xfs: don't use vfs writeback for pure metadata modifications

Under heavy multi-way parallel create workloads, the VFS struggles
to write back all the inodes that have been changed in age order.
The bdi flusher thread becomes CPU bound, spending 85% of it's time
in the VFS code, mostly traversing the superblock dirty inode list
to separate dirty inodes old enough to flush.

We already keep an index of all metadata changes in age order - in
the AIL - and continued log pressure will do age ordered writeback
without any extra overhead at all. If there is no pressure on the
log, the xfssyncd will periodically write back metadata in ascending
disk address offset order so will be very efficient.

Hence we can stop marking VFS inodes dirty during transaction commit
or when changing timestamps during transactions. This will keep the
inodes in the superblock dirty list to those containing data or
unlogged metadata changes.

However, the timstamp changes are slightly more complex than this -
there are a couple of places that do unlogged updates of the
timestamps, and the VFS need to be informed of these. Hence add a
new function xfs_trans_ichgtime() for transactional changes,
and leave xfs_ichgtime() for the non-transactional changes.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>

3 files changed, 2 insertions, 42 deletions

diff --git a/fs/xfs/linux-2.6/xfs_ioctl.c b/fs/xfs/linux-2.6/xfs_ioctl.c
index 03aa908a9cb9..10206be7a077 100644
--- a/fs/xfs/linux-2.6/xfs_ioctl.c
+++ b/fs/xfs/linux-2.6/xfs_ioctl.c
@@ -1088,8 +1088,8 @@ xfs_ioctl_setattr(
 		xfs_diflags_to_linux(ip);
 	}
 
+	xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
-	xfs_ichgtime(ip, XFS_ICHGTIME_CHG);
 
 	XFS_STATS_INC(xs_ig_attrchg);
 
diff --git a/fs/xfs/linux-2.6/xfs_iops.c b/fs/xfs/linux-2.6/xfs_iops.c
index b1fc2a6bfe83..a788f016d1fa 100644
--- a/fs/xfs/linux-2.6/xfs_iops.c
+++ b/fs/xfs/linux-2.6/xfs_iops.c
@@ -95,41 +95,6 @@ xfs_mark_inode_dirty(
 }
 
 /*
- * Change the requested timestamp in the given inode.
- * We don't lock across timestamp updates, and we don't log them but
- * we do record the fact that there is dirty information in core.
- */
-void
-xfs_ichgtime(
-	xfs_inode_t	*ip,
-	int		flags)
-{
-	struct inode	*inode = VFS_I(ip);
-	timespec_t	tv;
-	int		sync_it = 0;
-
-	tv = current_fs_time(inode->i_sb);
-
-	if ((flags & XFS_ICHGTIME_MOD) &&
-	    !timespec_equal(&inode->i_mtime, &tv)) {
-		inode->i_mtime = tv;
-		sync_it = 1;
-	}
-	if ((flags & XFS_ICHGTIME_CHG) &&
-	    !timespec_equal(&inode->i_ctime, &tv)) {
-		inode->i_ctime = tv;
-		sync_it = 1;
-	}
-
-	/*
-	 * Update complete - now make sure everyone knows that the inode
-	 * is dirty.
-	 */
-	if (sync_it)
-		xfs_mark_inode_dirty_sync(ip);
-}
-
-/*
  * Hook in SELinux.  This is not quite correct yet, what we really need
  * here (as we do for default ACLs) is a mechanism by which creation of
  * these attrs can be journalled at inode creation time (along with the
diff --git a/fs/xfs/linux-2.6/xfs_super.c b/fs/xfs/linux-2.6/xfs_super.c
index a4e07974955b..83154c0a3175 100644
--- a/fs/xfs/linux-2.6/xfs_super.c
+++ b/fs/xfs/linux-2.6/xfs_super.c
@@ -972,12 +972,7 @@ xfs_fs_inode_init_once(
 
 /*
  * Dirty the XFS inode when mark_inode_dirty_sync() is called so that
- * we catch unlogged VFS level updates to the inode. Care must be taken
- * here - the transaction code calls mark_inode_dirty_sync() to mark the
- * VFS inode dirty in a transaction and clears the i_update_core field;
- * it must clear the field after calling mark_inode_dirty_sync() to
- * correctly indicate that the dirty state has been propagated into the
- * inode log item.
+ * we catch unlogged VFS level updates to the inode.
  *
  * We need the barrier() to maintain correct ordering between unlogged
  * updates and the transaction commit code that clears the i_update_core