Linux Audio

Check our new training course

Linux debugging, profiling, tracing and performance analysis training

Mar 24-27, 2025, special US time zones
Register
Loading...
v6.2
  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 */
 28struct recovery_info
 29{
 30	tid_t		start_transaction;
 31	tid_t		end_transaction;
 32
 33	int		nr_replays;
 34	int		nr_revokes;
 35	int		nr_revoke_hits;
 36};
 37
 
 38static int do_one_pass(journal_t *journal,
 39				struct recovery_info *info, enum passtype pass);
 40static int scan_revoke_records(journal_t *, struct buffer_head *,
 41				tid_t, struct recovery_info *);
 42
 43#ifdef __KERNEL__
 44
 45/* Release readahead buffers after use */
 46static void journal_brelse_array(struct buffer_head *b[], int n)
 47{
 48	while (--n >= 0)
 49		brelse (b[n]);
 50}
 51
 52
 53/*
 54 * When reading from the journal, we are going through the block device
 55 * layer directly and so there is no readahead being done for us.  We
 56 * need to implement any readahead ourselves if we want it to happen at
 57 * all.  Recovery is basically one long sequential read, so make sure we
 58 * do the IO in reasonably large chunks.
 59 *
 60 * This is not so critical that we need to be enormously clever about
 61 * the readahead size, though.  128K is a purely arbitrary, good-enough
 62 * fixed value.
 63 */
 64
 65#define MAXBUF 8
 66static int do_readahead(journal_t *journal, unsigned int start)
 67{
 68	int err;
 69	unsigned int max, nbufs, next;
 70	unsigned long long blocknr;
 71	struct buffer_head *bh;
 72
 73	struct buffer_head * bufs[MAXBUF];
 74
 75	/* Do up to 128K of readahead */
 76	max = start + (128 * 1024 / journal->j_blocksize);
 77	if (max > journal->j_total_len)
 78		max = journal->j_total_len;
 79
 80	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
 81	 * a time to the block device IO layer. */
 82
 83	nbufs = 0;
 84
 85	for (next = start; next < max; next++) {
 86		err = jbd2_journal_bmap(journal, next, &blocknr);
 87
 88		if (err) {
 89			printk(KERN_ERR "JBD2: bad block at offset %u\n",
 90				next);
 91			goto failed;
 92		}
 93
 94		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 95		if (!bh) {
 96			err = -ENOMEM;
 97			goto failed;
 98		}
 99
100		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
101			bufs[nbufs++] = bh;
102			if (nbufs == MAXBUF) {
103				bh_readahead_batch(nbufs, bufs, 0);
104				journal_brelse_array(bufs, nbufs);
105				nbufs = 0;
106			}
107		} else
108			brelse(bh);
109	}
110
111	if (nbufs)
112		bh_readahead_batch(nbufs, bufs, 0);
113	err = 0;
114
115failed:
116	if (nbufs)
117		journal_brelse_array(bufs, nbufs);
118	return err;
119}
120
121#endif /* __KERNEL__ */
122
123
124/*
125 * Read a block from the journal
126 */
127
128static int jread(struct buffer_head **bhp, journal_t *journal,
129		 unsigned int offset)
130{
131	int err;
132	unsigned long long blocknr;
133	struct buffer_head *bh;
134
135	*bhp = NULL;
136
137	if (offset >= journal->j_total_len) {
138		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
139		return -EFSCORRUPTED;
140	}
141
142	err = jbd2_journal_bmap(journal, offset, &blocknr);
143
144	if (err) {
145		printk(KERN_ERR "JBD2: bad block at offset %u\n",
146			offset);
147		return err;
148	}
149
150	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
151	if (!bh)
152		return -ENOMEM;
153
154	if (!buffer_uptodate(bh)) {
155		/*
156		 * If this is a brand new buffer, start readahead.
157		 * Otherwise, we assume we are already reading it.
158		 */
159		bool need_readahead = !buffer_req(bh);
160
161		bh_read_nowait(bh, 0);
162		if (need_readahead)
163			do_readahead(journal, offset);
164		wait_on_buffer(bh);
165	}
166
167	if (!buffer_uptodate(bh)) {
168		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
169			offset);
170		brelse(bh);
171		return -EIO;
172	}
173
174	*bhp = bh;
175	return 0;
176}
177
178static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
179{
180	struct jbd2_journal_block_tail *tail;
181	__be32 provided;
182	__u32 calculated;
183
184	if (!jbd2_journal_has_csum_v2or3(j))
185		return 1;
186
187	tail = (struct jbd2_journal_block_tail *)((char *)buf +
188		j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
189	provided = tail->t_checksum;
190	tail->t_checksum = 0;
191	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
192	tail->t_checksum = provided;
193
194	return provided == cpu_to_be32(calculated);
195}
196
197/*
198 * Count the number of in-use tags in a journal descriptor block.
199 */
200
201static int count_tags(journal_t *journal, struct buffer_head *bh)
202{
203	char *			tagp;
204	journal_block_tag_t	tag;
205	int			nr = 0, size = journal->j_blocksize;
206	int			tag_bytes = journal_tag_bytes(journal);
207
208	if (jbd2_journal_has_csum_v2or3(journal))
209		size -= sizeof(struct jbd2_journal_block_tail);
210
211	tagp = &bh->b_data[sizeof(journal_header_t)];
212
213	while ((tagp - bh->b_data + tag_bytes) <= size) {
214		memcpy(&tag, tagp, sizeof(tag));
215
216		nr++;
217		tagp += tag_bytes;
218		if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
219			tagp += 16;
220
221		if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
222			break;
223	}
224
225	return nr;
226}
227
228
229/* Make sure we wrap around the log correctly! */
230#define wrap(journal, var)						\
231do {									\
232	unsigned long _wrap_last =					\
233		jbd2_has_feature_fast_commit(journal) ?			\
234			(journal)->j_fc_last : (journal)->j_last;	\
235									\
236	if (var >= _wrap_last)						\
237		var -= (_wrap_last - (journal)->j_first);		\
238} while (0)
239
240static int fc_do_one_pass(journal_t *journal,
241			  struct recovery_info *info, enum passtype pass)
242{
243	unsigned int expected_commit_id = info->end_transaction;
244	unsigned long next_fc_block;
245	struct buffer_head *bh;
246	int err = 0;
247
248	next_fc_block = journal->j_fc_first;
