From 453eac8a9aa417878a38bdfbccafd5f7ce4e8e4e Mon Sep 17 00:00:00 2001
From: Dave Chinner <david@fromorbit.com>
Date: Mon, 11 Jan 2010 11:49:58 +0000
Subject: xfs: Don't wake the aild once per second

Now that the AIL push algorithm is traversal safe, we don't need a
watchdog function in the xfsaild to catch pushes that fail to make
progress. Remove the watchdog timeout and make pushes purely driven
by demand. This will remove the once-per-second wakeup that is seen
when the filesystem is idle and make laptop power misers happy.

Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
---
 fs/xfs/xfs_trans_ail.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)

(limited to 'fs/xfs/xfs_trans_ail.c')

diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
index 2ffc570679be..063dfbdca94b 100644
--- a/fs/xfs/xfs_trans_ail.c
+++ b/fs/xfs/xfs_trans_ail.c
@@ -237,14 +237,15 @@ out:
 }
 
 /*
- * Function that does the work of pushing on the AIL
+ * xfsaild_push does the work of pushing on the AIL.  Returning a timeout of
+ * zero indicates that the caller should sleep until woken.
  */
 long
 xfsaild_push(
 	struct xfs_ail	*ailp,
 	xfs_lsn_t	*last_lsn)
 {
-	long		tout = 1000; /* milliseconds */
+	long		tout = 0;
 	xfs_lsn_t	last_pushed_lsn = *last_lsn;
 	xfs_lsn_t	target =  ailp->xa_target;
 	xfs_lsn_t	lsn;
@@ -262,7 +263,7 @@ xfsaild_push(
 		 */
 		xfs_trans_ail_cursor_done(ailp, cur);
 		spin_unlock(&ailp->xa_lock);
-		last_pushed_lsn = 0;
+		*last_lsn = 0;
 		return tout;
 	}
 
@@ -279,7 +280,6 @@ xfsaild_push(
 	 * prevents use from spinning when we can't do anything or there is
 	 * lots of contention on the AIL lists.
 	 */
-	tout = 10;
 	lsn = lip->li_lsn;
 	flush_log = stuck = count = 0;
 	while ((XFS_LSN_CMP(lip->li_lsn, target) < 0)) {
@@ -376,14 +376,14 @@ xfsaild_push(
 
 	if (!count) {
 		/* We're past our target or empty, so idle */
-		tout = 1000;
+		last_pushed_lsn = 0;
 	} else if (XFS_LSN_CMP(lsn, target) >= 0) {
 		/*
 		 * We reached the target so wait a bit longer for I/O to
 		 * complete and remove pushed items from the AIL before we
 		 * start the next scan from the start of the AIL.
 		 */
-		tout += 20;
+		tout = 50;
 		last_pushed_lsn = 0;
 	} else if ((stuck * 100) / count > 90) {
 		/*
@@ -395,11 +395,14 @@ xfsaild_push(
 		 * Backoff a bit more to allow some I/O to complete before
 		 * continuing from where we were.
 		 */
-		tout += 10;
+		tout = 20;
+	} else {
+		/* more to do, but wait a short while before continuing */
+		tout = 10;
 	}
 	*last_lsn = last_pushed_lsn;
 	return tout;
-}	/* xfsaild_push */
+}
 
 
 /*
-- 
cgit v1.2.3


From a14a348bff2f99471a28e5928eb6801224c053d8 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@infradead.org>
Date: Tue, 19 Jan 2010 09:56:46 +0000
Subject: xfs: cleanup up xfs_log_force calling conventions

Remove the XFS_LOG_FORCE argument which was always set, and the
XFS_LOG_URGE define, which was never used.

Split xfs_log_force into a two helpers - xfs_log_force which forces
the whole log, and xfs_log_force_lsn which forces up to the
specified LSN.  The underlying implementations already were entirely
separate, as were the users.

Also re-indent the new _xfs_log_force/_xfs_log_force which
previously had a weird coding style.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
---
 fs/xfs/linux-2.6/xfs_sync.c    |  17 +--
 fs/xfs/quota/xfs_dquot.c       |  10 +-
 fs/xfs/quota/xfs_dquot_item.c  |   9 +-
 fs/xfs/quota/xfs_qm_syscalls.c |   4 +-
 fs/xfs/xfs_alloc.c             |   2 +-
 fs/xfs/xfs_inode.c             |   9 +-
 fs/xfs/xfs_inode_item.c        |   7 +-
 fs/xfs/xfs_log.c               | 312 ++++++++++++++++++++---------------------
 fs/xfs/xfs_log.h               |  15 +-
 fs/xfs/xfs_log_recover.c       |   3 +-
 fs/xfs/xfs_mount.c             |   4 +-
 fs/xfs/xfs_trans.c             |   5 +-
 fs/xfs/xfs_trans_ail.c         |   2 +-
 fs/xfs/xfs_vnodeops.c          |   5 +-
 14 files changed, 193 insertions(+), 211 deletions(-)

(limited to 'fs/xfs/xfs_trans_ail.c')

diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c
index 58c24be72c65..c9b863eacab7 100644
--- a/fs/xfs/linux-2.6/xfs_sync.c
+++ b/fs/xfs/linux-2.6/xfs_sync.c
@@ -296,10 +296,7 @@ xfs_sync_data(
 	if (error)
 		return XFS_ERROR(error);
 
-	xfs_log_force(mp, 0,
-		      (flags & SYNC_WAIT) ?
-		       XFS_LOG_FORCE | XFS_LOG_SYNC :
-		       XFS_LOG_FORCE);
+	xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0);
 	return 0;
 }
 
@@ -325,10 +322,6 @@ xfs_commit_dummy_trans(
 	struct xfs_inode	*ip = mp->m_rootip;
 	struct xfs_trans	*tp;
 	int			error;
-	int			log_flags = XFS_LOG_FORCE;
-
-	if (flags & SYNC_WAIT)
-		log_flags |= XFS_LOG_SYNC;
 
 	/*
 	 * Put a dummy transaction in the log to tell recovery
@@ -350,7 +343,7 @@ xfs_commit_dummy_trans(
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 
 	/* the log force ensures this transaction is pushed to disk */
-	xfs_log_force(mp, 0, log_flags);
+	xfs_log_force(mp, (flags & SYNC_WAIT) ? XFS_LOG_SYNC : 0);
 	return error;
 }
 
@@ -390,7 +383,7 @@ xfs_sync_fsdata(
 		 * become pinned in between there and here.
 		 */
 		if (XFS_BUF_ISPINNED(bp))
-			xfs_log_force(mp, 0, XFS_LOG_FORCE);
+			xfs_log_force(mp, 0);
 	}
 
 
@@ -575,7 +568,7 @@ xfs_flush_inodes(
 	igrab(inode);
 	xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inodes_work, &completion);
 	wait_for_completion(&completion);
-	xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC);
+	xfs_log_force(ip->i_mount, XFS_LOG_SYNC);
 }
 
 /*
@@ -591,7 +584,7 @@ xfs_sync_worker(
 	int		error;
 
