journal.c

linux 内核源代码

}

static int flush_older_commits(struct super_block *s,
			       struct reiserfs_journal_list *jl)
{
	struct reiserfs_journal *journal = SB_JOURNAL(s);
	struct reiserfs_journal_list *other_jl;
	struct reiserfs_journal_list *first_jl;
	struct list_head *entry;
	unsigned long trans_id = jl->j_trans_id;
	unsigned long other_trans_id;
	unsigned long first_trans_id;

      find_first:
	/*
	 * first we walk backwards to find the oldest uncommitted transation
	 */
	first_jl = jl;
	entry = jl->j_list.prev;
	while (1) {
		other_jl = JOURNAL_LIST_ENTRY(entry);
		if (entry == &journal->j_journal_list ||
		    atomic_read(&other_jl->j_older_commits_done))
			break;

		first_jl = other_jl;
		entry = other_jl->j_list.prev;
	}

	/* if we didn't find any older uncommitted transactions, return now */
	if (first_jl == jl) {
		return 0;
	}

	first_trans_id = first_jl->j_trans_id;

	entry = &first_jl->j_list;
	while (1) {
		other_jl = JOURNAL_LIST_ENTRY(entry);
		other_trans_id = other_jl->j_trans_id;

		if (other_trans_id < trans_id) {
			if (atomic_read(&other_jl->j_commit_left) != 0) {
				flush_commit_list(s, other_jl, 0);

				/* list we were called with is gone, return */
				if (!journal_list_still_alive(s, trans_id))
					return 1;

				/* the one we just flushed is gone, this means all
				 * older lists are also gone, so first_jl is no longer
				 * valid either.  Go back to the beginning.
				 */
				if (!journal_list_still_alive
				    (s, other_trans_id)) {
					goto find_first;
				}
			}
			entry = entry->next;
			if (entry == &journal->j_journal_list)
				return 0;
		} else {
			return 0;
		}
	}
	return 0;
}

static int reiserfs_async_progress_wait(struct super_block *s)
{
	DEFINE_WAIT(wait);
	struct reiserfs_journal *j = SB_JOURNAL(s);
	if (atomic_read(&j->j_async_throttle))
		congestion_wait(WRITE, HZ / 10);
	return 0;
}

/*
** if this journal list still has commit blocks unflushed, send them to disk.
**
** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
** Before the commit block can by written, every other log block must be safely on disk
**
*/
static int flush_commit_list(struct super_block *s,
			     struct reiserfs_journal_list *jl, int flushall)
{
	int i;
	b_blocknr_t bn;
	struct buffer_head *tbh = NULL;
	unsigned long trans_id = jl->j_trans_id;
	struct reiserfs_journal *journal = SB_JOURNAL(s);
	int barrier = 0;
	int retval = 0;
	int write_len;

	reiserfs_check_lock_depth(s, "flush_commit_list");

	if (atomic_read(&jl->j_older_commits_done)) {
		return 0;
	}

	get_fs_excl();

	/* before we can put our commit blocks on disk, we have to make sure everyone older than
	 ** us is on disk too
	 */
	BUG_ON(jl->j_len <= 0);
	BUG_ON(trans_id == journal->j_trans_id);

	get_journal_list(jl);
	if (flushall) {
		if (flush_older_commits(s, jl) == 1) {
			/* list disappeared during flush_older_commits.  return */
			goto put_jl;
		}
	}

	/* make sure nobody is trying to flush this one at the same time */
	down(&jl->j_commit_lock);
	if (!journal_list_still_alive(s, trans_id)) {
		up(&jl->j_commit_lock);
		goto put_jl;
	}
	BUG_ON(jl->j_trans_id == 0);

	/* this commit is done, exit */
	if (atomic_read(&(jl->j_commit_left)) <= 0) {
		if (flushall) {
			atomic_set(&(jl->j_older_commits_done), 1);
		}
		up(&jl->j_commit_lock);
		goto put_jl;
	}

	if (!list_empty(&jl->j_bh_list)) {
		int ret;
		unlock_kernel();
		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
					    journal, jl, &jl->j_bh_list);
		if (ret < 0 && retval == 0)
			retval = ret;
		lock_kernel();
	}
	BUG_ON(!list_empty(&jl->j_bh_list));
	/*
	 * for the description block and all the log blocks, submit any buffers
	 * that haven't already reached the disk.  Try to write at least 256
	 * log blocks. later on, we will only wait on blocks that correspond
	 * to this transaction, but while we're unplugging we might as well
	 * get a chunk of data on there.
	 */
	atomic_inc(&journal->j_async_throttle);
	write_len = jl->j_len + 1;
	if (write_len < 256)
		write_len = 256;
	for (i = 0 ; i < write_len ; i++) {
		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
		    SB_ONDISK_JOURNAL_SIZE(s);
		tbh = journal_find_get_block(s, bn);
		if (tbh) {
			if (buffer_dirty(tbh))
			    ll_rw_block(WRITE, 1, &tbh) ;
			put_bh(tbh) ;
		}
	}
	atomic_dec(&journal->j_async_throttle);

	/* We're skipping the commit if there's an error */
	if (retval || reiserfs_is_journal_aborted(journal))
		barrier = 0;

	/* wait on everything written so far before writing the commit
	 * if we are in barrier mode, send the commit down now
	 */
	barrier = reiserfs_barrier_flush(s);
	if (barrier) {
		int ret;
		lock_buffer(jl->j_commit_bh);
		ret = submit_barrier_buffer(jl->j_commit_bh);
		if (ret == -EOPNOTSUPP) {
			set_buffer_uptodate(jl->j_commit_bh);
			disable_barrier(s);
			barrier = 0;
		}
	}
	for (i = 0; i < (jl->j_len + 1); i++) {
		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
		tbh = journal_find_get_block(s, bn);
		wait_on_buffer(tbh);
		// since we're using ll_rw_blk above, it might have skipped over
		// a locked buffer.  Double check here
		//
		if (buffer_dirty(tbh))	/* redundant, sync_dirty_buffer() checks */
			sync_dirty_buffer(tbh);
		if (unlikely(!buffer_uptodate(tbh))) {
#ifdef CONFIG_REISERFS_CHECK
			reiserfs_warning(s, "journal-601, buffer write failed");
#endif
			retval = -EIO;
		}
		put_bh(tbh);	/* once for journal_find_get_block */
		put_bh(tbh);	/* once due to original getblk in do_journal_end */
		atomic_dec(&(jl->j_commit_left));
	}

	BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);

	if (!barrier) {
		/* If there was a write error in the journal - we can't commit
		 * this transaction - it will be invalid and, if successful,
		 * will just end up propagating the write error out to
		 * the file system. */
		if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
			if (buffer_dirty(jl->j_commit_bh))
				BUG();
			mark_buffer_dirty(jl->j_commit_bh) ;
			sync_dirty_buffer(jl->j_commit_bh) ;
		}
	} else
		wait_on_buffer(jl->j_commit_bh);

	check_barrier_completion(s, jl->j_commit_bh);

	/* If there was a write error in the journal - we can't commit this
	 * transaction - it will be invalid and, if successful, will just end
	 * up propagating the write error out to the filesystem. */
	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
#ifdef CONFIG_REISERFS_CHECK
		reiserfs_warning(s, "journal-615: buffer write failed");
#endif
		retval = -EIO;
	}
	bforget(jl->j_commit_bh);
	if (journal->j_last_commit_id != 0 &&
	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
		reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
				 journal->j_last_commit_id, jl->j_trans_id);
	}
	journal->j_last_commit_id = jl->j_trans_id;

	/* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
	cleanup_freed_for_journal_list(s, jl);

	retval = retval ? retval : journal->j_errno;

	/* mark the metadata dirty */
	if (!retval)
		dirty_one_transaction(s, jl);
	atomic_dec(&(jl->j_commit_left));

	if (flushall) {
		atomic_set(&(jl->j_older_commits_done), 1);
	}
	up(&jl->j_commit_lock);
      put_jl:
	put_journal_list(s, jl);

	if (retval)
		reiserfs_abort(s, retval, "Journal write error in %s",
			       __FUNCTION__);
	put_fs_excl();
	return retval;
}

/*
** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or 
** returns NULL if it can't find anything 
*/
static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
							  reiserfs_journal_cnode
							  *cn)
{
	struct super_block *sb = cn->sb;
	b_blocknr_t blocknr = cn->blocknr;

	cn = cn->hprev;
	while (cn) {
		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
			return cn->jlist;
		}
		cn = cn->hprev;
	}
	return NULL;
}

static int newer_jl_done(struct reiserfs_journal_cnode *cn)
{
	struct super_block *sb = cn->sb;
	b_blocknr_t blocknr = cn->blocknr;

	cn = cn->hprev;
	while (cn) {
		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
		    atomic_read(&cn->jlist->j_commit_left) != 0)
				    return 0;
		cn = cn->hprev;
	}
	return 1;
}

static void remove_journal_hash(struct super_block *,
				struct reiserfs_journal_cnode **,
				struct reiserfs_journal_list *, unsigned long,
				int);

