Linux Audio

Check our new training course

Loading...
v3.1
 
  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#endif
 25
 26/*
 27 * Maintain information about the progress of the recovery job, so that
 28 * the different passes can carry information between them.
 29 */
 30struct recovery_info
 31{
 32	tid_t		start_transaction;
 33	tid_t		end_transaction;
 
 34
 35	int		nr_replays;
 36	int		nr_revokes;
 37	int		nr_revoke_hits;
 38};
 39
 40enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
 41static int do_one_pass(journal_t *journal,
 42				struct recovery_info *info, enum passtype pass);
 43static int scan_revoke_records(journal_t *, struct buffer_head *,
 44				tid_t, struct recovery_info *);
 45
 46#ifdef __KERNEL__
 47
 48/* Release readahead buffers after use */
 49static void journal_brelse_array(struct buffer_head *b[], int n)
 50{
 51	while (--n >= 0)
 52		brelse (b[n]);
 53}
 54
 55
 56/*
 57 * When reading from the journal, we are going through the block device
 58 * layer directly and so there is no readahead being done for us.  We
 59 * need to implement any readahead ourselves if we want it to happen at
 60 * all.  Recovery is basically one long sequential read, so make sure we
 61 * do the IO in reasonably large chunks.
 62 *
 63 * This is not so critical that we need to be enormously clever about
 64 * the readahead size, though.  128K is a purely arbitrary, good-enough
 65 * fixed value.
 66 */
 67
 68#define MAXBUF 8
 69static int do_readahead(journal_t *journal, unsigned int start)
 70{
 71	int err;
 72	unsigned int max, nbufs, next;
 73	unsigned long long blocknr;
 74	struct buffer_head *bh;
 75
 76	struct buffer_head * bufs[MAXBUF];
 77
 78	/* Do up to 128K of readahead */
 79	max = start + (128 * 1024 / journal->j_blocksize);
 80	if (max > journal->j_maxlen)
 81		max = journal->j_maxlen;
 82
 83	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
 84	 * a time to the block device IO layer. */
 85
 86	nbufs = 0;
 87
 88	for (next = start; next < max; next++) {
 89		err = jbd2_journal_bmap(journal, next, &blocknr);
 90
 91		if (err) {
 92			printk (KERN_ERR "JBD: bad block at offset %u\n",
 93				next);
 94			goto failed;
 95		}
 96
 97		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 98		if (!bh) {
 99			err = -ENOMEM;
100			goto failed;
101		}
102
103		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
104			bufs[nbufs++] = bh;
105			if (nbufs == MAXBUF) {
106				ll_rw_block(READ, nbufs, bufs);
107				journal_brelse_array(bufs, nbufs);
108				nbufs = 0;
109			}
110		} else
111			brelse(bh);
112	}
113
114	if (nbufs)
115		ll_rw_block(READ, nbufs, bufs);
116	err = 0;
117
118failed:
119	if (nbufs)
120		journal_brelse_array(bufs, nbufs);
121	return err;
122}
123
124#endif /* __KERNEL__ */
125
126
127/*
128 * Read a block from the journal
129 */
130
131static int jread(struct buffer_head **bhp, journal_t *journal,
132		 unsigned int offset)
133{
134	int err;
135	unsigned long long blocknr;
136	struct buffer_head *bh;
137
138	*bhp = NULL;
139
140	if (offset >= journal->j_maxlen) {
141		printk(KERN_ERR "JBD: corrupted journal superblock\n");
142		return -EIO;
143	}
144
145	err = jbd2_journal_bmap(journal, offset, &blocknr);
146
147	if (err) {
148		printk (KERN_ERR "JBD: bad block at offset %u\n",
149			offset);
150		return err;
151	}
152
153	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
154	if (!bh)
155		return -ENOMEM;
156
157	if (!buffer_uptodate(bh)) {
158		/* If this is a brand new buffer, start readahead.
159                   Otherwise, we assume we are already reading it.  */
160		if (!buffer_req(bh))
 
 
 
 
 
161			do_readahead(journal, offset);
162		wait_on_buffer(bh);
163	}
164
165	if (!buffer_uptodate(bh)) {
166		printk (KERN_ERR "JBD: Failed to read block at offset %u\n",
167			offset);
168		brelse(bh);
169		return -EIO;
170	}
171
172	*bhp = bh;
173	return 0;
174}
175
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
176
177/*
178 * Count the number of in-use tags in a journal descriptor block.
179 */
180
181static int count_tags(journal_t *journal, struct buffer_head *bh)
182{
183	char *			tagp;
184	journal_block_tag_t *	tag;
185	int			nr = 0, size = journal->j_blocksize;
186	int			tag_bytes = journal_tag_bytes(journal);
187
 
 
 
188	tagp = &bh->b_data[sizeof(journal_header_t)];
189
190	while ((tagp - bh->b_data + tag_bytes) <= size) {
191		tag = (journal_block_tag_t *) tagp;
192
193		nr++;
194		tagp += tag_bytes;
195		if (!(tag->t_flags & cpu_to_be32(JBD2_FLAG_SAME_UUID)))
196			tagp += 16;
197
198		if (tag->t_flags & cpu_to_be32(JBD2_FLAG_LAST_TAG))
199			break;
200	}
201
202	return nr;
203}
204
205
206/* Make sure we wrap around the log correctly! */
207#define wrap(journal, var)						\
208do {									\
209	if (var >= (journal)->j_last)					\
210		var -= ((journal)->j_last - (journal)->j_first);	\
211} while (0)
212
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
213/**
214 * jbd2_journal_recover - recovers a on-disk journal
215 * @journal: the journal to recover
216 *
217 * The primary function for recovering the log contents when mounting a
218 * journaled device.
219 *
220 * Recovery is done in three passes.  In the first pass, we look for the
221 * end of the log.  In the second, we assemble the list of revoke
222 * blocks.  In the third and final pass, we replay any un-revoked blocks
223 * in the log.
224 */
225int jbd2_journal_recover(journal_t *journal)
226{
227	int			err, err2;
228	journal_superblock_t *	sb;
229
230	struct recovery_info	info;
231
232	memset(&info, 0, sizeof(info));
233	sb = journal->j_superblock;
234
235	/*
236	 * The journal superblock's s_start field (the current log head)
237	 * is always zero if, and only if, the journal was cleanly
238	 * unmounted.
239	 */
240
241	if (!sb->s_start) {
242		jbd_debug(1, "No recovery required, last transaction %d\n",
243			  be32_to_cpu(sb->s_sequence));
244		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
 
