SIGKILL during zombie state following exit(2) changes exit status

[prattmic]

Description

If a SIGKILL is sent while a thread group is zombied it can change the exit status to SIGKILL if the last task exited with exit(2). If it exited with exit_group(2), then the exit status cannot change (matching Linux).

Steps to reproduce

See ExitTest.SigkillZombieThread in the incoming PR.

runsc version

No response

docker version (if using docker)

No response

uname

No response

kubectl (if using Kubernetes)

No response

repo state (if built from source)

No response

runsc debug logs (if available)

No response

[prattmic]

At a higher level, runsc sends a SIGKILL to all tasks when the gofer exits. The gofer exits when the init task exits. If the init task exits with exit(2), then the gofer shutdown SIGKILL changes the exit status to 9, making the entire container appear to have failed.