Re: [cicm] CICM BOF Summary

["Novikov, Lev" <[e-mail]>]

NOTE: Cross-posting from [e-mail] to [e-mail]
(Hat-tip: Kevin Wall)

Last week we had a BOF at IETF 81. Thanks to all who attended (in-person
and via Jabber). For those who couldn't make it, a summary:

--- Begin Summary ---
Dan Harkins and Dan Lanz were the BOF Chairs.

Sean Turner and Stephen Farrell are the Security ADs.

Vincent Roca presented slides about using CICM in a
High Assurance, High Performance Security Gateway.
Slides: http://www.ietf.org/proceedings/81/slides/cicm-1.pdf

Lev Novikov presented slides about CICM's logical model and how
security domain separation makes CICM different from other crypto APIs.
Slides: http://www.ietf.org/proceedings/81/slides/cicm-2.pdf

There were several points of discussion:

1. What about existing approaches:
   * Why can't you extend PKCS#11 so that crypto operations like
     encrypt always return TRUE?

     A few reasons were given:
     (a) CICM needs richer semantics (more and different kinds of
         inputs) than what is available in PKCS#11. Previous attempts
         at extending PKCS#11 became a mess.
     (b) Return values can be more complex than just TRUE (e.g., list
         of things that went wrong).

   * What about using an existing protocol as an interface?

     CICM could sit under such a protocol; it is also intended manage
     the crypto (note the large number of management commands), and not
     just the pipe (channel).

   * Which approach, C-style or object-oriented, was intended? The .NET
     crypto classes might be suitable for an object-oriented approach.

     CICM is defined in IDL for which one can generate bindings in many
     different languages including C, C++, Java, etc. We will have to
     investigate the .NET approach further.

 ** There was a request that folks on the list discuss these issues for
    the benefit of the community.

2. The charter is insufficient for a Working Group:

   * It was noted that there could be two goals:
     (a) to produce multi-vendor support for a standard interface
     (b) to introduce these concepts into existing IETF protocols

   * The charter appears to be too detailed; it should focus more
     on outlining the problem scope well.

   * CICM appears to address requirements that are not well explained
     in published documents.

   * How would CICM work with Authenticated Encryption with
     Authenticated Data [RFC 5116], TLS, or IPSEC? What are the
     consequences on other protocols?

The major consequence of these points is that we should re-write the
charter and write documents to address the:
 * larger problem scope
 * logical model (in more generic terms) and requirements
 * impact of this logical model on 2-3 existing protocols
 * details for an corresponding API (e.g., CICM)

--- End Summary ---

More on this to follow.

Lev