[Freeipa-devel] User Certificates in 4.2 - design and questions

[Simo Sorce]

On Mon, 2015-05-04 at 16:41 +0200, Martin Kosek wrote:
> On 05/04/2015 03:01 PM, Fraser Tweedale wrote:
> > On Mon, May 04, 2015 at 10:50:15AM +0200, Martin Kosek wrote:
> >> Hello,
> >>
> >> Please let me promote the design for one of the major FreeIPA 4.2 features, the
> >> (user) certificates and Smart Card integration:
> >>
> >> http://www.freeipa.org/page/V4/User_Certificates
> >>
> >> The design went through couple interim discussions between developers outside
> >> of this list, so there should not be too many objections. But still, please
> >> free to comment or improve the design yourself.
> >>
> >> For FreeIPA 4.2, I think this resolves in following, quite limited, scope of work:
> >> * Adding eq, pres indices for userCertificate
> >> * Applying new policy of storing certificate in userCertificate attribute,
> >> based on upcoming Certificate Profile feature by Fraser.
> >> * Making sure that multiple certificates can be added to userCertificate
> >> attribute manually by CLI and UI
> >>
> >> The "Certificate Identity Mapping" part will probably be moved to 4.3+, unless
> >> there is extra pool of development resources.
> >>
> >> There is still something to be resolved - how should the certificates be
> >> revoked in object-del or object-disable actions? Currently, certificate is
> >> always stored in userCertificate and it's serial is used for revoke operation
> >> in Dogtag. But that will not be true in 4.2+ since some certificates will not
> >> be stored in accounts.
> >>
> >> Do we only want to revoke those that have policy to be stored in the
> >> userCertificate attribute? Does not sound right to me. Or do we need a Dogtag
> >> API that would allow us to query (or even revoke directly) all certificates
> >> generated for a user/service/host and revoke them, regardless whether they are
> >> stored in userCertificate attribute or not?
> >>
> > If the DN or other searchable attributes bear a principal name,
> > existing APIs should be sufficient (if a little awkward).  But
> > Dogtag does not know about principals, it only knows what is on the
> > cert (and a few other things, like the profile that was used).
> 
> Kerberos principal SAN is added when the certificate is requested via
> Certmonger, but we do not add it when requested via cert-request command (yet).
> So we cannot depend on it.
> 
> > Later, when we implement GSSAPI and ACL enforcement in Dogtag
> > (https://fedorahosted.org/freeipa/ticket/5011) we can also store the
> > principal that issued the certificate for a concrete association in
> > Dogtag, which can be used to locate certificates for a principal.
> 
> Right, sounds as something we should do. I commented in the ticket.
> 
> > Considering known use cases in which one would not want to store the
> > issued cert in IPA, some of these have short lived certs so
> > revocation is not an issue.  With that in mind I think it would be
> > OK, for 4.2 at least, to not provide a way in IPA to revoke a cert
> > that was issued via IPA but not stored; it can still be revoked
> > using Dogtag directly, and we could provide pointers to Dogtag
> > documentation.  The important thing is to manage the user
> > expectations for 4.2.
> 
> Hm, maybe - Simo, if you disagree, please shout. In this case, we would need to
> make sure this side effect of the userCertificate policy is very well documented.

I do not disagree, in most cases when a user (or computer object) is
deleted, there is really no need to actually revoke the cert.
Keep in mind that revocation list growth is also a concern.
So I am fine *not* revoking certs automatically and instead documenting
best practices for certs lifecycle management (ie deleting certs when
not useful) and how to manually/explicitly revoke certs only when
actually compromised (for hosts), or when needed (user leaves
organization and may retain a copy of the private key, unlikly when the
cert was in a Smart Card which has been returned and wiped).

Simo.