xfs_buf_item.c

linux 内核源代码

		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
		ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
		ASSERT(XFS_BUF_ISSTALE(bp));
		ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
		xfs_buf_item_trace("UNPIN STALE", bip);
		xfs_buftrace("XFS_UNPIN STALE", bp);
		/*
		 * If we get called here because of an IO error, we may
		 * or may not have the item on the AIL. xfs_trans_delete_ail()
		 * will take care of that situation.
		 * xfs_trans_delete_ail() drops the AIL lock.
		 */
		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
			xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
			XFS_BUF_SET_FSPRIVATE(bp, NULL);
			XFS_BUF_CLR_IODONE_FUNC(bp);
		} else {
			AIL_LOCK(mp,s);
			xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip, s);
			xfs_buf_item_relse(bp);
			ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
		}
		xfs_buf_relse(bp);
	}
}

/*
 * this is called from uncommit in the forced-shutdown path.
 * we need to check to see if the reference count on the log item
 * is going to drop to zero.  If so, unpin will free the log item
 * so we need to free the item's descriptor (that points to the item)
 * in the transaction.
 */
STATIC void
xfs_buf_item_unpin_remove(
	xfs_buf_log_item_t	*bip,
	xfs_trans_t		*tp)
{
	xfs_buf_t		*bp;
	xfs_log_item_desc_t	*lidp;
	int			stale = 0;

	bp = bip->bli_buf;
	/*
	 * will xfs_buf_item_unpin() call xfs_buf_item_relse()?
	 */
	if ((atomic_read(&bip->bli_refcount) == 1) &&
	    (bip->bli_flags & XFS_BLI_STALE)) {
		ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0);
		xfs_buf_item_trace("UNPIN REMOVE", bip);
		xfs_buftrace("XFS_UNPIN_REMOVE", bp);
		/*
		 * yes -- clear the xaction descriptor in-use flag
		 * and free the chunk if required.  We can safely
		 * do some work here and then call buf_item_unpin
		 * to do the rest because if the if is true, then
		 * we are holding the buffer locked so no one else
		 * will be able to bump up the refcount.
		 */
		lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip);
		stale = lidp->lid_flags & XFS_LID_BUF_STALE;
		xfs_trans_free_item(tp, lidp);
		/*
		 * Since the transaction no longer refers to the buffer,
		 * the buffer should no longer refer to the transaction.
		 */
		XFS_BUF_SET_FSPRIVATE2(bp, NULL);
	}

	xfs_buf_item_unpin(bip, stale);

	return;
}

/*
 * This is called to attempt to lock the buffer associated with this
 * buf log item.  Don't sleep on the buffer lock.  If we can't get
 * the lock right away, return 0.  If we can get the lock, pull the
 * buffer from the free list, mark it busy, and return 1.
 */
STATIC uint
xfs_buf_item_trylock(
	xfs_buf_log_item_t	*bip)
{
	xfs_buf_t	*bp;

	bp = bip->bli_buf;

	if (XFS_BUF_ISPINNED(bp)) {
		return XFS_ITEM_PINNED;
	}

	if (!XFS_BUF_CPSEMA(bp)) {
		return XFS_ITEM_LOCKED;
	}

	/*
	 * Remove the buffer from the free list.  Only do this
	 * if it's on the free list.  Private buffers like the
	 * superblock buffer are not.
	 */
	XFS_BUF_HOLD(bp);

	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
	xfs_buf_item_trace("TRYLOCK SUCCESS", bip);
	return XFS_ITEM_SUCCESS;
}

/*
 * Release the buffer associated with the buf log item.
 * If there is no dirty logged data associated with the
 * buffer recorded in the buf log item, then free the
 * buf log item and remove the reference to it in the
 * buffer.
 *
 * This call ignores the recursion count.  It is only called
 * when the buffer should REALLY be unlocked, regardless
 * of the recursion count.
 *
 * If the XFS_BLI_HOLD flag is set in the buf log item, then
 * free the log item if necessary but do not unlock the buffer.
 * This is for support of xfs_trans_bhold(). Make sure the
 * XFS_BLI_HOLD field is cleared if we don't free the item.
 */
STATIC void
xfs_buf_item_unlock(
	xfs_buf_log_item_t	*bip)
{
	int		aborted;
	xfs_buf_t	*bp;
	uint		hold;

	bp = bip->bli_buf;
	xfs_buftrace("XFS_UNLOCK", bp);

