Re: removing MK_GSSAPI

On Mon, Aug 11, 2025 at 09:29:31AM -0700, Rick Macklem wrote:
> On Mon, Aug 11, 2025 at 8:49???AM Cy Schubert <Cy.Schubert@cschubert.com> wrote:
> >
> > In message <aJoEZhS5_JQbjLRK@freefall.freebsd.org>, Lexi Winter writes:
> > > hello,
> > >
> > > i've posted a review to remove the WITHOUT_GSSAPI src.conf knob in 15.0:
> > > https://reviews.freebsd.org/D51859.  my rationale for this is that since
> > > this is already a no-op for MIT Kerberos, there's no real need to keep
> > > it around for legacy Heimdal and removing it simplifies the build logic
> > > a little, since you either have Kerberos or you don't.
> > >
> > > Cy suggested that we might instead want to modify MIT Kerberos to make
> > > WITHOUT_GSSAPI do something.  i'm not sure this is useful, because
> > > without GSSAPI you can't really use Kerberos for anything even if you
> > > build it; in particular, both ssh and gssd require GSSAPI.  in the past,
> > > this knob might have made sense since base gssapi was separate from the
> > > implementation (Heimdal), but with MIT Kerberos, this is no longer true.
> > >
> > > but, perhaps there's a reason to keep this knob around?
> >
> > My thoughts,
> >
> > lib/libgssapi uses Heimdal data structures. GSS apps will run into trouble
> > when using MIT's GSSAPI with libgssapi.
> >
> > MIT's GSSAPI can replace our libgssapi. Installing it by itself is
> > pointless.  I agree with removing MK_GSSAPI. There are people here with
> > more history than I. I understand why we might want GSSAPI without
> > Kerberos, there could be other GSS providers. Is this realistic?
> Highly unlikely, I think. Once upon a time, GSSAPI was envisioned to
> someday have a variety of mechanisms, with Kerberos being one possible
> one. That has never happened, and since MIT sticks the GSSAPI in the
> same library as the Kerberos mechanism glue, I cannot see a GSSAPI
> that does not use Kerberos as ever likely to happen.
> 
> The only mumbling about other mechanisms I've heard is for Oauth2
> (or whatever that token stuff is called), but all that ever seemed to
> happen is getting an OID for it. (It would actually be nice to have, I
> think, but I'm not volunteered to write it.;-)

In principal Globus GSI has a GSSAPI interface for federated login, but
I'm not convinced it sees much use vs Oauth2 these days (it's still in
Ubuntu and Red Hat repo though).  I wouldn't worry about this since it's
had 20+ years to catch on and hasn't outside some niche environments.

-- Brooks