Different view on RH0: it is good to take out unmaintained networks

[Pekka Savola]

(Dropped off ipv6 at ietf.org, as this seems operational, and 
crossposting isn't very nice.)

On Mon, 14 May 2007, James Jun wrote:
...
> It's not a good thing when people ask you to do uRPF, and best your router
> can do for IPv6 is only strict-mode uRPF which doesn't work too well with
> multihomed customers or on peering and transit interfaces.  :)

(I've seen many generalizations like this over the years, but I'd like 
to see more detailed explanations why this is a problem.)

FWIW, we use strict-mode uRPF (with Juniper's Feasible paths toggle) 
successfully on multihomed customers.  Yes, this wouldn't work if the 
customer wanted to advertise only subset of their prefixes on our 
link, and another subset on the other link (or some other wacky TE 
scenarios), but we only accept consistent advertisements so we're not 
affected by this.

As Loose RPF doesn't really help with peering and transit interfaces 
in any case, you will need to set up manual ingress and egress 
filters.

On peering and upstream borders, we block packets leaving our network 
with source addresses that we don't advertise out with BGP (don't 
belong to us or our customers) *), and block packets entering our 
network with source addresses out of our addresses (excluding 
multihomed customers).  Works just fine and no RH0 problem :-)

*) with the exception of the address we use on the peering/transit 
point-to-point link.

Some of this is discussed in draft-savola-bcp84-urpf-experiences, some 
of it in draft-savola-rtgwg-backbone-attacks.

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings