mklassen/certbot-haproxy

HAProxy plugin for Let's Encrypt's Certbot

HAProxy plugin for Certbot

Table of Contents

• HAProxy plugin for Certbot
  • About
  • Installing: Requirements
  • Installing: Getting started
  • Development: Getting started
• Vagrant machines
  • Development: Running locally without sudo
  • Further time savers during development..
  • Setuptools version conflict
  • Making a .deb debian package

About

This is a certbot plugin for using certbot in combination with a HAProxy setup. Its advantage over using the standalone certbot is that it automatically places certificates in the correct directory and restarts HAProxy afterwards. It should also enable you to very easily do automatic certificate renewal.

Furthermore, you can configure HAProxy to handle Boulder's authentication using the HAProxy authenticator of this plugin.

It was created for use with Greenhost's shared hosting environment and can be useful to you in the following cases:

• If you use HAProxy and have several domains for which you want to enable Let's Encrypt certificates.
• If you yourself have a shared hosting platform that uses HAProxy to redirect to your client's websites.
• Actually any case in which you want to automatically restart HAProxy after you request a new certificate.

This plugin does not configure HAProxy for you, because HAProxy configurations can can vary a great deal. Please read the installation instructions on how to configure HAProxy for use with the plugin. If you have a good idea on how we can implement automatic HAProxy configuration, you are welcome to create a merge request or an issue.

Installing: Requirements

Currently this plugin has been tested on Debian Jessie, but it will most likely work on Ubuntu 14.04+ too. If you are running Debian Wheezy, you may need to take additional steps during the installation. Thus, the requirements are:

• Debian Jessie (or higher) or Ubuntu Trusty (or higher).
• Python 2.7 (2.6 is supported by certbot and our goal is to be compatible but it has not been tested yet).
• HAProxy 1.6+ (we will configure SNI, which is not strictly required)
• Certbot 0.8+

Installing: Getting started

The installation below assumes you are running Debian Jessie but it should be almost entirely the same process on Ubuntu.

First add the backports repo for Jessie to your apt sources.

Note

This will not work for Ubuntu, you will need to use another source, check which version comes with your version of Ubuntu, if it is a version below 0.8, you need to find a back port PPA or download certbot from source.

echo "deb http://ftp.debian.org/debian jessie-backports main" >> \
    /etc/apt/sources.list.d/jessie-backports.list

Now update, upgrade and install some requirements:

• Some utilities: sudo tcpdump ufw git curl wget
• OpenSSL and CA certificates: openssl ca-certificates
• Build dependencies: build-essential libffi-dev libssl-dev python-dev
• Python and related: python python-setuptools
• HAProxy: haproxy
• Python dependency managing: pip

apt-get update
apt-get upgrade -y
apt-get install -y \
    sudo tcpdump ufw git curl wget \
    openssl ca-certificates \
    build-essential libffi-dev libssl-dev python-dev \
    python python-setuptools \
    haproxy

easy_install pip
pip install --upgrade setuptools

We also installed a simple firewall above, but it is not yet configured, let's do that now:

ufw allow ssh
ufw allow http
ufw allow https
ufw default deny incoming
ufw --force enable

Warning

You probably want a little more protection for a production proxy than just this simple firewall, but it's out of the scope of this readme.

Now that we have all dependencies, it's time to start a process that may take quite some time to complete. HAProxy comes with a DH parameters file that is considered weak. We need to generate a new dhparams.pem file with a prime of at least 2048 bit length, you can also opt for 3072 or 4096. This can take hours on lower specification hardware, but will still take minutes on faster hardware, especially with 4096 bit primes. Run this is in a separate ssh session or use screen of tmux to allow this to run in the background.

openssl dhparam -out /opt/certbot/dhparams.pem 2048

Now set a hostname.

echo "[INSERT YOUR HOSTNAME HERE]" > /etc/hostname
hostname -F /etc/hostname

If you want to run Certbot in an unprivileged mode, keep reading, otherwise, skip to the installation of Certbot.

Certbot normally requires access to the /etc/ directory, which is owned by root and therefore, Certbot needs to run as root. However, we don't like it when processes run as root, most especially when they are opening ports on a public network interface..

