[linux.git] / fs / jbd / recovery.c

/*
 * linux/fs/jbd/recovery.c
 *
 * Written by Stephen C. Tweedie <[email protected]>, 1999
 *
 * Copyright 1999-2000 Red Hat Software --- All Rights Reserved
 *
 * This file is part of the Linux kernel and is made available under
 * the terms of the GNU General Public License, version 2, or at your
 * option, any later version, incorporated herein by reference.
 *
 * Journal recovery routines for the generic filesystem journaling code;
 * part of the ext2fs journaling system.
 */

#ifndef __KERNEL__
#include "jfs_user.h"
#else
#include <linux/time.h>
#include <linux/fs.h>
#include <linux/jbd.h>
#include <linux/errno.h>
#endif

/*
 * Maintain information about the progress of the recovery job, so that
 * the different passes can carry information between them.
 */
struct recovery_info
{
	tid_t		start_transaction;
	tid_t		end_transaction;

	int		nr_replays;
	int		nr_revokes;
	int		nr_revoke_hits;
};

enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
static int do_one_pass(journal_t *journal,
				struct recovery_info *info, enum passtype pass);
static int scan_revoke_records(journal_t *, struct buffer_head *,
				tid_t, struct recovery_info *);

#ifdef __KERNEL__

/* Release readahead buffers after use */
static void journal_brelse_array(struct buffer_head *b[], int n)
{
	while (--n >= 0)
		brelse (b[n]);
}


/*
 * When reading from the journal, we are going through the block device
 * layer directly and so there is no readahead being done for us.  We
 * need to implement any readahead ourselves if we want it to happen at
 * all.  Recovery is basically one long sequential read, so make sure we
 * do the IO in reasonably large chunks.
 *
 * This is not so critical that we need to be enormously clever about
 * the readahead size, though.  128K is a purely arbitrary, good-enough
 * fixed value.
 */

#define MAXBUF 8
static int do_readahead(journal_t *journal, unsigned int start)
{
	int err;
	unsigned int max, nbufs, next;
	unsigned int blocknr;
	struct buffer_head *bh;

	struct buffer_head * bufs[MAXBUF];

	/* Do up to 128K of readahead */
	max = start + (128 * 1024 / journal->j_blocksize);
	if (max > journal->j_maxlen)
		max = journal->j_maxlen;

	/* Do the readahead itself.  We'll submit MAXBUF buffer_heads at
	 * a time to the block device IO layer. */

	nbufs = 0;

	for (next = start; next < max; next++) {
		err = journal_bmap(journal, next, &blocknr);

		if (err) {
			printk (KERN_ERR "JBD: bad block at offset %u\n",
				next);
			goto failed;
		}

		bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
		if (!bh) {
			err = -ENOMEM;
			goto failed;
		}

		if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
			bufs[nbufs++] = bh;
			if (nbufs == MAXBUF) {
				ll_rw_block(READ, nbufs, bufs);
				journal_brelse_array(bufs, nbufs);
				nbufs = 0;
			}
		} else
			brelse(bh);
	}

	if (nbufs)
		ll_rw_block(READ, nbufs, bufs);
	err = 0;

failed:
	if (nbufs)
		journal_brelse_array(bufs, nbufs);
	return err;
}

#endif /* __KERNEL__ */


/*
 * Read a block from the journal
 */

static int jread(struct buffer_head **bhp, journal_t *journal,
		 unsigned int offset)
{
	int err;
	unsigned int blocknr;
	struct buffer_head *bh;

	*bhp = NULL;

	if (offset >= journal->j_maxlen) {
		printk(KERN_ERR "JBD: corrupted journal superblock\n");
		return -EIO;
	}

	err = journal_bmap(journal, offset, &blocknr);

	if (err) {
		printk (KERN_ERR "JBD: bad block at offset %u\n",
			offset);
		return err;
	}

	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
	if (!bh)
		return -ENOMEM;

	if (!buffer_uptodate(bh)) {
		/* If this is a brand new buffer, start readahead.
                   Otherwise, we assume we are already reading it.  */
		if (!buffer_req(bh))
			do_readahead(journal, offset);
		wait_on_buffer(bh);
	}

	if (!buffer_uptodate(bh)) {
		printk (KERN_ERR "JBD: Failed to read block at offset %u\n",
			offset);
		brelse(bh);
		return -EIO;
	}

	*bhp = bh;
	return 0;
}


/*
 * Count the number of in-use tags in a journal descriptor block.
 */

static int count_tags(struct buffer_head *bh, int size)
{
	char *			tagp;
	journal_block_tag_t *	tag;
	int			nr = 0;

	tagp = &bh->b_data[sizeof(journal_header_t)];

	while ((tagp - bh->b_data + sizeof(journal_block_tag_t)) <= size) {
		tag = (journal_block_tag_t *) tagp;

		nr++;
		tagp += sizeof(journal_block_tag_t);
		if (!(tag->t_flags & cpu_to_be32(JFS_FLAG_SAME_UUID)))
			tagp += 16;

		if (tag->t_flags & cpu_to_be32(JFS_FLAG_LAST_TAG))
			break;
	}

	return nr;
}


/* Make sure we wrap around the log correctly! */
#define wrap(journal, var)						\
do {									\
	if (var >= (journal)->j_last)					\
		var -= ((journal)->j_last - (journal)->j_first);	\
} while (0)

/**
 * journal_recover - recovers a on-disk journal
 * @journal: the journal to recover
 *
 * The primary function for recovering the log contents when mounting a
 * journaled device.
 *
 * Recovery is done in three passes.  In the first pass, we look for the
 * end of the log.  In the second, we assemble the list of revoke
 * blocks.  In the third and final pass, we replay any un-revoked blocks
 * in the log.
 */
int journal_recover(journal_t *journal)
{
	int			err, err2;
	journal_superblock_t *	sb;

	struct recovery_info	info;

	memset(&info, 0, sizeof(info));
	sb = journal->j_superblock;

	/*
	 * The journal superblock's s_start field (the current log head)
	 * is always zero if, and only if, the journal was cleanly
	 * unmounted.
	 */

	if (!sb->s_start) {
		jbd_debug(1, "No recovery required, last transaction %d\n",
			  be32_to_cpu(sb->s_sequence));
		journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
		return 0;
	}

	err = do_one_pass(journal, &info, PASS_SCAN);
	if (!err)
		err = do_one_pass(journal, &info, PASS_REVOKE);
	if (!err)
		err = do_one_pass(journal, &info, PASS_REPLAY);

	jbd_debug(1, "JBD: recovery, exit status %d, "
		  "recovered transactions %u to %u\n",
		  err, info.start_transaction, info.end_transaction);
	jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n",
		  info.nr_replays, info.nr_revoke_hits, info.nr_revokes);

	/* Restart the log at the next transaction ID, thus invalidating
	 * any existing commit records in the log. */
	journal->j_transaction_sequence = ++info.end_transaction;

	journal_clear_revoke(journal);
	err2 = sync_blockdev(journal->j_fs_dev);
	if (!err)
		err = err2;

	return err;
}

/**
 * journal_skip_recovery - Start journal and wipe exiting records
 * @journal: journal to startup
 *
 * Locate any valid recovery information from the journal and set up the
 * journal structures in memory to ignore it (presumably because the
 * caller has evidence that it is out of date).
 * This function does'nt appear to be exorted..
 *
 * We perform one pass over the journal to allow us to tell the user how
 * much recovery information is being erased, and to let us initialise
 * the journal transaction sequence numbers to the next unused ID.
 */
int journal_skip_recovery(journal_t *journal)
{
	int			err;
	struct recovery_info	info;

	memset (&info, 0, sizeof(info));

	err = do_one_pass(journal, &info, PASS_SCAN);

	if (err) {
		printk(KERN_ERR "JBD: error %d scanning journal\n", err);
		++journal->j_transaction_sequence;
	} else {
#ifdef CONFIG_JBD_DEBUG
		int dropped = info.end_transaction -
			      be32_to_cpu(journal->j_superblock->s_sequence);
		jbd_debug(1,
			  "JBD: ignoring %d transaction%s from the journal.\n",
			  dropped, (dropped == 1) ? "" : "s");
#endif
		journal->j_transaction_sequence = ++info.end_transaction;
	}

	journal->j_tail = 0;
	return err;
}

static int do_one_pass(journal_t *journal,
			struct recovery_info *info, enum passtype pass)
{
	unsigned int		first_commit_ID, next_commit_ID;
	unsigned int		next_log_block;
	int			err, success = 0;
	journal_superblock_t *	sb;
	journal_header_t *	tmp;
	struct buffer_head *	bh;
	unsigned int		sequence;
	int			blocktype;

	/*
	 * First thing is to establish what we expect to find in the log
	 * (in terms of transaction IDs), and where (in terms of log
	 * block offsets): query the superblock.
	 */

	sb = journal->j_superblock;
	next_commit_ID = be32_to_cpu(sb->s_sequence);
	next_log_block = be32_to_cpu(sb->s_start);

	first_commit_ID = next_commit_ID;
	if (pass == PASS_SCAN)
		info->start_transaction = first_commit_ID;

	jbd_debug(1, "Starting recovery pass %d\n", pass);

