manual update from upstream:

Applied Al's change 06a544971fad0992fe8b92c5647538d573089dd4
to new location of swiotlb.c

Signed-off-by: Tony Luck <tony.luck@intel.com>

1 files changed, 447 insertions, 1166 deletions

diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 3c654e06b08..6dda3860351 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -10,12 +10,12 @@
  *	one is available. Otherwise, it sleeps on the backlog queue
  *	(xprt_reserve).
  *  -	Next, the caller puts together the RPC message, stuffs it into
- *	the request struct, and calls xprt_call().
- *  -	xprt_call transmits the message and installs the caller on the
- *	socket's wait list. At the same time, it installs a timer that
+ *	the request struct, and calls xprt_transmit().
+ *  -	xprt_transmit sends the message and installs the caller on the
+ *	transport's wait list. At the same time, it installs a timer that
  *	is run after the packet's timeout has expired.
  *  -	When a packet arrives, the data_ready handler walks the list of
- *	pending requests for that socket. If a matching XID is found, the
+ *	pending requests for that transport. If a matching XID is found, the
  *	caller is woken up, and the timer removed.
  *  -	When no reply arrives within the timeout interval, the timer is
  *	fired by the kernel and runs xprt_timer(). It either adjusts the
@@ -33,36 +33,17 @@
  *
  *  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>
+ *  Transport switch API copyright (C) 2005, Chuck Lever <cel@netapp.com>
  */
 
+#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
@@ -73,81 +54,90 @@
 # define RPCDBG_FACILITY	RPCDBG_XPRT
 #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).
+ * The transport code maintains an estimate on the maximum number of out-
+ * standing RPC requests, using a smoothed version of the congestion
+ * avoidance implemented in 44BSD. This is basically the Van Jacobson
+ * congestion algorithm: If a retransmit occurs, the congestion window is
+ * halved; otherwise, it is incremented by 1/cwnd when
+ *
+ *	-	a reply is received and
+ *	-	a full number of requests are outstanding and
+ *	-	the congestion window hasn't been updated recently.
  */
-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
+#define RPC_CWNDSHIFT		(8U)
+#define RPC_CWNDSCALE		(1U << RPC_CWNDSHIFT)
+#define RPC_INITCWND		RPC_CWNDSCALE
+#define RPC_MAXCWND(xprt)	((xprt)->max_reqs << RPC_CWNDSHIFT)
 
-/*
- * Look up RPC transport given an INET socket
+#define RPCXPRT_CONGESTED(xprt) ((xprt)->cong >= (xprt)->cwnd)
+
+/**
+ * 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.
  */
-static inline struct rpc_xprt *
-xprt_from_sock(struct sock *sk)
+int xprt_reserve_xprt(struct rpc_task *task)
 {
-	return (struct rpc_xprt *) sk->sk_user_data;
+	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 sockets, in order to prevent different
- * requests from interfering with each other.
- * Also prevents TCP socket 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->sockstate)) {
+	if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) {
 		if (task == xprt->snd_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;
@@ -156,10 +146,10 @@ __xprt_lock_write(struct rpc_xprt *xprt, struct rpc_task *task)
 		return 1;
 	}
 	smp_mb__before_clear_bit();
-	clear_bit(XPRT_LOCKED, &xprt->sockstate);
+	clear_bit(XPRT_LOCKED, &xprt->state);
 	smp_mb__after_clear_bit();
 out_sleep:
-	dprintk("RPC: %4d failed to lock socket %p\n", task->tk_pid, xprt);
+	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)
@@ -169,26 +159,52 @@ 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->sock_lock);
-	retval = __xprt_lock_write(xprt, task);
-	spin_unlock_bh(&xprt->sock_lock);
+	spin_lock_bh(&xprt->transport_lock);
+	retval = xprt->ops->reserve_xprt(task);
+	spin_unlock_bh(&xprt->transport_lock);
 	return retval;
 }
 
+static void __xprt_lock_write_next(struct rpc_xprt *xprt)
+{
+	struct rpc_task *task;
+	struct rpc_rqst *req;
 
