About mail servers#

Simply put, a mail server sends and receives email. When ruby@protonmail.com emails klaasje@gmail.com, Protonmail’s mail server’s MTA sends the email to Google’s mail server’s MTA. Google’s MTA then passes the email on to Google’s MDA, which is responsible for storing the email. MDAs also run IMAP or POP3 servers so you can read your emails and send emails with an email client. IMAP and POP3 are protocols for email retrieval over TCP/IP. For sending email, you would use SMTP.

Or visualise it like this:

Ruby's email client via SMTP -> Protonmail's MTA -> Google's MTA -> Google's MDA -> Klaasje's email client via IMAP

A mail server is software which can be run on any computer, including yours. You can host a mail server on a server you already own that is hosting other services, so long as those other services aren’t using any of the mail ports.

Why run my own mail server?#

I’ll cut to the chase: the main reason why you’d want to run your own mail server is for related reasons of privacy and digital sovereignty. For privacy benefits, as much as you have control over your server, you can protect your email from the eyes of prying server admins (given that you yourself are the admin). Even email providers that market themselves around privacy (e.g. Protonmail) rely on trust that the provider is not reading your unencrypted incoming email. This is not an issue exclusive to any particular mail provider; if information arrives unencrypted at a server, those with access to the server (i.e. administrators) can read that information, simple as. And as nice as it would be if everyone used GPG end-to-end encryption for email, the vast majority of emails people receive are not end-to-end encrypted, and entirely legible to the mail servers involved. If you’re not exchanging E2EE email, you can’t mitigate the fact that your exchange is entirely legible to the mail server of the person you’re corresponding with, but you can at least eliminate your anxieties about the mail server you yourself are using.1

Running your own mail server also allows you to implement things your way, with the features you want. For instance, you can run a sieve filter for encrypting all incoming mail with a user’s public GPG key; for obvious reasons, public (as in, open to public sign-up) mail servers that implement sieve do not allow users to create their own executables for sieve filters.

Hosting your own mail server is not something I would universally recommend to people. While I’m very much against “nothing to hide, nothing to fear”, a combination of that factor alongside a low state threat model (i.e. there is little state interest in you, domestic or foreign), a lack of relevant knowledge, and a lack of interest in managing your own server/learning how to, likely make self-hosting email not a reasonable privacy suggestion.

When you host your own mail server, you are responsible for securing the server. If you entrust Google with your email, you can at least know that your email is secure, though not private; Google will hire people with the relevant knowledge and skills to secure a mail server. If you are not confident in your ability to do this and not interested in learning, you may want to find another solution.

There are also existing solutions that aim to “simplify” the process of managing a mail server; there have always been plenty of services offering mail server hosting which is managed for you, the customer, and there are also projects like Mailcow, which is an attempt to make a user-friendly distributable mail server stack based on Docker. It has a web GUI you can use to configure your mail server. This may be an option for people who value digital sovereignty but don’t want to get their hands dirty with config files. I only mention this as something you may want to look into; I have never used Mailcow and cannot recommend it personally.

Why this tutorial?#

There are many tutorials on the internet about how to set up a mail server. I don’t claim that mine is particularly better than anyone else’s; I’m mostly writing this for my own reference so that I can recreate my own setup on new machines.

There are existing mail server tutorials for Alpine Linux, including ones for Postfix and Dovecot. This tutorial aims to be “monolithic” (i.e. covers the entire mail server setup with all its components), making it easier to follow without trying to follow different tutorials that may not be 100% compatible with each other, and to be adapted to my own specific use-cases, which may not be yours.

I am hugely indebted to LinuxBabe’s mail server tutorial for Ubuntu, which is actually what I followed to set up my own server. Essentially, this tutorial could be thought of as an Alpine Linux adaptation of LinuxBabe’s tutorial. If you want to set up an Ubuntu or Debian mail server, I recommend following LinuxBabe’s tutorial, which is written very clearly and is easy to follow.

The mail server will be composed of the following software:

Component Software
Mail transfer agent Postfix
Mail delivery agent Dovecot
SPF authentication postfix-policyd-spf-perl
DKIM authentication and signing OpenDKIM
DMARC authentication OpenDMARC
Spam filter Amavis SpamAssassin
Antivirus ClamAV
MDA server-side filtering Pigeonhole

Postfix will be our SMTP server, and Dovecot will run an IMAP server for us.

Postfix and Dovecot are required for the minimum of what you’d expect from a working mail server (i.e. a user can log in through a standard SMTP/IMAP/POP3 email client, read their emails, and send emails). The rest is optional and modular, i.e. you can opt to have e.g. Pigeonhole but not Amavis.

We will end up with a small-scale mail server running on Alpine Linux with one domain, and we will use Unix user accounts as mail accounts. We will only set up IMAP, not POP3.

This tutorial was written for Alpine Linux 3.20.3, though will most likely work on other versions too.

This tutorial assumes no prior knowledge about mail servers.

You will need#

A server#

Any computer will do. This tutorial is for Alpine Linux specifically.

The relevant thing to consider for your server is that port 25 (the port for sending email) is not blocked. Most VPS hosts block port 25 because spammers commonly use VPSes to send spam, so you will need to find one that doesn’t block port 25. Some VPS hosts block port 25 by default, but will unblock it upon request, and re-block it if they find you are spamming.

A domain name#

Self-explanatory. You need a domain name and the ability to set its DNS records.

Before you start#

Set up your DNS records#

Firstly, pick a domain for your mail server. If you’re sending emails from domain.com, mail.domain.com is a common choice.

MX record#

An MX record denotes that your domain is used to send and receive email, and tells other MTAs the domain name of your mail server. We will use mail.domain.com for your MX record. For instance, my MX record looks like:

revsuine.xyz.                    14400 IN MX    0 mail.revsuine.xyz

Mail server records (A, AAAA, and/or CNAME)#

Now you need to set a record stating the IP address of your mail server (mail.domain.com above). Depending on your setup, you may want to create a CNAME record pointing to domain.com if the IP address of mail.domain.com is the same as the IP address of domain.com, or an A record if the IP address is not shared with another domain.

I use a CNAME record because the IP addresses of mail.revsuine.xyz and revsuine.xyz are the same, so my record is:

mail.revsuine.xyz.               14400 IN CNAME revsuine.xyz

If you use an A record, your record may look something like

mail.domain.com.                    14400 IN A     ip.address.here

If you use IPv6, you should also add an AAAA record, e.g.:

mail.domain.com.                    14400 IN AAAA  ip:address:here::

A note on my DNS records#

I use one server with one IP address for hosting several services under one apex domain. For instance, my static website is hosted at revsuine.xyz; my Nextcloud is hosted at cloud.revsuine.xyz; and my mail server is hosted at mail.revsuine.xyz. To handle this setup, I’ve created a subdomain master.revsuine.xyz (you can call the subdomain anything you want) with an A record (because your PTR record is expected to be an A record, not a CNAME record) to my server’s IP address.

I set my server’s hostname and PTR record to master.revsuine.xyz. I will also refer to this master.revsuine.xyz subdomain further down when configuring Postfix.

PTR record#

A PTR record is used for “reverse DNS”, or rDNS, lookup: instead of mapping a domain to an IP address, it maps an IP address to a domain. It is not managed through your DNS manager (e.g. whomever you bought your domain name from), but through whoever gives you your IP address. If you rent a server, be it a VPS or a dedicated server, your hosting provider whom you rent from will be able to manage PTR records. You may have an option to add a PTR record in the dashboard of your hosting provider, or you may have to contact their support to get a PTR record added.

For the above stated reasons, my PTR record for my single server (which hosts all my services) is master.revsuine.xyz.