245		return 0;
246	}
247
248	err = do_one_pass(journal, &info, PASS_SCAN);
249	if (!err)
250		err = do_one_pass(journal, &info, PASS_REVOKE);
251	if (!err)
252		err = do_one_pass(journal, &info, PASS_REPLAY);
253
254	jbd_debug(1, "JBD: recovery, exit status %d, "
255		  "recovered transactions %u to %u\n",
256		  err, info.start_transaction, info.end_transaction);
257	jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n",
258		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
259
260	/* Restart the log at the next transaction ID, thus invalidating
261	 * any existing commit records in the log. */
262	journal->j_transaction_sequence = ++info.end_transaction;
 
 
 
263
264	jbd2_journal_clear_revoke(journal);
265	err2 = sync_blockdev(journal->j_fs_dev);
266	if (!err)
267		err = err2;
268
 
 
 
 
 
 
 
 
269	return err;
270}
271
272/**
273 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
274 * @journal: journal to startup
275 *
276 * Locate any valid recovery information from the journal and set up the
277 * journal structures in memory to ignore it (presumably because the
278 * caller has evidence that it is out of date).
279 * This function does'nt appear to be exorted..
280 *
281 * We perform one pass over the journal to allow us to tell the user how
282 * much recovery information is being erased, and to let us initialise
283 * the journal transaction sequence numbers to the next unused ID.
284 */
285int jbd2_journal_skip_recovery(journal_t *journal)
286{
287	int			err;
288
289	struct recovery_info	info;
290
291	memset (&info, 0, sizeof(info));
292
293	err = do_one_pass(journal, &info, PASS_SCAN);
294
295	if (err) {
296		printk(KERN_ERR "JBD: error %d scanning journal\n", err);
297		++journal->j_transaction_sequence;
 
298	} else {
299#ifdef CONFIG_JBD2_DEBUG
300		int dropped = info.end_transaction - 
301			be32_to_cpu(journal->j_superblock->s_sequence);
302		jbd_debug(1,
303			  "JBD: ignoring %d transaction%s from the journal.\n",
304			  dropped, (dropped == 1) ? "" : "s");
305#endif
306		journal->j_transaction_sequence = ++info.end_transaction;
 
307	}
308
309	journal->j_tail = 0;
310	return err;
311}
312
313static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag_t *tag)
 
314{
315	unsigned long long block = be32_to_cpu(tag->t_blocknr);
316	if (tag_bytes > JBD2_TAG_SIZE32)
317		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
318	return block;
319}
320
321/*
322 * calc_chksums calculates the checksums for the blocks described in the
323 * descriptor block.
324 */
325static int calc_chksums(journal_t *journal, struct buffer_head *bh,
326			unsigned long *next_log_block, __u32 *crc32_sum)
327{
328	int i, num_blks, err;
329	unsigned long io_block;
330	struct buffer_head *obh;
331
332	num_blks = count_tags(journal, bh);
333	/* Calculate checksum of the descriptor block. */
334	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
335
336	for (i = 0; i < num_blks; i++) {
337		io_block = (*next_log_block)++;
338		wrap(journal, *next_log_block);
339		err = jread(&obh, journal, io_block);
340		if (err) {
341			printk(KERN_ERR "JBD: IO error %d recovering block "
342				"%lu in log\n", err, io_block);
343			return 1;
344		} else {
345			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
346				     obh->b_size);
347		}
348		put_bh(obh);
349	}
350	return 0;
351}
352
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
353static int do_one_pass(journal_t *journal,
354			struct recovery_info *info, enum passtype pass)
355{
356	unsigned int		first_commit_ID, next_commit_ID;
357	unsigned long		next_log_block;
358	int			err, success = 0;
359	journal_superblock_t *	sb;
360	journal_header_t *	tmp;
361	struct buffer_head *	bh;
362	unsigned int		sequence;
363	int			blocktype;
364	int			tag_bytes = journal_tag_bytes(journal);
365	__u32			crc32_sum = ~0; /* Transactional Checksums */
 
 
 
 
366
367	/*
368	 * First thing is to establish what we expect to find in the log
369	 * (in terms of transaction IDs), and where (in terms of log
370	 * block offsets): query the superblock.
371	 */
372
373	sb = journal->j_superblock;
374	next_commit_ID = be32_to_cpu(sb->s_sequence);
375	next_log_block = be32_to_cpu(sb->s_start);
 
376
377	first_commit_ID = next_commit_ID;
378	if (pass == PASS_SCAN)
379		info->start_transaction = first_commit_ID;
380
381	jbd_debug(1, "Starting recovery pass %d\n", pass);
382
383	/*
384	 * Now we walk through the log, transaction by transaction,
385	 * making sure that each transaction has a commit block in the
386	 * expected place.  Each complete transaction gets replayed back
387	 * into the main filesystem.
388	 */
389
390	while (1) {
391		int			flags;
392		char *			tagp;
393		journal_block_tag_t *	tag;
394		struct buffer_head *	obh;
395		struct buffer_head *	nbh;
396
397		cond_resched();
398
399		/* If we already know where to stop the log traversal,
400		 * check right now that we haven't gone past the end of
401		 * the log. */
402
403		if (pass != PASS_SCAN)
404			if (tid_geq(next_commit_ID, info->end_transaction))
405				break;
406
407		jbd_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
408			  next_commit_ID, next_log_block, journal->j_last);
409
410		/* Skip over each chunk of the transaction looking
411		 * either the next descriptor block or the final commit
412		 * record. */
413
414		jbd_debug(3, "JBD: checking block %ld\n", next_log_block);
415		err = jread(&bh, journal, next_log_block);
416		if (err)
417			goto failed;
418
419		next_log_block++;
420		wrap(journal, next_log_block);
421
422		/* What kind of buffer is it?
423		 *
424		 * If it is a descriptor block, check that it has the
425		 * expected sequence number.  Otherwise, we're all done
426		 * here. */
427
428		tmp = (journal_header_t *)bh->b_data;
429
430		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
431			brelse(bh);
432			break;
433		}
434
435		blocktype = be32_to_cpu(tmp->h_blocktype);
436		sequence = be32_to_cpu(tmp->h_sequence);
437		jbd_debug(3, "Found magic %d, sequence %d\n",
438			  blocktype, sequence);
439
440		if (sequence != next_commit_ID) {
441			brelse(bh);
442			break;
443		}
444
445		/* OK, we have a valid descriptor block which matches
446		 * all of the sequence number checks.  What are we going
447		 * to do with it?  That depends on the pass... */
448
449		switch(blocktype) {
450		case JBD2_DESCRIPTOR_BLOCK:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
451			/* If it is a valid descriptor block, replay it
452			 * in pass REPLAY; if journal_checksums enabled, then
453			 * calculate checksums in PASS_SCAN, otherwise,
454			 * just skip over the blocks it describes. */
455			if (pass != PASS_REPLAY) {
456				if (pass == PASS_SCAN &&
457				    JBD2_HAS_COMPAT_FEATURE(journal,
458					    JBD2_FEATURE_COMPAT_CHECKSUM) &&
459				    !info->end_transaction) {
460					if (calc_chksums(journal, bh,
461							&next_log_block,
462							&crc32_sum)) {
463						put_bh(bh);
464						break;
465					}
466					put_bh(bh);
467					continue;
468				}
469				next_log_block += count_tags(journal, bh);
470				wrap(journal, next_log_block);
471				put_bh(bh);
472				continue;
473			}
474
475			/* A descriptor block: we can now write all of
476			 * the data blocks.  Yay, useful work is finally
477			 * getting done here! */
478
479			tagp = &bh->b_data[sizeof(journal_header_t)];
480			while ((tagp - bh->b_data + tag_bytes)
481			       <= journal->j_blocksize) {
482				unsigned long io_block;
483
484				tag = (journal_block_tag_t *) tagp;
485				flags = be32_to_cpu(tag->t_flags);
486
487				io_block = next_log_block++;
488				wrap(journal, next_log_block);
489				err = jread(&obh, journal, io_block);
490				if (err) {
491					/* Recover what we can, but
492					 * report failure at the end. */
493					success = err;
494					printk (KERN_ERR
495						"JBD: IO error %d recovering "
496						"block %ld in log\n",
497						err, io_block);
498				} else {
499					unsigned long long blocknr;
500
501					J_ASSERT(obh != NULL);
502					blocknr = read_tag_block(tag_bytes,
503								 tag);
504
505					/* If the block has been
506					 * revoked, then we're all done
507					 * here. */
508					if (jbd2_journal_test_revoke
509					    (journal, blocknr,
510					     next_commit_ID)) {
511						brelse(obh);
512						++info->nr_revoke_hits;
513						goto skip_write;
514					}
515
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
516					/* Find a buffer for the new
517					 * data being restored */
518					nbh = __getblk(journal->j_fs_dev,
519							blocknr,
520							journal->j_blocksize);
521					if (nbh == NULL) {
522						printk(KERN_ERR
523						       "JBD: Out of memory "
524						       "during recovery.\n");
525						err = -ENOMEM;
526						brelse(bh);
527						brelse(obh);
528						goto failed;
529					}
530
531					lock_buffer(nbh);
532					memcpy(nbh->b_data, obh->b_data,
533							journal->j_blocksize);
534					if (flags & JBD2_FLAG_ESCAPE) {
535						*((__be32 *)nbh->b_data) =
536						cpu_to_be32(JBD2_MAGIC_NUMBER);
537					}
538
539					BUFFER_TRACE(nbh, "marking dirty");
540					set_buffer_uptodate(nbh);
541					mark_buffer_dirty(nbh);
542					BUFFER_TRACE(nbh, "marking uptodate");
543					++info->nr_replays;
544					/* ll_rw_block(WRITE, 1, &nbh); */
545					unlock_buffer(nbh);
546					brelse(obh);
547					brelse(nbh);
548				}
549
550			skip_write:
551				tagp += tag_bytes;
552				if (!(flags & JBD2_FLAG_SAME_UUID))
553					tagp += 16;
554
555				if (flags & JBD2_FLAG_LAST_TAG)
556					break;
557			}
558
559			brelse(bh);
560			continue;
561
562		case JBD2_COMMIT_BLOCK:
563			/*     How to differentiate between interrupted commit
564			 *               and journal corruption ?
565			 *
566			 * {nth transaction}
567			 *        Checksum Verification Failed
568			 *			 |
569			 *		 ____________________
570			 *		|		     |
571			 * 	async_commit             sync_commit
572			 *     		|                    |
573			 *		| GO TO NEXT    "Journal Corruption"
574			 *		| TRANSACTION
575			 *		|
576			 * {(n+1)th transanction}
577			 *		|
578			 * 	 _______|______________
579			 * 	|	 	      |
580			 * Commit block found	Commit block not found
581			 *      |		      |
582			 * "Journal Corruption"       |
583			 *		 _____________|_________
584			 *     		|	           	|
585			 *	nth trans corrupt	OR   nth trans
586			 *	and (n+1)th interrupted     interrupted
587			 *	before commit block
588			 *      could reach the disk.
589			 *	(Cannot find the difference in above
590			 *	 mentioned conditions. Hence assume
591			 *	 "Interrupted Commit".)
592			 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
593
594			/* Found an expected commit block: if checksums
595			 * are present verify them in PASS_SCAN; else not
 
596			 * much to do other than move on to the next sequence
597			 * number. */
 
598			if (pass == PASS_SCAN &&
599			    JBD2_HAS_COMPAT_FEATURE(journal,
600				    JBD2_FEATURE_COMPAT_CHECKSUM)) {
601				int chksum_err, chksum_seen;
602				struct commit_header *cbh =
603					(struct commit_header *)bh->b_data;
604				unsigned found_chksum =
605					be32_to_cpu(cbh->h_chksum[0]);
606
607				chksum_err = chksum_seen = 0;
608
609				if (info->end_transaction) {
610					journal->j_failed_commit =
611						info->end_transaction;
612					brelse(bh);
613					break;
614				}
615
616				if (crc32_sum == found_chksum &&
617				    cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
618				    cbh->h_chksum_size ==
619						JBD2_CRC32_CHKSUM_SIZE)
620				       chksum_seen = 1;
621				else if (!(cbh->h_chksum_type == 0 &&
622					     cbh->h_chksum_size == 0 &&
623					     found_chksum == 0 &&
624					     !chksum_seen))
625				/*
626				 * If fs is mounted using an old kernel and then
627				 * kernel with journal_chksum is used then we
628				 * get a situation where the journal flag has
629				 * checksum flag set but checksums are not
630				 * present i.e chksum = 0, in the individual
631				 * commit blocks.
632				 * Hence to avoid checksum failures, in this
633				 * situation, this extra check is added.
634				 */
635						chksum_err = 1;
636
637				if (chksum_err) {
638					info->end_transaction = next_commit_ID;
 