-static void
-__xprt_lock_write_next(struct rpc_xprt *xprt)
+	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
+		return;
+
+	task = rpc_wake_up_next(&xprt->resend);
+	if (!task) {
+		task = rpc_wake_up_next(&xprt->sending);
+		if (!task)
+			goto out_unlock;
+	}
+
+	req = task->tk_rqstp;
+	xprt->snd_task = task;
+	if (req) {
+		req->rq_bytes_sent = 0;
+		req->rq_ntrans++;
+	}
+	return;
+
+out_unlock:
+	smp_mb__before_clear_bit();
+	clear_bit(XPRT_LOCKED, &xprt->state);
+	smp_mb__after_clear_bit();
+}
+
+static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 {
 	struct rpc_task *task;
 
-	if (test_and_set_bit(XPRT_LOCKED, &xprt->sockstate))
+	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 		return;
-	if (!xprt->nocong && RPCXPRT_CONGESTED(xprt))
+	if (RPCXPRT_CONGESTED(xprt))
 		goto out_unlock;
 	task = rpc_wake_up_next(&xprt->resend);
 	if (!task) {
@@ -196,7 +212,7 @@ __xprt_lock_write_next(struct rpc_xprt *xprt)
 		if (!task)
 			goto out_unlock;
 	}
-	if (xprt->nocong || __xprt_get_cong(xprt, task)) {
+	if (__xprt_get_cong(xprt, task)) {
 		struct rpc_rqst *req = task->tk_rqstp;
 		xprt->snd_task = task;
 		if (req) {
@@ -207,87 +223,52 @@ __xprt_lock_write_next(struct rpc_xprt *xprt)
 	}
 out_unlock:
 	smp_mb__before_clear_bit();
-	clear_bit(XPRT_LOCKED, &xprt->sockstate);
+	clear_bit(XPRT_LOCKED, &xprt->state);
 	smp_mb__after_clear_bit();
 }
 
-/*
- * Releases the socket for use by other requests.
+/**
+ * xprt_release_xprt - allow other requests to use a transport
+ * @xprt: transport with other tasks potentially waiting
+ * @task: task that is releasing access to the transport
+ *
+ * Note that "task" can be NULL.  No congestion control is provided.
  */
-static void
-__xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
+void xprt_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 {
 	if (xprt->snd_task == task) {
 		xprt->snd_task = NULL;
 		smp_mb__before_clear_bit();
-		clear_bit(XPRT_LOCKED, &xprt->sockstate);
+		clear_bit(XPRT_LOCKED, &xprt->state);
 		smp_mb__after_clear_bit();
 		__xprt_lock_write_next(xprt);
 	}
 }
 
-static inline void
-xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
-{
-	spin_lock_bh(&xprt->sock_lock);
-	__xprt_release_write(xprt, task);
-	spin_unlock_bh(&xprt->sock_lock);
-}
-
-/*
- * Write data to socket.
+/**
+ * xprt_release_xprt_cong - allow other requests to use a transport
+ * @xprt: transport with other tasks potentially waiting
+ * @task: task that is releasing access to the transport
+ *
+ * Note that "task" can be NULL.  Another task is awoken to use the
+ * transport if the transport's congestion window allows it.
  */
-static inline int
-xprt_sendmsg(struct rpc_xprt *xprt, struct rpc_rqst *req)
+void xprt_release_xprt_cong(struct rpc_xprt *xprt, 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;
-
-	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);
+	if (xprt->snd_task == task) {
+		xprt->snd_task = NULL;
+		smp_mb__before_clear_bit();
+		clear_bit(XPRT_LOCKED, &xprt->state);
+		smp_mb__after_clear_bit();
+		__xprt_lock_write_next_cong(xprt);
 	}
-	/* 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 inline void xprt_release_write(struct rpc_xprt *xprt, struct rpc_task *task)
+{
+	spin_lock_bh(&xprt->transport_lock);
+	xprt->ops->release_xprt(xprt, task);
+	spin_unlock_bh(&xprt->transport_lock);
 }
 
 /*
@@ -321,26 +302,40 @@ __xprt_put_cong(struct rpc_xprt *xprt, struct rpc_rqst *req)
 		return;
 	req->rq_cong = 0;
 	xprt->cong -= RPC_CWNDSCALE;
-	__xprt_lock_write_next(xprt);
+	__xprt_lock_write_next_cong(xprt);
 }
 
