xfs_log_recover.c

linux 内核源代码

 * The next region to add is the start of a new region.  It could be
 * a whole region or it could be the first part of a new region.  Because
 * of this, the assumption here is that the type and size fields of all
 * format structures fit into the first 32 bits of the structure.
 *
 * This works because all regions must be 32 bit aligned.  Therefore, we
 * either have both fields or we have neither field.  In the case we have
 * neither field, the data part of the region is zero length.  We only have
 * a log_op_header and can throw away the header since a new one will appear
 * later.  If we have at least 4 bytes, then we can determine how many regions
 * will appear in the current log item.
 */
STATIC int
xlog_recover_add_to_trans(
	xlog_recover_t		*trans,
	xfs_caddr_t		dp,
	int			len)
{
	xfs_inode_log_format_t	*in_f;			/* any will do */
	xlog_recover_item_t	*item;
	xfs_caddr_t		ptr;

	if (!len)
		return 0;
	item = trans->r_itemq;
	if (item == NULL) {
		ASSERT(*(uint *)dp == XFS_TRANS_HEADER_MAGIC);
		if (len == sizeof(xfs_trans_header_t))
			xlog_recover_add_item(&trans->r_itemq);
		memcpy(&trans->r_theader, dp, len); /* d, s, l */
		return 0;
	}

	ptr = kmem_alloc(len, KM_SLEEP);
	memcpy(ptr, dp, len);
	in_f = (xfs_inode_log_format_t *)ptr;

	if (item->ri_prev->ri_total != 0 &&
	     item->ri_prev->ri_total == item->ri_prev->ri_cnt) {
		xlog_recover_add_item(&trans->r_itemq);
	}
	item = trans->r_itemq;
	item = item->ri_prev;

	if (item->ri_total == 0) {		/* first region to be added */
		item->ri_total	= in_f->ilf_size;
		ASSERT(item->ri_total <= XLOG_MAX_REGIONS_IN_ITEM);
		item->ri_buf = kmem_zalloc((item->ri_total *
					    sizeof(xfs_log_iovec_t)), KM_SLEEP);
	}
	ASSERT(item->ri_total > item->ri_cnt);
	/* Description region is ri_buf[0] */
	item->ri_buf[item->ri_cnt].i_addr = ptr;
	item->ri_buf[item->ri_cnt].i_len  = len;
	item->ri_cnt++;
	return 0;
}

STATIC void
xlog_recover_new_tid(
	xlog_recover_t		**q,
	xlog_tid_t		tid,
	xfs_lsn_t		lsn)
{
	xlog_recover_t		*trans;

	trans = kmem_zalloc(sizeof(xlog_recover_t), KM_SLEEP);
	trans->r_log_tid   = tid;
	trans->r_lsn	   = lsn;
	xlog_recover_put_hashq(q, trans);
}

STATIC int
xlog_recover_unlink_tid(
	xlog_recover_t		**q,
	xlog_recover_t		*trans)
{
	xlog_recover_t		*tp;
	int			found = 0;

	ASSERT(trans != NULL);
	if (trans == *q) {
		*q = (*q)->r_next;
	} else {
		tp = *q;
		while (tp) {
			if (tp->r_next == trans) {
				found = 1;
				break;
			}
			tp = tp->r_next;
		}
		if (!found) {
			xlog_warn(
			     "XFS: xlog_recover_unlink_tid: trans not found");
			ASSERT(0);
			return XFS_ERROR(EIO);
		}
		tp->r_next = tp->r_next->r_next;
	}
	return 0;
}

STATIC void
xlog_recover_insert_item_backq(
	xlog_recover_item_t	**q,
	xlog_recover_item_t	*item)
{
	if (*q == NULL) {
		item->ri_prev = item->ri_next = item;
		*q = item;
	} else {
		item->ri_next		= *q;
		item->ri_prev		= (*q)->ri_prev;
		(*q)->ri_prev		= item;
		item->ri_prev->ri_next	= item;
	}
}

STATIC void
xlog_recover_insert_item_frontq(
	xlog_recover_item_t	**q,
	xlog_recover_item_t	*item)
{
	xlog_recover_insert_item_backq(q, item);
	*q = item;
}

STATIC int
xlog_recover_reorder_trans(
	xlog_recover_t		*trans)
{
	xlog_recover_item_t	*first_item, *itemq, *itemq_next;
	xfs_buf_log_format_t	*buf_f;
	ushort			flags = 0;

	first_item = itemq = trans->r_itemq;
	trans->r_itemq = NULL;
	do {
		itemq_next = itemq->ri_next;
		buf_f = (xfs_buf_log_format_t *)itemq->ri_buf[0].i_addr;

		switch (ITEM_TYPE(itemq)) {
		case XFS_LI_BUF:
			flags = buf_f->blf_flags;
			if (!(flags & XFS_BLI_CANCEL)) {
				xlog_recover_insert_item_frontq(&trans->r_itemq,
								itemq);
				break;
			}
		case XFS_LI_INODE:
		case XFS_LI_DQUOT:
		case XFS_LI_QUOTAOFF:
		case XFS_LI_EFD:
		case XFS_LI_EFI:
			xlog_recover_insert_item_backq(&trans->r_itemq, itemq);
			break;
		default:
			xlog_warn(
	"XFS: xlog_recover_reorder_trans: unrecognized type of log operation");
			ASSERT(0);
			return XFS_ERROR(EIO);
		}
		itemq = itemq_next;
	} while (first_item != itemq);
	return 0;
}

