Linux Audio

Check our new training course

Open-source upstreaming

Need help get the support for your hardware in upstream Linux?
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}
v3.5.6
 
  1/*
  2 * linux/fs/jbd2/recovery.c
  3 *
  4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
  5 *
  6 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
  7 *
  8 * This file is part of the Linux kernel and is made available under
  9 * the terms of the GNU General Public License, version 2, or at your
 10 * option, any later version, incorporated herein by reference.
 11 *
 12 * Journal recovery routines for the generic filesystem journaling code;
 13 * part of the ext2fs journaling system.
 14 */
 15
 16#ifndef __KERNEL__
 17#include "jfs_user.h"
 18#else
 19#include <linux/time.h>
 20#include <linux/fs.h>
 21#include <linux/jbd2.h>
 22#include <linux/errno.h>
 23#include <linux/crc32.h>
 24#include <linux/blkdev.h>
 25#endif
 26
 27/*
 28 * Maintain information about the progress of the recovery job, so that
 29 * the different passes can carry information between them.
 30 */
 31struct recovery_info
 32{
 33	tid_t		start_transaction;
 34	tid_t		end_transaction;
 35
 36	int		nr_replays;
 37	int		nr_revokes;
 38	int		nr_revoke_hits;
 39};
 40
 41enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
 42static int do_one_pass(journal_t *journal,
 43				struct recovery_info *info, enum passtype pass);
 44static int scan_revoke_records(journal_t *, struct buffer_head *,
 45				tid_t, struct recovery_info *);
 46
 47#ifdef __KERNEL__
 48
 49/* Release readahead buffers after use */
 50static void journal_brelse_array(struct buffer_head *b[], int n)
 51{
 52	while (--n >= 0)
 53		brelse (b[n]);
 54}
 55
 56
 57/*
 58 * When reading from the journal, we are going through the block device
 59 * layer directly and so there is no readahead being done for us.  We
 60 * need to implement any readahead ourselves if we want it to happen at
 61 * all.  Recovery is basically one long sequential read, so make sure we
 62 * do the IO in reasonably large chunks.
 63 *
 64 * This is not so critical that we need to be enormously clever about
 65 * the readahead size, though.  128K is a purely arbitrary, good-enough
 66 * fixed value.
 67 */
 68
 69#define MAXBUF 8
 70static int do_readahead(journal_t *journal, unsigned int start)
 71{
 72	int err;
 73	unsigned int max, nbufs, next;
 74	unsigned long long blocknr;
 75	struct buffer_head *bh;
 76
 77	struct buffer_head * bufs[MAXBUF];
 78
 79	/* Do up to 128K of readahead */
 80	max = start + (128 * 1024 / journal->j_blocksize);
 81	if (max > journal->j_maxlen)
 82		max = journal->j_maxlen;
 83
 84	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
 85	 * a time to the block device IO layer. */
 86
 87	nbufs = 0;
 88
 89	for (next = start; next < max; next++) {
 90		err = jbd2_journal_bmap(journal, next, &blocknr);
 91
 92		if (err) {
 93			printk(KERN_ERR "JBD2: bad block at offset %u\n",
 94				next);
 95			goto failed;
 96		}
 97
 98		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 99		if (!bh) {
100			err = -ENOMEM;
101			goto failed;
102		}
103
104		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
105			bufs[nbufs++] = bh;
106			if (nbufs == MAXBUF) {
107				ll_rw_block(READ, nbufs, bufs);
108				journal_brelse_array(bufs, nbufs);
109				nbufs = 0;
110			}
111		} else
112			brelse(bh);
113	}
114
115	if (nbufs)
116		ll_rw_block(READ, nbufs, bufs);
117	err = 0;
118
119failed:
120	if (nbufs)
121		journal_brelse_array(bufs, nbufs);
122	return err;
123}
124
125#endif /* __KERNEL__ */
126
127
128/*
129 * Read a block from the journal
130 */
131
132static int jread(struct buffer_head **bhp, journal_t *journal,
133		 unsigned int offset)
134{
135	int err;
136	unsigned long long blocknr;
137	struct buffer_head *bh;
138
139	*bhp = NULL;
140
141	if (offset >= journal->j_maxlen) {
142		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
143		return -EIO;
144	}
145
146	err = jbd2_journal_bmap(journal, offset, &blocknr);
147
148	if (err) {
149		printk(KERN_ERR "JBD2: bad block at offset %u\n",
150			offset);
151		return err;
152	}
153
154	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
155	if (!bh)
156		return -ENOMEM;
157
158	if (!buffer_uptodate(bh)) {
159		/* If this is a brand new buffer, start readahead.
160                   Otherwise, we assume we are already reading it.  */
161		if (!buffer_req(bh))
 
 
 
 
 
162			do_readahead(journal, offset);
163		wait_on_buffer(bh);
164	}
165
166	if (!buffer_uptodate(bh)) {
167		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
168			offset);
169		brelse(bh);
170		return -EIO;
171	}
172
173	*bhp = bh;
174	return 0;
175}
176
177static int jbd2_descr_block_csum_verify(journal_t *j,
178					void *buf)
179{
180	struct jbd2_journal_block_tail *tail;
181	__u32 provided, calculated;
 
182
183	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
184		return 1;
185
186	tail = (struct jbd2_journal_block_tail *)(buf + j->j_blocksize -
187			sizeof(struct jbd2_journal_block_tail));
188	provided = tail->t_checksum;
189	tail->t_checksum = 0;
190	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
191	tail->t_checksum = provided;
192
193	provided = be32_to_cpu(provided);
194	return provided == 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_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
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		tag = (journal_block_tag_t *) tagp;
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	if (var >= (journal)->j_last)					\
233		var -= ((journal)->j_last - (journal)->j_first);	\
 
 
 
 
234} while (0)
235
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
236/**
237 * jbd2_journal_recover - recovers a on-disk journal
238 * @journal: the journal to recover
239 *
240 * The primary function for recovering the log contents when mounting a
241 * journaled device.
242 *
243 * Recovery is done in three passes.  In the first pass, we look for the
244 * end of the log.  In the second, we assemble the list of revoke
245 * blocks.  In the third and final pass, we replay any un-revoked blocks
246 * in the log.
247 */
248int jbd2_journal_recover(journal_t *journal)
249{
250	int			err, err2;
251	journal_superblock_t *	sb;
252
253	struct recovery_info	info;
254
255	memset(&info, 0, sizeof(info));
256	sb = journal->j_superblock;
257
258	/*
259	 * The journal superblock's s_start field (the current log head)
260	 * is always zero if, and only if, the journal was cleanly
261	 * unmounted.
262	 */
263
264	if (!sb->s_start) {
265		jbd_debug(1, "No recovery required, last transaction %d\n",
266			  be32_to_cpu(sb->s_sequence));
267		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
268		return 0;
269	}
270
271	err = do_one_pass(journal, &info, PASS_SCAN);
272	if (!err)
273		err = do_one_pass(journal, &info, PASS_REVOKE);
274	if (!err)
275		err = do_one_pass(journal, &info, PASS_REPLAY);
276
277	jbd_debug(1, "JBD2: recovery, exit status %d, "
278		  "recovered transactions %u to %u\n",
279		  err, info.start_transaction, info.end_transaction);
280	jbd_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
281		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
282
283	/* Restart the log at the next transaction ID, thus invalidating
284	 * any existing commit records in the log. */
285	journal->j_transaction_sequence = ++info.end_transaction;
286
287	jbd2_journal_clear_revoke(journal);
288	err2 = sync_blockdev(journal->j_fs_dev);
289	if (!err)
290		err = err2;
291	/* Make sure all replayed data is on permanent storage */
292	if (journal->j_flags & JBD2_BARRIER)
293		blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
 
 
 