If your hostname is mail.domain.com, your PTR record should be mail.domain.com.

Unblock your ports#

After unblocking ports from your internet provider (e.g. if your VPS host blocks outgoing port 25), make sure the following TCP ports are open on your firewall:

Port Usage
25 SMTP
143 IMAP
465 Email message submission over TLS
587 Email message submission
993 IMAPS (IMAP over TLS)
4190 ManageSieve

Obtain a TLS certificate#

To enable TLS encryption, you need a certificate. Let’s Encrypt provides free TLS certificates. To get a certificate from them, you can use certbot:

# apk add certbot

We will need a web server to use certbot. I’m going to use nginx for this guide, because nginx is what I use on my server, but the certbot website has instructions for a variety of setups. If you don’t already have an nginx server, install nginx and set it up now.

Install certbot-nginx with:

# apk add certbot-nginx

Add the following to your nginx config (for instance, inside http {} in /etc/nginx/nginx.conf, or in a dedicated virtual host file /etc/nginx/http.d/mail.domain.com.conf):

server {
    listen 80;
    listen [::]:80;
    server_name mail.domain.com;

    root /usr/share/nginx/html/;

    location ~ /.well-known/acme-challenge {
        allow all;
    }
}

Replace mail.domain.com with the FQDN of your mail server.

The root can be set to any extant directory on your system that you’re happy to publish to the web. You can just make an empty directory at /usr/share/nginx/html, or make this the directory of your website, etc.

Reload or restart nginx for the changes to take effect:

# rc-service nginx reload

Now run the following command to get your free TLS certificate:

# certbot certonly -a nginx --staple-ocsp --email your@email.here -d mail.domain.com

If you have several subdomains in your nginx config that you’d like covered by the same certificate, you can omit -d mail.domain.com and get a certificate covering all the domains in your nginx config. On my server, I have one certificate at /etc/letsencrypt/live/revsuine.xyz/ covering my apex domain and all subdomains. If you go for a certificate with only one domain name, e.g. for mail.domain.com, it will be at /etc/letsencrypt/live/mail.domain.com/.

Postfix#

Postfix is a mail transport agent (aka SMTP server). In its own words:

Postfix attempts to be fast, easy to administer, and secure. The outside has a definite Sendmail-ish flavor, but the inside is completely different.

Installing Postfix#

On your server, install Postfix with:

# apk add postfix

You likely also want to have Postfix documentation:

# apk add postfix-doc

Verify that Postfix is installed by checking its version:

$ postconf mail_version

Configuring Postfix#

Edit /etc/postfix/main.cf.

You should set myhostname to the hostname of your server; in my case, this is master.revsuine.xyz.

Now set mydomain to the domain you intend to send email from. For instance, my email addresses are name@revsuine.xyz, so mydomain is set to revsuine.xyz.

myorigin determines the domain name in the From: field of locally sent emails. So you could for instance set this to revsuine.xyz.

Set mydestination to the following:

mydestination = $myhostname, localhost.$mydomain, localhost, $mydomain

mydestination states the list of domains your machine will consider itself the destination for, e.g. if mydomain is set to revsuine.xyz then any emails sent to username@revsuine.xyz will be sent to my server according to the above configuration.

maillog_file denotes where Postfix’s log file is. By default this is /var/log/messages; you may want to configure Postfix to have a dedicated log file like /var/log/postfix.log.

You probably want to have logrotate rotate your Postfix log. If there isn’t already such a file, you want to create one at /etc/logrotate.d/postfix:

/var/log/postfix*.log
/var/log/mail*.log
{
    daily
    missingok
    notifempty
    rotate 7
}

Add the following TLS settings, replacing your.domain.com with your mail server’s FQDN, or otherwise where the TLS certificate we generated would be:

# Enable TLS encryption when Postfix receives incoming emails
smtpd_tls_cert_file = /etc/letsencrypt/live/your.domain.com/fullchain.pem
smtpd_tls_key_file = /etc/letsencrypt/live/your.domain.com/privkey.pem
smtpd_tls_security_level = may
smtpd_tls_loglevel = 1
smtpd_tls_session_cache_database = lmdb:${data_directory}/smtpd_scache

# Enable TLS encryption when Postfix sends outgoing emails
smtp_tls_security_level = may
smtp_tls_loglevel = 1
smtp_tls_session_cache_database = lmdb:${data_directory}/smtp_scache

# Enforce TLSv1.3 or TLSv1.2
smtpd_tls_mandatory_protocols = !SSLv2, !SSLv3, !TLSv1, !TLSv1.1
smtpd_tls_protocols = !SSLv2, !SSLv3, !TLSv1, !TLSv1.1
smtp_tls_mandatory_protocols = !SSLv2, !SSLv3, !TLSv1, !TLSv1.1
smtp_tls_protocols = !SSLv2, !SSLv3, !TLSv1, !TLSv1.1

# only offer authentication after STARTTLS
smtpd_tls_auth_only = yes

# disable SSL compression
tls_ssl_options = NO_COMPRESSION

# Configure the allowed cipher list
smtpd_tls_mandatory_ciphers = high
smtp_tls_mandatory_ciphers = high
smtpd_tls_ciphers = high
smtpd_tls_mandatory_ciphers = high
tls_high_cipherlist = ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256
tls_preempt_cipherlist = yes

The allowed cipher list is from Mailcow.

If you’re using this as a personal mail server, you may not want to have a mailbox size limit, so you can set:

mailbox_size_limit = 0

By default, mailbox_size_limit is 51200000. This number is in bytes. You can similarly set a message_size_limit.

Let’s hide some sensitive info that tends to get attached to email headers by clients. Edit /etc/postfix/header_checks, or create it if it doesn’t already exist, and add the following lines:

/^Received: .*/            IGNORE
/^User-Agent: .*/          IGNORE
/^X-Originating-IP: .*/    IGNORE
/^X-Mailer: .*/            IGNORE
/^Mime-Version: .*/        IGNORE

Now add the following to your /etc/postfix/main.cf:

header_checks = regexp:/etc/postfix/header_checks
smtp_header_checks = $header_checks
mime_header_checks = $header_checks

Finally, here are some various hardening settings you can add to your /etc/postfix/main.conf:

# connections rate limit: no of connections allowed per unit
# `postconf anvil_rate_time_unit` will give the time unit; by default it's
# 60 seconds, so 600/60=10 connections allowed per second
smtpd_client_connection_rate_limit = 600
# messages rate limit, again over same time limit
smtpd_client_message_rate_limit = 60
# VRFY command used to check if an email address exists
# not needed and can be used to find spam recipients
disable_vrfy_command = yes
# servers that don't use HELO or EHLO are either not properly configured
# or sending spam usually
smtpd_helo_required = yes
smtpd_delay_reject = yes
smtpd_helo_restrictions =
    permit_mynetworks,
    reject_invalid_helo_hostname,
    reject_unknown_helo_hostname,
    permit

Send your first email#

Have the postfix service auto-start upon boot, and start it during this session:

# rc-update add postfix default
# rc-service postfix start

You can now send an email with the following command:

$ echo "test email" | sendmail user@externaldomain.com

Send this email to your email account with an external server, e.g. a Gmail account. Note that Protonmail has quite stringent spam filters and this likely would be rejected by Protonmail, i.e. not even reach your spam folder.

Configure email aliases#

You can configure aliases for your mail server. Edit the /etc/postfix/aliases2 file.

You shouldn’t receive mail as root, so configure root to have an alias to your user, e.g.

root: revsuine

You also must have a MAILER-DAEMON and postmaster alias present:

MAILER-DAEMON: postmaster
postmaster:    root

Note how you can have referential aliases; mail to postmaster is aliased to root, which is aliased to revsuine, so ultimately revsuine will get postmaster’s mail.

You can continue to populate the aliases file with whatever aliases you want.

Enable Postfix submission and smtps service#

To send emails from email clients, you’ll need to enable Postfix’s submission service so that Postfix can receive emails to send via SMTP. Edit /etc/postfix/master.cf and ensure that the following lines are present:

submission inet n       -       n       -       -       smtpd
  -o syslog_name=postfix/submission
  -o smtpd_tls_security_level=encrypt
  -o smtpd_sasl_auth_enable=yes
  -o smtpd_relay_restrictions=permit_sasl_authenticated,reject
  -o smtpd_recipient_restrictions=permit_mynetworks,permit_sasl_authenticated,reject
  -o smtp_sasl_type=dovecot
  -o smtpd_sasl_path=private/auth

smtps     inet  n       -       n       -       -       smtpd
  -o syslog_name=postfix/smtps
  -o smtpd_tls_wrappermode=yes
  -o smtpd_sasl_auth_enable=yes
  -o smtpd_relay_restrictions=permit_sasl_authenticated,reject
  -o smtpd_recipient_restrictions=permit_mynetworks,permit_sasl_authenticated,reject
  -o smtpd_sasl_type=dovecot
  -o smtpd_sasl_path=private/auth

You may find them commented out, or partially present without some options, so make sure they are uncommented and complete as per above.

Restart Postfix.

# rc-service postfix restart

Dovecot#

Dovecot is a popular IMAP and POP3 server which we’ll be using for our MDA. Let’s install it:

# apk add dovecot

Check the Dovecot version with:

$ dovecot --version

Now let’s enable IMAP by editing /etc/dovecot/dovecot.conf. Find a protocols = line, or add one, and set it to:

protocols = imap

Logging#

By default, Dovecot logs to the syslog (/var/log/messages). It would be easier to monitor Dovecot if it logged to its own log file, so we’ll configure that at /etc/dovecot/conf.d/10-logging.conf:

log_path = /var/log/dovecot.log

The file also has various settings on logging verbosity you can configure. I set:

auth_verbose = yes
auth_debug = yes
mail_debug = yes

You should already have an /etc/logrotate.d/dovecot file. If not, create it with the following contents:

/var/log/dovecot*.log {
    daily
    missingok
    copytruncate
    rotate 7
    compress
    notifempty
    sharedscripts
    postrotate
        /etc/init.d/dovecot --quiet --ifstarted reopen
    endscript
}

Configure IMAP and email storage#

You probably want to use the Maildir format for storing emails, where each user’s mail is stored at ~/Maildir (this can be set to another location if desired).

In /etc/dovecot/conf.d/10-mail.conf, set:

mail_location = maildir:~/Maildir
mail_privileged_group = mail

mail_privileged_group tells us which group of Unix users can send mail; in this case, it’s anyone in the mail group. You can create the group with:

# addgroup mail
# adduser postfix mail
# adduser dovecot mail

We want to ensure that postfix and dovecot users have the right to access mail.

To change the Maildir directory, e.g. to set it to ~/mail, you would set the following:

/etc/dovecot/conf.d/10-mail.conf:

mail_location = maildir:~/mail

/etc/postfix/main.cf:

home_mailbox = mail/

Get emails with LMTP#

LMTP is a protocol which can be used for Postfix to pass incoming emails to Dovecot. To install it for Dovecot:

# apk add dovecot-lmtpd

Add lmtp to the supported protocols in /etc/dovecot/dovecot.conf:

protocols = imap lmtp

Now change the LMTP service (or add if it isn’t already there) in /etc/dovecot/conf.d/10-master.conf to:

service lmtp {
  unix_listener /var/spool/postfix/private/dovecot-lmtp {
    mode = 0600
    user = postfix
    group = postfix
  }

Postfix needs to be configured to use this socket. Edit /etc/postfix/main.cf with the following lines:

mailbox_transport = lmtp:unix:private/dovecot-lmtp
smtputf8_enable = no

Configuring authentication#

Edit /etc/dovecot/conf.d/10-auth.conf and uncomment the following line:

disable_plaintext_auth = yes

This disables plaintext authentication unless SSL/TLS is used.

In the same file, configure auth_username_format. As the variable name suggests, this denotes the format the server expects usernames in for authentication. Setting it to %n removes the domain, so to sign in to user@domain.com you’d enter your username as user. For this setup, you should set auth_username_format to %n, because we are using Unix user accounts for email accounts; Dovecot wouldn’t be able to find user@domain.com because the mailbox user is just user.

In the same file again, auth_mechanisms is a space-separated list of authentication mechanisms your server uses. Set this to

auth_mechanisms = plain login

login is mostly to support older email clients, but is optional.

Edit /etc/dovecot/conf.d/10-master.conf, and change service auth to the following:

service auth {
    unix_listener /var/spool/postfix/private/auth {
        mode = 0660
        user = postfix
        group = postfix
    }
}

Now we need to set the actual mechanism through which the server knows what password to expect, and what users exist. We will use a file at /etc/dovecot/passwd to manage this. Edit /etc/dovecot/conf.d/auth-passwdfile.conf.ext to be the following:

passdb {
    driver = passwd-file
    args = scheme=argon2id username_format=%n /etc/dovecot/passwd
}

userdb {
    driver = passwd
}

See this documentation to decide on a password scheme to use. I picked argon2id as the most secure option, however also the most expensive option, so it may be a poor option if you have many users. Dovecot recommends that, if using ARGON2ID, you set vsz_limit = 2G for the auth service. To do that with our setup, edit /etc/dovecot/conf.d/10-master.conf and add the line

vsz_limit = 2G

to the service auth {} section.

Now we want to include this file. Edit /etc/dovecot/10-auth.conf and ensure it includes this line, and that other !include lines for other auth-*.conf.ext files are commented out:

!include auth-passwdfile.conf.ext

Creating a user#

Now let’s create a user. We will be using a Unix user account, so create one with adduser if you don’t already have one. Their Unix username will be their email username, and what appears before the @ in their email address. They should also be in the mail group. So for instance:

# adduser revsuine
# adduser revsuine mail

We will not, however, be using their Unix password for authentication. Instead, we’ll set them a password dedicated for their email account. Create or edit the /etc/dovecot/passwd text file. Have a look at the following example:

revsuine:{ARGON2ID}$argon2id$v=19$m=65536,t=3,p=1$H1oyL7UdwUWiBuZGnyXorQ$3aW/cfyNdrjoHw3OK7HlOzwgKqdg61prln8QMtWJijg
muffalo:{ARGON2ID}$argon2id$v=19$m=65536,t=3,p=1$KLnLOiqlhbhOPLhmTUqllA$Raki8Rw/+eOgJzDSEXxtw0mqI+aYLyFf+gpi+MQTdfo
thrumbo:{ARGON2ID}$argon2id$v=19$m=65536,t=3,p=1$8R4rVQ2hlKmiZ0Zmyzwg0g$tkesrePjJfLAEKu1wyJY1tu6V+fR5+C6/etyKq6WJlQ

This lists three users, revsuine, muffalo, and thrumbo (those are their usernames). These will all be system users in the mail group. After the colon is the ARGON2ID hash of their password.

If you’re storing passwords with ARGON2ID, to get what goes after the colon in your passwd file, run this command:

# doveadm pw -s argon2id

You will be prompted to enter a password, and then it will output exactly what to put after the colon, such as {ARGON2ID}$argon2id$v=19$m=65536,t=3,p=1$H1oyL7UdwUWiBuZGnyXorQ$3aW/cfyNdrjoHw3OK7HlOzwgKqdg61prln8QMtWJijg. Each user is their own line of the file.

