags);
654 
655 				io_block = next_log_block++;
656 				wrap(journal, next_log_block);
657 				err = jread(&obh, journal, io_block);
658 				if (err) {
659 					/* Recover what we can, but
660 					 * report failure at the end. */
661 					success = err;
662 					printk(KERN_ERR
663 						"JBD2: IO error %d recovering "
664 						"block %ld in log\n",
665 						err, io_block);
666 				} else {
667 					unsigned long long blocknr;
668 
669 					J_ASSERT(obh != NULL);
670 					blocknr = read_tag_block(journal,
671 								 &tag);
672 
673 					/* If the block has been
674 					 * revoked, then we're all done
675 					 * here. */
676 					if (jbd2_journal_test_revoke
677 					    (journal, blocknr,
678 					     next_commit_ID)) {
679 						brelse(obh);
680 						++info->nr_revoke_hits;
681 						goto skip_write;
682 					}
683 
684 					/* Look for block corruption */
685 					if (!jbd2_block_tag_csum_verify(
686 			journal, &tag, (journal_block_tag3_t *)tagp,
687 			obh->b_data, be32_to_cpu(tmp->h_sequence))) {
688 						brelse(obh);
689 						success = -EFSBADCRC;
690 						printk(KERN_ERR "JBD2: Invalid "
691 						       "checksum recovering "
692 						       "data block %llu in "
693 						       "log\n", blocknr);
694 						block_error = 1;
695 						goto skip_write;
696 					}
697 
698 					/* Find a buffer for the new
699 					 * data being restored */
700 					nbh = __getblk(journal->j_fs_dev,
701 							blocknr,
702 							journal->j_blocksize);
703 					if (nbh == NULL) {
704 						printk(KERN_ERR
705 						       "JBD2: Out of memory "
706 						       "during recovery.\n");
707 						err = -ENOMEM;
708 						brelse(bh);
709 						brelse(obh);
710 						goto failed;
711 					}
712 
713 					lock_buffer(nbh);
714 					memcpy(nbh->b_data, obh->b_data,
715 							journal->j_blocksize);
716 					if (flags & JBD2_FLAG_ESCAPE) {
717 						*((__be32 *)nbh->b_data) =
718 						cpu_to_be32(JBD2_MAGIC_NUMBER);
719 					}
720 
721 					BUFFER_TRACE(nbh, "marking dirty");
722 					set_buffer_uptodate(nbh);
723 					mark_buffer_dirty(nbh);
724 					BUFFER_TRACE(nbh, "marking uptodate");
725 					++info->nr_replays;
726 					unlock_buffer(nbh);
727 					brelse(obh);
728 					brelse(nbh);
729 				}
730 
731 			skip_write:
732 				tagp += tag_bytes;
733 				if (!(flags & JBD2_FLAG_SAME_UUID))
734 					tagp += 16;
735 
736 				if (flags & JBD2_FLAG_LAST_TAG)
737 					break;
738 			}
739 
740 			brelse(bh);
741 			continue;
742 
743 		case JBD2_COMMIT_BLOCK:
744 			/*     How to differentiate between interrupted commit
745 			 *               and journal corruption ?
746 			 *
747 			 * {nth transaction}
748 			 *        Checksum Verification Failed
749 			 *			 |
750 			 *		 ____________________
751 			 *		|		     |
752 			 * 	async_commit             sync_commit
753 			 *     		|                    |
754 			 *		| GO TO NEXT    "Journal Corruption"
755 			 *		| TRANSACTION
756 			 *		|
757 			 * {(n+1)th transanction}
758 			 *		|
759 			 * 	 _______|______________
760 			 * 	|	 	      |
761 			 * Commit block found	Commit block not found
762 			 *      |		      |
763 			 * "Journal Corruption"       |
764 			 *		 _____________|_________
765 			 *     		|	           	|
766 			 *	nth trans corrupt	OR   nth trans
767 			 *	and (n+1)th interrupted     interrupted
768 			 *	before commit block
769 			 *      could reach the disk.
770 			 *	(Cannot find the difference in above
771 			 *	 mentioned conditions. Hence assume
772 			 *	 "Interrupted Commit".)
773 			 */
774 			commit_time = be64_to_cpu(
775 				((struct commit_header *)bh->b_data)->h_commit_sec);
776 			/*
777 			 * If need_check_commit_time is set, it means we are in
778 			 * PASS_SCAN and csum verify failed before. If
779 			 * commit_time is increasing, it's the same journal,
780 			 * otherwise it is stale journal block, just end this
781 			 * recovery.
782 			 */
783 			if (need_check_commit_time) {
784 				if (commit_time >= last_trans_commit_time) {
785 					pr_err("JBD2: Invalid checksum found in transaction %u\n",
786 					       next_commit_ID);
787 					err = -EFSBADCRC;
788 					brelse(bh);
789 					goto failed;
790 				}
791 			ignore_crc_mismatch:
792 				/*
793 				 * It likely does not belong to same journal,
794 				 * just end this recovery with success.
795 				 */
796 				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
797 					  next_commit_ID);
798 				brelse(bh);
799 				goto done;
800 			}
801 
802 			/*
803 			 * Found an expected commit block: if checksums
804 			 * are present, verify them in PASS_SCAN; else not
805 			 * much to do other than move on to the next sequence
806 			 * number.
