xref: /linux/fs/jbd2/recovery.c (revision 5c2a430e85994f4873ea5ec42091baa1153bc731)
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 *, enum passtype, 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 void do_readahead(journal_t *journal, unsigned int start)
69 {
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 		int 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 			goto failed;
98 
99 		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
100 			bufs[nbufs++] = bh;
101 			if (nbufs == MAXBUF) {
102 				bh_readahead_batch(nbufs, bufs, 0);
103 				journal_brelse_array(bufs, nbufs);
104 				nbufs = 0;
105 			}
106 		} else
107 			brelse(bh);
108 	}
109 
110 	if (nbufs)
111 		bh_readahead_batch(nbufs, bufs, 0);
112 
113 failed:
114 	if (nbufs)
115 		journal_brelse_array(bufs, nbufs);
116 }
117 
118 #endif /* __KERNEL__ */
119 
120 
121 /*
122  * Read a block from the journal
123  */
124 
jread(struct buffer_head ** bhp,journal_t * journal,unsigned int offset)125 static int jread(struct buffer_head **bhp, journal_t *journal,
126 		 unsigned int offset)
127 {
128 	int err;
129 	unsigned long long blocknr;
130 	struct buffer_head *bh;
131 
132 	*bhp = NULL;
133 
134 	if (offset >= journal->j_total_len) {
135 		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
136 		return -EFSCORRUPTED;
137 	}
138 
139 	err = jbd2_journal_bmap(journal, offset, &blocknr);
140 
141 	if (err) {
142 		printk(KERN_ERR "JBD2: bad block at offset %u\n",
143 			offset);
144 		return err;
145 	}
146 
147 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
148 	if (!bh)
149 		return -ENOMEM;
150 
151 	if (!buffer_uptodate(bh)) {
152 		/*
153 		 * If this is a brand new buffer, start readahead.
154 		 * Otherwise, we assume we are already reading it.
155 		 */
156 		bool need_readahead = !buffer_req(bh);
157 
158 		bh_read_nowait(bh, 0);
159 		if (need_readahead)
160 			do_readahead(journal, offset);
161 		wait_on_buffer(bh);
162 	}
163 
164 	if (!buffer_uptodate(bh)) {
165 		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
166 			offset);
167 		brelse(bh);
168 		return -EIO;
169 	}
170 
171 	*bhp = bh;
172 	return 0;
173 }
174 
jbd2_descriptor_block_csum_verify(journal_t * j,void * buf)175 static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
176 {
177 	struct jbd2_journal_block_tail *tail;
178 	__be32 provided;
179 	__u32 calculated;
180 
181 	if (!jbd2_journal_has_csum_v2or3(j))
182 		return 1;
183 
184 	tail = (struct jbd2_journal_block_tail *)((char *)buf +
185 		j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
186 	provided = tail->t_checksum;
187 	tail->t_checksum = 0;
188 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
189 	tail->t_checksum = provided;
190 
191 	return provided == cpu_to_be32(calculated);
192 }
193 
194 /*
195  * Count the number of in-use tags in a journal descriptor block.
