feat(docker,podman): add SELinux label support for bind mounts

[bergmannf]

Summary

Add optional selinux_label field to bind mount driver configs for both Docker and Podman drivers. On SELinux-enabled hosts (Fedora, RHEL), bind-mounted paths need relabelling so container processes can access them. This was missing from the bind mount support added in #1785.

Related Issue

Extends #1785

Changes

• Add shared SelinuxLabel enum (shared / private) to openshell-core::driver_mounts
• Docker driver: move user bind mounts from the structured Mount API (which lacks SELinux support) to the legacy string-format Binds field, appending :z or :Z when selinux_label is set
• Podman driver: add selinux_label to PodmanDriverMountConfig::Bind and push z/Z to the mount options vec
• Update docs/reference/sandbox-compute-drivers.mdx mount schema tables for both drivers

Testing

• cargo fmt --all -- --check passes
• cargo clippy --workspace passes
• Unit tests added/updated (5 new tests across Docker and Podman drivers)
• Manually tested with Podman on Fedora
• E2E tests added/updated (if applicable)

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

All contributors have signed the DCO ✍️ ✅
_{Posted by the DCO Assistant Lite bot.}

[bergmannf]

I have read the DCO document and I hereby sign the DCO.

[bergmannf]

recheck

[maxamillion]

/ok to test dc6a562

[maxamillion]

@drew @TaylorMutch I'm good with the podman changes here but I don't know enough about docker to know if falling back to the legacy string-format Binds is a problem or not and would like and extra +1 (cc @ericcurtin)

[ericcurtin]

@maxamillion could we do CI maybe? Using the new RHEL 10 runners?

https://github.blog/changelog/2026-06-25-red-hat-enterprise-linux-runner-images-are-now-in-public-preview/

Install Docker Engine:

curl -fsSL https://get.docker.com | bash
sed -i "s#/usr/bin/dockerd#/usr/bin/dockerd --ip-forward-no-drop#g" /usr/lib/systemd/system/docker.service 

Test on both podman/docker on RHEL 10 in CI (above is a change required to get podman and docker running on the same machine).

[ericcurtin]

But to answer the question:

I believe the string-format binds is the only way to do it correctly today in docker.

I still think the RHEL 10 podman/docker builds could be useful though to test cross-compatibility.

[TaylorMutch]

/ok to test 087148e

[ericcurtin]

Opened #2093 to add the RHEL 10 Docker/Podman/SELinux CI suite discussed above. It installs Docker alongside Podman on a bare RHEL 10 runner (the get.docker.com + --ip-forward-no-drop steps from #2092 (comment)), confirms SELinux is enforcing, and runs both e2e:docker and e2e:podman against the same host, failing on any AVC denials. It needs an org-provisioned RHEL 10 larger runner before it can actually execute — flagged in the PR description.

[maxamillion]

@ericcurtin good call, I would like to setup RHEL 10 runners here for CI as well. I'll see if we can get that added.

[krishicks]

Docker treats structured mounts and legacy binds differently with respect to missing source directories: it errors if the source path is missing on a structured mount, but silently creates the missing source as an empty, root-owned directory on a legacy bind. Since this PR moves bind mounts from the structured Mount API onto the legacy Binds field, should we make sure we preserve this fail-fast behavior, and any other beneficial behavior specific to the structured mount syntax?

From the docs:

If you use --volume to bind-mount a file or directory that does not yet exist on the Docker host, Docker automatically creates the directory on the host for you. It's always created as a directory.

By default, --mount does not automatically create a directory if the specified mount path does not exist on the host. Instead, it produces an error

[krishicks]

Instead of filtering out bind mounts here, could you instead do

Suggested change

	.filter(|m| !matches!(m, DockerDriverMountConfig::Bind { .. }))
	.map(docker_mount_from_config)
	.filter_map(|m| docker_mount_from_config(m).transpose())

and have docker_mount_from_config match DockerDriverMountConfig::Bind { .. } => Ok(None)?