> > AFAIU, you are essentially proposing to perform source address > > based routing by the hosts and by the routers in a multiprefix > > site, is that correct? > > I don't like the term, because I first do a destination lookup and > only look up the source address in certain cases. Kind of like the > previous comment on source routing, which in IEEE 802.5, DSR, and RFC > 791 IP means that the source specifies all or part of the routing > path. I think the term mis-states the case. > > But yes, in certain cases where there is a multipath route, the point > is that if the datagram is handed to the wrong ISP and the ISP is > doing ingress filtering, the datagram will be dropped, and hence I > suggest that we direct it toward the right ISP.
But still, you appear to propose to fundamentally change the forwarding process by inserting a relatively complex lookup operation into per-packet forwarding, and I think such a change needs quite a bit more in terms of justification. Instead, my inclination would be to "solve" this problem in a much simpler manner, simply by declaring it a configuration error. A site which receives prefixes from more than a single provider is clearly multihomed, and needs to have its providers make appropriate exceptions to a strict "I will only accept packets with source addresses from within the prefix I delegate" rule. Either that, or the domain in question needs to ensure via a combination of address selection and routing policy that one avoids being subjected to (presumably unwanted) RPF failures. Regards, - Håvard -------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