-/*
- * Adjust RPC congestion window
+/**
+ * xprt_release_rqst_cong - housekeeping when request is complete
+ * @task: RPC request that recently completed
+ *
+ * Useful for transports that require congestion control.
+ */
+void xprt_release_rqst_cong(struct rpc_task *task)
+{
+	__xprt_put_cong(task->tk_xprt, task->tk_rqstp);
+}
+
+/**
+ * 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
-xprt_adjust_cwnd(struct rpc_xprt *xprt, int result)
+void xprt_adjust_cwnd(struct rpc_task *task, 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. */
 		cwnd += (RPC_CWNDSCALE * RPC_CWNDSCALE + (cwnd >> 1)) / cwnd;
 		if (cwnd > RPC_MAXCWND(xprt))
 			cwnd = RPC_MAXCWND(xprt);
-		__xprt_lock_write_next(xprt);
+		__xprt_lock_write_next_cong(xprt);
 	} else if (result == -ETIMEDOUT) {
 		cwnd >>= 1;
 		if (cwnd < RPC_CWNDSCALE)
@@ -349,11 +344,89 @@ 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);
+}
+
+/**
+ * xprt_wake_pending_tasks - wake all tasks on a transport's pending queue
+ * @xprt: transport with waiting tasks
+ * @status: result code to plant in each task before waking it
+ *
+ */
+void xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status)
+{
+	if (status < 0)
+		rpc_wake_up_status(&xprt->pending, status);
+	else
+		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);
+}
+
+/**
+ * xprt_set_retrans_timeout_def - set a request's retransmit timeout
+ * @task: task whose timeout is to be set
+ *
+ * Set a request's retransmit timeout based on the transport's
+ * default timeout parameters.  Used by transports that don't adjust
+ * the retransmit timeout based on round-trip time estimation.
+ */
+void xprt_set_retrans_timeout_def(struct rpc_task *task)
+{
+	task->tk_timeout = task->tk_rqstp->rq_timeout;
 }
 
 /*
- * Reset the major timeout value
+ * xprt_set_retrans_timeout_rtt - set a request's retransmit timeout
+ * @task: task whose timeout is to be set
+ * 
+ * Set a request's retransmit timeout using the RTT estimator.
  */
+void xprt_set_retrans_timeout_rtt(struct rpc_task *task)
+{
+	int timer = task->tk_msg.rpc_proc->p_timer;
+	struct rpc_rtt *rtt = task->tk_client->cl_rtt;
+	struct rpc_rqst *req = task->tk_rqstp;
+	unsigned long max_timeout = req->rq_xprt->timeout.to_maxval;
+
+	task->tk_timeout = rpc_calc_rto(rtt, timer);
+	task->tk_timeout <<= rpc_ntimeo(rtt, timer) + req->rq_retries;
+	if (task->tk_timeout > max_timeout || task->tk_timeout == 0)
+		task->tk_timeout = max_timeout;
+}
+
 static void xprt_reset_majortimeo(struct rpc_rqst *req)
 {
 	struct rpc_timeout *to = &req->rq_xprt->timeout;
@@ -368,8 +441,10 @@ static void xprt_reset_majortimeo(struct rpc_rqst *req)
 	req->rq_majortimeo += jiffies;
 }
 
-/*
- * Adjust timeout values etc for next retransmit
+/**
+ * xprt_adjust_timeout - adjust timeout values for next retransmit
+ * @req: RPC request containing parameters to use for the adjustment
+ *
  */
 int xprt_adjust_timeout(struct rpc_rqst *req)
 {
@@ -391,9 +466,9 @@ int xprt_adjust_timeout(struct rpc_rqst *req)
 		req->rq_retries = 0;
 		xprt_reset_majortimeo(req);
 		/* Reset the RTT counters == "slow start" */
-		spin_lock_bh(&xprt->sock_lock);
+		spin_lock_bh(&xprt->transport_lock);
 		rpc_init_rtt(req->rq_task->tk_client->cl_rtt, to->to_initval);
-		spin_unlock_bh(&xprt->sock_lock);
+		spin_unlock_bh(&xprt->transport_lock);
 		pprintk("RPC: %lu timeout\n", jiffies);
 		status = -ETIMEDOUT;
 	}
@@ -405,133 +480,52 @@ 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)
+static void xprt_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
+/**
+ * xprt_disconnect - mark a transport as disconnected
+ * @xprt: transport to flag for disconnect
+ *
  */
-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);
+	spin_lock_bh(&xprt->transport_lock);
 	xprt_clear_connected(xprt);
-	rpc_wake_up_status(&xprt->pending, -ENOTCONN);
-	spin_unlock_bh(&xprt->sock_lock);
+	xprt_wake_pending_tasks(xprt, -ENOTCONN);
+	spin_unlock_bh(&xprt->transport_lock);
 }
 