294	return err;
295}
296
297/**
298 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
299 * @journal: journal to startup
300 *
301 * Locate any valid recovery information from the journal and set up the
302 * journal structures in memory to ignore it (presumably because the
303 * caller has evidence that it is out of date).
304 * This function does'nt appear to be exorted..
305 *
306 * We perform one pass over the journal to allow us to tell the user how
307 * much recovery information is being erased, and to let us initialise
308 * the journal transaction sequence numbers to the next unused ID.
309 */
310int jbd2_journal_skip_recovery(journal_t *journal)
311{
312	int			err;
313
314	struct recovery_info	info;
315
316	memset (&info, 0, sizeof(info));
317
318	err = do_one_pass(journal, &info, PASS_SCAN);
319
320	if (err) {
321		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
322		++journal->j_transaction_sequence;
323	} else {
324#ifdef CONFIG_JBD2_DEBUG
325		int dropped = info.end_transaction - 
326			be32_to_cpu(journal->j_superblock->s_sequence);
327		jbd_debug(1,
328			  "JBD2: ignoring %d transaction%s from the journal.\n",
329			  dropped, (dropped == 1) ? "" : "s");
330#endif
331		journal->j_transaction_sequence = ++info.end_transaction;
332	}
333
334	journal->j_tail = 0;
335	return err;
336}
337
338static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
 
339{
340	unsigned long long block = be32_to_cpu(tag->t_blocknr);
341	if (tag_bytes > JBD2_TAG_SIZE32)
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	__u32 provided, calculated;
 
382
383	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
384		return 1;
385
386	h = buf;
387	provided = h->h_chksum[0];
388	h->h_chksum[0] = 0;
389	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
390	h->h_chksum[0] = provided;
391
392	provided = be32_to_cpu(provided);
393	return provided == 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	__u32 provided, calculated;
 
400
401	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
402		return 1;
403
404	sequence = cpu_to_be32(sequence);
405	calculated = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&sequence,
406				 sizeof(sequence));
407	calculated = jbd2_chksum(j, calculated, buf, j->j_blocksize);
408	provided = be32_to_cpu(tag->t_checksum);
409
410	return provided == cpu_to_be32(calculated);
 
411}
412
413static int do_one_pass(journal_t *journal,
414			struct recovery_info *info, enum passtype pass)
415{
416	unsigned int		first_commit_ID, next_commit_ID;
417	unsigned long		next_log_block;
418	int			err, success = 0;
419	journal_superblock_t *	sb;
420	journal_header_t *	tmp;
421	struct buffer_head *	bh;
422	unsigned int		sequence;
423	int			blocktype;
424	int			tag_bytes = journal_tag_bytes(journal);
425	__u32			crc32_sum = ~0; /* Transactional Checksums */
426	int			descr_csum_size = 0;
 
 
 
427
428	/*
429	 * First thing is to establish what we expect to find in the log
430	 * (in terms of transaction IDs), and where (in terms of log
431	 * block offsets): query the superblock.
432	 */
433
434	sb = journal->j_superblock;
435	next_commit_ID = be32_to_cpu(sb->s_sequence);
436	next_log_block = be32_to_cpu(sb->s_start);
437
438	first_commit_ID = next_commit_ID;
439	if (pass == PASS_SCAN)
440		info->start_transaction = first_commit_ID;
441
442	jbd_debug(1, "Starting recovery pass %d\n", pass);
443
444	/*
445	 * Now we walk through the log, transaction by transaction,
446	 * making sure that each transaction has a commit block in the
447	 * expected place.  Each complete transaction gets replayed back
448	 * into the main filesystem.
449	 */
450
451	while (1) {
452		int			flags;
453		char *			tagp;
454		journal_block_tag_t *	tag;
455		struct buffer_head *	obh;
456		struct buffer_head *	nbh;
457
458		cond_resched();
459
460		/* If we already know where to stop the log traversal,
461		 * check right now that we haven't gone past the end of
462		 * the log. */
463
464		if (pass != PASS_SCAN)
465			if (tid_geq(next_commit_ID, info->end_transaction))
466				break;
467
468		jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
469			  next_commit_ID, next_log_block, journal->j_last);
 
