Dear Ben,

Thanks for your effort to re-ignite the discussion.

Personally, not speaking as a representative of my employer, I suggest two 
things to balance the interests:

  1.  Remove the 3rd section "Mozilla recognizes…" from 
https://wiki.mozilla.org/CA/Responding_To_An_Incident . This would IMHO clarify 
that no exceptions are allowed.
  2.  Introduce a third deadline of 15 (or 20) days in addition to the 24h and 
5days deadlines in TLS BR 4.9.1.1 to cover certificates that were correctly 
validated, have all the right technical attributes (OID, key length, EKU, …) 
but don't 100% comply with the TLS BR or the CAs CP/CPS.

Finding good and workable solutions is often a give-and-take from all involved 
parties. Maybe by taking away the leeway for exceptions but giving the 3rd 
deadline is something that leaves all sides both happy and unhappy enough to 
accept it? 😉

Rgds
Roman

From: 'Ben Wilson' via [email protected] 
<[email protected]>
Sent: Mittwoch, 24. Juli 2024 23:06
To: [email protected]
Cc: Mike Shaver <[email protected]>; Matt Palmer <[email protected]>; Tim 
Hollebeek <[email protected]>; Amir Omidi (aaomidi) 
<[email protected]>; Wayne <[email protected]>; Jeremy Rowley 
<[email protected]>
Subject: Re: Feasibility of a binding commitment to revoke before issuance


Thanks, everyone, for keeping this conversation going. It's essential that we 
continue because I believe the current framework is unworkable.

Ben


On Wed, Jul 24, 2024 at 2:53 PM Mike Shaver 
<[email protected]<mailto:[email protected]>> wrote:
On Wed, Jul 24, 2024 at 2:36 PM 'Ben Wilson' via 
[email protected]<mailto:[email protected]> 
<[email protected]<mailto:[email protected]>> wrote:

Personally, I currently favor extending the timeframe for the revocation of 
certificates that have no security impact,
I propose, tongue only slightly in cheek, that if a component of the 
certificate doesn't have security impact, it be removed from the certificate 
specification in the BRs. TLS certificates are precious space, and reducing 
their size by eliminating things that have no use in security contexts

 Are there any specific examples of what deviations of certificates would be 
deemed so minor that they can stay live on the web for 20 days, but still 
worthy of revocation? With the number of CAs out there, and the rate of 
certificate issuance and error related there-to, it would be a virtual 
guarantee that there are a number of flavours of "slightly wrong" certificates 
active on the web. That means that everything needs to handle such certificates 
existing, in order to operate as part of the web PKI, so we should just capture 
in the standard that this alternative shape is OK and let everyone issue in 
that expanded envelope all the time. Presumably the web would benefit from this 
in some way, if the CABF would entertain such changes, but I confess that I 
can't tell what that benefit is.

Similarly, I might ask: how much of a grace period should Firefox give for 
accepting a certificate after it has expired? I mean, what's 20 days? It 
expired naturally, after all...

More seriously, I don't think that we are generally in a position to be certain 
that no system exists which depends on a certain property of a certificate. Is 
there something out there that is gating access or acting differently on the 
basis of a case-sensitive country code match? If there is, the designers 
certainly weren't wrong to build it that way, IMO. The BRs are a commitment to 
the world that web PKI certificates will behave a certain way, and laxity on 
making sure that certificates actually do conform will mean that effectively 
the BRs are no longer really true or useful for their purpose.

In addition to whatever subjective assessment a CA might make (hardly as a 
disinterested party, I hasten to add) about the security implications of a 
given certificate's deviation from, there is also the concern of 
interoperability. A new entrant to the web (such as a browser like Ladybird, or 
a CA like the next Let's Encrypt, or some future CDN Fastflare) will need to 
not only implement to meet and handle the *specified* certificate forms and 
behaviours, but also somehow know about all the kinds of variants that are 
likely to be floating around at any given time. Mozilla and Firefox know 
first-hand and extensively what a barrier it can be when the standard says that 
things should be one way, but other parties produce or expect something 
different.

Finally, conformance to the standards and correct issuance is just not that 
hard, as regards the things that have been argued to be "too minor to revoke in 
5 days". They would virtually all have been caught by decent linting. I don't 
see how it helps the web to make these cases easier for CAs to handle. It seems 
only that it would benefit CAs who routinely misissue sloppy certificates in 
"minor" ways. If they can't get these little things right, how can we trust 
their key material management or background checks or entropy sources? It's not 
like we're seeing the raw audit reports, even though they are really for the 
benefit of the root programs.

Maybe the job of being a CA is too hard for some organizations that are doing 
it now. That's OK. The Web doesn't need all of the CAs we have today as much as 
it needs CAs that help move the integrity of web PKI *forward*, rather than 
weakening it a little bit at a time for their convenience when they have failed 
to meet their commitments.

Mike

--
You received this message because you are subscribed to the Google Groups 
"[email protected]<mailto:[email protected]>" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to 
[email protected]<mailto:[email protected]>.
To view this discussion on the web visit 
https://groups.google.com/a/mozilla.org/d/msgid/dev-security-policy/CA%2B1gtaYACVE_sN_OdvczL_MKTX-sVE8PyFEhxfCoPRxi7CG04g%40mail.gmail.com<https://groups.google.com/a/mozilla.org/d/msgid/dev-security-policy/CA%2B1gtaYACVE_sN_OdvczL_MKTX-sVE8PyFEhxfCoPRxi7CG04g%40mail.gmail.com?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups 
"[email protected]" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/a/mozilla.org/d/msgid/dev-security-policy/ZR0P278MB017048C6272071B8B12B1C09FAAB2%40ZR0P278MB0170.CHEP278.PROD.OUTLOOK.COM.

Reply via email to