Set up your /etc/dovecot/passwd file accordingly, making sure each user listed is also a system user, because we are storing their mail in their home directories.

Use your TLS certificate#

Edit /etc/dovecot/conf.d/10-ssl.conf and set the following options (some should already be present, so you should just change their values):

ssl = required
ssl_cert = </etc/letsencrypt/live/mail.domain.com/fullchain.pem
ssl_key = </etc/letsencrypt/live/mail.domain.com/privkey.pem
ssl_prefer_server_ciphers = yes
ssl_min_protocol = TLSv1.2
ssl_cipher_list = ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-CHACHA20-POLY1305:ECDHE-RSA-CHACHA20-POLY1305:DHE-RSA-AES128-GCM-SHA256:DHE-RSA-AES256-GCM-SHA384:DHE-RSA-CHACHA20-POLY1305

Replace mail.domain.com with the domain you obtained your TLS certificate for. You can check

$ ls /etc/letsencrypt/live

if you’re not sure.

The cipher list is from Mailcow.

OpenSSL 3.x has a FIPS provider which is incompatible with Dovecot, so it should be disabled. Edit /etc/ssl/openssl.conf and make sure the following line is commented out:

# providers = provider_sect

Configure “special” mailboxes#

The file /etc/dovecot/conf.d/15-mailboxes.conf denotes “special” mailboxes, such as your spam folder. You can configure these, including setting these to auto-create, in /etc/dovecot/conf.d/15-mailboxes.conf. I have mine set to:

# NOTE: Assumes "namespace inbox" has been defined in 10-mail.conf.
namespace inbox {
  # These mailboxes are widely used and could perhaps be created automatically:
  mailbox Drafts {
    special_use = \Drafts
    auto = create
  }
  mailbox Junk {
    special_use = \Junk
  }
  mailbox Spam {
    special_use = \Junk
    auto = create
  }
  mailbox Trash {
    special_use = \Trash
    auto = create
  }

  # For \Sent mailboxes there are two widely used names. We'll mark both of
  # them as \Sent. User typically deletes one of them if duplicates are created.
  mailbox Sent {
    special_use = \Sent
  }
  mailbox "Sent Messages" {
    special_use = \Sent
    auto = create
  }

  mailbox Drafts {
    special_use = \Drafts
    auto = create
  }
}

You can set the auto = create option for any mailbox you want to be auto-created for users.

Now enable and start the Dovecot service. Keep an eye on terminal output, as if there’s a configuration error, you’ll get an error message when restarting.

# rc-update add dovecot default
# rc-service dovecot start

mail_crypt#

Dovecot has a mail_crypt plugin which implements transparent encryption at rest for mail. By transparent, I mean “invisible” to email clients; you can use an IMAP client with your server with no changes, and no difference in user experience. Mail is decrypted on the server and sent over IMAP.

We will optionally set up global key mail_crypt encryption. This does not provide protection against an attacker with root access, or full disk access (which is basically root access), however it can protect against other processes reading our mail since they can’t read the private key.

If you want to implement it, declare usage of the mail_crypt plugin in /etc/dovecot/dovecot.conf:

mail_plugins = $mail_plugins mail_crypt

Now let’s generate some elliptic curve keys for this.

See what curves are available:

$ openssl ecparam -list_curves

If we pick prime256v1 as our curve, then run:

$ openssl ecparam -name prime256v1 -genkey | openssl pkey -out ecprivkey.pem

to generate the private key. To generate the public key:

$ openssl pkey -in ecprivkey.pem -pubout -out ecpubkey.pem

Now move these keys to /etc/dovecot/ and make sure they are owned by dovecot:

# mv ecpubkey.pem /etc/dovecot
# mv ecprivkey.pem /etc/dovecot
# chown dovecot:dovecot ecpubkey.pem ecprivkey.pem

Give them the correct permissions:

$ cd /etc/dovecot
# chmod 644 ecpubkey.pem
# chmod 600 ecprivkey.pem

Anyway, create and edit /etc/dovecot/conf.d/90-mail_crypt.conf and configure the plugin as follows:

plugin {
    mail_crypt_global_private_key = </etc/dovecot/ecprivkey.pem
    mail_crypt_global_public_key = </etc/dovecot/ecpubkey.pem
    mail_crypt_save_version = 2
}

Restart Dovecot for the changes to take effect:

# rc-service dovecot restart

Note that this will not make any difference to reading unencrypted email that was previously stored in your mailbox.

Use a local email client#

You are now ready to try logging in on a local email client such as Thunderbird, Evolution, Geary, KMail, etc.

To log in:

Enter your display name, email address (this is [Unix user]@[mydomain variable in Postfix config]), and the password you set for yourself in /etc/dovecot/passwd. Set the following settings:

Incoming Outgoing
Server type IMAP SMTP
Server mail.domain.com, or whatever your MX record is set to
Port 993 465
Username Your system username; this should be the contents of your email address before the @
Encryption SSL/TLS
Authentication Password PLAIN (shows up as Normal password in Thunderbird)

With our setup, we are also able to use STARTTLS on port 143 (incoming) and 587 (outgoing) too.

You should be able to send and receive emails as normal now.

Set up SPF, DKIM, and DMARC#

SPF, DKIM, and DMARC records are TXT DNS records that protect against people spoofing your domain. Having correctly configured SPF, DKIM, and DMARC records helps your mail avoid getting caught by spam filters.

SPF records indicate which hosts or IP addresses are allowed to send email from your domain.

DKIM digitally signs emails with a private key to verify that they came from your server. If someone else without your private key sends email from your domain, they wouldn’t be able to produce a valid DKIM signature. DKIM public keys are published in TXT DNS records, so that any other mail server can verify a signature of an email from your domain.

DMARC records state that your domain uses SPF and DKIM, and that emails sent from your domain should pass SPF and DKIM checks.

Sender Policy Framework#

Set up your DNS record#

Add a TXT record for your root domain with the contents v=spf1 mx ~all, like:

revsuine.xyz.                    900   IN TXT   "v=spf1 mx ~all"

In your DNS manager, you would enter:

Name @
Type TXT
TTL Default (mine is 900)
TXT data v=spf1 mx ~all

This is assuming you’re sending emails from domain.com instead of, e.g. sub.domain.com. If your email addresses are like user@sub.domain.com, you would use sub as the name in the DNS record instead of @, which denotes the apex domain. The same applies for all the DNS records in this section.

Breaking down the TXT data:

Part Explanation
v=spf1 This is an SPF record, and the SPF record version is SPF1.
mx All hosts listed in MX records are allowed to send emails from this domain.
~all Emails from your domain should only come from hosts specified by your SPF record. Can also be +all, -all, ?all, but those are rarely used. -all means that emails not from SPF-approved hosts should be rejected; however, this option can cause valid emails to be rejected if the recipient has multiple SMTP servers, and an email is relayed from one of their SMTP servers to another, for instance if one SMTP server goes down and they have a backup server.

Get Postfix to validate SPF#

We’re going to use a Postfix SMTPd policy server called postfix-policyd-spf-perl to check SPF of incoming emails. postfix-policyd-spf-perl is very simple and requires almost no configuration.

