xfs: on-stack delayed write buffer lists

Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.

This is now easily doable given that we have very few places left that write
delwri buffers:

 - log recovery:
	Only done at mount time, and already forcing out the buffers
	synchronously using xfs_flush_buftarg

 - quotacheck:
	Same story.

 - dquot reclaim:
	Writes out dirty dquots on the LRU under memory pressure.  We might
	want to look into doing more of this via xfsaild, but it's already
	more optimal than the synchronous inode reclaim that writes each
	buffer synchronously.

 - xfsaild:
	This is the main beneficiary of the change.  By keeping a local list
	of buffers to write we reduce latency of writing out buffers, and
	more importably we can remove all the delwri list promotions which
	were hitting the buffer cache hard under sustained metadata loads.

The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait.  Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list.  The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.

Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes.  The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes.  Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.

[ Dave Chinner:
	- rebase on previous patches.
	- improved comments for XBF_DELWRI_Q handling
	- fix XBF_ASYNC handling in queue submission (test 106 failure)
	- rename delwri submit function buffer list parameters for clarity
	- xfs_efd_item_push() should return XFS_ITEM_PINNED ]

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>

1 files changed, 33 insertions, 119 deletions

diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
index d3601ab75dd3..8aaebb2f9efa 100644
--- a/fs/xfs/xfs_inode_item.c
+++ b/fs/xfs/xfs_inode_item.c
@@ -480,25 +480,16 @@ xfs_inode_item_unpin(
 		wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);
 }
 
-/*
- * This is called to attempt to lock the inode associated with this
- * inode log item, in preparation for the push routine which does the actual
- * iflush.  Don't sleep on the inode lock or the flush lock.
- *
- * If the flush lock is already held, indicating that the inode has
- * been or is in the process of being flushed, then (ideally) we'd like to
- * see if the inode's buffer is still incore, and if so give it a nudge.
- * We delay doing so until the pushbuf routine, though, to avoid holding
- * the AIL lock across a call to the blackhole which is the buffer cache.
- * Also we don't want to sleep in any device strategy routines, which can happen
- * if we do the subsequent bawrite in here.
- */
 STATIC uint
-xfs_inode_item_trylock(
-	struct xfs_log_item	*lip)
+xfs_inode_item_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
 {
 	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
 	struct xfs_inode	*ip = iip->ili_inode;
+	struct xfs_buf		*bp = NULL;
+	uint			rval = XFS_ITEM_SUCCESS;
+	int			error;
 
 	if (xfs_ipincount(ip) > 0)
 		return XFS_ITEM_PINNED;
@@ -511,34 +502,45 @@ xfs_inode_item_trylock(
 	 * taking the ilock.
 	 */
 	if (xfs_ipincount(ip) > 0) {
-		xfs_iunlock(ip, XFS_ILOCK_SHARED);
-		return XFS_ITEM_PINNED;
+		rval = XFS_ITEM_PINNED;
+		goto out_unlock;
 	}
 
+	/*
+	 * Someone else is already flushing the inode.  Nothing we can do
+	 * here but wait for the flush to finish and remove the item from
+	 * the AIL.
+	 */
 	if (!xfs_iflock_nowait(ip)) {
-		/*
-		 * inode has already been flushed to the backing buffer,
-		 * leave it locked in shared mode, pushbuf routine will
-		 * unlock it.
-		 */
-		return XFS_ITEM_PUSHBUF;
+		rval = XFS_ITEM_FLUSHING;
+		goto out_unlock;
 	}
 
-	/* Stale items should force out the iclog */
+	/*
+	 * Stale inode items should force out the iclog.
+	 */
 	if (ip->i_flags & XFS_ISTALE) {
 		xfs_ifunlock(ip);
 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 		return XFS_ITEM_PINNED;
 	}
 
-#ifdef DEBUG
-	if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
-		ASSERT(iip->ili_fields != 0);
-		ASSERT(iip->ili_logged == 0);
-		ASSERT(lip->li_flags & XFS_LI_IN_AIL);
+	ASSERT(iip->ili_fields != 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
+	ASSERT(iip->ili_logged == 0 || XFS_FORCED_SHUTDOWN(ip->i_mount));
+
+	spin_unlock(&lip->li_ailp->xa_lock);
+
+	error = xfs_iflush(ip, &bp);
+	if (!error) {
+		if (!xfs_buf_delwri_queue(bp, buffer_list))
+			rval = XFS_ITEM_FLUSHING;
+		xfs_buf_relse(bp);
 	}
-#endif
-	return XFS_ITEM_SUCCESS;
+
+	spin_lock(&lip->li_ailp->xa_lock);
+out_unlock:
+	xfs_iunlock(ip, XFS_ILOCK_SHARED);
+	return rval;
 }
 
