[PATCH] RPC: introduce client-side transport switch

 Move the bulk of client-side socket-specific code into a separate source
 file, net/sunrpc/xprtsock.c.

 Test-plan:
 Millions of fsx operations.  Performance characterization such as "sio" or
 "iozone".  Destructive testing (unplugging the network temporarily, server
 reboots).  Connectathon with v2, v3, and v4.

 Version: Thu, 11 Aug 2005 16:03:38 -0400

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

8 files changed, 1101 insertions, 989 deletions

diff --git a/include/linux/sunrpc/xdr.h b/include/linux/sunrpc/xdr.h
index d8b7656bca41..5da968729cf8 100644
--- a/include/linux/sunrpc/xdr.h
+++ b/include/linux/sunrpc/xdr.h
@@ -165,11 +165,6 @@ extern int csum_partial_copy_to_xdr(struct xdr_buf *, struct sk_buff *);
 extern ssize_t xdr_partial_copy_from_skb(struct xdr_buf *, unsigned int,
 		skb_reader_t *, skb_read_actor_t);
 
-struct socket;
-struct sockaddr;
-extern int xdr_sendpages(struct socket *, struct sockaddr *, int,
-		struct xdr_buf *, unsigned int, int);
-
 extern int xdr_encode_word(struct xdr_buf *, int, u32);
 extern int xdr_decode_word(struct xdr_buf *, int, u32 *);
 
diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index e618c1649814..d82b47ab73cb 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -59,7 +59,13 @@ extern unsigned int xprt_tcp_slot_table_entries;
  */
 #define RPC_REESTABLISH_TIMEOUT	(15*HZ)
 
-/* RPC call and reply header size as number of 32bit words (verifier
+/*
+ * RPC transport idle timeout.
+ */
+#define RPC_IDLE_DISCONNECT_TIMEOUT	(5*60*HZ)
+
+/*
+ * RPC call and reply header size as number of 32bit words (verifier
  * size computed separately)
  */
 #define RPC_CALLHDRSIZE		6
@@ -121,12 +127,19 @@ struct rpc_rqst {
 #define rq_svec			rq_snd_buf.head
 #define rq_slen			rq_snd_buf.len
 
-#define XPRT_LAST_FRAG		(1 << 0)
-#define XPRT_COPY_RECM		(1 << 1)
-#define XPRT_COPY_XID		(1 << 2)
-#define XPRT_COPY_DATA		(1 << 3)
+struct rpc_task;
+struct rpc_xprt;
+
+struct rpc_xprt_ops {
+	void		(*set_buffer_size)(struct rpc_xprt *xprt);
+	void		(*connect)(struct rpc_task *task);
+	int		(*send_request)(struct rpc_task *task);
+	void		(*close)(struct rpc_xprt *xprt);
+	void		(*destroy)(struct rpc_xprt *xprt);
+};
 
 struct rpc_xprt {
+	struct rpc_xprt_ops *	ops;		/* transport methods */
 	struct socket *		sock;		/* BSD socket layer */
 	struct sock *		inet;		/* INET layer */
 
@@ -199,14 +212,22 @@ struct rpc_xprt {
 	wait_queue_head_t	cong_wait;
 };
 
+#define XPRT_LAST_FRAG		(1 << 0)
+#define XPRT_COPY_RECM		(1 << 1)
+#define XPRT_COPY_XID		(1 << 2)
+#define XPRT_COPY_DATA		(1 << 3)
+
 #ifdef __KERNEL__
 
 struct rpc_xprt *	xprt_create_proto(int proto, struct sockaddr_in *addr,
 					struct rpc_timeout *toparms);
+void			xprt_disconnect(struct rpc_xprt *);
 int			xprt_destroy(struct rpc_xprt *);
 void			xprt_set_timeout(struct rpc_timeout *, unsigned int,
 					unsigned long);
-
+struct rpc_rqst *	xprt_lookup_rqst(struct rpc_xprt *, u32);
+void			xprt_complete_rqst(struct rpc_xprt *,
+					struct rpc_rqst *, int);
 void			xprt_reserve(struct rpc_task *);
 int			xprt_prepare_transmit(struct rpc_task *);
 void			xprt_transmit(struct rpc_task *);
@@ -214,7 +235,10 @@ void			xprt_receive(struct rpc_task *);
 int			xprt_adjust_timeout(struct rpc_rqst *req);
 void			xprt_release(struct rpc_task *);
 void			xprt_connect(struct rpc_task *);
-void			xprt_sock_setbufsize(struct rpc_xprt *);
+int			xs_setup_udp(struct rpc_xprt *,
+					struct rpc_timeout *);
+int			xs_setup_tcp(struct rpc_xprt *,
+					struct rpc_timeout *);
 
 #define XPRT_LOCKED	0
 #define XPRT_CONNECT	1
diff --git a/net/sunrpc/Makefile b/net/sunrpc/Makefile
index f0a955627177..cdcab9ca4c60 100644
--- a/net/sunrpc/Makefile
+++ b/net/sunrpc/Makefile
@@ -6,7 +6,7 @@
 obj-$(CONFIG_SUNRPC) += sunrpc.o
 obj-$(CONFIG_SUNRPC_GSS) += auth_gss/
 
-sunrpc-y := clnt.o xprt.o socklib.o sched.o \
+sunrpc-y := clnt.o xprt.o socklib.o xprtsock.o sched.o \
 	    auth.o auth_null.o auth_unix.o \
 	    svc.o svcsock.o svcauth.o svcauth_unix.o \
 	    pmap_clnt.o timer.o xdr.o \
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 2d3cf0a52d82..ab50c3c9e6a8 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -525,8 +525,7 @@ rpc_setbufsize(struct rpc_clnt *clnt, unsigned int sndsize, unsigned int rcvsize
 	xprt->rcvsize = 0;
 	if (rcvsize)
 		xprt->rcvsize = rcvsize + RPC_SLACK_SPACE;
-	if (xprt_connected(xprt))
-		xprt_sock_setbufsize(xprt);
+	xprt->ops->set_buffer_size(xprt);
 }
 
 /*
diff --git a/net/sunrpc/sysctl.c b/net/sunrpc/sysctl.c
index 1b9616a12e24..ef483262f17f 100644
--- a/net/sunrpc/sysctl.c
+++ b/net/sunrpc/sysctl.c
@@ -119,6 +119,9 @@ done:
 	return 0;
 }
 
+unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
+unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
+
 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
 
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index 9cc12aeed22c..32df43372ee9 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -6,15 +6,12 @@
  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
  */
 
+#include <linux/module.h>
 #include <linux/types.h>
-#include <linux/socket.h>
 #include <linux/string.h>
 #include <linux/kernel.h>
 #include <linux/pagemap.h>
 #include <linux/errno.h>
-#include <linux/in.h>
-#include <linux/net.h>
-#include <net/sock.h>
 #include <linux/sunrpc/xdr.h>
 #include <linux/sunrpc/msg_prot.h>
 
@@ -177,103 +174,6 @@ xdr_inline_pages(struct xdr_buf *xdr, unsigned int offset,
 }
 
 
-int
-xdr_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
-		struct xdr_buf *xdr, unsigned int base, int msgflags)
-{
-	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);
-
-	len = xdr->head[0].iov_len;
-	if (base < len || (addr != NULL && base == 0)) {
-		struct kvec iov = {
-			.iov_base = xdr->head[0].iov_base + base,
-			.iov_len  = len - base,
-		};
-		struct msghdr msg = {
-			.msg_name    = addr,
-			.msg_namelen = addrlen,
-			.msg_flags   = msgflags,
-		};
-		if (xdr->len > len)
-			msg.msg_flags |= MSG_MORE;
-
-		if (iov.iov_len != 0)
-			err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
-		else
-			err = kernel_sendmsg(sock, &msg, NULL, 0, 0);
-		if (ret == 0)
-			ret = err;
-		else if (err > 0)
-			ret += err;
-		if (err != iov.iov_len)
-			goto out;
-		base = 0;
-	} else
-		base -= len;
-
-	if (pglen == 0)
-		goto copy_tail;
-	if (base >= pglen) {
-		base -= pglen;
-		goto copy_tail;
-	}
-	if (base || xdr->page_base) {
-		pglen -= base;
-		base  += xdr->page_base;
-		ppage += base >> PAGE_CACHE_SHIFT;
-		base &= ~PAGE_CACHE_MASK;
-	}
-
-	sendpage = sock->ops->sendpage ? : sock_no_sendpage;
-	do {
-		int flags = msgflags;
-
-		len = PAGE_CACHE_SIZE;
-		if (base)
-			len -= base;
-		if (pglen < len)
-			len = pglen;
-
-		if (pglen != len || xdr->tail[0].iov_len != 0)
-			flags |= MSG_MORE;
-
-		/* Hmm... We might be dealing with highmem pages */
-		if (PageHighMem(*ppage))
-			sendpage = sock_no_sendpage;
-		err = sendpage(sock, *ppage, base, len, flags);
-		if (ret == 0)
-			ret = err;
-		else if (err > 0)
-			ret += err;
-		if (err != len)
-			goto out;
-		base = 0;
-		ppage++;
-	} while ((pglen -= len) != 0);
-copy_tail:
-	len = xdr->tail[0].iov_len;
-	if (base < len) {
-		struct kvec iov = {
-			.iov_base = xdr->tail[0].iov_base + base,
-			.iov_len  = len - base,
-		};
-		struct msghdr msg = {
-			.msg_flags   = msgflags,
-		};
-		err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
-		if (ret == 0)
-			ret = err;
-		else if (err > 0)
-			ret += err;
-	}
-out:
-	return ret;
-}
-
-
 /*
  * Helper routines for doing 'memmove' like operations on a struct xdr_buf
  *
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 67444f494fea..4342acf4d1cd 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -32,37 +32,16 @@
  *  tasks that rely on callbacks.
  *
  *  Copyright (C) 1995-1997, Olaf Kirch <okir@monad.swb.de>
- *
- *  TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
- *  TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
- *  TCP NFS related read + write fixes
- *   (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
- *
- *  Rewrite of larges part of the code in order to stabilize TCP stuff.
- *  Fix behaviour when socket buffer is full.
- *   (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  */
 
