On 03/01/16(Sun) 23:16, Martin Pieuchot wrote:
> I don't understand why the check for reading ahead one block has been
> kept for clustering read when the Dynamic buffer cache got introduced.
>
> Since bread_cluster() replaces cluster_read() I believe this can go
> away.
Any ok?
> Index: ufs/ffs/ffs_vnops.c
> ===================================================================
> RCS file: /cvs/src/sys/ufs/ffs/ffs_vnops.c,v
> retrieving revision 1.80
> diff -u -p -r1.80 ffs_vnops.c
> --- ufs/ffs/ffs_vnops.c 14 Mar 2015 03:38:52 -0000 1.80
> +++ ufs/ffs/ffs_vnops.c 3 Jan 2016 21:42:22 -0000
> @@ -242,14 +242,11 @@ ffs_read(void *v)
>
> if (lblktosize(fs, nextlbn) >= DIP(ip, size))
> error = bread(vp, lbn, size, &bp);
> - else if (lbn - 1 == ip->i_ci.ci_lastr) {
> + else
> error = bread_cluster(vp, lbn, size, &bp);
> - } else
> - error = bread(vp, lbn, size, &bp);
>
> if (error)
> break;
> - ip->i_ci.ci_lastr = lbn;
>
> /*
> * We should only get non-zero b_resid when an I/O error
>
