computer bullshit

How I Set Up My Server (Outdated)

EDIT : I no longer use this setup to run my website. Instead, I'm using Bear. I want to keep this since it could be helpful to others who want to set up their site this way, or myself if I want to run a server with a similar setup again.

This is less of a blog post and more of a compilation of my notes for setting up my web server. Feel free to follow along or try this yourself, but I won't go into too much detail unless it will help me remember later. Additionally, this guide assumes you're running on Debian 11, so adjust any of the commands (mainly package manager ones) accordingly. Before we get started, we need to:

  1. Buy a domain
  2. Rent a VPS somewhere (I run this on Linode for $5 per month)

Secure Server with Sane Defaults

Before we can start serving content, we need to set up sane user permissions. We don't want anyone to log in as root (including us) via SSH, so we need to make our own account first. All of these tasks come from Linode's setup documentation.

# on local machine
ssh root@[host]
# on remote server, as root
apt update && apt upgrade
dpkg-reconfigure tzdata
hostnamectl set-hostname [name]
useradd -m [user] sudo --password [password] --shell bash

First, we update all system dependencies to their latest version (say, if we were using an older install) via apt update && apt upgrade. Then, we configure other system information like the timezone (dpkg-reconfigure tzdata)1 and the host name (hostnamectl set-hostname [name])2. Finally, we create a new user account that we'll use as our primary login account. No more running or logging as root.

Next, we'll copy our SSH public key (here are instructions for setting one up):

# on your local machine
ssh-copy-id [user]@[host]
# this will prompt you for your password

Then you're done! Log out of root and log back in using your newly-created account.

Configure SSH and Firewall

With our own account made, we can restrict root access and set up a firewall. As root, set the following values in /etc/sshd.config:

PasswordAuthentication no
PermitRootLogin no
AddressFamily inet

PasswordAuthentication no forces the use of SSH keys for authentication, PermitRootLogin no prevents someone from running ssh root@[host], and AddressFamily inet forces SSH to be over IPv43. Check out the sshd config man pages for more information.

Restart SSH to have it reflect your new changes via:

sudo systemctl restart sshd

ufw

ufw (uncomplicated firewall) is the firewall tool we'll be using. We'll install it and configure it to be only as open as we need it:

sudo apt-get install ufw
sudo ufw default allow outgoing
sudo ufw allow ssh
sudo ufw default deny incoming
sudo ufw enable
sudo ufw logging on

At least via TCP, ufw allows us to choose whether to block incoming or outgoing ports. We need SSH on (which listens on port 22), so we allow that. Any other incoming connections are blocked. Then, we have to enable the firewall, and optionally enable logging. Since this will be a web server, we'll need to open up some more ports, but we'll do that later.

fail2ban

TODO: write about fail2ban

Enable Git Access via SSH

My server also houses my Git repositories, which has push/pull access via SSH. Roughly, this advice follows the Git book's advice for Git Servers. We'll create a user to do all Git operations, then create a folder for the repositories to live:

# in case git isn't installed already (wasn't for me)
sudo apt-get install git
sudo useradd -m git
sudo mkdir /srv/git
sudo chown git:git /srv/git
sudo chmod a-rwx /srv/git
sudo chmod u+rwx /srv/git
sudo chmod g+rx /srv/git

Next, we need to allow ourselves to use the git user via SSH, which we'll do by copying our public key from our account into git user's account:

cp ~/.ssh/authorized_keys /home/git/.ssh/authorized_keys

If anyone else ever needs Git access, add their public key to this file.

Lock Down Git User

We have our git user set up, but anyone who has SSH access to it can run an interactive session under that account. There's no reason for this, so we'll prevent it by setting the git user's shell to git-shell:

# Check if git-shell is already configured
cat /etc/shells
# If it isn't run these commands:
which git-shell
sudo -e /etc/shells

Then, edit /etc/passwd and change the git user's shell to the output of which git-shell.

(Optional) Chroot the Git User

This isn't required (and I may remove this from my setup at some point), but I chroot'd the git user so it only has access to the Git repositories. Other users can reach in, but git can't reach out. I won't go into detail, but here's some more information if you're interested. While using git-shell prevents users from doing anything interactively, I wanted to prevent someone from getting access to other parts of the file system via a git clone command. This could probably be accomplished through stricter permissions.

Creating a Repository

mkdir /srv/git/my-repo.git
cd /srv/git/my-repo.git
git init --bare

And that's it! You can clone it like:

git clone git@[host]:/srv/git/my-repo.git

Install cgit

cgit is a web UI for Git repositories written in C. There's nothing fancy about it: it's just a way to make navigating repos a little bit nicer. Yes, GitHub does that, and there are other tools that are more conducive for collaboration and can be self-hosted (Gitea, Gogs, and GitLab to name a few), but I wanted something that:

There are some tweaks I'd like to make to my cgit instance to make it nicer looking, but that's besides the point. Let's set it up!

Installation

sudo apt install cgit
# Used for syntax highlighting
sudo apt install highlight

This creates a couple directories and files:

However, we're going to move these to an easier location to manage:

sudo mkdir -p /var/www/src.freedman.dev
sudo ln -s /usr/lib/cgit/cgit.cgi /var/www/src.freedman.dev/cgit.cgi
sudo mv /usr/share/webapps/cgit/* /var/www/src.freedman.dev
sudo chown -R www-data:www-data /var/www/src.freedman.dev

Global Configuration

First, let's modify /etc/cgitrc (for more info on each config option, the man page on the Arch Linux wiki is pretty good):

# Stylistic options
css=/cgit.css
logo=/cgit.png
virtual-root=/
root-title=Josh Freedman's Git Repositories
root-desc=My personal projects, probably unfinished or half-baked
root-readme=/srv/git/README.html
head-include=/srv/git/head-include.html

# Global Repository options
readme=trunk:README.html
readme=main:README.html
readme=master:README.html
readme=trunk:readme.html
readme=main:readme.html
readme=master:readme.html
enable-index-owner=1
source-filter=/usr/lib/cgit/filters/syntax-highlighting.sh
scan-path=/srv/git

Create /srv/git/head-include.html and put this content into it:

<meta name="og:site_name" content="Josh Freedman's Git Repositories" />
<meta name="og:image" content="https://src.freedman.dev/favicon.ico" />

Create a /srv/git/README.html file and put whatever in it, it will get displayed on the "About" page from the home page.

Repository-Specific Configuration

We've configured cgit globally, but repositories need a cgitrc specific to them to store any repo-specific data. We could manually add a cgitrc to the repository on the server, or we can add it to the template that Git uses to create the repository. Then, we'll create a little script that makes this process easier.

First, create a file at /usr/share/git-core/templates/cgitrc:

url=$REPO_FILE_NAME
name=$REPO_NAME
desc=$REPO_DESC
extra-head-content=<meta name="og:title" content="$REPO_NAME" /><meta name="og:description" content="$REPO_DESC" /><meta name="og:url" content="https://src.freedman.dev/$REPO_FILE_NAME" />
owner=Josh Freedman
image=/favicon.ico

Note that there's environment-variable syntax in there. This is because we'll expand them to the repo-specific values in our script:

#!/bin/bash

# RUN THIS AS ROOT!
# Helpful command: sudo -u root [path to this script]

REPOS_ROOT=/srv/git

# Read in values from stdin
read -p "File Name (.git will be added automatically): " REPO_FILE_NAME
REPO_FILE_NAME=$REPO_FILE_NAME.git
read -p "Name: " REPO_NAME
read -p "Description: " REPO_DESC

FULL_REPO_PATH="$REPOS_ROOT/$REPO_FILE_NAME"

# make them available to envsubst
export REPO_FILE_NAME
export REPO_NAME
export REPO_DESC

# Create the repository from the template
mkdir $FULL_REPO_PATH
cd $FULL_REPO_PATH 
git init --bare

# Expand environment variables in the default cgitrc
cat "$FULL_REPO_PATH/cgitrc" | envsubst > "$FULL_REPO_PATH/cgitrc_expanded"
rm "$FULL_REPO_PATH/cgitrc"
mv "$FULL_REPO_PATH/cgitrc_expanded" "$FULL_REPO_PATH/cgitrc"

# Set permissions appropriately
chown -R git:git "$FULL_REPO_PATH"
chmod g+rx -R "$FULL_REPO_PATH"

Set Up Personal Website

These are just static files in the /var/www/freedman.dev folder. To set this up:

sudo mkdir -p /var/www/freedman.dev
sudo chown -R www-data:www-data /var/www/freedman.dev

Then, copy the files to serve into this folder.

Set Up Web Server

Finally, the main event: the web server. This will route all requests made by a browser to the appropriate place. For my server, I want two sites:

  1. My personal site/blog: at freedman.dev (and www.freedman.dev)
  2. My Git repositories via cgit: at src.freedman.dev

My personal site will be entirely static files, and my Git repos will be served by cgit (which Apache runs under CGI). Let's get started!

DNS Configuration

DNS lets us use memorable names to refer to computers instead of IP addresses by creating A/AAAA records. These can be used for both the root domain and any subdomains. A records are for IPv4, and AAAA records are for IPv6. In your domain registrar, create the following records:

DNS Record Host Value
A freedman.dev IPv4
AAAA freedman.dev IPv6
A www.freedman.dev IPv4
AAAA www.freedman.dev IPv6
A src.freedman.dev IPv4
AAAA src.freedman.dev IPv6

Repeat this configuration in your cloud provider if necessary (I had to do so for Linode for some reason)4.

