[PATCH] RPC: separate TCP and UDP socket write paths

 Split the RPC client's main socket write path into a TCP version and a UDP
 version to eliminate another dependency on the "xprt->stream" variable.

 Compiler optimization removes unneeded code from xs_sendpages, as this
 function is now called with some constant arguments.

 We can now cleanly perform transport protocol-specific return code testing
 and error recovery in each path.

 Test-plan:
 Millions of fsx operations.  Performance characterization such as
 "sio" or "iozone".  Examine oprofile results for any changes before and
 after this patch is applied.

 Version: Thu, 11 Aug 2005 16:08:46 -0400

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

net/sunrpc/xprtsock.c

@@ -40,6 +40,12 @@
  */
 #define XS_MAX_RESVPORT		(800U)
 
+/*
+ * How many times to try sending a request on a socket before waiting
+ * for the socket buffer to clear.
+ */
+#define XS_SENDMSG_RETRY	(10U)
+
 #ifdef RPC_DEBUG
 # undef  RPC_DEBUG_DATA
 # define RPCDBG_FACILITY	RPCDBG_TRANS
@@ -114,13 +120,18 @@ static int xs_send_tail(struct socket *sock, struct xdr_buf *xdr, unsigned int b
  * @base: starting position in the buffer
  *
  */
-static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
+static inline int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 {
 	struct page **ppage = xdr->pages;
 	unsigned int len, pglen = xdr->page_len;
 	int err, ret = 0;
 	ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
 
+	if (unlikely(!sock))
+		return -ENOTCONN;
+
+	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
+
 	len = xdr->head[0].iov_len;
 	if (base < len || (addr != NULL && base == 0)) {
 		err = xs_send_head(sock, addr, addrlen, xdr, base, len);
@@ -187,140 +198,162 @@ out:
 }
 
 /**
- * xs_sendmsg - write an RPC request to a socket
- * @xprt: generic transport
- * @req: the RPC request to write
+ * xs_nospace - place task on wait queue if transmit was incomplete
+ * @task: task to put to sleep
  *
  */
-static int xs_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
+static void xs_nospace(struct rpc_task *task)
 {
-	struct socket *sock = xprt->sock;
-	struct xdr_buf *xdr = &req->rq_snd_buf;
-	struct sockaddr *addr = NULL;
-	int addrlen = 0;
-	unsigned int skip;
-	int result;
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_xprt *xprt = req->rq_xprt;
 
-	if (!sock)
-		return -ENOTCONN;
+	dprintk("RPC: %4d xmit incomplete (%u left of %u)\n",
+			task->tk_pid, req->rq_slen - req->rq_bytes_sent,
+			req->rq_slen);
+
+	if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
+		/* 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))
+			xprt_wait_for_buffer_space(task);
+
+		spin_unlock_bh(&xprt->transport_lock);
+	} else
+		/* Keep holding the socket if it is blocked */
+		rpc_delay(task, HZ>>4);
+}
+
+/**
+ * xs_udp_send_request - write an RPC request to a UDP socket
+ * @task: address of RPC task that manages the state of an RPC request
+ *
+ * Return values:
+ *        0:	The request has been sent
+ *   EAGAIN:	The socket was blocked, please call again later to
+ *		complete the request
+ * ENOTCONN:	Caller needs to invoke connect logic then call again
+ *    other:	Some other error occured, the request was not sent
+ */
+static int xs_udp_send_request(struct rpc_task *task)
+{
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_xprt *xprt = req->rq_xprt;
+	struct xdr_buf *xdr = &req->rq_snd_buf;
+	int status;
 
 	xs_pktdump("packet data:",
 				req->rq_svec->iov_base,
 				req->rq_svec->iov_len);
 
-	/* For UDP, we need to provide an address */
-	if (!xprt->stream) {
-		addr = (struct sockaddr *) &xprt->addr;
-		addrlen = sizeof(xprt->addr);
-	}
-	/* Don't repeat bytes */
-	skip = req->rq_bytes_sent;
+	req->rq_xtime = jiffies;
+	status = xs_sendpages(xprt->sock, (struct sockaddr *) &xprt->addr,
+				sizeof(xprt->addr), xdr, req->rq_bytes_sent);
 
-	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
-	result = xs_sendpages(sock, addr, addrlen, xdr, skip);
+	dprintk("RPC:      xs_udp_send_request(%u) = %d\n",
+			xdr->len - req->rq_bytes_sent, status);
 
-	dprintk("RPC:      xs_sendmsg(%d) = %d\n", xdr->len - skip, result);
+	if (likely(status >= (int) req->rq_slen))
+		return 0;
 
-	if (result >= 0)
-		return result;
+	/* Still some bytes left; set up for a retry later. */
+	if (status > 0)
+		status = -EAGAIN;
 
-	switch (result) {
+	switch (status) {
+	case -ENETUNREACH:
+	case -EPIPE:
 	case -ECONNREFUSED:
 		/* When the server has died, an ICMP port unreachable message
 		 * prompts ECONNREFUSED. */
-	case -EAGAIN:
 		break;
-	case -ECONNRESET:
-	case -ENOTCONN:
-	case -EPIPE:
-		/* connection broken */
-		if (xprt->stream)
-			result = -ENOTCONN;
+	case -EAGAIN:
+		xs_nospace(task);
 		break;
 	default:
+		dprintk("RPC:      sendmsg returned unrecognized error %d\n",
+			-status);
 		break;
 	}
-	return result;
+
+	return status;
 }
 
