Subject: Re: question: forcing IPV6 traffic down a IPV6/IPV4 tunnel
To: Jun-ichiro itojun Hagino <itojun@iijlab.net>
From: Robert Elz <kre@munnari.OZ.AU>
List: tech-net
Date: 03/11/2000 21:09:57
    Date:        Sat, 11 Mar 2000 18:07:20 +0900
    From:        Jun-ichiro itojun Hagino <itojun@iijlab.net>
    Message-ID:  <1419.952765640@lychee.itojun.org>

  | 	Are you okay with having tunnel's link-local address written in
  | 	the config file?

That's better than using a global address.   But using the interface
name is easier to manage and simpler.   Note that kernel interface
doesn't have to have a way to pass around interface names, there's
no reason why the route command given something like

	route add default -int gif0

can't go find what address is associated with gif0 and pass that in
as the default route destinatin (using the link-local addr if
appropriate, so it is immune to global addressing changes, and there
doesn't need to be a processes sitting watching for renumber events).

  | 	We can make it robust against ethernet card swaps, by
  | 	getting gif's link-local address by using
  | 	# route add -inet6 default `ifconfig gif0 | sed -e whatever`.

Just having the route command do the work would be easier.

  | 	Do you think we should try convincing ISPs to exchange routes with
  | 	downstream using routing protocols, rather than static route
  | 	configuration?

Yes, but that's a separate issue.  Even when the ISP is willing, some
customers that are on "pay by the byte" links aren't going to want that
extra traffic.   So, we should be looking for ways to avoid configuring
explicit IPv6 addresses everywhere possible - IPv6 addressing is
basically there already, it woudl be a shame to break it for the sake of
a page of code in the route command (and similar places).

kre