I think you should allow it as an explicit case. That is, you should block unknown IP addresses as a general policy, but allow it by static entry whenever anyone asks. As the direct access provider, you should be responsible for blocking spoofing. Your BGP peers, recognizing you as an ISP should accept that you are a significant provider and allow you to manage your own policies. Alternatively, they should ask you what your policies are, and block spoofing from your network only if you are not blocking spoofing yourself. As part of our business, we manage a voluntary community of test points, which primarily do remote traceroutes for the other members of the community. I've looked at a lot of outbound routes, and I can verify that your 5% number is in the right ballpark. Maybe a little high, but reasonable. Unconventional outbound routing policies are not always as clear as the simple case that you gave. I've seen outbound traffic split by an ISP between two upstreams, apparently based on some static mapping of destination IP address (not based on BGP routes...) I've also seen all outbound traffic routed to one backbone AS where the source addresses are part of a different backbone AS. But if you want to see something really weird, try this looking glass: http://lookingglass.glassberg.org/cgi-bin/lg.pl Trace to one of our addresses, if you like: 192.35.156.23 They seem to split their outbound traffic in some random way between Genuity and UU Net. If anyone can tell me what they are doing, I'd love to know. In any case, weirdness abounds. Accomodating customers who are weird is part of life, as long as they are willing to tell you what they are trying to do. Steve Schaefer Dashbit - The Leader In Internet Topology www.dashbit.com www.traceloop.com On Sun, 27 May 2001, Hank Nussbacher wrote:

I have recently been researching anti-spoofing for two ISPs and have an operational question or two.

We have found that about 5% of sites are connected to 2 ISPs but do not use BGP. They use two different sets of IP addresses and point default to one of the two ISPs. The return traffic will of course go to them via one of the two ISPs, but if you have anti-spoofing filters set up or try to set it up now, you will break their outgoing traffic.

Reasons to allow it:

1) Here we have multihomed customers, who are not eating up ASN space and are not asking for PI space and are happy the way they are working. By stopping them, we will force this 5% to ask for ASNs and PI space. So for the general well-being of the Internet - why not just let them be. 2) Anti-spoofing is set up to stop attacks from unknown IPs (RFC1918) or from an IP that doesn't belong to you. In this case, the IP can be traced back to the user (via ISP #2). 3) If you block it, the customer will leave and go to another ISP that does not block these IP addresses.

Reasons to not allow it:

1) If ISP #1 has blocked the customer due to being an open mail relay (example), and then that customer just sends the traffic out via ISP #2 (using ISP #1 IPs), they have circumvented the filter and blame will be placed on ISP #1 for not stopping an open email relay (this has actually happened once before). 2) I should not be announcing traffic for IPs that I am not announcing routing updates.

I am curious if others have found this 5% occurence and I am curious why no one else has raised this issue before. Could it be that almost no one is running uRPF and/or anti-spoofing filters?

Thanks, Hank