Re: [dane] Behavior in the face of no answer?

[Paul Hoffman <[e-mail]>]

On May 4, 2012, at 9:18 AM, Adam Langley wrote:

> On Fri, May 4, 2012 at 12:08 PM, Eric Rescorla <[e-mail]> wrote:
>> Before we discuss how to proceed, I think it would be useful to get agreement
>> on the security analysis. I claim that for Usages 0 and 1, treating
>> TLSA non-response
>> as if no TLSA records exist means that DANE adds minmal/no security
>> value for those
>> usages.
> 
> I believe that you're completely correct.
> 
> Browsers are not going to enable DANE hard fail in the short or medium
> term because of the fraction of clients that have broken DNS and can't
> fetch the records.
> 
> Therefore, browsers are not going to see a security benefit from DANE.
> The DNSSEC stapled certificates that Tom points to are addressing a
> very different use case: people who would otherwise use self-signed
> certificates and who get annoyed that we throw errors for them.
> 
> However, browsers are not the world. Other clients, which may benefit
> from a smaller, more controlled, deployment may be able to enforce
> DANE hard-fail and see security benefits.
> 
> I would personally suggest that the spec requires hard fail, or at
> least discusses the fact that, without hard-fail, the security
> benefits are moot. Implementations frequently ignore requirements in
> specs, but rarely add to them!

My preference is that the spec discuss the issue, propose hard-fail as an option, and explain that without hard-fail the benefits for usage 0 and 1 can be circumvented but that the client is no worse off than without DANE. It's important to emphasize that this is about usage 0 and 1, and that a different security analysis applies to types 2 and 3 (well, 3 certainly; I need to think about 2).

I would prefer that we not require hard fail while assuming some implementers will not follow our requirement, since that leads to us being dishonest about what we expect. If we believe (as Adam says) that implementations ignore these types of requirements, let's be honest about that and let the implementer know what it means to ignore the suggested action.

--Paul Hoffman