Re: [kitten] [Ietf-krb-wg] Channel bindings -- interop issue with GSS_C_AF_*

[Martin Rex <[e-mail]>]

Sam Hartman wrote:
> 
> I think RFC 2744 chose the wrong constant for nulladdr; it should have
> been 0 not 255.  I think we need to update RFC 2744 and 5554.
> I believe that
> 
> 1) Mechanism implementations should collapse unspecified address with no
> actual data and null address together
> 
> 2) The Kerberos mechanism should do so in a manner compatible with
> Microsoft
> 
> 3) We need to explicitly specify what applications should do here.
> 
> If someone argues that we cannot make incompatible changes to 2744, I
> respond that compatibility with the implementations we know about is
> more important to me than compatibility with the spec and if forced to
> choose I will choose the implementations.

In case I my messages were conceived as ambiguous:

I'm OK with updating the relevant specs to formally allow _and_ document
what the current installed base is doing, for all of the specs that
we know about.

Changing rfc2744 in a backwards incompatible fashion would be a bad idea,
but I don't think that is necessary.  Relaxing rfc-2744 should be
sufficient.

Two things should go into rfc-2744.

- We need to allow taggging an absent address
  (i.e. OctetString of size zero) with GSS_C_AF_UNSPEC(0).
  The current wording of rfc-2744 implies that absent address data
  needs to be tagged with GSS_C_AF_NULLADDR(255).

- We need to document which specs are going to require
  GSS_C_AF_UNSPEC(0) with absent network addresses and
  non-empty application_data channel bindings to match actual
  implementation practice and for interoperability with the
  installed base of that specs.


-Martin