[cabf_validation] Validation methods used for Wildcards/ADNs
Corey.Bonnell at digicert.com
Tue Feb 2 20:13:57 UTC 2021
This WG discussed the allowance for HTTP-based domain validation to be performed at the ADN in late 2018 as a potential item to address as part of SC25. Given that the group decided to not modify this allowance in SC25, I’m wondering what the motivation is for this change now, especially considering that you mentioned “we've become aware of some CAs having poorly evaluated the security risks in this space”. Is this referring to a new threat that has been identified, or something else?
From: Validation <validation-bounces at cabforum.org> On Behalf Of Ryan Sleevi via Validation
Sent: Wednesday, December 2, 2020 8:32 PM
To: CABforum3 <validation at cabforum.org>
Subject: [cabf_validation] Validation methods used for Wildcards/ADNs
I know we're not quite done with the certificate profile work, and I'm not wanting to distract from that too much. However, one of the long-standing items we had from our Herndon, VA validation summit (from Meeting 43) was in harmonizing the rules around what 188.8.131.52 methods can be used for Authorization Domain Names / Wildcard Domain Names.
I made an initial attempt at https://github.com/cabforum/servercert/compare/main...sleevi:2020-12-01_Wildcard_Rules to capture this. In effect, allowing validation as an ADN is conceptually "the same as" allowing a Wildcard Domain Name, since the ADN can authorize all children/grandchildren/etc of a domain, and a Wildcard is just a cert that works for all children of a domain.
As we've become aware of some CAs having poorly evaluated the security risks in this space, we'd like to try to close this gap. Here's the TL;DR summary
* 184.108.40.206.6: Agreed-upon Change to Website
* Sunset 2020-06-03 for new validations
* 220.127.116.11.18: Agreed-upon Change to Website v2
* 18.104.22.168.19: Agreed-upon Change to Website - ACME
(The other bits are just aligning some of the language, so that "MAY NOT" becomes a clearer "MUST NOT", even though we mean the same)
These methods are proposed to *only* authorize a single FQDN, because they only demonstrate control over a specific service/port on a specific FQDN, and not demonstration of control over the whole domain namespace. This aligns with 22.214.171.124.20 (TLS using ALPN), which also only demonstrates control over a single service/port on a single FQDN.
This doesn't touch 126.96.36.199.4 (Constructed Email to Domain Contact), although we identified that one as potentially messy. However, hopefully we'll see that one fully sunset separately, in favor of the improved CAA methods (.13 - .17).
It'd be useful to spend a few minutes on the call discussing folks initial reactions. The big question, as always, is going to be timelines for changes. If folks think more time is needed than "immediately", my request is that they'd share concrete data.
Since Ballot 190 (2017-09-19), CAs have been required to maintain records of the validation methods they use, so this "should" be as easy as scanning all unexpired validations for these three methods and identifying cetrs which have a SAN that doesn't equal the validated FQDN (e.g. a cert with "www.example.com <http://www.example.com> " when the method used was 188.8.131.52.6 for "example.com <http://example.com> "). Just sharing those numbers is useful to understand any challenges CAs might face.
* 30% of our certificates used 184.108.40.206.6. Of that 30%:
* 80% of our certificates had at least one FQDN validated by ADN, with 40% of that being "www.".
* Of the 20% that had >1, we saw an average of 7.3 additional FQDNs validated by FDN.
* 17% of our certificates used 220.127.116.11.18. of that 17% ....
* 80% of the FQDNs validated by ADNs were for domains that did not resolve (e.g. "internal.corp.foo.example"), and thus would have to switch to a new validation method or expose those services publicly.
This sort of concrete data helps understand the impact to CAs, and their customers, and thus indirectly, our users. It also helps figure out what reasonable time frames to phase in could be, in the unlikely event a phase-in became necessary.
This sunset "should" be fairly simple and uncontroversial, but since there are edge cases (like internal servers), concrete data like the above is useful if folks have concerns.
-------------- next part --------------
An HTML attachment was scrubbed...
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 4990 bytes
Desc: not available
More information about the Validation