 
 
 
 
 
 
 
639
640					if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
641					   JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
642						journal->j_failed_commit =
643							next_commit_ID;
644						brelse(bh);
645						break;
646					}
647				}
648				crc32_sum = ~0;
 
 
 
649			}
650			brelse(bh);
651			next_commit_ID++;
652			continue;
653
654		case JBD2_REVOKE_BLOCK:
 
 
 
 
 
 
 
 
 
 
 
655			/* If we aren't in the REVOKE pass, then we can
656			 * just skip over this block. */
657			if (pass != PASS_REVOKE) {
658				brelse(bh);
659				continue;
660			}
661
662			err = scan_revoke_records(journal, bh,
663						  next_commit_ID, info);
664			brelse(bh);
665			if (err)
666				goto failed;
667			continue;
668
669		default:
670			jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
671				  blocktype);
672			brelse(bh);
673			goto done;
674		}
675	}
676
677 done:
678	/*
679	 * We broke out of the log scan loop: either we came to the
680	 * known end of the log or we found an unexpected block in the
681	 * log.  If the latter happened, then we know that the "current"
682	 * transaction marks the end of the valid log.
683	 */
684
685	if (pass == PASS_SCAN) {
686		if (!info->end_transaction)
687			info->end_transaction = next_commit_ID;
 
 
688	} else {
689		/* It's really bad news if different passes end up at
690		 * different places (but possible due to IO errors). */
691		if (info->end_transaction != next_commit_ID) {
692			printk (KERN_ERR "JBD: recovery pass %d ended at "
693				"transaction %u, expected %u\n",
694				pass, next_commit_ID, info->end_transaction);
695			if (!success)
696				success = -EIO;
697		}
698	}
699
 
 
 
 
 
 
 
 
700	return success;
701
702 failed:
703	return err;
704}
705
706
707/* Scan a revoke record, marking all blocks mentioned as revoked. */
708
709static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
710			       tid_t sequence, struct recovery_info *info)
711{
712	jbd2_journal_revoke_header_t *header;
713	int offset, max;
 
 
714	int record_len = 4;
715
716	header = (jbd2_journal_revoke_header_t *) bh->b_data;
717	offset = sizeof(jbd2_journal_revoke_header_t);
718	max = be32_to_cpu(header->r_count);
 
 
 
 
 
 
719
720	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
721		record_len = 8;
722
723	while (offset + record_len <= max) {
724		unsigned long long blocknr;
725		int err;
726
727		if (record_len == 4)
728			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
729		else
730			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
731		offset += record_len;
732		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
733		if (err)
734			return err;
735		++info->nr_revokes;
736	}
737	return 0;
738}
v6.9.4
  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	unsigned long	head_block;
 33
 34	int		nr_replays;
 35	int		nr_revokes;
 36	int		nr_revoke_hits;
 37};
 38
 
