In message <199801051756.JAA17924@cypress.nwnet.net>, Tom Killalea writes:
A regular reader of your mailing list forwarded this to me :
I will also point out that many of the recent "smurf" attacks and similar problems people are having on the net would be gone if people would just carefully filter internal/external addresses on their border machines, that is, prevent packets claiming to be from "inside" networks from coming in from the "outside", and prevent packets claiming to be from "outside" networks from going out from the "inside". The latter will stop your network from *ever* being the source of a wide variety of packet forgery attacks, and is necessary to being a good network citizen. The former will stop your network from being the subject of a wide variety fo packet forgery attacks, and is necessary to make your customers even remotely safe on the net.
There are two chances of 'upholding the address space integrity' of the Internet; assuming the current service model with Customer --> ISP ----> Internet Core The first one is on the IGP level, where the addresses assigned inside the network of the ISP is routed towards the customer. These addresses should be enforced on the interface between the ISP and the customer; and they frequently are. The major obstacle for this are scaling issues related to routing and filtering. I am network manager for a pretty much medium-sized ISP, with around 1700 internal network blocks; 600 of which come from dynamic sources. (RADIUS; variuos routing protocols). Given that a stock router will run out of filter lists long before the 600 mark I see major scaling problems here. (Outside of our network we show around 30 BGP network aggregates). This must be database driven, properly authenticicated, and fast enough to be able to track re-routing in the network. This technology does not exist, and will have to be designed, implemented on standard hardware and rolled out into production networks to get proper address integrity on the Internet. The second chance is between the ISP and the Internet Core. Here BGP is used for interaction, and the BGP aggregates should be nailed up. Filter lists to match these are relatively easy to generate, but it means that some core routers will evaluate filter lists for some 10-100 megabits of traffic. Current routers can do that up to the low two-digit megabits, so for a medium-sized ISP far outside of the US we can use this approach; but for the large players this is a non-starter.
I strongly recommend such filtering in sections 5.7 and 5.8 of my "Security Expectations for Internet Service Providers" draft ftp://ds.internic.net/internet-drafts/draft-ietf-grip-isp-02.txt and we've heard Paul plug ftp://ds.internic.net/internet-drafts/draft-ferguson-ingress-filtering-03.txt here many times.
To answer Owen comments regarding the difficulty of filtering for transit providers, I argue that filtering should happen as close to the actual hosts as possible.
Tom. -- Tom Killalea (425) 649-7417 NorthWestNet tomk@nwnet.net
-- ___ === / / / __ ___ _/_ === Morten Reistad, Network Manager === /--- / / / / /__/ / === EUnet Norway AS, Sandakerveien 64, Oslo === /___ /__/ / / /__ / === <Morten.Reistad@Norway.EU.net> === Connecting Europe since 1982 === phone +47 2209 2940