posix-timers: Use sighand lock instead of tasklist_lock for task cloc…

…k sample

There is no need for the tasklist_lock just to take a process
wide clock sample.

All we need is to get a coherent sample that doesn't race with
exit() and exec():

* exit() may be concurrently reaping a task and flushing its time

* sighand is unstable under exit() and exec(), and the latter also
  result in group leader that can change

To protect against these, locking the target's sighand is enough.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Kosaki Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andrew Morton <akpm@linux-foundation.org>

kernel/posix-cpu-timers.c

@@ -271,12 +271,22 @@ static int posix_cpu_clock_get_task(struct task_struct *tsk,
 		if (same_thread_group(tsk, current))
 			err = cpu_clock_sample(which_clock, tsk, &rtn);
 	} else {
-		read_lock(&tasklist_lock);
+		unsigned long flags;
+		struct sighand_struct *sighand;
 
-		if (tsk->sighand && (tsk == current || thread_group_leader(tsk)))
+		/*
+		 * while_each_thread() is not yet entirely RCU safe,
+		 * keep locking the group while sampling process
+		 * clock for now.
+		 */
+		sighand = lock_task_sighand(tsk, &flags);
+		if (!sighand)
+			return err;
+
+		if (tsk == current || thread_group_leader(tsk))
 			err = cpu_clock_sample_group(which_clock, tsk, &rtn);
 
-		read_unlock(&tasklist_lock);
+		unlock_task_sighand(tsk, &flags);
 	}
 
 	if (!err)