 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_total_len)
 79		max = journal->j_total_len;
 80
 81	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
 82	 * a time to the block device IO layer. */
 83
 84	nbufs = 0;
 85
 86	for (next = start; next < max; next++) {
 87		err = jbd2_journal_bmap(journal, next, &blocknr);
 88
 89		if (err) {
 90			printk(KERN_ERR "JBD2: bad block at offset %u\n",
 91				next);
 92			goto failed;
 93		}
 94
 95		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 96		if (!bh) {
 97			err = -ENOMEM;
 98			goto failed;
 99		}
100
101		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
102			bufs[nbufs++] = bh;
103			if (nbufs == MAXBUF) {
104				bh_readahead_batch(nbufs, bufs, 0);
105				journal_brelse_array(bufs, nbufs);
106				nbufs = 0;
107			}
108		} else
109			brelse(bh);
110	}
111
112	if (nbufs)
113		bh_readahead_batch(nbufs, bufs, 0);
114	err = 0;
115
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_total_len) {
139		printk(KERN_ERR "JBD2: corrupted journal superblock\n");
140		return -EFSCORRUPTED;
141	}
142
143	err = jbd2_journal_bmap(journal, offset, &blocknr);
144
145	if (err) {
146		printk(KERN_ERR "JBD2: bad block at offset %u\n",
147			offset);
148		return err;
149	}
150
151	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
152	if (!bh)
153		return -ENOMEM;
154
155	if (!buffer_uptodate(bh)) {
156		/*
157		 * If this is a brand new buffer, start readahead.
158		 * Otherwise, we assume we are already reading it.
159		 */
160		bool need_readahead = !buffer_req(bh);
161
162		bh_read_nowait(bh, 0);
163		if (need_readahead)
164			do_readahead(journal, offset);
165		wait_on_buffer(bh);
166	}
167
168	if (!buffer_uptodate(bh)) {
169		printk(KERN_ERR "JBD2: Failed to read block at offset %u\n",
170			offset);
171		brelse(bh);
172		return -EIO;
173	}
174
175	*bhp = bh;
176	return 0;
177}
178
179static int jbd2_descriptor_block_csum_verify(journal_t *j, void *buf)
180{
181	struct jbd2_journal_block_tail *tail;
182	__be32 provided;
183	__u32 calculated;
184
185	if (!jbd2_journal_has_csum_v2or3(j))
186		return 1;
187
188	tail = (struct jbd2_journal_block_tail *)((char *)buf +
189		j->j_blocksize - sizeof(struct jbd2_journal_block_tail));
190	provided = tail->t_checksum;
191	tail->t_checksum = 0;
192	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
193	tail->t_checksum = provided;
194
195	return provided == cpu_to_be32(calculated);
196}
197
198/*
199 * Count the number of in-use tags in a journal descriptor block.
200 */
201
202static int count_tags(journal_t *journal, struct buffer_head *bh)
203{
204	char *			tagp;
205	journal_block_tag_t	tag;
206	int			nr = 0, size = journal->j_blocksize;
207	int			tag_bytes = journal_tag_bytes(journal);
208
209	if (jbd2_journal_has_csum_v2or3(journal))
210		size -= sizeof(struct jbd2_journal_block_tail);
211
212	tagp = &bh->b_data[sizeof(journal_header_t)];
213
214	while ((tagp - bh->b_data + tag_bytes) <= size) {
215		memcpy(&tag, tagp, sizeof(tag));
216
217		nr++;
218		tagp += tag_bytes;
219		if (!(tag.t_flags & cpu_to_be16(JBD2_FLAG_SAME_UUID)))
220			tagp += 16;
221
222		if (tag.t_flags & cpu_to_be16(JBD2_FLAG_LAST_TAG))
223			break;
224	}
225
226	return nr;
227}
228
229
230/* Make sure we wrap around the log correctly! */
231#define wrap(journal, var)						\
232do {									\
233	if (var >= (journal)->j_last)					\
234		var -= ((journal)->j_last - (journal)->j_first);	\
235} while (0)
236
237static int fc_do_one_pass(journal_t *journal,
238			  struct recovery_info *info, enum passtype pass)
239{
240	unsigned int expected_commit_id = info->end_transaction;
241	unsigned long next_fc_block;
242	struct buffer_head *bh;
243	int err = 0;
244
245	next_fc_block = journal->j_fc_first;
246	if (!journal->j_fc_replay_callback)
247		return 0;
248
249	while (next_fc_block <= journal->j_fc_last) {
250		jbd2_debug(3, "Fast commit replay: next block %ld\n",
251			  next_fc_block);
252		err = jread(&bh, journal, next_fc_block);
253		if (err) {
254			jbd2_debug(3, "Fast commit replay: read error\n");
255			break;
256		}
257
258		err = journal->j_fc_replay_callback(journal, bh, pass,
259					next_fc_block - journal->j_fc_first,
260					expected_commit_id);
261		brelse(bh);
262		next_fc_block++;
263		if (err < 0 || err == JBD2_FC_REPLAY_STOP)
264			break;
265		err = 0;
266	}
267
268	if (err)
269		jbd2_debug(3, "Fast commit replay failed, err = %d\n", err);
270
271	return err;
272}
273
274/**
275 * jbd2_journal_recover - recovers a on-disk journal
276 * @journal: the journal to recover
277 *
278 * The primary function for recovering the log contents when mounting a
279 * journaled device.
280 *
281 * Recovery is done in three passes.  In the first pass, we look for the
282 * end of the log.  In the second, we assemble the list of revoke
283 * blocks.  In the third and final pass, we replay any un-revoked blocks
284 * in the log.
285 */
286int jbd2_journal_recover(journal_t *journal)
287{
288	int			err, err2;
289	journal_superblock_t *	sb;
290
291	struct recovery_info	info;
292
293	memset(&info, 0, sizeof(info));
294	sb = journal->j_superblock;
295
296	/*
297	 * The journal superblock's s_start field (the current log head)
298	 * is always zero if, and only if, the journal was cleanly
299	 * unmounted.
300	 */
 
