Re: [Tcpcrypt] Initial questions

[Derek Fawcus <[e-mail]>]

On Fri, Jun 20, 2014 at 03:23:30PM -0700, Joe Touch wrote:
> 
> That's my view, and you've already disagreed with it; let's see what 
> others have to say.

I like the idea of the initiator being able to specify the crypto
algorithm, say an 8 bit value,  we have 1 or 2 intially defined.
If the receiver doesn't support the selected algorithm,
the whole tcpcrypt option will fail and we fallback to unencrypted.

Thereafter the initiator (application or host stack - not sure which)
can decide to terminate the connection,  and maybe retry with a different
algorithm,  continue unencryted,  or whatever.  i.e. local policy.

The idea being that we deploy with one recommended algorithm which
everyone supports,  and if a sucessful attack is found on it,  we
can switch over to the next algorithm.

I don't think the endpoints should negotiate the algorithm,
if the initiators choice is not supported,  the session is unencryted.
i.e. avoid round trips to agree/negotiate an algorithm,
and don't supply a list of choices in the initiation.


I'm not sure of the effect with simultaneous open,  but probably OK
until the recommended version changes.

.pdf