Home

2021-03-12 | On deploying a Phoenix application to a brand new remote server

(Turn your phone sideways if you are on mobile)

Update: Updated to Phoenix 1.7.7 (using esbuild and tailwind)

Goals for this article:

The goal here is: deploy a an existing Phoenix application to a remote server with some basic security considerations. We also expand on this initial goal by:

*** If you find a security vulnerability in the approach I suggest below, please let me know. I'm very interested in improving my craft. Please email me and I'll respond ASAP.

Why write this?:

I'm interested in improving my craft and learning things from first principles. A lot of times, writing a blog post like this is a fantastic way to learn more about something. That's why.

But... there are better options and solutions to this problem!

Yes, this can be cut down considerably by leveraging other technologies (containers, pre configured OS images) but I was interested in removing abstractions in this exercise and start from the ground up. Maybe a future part 2 could address these improvements.

## So, let's get started! ###

Domain name and DNS config: Buy a domain name

Head over to Namecheap (no affiliation, just what I have used in the past) and buy a cheap domain.

Domain name and DNS config: use Cloudflare to manage DNS records, SSL/TLS protocols and more

Now that we have a domain name, go to CloudFlare, and follow the instructions in this article on how to configure a new "site" on CloudFlare (written by Namecheap). Clicking through CloudFlare's setup will help you configure your DNS and a few other things that are useful and considered best practice. For example, the SSL/TLS config of always using https (great one).

Creating the machine: Create an Ubuntu 20.04 machine on Digital Ocean

Go to Digital Ocean (DO) and create a Ubuntu 20.04 (Ubuntu works well) droplet (machine from now on) and in the Authentication section, choose SSH. If you don't have an SSH key set up with DO, you will have an option to click and it will guide you through it. Select your the key you added and continue clicking through the process.

Security checkpoint: DO will have made access available only through SSH from you local machine. You will be able to login as root on that machine, with `ssh root@ip_for_your_new_box`. Once you select SSH for authentication, logging in with a password is disabled by DO (which is excellent). We also don't know the password for the root user on the machine, by default. We can verify that password logins are disabled by running:

$ cat /etc/ssh/sshd_config | grep "PasswordAuthentication"
PasswordAuthentication no
# PasswordAuthentication.  Depending on your PAM configuration,
# PAM authentication, then enable this but set PasswordAuthentication

Securing the machine: Configure firewall: ufw

The firewall is the first line of defense when it comes to networking. You can configure how the machine communicates with other members of the network. The approach here is to limit communication to only what we will need. Our setup is a good enough one.

$ ufw allow ssh
$ ufw allow http
$ ufw allow https
$ ufw enable

And you can verify things worked by running:

$ ufw status
Status: active

To                         Action      From
--                         ------      ----
22/tcp                     ALLOW       Anywhere
80/tcp                     ALLOW       Anywhere
443/tcp                    ALLOW       Anywhere
22/tcp (v6)                ALLOW       Anywhere (v6)
80/tcp (v6)                ALLOW       Anywhere (v6)
443/tcp (v6)               ALLOW       Anywhere (v6)

Security checkpoint: The only ports accessible to the external world are 22, 80 and 443. If we didn't have this and had a program listening on a certain port, an attacker could potentially interact with it. Now, the firewall will take care of that.

Securing the machine: Create an admin user

Let's create a user that is not root, add a password for it and add it to the `sudo` permissions group. The password for that user will be prompted every time this new user needs higher priviledge when executing a command they would need to know the password. Let's also add SSH access to the machine through that user by using the key we already have set up on our local machine. Let's call this new user admin_deploy.

# Think of a password and use it when running the below:
$ adduser admin_deploy
$ usermod -aG sudo admin_deploy

Security checkpoint: you are able to log in as the admin_deploy and root users by using ssh. The admin_deploy user requires a password to run priviledged commands while root can do anything on the machine (see encrypted passwords that can later be brute forced, install things on the machine, monitor usage, etc).

