sched/autogroup: Do not use autogroup->tg in zombie threads

Exactly because for_each_thread() in autogroup_move_group() can't see it
and update its ->sched_task_group before _put() and possibly free().

So the exiting task needs another sched_move_task() before exit_notify()
and we need to re-introduce the PF_EXITING (or similar) check removed by
the previous change for another reason.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: hartsjc@redhat.com
Cc: vbendel@redhat.com
Cc: vlovejoy@redhat.com
Link: http://lkml.kernel.org/r/20161114184612.GA15968@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

3 files changed, 22 insertions, 0 deletions

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 348f51b0ec92ed..e9c009dc3a4a35 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2567,6 +2567,7 @@ extern void sched_autogroup_create_attach(struct task_struct *p);
 extern void sched_autogroup_detach(struct task_struct *p);
 extern void sched_autogroup_fork(struct signal_struct *sig);
 extern void sched_autogroup_exit(struct signal_struct *sig);
+extern void sched_autogroup_exit_task(struct task_struct *p);
 #ifdef CONFIG_PROC_FS
 extern void proc_sched_autogroup_show_task(struct task_struct *p, struct seq_file *m);
 extern int proc_sched_autogroup_set_nice(struct task_struct *p, int nice);
@@ -2576,6 +2577,7 @@ static inline void sched_autogroup_create_attach(struct task_struct *p) { }
 static inline void sched_autogroup_detach(struct task_struct *p) { }
 static inline void sched_autogroup_fork(struct signal_struct *sig) { }
 static inline void sched_autogroup_exit(struct signal_struct *sig) { }
+static inline void sched_autogroup_exit_task(struct task_struct *p) { }
 #endif
 
 extern int yield_to(struct task_struct *p, bool preempt);
diff --git a/kernel/exit.c b/kernel/exit.c
index 9d68c45ebbe30e..3076f3089919b6 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -836,6 +836,7 @@ void __noreturn do_exit(long code)
 	 */
 	perf_event_exit_task(tsk);
 
+	sched_autogroup_exit_task(tsk);
 	cgroup_exit(tsk);
 
 	/*
diff --git a/kernel/sched/auto_group.c b/kernel/sched/auto_group.c
index ad2b19ad6ca026..f1c8fd5662464c 100644
--- a/kernel/sched/auto_group.c
+++ b/kernel/sched/auto_group.c
@@ -115,10 +115,26 @@ bool task_wants_autogroup(struct task_struct *p, struct task_group *tg)
 	 * If we race with autogroup_move_group() the caller can use the old
 	 * value of signal->autogroup but in this case sched_move_task() will
 	 * be called again before autogroup_kref_put().
+	 *
+	 * However, there is no way sched_autogroup_exit_task() could tell us
+	 * to avoid autogroup->tg, so we abuse PF_EXITING flag for this case.
 	 */
+	if (p->flags & PF_EXITING)
+		return false;
+
 	return true;
 }
 
+void sched_autogroup_exit_task(struct task_struct *p)
+{
+	/*
+	 * We are going to call exit_notify() and autogroup_move_group() can't
+	 * see this thread after that: we can no longer use signal->autogroup.
+	 * See the PF_EXITING check in task_wants_autogroup().
+	 */
+	sched_move_task(p);
+}
+
 static void
 autogroup_move_group(struct task_struct *p, struct autogroup *ag)
 {
@@ -142,6 +158,9 @@ autogroup_move_group(struct task_struct *p, struct autogroup *ag)
 	 * In the latter case for_each_thread() can not miss a migrating thread,
 	 * cpu_cgroup_attach() must not be possible after cgroup_exit() and it
 	 * can't be removed from thread list, we hold ->siglock.
+	 *
+	 * If an exiting thread was already removed from thread list we rely on
+	 * sched_autogroup_exit_task().
 	 */
 	for_each_thread(p, t)
 		sched_move_task(t);