SUNRPC: Fix a livelock problem in the xprt->backlog queue

This patch ensures that we throttle new RPC requests if there are
requests already waiting in the xprt->backlog queue. The reason for
doing this is to fix livelock issues that can occur when an existing
(high priority) task is waiting in the backlog queue, gets woken up
by xprt_free_slot(), but a new task then steals the slot.

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

1 files changed, 58 insertions, 3 deletions

diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index b7478d5e7ffd..745fca3cfd36 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -948,6 +948,34 @@ void xprt_transmit(struct rpc_task *task)
 	spin_unlock_bh(&xprt->transport_lock);
 }
 
+static void xprt_add_backlog(struct rpc_xprt *xprt, struct rpc_task *task)
+{
+	set_bit(XPRT_CONGESTED, &xprt->state);
+	rpc_sleep_on(&xprt->backlog, task, NULL);
+}
+
+static void xprt_wake_up_backlog(struct rpc_xprt *xprt)
+{
+	if (rpc_wake_up_next(&xprt->backlog) == NULL)
+		clear_bit(XPRT_CONGESTED, &xprt->state);
+}
+
+static bool xprt_throttle_congested(struct rpc_xprt *xprt, struct rpc_task *task)
+{
+	bool ret = false;
+
+	if (!test_bit(XPRT_CONGESTED, &xprt->state))
+		goto out;
+	spin_lock(&xprt->reserve_lock);
+	if (test_bit(XPRT_CONGESTED, &xprt->state)) {
+		rpc_sleep_on(&xprt->backlog, task, NULL);
+		ret = true;
+	}
+	spin_unlock(&xprt->reserve_lock);
+out:
+	return ret;
+}
+
 static struct rpc_rqst *xprt_dynamic_alloc_slot(struct rpc_xprt *xprt, gfp_t gfp_flags)
 {
 	struct rpc_rqst *req = ERR_PTR(-EAGAIN);
@@ -992,7 +1020,7 @@ void xprt_alloc_slot(struct rpc_xprt *xprt, struct rpc_task *task)
 		task->tk_status = -ENOMEM;
 		break;
 	case -EAGAIN:
-		rpc_sleep_on(&xprt->backlog, task, NULL);
+		xprt_add_backlog(xprt, task);
 		dprintk("RPC:       waiting for request slot\n");
 	default:
 		task->tk_status = -EAGAIN;
@@ -1028,7 +1056,7 @@ static void xprt_free_slot(struct rpc_xprt *xprt, struct rpc_rqst *req)
 		memset(req, 0, sizeof(*req));	/* mark unused */
 		list_add(&req->rq_list, &xprt->free);
 	}
-	rpc_wake_up_next(&xprt->backlog);
+	xprt_wake_up_backlog(xprt);
 	spin_unlock(&xprt->reserve_lock);
 }
 
@@ -1092,7 +1120,8 @@ EXPORT_SYMBOL_GPL(xprt_free);
  * xprt_reserve - allocate an RPC request slot
  * @task: RPC task requesting a slot allocation
  *
- * If no more slots are available, place the task on the transport's
+ * If the transport is marked as being congested, or if no more
+ * slots are available, place the task on the transport's
  * backlog queue.
  */
 void xprt_reserve(struct rpc_task *task)
@@ -1107,6 +1136,32 @@ void xprt_reserve(struct rpc_task *task)
 	task->tk_status = -EAGAIN;
 	rcu_read_lock();
 	xprt = rcu_dereference(task->tk_client->cl_xprt);
+	if (!xprt_throttle_congested(xprt, task))
+		xprt->ops->alloc_slot(xprt, task);
+	rcu_read_unlock();
+}
+
+/**
+ * xprt_retry_reserve - allocate an RPC request slot
+ * @task: RPC task requesting a slot allocation
+ *
+ * If no more slots are available, place the task on the transport's
+ * backlog queue.
+ * Note that the only difference with xprt_reserve is that we now
+ * ignore the value of the XPRT_CONGESTED flag.
+ */
+void xprt_retry_reserve(struct rpc_task *task)
+{
+	struct rpc_xprt	*xprt;
+
+	task->tk_status = 0;
+	if (task->tk_rqstp != NULL)
+		return;
+
+	task->tk_timeout = 0;
+	task->tk_status = -EAGAIN;
+	rcu_read_lock();
+	xprt = rcu_dereference(task->tk_client->cl_xprt);
 	xprt->ops->alloc_slot(xprt, task);
 	rcu_read_unlock();
 }