/*
** once all the real blocks have been flushed, it is safe to remove them from the
** journal list for this transaction.  Aside from freeing the cnode, this also allows the
** block to be reallocated for data blocks if it had been deleted.
*/
static void remove_all_from_journal_list(struct super_block *p_s_sb,
					 struct reiserfs_journal_list *jl,
					 int debug)
{
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	struct reiserfs_journal_cnode *cn, *last;
	cn = jl->j_realblock;

	/* which is better, to lock once around the whole loop, or
	 ** to lock for each call to remove_journal_hash?
	 */
	while (cn) {
		if (cn->blocknr != 0) {
			if (debug) {
				reiserfs_warning(p_s_sb,
						 "block %u, bh is %d, state %ld",
						 cn->blocknr, cn->bh ? 1 : 0,
						 cn->state);
			}
			cn->state = 0;
			remove_journal_hash(p_s_sb, journal->j_list_hash_table,
					    jl, cn->blocknr, 1);
		}
		last = cn;
		cn = cn->next;
		free_cnode(p_s_sb, last);
	}
	jl->j_realblock = NULL;
}

/*
** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
** releasing blocks in this transaction for reuse as data blocks.
** called by flush_journal_list, before it calls remove_all_from_journal_list
**
*/
static int _update_journal_header_block(struct super_block *p_s_sb,
					unsigned long offset,
					unsigned long trans_id)
{
	struct reiserfs_journal_header *jh;
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);

	if (reiserfs_is_journal_aborted(journal))
		return -EIO;

	if (trans_id >= journal->j_last_flush_trans_id) {
		if (buffer_locked((journal->j_header_bh))) {
			wait_on_buffer((journal->j_header_bh));
			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
#ifdef CONFIG_REISERFS_CHECK
				reiserfs_warning(p_s_sb,
						 "journal-699: buffer write failed");
#endif
				return -EIO;
			}
		}
		journal->j_last_flush_trans_id = trans_id;
		journal->j_first_unflushed_offset = offset;
		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
							b_data);
		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
		jh->j_first_unflushed_offset = cpu_to_le32(offset);
		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);

		if (reiserfs_barrier_flush(p_s_sb)) {
			int ret;
			lock_buffer(journal->j_header_bh);
			ret = submit_barrier_buffer(journal->j_header_bh);
			if (ret == -EOPNOTSUPP) {
				set_buffer_uptodate(journal->j_header_bh);
				disable_barrier(p_s_sb);
				goto sync;
			}
			wait_on_buffer(journal->j_header_bh);
			check_barrier_completion(p_s_sb, journal->j_header_bh);
		} else {
		      sync:
			set_buffer_dirty(journal->j_header_bh);
			sync_dirty_buffer(journal->j_header_bh);
		}
		if (!buffer_uptodate(journal->j_header_bh)) {
			reiserfs_warning(p_s_sb,
					 "journal-837: IO error during journal replay");
			return -EIO;
		}
	}
	return 0;
}

static int update_journal_header_block(struct super_block *p_s_sb,
				       unsigned long offset,
				       unsigned long trans_id)
{
	return _update_journal_header_block(p_s_sb, offset, trans_id);
}

/* 
** flush any and all journal lists older than you are 
** can only be called from flush_journal_list
*/
static int flush_older_journal_lists(struct super_block *p_s_sb,
				     struct reiserfs_journal_list *jl)
{
	struct list_head *entry;
	struct reiserfs_journal_list *other_jl;
	struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
	unsigned long trans_id = jl->j_trans_id;

	/* we know we are the only ones flushing things, no extra race
	 * protection is required.
	 */
      restart:
	entry = journal->j_journal_list.next;
	/* Did we wrap? */
	if (entry == &journal->j_journal_list)
		return 0;
	other_jl = JOURNAL_LIST_ENTRY(entry);
	if (other_jl->j_trans_id < trans_id) {
		BUG_ON(other_jl->j_refcount <= 0);
		/* do not flush all */
		flush_journal_list(p_s_sb, other_jl, 0);

		/* other_jl is now deleted from the list */
		goto restart;
	}
	return 0;
}

static void del_from_work_list(struct super_block *s,
			       struct reiserfs_journal_list *jl)
{
	struct reiserfs_journal *journal = SB_JOURNAL(s);
	if (!list_empty(&jl->j_working_list)) {
		list_del_init(&jl->j_working_list);
		journal->j_num_work_lists--;
	}
}

/* flush a journal list, both commit and real blocks
**
** always set flushall to 1, unless you are calling from inside
** flush_journal_list
**
** IMPORTANT.  This can only be called while there are no journal writers, 
** and the journal is locked.  That means it can only be called from 
** do_journal_end, or by journal_release
*/
static int flush_journal_list(struct super_block *s,
			      struct reiserfs_journal_list *jl, int flushall)
{