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, 36 insertions, 125 deletions

diff --git a/fs/xfs/xfs_dquot_item.c b/fs/xfs/xfs_dquot_item.c
index 8d8295814272..9c5d58d24e54 100644
--- a/fs/xfs/xfs_dquot_item.c
+++ b/fs/xfs/xfs_dquot_item.c
@@ -108,46 +108,6 @@ xfs_qm_dquot_logitem_unpin(
 		wake_up(&dqp->q_pinwait);
 }
 
-/*
- * Given the logitem, this writes the corresponding dquot entry to disk
- * asynchronously. This is called with the dquot entry securely locked;
- * we simply get xfs_qm_dqflush() to do the work, and unlock the dquot
- * at the end.
- */
-STATIC void
-xfs_qm_dquot_logitem_push(
-	struct xfs_log_item	*lip)
-{
-	struct xfs_dquot	*dqp = DQUOT_ITEM(lip)->qli_dquot;
-	struct xfs_buf		*bp = NULL;
-	int			error;
-
-	ASSERT(XFS_DQ_IS_LOCKED(dqp));
-	ASSERT(!completion_done(&dqp->q_flush));
-	ASSERT(atomic_read(&dqp->q_pincount) == 0);
-
-	/*
-	 * Since we were able to lock the dquot's flush lock and
-	 * we found it on the AIL, the dquot must be dirty.  This
-	 * is because the dquot is removed from the AIL while still
-	 * holding the flush lock in xfs_dqflush_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 dquot.
-	 */
-	error = xfs_qm_dqflush(dqp, &bp);
-	if (error) {
-		xfs_warn(dqp->q_mount, "%s: push error %d on dqp %p",
-			__func__, error, dqp);
-		goto out_unlock;
-	}
-
-	xfs_buf_delwri_queue(bp);
-	xfs_buf_relse(bp);
-out_unlock:
-	xfs_dqunlock(dqp);
-}
-
 STATIC xfs_lsn_t
 xfs_qm_dquot_logitem_committed(
 	struct xfs_log_item	*lip,
@@ -179,67 +139,15 @@ xfs_qm_dqunpin_wait(
 	wait_event(dqp->q_pinwait, (atomic_read(&dqp->q_pincount) == 0));
 }
 
-/*
- * This is called when IOP_TRYLOCK returns XFS_ITEM_PUSHBUF to indicate that
- * the dquot is locked by us, but the flush lock isn't. So, here we are
- * going to see if the relevant dquot buffer is incore, waiting on DELWRI.
- * If so, we want to push it out to help us take this item off the AIL as soon
- * as possible.
- *
- * We must not be holding the AIL lock at this point. Calling incore() to
- * search the buffer cache can be a time consuming thing, and AIL lock is a
- * spinlock.
- */
-STATIC bool
-xfs_qm_dquot_logitem_pushbuf(
-	struct xfs_log_item	*lip)
-{
-	struct xfs_dq_logitem	*qlip = DQUOT_ITEM(lip);
-	struct xfs_dquot	*dqp = qlip->qli_dquot;
-	struct xfs_buf		*bp;
-	bool			ret = true;
-
-	ASSERT(XFS_DQ_IS_LOCKED(dqp));
-
-	/*
-	 * If flushlock isn't locked anymore, chances are that the
-	 * inode flush completed and the inode was taken off the AIL.
-	 * So, just get out.
-	 */
-	if (completion_done(&dqp->q_flush) ||
-	    !(lip->li_flags & XFS_LI_IN_AIL)) {
-		xfs_dqunlock(dqp);
-		return true;
-	}
-
-	bp = xfs_incore(dqp->q_mount->m_ddev_targp, qlip->qli_format.qlf_blkno,
-			dqp->q_mount->m_quotainfo->qi_dqchunklen, XBF_TRYLOCK);
-	xfs_dqunlock(dqp);
-	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 attempt to lock the dquot associated with this
- * dquot log item.  Don't sleep on the dquot lock or the flush lock.
- * If the flush lock is already held, indicating that the dquot has
- * been or is in the process of being flushed, then see if we can
- * find the dquot's buffer in the buffer cache without sleeping.  If
- * we can and it is marked delayed write, then we want to send it out.
- * We delay doing so until the push routine, though, to avoid sleeping
- * in any device strategy routines.
- */
 STATIC uint
-xfs_qm_dquot_logitem_trylock(
-	struct xfs_log_item	*lip)
+xfs_qm_dquot_logitem_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
 {
 	struct xfs_dquot	*dqp = DQUOT_ITEM(lip)->qli_dquot;
+	struct xfs_buf		*bp = NULL;
+	uint			rval = XFS_ITEM_SUCCESS;
+	int			error;
 
 	if (atomic_read(&dqp->q_pincount) > 0)
 		return XFS_ITEM_PINNED;
@@ -252,20 +160,36 @@ xfs_qm_dquot_logitem_trylock(
 	 * taking the quota lock.
 	 */
 	if (atomic_read(&dqp->q_pincount) > 0) {
-		xfs_dqunlock(dqp);
-		return XFS_ITEM_PINNED;
+		rval = XFS_ITEM_PINNED;
+		goto out_unlock;
 	}
 
+	/*
+	 * Someone else is already flushing the dquot.  Nothing we can do
+	 * here but wait for the flush to finish and remove the item from
+	 * the AIL.
+	 */
 	if (!xfs_dqflock_nowait(dqp)) {
-		/*
-		 * dquot has already been flushed to the backing buffer,
-		 * leave it locked, pushbuf routine will unlock it.
-		 */
-		return XFS_ITEM_PUSHBUF;
+		rval = XFS_ITEM_FLUSHING;
+		goto out_unlock;
+	}
+
+	spin_unlock(&lip->li_ailp->xa_lock);
+
+	error = xfs_qm_dqflush(dqp, &bp);
+	if (error) {
+		xfs_warn(dqp->q_mount, "%s: push error %d on dqp %p",
+			__func__, error, dqp);
+	} else {
+		if (!xfs_buf_delwri_queue(bp, buffer_list))
+			rval = XFS_ITEM_FLUSHING;
+		xfs_buf_relse(bp);
 	}
 
-	ASSERT(lip->li_flags & XFS_LI_IN_AIL);
-	return XFS_ITEM_SUCCESS;
+	spin_lock(&lip->li_ailp->xa_lock);
+out_unlock:
+	xfs_dqunlock(dqp);
+	return rval;
 }
 
 /*
@@ -316,11 +240,9 @@ static const struct xfs_item_ops xfs_dquot_item_ops = {
 	.iop_format	= xfs_qm_dquot_logitem_format,
 	.iop_pin	= xfs_qm_dquot_logitem_pin,
 	.iop_unpin	= xfs_qm_dquot_logitem_unpin,
-	.iop_trylock	= xfs_qm_dquot_logitem_trylock,
 	.iop_unlock	= xfs_qm_dquot_logitem_unlock,
 	.iop_committed	= xfs_qm_dquot_logitem_committed,
 	.iop_push	= xfs_qm_dquot_logitem_push,
-	.iop_pushbuf	= xfs_qm_dquot_logitem_pushbuf,
 	.iop_committing = xfs_qm_dquot_logitem_committing
 };
 
@@ -415,11 +337,13 @@ xfs_qm_qoff_logitem_unpin(
 }
 
 /*
- * Quotaoff items have no locking, so just return success.
+ * There isn't much you can do to push a quotaoff item.  It is simply
+ * stuck waiting for the log to be flushed to disk.
  */
 STATIC uint
-xfs_qm_qoff_logitem_trylock(
-	struct xfs_log_item	*lip)
+xfs_qm_qoff_logitem_push(
+	struct xfs_log_item	*lip,
+	struct list_head	*buffer_list)
 {
 	return XFS_ITEM_LOCKED;
 }
@@ -446,17 +370,6 @@ xfs_qm_qoff_logitem_committed(
 	return lsn;
 }
 
-/*
- * There isn't much you can do to push on an quotaoff item.  It is simply
- * stuck waiting for the log to be flushed to disk.
- */
-STATIC void
-xfs_qm_qoff_logitem_push(
-	struct xfs_log_item	*lip)
-{
-}
-
-
 STATIC xfs_lsn_t
 xfs_qm_qoffend_logitem_committed(
 	struct xfs_log_item	*lip,
@@ -504,7 +417,6 @@ static const struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
 	.iop_format	= xfs_qm_qoff_logitem_format,
 	.iop_pin	= xfs_qm_qoff_logitem_pin,
 	.iop_unpin	= xfs_qm_qoff_logitem_unpin,
-	.iop_trylock	= xfs_qm_qoff_logitem_trylock,
 	.iop_unlock	= xfs_qm_qoff_logitem_unlock,
 	.iop_committed	= xfs_qm_qoffend_logitem_committed,
 	.iop_push	= xfs_qm_qoff_logitem_push,
@@ -519,7 +431,6 @@ static const struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
 	.iop_format	= xfs_qm_qoff_logitem_format,
 	.iop_pin	= xfs_qm_qoff_logitem_pin,
 	.iop_unpin	= xfs_qm_qoff_logitem_unpin,
-	.iop_trylock	= xfs_qm_qoff_logitem_trylock,
 	.iop_unlock	= xfs_qm_qoff_logitem_unlock,
 	.iop_committed	= xfs_qm_qoff_logitem_committed,
 	.iop_push	= xfs_qm_qoff_logitem_push,