SUNRPC: Clean up transport write space handling

Treat socket write space handling in the same way we now treat transport
congestion: by denying the XPRT_LOCK until the transport signals that it
has free buffer space.
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

include/linux/sunrpc/svc_xprt.h

@@ -84,7 +84,6 @@ struct svc_xprt {
 	struct sockaddr_storage	xpt_remote;	/* remote peer's address */
 	size_t			xpt_remotelen;	/* length of address */
 	char			xpt_remotebuf[INET6_ADDRSTRLEN + 10];
-	struct rpc_wait_queue	xpt_bc_pending;	/* backchannel wait queue */
 	struct list_head	xpt_users;	/* callbacks on free */
 
 	struct net		*xpt_net;

include/linux/sunrpc/xprt.h

@@ -387,8 +387,8 @@ int			xprt_load_transport(const char *);
 void			xprt_set_retrans_timeout_def(struct rpc_task *task);
 void			xprt_set_retrans_timeout_rtt(struct rpc_task *task);
 void			xprt_wake_pending_tasks(struct rpc_xprt *xprt, int status);
-void			xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action);
-void			xprt_write_space(struct rpc_xprt *xprt);
+void			xprt_wait_for_buffer_space(struct rpc_xprt *xprt);
+bool			xprt_write_space(struct rpc_xprt *xprt);
 void			xprt_adjust_cwnd(struct rpc_xprt *xprt, struct rpc_task *task, int result);
 struct rpc_rqst *	xprt_lookup_rqst(struct rpc_xprt *xprt, __be32 xid);
 void			xprt_update_rtt(struct rpc_task *task);
@@ -416,6 +416,7 @@ void			xprt_unlock_connect(struct rpc_xprt *, void *);
 #define XPRT_CLOSING		(6)
 #define XPRT_CONGESTED		(9)
 #define XPRT_CWND_WAIT		(10)
+#define XPRT_WRITE_SPACE	(11)
 
 static inline void xprt_set_connected(struct rpc_xprt *xprt)
 {

net/sunrpc/clnt.c

@@ -1964,13 +1964,14 @@ call_transmit(struct rpc_task *task)
 {
 	dprint_status(task);
 
+	task->tk_status = 0;
+	if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate)) {
+		if (!xprt_prepare_transmit(task))
+			return;
+		xprt_transmit(task);
+	}
 	task->tk_action = call_transmit_status;
-	if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
-		return;
-
-	if (!xprt_prepare_transmit(task))
-		return;
-	xprt_transmit(task);
+	xprt_end_transmit(task);
 }
 
 /*
@@ -1986,7 +1987,6 @@ call_transmit_status(struct rpc_task *task)
 	 * test first.
 	 */
 	if (task->tk_status == 0) {
-		xprt_end_transmit(task);
 		xprt_request_wait_receive(task);
 		return;
 	}
@@ -1994,15 +1994,8 @@ call_transmit_status(struct rpc_task *task)
 	switch (task->tk_status) {
 	default:
 		dprint_status(task);
-		xprt_end_transmit(task);
-		break;
-	case -EBADSLT:
-		xprt_end_transmit(task);
-		task->tk_action = call_transmit;
-		task->tk_status = 0;
 		break;
 	case -EBADMSG:
-		xprt_end_transmit(task);
 		task->tk_status = 0;
 		task->tk_action = call_encode;
 		break;
@@ -2015,6 +2008,7 @@ call_transmit_status(struct rpc_task *task)
 	case -ENOBUFS:
 		rpc_delay(task, HZ>>2);
 		/* fall through */
+	case -EBADSLT:
 	case -EAGAIN:
 		task->tk_action = call_transmit;
 		task->tk_status = 0;
@@ -2026,7 +2020,6 @@ call_transmit_status(struct rpc_task *task)
 	case -ENETUNREACH:
 	case -EPERM:
 		if (RPC_IS_SOFTCONN(task)) {
-			xprt_end_transmit(task);
 			if (!task->tk_msg.rpc_proc->p_proc)
 				trace_xprt_ping(task->tk_xprt,
 						task->tk_status);
@@ -2069,9 +2062,6 @@ call_bc_transmit(struct rpc_task *task)
 
 	xprt_transmit(task);
 