196  */
197 
count_tags(journal_t * journal,struct buffer_head * bh)198 static int count_tags(journal_t *journal, struct buffer_head *bh)
199 {
200 	char *			tagp;
201 	journal_block_tag_t	tag;
202 	int			nr = 0, size = journal->j_blocksize;
203 	int			tag_bytes = journal_tag_bytes(journal);
204 
205 	if (jbd2_journal_has_csum_v2or3(journal))
206 		size -= sizeof(struct jbd2_journal_block_tail);
207 
208 	tagp = &bh->b_data[sizeof(journal_header_t)];
209 
210 	while ((tagp - bh->b_data + tag_bytes) <= size) {
211 		memcpy(&tag, tagp, sizeof(tag));
212 
213 		nr++;
214 		tagp += tag_bytes;
215 		if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
216 			tagp += 16;
217 
218 		if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
219 			break;
220 	}
221 
222 	return nr;
223 }
224 
225 
226 /* Make sure we wrap around the log correctly! */
227 #define wrap(journal, var)						\
228 do {									\
229 	if (var >= (journal)->j_last)					\
230 		var -= ((journal)->j_last - (journal)->j_first);	\
231 } while (0)
232 
fc_do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)233 static int fc_do_one_pass(journal_t *journal,
234 			  struct recovery_info *info, enum passtype pass)
235 {
236 	unsigned int expected_commit_id = info->end_transaction;
237 	unsigned long next_fc_block;
238 	struct buffer_head *bh;
239 	int err = 0;
240 
241 	next_fc_block = journal->j_fc_first;
242 	if (!journal->j_fc_replay_callback)
243 		return 0;
244 
245 	while (next_fc_block <= journal->j_fc_last) {
246 		jbd2_debug(3, "Fast commit replay: next block %ld\n",
247 			  next_fc_block);
248 		err = jread(&bh, journal, next_fc_block);
249 		if (err) {
250 			jbd2_debug(3, "Fast commit replay: read error\n");
251 			break;
252 		}
253 
254 		err = journal->j_fc_replay_callback(journal, bh, pass,
255 					next_fc_block - journal->j_fc_first,
256 					expected_commit_id);
257 		brelse(bh);
258 		next_fc_block++;
259 		if (err < 0 || err == JBD2_FC_REPLAY_STOP)
260 			break;
261 		err = 0;
262 	}
263 
264 	if (err)
265 		jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
266 
267 	return err;
268 }
269 
270 /**
271  * jbd2_journal_recover - recovers a on-disk journal
272  * @journal: the journal to recover
273  *
274  * The primary function for recovering the log contents when mounting a
275  * journaled device.
276  *
277  * Recovery is done in three passes.  In the first pass, we look for the
278  * end of the log.  In the second, we assemble the list of revoke
279  * blocks.  In the third and final pass, we replay any un-revoked blocks
280  * in the log.
281  */
jbd2_journal_recover(journal_t * journal)282 int jbd2_journal_recover(journal_t *journal)
283 {
284 	int			err, err2;
285 	struct recovery_info	info;
286 
287 	memset(&info, 0, sizeof(info));
288 
289 	/*
290 	 * The journal superblock's s_start field (the current log head)
291 	 * is always zero if, and only if, the journal was cleanly
292 	 * unmounted. We use its in-memory version j_tail here because
293 	 * jbd2_journal_wipe() could have updated it without updating journal
294 	 * superblock.
295 	 */
296 	if (!journal->j_tail) {
297 		journal_superblock_t *sb = journal->j_superblock;
298 
299 		jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
300 			  be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
301 		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
302 		journal->j_head = be32_to_cpu(sb->s_head);
303 		return 0;
304 	}
305 
306 	err = do_one_pass(journal, &info, PASS_SCAN);
307 	if (!err)
308 		err = do_one_pass(journal, &info, PASS_REVOKE);
309 	if (!err)
310 		err = do_one_pass(journal, &info, PASS_REPLAY);
311 
312 	jbd2_debug(1, "JBD2: recovery, exit status %d, "
313 		  "recovered transactions %u to %u\n",
314 		  err, info.start_transaction, info.end_transaction);
315 	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
316 		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
317 
318 	/* Restart the log at the next transaction ID, thus invalidating
319 	 * any existing commit records in the log. */
320 	journal->j_transaction_sequence = ++info.end_transaction;
321 	journal->j_head = info.head_block;
322 	jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
323 		  journal->j_transaction_sequence, journal->j_head);
324 
325 	jbd2_journal_clear_revoke(journal);
326 	/* Free revoke table allocated for replay */
327 	if (journal->j_revoke != journal->j_revoke_table[0] &&
328 	    journal->j_revoke != journal->j_revoke_table[1]) {
329 		jbd2_journal_destroy_revoke_table(journal->j_revoke);
330 		journal->j_revoke = journal->j_revoke_table[1];
331 	}
332 	err2 = sync_blockdev(journal->j_fs_dev);
333 	if (!err)
334 		err = err2;
335 	err2 = jbd2_check_fs_dev_write_error(journal);
336 	if (!err)
337 		err = err2;
338 	/* Make sure all replayed data is on permanent storage */
339 	if (journal->j_flags & JBD2_BARRIER) {
340 		err2 = blkdev_issue_flush(journal->j_fs_dev);
341 		if (!err)
342 			err = err2;
343 	}
344 	return err;
345 }
346 
347 /**
348  * jbd2_journal_skip_recovery - Start journal and wipe exiting records
349  * @journal: journal to startup
350  *
351  * Locate any valid recovery information from the journal and set up the
352  * journal structures in memory to ignore it (presumably because the
353  * caller has evidence that it is out of date).