In order to let Certbot run as an unprivileged user, we will:

• Create a certbot user with a home directory on the system so the automatic renewal of certificates can be run by this user.
• Tell Certbot that the working directories are located in certbot's home directory.
• Optionally: add your own user account to the Certbot user's group so you can run Certbot manually.
• Allow HAProxy to access the certificates that are generated by Certbot.
• Allow the certbot user to restart the HAProxy server.

Lastly, to do automatic renewal of certificates, we will create a systemd timer and a service to start at every boot and every 12 hours, at a random time off the day, in order to not collectively DDOS Let's Encrypts service.

useradd -s /bin/bash -m -d /opt/certbot certbot
usermod -a -G certbot haproxy  # Allow HAProxy access to the certbot certs
mkdir -p /opt/certbot/logs
mkdir -p /opt/certbot/config
mkdir -p /opt/certbot/.config/letsencrypt

If you need to use Certbot from your user account, or if you have a daemon running on your proxy server, that configures domains on your proxy, e.g.: in a web hosting environment - you can add those users to the certbot group.

usermod -a -G certbot [ADD YOUR USER HERE]

You will also need to tell your user what the working directory of your Certbot setup is (/opt/certbot/). Certbot allows you to create a configuration file with default settings in the users' home dir: opt/certbot/.config/letsencrypt/cli.ini.

Besides the working directory.

mkdir -p /opt/certbot/.config/letsencrypt
cat <<EOF > /opt/certbot/.config/letsencrypt/cli.ini
work-dir=/opt/certbot/
logs-dir=/opt/certbot/logs/
config-dir=/opt/certbot/config
EOF

Next time you run Certbot, it will use our new working directory.

Now to allow the certbot user to restart HAProxy, put the following in the sudoers file:

cat <<EOF >> /etc/sudoers
%certbot ALL=NOPASSWD: /bin/systemctl restart haproxy
EOF

Now we haven't done one very essential thing yet, install certbot-haproxy. Since our plugin is in an alpha stage, we did not package it yet. You will need to get it from our Gitlab server.

git clone https://code.greenhost.net/open/certbot-haproxy.git
cd ./certbot-haproxy/
sudo pip install ./

Let's Encrypt's CA server will try to contact your proxy on port 80, which is most likely in use for your and/or your customers' websites. So we have configured our plugin to open port 8000 to verify control over the domain instead. Therefore we need to forward verification requests on port 80 to port 8000 internally.

The sample below contains all that is required for a working load-balancing HAProxy setup that also forwards these verification requests. But it is probably not "copy-paste compatible" with your setup. So you need to piece together a configuration that works for you.

cat <<EOF > /etc/haproxy/haproxy.cfg
global
    log /dev/log local0
    log /dev/log local1 notice
    chroot /var/lib/haproxy
    stats socket /run/haproxy/admin.sock mode 660 level admin
    stats timeout 30s
    user haproxy
    group haproxy
    daemon

    # Default ciphers to use on SSL-enabled listening sockets.
    # Cipher suites chosen by following logic:
    #  - Bits of security 128>256 (weighing performance vs added security)
    #  - Key exchange: EECDH>DHE (faster first)
    #  - Mode: GCM>CBC (streaming cipher over block cipher)
    #  - Ephemeral: All use ephemeral key exchanges
    #  - Explicitly disable weak ciphers and SSLv3
    ssl-default-bind-ciphers AES128+AESGCM+EECDH+SHA256:AES128+EECDH:AES128+AESGCM+DHE:AES128+EDH:AES256+AESGCM+EECDH:AES256+EECDH:AES256+AESGCM+EDH:AES256+EDH:-SHA:AES128+AESGCM+EECDH+SHA256:AES128+EECDH:AES128+AESGCM+DHE:AES128+EDH:AES256+AESGCM+EECDH:AES256+EECDH:AES256+AESGCM+EDH:AES256+EDH:!aNULL:!eNULL:!EXPORT:!DES:!RC4:!MD5:!PSK:!aECDH:!3DES:!DSS
    #ssl-default-bind-options no-sslv3 no-tls-tickets force-tlsv12
    ssl-default-bind-options no-sslv3 no-tls-tickets
    ssl-dh-param-file /opt/certbot/dhparams.pem

