Skip to content

Conversation

@metan-ucw
Copy link
Member

If the process we executed with ltx_handle_exec() forks children these did escape the attempt to kill these with ltx_handle_kill() because we only killed the parent, the children were reparented to init and may continue to run happily ever after. Since the parent process was killed we wait() on it and we got stuck in the read() that gets rest of the log from the pipe the stderr and stdout of the executed process is redirected to. That is because the pipe stil has writers active as long as any child process is active. If the main shell on the system was from busybox this happend for any executed command since busybox forks and executes in each case.

To fix it, we move the top process in the slot into a seprate process group and the kill command kills a process group rather than a single process. In order to make sure that the process is moved into right process group right after the fork we move it both in the parent and in the child.

We need it to move there in the parent in a case that subsequent command is kill and terminate the process before it has a chance to run.

We need to make the move in the child because in case it runs before the parent (ltx process) we need the process group right before we may possibly fork any children.

If the process we executed with ltx_handle_exec() forks children these
did escape the attempt to kill these with ltx_handle_kill() because we
only killed the parent, the children were reparented to init and may
continue to run happily ever after. Since the parent process was killed
we wait() on it and we got stuck in the read() that gets rest of the log
from the pipe the stderr and stdout of the executed process is
redirected to. That is because the pipe stil has writers active as long
as any child process is active. If the main shell on the system was from
busybox this happend for any executed command since busybox forks and
executes in each case.

To fix it, we move the top process in the slot into a seprate process
group and the kill command kills a process group rather than a single
process. In order to make sure that the process is moved into right
process group right after the fork we move it both in the parent and in
the child.

We need it to move there in the parent in a case that subsequent command
is kill and terminate the process before it has a chance to run.

We need to make the move in the child because in case it runs before the
parent (ltx process) we need the process group right before we may
possibly fork any children.

Signed-off-by: Cyril Hrubis <chrubis@suse.cz>
@acerv acerv merged commit ce7203b into linux-test-project:master Aug 26, 2025
1 check passed
acerv added a commit to linux-test-project/kirk that referenced this pull request Aug 26, 2025
LTX discovered a bug caused by not setting pid group while new EXEC
command are performed. This was causing *_stop tests to fail and it's
going to be fixed in the next release.

linux-test-project/ltx#2

Signed-off-by: Andrea Cervesato <andrea.cervesato@suse.com>
@pevik
Copy link
Member

pevik commented Aug 26, 2025

Thanks for fixing this. IMHO a candidate for yet another release bump, let's do it unless you plan to do more fixes soon.

@acerv
Copy link
Collaborator

acerv commented Aug 26, 2025

Thanks for fixing this. IMHO a candidate for yet another release bump, let's do it unless you plan to do more fixes soon.

Yes, this bug requires a new release for sure

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants