From: jpp@markv.com
Date: Fri, 9 Jul 93 23:15:55 PDT
To: cypherpunks@toad.com
Subject: Secure comm program, Sockets + LINK
In-Reply-To: <9306090653.AA23961@triton.unm.edu>
Message-ID: <9307092314.aa24145@hermix.markv.com>
MIME-Version: 1.0
Content-Type: text/plain


  Supose you opened up a socket on the local machine.  And that you
ran your usual telnet to connect to it.  The program listening on the
local socket would be responsible for running one end of a 'LINK' like
secure protocol.  It would connect to either the remote telnet socket,
or a special purpose socket at the remote end.  There either you can
use a pipe to a pty (standard telnet -> login shell -> LINK -> pty),
or in a special socket through LINK out the telnet socket.  (There is
an obvious extention with multiple hops through LINK-socket programs
which should provide the same kind of anonymity that is provided by
the CP remailers.)

  In this way, the data passing over the (presumed unsecure) net
connection (which might well be slip) is encrypted.  I am presuming
that both the local and remote kernels are 'secure' enough that you
would want to use them.

  The upside is that you get to use 'comercial grade' telnet==comm
programs.

  Of course this stratagy is no use for folks who don't have sockets
on both ends of the channel.  It is also likely to be obsoleted as
soon as secure-ip gets out.  Some ambitious sole might want to try the
analogous stratagy where a local pseudo-serial-device is created in
software, thus again getting 'comercial grade' comm==telnet programs.

  I sugest this stratagy, because it is one I am concidereing under MS
Windows.  I have a telnet which is acceptable, I have slip, and I have
the source to LINK, so...

(local-telnet) --kernel--> (local-LINK-SOCKET) ----unsecure-network---+
                                                                      |
                  (remote-telnet) <--kernel-- (remote-LINK-SOCKET) <--+
                                                                      |
(remote-pty) <--kernel-- (remote-LINK) <--kernel-- (remote-telnet) <--+

  Infact it would be nice to see some socket (perhaps 32?) become the
standard for the secure telnet service.

  Meanwhile, I did peruse the LINK source and am a little unhappy with
the actual protocol used in setting up the secure channel.  It is only
authenticated in one direction, rather than both (as I understand
it).  I would like to see two way authentication, and (perhaps)
Diffe-Helman key exchange.

j'