301	if (!sb->s_start) {
302		jbd2_debug(1, "No recovery required, last transaction %d, head block %u\n",
303			  be32_to_cpu(sb->s_sequence), be32_to_cpu(sb->s_head));
304		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
305		journal->j_head = be32_to_cpu(sb->s_head);
306		return 0;
307	}
308
309	err = do_one_pass(journal, &info, PASS_SCAN);
310	if (!err)
311		err = do_one_pass(journal, &info, PASS_REVOKE);
312	if (!err)
313		err = do_one_pass(journal, &info, PASS_REPLAY);
314
315	jbd2_debug(1, "JBD2: recovery, exit status %d, "
316		  "recovered transactions %u to %u\n",
317		  err, info.start_transaction, info.end_transaction);
318	jbd2_debug(1, "JBD2: Replayed %d and revoked %d/%d blocks\n",
319		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
320
321	/* Restart the log at the next transaction ID, thus invalidating
322	 * any existing commit records in the log. */
323	journal->j_transaction_sequence = ++info.end_transaction;
324	journal->j_head = info.head_block;
325	jbd2_debug(1, "JBD2: last transaction %d, head block %lu\n",
326		  journal->j_transaction_sequence, journal->j_head);
327
328	jbd2_journal_clear_revoke(journal);
329	err2 = sync_blockdev(journal->j_fs_dev);
330	if (!err)
331		err = err2;
332	err2 = jbd2_check_fs_dev_write_error(journal);
333	if (!err)
334		err = err2;
335	/* Make sure all replayed data is on permanent storage */
336	if (journal->j_flags & JBD2_BARRIER) {
337		err2 = blkdev_issue_flush(journal->j_fs_dev);
338		if (!err)
339			err = err2;
340	}
341	return err;
342}
343
344/**
345 * jbd2_journal_skip_recovery - Start journal and wipe exiting records
346 * @journal: journal to startup
347 *
348 * Locate any valid recovery information from the journal and set up the
349 * journal structures in memory to ignore it (presumably because the
350 * caller has evidence that it is out of date).
351 * This function doesn't appear to be exported..
352 *
353 * We perform one pass over the journal to allow us to tell the user how
354 * much recovery information is being erased, and to let us initialise
355 * the journal transaction sequence numbers to the next unused ID.
356 */
357int jbd2_journal_skip_recovery(journal_t *journal)
358{
359	int			err;
360
361	struct recovery_info	info;
362
363	memset (&info, 0, sizeof(info));
364
365	err = do_one_pass(journal, &info, PASS_SCAN);
366
367	if (err) {
368		printk(KERN_ERR "JBD2: error %d scanning journal\n", err);
369		++journal->j_transaction_sequence;
370		journal->j_head = journal->j_first;
371	} else {
372#ifdef CONFIG_JBD2_DEBUG
373		int dropped = info.end_transaction - 
374			be32_to_cpu(journal->j_superblock->s_sequence);
375		jbd2_debug(1,
376			  "JBD2: ignoring %d transaction%s from the journal.\n",
377			  dropped, (dropped == 1) ? "" : "s");
378#endif
379		journal->j_transaction_sequence = ++info.end_transaction;
380		journal->j_head = info.head_block;
381	}
382
383	journal->j_tail = 0;
384	return err;
385}
386
387static inline unsigned long long read_tag_block(journal_t *journal,
388						journal_block_tag_t *tag)
389{
390	unsigned long long block = be32_to_cpu(tag->t_blocknr);
391	if (jbd2_has_feature_64bit(journal))
392		block |= (u64)be32_to_cpu(tag->t_blocknr_high) << 32;
393	return block;
394}
395
396/*
397 * calc_chksums calculates the checksums for the blocks described in the
398 * descriptor block.
399 */
400static int calc_chksums(journal_t *journal, struct buffer_head *bh,
401			unsigned long *next_log_block, __u32 *crc32_sum)
402{
403	int i, num_blks, err;
404	unsigned long io_block;
405	struct buffer_head *obh;
406
407	num_blks = count_tags(journal, bh);
408	/* Calculate checksum of the descriptor block. */
409	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
410
411	for (i = 0; i < num_blks; i++) {
412		io_block = (*next_log_block)++;
413		wrap(journal, *next_log_block);
414		err = jread(&obh, journal, io_block);
415		if (err) {
416			printk(KERN_ERR "JBD2: IO error %d recovering block "
417				"%lu in log\n", err, io_block);
418			return 1;
419		} else {
420			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
421				     obh->b_size);
422		}
423		put_bh(obh);
424	}
425	return 0;
426}
427
428static int jbd2_commit_block_csum_verify(journal_t *j, void *buf)
429{
430	struct commit_header *h;
431	__be32 provided;
432	__u32 calculated;
433
434	if (!jbd2_journal_has_csum_v2or3(j))
435		return 1;
436
437	h = buf;
438	provided = h->h_chksum[0];
439	h->h_chksum[0] = 0;
440	calculated = jbd2_chksum(j, j->j_csum_seed, buf, j->j_blocksize);
441	h->h_chksum[0] = provided;
442
443	return provided == cpu_to_be32(calculated);
444}
445
446static int jbd2_block_tag_csum_verify(journal_t *j, journal_block_tag_t *tag,
447				      journal_block_tag3_t *tag3,
448				      void *buf, __u32 sequence)
449{
450	__u32 csum32;
451	__be32 seq;
452
453	if (!jbd2_journal_has_csum_v2or3(j))
454		return 1;
455
456	seq = cpu_to_be32(sequence);
457	csum32 = jbd2_chksum(j, j->j_csum_seed, (__u8 *)&seq, sizeof(seq));
458	csum32 = jbd2_chksum(j, csum32, buf, j->j_blocksize);
459
460	if (jbd2_has_feature_csum3(j))
461		return tag3->t_checksum == cpu_to_be32(csum32);
462	else
463		return tag->t_checksum == cpu_to_be16(csum32);
464}
465
466static int do_one_pass(journal_t *journal,
467			struct recovery_info *info, enum passtype pass)
468{
469	unsigned int		first_commit_ID, next_commit_ID;
470	unsigned long		next_log_block, head_block;
471	int			err, success = 0;
472	journal_superblock_t *	sb;
473	journal_header_t *	tmp;
474	struct buffer_head *	bh;
475	unsigned int		sequence;
476	int			blocktype;
477	int			tag_bytes = journal_tag_bytes(journal);
478	__u32			crc32_sum = ~0; /* Transactional Checksums */
479	int			descr_csum_size = 0;
480	int			block_error = 0;
481	bool			need_check_commit_time = false;
482	__u64			last_trans_commit_time = 0, commit_time;
483
484	/*
485	 * First thing is to establish what we expect to find in the log
486	 * (in terms of transaction IDs), and where (in terms of log
487	 * block offsets): query the superblock.
488	 */
489
490	sb = journal->j_superblock;
491	next_commit_ID = be32_to_cpu(sb->s_sequence);
492	next_log_block = be32_to_cpu(sb->s_start);
493	head_block = next_log_block;
494
495	first_commit_ID = next_commit_ID;
496	if (pass == PASS_SCAN)
497		info->start_transaction = first_commit_ID;
498
499	jbd2_debug(1, "Starting recovery pass %d\n", pass);
500
501	/*
502	 * Now we walk through the log, transaction by transaction,
503	 * making sure that each transaction has a commit block in the
504	 * expected place.  Each complete transaction gets replayed back
505	 * into the main filesystem.
506	 */
507
508	while (1) {
509		int			flags;
510		char *			tagp;
511		journal_block_tag_t	tag;
512		struct buffer_head *	obh;
513		struct buffer_head *	nbh;
514
515		cond_resched();
516
517		/* If we already know where to stop the log traversal,
518		 * check right now that we haven't gone past the end of
519		 * the log. */
520
521		if (pass != PASS_SCAN)
522			if (tid_geq(next_commit_ID, info->end_transaction))
523				break;
524
525		jbd2_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
526			  next_commit_ID, next_log_block, journal->j_last);
527
528		/* Skip over each chunk of the transaction looking
529		 * either the next descriptor block or the final commit
530		 * record. */
531
532		jbd2_debug(3, "JBD2: checking block %ld\n", next_log_block);
533		err = jread(&bh, journal, next_log_block);
534		if (err)
535			goto failed;
536
537		next_log_block++;
538		wrap(journal, next_log_block);
539
540		/* What kind of buffer is it?
541		 *
542		 * If it is a descriptor block, check that it has the
543		 * expected sequence number.  Otherwise, we're all done
544		 * here. */
545
546		tmp = (journal_header_t *)bh->b_data;
547
548		if (tmp->h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER)) {
549			brelse(bh);
550			break;
551		}
552
553		blocktype = be32_to_cpu(tmp->h_blocktype);
554		sequence = be32_to_cpu(tmp->h_sequence);
555		jbd2_debug(3, "Found magic %d, sequence %d\n",
556			  blocktype, sequence);
557
558		if (sequence != next_commit_ID) {
559			brelse(bh);
560			break;
561		}
562
563		/* OK, we have a valid descriptor block which matches
564		 * all of the sequence number checks.  What are we going
565		 * to do with it?  That depends on the pass... */
566
567		switch(blocktype) {
568		case JBD2_DESCRIPTOR_BLOCK:
569			/* Verify checksum first */
570			if (jbd2_journal_has_csum_v2or3(journal))
571				descr_csum_size =
572					sizeof(struct jbd2_journal_block_tail);
573			if (descr_csum_size > 0 &&
574			    !jbd2_descriptor_block_csum_verify(journal,
575							       bh->b_data)) {
576				/*
577				 * PASS_SCAN can see stale blocks due to lazy
578				 * journal init. Don't error out on those yet.
579				 */
580				if (pass != PASS_SCAN) {
581					pr_err("JBD2: Invalid checksum recovering block %lu in log\n",
582					       next_log_block);
583					err = -EFSBADCRC;
584					brelse(bh);
585					goto failed;
586				}
587				need_check_commit_time = true;
588				jbd2_debug(1,
589					"invalid descriptor block found in %lu\n",
590					next_log_block);
591			}
592
593			/* If it is a valid descriptor block, replay it
594			 * in pass REPLAY; if journal_checksums enabled, then
595			 * calculate checksums in PASS_SCAN, otherwise,
596			 * just skip over the blocks it describes. */
597			if (pass != PASS_REPLAY) {
598				if (pass == PASS_SCAN &&
599				    jbd2_has_feature_checksum(journal) &&
600				    !need_check_commit_time &&
601				    !info->end_transaction) {
602					if (calc_chksums(journal, bh,
603							&next_log_block,
604							&crc32_sum)) {
605						put_bh(bh);
606						break;
607					}
608					put_bh(bh);
609					continue;
610				}
611				next_log_block += count_tags(journal, bh);
612				wrap(journal, next_log_block);
613				put_bh(bh);
614				continue;
615			}
616
617			/* A descriptor block: we can now write all of
618			 * the data blocks.  Yay, useful work is finally
619			 * getting done here! */
620
621			tagp = &bh->b_data[sizeof(journal_header_t)];
622			while ((tagp - bh->b_data + tag_bytes)
623			       <= journal->j_blocksize - descr_csum_size) {
624				unsigned long io_block;
625
626				memcpy(&tag, tagp, sizeof(tag));
627				flags = be16_to_cpu(tag.t_flags);
628
629				io_block = next_log_block++;
630				wrap(journal, next_log_block);
631				err = jread(&obh, journal, io_block);
632				if (err) {
633					/* Recover what we can, but
634					 * report failure at the end. */
635					success = err;
636					printk(KERN_ERR
637						"JBD2: IO error %d recovering "
638						"block %lu in log\n",
639						err, io_block);
640				} else {
641					unsigned long long blocknr;
642
643					J_ASSERT(obh != NULL);
644					blocknr = read_tag_block(journal,
645								 &tag);
646
647					/* If the block has been
648					 * revoked, then we're all done
649					 * here. */
650					if (jbd2_journal_test_revoke
651					    (journal, blocknr,
652					     next_commit_ID)) {
653						brelse(obh);
654						++info->nr_revoke_hits;
655						goto skip_write;
656					}
657
658					/* Look for block corruption */
659					if (!jbd2_block_tag_csum_verify(
660			journal, &tag, (journal_block_tag3_t *)tagp,
661			obh->b_data, be32_to_cpu(tmp->h_sequence))) {
662						brelse(obh);
663						success = -EFSBADCRC;
664						printk(KERN_ERR "JBD2: Invalid "
665						       "checksum recovering "
666						       "data block %llu in "
667						       "journal block %lu\n",
668						       blocknr, io_block);
669						block_error = 1;
670						goto skip_write;
671					}
672
673					/* Find a buffer for the new
674					 * data being restored */
675					nbh = __getblk(journal->j_fs_dev,
676							blocknr,
677							journal->j_blocksize);
678					if (nbh == NULL) {
679						printk(KERN_ERR
680						       "JBD2: Out of memory "
681						       "during recovery.\n");
682						err = -ENOMEM;
683						brelse(bh);
684						brelse(obh);
685						goto failed;
686					}
687
688					lock_buffer(nbh);
689					memcpy(nbh->b_data, obh->b_data,
690							journal->j_blocksize);
691					if (flags & JBD2_FLAG_ESCAPE) {
692						*((__be32 *)nbh->b_data) =
693						cpu_to_be32(JBD2_MAGIC_NUMBER);
694					}
695
696					BUFFER_TRACE(nbh, "marking dirty");
697					set_buffer_uptodate(nbh);
698					mark_buffer_dirty(nbh);
699					BUFFER_TRACE(nbh, "marking uptodate");
700					++info->nr_replays;
 