 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+		xfs_log_force(mp, 0);
 		xfs_reclaim_inodes(mp, XFS_IFLUSH_DELWRI_ELSE_ASYNC);
 		/* dgc: errors ignored here */
 		error = xfs_qm_sync(mp, SYNC_TRYLOCK);
diff --git a/fs/xfs/quota/xfs_dquot.c b/fs/xfs/quota/xfs_dquot.c
index 5756392ffdee..f9baeedbfdfe 100644
--- a/fs/xfs/quota/xfs_dquot.c
+++ b/fs/xfs/quota/xfs_dquot.c
@@ -1248,7 +1248,7 @@ xfs_qm_dqflush(
 	 */
 	if (XFS_BUF_ISPINNED(bp)) {
 		trace_xfs_dqflush_force(dqp);
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+		xfs_log_force(mp, 0);
 	}
 
 	if (flags & XFS_QMOPT_DELWRI) {
@@ -1531,11 +1531,9 @@ xfs_qm_dqflock_pushbuf_wait(
 	if (bp != NULL) {
 		if (XFS_BUF_ISDELAYWRITE(bp)) {
 			int	error;
-			if (XFS_BUF_ISPINNED(bp)) {
-				xfs_log_force(dqp->q_mount,
-					      (xfs_lsn_t)0,
-					      XFS_LOG_FORCE);
-			}
+
+			if (XFS_BUF_ISPINNED(bp))
+				xfs_log_force(dqp->q_mount, 0);
 			error = xfs_bawrite(dqp->q_mount, bp);
 			if (error)
 				xfs_fs_cmn_err(CE_WARN, dqp->q_mount,
diff --git a/fs/xfs/quota/xfs_dquot_item.c b/fs/xfs/quota/xfs_dquot_item.c
index 116580d52fae..1b564376d50c 100644
--- a/fs/xfs/quota/xfs_dquot_item.c
+++ b/fs/xfs/quota/xfs_dquot_item.c
@@ -190,7 +190,7 @@ xfs_qm_dqunpin_wait(
 	/*
 	 * Give the log a push so we don't wait here too long.
 	 */
-	xfs_log_force(dqp->q_mount, (xfs_lsn_t)0, XFS_LOG_FORCE);
+	xfs_log_force(dqp->q_mount, 0);
 	wait_event(dqp->q_pinwait, (atomic_read(&dqp->q_pincount) == 0));
 }
 
@@ -245,10 +245,9 @@ xfs_qm_dquot_logitem_pushbuf(
 			qip->qli_pushbuf_flag = 0;
 			xfs_dqunlock(dqp);
 
-			if (XFS_BUF_ISPINNED(bp)) {
-				xfs_log_force(mp, (xfs_lsn_t)0,
-					      XFS_LOG_FORCE);
-			}
+			if (XFS_BUF_ISPINNED(bp))
+				xfs_log_force(mp, 0);
+
 			if (dopush) {
 				int	error;
 #ifdef XFSRACEDEBUG
diff --git a/fs/xfs/quota/xfs_qm_syscalls.c b/fs/xfs/quota/xfs_qm_syscalls.c
index 873e07e29074..5d0ee8d492db 100644
--- a/fs/xfs/quota/xfs_qm_syscalls.c
+++ b/fs/xfs/quota/xfs_qm_syscalls.c
@@ -1192,9 +1192,9 @@ xfs_qm_internalqcheck(
 	if (! XFS_IS_QUOTA_ON(mp))
 		return XFS_ERROR(ESRCH);
 
-	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
+	xfs_log_force(mp, XFS_LOG_SYNC);
 	XFS_bflush(mp->m_ddev_targp);
-	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
+	xfs_log_force(mp, XFS_LOG_SYNC);
 	XFS_bflush(mp->m_ddev_targp);
 
 	mutex_lock(&qcheck_lock);
diff --git a/fs/xfs/xfs_alloc.c b/fs/xfs/xfs_alloc.c
index a27aeb7d9e74..94cddbfb2560 100644
--- a/fs/xfs/xfs_alloc.c
+++ b/fs/xfs/xfs_alloc.c
@@ -2601,5 +2601,5 @@ xfs_alloc_search_busy(xfs_trans_t *tp,
 	 * transaction that freed the block
 	 */
 	if (lsn)
-		xfs_log_force(tp->t_mountp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC);
+		xfs_log_force_lsn(tp->t_mountp, lsn, XFS_LOG_SYNC);
 }
diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index bbb3bee8e936..d0d1b5a05183 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -2484,8 +2484,11 @@ __xfs_iunpin_wait(
 		return;
 
 	/* Give the log a push to start the unpinning I/O */
-	xfs_log_force(ip->i_mount, (iip && iip->ili_last_lsn) ?
-				iip->ili_last_lsn : 0, XFS_LOG_FORCE);
+	if (iip && iip->ili_last_lsn)
+		xfs_log_force_lsn(ip->i_mount, iip->ili_last_lsn, 0);
+	else
+		xfs_log_force(ip->i_mount, 0);
+
 	if (wait)
 		wait_event(ip->i_ipin_wait, (atomic_read(&ip->i_pincount) == 0));
 }
@@ -2970,7 +2973,7 @@ xfs_iflush(
 	 * get stuck waiting in the write for too long.
 	 */
 	if (XFS_BUF_ISPINNED(bp))
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+		xfs_log_force(mp, 0);
 
 	/*
 	 * inode clustering:
diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
index da4cac67bdae..48ec1c0b23ce 100644
--- a/fs/xfs/xfs_inode_item.c
+++ b/fs/xfs/xfs_inode_item.c
@@ -804,10 +804,9 @@ xfs_inode_item_pushbuf(
 
 			trace_xfs_inode_item_push(bp, _RET_IP_);
 
-			if (XFS_BUF_ISPINNED(bp)) {
-				xfs_log_force(mp, (xfs_lsn_t)0,
-					      XFS_LOG_FORCE);
-			}
+			if (XFS_BUF_ISPINNED(bp))
+				xfs_log_force(mp, 0);
+
 			if (dopush) {
 				int	error;
 				error = xfs_bawrite(mp, bp);
diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index 20118ddadef6..4f16be4b6ee5 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -79,11 +79,6 @@ STATIC int  xlog_state_release_iclog(xlog_t		*log,
 STATIC void xlog_state_switch_iclogs(xlog_t		*log,
 				     xlog_in_core_t *iclog,
 				     int		eventual_size);
-STATIC int  xlog_state_sync(xlog_t			*log,
-			    xfs_lsn_t 			lsn,
-			    uint			flags,
-			    int				*log_flushed);
-STATIC int  xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed);
 STATIC void xlog_state_want_sync(xlog_t	*log, xlog_in_core_t *iclog);
 
 /* local functions to manipulate grant head */
@@ -296,65 +291,6 @@ xfs_log_done(xfs_mount_t	*mp,
 	return lsn;
 }	/* xfs_log_done */
 
