3 files changed, 65 insertions, 16 deletions

diff --git a/net/sunrpc/sched.c b/net/sunrpc/sched.c
index 243fc09b164..3fc8624fcd1 100644
--- a/net/sunrpc/sched.c
+++ b/net/sunrpc/sched.c
@@ -252,23 +252,37 @@ static void rpc_set_active(struct rpc_task *task)
 
 /*
  * Mark an RPC call as having completed by clearing the 'active' bit
+ * and then waking up all tasks that were sleeping.
  */
-static void rpc_mark_complete_task(struct rpc_task *task)
+static int rpc_complete_task(struct rpc_task *task)
 {
-	smp_mb__before_clear_bit();
+	void *m = &task->tk_runstate;
+	wait_queue_head_t *wq = bit_waitqueue(m, RPC_TASK_ACTIVE);
+	struct wait_bit_key k = __WAIT_BIT_KEY_INITIALIZER(m, RPC_TASK_ACTIVE);
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&wq->lock, flags);
 	clear_bit(RPC_TASK_ACTIVE, &task->tk_runstate);
-	smp_mb__after_clear_bit();
-	wake_up_bit(&task->tk_runstate, RPC_TASK_ACTIVE);
+	ret = atomic_dec_and_test(&task->tk_count);
+	if (waitqueue_active(wq))
+		__wake_up_locked_key(wq, TASK_NORMAL, &k);
+	spin_unlock_irqrestore(&wq->lock, flags);
+	return ret;
 }
 
 /*
  * Allow callers to wait for completion of an RPC call
+ *
+ * Note the use of out_of_line_wait_on_bit() rather than wait_on_bit()
+ * to enforce taking of the wq->lock and hence avoid races with
+ * rpc_complete_task().
  */
 int __rpc_wait_for_completion_task(struct rpc_task *task, int (*action)(void *))
 {
 	if (action == NULL)
 		action = rpc_wait_bit_killable;
-	return wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
+	return out_of_line_wait_on_bit(&task->tk_runstate, RPC_TASK_ACTIVE,
 			action, TASK_KILLABLE);
 }
 EXPORT_SYMBOL_GPL(__rpc_wait_for_completion_task);
@@ -857,34 +871,67 @@ static void rpc_async_release(struct work_struct *work)
 	rpc_free_task(container_of(work, struct rpc_task, u.tk_work));
 }
 
-void rpc_put_task(struct rpc_task *task)
+static void rpc_release_resources_task(struct rpc_task *task)
 {
-	if (!atomic_dec_and_test(&task->tk_count))
-		return;
-	/* Release resources */
 	if (task->tk_rqstp)
 		xprt_release(task);
 	if (task->tk_msg.rpc_cred)
 		put_rpccred(task->tk_msg.rpc_cred);
 	rpc_task_release_client(task);
-	if (task->tk_workqueue != NULL) {
+}
+
+static void rpc_final_put_task(struct rpc_task *task,
+		struct workqueue_struct *q)
+{
+	if (q != NULL) {
 		INIT_WORK(&task->u.tk_work, rpc_async_release);
-		queue_work(task->tk_workqueue, &task->u.tk_work);
+		queue_work(q, &task->u.tk_work);
 	} else
 		rpc_free_task(task);
 }
+
+static void rpc_do_put_task(struct rpc_task *task, struct workqueue_struct *q)
+{
+	if (atomic_dec_and_test(&task->tk_count)) {
+		rpc_release_resources_task(task);
+		rpc_final_put_task(task, q);
+	}
+}
+
+void rpc_put_task(struct rpc_task *task)
+{
+	rpc_do_put_task(task, NULL);
+}
 EXPORT_SYMBOL_GPL(rpc_put_task);
 
+void rpc_put_task_async(struct rpc_task *task)
+{
+	rpc_do_put_task(task, task->tk_workqueue);
+}
+EXPORT_SYMBOL_GPL(rpc_put_task_async);
+
 static void rpc_release_task(struct rpc_task *task)
 {
 	dprintk("RPC: %5u release task\n", task->tk_pid);
 
 	BUG_ON (RPC_IS_QUEUED(task));
 
-	/* Wake up anyone who is waiting for task completion */
-	rpc_mark_complete_task(task);
+	rpc_release_resources_task(task);
 
-	rpc_put_task(task);
+	/*
+	 * Note: at this point we have been removed from rpc_clnt->cl_tasks,
+	 * so it should be safe to use task->tk_count as a test for whether
+	 * or not any other processes still hold references to our rpc_task.
+	 */
+	if (atomic_read(&task->tk_count) != 1 + !RPC_IS_ASYNC(task)) {
+		/* Wake up anyone who may be waiting for task completion */
+		if (!rpc_complete_task(task))
+			return;
+	} else {
+		if (!atomic_dec_and_test(&task->tk_count))
+			return;
+	}
+	rpc_final_put_task(task, task->tk_workqueue);
 }
 
 int rpciod_up(void)
@@ -908,7 +955,7 @@ static int rpciod_start(void)
 	 * Create the rpciod thread and wait for it to start.
 	 */
 	dprintk("RPC:       creating workqueue rpciod\n");
-	wq = alloc_workqueue("rpciod", WQ_RESCUER, 0);
+	wq = alloc_workqueue("rpciod", WQ_MEM_RECLAIM, 0);
 	rpciod_workqueue = wq;
 	return rpciod_workqueue != NULL;
 }
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 9df1eadc912..1a10dcd999e 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -1335,6 +1335,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
 					    p, 0, length, DMA_FROM_DEVICE);
 	if (ib_dma_mapping_error(xprt->sc_cm_id->device, ctxt->sge[0].addr)) {
 		put_page(p);
+		svc_rdma_put_context(ctxt, 1);
 		return;
 	}
 	atomic_inc(&xprt->sc_dma_used);
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index c431f5a5796..be96d429b47 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -1631,7 +1631,8 @@ static struct socket *xs_create_sock(struct rpc_xprt *xprt,
 	}
 	xs_reclassify_socket(family, sock);
 
-	if (xs_bind(transport, sock)) {
+	err = xs_bind(transport, sock);
+	if (err) {
 		sock_release(sock);
 		goto out;
 	}