[cabfpub] Defining BR scope

[Erwann Abalea]

Bonjour,


Le 2 févr. 2016 à 20:02, Peter Bowen <pzb at amzn.com<mailto:pzb at amzn.com>> a écrit :

On Feb 2, 2016, at 10:06 AM, Ryan Sleevi <sleevi at google.com<mailto:sleevi at google.com>> wrote:
On Tue, Feb 2, 2016 at 4:20 AM, Peter Bowen <pzbowen at gmail.com<mailto:pzbowen at gmail.com>> wrote:

Do they do this in the presence of a SAN extension or just the absence?

Modern browsers: Only when the SAN is bereft of DNS/IP addresses
Most every other software library: In addition to, and generally as the first match. I have seen code (and filed bugs) against implementations in PHP, Python, Ruby, and C that all opted to check CN, and if CN didn't match, check SAN. I've pointed them to RFC 6125 and explained the risks.

>> If so, are we anywhere near the point where
>> they could stop doing this?
>
> Well, we mandated that SANs should mirror CN quite a while back, so
> there may be scope for this soon for publicly-trusted certs. I believe
> Ryan had some views here...

Modern browsers: If they're willing to break the certs issued by some CAs that had 5-10 year validities right before the BRs came into effect, it should be reasonably accomplishable. However, this can *only* be accomplished if browsers recognize the distinction between "Chains to a BR compliant CA" and "Chains to a non-BR compliant CA" (e.g. enterprise CA). The former category should be mostly safe for CNs, save for the long-lived zombie certs. The latter category is dominated by CNs w/o SANs.

Other software: It will be a decade+ before such applications stop checking CNs.

And it is certs like https://crt.sh/?id=11880495 that prove that this is true.  That certificate was issued in the last month, has no SAN or EKU, has a FQDN in the CN, and chains back to one or more roots trusted by every browser (https://gist.github.com/pzb/c55a802c283c7b44002d).  https://www.ssllabs.com/ssltest/analyze.html?d=aflsa.jag.af.mil shows that the certificate is in use at the FQDN in the CN.

According to the ssllabs analysis, the chain should fail due to policy constraints, but I’m not sure if anything enforces such constraints.

Correct.

From top to bottom, the chain is:

  *   DST ACES CA X6 (trust anchor)
  *   IdenTrust ACES CA 1
  *   Federal Bridge CA 2013
  *   Federal Common Policy CA
  *   SHA-1 Federal Root CA
  *   DoD Interoperability Root CA 1
  *   DoD Root CA 2
  *   DOD CA-28
  *   aflsa.jag.af.mil

Using RFC5280 wording, valid_policy_tree is equal to NULL at « SHA-1 Federal Root CA », and the chain fails at the « DoD Root CA 2 » certificate, because now explicit_policy is equal to 0 (and valid_policy_tree is NULL).
Strangely, certificate for « DoD Interoperability Root CA 1 » has a PolicyMappings extension, but policy mapping is forbidden since « SHA-1 Federal Root CA ».
Strangely again, the PolicyConstraints extension contained in « DoD Root CA 2 » is not critical. It’s a MUST for RFC5280, and a MAY for X.509 (with a recommendation to set it critical). It’s the extension that requires an explicit policy and makes the chain fail.

OpenSSL is able to reject such a chain (it doesn’t seem to handle the policy mappings), I’ve been told that MS can correctly reject it, I’m pretty sure that the recent Mozilla PKIX lib accepts it while the previous one rejects it. (And I wouldn’t bet a cent on GnuTLS to do such things correctly)


Cordialement;
Erwann Abalea

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cabforum.org/pipermail/public/attachments/20160204/332de89c/attachment-0003.html>