core: process refused TCP data immediately once the connection is ready

If the first TCP data chunk was refused, perhaps due to the
connection was under connecting state (it's almost always the
case), waiting for timers to process the refused data would
introduce a ~250ms (TCP_FAST_INTERVAL) delay for the connection.
(from eycorsican/go-tun2socks)

core/c/include/lwip/tcp.h

@@ -479,6 +479,8 @@ err_t            tcp_output  (struct tcp_pcb *pcb);
 
 err_t            tcp_tcp_get_tcp_addrinfo(struct tcp_pcb *pcb, int local, ip_addr_t *addr, u16_t *port);
 
+err_t             tcp_process_refused_data(struct tcp_pcb *pcb);
+
 #define tcp_dbg_get_tcp_state(pcb) ((pcb)->state)
 
 /* for compatibility with older implementation */

core/tcp_conn.go

@@ -64,7 +64,7 @@ type tcpConn struct {
 	localAddr     *net.TCPAddr
 	connKeyArg    unsafe.Pointer
 	connKey       uint32
-	canWrite      *sync.Cond // Condition variable to implement TCP back pressure.
+	canWrite      *sync.Cond // Condition variable to implement TCP backpressure.
 	state         tcpConnState
 	sndPipeReader *io.PipeReader
 	sndPipeWriter *io.PipeWriter
@@ -116,6 +116,12 @@ func newTCPConn(pcb *C.struct_tcp_pcb, handler TCPConnHandler) (TCPConn, error)
 			conn.Lock()
 			conn.state = tcpConnected
 			conn.Unlock()
+
+			lwipMutex.Lock()
+			if pcb.refused_data != nil {
+				C.tcp_process_refused_data(pcb)
+			}
+			lwipMutex.Unlock()
 		}
 	}()
 
@@ -133,11 +139,9 @@ func (conn *tcpConn) LocalAddr() net.Addr {
 func (conn *tcpConn) SetDeadline(t time.Time) error {
 	return nil
 }
-
 func (conn *tcpConn) SetReadDeadline(t time.Time) error {
 	return nil
 }
-
 func (conn *tcpConn) SetWriteDeadline(t time.Time) error {
 	return nil
 }