 
470
471		/* Skip over each chunk of the transaction looking
472		 * either the next descriptor block or the final commit
473		 * record. */
474
475		jbd_debug(3, "JBD2: checking block %ld\n", next_log_block);
476		err = jread(&bh, journal, next_log_block);
477		if (err)
478			goto failed;
479
480		next_log_block++;
481		wrap(journal, next_log_block);
482
483		/* What kind of buffer is it?
484		 *
485		 * If it is a descriptor block, check that it has the
486		 * expected sequence number.  Otherwise, we're all done
487		 * here. */
488
489		tmp = (journal_header_t *)bh->b_data;
490
491		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
492			brelse(bh);
493			break;
494		}
495
496		blocktype = be32_to_cpu(tmp->h_blocktype);
497		sequence = be32_to_cpu(tmp->h_sequence);
498		jbd_debug(3, "Found magic %d, sequence %d\n",
499			  blocktype, sequence);
500
501		if (sequence != next_commit_ID) {
502			brelse(bh);
503			break;
504		}
505
506		/* OK, we have a valid descriptor block which matches
507		 * all of the sequence number checks.  What are we going
508		 * to do with it?  That depends on the pass... */
509
510		switch(blocktype) {
511		case JBD2_DESCRIPTOR_BLOCK:
512			/* Verify checksum first */
513			if (JBD2_HAS_INCOMPAT_FEATURE(journal,
514					JBD2_FEATURE_INCOMPAT_CSUM_V2))
515				descr_csum_size =
516					sizeof(struct jbd2_journal_block_tail);
517			if (descr_csum_size > 0 &&
518			    !jbd2_descr_block_csum_verify(journal,
519							  bh->b_data)) {
520				err = -EIO;
521				goto failed;
 
 
 
 
 
 
 
 
 
 
 
 
 
522			}
523
524			/* If it is a valid descriptor block, replay it
525			 * in pass REPLAY; if journal_checksums enabled, then
526			 * calculate checksums in PASS_SCAN, otherwise,
527			 * just skip over the blocks it describes. */
528			if (pass != PASS_REPLAY) {
529				if (pass == PASS_SCAN &&
530				    JBD2_HAS_COMPAT_FEATURE(journal,
531					    JBD2_FEATURE_COMPAT_CHECKSUM) &&
532				    !info->end_transaction) {
533					if (calc_chksums(journal, bh,
534							&next_log_block,
535							&crc32_sum)) {
536						put_bh(bh);
537						break;
538					}
539					put_bh(bh);
540					continue;
541				}
542				next_log_block += count_tags(journal, bh);
543				wrap(journal, next_log_block);
544				put_bh(bh);
545				continue;
546			}
547
548			/* A descriptor block: we can now write all of
549			 * the data blocks.  Yay, useful work is finally
550			 * getting done here! */
551
552			tagp = &bh->b_data[sizeof(journal_header_t)];
553			while ((tagp - bh->b_data + tag_bytes)
554			       <= journal->j_blocksize - descr_csum_size) {
555				unsigned long io_block;
556
557				tag = (journal_block_tag_t *) tagp;
558				flags = be16_to_cpu(tag->t_flags);
559
560				io_block = next_log_block++;
561				wrap(journal, next_log_block);
562				err = jread(&obh, journal, io_block);
563				if (err) {
564					/* Recover what we can, but
565					 * report failure at the end. */
566					success = err;
567					printk(KERN_ERR
568						"JBD2: IO error %d recovering "
569						"block %ld in log\n",
570						err, io_block);
571				} else {
572					unsigned long long blocknr;
573
574					J_ASSERT(obh != NULL);
575					blocknr = read_tag_block(tag_bytes,
576								 tag);
577
578					/* If the block has been
579					 * revoked, then we're all done
580					 * here. */
581					if (jbd2_journal_test_revoke
582					    (journal, blocknr,
583					     next_commit_ID)) {
584						brelse(obh);
585						++info->nr_revoke_hits;
586						goto skip_write;
587					}
588
589					/* Look for block corruption */
590					if (!jbd2_block_tag_csum_verify(
591						journal, tag, obh->b_data,
592						be32_to_cpu(tmp->h_sequence))) {
593						brelse(obh);
594						success = -EIO;
595						printk(KERN_ERR "JBD: Invalid "
596						       "checksum recovering "
597						       "block %llu in log\n",
598						       blocknr);
599						continue;
 
