[Top] [All Lists]

Re: Kernel 2.6 IPV6 Busted

To: Quantum Scientific <Info@xxxxxxxxxxxxxxx>
Subject: Re: Kernel 2.6 IPV6 Busted
From: Andre Tomt <andre@xxxxxxxx>
Date: Tue, 01 Mar 2005 22:50:25 +0100
Cc: netdev@xxxxxxxxxxx
In-reply-to: <>
References: <> <> <>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 1.0 (Windows/20041206)
Quantum Scientific wrote:
On Sunday 27 February 2005 11:40, Andre Tomt wrote:
You seem to be fixed on the idea that a ipv6 stack has to have stateful firewalling, or else its utter crap, correct? :-)

No, I'll try to say this clearer.

The stack works fine in. And out. But for a useful virtual circuit you must have something like connection tracking.

Remember what my issue is: - I have a very tight firewall,
- I ping6 out,
- The firewall blocks the reply back, because the connection is stateless!

Never, ever, filter ICMP. Or at least be extremely careful doing so. You may end up breaking things like PMTU and error notification mechanisms.

- Same with http, etc.

This means that I have to open for incoming, virtually every port I send outgoing to, or else I do not get any replies. This is what I call non-functional, because one does not open incoming ports, for the most part.

Why are you not having this problem?

Because I tend to use the oldskool way of doing it when there is not other option, by matching on SYN. It's a bit trickier with UDP, but doable for most UDP based protocols.

Also on a per-system basis I tend to prefer to secure services rather than firewall them; by for example just shutting them off/uninstalling them if not used, binding to localhost, use tcpwrappers.. that sort of thing.

Don't get me wrong; I'd *love* to see connection tracking integrated with ipv6 netfilter. It would simplify some of my setups greatly. But it would also be out of the question on a lot of my other setups; as connection tracking is a *severe* bottleneck when faced with any real amounts of load.

It's not The universal solution, and the lack of it is not *that* bad.

Connection tracking is on the way, currently a implementation exists in the patch-o-matic svn.

Is this reasonably solid? Does this operate on Layer 3, rather than Layer 2?

It operates like the IPv4 state matches. Solid? Well, I guess testers are welcome :)

<Prev in Thread] Current Thread [Next in Thread>