Subject: Re: IPv6 autoconfig for a multi-interface host
To: Ignatios Souvatzis <>
From: Bill Studenmund <>
List: tech-net
Date: 01/10/2005 16:58:16
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Sat, Jan 08, 2005 at 02:26:01PM +0100, Ignatios Souvatzis wrote:
> On Fri, Jan 07, 2005 at 10:51:42PM -0500, William Allen Simpson wrote:
> > Bill Studenmund wrote:
> >=20
> > >Well, not being able to use some form of autoconfig is lame. Think abo=
> > >all the motherboards with two ethernet interfaces. Hook them both up t=
> > >two nets, and v4 works.
> Bill St., are you sure? When you use "rdisc" on both interfaces? See belo=

I'm going by what the man pages say and what I can glean from the=20

> > >If v6 can't work in the same way, v6 is less=20
> > >useful than v4. :-(
> > When I originally designed IPv6 Neighbor Discovery, it was explicitly
> > able to autoconfigure on multiple interfaces as a design requirement,
> > and in the original code I wrote.
> I'm not so sure _Neighbour_ Discovery as such is broken; but I expect
> that the router discovery on top of it is broken, and can't be unbroken,
> unless we have a way to handle multiple default routes in NetBSD.
> Last I looked, we only can have one of them per address family.
> Maybe a NetBSD autohost can have address A in network a on interface ia0
> and address B on network b, interface ib0, and even chose the right source
> addresses for each of them, but what about off-link addresses? It needs to
> decide whether to use router-of(a) or router-of(b). Same problem as for
> IPv4, only that there you normally either use static configuration or DHC=
> and hopefully the dhcpd(a) and dhcpd(b) are coordinated in a way that you=
> multiple-interface host gets one working default router.
> Using routed in router-discovery-only mode for IPv4 would have the same
> sort of ambiguity, and our documentation advises that no_rdisc might be
> needed in that case.

The documentation is a bit vague about no_rdisc

> (Actually, I think IPv6 autoconfiguration should be fine for multihomed
> hosts if different router lifetime fields (e.g., all but one zero) would
> allow selection of the "best" router by our code, but I don't know if our
> code does this.)

One issue with this is we're implicitly adding policy. I understand the=20
kernel has to be able to implement it. I'd be fine with using whichever=20
route was entered first, as long as it was still valid.

> I'm not familiar with DHCPv6, so can't tell whether it (will) allow(s)=20
> to handle this situation better.

So what do we actively suggest? Just to reitterate, I want to support IPv6=
autoconfig on a multi-interface host that does not forward packets.

Take care,


Content-Type: application/pgp-signature
Content-Disposition: inline

Version: GnuPG v1.2.3 (NetBSD)