Proposal 176: Proposed version-3 link handshake for Tor

Nick Mathewson nickm at freehaven.net
Wed Feb 2 17:16:26 UTC 2011


On Mon, Jan 31, 2011 at 10:46 PM, Adam Langley <agl at imperialviolet.org> wrote:
> On Mon, Jan 31, 2011 at 9:50 PM, Nick Mathewson <nickm at freehaven.net> wrote:
>>   certificates.  The security properties of this protocol are just
>>   fine; the problem was that our behavior of sending
>>   two-certificate chains made Tor easy to identify.
>
> Actually, two (or more) certificates are very common when talking to
> HTTPS servers. (See mail.google.com:443 for example.)
>>   And the cell-based part of the V3 handshake, in summary, is:
>>
>>    C<->S: TLS handshake where S sends a "v3 certificate"
>>
>>    In TLS:
>>
>>       C->S: VERSIONS cell
>>       S->C: VERSIONS cell, CERT cell, AUTH_CHALLENGE cell, NETINFO cell
>>
>>       C->S: Optionally: CERT cell, AUTHENTICATE cell
>
> Forgive my ignorance here. I have only a passing knowledge of the Tor
> protocol these days.
>
> If you wish to prevent scanning for Tor nodes then you could have the
> client put the SHA256 of the server's identity key in its initial
> cell. This supposes that the client always knows the identity of the
> server that it's connecting to; which may not be the case.

Alas it isn't; for usability reasons we require that clients can
connect to bridges without knowing their ID.  But maybe we could do
something like that for connections to non-bridges.  One problem is
that if we had a cell type like this, we'd want to have it appear
*before* the versions cell, which means you can't really have it
negotiated as part of the protocol.

> If the client doesn't care about authenticating to the server, then it
> could optimistically send cells predicated on a correct version
> prediction. SSH does this (see RFC 4253, section 7.1,
> 'first_kex_packet_follows'). The server will know if the prediction
> was correct once it sees the client's version and can discard
> optimistic cells.

That's a pretty neat idea.  I wonder what cells we'd actually want to
send, though: we don't want to send any actual data to the server
until we've authenticated the server, and we can't do that until we
get the CERTS cell from the server, which doesn't come until after the
server sends us its VERSIONS cell.

-- 
Nick



More information about the tor-dev mailing list