	/*
	 * Now we walk through the log, transaction by transaction,
	 * making sure that each transaction has a commit block in the
	 * expected place.  Each complete transaction gets replayed back
	 * into the main filesystem.
	 */

	while (1) {
		int			flags;
		char *			tagp;
		journal_block_tag_t *	tag;
		struct buffer_head *	obh;
		struct buffer_head *	nbh;

		cond_resched();

		/* If we already know where to stop the log traversal,
		 * check right now that we haven't gone past the end of
		 * the log. */

		if (pass != PASS_SCAN)
			if (tid_geq(next_commit_ID, info->end_transaction))
				break;

		jbd_debug(2, "Scanning for sequence ID %u at %u/%u\n",
			  next_commit_ID, next_log_block, journal->j_last);

		/* Skip over each chunk of the transaction looking
		 * either the next descriptor block or the final commit
		 * record. */

		jbd_debug(3, "JBD: checking block %u\n", next_log_block);
		err = jread(&bh, journal, next_log_block);
		if (err)
			goto failed;

		next_log_block++;
		wrap(journal, next_log_block);

		/* What kind of buffer is it?
		 *
		 * If it is a descriptor block, check that it has the
		 * expected sequence number.  Otherwise, we're all done
		 * here. */

		tmp = (journal_header_t *)bh->b_data;

		if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) {
			brelse(bh);
			break;
		}

		blocktype = be32_to_cpu(tmp->h_blocktype);
		sequence = be32_to_cpu(tmp->h_sequence);
		jbd_debug(3, "Found magic %d, sequence %d\n",
			  blocktype, sequence);

		if (sequence != next_commit_ID) {
			brelse(bh);
			break;
		}

		/* OK, we have a valid descriptor block which matches
		 * all of the sequence number checks.  What are we going
		 * to do with it?  That depends on the pass... */

		switch(blocktype) {
		case JFS_DESCRIPTOR_BLOCK:
			/* If it is a valid descriptor block, replay it
			 * in pass REPLAY; otherwise, just skip over the
			 * blocks it describes. */
			if (pass != PASS_REPLAY) {
				next_log_block +=
					count_tags(bh, journal->j_blocksize);
				wrap(journal, next_log_block);
				brelse(bh);
				continue;
			}

			/* A descriptor block: we can now write all of
			 * the data blocks.  Yay, useful work is finally
			 * getting done here! */

			tagp = &bh->b_data[sizeof(journal_header_t)];
			while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
			       <= journal->j_blocksize) {
				unsigned int io_block;

				tag = (journal_block_tag_t *) tagp;
				flags = be32_to_cpu(tag->t_flags);

				io_block = next_log_block++;
				wrap(journal, next_log_block);
				err = jread(&obh, journal, io_block);
				if (err) {
					/* Recover what we can, but
					 * report failure at the end. */
					success = err;
					printk (KERN_ERR
						"JBD: IO error %d recovering "
						"block %u in log\n",
						err, io_block);
				} else {
					unsigned int blocknr;

					J_ASSERT(obh != NULL);
					blocknr = be32_to_cpu(tag->t_blocknr);

					/* If the block has been
					 * revoked, then we're all done
					 * here. */
					if (journal_test_revoke
					    (journal, blocknr,
					     next_commit_ID)) {
						brelse(obh);
						++info->nr_revoke_hits;
						goto skip_write;
					}

					/* Find a buffer for the new
					 * data being restored */
					nbh = __getblk(journal->j_fs_dev,
							blocknr,
							journal->j_blocksize);
					if (nbh == NULL) {
						printk(KERN_ERR
						       "JBD: Out of memory "
						       "during recovery.\n");
						err = -ENOMEM;
						brelse(bh);
						brelse(obh);
						goto failed;
					}

					lock_buffer(nbh);
					memcpy(nbh->b_data, obh->b_data,
							journal->j_blocksize);
					if (flags & JFS_FLAG_ESCAPE) {
						*((__be32 *)nbh->b_data) =
						cpu_to_be32(JFS_MAGIC_NUMBER);
					}

