Dockerfile
builds the vulnerable container image
Defines an Ubuntu-based image that creates directories and symlinks to set up a TOCTOU race condition for the NVIDIA Container Toolkit's library mounting mechanism.
Dockerfile:1-13r0binak/CVE-2024-0132
PoC files
2 filesFile viewing is interactive and short-lived. Downloads are password-protected ZIP archives using password eip.
Docker lab environments
1GitHubRepository root
Created Vuln labCVE-2024-0132Dockerfile
Analysis
Environment assessment
A single Dockerfile based on Ubuntu that creates a directory structure and symlinks to demonstrate a TOCTOU vulnerability in the NVIDIA Container Toolkit (CVE-2024-0132). The README explains the exploit mechanism.
Dockerfile:1-13README.md:1-51Lab assessment
Vulnerability labThe repository is explicitly named after CVE-2024-0132 and the README describes a proof-of-concept (PoC) for a TOCTOU vulnerability in the NVIDIA Container Toolkit. The Dockerfile constructs the necessary directory and symlink layout to exploit the described vulnerability.
README.md:1-2README.md:5-51Lab shapeDockerfile
ServicesUnknown
Compose manifests0
Dockerfiles1
Evidence & reasoningComponents · CVE assessment · exercise context · review boundaries
Services and files described by the evidence
README.md
documents the vulnerability and PoC
Explains the CVE-2024-0132 vulnerability, the TOCTOU attack vector, and the purpose of each Dockerfile instruction.
README.md:1-51How the supplied evidence relates each vulnerability
CVE-2024-0132
Supported by supplied evidence
The repository name, README title, and the described exploit mechanism (TOCTOU in NVIDIA Container Toolkit's /usr/local/cuda/compat/ handling) directly correspond to CVE-2024-0132. The Dockerfile implements the described directory and symlink layout.
README.md:1-2README.md:5-51Dockerfile:1-13Requirements and sequence described by the evidence
Prerequisites
- The container must be run with the NVIDIA Container Toolkit (nvidia-docker) and a GPU driver installed on the host, so that the toolkit attempts to mount libraries from /usr/local/cuda/compat/.
README.md:9-11 - The host must have a real NVIDIA driver version different from 1337, so the library name libdxcore.so.1337 passes the toolkit's filter but does not match the host driver version.
README.md:37
Evidence-described exercise path
- Build the Docker image using the provided Dockerfile.
Dockerfile:1-13 - Run the container with the NVIDIA Container Toolkit (e.g., docker run --gpus all ...).
README.md:9-11 - The toolkit checks /usr/local/cuda/compat/ and resolves the symlinks. Due to the TOCTOU race, the host filesystem is mounted inside the container at /usr/lib/x86_64-linux-gnu/libdxcore.so.1337.hostfs/.
README.md:43-47
Behaviors behind the stored safety assessment
No harmful behavior observed
The Dockerfile only creates directories, a test file, and symlinks. It does not execute any commands, download payloads, or establish network connections. The described exploit targets the container's own host filesystem mount point, which is the intended lab target for a container-escape demonstration. No hidden or unrelated harmful behavior is present.
Dockerfile:1-13README.md:43-47What the analysis did not establish
- The packet contains only a Dockerfile and README; no runtime exploit script or binary is included.
- The actual success of the TOCTOU race condition depends on timing and the specific NVIDIA Container Toolkit version, which is not verified by the supplied evidence.
- The README contains a typo in the last Dockerfile instruction (line 49 uses /usr/lib64/ instead of /usr/lib/x86_64-linux-gnu/), which may affect the exploit's functionality.
Model interpretation
This review is limited to the supplied lab evidence packet. It does not assert that the environment runs, reproduces a vulnerability, or is safe to execute. Contract: eip-docker-lab-analysis-v1.