-
-/*
- * Force the in-core log to disk.  If flags == XFS_LOG_SYNC,
- *	the force is done synchronously.
- *
- * Asynchronous forces are implemented by setting the WANT_SYNC
- * bit in the appropriate in-core log and then returning.
- *
- * Synchronous forces are implemented with a signal variable. All callers
- * to force a given lsn to disk will wait on a the sv attached to the
- * specific in-core log.  When given in-core log finally completes its
- * write to disk, that thread will wake up all threads waiting on the
- * sv.
- */
-int
-_xfs_log_force(
-	xfs_mount_t	*mp,
-	xfs_lsn_t	lsn,
-	uint		flags,
-	int		*log_flushed)
-{
-	xlog_t		*log = mp->m_log;
-	int		dummy;
-
-	if (!log_flushed)
-		log_flushed = &dummy;
-
-	ASSERT(flags & XFS_LOG_FORCE);
-
-	XFS_STATS_INC(xs_log_force);
-
-	if (log->l_flags & XLOG_IO_ERROR)
-		return XFS_ERROR(EIO);
-	if (lsn == 0)
-		return xlog_state_sync_all(log, flags, log_flushed);
-	else
-		return xlog_state_sync(log, lsn, flags, log_flushed);
-}	/* _xfs_log_force */
-
-/*
- * Wrapper for _xfs_log_force(), to be used when caller doesn't care
- * about errors or whether the log was flushed or not. This is the normal
- * interface to use when trying to unpin items or move the log forward.
- */
-void
-xfs_log_force(
-	xfs_mount_t	*mp,
-	xfs_lsn_t	lsn,
-	uint		flags)
-{
-	int	error;
-	error = _xfs_log_force(mp, lsn, flags, NULL);
-	if (error) {
-		xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
-			"error %d returned.", error);
-	}
-}
-
-
 /*
  * Attaches a new iclog I/O completion callback routine during
  * transaction commit.  If the log is in error state, a non-zero
@@ -601,7 +537,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
 	if (mp->m_flags & XFS_MOUNT_RDONLY)
 		return 0;
 
-	error = _xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC, NULL);
+	error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
 	ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
 
 #ifdef DEBUG
@@ -2853,7 +2789,6 @@ xlog_state_switch_iclogs(xlog_t		*log,
 	log->l_iclog = iclog->ic_next;
 }	/* xlog_state_switch_iclogs */
 
-
 /*
  * Write out all data in the in-core log as of this exact moment in time.
  *
@@ -2881,11 +2816,17 @@ xlog_state_switch_iclogs(xlog_t		*log,
  *	   b) when we return from flushing out this iclog, it is still
  *		not in the active nor dirty state.
  */
-STATIC int
-xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed)
+int
+_xfs_log_force(
+	struct xfs_mount	*mp,
+	uint			flags,
+	int			*log_flushed)
 {
-	xlog_in_core_t	*iclog;
-	xfs_lsn_t	lsn;
+	struct log		*log = mp->m_log;
+	struct xlog_in_core	*iclog;
+	xfs_lsn_t		lsn;
+
+	XFS_STATS_INC(xs_log_force);
 
 	spin_lock(&log->l_icloglock);
 
@@ -2931,7 +2872,9 @@ xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed)
 
 				if (xlog_state_release_iclog(log, iclog))
 					return XFS_ERROR(EIO);
-				*log_flushed = 1;
+
+				if (log_flushed)
+					*log_flushed = 1;
 				spin_lock(&log->l_icloglock);
 				if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
 				    iclog->ic_state != XLOG_STATE_DIRTY)
@@ -2975,19 +2918,37 @@ maybe_sleep:
 		 */
 		if (iclog->ic_state & XLOG_STATE_IOERROR)
 			return XFS_ERROR(EIO);
-		*log_flushed = 1;
-
+		if (log_flushed)
+			*log_flushed = 1;
 	} else {
 
 no_sleep:
 		spin_unlock(&log->l_icloglock);
 	}
 	return 0;
-}	/* xlog_state_sync_all */
+}
 
+/*
+ * Wrapper for _xfs_log_force(), to be used when caller doesn't care
+ * about errors or whether the log was flushed or not. This is the normal
+ * interface to use when trying to unpin items or move the log forward.
+ */
+void
+xfs_log_force(
+	xfs_mount_t	*mp,
+	uint		flags)
+{
+	int	error;
+
+	error = _xfs_log_force(mp, flags, NULL);
+	if (error) {
+		xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
+			"error %d returned.", error);
+	}
+}
 
 /*
- * Used by code which implements synchronous log forces.
+ * Force the in-core log to disk for a specific LSN.
  *
  * Find in-core log with lsn.
  *	If it is in the DIRTY state, just return.
@@ -2995,109 +2956,142 @@ no_sleep:
  *		state and go to sleep or return.
  *	If it is in any other state, go to sleep or return.
  *
- * If filesystem activity goes to zero, the iclog will get flushed only by
- * bdflush().
+ * Synchronous forces are implemented with a signal variable. All callers
+ * to force a given lsn to disk will wait on a the sv attached to the
+ * specific in-core log.  When given in-core log finally completes its
+ * write to disk, that thread will wake up all threads waiting on the
+ * sv.
  */
-STATIC int
-xlog_state_sync(xlog_t	  *log,
-		xfs_lsn_t lsn,
-		uint	  flags,
-		int	  *log_flushed)
+int
+_xfs_log_force_lsn(
+	struct xfs_mount	*mp,
+	xfs_lsn_t		lsn,
+	uint			flags,
+	int			*log_flushed)
 {
-    xlog_in_core_t	*iclog;
-    int			already_slept = 0;
-
-try_again:
-    spin_lock(&log->l_icloglock);
-    iclog = log->l_iclog;
+	struct log		*log = mp->m_log;
+	struct xlog_in_core	*iclog;
+	int			already_slept = 0;
 
-    if (iclog->ic_state & XLOG_STATE_IOERROR) {
-	    spin_unlock(&log->l_icloglock);
-	    return XFS_ERROR(EIO);
-    }
+	ASSERT(lsn != 0);
 
-    do {
-	if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
-		iclog = iclog->ic_next;
-		continue;
-	}
+	XFS_STATS_INC(xs_log_force);
 
-	if (iclog->ic_state == XLOG_STATE_DIRTY) {
+try_again:
+	spin_lock(&log->l_icloglock);
+	iclog = log->l_iclog;
+	if (iclog->ic_state & XLOG_STATE_IOERROR) {
 		spin_unlock(&log->l_icloglock);
-		return 0;
+		return XFS_ERROR(EIO);
 	}
 