-	if (task->tk_status == -EAGAIN)
-		goto out_retry;
-
 	xprt_end_transmit(task);
 	dprint_status(task);
 	switch (task->tk_status) {
@@ -2087,6 +2077,8 @@ call_bc_transmit(struct rpc_task *task)
 	case -ENOTCONN:
 	case -EPIPE:
 		break;
+	case -EAGAIN:
+		goto out_retry;
 	case -ETIMEDOUT:
 		/*
 		 * Problem reaching the server.  Disconnect and let the

net/sunrpc/svc_xprt.c

@@ -171,7 +171,6 @@ void svc_xprt_init(struct net *net, struct svc_xprt_class *xcl,
 	mutex_init(&xprt->xpt_mutex);
 	spin_lock_init(&xprt->xpt_lock);
 	set_bit(XPT_BUSY, &xprt->xpt_flags);
-	rpc_init_wait_queue(&xprt->xpt_bc_pending, "xpt_bc_pending");
 	xprt->xpt_net = get_net(net);
 	strcpy(xprt->xpt_remotebuf, "uninitialized");
 }
@@ -895,7 +894,6 @@ int svc_send(struct svc_rqst *rqstp)
 	else
 		len = xprt->xpt_ops->xpo_sendto(rqstp);
 	mutex_unlock(&xprt->xpt_mutex);
-	rpc_wake_up(&xprt->xpt_bc_pending);
 	trace_svc_send(rqstp, len);
 	svc_xprt_release(rqstp);
 

net/sunrpc/xprt.c

@@ -169,6 +169,17 @@ int xprt_load_transport(const char *transport_name)
 }
 EXPORT_SYMBOL_GPL(xprt_load_transport);
 
+static void xprt_clear_locked(struct rpc_xprt *xprt)
+{
+	xprt->snd_task = NULL;
+	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
+		smp_mb__before_atomic();
+		clear_bit(XPRT_LOCKED, &xprt->state);
+		smp_mb__after_atomic();
+	} else
+		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
+}
+
 /**
  * xprt_reserve_xprt - serialize write access to transports
  * @task: task that is requesting access to the transport
@@ -188,10 +199,14 @@ int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 			return 1;
 		goto out_sleep;
 	}
+	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
+		goto out_unlock;
 	xprt->snd_task = task;
 
 	return 1;
 
+out_unlock:
+	xprt_clear_locked(xprt);
 out_sleep:
 	dprintk("RPC: %5u failed to lock transport %p\n",
 			task->tk_pid, xprt);
@@ -208,17 +223,6 @@ int xprt_reserve_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 }
 EXPORT_SYMBOL_GPL(xprt_reserve_xprt);
 
-static void xprt_clear_locked(struct rpc_xprt *xprt)
-{
-	xprt->snd_task = NULL;
-	if (!test_bit(XPRT_CLOSE_WAIT, &xprt->state)) {
-		smp_mb__before_atomic();
-		clear_bit(XPRT_LOCKED, &xprt->state);
-		smp_mb__after_atomic();
-	} else
-		queue_work(xprtiod_workqueue, &xprt->task_cleanup);
-}
-
 static bool
 xprt_need_congestion_window_wait(struct rpc_xprt *xprt)
 {
@@ -267,10 +271,13 @@ int xprt_reserve_xprt_cong(struct rpc_xprt *xprt, struct rpc_task *task)
 		xprt->snd_task = task;
 		return 1;
 	}
+	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
+		goto out_unlock;
 	if (!xprt_need_congestion_window_wait(xprt)) {
 		xprt->snd_task = task;
 		return 1;
 	}
+out_unlock:
 	xprt_clear_locked(xprt);
 out_sleep:
 	dprintk("RPC: %5u failed to lock transport %p\n", task->tk_pid, xprt);
@@ -309,10 +316,12 @@ static void __xprt_lock_write_next(struct rpc_xprt *xprt)
 {
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 		return;
-
+	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
+		goto out_unlock;
 	if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
 				__xprt_lock_write_func, xprt))
 		return;
+out_unlock:
 	xprt_clear_locked(xprt);
 }
 