Apache

Let's start by installing Apache and opening up the firewall to the appropriate ports (80 for HTTP and 443 for HTTPS)5:

sudo apt install apache2
sudo ufw allow http
sudo ufw allow https

This creates a bunch of directories and files:

And a ton of other stuff too, but those are the big ones. Note that all of these are owned or ran by the www-data user in the www-data group. This will be important later.

Next, we'll set up some Virtual Hosts; they'll both go into /etc/apache2/sites-available. First will go under 001-mysites.conf:

<VirtualHost *:80>
    ServerName freedman.dev
    ServerAlias www.freedman.dev
    DocumentRoot /var/www/freedman.dev
    ErrorLog ${APACHE_LOG_DIR}/error.log
    CustomLog ${APACHE_LOG_DIR}/access.log combined
</VirtualHost>

And another under 002-src.conf:

<VirtualHost *:80>
    ServerName src.freedman.dev
    DocumentRoot /var/www/src.freedman.dev
    ErrorLog ${APACHE_LOG_DIR}/error.log
    CustomLog ${APACHE_LOG_DIR}/access.log combined
    
    <Directory "/var/www/src.freedman.dev">
        Options +ExecCGI +FollowSymLinks +SymLinksIfOwnerMatch
        AddHandler cgi-script .cgi
        DirectoryIndex cgit.cgi

        RewriteEngine on
        RewriteCond %{REQUEST_FILENAME} !-f
        RewriteCond %{REQUEST_FILENAME} !-d
        RewriteRule (.*) /cgit.cgi/$1 [END,QSA]

        RewriteCond %{QUERY_STRING} service=git-receive-pack
        RewriteRule .* - [END,F]
    </Directory>
    <Directory "/srv/git">
        Require all granted
    </Directory>
</VirtualHost>

Before going any further, let's break down a couple things here:

  1. No content is being served via HTTPS yet since we're only listening on port 80 (regular, unencrypted HTTP). We'll run certbot on these configurations and it will generate HTTPS configs from these ones.
  2. The cgit configuration is pulled straight from Andrew Marchetta's great blog post. Highly recommend checking it out!

Enable the Sites and Modules

# add the www-data user to the git group so it can
# read/execute the /srv/git folder
sudo usermod -a -G git www-data
# enable CGI module for cgit
sudo a2enmod cgi.load
# enable our website configurations
sudo a2ensite 001-mysites.conf
sudo a2ensite 002-src.conf
# alternatively, manually symlink the sites from /etc/apache2/[xyz]-available to [xyz]-enabled

HTTPS via Certbot

HTTPS is a must for all websites, even sites like these that have no authentication and no sensitive data being transmitted between the browser and my servers. Certbot is a free tool that will generate HTTPS certificates so browsers can trust us. The process for installation and setup goes as follows:

sudo apt install snapd
sudo snap install core
sudo snap refresh core
sudo snap install --classic certbot
sudo ln -s /snap/bin/certbot /usr/bin/certbot
sudo certbot --apache
# if all is successful, restart apache for the changes to take effect
sudo systemctl restart apache2

Cerbot should automatically pick up all the domains/subdomains in your Apache config and be able to make the modifications necessary.

Optional Next Steps

While this should result in a working web server, there are some potential areas of improvement:

  1. Instead of relying on manual configuration of the git user, use something like Gitolite.
  2. Set up continuous integration, either stitched together using post-receive hooks that trigger the at command, or through a real CI system.
  3. Make cgit's UI look nicer, namely around mobile-friendly UI and dark mode (the latter of which appears to be included in the latest commits of cgit).
  4. Look into managing environments with Nix. Maybe even run the server with NixOS and have all configuration stored as a file?

Thanks

I'd like to thank my friend Justin and my girlfriend Amaya6 for listening to me rant about the various issues I ran into. And if you made it all the way down here, thanks to you for reading!

  1. This was already the appropriate default, so this may not be necessary.

  2. I don't remember this working for me (not sure what the issue was), so I didn't run it. Linode suggests it, though.

  3. Might be a good idea to allow SSH over IPv6 too. I don't remember why I explicitly set this option.

  4. Google Domains shows the whole domain, but Linode only showed the subdomain part. So, the first two records would be blank, but then the last 2 would just contain "src".

  5. The "apache2" verbiage is Debian-specific. Some distros refer to it as "httpd". It's the same software with the same configuration, just different names.

  6. "I didn't even listen, you just talked at me for 20 minutes straight without letting me get a word in" -Amaya

#apache #cgit #linode #linux