	/*
	 * Clear the buffer's association with this transaction.
	 */
	XFS_BUF_SET_FSPRIVATE2(bp, NULL);

	/*
	 * If this is a transaction abort, don't return early.
	 * Instead, allow the brelse to happen.
	 * Normally it would be done for stale (cancelled) buffers
	 * at unpin time, but we'll never go through the pin/unpin
	 * cycle if we abort inside commit.
	 */
	aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;

	/*
	 * If the buf item is marked stale, then don't do anything.
	 * We'll unlock the buffer and free the buf item when the
	 * buffer is unpinned for the last time.
	 */
	if (bip->bli_flags & XFS_BLI_STALE) {
		bip->bli_flags &= ~XFS_BLI_LOGGED;
		xfs_buf_item_trace("UNLOCK STALE", bip);
		ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
		if (!aborted)
			return;
	}

	/*
	 * Drop the transaction's reference to the log item if
	 * it was not logged as part of the transaction.  Otherwise
	 * we'll drop the reference in xfs_buf_item_unpin() when
	 * the transaction is really through with the buffer.
	 */
	if (!(bip->bli_flags & XFS_BLI_LOGGED)) {
		atomic_dec(&bip->bli_refcount);
	} else {
		/*
		 * Clear the logged flag since this is per
		 * transaction state.
		 */
		bip->bli_flags &= ~XFS_BLI_LOGGED;
	}

	/*
	 * Before possibly freeing the buf item, determine if we should
	 * release the buffer at the end of this routine.
	 */
	hold = bip->bli_flags & XFS_BLI_HOLD;
	xfs_buf_item_trace("UNLOCK", bip);

	/*
	 * If the buf item isn't tracking any data, free it.
	 * Otherwise, if XFS_BLI_HOLD is set clear it.
	 */
	if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
			     bip->bli_format.blf_map_size)) {
		xfs_buf_item_relse(bp);
	} else if (hold) {
		bip->bli_flags &= ~XFS_BLI_HOLD;
	}

	/*
	 * Release the buffer if XFS_BLI_HOLD was not set.
	 */
	if (!hold) {
		xfs_buf_relse(bp);
	}
}

/*
 * This is called to find out where the oldest active copy of the
 * buf log item in the on disk log resides now that the last log
 * write of it completed at the given lsn.
 * We always re-log all the dirty data in a buffer, so usually the
 * latest copy in the on disk log is the only one that matters.  For
 * those cases we simply return the given lsn.
 *
 * The one exception to this is for buffers full of newly allocated
 * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
 * flag set, indicating that only the di_next_unlinked fields from the
 * inodes in the buffers will be replayed during recovery.  If the
 * original newly allocated inode images have not yet been flushed
 * when the buffer is so relogged, then we need to make sure that we
 * keep the old images in the 'active' portion of the log.  We do this
 * by returning the original lsn of that transaction here rather than
 * the current one.
 */
STATIC xfs_lsn_t
xfs_buf_item_committed(
	xfs_buf_log_item_t	*bip,
	xfs_lsn_t		lsn)
{
	xfs_buf_item_trace("COMMITTED", bip);
	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
	    (bip->bli_item.li_lsn != 0)) {
		return bip->bli_item.li_lsn;
	}
	return (lsn);
}

/*
 * This is called to asynchronously write the buffer associated with this
 * buf log item out to disk. The buffer will already have been locked by
 * a successful call to xfs_buf_item_trylock().  If the buffer still has
 * B_DELWRI set, then get it going out to disk with a call to bawrite().
 * If not, then just release the buffer.
 */
STATIC void
xfs_buf_item_push(
	xfs_buf_log_item_t	*bip)
{
	xfs_buf_t	*bp;

	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
	xfs_buf_item_trace("PUSH", bip);

	bp = bip->bli_buf;

	if (XFS_BUF_ISDELAYWRITE(bp)) {
		xfs_bawrite(bip->bli_item.li_mountp, bp);
	} else {
		xfs_buf_relse(bp);
	}
}

/* ARGSUSED */
STATIC void
xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
{
}

/*
 * This is the ops vector shared by all buf log items.
 */
static struct xfs_item_ops xfs_buf_item_ops = {
	.iop_size	= (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
					xfs_buf_item_format,
	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
					xfs_buf_item_unpin_remove,
	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
	.iop_unlock	= (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
	.iop_committed	= (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
					xfs_buf_item_committed,
	.iop_push	= (void(*)(xfs_log_item_t*))xfs_buf_item_push,
	.iop_pushbuf	= NULL,
	.iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
					xfs_buf_item_committing
};


/*
 * Allocate a new buf log item to go with the given buffer.
 * Set the buffer's b_fsprivate field to point to the new
 * buf log item.  If there are other item's attached to the
 * buffer (see xfs_buf_attach_iodone() below), then put the
 * buf log item at the front.
 */
void
xfs_buf_item_init(
	xfs_buf_t	*bp,
	xfs_mount_t	*mp)
{
	xfs_log_item_t		*lip;
	xfs_buf_log_item_t	*bip;
	int			chunks;
	int			map_size;

	/*
	 * Check to see if there is already a buf log item for
	 * this buffer.  If there is, it is guaranteed to be
	 * the first.  If we do already have one, there is
	 * nothing to do here so return.
	 */
	if (XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *) != mp)
		XFS_BUF_SET_FSPRIVATE3(bp, mp);
	XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
	if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
		lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
		if (lip->li_type == XFS_LI_BUF) {
			return;
		}
	}

	/*
	 * chunks is the number of XFS_BLI_CHUNK size pieces
	 * the buffer can be divided into. Make sure not to
	 * truncate any pieces.  map_size is the size of the
	 * bitmap needed to describe the chunks of the buffer.
	 */
	chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT);
	map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);

	bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
						    KM_SLEEP);
	bip->bli_item.li_type = XFS_LI_BUF;
	bip->bli_item.li_ops = &xfs_buf_item_ops;
	bip->bli_item.li_mountp = mp;
	bip->bli_buf = bp;
	bip->bli_format.blf_type = XFS_LI_BUF;
	bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
	bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
	bip->bli_format.blf_map_size = map_size;
#ifdef XFS_BLI_TRACE
	bip->bli_trace = ktrace_alloc(XFS_BLI_TRACE_SIZE, KM_SLEEP);
#endif

#ifdef XFS_TRANS_DEBUG
	/*
	 * Allocate the arrays for tracking what needs to be logged
	 * and what our callers request to be logged.  bli_orig
	 * holds a copy of the original, clean buffer for comparison
	 * against, and bli_logged keeps a 1 bit flag per byte in
	 * the buffer to indicate which bytes the callers have asked
	 * to have logged.
	 */
	bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
	memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
	bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
#endif

	/*
	 * Put the buf item into the list of items attached to the
	 * buffer at the front.
	 */
	if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
		bip->bli_item.li_bio_list =
				XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
	}
	XFS_BUF_SET_FSPRIVATE(bp, bip);
}


/*
 * Mark bytes first through last inclusive as dirty in the buf
 * item's bitmap.
 */
void
xfs_buf_item_log(
	xfs_buf_log_item_t	*bip,
	uint			first,
	uint			last)
{
	uint		first_bit;
	uint		last_bit;
	uint		bits_to_set;
	uint		bits_set;
	uint		word_num;
	uint		*wordp;
	uint		bit;
	uint		end_bit;
	uint		mask;

	/*
	 * Mark the item as having some dirty data for
	 * quick reference in xfs_buf_item_dirty.
	 */