 /*
@@ -623,92 +625,6 @@ xfs_inode_item_committed(
 }
 
 /*
- * This gets called by xfs_trans_push_ail(), when IOP_TRYLOCK
- * failed to get the inode flush lock but did get the inode locked SHARED.
- * Here we're trying to see if the inode buffer is incore, and if so whether it's
- * marked delayed write. If that's the case, we'll promote it and that will
- * allow the caller to write the buffer by triggering the xfsbufd to run.
- */
-STATIC bool
-xfs_inode_item_pushbuf(
-	struct xfs_log_item	*lip)
-{
-	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
-	struct xfs_inode	*ip = iip->ili_inode;
-	struct xfs_buf		*bp;
-	bool			ret = true;
-
-	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
-
-	/*
-	 * If a flush is not in progress anymore, chances are that the
-	 * inode was taken off the AIL. So, just get out.
-	 */
-	if (!xfs_isiflocked(ip) ||
-	    !(lip->li_flags & XFS_LI_IN_AIL)) {
-		xfs_iunlock(ip, XFS_ILOCK_SHARED);
-		return true;
-	}
-
-	bp = xfs_incore(ip->i_mount->m_ddev_targp, iip->ili_format.ilf_blkno,
-			iip->ili_format.ilf_len, XBF_TRYLOCK);
-
-	xfs_iunlock(ip, XFS_ILOCK_SHARED);
-	if (!bp)
-		return true;
-	if (XFS_BUF_ISDELAYWRITE(bp))
-		xfs_buf_delwri_promote(bp);
-	if (xfs_buf_ispinned(bp))
-		ret = false;
-	xfs_buf_relse(bp);
-	return ret;
-}
-
-/*
- * This is called to asynchronously write the inode associated with this
- * inode log item out to disk. The inode will already have been locked by
- * a successful call to xfs_inode_item_trylock().
- */
-STATIC void
-xfs_inode_item_push(
-	struct xfs_log_item	*lip)
-{
-	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
-	struct xfs_inode	*ip = iip->ili_inode;
-	struct xfs_buf		*bp = NULL;
-	int			error;
-
-	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
-	ASSERT(xfs_isiflocked(ip));
-
-	/*
-	 * Since we were able to lock the inode's flush lock and
-	 * we found it on the AIL, the inode must be dirty.  This
-	 * is because the inode is removed from the AIL while still
-	 * holding the flush lock in xfs_iflush_done().  Thus, if
-	 * we found it in the AIL and were able to obtain the flush
-	 * lock without sleeping, then there must not have been
-	 * anyone in the process of flushing the inode.
-	 */
-	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || iip->ili_fields != 0);
-
-	/*
-	 * Push the inode to it's backing buffer. This will not remove the
-	 * inode from the AIL - a further push will be required to trigger a
-	 * buffer push. However, this allows all the dirty inodes to be pushed
-	 * to the buffer before it is pushed to disk. The buffer IO completion
-	 * will pull the inode from the AIL, mark it clean and unlock the flush
-	 * lock.
-	 */
-	error = xfs_iflush(ip, &bp);
-	if (!error) {
-		xfs_buf_delwri_queue(bp);
-		xfs_buf_relse(bp);
-	}
-	xfs_iunlock(ip, XFS_ILOCK_SHARED);
-}
-
-/*
  * XXX rcc - this one really has to do something.  Probably needs
  * to stamp in a new field in the incore inode.
  */
@@ -728,11 +644,9 @@ static const struct xfs_item_ops xfs_inode_item_ops = {
 	.iop_format	= xfs_inode_item_format,
 	.iop_pin	= xfs_inode_item_pin,
 	.iop_unpin	= xfs_inode_item_unpin,
-	.iop_trylock	= xfs_inode_item_trylock,
 	.iop_unlock	= xfs_inode_item_unlock,
 	.iop_committed	= xfs_inode_item_committed,
 	.iop_push	= xfs_inode_item_push,
-	.iop_pushbuf	= xfs_inode_item_pushbuf,
 	.iop_committing = xfs_inode_item_committing
 };