Securing the machine: Configure SSH for new admin user

Now that we have the user, let's set up SSH for it so we can connect to the machine without connecting as root.

# Change to the admin_deploy user:
$ su admin_deploy

# Go to the correct home directory and create a file:
$ cd ~
$ mkdir .ssh
$ touch .ssh/authorized_keys

# Copy your local public key to the remote machine. (** Do this on your local machine **).
# Assumes you have one already since you used it for creating the droplet. You
# can also use a different one, but for simplicity purposes we will use the
# same one.
$ cat ~/.ssh/id_rsa.pub | xclip -sel c

# Then paste your public SSH key that you have copied on your clipboard to the
# remote machine
$ vim .ssh/authorized_keys

# then, restart the ssh service:
$ service ssh restart

You should be able to SSH from your local machine as the admin_deploy user. Do so from now on, this is the user that will be used as our main user on the machine.

Security checkpoint: No change. We just created a new secure door to connect to the system. Only your local machine can use that door.

Securing the machine: Disable SSH for root

Now that we have a more secure door to go through (admin_deploy SSH), let's kill the one that let's you do anything on the machine (root SSH).

$ sudo vim /etc/ssh/sshd_config
# Find this line:
# PermitRootLogin yes
# change it to `no`, so it looks like this:
# PermitRootLogin no

# Then run
$ service ssh restart

Security checkpoint: Big change. Now you can no longer log in as root (the user that could do whatever it wanted in the machine) at all. You can still do whatever you want when logged in as admin_deploy because you have sudo priviledges, but you need to know the password you set for the user. We now have a couple layers of security.

Preparing the machine: Prepare the machine to run our Phoenix application

Now that we have basic security in place, let's prepare the machine so it has everything we need to run our Phoenix application. The only abstraction we are using here is the fact that we are using a runtime version manager for the languages we need (Erlang and Elixir) instead of compiling and bulding the code locally. We need the following on the machine to run the Phoenix application I had built:

And since we want to use nginx to reverse proxy our traffic, we need to install it as well.

I've put together this script because I set up a few Ubuntu machines for development and it gets us everything we need to be able to write, build and deploy our app. Feel free to manually download the above, or you can copy some parts of the script as you'd like.

Preparing the machine: Install and configure Nginx

Now that we have the tools we need installed on the machine, let's configure a great tool to handle the network traffic that comes to the machine: nginx. We don't need this, we could have set up things differently, but, I wanted this additional layer of control for this exercise.

Here is a config file that fits the purpose of this exercise. We listen to port 80 on the machine and we redirect accordingly to the port the Phoenix application is will be listening (the default for Phoenix is 4000). We also handle socket connections (therefore liveview) correctly. For brevity, I will use the fact that I cloned the repo where I have these utility scripts (dotfiles) and copy the config file to the appropriate location.

$ sudo rm /etc/nginx/sites-available/default
$ sudo cp dotfiles/available-sites-example /etc/nginx/sites-available/default
$ sudo cat /etc/nginx/sites-available/default
$ sudo nginx -s reload

You can also manually copy/edit the file (/etc/nginx/sites-available/default) with the following:

server {
  listen 80;
  server_name phoenix 0.0.0.0;

  location / {
    proxy_set_header X-Real-IP $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header Host $http_host;
    proxy_redirect off;
    proxy_pass http://localhost:4000;
    proxy_set_header Upgrade $http_upgrade;
    proxy_set_header Connection "upgrade";
  }

  location /live {
    proxy_pass http://localhost:4000$request_uri;
    proxy_http_version 1.1;
    proxy_set_header   Upgrade $http_upgrade;
    proxy_set_header   Connection "Upgrade";
    proxy_set_header   Host $host;
  }

  location /socket {
    proxy_pass http://localhost:4000$request_uri;
    proxy_http_version 1.1;
    proxy_set_header   Upgrade $http_upgrade;
    proxy_set_header   Connection "Upgrade";
    proxy_set_header   Host $host;
  }
}

The user that nginx creates and uses when creating OS processes for each of the reverse proxy requests does not have sudo privileges. You can check which user nginx will use when creating OS process by:

# Verify the default user, created by a default nginx install
$ cat /etc/nginx/nginx.conf | ag "user " -Q
user www-data;

And verify it does not have sudo privileges:

$ sudo -l -U www-data
User www-data is not allowed to run sudo on ubuntu-s-1vcpu-2gb-amd-nyc3-01.

