> On May 11, 2016, at 1:02 AM, Aaron Zauner <[email protected]> wrote:
> 
>> With only one MTA open-source supporting DANE at present, and only with
>> relatively recent releases, and OpenSSL 1.1.0 still (for a couple more
>> weeks) in beta, it is not surprising that deployment is still light.
>> We can't expect deployment when supporting code is not yet widely
>> available.
> 
> That's more to the point, thanks for the data. Are your raw data-sets
> public anywhere (e.g. scans.io)?

Since more than one of my sources is non-public, I think I am not at
liberty to publish the raw underlying data. :-(  Sorry about that.

The main thrust of the survey is to find DANE-enabled domains whose
records go stale, and nudge them to fix.  Ditto with DNSSEC zones
that drop TLSA queries or mangle denial of existence.

This effort has payed off in that both sets of problems are quite
rare, and would not have been otherwise.

With the early adoption cycle protracted, and early enthusiasm
to publish records not necessarily translating to operational
discipline, it is important to artificially maintain hygiene
until real incentives kick in due to broader adoption.

Once more senders enable DANE outbound, receiving systems with
broken records will have real incentives to fix problems quickly.
However, if the early adopters were all messed up, there'd be
strong disincentives for broader adoption.  So I've lowered
the barriers to adoption by keeping the error rates much lower
than they would naturally have been at this stage.

Making the data public was not a primary goal, though I have
posted some aggregate summaries from time to time at
[email protected], [email protected], [email protected]
and this list.

There are still more HOWTO documents to write, a bit of time
to invest in improving the experimental support in Exim, perhaps
more TLS libraries to augment beyond OpenSSL, ...

This takes time, but so far so good.  The German and Dutch
domains are an adequate incubator.  The lessons will be
applied to better deployment guides for those who deploy
later.

It me a while to realize that a combination of "3 1 1"
and "2 1 1" records provides optimal operational resilience.
This allows decoupling of DNS updates from cert chain
updates, provided that at any given time at least one of
the issuer CA or server public keys remains unchanged.

One can deploy a new server keypair with a cert signed
by the same issuer without pre-publishing an updated
TLSA RRset, and then update the TLSA RRset some time
after.  Or deploy a new cert for the same server keypair
from the same issuer.

This is especially well suited for Let's Encrypt, but
generalizes also to use of a private issuing CA.

This will be written up in more detail in the next
few months, bundled with Postfix docs, and ideally
also as an informational RFC on DANE best-practice.

-- 
-- 
        Viktor.

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

Reply via email to