sr229 / wsl-systemd

A script that spawns a namespace with systemd as PID1, and allows you to spawn interactive sessions or exec commands inside it

systemd Helper Script for WSL2

This is a script designed to allow you to use systemd in WSL2 by faking PID1 using namespaces - aka "the poor man's container".

The main motivation behind this is you can run systemctl or your usual systemd stuff in WSL2 - all without waiting for Microsoft to wait for the official integration for custom inits (for now).

Why Another?

Most of the approaches done is more complicated to set up. This was created based on my frustration on the existing approaches:

• Current Existing Approach does not preserve PWD when walking to the namespace. the Snaps guide is a example of this, while that guide was the basis of this project, I took the liberties to improve the script so it works seamlessly with WSL2.

• Almost zero dependency - All you need is util-linux, nsenter and a sane POSIX-compatible shell. Approaches like genie requires you to install .NET Core just to do this which is rather - bloaty. You're just spawning a shared namespace, you don't need to write it in C#.

Installing

Simply download this script and add it on your /usr/bin or wherever you like.

However, to make sure this script truly works, make sure you have the following dependencies:

• util-linux (comes standard in all linux distros)
• nsenter (comes standard in all linux distros)

If you have them all, congrats! You're ready to use this.

Usage

The standard way

Usage: ./init-systemd -u (USER) -c (command)
This script spawns a namespace to be able to use systemd in a specified user.
This requires unshare (from util-linux).
If -c is provided, the command will be executed in the namespace's context non-interactively.

Execute during login

It's very much possible to execute this during login. However, I am not responsible if you can't log in normally in your system if something breaks.

Setting the sudoers

First of all is setting the sudoers so it can be executed by root without any hitches.

# Change this to where your init-systemd is.
%sudo ALL=(ALL) NOPASSWD: /usr/sbin/init-systemd

Then finally, let's edit /etc/bash.bashrc

/usr/sbin/init-systemd -u 1000

Now your system should enter the namespace whenever a interactive session is opened.

Limitations

• Currently we can't kill the namespaces properly, so your ps tree might show multiple instances of unshare. That's going to be fixed soon once I figured out how to share only ONE systemd namespace instead of spawning another.

• namespaces can't be killed for some reason. That's under investigation since even SIGKILL does not terminate a unshare namespace.

Contrbuting

There's more laywork to be done so it will be desirable for everyone, but in my own current usage, I say its ready for my own intended needs.

If you think you can make this script better please send patches via PR, or open some issues so I'm aware of the outstanding problems.

The entire script is written to be POSIX as possible. Bash-specific syntax should be prevented at all costs.

If you want to support me financially, my Ko-fi is available for you to donate to.

Copyright

This script is Copyright 2020 © Ayane "Capuccino" Satomi. Licensed under MIT.

This is inspired of this article from Snapcraft.io.

wsl-systemd's Issues

Error on the first run

When I run init-systemd for the first time when I "booted" into WSL2, it will throw an error nsenter: failed to parse PID: ''. If I run it again, systemd will be set up correctly.

Extra Information
lsb_release -sa output: n/a AOSC OS "AOSC OS" 7.1.1 Gumblex

Optimizations

Hi,

I wonder if you can replace the dependency on deamonize by running the command with stdin closed and backgrounded: ...systemd... <&- &

I believe bash will behave similiar to daemonize then.

You could also combine the grep+awk into just awk with '/regex/ && !/regex/ {print $2}'

On mobile, sorry