Hi all,
I found a hobby in trying to secure my Linux server, maybe even beyond reasonable means.
Currently, my system is heavily locked down with user permissions. Every file has a group owner, and every server application has its own user. Each user will only have access to files it is explicitly added to.
My server is only accessible from LAN or VPN (though I’ve been interested in hosting publicly accessible stuff). I have TLS certs for most everything they can use it (albeit they’re self signed certs, which some people don’t like), and ssh is only via ssh keys that are passphrase protected.
What are some suggestions for things I can do to further improve my security? It doesn’t have to be super useful, as this is also fun for me.
Some things in mind:
- 2 factor auth for SSH (and maybe all shell sessions if I can)
- look into firejail, nsjail, etc.
- look into access control lists
- network namespace and vlan to prevent server applications from accessing the internal network when they don’t need to
- considering containerization, but so far, I find it not worth foregoing the benefits I get of a single package manager for the entire server
Other questions:
- Is there a way for me to be “notified” if shell access of any form is gained by someone? Or somehow block all shell access that is not 2FA’d?
- my system currently secures files on the device. But all applications can see all process PIDs. Do I need to protect against this?
threat model
- attacker gains shell access
- attacker influences server application to perform unauthorized actions
- not in my threat model: physical access
Maybe not 100% in the subject, but I just deployed a Wazuh instance to let me know how any of my hosts, containers and computers may have vulnerabilities. I found a crap load of holes in my services, and I’m halfway through squashing all of them.
If this is a hobby, that’s sure to keep you entertained for quite some time.
AppArmor or SELinux, OSSEC, TPM and SecureBoot boot chain.
Skip the TPM and secureboot as those are pretty useless
That’s patently false.
How so? They clearly say physical access is not in there threat model. If someone has root it is game over.
It can still prevent vectors of persistency.
I would reconsider docker because if a specific application leaks some sort of shell access or system file access you’ll be protected out side of container host escalation.
Unrelated to security, I prefer docker because it leaves the server very clean if you remove different apps. Can also save time configuring more complex applications or applications that conflict with system libraries.
Add fail2ban on your list of applications it watches logs for invalid logins and puts them on firewall block rules after so many failed attempts.
Docker performs some syscall filtering as well which may reduce the kernel attack surface. It can be pain to set up services this way, but it could help frustrate an attacker moving laterally in the system.
Processes in the container cannot see external processes for example as I think interested the OP.
I really wish there was a system wide package manager for docker containers, which would update software in all your containers at once similar to how a typical package manager would.
I did not completely rule out docker, but I wonder if I can obtain most of its benefits without this major con with package management. I mean I know it’s possible, since its mostly kernel features, but it would be difficult to simulate and the tooling is probably lacking (maybe nsjail can get me closer).
You can have a look at systemd-nspawn and machinectl actually. Sounds like exactly what you’re looking for :)
I am really interested in systemd-nspawn. Unfortunately I have openRC now (I liked it’s simplicity) so can’t try out systemd yet.
Is machinectl tied to systemd also?
You could give bubblewrap a try instead. It is quite similar to systemd-nspawn.
Yes machinectl is the interface for nspawn
Absolutely essential is using a firewall and set it as strict as possible. Use MAC like SELinux or Apparmor. This is extremely overkill for a personal server, but you may also compile everything yourself and enable as many hardening flags as possible and compile your own kernel with as many mitigations and hardening flags enabled (also stripped out of features you don’t need)
This is extremely overkill…
I actually do all of that, thanks to Gentoo :')
There are entire books dating back to the 80’s that go into this, that are still fairly valid to this day.
If you want to take things further at your own risk, look into how to use TPM and Secure Boot to your advantage. It’s tricky, but worth a delve.
For network security, you’re only going to be as effective as the attack hitting you, and self-hosting is not where you want to get tested. Cloudflare is a fine and cheap solution for that. VLANS won’t save you, and an on-prem attack won’t save you here. Look into Crowdsec.
Disable any wireless comms. Use your BIOS to ensure things like Bluetooth is disabled…you get the idea. Use RFKill to ensure the OS respects the disablement of wireless devices.
At the end of the day, every single OS in existence is only as secure as the attack vectors you allow it to have. Eventually, somebody can get in. Just removing the obvious entry points is the best you can do.
What’s the issue with VLAN?
VLANs are for organizing traffic, not authorization of traffic.
Can be pretty easily spoofed by packet.
Only if you don’t set it up correctly. You should set which devices are allowed to set which vlans and then make sure client devices aren’t authorized to send or receive tagged packets.
You then combine that with a firewall only needed traffic allowed.
If you set it up incorrectly you can perform an attack called vlan hoping.
You also need to setup Firewall rules to properly isolate zones
Cis level 2 hardening
-
SElinux
-
monitoring
-
proper containers (ideally rootless)
-
separate accounts for each function and permission set. Your containers should run as a low privileged user
-
Get your firewall right then maybe add fail2ban.
You could also consider IDs/IPs on your primary router/firewall if this is internal. If not you can install surricata on a public server. Obviously if you go with something as powerful as surricata you no longer need fail2ban.
Keep a sharp eye on any users with sudo. Beyond that consider docker as others have mentioned.
It does add to security because it allows the developers a bit more control of what packages are utilized for their applications. It creates a more predictable environment.
deleted by creator
That sounds extremely painful to manage and prone to error if you aren’t using containers.
It does require some effort to manage, but I would argue it’s easier to keep all packages (including dependencies) up-to-date across the system, which is a huge security benefit imo.
The permission system, once you set it up, you never need to change it unless you’re changing something.
I am not sure what you are talking about. None of the stuff OP talked about are related to containers. Also containers complicate networking a lot, so i would avoid them at all costs and use VMs
Great that you included your threat model, but you should have specified the type of services that you host/provide.
One thing i would look into is disabling any port that is not necessary (like 80 and 443) and disable ssh on the wider network.
Host a wireguard endpoint in the internal network that acts like a bastion and allows you to ssh-jump to any other host and VM on the network.
Wireguard is more secure than ssh, assuming sound crypto and hygiene for both, because you can’t probe a host from the outside and know if wireguard is running or not