Install postfix-policyd-spf-perl and create a user, policyd-spf for it:

# apk add postfix-policyd-spf-perl
# adduser -S -s /sbin/nologin -h /dev/null -H policyd-spf

Explanation of adduser flags:

Option Explanation
-S Create a system user
-s /sbin/nologin Set shell to /sbin/nologin so the user doesn't have a shell
-h /dev/null Set home directory to /dev/null
-H Don't create a home directory (if you try to create /dev/null and assign it to policyd-spf there will be all sorts of permissions issues)

Now edit /etc/postfix/master.cf to tell Postfix to start up the postfix-policyd-spf-perl daemon:

policyd-spf   unix -     n       n       -       0       spawn
  user=policyd-spf argv=/usr/bin/postfix-policyd-spf-perl

Now get Postfix to use postfix-policyd-spf-perl in /etc/postfix/main.cf by adding the following lines:

smtpd_recipient_restrictions =
    permit_mynetworks,
    reject_unauth_destination,
    check_policy_service unix:private/policyd-spf
policyd-spf_time_limit = 3600

postfix-policyd-spf-perl is now set up, and you can test it by sending yourself an email from a mainstream email provider (which ought to have an SPF record) and checking for the presence of this header:

Received-SPF: pass (protonmail.com: Sender is authorized to use 'revsuine@protonmail.com' in 'mfrom' identity (mechanism 'include:_spf.protonmail.ch' matched))

DomainKeys Identified Mail#

Configure OpenDKIM#

OpenDKIM is an open-source implementation of DKIM signing and authentication. You can install it with:

# apk add opendkim opendkim-utils

You may also want opendkim-doc for documentation.

Enable the service:

# rc-update add opendkim default
# rc-service opendkim start

Add the postfix user to the opendkim group:

# adduser postfix opendkim

Edit your OpenDKIM config file at /etc/opendkim/opendkim.conf. Ensure the following is present (Canonicalization is likely already there, so just change and uncomment things, and add lines where necessary):

Canonicalization        relaxed/simple
Mode                    sv
SubDomains              no
AutoRestart             yes
AutoRestartRate         10/1M
Background              yes
DNSTimeout              5
SignatureAlgorithm      rsa-sha256

Add the following to the config file:

# OpenDKIM user
# Remember to add user postfix to group opendkim
UserID             opendkim

# Map domains in From addresses to keys used to sign messages
KeyTable           refile:/etc/opendkim/key.table
SigningTable       refile:/etc/opendkim/signing.table

# Hosts to ignore when verifying signatures
ExternalIgnoreList  /etc/opendkim/trusted.hosts

# A set of internal hosts whose mail should be signed
InternalHosts       /etc/opendkim/trusted.hosts

Create a directory for OpenDKIM’s keys at /etc/opendkim/keys/, and ensure the directory is owned by opendkim and that no one else has read or write access to the keys directory:

# mkdir /etc/opendkim/keys
# chown -R opendkim:opendkim /etc/opendkim
# chmod go-rw /etc/opendkim/keys

Create /etc/opendkim/signing.table and set its contents to the following:

*@domain.com      default._domainkey.domain.com
*@*.domain.com    default._domainkey.domain.com

This tells OpenDKIM to use your domain’s private key to sign any emails originating from domain.com and its subdomains. Replace domain.com with your domain.

Now create /etc/opendkim/key.table and set its contents to:

default._domainkey.domain.com     domain.com:default:/etc/opendkim/keys/domain.com/default.private

Replacing domain.com with your domain. This tells OpenDKIM where your private key is stored.

Create /etc/opendkim/trusted.hosts and set its contents to

127.0.0.1
localhost

.domain.com

Similarly, replace domain.com with your domain.

Generate your keys#

Create a directory for your domain’s keys at /etc/opendkim/keys/domain.com. Use opendkim-genkey to generate your keys:

# opendkim-genkey -b 2048 -d domain.com -D /etc/opendkim/keys/domain.com -s default -v

Replace domain.com with your domain.

Set the owner and permissions of the private key so that only opendkim has read and write access to it:

# chown opendkim:opendkim /etc/opendkim/keys/domain.com/default.private
# chmod 600 /etc/opendkim/keys/domain.com/default.private

Publish your public key in your DNS records#

Look at your public key:

# cat /etc/opendkim/keys/domain.com/default.txt

You’ll see something like this:

default._domainkey	IN	TXT	( "v=DKIM1; k=rsa; "
	 "p=MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEAu7QJxihVsEGhsZBLMxGKCVecIhjINp/zMGH0ExHN5QelxRIKdAHpJ33cZUBEsL0H27/4E8BK9RZ4AOaDMHfIGJ9FybyyD+gHhvKFFHzyts0QpkwLV+3bApsMqSR4ZsSzt3pZvRhQrXaoUds+9CyQBuEGx16PcA2O04/vWlvWhwStJrZWWHDYl2PX3QHfFFSsJaTsiFMwnMECvg"
	 "a6TdBTcsbcHwfedkoeMv5RQFlKisqCEiJkUYbJaczDA88fcZm21eGW8HiCSwq3324ExfWSwPGVsXAn1Blq/oKklpjXQIwwrphglK//G6AF7M3AAfqBdCvUbkAjs28fJwGqiVxaCwIDAQAB" )  ; ----- DKIM key default for revsuine.xyz

This is what your DNS record should look like. Copy everything between the brackets and add it to your DNS record:

Name default._domainkey
Type TXT
TTL Default value (mine is 900)
TXT data What you copied from between the brackets, with all double quotes and extra whitespace (beyond a single space character) removed

So with the above public key, my DNS record is

default._domainkey.revsuine.xyz. 900   IN TXT   (
                                                  "v=DKIM1; k=rsa; p=MIIBIjANBgkqhkiG9w0BAQ"
                                                  "EFAAOCAQ8AMIIBCgKCAQEAu7QJxihVsEGhsZBLMx"
                                                  "GKCVecIhjINp/zMGH0ExHN5QelxRIKdAHpJ33cZU"
                                                  "BEsL0H27/4E8BK9RZ4AOaDMHfIGJ9FybyyD+gHhv"
                                                  "KFFHzyts0QpkwLV+3bApsMqSR4ZsSzt3pZvRhQrX"
                                                  "aoUds+9CyQBuEGx16PcA2O04/vWlvWhwStJrZWWH"
                                                  "DYl2PX3QHfFFSsJaTsiFMwnMECvg a6TdBTcsbcH"
                                                  "wfedkoeMv5RQFlKisqCEiJkUYbJaczDA88fcZm21"
                                                  "eGW8HiCSwq3324ExfWSwPGVsXAn1Blq/oKklpjXQ"
                                                  "IwwrphglK//G6AF7M3AAfqBdCvUbkAjs28fJwGqi"
                                                  "VxaCwIDAQAB"
                                                )

and what I entered into the data field of my DNS manager when adding the record was

v=DKIM1; k=rsa; p=MIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKCAQEAu7QJxihVsEGhsZBLMxGKCVecIhjINp/zMGH0ExHN5QelxRIKdAHpJ33cZUBEsL0H27/4E8BK9RZ4AOaDMHfIGJ9FybyyD+gHhvKFFHzyts0QpkwLV+3bApsMqSR4ZsSzt3pZvRhQrXaoUds+9CyQBuEGx16PcA2O04/vWlvWhwStJrZWWHDYl2PX3QHfFFSsJaTsiFMwnMECvg a6TdBTcsbcHwfedkoeMv5RQFlKisqCEiJkUYbJaczDA88fcZm21eGW8HiCSwq3324ExfWSwPGVsXAn1Blq/oKklpjXQIwwrphglK//G6AF7M3AAfqBdCvUbkAjs28fJwGqiVxaCwIDAQAB

You can now test your DKIM key with:

# opendkim-testkey -d domain.com -s default -vvv

Don’t worry about

opendkim-testkey: key not secure

This just means that DNSSEC isn’t enabled for your domain name.

Use OpenDKIM with Postfix#

You can use a Unix socket for this, but I couldn’t get the Unix socket to work, so I’ll show you how to connect Postfix to OpenDKIM using a TCP/IP socket on port 8891 instead.

