lwt_bpf: remove unnecessary rcu_read_lock in run_lwt_bpf
authorTaehee Yoo <ap420073@gmail.com>
Mon, 30 Jul 2018 13:22:59 +0000 (22:22 +0900)
committerDaniel Borkmann <daniel@iogearbox.net>
Tue, 31 Jul 2018 07:26:34 +0000 (09:26 +0200)
run_lwt_bpf is called by bpf_{input/output/xmit}.
These functions are already protected by rcu_read_lock.
because lwtunnel_{input/output/xmit} holds rcu_read_lock
and then calls bpf_{input/output/xmit}.
So that rcu_read_lock in the run_lwt_bpf is unnecessary.

Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Acked-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
net/core/lwt_bpf.c

index e7e626fb87bb34f89014cb16e58aaaf132c65417..a49c7baf62f853703592b573496bdce2ef3fbf2d 100644 (file)
@@ -50,10 +50,8 @@ static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
         * mixing with BH RCU lock doesn't work.
         */
        preempt_disable();
-       rcu_read_lock();
        bpf_compute_data_pointers(skb);
        ret = bpf_prog_run_save_cb(lwt->prog, skb);
-       rcu_read_unlock();
 
        switch (ret) {
        case BPF_OK: