Dipping My Feet Into DKIM and DMARC
Apr 10, 2023, 10:56 AM
For a very long time now, I've had my mail set up in a grandfathered-in free Google Whatever-It's-Called-Now account, which, despite its creepiness, serves me well. It's readily supported by everything and it takes almost all of the mail-hosting hassle out of my hands.
Not all of the hassle, though, and over the past couple weeks I decided that I should look into configuring DKIM and DMARC, first for my personal mail and (if it doesn't blow up) for my company mail. I had set up SPF a couple years back, and I figured it was high time to finish the rest.
As with any admin-related post, keep in mind that I'm just tinkering with this stuff. I Am Not A Lawyer, and so forth.
DKIM is a neat little standard. It's sort of like S/MIME's mail-signing capabilities, except less hierarchical and more commonly enforced on the server than on the client. That "sort of" does some heavy lifting, but it should suit to think of it like that. What you do is have your server generate a keypair (Google has a system for this), take the public key from that, and stick it in your DNS configuration. The sending server will then add a header to outgoing messages with a signature and a lookup key - in turn, the receiving server can choose to look up the key in the claimed DNS to verify it. If the key exists in DNS and the signature is valid, then the receiver can be fairly certain that the receiver can at least be confident that the sender is who they say they are (in the sense of having control of a sending server and DNS, anyway). Since this signing is server-based, it requires a lot less setup than S/MIME or GPG for mail users, though it also doesn't confer all the benefits. Neat, though.
DMARC is an interesting thing. It kind of sits on top of SPF and DKIM and allows an admin to define some requested handling of mail for their domain. You can explicitly state that you expect your SPF and DKIM records to be enforced and provide some guidance for recipient servers to do so. For example, you might own "foo.com" and go whole-hog: declare that your definitions are complete and that remote servers should outright reject 100% of email claiming to be from "foo.com" but either didn't come from a server enumerated in your SPF record or lack a valid DKIM signature. Most likely, at least when rolling it out, you'll start softer, maybe saying to not reject anything, or to quarantine some percentage of failing messages. It's a whole process, but it's good that gradual adoption is built in.
Interestingly, DMARC also lets you request that servers that received mail from "you" email you summaries from time to time. These generally (always?) take the form of a ZIP attachment containing an XML file. In there, you'll get a list of servers that contacted them claiming to be you and a summary of the pass/fail state of SPF and DKIM for them. This has been useful - I found that I had to do a little tweaking to SPF for known-good servers. This is vital for a slow roll-out, since it's very difficult to be completely sure you got everything when you first start setting this stuff up, and you don't want to too-eagerly poison your outgoing mail.
Really, configuring this stuff wasn't bad. I mostly followed Google's guides for DKIM and DMARC, which are pretty clear and give you a good plan for a slow rollout.
Though Google is my main sender, I still have some older agents that might send out mail for my old ID from time to time from Domino, so I wanted to make sure that was covered too. Fortunately, Domino supports DKIM as well, and it wasn't too bad. Admittedly, the process is a little more "raw" than with Google's admin site, but it's not too bad. It's not like I'm uncomfortable with a CLI-based approach, and it's in line with other recent-era security additions using the
keymgmt tool, like shared DAOS encryption.
It just came down to following the instructions in HCL's docs and it worked swimmingly. If you have a document in your cred store that matches an INI-configured "domain to ID" value for outgoing mail, Domino will use it. Like how DMARC has a slow-roll-out system built in, Domino lets you choose between signing mail just when available or being harsher about it, and refusing to send out any mail it doesn't know how to sign. I'll probably switch to the second option eventually, since it sounds like a good way to ensure that your server is being a good citizen across the board.
In any event, this is all pretty neat. It's outside my bailiwick, but it's good to know about it, and it also helps reinforce a pub-key mental model similar to things like OIDC. It also, as always, just feels good to check a couple more boxes for being a good modern server.