A few days ago at our family dinner I talked about how Alexa Bliss was setting off Amazon Echos during her matches. This is a slightly funnier, and less expensive, version of the TV Report prompts Amazon Echos to buy dollhouses story. I showed my wife a video of how the commentators were saying her name over and over, and an Echo was responding.
My youngest son said it would be cool to have one, and asked if we could get one. I said no. My wife and I are on the same page about this, but the idea of a device, which I have no control over, listening to everything being said is not something we want in our house. It's not just me not liking the Amazon Echo, either - I don't want a Google Home in the house either.
That lead to a discussion about why having a listening device in the home is bad. We expect a certain amount of privacy in our own home regardless of the fact that we are not doing anything against the law. I just do not want my private conversations overheard by a device that sends all of that back to a server, where it sits forever. Police have already tried to get Echo recordings for a murder, though if Amazon is to be believed unless someone said "Alexa, help me!" nothing should have been recorded. Even if it had recorded something, Amazon states that such voice recordings are encrypted.
Knowing how well software is built and how often "encrypted" data gets accessed means I do not want my words recorded and stored on Amazon's, or anyone else's, servers. Hell, I work for a company who designs and sells a network appliance to find bad traffic on networks. When someone has access to servers or the network, getting access to information is trivial. Amazon now also sells Echo Look, which is a camera that currently helps you dress fashionably. I do not even have to talk about how creeped out that would make me feel.
We grow increasingly reliant upon companies that make our lives more convenient. I've used Google's e-mail, calendaring, and document storage services for years because it was easy to use, worked directly with my phone, and meant I did not have to worry about e-mail. There are some nice perks to that, like online document editing, having airline data directly parsed and made available, intelligent spam filtering, and device syncing, all to name a few.
If I do not want my speech hosted on Amazon or Google servers... why my textual life hosted and sifted through by Google?
Taking Back My E-mail
The first thing I've decided to move off of Google, and back into my own control, is my e-mail.
I have a lot of e-mail addresses, and I have been attempting to consolidate them into just a few. Google made that pretty easy. I'm grandfathered into the old G Suite setup of it being free for 100 users, but I took liberal advantage of domain aliases and catchall e-mail addresses.
I looked at services like FastMail, ProtonMail, and Kolab Now. All three of them are highly regarded, with Kolab and ProtonMail being open source projects. Moving my domains and setting up aliases though, that would end up being very, very costly. Kolab charges around $50 for just setting up a single domain alias. FastMail and ProtonMail would start to get very pricy as I moved all my domains over.
ProtonMail also lost points as I would have to use a web browser on my desktop. I want my e-mail in any app of my chosing. I am not paranoid enough to think someone is trying to get into my e-mail, so the security aspect of ProtonMail was not a huge selling point.
I decided to host my own e-mail.
Running My Own Server
"Email is one of the bastions of the decentralised Internet and we should hang onto it" - Nux, Hacker News
I'm not afraid of servers or their maintenance at all. My career started with maintaining servers and dealing with configuring them, so why not just run my own e-mail server?
I know, I know. I should not run my own e-mail server because:
- There are lots of moving parts
- It's not just e-mail, its virus scanning, spam filtering, e-mail access
- Maintenance is time consuming
- Blacklist maintainers are cold, heartless beings that never remove IPs
- Russians will hack me
- E-mail isn't secure
- I have to trust my host
Frankly, most of the above is FUD. If we, as developers, are telling people to run things like Docker or set up their own VPS because "it's the right way to run a web app," then running an e-mail server should not be some scary thing. Granted, I am not going into this blind as I've set up an e-mail server before, but come on people. It isn't that bad.
I do want to cut down on the amount of work I have to do. I first looked at Mail-in-a-Box, which is a set of scripts that sets up a mail server. I decided against it as it is pretty much all or nothing. You run and set up the box the way it wants to be set up and that's it. Want to do something else with the box? Too bad.
I then found sovereign. It is a set of Ansible playbooks that set up a server that includes e-mail as one of the various services. Since it is just based on Ansible configuration and I know how to work with that, I decided on sovereign.
Setting up the Server
I use Digital Ocean for a lot of projects. As I said before, privacy from foreign powers is not a current concern I have so hosting a server in the US is fine for the moment. I created a VPS with Debian 8 as that was what sovereign recommended.
The next thing I did was check the assigned IP on http://multirbl.valli.org/. This site will check a bunch of well used DNS blacklists to see if the IP that Digital Ocean gave me has had a shady history. The first one... well, once it hit twenty blacklists I deleted the VM and rebuilt it on a different server.
The second one was only on four blacklists. That is a much more manageable number to deal with. Most blacklists are fairly easy to get removed from, and if I'm only on four I will take my chances.
With that sorted out I followed through the rest of the instructions in the sovereign README file. It took only a few minutes of prep before running the Ansible playbooks.
I started off with a domain that did not previously have e-mail associated with it, to test things out. That way if it all went to Hell I wouldn't lose any e-mail. Ran the scripts and after about 15 minutes ran out of memory on the server.
I tried to work around it, but with everything running 512mb was not big enough. I deleted the server and reprovisioned a bigger one. Not only did it have more memory, it also had more hard drive space.
That worked better. About 20 minutes later I had a server up and running!
Shutting down Services
sovereign comes with a bunch of services installed, and since this was my first run through I let it install everything. Once I confirmed everything was working well, I SSH'd into the server and disabled a bunch of stuff I did not need, like ZNC. I happily pay IRCCloud for IRC bouncing.
Most servers are compromised because of services running on the box. It is rare that an actual OS exploit is the problem. I removed the services I did not need from the
site.yml file, and shut down services I did not need.
I did want to keep the webmail so I just disabled a bunch of vhosts as well. So far so good.
sovereign actually makes it pretty simple to set up multiple domains on a single install.
group_vars/sovereign houses all of the domains and accounts you want to set up. Adding a second domain was a simple as adding a new entry under
mail_virtual_domains, and the associated accounts under
Another Ansible run, and my legit domains I wanted to move off of Google were all set up. I tested logging in via Evolution, the e-mail client that comes with GNOME and what I use on my desktop and laptop. Auto config did not work, but I manually set up IMAP+ with no issues. I could send e-mail to and from accounts without a problem.
That left me figuring out how to get catchall e-mail addresses to work. There was an open issue on the Github project, so I dug around a bit. sovereign uses a Postgresql-backed e-mail system for the users, so finding how to do catchall addresses was a bit of a pain. Turns out it is really hard and not well documented. This wasn't a problem with sovereign, but postfix itself.
I found instructions for how to do it at https://workaround.org/ispmail/wheezy/connecting-postfix-to-the-database. I created a new file at
roles/mailserver/templates/etc_postfix_pgsql-email2email.cf.j2 and modified the Ansible scripts to use it per the instructions on workaround.org.
Another Ansible deploy, and I tested it from my old Hotmail address.
I did not get my e-mails.
Checking the logs I was getting greylisting errors. Turns out Hotmail/Outlook.com get flagged quite regularly for spam, so my server was greylisting them. I added the following to
/etc/postgrey/whitelist_clients and restarted postgrey:
# Outlook.com 184.108.40.206/17 220.127.116.11/16 /.*outbound.protection.outlook.com$/ /outlook/
I sent another e-mail, and my catchall started working! Well, technically, it was working before, just my greylist service was slowing Outlook.com down.
Moving from Google
After all my testing, I was ready. I went into my DNS providers and added the needed DKIM, DMARC, and MX records to point to my new server. I waited about fifteen minutes, as the TTL on all the records was 900 seconds, and tried to send an e-mail. It showed up in my new inbox.
I actually started recieving legitimate e-mail as well. I noticed some, like e-mails from Twitter, were coming in about 2 hours later than their timestamp. Quick look at the logs showed I'm greylisting Twitter's servers as well. Everything was working though, as grey listing is a normal part of day-to-day e-mail. If I'm greylisting someone and it's important, there are many other ways to get in touch with me ASAP.
I have years worth of e-mail sitting in GMail though. I wanted to move all of that over.
After some searching I came across
imapsync, which is an open source tool that syncs mail from one IMAP server to another. I followed the directions at http://blog.jgrossi.com/2013/migrating-emails-using-imap-imapsync-tofrom-gmail-yahoo-etc/ on compiling and setting it up on my Ubuntu 17.04 desktop.
I then followed the directions at https://imapsync.lamiral.info/FAQ.d/FAQ.Gmail.txt for syncing from GMail to my local server. I settled on the following command to run:
imapsync \ --host1 imap.gmail.com \ --ssl1 \ --user1 email@example.com \ --password1 p@ssw0rd \ --authmech1 plain \ --host2 mail.newmailserver.com \ --ssl2 \ --user2 firstname.lastname@example.org \ --password2 n3wp@ssw0rd \ --useheader="X-Gmail-Received" \ --useheader "Message-Id" \ --automap \ --regextrans2 "s,\[Gmail\].,," \ --skipcrossduplicates \ --folderlast "[Gmail]/All Mail"
GMail has a 2.5GB limit on mail transfer per day, but I was below that limit. I fired up the command and was immediately shut down by Google. They consider PLAIN authentication an insecure way to authenticate (for good reason), but they provided a link and explanation. I followed the directions and ran the command again.
Nearly 48 hours to download all of the e-mail. It worked though. I started to see all of my folders and e-mail show up in my new server.
With that, I was off of Google's mail servers.
E-mail is not secure. It was never designed to be. Even running something like ProtonMail, which touts it's encryption, does nothing to encrypt e-mails once it leaves their servers. Anyone can sniff e-mail on the wire. That's the nature of e-mail.
What is a concern is authentication, and access to the box.
SSH access is locked down to key-based authentication. No users have passwords. sovereign also sets up fail2ban, which should stop any brute force attacks. I'll probably supplement that with ossec. I should be able to get that installed with a new Ansible role.
For any virtual hosts on the machine as well as IMAP, sovereign sets up Let's Encrypt for SSL certificates, as well as scripts to renew them when needed. sovereign sets up Roundcube for web mail, which is protected with this, and any new subdomains it activates will be protected as well (with the appropriate changes to Ansible).
E-mail access and sending require authentication. Most servers get blacklisted due to the lack of authentication on the sending portion. Authentication is set up by default with sovereign, and all of the authentication happens over SSL/TLS.
My only main job is to update the base OS and packages every so often. I think I'm pretty well set up other than that.
Step One Completed
It's been a few days now and so far so good. The only hard thing thus far was setting up the catchall addresses. I'm getting e-mail on my laptop, desktop, and phone without an issue. I've tested sending mail to different services and so far have not been blocked. The e-mail transfer from GMail to the new server has been taking a while, but it's pretty hands off once it starts.
I am not totally off of Google yet. Next step is to move all of my calenders, which I believe I can do with ownCloud. ownCloud is an open source file, calendar, and contacts, storage/sharing service that gets installed as part of sovereign. ownCloud should actually handle both moving my calendar from Google Calendar, but also my files from Google Drive.
I also have a few patches that I want to clean up and send to sovereign. One nice one is the catchall setup, but then I've also been working with the Ansible scripts a bit to make it smaller to run. By default it runs all the tasks, but for something like adding a single e-mail address that means a 15-20 minute run.
So far I've been impressed with sovereign. I'd highly suggest looking into it if you want to run your own server.