[PATCH] RPC: separate TCP and UDP write space callbacks

 Split the socket write space callback function into a TCP version and UDP
 version, eliminating one dependence on the "xprt->stream" variable.

 Keep the common pieces of this path in xprt.c so other transports can use
 it too.

 Test-plan:
 Write-intensive workload on a single mount point.

 Version: Thu, 11 Aug 2005 16:07:51 -0400

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

2 files changed, 87 insertions, 31 deletions

diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 247fa1e..31ef7dc 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -241,6 +241,40 @@ void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
 		rpc_wake_up(&xprt->pending);
 }
 
+/**
+ * xprt_wait_for_buffer_space - wait for transport output buffer to clear
+ * @task: task to be put to sleep
+ *
+ */
+void xprt_wait_for_buffer_space(struct rpc_task *task)
+{
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_xprt *xprt = req->rq_xprt;
+
+	task->tk_timeout = req->rq_timeout;
+	rpc_sleep_on(&xprt->pending, task, NULL, NULL);
+}
+
+/**
+ * xprt_write_space - wake the task waiting for transport output buffer space
+ * @xprt: transport with waiting tasks
+ *
+ * Can be called in a soft IRQ context, so xprt_write_space never sleeps.
+ */
+void xprt_write_space(struct rpc_xprt *xprt)
+{
+	if (unlikely(xprt->shutdown))
+		return;
+
+	spin_lock_bh(&xprt->transport_lock);
+	if (xprt->snd_task) {
+		dprintk("RPC:      write space: waking waiting task on xprt %p\n",
+				xprt);
+		rpc_wake_up_task(xprt->snd_task);
+	}
+	spin_unlock_bh(&xprt->transport_lock);
+}
+
 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 {
 	struct rpc_timeout *to = &req->rq_xprt->timeout;
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 7f0b9f7..70a772d 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -308,15 +308,13 @@ static int xs_send_request(struct rpc_task *task)
 
 	if (status == -EAGAIN) {
 		if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
-			/* Protect against races with xs_write_space */
+			/* Protect against races with write_space */
 			spin_lock_bh(&xprt->transport_lock);
 			/* Don't race with disconnect */
 			if (!xprt_connected(xprt))
 				task->tk_status = -ENOTCONN;
-			else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags)) {
-				task->tk_timeout = req->rq_timeout;
-				rpc_sleep_on(&xprt->pending, task, NULL, NULL);
-			}
+			else if (test_bit(SOCK_NOSPACE, &xprt->sock->flags))
+				xprt_wait_for_buffer_space(task);
 			spin_unlock_bh(&xprt->transport_lock);
 			return status;
 		}
@@ -721,45 +719,68 @@ static void xs_tcp_state_change(struct sock *sk)
 }
 
 /**
- * xs_write_space - callback invoked when socket buffer space becomes
- *                         available
+ * xs_udp_write_space - callback invoked when socket buffer space
+ *                             becomes available
  * @sk: socket whose state has changed
  *
  * Called when more output buffer space is available for this socket.
  * We try not to wake our writers until they can make "significant"
- * progress, otherwise we'll waste resources thrashing sock_sendmsg
+ * progress, otherwise we'll waste resources thrashing kernel_sendmsg
  * with a bunch of small requests.
  */
-static void xs_write_space(struct sock *sk)
+static void xs_udp_write_space(struct sock *sk)
 {
-	struct rpc_xprt *xprt;
-	struct socket *sock;
-
 	read_lock(&sk->sk_callback_lock);
-	if (!(xprt = xprt_from_sock(sk)) || !(sock = sk->sk_socket))
-		goto out;
-	if (xprt->shutdown)
-		goto out;
 
-	/* Wait until we have enough socket memory */
-	if (xprt->stream) {
-		/* from net/core/stream.c:sk_stream_write_space */
-		if (sk_stream_wspace(sk) < sk_stream_min_wspace(sk))
+	/* from net/core/sock.c:sock_def_write_space */
+	if (sock_writeable(sk)) {
+		struct socket *sock;
+		struct rpc_xprt *xprt;
+
+		if (unlikely(!(sock = sk->sk_socket)))
 			goto out;
-	} else {
-		/* from net/core/sock.c:sock_def_write_space */
-		if (!sock_writeable(sk))
+		if (unlikely(!(xprt = xprt_from_sock(sk))))
+			goto out;
+		if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
 			goto out;
+
+		xprt_write_space(xprt);
 	}
 
-	if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
-		goto out;
+ out:
+	read_unlock(&sk->sk_callback_lock);
+}
 
-	spin_lock_bh(&xprt->transport_lock);
-	if (xprt->snd_task)
-		rpc_wake_up_task(xprt->snd_task);
-	spin_unlock_bh(&xprt->transport_lock);
-out:
+/**
+ * xs_tcp_write_space - callback invoked when socket buffer space
+ *                             becomes available
+ * @sk: socket whose state has changed
+ *
+ * Called when more output buffer space is available for this socket.
+ * We try not to wake our writers until they can make "significant"
+ * progress, otherwise we'll waste resources thrashing kernel_sendmsg
+ * with a bunch of small requests.
+ */
+static void xs_tcp_write_space(struct sock *sk)
+{
+	read_lock(&sk->sk_callback_lock);
+
+	/* from net/core/stream.c:sk_stream_write_space */
+	if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk)) {
+		struct socket *sock;
+		struct rpc_xprt *xprt;
+
+		if (unlikely(!(sock = sk->sk_socket)))
+			goto out;
+		if (unlikely(!(xprt = xprt_from_sock(sk))))
+			goto out;
+		if (unlikely(!test_and_clear_bit(SOCK_NOSPACE, &sock->flags)))
+			goto out;
+
+		xprt_write_space(xprt);
+	}
+
+ out:
 	read_unlock(&sk->sk_callback_lock);
 }
 
@@ -855,15 +876,16 @@ static void xs_bind(struct rpc_xprt *xprt, struct socket *sock)
 	xprt->old_write_space = sk->sk_write_space;
 	if (xprt->prot == IPPROTO_UDP) {
 		sk->sk_data_ready = xs_udp_data_ready;
+		sk->sk_write_space = xs_udp_write_space;
 		sk->sk_no_check = UDP_CSUM_NORCV;
 		xprt_set_connected(xprt);
 	} else {
 		tcp_sk(sk)->nonagle = 1;	/* disable Nagle's algorithm */
 		sk->sk_data_ready = xs_tcp_data_ready;
 		sk->sk_state_change = xs_tcp_state_change;
+		sk->sk_write_space = xs_tcp_write_space;
 		xprt_clear_connected(xprt);
 	}
-	sk->sk_write_space = xs_write_space;
 
 	/* Reset to new socket */
 	xprt->sock = sock;