249	if (!journal->j_fc_replay_callback)
250		return 0;
251
252	while (next_fc_block <= journal->j_fc_last) {
253		jbd2_debug(3, "Fast commit replay: next block %ld\n",
254			  next_fc_block);
255		err = jread(&bh, journal, next_fc_block);
256		if (err) {
257			jbd2_debug(3, "Fast commit replay: read error\n");
258			break;
259		}
260
261		err = journal->j_fc_replay_callback(journal, bh, pass,
262					next_fc_block - journal->j_fc_first,
263					expected_commit_id);
264		brelse(bh);
265		next_fc_block++;
266		if (err < 0 || err == JBD2_FC_REPLAY_STOP)
267			break;
268		err = 0;
269	}
270
271	if (err)
272		jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
273
274	return err;
275}
276
277/**
278 * jbd2_journal_recover - recovers a on-disk journal
279 * @journal: the journal to recover
280 *
281 * The primary function for recovering the log contents when mounting a
282 * journaled device.
283 *
284 * Recovery is done in three passes.  In the first pass, we look for the
285 * end of the log.  In the second, we assemble the list of revoke
286 * blocks.  In the third and final pass, we replay any un-revoked blocks
287 * in the log.
288 */
289int jbd2_journal_recover(journal_t *journal)
290{
291	int			err, err2;
292	journal_superblock_t *	sb;
293
294	struct recovery_info	info;
295
296	memset(&info, 0, sizeof(info));
297	sb = journal->j_superblock;
298
299	/*
300	 * The journal superblock's s_start field (the current log head)
301	 * is always zero if, and only if, the journal was cleanly
302	 * unmounted.
303	 */
304
305	if (!sb->s_start) {
306		jbd2_debug(1, "No recovery required, last transaction %d\n",
307			  be32_to_cpu(sb->s_sequence));
308		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
309		return 0;
310	}
311
312	err = do_one_pass(journal, &info, PASS_SCAN);
313	if (!err)
314		err = do_one_pass(journal, &info, PASS_REVOKE);
315	if (!err)
316		err = do_one_pass(journal, &info, PASS_REPLAY);
317
318	jbd2_debug(1, "JBD2: recovery, exit status %d, "
319		  "recovered transactions %u to %u\n",
320		  err, info.start_transaction, info.end_transaction);
321	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
322		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
323
324	/* Restart the log at the next transaction ID, thus invalidating
325	 * any existing commit records in the log. */
326	journal->j_transaction_sequence = ++info.end_transaction;
327
328	jbd2_journal_clear_revoke(journal);
329	err2 = sync_blockdev(journal->j_fs_dev);
330	if (!err)
331		err = err2;
332	/* Make sure all replayed data is on permanent storage */
333	if (journal->j_flags & JBD2_BARRIER) {
334		err2 = blkdev_issue_flush(journal->j_fs_dev);
335		if (!err)
336			err = err2;
337	}
338	return err;
339}
340
341/**
342 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
343 * @journal: journal to startup
344 *
345 * Locate any valid recovery information from the journal and set up the
346 * journal structures in memory to ignore it (presumably because the
347 * caller has evidence that it is out of date).
348 * This function doesn't appear to be exported..
349 *
350 * We perform one pass over the journal to allow us to tell the user how
351 * much recovery information is being erased, and to let us initialise
352 * the journal transaction sequence numbers to the next unused ID.
353 */
354int jbd2_journal_skip_recovery(journal_t *journal)
355{
356	int			err;
357
358	struct recovery_info	info;
359
360	memset (&info, 0, sizeof(info));
361
362	err = do_one_pass(journal, &info, PASS_SCAN);
363
364	if (err) {
365		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
366		++journal->j_transaction_sequence;
367	} else {
368#ifdef CONFIG_JBD2_DEBUG
369		int dropped = info.end_transaction - 
370			be32_to_cpu(journal->j_superblock->s_sequence);
371		jbd2_debug(1,
372			  "JBD2: ignoring %d transaction%s from the journal.\n",
373			  dropped, (dropped == 1) ? "" : "s");
374#endif
375		journal->j_transaction_sequence = ++info.end_transaction;
376	}
377
378	journal->j_tail = 0;
379	return err;
380}
381
382static inline unsigned long long read_tag_block(journal_t *journal,
383						journal_block_tag_t *tag)
384{
385	unsigned long long block = be32_to_cpu(tag->t_blocknr);
386	if (jbd2_has_feature_64bit(journal))
387		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
388	return block;
389}
390
391/*
392 * calc_chksums calculates the checksums for the blocks described in the
393 * descriptor block.
394 */
395static int calc_chksums(journal_t *journal, struct buffer_head *bh,
396			unsigned long *next_log_block, __u32 *crc32_sum)
397{
398	int i, num_blks, err;
399	unsigned long io_block;
400	struct buffer_head *obh;
401
402	num_blks = count_tags(journal, bh);
403	/* Calculate checksum of the descriptor block. */
404	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
405
406	for (i = 0; i < num_blks; i++) {
407		io_block = (*next_log_block)++;
408		wrap(journal, *next_log_block);
409		err = jread(&obh, journal, io_block);
410		if (err) {
411			printk(KERN_ERR "JBD2: IO error %d recovering block "
412				"%lu in log\n", err, io_block);
413			return 1;
414		} else {
415			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
416				     obh->b_size);
417		}
418		put_bh(obh);
419	}
420	return 0;
421}
422
423static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
424{
425	struct commit_header *h;
426	__be32 provided;
427	__u32 calculated;
428
429	if (!jbd2_journal_has_csum_v2or3(j))
430		return 1;
431
432	h = buf;
433	provided = h->h_chksum[0];
434	h->h_chksum[0] = 0;
435	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
436	h->h_chksum[0] = provided;
437
438	return provided == cpu_to_be32(calculated);
439}
440
441static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
442				      journal_block_tag3_t *tag3,
443				      void *buf, __u32 sequence)
444{
 
