Re: [TLS] TLS renegotiation issue

[Nicolas Williams <[e-mail]>]

On Thu, Nov 05, 2009 at 06:11:41PM +0100, Martin Rex wrote:
> One conceivable approach would be a TLS extension for secure renegotiation
> with roughly the following semantics:
> 
> 
>  - on initial TLS handshakes (aka Re-nego of a TLS_NULL_WITH_NULL_NULL)
>    the client that supports secure renegotiation should send the
>    TLS extension in the client hello with an empty extension data
>    to signal to the server that it support secure renegotation.
> 
>  - on TLS session renegotiations, the client should sent the TLS extension
>    with the extension data containing the client.random and server.random
>    from the existing session.

The MITM can make sure they match.

Why not just send an extension with the tls-unique channel binding for
the first/outer connection, and then make sure that that gets fed as an
input for the PRF in the second/inner connection?  I.e., proper channel
binding.  The MITM _cannot_ make sure that the tls-unique CB of the
client<->MITM and MITM<->server outer connections match.

>  - A server that receives a ClientHello without the TLS extension for
>    secure renegotiation should NOT perform old-style renegotiation
>    for that session (and get apps with flawed assumptions into trouble),
>    i.e. NOT support delayed authentication for those TLS clients.

Right.

>  - A server that receives a ClientHello on an initial TLS handshake
>    with a TLS extension (empty data) for secure renegotiation may decide to
>    delay client-cert authentication to a securely renegotiated session

Right.

>  - A server that receives a ClientHello with a TLS extension for secure
>    negotiation and a reference to a previous session MUST compare the
>    client.random and server.random from the extension data to that of
>    the current session (and abort if they do not match).  And the
>    server should probably insist on doing a ChangeCipherSpec on both
>    directions in the renegotiation handshake.

See above.

> Since Larry is looking for a means to uniquely identify a single
> TLS "connection" independent of whether its a single TLS handshake
> or renegotiated, we should check whether we can carry-over some
> information form the initial handshake on a connection along
> with the client.random&server.random.  Technically there is no
> limit on the number of renegotiations, so a simple pointer
> only one TLS session into the past does not seem sufficient
> for that purpose.

See above.  Use tls-unique.

Nico
--