defaults
    log     global
    mode    http
    option  httplog
    option  dontlognull
    timeout connect 5000
    timeout client  50000
    timeout server  50000
    errorfile 400 /etc/haproxy/errors/400.http
    errorfile 403 /etc/haproxy/errors/403.http
    errorfile 408 /etc/haproxy/errors/408.http
    errorfile 500 /etc/haproxy/errors/500.http
    errorfile 502 /etc/haproxy/errors/502.http
    errorfile 503 /etc/haproxy/errors/503.http
    errorfile 504 /etc/haproxy/errors/504.http

frontend http-in
    # Listen on port 80
    bind \*:80
    # Listen on port 443
    # Uncomment after running certbot for the first time, a certificate
    # needs to be installed *before* HAProxy will be able to start when this
    # directive is not commented.
    #
    bind \*:443 ssl crt /opt/certbot/haproxy_fullchains/__fallback.pem crt /opt/certbot/haproxy_fullchains

    # Forward Certbot verification requests to the certbot-haproxy plugin
    acl is_certbot path_beg -i /.well-known/acme-challenge
    rspadd Strict-Transport-Security:\ max-age=31536000;\ includeSubDomains;\ preload
    rspadd X-Frame-Options:\ DENY
    use_backend certbot if is_certbot
    # The default backend is a cluster of 4 Apache servers that you need to
    # host.
    default_backend nodes

backend certbot
    log global
    mode http
    server certbot 127.0.0.1:8000

    # You can also configure separate domains to force a redirect from port 80
    # to 443 like this:
    # redirect scheme https if !{ ssl_fc } and [PUT YOUR DOMAIN NAME HERE]

backend nodes
    log global
    balance roundrobin
    option forwardfor
    option http-server-close
    option httpclose
    http-request set-header X-Forwarded-Port %[dst_port]
    http-request add-header X-Forwarded-Proto https if { ssl_fc }
    option httpchk HEAD / HTTP/1.1\r\nHost:localhost
    server node1 127.0.0.1:8080 check
    server node2 127.0.0.1:8080 check
    server node3 127.0.0.1:8080 check
    server node4 127.0.0.1:8080 check
    # If redirection from port 80 to 443 is to be forced, uncomment the next
    # line. Keep in mind that the bind \*:443 line should be uncommented and a
    # certificate should be present for all domains
    redirect scheme https if !{ ssl_fc }

EOF

systemctl restart haproxy

Now you can try to run Certbot with the plugin as the Authenticator and Installer, if you already have websites configured in your HAProxy setup, you may try to install a certificate now.

certbot run --authenticator certbot-haproxy:haproxy-authenticator \
    --installer certbot-haproxy:haproxy-installer

If you want your certbot to always use our Installer and Authenticator, you can add this to your configuration file:

cat <<EOF >> $HOME/.config/letsencrypt/cli.ini
authenticator=certbot-haproxy:haproxy-authenticator
installer=certbot-haproxy:haproxy-installer
EOF

If you need to run in unattended mode, there are a bunch of arguments you need to set in order for Certbot to generate a certificate for you.

• --domain [DOMAIN NAME] The domain name you want SSL to be enabled for.
• --agree-tos Tell Certbot you agree with its TOS
• --email [EMAIL ADDRESS] An e-mail address where issues with certificates can be sent to, as well as changes in the TOS. Or you could supply --register-unsafely-without-email but this is not recommended.

After you run certbot successfully once, there will be 2 certificate files in the certificate directory. This is a pre-requisite for HAProxy to start with the bind *:443 [..] directive in the configuration.

