On Dec 12, 2011, at 7:19 AM, Yoav Nir wrote:

> 
> On Dec 12, 2011, at 5:06 PM, Paul Hoffman wrote:
> 
>> 
>> On Dec 12, 2011, at 6:57 AM, Yoav Nir wrote:
>> 
>>> Section 2.3 does say that pins must only be noted if they come over an 
>>> error-free HTTPS connection, where the chain contains the pinned SPKI. So 
>>> the MITM would have to be able to take over the DNS (at least for a while) 
>>> as in the attack described by David on 11-Dec.
>> 
>> I don't agree that the MITM must control the DNS for this attack to work. As 
>> long as the MITM can intercept the TLS handshake, and the client isn't doing 
>> OCSP (or the MITM intercepts that too), using a rogue cert works; thus the 
>> rogue pinning works as well.
> 
> The rogue cert works only if it validates. So the attacker will have to have 
> compromised the CA as well.

Of course. That is exactly the scenario (and pretty much the only scenario) 
where key pinning gives the relying party better security.

> Besides, one could argue that using a certificate without OCSP or checking 
> the CRL is not "error-free", but that would have to be clarified.

We disagree. OCSP and CRL checking only bring benefit to the relying party if a 
certificate is revoked. The key pinning scenario is about a rogue CA issuing a 
certificate they should not have. There is no revocation issue in that 
scenario, is there?

>>> Maybe we can mitigate this with a version of this header (or another one) 
>>> that says "no pins for the next x seconds"
>> 
>> That still doesn't deal with the site that hears about the pinning header 
>> after the MITM does.
> 
> Agree, but it allows me to make my site safe from rogue pinning attacks now.

Sure, but I don't think that is a good enough reason to argue against more 
information in the spec about this problem, which is what I am requesting.

--Paul Hoffman

_______________________________________________
websec mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/websec

Reply via email to