-	if (iclog->ic_state == XLOG_STATE_ACTIVE) {
-		/*
-		 * We sleep here if we haven't already slept (e.g.
-		 * this is the first time we've looked at the correct
-		 * iclog buf) and the buffer before us is going to
-		 * be sync'ed. The reason for this is that if we
-		 * are doing sync transactions here, by waiting for
-		 * the previous I/O to complete, we can allow a few
-		 * more transactions into this iclog before we close
-		 * it down.
-		 *
-		 * Otherwise, we mark the buffer WANT_SYNC, and bump
-		 * up the refcnt so we can release the log (which drops
-		 * the ref count).  The state switch keeps new transaction
-		 * commits from using this buffer.  When the current commits
-		 * finish writing into the buffer, the refcount will drop to
-		 * zero and the buffer will go out then.
-		 */
-		if (!already_slept &&
-		    (iclog->ic_prev->ic_state & (XLOG_STATE_WANT_SYNC |
-						 XLOG_STATE_SYNCING))) {
-			ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
-			XFS_STATS_INC(xs_log_force_sleep);
-			sv_wait(&iclog->ic_prev->ic_write_wait, PSWP,
-				&log->l_icloglock, s);
-			*log_flushed = 1;
-			already_slept = 1;
-			goto try_again;
-		} else {
+	do {
+		if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
+			iclog = iclog->ic_next;
+			continue;
+		}
+
+		if (iclog->ic_state == XLOG_STATE_DIRTY) {
+			spin_unlock(&log->l_icloglock);
+			return 0;
+		}
+
+		if (iclog->ic_state == XLOG_STATE_ACTIVE) {
+			/*
+			 * We sleep here if we haven't already slept (e.g.
+			 * this is the first time we've looked at the correct
+			 * iclog buf) and the buffer before us is going to
+			 * be sync'ed. The reason for this is that if we
+			 * are doing sync transactions here, by waiting for
+			 * the previous I/O to complete, we can allow a few
+			 * more transactions into this iclog before we close
+			 * it down.
+			 *
+			 * Otherwise, we mark the buffer WANT_SYNC, and bump
+			 * up the refcnt so we can release the log (which
+			 * drops the ref count).  The state switch keeps new
+			 * transaction commits from using this buffer.  When
+			 * the current commits finish writing into the buffer,
+			 * the refcount will drop to zero and the buffer will
+			 * go out then.
+			 */
+			if (!already_slept &&
+			    (iclog->ic_prev->ic_state &
+			     (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
+				ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
+
+				XFS_STATS_INC(xs_log_force_sleep);
+
+				sv_wait(&iclog->ic_prev->ic_write_wait,
+					PSWP, &log->l_icloglock, s);
+				if (log_flushed)
+					*log_flushed = 1;
+				already_slept = 1;
+				goto try_again;
+			}
 			atomic_inc(&iclog->ic_refcnt);
 			xlog_state_switch_iclogs(log, iclog, 0);
 			spin_unlock(&log->l_icloglock);
 			if (xlog_state_release_iclog(log, iclog))
 				return XFS_ERROR(EIO);
-			*log_flushed = 1;
+			if (log_flushed)
+				*log_flushed = 1;
 			spin_lock(&log->l_icloglock);
 		}
-	}
 
-	if ((flags & XFS_LOG_SYNC) && /* sleep */
-	    !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
+		if ((flags & XFS_LOG_SYNC) && /* sleep */
+		    !(iclog->ic_state &
+		      (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
+			/*
+			 * Don't wait on completion if we know that we've
+			 * gotten a log write error.
+			 */
+			if (iclog->ic_state & XLOG_STATE_IOERROR) {
+				spin_unlock(&log->l_icloglock);
+				return XFS_ERROR(EIO);
+			}
+			XFS_STATS_INC(xs_log_force_sleep);
+			sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
+			/*
+			 * No need to grab the log lock here since we're
+			 * only deciding whether or not to return EIO
+			 * and the memory read should be atomic.
+			 */
+			if (iclog->ic_state & XLOG_STATE_IOERROR)
+				return XFS_ERROR(EIO);
 
-		/*
-		 * Don't wait on completion if we know that we've
-		 * gotten a log write error.
-		 */
-		if (iclog->ic_state & XLOG_STATE_IOERROR) {
+			if (log_flushed)
+				*log_flushed = 1;
+		} else {		/* just return */
 			spin_unlock(&log->l_icloglock);
-			return XFS_ERROR(EIO);
 		}
-		XFS_STATS_INC(xs_log_force_sleep);
-		sv_wait(&iclog->ic_force_wait, PSWP, &log->l_icloglock, s);
-		/*
-		 * No need to grab the log lock here since we're
-		 * only deciding whether or not to return EIO
-		 * and the memory read should be atomic.
-		 */
-		if (iclog->ic_state & XLOG_STATE_IOERROR)
-			return XFS_ERROR(EIO);
-		*log_flushed = 1;
-	} else {		/* just return */
-		spin_unlock(&log->l_icloglock);
-	}
-	return 0;
 
-    } while (iclog != log->l_iclog);
+		return 0;
+	} while (iclog != log->l_iclog);
 
-    spin_unlock(&log->l_icloglock);
-    return 0;
-}	/* xlog_state_sync */
+	spin_unlock(&log->l_icloglock);
+	return 0;
+}
+
+/*
+ * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
+ * about errors or whether the log was flushed or not. This is the normal
+ * interface to use when trying to unpin items or move the log forward.
+ */
+void
+xfs_log_force_lsn(
+	xfs_mount_t	*mp,
+	xfs_lsn_t	lsn,
+	uint		flags)
+{
+	int	error;
 
+	error = _xfs_log_force_lsn(mp, lsn, flags, NULL);
+	if (error) {
+		xfs_fs_cmn_err(CE_WARN, mp, "xfs_log_force: "
+			"error %d returned.", error);
+	}
+}
 
 /*
  * Called when we want to mark the current iclog as being ready to sync to
@@ -3462,7 +3456,6 @@ xfs_log_force_umount(
 	xlog_ticket_t	*tic;
 	xlog_t		*log;
 	int		retval;
-	int		dummy;
 
 	log = mp->m_log;
 
@@ -3536,13 +3529,14 @@ xfs_log_force_umount(
 	}
 	spin_unlock(&log->l_grant_lock);
 
-	if (! (log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
+	if (!(log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
 		ASSERT(!logerror);
 		/*
 		 * Force the incore logs to disk before shutting the
 		 * log down completely.
 		 */
-		xlog_state_sync_all(log, XFS_LOG_FORCE|XFS_LOG_SYNC, &dummy);
+		_xfs_log_force(mp, XFS_LOG_SYNC, NULL);
+
 		spin_lock(&log->l_icloglock);
 		retval = xlog_state_ioerror(log);
 		spin_unlock(&log->l_icloglock);
diff --git a/fs/xfs/xfs_log.h b/fs/xfs/xfs_log.h
index 811ccf4d8b3e..7074be9d13e9 100644
--- a/fs/xfs/xfs_log.h
+++ b/fs/xfs/xfs_log.h
@@ -70,14 +70,8 @@ static inline xfs_lsn_t	_lsn_cmp(xfs_lsn_t lsn1, xfs_lsn_t lsn2)
  * Flags to xfs_log_force()
  *
  *	XFS_LOG_SYNC:	Synchronous force in-core log to disk
- *	XFS_LOG_FORCE:	Start in-core log write now.
- *	XFS_LOG_URGE:	Start write within some window of time.
- *
- * Note: Either XFS_LOG_FORCE or XFS_LOG_URGE must be set.
  */
 #define XFS_LOG_SYNC		0x1
-#define XFS_LOG_FORCE		0x2
-#define XFS_LOG_URGE		0x4
 
