Hi, Geoff At Tue, 27 Apr 2010 14:07:30 +1000, Geoff Huston wrote: > > I dug out the original exchange from 3 June 2007 and I think you were > proposing some complications to the mechanism at the time....
No doubt. Thanks for the archives, I had forgotten some of the details, but I still think that we (for some definition of "we") went a bit overboard at the time. You're correct that perfect avoidance of all possible replay would require a counter or clock that increments at least as fast as the signing rate, whatever that might be. On the other hand, for the one attack I know of at the moment (replay of an old issuance request with a compromised key after that key has been revoked), a clock check with one-second granularity is almost certainly good enough: for this attack to work in this case, the revocation request would have to come less than a second after the initial issuance request, which seems unlikely. The nanosecond timer isn't quite the same thing as the sequence number I was talking about yesterday: the sequence number would be intended to catch dropped messages, a timer wouldn't do that. Interesting that Steve Bellovin seems to have been advocating the clock-based approach back when as a way of avoiding state; he was correct about that, of course, but he's also the person who suggested the sequence number this time, so I may need to get him into an argument with himself to find out who wins that one. In any case: I still suspect that the trivial timer check I proposed yesterday would suffice for the one attack we know about today. _______________________________________________ sidr mailing list [email protected] https://www.ietf.org/mailman/listinfo/sidr
