Re: Names in Proxy Certificate

["Housley, Russ" <rhousley@xxxxxxxxxxxxxxx>]

Steve Tuecke:

> > In draft-ietf-pkix-proxy-01, section 6.4, the authors talk about end 
> > entity names.  Mostly the specification talks about an approach that was 
> > rejected.  Finally, I think that the conclusion is that the parent of the 
> > proxy certificate contains the real identity information.
>
> Note that section 6.4 is the commentary section.  It is not intended to be 
> the specification, but rather to help give some insight on the 
> specification (e.g. what was considered but rejected).  At some point it 
> may make sense to move this out to a separate informational draft, but we 
> figured it was better to keep it all in one spot to begin with.
>
> The relevant portions of the spec here are sections 3.1 and 3.2.

Section 6.4 was the place that these problems jumped out at me.  I will 
take a look at these sections in the future.

> > If I have it right, then have a few concerns.
> >
> > First, in PKIX, we require that all issuers have a subject DN.  So, empty 
> > DNs cannot be used in the long-term certificates if proxy certificates 
> > are to be associated with them.  Further, the proxy certificate needs to 
> > include a subject DN if proxy certificates are permitted to issue 
> > subordinate proxy certificates.
>
> First, let me say that we're not at all married to this particular 
> approach to subjects and issuers.  What is in this draft is, in fact, 
> different from what we did in the Globus Toolkit.  In writing this draft, 
> we changed the approach (though not the proxy concept) to try to make it 
> live better with son-of-2459.  But in going back to son-of-2459, I see I 
> missed the relevant statement here in 4.1.2.4, of "The issuer field MUST 
> contain a non-empty distinguished name (DN)."  I don't know how I missed 
> this one!

This is a very important requirement.  It is in RFC 2459; it is not new in 
Son-of-2459.  If you want all of the discussion, you will have to review 
the mail list archive.  Off the top of my head, I recall two major 
points.  First, CRL entries are identified by issuer DN and serial 
number.  Second, S/MIME v2 identifies the recipient public key only with 
issuer DN and serial number.  S/MIME v3 continues to supports this method 
as well as the subject public key identifiers.

> The problem (and perhaps why I subconsciously ignored this issuer 
> requirement :-) is that I think the current son-of-2459 language puts us 
> between a rock and a hard place.  As your comment suggests, what if the 
> End Entity Certificate (EEC, or what I think you are referring to as the 
> long-term cert) has an empty subject field, and instead only has a 
> subjectAltName?  What do we use as the issuer in proxy that is signed by 
> that EEC?

Yes, I see the end entity certificate as long-lived, and the proxy 
certificate as short-lived.

> Maybe the best thing to do is explain what we are trying to accomplish, 
> what I observe the constraints to be, and you might be able to advise us 
> of an approach that lives better with son-of-2459.
>
> First, what are we trying to accomplish...  The point of proxies is that 
> we are specifically not trying to create a new identity.  Rather, we are 
> trying to define a mechanism by which a proxy can be created that is 
> associated back only to the subject or subjectAltName of the EEC that is 
> at the top of the proxy signing chain.  In essence, we want to bind a new 
> key pair back to the subject of the EEC, so that the holder of the new 
> private key can make a verifiable claim on the EEC subject name.
>
> What is the nature of that claim?  That's really up to the authorizing 
> party.  Having authenticated with a party that is using a proxy 
> certificate, and verified its path up through the proxy chain, EEC, and 
> CA, I know that this party has a valid proxy that was issued by an EEC 
> with a particular subject name.  In practice, the way that we use this is 
> as a way to delegate authorization rights.  That is, if I have 
> authenticated a proxy that is tied to an EEC with a subject of "Steve", 
> then I apply whatever authorization that I would normally apply to "Steve".
>
> And this is where the proxyRestriction field (3.6.1.3) comes into 
> play.  What the holder of an EEC typically wants to do is issue a proxy 
> that implies a subset of the rights normally associated with that EEC 
> subject.  So when such a proxy is presented to me, I would apply the 
> intersection of whatever authorization rights I would normally grant to 
> "Steve", with those that are defined in the proxyRestriction field of all 
> proxies in the signing path.
>
> In the current Globus Toolkit code, the way we handle subject and issuers 
> is that we require that the EEC subject be a DN, and then we extend that 
> DN with a "/CN=proxy" attribute in each proxy certificate.  Then in the 
> path validation we check that the proxy subject is the same as the issuer, 
> except for an extra "/CN=proxy" attribute.

In PEM (see RFC 1114 and RFC 1422), similar name subordination rules were 
specified.  Such rules were defined because version 1 certificates were 
employed.  Version 3 extensions allow much better answers (i.e., name 
constraints extensions).  I suggest that you avoid name-oriented conventions.

> So now what are the issues and constraints in making this play nicely with 
> son-of-2459:
>
>   * What do we do if the EEC does not have a subject, but instead only 
> has a subjectAltName?  Obviously, the approach that we used in the Globus 
> Toolkit doesn't work in this case.  And it doesn't seem particularly 
> satisfying to say that you can only create a proxy certificate from an 
> EEC that has a subject DN.

Some implementations use names to construct paths.  Others use public keys 
(or identifiers of them).  So, if you want all currently fielded software 
to be able to construct a certification path from the proxy certificate 
(PC) to the trust anchor, then you should mandate the subject DN.  On the 
other hand, if you want such software to only construct the certification 
path from the end entity certificate (EEC), then you have more design 
freedom; however, your applications will have to include their own software 
to validate the PCs.

>   * One of the uses of the subject is for path discovery.  So whatever 
> approach we choose needs to address this.  What we did in this draft is 
> give the proxy subjectAltName a (statistically) unique name, so that the 
> path can be discovered from these names.  Another option would be to use 
> (or invent) a different field for path discovery, such as 
> SubjectKeyIdentifier and AuthorityKeyIdentifier.

As I said above, there is fielded software that works with both 
approaches.  I would prefer both approaches to be supported by the 
specification.

>   * If we choose a different field for path discovery, this obviously 
> liberates us some with respect to the subject and issuer.  Nonetheless, 
> requiring a non-empty issuer still seems like a problem.  But ignoring 
> that for the moment, what would we want for the subject or 
> subjectAltName?  Would it just be a duplicate of the EEC's?  Are there 
> any issues with have multiple certificates with the same subject, even 
> though the proxies are clearly marked as such with a critical extensions?

Self-signed certificates are part of the architecture in several 
places.  So far, I do not see any reason to invent a new concept.

I think that the PC should contain the identity.  Looking up the chain for 
the first non-PC seems like work that can easily be avoided.  Further, I 
think that you would want to minimize any special processing.  In other 
words, the processing of a PC and an EEC should be kept as similar as possible.

>   * We must be very careful to not allow proxies to carry, or somehow 
> imply, the subject of some other EEC than the one at the top of the proxy 
> signing chain.  It seems we have basically 2 choices.  Either make the 
> subject (or subjectAltName) meaningless and ignore it.  Or give it 
> meaning (e.g. carry along the EEC subject) and check it in the path 
> validation code so that some other subject cannot be substituted.  We 
> went with the former (modulo path discovery), in the interest of avoiding 
> yet another thing that needs to be checked in the path validation 
> code.  But I'd also be comfortable with the latter.

Again, I recommend the inclusion of the EEC names in the PC.

> Anyway, I'm eager to hear if you have any suggestions for alternative 
> approaches.

If you are going to be at the IETF in Salt Lake City, we can discuss these 
issues further.

> > Second, in PKIX, we allow attribute certificates to be associated with 
> > subject DN or issuer/serial.  Wouldn't it be useful for authorizations in 
> > attribute certificates issued against the long-term certificate to apply 
> > to the proxy certificates as well?  This seems possible if the attribute 
> > certificate uses the subject DN and the proxy certificates also contain 
> > this same DN.
>
> Yes, I can certainly imagine this being useful.
>
> Perhaps this is motivation for changing the approach to proxy subject 
> names.  For example, we might consider carrying the EEC subject in the 
> proxy unchanged, check it in the path validation to make sure it 
> identical, and use the KeyIdentifiers for path discovery.  But there is 
> still the question of what goes in the issuer field.

If you require the EEC to contain a subject DN, then this gets pretty simple.

> But an interesting question with the combination of attribute certs and 
> proxy certs is what constraints might you want to put on this?  Do you 
> want to be able to associate attribute certs with just EEC's, but not 
> their proxies.  Or just with proxies?  Or just with a particular 
> proxy?  Or a set of proxies signed by a particular proxy?   Or some other 
> set of proxies?  I can imagine arguments for all of these.  This would 
> clearly be an interesting avenue to pursue.

If the Attribute Authority (AA) wants to associate a set of attributes with 
the EEC, or with a particular PC, then the AA can issue the Attribute 
Certificate (AC) with the holder set to the appropriate issuer DN/serial 
number pair.  If the AA wants to associate a set of attributes with the 
identity, then the AA can issue the AC with the holder set to the subject DN.

Russ