Security checkpoint: Our nginx reverse proxy is configured and routes all network traffic to port 4000 (where the Phoenix application will use). The OS processes that handle the requests are created by the www-data user, who does not have sudo privileges.

Preparing the machine: Configure PostgreSQL

PostgreSQL configures a user (called postgres) during installation and all PostgreSQL OS processes run under its ownership. This user doesn't have sudo privileges. Since processes will be run under the postgres user, PostgreSQL creates a user inside PostgreSQL itself, called postgres :) Let's set the password for the postgres user inside PostgreSQL itself to something we know and can refer to.

After writing this section, I realized how similar the names are and how that can cause confusion. Definitely annoying for folks starting out. Hope it is clear enough.

# Let's change to the postgres user
$ sudo su postgres

# Let's change the password for the postgres user to... drumroll... postgres
$ psql -c "ALTER USER postgres WITH PASSWORD 'postgres';"

# now you exited the psql shell and have a shell under the postgres user, which
# we want to exit back to admin_deploy
$ exit

# we now have a shell under admin_deploy, you can verify with:
$ whoami
admin_deploy

Security checkpoint: We could have set up the password to be something else. But, if an attacker had access to our machine under the admin_user they could change the password for the database anyways since they could `sudo su postgres` and change the password themselves. Maybe there is room for improvement in this step, I will update the post as I learn more about it.

Preparing the machine: Install and configure Fail2ban

I came across a good guide on how to install and configure Fail2ban on the machine. I subscribed to their newsletter, let's hope it is good :). They give great details about fail2ban, but the main parts are:

# Install it on the machine:
$ sudo apt install fail2ban

# Copy the default files to avoid issues when there are updates.
$ sudo cp /etc/fail2ban/fail2ban.conf /etc/fail2ban/fail2ban.local
$ sudo cp /etc/fail2ban/jail.conf /etc/fail2ban/jail.local

# Let's edit the /etc/fail2ban/jail.local file so we can make the "maxretry"
# variable smaller, meaning we immediately go for the ban if we have a bad ssh
# attempt
$ sudo vim /etc/fail2ban/jail.local

# and change maxretry in the Default section, from 5, to 1. Save and close the file.
maxretry = 1

# And finally, let's turn it on:
$ sudo systemctl start fail2ban
$ sudo systemctl enable fail2ban

Security checkpoint: Fail2ban will help deter bot activity that keep probing the ssh port.

After monitoring some of the incoming http requests, you see a pattern where bots attempt known vulnerabilities in other technology (wordpress, php websites) and keep issuing those bogus requests to our machine. Currently, we are good ssh wise, but let's make it so we are good on the http side as well. At least, better than we were before. Fail2ban comes with a pre-set "jail" called `[nginx-http-auth]` to handle bad nginx login attempts. We will enable the existing one and add a couple more specific jails and filters for these jails. We will use the strategies detailed here: here, an excellent article by Digital Ocean. For brevity, I'm going to copy over (from the dotfiles repo) the filter files that fail2ban will use after we the directives below:

# Open the file with an editor:
$ sudo vim /etc/fail2ban/jail.local

# Find the [nginx-http-auth] directive and set it to enabled, add the port and the logpath:
[nginx-http-auth]

enabled  = true
filter   = nginx-http-auth
port     = http,https
logpath  = /var/log/nginx/error.log

# Add these other 4 directives:
[nginx-noscript]

enabled  = true
port     = http,https
filter   = nginx-noscript
logpath  = /var/log/nginx/access.log
maxretry = 1

[nginx-badbots]

enabled  = true
port     = http,https
filter   = nginx-badbots
logpath  = /var/log/nginx/access.log
maxretry = 1

[nginx-nohome]

enabled  = true
port     = http,https
filter   = nginx-nohome
logpath  = /var/log/nginx/access.log
maxretry = 1

[nginx-noproxy]

enabled  = true
port     = http,https
filter   = nginx-noproxy
logpath  = /var/log/nginx/access.log
maxretry = 1

Now we add the filters for these jails. As I said before, I'm just moving them from my dotfiles. Please inspect them before doing this.