701					unlock_buffer(nbh);
702					brelse(obh);
703					brelse(nbh);
704				}
705
706			skip_write:
707				tagp += tag_bytes;
708				if (!(flags & JBD2_FLAG_SAME_UUID))
709					tagp += 16;
710
711				if (flags & JBD2_FLAG_LAST_TAG)
712					break;
713			}
714
715			brelse(bh);
716			continue;
717
718		case JBD2_COMMIT_BLOCK:
719			/*     How to differentiate between interrupted commit
720			 *               and journal corruption ?
721			 *
722			 * {nth transaction}
723			 *        Checksum Verification Failed
724			 *			 |
725			 *		 ____________________
726			 *		|		     |
727			 * 	async_commit             sync_commit
728			 *     		|                    |
729			 *		| GO TO NEXT    "Journal Corruption"
730			 *		| TRANSACTION
731			 *		|
732			 * {(n+1)th transanction}
733			 *		|
734			 * 	 _______|______________
735			 * 	|	 	      |
736			 * Commit block found	Commit block not found
737			 *      |		      |
738			 * "Journal Corruption"       |
739			 *		 _____________|_________
740			 *     		|	           	|
741			 *	nth trans corrupt	OR   nth trans
742			 *	and (n+1)th interrupted     interrupted
743			 *	before commit block
744			 *      could reach the disk.
745			 *	(Cannot find the difference in above
746			 *	 mentioned conditions. Hence assume
747			 *	 "Interrupted Commit".)
748			 */
749			commit_time = be64_to_cpu(
750				((struct commit_header *)bh->b_data)->h_commit_sec);
751			/*
752			 * If need_check_commit_time is set, it means we are in
753			 * PASS_SCAN and csum verify failed before. If
754			 * commit_time is increasing, it's the same journal,
755			 * otherwise it is stale journal block, just end this
756			 * recovery.
757			 */
758			if (need_check_commit_time) {
759				if (commit_time >= last_trans_commit_time) {
760					pr_err("JBD2: Invalid checksum found in transaction %u\n",
761					       next_commit_ID);
762					err = -EFSBADCRC;
763					brelse(bh);
764					goto failed;
765				}
766			ignore_crc_mismatch:
767				/*
768				 * It likely does not belong to same journal,
769				 * just end this recovery with success.
770				 */
771				jbd2_debug(1, "JBD2: Invalid checksum ignored in transaction %u, likely stale data\n",
772					  next_commit_ID);
773				brelse(bh);
774				goto done;
775			}
776
777			/*
778			 * Found an expected commit block: if checksums
779			 * are present, verify them in PASS_SCAN; else not
780			 * much to do other than move on to the next sequence
781			 * number.
782			 */
783			if (pass == PASS_SCAN &&
784			    jbd2_has_feature_checksum(journal)) {
 