Edit your OpenDKIM config /etc/opendkim/opendkim.conf, and ensure that the following line is present:

Socket                  inet:8891@localhost

Also make sure there are no other uncommented Socket lines.

Now configure Postfix at /etc/postfix/main.cf, and add the following lines:

milter_default_action = accept
milter_protocol = 6
smtpd_milters = inet:127.0.0.1:8891
non_smtpd_milters = $smtpd_milters

This uses the Milter extension, which is something that can be used to process mail; in this case, to add headers to emails relating to DKIM.

You can, again, test this on both incoming and outgoing mail. On outgoing mail, there should be a DKIM-Signature: header present. On incoming mail from domains implementing DKIM, there should be a Authentication-Results: master.revsuine.xyz; header (obviously replacing master.revsuine.xyz with your hostname) indicating whether or not the email has passed DKIM authentication.

Domain-based Message Authentication, Reporting, and Conformance#

Ensure your domains are aligned in email headers#

Send a test email from your domain and look at the email headers of the sent email.

Return-Path: <pid1@revsuine.xyz>
Received: from master.revsuine.xyz (master.revsuine.xyz. [93.113.25.226])
        by mx.google.com with ESMTPS id ffacd0b85a97d-3825fb5a132si1538595f8f.66.2024.11.22.08.53.01
        for <revsuine@gmail.com>
        (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256);
        Fri, 22 Nov 2024 08:53:01 -0800 (PST)
Received-SPF: pass (google.com: domain of pid1@revsuine.xyz designates 93.113.25.226 as permitted sender) client-ip=93.113.25.226;
Authentication-Results: mx.google.com;
       dkim=pass header.i=@revsuine.xyz header.s=default header.b=HRewRpbu;
       spf=pass (google.com: domain of pid1@revsuine.xyz designates 93.113.25.226 as permitted sender) smtp.mailfrom=pid1@revsuine.xyz;
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=revsuine.xyz;
        s=default; t=1732294380;
        bh=Q6d5/50T+3BnL4k+i/ARiher4CIHkBrfcsHufAU/i4E=;
        h=Subject:From:To:Date;
        b=HRewRpbuXhaI4vf8ZXc0PzepcNAd5mGDvWh38mAF+JYvefM7pF+q0dNYrA0bCrCys
         a6qev9C++7hJC7oLYKgcXoewHPiGQzfgjj5oIVuIE10CSuSNyk4vd4Nf37P8k1A8lO
         jy+g+kv7VbX5N+ElzQbqXL1PTjgd9seDYoPn5FTTWGfVbLhh6gWfU3xeb88UZQCFJ+
         kgWMFOUEjoTffzqNBe/NICvcoX3ZJvr+vCdZVANHEV1AhVExDB9FM/0TRFnybhTVo0
         eGDCEtvkxgNQYVFVnJUxWZ3MCt+cDVuYVsHw0SXQMToiNSPlsn65iV2+/giw7DmSCo
         fgOvYG+eHu10w==
Message-ID: <47971a581909651b6e089287b5b408fd56352b82.camel@revsuine.xyz>
Subject: Test
From: revsuine <pid1@revsuine.xyz>

Ensure that the domains in Return-Path, DKIM-Signature, and From: are the same. If they are, they are strictly aligned. Or if the Return-Path or DKIM-Signature’s d= parameter use a subdomain instead of the main domain name in the From: field, they have relaxed alignment. Either way, they are aligned.

By default, DMARC uses relaxed alignment. Your domains have to be aligned in order to pass DMARC checks.

Create a DMARC DNS record#

Create a TXT record with _dmarc as the record name, and

v=DMARC1; p=none; pct=100; fo=1; rua=mailto:dmarc@domain.com

as the TXT data, replacing dmarc@domain.com with an email address you want mail servers to send their DMARC reports to.

Policies#

p=none indicates the policy for our DMARC record. The available policies are:

Policy Meaning
none Don't do anything if the DMARC check fails
quarantine Quarantine emails if DMARC check fails; an admin must approve the email before it reaches the recipient's inbox
reject Reject the email if DMARC check fails

It’s recommended to start with none or quarantine until you’re sure that your emails are passing DMARC. none is more appropriate for domains that have already sent some emails; if you’ve never sent email from your domain before, try quarantine with pct=30 (see below) as a start.

If your domain is not going to send emails, set p=reject.

Percentage#

pct=100 means that a mail server should expect 100% of incoming mail from your domain to pass DMARC checks. As you increase the strictness of your policy, you should also decrease pct until you are sure that your email is passing checks.

LinuxBabe recommends the following “progression”, moving to the next row in the table as you’ve tested the previous row and haven’t seen legitimate mail rejected:

Policy Percentage
none 100
quarantine 30
quarantine 70
quarantine 100
reject 30
reject 70
reject 100

Email reports#

You set an email address for DMARC reports to be sent to with rua=mailto:dmarc@domain.com. This means that when a mail server receives email from your domain, if it runs a DMARC check, it will send the results to that email.

The fo tag indicates when you would like to receive reports. The options are:

Option Meaning
0 Generate reports if all authentication mechanisms fail
1 Generate reports if any authentication mechanisms fail
d Generate reports if DKIM authentication fails
s Generate reports if SPF authentication fails

OpenDMARC3#

We can use software called OpenDMARC to enforce DMARC policies for incoming mail. OpenDMARC is another milter. Let’s install it and enable its service:

# apk add opendmarc
# rc-update add opendmarc
# rc-service opendmarc start

Edit the OpenDMARC config at /etc/opendmarc/opendmarc.conf.

Change

AuthservID HOSTNAME

to

AuthservID OpenDMARC

This is so that the Authentication-Results header from OpenDKIM authentication. This will also make it clear which program adds which Authentication-Results header.

Add the following line, replacing mail.domain.com with your hostname (so in my instance, this is master.revsuine.xyz).

TrustedAuthservIDs mail.domain.com

This specifies that OpenDMARC should trust authentication results from mail.domain.com. Otherwise you would get the following error message in your syslog:

ignoring Authentication-Results at 1 from mail.domain.com

Enable RejectFailures, which means your server will comply with p=reject in DMARC DNS records.

RejectFailures true

You also probably want to enable RequiredHeaders, which rejects emails that don’t conform to RFC5322 standards, e.g. are missing a From: header.

RequiredHeaders true

In case external SPF validation fails (as in, no SPF results are placed in the message header), you probably want to add

SPFSelfValidate true

which tells OpenDMARC to perform the SPF check itself if it can’t find SPF results in the message header.

Now provide OpenDMARC with a socket to use for communication with sendmail. We will use a TCP socket on port 8893:

Socket inet:8893@localhost

For a Unix socket, you’d use the following format:

Socket local:/var/run/opendmarc/opendmarc.sock

By default, you will have the line

IgnoreHosts /etc/opendmarc/ignore.hosts

in /etc/opendmarc/opendmarc.conf. This tells OpenDMARC to not authenticate the list of hosts in /etc/opendmarc/ignore.hosts. An example ignore.hosts is

127.0.0.1
93.113.25.226

Keep in mind that if you have specified IgnoreHosts, this file needs to exist in order for OpenDMARC to run. If you have the option set, make sure to touch /etc/opendmarc/ignore.hosts (or whatever filepath you’ve specified). Alternatively, comment out this option in order to use the default, which is to not authenticate mail coming from 127.0.0.1.