+#include <linux/module.h>
+
 #include <linux/types.h>
-#include <linux/slab.h>
-#include <linux/capability.h>
-#include <linux/sched.h>
-#include <linux/errno.h>
-#include <linux/socket.h>
-#include <linux/in.h>
-#include <linux/net.h>
-#include <linux/mm.h>
-#include <linux/udp.h>
-#include <linux/tcp.h>
-#include <linux/sunrpc/clnt.h>
-#include <linux/file.h>
+#include <linux/interrupt.h>
 #include <linux/workqueue.h>
 #include <linux/random.h>
 
-#include <net/sock.h>
-#include <net/checksum.h>
-#include <net/udp.h>
-#include <net/tcp.h>
+#include <linux/sunrpc/clnt.h>
 
 /*
  * Local variables
@@ -74,64 +53,17 @@
 #endif
 
 #define XPRT_MAX_BACKOFF	(8)
-#define XPRT_IDLE_TIMEOUT	(5*60*HZ)
-#define XPRT_MAX_RESVPORT	(800)
 
 /*
  * Local functions
  */
 static void	xprt_request_init(struct rpc_task *, struct rpc_xprt *);
 static inline void	do_xprt_reserve(struct rpc_task *);
-static void	xprt_disconnect(struct rpc_xprt *);
 static void	xprt_connect_status(struct rpc_task *task);
-static struct rpc_xprt * xprt_setup(int proto, struct sockaddr_in *ap,
-						struct rpc_timeout *to);
-static struct socket *xprt_create_socket(struct rpc_xprt *, int, int);
-static void	xprt_bind_socket(struct rpc_xprt *, struct socket *);
 static int      __xprt_get_cong(struct rpc_xprt *, struct rpc_task *);
 
 static int	xprt_clear_backlog(struct rpc_xprt *xprt);
 
-#ifdef RPC_DEBUG_DATA
-/*
- * Print the buffer contents (first 128 bytes only--just enough for
- * diropres return).
- */
-static void
-xprt_pktdump(char *msg, u32 *packet, unsigned int count)
-{
-	u8	*buf = (u8 *) packet;
-	int	j;
-
-	dprintk("RPC:      %s\n", msg);
-	for (j = 0; j < count && j < 128; j += 4) {
-		if (!(j & 31)) {
-			if (j)
-				dprintk("\n");
-			dprintk("0x%04x ", j);
-		}
-		dprintk("%02x%02x%02x%02x ",
-			buf[j], buf[j+1], buf[j+2], buf[j+3]);
-	}
-	dprintk("\n");
-}
-#else
-static inline void
-xprt_pktdump(char *msg, u32 *packet, unsigned int count)
-{
-	/* NOP */
-}
-#endif
-
-/*
- * Look up RPC transport given an INET socket
- */
-static inline struct rpc_xprt *
-xprt_from_sock(struct sock *sk)
-{
-	return (struct rpc_xprt *) sk->sk_user_data;
-}
-
 /*
  * Serialize write access to sockets, in order to prevent different
  * requests from interfering with each other.
@@ -235,62 +167,6 @@ xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
 }
 
 /*
- * Write data to socket.
- */
-static inline int
-xprt_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
-{
-	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;
-
-	if (!sock)
-		return -ENOTCONN;
-
-	xprt_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);
-	}
-	/* Dont repeat bytes */
-	skip = req->rq_bytes_sent;
-
-	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
-	result = xdr_sendpages(sock, addr, addrlen, xdr, skip, MSG_DONTWAIT);
-
-	dprintk("RPC:      xprt_sendmsg(%d) = %d\n", xdr->len - skip, result);
-
-	if (result >= 0)
-		return result;
-
-	switch (result) {
-	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;
-		break;
-	default:
-		printk(KERN_NOTICE "RPC: sendmsg returned error %d\n", -result);
-	}
-	return result;
-}
-
-/*
  * Van Jacobson congestion avoidance. Check if the congestion window
  * overflowed. Put the task to sleep if this is the case.
  */
@@ -405,48 +281,20 @@ int xprt_adjust_timeout(struct rpc_rqst *req)
 	return status;
 }
 
-/*
- * Close down a transport socket
- */
-static void
-xprt_close(struct rpc_xprt *xprt)
-{
-	struct socket	*sock = xprt->sock;
-	struct sock	*sk = xprt->inet;
-
-	if (!sk)
-		return;
-
-	write_lock_bh(&sk->sk_callback_lock);
-	xprt->inet = NULL;
-	xprt->sock = NULL;
-
-	sk->sk_user_data    = NULL;
-	sk->sk_data_ready   = xprt->old_data_ready;
-	sk->sk_state_change = xprt->old_state_change;
-	sk->sk_write_space  = xprt->old_write_space;
-	write_unlock_bh(&sk->sk_callback_lock);
-
-	sk->sk_no_check	 = 0;
-
-	sock_release(sock);
-}
-
 static void
 xprt_socket_autoclose(void *args)
 {
 	struct rpc_xprt *xprt = (struct rpc_xprt *)args;
 
 	xprt_disconnect(xprt);
-	xprt_close(xprt);
+	xprt->ops->close(xprt);
 	xprt_release_write(xprt, NULL);
 }
 
 /*
  * Mark a transport as disconnected
  */
-static void
-xprt_disconnect(struct rpc_xprt *xprt)
+void xprt_disconnect(struct rpc_xprt *xprt)
 {
 	dprintk("RPC:      disconnected transport %p\n", xprt);
 	spin_lock_bh(&xprt->sock_lock);
@@ -479,57 +327,6 @@ out_abort:
 	spin_unlock(&xprt->sock_lock);
 }
 
-static void xprt_socket_connect(void *args)
-{
-	struct rpc_xprt *xprt = (struct rpc_xprt *)args;
-	struct socket *sock = xprt->sock;
-	int status = -EIO;
-
-	if (xprt->shutdown || xprt->addr.sin_port == 0)
-		goto out;
-
-	/*
-	 * Start by resetting any existing state
-	 */
-	xprt_close(xprt);
-	sock = xprt_create_socket(xprt, xprt->prot, xprt->resvport);
-	if (sock == NULL) {
-		/* couldn't create socket or bind to reserved port;
-		 * this is likely a permanent error, so cause an abort */
-		goto out;
-	}
-	xprt_bind_socket(xprt, sock);
-	xprt_sock_setbufsize(xprt);
-
-	status = 0;
-	if (!xprt->stream)
-		goto out;
-
-	/*
-	 * Tell the socket layer to start connecting...
-	 */
-	status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
-			sizeof(xprt->addr), O_NONBLOCK);
-	dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
-			xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
-	if (status < 0) {
-		switch (status) {
-			case -EINPROGRESS:
-			case -EALREADY:
-				goto out_clear;
-		}
-	}
-out:
-	if (status < 0)
-		rpc_wake_up_status(&xprt->pending, status);
-	else
-		rpc_wake_up(&xprt->pending);
-out_clear:
-	smp_mb__before_clear_bit();
-	clear_bit(XPRT_CONNECTING, &xprt->sockstate);
-	smp_mb__after_clear_bit();
-}
-
 /*
  * Attempt to connect a TCP socket.
  *
@@ -552,30 +349,16 @@ void xprt_connect(struct rpc_task *task)
 	if (!xprt_lock_write(xprt, task))
 		return;
 	if (xprt_connected(xprt))
-		goto out_write;
+		xprt_release_write(xprt, task);
+	else {
+		if (task->tk_rqstp)
+			task->tk_rqstp->rq_bytes_sent = 0;
 
-	if (task->tk_rqstp)
-		task->tk_rqstp->rq_bytes_sent = 0;
-
-	task->tk_timeout = RPC_CONNECT_TIMEOUT;
-	rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
-	if (!test_and_set_bit(XPRT_CONNECTING, &xprt->sockstate)) {
-		/* Note: if we are here due to a dropped connection
-		 * 	 we delay reconnecting by RPC_REESTABLISH_TIMEOUT/HZ
-		 * 	 seconds
-		 */
-		if (xprt->sock != NULL)
-			schedule_delayed_work(&xprt->sock_connect,
-					RPC_REESTABLISH_TIMEOUT);
-		else {
-			schedule_work(&xprt->sock_connect);
-			if (!RPC_IS_ASYNC(task))
-				flush_scheduled_work();
-		}
+		task->tk_timeout = RPC_CONNECT_TIMEOUT;
+		rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
+		xprt->ops->connect(task);
 	}
 	return;
- out_write:
-	xprt_release_write(xprt, task);
 }
 
