Re: [websec] #58: Should we pin only SPKI, or also names

[Tobias Gondrom <[e-mail]>]

On 11/08/13 16:54, Trevor Perrin wrote:
> On Sun, Aug 11, 2013 at 4:34 AM, Tobias Gondrom
> <[e-mail]> wrote:
>> Hi all,
>>
>> <no hats>
>>
>> A small question about pinning to names and uniqueness of "pinned names":
>> Under which conditions could the following attack scenario be a problem
>> and what would we do about it?
>>
>> Domain A has bought a cert from CA-11 with the name "super safe" and
>> pins to the name instead of the cert.
>> CA-11 could be an intermediate for CA-1 with the name "uber super safe".
>> Which names could we pin to?
>> - the intermediate CA-1 and/or CA-11?
>>
>> Now wondering whether the following is a problem:
>> attacker gains control over CA-2 (either through an attack or through
>> government influence) and issues a certificate for an intermediate
>> CA-11' with the name "super safe".
>
> Yes, that's a problem.  Gerv also brought up the "DigiCert" name collision.
>
> So using names as a "layer of indirection" to point to a set of
> CA-declared keys seems better than trying to pin to names as they
> appear in certs.
>
>
> Trevor

Thank you for the clarification.
In that case I would prefer not to pin to names. As the above scenario
is exactly the one we wanted to avoid in the first place.
We need to allow a domain to link to one specific cert or a specific set
of certs (like from one specified CA), and by this avoid that they are
exposed to risks by a breach of any other CA.

If we really need a to pin to a group of certs, maybe one other idea
might be to allow to pin to a top-node of a CA directly (but not
intermediaries as we would have the same attack scenario with them, too.)

Best regards, Tobias