445	__u32 csum32;
446	__be32 seq;
447
448	if (!jbd2_journal_has_csum_v2or3(j))
449		return 1;
450
451	seq = cpu_to_be32(sequence);
452	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
453	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
454
455	if (jbd2_has_feature_csum3(j))
456		return tag3->t_checksum == cpu_to_be32(csum32);
457	else
458		return tag->t_checksum == cpu_to_be16(csum32);
459}
460
461static int do_one_pass(journal_t *journal,
462			struct recovery_info *info, enum passtype pass)
463{
464	unsigned int		first_commit_ID, next_commit_ID;
465	unsigned long		next_log_block;
466	int			err, success = 0;
467	journal_superblock_t *	sb;
468	journal_header_t *	tmp;
469	struct buffer_head *	bh;
470	unsigned int		sequence;
471	int			blocktype;
472	int			tag_bytes = journal_tag_bytes(journal);
473	__u32			crc32_sum = ~0; /* Transactional Checksums */
474	int			descr_csum_size = 0;
475	int			block_error = 0;
476	bool			need_check_commit_time = false;
477	__u64			last_trans_commit_time = 0, commit_time;
478
479	/*
480	 * First thing is to establish what we expect to find in the log
481	 * (in terms of transaction IDs), and where (in terms of log
482	 * block offsets): query the superblock.
483	 */
484
485	sb = journal->j_superblock;
486	next_commit_ID = be32_to_cpu(sb->s_sequence);
487	next_log_block = be32_to_cpu(sb->s_start);
488
489	first_commit_ID = next_commit_ID;
490	if (pass == PASS_SCAN)
491		info->start_transaction = first_commit_ID;
492
493	jbd2_debug(1, "Starting recovery pass %d\n", pass);
494
495	/*
496	 * Now we walk through the log, transaction by transaction,
497	 * making sure that each transaction has a commit block in the
498	 * expected place.  Each complete transaction gets replayed back
499	 * into the main filesystem.
500	 */
501
502	while (1) {
503		int			flags;
504		char *			tagp;
505		journal_block_tag_t	tag;
506		struct buffer_head *	obh;
507		struct buffer_head *	nbh;
508
509		cond_resched();
510
511		/* If we already know where to stop the log traversal,
512		 * check right now that we haven't gone past the end of
513		 * the log. */
514
515		if (pass != PASS_SCAN)
516			if (tid_geq(next_commit_ID, info->end_transaction))
517				break;
518
519		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
520			  next_commit_ID, next_log_block,
521			  jbd2_has_feature_fast_commit(journal) ?
522			  journal->j_fc_last : journal->j_last);
523
524		/* Skip over each chunk of the transaction looking
525		 * either the next descriptor block or the final commit
526		 * record. */
527
528		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
529		err = jread(&bh, journal, next_log_block);
530		if (err)
531			goto failed;
532
533		next_log_block++;
534		wrap(journal, next_log_block);
535
536		/* What kind of buffer is it?
537		 *
538		 * If it is a descriptor block, check that it has the
539		 * expected sequence number.  Otherwise, we're all done
540		 * here. */
541
542		tmp = (journal_header_t *)bh->b_data;
543
544		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
545			brelse(bh);
546			break;
547		}
548
549		blocktype = be32_to_cpu(tmp->h_blocktype);
550		sequence = be32_to_cpu(tmp->h_sequence);
551		jbd2_debug(3, "Found magic %d, sequence %d\n",
552			  blocktype, sequence);
553
554		if (sequence != next_commit_ID) {
555			brelse(bh);
556			break;
557		}
558
559		/* OK, we have a valid descriptor block which matches
560		 * all of the sequence number checks.  What are we going
561		 * to do with it?  That depends on the pass... */
562
563		switch(blocktype) {
564		case JBD2_DESCRIPTOR_BLOCK:
565			/* Verify checksum first */
566			if (jbd2_journal_has_csum_v2or3(journal))
567				descr_csum_size =
568					sizeof(struct jbd2_journal_block_tail);
569			if (descr_csum_size > 0 &&
570			    !jbd2_descriptor_block_csum_verify(journal,
571							       bh->b_data)) {
572				/*
573				 * PASS_SCAN can see stale blocks due to lazy
574				 * journal init. Don't error out on those yet.
575				 */
576				if (pass != PASS_SCAN) {
577					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
578					       next_log_block);
579					err = -EFSBADCRC;
580					brelse(bh);
581					goto failed;
582				}
583				need_check_commit_time = true;
584				jbd2_debug(1,
585					"invalid descriptor block found in %lu\n",
586					next_log_block);
587			}
588
589			/* If it is a valid descriptor block, replay it
590			 * in pass REPLAY; if journal_checksums enabled, then
591			 * calculate checksums in PASS_SCAN, otherwise,
592			 * just skip over the blocks it describes. */
593			if (pass != PASS_REPLAY) {
594				if (pass == PASS_SCAN &&
595				    jbd2_has_feature_checksum(journal) &&
596				    !need_check_commit_time &&
597				    !info->end_transaction) {
598					if (calc_chksums(journal, bh,
599							&next_log_block,
600							&crc32_sum)) {
601						put_bh(bh);
602						break;
603					}
604					put_bh(bh);
605					continue;
606				}
607				next_log_block += count_tags(journal, bh);
608				wrap(journal, next_log_block);
609				put_bh(bh);
610				continue;
611			}
612
613			/* A descriptor block: we can now write all of
614			 * the data blocks.  Yay, useful work is finally
615			 * getting done here! */
616
617			tagp = &bh->b_data[sizeof(journal_header_t)];
618			while ((tagp - bh->b_data + tag_bytes)
619			       <= journal->j_blocksize - descr_csum_size) {
620				unsigned long io_block;
621
622				memcpy(&tag, tagp, sizeof(tag));
623				flags = be16_to_cpu(tag.t_flags);
624
625				io_block = next_log_block++;
626				wrap(journal, next_log_block);
627				err = jread(&obh, journal, io_block);
628				if (err) {
629					/* Recover what we can, but
630					 * report failure at the end. */
631					success = err;
632					printk(KERN_ERR
633						"JBD2: IO error %d recovering "
634						"block %ld in log\n",
635						err, io_block);
636				} else {
637					unsigned long long blocknr;
638
639					J_ASSERT(obh != NULL);
640					blocknr = read_tag_block(journal,
641								 &tag);
642
643					/* If the block has been
644					 * revoked, then we're all done
645					 * here. */
646					if (jbd2_journal_test_revoke
647					    (journal, blocknr,
648					     next_commit_ID)) {
649						brelse(obh);
650						++info->nr_revoke_hits;
651						goto skip_write;
652					}
653
654					/* Look for block corruption */
655					if (!jbd2_block_tag_csum_verify(
656			journal, &tag, (journal_block_tag3_t *)tagp,
657			obh->b_data, be32_to_cpu(tmp->h_sequence))) {
658						brelse(obh);
659						success = -EFSBADCRC;
660						printk(KERN_ERR "JBD2: Invalid "
661						       "checksum recovering "
662						       "data block %llu in "
663						       "log\n", blocknr);
664						block_error = 1;
665						goto skip_write;
666					}
667
668					/* Find a buffer for the new
669					 * data being restored */
670					nbh = __getblk(journal->j_fs_dev,
671							blocknr,
672							journal->j_blocksize);
673					if (nbh == NULL) {
674						printk(KERN_ERR
675						       "JBD2: Out of memory "
676						       "during recovery.\n");
677						err = -ENOMEM;
678						brelse(bh);
679						brelse(obh);
680						goto failed;
681					}
682
683					lock_buffer(nbh);
684					memcpy(nbh->b_data, obh->b_data,
685							journal->j_blocksize);
686					if (flags & JBD2_FLAG_ESCAPE) {
687						*((__be32 *)nbh->b_data) =
688						cpu_to_be32(JBD2_MAGIC_NUMBER);
689					}
690
691					BUFFER_TRACE(nbh, "marking dirty");
692					set_buffer_uptodate(nbh);
693					mark_buffer_dirty(nbh);
694					BUFFER_TRACE(nbh, "marking uptodate");
695					++info->nr_replays;
 
