Dangerzone-image

This repository contains the dangerzone container image that is used to perform "document to pixels" conversions. This container is used by dangerzone to securely convert its documents.

Using the container image

The image is published on a monthly basis on the container registry, alongside their Cosign signatures. Additionally, nightly and development branches are published under the dangerzone-testing namespace.

What this container provides

This container provides a way to convert documents to pixel buffers, using a secure sandbox.

The security of the sandbox is provided by different layers:

• The container uses gVisor, an application Kernel that provides a strong layer of isolation between running applications and the host operating system. It is written in a memory-safe language (Go) and runs in userspace.
• Additionally, it is expected that this container is run with specific flags and a specific seccomp policy, to unsure that users are not mapped in the container, that no network is available in the container, etc. See the "how to use" section.

We also provide the following guarantees, related to the distribution of the image:

• The container is signed in an auditable way, using Cosign
• Ultimately, the container is reproducible, and so one can verify that it can be rebuilt, resulting to the same digests.

How to use this container?

The recommended way to use this container is via these flags. They require to defined a specific seccomp policy. Seccomp policies is a way to define which system calls are authorized inside the container.

Here is a podman command with the proper flags, and the gvisor seccomp policy.

podman run \
    --log-driver none \
    --security-opt no-new-privileges \
    --userns nomap \
    --security-opt seccomp=tests/share/seccomp.gvisor.json \
    --cap-drop all \
    --cap-add SYS_CHROOT \
    --security-opt label=type:container_engine_t \
    --network=none \
    -u dangerzone \
    --rm -i ghcr.io/freedomofpress/dangerzone/v1 \
    /usr/bin/python3 -m dangerzone.conversion.doc_to_pixels

Output Format

The output of the container is streamed to stdout in a custom binary format:

1. Total Pages: A 4-byte unsigned integer representing the total number of pages in the converted document.
2. For each page: a. Page Width: A 4-byte unsigned integer representing the width of the page in pixels. b. Page Height: A 4-byte unsigned integer representing the height of the page in pixels. c. Pixel Data: bytes of raw RGB pixel data - Length is width x height x 3 color channels

dangerzone-insecure-conversion python package

[!WARNING] Do not use this unless you are certain about what you are doing. Do not use this to convert documents that should be processed safely!

The python code that runs inside the container is packaged under the name "dangerzone-insecure-conversion". It's considered insecure because the intended way to run dangerzone is by using a hardened sandbox, which is provided by dangerzone.

With that being said, there are situations where it's useful to run this code on its own, for instance when adding new file formats.

Running the tests

uv pip install -e .
uv run pytest

# Or, if you prefer to run the tests outside the sandbox:
uv run pytest --local

# It's also possible to run tests in parallel if you have multiple cores:
uv run --with pytest-xdist pytest -n 6

Building and Reproducing the Image

To build the Dangerzone container image, use the build-image.py script:

python3 build-image.py [OPTIONS]

Common Options:

• --platform <PLATFORM>: Specify the build platform (e.g., linux/amd64, linux/arm64). Defaults to the current platform.
• --runtime <RUNTIME>: Specify the container runtime (docker or podman). Defaults to podman.
• --debian-archive-date <YYYYMMDD>: Use a specific Debian snapshot archive date for reproducibility.

Example:

python3 build-image.py --platform linux/amd64 --debian-archive-date 20231026

To verify the reproducibility of a Dangerzone container image, follow these instructions.