Re: [dane] DANE Testing

[Stephen Nightingale <[e-mail]>]

On 2/21/2014 4:38 PM, Viktor Dukhovni wrote:
> On Fri, Feb 21, 2014 at 03:19:10PM -0500, Stephen Nightingale wrote:
>
>> It is now possible to test from both TLSlite based and GnuTLS based
>> clients.
> The GnuTLS DANE implementation does not implement DANE-TA(2) or
> correctly.  It may also IIRC not do PKIX-TA(0) right either.  The
> issue is that the TA is incorrectly constrained to be the immediate
> issuer of the leaf certificate.  There may be other problems, I
> did not perform a comprehensive code review.
>
> So results from GnuTLS DANE verification can be misleading.
I should point out that I am not using GnuTLS idea of DANE verification. 
I'm getting the cert chain passed up and doing my own DANE verification. 
For this purpose I had to modify the python-gnutls-1.2.5 interface to 
GnuTLS, to pass in a certificate chain, since in the download it only 
passes in a single end cert.



>
>> A golden, canonical CA list would be nice to find. But I guess that its
>> non-universal availability is one of the problems of the CA system
>> that DANE is aiming squarely at.
> There is no such beast.  A form of Goedel's incompleteness theorem
> applies: any list of CAs is either incomplete or untrustworthy (or
> both).
>
>> The differences between TLSlite and GnuTLS clients highlight the
>> fact that there are unresolved interoperability issues among TLS
>> implementations.
> Default configurations of GnuTLS typically enforce unreasonable
> minimum sizes on EDH primes.  Applications have to work around
> these with policy overrides.
>
>> I look forward to seeing the newly upgraded OpenSSL
>> client with added DANE.  It is quite possible that as an interim step
>> before its appearance I will add this DANE-in-the-App implementation
>> to pyOpenSSL and/or Twisted.
> Such a project should not be undertaken lightly, it is too easy to
> get it wrong.
>
>> If you find any glaring errors, I will be embarrassed but thankful.
>> If you find any subtle errors I will be impressed and thankful.
> A code review is required to rule out subtle errors, glaring errors
> may show up in tests if the test bed is sufficiently comprehensive.
>