I Switched My Operations to Caddy Web Server

Good Morning from my Robotics Lab! This is Shadow_8472, and today I am rebuilding my home server, Button Mash, from the operating system up. Let’s get started!

Caddy Over Nginx

I spent well over a month obsessing over Nginx, so why would I start over now? As far as I am concerned, Caddy is the piece of software for my use case. While I am sure Nginx is great at what it does, I keep slamming into its learning curve – especially with integrating Let’s Encrypt, a tool for automating SSL encryption (HTTPS/the green padlock). Caddy builds that functionality in while still doing everything I wanted Nginx for.

The official Caddy install instructions [1] for Fedora, Red Hat, and CentOS systems are as follows:

$ dnf install ‘dnf-command(copr)’
$ dnf copr enable @caddy/caddy
$ dnf install caddy

First of all, new command: copr. Background research time! COPR (Cool Other Package Repositories) is a Fedora project I feel comfortable comparing to the Arch User Repository (AUR) or Personal Package Archive (PPA): it lets users make their own software repositories.

Installation went smoothly. When I enabled the repository, I had to accept a GPG key that wasn’t mentioned in the instructions at all. From a user point of view, they appear to fill a similar purpose here to a SSH keys: special numbers use math to prove you are still you in case you get lost.

Caddy uses an HTML interface (a REST API – Don’t ask, I don’t understand myself) on the computer’s internal network known as loopback or localhost on port 2019. Caddy additionally serves everything over HTTPS by default. If it cannot convince Let’s Encrypt to give it a security certificate, it will sign one itself and tell the operating system to trust it. In other words, if I were not running ButtonMash headless (without a graphical interface), I’d be able to try connecting to localhost:2019 with a favorite browser, like at least one of the limited supply of Caddy tutorials did.

IP Range Transplant

I should have just done my experimentation on DerpyChips or something. Instead, I pressed on with trying to point a family-owned domain name at Button Mash. This side adventure sprouted into last week’s post. In short: ButtonMash’s static IP kept was in conflict with what my ISP-provided equipment kept trying to assign it, resulting in an estimated 50% downtime from a confused router. Upgrading to the next gateway may have allowed us to free up the IP range for the gaming router’s use, but it’s not out for our area yet. My father and I switched our network connections over to a “gaming router” we had laying about and enabled bridge mode on the gateway to supposedly disable its router part. I have my doubts about how it’s actually implemented.

Most of our computers gladly accepted the new IP range, but GoldenOakLibry and ButtonMash –having static IP’s– were holdouts. I temporarily reactivated a few lines of configuration on my laptop to set a static IP so I could talk with them directly and manually transfer them over to the new IP range, breaking NFS shares and Vaultwarden on them respectively.

In the confusion, ButtonMash lost its DNS settings; those were easy enough to fix by copying a config line to point those requests to the router. GoldenOakLibry took a bit longer to figure out because the NFS shares themselves had to accept traffic from the new IP range with settings buried deep within the web interface. Once that was sorted, I had to adjust the .mount files in or around /etc/systemd/system on several computers. Editing note: While trying to upload, I found I could not access GoldenOakLibry on at least a couple of my machines. Note 2: I had to change the DHCP settings to the new IP range on my Raspberry Pi reverse Wi-Fi router. Systemd on both goofed systems needed a “swift kick” to fix them.

sudo systemctl start <mount-path-with-hyphens>.mount

Repairs Incomplete

That left Vaultwarden. I was already in a it’s-broken:-fix-it-properly mentality from the modem/router spinoff project. I got as far as briefly forwarding the needed ports for an incompletely configured Caddy to respond with an error message before deciding I wanted to ensure Bitwarden was locked down tightly before exposing it to the Internet. That wasn’t happening without learning Caddy’s reverse proxy, as I put Vaultwarden exclusively onto a loopback port.

Speaking of loopback, I found the official Caddy tutorials lacking. They –like many others after them– never consider a pupil with a headless server. I have not yet figured out how to properly convince my other computers to trust Caddy’s self-signed certificates and open up the administration endpoint. That will come in another post. I did get it to serve stuff over HTTP by listing IP’s as http://<LAN address>, but Bitwarden/Vaultwarden won’t let me log in on plain HTTP, even over a trusted network and confine the annoying log to a file.

As far as I can tell, the administration API on port 2019 does not serve a normal web page. Despite my efforts, the most access I have gotten to it was for it to error out with “host not allowed.” I haven’t made total sense of it yet. I recognize some of the jargon being used, but its exact mechanics are beyond me for the time being.

Takeaway

Caddy is a powerful tool. The documentation is aesthetically presented and easy enough to understand if you aren’t skimming. But you will have a much better time teaching yourself when you aren’t trying to learn it over the network like I did.

Final Question

Do you know Caddy? I can tell I’m close, but I can’t know for sure if I’m there in terms of the HTTP API and just don’t recognize it yet. I look forward hearing from you in the comments below or on my Discord server.

Works Cited

[1] “Install,” Caddy Documentation. [Online], Available: https://caddyserver.com/docs/install. [Accessed: June 6, 2022].

Leave a Reply