Re: [dane] Improving DANE S/MIME Privacy

[Viktor Dukhovni <[e-mail]>]

> On Apr 12, 2017, at 11:59 PM, Phil Pennock <[e-mail]> wrote:
> 
>> I should note that one can of course implement one's SMIMEA deployment
>> in exactly this way, something along the lines of:
>> 
>>   *._smimecert.example.net. IN SMIMEA 2 1 1 e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
>> 
>> would associate the same TA public key digest with every user, and would
>> not enable user enumeration.
> 
> FWIW, I did something similar to this.  Of course, there's a chicken/egg
> problem in _getting_ the CA cert for private CAs, unless you put the
> whole cert into DNS.  And wildcards for something that large would be
> cache-unpleasant, but I did the same thing I do for TLSA records: define
> one SMIMEA record and CNAME to it elsewhere.  Much more cache friendly.
> 
> I never thought I'd see the day that I willingly chose to deploy a
> wildcard CNAME.   :^D
> 
> $ORIGIN spodhuis.org.
> *._smimecert            CNAME   _globnix-smimea
> _globnix-smimea         SMIMEA  (       ; GlobnixCA5 PKIX-less trust anchor
>        02 00 00
> ; SKIP: binary blob for an ECDSA CA
>        ; SMIMEA DANE-TA CERT FULL
>        )
> 
> I can't attest that this _works_.  As far as I know, it's entirely
> draft-spec compliant.

Note, that provided your MUA includes the issuing certificate in the
signature block (you can always make it an intermediate CA issued
by a throw-away root whose private key has been destroyed, and then
the MUA should include at least all the intermediates), there's no
need to use "SMIMEA 2 0 0" with the "large" DNS payloads that this
entails.  You get the same mileage from "SMIMEA 2 1 1".

That said, ECDSA certificates are often considerably smaller than
is the case with RSA, so "2 0 0" may be sufficient small to avoid
issues with UDP MTUs.

Still, I'd just go with "SMIMEA 2 1 1".

-- 
	Viktor.