[PATCH] xfs: optimize log flushing in xfs_fsync

[Alex Elder]

On Fri, 2010-02-05 at 04:57 -0500, Christoph Hellwig wrote:
> If we have a pinned inode it must have a log item attached to it.
> Usually that log item will have ili_last_lsn already set, in which
> case we only need to flush the log up to that LSN instead of doing
> a full log force.  This gives speedups of about 5% in some fsync
> heavy workloads.

This looks good.  I was going to suggest:
    ASSERT(ip->i_itemp != NULL);
before it's dereferenced, but a pinned inode had better
have an item pointer...

Reviewed-by: Alex Elder <aelder at sgi.com>

> Signed-off-by: Christoph Hellwig <hch at lst.de>
> 
> Index: xfs/fs/xfs/xfs_vnodeops.c
> ===================================================================
> --- xfs.orig/fs/xfs/xfs_vnodeops.c	2010-02-04 17:38:33.679254119 +0100
> +++ xfs/fs/xfs/xfs_vnodeops.c	2010-02-04 17:38:51.606006156 +0100
> @@ -626,8 +626,14 @@ xfs_fsync(
>  		 * force the log.
>  		 */
>  		if (xfs_ipincount(ip)) {
> -			error = _xfs_log_force(ip->i_mount, XFS_LOG_SYNC,
> -					       &log_flushed);
> +			if (ip->i_itemp->ili_last_lsn) {
> +				error = _xfs_log_force_lsn(ip->i_mount,
> +						ip->i_itemp->ili_last_lsn,
> +						XFS_LOG_SYNC, &log_flushed);
> +			} else {
> +				error = _xfs_log_force(ip->i_mount,
> +						XFS_LOG_SYNC, &log_flushed);
> +			}
>  		}
>  	} else	{
>  		/*
> 
> _______________________________________________
> xfs mailing list
> xfs at oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs