[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-tiny
Subject:    NFS problems (nfsroot actually)
From:       mpm () selenic ! com (Matt Mackall)
Date:       2004-07-28 11:24:11
Message-ID: 20040728162407.GA16310 () waste ! org
[Download RAW message or body]

On Wed, Jul 28, 2004 at 05:28:42PM +0300, Vassilis Virvilis wrote:
> I wouldn't report this without checking with vanilla 2.6.7 but since 
> both kernels behave erratically but in a different way I thought it 
> would be nice to report it.
> 
> The goal is to boot a Cyrix486/8MB from the network. It has a 10Mbit ISA 
> card while the rest of the network is on 100Mbit cards.
> 
> 2.4.27pre2 UDP works so we know that the setup is not screwed.
> 
> 2.6.7 tiny NFSroot, UDP
> hangs when it tries to connect to portmapper to get the ports for the 
> mount services. The portmapper answers but somehow the 486 drops the 
> answer, timeouts, and then continues to default ports for mountd. At 
> that point my firewall blocks any further communication.
> 
> I run ethereal and I have evidence that the NFS server does respond to 
> the portmapper request. It seems however that the client doesn't listen 
> properly.

I presume you're running ethereal on the server. Is there any way to
boot userspace on the client, run tcpdump, and then do an NFS mount?
 
> I have read a thread in LKML saying that UDP is not reliable especially 
> in 10/100 combination. I do not know enough in order to disagree with 
> this statement but I cannot wonder why 2.4.x works. Is 2.6 so faster 
> that triggers the NFS design bugs? Weird.

It is going to be fairly dependent on your switch.

> 2.6.7 vanilla NFSroot, UDP
> I boots alright, it mounts the / partition but it hangs when it tries to 
> mount the other exports /usr etc... Looks like dropped packets again.
> 
> 2.6.7 tiny NFSroot, TCP
> kernel panics, NULL pointer dereferences etc... I think every time 
> different crash, but I am not 100% sure.

If you can get netconsole or serial console working and capture some
of these oopses, it'd be useful.

Also, you can try turning off all the tiny features (which essentially
gets you back to mainline) and turn on a bunch at a time until it
starts misbehaving to narrow down the problem.

-- 
Mathematics is the supreme nostalgia of our time.

[prev in list] [next in list] [prev in thread] [next in thread]