xt_gateway 20070605 (kernel)

[Amin Azez]

I apologise again for top-posting and all the sins of pocket outlook quoting.

As best as I can recall without the source in front of me a route cache item attached to the skb on routing is cast to a neighbour table entry. (they are the same waist up). 

There is no way at forward or postrouting time to tell if the skb's attached neighbour table member is (was once) a route cache entry or just a plain neighbour table entry.

This means it is not safe (hard to know when it is safe) to cast it back to a route cache item to match what you need.

Perhaps yours is a case to use realm match as Patrick has mentioned.

What you want requires a change to the routing code so that it leaves evidence that the next hop is a route entry.

BTW the ROUTE target doesn't route so much as transmit. I hope to alter it so that in PREROUTING (or later) it actually adds a route to the skb, so that postrouting will work.

Sam

-----Original Message-----
From: "Phil Oester" <kernel at linuxace.com>
To: "Jan Engelhardt" <jengelh at linux01.gwdg.de>
Cc: "Amin Azez" <azez at ufomechanic.net>; "Netfilter Developer Mailing List" <netfilter-devel at lists.netfilter.org>
Sent: 05/06/07 23:24
Subject: Re: xt_gateway 20070605 (kernel)

On Tue, Jun 05, 2007 at 01:17:57PM +0200, Jan Engelhardt wrote:
> 
> Originally from Amin Azez <azez at ufomechanic.net>,
> http://lists.netfilter.org/pipermail/netfilter-devel/2007-June/027954.html
> 
> This adds a gateway match to iptables that lets you match against the
> routed ipv4 gateway, it's very useful for SNAT if you want to avoid
> replicating your routing in your SNAT table.

Just a suggestion...for a while I've been needing the ability to match
on the routing table (but not just the gateway).  Could we perhaps name
this match 'route' instead (similar to the ROUTE target)?

Some of the things I'd like to be able to do is match on the length
of a route.  For instance we use lots of 10.x.x.x/30 nets internally
and I'd like to be able to match on them.  I haven't quite gotten around
to figuring out how to do this given the route cache doesn't include
prefix length, but I do think it would be useful.  I could see how
it could be combined with this gateway match.

Thoughts?

Phil