Request for allocation regarding RFC4752 (SASL/GSSAPI) interop problem

[Nicolas Williams]

There is an interop problem between various implementations of RFC4752.

To remind you, RFC4752 (SASL/GSSAPI) looks like:

    <app-layer SASL mechanism negotiation>
    C->S: <initial security context token for GSS mechanism>
   [S->C: <reply security context token>]
   [C->S: <reply security context token>]
   [...]
  [[C->S: <empty>]]
    S->C: <wrap token containing a) a bitmask of supported security
           layers, b) the server's max message size>
    C->S: <wrap token containing a) the security layer requested by the
           client, b) the client's max message size, c) requested
	   authzid>
    S->C: <app-layer outcome of authentication message>


The interop problem relates to (a) in the C->S wrap token.  That item is
a bitmask, with three bits defined:

"
          1 No security layer
          2 Integrity protection.
            Sender calls GSS_Wrap with conf_flag set to FALSE
          4 Confidentiality protection.
            Sender calls GSS_Wrap with conf_flag set to TRUE
"

Now, one group of implementors interprets this to mean that in order to
require integrity _and_ confidentiality protection the client must set
both, 2 and 4, while another group of implementors interprets this to
mean that only 4 must be sent -- after all, the GSS-API requires that
integrity protection be provided when confidentiality protection is
requested, and anyways, it's not possible to set conf_flag to both,
FALSE and TRUE.

For example, Microsoft's Active Directory requires that clients send '6'
when the server is configured to require "LDAP signing and sealing".
So does the Perl CPAN RFC4752 implementation.  Whereas Cyrus SASL, for
example, requires '4'.

Whatever the rationales each implementor might give for one or the
other, this interop bug exists in the field, and we need to work around
it.

Clients cannot tell which interpretation a server uses.  They can only
be told by the application a priori.  Servers can choose to accept 4 and
6 as equivalent choices by the client.

Therefore I propose:

a) that Cyrus SASL's 'gss' plugin treat '4' and '6' as equivalent, on
the server side;

b) that Cyrus SASL add a property for specifying which to send on the
client side;

and

c) that applications that can re-bind or re-connect try it one way, then
the other, while applications that cannot should make the choice of
behavior configurable.

OpenSolaris will be implementing the above.  Therefore we'd like at
least a property name and number to be allocated for (b).

(Yes, I'll be sending e-mail about this to the SASL WG list too, but
separately.  No, I'm not interested in an update to RFC4752 -- SASL/GS2
should just replace it, but the WG should be aware of this issue.)

Nico
--