Hello,
at our site we are in the process of decommissioning Active Directory Domain Services and Certificate Services as we have 0 Windows systems left, except for the ones making up the current Identity and PKI infra.
We want to be sure of the design we are drafting as fixing it afterwards can be very difficult.
We are heavy consumers of digital certificates, both for Server Authentication and Client Authentication scenarios (including smart cards for administrators, and someday if we get our IdP right even for users).
Currently implemented we have a Root CA (with it's OCSP responder) with multiple Subordinate CAs (the "Enterprise" Issuing for AD DS, the firewalls, ..).
My experience with IPA so far is that all interaction with DogTag happens via the IPA API (the tomcat web interfaces of DogTag for Agents and End Entities are actually inaccessible by default) and the IPA API does not permit to issue certificates for principals who are not known to IPA itself.
To me, it seems like the IPA API is designed around the concept of delivering end entity certificates to Kerberos prinicpals who can authenticate via Kerberos token (be it obtained interactively by a user or via Keytab for a host) and automatically issue certificates based on DogTag templates that contains only names that are registered SPNs "services" for the requesting entity.
I see on the Internet that many people try to circumvent this by issuing certificates using the pki command line directly (to bypass IPA API) and/or registering dummy hosts with dummy services to have the API release certificates for hosts that are not member of the IPA domain.
This does not seems to us a good idea.
We have come to the conclusion that the best architecture for us would be to have a standalone DogTag CA with integrated OCSP as root; that root will sign the IPA-CA (which comes with it's own OCSP), another DogTag as child of the root to issue all the certificates for other purposes (network equipment, clusters, ...) and finally the existing issuing CAs will become subordinate to this new root (firewalls,...).
In a future, the Token Management System of DogTag is attractive to us, and I think this will give us the flexibility we want.
The point, and the question is: DogTag standalone has a pretty steep learning curve (to put it mildly...). Do we have misunderstood IPA-CA architecture and we are over complicating things for no reason?
If we have to suck it up (DogTag learning curve) so be it, but first I would like some confirmation that our assumptions on IPA-CA are broadly right.
Thank you very much,
Luca