fix(vm-driver): fixes BYOC sandbox creation failing with ext4-fs writ…

[bornav]

Summary

Fix BYOC sandbox creation failing with "EXT4-fs (vdc): write access unavailable, cannot proceed" under the VM driver by cleanly unmounting the overlay filesystem before the image-prep VM exits, preventing the ext4 journal from being left in a dirty state. Adds noload to the image-cache mount and attaches the disk as hypervisor-writable as belt-and-suspenders so a dirty journal never blocks the mount.

Related Issue

Closes #2149

Changes

changing how VM driver creates container

Testing

• mise run pre-commit passes | check comment

• Unit tests added/updated

• E2E tests added/updated (if applicable)

• manually tested before and after patch if it fixes the issue

Checklist

• Follows Conventional Commits
• Commits are signed off (DCO)
• Architecture docs updated (if applicable)

[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.}

[bornav]

mise run pre-commit fails both on main and with my patch with error:

[test:rust] test l7::rest::tests::relay_response_no_framing_event_stream_survives_idle_gap ... ok
[test:rust]
[test:rust] failures:
[test:rust]
[test:rust] ---- procfs::tests::binary_path_preserves_live_deleted_basename stdout ----
[test:rust]
[test:rust] thread 'procfs::tests::binary_path_preserves_live_deleted_basename' (303235) panicked at crates/openshell-supervisor-network/src/procfs.rs:662:48:
[test:rust] called `Result::unwrap()` on an `Err` value: Os { code: 2, kind: NotFound, message: "No such file or directory" }
[test:rust] note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
[test:rust]
[test:rust] ---- procfs::tests::binary_path_strips_deleted_suffix stdout ----
[test:rust]
[test:rust] thread 'procfs::tests::binary_path_strips_deleted_suffix' (303237) panicked at crates/openshell-supervisor-network/src/procfs.rs:608:48:
[test:rust] called `Result::unwrap()` on an `Err` value: Os { code: 2, kind: NotFound, message: "No such file or directory" }
[test:rust]
[test:rust] ---- procfs::tests::binary_path_strips_suffix_for_non_utf8_filename stdout ----
[test:rust]
[test:rust] thread 'procfs::tests::binary_path_strips_suffix_for_non_utf8_filename' (303238) panicked at crates/openshell-supervisor-network/src/procfs.rs:705:48:
[test:rust] called `Result::unwrap()` on an `Err` value: Os { code: 2, kind: NotFound, message: "No such file or directory" }
[test:rust]
[test:rust]
[test:rust] failures:
[test:rust]     procfs::tests::binary_path_preserves_live_deleted_basename
[test:rust]     procfs::tests::binary_path_strips_deleted_suffix
[test:rust]     procfs::tests::binary_path_strips_suffix_for_non_utf8_filename
[test:rust]
[test:rust] test result: FAILED. 829 passed; 3 failed; 0 ignored; 0 measured; 0 filtered out; finished in 6.02s
[test:rust]
[test:rust] error: test failed, to rerun pass `-p openshell-supervisor-network --lib`
Finished in 48.05s
[test:rust] ERROR task failed

[bornav]

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

[bornav]

recheck

[drew]

/ok to test 6bde630

[drew]

Could we keep the prepared image disk hypervisor-read-only? Since this is a shared cache, attaching it writable allows guest root to modify /dev/vdc directly despite the ro,noload filesystem mount, potentially corrupting or poisoning the cache for other sandboxes.

How about making umount /overlay mandatory in image-prep mode and failing preparation if it cannot cleanly unmount?

[bornav]

@drew updated the pr, tested and still works, but think there is still a situation where disk can be left in "locked" state

unfortunatly cant test permissions properly as policies are currently not working think this #1998

[drew]

Hi @bornav, #1998 should be fixed if you want to try and retest.

[bornav]

@drew after rebasing from main retested both scenarios again, still the patch here is neccesary for BYOC

but now with latest main permissions work well and afaics nothing is broken