 
785				struct commit_header *cbh =
786					(struct commit_header *)bh->b_data;
787				unsigned found_chksum =
788					be32_to_cpu(cbh->h_chksum[0]);
789
 
 
790				if (info->end_transaction) {
791					journal->j_failed_commit =
792						info->end_transaction;
793					brelse(bh);
794					break;
795				}
796
797				/* Neither checksum match nor unused? */
798				if (!((crc32_sum == found_chksum &&
799				       cbh->h_chksum_type ==
800						JBD2_CRC32_CHKSUM &&
801				       cbh->h_chksum_size ==
802						JBD2_CRC32_CHKSUM_SIZE) ||
803				      (cbh->h_chksum_type == 0 &&
804				       cbh->h_chksum_size == 0 &&
805				       found_chksum == 0)))
806					goto chksum_error;
 
 
 
 
 
 
 
 
 
 
807
808				crc32_sum = ~0;
809			}
810			if (pass == PASS_SCAN &&
811			    !jbd2_commit_block_csum_verify(journal,
812							   bh->b_data)) {
813			chksum_error:
814				if (commit_time < last_trans_commit_time)
815					goto ignore_crc_mismatch;
816				info->end_transaction = next_commit_ID;
817				info->head_block = head_block;
818
819				if (!jbd2_has_feature_async_commit(journal)) {
820					journal->j_failed_commit =
821						next_commit_ID;
822					brelse(bh);
823					break;
 
 
824				}
825			}
826			if (pass == PASS_SCAN) {
827				last_trans_commit_time = commit_time;
828				head_block = next_log_block;
829			}
830			brelse(bh);
831			next_commit_ID++;
832			continue;
833
834		case JBD2_REVOKE_BLOCK:
835			/*
836			 * Check revoke block crc in pass_scan, if csum verify
837			 * failed, check commit block time later.
838			 */
839			if (pass == PASS_SCAN &&
840			    !jbd2_descriptor_block_csum_verify(journal,
841							       bh->b_data)) {
842				jbd2_debug(1, "JBD2: invalid revoke block found in %lu\n",
843					  next_log_block);
844				need_check_commit_time = true;
845			}
846			/* If we aren't in the REVOKE pass, then we can
847			 * just skip over this block. */
848			if (pass != PASS_REVOKE) {
849				brelse(bh);
850				continue;
851			}
852
853			err = scan_revoke_records(journal, bh,
854						  next_commit_ID, info);
855			brelse(bh);
856			if (err)
857				goto failed;
858			continue;
859
860		default:
861			jbd2_debug(3, "Unrecognised magic %d, end of scan.\n",
862				  blocktype);
863			brelse(bh);
864			goto done;
865		}
866	}
867
868 done:
869	/*
870	 * We broke out of the log scan loop: either we came to the
871	 * known end of the log or we found an unexpected block in the
872	 * log.  If the latter happened, then we know that the "current"
873	 * transaction marks the end of the valid log.
874	 */
875
876	if (pass == PASS_SCAN) {
877		if (!info->end_transaction)
878			info->end_transaction = next_commit_ID;
879		if (!info->head_block)
880			info->head_block = head_block;
881	} else {
882		/* It's really bad news if different passes end up at
883		 * different places (but possible due to IO errors). */
884		if (info->end_transaction != next_commit_ID) {
885			printk(KERN_ERR "JBD2: recovery pass %d ended at "
886				"transaction %u, expected %u\n",
887				pass, next_commit_ID, info->end_transaction);
888			if (!success)
889				success = -EIO;
890		}
891	}
892
893	if (jbd2_has_feature_fast_commit(journal) &&  pass != PASS_REVOKE) {
894		err = fc_do_one_pass(journal, info, pass);
895		if (err)
896			success = err;
897	}
898
899	if (block_error && success == 0)
900		success = -EIO;
901	return success;
902
903 failed:
904	return err;
905}
906
 
907/* Scan a revoke record, marking all blocks mentioned as revoked. */
908
909static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
910			       tid_t sequence, struct recovery_info *info)
911{
912	jbd2_journal_revoke_header_t *header;
913	int offset, max;
914	unsigned csum_size = 0;
915	__u32 rcount;
916	int record_len = 4;
917
918	header = (jbd2_journal_revoke_header_t *) bh->b_data;
919	offset = sizeof(jbd2_journal_revoke_header_t);
920	rcount = be32_to_cpu(header->r_count);
921
922	if (jbd2_journal_has_csum_v2or3(journal))
923		csum_size = sizeof(struct jbd2_journal_block_tail);
924	if (rcount > journal->j_blocksize - csum_size)
925		return -EINVAL;
926	max = rcount;
927
928	if (jbd2_has_feature_64bit(journal))
929		record_len = 8;
930
931	while (offset + record_len <= max) {
932		unsigned long long blocknr;
933		int err;
934
935		if (record_len == 4)
936			blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
937		else
938			blocknr = be64_to_cpu(* ((__be64 *) (bh->b_data+offset)));
939		offset += record_len;
940		err = jbd2_journal_set_revoke(journal, blocknr, sequence);
941		if (err)
942			return err;
943		++info->nr_revokes;
944	}
945	return 0;
946}