SUNRPC: Add basic load balancing to the transport switch

For now, just count the queue length. It is less accurate than counting
number of bytes queued, but easier to implement.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

include/linux/sunrpc/xprt.h

@@ -238,6 +238,7 @@ struct rpc_xprt {
 	/*
 	 * Send stuff
 	 */
+	atomic_long_t		queuelen;
 	spinlock_t		transport_lock;	/* lock transport info */
 	spinlock_t		reserve_lock;	/* lock slot table */
 	spinlock_t		queue_lock;	/* send/receive queue lock */

include/linux/sunrpc/xprtmultipath.h

@@ -15,6 +15,8 @@ struct rpc_xprt_switch {
 	struct kref		xps_kref;
 
 	unsigned int		xps_nxprts;
+	unsigned int		xps_nactive;
+	atomic_long_t		xps_queuelen;
 	struct list_head	xps_xprt_list;
 
 	struct net *		xps_net;

net/sunrpc/clnt.c

@@ -968,13 +968,47 @@ struct rpc_clnt *rpc_bind_new_program(struct rpc_clnt *old,
 }
 EXPORT_SYMBOL_GPL(rpc_bind_new_program);
 
+static struct rpc_xprt *
+rpc_task_get_xprt(struct rpc_clnt *clnt)
+{
+	struct rpc_xprt_switch *xps;
+	struct rpc_xprt *xprt= xprt_iter_get_next(&clnt->cl_xpi);
+
+	if (!xprt)
+		return NULL;
+	rcu_read_lock();
+	xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch);
+	atomic_long_inc(&xps->xps_queuelen);
+	rcu_read_unlock();
+	atomic_long_inc(&xprt->queuelen);
+
+	return xprt;
+}
+
+static void
+rpc_task_release_xprt(struct rpc_clnt *clnt, struct rpc_xprt *xprt)
+{
+	struct rpc_xprt_switch *xps;
+
+	atomic_long_dec(&xprt->queuelen);
+	rcu_read_lock();
+	xps = rcu_dereference(clnt->cl_xpi.xpi_xpswitch);
+	atomic_long_dec(&xps->xps_queuelen);
+	rcu_read_unlock();
+
+	xprt_put(xprt);
+}
+
 void rpc_task_release_transport(struct rpc_task *task)
 {
 	struct rpc_xprt *xprt = task->tk_xprt;
 
 	if (xprt) {
 		task->tk_xprt = NULL;
-		xprt_put(xprt);
+		if (task->tk_client)
+			rpc_task_release_xprt(task->tk_client, xprt);
+		else
+			xprt_put(xprt);
 	}
 }
 EXPORT_SYMBOL_GPL(rpc_task_release_transport);
@@ -983,6 +1017,7 @@ void rpc_task_release_client(struct rpc_task *task)
 {
 	struct rpc_clnt *clnt = task->tk_client;
 
+	rpc_task_release_transport(task);
 	if (clnt != NULL) {
 		/* Remove from client task list */
 		spin_lock(&clnt->cl_lock);
@@ -992,14 +1027,13 @@ void rpc_task_release_client(struct rpc_task *task)
 
 		rpc_release_client(clnt);
 	}
-	rpc_task_release_transport(task);
 }
 
 static
 void rpc_task_set_transport(struct rpc_task *task, struct rpc_clnt *clnt)
 {
 	if (!task->tk_xprt)
-		task->tk_xprt = xprt_iter_get_next(&clnt->cl_xpi);
+		task->tk_xprt = rpc_task_get_xprt(clnt);
 }
 
 static

net/sunrpc/xprtmultipath.c

@@ -36,6 +36,7 @@ static void xprt_switch_add_xprt_locked(struct rpc_xprt_switch *xps,
 	if (xps->xps_nxprts == 0)
 		xps->xps_net = xprt->xprt_net;
 	xps->xps_nxprts++;
+	xps->xps_nactive++;
 }
 
 /**
@@ -62,6 +63,7 @@ static void xprt_switch_remove_xprt_locked(struct rpc_xprt_switch *xps,
 {
 	if (unlikely(xprt == NULL))
 		return;
+	xps->xps_nactive--;
 	xps->xps_nxprts--;
 	if (xps->xps_nxprts == 0)
 		xps->xps_net = NULL;
@@ -317,8 +319,24 @@ struct rpc_xprt *xprt_switch_find_next_entry_roundrobin(struct list_head *head,
 static
 struct rpc_xprt *xprt_iter_next_entry_roundrobin(struct rpc_xprt_iter *xpi)
 {
-	return xprt_iter_next_entry_multiple(xpi,
+	struct rpc_xprt_switch *xps = rcu_dereference(xpi->xpi_xpswitch);
+	struct rpc_xprt *xprt;
+	unsigned long xprt_queuelen;
+	unsigned long xps_queuelen;
+	unsigned long xps_avglen;
+
+	do {
+		xprt = xprt_iter_next_entry_multiple(xpi,
 			xprt_switch_find_next_entry_roundrobin);
+		if (xprt == NULL)
+			break;
+		xprt_queuelen = atomic_long_read(&xprt->queuelen);
+		if (xprt_queuelen <= 2)
+			break;
+		xps_queuelen = atomic_long_read(&xps->xps_queuelen);
+		xps_avglen = DIV_ROUND_UP(xps_queuelen, xps->xps_nactive);
+	} while (xprt_queuelen > xps_avglen);
+	return xprt;
 }
 
 static