[PATCH tip] perf/events/core: fix race in bpf program unregister

[Alexei Starovoitov <ast@plumgrid.com>]

there is a race between perf_event_free_bpf_prog() and free_trace_kprobe():

__free_event()
  event->destroy(event)
    tp_perf_event_destroy()
      perf_trace_destroy()
        perf_trace_event_unreg()

which is dropping event->tp_event->perf_refcount and allows to proceed in:

unregister_trace_kprobe()
  unregister_kprobe_event()
      trace_remove_event_call()
            probe_remove_event_call()
free_trace_kprobe()

while __free_event does:
call_rcu(&event->rcu_head, free_event_rcu);
  free_event_rcu()
    perf_event_free_bpf_prog()

To fix the race simply move perf_event_free_bpf_prog() before
event->destroy(), since event->tp_event is still valid at that point.

Note, perf_trace_destroy() is not racing with trace_remove_event_call()
since they both grab event_mutex.

Fixes: 2541517c32be ("tracing, perf: Implement BPF programs attached to kprobes")
Reported-by: Wang Nan <wangnan0@huawei.com>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
---
Wang, please test the fix to double check.

 kernel/events/core.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 84231a146dd5..b27bdc8b3558 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3415,7 +3415,6 @@ static void free_event_rcu(struct rcu_head *head)
 	if (event->ns)
 		put_pid_ns(event->ns);
 	perf_event_free_filter(event);
-	perf_event_free_bpf_prog(event);
 	kfree(event);
 }
 
@@ -3546,6 +3545,8 @@ static void __free_event(struct perf_event *event)
 			put_callchain_buffers();
 	}
 
+	perf_event_free_bpf_prog(event);
+
 	if (event->destroy)
 		event->destroy(event);
 
-- 
1.7.9.5