On Mon, Jul 22, 2013 at 1:27 PM, Trevor Perrin <[email protected]> wrote:

> On Mon, Jul 22, 2013 at 10:20 AM, Joseph Bonneau <[email protected]>
> wrote:
> >
> >> It would prevent that because the transmitted cookie from the
> >> legitimate browser is bound to that browser's TLS connection, via a
> >> MAC.  So the MITM can't reuse the cookie.
>
> > What prevents this, which seems like an attack the system is designed to
> > guard against?
>
> The "cookie_key", which is derived from the original TLS connection
> where the server sent the cookie.
>
> This key is never transmitted, so it won't be known to a MITM attacker.


Thanks, this makes more sense now. This seems to means the server either
needs to store (and verify) the correct cookie_key for every connection it
has ever sent a cookie on, or use the "stateless" variant you mentioned. I
think simplifying the proposal and only having the "stateless" variant
makes more sense-really this is just assuming the server has a long-lived
cookie-minting secret, which is also assumed in ChannelID and other
proposals. The difference here is that instead of binding to an
Origin-bound certificate to lock to a particular client, you're binding to
the original TLS master secret on which the cookie was set. Makes sense to
me now but I think this could be spelled out much more clearly.


>
>
> Trevor
>
_______________________________________________
websec mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/websec

Reply via email to