696					unlock_buffer(nbh);
697					brelse(obh);
698					brelse(nbh);
699				}
700
701			skip_write:
702				tagp += tag_bytes;
703				if (!(flags & JBD2_FLAG_SAME_UUID))
704					tagp += 16;
705
706				if (flags & JBD2_FLAG_LAST_TAG)
707					break;
708			}
709
710			brelse(bh);
711			continue;
712
713		case JBD2_COMMIT_BLOCK:
714			/*     How to differentiate between interrupted commit
715			 *               and journal corruption ?
716			 *
717			 * {nth transaction}
718			 *        Checksum Verification Failed
719			 *			 |
720			 *		 ____________________
721			 *		|		     |
722			 * 	async_commit             sync_commit
723			 *     		|                    |
724			 *		| GO TO NEXT    "Journal Corruption"
725			 *		| TRANSACTION
726			 *		|
727			 * {(n+1)th transanction}
728			 *		|
729			 * 	 _______|______________
730			 * 	|	 	      |
731			 * Commit block found	Commit block not found
732			 *      |		      |
733			 * "Journal Corruption"       |
734			 *		 _____________|_________
735			 *     		|	           	|
736			 *	nth trans corrupt	OR   nth trans
737			 *	and (n+1)th interrupted     interrupted
738			 *	before commit block
739			 *      could reach the disk.
740			 *	(Cannot find the difference in above
741			 *	 mentioned conditions. Hence assume
742			 *	 "Interrupted Commit".)
743			 */
744			commit_time = be64_to_cpu(
745				((struct commit_header *)bh->b_data)->h_commit_sec);
746			/*
747			 * If need_check_commit_time is set, it means we are in
748			 * PASS_SCAN and csum verify failed before. If
749			 * commit_time is increasing, it's the same journal,
750			 * otherwise it is stale journal block, just end this
751			 * recovery.
752			 */
753			if (need_check_commit_time) {
754				if (commit_time >= last_trans_commit_time) {
755					pr_err("JBD2: Invalid checksum found in transaction %u\n",
756					       next_commit_ID);
757					err = -EFSBADCRC;
758					brelse(bh);
759					goto failed;
760				}
761			ignore_crc_mismatch:
762				/*
763				 * It likely does not belong to same journal,
764				 * just end this recovery with success.
765				 */
766				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
767					  next_commit_ID);
768				brelse(bh);
769				goto done;
770			}
771
772			/*
773			 * Found an expected commit block: if checksums
774			 * are present, verify them in PASS_SCAN; else not
775			 * much to do other than move on to the next sequence
776			 * number.
777			 */
778			if (pass == PASS_SCAN &&
779			    jbd2_has_feature_checksum(journal)) {
780				struct commit_header *cbh =
781					(struct commit_header *)bh->b_data;
782				unsigned found_chksum =
783					be32_to_cpu(cbh->h_chksum[0]);
784
785				if (info->end_transaction) {
786					journal->j_failed_commit =
787						info->end_transaction;
788					brelse(bh);
789					break;
790				}
791
792				/* Neither checksum match nor unused? */
793				if (!((crc32_sum == found_chksum &&
794				       cbh->h_chksum_type ==
795						JBD2_CRC32_CHKSUM &&
796				       cbh->h_chksum_size ==
797						JBD2_CRC32_CHKSUM_SIZE) ||
798				      (cbh->h_chksum_type == 0 &&
799				       cbh->h_chksum_size == 0 &&
800				       found_chksum == 0)))
801					goto chksum_error;
802
803				crc32_sum = ~0;
804			}
805			if (pass == PASS_SCAN &&
806			    !jbd2_commit_block_csum_verify(journal,
807							   bh->b_data)) {
808			chksum_error:
809				if (commit_time < last_trans_commit_time)
810					goto ignore_crc_mismatch;
811				info->end_transaction = next_commit_ID;
812
813				if (!jbd2_has_feature_async_commit(journal)) {
814					journal->j_failed_commit =
815						next_commit_ID;
816					brelse(bh);
817					break;
818				}
819			}
820			if (pass == PASS_SCAN)
821				last_trans_commit_time = commit_time;
822			brelse(bh);
823			next_commit_ID++;
824			continue;
825
826		case JBD2_REVOKE_BLOCK:
827			/*
828			 * Check revoke block crc in pass_scan, if csum verify
829			 * failed, check commit block time later.
830			 */
831			if (pass == PASS_SCAN &&
832			    !jbd2_descriptor_block_csum_verify(journal,
833							       bh->b_data)) {
834				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
835					  next_log_block);
836				need_check_commit_time = true;
837			}
838			/* If we aren't in the REVOKE pass, then we can
839			 * just skip over this block. */
840			if (pass != PASS_REVOKE) {
841				brelse(bh);
842				continue;
843			}
844
845			err = scan_revoke_records(journal, bh,
846						  next_commit_ID, info);
847			brelse(bh);
848			if (err)
849				goto failed;
850			continue;
851
852		default:
853			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
854				  blocktype);
855			brelse(bh);
856			goto done;
857		}
858	}
859
860 done:
861	/*
862	 * We broke out of the log scan loop: either we came to the
863	 * known end of the log or we found an unexpected block in the
864	 * log.  If the latter happened, then we know that the "current"
865	 * transaction marks the end of the valid log.
866	 */
867
868	if (pass == PASS_SCAN) {
869		if (!info->end_transaction)
870			info->end_transaction = next_commit_ID;
871	} else {
872		/* It's really bad news if different passes end up at
873		 * different places (but possible due to IO errors). */
874		if (info->end_transaction != next_commit_ID) {
875			printk(KERN_ERR "JBD2: recovery pass %d ended at "
876				"transaction %u, expected %u\n",
877				pass, next_commit_ID, info->end_transaction);
878			if (!success)
879				success = -EIO;
880		}
881	}
882
883	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
884		err = fc_do_one_pass(journal, info, pass);
885		if (err)
886			success = err;
887	}
888
889	if (block_error && success == 0)
890		success = -EIO;
891	return success;
892
893 failed:
894	return err;
895}
896
897/* Scan a revoke record, marking all blocks mentioned as revoked. */
898
899static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
900			       tid_t sequence, struct recovery_info *info)
901{
902	jbd2_journal_revoke_header_t *header;
903	int offset, max;
904	unsigned csum_size = 0;
905	__u32 rcount;
906	int record_len = 4;
907
908	header = (jbd2_journal_revoke_header_t *) bh->b_data;
909	offset = sizeof(jbd2_journal_revoke_header_t);
910	rcount = be32_to_cpu(header->r_count);
 