# Copy an existing bot filter file:
$ sudo cp /etc/fail2ban/filter.d/apache-badbots.conf /etc/fail2ban/filter.d/nginx-badbots.conf

# Copy over the filters for each directive
$ sudo cp dotfiles/fail2ban_nginx-http-auth.conf /etc/fail2ban/filter.d/nginx-http-auth.conf
$ sudo cp dotfiles/fail2ban_nginx-noscript.conf /etc/fail2ban/filter.d/nginx-noscript.conf
$ sudo cp dotfiles/fail2ban_nginx-nohome.conf /etc/fail2ban/filter.d/nginx-nohome.conf
$ sudo cp dotfiles/fail2ban_nginx-noproxy.conf /etc/fail2ban/filter.d/nginx-noproxy.conf

# And finally, restart the service:
$ sudo systemctl restart fail2ban

# you can confirm the fail2ban status with:
$ sudo systemctl status fail2ban

# and confirm a specific fail2ban client:
$ sudo fail2ban-client status nginx-nohome

Security checkpoint: Fail2ban will also refuse requests from malicious users trying sneaky things with http requests. Our fail2ban log will hopefully have a lot of bans now.

Preparing the machine: Create a user that will be responsible for running the Phoenix application, without sudo privileges

We should create a user responsible for running the Phoenix application. This user should not have sudo privileges. We will use this new user to run the Phoenix application later on.

# Let's add the app user
$ sudo adduser app
# And set a password for the app user

# We can confirm the app user doesn't have sudo privileges with:
$ sudo -l -U app
[sudo] password for admin_deploy: # enter password
User app is not allowed to run sudo on ubuntu-s-1vcpu-2gb-amd-nyc3-01.

Security checkpoint: We created a user so that even if a malicious attacker somehow finds a way to execute remote commands through the Phoenix application (who knows, maybe the application runs commands directly on the OS) the user that would run those commands would not have sudo privileges.

Preparing the deploy: Create a deploy user SSH key for this machine to be used with Github/Gitlab/etc

In this case, we use Github and the repo is private. A good solution is to create an SSH key on the machine and configure a deploy token. With it in place, we can clone the repo to the machine in a safe and secure way. Follow the steps the link to set that up, select "Read only" when adding the key to the repo. Note that the SSH key should be created for the admin_deploy user, the one that still has sudo privileges.

Clone the repo to the remote machine. If the key was set up correctly, you should have the source code now.

Security checkpoint: If someone is able to get your private SSH key on your local machine and they can log into this remote machine and would also be able to access to the repo you just cloned. Honestly, if your private SSH key is compromised, I'd bet this is the least of your worries, but I thought I should mention it. Hardly no change to security in this step.

Preparing the deploy: Plumbing now that you have the source code

Now that we have the source code, we can get ready to run the server in "production". Due to our security focus, we want to run the server as a user that does not have sudo access (the app user). Elixir helps us in that regard by providing us an abstraction called releases, where we create a self-contained package with all of the code we need. So, we will create a release for the application and have the app user run it in the end.