354  * This function doesn't appear to be exported..
355  *
356  * We perform one pass over the journal to allow us to tell the user how
357  * much recovery information is being erased, and to let us initialise
358  * the journal transaction sequence numbers to the next unused ID.
359  */
jbd2_journal_skip_recovery(journal_t * journal)360 int jbd2_journal_skip_recovery(journal_t *journal)
361 {
362 	int			err;
363 
364 	struct recovery_info	info;
365 
366 	memset (&info, 0, sizeof(info));
367 
368 	err = do_one_pass(journal, &info, PASS_SCAN);
369 
370 	if (err) {
371 		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
372 		++journal->j_transaction_sequence;
373 		journal->j_head = journal->j_first;
374 	} else {
375 #ifdef CONFIG_JBD2_DEBUG
376 		int dropped = info.end_transaction -
377 			be32_to_cpu(journal->j_superblock->s_sequence);
378 		jbd2_debug(1,
379 			  "JBD2: ignoring %d transaction%s from the journal.\n",
380 			  dropped, str_plural(dropped));
381 #endif
382 		journal->j_transaction_sequence = ++info.end_transaction;
383 		journal->j_head = info.head_block;
384 	}
385 
386 	journal->j_tail = 0;
387 	return err;
388 }
389 
read_tag_block(journal_t * journal,journal_block_tag_t * tag)390 static inline unsigned long long read_tag_block(journal_t *journal,
391 						journal_block_tag_t *tag)
392 {
393 	unsigned long long block = be32_to_cpu(tag->t_blocknr);
394 	if (jbd2_has_feature_64bit(journal))
395 		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
396 	return block;
397 }
398 
399 /*
400  * calc_chksums calculates the checksums for the blocks described in the
401  * descriptor block.
402  */
calc_chksums(journal_t * journal,struct buffer_head * bh,unsigned long * next_log_block,__u32 * crc32_sum)403 static int calc_chksums(journal_t *journal, struct buffer_head *bh,
404 			unsigned long *next_log_block, __u32 *crc32_sum)
405 {
406 	int i, num_blks, err;
407 	unsigned long io_block;
408 	struct buffer_head *obh;
409 
410 	num_blks = count_tags(journal, bh);
411 	/* Calculate checksum of the descriptor block. */
412 	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
413 
414 	for (i = 0; i < num_blks; i++) {
415 		io_block = (*next_log_block)++;
416 		wrap(journal, *next_log_block);
417 		err = jread(&obh, journal, io_block);
418 		if (err) {
419 			printk(KERN_ERR "JBD2: IO error %d recovering block "
420 				"%lu in log\n", err, io_block);
421 			return 1;
422 		} else {
423 			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
424 				     obh->b_size);
425 		}
426 		put_bh(obh);
427 	}
428 	return 0;
429 }
430 
jbd2_commit_block_csum_verify(journal_t * j,void * buf)431 static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
432 {
433 	struct commit_header *h;
434 	__be32 provided;
435 	__u32 calculated;
436 
437 	if (!jbd2_journal_has_csum_v2or3(j))
438 		return 1;
439 
440 	h = buf;
441 	provided = h->h_chksum[0];
442 	h->h_chksum[0] = 0;
443 	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
444 	h->h_chksum[0] = provided;
445 
446 	return provided == cpu_to_be32(calculated);
447 }
448 
jbd2_commit_block_csum_verify_partial(journal_t * j,void * buf)449 static bool jbd2_commit_block_csum_verify_partial(journal_t *j, void *buf)
450 {
451 	struct commit_header *h;
452 	__be32 provided;
453 	__u32 calculated;
454 	void *tmpbuf;
455 
456 	tmpbuf = kzalloc(j->j_blocksize, GFP_KERNEL);
457 	if (!tmpbuf)
458 		return false;
459 
460 	memcpy(tmpbuf, buf, sizeof(struct commit_header));
461 	h = tmpbuf;
462 	provided = h->h_chksum[0];
463 	h->h_chksum[0] = 0;
464 	calculated = jbd2_chksum(j, j->j_csum_seed, tmpbuf, j->j_blocksize);
465 	kfree(tmpbuf);
466 
467 	return provided == cpu_to_be32(calculated);
468 }
469 
jbd2_block_tag_csum_verify(journal_t * j,journal_block_tag_t * tag,journal_block_tag3_t * tag3,void * buf,__u32 sequence)470 static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
471 				      journal_block_tag3_t *tag3,
472 				      void *buf, __u32 sequence)
473 {
474 	__u32 csum32;
475 	__be32 seq;
476 
477 	if (!jbd2_journal_has_csum_v2or3(j))
478 		return 1;
479 
480 	seq = cpu_to_be32(sequence);
481 	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
482 	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
483 
484 	if (jbd2_has_feature_csum3(j))
485 		return tag3->t_checksum == cpu_to_be32(csum32);
486 	else
487 		return tag->t_checksum == cpu_to_be16(csum32);
488 }
489 
jbd2_do_replay(journal_t * journal,struct recovery_info * info,struct buffer_head * bh,unsigned long * next_log_block,unsigned int next_commit_ID)490 static __always_inline int jbd2_do_replay(journal_t *journal,
491 					  struct recovery_info *info,
492 					  struct buffer_head *bh,
493 					  unsigned long *next_log_block,
494 					  unsigned int next_commit_ID)
495 {
496 	char *tagp;
497 	int flags;
498 	int ret = 0;
499 	int tag_bytes = journal_tag_bytes(journal);
500 	int descr_csum_size = 0;
501 	unsigned long io_block;
502 	journal_block_tag_t tag;
503 	struct buffer_head *obh;
504 	struct buffer_head *nbh;
505 
506 	if (jbd2_journal_has_csum_v2or3(journal))
507 		descr_csum_size = sizeof(struct jbd2_journal_block_tail);
508 
509 	tagp = &bh->b_data[sizeof(journal_header_t)];
510 	while (tagp - bh->b_data + tag_bytes <=
511 	       journal->j_blocksize - descr_csum_size) {
512 		int err;
513 
514 		memcpy(&tag, tagp, sizeof(tag));
515 		flags = be16_to_cpu(tag.t_flags);
516 
517 		io_block = (*next_log_block)++;
518 		wrap(journal, *next_log_block);
519 		err = jread(&obh, journal, io_block);
520 		if (err) {
521 			/* Recover what we can, but report failure at the end. */
522 			ret = err;
523 			pr_err("JBD2: IO error %d recovering block %lu in log\n",
524 			      err, io_block);
525 		} else {
526 			unsigned long long blocknr;
527 
528 			J_ASSERT(obh != NULL);
529 			blocknr = read_tag_block(journal, &tag);
530 
531 			/* If the block has been revoked, then we're all done here. */
532 			if (jbd2_journal_test_revoke(journal, blocknr,
533 						     next_commit_ID)) {
534 				brelse(obh);
535 				++info->nr_revoke_hits;
536 				goto skip_write;
537 			}
538 
539 			/* Look for block corruption */
540 			if (!jbd2_block_tag_csum_verify(journal, &tag,
541 					(journal_block_tag3_t *)tagp,
542 					obh->b_data, next_commit_ID)) {
543 				brelse(obh);
544 				ret = -EFSBADCRC;
545 				pr_err("JBD2: Invalid checksum recovering data block %llu in journal block %lu\n",
546 				      blocknr, io_block);
547 				goto skip_write;
548 			}
