Alternative proxy-creds API for constrained-delegation
iboukris at gmail.com
Wed Jun 3 19:53:23 EDT 2020
On Thu, Jun 4, 2020 at 1:26 AM Nico Williams <nico at cryptonector.com> wrote:
> > > Now I'm thinking it'd be nice to have separate sshd_config params for
> > > acceptor acquisition from cred_store and storing of
> > > deleg_cred_handles...
> > So if we go with gss_acquire_cred_from(), we can add a new store
> > option "delegation-policy: client-tgt,client-ticket" which will
> > override the corresponding krb5.conf option, which will default to
> > "client-tgt,proxy-creds". Then one could add GssCredStoreKeyValue
> > delegation-policy ...
> We could certainly add this on the acceptor credential acquisition side.
> We could also add this on the credential storing side.
> Since I already have application support for this on the credential
> storing side, that's my preference: I won't have to change my sshd at
> Yes, that means sshd will get a transient S4U2Proxy deleg_cred_handle
> when no cred was delegated that sshd will fail to store if
> delegation-policy = client-tgt, but my sshd ignores (logs, but otherwise
> ignores) failures to store deleg_cred_handles, so that's not a problem
> at all.
> That said, I'll probably end up adding code to call
> gss_acquire_cred_from(desired_name=GSS_C_NO_NAME, cred_store=...) in
> sshd, so this could be an option on both sides.
> Note that we could decompose this into an option for each case:
> - a gss_acquire_cred_from() option to say whether to use S4U2Proxy when
> a cred is not delegated (or even always!)
Right, like the above delegation-policy.
> - a gss_store_cred_into*() option to say whether and how to store an
> S4U2Proxy cred if there were multiple options for that
Do we really need that this part?
> Note that with gss_acquire_cred_impersonate_name() you can get an
> S4U2Self (+ S4U2Proxy) cred without having to have called
> gss_accept_sec_context() that could then be stored with
Or I can acquire impersonator creds with
The krb5.conf though should only apply to gss_accept() though.
More information about the krbdev