 
 
911
912	if (jbd2_journal_has_csum_v2or3(journal))
913		csum_size = sizeof(struct jbd2_journal_block_tail);
914	if (rcount > journal->j_blocksize - csum_size)
915		return -EINVAL;
916	max = rcount;
917
918	if (jbd2_has_feature_64bit(journal))
919		record_len = 8;
920
921	while (offset + record_len <= max) {
922		unsigned long long blocknr;
923		int err;
924
925		if (record_len == 4)
926			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
927		else
928			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
929		offset += record_len;
930		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
931		if (err)
932			return err;
933		++info->nr_revokes;
934	}
935	return 0;
936}
v5.9
  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 */
 28struct recovery_info
 29{
 30	tid_t		start_transaction;
 31	tid_t		end_transaction;
 32
 33	int		nr_replays;
 34	int		nr_revokes;
 35	int		nr_revoke_hits;
 36};
 37
 38enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
 39static int do_one_pass(journal_t *journal,
 40				struct recovery_info *info, enum passtype pass);
 41static 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 */
 47static 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
 67static 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_maxlen)
 79		max = journal->j_maxlen;
 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				ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
105				journal_brelse_array(bufs, nbufs);
106				nbufs = 0;
107			}
108		} else
109			brelse(bh);
110	}
111
112	if (nbufs)
113		ll_rw_block(REQ_OP_READ, 0, nbufs, bufs);
114	err = 0;
115
116failed:
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
129static 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_maxlen) {
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		/* If this is a brand new buffer, start readahead.
157                   Otherwise, we assume we are already reading it.  */
158		if (!buffer_req(bh))
 
 
 
 
 
159			do_readahead(journal, offset);
160		wait_on_buffer(bh);
161	}
162
163	if (!buffer_uptodate(bh)) {
164		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
165			offset);
166		brelse(bh);
167		return -EIO;
168	}
169
170	*bhp = bh;
171	return 0;
172}
173
174static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
175{
176	struct jbd2_journal_block_tail *tail;
177	__be32 provided;
178	__u32 calculated;
179
180	if (!jbd2_journal_has_csum_v2or3(j))
181		return 1;
182
183	tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
184			sizeof(struct jbd2_journal_block_tail));
185	provided = tail->t_checksum;
186	tail->t_checksum = 0;
187	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
188	tail->t_checksum = provided;
189
190	return provided == cpu_to_be32(calculated);
191}
192
193/*
194 * Count the number of in-use tags in a journal descriptor block.
195 */
196
197static int count_tags(journal_t *journal, struct buffer_head *bh)
198{
199	char *			tagp;
200	journal_block_tag_t *	tag;
201	int			nr = 0, size = journal->j_blocksize;
202	int			tag_bytes = journal_tag_bytes(journal);
203
204	if (jbd2_journal_has_csum_v2or3(journal))
205		size -= sizeof(struct jbd2_journal_block_tail);
206
207	tagp = &bh->b_data[sizeof(journal_header_t)];
208
209	while ((tagp - bh->b_data + tag_bytes) <= size) {
210		tag = (journal_block_tag_t *) tagp;
211
212		nr++;
213		tagp += tag_bytes;
214		if (!(tag->t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
215			tagp += 16;
216
217		if (tag->t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
218			break;
219	}
220
221	return nr;
222}
223
224
225/* Make sure we wrap around the log correctly! */
226#define wrap(journal, var)						\
227do {									\
228	if (var >= (journal)->j_last)					\
229		var -= ((journal)->j_last - (journal)->j_first);	\
 
 
 
 
230} while (0)
231
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
232/**
233 * jbd2_journal_recover - recovers a on-disk journal
234 * @journal: the journal to recover
235 *
236 * The primary function for recovering the log contents when mounting a
237 * journaled device.
238 *
239 * Recovery is done in three passes.  In the first pass, we look for the
240 * end of the log.  In the second, we assemble the list of revoke
241 * blocks.  In the third and final pass, we replay any un-revoked blocks
242 * in the log.
243 */
244int jbd2_journal_recover(journal_t *journal)
245{
246	int			err, err2;
247	journal_superblock_t *	sb;
248
249	struct recovery_info	info;
250
251	memset(&info, 0, sizeof(info));
252	sb = journal->j_superblock;
253
254	/*
255	 * The journal superblock's s_start field (the current log head)
256	 * is always zero if, and only if, the journal was cleanly
257	 * unmounted.
258	 */
259
260	if (!sb->s_start) {
261		jbd_debug(1, "No recovery required, last transaction %d\n",
262			  be32_to_cpu(sb->s_sequence));
263		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
264		return 0;
265	}
266
267	err = do_one_pass(journal, &info, PASS_SCAN);
268	if (!err)
269		err = do_one_pass(journal, &info, PASS_REVOKE);
270	if (!err)
271		err = do_one_pass(journal, &info, PASS_REPLAY);
272
273	jbd_debug(1, "JBD2: recovery, exit status %d, "
274		  "recovered transactions %u to %u\n",
275		  err, info.start_transaction, info.end_transaction);
276	jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
277		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
278
279	/* Restart the log at the next transaction ID, thus invalidating
280	 * any existing commit records in the log. */
281	journal->j_transaction_sequence = ++info.end_transaction;
282
283	jbd2_journal_clear_revoke(journal);
284	err2 = sync_blockdev(journal->j_fs_dev);
285	if (!err)
286		err = err2;
287	/* Make sure all replayed data is on permanent storage */
288	if (journal->j_flags & JBD2_BARRIER) {
289		err2 = blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL);
290		if (!err)
291			err = err2;
292	}
293	return err;
294}
295
296/**
297 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
298 * @journal: journal to startup
299 *
300 * Locate any valid recovery information from the journal and set up the
301 * journal structures in memory to ignore it (presumably because the
302 * caller has evidence that it is out of date).
303 * This function doesn't appear to be exported..
304 *
305 * We perform one pass over the journal to allow us to tell the user how
306 * much recovery information is being erased, and to let us initialise
307 * the journal transaction sequence numbers to the next unused ID.
308 */
309int jbd2_journal_skip_recovery(journal_t *journal)
310{
311	int			err;
312
313	struct recovery_info	info;
314
315	memset (&info, 0, sizeof(info));
316
317	err = do_one_pass(journal, &info, PASS_SCAN);
318
319	if (err) {
320		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
321		++journal->j_transaction_sequence;
322	} else {
323#ifdef CONFIG_JBD2_DEBUG
324		int dropped = info.end_transaction - 
325			be32_to_cpu(journal->j_superblock->s_sequence);
326		jbd_debug(1,
327			  "JBD2: ignoring %d transaction%s from the journal.\n",
328			  dropped, (dropped == 1) ? "" : "s");
329#endif
330		journal->j_transaction_sequence = ++info.end_transaction;
331	}
332
333	journal->j_tail = 0;
334	return err;
335}
336
337static inline unsigned long long read_tag_block(journal_t *journal,
338						journal_block_tag_t *tag)
339{
340	unsigned long long block = be32_to_cpu(tag->t_blocknr);
341	if (jbd2_has_feature_64bit(journal))
342		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
343	return block;
344}
345
346/*
347 * calc_chksums calculates the checksums for the blocks described in the
348 * descriptor block.
349 */
350static int calc_chksums(journal_t *journal, struct buffer_head *bh,
351			unsigned long *next_log_block, __u32 *crc32_sum)
352{
353	int i, num_blks, err;
354	unsigned long io_block;
355	struct buffer_head *obh;
356
357	num_blks = count_tags(journal, bh);
358	/* Calculate checksum of the descriptor block. */
359	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
360
361	for (i = 0; i < num_blks; i++) {
362		io_block = (*next_log_block)++;
363		wrap(journal, *next_log_block);
364		err = jread(&obh, journal, io_block);
365		if (err) {
366			printk(KERN_ERR "JBD2: IO error %d recovering block "
367				"%lu in log\n", err, io_block);
368			return 1;
369		} else {
370			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
371				     obh->b_size);
372		}
373		put_bh(obh);
374	}
375	return 0;
376}
377
378static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
379{
380	struct commit_header *h;
381	__be32 provided;
382	__u32 calculated;
383
384	if (!jbd2_journal_has_csum_v2or3(j))
385		return 1;
386
387	h = buf;
388	provided = h->h_chksum[0];
389	h->h_chksum[0] = 0;
390	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
391	h->h_chksum[0] = provided;
392
393	return provided == cpu_to_be32(calculated);
394}
395
396static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
 
397				      void *buf, __u32 sequence)
398{
399	journal_block_tag3_t *tag3 = (journal_block_tag3_t *)tag;
400	__u32 csum32;
401	__be32 seq;
402
403	if (!jbd2_journal_has_csum_v2or3(j))
404		return 1;
405
406	seq = cpu_to_be32(sequence);
407	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
408	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
409
410	if (jbd2_has_feature_csum3(j))
411		return tag3->t_checksum == cpu_to_be32(csum32);
412	else
413		return tag->t_checksum == cpu_to_be16(csum32);
414}
415
416static int do_one_pass(journal_t *journal,
417			struct recovery_info *info, enum passtype pass)
418{
419	unsigned int		first_commit_ID, next_commit_ID;
420	unsigned long		next_log_block;
421	int			err, success = 0;
422	journal_superblock_t *	sb;
423	journal_header_t *	tmp;
424	struct buffer_head *	bh;
425	unsigned int		sequence;
426	int			blocktype;
427	int			tag_bytes = journal_tag_bytes(journal);
428	__u32			crc32_sum = ~0; /* Transactional Checksums */
429	int			descr_csum_size = 0;
430	int			block_error = 0;
 