-/*
- * Used to allow disconnection when we've been idle
- */
 static void
 xprt_init_autodisconnect(unsigned long data)
 {
 	struct rpc_xprt *xprt = (struct rpc_xprt *)data;
 
-	spin_lock(&xprt->sock_lock);
+	spin_lock(&xprt->transport_lock);
 	if (!list_empty(&xprt->recv) || xprt->shutdown)
 		goto out_abort;
-	if (test_and_set_bit(XPRT_LOCKED, &xprt->sockstate))
+	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 		goto out_abort;
-	spin_unlock(&xprt->sock_lock);
-	/* Let keventd close the socket */
-	if (test_bit(XPRT_CONNECTING, &xprt->sockstate) != 0)
+	spin_unlock(&xprt->transport_lock);
+	if (xprt_connecting(xprt))
 		xprt_release_write(xprt, NULL);
 	else
 		schedule_work(&xprt->task_cleanup);
 	return;
 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();
+	spin_unlock(&xprt->transport_lock);
 }
 
-/*
- * Attempt to connect a TCP socket.
+/**
+ * xprt_connect - schedule a transport connect operation
+ * @task: RPC task that is requesting the connect
  *
  */
 void xprt_connect(struct rpc_task *task)
@@ -552,37 +546,19 @@ 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 = xprt->connect_timeout;
+		rpc_sleep_on(&xprt->pending, task, xprt_connect_status, NULL);
+		xprt->ops->connect(task);
 	}
 	return;
- out_write:
-	xprt_release_write(xprt, task);
 }
 
-/*
- * We arrive here when awoken from waiting on connection establishment.
- */
-static void
-xprt_connect_status(struct rpc_task *task)
+static void xprt_connect_status(struct rpc_task *task)
 {
 	struct rpc_xprt	*xprt = task->tk_xprt;
 
@@ -592,31 +568,42 @@ xprt_connect_status(struct rpc_task *task)
 		return;
 	}
 
-	/* if soft mounted, just cause this RPC to fail */
-	if (RPC_IS_SOFT(task))
-		task->tk_status = -EIO;
-
 	switch (task->tk_status) {
 	case -ECONNREFUSED:
 	case -ECONNRESET:
+		dprintk("RPC: %4d xprt_connect_status: server %s refused connection\n",
+				task->tk_pid, task->tk_client->cl_server);
+		break;
 	case -ENOTCONN:
-		return;
+		dprintk("RPC: %4d xprt_connect_status: connection broken\n",
+				task->tk_pid);
+		break;
 	case -ETIMEDOUT:
-		dprintk("RPC: %4d xprt_connect_status: timed out\n",
+		dprintk("RPC: %4d xprt_connect_status: connect attempt timed out\n",
 				task->tk_pid);
 		break;
 	default:
-		printk(KERN_ERR "RPC: error %d connecting to server %s\n",
-				-task->tk_status, task->tk_client->cl_server);
+		dprintk("RPC: %4d xprt_connect_status: error %d connecting to server %s\n",
+				task->tk_pid, -task->tk_status, task->tk_client->cl_server);
+		xprt_release_write(xprt, task);
+		task->tk_status = -EIO;
+		return;
+	}
+
+	/* if soft mounted, just cause this RPC to fail */
+	if (RPC_IS_SOFT(task)) {
+		xprt_release_write(xprt, task);
+		task->tk_status = -EIO;
 	}
-	xprt_release_write(xprt, task);
 }
 
