Re: [certid] open issue: wildcards in component fragments

[Martin Rex <[e-mail]>]

Matt McCutchen wrote:
> 
> On Thu, 2010-10-07 at 15:26 -0600, Peter Saint-Andre wrote:
> > The intended status of this document is BCP = Best Current Practice.
> > That "B" is in there for a reason. The question is: does allowing *oo
> > and f*o and foo* (and maybe even f*b*r) as wildcard fragments truly
> > represent a "best" current practice, or only "current" practice? It
> > doesn't strike me as a most excellent, effective, desirable, suitable,
> > appropriate, or useful practice, although I freely grant that it is in
> > common and general use at the present time.
> 
> I have never seen a certificate with a wildcard that is not a
> whole label on a public web site.

The idea behind writing documents such as rfc-2818 is, that implementers
don't have to do public opinion polls an apply heuristics, but instead
follow the implementation guidelines in the document.

It is pretty obvious that rfc-2818 meant to standardize substring wildcard
matching with a single wildcard character in the leftmost dns label.


What is _not_ clear, is whether rfc-2818 meant to also allow more
than one wildcard character in a CN-ID or DNS-ID, wildcards in
more than one DNS label or wildcards in other than the leftmost label.


I believe that shipping a TLS client that matches a server cert
with CN-ID or DNS-ID equal to "*.*.com" is neither reasonable nor
responsible, and it I don't care whether a negilgent CA issued
such a certificate on purpose or whether an attacker managed
to get such a cert issued--there were two bugs published that
made this possible: OID integer wraparound and embedded NUL chars
in the Subject DName, and a significant part of the installed
base does not check server certs for revocation (MSIE on XP/2003).

(The folks on domains such as "*.co.uk" or "*.co.tw" are in a
 slightly more difficult position).


-Martin