 /*
@@ -624,8 +407,7 @@ xprt_connect_status(struct rpc_task *task)
 /*
  * Look up the RPC request corresponding to a reply, and then lock it.
  */
-static inline struct rpc_rqst *
-xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
+struct rpc_rqst *xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 {
 	struct list_head *pos;
 	struct rpc_rqst	*req = NULL;
@@ -644,8 +426,7 @@ xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
  * Complete reply received.
  * The TCP code relies on us to remove the request from xprt->pending.
  */
-static void
-xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
+void xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 {
 	struct rpc_task	*task = req->rq_task;
 	struct rpc_clnt *clnt = task->tk_client;
@@ -692,409 +473,6 @@ xprt_complete_rqst(struct rpc_xprt *xprt, struct rpc_rqst *req, int copied)
 }
 
 /*
- * Input handler for RPC replies. Called from a bottom half and hence
- * atomic.
- */
-static void
-udp_data_ready(struct sock *sk, int len)
-{
-	struct rpc_task	*task;
-	struct rpc_xprt	*xprt;
-	struct rpc_rqst *rovr;
-	struct sk_buff	*skb;
-	int err, repsize, copied;
-	u32 _xid, *xp;
-
-	read_lock(&sk->sk_callback_lock);
-	dprintk("RPC:      udp_data_ready...\n");
-	if (!(xprt = xprt_from_sock(sk))) {
-		printk("RPC:      udp_data_ready request not found!\n");
-		goto out;
-	}
-
-	dprintk("RPC:      udp_data_ready client %p\n", xprt);
-
-	if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
-		goto out;
-
-	if (xprt->shutdown)
-		goto dropit;
-
-	repsize = skb->len - sizeof(struct udphdr);
-	if (repsize < 4) {
-		printk("RPC: impossible RPC reply size %d!\n", repsize);
-		goto dropit;
-	}
-
-	/* Copy the XID from the skb... */
-	xp = skb_header_pointer(skb, sizeof(struct udphdr),
-				sizeof(_xid), &_xid);
-	if (xp == NULL)
-		goto dropit;
-
-	/* Look up and lock the request corresponding to the given XID */
-	spin_lock(&xprt->sock_lock);
-	rovr = xprt_lookup_rqst(xprt, *xp);
-	if (!rovr)
-		goto out_unlock;
-	task = rovr->rq_task;
-
-	dprintk("RPC: %4d received reply\n", task->tk_pid);
-
-	if ((copied = rovr->rq_private_buf.buflen) > repsize)
-		copied = repsize;
-
-	/* Suck it into the iovec, verify checksum if not done by hw. */
-	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
-		goto out_unlock;
-
-	/* Something worked... */
-	dst_confirm(skb->dst);
-
-	xprt_complete_rqst(xprt, rovr, copied);
-
- out_unlock:
-	spin_unlock(&xprt->sock_lock);
- dropit:
-	skb_free_datagram(sk, skb);
- out:
-	read_unlock(&sk->sk_callback_lock);
-}
-
-/*
- * Copy from an skb into memory and shrink the skb.
- */
-static inline size_t
-tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
-{
-	if (len > desc->count)
-		len = desc->count;
-	if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
-		dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
-				len, desc->count);
-		return 0;
-	}
-	desc->offset += len;
-	desc->count -= len;
-	dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
-			len, desc->count);
-	return len;
-}
-
-/*
- * TCP read fragment marker
- */
-static inline void
-tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
-{
-	size_t len, used;
-	char *p;
-
-	p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
-	len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
-	used = tcp_copy_data(desc, p, len);
-	xprt->tcp_offset += used;
-	if (used != len)
-		return;
-	xprt->tcp_reclen = ntohl(xprt->tcp_recm);
-	if (xprt->tcp_reclen & 0x80000000)
-		xprt->tcp_flags |= XPRT_LAST_FRAG;
-	else
-		xprt->tcp_flags &= ~XPRT_LAST_FRAG;
-	xprt->tcp_reclen &= 0x7fffffff;
-	xprt->tcp_flags &= ~XPRT_COPY_RECM;
-	xprt->tcp_offset = 0;
-	/* Sanity check of the record length */
-	if (xprt->tcp_reclen < 4) {
-		printk(KERN_ERR "RPC: Invalid TCP record fragment length\n");
-		xprt_disconnect(xprt);
-	}
-	dprintk("RPC:      reading TCP record fragment of length %d\n",
-			xprt->tcp_reclen);
-}
-
-static void
-tcp_check_recm(struct rpc_xprt *xprt)
-{
-	dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
-			xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
-	if (xprt->tcp_offset == xprt->tcp_reclen) {
-		xprt->tcp_flags |= XPRT_COPY_RECM;
-		xprt->tcp_offset = 0;
-		if (xprt->tcp_flags & XPRT_LAST_FRAG) {
-			xprt->tcp_flags &= ~XPRT_COPY_DATA;
-			xprt->tcp_flags |= XPRT_COPY_XID;
-			xprt->tcp_copied = 0;
-		}
-	}
-}
-
-/*
- * TCP read xid
- */
-static inline void
-tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
-{
-	size_t len, used;
-	char *p;
-
-	len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
-	dprintk("RPC:      reading XID (%Zu bytes)\n", len);
-	p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
-	used = tcp_copy_data(desc, p, len);
-	xprt->tcp_offset += used;
-	if (used != len)
-		return;
-	xprt->tcp_flags &= ~XPRT_COPY_XID;
-	xprt->tcp_flags |= XPRT_COPY_DATA;
-	xprt->tcp_copied = 4;
-	dprintk("RPC:      reading reply for XID %08x\n",
-						ntohl(xprt->tcp_xid));
-	tcp_check_recm(xprt);
-}
-
-/*
- * TCP read and complete request
- */
-static inline void
-tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
-{
-	struct rpc_rqst *req;
-	struct xdr_buf *rcvbuf;
-	size_t len;
-	ssize_t r;
-
-	/* Find and lock the request corresponding to this xid */
-	spin_lock(&xprt->sock_lock);
-	req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
-	if (!req) {
-		xprt->tcp_flags &= ~XPRT_COPY_DATA;
-		dprintk("RPC:      XID %08x request not found!\n",
-				ntohl(xprt->tcp_xid));
-		spin_unlock(&xprt->sock_lock);
-		return;
-	}
-
-	rcvbuf = &req->rq_private_buf;
-	len = desc->count;
-	if (len > xprt->tcp_reclen - xprt->tcp_offset) {
-		skb_reader_t my_desc;
-
-		len = xprt->tcp_reclen - xprt->tcp_offset;
-		memcpy(&my_desc, desc, sizeof(my_desc));
-		my_desc.count = len;
-		r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
-					  &my_desc, tcp_copy_data);
-		desc->count -= r;
-		desc->offset += r;
-	} else
-		r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
-					  desc, tcp_copy_data);
-
-	if (r > 0) {
-		xprt->tcp_copied += r;
-		xprt->tcp_offset += r;
-	}
-	if (r != len) {
-		/* Error when copying to the receive buffer,
-		 * usually because we weren't able to allocate
-		 * additional buffer pages. All we can do now
-		 * is turn off XPRT_COPY_DATA, so the request
-		 * will not receive any additional updates,
-		 * and time out.
-		 * Any remaining data from this record will
-		 * be discarded.
-		 */
-		xprt->tcp_flags &= ~XPRT_COPY_DATA;
-		dprintk("RPC:      XID %08x truncated request\n",
-				ntohl(xprt->tcp_xid));
-		dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
-				xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
-		goto out;
-	}
-
-	dprintk("RPC:      XID %08x read %Zd bytes\n",
-			ntohl(xprt->tcp_xid), r);
-	dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
-			xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
-
-	if (xprt->tcp_copied == req->rq_private_buf.buflen)
-		xprt->tcp_flags &= ~XPRT_COPY_DATA;
-	else if (xprt->tcp_offset == xprt->tcp_reclen) {
-		if (xprt->tcp_flags & XPRT_LAST_FRAG)
-			xprt->tcp_flags &= ~XPRT_COPY_DATA;
-	}
-
-out:
-	if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
-		dprintk("RPC: %4d received reply complete\n",
-				req->rq_task->tk_pid);
-		xprt_complete_rqst(xprt, req, xprt->tcp_copied);
-	}
-	spin_unlock(&xprt->sock_lock);
-	tcp_check_recm(xprt);
-}
-
-/*
- * TCP discard extra bytes from a short read
- */
-static inline void
-tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
-{
-	size_t len;
-
-	len = xprt->tcp_reclen - xprt->tcp_offset;
-	if (len > desc->count)
-		len = desc->count;
-	desc->count -= len;
-	desc->offset += len;
-	xprt->tcp_offset += len;
-	dprintk("RPC:      discarded %Zu bytes\n", len);
-	tcp_check_recm(xprt);
-}
-
-/*
- * TCP record receive routine
- * We first have to grab the record marker, then the XID, then the data.
- */
-static int
-tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
-		unsigned int offset, size_t len)
-{
-	struct rpc_xprt *xprt = rd_desc->arg.data;
-	skb_reader_t desc = {
-		.skb	= skb,
-		.offset	= offset,
-		.count	= len,
-		.csum	= 0
-       	};
-
-	dprintk("RPC:      tcp_data_recv\n");
-	do {
-		/* Read in a new fragment marker if necessary */
-		/* Can we ever really expect to get completely empty fragments? */
-		if (xprt->tcp_flags & XPRT_COPY_RECM) {
-			tcp_read_fraghdr(xprt, &desc);
-			continue;
-		}
-		/* Read in the xid if necessary */
-		if (xprt->tcp_flags & XPRT_COPY_XID) {
-			tcp_read_xid(xprt, &desc);
-			continue;
-		}
-		/* Read in the request data */
-		if (xprt->tcp_flags & XPRT_COPY_DATA) {
-			tcp_read_request(xprt, &desc);
-			continue;
-		}
-		/* Skip over any trailing bytes on short reads */
-		tcp_read_discard(xprt, &desc);
-	} while (desc.count);
-	dprintk("RPC:      tcp_data_recv done\n");
-	return len - desc.count;
-}
-
-static void tcp_data_ready(struct sock *sk, int bytes)
-{
-	struct rpc_xprt *xprt;
-	read_descriptor_t rd_desc;
-
-	read_lock(&sk->sk_callback_lock);
-	dprintk("RPC:      tcp_data_ready...\n");
-	if (!(xprt = xprt_from_sock(sk))) {
-		printk("RPC:      tcp_data_ready socket info not found!\n");
-		goto out;
-	}
-	if (xprt->shutdown)
-		goto out;
-
-	/* We use rd_desc to pass struct xprt to tcp_data_recv */
-	rd_desc.arg.data = xprt;
-	rd_desc.count = 65536;
-	tcp_read_sock(sk, &rd_desc, tcp_data_recv);
-out:
-	read_unlock(&sk->sk_callback_lock);
-}
-
-static void
-tcp_state_change(struct sock *sk)
-{
-	struct rpc_xprt	*xprt;
-
-	read_lock(&sk->sk_callback_lock);
-	if (!(xprt = xprt_from_sock(sk)))
-		goto out;
-	dprintk("RPC:      tcp_state_change client %p...\n", xprt);
-	dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
-				sk->sk_state, xprt_connected(xprt),
-				sock_flag(sk, SOCK_DEAD),
-				sock_flag(sk, SOCK_ZAPPED));
-
-	switch (sk->sk_state) {
-	case TCP_ESTABLISHED:
-		spin_lock_bh(&xprt->sock_lock);
-		if (!xprt_test_and_set_connected(xprt)) {
-			/* Reset TCP record info */
-			xprt->tcp_offset = 0;
-			xprt->tcp_reclen = 0;
-			xprt->tcp_copied = 0;
-			xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
-			rpc_wake_up(&xprt->pending);
-		}
-		spin_unlock_bh(&xprt->sock_lock);
-		break;
-	case TCP_SYN_SENT:
-	case TCP_SYN_RECV:
-		break;
-	default:
-		xprt_disconnect(xprt);
-		break;
-	}
- out:
-	read_unlock(&sk->sk_callback_lock);
-}
-
-/*
- * 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
- * with a bunch of small requests.
- */
-static void
-xprt_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))
-			goto out;
-	} else {
-		/* from net/core/sock.c:sock_def_write_space */
-		if (!sock_writeable(sk))
-			goto out;
-	}
-
-	if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
-		goto out;
-
-	spin_lock_bh(&xprt->sock_lock);
-	if (xprt->snd_task)
-		rpc_wake_up_task(xprt->snd_task);
-	spin_unlock_bh(&xprt->sock_lock);
-out:
-	read_unlock(&sk->sk_callback_lock);
-}
-
-/*
  * RPC receive timeout handler.
  */
 static void
@@ -1161,19 +539,10 @@ xprt_transmit(struct rpc_task *task)
 	struct rpc_clnt *clnt = task->tk_client;
 	struct rpc_rqst	*req = task->tk_rqstp;
 	struct rpc_xprt	*xprt = req->rq_xprt;
-	int status, retry = 0;
-
+	int status;
 
