Actually, the inheritance mechanism should probably be prohibited
altogether.

As things stand it allows a CA to poison a cert issued by another CA!

Let us imagine CA X issues a cert for Alice, the chain is X -> XI -> A. Now
imagine that CA Y issues a cross cert for XI, we now have a chain Y-> XI ->
A

The parameters for the key now depend on the path! Yuk!


I don't know if there is a practical attack possible , this just occurred
to me. But even if there is no attack there is potential for error.

[The sort of attack I am thinking of is a kleptography like attack when
country A sells sabotaged crypto equipment to country B. Country B may
verify that only certs issued through the sub CA XI they control are used
but this gives a way to defeat that control.]


On Tue, Dec 13, 2011 at 10:20 AM, Adam Langley <[email protected]> wrote:

> On Tue, Dec 13, 2011 at 7:56 AM, Phillip Hallam-Baker <[email protected]>
> wrote:
> > I don't like a solution for pinning that depends on the CA delivering the
> > 'right' sort of cert. I would prefer to add in a second hash over the
> > parameter values or specify them explicitly in the pin or to have the
> hash
> > be over what the values would be if completely specified in the Key Info.
>
> In the case of ECDSA, it'll be very rare for the parameters to be
> omitted since they can be compactly represented by a named curve. In
> fact, I've not seen code in SSL libraries that'll go hunting for EC
> parameters in a CA key - I strongly suspect that support for this is
> minimal at best.
>
> Therefore I'm happy to simply exclude the possibility in the spec and
> save the complexity.
>
>
> Cheers
>
> AGL
>



-- 
Website: http://hallambaker.com/
_______________________________________________
websec mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/websec

Reply via email to