On Thu, Dec 27, 2018 at 9:09 AM Ryan Sleevi <[email protected]> wrote:

>
>
> On Thu, Dec 27, 2018 at 11:43 AM Rob Stradling <[email protected]> wrote:
>
>>
>> Proposed text:
>> https://github.com/google/certificate-transparency-rfcs/pull/305
>>
>> This PR takes an axe to the "Accepting Submissions" section, splitting
>> it into two subsections in order to (I think) more clearly specify (1)
>> what are the minimum rules for acceptable submissions and (2) what's
>> left to the log's discretion.
>>
>> I've added text to the minimum rules subsection regarding checking the
>> Basic Constraints and Key Usage extensions.
>>
>> I'd like to hear your thoughts about the general approach of this PR as
>> well as its detail.  Thanks!
>>
>
> Just to highlight and ensure it's intentional:
> 1) This would prohibit X.509v1 and X.509v2 (by virtue of requiring at
> least one extension)
> 2) This would prohibit RFC 3820 Proxy Certificates
>
> I totally accept that this is a wobbly gray area - some arguing only
> properly DER-encoded, fully-profile-conforming certificates should be
> accepted (which makes writing MUSTs easier and, by proxy, is assumed to
> clean up the ecosystem), others wanting to capture the world-as-it-is, not
> the world-as-it-should-be. Is CT a means of achieving the PKIX Protocol
> Police, or is it the camera that merely records all, without passing
> judgement or opinion?
>
> That said, I'm not sure the motivations for why these requirements are
> there are supported by the text. That is, expectations about what a monitor
> will find in the log aren't necessarily set (because the trust anchors may
> change over the lifetime of the log), and likewise, the attribution issue
> is problematic because, again, the trust anchors accepted by the log may
> chance over time. The primary reason for any of this largely is about
> preventing Log DoS (whether from spam or content/profile, collectively,
> "invalid submissions"). Basic constraints and usage are really just
> variants of DoS prevention, right?
>

What's fraught about this from my perspective is that the current text
requires that the server *reject* certificates which don't conform to
certain rules, apparently in the name of DoS protection. As a result:

- From a formal perspective, the set of certificates which must be rejected
needs to be unambiguous
- From a technical perspective, we should ensure that it's not trivial to
create DoS attacks with certificates that are not required to be rejected
(otherwise, the requirement is pointless)

-Ekr
_______________________________________________
Trans mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/trans

Reply via email to