Re: [Trans] responses to Ryan's detailed comments on draft-ietf-trans-threat-analysis-15

[Tim Hollebeek <[e-mail]>]

My comment is correct for the situation as it largely exists today.



That it might be different in the future is a fair point.



-Tim



From: Ryan Sleevi <[e-mail]>
Sent: Friday, September 28, 2018 9:04 AM
To: Tim Hollebeek <[e-mail]>
Cc: Ben Kaduk <[e-mail]>; Ryan Sleevi <[e-mail]>; Trans 
<[e-mail]>; [e-mail]
Subject: Re: [Trans] responses to Ryan's detailed comments on 
draft-ietf-trans-threat-analysis-15





On Fri, Sep 28, 2018 at 10:37 AM Tim Hollebeek <[e-mail] 
<mailto:[e-mail]> > wrote:


> For what it's worth, I do not read 6962-bis as "very much being focused"
on
> CA-based logging.  Consider, for example, certificate subjects submitting
> certificates to logs, something that is done without CA involvement and
can be
> done in response to (e.g.) Logs being distrusted or browsers increasing
the
> required number of SCTs.  It's unclear that CAs have as much incentive as
> subjects to be responsive to changing events in this way.

SCTs have to be included in the certificate so logging by third parties does
not help with that problem.



This is not correct. Conforming clients MUST support all three methods of 
delivery of SCTs. No policy or statement in 6962-bis requires that SCTs "have 
to be included in the certificate". Perhaps you're conflating the requirement 
with SCTs for precertificates, which has been substantially overhauled in 
6962-bis in light of the concerns around precertificates?



I'm not sure where this view that the dominant form is or should be CA 
initiated logging, or that the intent of 6962-bis is to only countenance that 
scenario. Over the past 28 days, 46% of the SCTs Chrome has observed have come 
from the TLS extension, and 53% of them embedded within certificates. 0.01% 
have come from OCSP responses. This latter number is no doubt driven by at 
least three CAs who have a largely homogenous user base (of government and 
public sector users) running on Microsoft-based services, that they were 
confident enough that they only needed to support OCSP embedding for their 
subscribers.



Any threat model design needs to consider 6962-bis as specified, which is to 
consider that these different approaches are all equally valid.