Re: [sasl] [TLS] lasgt call comments (st Call: draft-altman-tls-channel-bindings (Channel Bindings for TLS) to Proposed Standard)

[Nicolas Williams <Nicolas.Williams at sun.com>]

On Wed, Oct 28, 2009 at 08:59:53PM +0000, Larry Zhu wrote:
> How about
> 
>    The client's TLS Finished message (note: the Finished 
>    struct) in the clear text form from the first handshake of 
>    the TLS session as identified by the 
>    session ID in the server 
>    hello message, as defined in 7.4.1.3 of RFC5246, 
>     of the last handshake in the current active TLS session.
>    
> 
> Can we label this as "tls-session-unique"? The existing deployment
> seems to have a different interpretation but we do not have a real
> usage case for that. It does not hurt to create another label just to
> avoid potential interoperability issues.
> 
> This seems to be the most intuitive definition and we do not have any
> ambiguity here.

Larry and I just spoke on the phone.

The real issue is the word "connection".  Apparently RFC5246 uses it in
two senses, one of them being the-transport-layer-above-IP (i.e., TCP
for TLS and UDP for DTLS, typically), and that is confusing.

I explained to Larry that using "session" instead, as in his proposed
text above is nearly impossible to implement, because it means updating
session resumption state caches -- a very intrusive change to existing
TLS implementations.

What Larry really wants is to find a way to refer to what Microsoft's
SSPI-based implementation of TLS calls a "security context".  Advice on
what would be the best TLS-specific term for "security context" is
welcome.

Nico
--