Hi Thomas, On Wed, 11 May 2011 17:12:55 -0400 Thomas Narten <nar...@us.ibm.com> wrote:
> Ole Troan <otr...@employees.org> writes: > > > That is, a CPE router with multiple LAN interfaces is not allowed to > > > advertise itself as a default router (when it loses internet > > > connectivity), when that is arguably a fine thing for it to do. > > > > > > I have to wonder whether RFC 6204 has gotten this wrong. It is very > > > CPE centric, assuming it is the only router on the home network. It > > > has conflated the notion of a router being fine as a default router > > > for the hosts attached to the same link (i.e., via RAs) and the the > > > more traditional router notion of default route. > > > > > > I.e., if the CPE router has multiple LAN interfaces, each with a ULA > > > (or other prefix) assigned to it, advertising itself as a default > > > router seems like a perfectly reasonable thing to do. > > > indeed, but unfortunately that causes IPv6 breakage. this is not at > > all CPE centric, from the CPE perspective advertising itself as a > > default router is obviously not an issue. the problem is host > > implementations that have a ULA address and a default route (but no > > connectivity to the IPv6 Internet). these hosts will choose IPv6 > > first when connecting to dual-stack destinations, this leads to > > multi-minute time outs. > > Let me be sure I understand. Which 3484 rule comes into play that has > an impact here? Is it: > > Rule 1: Avoid unusable destinations. > If DB is known to be unreachable or if Source(DB) is undefined, then > prefer DA. Similarly, if DA is known to be unreachable or if > Source(DA) is undefined, then prefer DB. > > I.e., if trying to reach a site, and the IPv6 address is "unreachable" > per above, choose the v4 address? (Sounds great in theory!). So the > CPE rule that says you MUST NOT advertise yourself as a default router > if you don't have real IPv6 connectivity is intended to result in all > destinations (other than local ones) showing up as "unreachable" when > the above rule happens? > > Is this the intention, or am I looking at the wrong scenario here? > I think what caught people a bit was a particular CPE which doesn't generate an ICMPv6 destination unreachable when the default IPv6 route disappeared or wasn't present (or didn't at the time - this caused my employer's customers a lot of grief when we started shipping IPv6 enabled by default in them). I tested this scenario with another CPE that did generate ICMPv6 destination reachables when the default route disappeared, and found that my end-node immediately reverted to using IPv4 when both AAAAs and As were available and the AAAA was offsite. However, as generating ICMPv6 dest unreachables is a SHOULD, not a MUST, and they can be rate limited, it seems to me that using RA RIOs to convey local IPv6 topology to end nodes is a more robust way of trying to ensure local IPv6 reachability across local subnets when the offsite default route disappears. So I agree with making supporting them a node requirement. Regards, Mark. -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