 	dprintk("RPC: %4d xprt_transmit(%u)\n", task->tk_pid, req->rq_slen);
 
-	/* 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)));
-	}
-
 	smp_rmb();
 	if (!req->rq_received) {
 		if (list_empty(&req->rq_list)) {
@@ -1191,41 +560,9 @@ xprt_transmit(struct rpc_task *task)
 	} else if (!req->rq_bytes_sent)
 		return;
 
-	/* Continue transmitting the packet/record. We must be careful
-	 * to cope with writespace callbacks arriving _after_ we have
-	 * called xprt_sendmsg().
-	 */
-	while (1) {
-		req->rq_xtime = jiffies;
-		status = xprt_sendmsg(xprt, req);
-
-		if (status < 0)
-			break;
-
-		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;
-				goto out_receive;
-			}
-		} else {
-			if (status >= req->rq_slen)
-				goto out_receive;
-			status = -EAGAIN;
-			break;
-		}
-
-		dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
-				task->tk_pid, req->rq_slen - req->rq_bytes_sent,
-				req->rq_slen);
-
-		status = -EAGAIN;
-		if (retry++ > 50)
-			break;
-	}
+	status = xprt->ops->send_request(task);
+	if (!status)
+		goto out_receive;
 
 	/* Note: at this point, task->tk_sleeping has not yet been set,
 	 *	 hence there is no danger of the waking up task being put on
@@ -1234,26 +571,10 @@ xprt_transmit(struct rpc_task *task)
 	task->tk_status = status;
 
 	switch (status) {
-	case -EAGAIN:
-		if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
-			/* Protect against races with xprt_write_space */
-			spin_lock_bh(&xprt->sock_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);
-			}
-			spin_unlock_bh(&xprt->sock_lock);
-			return;
-		}
-		/* Keep holding the socket if it is blocked */
-		rpc_delay(task, HZ>>4);
-		return;
 	case -ECONNREFUSED:
 		task->tk_timeout = RPC_REESTABLISH_TIMEOUT;
 		rpc_sleep_on(&xprt->sending, task, NULL, NULL);
+	case -EAGAIN:
 	case -ENOTCONN:
 		return;
 	default:
@@ -1367,7 +688,8 @@ xprt_release(struct rpc_task *task)
 		list_del(&req->rq_list);
 	xprt->last_used = jiffies;
 	if (list_empty(&xprt->recv) && !xprt->shutdown)
-		mod_timer(&xprt->timer, xprt->last_used + XPRT_IDLE_TIMEOUT);
+		mod_timer(&xprt->timer,
+				xprt->last_used + RPC_IDLE_DISCONNECT_TIMEOUT);
 	spin_unlock_bh(&xprt->sock_lock);
 	task->tk_rqstp = NULL;
 	memset(req, 0, sizeof(*req));	/* mark unused */
@@ -1381,18 +703,6 @@ xprt_release(struct rpc_task *task)
 }
 
 /*
- * Set default timeout parameters
- */
-static void
-xprt_default_timeout(struct rpc_timeout *to, int proto)
-{
-	if (proto == IPPROTO_UDP)
-		xprt_set_timeout(to, 5,  5 * HZ);
-	else
-		xprt_set_timeout(to, 2, 60 * HZ);
-}
-
-/*
  * Set constant timeout
  */
 void
@@ -1405,68 +715,51 @@ xprt_set_timeout(struct rpc_timeout *to, unsigned int retr, unsigned long incr)
 	to->to_exponential = 0;
 }
 
-unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
-unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
-
 /*
  * Initialize an RPC client
  */
 static struct rpc_xprt *
 xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 {
+	int result;
 	struct rpc_xprt	*xprt;
-	unsigned int entries;
-	size_t slot_table_size;
 	struct rpc_rqst	*req;
 
-	dprintk("RPC:      setting up %s transport...\n",
-				proto == IPPROTO_UDP? "UDP" : "TCP");
-
-	entries = (proto == IPPROTO_TCP)?
-		xprt_tcp_slot_table_entries : xprt_udp_slot_table_entries;
-
 	if ((xprt = kmalloc(sizeof(struct rpc_xprt), GFP_KERNEL)) == NULL)
 		return ERR_PTR(-ENOMEM);
 	memset(xprt, 0, sizeof(*xprt)); /* Nnnngh! */
-	xprt->max_reqs = entries;
-	slot_table_size = entries * sizeof(xprt->slot[0]);
-	xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
-	if (xprt->slot == NULL) {
-		kfree(xprt);
-		return ERR_PTR(-ENOMEM);
-	}
-	memset(xprt->slot, 0, slot_table_size);
 
 	xprt->addr = *ap;
-	xprt->prot = proto;
-	xprt->stream = (proto == IPPROTO_TCP)? 1 : 0;
-	if (xprt->stream) {
-		xprt->cwnd = RPC_MAXCWND(xprt);
-		xprt->nocong = 1;
-		xprt->max_payload = (1U << 31) - 1;
-	} else {
-		xprt->cwnd = RPC_INITCWND;
-		xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
+
+	switch (proto) {
+	case IPPROTO_UDP:
+		result = xs_setup_udp(xprt, to);
+		break;
+	case IPPROTO_TCP:
+		result = xs_setup_tcp(xprt, to);
+		break;
+	default:
+		printk(KERN_ERR "RPC: unrecognized transport protocol: %d\n",
+				proto);
+		result = -EIO;
+		break;
+	}
+	if (result) {
+		kfree(xprt);
+		return ERR_PTR(result);
 	}
+
 	spin_lock_init(&xprt->sock_lock);
 	spin_lock_init(&xprt->xprt_lock);
 	init_waitqueue_head(&xprt->cong_wait);
 
