Re: [TLS] What would make TLS cryptographically better for TLS 1.3

[Yoav Nir <[e-mail]>]

On Nov 3, 2013, at 9:03 AM, Ralf Skyper Kaiser <[e-mail]> wrote:

> Hi,
> 
> avoid renegotiation. It serves no purpose and only adds complexity. It is so much more secure to kill and re-establish the TLS if the counters run out instead of renegotiating.
> 

Hi, Ralf

The one use of renegotiation that I'm aware of, is for overcoming a UI issue in browsers. If you do a TLS handshake with mutual authentication (so the server sends a CertReq), the browser pops up a dialog box with all the certificates you might have. Website designers with to avoid that, especially on the welcome screen, so the web server does not send a CertReq. Instead, they present a welcome screen with a button or link that says "Login with certificates" Clicking that performs a regular SSL handshake (or does nothing at all if the connection is already established), but when the request comes in ("GET /login_with_certs HTTP/1.1"), the web server sends a HELLO_REQUEST, and in the resulting handshake it sends the CertReq, so the pop-up appears when the user *is* expecting it.

I totally agree that renegotiation for rekeying is useless for people who are not doing DES-CBC and 3DES-CBC. It's even superfluous for them in most cases (you're pretty save doing 500,000,000 blocks, and that's 4 GB in 3DES. How many sites do you browse with 4 GB?  Maybe downloading stuff…)  

But before we can drop renegotiation from the standards, or recommend that implementors don't implement it, we need an alternate mechanism to upgrade from server-authenticated to mutually-authenticated within the same session. That is a real market need. How about allowing a CertReq sent from the server to the client in the middle of a connection, followed by the client sending a Certificate and Certificate Verify. For simplicity, we could always do that after the Finished, so that it's always Server-authenticated session when the Finished is sent.

Yoav