Les: Hi! The risk is not in the mechanism itself, assuming two routers with the correct implementation. The risk is in the fact that a compromised router can craft a packet to force the auto-config routers to continuously restart. All we’re asking for is a recognition that the risk exists. Thanks! Alvaro. On 4/7/17, 6:44 PM, "Les Ginsberg (ginsberg)" <ginsberg@xxxxxxxxx> wrote: Alvaro - Not sure how long we should debate this as the text you are asking for is modest - but one more attempt on my part - not so much to debate whether to add text or not - but to come to a common understanding. > -----Original Message----- > From: Alvaro Retana (aretana) > Sent: Friday, April 07, 2017 2:45 PM > To: Les Ginsberg (ginsberg); Robert Sparks; gen-art@xxxxxxxx > Cc: draft-ietf-isis-auto-conf.all@xxxxxxxx; ietf@xxxxxxxx; isis-wg@xxxxxxxx > Subject: Re: Genart last call review of draft-ietf-isis-auto-conf-04 > > On 4/7/17, 5:30 PM, "Les Ginsberg (ginsberg)" <ginsberg@xxxxxxxxx> wrote: > > Les: > > Hi! > > > System-id duplication is a problem for any deployment - not just > > autoconfig deployments. And it will be disruptive in any network until it is > resolved. > > > > The only thing autoconfig has added is a way to resolve this w/o > > manual intervention. This in no way increases the vulnerability nor > > the disruption the attacker can produce. (Yes - I state that quite > intentionally). > > I don’t know about Robert, but that is part of the discussion I would like to > see. > > Yes, duplicate system-ids have always been a potential problem, but this > document introduces a new de-duplication mechanism that results not just > in unsync’d databases, but in restarting adjacencies – so at least the > manifestation of the problem is different. > [Les:] The "problem" exists as long as the duplicate system-ids exist. The recovery mechanism does not introduce new problems - it actually acts to minimize the duration of the problem. One could argue (and I am NOT doing so) that automated resolution could be a benefit in all deployments. I think the issue in non-autoconfig deployments is that since systems have been manually provisioned we cannot assume that network operators want us to automatically change the config on one of their boxes. Though, who knows - maybe we will get asked to use this extension outside of autoconfig deployments. It is necessary to have such a mechanism in autoconfig deployments since there is - by default - no manual intervention. But to characterize this as adding risk is incorrect IMO. Les > > So you are asking us to repeat a discussion which has already been > > held in the context of RFC 5304 and RFC 5310. > > > > It would be more appropriate to add the normal reference to RFC > > 5304/5310 in the Security section than what you propose. > > I don’t think it hurts to add a reference to those RFCs, but they are both > about adding authentication – the problem in this document is exacerbated > by the fact that there’s no authentication by default. > > The lower layer authentication mechanisms are quite weak, specially > knowing that, if in a home environment, for example, it may be relatively > easy to connect to the WiFi network. > > Alvaro. >