[Smcwg-public] CAA and S/MIME

[Paul van Brouwershaven]

While the BR only specifies how CAA must be implemented/used for TLS certificates the CAA RFC is not limited to just TLS certificates, the RFC 8659 (and previously RFC 6844) begins with:


The Certification Authority Authorization (CAA) DNS Resource Record
   allows a DNS domain name holder to specify one or more Certification
   Authorities (CAs) authorized to issue certificates for that domain
   name.

This does make sense as this would create a `deny all, except` principle where you need to give explicit permission like as in a good firewall configuration. But I agree that there should be a possibility to change permission per certificate type and to drop restrictions where needed (such as for S/MIME with shared mail providers).

I'm currently not in favor of having a new S/MIME specific property, and one for client certs, document signing, etc. as where does it stop. It would also not allow to have separate `iodef` settings for example (or only enable them for TLS).

We could define one new parameter to define the certificate type, the standard already has a reversed policy property which was used in the draft RFC to limit issuance on policy OID. Rob pointed me at the draft CAA specification that had a policy property value instead of a CA domain name.
https://datatracker.ietf.org/doc/html/draft-hallambaker-donotissue-04#section-3.1.2

This could work with cabforum defined OID's, the advantage from using OID's is that it would support an OID prefix, and it would be easier for CAs to enforce. But it's not very user friendly..?

This would allow:


   $ORIGIN example.com

   .       CAA 0 issue "ca1.example.net; policy=2.23.140.1"   // Only cabforum

   .       CAA 0 issue "ca2.example.net; policy=2.23.140.1.5" // SMIME

   .       CAA 0 issue "ca3.example.net; policy=2.23.140.1.2" // DV, OV, IV

   .       CAA 0 issue "ca4.example.net; policy=2.23.140.1.1" // EV


For user friendliness, we could define a new parameter such as `type` with the same intention but based on a name value:

This would allow:


   $ORIGIN example.com
   .       CAA 0 issue "ca1.example.net"

   .       CAA 0 issue "ca2.example.net; type=smime"

   .       CAA 0 issue "ca3.example.net; type=tls"

   .       CAA 0 issue "ca4.example.net; type=tls-ev"

Where:

  *   ca1 could issue any type of certificate not explicitly specified (so no S/MIME, or TLS certificates in this example)
  *   ca2 could issue only smime certificates
  *

ca3 could issue any type of TLS certificate except for EV

  *

ca4 could issue only EV TLS certificates


Alternatively, we could define two parameters, one for the certificate type and one for the assurance level, this would give:


   $ORIGIN example.com
   .       CAA 0 issue "ca1.example.net"

   .       CAA 0 issue "ca2.example.net; type=smime"

   .       CAA 0 issue "ca3.example.net; type=tls"

   .       CAA 0 issue "ca4.example.net; type=tls; level=ev;"

The challenge for all these methods is how do we drop CAA limitations, as we want something like:


   $ORIGIN example.com
   .       CAA 0 issue "ca1.example.net"

   .       CAA 0 issue "*; type=smime"


But that is not allowed by the RFC.


If we would define a separate property per certificate type but follow the RFC and honoring the inheritance, we would still have the same challenge of stopping the inheritance .


   $ORIGIN example.com
   .       CAA 0 issue "ca1.example.net"

   .       CAA 0 issuesmime "ca2.example.net"

   .       CAA 0 issuetls "ca3.example.net"

   .       CAA 0 issuetlsev "ca4.example.net"


Maybe we could simply create an 'unrestricted' parameter to overrule the RFC?:


   $ORIGIN example.com
   .       CAA 0 issue "ca1.example.net"

   .       CAA 0 issue "; type=smime; unrestricted=true"


Paul

________________________________
From: Smcwg-public <smcwg-public-bounces at cabforum.org> on behalf of Tim Hollebeek via Smcwg-public <smcwg-public at cabforum.org>
Sent: Monday, October 26, 2020 15:25
To: Neil Dunbar <ndunbar at trustcorsystems.com>; SMIME Certificate Working Group <smcwg-public at cabforum.org>
Subject: [EXTERNAL]Re: [Smcwg-public] CAA and S/MIME


This is how I feel about the issue.  CAA is potentially an interesting improvement to the S/MIME ecosystem, but the current tags and implementation were meant for TLS, and shouldn’t be reused.



The RFC has an extension mechanism which can easily be used to add new tags for S/MIME issuance, and issuance of other kinds of non-TLS certificates.



-Tim



From: Smcwg-public <smcwg-public-bounces at cabforum.org> On Behalf Of Neil Dunbar via Smcwg-public
Sent: Monday, October 26, 2020 6:03 AM
To: smcwg-public at cabforum.org
Subject: Re: [Smcwg-public] CAA and S/MIME





On 24/10/2020 16:21, Stephen Davidson via Smcwg-public wrote:

Hello:



The topic of Certification Authority Authorisation (CAA) has arisen a number of times in relation to the evolving S/MIME Baseline.

I highlight a discussion on that subject related to the Mozilla policy: https://github.com/mozilla/pkipolicy/issues/135

A significant number of email providers – such as gmail.com, outlook.com, protonmail.com, and others – have CAA records.

Questions for us to address later in our discussions:



  *   Is CAA a desired requirement of the S/MIME Baseline?
  *   Should the S/MIME Baseline rely upon the existing requirements stated in the TLS BR, or is the S/MIME use case sufficiently different to merit a separate CAA tag?



Generally, I'm a fan of allowing organisations (however defined) to specify their policy requirements for publicly trusted certificates via CAA records; so I would say "yes", it is a desired requirement of the S/MIME baseline. I would certainly expect it to make its way into the root program requirements at some point, and having a pan-root program consensus on those requirements beats having overlapping or potentially conflicting requirements.

That said, I'm not a fan of ninja semantics (changing the meaning of a deployed resource where the deployer might not have considered its eventual full scope) - it seems to me that the "issue" and "issuewild" tags were framed with TLS certificates in mind[*], and I think extending "issue" to cover S/MIME could have effects on domain owners which were not expected. In other words, we would be saying to them that all certificates are hereby covered, without them having any means of expressing the policy "I want CA X to issue TLS certificates, but any CA could issue S/MIME certificates"; so I'm less of a fan of reducing the expressive potential of domain owners.

To that extent, I think that I'd prefer tags like "issue-tls", "issue-tls-wildcard", "issue-email", and so on, with similar semantics which work over "issue" and "issuewild" right now. Once those are in effect, then you could extend the "issue" tag to mean "all certificates" as a shorthand, while leaving finer detailed policy expressions. However, that goes further than anything the S/MIME WG could reasonably pronounce upon. But "issue-email" to cover S/MIME certs falls within its charter and seems to have a clearer scope. There's even an opportunity to allow domain owners to specify the validation methods permissible for issuance, but that's a whole different discussion.

Just my opinion, of course.

Neil

[*] Genuine question: would "issuewild" have any meaning outside of TLS certificates?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cabforum.org/pipermail/smcwg-public/attachments/20210129/54060d7c/attachment-0001.html>