[PATCH] RPC: expose API for serializing access to RPC transports

 The next several patches introduce an API that allows transports to
 choose whether the RPC client provides congestion control or whether
 the transport itself provides it.

 The first method we abstract is the one that serializes access to the
 RPC transport to prevent the bytes from different requests from mingling
 together.  This method provides proper request serialization and the
 opportunity to prevent new requests from being started because the
 transport is congested.

 The normal situation is for the transport to handle congestion control
 itself.  Although NFS over UDP was first, it has been recognized after
 years of experience that having the transport provide congestion control
 is much better than doing it in the RPC client.  Thus TCP, and probably
 every future transport implementation, will use the default method,
 xprt_lock_write, provided in xprt.c, which does not provide any kind
 of congestion control.  UDP can continue using the xprt.c-provided
 Van Jacobson congestion avoidance implementation.

 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>

3 files changed, 57 insertions, 12 deletions

diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index ac08e99..eee1c68 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -132,6 +132,7 @@ struct rpc_xprt;
 
 struct rpc_xprt_ops {
 	void		(*set_buffer_size)(struct rpc_xprt *xprt);
+	int		(*reserve_xprt)(struct rpc_task *task);
 	void		(*connect)(struct rpc_task *task);
 	int		(*send_request)(struct rpc_task *task);
 	void		(*set_retrans_timeout)(struct rpc_task *task);
@@ -232,6 +233,8 @@ void			xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long
  */
 void			xprt_connect(struct rpc_task *task);
 void			xprt_reserve(struct rpc_task *task);
+int			xprt_reserve_xprt(struct rpc_task *task);
+int			xprt_reserve_xprt_cong(struct rpc_task *task);
 int			xprt_prepare_transmit(struct rpc_task *task);
 void			xprt_transmit(struct rpc_task *task);
 int			xprt_adjust_timeout(struct rpc_rqst *req);
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 1ac2fbe..2d1e8b8 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -64,14 +64,56 @@ static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
 
 static int	xprt_clear_backlog(struct rpc_xprt *xprt);
 
+/**
+ * xprt_reserve_xprt - serialize write access to transports
+ * @task: task that is requesting access to the transport
+ *
+ * This prevents mixing the payload of separate requests, and prevents
+ * transport connects from colliding with writes.  No congestion control
+ * is provided.
+ */
+int xprt_reserve_xprt(struct rpc_task *task)
+{
+	struct rpc_xprt	*xprt = task->tk_xprt;
+	struct rpc_rqst *req = task->tk_rqstp;
+
+	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
+		if (task == xprt->snd_task)
+			return 1;
+		if (task == NULL)
+			return 0;
+		goto out_sleep;
+	}
+	xprt->snd_task = task;
+	if (req) {
+		req->rq_bytes_sent = 0;
+		req->rq_ntrans++;
+	}
+	return 1;
+
+out_sleep:
+	dprintk("RPC: %4d failed to lock transport %p\n",
+			task->tk_pid, xprt);
+	task->tk_timeout = 0;
+	task->tk_status = -EAGAIN;
+	if (req && req->rq_ntrans)
+		rpc_sleep_on(&xprt->resend, task, NULL, NULL);
+	else
+		rpc_sleep_on(&xprt->sending, task, NULL, NULL);
+	return 0;
+}
+
 /*
- * Serialize write access to transports, in order to prevent different
- * requests from interfering with each other.
- * Also prevents transport connects from colliding with writes.
+ * xprt_reserve_xprt_cong - serialize write access to transports
+ * @task: task that is requesting access to the transport
+ *
+ * Same as xprt_reserve_xprt, but Van Jacobson congestion control is
+ * integrated into the decision of whether a request is allowed to be
+ * woken up and given access to the transport.
  */
-static int
-__xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
+int xprt_reserve_xprt_cong(struct rpc_task *task)
 {
+	struct rpc_xprt	*xprt = task->tk_xprt;
 	struct rpc_rqst *req = task->tk_rqstp;
 
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
@@ -79,7 +121,7 @@ __xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 			return 1;
 		goto out_sleep;
 	}
-	if (xprt->nocong || __xprt_get_cong(xprt, task)) {
+	if (__xprt_get_cong(xprt, task)) {
 		xprt->snd_task = task;
 		if (req) {
 			req->rq_bytes_sent = 0;
@@ -101,20 +143,18 @@ out_sleep:
 	return 0;
 }
 
-static inline int
-xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
+static inline int xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 {
 	int retval;
 
 	spin_lock_bh(&xprt->transport_lock);
-	retval = __xprt_lock_write(xprt, task);
+	retval = xprt->ops->reserve_xprt(task);
 	spin_unlock_bh(&xprt->transport_lock);
 	return retval;
 }
 
 
-static void
-__xprt_lock_write_next(struct rpc_xprt *xprt)
+static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 {
 	struct rpc_task *task;
 
@@ -598,7 +638,7 @@ int xprt_prepare_transmit(struct rpc_task *task)
 		err = req->rq_received;
 		goto out_unlock;
 	}
-	if (!__xprt_lock_write(xprt, task)) {
+	if (!xprt->ops->reserve_xprt(task)) {
 		err = -EAGAIN;
 		goto out_unlock;
 	}
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 79433ff..fc4fbe8 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -1045,6 +1045,7 @@ static void xs_connect(struct rpc_task *task)
 
 static struct rpc_xprt_ops xs_udp_ops = {
 	.set_buffer_size	= xs_udp_set_buffer_size,
+	.reserve_xprt		= xprt_reserve_xprt_cong,
 	.connect		= xs_connect,
 	.send_request		= xs_udp_send_request,
 	.set_retrans_timeout	= xprt_set_retrans_timeout_rtt,
@@ -1054,6 +1055,7 @@ static struct rpc_xprt_ops xs_udp_ops = {
 
 static struct rpc_xprt_ops xs_tcp_ops = {
 	.set_buffer_size	= xs_tcp_set_buffer_size,
+	.reserve_xprt		= xprt_reserve_xprt,
 	.connect		= xs_connect,
 	.send_request		= xs_tcp_send_request,
 	.set_retrans_timeout	= xprt_set_retrans_timeout_def,