[PATCH] RPC: separate xprt_timer implementations

Allow transports to hook the retransmit timer interrupt.  Some transports
 calculate their congestion window here so that a retransmit timeout has
 immediate effect on the congestion window.

 Test-plan:
 Use WAN simulation to cause sporadic bursty packet loss.  Look for significant
 regression in performance or client stability.

 Signed-off-by: Chuck Lever <cel@netapp.com>
 Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

include/linux/sunrpc/xprt.h

@@ -137,6 +137,7 @@ struct rpc_xprt_ops {
 	void		(*connect)(struct rpc_task *task);
 	int		(*send_request)(struct rpc_task *task);
 	void		(*set_retrans_timeout)(struct rpc_task *task);
+	void		(*timer)(struct rpc_task *task);
 	void		(*close)(struct rpc_xprt *xprt);
 	void		(*destroy)(struct rpc_xprt *xprt);
 };
@@ -257,6 +258,7 @@ void			xprt_set_retrans_timeout_rtt(struct rpc_task *task);
 void			xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status);
 void			xprt_wait_for_buffer_space(struct rpc_task *task);
 void			xprt_write_space(struct rpc_xprt *xprt);
+void			xprt_adjust_cwnd(struct rpc_task *task, int result);
 struct rpc_rqst *	xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid);
 void			xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied);
 void			xprt_disconnect(struct rpc_xprt *xprt);

net/sunrpc/xprt.c

@@ -289,16 +289,19 @@ __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 	__xprt_lock_write_next_cong(xprt);
 }
 
-/*
+/**
- * Adjust RPC congestion window
+ * xprt_adjust_cwnd - adjust transport congestion window
+ * @task: recently completed RPC request used to adjust window
+ * @result: result code of completed RPC request
+ *
  * We use a time-smoothed congestion estimator to avoid heavy oscillation.
  */
-static void
+void xprt_adjust_cwnd(struct rpc_task *task, int result)
-xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
 {
-	unsigned long	cwnd;
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_xprt *xprt = task->tk_xprt;
+	unsigned long cwnd = xprt->cwnd;
 
-	cwnd = xprt->cwnd;
 	if (result >= 0 && cwnd <= xprt->cong) {
 		/* The (cwnd >> 1) term makes sure
 		 * the result gets rounded properly. */
@@ -314,6 +317,7 @@ xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
 	dprintk("RPC:      cong %ld, cwnd was %ld, now %ld\n",
 			xprt->cong, xprt->cwnd, cwnd);
 	xprt->cwnd = cwnd;
+	__xprt_put_cong(xprt, req);
 }
 
 /**
@@ -602,8 +606,7 @@ void xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 	/* Adjust congestion window */
 	if (!xprt->nocong) {
 		unsigned timer = task->tk_msg.rpc_proc->p_timer;
-		xprt_adjust_cwnd(xprt, copied);
+		xprt_adjust_cwnd(task, copied);
-		__xprt_put_cong(xprt, req);
 		if (timer) {
 			if (req->rq_ntrans == 1)
 				rpc_update_rtt(clnt->cl_rtt, timer,
@@ -640,27 +643,19 @@ void xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 	return;
 }
 
-/*
+static void xprt_timer(struct rpc_task *task)
- * RPC receive timeout handler.
- */
-static void
-xprt_timer(struct rpc_task *task)
 {
-	struct rpc_rqst	*req = task->tk_rqstp;
+	struct rpc_rqst *req = task->tk_rqstp;
 	struct rpc_xprt *xprt = req->rq_xprt;
 
+	dprintk("RPC: %4d xprt_timer\n", task->tk_pid);
+
 	spin_lock(&xprt->transport_lock);
-	if (req->rq_received)
+	if (!req->rq_received) {
-		goto out;
+		if (xprt->ops->timer)
-
+			xprt->ops->timer(task);
-	xprt_adjust_cwnd(req->rq_xprt, -ETIMEDOUT);
+		task->tk_status = -ETIMEDOUT;
-	__xprt_put_cong(xprt, req);
+	}
-
-	dprintk("RPC: %4d xprt_timer (%s request)\n",
-		task->tk_pid, req ? "pending" : "backlogged");
-
-	task->tk_status  = -ETIMEDOUT;
-out:
 	task->tk_timeout = 0;
 	rpc_wake_up_task(task);
 	spin_unlock(&xprt->transport_lock);

net/sunrpc/xprtsock.c

@@ -860,6 +860,17 @@ static void xs_tcp_set_buffer_size(struct rpc_xprt *xprt)
 	return;
 }
 
+/**
+ * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
+ * @task: task that timed out
+ *
+ * Adjust the congestion window after a retransmit timeout has occurred.
+ */
+static void xs_udp_timer(struct rpc_task *task)
+{
+	xprt_adjust_cwnd(task, -ETIMEDOUT);
+}
+
 static int xs_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
 {
 	struct sockaddr_in myaddr = {
@@ -1050,6 +1061,7 @@ static struct rpc_xprt_ops xs_udp_ops = {
 	.connect		= xs_connect,
 	.send_request		= xs_udp_send_request,
 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
+	.timer			= xs_udp_timer,
 	.close			= xs_close,
 	.destroy		= xs_destroy,
 };