Re: [PATCH tip/core/urgent 3/3] sched: protect__sched_setscheduler() access to cgroups

[Peter Zijlstra]

On Thu, 2010-04-22 at 12:54 -0700, Paul E. McKenney wrote:
> A given task's cgroups structures must remain while that task is running
> due to reference counting, so this is presumably a false positive.
> Updated to reflect feedback from Tetsuo Handa.

I think its not a false positive, I think we can race with the task
being placed in another cgroup. We don't hold task_lock() [our other
discussion] nor does it hold rq->lock [used by the sched ->attach()
method].

That said, we should probably cure the race condition of
sched_setscheduler() vs ->attach().

Something like the below perhaps?

Signed-off-by: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
---
 kernel/sched.c |   38 ++++++++++++++++++++++++++------------
 1 files changed, 26 insertions(+), 12 deletions(-)

diff --git a/kernel/sched.c b/kernel/sched.c
index 95eaecc..345df67 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -4425,16 +4425,6 @@ recheck:
 	}
 
 	if (user) {
-#ifdef CONFIG_RT_GROUP_SCHED
-		/*
-		 * Do not allow realtime tasks into groups that have no runtime
-		 * assigned.
-		 */
-		if (rt_bandwidth_enabled() && rt_policy(policy) &&
-				task_group(p)->rt_bandwidth.rt_runtime == 0)
-			return -EPERM;
-#endif
-
 		retval = security_task_setscheduler(p, policy, param);
 		if (retval)
 			return retval;
@@ -4450,6 +4440,28 @@ recheck:
 	 * runqueue lock must be held.
 	 */
 	rq = __task_rq_lock(p);
+	retval = 0;
+#ifdef CONFIG_RT_GROUP_SCHED
+	if (user) {
+		/*
+		 * Do not allow realtime tasks into groups that have no runtime
+		 * assigned.
+		 *
+		 * RCU read lock not strictly required but here for PROVE_RCU,
+		 * the task is pinned by holding rq->lock which avoids races
+		 * with ->attach().
+		 */
+		rcu_read_lock();
+		if (rt_bandwidth_enabled() && rt_policy(policy) &&
+				task_group(p)->rt_bandwidth.rt_runtime == 0)
+			retval = -EPERM;
+		rcu_read_unlock();
+
+		if (retval)
+			goto unlock;
+	}
+#endif
+
 	/* recheck policy now with rq lock held */
 	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
 		policy = oldpolicy = -1;
@@ -4477,12 +4489,14 @@ recheck:
 
 		check_class_changed(rq, p, prev_class, oldprio, running);
 	}
+unlock:
 	__task_rq_unlock(rq);
 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
 
-	rt_mutex_adjust_pi(p);
+	if (!retval)
+		rt_mutex_adjust_pi(p);
 
-	return 0;
+	return retval;
 }
 
 /**


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/