 
431
432	/*
433	 * First thing is to establish what we expect to find in the log
434	 * (in terms of transaction IDs), and where (in terms of log
435	 * block offsets): query the superblock.
436	 */
437
438	sb = journal->j_superblock;
439	next_commit_ID = be32_to_cpu(sb->s_sequence);
440	next_log_block = be32_to_cpu(sb->s_start);
441
442	first_commit_ID = next_commit_ID;
443	if (pass == PASS_SCAN)
444		info->start_transaction = first_commit_ID;
445
446	jbd_debug(1, "Starting recovery pass %d\n", pass);
447
448	/*
449	 * Now we walk through the log, transaction by transaction,
450	 * making sure that each transaction has a commit block in the
451	 * expected place.  Each complete transaction gets replayed back
452	 * into the main filesystem.
453	 */
454
455	while (1) {
456		int			flags;
457		char *			tagp;
458		journal_block_tag_t *	tag;
459		struct buffer_head *	obh;
460		struct buffer_head *	nbh;
461
462		cond_resched();
463
464		/* If we already know where to stop the log traversal,
465		 * check right now that we haven't gone past the end of
466		 * the log. */
467
468		if (pass != PASS_SCAN)
469			if (tid_geq(next_commit_ID, info->end_transaction))
470				break;
471
472		jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
473			  next_commit_ID, next_log_block, journal->j_last);
 
 
474
475		/* Skip over each chunk of the transaction looking
476		 * either the next descriptor block or the final commit
477		 * record. */
478
479		jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
480		err = jread(&bh, journal, next_log_block);
481		if (err)
482			goto failed;
483
484		next_log_block++;
485		wrap(journal, next_log_block);
486
487		/* What kind of buffer is it?
488		 *
489		 * If it is a descriptor block, check that it has the
490		 * expected sequence number.  Otherwise, we're all done
491		 * here. */
492
493		tmp = (journal_header_t *)bh->b_data;
494
495		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
496			brelse(bh);
497			break;
498		}
499
500		blocktype = be32_to_cpu(tmp->h_blocktype);
501		sequence = be32_to_cpu(tmp->h_sequence);
502		jbd_debug(3, "Found magic %d, sequence %d\n",
503			  blocktype, sequence);
504
505		if (sequence != next_commit_ID) {
506			brelse(bh);
507			break;
508		}
509
510		/* OK, we have a valid descriptor block which matches
511		 * all of the sequence number checks.  What are we going
512		 * to do with it?  That depends on the pass... */
513
514		switch(blocktype) {
515		case JBD2_DESCRIPTOR_BLOCK:
516			/* Verify checksum first */
517			if (jbd2_journal_has_csum_v2or3(journal))
518				descr_csum_size =
519					sizeof(struct jbd2_journal_block_tail);
520			if (descr_csum_size > 0 &&
521			    !jbd2_descriptor_block_csum_verify(journal,
522							       bh->b_data)) {
523				printk(KERN_ERR "JBD2: Invalid checksum "
524				       "recovering block %lu in log\n",
525				       next_log_block);
526				err = -EFSBADCRC;
527				brelse(bh);
528				goto failed;
 
 
 
 
 
 
 
 
 
529			}
530
531			/* If it is a valid descriptor block, replay it
532			 * in pass REPLAY; if journal_checksums enabled, then
533			 * calculate checksums in PASS_SCAN, otherwise,
534			 * just skip over the blocks it describes. */
535			if (pass != PASS_REPLAY) {
536				if (pass == PASS_SCAN &&
537				    jbd2_has_feature_checksum(journal) &&
 
538				    !info->end_transaction) {
539					if (calc_chksums(journal, bh,
540							&next_log_block,
541							&crc32_sum)) {
542						put_bh(bh);
543						break;
544					}
545					put_bh(bh);
546					continue;
547				}
548				next_log_block += count_tags(journal, bh);
549				wrap(journal, next_log_block);
550				put_bh(bh);
551				continue;
552			}
553
554			/* A descriptor block: we can now write all of
555			 * the data blocks.  Yay, useful work is finally
556			 * getting done here! */
557
558			tagp = &bh->b_data[sizeof(journal_header_t)];
559			while ((tagp - bh->b_data + tag_bytes)
560			       <= journal->j_blocksize - descr_csum_size) {
561				unsigned long io_block;
562
563				tag = (journal_block_tag_t *) tagp;
564				flags = be16_to_cpu(tag->t_flags);
565
566				io_block = next_log_block++;
567				wrap(journal, next_log_block);
568				err = jread(&obh, journal, io_block);
569				if (err) {
570					/* Recover what we can, but
571					 * report failure at the end. */
572					success = err;
573					printk(KERN_ERR
574						"JBD2: IO error %d recovering "
575						"block %ld in log\n",
576						err, io_block);
577				} else {
578					unsigned long long blocknr;
579
580					J_ASSERT(obh != NULL);
581					blocknr = read_tag_block(journal,
582								 tag);
583
584					/* If the block has been
585					 * revoked, then we're all done
586					 * here. */
587					if (jbd2_journal_test_revoke
588					    (journal, blocknr,
589					     next_commit_ID)) {
590						brelse(obh);
591						++info->nr_revoke_hits;
592						goto skip_write;
593					}
594
595					/* Look for block corruption */
596					if (!jbd2_block_tag_csum_verify(
597						journal, tag, obh->b_data,
598						be32_to_cpu(tmp->h_sequence))) {
599						brelse(obh);
600						success = -EFSBADCRC;
601						printk(KERN_ERR "JBD2: Invalid "
602						       "checksum recovering "
603						       "data block %llu in "
604						       "log\n", blocknr);
605						block_error = 1;
606						goto skip_write;
607					}
608
609					/* Find a buffer for the new
610					 * data being restored */
611					nbh = __getblk(journal->j_fs_dev,
612							blocknr,
613							journal->j_blocksize);
614					if (nbh == NULL) {
615						printk(KERN_ERR
616						       "JBD2: Out of memory "
617						       "during recovery.\n");
618						err = -ENOMEM;
619						brelse(bh);
620						brelse(obh);
621						goto failed;
622					}
623
624					lock_buffer(nbh);
625					memcpy(nbh->b_data, obh->b_data,
626							journal->j_blocksize);
627					if (flags & JBD2_FLAG_ESCAPE) {
628						*((__be32 *)nbh->b_data) =
629						cpu_to_be32(JBD2_MAGIC_NUMBER);
630					}
631
632					BUFFER_TRACE(nbh, "marking dirty");
633					set_buffer_uptodate(nbh);
634					mark_buffer_dirty(nbh);
635					BUFFER_TRACE(nbh, "marking uptodate");
636					++info->nr_replays;
637					/* ll_rw_block(WRITE, 1, &nbh); */
638					unlock_buffer(nbh);
639					brelse(obh);
640					brelse(nbh);
641				}
642
643			skip_write:
644				tagp += tag_bytes;
645				if (!(flags & JBD2_FLAG_SAME_UUID))
646					tagp += 16;
647
648				if (flags & JBD2_FLAG_LAST_TAG)
649					break;
650			}
651
652			brelse(bh);
653			continue;
654
655		case JBD2_COMMIT_BLOCK:
656			/*     How to differentiate between interrupted commit
657			 *               and journal corruption ?
658			 *
659			 * {nth transaction}
660			 *        Checksum Verification Failed
661			 *			 |
662			 *		 ____________________
663			 *		|		     |
664			 * 	async_commit             sync_commit
665			 *     		|                    |
666			 *		| GO TO NEXT    "Journal Corruption"
667			 *		| TRANSACTION
668			 *		|
669			 * {(n+1)th transanction}
670			 *		|
671			 * 	 _______|______________
672			 * 	|	 	      |
673			 * Commit block found	Commit block not found
674			 *      |		      |
675			 * "Journal Corruption"       |
676			 *		 _____________|_________
677			 *     		|	           	|
678			 *	nth trans corrupt	OR   nth trans
679			 *	and (n+1)th interrupted     interrupted
680			 *	before commit block
681			 *      could reach the disk.
682			 *	(Cannot find the difference in above
683			 *	 mentioned conditions. Hence assume
684			 *	 "Interrupted Commit".)
685			 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
686
687			/* Found an expected commit block: if checksums
688			 * are present verify them in PASS_SCAN; else not
 
