Patch to issue pure flushes directly (Was: Re: [PATCH v6 0/4] ext4: Coordinate data-only flush requests sent) by fsync

[Ted Ts'o <tytso@mit.edu>]

This patch seemed like a good idea to me; I just checked linux-next,
and looks like nothing like this is planned to be merged.  Just
thought I would send a prod-o-gram to see what the current thinking
was around adding something like this.

Thanks,

						- Ted


On Tue, Nov 30, 2010 at 05:41:24PM +0100, Christoph Hellwig wrote:
> FYI, here's an updated version of my patch to not run pure flushes
> that also works on SCSI/ATA and not just virtio.  I suspect the patch
> alone might not be enough, but together with a variant of Neil's
> suggestion might do the trick, with my patch taking care of the
> highend-devices and Neil's scheme of taking care of stupid SATA disks.
> 
> 
> Index: linux-2.6/block/blk-flush.c
> ===================================================================
> --- linux-2.6.orig/block/blk-flush.c	2010-11-30 17:27:33.108254088 +0100
> +++ linux-2.6/block/blk-flush.c	2010-11-30 17:27:38.790004333 +0100
> @@ -143,6 +143,17 @@ struct request *blk_do_flush(struct requ
>  	unsigned skip = 0;
>  
>  	/*
> +	 * Just issue pure flushes directly.
> +	 */
> +	if (!blk_rq_sectors(rq)) {
> +		if (!do_preflush) {
> +			__blk_end_request_all(rq, 0);
> +			return NULL;
> +		}
> +		return rq;
> +	}
> +
> +	/*
>  	 * Special case.  If there's data but flush is not necessary,
>  	 * the request can be issued directly.
>  	 *
> Index: linux-2.6/drivers/scsi/scsi_lib.c
> ===================================================================
> --- linux-2.6.orig/drivers/scsi/scsi_lib.c	2010-11-30 17:27:33.120254298 +0100
> +++ linux-2.6/drivers/scsi/scsi_lib.c	2010-11-30 17:27:38.791003634 +0100
> @@ -1060,7 +1060,7 @@ int scsi_setup_blk_pc_cmnd(struct scsi_d
>  	 * that does not transfer data, in which case they may optionally
>  	 * submit a request without an attached bio.
>  	 */
> -	if (req->bio) {
> +	if (req->bio && !(req->cmd_flags & REQ_FLUSH)) {
>  		int ret;
>  
>  		BUG_ON(!req->nr_phys_segments);