-/*
- * Look up the RPC request corresponding to a reply, and then lock it.
+/**
+ * xprt_lookup_rqst - find an RPC request corresponding to an XID
+ * @xprt: transport on which the original request was transmitted
+ * @xid: RPC XID of incoming reply
+ *
  */
-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;
@@ -631,556 +618,68 @@ xprt_lookup_rqst(struct rpc_xprt *xprt, u32 xid)
 	return req;
 }
 
-/*
- * 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)
-{
-	struct rpc_task	*task = req->rq_task;
-	struct rpc_clnt *clnt = task->tk_client;
-
-	/* Adjust congestion window */
-	if (!xprt->nocong) {
-		unsigned timer = task->tk_msg.rpc_proc->p_timer;
-		xprt_adjust_cwnd(xprt, copied);
-		__xprt_put_cong(xprt, req);
-		if (timer) {
-			if (req->rq_ntrans == 1)
-				rpc_update_rtt(clnt->cl_rtt, timer,
-						(long)jiffies - req->rq_xtime);
-			rpc_set_timeo(clnt->cl_rtt, timer, req->rq_ntrans - 1);
-		}
-	}
-
-#ifdef RPC_PROFILE
-	/* Profile only reads for now */
-	if (copied > 1024) {
-		static unsigned long	nextstat;
-		static unsigned long	pkt_rtt, pkt_len, pkt_cnt;
-
-		pkt_cnt++;
-		pkt_len += req->rq_slen + copied;
-		pkt_rtt += jiffies - req->rq_xtime;
-		if (time_before(nextstat, jiffies)) {
-			printk("RPC: %lu %ld cwnd\n", jiffies, xprt->cwnd);
-			printk("RPC: %ld %ld %ld %ld stat\n",
-					jiffies, pkt_cnt, pkt_len, pkt_rtt);
-			pkt_rtt = pkt_len = pkt_cnt = 0;
-			nextstat = jiffies + 5 * HZ;
-		}
-	}
-#endif
-
-	dprintk("RPC: %4d has input (%d bytes)\n", task->tk_pid, copied);
-	list_del_init(&req->rq_list);
-	req->rq_received = req->rq_private_buf.len = copied;
-
-	/* ... and wake up the process. */
-	rpc_wake_up_task(task);
-	return;
-}
-
-static size_t
-skb_read_bits(skb_reader_t *desc, void *to, size_t len)
-{
-	if (len > desc->count)
-		len = desc->count;
-	if (skb_copy_bits(desc->skb, desc->offset, to, len))
-		return 0;
-	desc->count -= len;
-	desc->offset += len;
-	return len;
-}
-
-static size_t
-skb_read_and_csum_bits(skb_reader_t *desc, void *to, size_t len)
-{
-	unsigned int csum2, pos;
-
-	if (len > desc->count)
-		len = desc->count;
-	pos = desc->offset;
-	csum2 = skb_copy_and_csum_bits(desc->skb, pos, to, len, 0);
-	desc->csum = csum_block_add(desc->csum, csum2, pos);
-	desc->count -= len;
-	desc->offset += len;
-	return len;
-}
-
-/*
- * We have set things up such that we perform the checksum of the UDP
- * packet in parallel with the copies into the RPC client iovec.  -DaveM
- */
-int
-csum_partial_copy_to_xdr(struct xdr_buf *xdr, struct sk_buff *skb)
-{
-	skb_reader_t desc;
-
-	desc.skb = skb;
-	desc.offset = sizeof(struct udphdr);
-	desc.count = skb->len - desc.offset;
-
-	if (skb->ip_summed == CHECKSUM_UNNECESSARY)
-		goto no_checksum;
-
-	desc.csum = csum_partial(skb->data, desc.offset, skb->csum);
-	if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_and_csum_bits) < 0)
-		return -1;
-	if (desc.offset != skb->len) {
-		unsigned int csum2;
-		csum2 = skb_checksum(skb, desc.offset, skb->len - desc.offset, 0);
-		desc.csum = csum_block_add(desc.csum, csum2, desc.offset);
-	}
-	if (desc.count)
-		return -1;
-	if ((unsigned short)csum_fold(desc.csum))
-		return -1;
-	return 0;
-no_checksum:
-	if (xdr_partial_copy_from_skb(xdr, 0, &desc, skb_read_bits) < 0)
-		return -1;
-	if (desc.count)
-		return -1;
-	return 0;
-}
-
-/*
- * 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->coun