549 
550 			/* Find a buffer for the new data being restored */
551 			nbh = __getblk(journal->j_fs_dev, blocknr,
552 				       journal->j_blocksize);
553 			if (nbh == NULL) {
554 				pr_err("JBD2: Out of memory during recovery.\n");
555 				brelse(obh);
556 				return -ENOMEM;
557 			}
558 
559 			lock_buffer(nbh);
560 			memcpy(nbh->b_data, obh->b_data, journal->j_blocksize);
561 			if (flags & JBD2_FLAG_ESCAPE) {
562 				*((__be32 *)nbh->b_data) =
563 				cpu_to_be32(JBD2_MAGIC_NUMBER);
564 			}
565 
566 			BUFFER_TRACE(nbh, "marking dirty");
567 			set_buffer_uptodate(nbh);
568 			mark_buffer_dirty(nbh);
569 			BUFFER_TRACE(nbh, "marking uptodate");
570 			++info->nr_replays;
571 			unlock_buffer(nbh);
572 			brelse(obh);
573 			brelse(nbh);
574 		}
575 
576 skip_write:
577 		tagp += tag_bytes;
578 		if (!(flags & JBD2_FLAG_SAME_UUID))
579 			tagp += 16;
580 
581 		if (flags & JBD2_FLAG_LAST_TAG)
582 			break;
583 	}
584 
585 	return ret;
586 }
587 
do_one_pass(journal_t * journal,struct recovery_info * info,enum passtype pass)588 static int do_one_pass(journal_t *journal,
589 			struct recovery_info *info, enum passtype pass)
590 {
591 	unsigned int		first_commit_ID, next_commit_ID;
592 	unsigned long		next_log_block, head_block;
593 	int			err, success = 0;
594 	journal_superblock_t *	sb;
595 	journal_header_t *	tmp;
596 	struct buffer_head	*bh = NULL;
597 	unsigned int		sequence;
598 	int			blocktype;
599 	__u32			crc32_sum = ~0; /* Transactional Checksums */
600 	bool			need_check_commit_time = false;
601 	__u64			last_trans_commit_time = 0, commit_time;
602 
603 	/*
604 	 * First thing is to establish what we expect to find in the log
605 	 * (in terms of transaction IDs), and where (in terms of log
606 	 * block offsets): query the superblock.
607 	 */
608 
609 	sb = journal->j_superblock;
610 	next_commit_ID = be32_to_cpu(sb->s_sequence);
611 	next_log_block = be32_to_cpu(sb->s_start);
612 	head_block = next_log_block;
613 
614 	first_commit_ID = next_commit_ID;
615 	if (pass == PASS_SCAN)
616 		info->start_transaction = first_commit_ID;
617 	else if (pass == PASS_REVOKE) {
618 		/*
619 		 * Would the default revoke table have too long hash chains
620 		 * during replay?
621 		 */
622 		if (info->nr_revokes > JOURNAL_REVOKE_DEFAULT_HASH * 16) {
623 			unsigned int hash_size;
624 
625 			/*
626 			 * Aim for average chain length of 8, limit at 1M
627 			 * entries to avoid problems with malicious
628 			 * filesystems.
629 			 */
630 			hash_size = min(roundup_pow_of_two(info->nr_revokes / 8),
631 					1U << 20);
632 			journal->j_revoke =
633 				jbd2_journal_init_revoke_table(hash_size);
634 			if (!journal->j_revoke) {
635 				printk(KERN_ERR
636 				       "JBD2: failed to allocate revoke table for replay with %u entries. "
637 				       "Journal replay may be slow.\n", hash_size);
638 				journal->j_revoke = journal->j_revoke_table[1];
639 			}
640 		}
641 	}
642 
643 	jbd2_debug(1, "Starting recovery pass %d\n", pass);
644 
645 	/*
646 	 * Now we walk through the log, transaction by transaction,
647 	 * making sure that each transaction has a commit block in the
648 	 * expected place.  Each complete transaction gets replayed back
649 	 * into the main filesystem.
