Re: [tcpm] tcp-auth-opt issue: support for NATs

["Adam Langley" <[e-mail]>]

On Tue, Aug 5, 2008 at 7:02 PM, Eric Rescorla <[e-mail]> wrote:
> Isn't the whole point of the key-ids to avoid requiring this?

Well, keyids would be required too. Here's how I see it working.

Define a synchronisation range, R, say 1 minute. We require that the
maximum skew between two clocks is R.

Define a key rotation timespan, T, say 1 minute. Let N be the maximum
number of keys that exist in a timespan of 2R. In our example, this is
two.

Define a function K(master, t) that maps a timestamp to a key derived
from a master key and a keyid in [0, N). For example, F(master, t) =
SHA1(master + uint32be (t/T)) with keyid (t/T) `mod` N.

At any time, a client transmits with key and keyid from F(master,
currentTime). At any time, the server accepts SYN signed with any
keyid and key in a span of 2R centered around the current time.

So one can only connect if one's clock skew is acceptable and one
knows the master key.

The keys for the resulting connection would be connection specific,
probably hashing in the SEQ/ACK numbers, other stuff etc.


AGL

-- 
Adam Langley [e-mail] http://www.imperialviolet.org
_______________________________________________
tcpm mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/tcpm