socker: A wrapper for secure running of Docker containers on Slurm

Introduction

Socker is secure for enabling unprivileged users to run Docker containers. It mainly does two things:

• It enforces running containers within as the user not as root
• When it is called inside a Slurm job, it enforces the inclusion of containers in the cgroups assigned by Slurm to the parent jobs

Design

Socker is composed of one binary that is a compiled python script. While system administrators can be members of the docker group, regular users can use Docker via Socker.

Usage

• Install Nuitka with it's prerequisites (python and gcc)
• Compile socker:

nuitka --recurse-on socker.py

• Change the owner of the binary to root and enable SUID:

mv socker.exe socker
sudo chown 0:0 socker
sudo chmod +s socker

• Create a list of authorized images as root (you need to fix the path to the images file in socker before compiting):

sudo vim socker-images

• Options:

socker --help 

NAME
	socker - Secure runner for Docker containers

SYNOPSIS
	socker run <docker-image> <command>

OPTIONS
	--version
		show the version number and exit
	-h, --help
		show this help message and exit
	-v, --verbose
		run in verbose mode
	images
		List the authorized Docker images (found in socker-images)
	run IMAGE COMMAND
		start a container from IMAGE executing COMMAND as the user

EXAMPLES
	List available images
		$ socker images
	Run a CentOS container and print the system release
		$ socker run centos cat /etc/system-release
	Run the previous command in verbose mode
		$ socker -v run centos cat /etc/system-release

SUPPORT
	Contact [email protected]

Prerequisites

• Docker 1.6+
• You MUST have a group docker and a user dockerroot who is member of ONLY the docker group. The docker run command will be executed as dockerroot. Here is an installation guide for CentOS 6 and BeeGFS
• Slurm is not a prerequisite, but if you run socker inside a Slurm job, it will put the container under Slurm's control

Support and Bug Reports

Report an issue on the issues section or send an email to [email protected]