 	INIT_LIST_HEAD(&xprt->free);
 	INIT_LIST_HEAD(&xprt->recv);
-	INIT_WORK(&xprt->sock_connect, xprt_socket_connect, xprt);
 	INIT_WORK(&xprt->task_cleanup, xprt_socket_autoclose, xprt);
 	init_timer(&xprt->timer);
 	xprt->timer.function = xprt_init_autodisconnect;
 	xprt->timer.data = (unsigned long) xprt;
 	xprt->last_used = jiffies;
-	xprt->port = XPRT_MAX_RESVPORT;
-
-	/* Set timeout parameters */
-	if (to) {
-		xprt->timeout = *to;
-	} else
-		xprt_default_timeout(&xprt->timeout, xprt->prot);
 
 	rpc_init_wait_queue(&xprt->pending, "xprt_pending");
 	rpc_init_wait_queue(&xprt->sending, "xprt_sending");
@@ -1474,14 +767,11 @@ xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 	rpc_init_priority_wait_queue(&xprt->backlog, "xprt_backlog");
 
 	/* initialize free list */
-	for (req = &xprt->slot[entries-1]; req >= &xprt->slot[0]; req--)
+	for (req = &xprt->slot[xprt->max_reqs-1]; req >= &xprt->slot[0]; req--)
 		list_add(&req->rq_list, &xprt->free);
 
 	xprt_init_xid(xprt);
 
-	/* Check whether we want to use a reserved port */
-	xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
-
 	dprintk("RPC:      created transport %p with %u slots\n", xprt,
 			xprt->max_reqs);
 	
@@ -1489,120 +779,6 @@ xprt_setup(int proto, struct sockaddr_in *ap, struct rpc_timeout *to)
 }
 
 /*
- * Bind to a reserved port
- */
-static inline int xprt_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
-{
-	struct sockaddr_in myaddr = {
-		.sin_family = AF_INET,
-	};
-	int		err, port;
-
-	/* Were we already bound to a given port? Try to reuse it */
-	port = xprt->port;
-	do {
-		myaddr.sin_port = htons(port);
-		err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
-						sizeof(myaddr));
-		if (err == 0) {
-			xprt->port = port;
-			return 0;
-		}
-		if (--port == 0)
-			port = XPRT_MAX_RESVPORT;
-	} while (err == -EADDRINUSE && port != xprt->port);
-
-	printk("RPC: Can't bind to reserved port (%d).\n", -err);
-	return err;
-}
-
-static void
-xprt_bind_socket(struct rpc_xprt *xprt, struct socket *sock)
-{
-	struct sock	*sk = sock->sk;
-
-	if (xprt->inet)
-		return;
-
-	write_lock_bh(&sk->sk_callback_lock);
-	sk->sk_user_data = xprt;
-	xprt->old_data_ready = sk->sk_data_ready;
-	xprt->old_state_change = sk->sk_state_change;
-	xprt->old_write_space = sk->sk_write_space;
-	if (xprt->prot == IPPROTO_UDP) {
-		sk->sk_data_ready = udp_data_ready;
-		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 = tcp_data_ready;
-		sk->sk_state_change = tcp_state_change;
-		xprt_clear_connected(xprt);
-	}
-	sk->sk_write_space = xprt_write_space;
-
-	/* Reset to new socket */
-	xprt->sock = sock;
-	xprt->inet = sk;
-	write_unlock_bh(&sk->sk_callback_lock);
-
-	return;
-}
-
-/*
- * Set socket buffer length
- */
-void
-xprt_sock_setbufsize(struct rpc_xprt *xprt)
-{
-	struct sock *sk = xprt->inet;
-
-	if (xprt->stream)
-		return;
-	if (xprt->rcvsize) {
-		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
-		sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
-	}
-	if (xprt->sndsize) {
-		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
-		sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
-		sk->sk_write_space(sk);
-	}
-}
-
-/*
- * Datastream sockets are created here, but xprt_connect will create
- * and connect stream sockets.
- */
-static struct socket * xprt_create_socket(struct rpc_xprt *xprt, int proto, int resvport)
-{
-	struct socket	*sock;
-	int		type, err;
-
-	dprintk("RPC:      xprt_create_socket(%s %d)\n",
-			   (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
-
-	type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
-
-	if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) {
-		printk("RPC: can't create socket (%d).\n", -err);
-		return NULL;
-	}
-
-	/* If the caller has the capability, bind to a reserved port */
-	if (resvport && xprt_bindresvport(xprt, sock) < 0) {
-		printk("RPC: can't bind to reserved port.\n");
-		goto failed;
-	}
-
-	return sock;
-
-failed:
-	sock_release(sock);
-	return NULL;
-}
-
-/*
  * Create an RPC client transport given the protocol and peer address.
  */
 struct rpc_xprt *
@@ -1631,10 +807,6 @@ xprt_shutdown(struct rpc_xprt *xprt)
 	rpc_wake_up(&xprt->backlog);
 	wake_up(&xprt->cong_wait);
 	del_timer_sync(&xprt->timer);
-
-	/* synchronously wait for connect worker to finish */
-	cancel_delayed_work(&xprt->sock_connect);
-	flush_scheduled_work();
 }
 
 /*
@@ -1655,9 +827,7 @@ xprt_destroy(struct rpc_xprt *xprt)
 {
 	dprintk("RPC:      destroying transport %p\n", xprt);
 	xprt_shutdown(xprt);
-	xprt_disconnect(xprt);
-	xprt_close(xprt);
-	kfree(xprt->slot);
+	xprt->ops->destroy(xprt);
 	kfree(xprt);
 
