Followup on the referral discussion

Isaac Boukris iboukris at gmail.com
Sun Dec 22 05:58:42 EST 2019


On Sun, Dec 22, 2019 at 8:32 AM Greg Hudson <ghudson at mit.edu> wrote:
>
> On 12/21/19 2:37 PM, Isaac Boukris wrote:
> > Later last week, I had a call with metze in which he corrected me
> > about a couple of things I mentioned in our discussion. In short,
> > unlike what I said, referrals should always work in windows env, for
> > both forest and external trusts.
>
> I can interpret this in two different ways, and I'm not sure which is meant:
>
> 1. For an external trust, the client has to be told the external realm,
> but will then follow any referrals issued by that realm.
>
> 2. The local KDC knows how to issue referrals to external realms (how
> does it decide which SPNs to do this for?), so the client doesn't behave
> any differently than it would for a forest trust.

The latter. To my understanding the KDC decides how to refer based on
spn suffix table negotiated at trust establishment (short spns for
instance can't be referred, and i think might be duplicated).

> > He also suggested that making assumption based on the name type, on
> > the client side is not correct, and that we should not override the
> > realm when requested but rather chase referrals to krbtgt/srealm and
> > then chase again referrals to server (and that could be made to work
> > with netbios realms, if canonicalize is set).
>
> "Chase referrals to krbtgt/srealm" means asking the local KDC for
> krbtgt/srealm, and keep following the issued TGTs until we get there?
> If so, that isn't actually a referrals chase in the sense of requiring
> RFC 6806 extensions; following alternate TGTs to a realm is specified in
> RFC 4120 section 3.3.3.  The second part (following referrals issued by
> srealm for the server name) of course requires RFC 6806.

Right, alternate TGTs.

> MIT krb5 implements the first part in a pretty complicated way, because
> it is supposed to work if either the client or the KDC knows a path to
> the specified realm (via [capaths] configuration).

Yeah, it is like MIT in that we only follow referrals, and respect in
a way the app requested realm.

I still can't think of a clean way to apply it to rbcd, say we follow
alternate TGTs to proxy realm (normally), then what do we do?


More information about the krbdev mailing list