On 11/28, Eric W. Biederman wrote: > > Oleg Nesterov <oleg@xxxxxxxxxx> writes: > > > On 11/27, Jürg Billeter wrote: > >> > >> @@ -704,6 +713,9 @@ static void exit_notify(struct task_struct *tsk, int group_dead) > >> struct task_struct *p, *n; > >> LIST_HEAD(dead); > >> > >> + if (group_dead && tsk->signal->kill_descendants_on_exit) > >> + walk_process_tree(tsk, kill_descendant_visitor, NULL); > > > > Well, this is not exactly right, at least this is suboptimal in that > > other sub-threads can too call walk_process_tree(kill_descendant_visitor) > > later for no reason. > > Oleg I think I am missing something. No, it is stupid me who can't read, > Reading kernel/exit.c I see "group_dead = atomic_dec_and_test(&tsk->signal->live)". > Which seems like enough to ensure exactly one task/thread calls walk_process_tree. Of course you right, sorry for confusion. To me it would be more clean to call walk_process_tree(kill_descendant_visitor) unconditionally in find_new_reaper() right before "if (has_child_subreaper)", but then we will need to shift read_lock(tasklist) from walk_process_tree(). So I think the patch is mostly fine, the only problem I can see is that PR_SET_KILL_DESCENDANTS_ON_EXIT can race with PR_SET_CHILD_SUBREAPER, they both need to update the bits in the same word. Oleg.