650 	 */
651 
652 	while (1) {
653 		cond_resched();
654 
655 		/* If we already know where to stop the log traversal,
656 		 * check right now that we haven't gone past the end of
657 		 * the log. */
658 
659 		if (pass != PASS_SCAN)
660 			if (tid_geq(next_commit_ID, info->end_transaction))
661 				break;
662 
663 		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
664 			  next_commit_ID, next_log_block, journal->j_last);
665 
666 		/* Skip over each chunk of the transaction looking
667 		 * either the next descriptor block or the final commit
668 		 * record. */
669 
670 		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
671 		brelse(bh);
672 		bh = NULL;
673 		err = jread(&bh, journal, next_log_block);
674 		if (err)
675 			goto failed;
676 
677 		next_log_block++;
678 		wrap(journal, next_log_block);
679 
680 		/* What kind of buffer is it?
681 		 *
682 		 * If it is a descriptor block, check that it has the
683 		 * expected sequence number.  Otherwise, we're all done
684 		 * here. */
685 
686 		tmp = (journal_header_t *)bh->b_data;
687 
688 		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER))
689 			break;
690 
691 		blocktype = be32_to_cpu(tmp->h_blocktype);
692 		sequence = be32_to_cpu(tmp->h_sequence);
693 		jbd2_debug(3, "Found magic %d, sequence %d\n",
694 			  blocktype, sequence);
695 
696 		if (sequence != next_commit_ID)
697 			break;
698 
699 		/* OK, we have a valid descriptor block which matches
700 		 * all of the sequence number checks.  What are we going
701 		 * to do with it?  That depends on the pass... */
702 
703 		switch(blocktype) {
704 		case JBD2_DESCRIPTOR_BLOCK:
705 			/* Verify checksum first */
706 			if (!jbd2_descriptor_block_csum_verify(journal,
707 							       bh->b_data)) {
708 				/*
709 				 * PASS_SCAN can see stale blocks due to lazy
710 				 * journal init. Don't error out on those yet.
711 				 */
712 				if (pass != PASS_SCAN) {
713 					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
714 					       next_log_block);
715 					err = -EFSBADCRC;
716 					goto failed;
717 				}
718 				need_check_commit_time = true;
719 				jbd2_debug(1,
720 					"invalid descriptor block found in %lu\n",
721 					next_log_block);
722 			}
723 
724 			/* If it is a valid descriptor block, replay it
725 			 * in pass REPLAY; if journal_checksums enabled, then
726 			 * calculate checksums in PASS_SCAN, otherwise,
727 			 * just skip over the blocks it describes. */
728 			if (pass != PASS_REPLAY) {
729 				if (pass == PASS_SCAN &&
730 				    jbd2_has_feature_checksum(journal) &&
731 				    !info->end_transaction) {
732 					if (calc_chksums(journal, bh,
733 							&next_log_block,
734 							&crc32_sum))
735 						break;
736 					continue;
737 				}
738 				next_log_block += count_tags(journal, bh);
739 				wrap(journal, next_log_block);
740 				continue;
741 			}
742 
743 			/*
744 			 * A descriptor block: we can now write all of the
745 			 * data blocks. Yay, useful work is finally getting
746 			 * done here!
747 			 */
748 			err = jbd2_do_replay(journal, info, bh, &next_log_block,
749 					     next_commit_ID);
750 			if (err) {
751 				if (err == -ENOMEM)
752 					goto failed;
753 				success = err;
754 			}
755 
756 			continue;
757 
758 		case JBD2_COMMIT_BLOCK:
759 			if (pass != PASS_SCAN) {
760 				next_commit_ID++;
761 				continue;
762 			}
763 
764 			/*     How to differentiate between interrupted commit
765 			 *               and journal corruption ?
