Feedback wanted on Docker hardening flags in my install scripts

[PacifAIst]

I've written a tool that starts a sandboxed container with GPU access and exposes a terminal over a link. Before I ask anyone else to run it, I'd like people who know Docker security better than I do to look at how I'm launching the container.

The two install scripts are here: https://github.com/PacifAIst/runsnack (Apache 2.0)

The relevant part is the docker run invocation — read-only root filesystem, all capabilities dropped, no-new-privileges, non-root user, with --gpus passthrough.

What I'm unsure about:

1. GPU passthrough widens the attack surface since the container talks to the host's kernel driver. Is there anything else worth restricting when --gpus is in play, or is that boundary just inherently weaker?
2. Am I missing flags that should be standard here — seccomp profile, user namespace remapping, anything around resource limits?
3. Does the script handle the Docker-not-running and missing-NVIDIA-toolkit cases sensibly, or is the failure mode confusing?

Also happy to hear if anyone tries it on a setup I haven't — ARM, WSL2, non-NVIDIA — but the flags are what I most want checked.

For transparency: the scripts are open, the agent inside the container is closed source and ships separately. The README says where that line falls.

Guidelines

• I have read and understood this category's guidelines before making this post.

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[tgeruzov]

I checked the docker run in snackup.sh. The baseline is already pretty solid: read-only rootfs, dropped capabilities, no-new-privileges, non-root execution, CPU/memory limits and bounded tmpfs mounts.

I wouldn't add a custom seccomp profile just for the sake of having one — Docker uses its built-in seccomp profile by default unless it's explicitly disabled.

A couple of things I'd still consider:

• add --pids-limit so a session can't exhaust the host's process table;
• if the user doesn't actually need every GPU, expose a specific GPU with --gpus '"device=0"' instead of --gpus all;
• for a stronger host boundary on Linux, rootless Docker or daemon-level userns-remap is worth considering.

The GPU boundary is inherently weaker than CPU-only isolation because the workload has to talk to the host NVIDIA driver, so I wouldn't describe GPU passthrough as equivalent to a VM security boundary.

One thing I'd change in the detection logic: docker info | grep -qi nvidia is a fairly indirect test. I'd distinguish "Docker isn't running" from "GPU runtime/toolkit isn't usable" and report the latter separately rather than silently treating it as CPU-only.