default behavior & skb_linearize

default behavior & skb_linearize

[Brad Parker]

Part of this may be better for the users list, but part of it is internals
so bear with me... (or is that "bare" with me? :-)

I want to so some work with iptables in a 2.4.20 kernel.  I'm booting
with a NFS root.  The stock kernel works fine and the nfs client works
fine.

But, when I turn on netfilter, the nfs client stops workings - sort of.

aside: I am assuming that if I turn on netfilter and connection tracking
the "default" (i.e. no external rules loaded via iptables) should pass
everything and the kernel nfs client should continue to work.  This is
true, yes?

I can send the interesting parts of my .config if needed.

It looks to me like nfs continues to work but the nfs read response
udp fragements cause some sort of problem...

I attempt to short circuit nf_hook_slow and this works, but only if I
comment out the call to "skb_linearize".  Any idea why that is?

(it's obvious, but I'll point out that if I short circuit nf_hook_slow -
by calling okfn and returning - the nfs client works fine)

I'll keep digging but I thought there might be something obvious I'm
missing...

-brad

Re: default behavior & skb_linearize

[Patrick Schaaf]

> I want to so some work with iptables in a 2.4.20 kernel.

Try a current kernel. Apply patch-o-matic/pending from CVS.

> aside: I am assuming that if I turn on netfilter and connection tracking
> the "default" (i.e. no external rules loaded via iptables) should pass
> everything and the kernel nfs client should continue to work.  This is
> true, yes?

An emphatic YES: that's how it _should_ be.

> I can send the interesting parts of my .config if needed.

Only with a current kernel, I fear. 2.4.20, as far as I remember,
is supposed to be broken in the iptables area.

best regards
  Patrick

Re: default behavior & skb_linearize

[Patrick McHardy]

Brad Parker wrote:

>But, when I turn on netfilter, the nfs client stops workings - sort of.
>

Are you using RTL8139 NICs ?