 #endif	/* __KERNEL__ */
 
@@ -138,12 +132,17 @@ xfs_lsn_t xfs_log_done(struct xfs_mount *mp,
 		       void		**iclog,
 		       uint		flags);
 int	  _xfs_log_force(struct xfs_mount *mp,
-			 xfs_lsn_t	lsn,
 			 uint		flags,
 			 int		*log_forced);
 void	  xfs_log_force(struct xfs_mount	*mp,
-			xfs_lsn_t		lsn,
 			uint			flags);
+int	  _xfs_log_force_lsn(struct xfs_mount *mp,
+			     xfs_lsn_t		lsn,
+			     uint		flags,
+			     int		*log_forced);
+void	  xfs_log_force_lsn(struct xfs_mount	*mp,
+			    xfs_lsn_t		lsn,
+			    uint		flags);
 int	  xfs_log_mount(struct xfs_mount	*mp,
 			struct xfs_buftarg	*log_target,
 			xfs_daddr_t		start_block,
diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c
index 97148f0c4bdd..22e6efdc17ea 100644
--- a/fs/xfs/xfs_log_recover.c
+++ b/fs/xfs/xfs_log_recover.c
@@ -3913,8 +3913,7 @@ xlog_recover_finish(
 		 * case the unlink transactions would have problems
 		 * pushing the EFIs out of the way.
 		 */
-		xfs_log_force(log->l_mp, (xfs_lsn_t)0,
-			      (XFS_LOG_FORCE | XFS_LOG_SYNC));
+		xfs_log_force(log->l_mp, XFS_LOG_SYNC);
 
 		xlog_recover_process_iunlinks(log);
 
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index bb0154047e85..7f81ed72c875 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -1455,7 +1455,7 @@ xfs_unmountfs(
 	 * push out the iclog we will never get that unlocked. hence we
 	 * need to force the log first.
 	 */
-	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
+	xfs_log_force(mp, XFS_LOG_SYNC);
 	xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC);
 
 	xfs_qm_unmount(mp);
@@ -1465,7 +1465,7 @@ xfs_unmountfs(
 	 * that nothing is pinned.  This is important because bflush()
 	 * will skip pinned buffers.
 	 */
-	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
+	xfs_log_force(mp, XFS_LOG_SYNC);
 
 	xfs_binval(mp->m_ddev_targp);
 	if (mp->m_rtdev_targp) {
diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
index 7dbe3c3051db..be942d4e3324 100644
--- a/fs/xfs/xfs_trans.c
+++ b/fs/xfs/xfs_trans.c
@@ -981,9 +981,8 @@ shut_us_down:
 	 */
 	if (sync) {
 		if (!error) {
-			error = _xfs_log_force(mp, commit_lsn,
-				      XFS_LOG_FORCE | XFS_LOG_SYNC,
-				      log_flushed);
+			error = _xfs_log_force_lsn(mp, commit_lsn,
+				      XFS_LOG_SYNC, log_flushed);
 		}
 		XFS_STATS_INC(xs_trans_sync);
 	} else {
diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
index 063dfbdca94b..d7b1af8a832d 100644
--- a/fs/xfs/xfs_trans_ail.c
+++ b/fs/xfs/xfs_trans_ail.c
@@ -371,7 +371,7 @@ xfsaild_push(
 		 * move forward in the AIL.
 		 */
 		XFS_STATS_INC(xs_push_ail_flush);
-		xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
+		xfs_log_force(mp, 0);
 	}
 
 	if (!count) {
diff --git a/fs/xfs/xfs_vnodeops.c b/fs/xfs/xfs_vnodeops.c
index 4da96cdffb76..fd108b738559 100644
--- a/fs/xfs/xfs_vnodeops.c
+++ b/fs/xfs/xfs_vnodeops.c
@@ -631,9 +631,8 @@ xfs_fsync(
 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 
 		if (xfs_ipincount(ip)) {
-			error = _xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
-				      XFS_LOG_FORCE | XFS_LOG_SYNC,
-				      &log_flushed);
+			error = _xfs_log_force(ip->i_mount, XFS_LOG_SYNC,
+					       &log_flushed);
 		} else {
 			/*
 			 * If the inode is not pinned and nothing has changed
-- 
cgit v1.2.3


From d808f617ad00a413585b806de340feda5ad9a2da Mon Sep 17 00:00:00 2001
From: Dave Chinner <david@fromorbit.com>
Date: Tue, 2 Feb 2010 10:13:42 +1100
Subject: xfs: Don't issue buffer IO direct from AIL push V2

All buffers logged into the AIL are marked as delayed write.
When the AIL needs to push the buffer out, it issues an async write of the
buffer. This means that IO patterns are dependent on the order of
buffers in the AIL.

Instead of flushing the buffer, promote the buffer in the delayed
write list so that the next time the xfsbufd is run the buffer will
be flushed by the xfsbufd. Return the state to the xfsaild that the
buffer was promoted so that the xfsaild knows that it needs to cause
the xfsbufd to run to flush the buffers that were promoted.

Using the xfsbufd for issuing the IO allows us to dispatch all
buffer IO from the one queue. This means that we can make much more
enlightened decisions on what order to flush buffers to disk as
we don't have multiple places issuing IO. Optimisations to xfsbufd
will be in a future patch.

Version 2
- kill XFS_ITEM_FLUSHING as it is now unused.

Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/linux-2.6/xfs_buf.c    | 29 +++++++++++++
 fs/xfs/linux-2.6/xfs_buf.h    |  2 +
 fs/xfs/linux-2.6/xfs_trace.h  |  1 +
 fs/xfs/quota/xfs_dquot_item.c | 85 ++++++-------------------------------
 fs/xfs/quota/xfs_dquot_item.h |  4 --
 fs/xfs/xfs_buf_item.c         | 64 +++++++++++++++-------------
 fs/xfs/xfs_inode_item.c       | 98 +++++++------------------------------------
 fs/xfs/xfs_inode_item.h       |  6 ---
 fs/xfs/xfs_trans.h            |  3 +-
 fs/xfs/xfs_trans_ail.c        | 13 +++---
 10 files changed, 102 insertions(+), 203 deletions(-)

(limited to 'fs/xfs/xfs_trans_ail.c')

diff --git a/fs/xfs/linux-2.6/xfs_buf.c b/fs/xfs/linux-2.6/xfs_buf.c
index 44e20e578ba0..b306265caa33 100644
--- a/fs/xfs/linux-2.6/xfs_buf.c
+++ b/fs/xfs/linux-2.6/xfs_buf.c
@@ -1778,6 +1778,35 @@ xfs_buf_delwri_dequeue(
 	trace_xfs_buf_delwri_dequeue(bp, _RET_IP_);
 }
 
+/*
+ * If a delwri buffer needs to be pushed before it has aged out, then promote
+ * it to the head of the delwri queue so that it will be flushed on the next
+ * xfsbufd run. We do this by resetting the queuetime of the buffer to be older
+ * than the age currently needed to flush the buffer. Hence the next time the
+ * xfsbufd sees it is guaranteed to be considered old enough to flush.
+ */
+void
+xfs_buf_delwri_promote(
+	struct xfs_buf	*bp)
+{
+	struct xfs_buftarg *btp = bp->b_target;
+	long		age = xfs_buf_age_centisecs * msecs_to_jiffies(10) + 1;
+
+	ASSERT(bp->b_flags & XBF_DELWRI);
+	ASSERT(bp->b_flags & _XBF_DELWRI_Q);
+
+	/*
+	 * Check the buffer age before locking the delayed write queue as we
+	 * don't need to promote buffers that are already past the flush age.
+	 */
+	if (bp->b_queuetime < jiffies - age)
+		return;
+	bp->b_queuetime = jiffies - age;
+	spin_lock(&btp->bt_delwrite_lock);
+	list_move(&bp->b_list, &btp->bt_delwrite_queue);
+	spin_unlock(&btp->bt_delwrite_lock);
+}
+
 STATIC void
 xfs_buf_runall_queues(
 	struct workqueue_struct	*queue)
diff --git a/fs/xfs/linux-2.6/xfs_buf.h b/fs/xfs/linux-2.6/xfs_buf.h
index ea8c198f0c39..be45e8c5768d 100644
--- a/fs/xfs/linux-2.6/xfs_buf.h
+++ b/fs/xfs/linux-2.6/xfs_buf.h
@@ -266,6 +266,7 @@ extern int xfs_buf_ispin(xfs_buf_t *);
 