 	return 0;
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
new file mode 100644
index 000000000000..fa1180ac4823
--- /dev/null
+++ b/net/sunrpc/xprtsock.c
@@ -0,0 +1,1021 @@
+/*
+ * linux/net/sunrpc/xprtsock.c
+ *
+ * Client-side transport implementation for sockets.
+ *
+ * TCP callback races fixes (C) 1998 Red Hat Software <alan@redhat.com>
+ * TCP send fixes (C) 1998 Red Hat Software <alan@redhat.com>
+ * TCP NFS related read + write fixes
+ *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
+ *
+ * Rewrite of larges part of the code in order to stabilize TCP stuff.
+ * Fix behaviour when socket buffer is full.
+ *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/capability.h>
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/errno.h>
+#include <linux/socket.h>
+#include <linux/in.h>
+#include <linux/net.h>
+#include <linux/mm.h>
+#include <linux/udp.h>
+#include <linux/tcp.h>
+#include <linux/sunrpc/clnt.h>
+#include <linux/file.h>
+
+#include <net/sock.h>
+#include <net/checksum.h>
+#include <net/udp.h>
+#include <net/tcp.h>
+
+#ifdef RPC_DEBUG
+# undef  RPC_DEBUG_DATA
+# define RPCDBG_FACILITY	RPCDBG_XPRT
+#endif
+
+#define XPRT_MAX_RESVPORT	(800)
+
+#ifdef RPC_DEBUG_DATA
+/*
+ * Print the buffer contents (first 128 bytes only--just enough for
+ * diropres return).
+ */
+static void
+xprt_pktdump(char *msg, u32 *packet, unsigned int count)
+{
+	u8	*buf = (u8 *) packet;
+	int	j;
+
+	dprintk("RPC:      %s\n", msg);
+	for (j = 0; j < count && j < 128; j += 4) {
+		if (!(j & 31)) {
+			if (j)
+				dprintk("\n");
+			dprintk("0x%04x ", j);
+		}
+		dprintk("%02x%02x%02x%02x ",
+			buf[j], buf[j+1], buf[j+2], buf[j+3]);
+	}
+	dprintk("\n");
+}
+#else
+static inline void
+xprt_pktdump(char *msg, u32 *packet, unsigned int count)
+{
+	/* NOP */
+}
+#endif
+
+/*
+ * Look up RPC transport given an INET socket
+ */
+static inline struct rpc_xprt *
+xprt_from_sock(struct sock *sk)
+{
+	return (struct rpc_xprt *) sk->sk_user_data;
+}
+
+static int
+xdr_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
+		struct xdr_buf *xdr, unsigned int base, int msgflags)
+{
+	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);
+
+	len = xdr->head[0].iov_len;
+	if (base < len || (addr != NULL && base == 0)) {
+		struct kvec iov = {
+			.iov_base = xdr->head[0].iov_base + base,
+			.iov_len  = len - base,
+		};
+		struct msghdr msg = {
+			.msg_name    = addr,
+			.msg_namelen = addrlen,
+			.msg_flags   = msgflags,
+		};
+		if (xdr->len > len)
+			msg.msg_flags |= MSG_MORE;
+
+		if (iov.iov_len != 0)
+			err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
+		else
+			err = kernel_sendmsg(sock, &msg, NULL, 0, 0);
+		if (ret == 0)
+			ret = err;
+		else if (err > 0)
+			ret += err;
+		if (err != iov.iov_len)
+			goto out;
+		base = 0;
+	} else
+		base -= len;
+
+	if (pglen == 0)
+		goto copy_tail;
+	if (base >= pglen) {
+		base -= pglen;
+		goto copy_tail;
+	}
+	if (base || xdr->page_base) {
+		pglen -= base;
+		base  += xdr->page_base;
+		ppage += base >> PAGE_CACHE_SHIFT;
+		base &= ~PAGE_CACHE_MASK;
+	}
+
+	sendpage = sock->ops->sendpage ? : sock_no_sendpage;
+	do {
+		int flags = msgflags;
+
+		len = PAGE_CACHE_SIZE;
+		if (base)
+			len -= base;
+		if (pglen < len)
+			len = pglen;
+
+		if (pglen != len || xdr->tail[0].iov_len != 0)
+			flags |= MSG_MORE;
+
+		/* Hmm... We might be dealing with highmem pages */
+		if (PageHighMem(*ppage))
+			sendpage = sock_no_sendpage;
+		err = sendpage(sock, *ppage, base, len, flags);
+		if (ret == 0)
+			ret = err;
+		else if (err > 0)
+			ret += err;
+		if (err != len)
+			goto out;
+		base = 0;
+		ppage++;
+	} while ((pglen -= len) != 0);
+copy_tail:
+	len = xdr->tail[0].iov_len;
+	if (base < len) {
+		struct kvec iov = {
+			.iov_base = xdr->tail[0].iov_base + base,
+			.iov_len  = len - base,
+		};
+		struct msghdr msg = {
+			.msg_flags   = msgflags,
+		};
+		err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
+		if (ret == 0)
+			ret = err;
+		else if (err > 0)
+			ret += err;
+	}
+out:
+	return ret;
+}
+
+/*
+ * Write data to socket.
+ */
+static inline int
+xprt_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
+{
+	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;
+
+	if (!sock)
+		return -ENOTCONN;
+
+	xprt_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);
+	}
+	/* Dont repeat bytes */
+	skip = req->rq_bytes_sent;
+
+	clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
+	result = xdr_sendpages(sock, addr, addrlen, xdr, skip, MSG_DONTWAIT);
+
+	dprintk("RPC:      xprt_sendmsg(%d) = %d\n", xdr->len - skip, result);
+
+	if (result >= 0)
+		return result;
+
+	switch (result) {
+	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;
+		break;
+	default:
+		printk(KERN_NOTICE "RPC: sendmsg returned error %d\n", -result);
+	}
+	return result;
+}
+
+static int
+xprt_send_request(struct rpc_task *task)
+{
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_xprt *xprt = req->rq_xprt;
+	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)));
+	}
+
+	/* Continue transmitting the packet/record. We must be careful
+	 * to cope with writespace callbacks arriving _after_ we have
+	 * called xprt_sendmsg().
+	 */
+	while (1) {
+		req->rq_xtime = jiffies;
+		status = xprt_sendmsg(xprt, req);
+
+		if (status < 0)
+			break;
+
+		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;
+			break;
+		}
+
+		dprintk("RPC: %4d xmit incomplete (%d left of %d)\n",
+				task->tk_pid, req->rq_slen - req->rq_bytes_sent,
+				req->rq_slen);
+
+		status = -EAGAIN;
+		if (retry++ > 50)
+			break;
+	}
+
+	if (status == -EAGAIN) {
+		if (test_bit(SOCK_ASYNC_NOSPACE, &xprt->sock->flags)) {
+			/* Protect against races with xprt_write_space */
+			spin_lock_bh(&xprt->sock_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);
+			}
+			spin_unlock_bh(&xprt->sock_lock);
+			return status;
+		}
+		/* Keep holding the socket if it is blocked */
+		rpc_delay(task, HZ>>4);
+	}
+	return status;
+}
+
+/*
+ * Close down a transport socket
+ */
+static void
+xprt_close(struct rpc_xprt *xprt)
+{
+	struct socket	*sock = xprt->sock;
+	struct sock	*sk = xprt->inet;
+
+	if (!sk)
+		return;
+
+	write_lock_bh(&sk->sk_callback_lock);
+	xprt->inet = NULL;
+	xprt->sock = NULL;
+
+	sk->sk_user_data    = NULL;
+	sk->sk_data_ready   = xprt->old_data_ready;
+	sk->sk_state_change = xprt->old_state_change;
+	sk->sk_write_space  = xprt->old_write_space;
+	write_unlock_bh(&sk->sk_callback_lock);
+
+	sk->sk_no_check	 = 0;
+
+	sock_release(sock);
+}
+
+static void xprt_socket_destroy(struct rpc_xprt *xprt)
+{
+	cancel_delayed_work(&xprt->sock_connect);
+	flush_scheduled_work();
+
+	xprt_disconnect(xprt);
+	xprt_close(xprt);
+	kfree(xprt->slot);
+}
+
+/*
+ * Input handler for RPC replies. Called from a bottom half and hence
+ * atomic.
+ */
+static void
+udp_data_ready(struct sock *sk, int len)
+{
+	struct rpc_task	*task;
+	struct rpc_xprt	*xprt;
+	struct rpc_rqst *rovr;
+	struct sk_buff	*skb;
+	int err, repsize, copied;
+	u32 _xid, *xp;
+
+	read_lock(&sk->sk_callback_lock);
+	dprintk("RPC:      udp_data_ready...\n");
+	if (!(xprt = xprt_from_sock(sk))) {
+		printk("RPC:      udp_data_ready request not found!\n");
+		goto out;
+	}
+
+	dprintk("RPC:      udp_data_ready client %p\n", xprt);
+
+	if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
+		goto out;
+
+	if (xprt->shutdown)
+		goto dropit;
+
+	repsize = skb->len - sizeof(struct udphdr);
+	if (repsize < 4) {
+		printk("RPC: impossible RPC reply size %d!\n", repsize);
+		goto dropit;
+	}
+
+	/* Copy the XID from the skb... */
+	xp = skb_header_pointer(skb, sizeof(struct udphdr),
+				sizeof(_xid), &_xid);
+	if (xp == NULL)
+		goto dropit;
+
+	/* Look up and lock the request corresponding to the given XID */
+	spin_lock(&xprt->sock_lock);
+	rovr = xprt_lookup_rqst(xprt, *xp);
+	if (!rovr)
+		goto out_unlock;
+	task = rovr->rq_task;
+