766 			 *
767 			 * {nth transaction}
768 			 *        Checksum Verification Failed
769 			 *			 |
770 			 *		 ____________________
771 			 *		|		     |
772 			 * 	async_commit             sync_commit
773 			 *     		|                    |
774 			 *		| GO TO NEXT    "Journal Corruption"
775 			 *		| TRANSACTION
776 			 *		|
777 			 * {(n+1)th transanction}
778 			 *		|
779 			 * 	 _______|______________
780 			 * 	|	 	      |
781 			 * Commit block found	Commit block not found
782 			 *      |		      |
783 			 * "Journal Corruption"       |
784 			 *		 _____________|_________
785 			 *     		|	           	|
786 			 *	nth trans corrupt	OR   nth trans
787 			 *	and (n+1)th interrupted     interrupted
788 			 *	before commit block
789 			 *      could reach the disk.
790 			 *	(Cannot find the difference in above
791 			 *	 mentioned conditions. Hence assume
792 			 *	 "Interrupted Commit".)
793 			 */
794 			commit_time = be64_to_cpu(
795 				((struct commit_header *)bh->b_data)->h_commit_sec);
796 			/*
797 			 * If need_check_commit_time is set, it means we are in
798 			 * PASS_SCAN and csum verify failed before. If
799 			 * commit_time is increasing, it's the same journal,
800 			 * otherwise it is stale journal block, just end this
801 			 * recovery.
802 			 */
803 			if (need_check_commit_time) {
804 				if (commit_time >= last_trans_commit_time) {
805 					pr_err("JBD2: Invalid checksum found in transaction %u\n",
806 					       next_commit_ID);
807 					err = -EFSBADCRC;
808 					goto failed;
809 				}
810 			ignore_crc_mismatch:
811 				/*
812 				 * It likely does not belong to same journal,
813 				 * just end this recovery with success.
814 				 */
815 				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
816 					  next_commit_ID);
817 				goto done;
818 			}
819 
820 			/*
821 			 * Found an expected commit block: if checksums
822 			 * are present, verify them in PASS_SCAN; else not
823 			 * much to do other than move on to the next sequence
824 			 * number.
825 			 */
826 			if (jbd2_has_feature_checksum(journal)) {
827 				struct commit_header *cbh =
828 					(struct commit_header *)bh->b_data;
829 				unsigned found_chksum =
830 					be32_to_cpu(cbh->h_chksum[0]);
831 
832 				if (info->end_transaction) {
833 					journal->j_failed_commit =
834 						info->end_transaction;
835 					break;
836 				}
837 
838 				/* Neither checksum match nor unused? */
839 				if (!((crc32_sum == found_chksum &&
840 				       cbh->h_chksum_type ==
841 						JBD2_CRC32_CHKSUM &&
842 				       cbh->h_chksum_size ==
843 						JBD2_CRC32_CHKSUM_SIZE) ||
844 				      (cbh->h_chksum_type == 0 &&
845 				       cbh->h_chksum_size == 0 &&
846 				       found_chksum == 0)))
847 					goto chksum_error;
848 
849 				crc32_sum = ~0;
850 				goto chksum_ok;
851 			}
852 
853 			if (jbd2_commit_block_csum_verify(journal, bh->b_data))
854 				goto chksum_ok;
855 
856 			if (jbd2_commit_block_csum_verify_partial(journal,
857 								  bh->b_data)) {
858 				pr_notice("JBD2: Find incomplete commit block in transaction %u block %lu\n",
859 					  next_commit_ID, next_log_block);
860 				goto chksum_ok;
861 			}
862 
863 chksum_error:
864 			if (commit_time < last_trans_commit_time)
865 				goto ignore_crc_mismatch;
866 			info->end_transaction = next_commit_ID;
867 			info->head_block = head_block;
868 
869 			if (!jbd2_has_feature_async_commit(journal)) {
870 				journal->j_failed_commit = next_commit_ID;
871 				break;
872 			}
873 
874 chksum_ok:
875 			last_trans_commit_time = commit_time;
876 			head_block = next_log_block;
877 			next_commit_ID++;
878 			continue;
879 
880 		case JBD2_REVOKE_BLOCK:
881 			/*
882 			 * If we aren't in the SCAN or REVOKE pass, then we can
883 			 * just skip over this block.
884 			 */
885 			if (pass != PASS_REVOKE && pass != PASS_SCAN)
886 				continue;
887 
888 			/*
889 			 * Check revoke block crc in pass_scan, if csum verify
890 			 * failed, check commit block time later.
891 			 */
892 			if (pass == PASS_SCAN &&
893 			    !jbd2_descriptor_block_csum_verify(journal,
894 							       bh->b_data)) {
895 				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
896 					  next_log_block);
897 				need_check_commit_time = true;
898 			}
899 
900 			err = scan_revoke_records(journal, pass, bh,
901 						  next_commit_ID, info);
902 			if (err)
903 				goto failed;
904 			continue;
905 
906 		default:
907 			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
908 				  blocktype);
909 			goto done;
910 		}
911 	}
912 
913  done:
914 	brelse(bh);
915 	/*
916 	 * We broke out of the log scan loop: either we came to the
917 	 * known end of the log or we found an unexpected block in the
918 	 * log.  If the latter happened, then we know that the "current"
919 	 * transaction marks the end of the valid log.
920 	 */
921 
922 	if (pass == PASS_SCAN) {
923 		if (!info->end_transaction)
924 			info->end_transaction = next_commit_ID;
925 		if (!info->head_block)
926 			info->head_block = head_block;
927 	} else {
928 		/* It's really bad news if different passes end up at
929 		 * different places (but possible due to IO errors). */
930 		if (info->end_transaction != next_commit_ID) {
931 			printk(KERN_ERR "JBD2: recovery pass %d ended at "
932 				"transaction %u, expected %u\n",
933 				pass, next_commit_ID, info->end_transaction);
934 			if (!success)
935 				success = -EIO;
936 		}
937 	}
938 
939 	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
940 		err = fc_do_one_pass(journal, info, pass);
941 		if (err)
942 			success = err;
943 	}
944 
945 	return success;
946 
947  failed:
948 	brelse(bh);
949 	return err;
950 }
951 
952 /* Scan a revoke record, marking all blocks mentioned as revoked. */
953 
scan_revoke_records(journal_t * journal,enum passtype pass,struct buffer_head * bh,tid_t sequence,struct recovery_info * info)954 static int scan_revoke_records(journal_t *journal, enum passtype pass,
955 			       struct buffer_head *bh, tid_t sequence,
956 			       struct recovery_info *info)
957 {
958 	jbd2_journal_revoke_header_t *header;
959 	int offset, max;
960 	unsigned csum_size = 0;
961 	__u32 rcount;
962 	int record_len = 4;
963 
964 	header = (jbd2_journal_revoke_header_t *) bh->b_data;
965 	offset = sizeof(jbd2_journal_revoke_header_t);
966 	rcount = be32_to_cpu(header->r_count);
967 
968 	if (jbd2_journal_has_csum_v2or3(journal))
969 		csum_size = sizeof(struct jbd2_journal_block_tail);
970 	if (rcount > journal->j_blocksize - csum_size)
971 		return -EINVAL;
972 	max = rcount;
973 
974 	if (jbd2_has_feature_64bit(journal))
975 		record_len = 8;
976 
977 	if (pass == PASS_SCAN) {
978 		info->nr_revokes += (max - offset) / record_len;
979 		return 0;
980 	}
981 
982 	while (offset + record_len <= max) {
983 		unsigned long long blocknr;
984 		int err;
985 
986 		if (record_len == 4)
987 			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
988 		else
989 			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
990 		offset += record_len;
991 		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
992 		if (err)
993 			return err;
994 	}
995 	return 0;
996 }
997