Restart OpenDMARC for these changes to take effect:

# rc-service opendmarc restart

To have Postfix use the OpenDMARC milter, it’s simple as adding the socket to the smptd_milters and non_smtpd_milters variable in /etc/postfix/main.cf:

milter_default_action = accept
milter_protocol = 6
smtpd_milters = inet:127.0.0.1:8891,inet:127.0.0.1:8893
non_smtpd_milters = $smtpd_milters

Restart Postfix for the changes to take effect:

# rc-service postfix restart

And when you receive emails from a legitimate source that implements DMARC, you should see the following headers in your emails:

Received-SPF: pass (protonmail.com: Sender is authorized to use 'revsuine@protonmail.com' in 'mfrom' identity (mechanism 'include:_spf.protonmail.ch' matched)) receiver=master.revsuine.xyz; identity=mailfrom; envelope-from="revsuine@protonmail.com"; helo=mail-40130.protonmail.ch; client-ip=185.70.40.130
DMARC-Filter: OpenDMARC Filter v1.4.2 master.revsuine.xyz 88CFF1288D1
Authentication-Results: OpenDMARC; dmarc=pass (p=quarantine dis=none) header.from=protonmail.com
Authentication-Results: OpenDMARC; spf=pass smtp.mailfrom=protonmail.com
Authentication-Results: master.revsuine.xyz;
        dkim=pass (2048-bit key; secure) header.d=protonmail.com header.i=@protonmail.com header.a=rsa-sha256 header.s=protonmail3 header.b=nc4YWVM/

Test SPF, DKIM, and DMARC#

You can use mail-tester.com and send an email from your domain to check that SPF, DKIM, and DMARC are all working for your server.

Screenshot of mail-tester.com authentication results passing all checks

Failure#

There are a few things that can cause SPF, DKIM, and consequently DMARC to fail.

One common scenario is email relaying causing SPF to fail, because the recipient mail server does not receive the email directly from your mail server.

Mailing Lists#

“Announcement” type mailing lists can cause DKIM failure, because usually additional headers are added to your emails. “Discussion” type mailing lists are more problematic; they can also cause SPF failures because emails are coming from all different mail servers through the mailing list server, and the central mailing list server cannot sign other domain’s DKIM signatures.

For announcements, if you are using an external mailing list provider, just add their server to your SPF record. For discussion mailing lists, resolving authentication issues is more complex. GNU Mailman has a wiki article with some suggestions.

Amavis#

Amavis is a high-performance interface between MTAs and content checkers, such as virus scanners and spam filters. We will use Amavis as a Postfix content filter for virus scanning and spam filtering, with ClamAV and SpamAssassin respectively.

Install Amavis and enable the service:

# apk add amavis
# rc-update add amavisd default
# rc-service amavisd start

Amavis and Postfix#

Let’s set up Amavis as an SMTP proxy. Edit your Postfix config at /etc/postfix/main.cf, and add the following to the end of the file:

# amavis filtering
# gets overridden by submission & smtps services in master.cf:
content_filter = smtp-amavis:[127.0.0.1]:10024
# delays postfix connection to content filter until entire email message has arrived
smtpd_proxy_options = speed_adjust

Amavis listens on port 10024 by default, so this tells Postfix to use Amavis as a content filter.

Now edit /etc/postfix/master.cf and add the following lines:

smtp-amavis   unix -     -       n       -       2       smtp
  -o syslog_name=postfix/amavis
  -o smtp_data_done_timeout=1200
  -o smtp_send_xforward_command=yes
  -o smtp_dns_support_level=disabled
  -o max_use=20
  -o smtp_tls_security_level=none

127.0.0.1:10025   inet   n    -     n     -     -    smtpd
  -o syslog_name=postfix/10025
  -o content_filter=
  -o mynetworks_style=host
  -o mynetworks=127.0.0.0/8
  -o local_recipient_maps=
  -o relay_recipient_maps=
  -o strict_rfc821_envelopes=yes
  -o smtp_tls_security_level=none
  -o smtpd_tls_security_level=none
  -o smtpd_restriction_classes=
  -o smtpd_delay_reject=no
  -o smtpd_client_restrictions=permit_mynetworks,reject
  -o smtpd_helo_restrictions=
  -o smtpd_sender_restrictions=
  -o smtpd_recipient_restrictions=permit_mynetworks,reject
  -o smtpd_end_of_data_restrictions=
  -o smtpd_error_sleep_time=0
  -o smtpd_soft_error_limit=1001
  -o smtpd_hard_error_limit=1000
  -o smtpd_client_connection_count_limit=0
  -o smtpd_client_connection_rate_limit=0
  -o receive_override_options=no_header_body_checks,no_unknown_recipient_checks,no_address_mappings
  # avoid double dkim signing by setting smtpd_milters to empty
  # otherwise will run all milters again after amavis
  -o smtpd_milters=

The first block tells Postfix to send emails to Amavis, and the second block tells Postfix to run an extra smtpd daemon on port 10025 to receive emails back from Amavis. Restart Postfix for the changes to take effect:

# rc-service postfix restart

It’s good practice to use a different port for email submissions from authenticated users. Let’s use port 10026 for this. Edit /etc/amavisd.conf and set $inet_socket_port to:

$inet_socket_port = [10024,10026];

to listen on multiple TCP ports.

We’ll set the policy to ORIGINATING for port 10026 in the same Amavis config file:

$interface_policy{'10026'} = 'ORIGINATING';

Then define the ORIGINATING policy by adding the following lines:

$policy_bank{'ORIGINATING'} = {  # mail supposedly originating from our users
  originating => 1,  # declare that mail was submitted by our smtp client
  allow_disclaimers => 1,  # enables disclaimer insertion if available
  # notify administrator of locally originating malware
  virus_admin_maps => ["virusalert\@$mydomain"],
  spam_admin_maps  => ["virusalert\@$mydomain"],
  warnbadhsender   => 1,
  # force MTA conversion to 7-bit (e.g. before DKIM signing)
  smtpd_discard_ehlo_keywords => ['8BITMIME'],
  bypass_banned_checks_maps => [1],  # allow sending any file names and types
  terminate_dsn_on_notify_success => 0,  # don't remove NOTIFY=SUCCESS option
};

Note that this is configured to send virus alerts to virusalert@domain.com. This should be a real email address, not an alias, because ClamAV bypasses Postfix and sends emails straight to Dovecot, which doesn’t have access to Postfix aliases.

Restart Amavis for the change to take effect:

# rc-service amavisd restart

Edit /etc/postfix/master.cf now and add the following to the submission and smtps services:

  # emails from authenticated SMTP clients will be passed to Amavis listening
  # on port 10026:
  -o content_filter=smtp-amavis:[127.0.0.1]:10026

Restart Postfix again for the changes to take effect:

# rc-service postfix restart

Amavis and ClamAV#

Install ClamAV and enable its daemon:

# apk add clamav clamav-daemon
# rc-update add clamd default
# rc-service clamd start

Enable virus-checking in Amavis by editing /etc/amavisd.conf and setting @bypass_virus_checks_maps to the following:

# enable virus checking
@bypass_virus_checks_maps = (
    \%bypass_virus_checks, \@bypass_virus_checks_acl, \$bypass_virus_checks_re);

Add clamav to the amavis group:

# adduser clamav amavis

Restart the amavisd and clamd daemons:

# rc-service amavisd restart
# rc-service clamd restart

Amavis and SpamAssassin#

Install SpamAssassin:

# apk add spamassassin

You may also want to install spamassassin-doc.

Let’s configure SpamAssassin. SpamAssassin is configured at /etc/mail/spamassassin/local.cf. You may want to configure the required_score option, which defaults to 5.0. This is the spam score required for an email to be marked spam. 5.0 is a sensible default, but you can adjust this if you find that your spam filter needs to be more or less aggressive.

You can also set options like rewrite_header to rewrite headers of a message marked spam, e.g.

rewrite_header Subject [SPAM]

prepends [SPAM] to the subject line of a spam message.

Enable the service:

# rc-update add spamd default
# rc-service spamd start

Enable spam filtering by setting @bypass_spam_checks_maps in your /etc/amavisd.conf:

# enable spam filtering
@bypass_spam_checks_maps = (
    \%bypass_spam_checks, \@bypass_spam_checks_acl, $bypass_spam_checks_re);

Restart Amavis:

# rc-service amavisd restart

Now let’s test your spam filter. Send yourself an email containing the following string somewhere in the body:

XJS*C4JDBQADN1.NSBN3*2IDNEN*GTUBE-STANDARD-ANTI-UBE-TEST-EMAIL*C.34X

You should see the email arrive with the following headers:

X-Spam-Flag: YES
X-Spam-Score: 999.802
X-Spam-Level: ****************************************************************
X-Spam-Status: Yes, score=999.802 tagged_above=2 required=6.2
 tests=[DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1,
 DKIM_VALID_EF=-0.1, GTUBE=1000, NO_RECEIVED=-0.001, NO_RELAYS=-0.001,
 TVD_SPACE_RATIO=0.001, URIBL_BLOCKED=0.001, URIBL_DBL_BLOCKED_OPENDNS=0.001,
 URIBL_ZEN_BLOCKED_OPENDNS=0.001] autolearn=no autolearn_force=no

Pigeonhole#

Dovecot can do server-side mail filtering with sieve scripts. These are user scripts that can perform actions on mail based on particular criteria, e.g.

require "fileinto";

if address :is "to" "postmaster@revsuine.xyz" {
    fileinto "Postmaster";
}

Places mail in the Postmaster folder if the To: field is postmaster@revsuine.xyz. You also can do things unconditionally, like

redirect postmaster@revsuine.xyz;

unconditionally redirects all mail to postmaster@revsuine.xyz.

Sieve scripts can be both per-user and system-wide.

For more examples, this page has some good examples.

Installing and setting up Pigeonhole#

To use Sieve, install dovecot-pigeonhole-plugin:

# apk add dovecot-pigeonhole-plugin

Then edit /etc/dovecot/conf.d/20-lmtp.conf, and add the sieve plugin like so:

protocol lmtp {
  # Space separated list of plugins to load (default is global mail_plugins).
  mail_plugins = $mail_plugins sieve
}

To configure Pigeonhole and sieve, edit /etc/dovecot/conf.d/90-sieve.conf. Sieve’s options will be configured in the plugin {} block in this file.

We can set the location of user sieve scripts with the sieve option.

sieve = file:~/sieve;active=~/.dovecot.sieve

means that ~/sieve is a directory of sieve scripts, whilst ~/.dovecot.sieve is a symlink to the “active” one, e.g.

sieve
├── script1.sieve
├── script2.sieve
└── script3.sieve

could be your ~/sieve/ directory, and to make script2.sieve active, you would do

$ ln -s ~/sieve/script2.sieve ~/.dovecot.sieve

sieve_before defines a directory of sieve scripts which will be executed prior to any user scripts. e.g.

sieve_before = /etc/dovecot/sieve

means that the sieve scripts in /etc/dovecot/sieve will be executed first, then the user’s personal scripts at ~/.dovecot.sieve.

You can specify multiple directories in order, like so:

sieve_before = /var/lib/dovecot/sieve.d/
sieve_before2 = ldap:/etc/sieve-ldap.conf;name=ldap-domain
sieve_before3 = /etc/dovecot/sieve

etc. The sieve_after option also exists, and works the same way.

This is not the same as sieve_default, which is overridden by user sieve scripts and only executes when a user has no sieve script.

You can review this documentation for a full list of Pigeonhole options.

ManageSieve#

Users can configure their own user sieve scripts using a protocol called ManageSieve. Like how IMAP allows users to read their emails without having shell access to the mail server, ManageSieve allows users to write sieve scripts without requiring shell access.

To enable ManageSieve, edit /etc/dovecot/conf.d/20-managesieve.conf. Make sure the following line is uncommented:

protocols = $protocols sieve

By default, ManageSieve will listen on port 4190.

Sieve scripts for spam filtering#

Let’s use a system-wide sieve script to file SpamAssassin-marked spam into a Spam folder. Create an /etc/dovecot/sieve/ directory, and add it to your sieve_before settings:

plugin {
  ...
  sieve_before = /etc/dovecot/sieve/
  ...
}

Now create a new sieve script, /etc/dovecot/sieve/spam_folder.sieve:

require ["fileinto", "mailbox"];
if header :contains "X-Spam-Flag" "YES" {
  fileinto :create "Spam";
}

Replace "Spam" with the name of your spam folder. If it’s a subfolder, e.g.

Inbox
└── Spam

you would write "Inbox.Spam" in that case.

Miscellaneous suggestions#

To use fail2ban for your mail server, create /etc/fail2ban/jail.d/mail.local:

[postfix]
enabled  = true
logpath  = /var/log/postfix.log

[dovecot]
enabled  = true
logpath  = /var/log/dovecot.log

[sieve]
enabled  = true
logpath  = /var/log/dovecot.log

These are the log file locations we configured above, but you can set them to the syslog if you didn’t give Postfix and Dovecot their own log files.

You may want to get your domain whitelisted on dnswl.org, an email whitelist service where admins can submit their domain and IP address to indicate trustworthiness.

Finally, you can PGP-encrypt users’ incoming mail with their public keys, much like Protonmail.


  1. This is only true to the extent that your server is not compromised. You could say there’s an order of server trust-ability that goes:

    VPS < rented dedicated server < server you yourself physically own, store, and manage
    

    There is little that can be done to secure a VM running on a compromised host. Even with full-disk encryption, the host can dump the encryption key from RAM, because the encryption key must be stored in memory whilst a full-disk-encrypted system is booted.

    For a dedicated server you rent, there are at least no concerns about a compromised host, but an attacker with physical access (in this case, the untrusted people you rent the dedicated server from) can attempt evil maid attacks. You are hopefully able to implement mechanisms to detect this, though.

    There are reasons you may want to go with a rented server instead of one you own, though. For instance, if you live in a jurisdiction known for terrible privacy laws such as a 5/14 eyes country, or if you are a political dissident with domestic state interest in you, you likely want to go offshore for server hosting. Changing the jurisdiction can protect you if the jurisdiction you choose won’t work with your national intelligence agencies. ↩︎

  2. Your aliases file will most likely be in this location by default, but you can run

    $ postconf alias_maps
    

    to find out where this file should be. ↩︎

  3. The OpenDMARC milter will be run before the SPF policy daemon, meaning that OpenDMARC must do its own SPF checking. This makes postfix-policyd-spf-perl redundant, and results in multiple SPF checks in our email headers (e.g. see email headers here).

    To get around this, you’d probably have to use a milter to validate SPF, and just list the milter before OpenDMARC. I had a bit of a look around for SPF milters and found ACME’s spfmilter, but had issues compiling it on Alpine; it complained about

    spfmilter.c:1623:2: error: #error "neither libspf nor libspf2 is present - please provide one"
     1623 | #error "neither libspf nor libspf2 is present - please provide one"
          |  ^~~~~
    

    despite having libspf2 installed. Potentially this is a musl issue, but I’m doubtful that a project like this would be tied to glibc specifically.

    I tried compiling it on Artix Linux and it compiled just fine with the right dependencies installed, so the program does compile.

    If anyone can either suggest a way to get ACME’s spfmilter to compile on Alpine, or an alternative solution to this problem, that would be helpful. ↩︎