 /**
- * xs_send_request - write an RPC request to a socket
+ * xs_tcp_send_request - write an RPC request to a TCP socket
  * @task: address of RPC task that manages the state of an RPC request
  *
  * Return values:
- *      0:  The request has been sent
- * EAGAIN:  The socket was blocked, please call again later to
- *          complete the request
- *  other:  Some other error occured, the request was not sent
+ *        0:	The request has been sent
+ *   EAGAIN:	The socket was blocked, please call again later to
+ *		complete the request
+ * ENOTCONN:	Caller needs to invoke connect logic then call again
+ *    other:	Some other error occured, the request was not sent
  *
  * XXX: In the case of soft timeouts, should we eventually give up
- *      if the socket is not able to make progress?
+ *	if sendmsg is not able to make progress?
  */
-static int xs_send_request(struct rpc_task *task)
+static int xs_tcp_send_request(struct rpc_task *task)
 {
 	struct rpc_rqst *req = task->tk_rqstp;
 	struct rpc_xprt *xprt = req->rq_xprt;
+	struct xdr_buf *xdr = &req->rq_snd_buf;
+	u32 *marker = req->rq_svec[0].iov_base;
 	int status, retry = 0;
 
-	/* set up everything as needed. */
 	/* Write the record marker */
-	if (xprt->stream) {
-		u32 *marker = req->rq_svec[0].iov_base;
+	*marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
 
-		*marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
-	}
+	xs_pktdump("packet data:",
+				req->rq_svec->iov_base,
+				req->rq_svec->iov_len);
 
 	/* Continue transmitting the packet/record. We must be careful
 	 * to cope with writespace callbacks arriving _after_ we have
-	 * called sendmsg().
-	 */
+	 * called sendmsg(). */
 	while (1) {
 		req->rq_xtime = jiffies;
-		status = xs_sendmsg(xprt, req);
+		status = xs_sendpages(xprt->sock, NULL, 0, xdr,
+						req->rq_bytes_sent);
 
-		if (status < 0)
-			break;
+		dprintk("RPC:      xs_tcp_send_request(%u) = %d\n",
+				xdr->len - req->rq_bytes_sent, status);
 
-		if (xprt->stream) {
-			req->rq_bytes_sent += status;
-
-			/* If we've sent the entire packet, immediately
-			 * reset the count of bytes sent. */
-			if (req->rq_bytes_sent >= req->rq_slen) {
-				req->rq_bytes_sent = 0;
-				return 0;
-			}
-		} else {
-			if (status >= req->rq_slen)
-				return 0;
-			status = -EAGAIN;
+		if (unlikely(status < 0))
 			break;
-		}
 
-		dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
-				task->tk_pid, req->rq_slen - req->rq_bytes_sent,
-				req->rq_slen);
+		/* If we've sent the entire packet, immediately
+		 * reset the count of bytes sent. */
+		req->rq_bytes_sent += status;
+		if (likely(req->rq_bytes_sent >= req->rq_slen)) {
+			req->rq_bytes_sent = 0;
+			return 0;
+		}
 
 		status = -EAGAIN;
-		if (retry++ > 50)
+		if (retry++ > XS_SENDMSG_RETRY)
 			break;
 	}
 
-	if (status == -EAGAIN) {
-		if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
-			/* 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))
-				xprt_wait_for_buffer_space(task);
-			spin_unlock_bh(&xprt->transport_lock);
-			return status;
-		}
-		/* Keep holding the socket if it is blocked */
-		rpc_delay(task, HZ>>4);
+	switch (status) {
+	case -EAGAIN:
+		xs_nospace(task);
+		break;
+	case -ECONNREFUSED:
+	case -ECONNRESET:
+	case -ENOTCONN:
+	case -EPIPE:
+		status = -ENOTCONN;
+		break;
+	default:
+		dprintk("RPC:      sendmsg returned unrecognized error %d\n",
+			-status);
+		break;
 	}
+
 	return status;
 }
 
@@ -992,10 +1025,18 @@ static void xs_connect(struct rpc_task *task)
 	}
 }
 
-static struct rpc_xprt_ops xs_ops = {
+static struct rpc_xprt_ops xs_udp_ops = {
+	.set_buffer_size	= xs_set_buffer_size,
+	.connect		= xs_connect,
+	.send_request		= xs_udp_send_request,
+	.close			= xs_close,
+	.destroy		= xs_destroy,
+};
+
+static struct rpc_xprt_ops xs_tcp_ops = {
 	.set_buffer_size	= xs_set_buffer_size,
 	.connect		= xs_connect,
-	.send_request		= xs_send_request,
+	.send_request		= xs_tcp_send_request,
 	.close			= xs_close,
 	.destroy		= xs_destroy,
 };
@@ -1033,7 +1074,7 @@ int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
 
 	INIT_WORK(&xprt->connect_worker, xs_udp_connect_worker, xprt);
 
-	xprt->ops = &xs_ops;
+	xprt->ops = &xs_udp_ops;
 
 	if (to)
 		xprt->timeout = *to;
@@ -1072,7 +1113,7 @@ int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
 
 	INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);
 
-	xprt->ops = &xs_ops;
+	xprt->ops = &xs_tcp_ops;
 
 	if (to)
 		xprt->timeout = *to;