Self-Hosted Email services on OpenBSD


Looking at my notes , it seems I haven’t setup an email services server from scratch since 2015. Of course, mine have evolved following OpenBSD updates and upgrades.

Let’s benefits from the fact that I’m migrating from Vultr to OpenBSD Amsterdam to write a few notes about the mail server (re)creation. At the time of writing, OpenBSD is available in version 7.2.

“Setting up a mail server” means all and nothing. So to clarify, here’s what I want to be able to do:

I also want to manage adressbooks and calendars, but that’s another story ; and won’t be convered here.


In the ancient times, I did a few things quite differently than I will today. So don’t expect an updated version of the previous posts I already did. My configuration is now minimal and enough to deal with 5-10 users.


I don’t manage a lot of users. So I just create them by hand. Using an SSH connexion to the server, issue a simple classical command:

# useradd -m -c "John Doe" -g users -s /sbin/nologin -u 10000 jdoe

Then configure the initial user password.

# passwd jdoe

One way to allow users to change their own password is to create SSH keys for them, turn their shell to /bin/sh and let them run passwd.

Another way is to limit the commands they can run (read the SSH-BASED VIRTUAL PRIVATE NETWORKS section from ssh(1)) but allowing only passwd and configuring an SSH client that will run that command. But this is out of scope here.

All the mail services daemon use /etc/passwd to authenticate users.

Email domains and addresses

Big Tech and most companies offer authentication using email addresses. It is very user friendly. But I don’t like this. Because since you know the name of a person, you can guess their login. And that’s one less thing to guess for the attackers.

An OpenSMTPD table lists all email domains the server should accept for local delivery.

An OpenSMTPD table lists all email addresses that are legit. It also references every email aliases that I want / need to manage. All those email addresses point to one of the several local users ; those created previously. Read tables(5) for more information.

# vi /etc/mail/vdomains

# vi /etc/mail/vusers
abuse©tumfatig•net      root©tumfatig•net
postmaster©tumfatig•net root©tumfatig•net
webmaster©tumfatig•net  root©tumfatig•net
root©tumfatig•net       jdoe
john•doe©carnat•net     jdoe
jane•smith©carnat•net   jsmith

Note that email and DNS format is not correct. Read and learn, do not simply copy/paste ;)

TLS certificate

To secure the connexions to the server and provide a bit of privacy, a TLS certificate has to be obtained and used by the email daemons. I’ll be using Let’s Encrypt certificates. Read acme-client(1) for more detailed informations.

# cp /etc/examples/httpd.conf /etc/
# vim /etc/httpd.conf
# rcctl enable httpd
# rcctl start httpd

# cp /etc/examples/acme-client.conf /etc/
# vim /etc/acme-client.conf
# acme-client -v mail•carnat•net
acme-client: /etc/ssl/mail•carnat•net•crt: created

Receive and send emails


Receiving email is dependant on rspamd, for spam checking. And sending email is depending on rspamd for DKIM signing. So first thing to do is install and configure rspamd.

Install the required packages. Note that redis is not strictly required but it will smooth rspamd reactions.

# pkg_add rspamd--hyperscan opensmtpd-filter-rspamd redis

# rcctl enable redis
# rcctl start redis

Configure rspamd to actually use redis.

# vi /etc/rspamd/local.d/redis.conf
servers = "";

# vi /etc/rspamd/local.d/greylist.conf
servers = "";

Configure rspamd SPF feature.

# vi /etc/rspamd/local.d/spf.conf
spf_cache_size = 1k;
spf_cache_expire = 1d;
max_dns_nesting = 10;
max_dns_requests = 30;
min_cache_ttl = 5m;

Configure rspamd DKIM feature.

# vi /etc/rspamd/local.d/dkim_signing.conf
domain {
  carnat•net {
    path = "/etc/mail/dkim-carnat•net•key";
    selector = "default";

The documentation for DKIM key generation is available here .

With all this done, rspamd can be started and used.

# rcctl enable rspamd
# rcctl start rspamd


OpenSMTPD uses rspamd and senderscore filters.

# pkg_add opensmtpd-filter-senderscore

The hostname of the server is not the one the SMTP server is known as. So let’s tell OpenSMTPD to use the proper (HELO) name.

# vi /etc/mail/mailname

Now configure OpenSMTPD.

# vim /etc/mail/smtpd.conf
# TLS certificate ------------------------------------------------------
pki mail cert "/etc/ssl/mail•carnat•net•crt"
pki mail key "/etc/ssl/private/mail•carnat•net•key"

# Incoming mail filters ------------------------------------------------
filter "rdns" phase connect match !rdns disconnect "550 Nope."
filter "fcrdns" phase connect match !fcrdns disconnect "550 Nope."
filter "senderscore" proc-exec \
  "/usr/local/libexec/smtpd/filter-senderscore -blockBelow 10 -junkBelow 70 -slowFactor 1000"
filter "rspamd" proc-exec "filter-rspamd"

# Domain and email tables ----------------------------------------------
table vdomains file:/etc/mail/vdomains
table vusers   file:/etc/mail/vusers

# Listeners ------------------------------------------------------------
listen on socket
listen on all tls pki mail filter { "rdns", "fcrdns", "senderscore", "rspamd" }
listen on all port submission tls-require pki mail auth mask-src

# Actions to process ---------------------------------------------------
action "local_mail" maildir junk virtual <vusers>
action "outbound"   relay helo mail•carnat•net

match from any       for domain <vdomains>   action "local_mail"
match from any auth  for any                 action "outbound"
match from local     for local               action "local_mail"
match from local     for any                 action "outbound"

I thought DNS filtering was overkill. But looking at my logs, there are about 200 hits per day for my current rdns and fcrdns filters. So let’s keep them.