+	dprintk("RPC: %4d received reply\n", task->tk_pid);
+
+	if ((copied = rovr->rq_private_buf.buflen) > repsize)
+		copied = repsize;
+
+	/* Suck it into the iovec, verify checksum if not done by hw. */
+	if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb))
+		goto out_unlock;
+
+	/* Something worked... */
+	dst_confirm(skb->dst);
+
+	xprt_complete_rqst(xprt, rovr, copied);
+
+ out_unlock:
+	spin_unlock(&xprt->sock_lock);
+ dropit:
+	skb_free_datagram(sk, skb);
+ out:
+	read_unlock(&sk->sk_callback_lock);
+}
+
+/*
+ * Copy from an skb into memory and shrink the skb.
+ */
+static inline size_t
+tcp_copy_data(skb_reader_t *desc, void *p, size_t len)
+{
+	if (len > desc->count)
+		len = desc->count;
+	if (skb_copy_bits(desc->skb, desc->offset, p, len)) {
+		dprintk("RPC:      failed to copy %zu bytes from skb. %zu bytes remain\n",
+				len, desc->count);
+		return 0;
+	}
+	desc->offset += len;
+	desc->count -= len;
+	dprintk("RPC:      copied %zu bytes from skb. %zu bytes remain\n",
+			len, desc->count);
+	return len;
+}
+
+/*
+ * TCP read fragment marker
+ */
+static inline void
+tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc)
+{
+	size_t len, used;
+	char *p;
+
+	p = ((char *) &xprt->tcp_recm) + xprt->tcp_offset;
+	len = sizeof(xprt->tcp_recm) - xprt->tcp_offset;
+	used = tcp_copy_data(desc, p, len);
+	xprt->tcp_offset += used;
+	if (used != len)
+		return;
+	xprt->tcp_reclen = ntohl(xprt->tcp_recm);
+	if (xprt->tcp_reclen & 0x80000000)
+		xprt->tcp_flags |= XPRT_LAST_FRAG;
+	else
+		xprt->tcp_flags &= ~XPRT_LAST_FRAG;
+	xprt->tcp_reclen &= 0x7fffffff;
+	xprt->tcp_flags &= ~XPRT_COPY_RECM;
+	xprt->tcp_offset = 0;
+	/* Sanity check of the record length */
+	if (xprt->tcp_reclen < 4) {
+		printk(KERN_ERR "RPC: Invalid TCP record fragment length\n");
+		xprt_disconnect(xprt);
+	}
+	dprintk("RPC:      reading TCP record fragment of length %d\n",
+			xprt->tcp_reclen);
+}
+
+static void
+tcp_check_recm(struct rpc_xprt *xprt)
+{
+	dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u, tcp_flags = %lx\n",
+			xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen, xprt->tcp_flags);
+	if (xprt->tcp_offset == xprt->tcp_reclen) {
+		xprt->tcp_flags |= XPRT_COPY_RECM;
+		xprt->tcp_offset = 0;
+		if (xprt->tcp_flags & XPRT_LAST_FRAG) {
+			xprt->tcp_flags &= ~XPRT_COPY_DATA;
+			xprt->tcp_flags |= XPRT_COPY_XID;
+			xprt->tcp_copied = 0;
+		}
+	}
+}
+
+/*
+ * TCP read xid
+ */
+static inline void
+tcp_read_xid(struct rpc_xprt *xprt, skb_reader_t *desc)
+{
+	size_t len, used;
+	char *p;
+
+	len = sizeof(xprt->tcp_xid) - xprt->tcp_offset;
+	dprintk("RPC:      reading XID (%Zu bytes)\n", len);
+	p = ((char *) &xprt->tcp_xid) + xprt->tcp_offset;
+	used = tcp_copy_data(desc, p, len);
+	xprt->tcp_offset += used;
+	if (used != len)
+		return;
+	xprt->tcp_flags &= ~XPRT_COPY_XID;
+	xprt->tcp_flags |= XPRT_COPY_DATA;
+	xprt->tcp_copied = 4;
+	dprintk("RPC:      reading reply for XID %08x\n",
+						ntohl(xprt->tcp_xid));
+	tcp_check_recm(xprt);
+}
+
+/*
+ * TCP read and complete request
+ */
+static inline void
+tcp_read_request(struct rpc_xprt *xprt, skb_reader_t *desc)
+{
+	struct rpc_rqst *req;
+	struct xdr_buf *rcvbuf;
+	size_t len;
+	ssize_t r;
+
+	/* Find and lock the request corresponding to this xid */
+	spin_lock(&xprt->sock_lock);
+	req = xprt_lookup_rqst(xprt, xprt->tcp_xid);
+	if (!req) {
+		xprt->tcp_flags &= ~XPRT_COPY_DATA;
+		dprintk("RPC:      XID %08x request not found!\n",
+				ntohl(xprt->tcp_xid));
+		spin_unlock(&xprt->sock_lock);
+		return;
+	}
+
+	rcvbuf = &req->rq_private_buf;
+	len = desc->count;
+	if (len > xprt->tcp_reclen - xprt->tcp_offset) {
+		skb_reader_t my_desc;
+
+		len = xprt->tcp_reclen - xprt->tcp_offset;
+		memcpy(&my_desc, desc, sizeof(my_desc));
+		my_desc.count = len;
+		r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
+					  &my_desc, tcp_copy_data);
+		desc->count -= r;
+		desc->offset += r;
+	} else
+		r = xdr_partial_copy_from_skb(rcvbuf, xprt->tcp_copied,
+					  desc, tcp_copy_data);
+
+	if (r > 0) {
+		xprt->tcp_copied += r;
+		xprt->tcp_offset += r;
+	}
+	if (r != len) {
+		/* Error when copying to the receive buffer,
+		 * usually because we weren't able to allocate
+		 * additional buffer pages. All we can do now
+		 * is turn off XPRT_COPY_DATA, so the request
+		 * will not receive any additional updates,
+		 * and time out.
+		 * Any remaining data from this record will
+		 * be discarded.
+		 */
+		xprt->tcp_flags &= ~XPRT_COPY_DATA;
+		dprintk("RPC:      XID %08x truncated request\n",
+				ntohl(xprt->tcp_xid));
+		dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
+				xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
+		goto out;
+	}
+
+	dprintk("RPC:      XID %08x read %Zd bytes\n",
+			ntohl(xprt->tcp_xid), r);
+	dprintk("RPC:      xprt = %p, tcp_copied = %lu, tcp_offset = %u, tcp_reclen = %u\n",
+			xprt, xprt->tcp_copied, xprt->tcp_offset, xprt->tcp_reclen);
+
+	if (xprt->tcp_copied == req->rq_private_buf.buflen)
+		xprt->tcp_flags &= ~XPRT_COPY_DATA;
+	else if (xprt->tcp_offset == xprt->tcp_reclen) {
+		if (xprt->tcp_flags & XPRT_LAST_FRAG)
+			xprt->tcp_flags &= ~XPRT_COPY_DATA;
+	}
+
+out:
+	if (!(xprt->tcp_flags & XPRT_COPY_DATA)) {
+		dprintk("RPC: %4d received reply complete\n",
+				req->rq_task->tk_pid);
+		xprt_complete_rqst(xprt, req, xprt->tcp_copied);
+	}
+	spin_unlock(&xprt->sock_lock);
+	tcp_check_recm(xprt);
+}
+
+/*
+ * TCP discard extra bytes from a short read
+ */
+static inline void
+tcp_read_discard(struct rpc_xprt *xprt, skb_reader_t *desc)
+{
+	size_t len;
+
+	len = xprt->tcp_reclen - xprt->tcp_offset;
+	if (len > desc->count)
+		len = desc->count;
+	desc->count -= len;
+	desc->offset += len;
+	xprt->tcp_offset += len;
+	dprintk("RPC:      discarded %Zu bytes\n", len);
+	tcp_check_recm(xprt);
+}
+
+/*
+ * TCP record receive routine
+ * We first have to grab the record marker, then the XID, then the data.
+ */
+static int
+tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
+		unsigned int offset, size_t len)
+{
+	struct rpc_xprt *xprt = rd_desc->arg.data;
+	skb_reader_t desc = {
+		.skb	= skb,
+		.offset	= offset,
+		.count	= len,
+		.csum	= 0
+       	};
+
+	dprintk("RPC:      tcp_data_recv\n");
+	do {
+		/* Read in a new fragment marker if necessary */
+		/* Can we ever really expect to get completely empty fragments? */
+		if (xprt->tcp_flags & XPRT_COPY_RECM) {
+			tcp_read_fraghdr(xprt, &desc);
+			continue;
+		}
+		/* Read in the xid if necessary */
+		if (xprt->tcp_flags & XPRT_COPY_XID) {
+			tcp_read_xid(xprt, &desc);
+			continue;
+		}
+		/* Read in the request data */
+		if (xprt->tcp_flags & XPRT_COPY_DATA) {
+			tcp_read_request(xprt, &desc);
+			continue;
+		}
+		/* Skip over any trailing bytes on short reads */
+		tcp_read_discard(xprt, &desc);
+	} while (desc.count);
+	dprintk("RPC:      tcp_data_recv done\n");
+	return len - desc.count;
+}
+
+static void tcp_data_ready(struct sock *sk, int bytes)
+{
+	struct rpc_xprt *xprt;
+	read_descriptor_t rd_desc;
+
+	read_lock(&sk->sk_callback_lock);
+	dprintk("RPC:      tcp_data_ready...\n");
+	if (!(xprt = xprt_from_sock(sk))) {
+		printk("RPC:      tcp_data_ready socket info not found!\n");
+		goto out;
+	}
+	if (xprt->shutdown)
+		goto out;
+
+	/* We use rd_desc to pass struct xprt to tcp_data_recv */
+	rd_desc.arg.data = xprt;
+	rd_desc.count = 65536;
+	tcp_read_sock(sk, &rd_desc, tcp_data_recv);
+out:
+	read_unlock(&sk->sk_callback_lock);
+}
+
+static void
+tcp_state_change(struct sock *sk)
+{
+	struct rpc_xprt	*xprt;
+
+	read_lock(&sk->sk_callback_lock);
+	if (!(xprt = xprt_from_sock(sk)))
+		goto out;
+	dprintk("RPC:      tcp_state_change client %p...\n", xprt);
+	dprintk("RPC:      state %x conn %d dead %d zapped %d\n",
+				sk->sk_state, xprt_connected(xprt),
+				sock_flag(sk, SOCK_DEAD),
+				sock_flag(sk, SOCK_ZAPPED));
+
+	switch (sk->sk_state) {
+	case TCP_ESTABLISHED:
