Advanced Container Escape Techniques

Containers are one of the most common ways modern applications get packaged and run. Docker alone sits underneath a huge share of production infrastructure, and most teams treat the isolation a container provides as a hard security boundary rather than the thin, misconfigurable layer it actually is. That gap between assumed and actual isolation shows up constantly on real assessments, and it's exactly what this room is built to demonstrate.

As a pentester, you'll work through four container escape techniques back to back: a shared volume nobody scoped properly, a --privileged flag that hands over far more than it should, a gap between Linux capabilities and plain file permissions, and a single narrow capability that grants full host code execution. Each technique builds on the one before it, four containers, one unbroken chain, so treat this less like four standalone puzzles and more like one continuous pivot.

Here's the scenario: you already have access to container-alpha, one small piece of a larger internal Docker deployment, from earlier in this engagement. It's a thin foothold, no sudo, no obvious misconfiguration sitting in plain sight, no --privileged flag anywhere, no exposed docker.sock waiting to be found, but it's enough to start finding your way out. Your objective for the rest of this room: turn that one thin shell into a full compromise of every container on this network, using nothing but what a low-privilege shell can already reach.

Learning Objectives

• Identify an unintended shared Docker volume between containers of different trust levels and recognize why permissive shared-mount permissions are a container-native misconfiguration
• Exploit a world-writable shared volume to pivot from a low-privilege, no-capability container into a root-owned neighboring container
• Explain what the --privileged flag exposes beyond capabilities and use that access to mount and read the host's raw block device from inside a container
• Distinguish capability-gated syscalls from DAC and device-cgroup-gated raw device reads, and exploit that gap in a hardened, capability-dropped container
• Compile and load a malicious kernel module to prove that a single narrow capability grants unrestricted host code execution regardless of container or namespace boundaries
• Map each escape class covered here to a concrete, defense-in-depth remediation

Prerequisites

• Container fundamentals: completion of TryHackMe's Container Vulnerabilities room is recommended. That room covers capability and --privileged abuse, cgroup v1 release_agent abuse, exposed docker.sock abuse, and namespace abuse via nsenter. This room assumes familiarity with those concepts and doesn't re-teach them.
• Linux CLI comfort: completion of TryHackMe's Linux Fundamentals Part 1 room is recommended if you're not yet comfortable navigating a Linux shell over SSH, reading file permissions, and running basic commands like mount, ls, and ps.

You're still sitting inside container-alpha, one of several containers on a small internal Docker bridge network with no route out to the internet. Everything from here happens laterally, one pivot at a time, entirely inside this environment.

Machine Access

Start the lab by clicking the Start Machine button below. The Docker host at MACHINE_IP needs about two minutes to fully boot. All exercises run from the AttackBox against this lab machine.

Set up your virtual environment

To successfully complete this room, you'll need to set up your virtual environment. This involves starting the Lab Machine, ensuring you're equipped with the necessary tools and access to tackle the challenges ahead.

Lab machine

Status:Off

Start Lab Machine

Credentials for the lab machine are below:

Once you're up, open a terminal and SSH straight into container-alpha using the credentials above, the same way you'd land on any other target host in an engagement.

root@tryhackme:~# ssh alpha@MACHINE_IP
alpha@MACHINE_IP's password:
alpha@f34c12183748:/$

That prompt is where every technique in this room begins.