You can auto renew certificates by using the systemd service and timer below. They are set to run every 12 hours because certificates that will not expire soon will not be replaced but certificates that will expire soon, will be replaced in a timely manner. The timer also starts the renewal process 2 minutes after the server boots, this is done so renewal starts immediately after the server has been offline for a long time.

cat <<EOF > /etc/systemd/system/letsencrypt.timer
[Unit]
Description=Run Let's Encrypt every 12 hours

[Timer]
# Time to wait after booting before we run first time
OnBootSec=2min
# Time between running each consecutive time
OnUnitActiveSec=12h
Unit=letsencrypt.service

[Install]
WantedBy=timers.target
EOF

cat <<EOF > /etc/systemd/system/letsencrypt.service
[Unit]
Description=Renew Let's Encrypt Certificates

[Service]
Type=simple
User=certbot
ExecStart=/usr/bin/certbot renew -q
EOF

# Enable the timer and start it, this is not necessary for the service,
# since the timer starts it.
systemctl enable letsencrypt.timer
systemctl start letsencrypt.timer

Development: Getting started

In order to run tests against the Let's Encrypt API we will run a Boulder server, which is the exact same server Let's Encrypt is running. The server is started in Virtual Box using Vagrant. To prevent the installation of any components and dependencies from cluttering up your computer there is also a client Virtual Box instance. Both of these machines can be setup and started by running the dev_start.sh script. This sets up a local boulder server and the letsencrypt client, so don't worry if it takes more than an hour.

Vagrant machines

The dev_start.sh script boots two virtual machines. The first is named 'boulder' and runs a development instance of the boulder server. The second is 'lehaproxy' and runs the client. To test if the machines are setup correctly, you can SSH into the 'lehaproxy' machine, by running vagrant ssh lehaproxy. Next, go to the /lehaproxy directory and run ./tests/boulder-integration.sh. This runs a modified version of certbot's boulder-integration test, which tests the HAProxy plugin. If the test succeeds, your development environment is setup correctly.

Development: Running locally without sudo

You can't run certbot without root privileges because it needs to access /etc/letsencrypt, however you can tell it not to use /etc/ and use some other path in your home directory.

mkdir ~/projects/certbot-haproxy/working
mkdir ~/projects/certbot-haproxy/working/config
mkdir ~/projects/certbot-haproxy/working/logs
cat <<EOF >> ~/.config/letsencrypt/cli.ini
work-dir=~/projects/certbot-haproxy/working/
logs-dir=~/projects/certbot-haproxy/working/logs/
config-dir=~/projects/certbot-haproxy/working/config
EOF

Now you can run Certbot without root privileges.

Further time savers during development..

The following options can be saved in the cli.ini file for the following reasons.

• agree-tos: During each request for a certificate you need to agree to the terms of service of Let's Encrypt, automatically accept them every time.
• no-self-upgrade: Tell LE to not upgrade itself. Could be very annoying when stuff starts to suddenly break, that worked just fine before.
• register-unsafely-without-email: Tell LE that you don't want to be notified by e-mail when certificates are about to expire or when the TOS changes, if you don't you will need to enter a valid e-mail address for every test run.
• text: Disable the curses UI, and use the plain CLI version instead.
• domain example.org: Enter a default domain name to request a certificate for, so you don't have to specify it every time.
• configurator certbot-haproxy:haproxy: Test with the HAProxy plugin every time.

cat <<EOF >> ~/.config/letsencrypt/cli.ini
agree-tos=True
no-self-upgrade=True
register-unsafely-without-email=True
text=True
domain=example.org
authenticator=certbot-haproxy:haproxy-authenticator
installer=certbot-haproxy:haproxy-installer
EOF

Setuptools version conflict

Most likely the python-setuptools version in your os's repositories is quite outdated. You will need to install a newer version, to do this you can run:

pip install --upgrade setuptools

Since pip is part of python-setuptools, you need to have it installed before you can update.

Making a .deb debian package

Requirements:

• python stdeb: pip install --upgrade stdeb
• dh clean: apt-get install dh-make

Run the following commands in your vagrant machine:

apt-file update
python setup.py sdist
# py2dsc has a problem with vbox mounted folders
mv dist/certbot-haproxy-<version>.tar.gz ~
cd ~
py2dsc certbot-haproxy-<version>.tar.gz
cd deb_dist/certbot-haproxy-<version>
# NOTE: Not signed, no signed changes (with -uc and -us)
# NOTE: Add the package to the ghtools repo
dpkg-buildpackage -rfakeroot -uc -us