+		spin_lock_bh(&xprt->sock_lock);
+		if (!xprt_test_and_set_connected(xprt)) {
+			/* Reset TCP record info */
+			xprt->tcp_offset = 0;
+			xprt->tcp_reclen = 0;
+			xprt->tcp_copied = 0;
+			xprt->tcp_flags = XPRT_COPY_RECM | XPRT_COPY_XID;
+			rpc_wake_up(&xprt->pending);
+		}
+		spin_unlock_bh(&xprt->sock_lock);
+		break;
+	case TCP_SYN_SENT:
+	case TCP_SYN_RECV:
+		break;
+	default:
+		xprt_disconnect(xprt);
+		break;
+	}
+ out:
+	read_unlock(&sk->sk_callback_lock);
+}
+
+/*
+ * 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
+ * with a bunch of small requests.
+ */
+static void
+xprt_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))
+			goto out;
+	} else {
+		/* from net/core/sock.c:sock_def_write_space */
+		if (!sock_writeable(sk))
+			goto out;
+	}
+
+	if (!test_and_clear_bit(SOCK_NOSPACE, &sock->flags))
+		goto out;
+
+	spin_lock_bh(&xprt->sock_lock);
+	if (xprt->snd_task)
+		rpc_wake_up_task(xprt->snd_task);
+	spin_unlock_bh(&xprt->sock_lock);
+out:
+	read_unlock(&sk->sk_callback_lock);
+}
+
+/*
+ * Set socket buffer length
+ */
+static void
+xprt_sock_setbufsize(struct rpc_xprt *xprt)
+{
+	struct sock *sk = xprt->inet;
+
+	if (xprt->stream)
+		return;
+	if (xprt->rcvsize) {
+		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
+		sk->sk_rcvbuf = xprt->rcvsize * xprt->max_reqs *  2;
+	}
+	if (xprt->sndsize) {
+		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
+		sk->sk_sndbuf = xprt->sndsize * xprt->max_reqs * 2;
+		sk->sk_write_space(sk);
+	}
+}
+
+/*
+ * Bind to a reserved port
+ */
+static inline int xprt_bindresvport(struct rpc_xprt *xprt, struct socket *sock)
+{
+	struct sockaddr_in myaddr = {
+		.sin_family = AF_INET,
+	};
+	int		err, port;
+
+	/* Were we already bound to a given port? Try to reuse it */
+	port = xprt->port;
+	do {
+		myaddr.sin_port = htons(port);
+		err = sock->ops->bind(sock, (struct sockaddr *) &myaddr,
+						sizeof(myaddr));
+		if (err == 0) {
+			xprt->port = port;
+			return 0;
+		}
+		if (--port == 0)
+			port = XPRT_MAX_RESVPORT;
+	} while (err == -EADDRINUSE && port != xprt->port);
+
+	printk("RPC: Can't bind to reserved port (%d).\n", -err);
+	return err;
+}
+
+static void
+xprt_bind_socket(struct rpc_xprt *xprt, struct socket *sock)
+{
+	struct sock	*sk = sock->sk;
+
+	if (xprt->inet)
+		return;
+
+	write_lock_bh(&sk->sk_callback_lock);
+	sk->sk_user_data = xprt;
+	xprt->old_data_ready = sk->sk_data_ready;
+	xprt->old_state_change = sk->sk_state_change;
+	xprt->old_write_space = sk->sk_write_space;
+	if (xprt->prot == IPPROTO_UDP) {
+		sk->sk_data_ready = udp_data_ready;
+		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 = tcp_data_ready;
+		sk->sk_state_change = tcp_state_change;
+		xprt_clear_connected(xprt);
+	}
+	sk->sk_write_space = xprt_write_space;
+
+	/* Reset to new socket */
+	xprt->sock = sock;
+	xprt->inet = sk;
+	write_unlock_bh(&sk->sk_callback_lock);
+
+	return;
+}
+
+/*
+ * Datastream sockets are created here, but xprt_connect will create
+ * and connect stream sockets.
+ */
+static struct socket * xprt_create_socket(struct rpc_xprt *xprt, int proto, int resvport)
+{
+	struct socket	*sock;
+	int		type, err;
+
+	dprintk("RPC:      xprt_create_socket(%s %d)\n",
+			   (proto == IPPROTO_UDP)? "udp" : "tcp", proto);
+
+	type = (proto == IPPROTO_UDP)? SOCK_DGRAM : SOCK_STREAM;
+
+	if ((err = sock_create_kern(PF_INET, type, proto, &sock)) < 0) {
+		printk("RPC: can't create socket (%d).\n", -err);
+		return NULL;
+	}
+
+	/* If the caller has the capability, bind to a reserved port */
+	if (resvport && xprt_bindresvport(xprt, sock) < 0) {
+		printk("RPC: can't bind to reserved port.\n");
+		goto failed;
+	}
+
+	return sock;
+
+failed:
+	sock_release(sock);
+	return NULL;
+}
+
+static void xprt_socket_connect(void *args)
+{
+	struct rpc_xprt *xprt = (struct rpc_xprt *)args;
+	struct socket *sock = xprt->sock;
+	int status = -EIO;
+
+	if (xprt->shutdown || xprt->addr.sin_port == 0)
+		goto out;
+
+	/*
+	 * Start by resetting any existing state
+	 */
+	xprt_close(xprt);
+	sock = xprt_create_socket(xprt, xprt->prot, xprt->resvport);
+	if (sock == NULL) {
+		/* couldn't create socket or bind to reserved port;
+		 * this is likely a permanent error, so cause an abort */
+		goto out;
+	}
+	xprt_bind_socket(xprt, sock);
+	xprt_sock_setbufsize(xprt);
+
+	status = 0;
+	if (!xprt->stream)
+		goto out;
+
+	/*
+	 * Tell the socket layer to start connecting...
+	 */
+	status = sock->ops->connect(sock, (struct sockaddr *) &xprt->addr,
+			sizeof(xprt->addr), O_NONBLOCK);
+	dprintk("RPC: %p  connect status %d connected %d sock state %d\n",
+			xprt, -status, xprt_connected(xprt), sock->sk->sk_state);
+	if (status < 0) {
+		switch (status) {
+			case -EINPROGRESS:
+			case -EALREADY:
+				goto out_clear;
+		}
+	}
+out:
+	if (status < 0)
+		rpc_wake_up_status(&xprt->pending, status);
+	else
+		rpc_wake_up(&xprt->pending);
+out_clear:
+	smp_mb__before_clear_bit();
+	clear_bit(XPRT_CONNECTING, &xprt->sockstate);
+	smp_mb__after_clear_bit();
+}
+
+static void
+xprt_connect_sock(struct rpc_task *task)
+{
+	struct rpc_xprt *xprt = task->tk_xprt;
+
+	if (!test_and_set_bit(XPRT_CONNECTING, &xprt->sockstate)) {
+		/* Note: if we are here due to a dropped connection
+		 * 	 we delay reconnecting by RPC_REESTABLISH_TIMEOUT/HZ
+		 * 	 seconds
+		 */
+		if (xprt->sock != NULL)
+			schedule_delayed_work(&xprt->sock_connect,
+					RPC_REESTABLISH_TIMEOUT);
+		else {
+			schedule_work(&xprt->sock_connect);
+			/* flush_scheduled_work can sleep... */
+			if (!RPC_IS_ASYNC(task))
+				flush_scheduled_work();
+		}
+	}
+}
+
+/*
+ * Set default timeout parameters
+ */
+static void
+xprt_default_timeout(struct rpc_timeout *to, int proto)
+{
+	if (proto == IPPROTO_UDP)
+		xprt_set_timeout(to, 5,  5 * HZ);
+	else
+		xprt_set_timeout(to, 2, 60 * HZ);
+}
+
+static struct rpc_xprt_ops xprt_socket_ops = {
+	.set_buffer_size	= xprt_sock_setbufsize,
+	.connect		= xprt_connect_sock,
+	.send_request		= xprt_send_request,
+	.close			= xprt_close,
+	.destroy		= xprt_socket_destroy,
+};
+
+extern unsigned int xprt_udp_slot_table_entries;
+extern unsigned int xprt_tcp_slot_table_entries;
+
+int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
+{
+	size_t slot_table_size;
+
+	dprintk("RPC:      setting up udp-ipv4 transport...\n");
+
+	xprt->max_reqs = xprt_udp_slot_table_entries;
+	slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
+	xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
+	if (xprt->slot == NULL)
+		return -ENOMEM;
+	memset(xprt->slot, 0, slot_table_size);
+
+	xprt->prot = IPPROTO_UDP;
+	xprt->port = XPRT_MAX_RESVPORT;
+	xprt->stream = 0;
+	xprt->nocong = 0;
+	xprt->cwnd = RPC_INITCWND;
+	xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
+	/* XXX: header size can vary due to auth type, IPv6, etc. */
+	xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
+
+	INIT_WORK(&xprt->sock_connect, xprt_socket_connect, xprt);
+
+	xprt->ops = &xprt_socket_ops;
+
+	if (to)
+		xprt->timeout = *to;
+	else
+		xprt_default_timeout(to, xprt->prot);
+
+	return 0;
+}
+
+int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
+{
+	size_t slot_table_size;
+
+	dprintk("RPC:      setting up tcp-ipv4 transport...\n");
+
+	xprt->max_reqs = xprt_tcp_slot_table_entries;
+	slot_table_size = xprt->max_reqs * sizeof(xprt->slot[0]);
+	xprt->slot = kmalloc(slot_table_size, GFP_KERNEL);
+	if (xprt->slot == NULL)
+		return -ENOMEM;
+	memset(xprt->slot, 0, slot_table_size);
+
+	xprt->prot = IPPROTO_TCP;
+	xprt->port = XPRT_MAX_RESVPORT;
+	xprt->stream = 1;
+	xprt->nocong = 1;
+	xprt->cwnd = RPC_MAXCWND(xprt);
+	xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
+	xprt->max_payload = (1U << 31) - 1;
+
+	INIT_WORK(&xprt->sock_connect, xprt_socket_connect, xprt);
+
+	xprt->ops = &xprt_socket_ops;
+
+	if (to)
+		xprt->timeout = *to;
+	else
+		xprt_default_timeout(to, xprt->prot);
+
+	return 0;
+}