[PATCH 3.2 107/125] perf: Fix race in removing an event

[Ben Hutchings]

3.2.61-rc1 review patch.  If anyone has any objections, please let me know.

------------------

From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>

commit 46ce0fe97a6be7532ce6126bb26ce89fed81528c upstream.

When removing a (sibling) event we do:

	raw_spin_lock_irq(&ctx->lock);
	perf_group_detach(event);
	raw_spin_unlock_irq(&ctx->lock);

	<hole>

	perf_remove_from_context(event);
		raw_spin_lock_irq(&ctx->lock);
		...
		raw_spin_unlock_irq(&ctx->lock);

Now, assuming the event is a sibling, it will be 'unreachable' for
things like ctx_sched_out() because that iterates the
groups->siblings, and we just unhooked the sibling.

So, if during <hole> we get ctx_sched_out(), it will miss the event
and not call event_sched_out() on it, leaving it programmed on the
PMU.

The subsequent perf_remove_from_context() call will find the ctx is
inactive and only call list_del_event() to remove the event from all
other lists.

Hereafter we can proceed to free the event; while still programmed!

Close this hole by moving perf_group_detach() inside the same
ctx->lock region(s) perf_remove_from_context() has.

The condition on inherited events only in __perf_event_exit_task() is
likely complete crap because non-inherited events are part of groups
too and we're tearing down just the same. But leave that for another
patch.

Most-likely-Fixes: e03a9a55b4e ("perf: Change close() semantics for group events")
Reported-by: Vince Weaver <vincent.weaver@xxxxxxxxx>
Tested-by: Vince Weaver <vincent.weaver@xxxxxxxxx>
Much-staring-at-traces-by: Vince Weaver <vincent.weaver@xxxxxxxxx>
Much-staring-at-traces-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Link: http://lkml.kernel.org/r/20140505093124.GN17778@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
[bwh: Backported to 3.2: drop change in perf_pmu_migrate_context()]
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
---
 kernel/events/core.c | 47 ++++++++++++++++++++++++++---------------------
 1 file changed, 26 insertions(+), 21 deletions(-)

--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -1180,6 +1180,11 @@ group_sched_out(struct perf_event *group
 		cpuctx->exclusive = 0;
 }
 
+struct remove_event {
+	struct perf_event *event;
+	bool detach_group;
+};
+
 /*
  * Cross CPU call to remove a performance event
  *
@@ -1188,12 +1193,15 @@ group_sched_out(struct perf_event *group
  */
 static int __perf_remove_from_context(void *info)
 {
-	struct perf_event *event = info;
+	struct remove_event *re = info;
+	struct perf_event *event = re->event;
 	struct perf_event_context *ctx = event->ctx;
 	struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
 
 	raw_spin_lock(&ctx->lock);
 	event_sched_out(event, cpuctx, ctx);
+	if (re->detach_group)
+		perf_group_detach(event);
 	list_del_event(event, ctx);
 	if (!ctx->nr_events && cpuctx->task_ctx == ctx) {
 		ctx->is_active = 0;
@@ -1218,10 +1226,14 @@ static int __perf_remove_from_context(vo
  * When called from perf_event_exit_task, it's OK because the
  * context has been detached from its task.
  */
-static void perf_remove_from_context(struct perf_event *event)
+static void perf_remove_from_context(struct perf_event *event, bool detach_group)
 {
 	struct perf_event_context *ctx = event->ctx;
 	struct task_struct *task = ctx->task;
+	struct remove_event re = {
+		.event = event,
+		.detach_group = detach_group,
+	};
 
 	lockdep_assert_held(&ctx->mutex);
 
@@ -1230,12 +1242,12 @@ static void perf_remove_from_context(str
 		 * Per cpu events are removed via an smp call and
 		 * the removal is always successful.
 		 */
-		cpu_function_call(event->cpu, __perf_remove_from_context, event);
+		cpu_function_call(event->cpu, __perf_remove_from_context, &re);
 		return;
 	}
 
 retry:
-	if (!task_function_call(task, __perf_remove_from_context, event))
+	if (!task_function_call(task, __perf_remove_from_context, &re))
 		return;
 
 	raw_spin_lock_irq(&ctx->lock);
@@ -1252,6 +1264,8 @@ retry:
 	 * Since the task isn't running, its safe to remove the event, us
 	 * holding the ctx->lock ensures the task won't get scheduled in.
 	 */
+	if (detach_group)
+		perf_group_detach(event);
 	list_del_event(event, ctx);
 	raw_spin_unlock_irq(&ctx->lock);
 }
@@ -3046,10 +3060,7 @@ int perf_event_release_kernel(struct per
 	 *     to trigger the AB-BA case.
 	 */
 	mutex_lock_nested(&ctx->mutex, SINGLE_DEPTH_NESTING);
-	raw_spin_lock_irq(&ctx->lock);
-	perf_group_detach(event);
-	raw_spin_unlock_irq(&ctx->lock);
-	perf_remove_from_context(event);
+	perf_remove_from_context(event, true);
 	mutex_unlock(&ctx->mutex);
 
 	free_event(event);
@@ -6459,7 +6470,7 @@ SYSCALL_DEFINE5(perf_event_open,
 		struct perf_event_context *gctx = group_leader->ctx;
 
 		mutex_lock(&gctx->mutex);
-		perf_remove_from_context(group_leader);
+		perf_remove_from_context(group_leader, false);
 
 		/*
 		 * Removing from the context ends up with disabled
@@ -6469,7 +6480,7 @@ SYSCALL_DEFINE5(perf_event_open,
 		perf_event__state_init(group_leader);
 		list_for_each_entry(sibling, &group_leader->sibling_list,
 				    group_entry) {
-			perf_remove_from_context(sibling);
+			perf_remove_from_context(sibling, false);
 			perf_event__state_init(sibling);
 			put_ctx(gctx);
 		}
@@ -6622,13 +6633,7 @@ __perf_event_exit_task(struct perf_event
 			 struct perf_event_context *child_ctx,
 			 struct task_struct *child)
 {
-	if (child_event->parent) {
-		raw_spin_lock_irq(&child_ctx->lock);
-		perf_group_detach(child_event);
-		raw_spin_unlock_irq(&child_ctx->lock);
-	}
-
-	perf_remove_from_context(child_event);
+	perf_remove_from_context(child_event, !!child_event->parent);
 
 	/*
 	 * It can happen that the parent exits first, and has events
@@ -7113,14 +7118,14 @@ static void perf_pmu_rotate_stop(struct
 
 static void __perf_event_exit_context(void *__info)
 {
+	struct remove_event re = { .detach_group = false };
 	struct perf_event_context *ctx = __info;
-	struct perf_event *event;
 
 	perf_pmu_rotate_stop(ctx->pmu);
 
 	rcu_read_lock();
-	list_for_each_entry_rcu(event, &ctx->event_list, event_entry)
-		__perf_remove_from_context(event);
+	list_for_each_entry_rcu(re.event, &ctx->event_list, event_entry)
+		__perf_remove_from_context(&re);
 	rcu_read_unlock();
 }
 

--
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/