Init: Graceful termination and signal propagation #8

New issue

Closed

opened 2026-02-11 19:21:19 +00:00 by thabeta · 1 comment

thabeta commented

2026-02-11 19:21:19 +00:00

Owner

The chvm-init binary (pid 1) handles signals in signal.rs. The implementation needs to ensure atomic cleanup of mounts and termination of child processes.

Proposed Changes:

Ensure signals are propagated to the entire process group.
Implement a timeout for processes to exit before sending force kill signals.
Verify filesystem integrity by ensuring proper unmounting before exit.

The `chvm-init` binary (pid 1) handles signals in `signal.rs`. The implementation needs to ensure atomic cleanup of mounts and termination of child processes. **Proposed Changes:** 1. Ensure signals are propagated to the entire process group. 2. Implement a timeout for processes to exit before sending force kill signals. 3. Verify filesystem integrity by ensuring proper unmounting before exit.

rawan self-assigned this

2026-03-09 10:09:57 +00:00

rawan commented

2026-03-10 15:26:25 +00:00

Member

Changes implemented:

Re-entrancy guard on poweroff
Added AtomicBool flag. The first thread to call poweroff() claims it. Any subsequent callers loop on pause() forever, preventing concurrent shutdown races on unmounts and the reboot syscall.
Child process termination before unmounting
Added terminate_all_children() which runs before sync() and unmounts:
- Sends SIGTERM to all child processes
- Polls for up to 3 seconds, reaping exited children every 50ms
- If any children survive, sends SIGKILL and reaps for up to 1 second

This ensures all file handles on mounts are released before umount runs

Changes implemented: 1. Re-entrancy guard on `poweroff` Added AtomicBool flag. The first thread to call poweroff() claims it. Any subsequent callers loop on pause() forever, preventing concurrent shutdown races on unmounts and the reboot syscall. 2. Child process termination before unmounting Added `terminate_all_children()` which runs before sync() and unmounts: - Sends SIGTERM to all child processes - Polls for up to 3 seconds, reaping exited children every 50ms - If any children survive, sends SIGKILL and reaps for up to 1 second This ensures all file handles on mounts are released before umount runs

rawan referenced this issue

2026-03-10 22:02:16 +00:00

Init: Graceful termination and signal propagation #29