Re: [TLS] Review of PR #209

[Geoffrey Keating <[e-mail]>]

"[e-mail]" <[e-mail]> writes:

> > On 22 Sep 2015, at 01:40, Geoffrey Keating <[e-mail]> wrote:
> > 
> > Daniel Kahn Gillmor <[e-mail]> writes:
> > 
> >> Consider a server has an ongoing session wrapped in TLS that uses client
> >> authentication to approve or deny some requests from the client.  It
> >> remembers what requests the client has made as some sort of relevant
> >> state.  Let's take an imap server working with a client that has state
> >> of a "currently-examined folder", but this applies to servers and
> >> clients with much more complex state as well.
> > ...
> > 
> > I think for such a protocol, you'd want to say that authentication is
> > not retroactive.
> > 
> > For other protocols, you might want something else.  For example, in a
> > protocol which uses client authentication for billing, you might want
> > to treat an authentication half-way through the session as the account
> > to bill for the entire session.
> > 
> > Some protocols will also want to support multiple identities, and some
> > will not.  For some protocols a new authentication might want to in
> > some fashion dis-trust previous authentications, other protocols might
> > say that all previous authentications are valid until the end of the
> > session.
> 
> If I get this right: 
> 1) it will be possible on the same connection to authenticate
> with multiple different certificates,

I think this is still under discussion.  In the most recent draft, no,
but in that draft you can only send a client certificate before
sending application data.  However a possibility under discussion is
that you could send a client certificate at any point in the stream.

> 2) the different identities won't ( necessarily ? ) be cumulative
> ie, a server getting the identity I1 and then I2 on the same TLS
> connection won't be able to conclude that the referent of I1 is the
> same as the referent of I2 ?

IMO, this is outside the scope of TLS, a higher-layer protocol will
have to determine what it means.  TLS need only report 'you got a client
certificate here and another client certificate here'.

> If that were to work correctly would one not also have to change the
> encryption for each user?

No.  Even if the client private key is possessed by an untrusted
entity, all that needs to be revealed to that entity to perform the
signature is the current value of the handshake hash.