 /* Delayed Write Buffer Routines */
 extern void xfs_buf_delwri_dequeue(xfs_buf_t *);
+extern void xfs_buf_delwri_promote(xfs_buf_t *);
 
 /* Buffer Daemon Setup Routines */
 extern int xfs_buf_init(void);
@@ -395,6 +396,7 @@ extern void xfs_free_buftarg(struct xfs_mount *, struct xfs_buftarg *);
 extern void xfs_wait_buftarg(xfs_buftarg_t *);
 extern int xfs_setsize_buftarg(xfs_buftarg_t *, unsigned int, unsigned int);
 extern int xfs_flush_buftarg(xfs_buftarg_t *, int);
+
 #ifdef CONFIG_KDB_MODULES
 extern struct list_head *xfs_get_buftarg_list(void);
 #endif
diff --git a/fs/xfs/linux-2.6/xfs_trace.h b/fs/xfs/linux-2.6/xfs_trace.h
index 1bb09e70b2eb..a4574dcf5065 100644
--- a/fs/xfs/linux-2.6/xfs_trace.h
+++ b/fs/xfs/linux-2.6/xfs_trace.h
@@ -483,6 +483,7 @@ DEFINE_BUF_ITEM_EVENT(xfs_buf_item_unlock);
 DEFINE_BUF_ITEM_EVENT(xfs_buf_item_unlock_stale);
 DEFINE_BUF_ITEM_EVENT(xfs_buf_item_committed);
 DEFINE_BUF_ITEM_EVENT(xfs_buf_item_push);
+DEFINE_BUF_ITEM_EVENT(xfs_buf_item_pushbuf);
 DEFINE_BUF_ITEM_EVENT(xfs_trans_get_buf);
 DEFINE_BUF_ITEM_EVENT(xfs_trans_get_buf_recur);
 DEFINE_BUF_ITEM_EVENT(xfs_trans_getsb);
diff --git a/fs/xfs/quota/xfs_dquot_item.c b/fs/xfs/quota/xfs_dquot_item.c
index 1b564376d50c..dda0fb045c8a 100644
--- a/fs/xfs/quota/xfs_dquot_item.c
+++ b/fs/xfs/quota/xfs_dquot_item.c
@@ -212,18 +212,10 @@ xfs_qm_dquot_logitem_pushbuf(
 	xfs_dquot_t	*dqp;
 	xfs_mount_t	*mp;
 	xfs_buf_t	*bp;
-	uint		dopush;
 
 	dqp = qip->qli_dquot;
 	ASSERT(XFS_DQ_IS_LOCKED(dqp));
 
-	/*
-	 * The qli_pushbuf_flag keeps others from
-	 * trying to duplicate our effort.
-	 */
-	ASSERT(qip->qli_pushbuf_flag != 0);
-	ASSERT(qip->qli_push_owner == current_pid());
-
 	/*
 	 * If flushlock isn't locked anymore, chances are that the
 	 * inode flush completed and the inode was taken off the AIL.
@@ -231,47 +223,20 @@ xfs_qm_dquot_logitem_pushbuf(
 	 */
 	if (completion_done(&dqp->q_flush)  ||
 	    ((qip->qli_item.li_flags & XFS_LI_IN_AIL) == 0)) {
-		qip->qli_pushbuf_flag = 0;
 		xfs_dqunlock(dqp);
 		return;
 	}
 	mp = dqp->q_mount;
 	bp = xfs_incore(mp->m_ddev_targp, qip->qli_format.qlf_blkno,
 		    XFS_QI_DQCHUNKLEN(mp), XBF_TRYLOCK);
-	if (bp != NULL) {
-		if (XFS_BUF_ISDELAYWRITE(bp)) {
-			dopush = ((qip->qli_item.li_flags & XFS_LI_IN_AIL) &&
-				  !completion_done(&dqp->q_flush));
-			qip->qli_pushbuf_flag = 0;
-			xfs_dqunlock(dqp);
-
-			if (XFS_BUF_ISPINNED(bp))
-				xfs_log_force(mp, 0);
-
-			if (dopush) {
-				int	error;
-#ifdef XFSRACEDEBUG
-				delay_for_intr();
-				delay(300);
-#endif
-				error = xfs_bawrite(mp, bp);
-				if (error)
-					xfs_fs_cmn_err(CE_WARN, mp,
-	"xfs_qm_dquot_logitem_pushbuf: pushbuf error %d on qip %p, bp %p",
-							error, qip, bp);
-			} else {
-				xfs_buf_relse(bp);
-			}
-		} else {
-			qip->qli_pushbuf_flag = 0;
-			xfs_dqunlock(dqp);
-			xfs_buf_relse(bp);
-		}
+	xfs_dqunlock(dqp);
+	if (!bp)
 		return;
-	}
+	if (XFS_BUF_ISDELAYWRITE(bp))
+		xfs_buf_delwri_promote(bp);
+	xfs_buf_relse(bp);
+	return;
 
-	qip->qli_pushbuf_flag = 0;
-	xfs_dqunlock(dqp);
 }
 
 /*
@@ -289,50 +254,24 @@ xfs_qm_dquot_logitem_trylock(
 	xfs_dq_logitem_t	*qip)
 {
 	xfs_dquot_t		*dqp;
-	uint			retval;
 
 	dqp = qip->qli_dquot;
 	if (atomic_read(&dqp->q_pincount) > 0)
-		return (XFS_ITEM_PINNED);
+		return XFS_ITEM_PINNED;
 
 	if (! xfs_qm_dqlock_nowait(dqp))
-		return (XFS_ITEM_LOCKED);
+		return XFS_ITEM_LOCKED;
 
