Re: [PATCH v2 mptcp-next 1/8] mptcp: enable busypoll from mptcp receive path

[Florian Westphal <fw@strlen.de>]

Paolo Abeni <pabeni@redhat.com> wrote:
> On Tue, 2021-05-11 at 15:36 +0200, Florian Westphal wrote:
> > +	if (sk_can_busy_loop(sk) &&
> > +	    skb_queue_empty_lockless(&msk->receive_queue) &&
> > +	    skb_queue_empty_lockless(&sk->sk_receive_queue) &&
> > +	    inet_sk_state_load(sk) == TCP_ESTABLISHED)
> > +		sk_busy_loop(sk, nonblock);
> 
> If I understand the code correctly, here we need a valid sk-
> >sk_napi_id. That field is set by the TCP input path
> in tcp_finish_connect()/tcp_v{4,6}_do_rcv()/tcp_child_process() for the
> subflow ssk.
> 
> I think sk_napi_id is not currently available for the msk?!? 
> Copying it from the subflow requires some care, as the msk could end-up 
> busy polling on the 'wrong' napi instance ?!?

What about:

diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
--- a/net/mptcp/protocol.c
+++ b/net/mptcp/protocol.c
@@ -303,6 +303,7 @@ static bool __mptcp_move_skb(struct mptcp_sock *msk, struct sock *ssk,
 		if (tail && mptcp_try_coalesce(sk, tail, skb))
 			return true;
 
+		sk_mark_napi_id_once(sk, skb);
 		skb_set_owner_r(skb, sk);
 		__skb_queue_tail(&sk->sk_receive_queue, skb);
 		return true;

This sets the napi id to one that provided the last mptcp-in-sequence skb.