[TLS] Forward secrecy with resumption, and 0-RTT security

[Bill Cox <[e-mail]>]

AFAIK, there has never been a session resumed with forward secrecy.  Is
this correct?

In the past, there were two cases: resumption using session IDs, and
resumption with session tickets.  Using session IDs loses forward secrecy,
because the server always has session keys in a session cache, which could
be used to decrypt the prior sessions.  Using tickets did not work either,
because the server always kept a ticket decryption key which could be used
to decrypt all resumed sessions since the key was last rotated.

My first question is: Do we care?

The original definition of forward secrecy seemed to mean forward secrecy
for all sessions older than the key rotation interval.  This is what
Twitter had when claiming support for forward secrecy.  Do I understand
this correctly?  However, the new definition we seem to be using now is
that no past session can be decrypted using server-side state that persists
after the connection ends.  Under this definition, AFAIK, there has never
been a resumable session with forward secrecy.  Is this definition taking
forward secrecy too far?

To have forward secrecy under the strict definition using the new PSK
scheme, we need to use what the new name implies: pre _shared_ keys.  The
new name implies that the server remembers a unique shared key for the
connection, which is server-side state per resumable connection.  If we
only store the PSK in the ticket, and go stateless on the server, then the
server-side ticket decryption key again defeats strict forward secrecy.
With a remembered server-side pre-shared secret, we can have a 0-RTT resume
with strict forward secrecy.  Is this already implied in the spec?  If so,
the spec could be improved, by stating this explicitly.  However, the
current statement that 0-RTT does not provide forward secrecy seems to be
wrong in the case that we resume a 0-RTT connections with a
server-remembered PSK.

To have strict forward secrecy, given that the server already has to have
server-side state (the pre-shared key), I find the old session ID scheme
more palatable than the new tickets.  The main point of session tickets was
to eliminate the server-side session cache, wasn't it?

I think the current spec does not describe well enough how to implement
secure 0-RTT infrastructure.  Instead, it seems to recommend against using
0-RTT, with a pretty dire warning about the insecurity of 0-RTT.  I think
that nearly the entire world will switch to 0-RTT, regardless of security
limitations and warnings.  Companies won’t be able to resist the improved
connection time.  IMO, the spec should describe how to address the security
issues with 0-RTT, rather than just warning about security issues.

So, here are my dumb thoughts about practical 0-RTT security
infrastructure...

One way is to bind resumption to orbits, by having only the orbit that
issues a PSK remember the secret key.  Something like this is needed in any
case to provide 0-RTT replay protection.  Without at least orbit-based
replay protection, I think it does not make sense to allow a ClientVerify
in a 0-RTT connection.

There was a TODO in the last version I read asking what the client would
sign in a 0-RTT ClientVerify.  If orbit-based replay protection is in place
using PSKs, then signing anything that identifies what PSK to use is
enough. However, the handshake has to somehow be bound to the orbit to
prevent the signature from being replayed in a MitM attack. Having a ticket
that can only be decrypted using the orbit-specific ticket decryption key
would work. Using a source address token bound to the orbit would also
work.  Signing the server Finish is not required to insure the signature
will only be accepted once.

There are two parts to a proof-of-possession: uniqueness and freshness.
Uniqueness can be had using orbit-bound PSKs.  Insuring freshness is
harder.  This could be done by adding timestamps to the material signed.
With this, we would not get a complete proof of possession, because an
attacker can attack the client's clock, unlike a normal proof of
possession.  However, if steps are taken to secure clocks, which is needed
for TLS in any case, then we can get what I think of as "compelling
evidence" of possession, which might be good enough.

I think I've partially covered all three security concerns listed for 0-RTT
connections: forward secrecy, replay protection, and client certs.  The
limitations seem to be:


   -

   0-RTT resumption using PSK can address all 3, but not 0-RTT connections
   -

   Strict forward secrecy requires server-side state, so why are we using
   tickets?
   -

   Replay protection based on orbits is only partial protection
   -

   ClientVerify can only provide “compelling evidence”, rather than proof
   of possession


It concerns me that the TLS 1.3 spec seems to say "Don't use 0-RTT!  It's
not secure!" when it is clear that the world will go in that direction.  I
would feel a lot better if the spec had more discussion about how to do
0-RTT securely.

Thoughts?

Bill