807 			 */
808 			if (pass == PASS_SCAN &&
809 			    jbd2_has_feature_checksum(journal)) {
810 				struct commit_header *cbh =
811 					(struct commit_header *)bh->b_data;
812 				unsigned found_chksum =
813 					be32_to_cpu(cbh->h_chksum[0]);
814 
815 				if (info->end_transaction) {
816 					journal->j_failed_commit =
817 						info->end_transaction;
818 					brelse(bh);
819 					break;
820 				}
821 
822 				/* Neither checksum match nor unused? */
823 				if (!((crc32_sum == found_chksum &&
824 				       cbh->h_chksum_type ==
825 						JBD2_CRC32_CHKSUM &&
826 				       cbh->h_chksum_size ==
827 						JBD2_CRC32_CHKSUM_SIZE) ||
828 				      (cbh->h_chksum_type == 0 &&
829 				       cbh->h_chksum_size == 0 &&
830 				       found_chksum == 0)))
831 					goto chksum_error;
832 
833 				crc32_sum = ~0;
834 			}
835 			if (pass == PASS_SCAN &&
836 			    !jbd2_commit_block_csum_verify(journal,
837 							   bh->b_data)) {
838 				if (jbd2_commit_block_csum_verify_partial(
839 								  journal,
840 								  bh->b_data)) {
841 					pr_notice("JBD2: Find incomplete commit block in transaction %u block %lu\n",
842 						  next_commit_ID, next_log_block);
843 					goto chksum_ok;
844 				}
845 			chksum_error:
846 				if (commit_time < last_trans_commit_time)
847 					goto ignore_crc_mismatch;
848 				info->end_transaction = next_commit_ID;
849 				info->head_block = head_block;
850 
851 				if (!jbd2_has_feature_async_commit(journal)) {
852 					journal->j_failed_commit =
853 						next_commit_ID;
854 					brelse(bh);
855 					break;
856 				}
857 			}
858 			if (pass == PASS_SCAN) {
859 			chksum_ok:
860 				last_trans_commit_time = commit_time;
861 				head_block = next_log_block;
862 			}
863 			brelse(bh);
864 			next_commit_ID++;
865 			continue;
866 
867 		case JBD2_REVOKE_BLOCK:
868 			/*
869 			 * Check revoke block crc in pass_scan, if csum verify
870 			 * failed, check commit block time later.
871 			 */
872 			if (pass == PASS_SCAN &&
873 			    !jbd2_descriptor_block_csum_verify(journal,
874 							       bh->b_data)) {
875 				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
876 					  next_log_block);
877 				need_check_commit_time = true;
878 			}
879 
880 			/* If we aren't in the REVOKE pass, then we can
881 			 * just skip over this block. */
882 			if (pass != PASS_REVOKE) {
883 				brelse(bh);
884 				continue;
885 			}
886 
887 			err = scan_revoke_records(journal, bh,
888 						  next_commit_ID, info);
889 			brelse(bh);
890 			if (err)
891 				goto failed;
892 			continue;
893 
894 		default:
895 			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
896 				  blocktype);
897 			brelse(bh);
898 			goto done;
899 		}
900 	}
901 
902  done:
903 	/*
904 	 * We broke out of the log scan loop: either we came to the
905 	 * known end of the log or we found an unexpected block in the
906 	 * log.  If the latter happened, then we know that the "current"
907 	 * transaction marks the end of the valid log.
908 	 */
909 
910 	if (pass == PASS_SCAN) {
911 		if (!info->end_transaction)
912 			info->end_transaction = next_commit_ID;
913 		if (!info->head_block)
914 			info->head_block = head_block;
915 	} else {
916 		/* It's really bad news if different passes end up at
917 		 * different places (but possible due to IO errors). */
918 		if (info->end_transaction != next_commit_ID) {
919 			printk(KERN_ERR "JBD2: recovery pass %d ended at "
920 				"transaction %u, expected %u\n",
921 				pass, next_commit_ID, info->end_transaction);
922 			if (!success)
923 				success = -EIO;
924 		}
925 	}
926 
927 	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
928 		err = fc_do_one_pass(journal, info, pass);
929 		if (err)
930 			success = err;
931 	}
932 
933 	if (block_error && success == 0)
934 		success = -EIO;
935 	return success;
936 
937  failed:
938 	return err;
939 }
940 
941 /* Scan a revoke record, marking all blocks mentioned as revoked. */
942 
scan_revoke_records(journal_t * journal,struct buffer_head * bh,tid_t sequence,struct recovery_info * info)943 static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
944 			       tid_t sequence, struct recovery_info *info)
945 {
946 	jbd2_journal_revoke_header_t *header;
947 	int offset, max;
948 	unsigned csum_size = 0;
949 	__u32 rcount;
950 	int record_len = 4;
951 
952 	header = (jbd2_journal_revoke_header_t *) bh->b_data;
953 	offset = sizeof(jbd2_journal_revoke_header_t);
954 	rcount = be32_to_cpu(header->r_count);
955 
956 	if (jbd2_journal_has_csum_v2or3(journal))
957 		csum_size = sizeof(struct jbd2_journal_block_tail);
958 	if (rcount > journal->j_blocksize - csum_size)
959 		return -EINVAL;
960 	max = rcount;
961 
962 	if (jbd2_has_feature_64bit(journal))
963 		record_len = 8;
964 
965 	while (offset + record_len <= max) {
966 		unsigned long long blocknr;
967 		int err;
968 
969 		if (record_len == 4)
970 			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
971 		else
972 			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
973 		offset += record_len;
974 		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
975 		if (err)
976 			return err;
977 		++info->nr_revokes;
978 	}
979 	return 0;
980 }
981