Subject: Re: hangs with NFS
To: None <>
From: Thilo Manske <Thilo.Manske@HEH.Uni-Oldenburg.DE>
List: current-users
Date: 03/28/2001 21:10:31
On Thu, Mar 29 2001 at 03:16:39 +0900, Izumi Tsutsui wrote:
> In <>
> wrote:
> > with recent -current kernels I experience strange hangs when NFS is used.
> > NFS server is a NetBSD 1.5, client -current as of today. My home dir is on
> > the server.
>  :
> > Anyone else seen this ?
> I'm annoyed by the same problem, too.
> My server is NetBSD/arm32 1.5S (CATS) and
> NetBSD/hp300 1.5S and NetBSD/newsmips 1.5S hang
> when they try to read files on the NFS server.
> news68k, i386, pmax and macppc (they are also 1.5S) do not
> have the problem, and 1.5.1_ALPHA i386 also hangs on access to
> 1.5S i386 server. Before upgrading server to 1.5S from 1.5,
> there was no problem, so I guess it is not client problem.
(Since you mention newsmips.)
Here's my collection of NFS problems, maybe related or not.
I was to lazy to investigate them further:

I had (have) some NFS problems as well with newsmips as client, they started
after I upgraded my NFS server (i386) from 1.5BETA2 to 1.5S as well.  Old
newsmips kernels (befor 1.5 was released) still work fine.

My workaround is NFS_BOOT_RWSIZE=1024 in the kernel config file
(for diskless boots) and option -r=1024 for nfs_mount, that get's it partly
working (the newsmips machine crashes after some filesystem activity (e.g
make includes), but maybe that's not related to this problem).

My sparcs still work without that, but one (a SLC) shows now a *long* delay
during boot, the others not, but I haven't rebooted all of them to try that.
After boot everything seems to be ok.

My arm32 (RISC PC) had dying NFS mounts as well, but a patch from Mike Pumfort
for the network interface driver (ne - See ) got it going.
The sources have changed now, so you have to aplly some parts of the patch

I currently have problems with -current and my diskless vax, but an old
GENERIC kernel from 1.5 can stay up for a week.  (-current crashes during
boot, but my current vax is to slow to investigate that (kernel compilation
needs a week and there's no vax-crosscompilation package :-(), I'll get a
faster vax in a few days.)

My alpha, i386, macppc didn't show any oddities.

Too bad I don't have a 1.5ish sytem around anymore that I can use as boot
and file server to see if that could "fix" the problems with newsmips and
vax. Maybe 1.5S as NFS server triggers a bug for some architectures?

Dies ist Thilos Unix Signature! Viel Spass damit.