-	retval = XFS_ITEM_SUCCESS;
 	if (!xfs_dqflock_nowait(dqp)) {
 		/*
-		 * The dquot is already being flushed.	It may have been
-		 * flushed delayed write, however, and we don't want to
-		 * get stuck waiting for that to complete.  So, we want to check
-		 * to see if we can lock the dquot's buffer without sleeping.
-		 * If we can and it is marked for delayed write, then we
-		 * hold it and send it out from the push routine.  We don't
-		 * want to do that now since we might sleep in the device
-		 * strategy routine.  We also don't want to grab the buffer lock
-		 * here because we'd like not to call into the buffer cache
-		 * while holding the AIL lock.
-		 * Make sure to only return PUSHBUF if we set pushbuf_flag
-		 * ourselves.  If someone else is doing it then we don't
-		 * want to go to the push routine and duplicate their efforts.
+		 * dquot has already been flushed to the backing buffer,
+		 * leave it locked, pushbuf routine will unlock it.
 		 */
-		if (qip->qli_pushbuf_flag == 0) {
-			qip->qli_pushbuf_flag = 1;
-			ASSERT(qip->qli_format.qlf_blkno == dqp->q_blkno);
-#ifdef DEBUG
-			qip->qli_push_owner = current_pid();
-#endif
-			/*
-			 * The dquot is left locked.
-			 */
-			retval = XFS_ITEM_PUSHBUF;
-		} else {
-			retval = XFS_ITEM_FLUSHING;
-			xfs_dqunlock_nonotify(dqp);
-		}
+		return XFS_ITEM_PUSHBUF;
 	}
 
 	ASSERT(qip->qli_item.li_flags & XFS_LI_IN_AIL);
-	return (retval);
+	return XFS_ITEM_SUCCESS;
 }
 
 
diff --git a/fs/xfs/quota/xfs_dquot_item.h b/fs/xfs/quota/xfs_dquot_item.h
index 5a632531f843..5acae2ada70b 100644
--- a/fs/xfs/quota/xfs_dquot_item.h
+++ b/fs/xfs/quota/xfs_dquot_item.h
@@ -27,10 +27,6 @@ typedef struct xfs_dq_logitem {
 	xfs_log_item_t		 qli_item;	   /* common portion */
 	struct xfs_dquot	*qli_dquot;	   /* dquot ptr */
 	xfs_lsn_t		 qli_flush_lsn;	   /* lsn at last flush */
-	unsigned short		 qli_pushbuf_flag; /* 1 bit used in push_ail */
-#ifdef DEBUG
-	uint64_t		 qli_push_owner;
-#endif
 	xfs_dq_logformat_t	 qli_format;	   /* logged structure */
 } xfs_dq_logitem_t;
 
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
index e0a11583ce5a..f3c49e69eab9 100644
--- a/fs/xfs/xfs_buf_item.c
+++ b/fs/xfs/xfs_buf_item.c
@@ -467,8 +467,10 @@ xfs_buf_item_unpin_remove(
 /*
  * 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.
+ * the lock right away, return 0.  If we can get the lock, take a
+ * reference to the buffer. If this is a delayed write buffer that
+ * needs AIL help to be written back, invoke the pushbuf routine
+ * rather than the normal success path.
  */
 STATIC uint
 xfs_buf_item_trylock(
@@ -477,24 +479,18 @@ xfs_buf_item_trylock(
 	xfs_buf_t	*bp;
 
 	bp = bip->bli_buf;
-
-	if (XFS_BUF_ISPINNED(bp)) {
+	if (XFS_BUF_ISPINNED(bp))
 		return XFS_ITEM_PINNED;
-	}
-
-	if (!XFS_BUF_CPSEMA(bp)) {
+	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.
-	 */
+	/* take a reference to the buffer.  */
 	XFS_BUF_HOLD(bp);
 
 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 	trace_xfs_buf_item_trylock(bip);
+	if (XFS_BUF_ISDELAYWRITE(bp))
+		return XFS_ITEM_PUSHBUF;
 	return XFS_ITEM_SUCCESS;
 }
 
@@ -626,11 +622,9 @@ xfs_buf_item_committed(
 }
 
 /*
- * 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.
+ * The buffer is locked, but is not a delayed write buffer. This happens
+ * if we race with IO completion and hence we don't want to try to write it
+ * again. Just release the buffer.
  */
 STATIC void
 xfs_buf_item_push(
@@ -642,17 +636,29 @@ xfs_buf_item_push(
 	trace_xfs_buf_item_push(bip);
 
 	bp = bip->bli_buf;
+	ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
+	xfs_buf_relse(bp);
+}
 
-	if (XFS_BUF_ISDELAYWRITE(bp)) {
-		int	error;
-		error = xfs_bawrite(bip->bli_item.li_mountp, bp);
-		if (error)
-			xfs_fs_cmn_err(CE_WARN, bip->bli_item.li_mountp,
-			"xfs_buf_item_push: pushbuf error %d on bip %p, bp %p",
-					error, bip, bp);
-	} else {
-		xfs_buf_relse(bp);
-	}
+/*
+ * The buffer is locked and is a delayed write buffer. Promote the buffer
+ * in the delayed write queue as the caller knows that they must invoke
+ * the xfsbufd to get this buffer written. We have to unlock the buffer
+ * to allow the xfsbufd to write it, too.
+ */
+STATIC void
+xfs_buf_item_pushbuf(
+	xfs_buf_log_item_t	*bip)
+{
+	xfs_buf_t	*bp;
+
+	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
+	trace_xfs_buf_item_pushbuf(bip);
+
+	bp = bip->bli_buf;
+	ASSERT(XFS_BUF_ISDELAYWRITE(bp));
+	xfs_buf_delwri_promote(bp);
+	xfs_buf_relse(bp);
 }
 
 /* ARGSUSED */
@@ -677,7 +683,7 @@ static struct xfs_item_ops xfs_buf_item_ops = {
 	.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_pushbuf	= (void(*)(xfs_log_item_t*))xfs_buf_item_pushbuf,
 	.iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
 					xfs_buf_item_committing
 };
diff --git a/fs/xfs/xfs_inode_item.c b/fs/xfs/xfs_inode_item.c
index 207553e82954..d4dc063111f8 100644
--- a/fs/xfs/xfs_inode_item.c
+++ b/fs/xfs/xfs_inode_item.c
@@ -602,33 +602,20 @@ xfs_inode_item_trylock(
 
 	if (!xfs_iflock_nowait(ip)) {
 		/*
-		 * If someone else isn't already trying to push the inode
-		 * buffer, we get to do it.
+		 * inode has already been flushed to the backing buffer,
+		 * leave it locked in shared mode, pushbuf routine will
+		 * unlock it.
 		 */
-		if (iip->ili_pushbuf_flag == 0) {
-			iip->ili_pushbuf_flag = 1;
-#ifdef DEBUG
-			iip->ili_push_owner = current_pid();
-#endif
-			/*
-			 * Inode is left locked in shared mode.
-			 * Pushbuf routine gets to unlock it.
-			 */
-			return XFS_ITEM_PUSHBUF;
-		} else {
-			/*
-			 * We hold the AIL lock, so we must specify the
-			 * NONOTIFY flag so that we won't double trip.
-			 */
-			xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
-			return XFS_ITEM_FLUSHING;
-		}
-		/* NOTREACHED */
+		return XFS_ITEM_PUSHBUF;
 	}
 