					BUFFER_TRACE(nbh, "marking dirty");
					set_buffer_uptodate(nbh);
					mark_buffer_dirty(nbh);
					BUFFER_TRACE(nbh, "marking uptodate");
					++info->nr_replays;
					/* ll_rw_block(WRITE, 1, &nbh); */
					unlock_buffer(nbh);
					brelse(obh);
					brelse(nbh);
				}

			skip_write:
				tagp += sizeof(journal_block_tag_t);
				if (!(flags & JFS_FLAG_SAME_UUID))
					tagp += 16;

				if (flags & JFS_FLAG_LAST_TAG)
					break;
			}

			brelse(bh);
			continue;

		case JFS_COMMIT_BLOCK:
			/* Found an expected commit block: not much to
			 * do other than move on to the next sequence
			 * number. */
			brelse(bh);
			next_commit_ID++;
			continue;

		case JFS_REVOKE_BLOCK:
			/* If we aren't in the REVOKE pass, then we can
			 * just skip over this block. */
			if (pass != PASS_REVOKE) {
				brelse(bh);
				continue;
			}

			err = scan_revoke_records(journal, bh,
						  next_commit_ID, info);
			brelse(bh);
			if (err)
				goto failed;
			continue;

		default:
			jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
				  blocktype);
			brelse(bh);
			goto done;
		}
	}

 done:
	/*
	 * We broke out of the log scan loop: either we came to the
	 * known end of the log or we found an unexpected block in the
	 * log.  If the latter happened, then we know that the "current"
	 * transaction marks the end of the valid log.
	 */

	if (pass == PASS_SCAN)
		info->end_transaction = next_commit_ID;
	else {
		/* It's really bad news if different passes end up at
		 * different places (but possible due to IO errors). */
		if (info->end_transaction != next_commit_ID) {
			printk (KERN_ERR "JBD: recovery pass %d ended at "
				"transaction %u, expected %u\n",
				pass, next_commit_ID, info->end_transaction);
			if (!success)
				success = -EIO;
		}
	}

	return success;

 failed:
	return err;
}


/* Scan a revoke record, marking all blocks mentioned as revoked. */

static int scan_revoke_records(journal_t *journal, struct buffer_head *bh,
			       tid_t sequence, struct recovery_info *info)
{
	journal_revoke_header_t *header;
	int offset, max;

	header = (journal_revoke_header_t *) bh->b_data;
	offset = sizeof(journal_revoke_header_t);
	max = be32_to_cpu(header->r_count);

	while (offset < max) {
		unsigned int blocknr;
		int err;

		blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
		offset += 4;
		err = journal_set_revoke(journal, blocknr, sequence);
		if (err)
			return err;
		++info->nr_revokes;
	}
	return 0;
}
Commit	Line	Data
1da177e4	1	/*
58862699	2	* linux/fs/jbd/recovery.c
ae6ddcc5	3	*
1da177e4 LT	4	* Written by Stephen C. Tweedie <[email protected]>, 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;
ae6ddcc5	13	* part of the ext2fs journaling system.
1da177e4 LT	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/jbd.h>
	22	#include <linux/errno.h>
1da177e4 LT	23	#endif
	24
	25	/*
	26	* Maintain information about the progress of the recovery job, so that
ae6ddcc5	27	* the different passes can carry information between them.
1da177e4	28	*/
ae6ddcc5	29	struct recovery_info
1da177e4 LT	30	{
	31	tid_t start_transaction;
	32	tid_t end_transaction;
	33
	34	int nr_replays;
	35	int nr_revokes;
	36	int nr_revoke_hits;
	37	};
	38
	39	enum passtype {PASS_SCAN, PASS_REVOKE, PASS_REPLAY};
	40	static int do_one_pass(journal_t *journal,
	41	struct recovery_info *info, enum passtype pass);
	42	static int scan_revoke_records(journal_t , struct buffer_head ,
	43	tid_t, struct recovery_info *);
	44
	45	#ifdef __KERNEL__
	46
	47	/* Release readahead buffers after use */
f71b2f10	48	static void journal_brelse_array(struct buffer_head *b[], int n)
1da177e4 LT	49	{
	50	while (--n >= 0)
	51	brelse (b[n]);
	52	}
	53
	54
	55	/*
	56	* When reading from the journal, we are going through the block device
	57	* layer directly and so there is no readahead being done for us. We
	58	* need to implement any readahead ourselves if we want it to happen at
	59	* all. Recovery is basically one long sequential read, so make sure we
	60	* do the IO in reasonably large chunks.
	61	*
	62	* This is not so critical that we need to be enormously clever about
	63	* the readahead size, though. 128K is a purely arbitrary, good-enough
	64	* fixed value.
	65	*/
	66
	67	#define MAXBUF 8
	68	static int do_readahead(journal_t *journal, unsigned int start)
	69	{
	70	int err;
	71	unsigned int max, nbufs, next;
9c28cbcc	72	unsigned int blocknr;
1da177e4 LT	73	struct buffer_head *bh;
	74
	75	struct buffer_head * bufs[MAXBUF];
	76
	77	/* Do up to 128K of readahead */
	78	max = start + (128 * 1024 / journal->j_blocksize);
	79	if (max > journal->j_maxlen)
	80	max = journal->j_maxlen;
	81
	82	/* Do the readahead itself. We'll submit MAXBUF buffer_heads at
	83	* a time to the block device IO layer. */
	84
	85	nbufs = 0;
	86
	87	for (next = start; next < max; next++) {
	88	err = journal_bmap(journal, next, &blocknr);
	89
	90	if (err) {
	91	printk (KERN_ERR "JBD: bad block at offset %u\n",
	92	next);
	93	goto failed;
	94	}
	95
	96	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
	97	if (!bh) {
	98	err = -ENOMEM;
	99	goto failed;
	100	}
	101
	102	if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
	103	bufs[nbufs++] = bh;
	104	if (nbufs == MAXBUF) {
	105	ll_rw_block(READ, nbufs, bufs);
	106	journal_brelse_array(bufs, nbufs);
	107	nbufs = 0;
	108	}
	109	} else
	110	brelse(bh);
	111	}
	112
	113	if (nbufs)
	114	ll_rw_block(READ, nbufs, bufs);
	115	err = 0;
	116
	117	failed:
ae6ddcc5	118	if (nbufs)
1da177e4 LT	119	journal_brelse_array(bufs, nbufs);
	120	return err;
	121	}
	122
	123	#endif /* __KERNEL__ */
	124
	125
	126	/*
	127	* Read a block from the journal
	128	*/
	129
ae6ddcc5	130	static int jread(struct buffer_head *bhp, journal_t journal,
1da177e4 LT	131	unsigned int offset)
	132	{
	133	int err;
9c28cbcc	134	unsigned int blocknr;
1da177e4 LT	135	struct buffer_head *bh;
	136
	137	*bhp = NULL;
	138
	139	if (offset >= journal->j_maxlen) {
	140	printk(KERN_ERR "JBD: corrupted journal superblock\n");
	141	return -EIO;
	142	}
	143
	144	err = journal_bmap(journal, offset, &blocknr);
	145
	146	if (err) {
	147	printk (KERN_ERR "JBD: bad block at offset %u\n",
	148	offset);
	149	return err;
	150	}
	151
	152	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
	153	if (!bh)
	154	return -ENOMEM;
	155
	156	if (!buffer_uptodate(bh)) {
	157	/* If this is a brand new buffer, start readahead.
	158	Otherwise, we assume we are already reading it. */
	159	if (!buffer_req(bh))
	160	do_readahead(journal, offset);
