Re: Requirements for Secondary Certificates (#522)

On Wed, Apr 11, 2018 at 1:01 AM, Amos Jeffries <squid3@treenet.co.nz> wrote:

> On 11/04/18 08:20, Jeffrey Yasskin wrote:
> >
> > If the main problem is that Secondary Certificates make detecting
> > compromise more difficult, would it help to have clients make a parallel
> > connection to the DNS-discovered IP address that simply reports who's
> > using the server's identity? I think it's safe for this connection to
> > fail open, since if the attacker's network-privileged, they didn't need
> > to use Secondary Certs.
>
> I don't think it would help at all. Recall that DNS/BGP is already
> compromised in order to perform the attack at all. So any side
> connection is just as easily caught and faked as the original was.
>

I think there may have been a misunderstanding. In the Secondary
Certificates case, with key compromise, you can mount the attack without a
DNS/BGP compromise. That is, total effort to mount attack = (key
compromise). Jeffrey's suggestion is to force a connection establishment -
even in the case of a Secondary Certificate offer - such that the effective
steps to compromise require (key compromise) + (BGP hijack, DNS compromise,
on-path presence). Which is roughly the status quo for today in a
non-Secondary Certificate case.