/*
 * Build up the table of buf cancel records so that we don't replay
 * cancelled data in the second pass.  For buffer records that are
 * not cancel records, there is nothing to do here so we just return.
 *
 * If we get a cancel record which is already in the table, this indicates
 * that the buffer was cancelled multiple times.  In order to ensure
 * that during pass 2 we keep the record in the table until we reach its
 * last occurrence in the log, we keep a reference count in the cancel
 * record in the table to tell us how many times we expect to see this
 * record during the second pass.
 */
STATIC void
xlog_recover_do_buffer_pass1(
	xlog_t			*log,
	xfs_buf_log_format_t	*buf_f)
{
	xfs_buf_cancel_t	*bcp;
	xfs_buf_cancel_t	*nextp;
	xfs_buf_cancel_t	*prevp;
	xfs_buf_cancel_t	**bucket;
	xfs_daddr_t		blkno = 0;
	uint			len = 0;
	ushort			flags = 0;

	switch (buf_f->blf_type) {
	case XFS_LI_BUF:
		blkno = buf_f->blf_blkno;
		len = buf_f->blf_len;
		flags = buf_f->blf_flags;
		break;
	}

	/*
	 * If this isn't a cancel buffer item, then just return.
	 */
	if (!(flags & XFS_BLI_CANCEL))
		return;

	/*
	 * Insert an xfs_buf_cancel record into the hash table of
	 * them.  If there is already an identical record, bump
	 * its reference count.
	 */
	bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
					  XLOG_BC_TABLE_SIZE];
	/*
	 * If the hash bucket is empty then just insert a new record into
	 * the bucket.
	 */
	if (*bucket == NULL) {
		bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
						     KM_SLEEP);
		bcp->bc_blkno = blkno;
		bcp->bc_len = len;
		bcp->bc_refcount = 1;
		bcp->bc_next = NULL;
		*bucket = bcp;
		return;
	}

	/*
	 * The hash bucket is not empty, so search for duplicates of our
	 * record.  If we find one them just bump its refcount.  If not
	 * then add us at the end of the list.
	 */
	prevp = NULL;
	nextp = *bucket;
	while (nextp != NULL) {
		if (nextp->bc_blkno == blkno && nextp->bc_len == len) {
			nextp->bc_refcount++;
			return;
		}
		prevp = nextp;
		nextp = nextp->bc_next;
	}
	ASSERT(prevp != NULL);
	bcp = (xfs_buf_cancel_t *)kmem_alloc(sizeof(xfs_buf_cancel_t),
					     KM_SLEEP);
	bcp->bc_blkno = blkno;
	bcp->bc_len = len;
	bcp->bc_refcount = 1;
	bcp->bc_next = NULL;
	prevp->bc_next = bcp;
}

/*
 * Check to see whether the buffer being recovered has a corresponding
 * entry in the buffer cancel record table.  If it does then return 1
 * so that it will be cancelled, otherwise return 0.  If the buffer is
 * actually a buffer cancel item (XFS_BLI_CANCEL is set), then decrement
 * the refcount on the entry in the table and remove it from the table
 * if this is the last reference.
 *
 * We remove the cancel record from the table when we encounter its
 * last occurrence in the log so that if the same buffer is re-used
 * again after its last cancellation we actually replay the changes
 * made at that point.
 */
STATIC int
xlog_check_buffer_cancelled(
	xlog_t			*log,
	xfs_daddr_t		blkno,
	uint			len,
	ushort			flags)
{
	xfs_buf_cancel_t	*bcp;
	xfs_buf_cancel_t	*prevp;
	xfs_buf_cancel_t	**bucket;

	if (log->l_buf_cancel_table == NULL) {
		/*
		 * There is nothing in the table built in pass one,
		 * so this buffer must not be cancelled.
		 */
		ASSERT(!(flags & XFS_BLI_CANCEL));
		return 0;
	}

	bucket = &log->l_buf_cancel_table[(__uint64_t)blkno %
					  XLOG_BC_TABLE_SIZE];
	bcp = *bucket;
	if (bcp == NULL) {
		/*
		 * There is no corresponding entry in the table built
		 * in pass one, so this buffer has not been cancelled.
		 */
		ASSERT(!(flags & XFS_BLI_CANCEL));
		return 0;
	}

	/*
	 * Search for an entry in the buffer cancel table that
	 * matches our buffer.
	 */
	prevp = NULL;
	while (bcp != NULL) {
		if (bcp->bc_blkno == blkno && bcp->bc_len == len) {
			/*
			 * We've go a match, so return 1 so that the
			 * recovery of this buffer is cancelled.
			 * If this buffer is actually a buffer cancel
			 * log item, then decrement the refcount on the
			 * one in the table and remove it if this is the
			 * last reference.
			 */
			if (flags & XFS_BLI_CANCEL) {
				bcp->bc_refcount--;
				if (bcp->bc_refcount == 0) {
					if (prevp == NULL) {
						*bucket = bcp->bc_next;
					} else {
						prevp->bc_next = bcp->bc_next;
					}
					kmem_free(bcp,
						  sizeof(xfs_buf_cancel_t));
				}
			}
			return 1;
		}
		prevp = bcp;
		bcp = bcp->bc_next;
	}
	/*
	 * We didn't find a corresponding entry in the table, so
	 * return 0 so that the buffer is NOT cancelled.
	 */
	ASSERT(!(flags & XFS_BLI_CANCEL));
	return 0;
}

STATIC int
xlog_recover_do_buffer_pass2(
	xlog_t			*log,
	xfs_buf_log_format_t	*buf_f)
{
	xfs_daddr_t		blkno = 0;
	ushort			flags = 0;
	uint			len = 0;

	switch (buf_f->blf_type) {
	case XFS_LI_BUF:
		blkno = buf_f->blf_blkno;
		flags = buf_f->blf_flags;
		len = buf_f->blf_len;
		break;
	}

	return xlog_check_buffer_cancelled(log, blkno, len, flags);
}

/*
 * Perform recovery for a buffer full of inodes.  In these buffers,
 * the only data which should be recovered is that which corresponds
 * to the di_next_unlinked pointers in the on disk inode structures.
 * The rest of the data for the inodes is always logged through the
 * inodes themselves rather than the inode buffer and is recovered
 * in xlog_recover_do_inode_trans().
 *
 * The only time when buffers full of inodes are fully recovered is
 * when the buffer is full of newly allocated inodes.  In this case
 * the buffer will not be marked as an inode buffer and so will be
 * sent to xlog_recover_do_reg_buffer() below during recovery.
 */
STATIC int
xlog_recover_do_inode_buffer(
	xfs_mount_t		*mp,
	xlog_recover_item_t	*item,
	xfs_buf_t		*bp,
	xfs_buf_log_format_t	*buf_f)
{
	int			i;
	int			item_index;
	int			bit;
	int			nbits;
	int			reg_buf_offset;
	int			reg_buf_bytes;
	int			next_unlinked_offset;
	int			inodes_per_buf;
	xfs_agino_t		*logged_nextp;
	xfs_agino_t		*buffer_nextp;
	unsigned int		*data_map = NULL;
	unsigned int		map_size = 0;

	switch (buf_f->blf_type) {
	case XFS_LI_BUF:
		data_map = buf_f->blf_data_map;
		map_size = buf_f->blf_map_size;
		break;
	}
	/*
	 * Set the variables corresponding to the current region to
	 * 0 so that we'll initialize them on the first pass through
	 * the loop.
	 */
	reg_buf_offset = 0;
	reg_buf_bytes = 0;
	bit = 0;
	nbits = 0;
	item_index = 0;
	inodes_per_buf = XFS_BUF_COUNT(bp) >> mp->m_sb.sb_inodelog;
	for (i = 0; i < inodes_per_buf; i++) {
		next_unlinked_offset = (i * mp->m_sb.sb_inodesize) +
			offsetof(xfs_dinode_t, di_next_unlinked);

		while (next_unlinked_offset >=
		       (reg_buf_offset + reg_buf_bytes)) {
			/*
			 * The next di_next_unlinked field is beyond
			 * the current logged region.  Find the next
			 * logged region that contains or is beyond
			 * the current di_next_unlinked field.
			 */
			bit += nbits;
			bit = xfs_next_bit(data_map, map_size, bit);

			/*
			 * If there are no more logged regions in the
			 * buffer, then we're done.
			 */
			if (bit == -1) {
				return 0;
			}

			nbits = xfs_contig_bits(data_map, map_size,
							 bit);
			ASSERT(nbits > 0);
			reg_buf_offset = bit << XFS_BLI_SHIFT;
			reg_buf_bytes = nbits << XFS_BLI_SHIFT;
			item_index++;
		}

		/*
		 * If the current logged region starts after the current
		 * di_next_unlinked field, then move on to the next
		 * di_next_unlinked field.
		 */
		if (next_unlinked_offset < reg_buf_offset) {
			continue;
		}

		ASSERT(item->ri_buf[item_index].i_addr != NULL);
		ASSERT((item->ri_buf[item_index].i_len % XFS_BLI_CHUNK) == 0);
		ASSERT((reg_buf_offset + reg_buf_bytes) <= XFS_BUF_COUNT(bp));

		/*
		 * The current logged region contains a copy of the
		 * current di_next_unlinked field.  Extract its value
		 * and copy it to the buffer copy.
		 */
		logged_nextp = (xfs_agino_t *)
			       ((char *)(item->ri_buf[item_index].i_addr) +
				(next_unlinked_offset - reg_buf_offset));
		if (unlikely(*logged_nextp == 0)) {
			xfs_fs_cmn_err(CE_ALERT, mp,