[resend]

On Mar 31, 2011 3:03 AM, "Sam Hartman" <[email protected]> wrote:
> Folks, during today's meeting we discussed the need for protecting
> information exchanged during the context exchange.
>
> An example of this need would be protecting context flags from the
> client to the server.
> Some server implementations require that certain context flags be set.
> As an example ssh servers following RFC 4462 require the mutual flag be
> set.
> This needs to be integrity protected.

If this were the only unauthenticated plaintext requiring
protection... then you could require that all such flags be set that
the mechanism supports.

> There are a number of possible options:
>
> 1) Integrity protect each token separately. Down side: more complex
> especially if tokens need integrity protection that are exchanged before
> a key is available.

You'd have to store or re-create those messages that come before kwy exchange.

> 2) Extend our mechanism to depend on a specific hash function.
> Disadvantage: requires us dealing with crypto primitives directly . Adds
> complexity to specificiation of the mechanisms.

Also, hash agility.  TLS 1.2 didn't do this in a well performing way
and you'd have the same problem (you can't know a priori which hash
function you'll negotiate, so you either use them all as you go or
keep the messages.

> 3) Provide a gss_getmic or similar of the entire conversation.  The
> disadvantage here is that the client needs to maintain state sufficient
> to hold a copy of the conversation. If there is a stateless server, this
> ever-increasing state needs to be transported back and forth for each
> message.

4) Re-construct those messages near the end of the exchange, storing
only iyems such as nonces in the interim.

Nico
--
_______________________________________________
abfab mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/abfab

Reply via email to