689			 * much to do other than move on to the next sequence
690			 * number. */
 
691			if (pass == PASS_SCAN &&
692			    jbd2_has_feature_checksum(journal)) {
693				struct commit_header *cbh =
694					(struct commit_header *)bh->b_data;
695				unsigned found_chksum =
696					be32_to_cpu(cbh->h_chksum[0]);
697
698				if (info->end_transaction) {
699					journal->j_failed_commit =
700						info->end_transaction;
701					brelse(bh);
702					break;
703				}
704
705				/* Neither checksum match nor unused? */
706				if (!((crc32_sum == found_chksum &&
707				       cbh->h_chksum_type ==
708						JBD2_CRC32_CHKSUM &&
709				       cbh->h_chksum_size ==
710						JBD2_CRC32_CHKSUM_SIZE) ||
711				      (cbh->h_chksum_type == 0 &&
712				       cbh->h_chksum_size == 0 &&
713				       found_chksum == 0)))
714					goto chksum_error;
715
716				crc32_sum = ~0;
717			}
718			if (pass == PASS_SCAN &&
719			    !jbd2_commit_block_csum_verify(journal,
720							   bh->b_data)) {
721			chksum_error:
 
 
722				info->end_transaction = next_commit_ID;
723
724				if (!jbd2_has_feature_async_commit(journal)) {
725					journal->j_failed_commit =
726						next_commit_ID;
727					brelse(bh);
728					break;
729				}
730			}
 
 
731			brelse(bh);
732			next_commit_ID++;
733			continue;
734
735		case JBD2_REVOKE_BLOCK:
 
 
 
 
 
 
 
 
 
 
 
736			/* If we aren't in the REVOKE pass, then we can
737			 * just skip over this block. */
738			if (pass != PASS_REVOKE) {
739				brelse(bh);
740				continue;
741			}
742
743			err = scan_revoke_records(journal, bh,
744						  next_commit_ID, info);
745			brelse(bh);
746			if (err)
747				goto failed;
748			continue;
749
750		default:
751			jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
752				  blocktype);
753			brelse(bh);
754			goto done;
755		}
756	}
757
758 done:
759	/*
760	 * We broke out of the log scan loop: either we came to the
761	 * known end of the log or we found an unexpected block in the
762	 * log.  If the latter happened, then we know that the "current"
763	 * transaction marks the end of the valid log.
764	 */
765
766	if (pass == PASS_SCAN) {
767		if (!info->end_transaction)
768			info->end_transaction = next_commit_ID;
769	} else {
770		/* It's really bad news if different passes end up at
771		 * different places (but possible due to IO errors). */
772		if (info->end_transaction != next_commit_ID) {
773			printk(KERN_ERR "JBD2: recovery pass %d ended at "
774				"transaction %u, expected %u\n",
775				pass, next_commit_ID, info->end_transaction);
776			if (!success)
777				success = -EIO;
778		}
779	}
 
 
 
 
 
 
 
780	if (block_error && success == 0)
781		success = -EIO;
782	return success;
783
784 failed:
785	return err;
786}
787
788/* Scan a revoke record, marking all blocks mentioned as revoked. */
789
790static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
791			       tid_t sequence, struct recovery_info *info)
792{
793	jbd2_journal_revoke_header_t *header;
794	int offset, max;
795	int csum_size = 0;
796	__u32 rcount;
797	int record_len = 4;
798
799	header = (jbd2_journal_revoke_header_t *) bh->b_data;
800	offset = sizeof(jbd2_journal_revoke_header_t);
801	rcount = be32_to_cpu(header->r_count);
802
803	if (!jbd2_descriptor_block_csum_verify(journal, header))
804		return -EFSBADCRC;
805
806	if (jbd2_journal_has_csum_v2or3(journal))
807		csum_size = sizeof(struct jbd2_journal_block_tail);
808	if (rcount > journal->j_blocksize - csum_size)
809		return -EINVAL;
810	max = rcount;
811
812	if (jbd2_has_feature_64bit(journal))
813		record_len = 8;
814
815	while (offset + record_len <= max) {
816		unsigned long long blocknr;
817		int err;
818
819		if (record_len == 4)
820			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
821		else
822			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
823		offset += record_len;
824		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
825		if (err)
826			return err;
827		++info->nr_revokes;
828	}
829	return 0;
830}