@@ -320,6 +329,8 @@ static void __xprt_lock_write_next_cong(struct rpc_xprt *xprt)
 {
 	if (test_and_set_bit(XPRT_LOCKED, &xprt->state))
 		return;
+	if (test_bit(XPRT_WRITE_SPACE, &xprt->state))
+		goto out_unlock;
 	if (xprt_need_congestion_window_wait(xprt))
 		goto out_unlock;
 	if (rpc_wake_up_first_on_wq(xprtiod_workqueue, &xprt->sending,
@@ -510,39 +521,46 @@ EXPORT_SYMBOL_GPL(xprt_wake_pending_tasks);
 
 /**
  * xprt_wait_for_buffer_space - wait for transport output buffer to clear
- * @task: task to be put to sleep
- * @action: function pointer to be executed after wait
+ * @xprt: transport
  *
  * Note that we only set the timer for the case of RPC_IS_SOFT(), since
  * we don't in general want to force a socket disconnection due to
  * an incomplete RPC call transmission.
  */
-void xprt_wait_for_buffer_space(struct rpc_task *task, rpc_action action)
+void xprt_wait_for_buffer_space(struct rpc_xprt *xprt)
 {
-	struct rpc_rqst *req = task->tk_rqstp;
-	struct rpc_xprt *xprt = req->rq_xprt;
-
-	task->tk_timeout = RPC_IS_SOFT(task) ? req->rq_timeout : 0;
-	rpc_sleep_on(&xprt->pending, task, action);
+	set_bit(XPRT_WRITE_SPACE, &xprt->state);
 }
 EXPORT_SYMBOL_GPL(xprt_wait_for_buffer_space);
 
+static bool
+xprt_clear_write_space_locked(struct rpc_xprt *xprt)
+{
+	if (test_and_clear_bit(XPRT_WRITE_SPACE, &xprt->state)) {
+		__xprt_lock_write_next(xprt);
+		dprintk("RPC:       write space: waking waiting task on "
+				"xprt %p\n", xprt);
+		return true;
+	}
+	return false;
+}
+
 /**
  * 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)
+bool xprt_write_space(struct rpc_xprt *xprt)
 {
+	bool ret;
+
+	if (!test_bit(XPRT_WRITE_SPACE, &xprt->state))
+		return false;
 	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_queued_task_on_wq(xprtiod_workqueue,
-				&xprt->pending, xprt->snd_task);
-	}
+	ret = xprt_clear_write_space_locked(xprt);
 	spin_unlock_bh(&xprt->transport_lock);
+	return ret;
 }
 EXPORT_SYMBOL_GPL(xprt_write_space);
 
@@ -653,6 +671,7 @@ void xprt_disconnect_done(struct rpc_xprt *xprt)
 	dprintk("RPC:       disconnected transport %p\n", xprt);
 	spin_lock_bh(&xprt->transport_lock);
 	xprt_clear_connected(xprt);
+	xprt_clear_write_space_locked(xprt);
 	xprt_wake_pending_tasks(xprt, -EAGAIN);
 	spin_unlock_bh(&xprt->transport_lock);
 }
@@ -1326,9 +1345,7 @@ xprt_transmit(struct rpc_task *task)
 			if (!xprt_request_data_received(task) ||
 			    test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
 				continue;
-		} else if (!test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
-			rpc_wake_up_queued_task(&xprt->pending, task);
-		else
+		} else if (test_bit(RPC_TASK_NEED_XMIT, &task->tk_runstate))
 			task->tk_status = status;
 		break;
 	}

net/sunrpc/xprtrdma/rpc_rdma.c

@@ -866,7 +866,7 @@ rpcrdma_marshal_req(struct rpcrdma_xprt *r_xprt, struct rpc_rqst *rqst)
 out_err:
 	switch (ret) {
 	case -EAGAIN:
-		xprt_wait_for_buffer_space(rqst->rq_task, NULL);
+		xprt_wait_for_buffer_space(rqst->rq_xprt);
 		break;
 	case -ENOBUFS:
 		break;

net/sunrpc/xprtrdma/svc_rdma_backchannel.c

@@ -224,12 +224,7 @@ xprt_rdma_bc_send_request(struct rpc_rqst *rqst, struct rpc_task *task)
 	dprintk("svcrdma: sending bc call with xid: %08x\n",
 		be32_to_cpu(rqst->rq_xid));
 
-	if (!mutex_trylock(&sxprt->xpt_mutex)) {
-		rpc_sleep_on(&sxprt->xpt_bc_pending, task, NULL);
-		if (!mutex_trylock(&sxprt->xpt_mutex))
-			return -EAGAIN;
-		rpc_wake_up_queued_task(&sxprt->xpt_bc_pending, task);
-	}
+	mutex_lock(&sxprt->xpt_mutex);
 
 	ret = -ENOTCONN;
 	rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt);

net/sunrpc/xprtsock.c

@@ -440,20 +440,12 @@ static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
 	return err;
 }
 
-static void xs_nospace_callback(struct rpc_task *task)
-{
-	struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
-
-	transport->inet->sk_write_pending--;
-}
-
 /**
- * xs_nospace - place task on wait queue if transmit was incomplete
+ * xs_nospace - handle transmit was incomplete
  * @req: pointer to RPC request
- * @task: task to put to sleep
  *
  */
-static int xs_nospace(struct rpc_rqst *req, struct rpc_task *task)
+static int xs_nospace(struct rpc_rqst *req)
 {
 	struct rpc_xprt *xprt = req->rq_xprt;
 	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
@@ -461,7 +453,8 @@ static int xs_nospace(struct rpc_rqst *req, struct rpc_task *task)
 	int ret = -EAGAIN;
 
 	dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
-			task->tk_pid, req->rq_slen - transport->xmit.offset,
+			req->rq_task->tk_pid,
+			req->rq_slen - transport->xmit.offset,
 			req->rq_slen);
 
 	/* Protect against races with write_space */
@@ -471,7 +464,7 @@ static int xs_nospace(struct rpc_rqst *req, struct rpc_task *task)
 	if (xprt_connected(xprt)) {
 		/* wait for more buffer space */
 		sk->sk_write_pending++;
-		xprt_wait_for_buffer_space(task, xs_nospace_callback);
+		xprt_wait_for_buffer_space(xprt);
 	} else
 		ret = -ENOTCONN;
 
@@ -569,7 +562,7 @@ static int xs_local_send_request(struct rpc_rqst *req, struct rpc_task *task)
 	case -ENOBUFS:
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req, task);
+		status = xs_nospace(req);
 		break;
 	default:
 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
@@ -642,7 +635,7 @@ static int xs_udp_send_request(struct rpc_rqst *req, struct rpc_task *task)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req, task);
+		status = xs_nospace(req);
 		break;
 	case -ENETUNREACH:
 	case -ENOBUFS:
@@ -765,7 +758,7 @@ static int xs_tcp_send_request(struct rpc_rqst *req, struct rpc_task *task)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req, task);
+		status = xs_nospace(req);
 		break;
 	case -ECONNRESET:
 	case -ECONNREFUSED:
@@ -1672,7 +1665,8 @@ static void xs_write_space(struct sock *sk)
 	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
 		goto out;
 
-	xprt_write_space(xprt);
+	if (xprt_write_space(xprt))
+		sk->sk_write_pending--;
 out:
 	rcu_read_unlock();
 }
@@ -2725,12 +2719,7 @@ static int bc_send_request(struct rpc_rqst *req, struct rpc_task *task)
 	 * Grab the mutex to serialize data as the connection is shared
 	 * with the fore channel
 	 */
-	if (!mutex_trylock(&xprt->xpt_mutex)) {
-		rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
-		if (!mutex_trylock(&xprt->xpt_mutex))
-			return -EAGAIN;
-		rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
-	}
+	mutex_lock(&xprt->xpt_mutex);
 	if (test_bit(XPT_DEAD, &xprt->xpt_flags))
 		len = -ENOTCONN;
 	else