Re: Routing issue: several interfaces with the same inet6 prefix

On Sun, Jun 05, 2005 at 09:03:54PM +0200, Olivier Mehani wrote:
| This is not an OpenBSD-only problem, but I'm facing it will trying to
| configure an OpenBSD based router.
|
| I have the followind network configuration:
|
|           sis0 /-----------\ sis1
| (outside) -----|OpenBSD box|------ (local wired net)
|                \-----------/
|                     rtw0 \.oOo. (local wireless net)
|
| With this in sysctl.conf:
| net.inet6.ip6.forwarding=1
| net.inet6.ip6.accept_rtadv=1
|
| sis0's IPv6 is configured via stateless autoconfiguration, with a /64
| prefix but (first problem) doesn't get a default route (I suspect this
| is due to net.inet6.ip6.forwarding).
|
| The machine is then advertising the same prefix on sis1 and rtw0 to
| autoconfigure the boxen behind the router. They properly get their
| autoconf IP and default routes.
|
| (second and main problem) The router correctly passes request from the
| local net to the outside, but then doesn't know where to send them back.
| The following type of packets is sent on sis0:
|
| 20:52:28.945616 (THE_ROUTERS_IPV6) > ff02::1:ff8b:234a: icmp6:
| neighbor sol: who has (THE_CLIENT_ON_rtw0)
|
| "8b:234a" is part of the client's autoconfigured IPv6.
|
| How can I make the router send the packets to the correct interface ? Is
| there a way to update the routing tables when receiving packets from the
| hosts ?
|
| I guess the same problem will arrise when trying to communicate from a
| machine on rtw0's net to a machine on sis1's net.
|
| Any ideas ?

I think you should be looking at bridging or at getting more address
space. You shouldn't

a) use router sollicitation on a(n IPv6) router
b) distribute one IPv6 network over separate logical networks

To solve a) simply configure your interface with eui64 and add a
default IPv6 gateway. To solve b) either get more address space and
split that over your logical networks (this can be done without the
extra address space but then you will not be able to use rtadvd) or
bridge your interfaces to put them in the same logical network.

I'd suggest to get more address space. Your setup looks somewhat
awkward with the same IPv6 subnet on all your interfaces. You are
advertising the same prefix on sis1 and rtw0 as your v6-provider sent
you on sis0 - baaaaaad.

You shouldn't be able to configure your system with both
net.inet6.ip6.forwarding=1 and net.inet6.ip6.accept_rtadv=1, these are
conflicting setting one will disable the other and vice versa. Don't
do that.

If you want to get really nasty (but (IMO) less so then what you're
doing now) then you could use pf(4) to NAT for IPv6 and use
(deprecated) site-local addresses on sis1 and rtw0 :

/etc/hostname.sis0:
inet6 <YOUR_PREFIX>:: 64 eui64
!/bin/route add -inet6 default <GATEWAY_ADDRESS>

/etc/hostname.sis1:
inet6 fec0:1:: 64 eui64

/etc/hostname.rtw0:
inet6 fec0:2:: 64 eui64

/etc/pf.conf:
nat on sis0 inet6 from fec0::/16 -> <IPV6_ON_SIS0>

/etc/rc.conf.local:
rtadvd_flags="sis1 rtw0"

Cheers,

Paul 'WEiRD' de Weerd

--
>++++++++[<++++++++++>-]<+++++++.>+++[<------>-]<.>+++[<+
+++++++++++>-]<.>++[<------------>-]<+.--------------.[-]
                 http://www.weirdnet.nl/

[demime 1.01d removed an attachment of type application/pgp-signature]