Re: [PATCH] pid: Add a judgment for ns null in pid_nr_ns

[Oleg Nesterov <oleg@redhat.com>]

On 08/04, Oleg Nesterov wrote:
>
> But. I need to re-check, but I just realized that pid_alive() can't
> really help, tsk->thread_pid is not stable even if tsk == current.
>
> This means that, say, task_ppid_nr_ns() is buggy.

After the quick grep I don't see the problematic users, but if a zombie
task T does task_ppid_nr_ns(current, NULL) the kernel can crash:

	- pid_alive() succeeds, the task is not reaped yet
	
	- the parent/debugger does wait()->release_task(T), T->thread_pid
	  is NULL now

	- T calls task_tgid_nr_ns()-> ... pid_nr_ns(ns => NULL) because
	  task_active_pid_ns(T) returns NULL

Do you think this worth fixing?


In any case, I think that task_state(), sched_show_task(), bacct_add_tsk()
should be changed to use task_ppid_nr_ns(tsk).

Oleg.