 	/* Stale items should force out the iclog */
 	if (ip->i_flags & XFS_ISTALE) {
 		xfs_ifunlock(ip);
+		/*
+		 * we hold the AIL lock - notify the unlock routine of this
+		 * so it doesn't try to get the lock again.
+		 */
 		xfs_iunlock(ip, XFS_ILOCK_SHARED|XFS_IUNLOCK_NONOTIFY);
 		return XFS_ITEM_PINNED;
 	}
@@ -746,11 +733,8 @@ 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 initiate a bawrite on that
- * buffer to expedite the process.
- *
- * We aren't holding the AIL lock (or the flush lock) when this gets called,
- * so it is inherently race-y.
+ * 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 void
 xfs_inode_item_pushbuf(
@@ -759,26 +743,16 @@ xfs_inode_item_pushbuf(
 	xfs_inode_t	*ip;
 	xfs_mount_t	*mp;
 	xfs_buf_t	*bp;
-	uint		dopush;
 
 	ip = iip->ili_inode;
-
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED));
 
-	/*
-	 * The ili_pushbuf_flag keeps others from
-	 * trying to duplicate our effort.
-	 */
-	ASSERT(iip->ili_pushbuf_flag != 0);
-	ASSERT(iip->ili_push_owner == current_pid());
-
 	/*
 	 * If a flush is not in progress anymore, chances are that the
 	 * inode was taken off the AIL. So, just get out.
 	 */
 	if (completion_done(&ip->i_flush) ||
 	    ((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0)) {
-		iip->ili_pushbuf_flag = 0;
 		xfs_iunlock(ip, XFS_ILOCK_SHARED);
 		return;
 	}
@@ -787,53 +761,12 @@ xfs_inode_item_pushbuf(
 	bp = xfs_incore(mp->m_ddev_targp, iip->ili_format.ilf_blkno,
 		    iip->ili_format.ilf_len, XBF_TRYLOCK);
 
-	if (bp != NULL) {
-		if (XFS_BUF_ISDELAYWRITE(bp)) {
-			/*
-			 * We were racing with iflush because we don't hold
-			 * the AIL lock or the flush lock. However, at this point,
-			 * we have the buffer, and we know that it's dirty.
-			 * So, it's possible that iflush raced with us, and
-			 * this item is already taken off the AIL.
-			 * If not, we can flush it async.
-			 */
-			dopush = ((iip->ili_item.li_flags & XFS_LI_IN_AIL) &&
-				  !completion_done(&ip->i_flush));
-			iip->ili_pushbuf_flag = 0;
-			xfs_iunlock(ip, XFS_ILOCK_SHARED);
-
-			trace_xfs_inode_item_push(bp, _RET_IP_);
-
-			if (XFS_BUF_ISPINNED(bp))
-				xfs_log_force(mp, 0);
-
-			if (dopush) {
-				int	error;
-				error = xfs_bawrite(mp, bp);
-				if (error)
-					xfs_fs_cmn_err(CE_WARN, mp,
-		"xfs_inode_item_pushbuf: pushbuf error %d on iip %p, bp %p",
-							error, iip, bp);
-			} else {
-				xfs_buf_relse(bp);
-			}
-		} else {
-			iip->ili_pushbuf_flag = 0;
-			xfs_iunlock(ip, XFS_ILOCK_SHARED);
-			xfs_buf_relse(bp);
-		}
-		return;
-	}
-	/*
-	 * We have to be careful about resetting pushbuf flag too early (above).
-	 * Even though in theory we can do it as soon as we have the buflock,
-	 * we don't want others to be doing work needlessly. They'll come to
-	 * this function thinking that pushing the buffer is their
-	 * responsibility only to find that the buffer is still locked by
-	 * another doing the same thing
-	 */
-	iip->ili_pushbuf_flag = 0;
 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
+	if (!bp)
+		return;
+	if (XFS_BUF_ISDELAYWRITE(bp))
+		xfs_buf_delwri_promote(bp);
+	xfs_buf_relse(bp);
 	return;
 }
 
@@ -937,7 +870,6 @@ xfs_inode_item_init(
 	/*
 	   We have zeroed memory. No need ...
 	   iip->ili_extents_buf = NULL;
-	   iip->ili_pushbuf_flag = 0;
 	 */
 
 	iip->ili_format.ilf_type = XFS_LI_INODE;
diff --git a/fs/xfs/xfs_inode_item.h b/fs/xfs/xfs_inode_item.h
index cc8df1ac7783..9a467958ecdd 100644
--- a/fs/xfs/xfs_inode_item.h
+++ b/fs/xfs/xfs_inode_item.h
@@ -144,12 +144,6 @@ typedef struct xfs_inode_log_item {
 						      data exts */
 	struct xfs_bmbt_rec	*ili_aextents_buf; /* array of logged
 						      attr exts */
-	unsigned int            ili_pushbuf_flag;  /* one bit used in push_ail */
-
-#ifdef DEBUG
-	uint64_t                ili_push_owner;    /* one who sets pushbuf_flag
-						      above gets to push the buf */
-#endif
 #ifdef XFS_TRANS_DEBUG
 	int			ili_root_size;
 	char			*ili_orig_root;
diff --git a/fs/xfs/xfs_trans.h b/fs/xfs/xfs_trans.h
index ca64f33c63a3..c93e3a102857 100644
--- a/fs/xfs/xfs_trans.h
+++ b/fs/xfs/xfs_trans.h
@@ -861,8 +861,7 @@ typedef struct xfs_item_ops {
 #define	XFS_ITEM_SUCCESS	0
 #define	XFS_ITEM_PINNED		1
 #define	XFS_ITEM_LOCKED		2
-#define	XFS_ITEM_FLUSHING	3
-#define XFS_ITEM_PUSHBUF	4
+#define XFS_ITEM_PUSHBUF	3
 
 /*
  * This structure is used to maintain a list of block ranges that have been
diff --git a/fs/xfs/xfs_trans_ail.c b/fs/xfs/xfs_trans_ail.c
index d7b1af8a832d..e799824f7245 100644
--- a/fs/xfs/xfs_trans_ail.c
+++ b/fs/xfs/xfs_trans_ail.c
@@ -253,6 +253,7 @@ xfsaild_push(
 	int		flush_log, count, stuck;
 	xfs_mount_t	*mp = ailp->xa_mount;
 	struct xfs_ail_cursor	*cur = &ailp->xa_cursors;
+	int		push_xfsbufd = 0;
 
 	spin_lock(&ailp->xa_lock);
 	xfs_trans_ail_cursor_init(ailp, cur);
@@ -308,6 +309,7 @@ xfsaild_push(
 			XFS_STATS_INC(xs_push_ail_pushbuf);
 			IOP_PUSHBUF(lip);
 			last_pushed_lsn = lsn;
+			push_xfsbufd = 1;
 			break;
 
 		case XFS_ITEM_PINNED:
@@ -322,12 +324,6 @@ xfsaild_push(
 			stuck++;
 			break;
 
-		case XFS_ITEM_FLUSHING:
-			XFS_STATS_INC(xs_push_ail_flushing);
-			last_pushed_lsn = lsn;
-			stuck++;
-			break;
-
 		default:
 			ASSERT(0);
 			break;
@@ -374,6 +370,11 @@ xfsaild_push(
 		xfs_log_force(mp, 0);
 	}
 
+	if (push_xfsbufd) {
+		/* we've got delayed write buffers to flush */
+		wake_up_process(mp->m_ddev_targp->bt_task);
+	}
+
 	if (!count) {
 		/* We're past our target or empty, so idle */
 		last_pushed_lsn = 0;
-- 
cgit v1.2.3