[Cluster-devel] [GFS2 Patch] GFS2: Add readahead to sequential directory traversal

[Steven Whitehouse <swhiteho@redhat.com>]

Hi,

Were we not intending to make this turn itself off in cases where it
produces no benefit? I thought the plan was to track the state via the
file descriptor in order to avoid readingahead the same blocks over and
over again too,

Steve.

On Tue, 2011-10-04 at 12:39 -0400, Bob Peterson wrote:
> Hi,
> 
> This patch adds read-ahead capability to GFS2's
> directory hash table management.  It greatly improves
> performance for some directory operations.  For example:
> In one of my file systems that has 1000 directories, each
> of which has 1000 files, time to execute a recursive
> ls (time ls -fR /mnt/gfs2 > /dev/null) was reduced
> from 2m2.814s on a stock kernel to 0m45.938s.
> 
> Regards,
> 
> Bob Peterson
> Red Hat File Systems
> 
> Signed-off-by: Bob Peterson <rpeterso@redhat.com> 
> --
>  fs/gfs2/dir.c |   33 +++++++++++++++++++++++++++++++++
>  1 files changed, 33 insertions(+), 0 deletions(-)
> 
> diff --git a/fs/gfs2/dir.c b/fs/gfs2/dir.c
> index 2045d70..9b4262e 100644
> --- a/fs/gfs2/dir.c
> +++ b/fs/gfs2/dir.c
> @@ -1376,6 +1376,37 @@ out:
>  	return error;
>  }
>  
> +static void gfs2_dir_readahead(struct inode *inode, __be64 *ht, unsigned hsize,
> +			       u32 index)
> +{
> +	struct gfs2_inode *ip = GFS2_I(inode);
> +	struct gfs2_glock *gl = ip->i_gl;
> +	struct buffer_head *bh;
> +	u64 blocknr = 0, last;
> +	unsigned count = 0;
> +
> +	while (index < hsize) {
> +		last = blocknr;
> +		blocknr = be64_to_cpu(ht[index++]);
> +		if (blocknr == last)
> +			continue;
> +		count++;
> +		if (count > 128)
> +			break;
> +		bh = gfs2_getbuf(gl, blocknr, 1);
> +		if (trylock_buffer(bh)) {
> +			if (buffer_uptodate(bh)) {
> +				unlock_buffer(bh);
> +				brelse(bh);
> +				continue;
> +			}
> +			bh->b_end_io = end_buffer_read_sync;
> +			submit_bh(READA | REQ_META, bh);
> +			continue;
> +		}
> +		brelse(bh);
> +	}
> +}
>  
>  /**
>   * dir_e_read - Reads the entries from a directory into a filldir buffer
> @@ -1406,6 +1437,8 @@ static int dir_e_read(struct inode *inode, u64 *offset, void *opaque,
>  	if (IS_ERR(lp))
>  		return PTR_ERR(lp);
>  
> +	gfs2_dir_readahead(inode, lp, hsize, index);
> +
>  	while (index < hsize) {
>  		error = gfs2_dir_read_leaf(inode, offset, opaque, filldir,
>  					   &copied, &depth,
>