Hi Ryan S.,

 

> Isn’t this effectively suggesting that a certificate, not a key, signs a 
> response? That doesn’t seem to square with any of the technologies in play.

 

> That is, such an OCSP response would still be signed by the root’s private 
> key, which is the key issue here.

 

My understanding is that the requirements surrounding the inclusion of the 
digitalSignature KU have in mind a theoretical client that will reject OCSP 
responses unless the client can build a certification path (through being 
presented such certificates through the “certs” field of the BasicOCSPResponse, 
local policy, etc.) where the signing key is certified to create digital 
signatures (i.e., KU includes digitalSignature). CAs are under an obligation to 
serve OCSP responses that will be accepted by this theoretical client, but we 
know from BR section 4.9.9 that CAs may employ OCSP delegated signing 
certificates to accomplish this.

 

Do you agree that this is the goal of the requirements, or do you see it 
differently?

 

> This approach would have significant comparability risk, for all the issues 
> that doppelgängers have to begin with, so it doesn’t seem to be functionally 
> any different. It seems like the goal is to work around the CPS issue, but 
> that wouldn’t address the underlying compliance issue, right?

 

This solution would provide the theoretical client described above with 
sufficient information to verify the signature on the OCSP response, so I 
believe it does resolve both the compliance issue and the practical client 
interoperability issue.

 

> Note: While the key reuse issue is the functional root problem, even ignoring 
> it, this option would have issues, in that the certs field of the 
> BasicOCSPResponse isn’t part of the signed data - it’s supplemental data. So 
> the signature is still not bound to the “new” certificate.

 

Not sure why this matters here, because this is a general issue for all 
delegated OCSP signing; an attacker can mangle the response to strip out such 
information so that the client rejects the response. But if they can do that, 
they can likely just block the response entirely from ever being received by 
the client, achieving the same goal.

 

Thanks,

Corey 

 

From: Ryan Sleevi <[email protected]> 
Sent: Wednesday, December 22, 2021 2:02 PM
To: Corey Bonnell <[email protected]>
Cc: Ryan Dickson <[email protected]>; [email protected]
Subject: Re: Root Replacement with digitalSignature Key Usage to Sign OCSP 
responses

 

 

 

On Wed, Dec 22, 2021 at 1:37 PM 'Corey Bonnell' via 
[email protected] <mailto:[email protected]>  
<[email protected] <mailto:[email protected]> > 
wrote:

There may be a third option which avoids the complexity of OCSP delegated 
responder key management but resolves the compliance and potential client 
interoperability issues. The root can issue a self-signed end-entity OCSP 
responder certificate (same name + key as the root, but BC cA=False and EKU = 
id-kp-OCSPSigning and KU=digitalSignature, etc.) and all subsequently generated 
OCSP responses will include this certificate in the “certs” field of the 
BasicOCSPResponse. This will provide clients with the certificates necessary to 
validate the OCSP response signature, as they can build a certification path of 
root -> OCSP delegated responder certificate. Would such a solution also 
remediate the issues?

Isn’t this effectively suggesting that a certificate, not a key, signs a 
response? That doesn’t seem to square with any of the technologies in play.

 

That is, such an OCSP response would still be signed by the root’s private key, 
which is the key issue here.

 

This approach would have significant comparability risk, for all the issues 
that doppelgängers have to begin with, so it doesn’t seem to be functionally 
any different. It seems like the goal is to work around the CPS issue, but that 
wouldn’t address the underlying compliance issue, right?

 

Note: While the key reuse issue is the functional root problem, even ignoring 
it, this option would have issues, in that the certs field of the 
BasicOCSPResponse isn’t part of the signed data - it’s supplemental data. So 
the signature is still not bound to the “new” certificate.

-- 
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/DM6PR14MB2186BB293880AC62EDBA14EF927D9%40DM6PR14MB2186.namprd14.prod.outlook.com.

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to