]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
exit: fix the usage of delay_group_leader->exit_code in do_notify_parent() and pidfs_...
authorOleg Nesterov <oleg@redhat.com>
Mon, 24 Mar 2025 17:19:41 +0000 (18:19 +0100)
committerChristian Brauner <brauner@kernel.org>
Tue, 25 Mar 2025 14:56:22 +0000 (15:56 +0100)
commit9133607de37a4887c6f89ed937176a0a0c1ebb17
treebb871df6b63ffa7344bedc135dda2e331435684a
parent1243045c9448cd3f29e9d075de58dc81a0c2c3d9
exit: fix the usage of delay_group_leader->exit_code in do_notify_parent() and pidfs_exit()

Consider a process with a group leader L and a sub-thread T.
L does sys_exit(1), then T does sys_exit_group(2).

In this case wait_task_zombie(L) will notice SIGNAL_GROUP_EXIT and use
L->signal->group_exit_code, this is correct.

But, before that, do_notify_parent(L) called by release_task(T) will use
L->exit_code != L->signal->group_exit_code, and this is not consistent.
We don't really care, I think that nobody relies on the info which comes
with SIGCHLD, if nothing else SIGCHLD < SIGRTMIN can be queued only once.

But pidfs_exit() is more problematic, I think pidfs_exit_info->exit_code
should report ->group_exit_code in this case, just like wait_task_zombie().

TODO: with this change we can hopefully cleanup (or may be even kill) the
similar SIGNAL_GROUP_EXIT checks, at least in wait_task_zombie().

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Link: https://lore.kernel.org/r/20250324171941.GA13114@redhat.com
Signed-off-by: Christian Brauner <brauner@kernel.org>
kernel/exit.c