	161	wait_on_buffer(bh);
	162	}
	163
	164	if (!buffer_uptodate(bh)) {
	165	printk (KERN_ERR "JBD: Failed to read block at offset %u\n",
	166	offset);
	167	brelse(bh);
	168	return -EIO;
	169	}
	170
	171	*bhp = bh;
	172	return 0;
	173	}
	174
	175
	176	/*
	177	* Count the number of in-use tags in a journal descriptor block.
	178	*/
	179
	180	static int count_tags(struct buffer_head *bh, int size)
	181	{
	182	char * tagp;
	183	journal_block_tag_t * tag;
	184	int nr = 0;
	185
	186	tagp = &bh->b_data[sizeof(journal_header_t)];
	187
	188	while ((tagp - bh->b_data + sizeof(journal_block_tag_t)) <= size) {
	189	tag = (journal_block_tag_t *) tagp;
	190
	191	nr++;
	192	tagp += sizeof(journal_block_tag_t);
	193	if (!(tag->t_flags & cpu_to_be32(JFS_FLAG_SAME_UUID)))
	194	tagp += 16;
	195
	196	if (tag->t_flags & cpu_to_be32(JFS_FLAG_LAST_TAG))
	197	break;
	198	}
199
200	return nr;
201	}
202
203
204	/* Make sure we wrap around the log correctly! */
205	#define wrap(journal, var) \
206	do { \
207	if (var >= (journal)->j_last) \
208	var -= ((journal)->j_last - (journal)->j_first); \
209	} while (0)
210
211	/**
6c8bec6d	212	* journal_recover - recovers a on-disk journal
1da177e4	213	* @journal: the journal to recover
ae6ddcc5	214	*
1da177e4	215	* The primary function for recovering the log contents when mounting a
ae6ddcc5	216	* journaled device.
1da177e4 LT	217	*
	218	* Recovery is done in three passes. In the first pass, we look for the
	219	* end of the log. In the second, we assemble the list of revoke
	220	* blocks. In the third and final pass, we replay any un-revoked blocks
ae6ddcc5	221	* in the log.
1da177e4 LT	222	*/
	223	int journal_recover(journal_t *journal)
	224	{
4afe9785	225	int err, err2;
1da177e4 LT	226	journal_superblock_t * sb;
	227
	228	struct recovery_info info;
	229
	230	memset(&info, 0, sizeof(info));
	231	sb = journal->j_superblock;
	232
ae6ddcc5	233	/*
1da177e4 LT	234	* The journal superblock's s_start field (the current log head)
1da177e4 LT	235	* is always zero if, and only if, the journal was cleanly
ae6ddcc5	236	* unmounted.
1da177e4 LT	237	*/
	238
	239	if (!sb->s_start) {
	240	jbd_debug(1, "No recovery required, last transaction %d\n",
	241	be32_to_cpu(sb->s_sequence));
	242	journal->j_transaction_sequence = be32_to_cpu(sb->s_sequence) + 1;
	243	return 0;
	244	}
	245
	246	err = do_one_pass(journal, &info, PASS_SCAN);
	247	if (!err)
	248	err = do_one_pass(journal, &info, PASS_REVOKE);
	249	if (!err)
	250	err = do_one_pass(journal, &info, PASS_REPLAY);
	251
9ad163ae	252	jbd_debug(1, "JBD: recovery, exit status %d, "
1da177e4 LT	253	"recovered transactions %u to %u\n",
1da177e4 LT	254	err, info.start_transaction, info.end_transaction);
9ad163ae	255	jbd_debug(1, "JBD: Replayed %d and revoked %d/%d blocks\n",
1da177e4 LT	256	info.nr_replays, info.nr_revoke_hits, info.nr_revokes);
	257
	258	/* Restart the log at the next transaction ID, thus invalidating
	259	* any existing commit records in the log. */
	260	journal->j_transaction_sequence = ++info.end_transaction;
	261
	262	journal_clear_revoke(journal);
4afe9785 HK	263	err2 = sync_blockdev(journal->j_fs_dev);
	264	if (!err)
	265	err = err2;
	266
1da177e4 LT	267	return err;
	268	}
	269
	270	/**
6c8bec6d	271	* journal_skip_recovery - Start journal and wipe exiting records
1da177e4	272	* @journal: journal to startup
ae6ddcc5	273	*
1da177e4 LT	274	* Locate any valid recovery information from the journal and set up the
1da177e4 LT	275	* journal structures in memory to ignore it (presumably because the
ae6ddcc5	276	* caller has evidence that it is out of date).
1da177e4 LT	277	* This function does'nt appear to be exorted..
	278	*
	279	* We perform one pass over the journal to allow us to tell the user how
	280	* much recovery information is being erased, and to let us initialise
ae6ddcc5	281	* the journal transaction sequence numbers to the next unused ID.
1da177e4 LT	282	*/
	283	int journal_skip_recovery(journal_t *journal)
	284	{
	285	int err;
1da177e4 LT	286	struct recovery_info info;
	287
	288	memset (&info, 0, sizeof(info));
1da177e4 LT	289
	290	err = do_one_pass(journal, &info, PASS_SCAN);
	291
	292	if (err) {
	293	printk(KERN_ERR "JBD: error %d scanning journal\n", err);
	294	++journal->j_transaction_sequence;
	295	} else {
	296	#ifdef CONFIG_JBD_DEBUG
0411ba79 AK	297	int dropped = info.end_transaction -
0411ba79 AK	298	be32_to_cpu(journal->j_superblock->s_sequence);
9ad163ae	299	jbd_debug(1,
1da177e4 LT	300	"JBD: ignoring %d transaction%s from the journal.\n",
1da177e4 LT	301	dropped, (dropped == 1) ? "" : "s");
c5639bef	302	#endif
1da177e4 LT	303	journal->j_transaction_sequence = ++info.end_transaction;
	304	}
	305
	306	journal->j_tail = 0;
	307	return err;
	308	}
	309
	310	static int do_one_pass(journal_t *journal,
	311	struct recovery_info *info, enum passtype pass)
	312	{
	313	unsigned int first_commit_ID, next_commit_ID;
9c28cbcc	314	unsigned int next_log_block;
1da177e4 LT	315	int err, success = 0;
1da177e4 LT	316	journal_superblock_t * sb;
e9ad5620	317	journal_header_t * tmp;
1da177e4 LT	318	struct buffer_head * bh;
	319	unsigned int sequence;
	320	int blocktype;
	321
ae6ddcc5	322	/*
1da177e4 LT	323	* First thing is to establish what we expect to find in the log
1da177e4 LT	324	* (in terms of transaction IDs), and where (in terms of log
ae6ddcc5	325	* block offsets): query the superblock.
1da177e4 LT	326	*/
	327
	328	sb = journal->j_superblock;
	329	next_commit_ID = be32_to_cpu(sb->s_sequence);
	330	next_log_block = be32_to_cpu(sb->s_start);
	331
	332	first_commit_ID = next_commit_ID;
	333	if (pass == PASS_SCAN)
	334	info->start_transaction = first_commit_ID;
	335
	336	jbd_debug(1, "Starting recovery pass %d\n", pass);
	337
	338	/*
	339	* Now we walk through the log, transaction by transaction,
	340	* making sure that each transaction has a commit block in the
	341	* expected place. Each complete transaction gets replayed back
ae6ddcc5	342	* into the main filesystem.
1da177e4 LT	343	*/
	344
	345	while (1) {
	346	int flags;
	347	char * tagp;
	348	journal_block_tag_t * tag;
	349	struct buffer_head * obh;
	350	struct buffer_head * nbh;
	351
e86e1438	352	cond_resched();
1da177e4 LT	353
	354	/* If we already know where to stop the log traversal,
	355	* check right now that we haven't gone past the end of
	356	* the log. */
	357
	358	if (pass != PASS_SCAN)
	359	if (tid_geq(next_commit_ID, info->end_transaction))
	360	break;
	361
9c28cbcc	362	jbd_debug(2, "Scanning for sequence ID %u at %u/%u\n",
1da177e4 LT	363	next_commit_ID, next_log_block, journal->j_last);
	364
	365	/* Skip over each chunk of the transaction looking
	366	* either the next descriptor block or the final commit
	367	* record. */
	368
9c28cbcc	369	jbd_debug(3, "JBD: checking block %u\n", next_log_block);
1da177e4 LT	370	err = jread(&bh, journal, next_log_block);
	371	if (err)
	372	goto failed;
	373
	374	next_log_block++;
	375	wrap(journal, next_log_block);
	376
ae6ddcc5 MC	377	/* What kind of buffer is it?
ae6ddcc5 MC	378	*
1da177e4 LT	379	* If it is a descriptor block, check that it has the
	380	* expected sequence number. Otherwise, we're all done
	381	* here. */
	382
	383	tmp = (journal_header_t *)bh->b_data;
	384
	385	if (tmp->h_magic != cpu_to_be32(JFS_MAGIC_NUMBER)) {
	386	brelse(bh);
	387	break;
	388	}
	389
	390	blocktype = be32_to_cpu(tmp->h_blocktype);
	391	sequence = be32_to_cpu(tmp->h_sequence);
ae6ddcc5	392	jbd_debug(3, "Found magic %d, sequence %d\n",
1da177e4 LT	393	blocktype, sequence);
	394
	395	if (sequence != next_commit_ID) {
	396	brelse(bh);
	397	break;
	398	}
	399
	400	/* OK, we have a valid descriptor block which matches
	401	* all of the sequence number checks. What are we going
	402	* to do with it? That depends on the pass... */
	403
	404	switch(blocktype) {
	405	case JFS_DESCRIPTOR_BLOCK:
	406	/* If it is a valid descriptor block, replay it
	407	* in pass REPLAY; otherwise, just skip over the
	408	* blocks it describes. */
	409	if (pass != PASS_REPLAY) {
	410	next_log_block +=
	411	count_tags(bh, journal->j_blocksize);
	412	wrap(journal, next_log_block);
	413	brelse(bh);
	414	continue;
	415	}
	416
	417	/* A descriptor block: we can now write all of
	418	* the data blocks. Yay, useful work is finally
	419	* getting done here! */
	420
	421	tagp = &bh->b_data[sizeof(journal_header_t)];
	422	while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
	423	<= journal->j_blocksize) {
9c28cbcc	424	unsigned int io_block;
1da177e4 LT	425
	426	tag = (journal_block_tag_t *) tagp;
	427	flags = be32_to_cpu(tag->t_flags);
	428
	429	io_block = next_log_block++;
	430	wrap(journal, next_log_block);
	431	err = jread(&obh, journal, io_block);
	432	if (err) {
	433	/* Recover what we can, but
	434	* report failure at the end. */
	435	success = err;
ae6ddcc5	436	printk (KERN_ERR
1da177e4	437	"JBD: IO error %d recovering "
9c28cbcc	438	"block %u in log\n",
1da177e4 LT	439	err, io_block);
1da177e4 LT	440	} else {
9c28cbcc	441	unsigned int blocknr;
1da177e4 LT	442
	443	J_ASSERT(obh != NULL);
	444	blocknr = be32_to_cpu(tag->t_blocknr);
	445
	446	/* If the block has been
	447	* revoked, then we're all done
	448	* here. */
	449	if (journal_test_revoke
ae6ddcc5	450	(journal, blocknr,
1da177e4 LT	451	next_commit_ID)) {
	452	brelse(obh);
	453	++info->nr_revoke_hits;
	454	goto skip_write;
	455	}
	456
	457	/* Find a buffer for the new
	458	* data being restored */
	459	nbh = __getblk(journal->j_fs_dev,
	460	blocknr,
	461	journal->j_blocksize);
	462	if (nbh == NULL) {
ae6ddcc5	463	printk(KERN_ERR
1da177e4 LT	464	"JBD: Out of memory "
	465	"during recovery.\n");
	466	err = -ENOMEM;
	467	brelse(bh);
	468	brelse(obh);
	469	goto failed;
	470	}
	471
	472	lock_buffer(nbh);
	473	memcpy(nbh->b_data, obh->b_data,
	474	journal->j_blocksize);
	475	if (flags & JFS_FLAG_ESCAPE) {
439aeec6	476	((__be32 )nbh->b_data) =
1da177e4 LT	477	cpu_to_be32(JFS_MAGIC_NUMBER);
	478	}
	479
	480	BUFFER_TRACE(nbh, "marking dirty");
	481	set_buffer_uptodate(nbh);
	482	mark_buffer_dirty(nbh);
	483	BUFFER_TRACE(nbh, "marking uptodate");
	484	++info->nr_replays;
	485	/* ll_rw_block(WRITE, 1, &nbh); */
	486	unlock_buffer(nbh);
	487	brelse(obh);
	488	brelse(nbh);
	489	}
	490
	491	skip_write:
	492	tagp += sizeof(journal_block_tag_t);
	493	if (!(flags & JFS_FLAG_SAME_UUID))
	494	tagp += 16;
	495
	496	if (flags & JFS_FLAG_LAST_TAG)
	497	break;
	498	}
	499
	500	brelse(bh);
	501	continue;
	502
	503	case JFS_COMMIT_BLOCK:
	504	/* Found an expected commit block: not much to
	505	* do other than move on to the next sequence
	506	* number. */
	507	brelse(bh);
	508	next_commit_ID++;
	509	continue;
	510
	511	case JFS_REVOKE_BLOCK:
	512	/* If we aren't in the REVOKE pass, then we can
	513	* just skip over this block. */
	514	if (pass != PASS_REVOKE) {
	515	brelse(bh);
	516	continue;
	517	}
	518
	519	err = scan_revoke_records(journal, bh,
	520	next_commit_ID, info);
	521	brelse(bh);
	522	if (err)
	523	goto failed;
	524	continue;
	525
	526	default:
	527	jbd_debug(3, "Unrecognised magic %d, end of scan.\n",
	528	blocktype);
e8f1c622	529	brelse(bh);
1da177e4 LT	530	goto done;
	531	}
	532	}
	533
	534	done:
ae6ddcc5	535	/*
1da177e4 LT	536	* We broke out of the log scan loop: either we came to the
	537	* known end of the log or we found an unexpected block in the
	538	* log. If the latter happened, then we know that the "current"
	539	* transaction marks the end of the valid log.
	540	*/
	541
	542	if (pass == PASS_SCAN)
	543	info->end_transaction = next_commit_ID;
	544	else {
	545	/* It's really bad news if different passes end up at
	546	* different places (but possible due to IO errors). */
	547	if (info->end_transaction != next_commit_ID) {
	548	printk (KERN_ERR "JBD: recovery pass %d ended at "
	549	"transaction %u, expected %u\n",
	550	pass, next_commit_ID, info->end_transaction);
	551	if (!success)
	552	success = -EIO;
	553	}
	554	}
	555
	556	return success;
	557
	558	failed:
	559	return err;
	560	}
	561
	562
	563	/* Scan a revoke record, marking all blocks mentioned as revoked. */
	564
ae6ddcc5	565	static int scan_revoke_records(journal_t journal, struct buffer_head bh,
1da177e4 LT	566	tid_t sequence, struct recovery_info *info)
	567	{
	568	journal_revoke_header_t *header;
	569	int offset, max;
	570
	571	header = (journal_revoke_header_t *) bh->b_data;
	572	offset = sizeof(journal_revoke_header_t);
	573	max = be32_to_cpu(header->r_count);
	574
	575	while (offset < max) {
9c28cbcc	576	unsigned int blocknr;
1da177e4 LT	577	int err;
	578
	579	blocknr = be32_to_cpu(* ((__be32 *) (bh->b_data+offset)));
	580	offset += 4;
	581	err = journal_set_revoke(journal, blocknr, sequence);
	582	if (err)
	583	return err;
	584	++info->nr_revokes;
	585	}
	586	return 0;
	587	}