600					}
601
602					/* Find a buffer for the new
603					 * data being restored */
604					nbh = __getblk(journal->j_fs_dev,
605							blocknr,
606							journal->j_blocksize);
607					if (nbh == NULL) {
608						printk(KERN_ERR
609						       "JBD2: Out of memory "
610						       "during recovery.\n");
611						err = -ENOMEM;
612						brelse(bh);
613						brelse(obh);
614						goto failed;
615					}
616
617					lock_buffer(nbh);
618					memcpy(nbh->b_data, obh->b_data,
619							journal->j_blocksize);
620					if (flags & JBD2_FLAG_ESCAPE) {
621						*((__be32 *)nbh->b_data) =
622						cpu_to_be32(JBD2_MAGIC_NUMBER);
623					}
624
625					BUFFER_TRACE(nbh, "marking dirty");
626					set_buffer_uptodate(nbh);
627					mark_buffer_dirty(nbh);
628					BUFFER_TRACE(nbh, "marking uptodate");
629					++info->nr_replays;
630					/* ll_rw_block(WRITE, 1, &nbh); */
631					unlock_buffer(nbh);
632					brelse(obh);
633					brelse(nbh);
634				}
635
636			skip_write:
637				tagp += tag_bytes;
638				if (!(flags & JBD2_FLAG_SAME_UUID))
639					tagp += 16;
640
641				if (flags & JBD2_FLAG_LAST_TAG)
642					break;
643			}
644
645			brelse(bh);
646			continue;
647
648		case JBD2_COMMIT_BLOCK:
649			/*     How to differentiate between interrupted commit
650			 *               and journal corruption ?
651			 *
652			 * {nth transaction}
653			 *        Checksum Verification Failed
654			 *			 |
655			 *		 ____________________
656			 *		|		     |
657			 * 	async_commit             sync_commit
658			 *     		|                    |
659			 *		| GO TO NEXT    "Journal Corruption"
660			 *		| TRANSACTION
661			 *		|
662			 * {(n+1)th transanction}
663			 *		|
664			 * 	 _______|______________
665			 * 	|	 	      |
666			 * Commit block found	Commit block not found
667			 *      |		      |
668			 * "Journal Corruption"       |
669			 *		 _____________|_________
670			 *     		|	           	|
671			 *	nth trans corrupt	OR   nth trans
672			 *	and (n+1)th interrupted     interrupted
673			 *	before commit block
674			 *      could reach the disk.
675			 *	(Cannot find the difference in above
676			 *	 mentioned conditions. Hence assume
677			 *	 "Interrupted Commit".)
678			 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
679
680			/* Found an expected commit block: if checksums
681			 * are present verify them in PASS_SCAN; else not
 
682			 * much to do other than move on to the next sequence
683			 * number. */
 
684			if (pass == PASS_SCAN &&
685			    JBD2_HAS_COMPAT_FEATURE(journal,
686				    JBD2_FEATURE_COMPAT_CHECKSUM)) {
687				int chksum_err, chksum_seen;
688				struct commit_header *cbh =
689					(struct commit_header *)bh->b_data;
690				unsigned found_chksum =
691					be32_to_cpu(cbh->h_chksum[0]);
692
693				chksum_err = chksum_seen = 0;
694
695				if (info->end_transaction) {
696					journal->j_failed_commit =
697						info->end_transaction;
698					brelse(bh);
699					break;
700				}
701
702				if (crc32_sum == found_chksum &&
703				    cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
704				    cbh->h_chksum_size ==
705						JBD2_CRC32_CHKSUM_SIZE)
706				       chksum_seen = 1;
707				else if (!(cbh->h_chksum_type == 0 &&
708					     cbh->h_chksum_size == 0 &&
709					     found_chksum == 0 &&
710					     !chksum_seen))
711				/*
712				 * If fs is mounted using an old kernel and then
713				 * kernel with journal_chksum is used then we
714				 * get a situation where the journal flag has
715				 * checksum flag set but checksums are not
716				 * present i.e chksum = 0, in the individual
717				 * commit blocks.
718				 * Hence to avoid checksum failures, in this
719				 * situation, this extra check is added.
720				 */
721						chksum_err = 1;
722
723				if (chksum_err) {
724					info->end_transaction = next_commit_ID;
725
726					if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
727					   JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
728						journal->j_failed_commit =
729							next_commit_ID;
730						brelse(bh);
731						break;
732					}
733				}
734				crc32_sum = ~0;
735			}
736			if (pass == PASS_SCAN &&
737			    !jbd2_commit_block_csum_verify(journal,
738							   bh->b_data)) {
 
