Merge tag 'sched_ext-for-6.19-rc8-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext

Pull sched_ext fix from Tejun Heo:

 - Fix race where sched_class operations (sched_setscheduler() and
   friends) could be invoked on dead tasks after sched_ext_dead()
   already ran, causing invalid SCX task state transitions and NULL
   pointer dereferences.

   This was a regression from the cgroup exit ordering fix which
   moved sched_ext_free() to finish_task_switch().

* tag 'sched_ext-for-6.19-rc8-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext:
  sched_ext: Short-circuit sched_class operations on dead tasks

1 files changed, 48 insertions, 0 deletions

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index afe28c04d5aa..0bb8fa927e9e 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -194,6 +194,7 @@ MODULE_PARM_DESC(bypass_lb_intv_us, "bypass load balance interval in microsecond
 #include <trace/events/sched_ext.h>
 
 static void process_ddsp_deferred_locals(struct rq *rq);
+static bool task_dead_and_done(struct task_struct *p);
 static u32 reenq_local(struct rq *rq);
 static void scx_kick_cpu(struct scx_sched *sch, s32 cpu, u64 flags);
 static bool scx_vexit(struct scx_sched *sch, enum scx_exit_kind kind,
@@ -2619,6 +2620,9 @@ static void set_cpus_allowed_scx(struct task_struct *p,
 
 	set_cpus_allowed_common(p, ac);
 
+	if (task_dead_and_done(p))
+		return;
+
 	/*
 	 * The effective cpumask is stored in @p->cpus_ptr which may temporarily
 	 * differ from the configured one in @p->cpus_mask. Always tell the bpf
@@ -3034,10 +3038,45 @@ void scx_cancel_fork(struct task_struct *p)
 	percpu_up_read(&scx_fork_rwsem);
 }
 
+/**
+ * task_dead_and_done - Is a task dead and done running?
+ * @p: target task
+ *
+ * Once sched_ext_dead() removes the dead task from scx_tasks and exits it, the
+ * task no longer exists from SCX's POV. However, certain sched_class ops may be
+ * invoked on these dead tasks leading to failures - e.g. sched_setscheduler()
+ * may try to switch a task which finished sched_ext_dead() back into SCX
+ * triggering invalid SCX task state transitions and worse.
+ *
+ * Once a task has finished the final switch, sched_ext_dead() is the only thing
+ * that needs to happen on the task. Use this test to short-circuit sched_class
+ * operations which may be called on dead tasks.
+ */
+static bool task_dead_and_done(struct task_struct *p)
+{
+	struct rq *rq = task_rq(p);
+
+	lockdep_assert_rq_held(rq);
+
+	/*
+	 * In do_task_dead(), a dying task sets %TASK_DEAD with preemption
+	 * disabled and __schedule(). If @p has %TASK_DEAD set and off CPU, @p
+	 * won't ever run again.
+	 */
+	return unlikely(READ_ONCE(p->__state) == TASK_DEAD) &&
+		!task_on_cpu(rq, p);
+}
+
 void sched_ext_dead(struct task_struct *p)
 {
 	unsigned long flags;
 
+	/*
+	 * By the time control reaches here, @p has %TASK_DEAD set, switched out
+	 * for the last time and then dropped the rq lock - task_dead_and_done()
+	 * should be returning %true nullifying the straggling sched_class ops.
+	 * Remove from scx_tasks and exit @p.
+	 */
 	raw_spin_lock_irqsave(&scx_tasks_lock, flags);
 	list_del_init(&p->scx.tasks_node);
 	raw_spin_unlock_irqrestore(&scx_tasks_lock, flags);
@@ -3063,6 +3102,9 @@ static void reweight_task_scx(struct rq *rq, struct task_struct *p,
 
 	lockdep_assert_rq_held(task_rq(p));
 
+	if (task_dead_and_done(p))
+		return;
+
 	p->scx.weight = sched_weight_to_cgroup(scale_load_down(lw->weight));
 	if (SCX_HAS_OP(sch, set_weight))
 		SCX_CALL_OP_TASK(sch, SCX_KF_REST, set_weight, rq,
@@ -3077,6 +3119,9 @@ static void switching_to_scx(struct rq *rq, struct task_struct *p)
 {
 	struct scx_sched *sch = scx_root;
 
+	if (task_dead_and_done(p))
+		return;
+
 	scx_enable_task(p);
 
 	/*
@@ -3090,6 +3135,9 @@ static void switching_to_scx(struct rq *rq, struct task_struct *p)
 
 static void switched_from_scx(struct rq *rq, struct task_struct *p)
 {
+	if (task_dead_and_done(p))
+		return;
+
 	scx_disable_task(p);
 }