[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Comments on CRACK



Vipul Gupta wrote:
> In message <3815F49E.BFABF7C9@xxxxxxxxx>, Roy Pereira writes:
>
> >
> > Let me ask everyone who is interested;  How do we support existing
> > legacy user authentication within IKE without using a PKI ?
>
> With a protocol that lets the customer download an encrypted private key/
> certificate pair from a server, followed by ordinary IKE.
>
>               --Steve Bellovin
>

  A perfect lead-in for what I've been thinking about for some time
  now :-)

  How about using an HTML forms based interaction over HTTPS between
  a webserver and a user to accomplish what you state.

           Internet                           Intranet

                               |
                               |          +--> Legacy Auth server
           SSL/TLS protected   |         /
     user =================== HTTPS <---+
                              server
                               |
                               |

   This interaction can easily accomodate legacy user auth mechanisms
   like SecureID, DES Gold, OTP, CHAP because the HTTPS server has access
   to authentication tokens in the clear. Even multiple rounds don't
   pose a problem. After the Auth server responds with "OK", the
   HTTP server can squirt out a special MIME datatype and the browser
   could be set up to automatically invoke the IKE daemon (or companion
   software) to handle that MIME type. The HTTPS may need to coordinate
   with the IPSec gateway on the Intranet side.

   This could be a reasonable solution for the road warrior VPN scenario.
   I've heard Paul Hoffman use the term "user authentication in Phase 0.5"
   for an approach like this (in contrast to Hybrid's Phase 1.5).

   (Maybe now's a good time to go look for that fire extingusher :-)).

   vipul

That's not a bad idea in itself, although I'd rather not have the requirement
to implement HTTPS just to be able to implement legacy authentication
support in IKE!

However, let me quote an earlier email that I sent:

There's another architectural thing you should consider. What about modifying
the protocol so that when the server starts believing in the authenticity of the
client, the server issues the client's public key a certificate? This certificate
would have a very limited life-time, just enough for the purpose at hand.
It would be transported to the client in the 'last' message, whatever that is.

Although this creates more public key operations, the legacy authentication
functionality could be located on a different physical box than the actual
security gateway.. This achieves a very similar function to the Kerberos ticket
granting server, and the certificate is similar to Kerberos tickets. You'd of
course have to set up the trust relations appropriately.

There could also exist "one time certificates" that can be used only once
during their life-time to gain access, similar to one time passwords. Some
way or another they would be revoked the moment they are used.


If you wanted, you could transport such a certificate through HTTPS to the client.
Although, as said, I'd rather not have HTTPS in the picture.

--
Ari Huttunen                   phone: +358 9 859 900
Senior Software Engineer       fax  : +358 9 8599 0452

Data Fellows Corporation       http://www.DataFellows.com

F-Secure products: Integrated Solutions for Enterprise Security