Let's run the following to prepare the project for the release:

# Install Phoenix
$ mix archive.install hex phx_new 1.5.8

#
# `cd` to the project's directory, and from there, continue below
#

# Fetch dependencies
$ mix deps.get

# Prepare the assets to be served and create the correct config files.
$ mix assets.setup && mix assets.build

# Create a secret that Phoenix will use to sign and encrypt data
# Copy the output when you get it.
$ mix phx.gen.secret

Preparing the deploy: Let's create some files to help with the process

Let's do the following:
  1. Create a `.env` file
  2. Create a `build_release.sh` file
  3. Create an `entrypoint.sh` file
  4. Create a log file to hold the Phoenix application's logs
#
# As admin_deploy, `cd` to the project's root directory
#

# 1) Create a `.env` file with the following:
DATABASE_URL=ecto://postgres:postgres@localhost/database
PHX_SERVER=true
SECRET_KEY_BASE=the_secret_you_copied_above


# 2) Create a `build_release.sh` file and add this to the file:

#!/bin/bash
echo "Building Phoenix release"
git pull origin main
export $(cat .env | xargs)
mix deps.get --only prod
MIX_ENV=prod mix ecto.migrate
MIX_ENV=prod mix compile
MIX_ENV=prod mix assets.build
MIX_ENV=prod mix assets.deploy
MIX_ENV=prod mix release


# make it executable
$ sudo chmod +x build_release.sh

# 3) Create script to run the server as the app user, `entrypoint.sh and `add
this to the file. Make sure to replace `your_app_name` with... your app name.

#!/bin/bash
echo "Starting the Phoenix server as $(whoami)"
cd ~
_build/prod/rel/your_app_name/bin/your_app_name start >> logs/log.log


# make it executable
$ sudo chmod +x entrypoint.sh

# 4) Set up logs in the app user directory
$ sudo mkdir /home/app/logs
$ sudo touch /home/app/logs/log.log
$ sudo chown app /home/app/logs/log.log

If you don't have the code prepared for Phoenix releases, the only change you need to make is the following, documented here: you need to add `server: true` to the Endpoint configuration in `config/prod.secret.exs`. Also, since we have a domain, plug it into `config/prod.exs` instead of `example.com`. Now that this is done, let's continue:

Preparing the deploy: Let's deploy!

#
# As admin_deploy, `cd` to the project's root directory
#

# Export env vars
$ export $(cat .env | xargs)

# Create the database. We will only do this one time, so we are doing it here:
$ MIX_ENV=prod mix ecto.create

# Set up tailwind and esbuild
$ mix assets.setup

# Build the release
$ ./build_release.sh

# Copy release to /home/app/_build so we can run it as the app user
$ sudo cp -r _build/ /home/app/_build/

# And give access to app:
$ sudo chown app /home/app/_build/

#
# And finally:
#

# Run the entrypoint file, which runs the server as the app user, outputting to
# the log file. From the root of the project, as admin_deploy, run:
$ sudo -u app "./entrypoint.sh"

Security checkpoint: Note how we run the Phoenix server as the app user, a user without sudo privileges. You can verify that the Phoenix application is run as the app user by doing `ps aux | ag beam.smp` and you will see the BEAM process under the app user's ownership.

Post deploy: Basic monitoring

Now that the Phoenix application is deployed, we can monitor some interesting things.

# Monitor nginx traffic logs
$ sudo tail -f /var/log/nginx/access.log

# Monitor the Phoenix application's logs
$ sudo tail -f /home/app/logs/log.log

# Monitor authentication attempts on the machine
$ sudo tail -f /var/log/auth.log

# Monitor Fail2ban logs
$ sudo tail -f /var/log/fail2ban.log

# IEx shell on the release, from the project's root:
_build/prod/rel/your_app_name/bin/your_app_name remote

# Run htop to see how the machine is doing
$ htop

I use tmux and have the above running on different windows. There is a lot, a lot of bot activity probing the machine. This likely is a blog post in itself, where I show some of the most common attacks that one receives after putting a machine online. The types of attacks the bots attempt make me wonder if folks even try to secure a machine. My new hobby is tailing the fail2ban logs and watching bots get banned.

Post deploy: Future deployments

We would use the `build_release.sh` script and copy over the release to the app user directory. Should be good enough for now.