Check configuration and start the daemon.

# smtpd -dvn                                 
configuration OK
# rcctl restart smtpd

Emails will now be delivered to the ~/Maildir directory of the configured users. If rspamd is not up when a mail is received, OpenSMTPD will not accept the mail and issue a “451 Try again later” error message. The email will have an opportunity to be delivered later on. As no DNS is properly configured yet, expect to receive only SPAM. Also, email clients configured to use this server right now - to send emails - shall not pass DKIM/DMARC tests.

Read and manage email

Accessing emails with a decent mail client will be provided by Dovecot.

# pkg_add dovecot

# vi /etc/dovecot/conf.d/10-mail.conf
mail_location = maildir:~/Maildir

# vi /etc/dovecot/conf.d/10-master.conf
 service imap-login {
   inet_listener imap {
    port = 0

# vi /etc/dovecot/conf.d/10-ssl.conf
ssl = yes
ssl_cert = </etc/ssl/mail•carnat•net.crt
ssl_key = </etc/ssl/private/mail•carnat•net.key

# vi /etc/dovecot/conf.d/20-imap.conf
mail_plugins = $mail_plugins imap_sieve zlib
mail_max_userip_connections = 50

# vi /etc/dovecot/conf.d/90-plugin.conf
plugin {
  zlib_save = zstd

Using the rspamd configuration described above, some email may be identified as SPAM and moved to the Junk folder, automatically. But what if there are false-positives. And what if rspamd does not recognise all the SPAM I receive and let some land to my INBOX. Well a way to train rspamd is by using sieve scripts that will be targetted when email clients move emails to / from the Junk mailbox. Moving a mail from INBOX to Junk (or tagging an email as SPAM) teaches rspamd that this is a SPAM. Moving a mail out of Junk teaches rspamd that this mail wasn’t a SPAM.

# pkg_add dovecot-pigeonhole

# vi /etc/dovecot/conf.d/90-plugin.conf
 plugin {
  zlib_save = zstd

 sieve_plugins = sieve_imapsieve sieve_extprograms
 sieve_global_extensions = +vnd.dovecot.pipe +vnd.dovecot.environment

 # Spam: move from anywhere to Junk or flag changed in Junk
 imapsieve_mailbox1_name = Junk
 imapsieve_mailbox1_causes = COPY APPEND FLAG
 imapsieve_mailbox1_before = file:/usr/local/lib/dovecot/sieve/report-spam.sieve

 # Ham: move from Junk to anywhere
 imapsieve_mailbox2_name = *
 imapsieve_mailbox2_from = Junk
 imapsieve_mailbox2_causes = COPY
 imapsieve_mailbox2_before = file:/usr/local/lib/dovecot/sieve/report-ham.sieve

 sieve_pipe_bin_dir = /usr/local/lib/dovecot/sieve

# vi /usr/local/lib/dovecot/sieve/report-spam.sieve
require ["vnd.dovecot.pipe", "copy", "imapsieve", "environment", "variables"];
if environment :matches "imap.user" "*" { set "username" "${1}"; }
pipe :copy "" [ "${username}" ];

# vi /usr/local/lib/dovecot/sieve/report-ham.sieve
require ["vnd.dovecot.pipe", "copy", "imapsieve", "environment", "variables"];
if environment :matches "imap.user" "*" { set "username" "${1}"; }
pipe :copy "" [ "${username}" ];

# vi /usr/local/lib/dovecot/sieve/
exec /usr/local/bin/rspamc learn_spam

# vi /usr/local/lib/dovecot/sieve/
exec /usr/local/bin/rspamc learn_ham

# cd /usr/local/lib/dovecot/sieve/
# chmod 0755 *sh
# sievec report-ham.sieve
# sievec report-spam.sieve

The previous magic spells tell Dovecot to run some sieve scripts when certain mail movement condition are met ; like mark as SPAM or move out of Junk folder.

The sieve scripts end piping the emails to some shell scripts.

The shell scripts call the rspamd commands required to learn a new SPAM or a new HAM message.

It is then time to use the complete Dovecot configuration.

# rcctl enable dovecot
# rcctl start dovecot

The email clients can now be pointing at the IMAPS port of the server.

It’s not DNS

Sometimes, it is DNS. To be able to receive email from the Wild Wild World, the mail domains will require proper MX records. And to be able to send email to the World lowering the chances of being rejected as SPAM, the mail domains need a couple of DNS records.

Setup the server’s DNS records:

Tell the world that the server can receive emails for your domain(s):

If the server’s name is resolvable using IPv4 and IPv6, then you shall receive email from servers using both stack.

Tell the world that the server is legit to send emails for your domain(s) ; and that others aren’t.

You can have a look at what others do. At least to see if you understood the docs properly and planned TXT records that seem classical.

# dig +short txt
"v=spf1 mx ~all"

# dig +short txt 

# dig +short txt        

# dig +short txt   
"v=DMARC1; p=none; sp=quarantine;"

# dig +short TXT
"v=spf1 ~all"

Once done, you may use tools such as DMARCLY to verify that your DNS records are configured properly.

One more thing

rspamd has a nice reporting web GUI. By default, it only listens on localhost and access is granted unauthenticated. Steps are required to set a password. One way to access the Web page is to use SSH port forwarding feature.

# ssh -L 11334:localhost:11334 <mail server>

Browse to http://localhost:11334/ and enjoy.

There probably are ways to grab those metrics and send them to InfluxDB so that they can be rendered using Grafana. So is the case of smtpd logs. But that’s another story.

Happy mailing!