On Tue, Aug 5, 2014 at 3:44 PM, Christian Huitema <[email protected]> wrote:
>> More things worth beating this horse past death with:
>>
>> - authentication is difficult to do scalably, but unauthenticated key
>>   exchange is trivial, therefore
>>
>> - having an option to do unauthenticated key exchange, as a
>>   middle-of-the-road choice between no-security and authentication, is
>>   a very good thing
>>
>> - authentication can always be added later (the charter says this!)
>>
>> Is this horse dead yet?  I think so.
>
> Absolutely. By the way, having hooks like the unique session-ID of TCP Crypt
> is essential. It allows applications to implement a simple MITM detection as
> part of an end-to-end authentication process. All applications may not
> implement that, but some will. That creates lots of uncertainty for any MITM
> attacker, because they now have a clear risk of being detected.

Yes.  We call that "channel binding".  See RFC5056.  It's important to
not get it wrong.  A unique session ID is not sufficient (e.g., TLS
session IDs are not a channel binding).

_______________________________________________
Tcpinc mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tcpinc

Reply via email to