$ A one liner would look something like this, from the project's root::
sudo rm -rf /home/app/_build/ && ./build_release.sh && sudo cp -r _build/ /home/app/_build/

More security: Restrict the app user so all it knows is how to run the release

Let's attempt to make the app user even less capable. In this case, let's give it only what it needs to run the release, nothing else. Here is some more info on how to restrict users and attempt to restrict their experience. The overall idea is to only list commands that you'd like the user to issue and restrict everything else. So, we create a ".bashrc" file for this purpose. I have it here in case you'd like to copy it directly, or, you can create one with these contents however you'd like. I prefer to pull the dotfiles directory and do a `cp`. Here are the file contents for the `.bashrc`:

alias cd="printf 'not today \n'"
alias exit="printf 'not today \n'"

PATH=$HOME/bin
export PATH

If you cloned the dotfiles repository, do this:

$ sudo cp dotfiles/.deploy_user_bashrc /home/app/.bashrc

# make it so `app` can't edit the file
$ sudo chattr +i /home/app/.bashrc

With the file created, all new shells for the app user will use the special `.bashrc` file and therefore do very little.

# Make user use restricted bash
$ sudo chsh -s /bin/rbash app

# Create a bin directory for the app user
$ sudo mkdir /home/app/bin
$ sudo chmod 755 /home/app/bin

# We need to symlink only the tools the user needs to run the release I
# listed them here:
# https://github.com/pdgonzalez872/dotfiles/blob/master/ln_for_app_user.sh
# If you have the dotfiles repo locally, you can run `./ln_for_app_user.sh`.
# If not, do the below:

# We allow this to print the current user when running the release. Not needed
# to run the release per say.
$ sudo ln -s /usr/bin/whoami /home/app/bin/

# These are needed to run the release:
$ sudo ln -s /usr/bin/readlink /home/app/bin
$ sudo ln -s /usr/bin/dirname /home/app/bin
$ sudo ln -s /usr/bin/cut /home/app/bin
$ sudo ln -s /usr/bin/sed /home/app/bin
$ sudo ln -s /usr/bin/cat /home/app/bin
$ sudo ln -s /usr/bin/grep /home/app/bin
$ sudo ln -s /usr/bin/basename /home/app/bin

Security checkpoint: The app user shell is very limited, only has enough to run the release.

More security: let's lock some known users

If an attacker gets shell access in our system, they could potentially attempt to brute force other accounts using `su postgres` for example. For that reason, let's lock some of the users we dealt with in this article, so there is no possible way of using password logins within the machine.

I chose to keep the admin_deploy user unlocked meaning you still need a password to get in and if you know the password, it would let you in. Unlike the locked option, where even with the correct passwod, it would not let you in. Tradeoffs. I preferred to not have a passwordless sudo user in the system.

$ sudo /usr/bin/passwd --lock app
$ sudo /usr/bin/passwd --lock root
$ sudo /usr/bin/passwd --lock postgres
$ sudo /usr/bin/passwd --lock www-data

Security checkpoint: One can't log into the accounts above. An attacker could still attempt to brute force the admin_deploy user password if they get access to a shell that is not restricted like the app user is. I don't know how one would do this, but I'm not a hacker and there are pros out there. Everything is possible. :)

Conclusion

After all of this, your application should be up, with all that it needs to work in a fairly safe environment. With that said, I'd choose a provider for this type of work (such as Heroku) if I were to deliver a production ready project. There are a few others you can choose from after doing a little bit of searching online. There are guides to deploy to them on the Phoenix Guides page.

General Drawbacks

As I mentioned in the monitoring section, there is a lot of bot activity when you are creating a machine that can be accessible to the whole internet. I'll write a blog post sometime analyzing the type of activity we get just by putting up a machine online. It's a lot. Folks will try to access the machine directly with different usernames/passwords (our machine is not accessible through passwords, so this doesn't work), send malicious http requests to the machine attempting to gain useful info, among other interesting things. Fail2ban helps with this issue. I still think it would be interesting to go through this article again, skip Fail2ban protection, stand a hello world Phoenix app, wait for a few days to collect data and then analyze that data. Maybe for a future blog post.

Special thanks to my friends that have helped with this post in some fashion:

--------------

Thanks for reading, PDG

Home