At 3:15 PM -0400 9/9/10, David A. Cooper wrote:
There is still a problem in this document that needs to be fixed.
Section 3.1 of the document explicitly states that there is no
requirement for subject names to be globally unique in the RPKI. As
I have pointed out several times before, this is a violation of
X.509, which says that "CAs are unambiguously defined by a
distinguished name" and that a directory name is "A construct that
singles out a particular object from all other objects. A name shall
be unambiguous (that is, denote just one object); however, it need
not be unique (that is, be the only name which unambiguously denotes
the object)."
David,
As I have noted previously, in the RPKI it would be silly (at best)
to assume name uniqueness on a global basis. This is because of the
very large number of entities that will act as CAs, and the reality
that not all of them ought be trusted. The RPKI, unlike most PKIs,
does not bind a key to a name as its principle secruity function. It
binds a key to a resource set. The RPKI is more secure than typical
PKIs, with regard to its principle function. This is because the 3779
extension imposes a subset relationship on cert paths analogous to
what the name constraints extension does. But, as we both know, the
name constraints extension is very rarely used in PKIs.
You ask what not put the requirement for name uniqueness into the
SIDR specs, even if we know that it would be dangerous to rely on it.
If one puts a requirement of this sort into a spec, one is
encouraging developers to rely on the requirement, and we don not
want to do that, for the reasons noted above.
I note that all of the developers of RPKI CA software are employing
techniques to generate names that should minimize the likelihood of
name collisions. I'm in favor of including some of the text you
generated on how to do this in the cert/CRL profile, as
recommendations in the design note section.
I also omise that, before the cert/CRL profile is finalized, it will
include text to ensure that the concern you cite will be addressed,
i.e., to prevent an RP from (naively) assuming that a name match (vs.
key match) is sufficient when checking CRLs. because of the
constraints already imposed in the RPKI (re CRL issuance) this is a
valid check to perform, and it will be called out.
I do not propose to change the CP as you suggest, for the reasons cited above.
I'd like to hear the Security ADs opinions on this matter.
Steve
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr