dmarc

A bunch of email providers announced DMARC which builds on top of SPF and DKIM to allow domains to specify more policy when SPF and/or DKIM fail.

I already have SPF and DKIM set up on my personal domain, hawaga.org.uk, which has been round for over a decade. I run mail servers for various other domains, but those are much younger and much less widely used.

Its been hard to quantify how much this has helped/not helped. I don't get complaints about spam originating from my address. I used to get a lost of postmaster backscatter but not any more - not sure why, though I can invent various possible reasons.

One of the interesting things with DMARC is that it claims to provide feedback about what filtering is happening, from receiving/filtering parties - that makes it especially interesting, I think.

So, given that I already have DKIM and SPF, what extra do I need to do to get something useful from DMARC?

I need to publish a policy in DNS, under my sending domain. (this is also how SPF and DKIM do things)

So I've put in this policy on the 4th of Feb:
_dmarc.hawaga.org.uk. 3583 IN TXT "v=DMARC1\;p=none\;rua=mailto:benc@hawaga.org.uk\;ruf=mailto:benc@hawaga.org.uk\;ri=3600"
That says to not enforce any policy, but to email benc@hawaga.org.uk with reports every 3600 seconds (= 1 hour).

I set this up at about 5pm on a Saturday and about 11am on Sunday morning my first report arrived, with a timestamp range of a day, which must extend back before I turned this on...

In there, three messages from my main outbound mail server, and no others.

Lets see what else I get...

A couple of weeks later...

I got daily reports most days from Google (I think maybe the day I didn't get a report was because I hadn't sent any mail into google all(?)).

A few days after the above I added in two other domains: my company domain with only occasionally sends mail, and my girlfriend's vanity domain. Neither of those have SPF or DKIM on them, even though they come from the same mail servers as hawaga.org.uk.

There was a noticeable lack of reports from anyone other than Google. I asked around (on Google+) to see if anyone had reports from elsewhere (eg AOL or Yahoo, because those were also listed) but no one said yes.

So what about the reports?

Well, there were surprisingly more mail servers than I expected: along with my own two outbound servers, there were about 10 other servers, being the outbound mail servers of a handful of research institutes that I work with. Those reports were tagged by google as being via a mailing list. Its not clear to me what defines a message as being via a mailing list, but I guess it would mean that they'll put less weight on my SPF records? It also highlights how a naive interaction between mailing lists and SPF can result in your message being treated as spam.

I also got some DKIM fails reported from my own legitimate mail server. The best I have been able to diagnose there is that I had sendmail set to deliver mail without a DKIM signature if the dkim milter timed out; but if that's going to contribute negatively to spam treatment, then I think a better configuration is to have the milter set to retry later, resulting in more delayed mail, but more DKIM-signed mail.

The extra domains I added had no DKIM on them, but those weren't treated as DKIM-fails. Instead they were reported as DKIM 'none'. I'm not sure what causes none rather than fail, but my guess is its something to do with the fact that hawaga.org.uk has DKIM records in its DNS, and thats being treated as an indication that there should be DKIM signatures on messages. I think that's extra meaning that I hadn't understood DKIM DNS records to mean.

I have a similar confusion with the interaction between SPF and DMARC: SPF has multiple output states, not just pass or fail, and its not clear to me how those are treated by DMARC.

Processingwise: the reports come as zipped XML documents. It was relatively straightforward to munge these like any other XML (though I made it harder for myself by learning a new Haskell XML library rather than using ones I already knew).

Its unclear to me how I know that a report really is from a particular sender, and what the threat model is for people injecting false DKIM reports - perhaps injecting them to suggest that people's use of DKIM and SPF is causing their mail to be dropped, and thus encouraging them to turn off SPF and DKIM?

So for now, I'll keep this switched on, in monitoring-only mode. I don't feel I understand it well enough to turn it on in enforcement mode (especially as I'm not the only user sending mail under hawaga.org.uk. I think its very interesting and probably useful to be able to specify policy this way; but the policy language at the moment feels either vaguely defined, or at the least not concisely described, in a way that makes me comfortable.