 
 
739				info->end_transaction = next_commit_ID;
740
741				if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
742				     JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
743					journal->j_failed_commit =
744						next_commit_ID;
745					brelse(bh);
746					break;
747				}
748			}
 
 
749			brelse(bh);
750			next_commit_ID++;
751			continue;
752
753		case JBD2_REVOKE_BLOCK:
 
 
 
 
 
 
 
 
 
 
 
754			/* If we aren't in the REVOKE pass, then we can
755			 * just skip over this block. */
756			if (pass != PASS_REVOKE) {
757				brelse(bh);
758				continue;
759			}
760
761			err = scan_revoke_records(journal, bh,
762						  next_commit_ID, info);
763			brelse(bh);
764			if (err)
765				goto failed;
766			continue;
767
768		default:
769			jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
770				  blocktype);
771			brelse(bh);
772			goto done;
773		}
774	}
775
776 done:
777	/*
778	 * We broke out of the log scan loop: either we came to the
779	 * known end of the log or we found an unexpected block in the
780	 * log.  If the latter happened, then we know that the "current"
781	 * transaction marks the end of the valid log.
782	 */
783
784	if (pass == PASS_SCAN) {
785		if (!info->end_transaction)
786			info->end_transaction = next_commit_ID;
787	} else {
788		/* It's really bad news if different passes end up at
789		 * different places (but possible due to IO errors). */
790		if (info->end_transaction != next_commit_ID) {
791			printk(KERN_ERR "JBD2: recovery pass %d ended at "
792				"transaction %u, expected %u\n",
793				pass, next_commit_ID, info->end_transaction);
794			if (!success)
795				success = -EIO;
796		}
797	}
798
 
 
 
 
 
 
 
 
799	return success;
800
801 failed:
802	return err;
803}
804
805static int jbd2_revoke_block_csum_verify(journal_t *j,
806					 void *buf)
807{
808	struct jbd2_journal_revoke_tail *tail;
809	__u32 provided, calculated;
810
811	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
812		return 1;
813
814	tail = (struct jbd2_journal_revoke_tail *)(buf + j->j_blocksize -
815			sizeof(struct jbd2_journal_revoke_tail));
816	provided = tail->r_checksum;
817	tail->r_checksum = 0;
818	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
819	tail->r_checksum = provided;
820
821	provided = be32_to_cpu(provided);
822	return provided == calculated;
823}
824
825/* Scan a revoke record, marking all blocks mentioned as revoked. */
826
827static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
828			       tid_t sequence, struct recovery_info *info)
829{
830	jbd2_journal_revoke_header_t *header;
831	int offset, max;
 
 
832	int record_len = 4;
833
834	header = (jbd2_journal_revoke_header_t *) bh->b_data;
835	offset = sizeof(jbd2_journal_revoke_header_t);
836	max = be32_to_cpu(header->r_count);
837
838	if (!jbd2_revoke_block_csum_verify(journal, header))
 
 
839		return -EINVAL;
 
840
841	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
842		record_len = 8;
843
844	while (offset + record_len <= max) {
845		unsigned long long blocknr;
846		int err;
847
848		if (record_len == 4)
849			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
850		else
851			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
852		offset += record_len;
853		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
854		if (err)
855			return err;
856		++info->nr_revokes;
857	}
858	return 0;
859}