On Tue, Oct 09, 2012 at 05:59:06PM -0700, Chuck Silvers wrote:
> > [...]
> > with a 'cat big_file > /dev/null'
> > writes are still limited to 64k ...
>
> I would hope that cat'ing a file to /dev/null wouldn't result in any writes.
> :-)
> I assume you meant 'cat big_file > other_file' ?
I use: dd if=/dev/zero of=bigfile bs=1g count=7
>
> if so, then the reason for the 64k writes would be this block of code in
> ffs_write():
>
> if (!async && oldoff >> 16 != uio->uio_offset >> 16) {
> mutex_enter(vp->v_interlock);
> error = VOP_PUTPAGES(vp, (oldoff >> 16) << 16,
> (uio->uio_offset >> 16) << 16,
> PGO_CLEANIT | PGO_JOURNALLOCKED | PGO_LAZY);
> if (error)
> break;
> }
>
that's it. I did s/16/32/g in the code above and now I get 128k writes.
>
> there's a similar block in many file systems at this point. when I wrote that
> I intended to replace it with something better before very long, but I just
> never
> got back to it, alas.
I'm not sure what the best way to handle this would be.
If we assume that maxphys is a power of 2, we could use a maxphys-derived
mask here. Otherwise, maybe we should compute and cache the largest power-of-2
value below maxphys in v_mount, as is done in vfs_vnops.c:vn_ra_allocctx()
(actually, this would remove ra_iochunk as we could use the mount point's
value)
--
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
NetBSD: 26 ans d'experience feront toujours la difference
--