tcp: add tcp_sock_set_nodelay

Add a helper to directly set the TCP_NODELAY sockopt from kernel space
without going through a fake uaccess.  Cleanup the callers to avoid
pointless wrappers now that this is a simple function call.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Sagi Grimberg <sagi@grimberg.me>
Acked-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/block/drbd/drbd_int.h

@@ -1570,13 +1570,6 @@ extern void drbd_set_recv_tcq(struct drbd_device *device, int tcq_enabled);
 extern void _drbd_clear_done_ee(struct drbd_device *device, struct list_head *to_be_freed);
 extern int drbd_connected(struct drbd_peer_device *);
 
-static inline void drbd_tcp_nodelay(struct socket *sock)
-{
-	int val = 1;
-	(void) kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
-			(char*)&val, sizeof(val));
-}
-
 static inline void drbd_tcp_quickack(struct socket *sock)
 {
 	int val = 2;

drivers/block/drbd/drbd_main.c

@@ -660,7 +660,7 @@ static int __send_command(struct drbd_connection *connection, int vnr,
 	/* DRBD protocol "pings" are latency critical.
 	 * This is supposed to trigger tcp_push_pending_frames() */
 	if (!err && (cmd == P_PING || cmd == P_PING_ACK))
-		drbd_tcp_nodelay(sock->socket);
+		tcp_sock_set_nodelay(sock->socket->sk);
 
 	return err;
 }

drivers/block/drbd/drbd_receiver.c

@@ -1051,8 +1051,8 @@ randomize:
 
 	/* we don't want delays.
 	 * we use TCP_CORK where appropriate, though */
-	drbd_tcp_nodelay(sock.socket);
+	tcp_sock_set_nodelay(sock.socket->sk);
-	drbd_tcp_nodelay(msock.socket);
+	tcp_sock_set_nodelay(msock.socket->sk);
 
 	connection->data.socket = sock.socket;
 	connection->meta.socket = msock.socket;

drivers/infiniband/sw/siw/siw_cm.c

@@ -947,16 +947,8 @@ static void siw_accept_newconn(struct siw_cep *cep)
 	siw_cep_get(new_cep);
 	new_s->sk->sk_user_data = new_cep;
 
-	if (siw_tcp_nagle == false) {
+	if (siw_tcp_nagle == false)
-		int val = 1;
+		tcp_sock_set_nodelay(new_s->sk);
-
-		rv = kernel_setsockopt(new_s, SOL_TCP, TCP_NODELAY,
-				       (char *)&val, sizeof(val));
-		if (rv) {
-			siw_dbg_cep(cep, "setsockopt NODELAY error: %d\n", rv);
-			goto error;
-		}
-	}
 	new_cep->state = SIW_EPSTATE_AWAIT_MPAREQ;
 
 	rv = siw_cm_queue_work(new_cep, SIW_CM_WORK_MPATIMEOUT);
@@ -1386,16 +1378,8 @@ int siw_connect(struct iw_cm_id *id, struct iw_cm_conn_param *params)
 		siw_dbg_qp(qp, "kernel_bindconnect: error %d\n", rv);
 		goto error;
 	}
-	if (siw_tcp_nagle == false) {
+	if (siw_tcp_nagle == false)
-		int val = 1;
+		tcp_sock_set_nodelay(s->sk);
-
-		rv = kernel_setsockopt(s, SOL_TCP, TCP_NODELAY, (char *)&val,
-				       sizeof(val));
-		if (rv) {
-			siw_dbg_qp(qp, "setsockopt NODELAY error: %d\n", rv);
-			goto error;
-		}
-	}
 	cep = siw_cep_alloc(sdev);
 	if (!cep) {
 		rv = -ENOMEM;

drivers/nvme/host/tcp.c

@@ -1346,14 +1346,7 @@ static int nvme_tcp_alloc_queue(struct nvme_ctrl *nctrl,
 	}
 
 	/* Set TCP no delay */
-	opt = 1;
+	tcp_sock_set_nodelay(queue->sock->sk);
-	ret = kernel_setsockopt(queue->sock, IPPROTO_TCP,
-			TCP_NODELAY, (char *)&opt, sizeof(opt));
-	if (ret) {
-		dev_err(nctrl->device,
-			"failed to set TCP_NODELAY sock opt %d\n", ret);
-		goto err_sock;
-	}
 
 	/*
 	 * Cleanup whatever is sitting in the TCP transmit queue on socket

drivers/nvme/target/tcp.c

@@ -1580,7 +1580,7 @@ static int nvmet_tcp_add_port(struct nvmet_port *nport)
 {
 	struct nvmet_tcp_port *port;
 	__kernel_sa_family_t af;
-	int opt, ret;
+	int ret;
 
 	port = kzalloc(sizeof(*port), GFP_KERNEL);
 	if (!port)
@@ -1625,15 +1625,7 @@ static int nvmet_tcp_add_port(struct nvmet_port *nport)
 	port->data_ready = port->sock->sk->sk_data_ready;
 	port->sock->sk->sk_data_ready = nvmet_tcp_listen_data_ready;
 	sock_set_reuseaddr(port->sock->sk);
-
+	tcp_sock_set_nodelay(port->sock->sk);
-	opt = 1;
-	ret = kernel_setsockopt(port->sock, IPPROTO_TCP,
-			TCP_NODELAY, (char *)&opt, sizeof(opt));
-	if (ret) {
-		pr_err("failed to set TCP_NODELAY sock opt %d\n", ret);
-		goto err_sock;
-	}
-
 	if (so_priority > 0)
 		sock_set_priority(port->sock->sk, so_priority);
 

drivers/target/iscsi/iscsi_target_login.c

@@ -897,20 +897,11 @@ int iscsit_setup_np(
 	/*
 	 * Set SO_REUSEADDR, and disable Nagel Algorithm with TCP_NODELAY.
 	 */
-	/* FIXME: Someone please explain why this is endian-safe */
+	if (np->np_network_transport == ISCSI_TCP)
-	opt = 1;
+		tcp_sock_set_nodelay(sock->sk);
-	if (np->np_network_transport == ISCSI_TCP) {
-		ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_NODELAY,
-				(char *)&opt, sizeof(opt));
-		if (ret < 0) {
-			pr_err("kernel_setsockopt() for TCP_NODELAY"
-				" failed: %d\n", ret);
-			goto fail;
-		}
-	}
-
 	sock_set_reuseaddr(sock->sk);
 
+	opt = 1;
 	ret = kernel_setsockopt(sock, IPPROTO_IP, IP_FREEBIND,
 			(char *)&opt, sizeof(opt));
 	if (ret < 0) {

fs/cifs/connect.c

@@ -3929,14 +3929,8 @@ generic_ip_connect(struct TCP_Server_Info *server)
 			socket->sk->sk_rcvbuf = 140 * 1024;
 	}
 
-	if (server->tcp_nodelay) {
+	if (server->tcp_nodelay)
-		int val = 1;
+		tcp_sock_set_nodelay(socket->sk);
-		rc = kernel_setsockopt(socket, SOL_TCP, TCP_NODELAY,
-				(char *)&val, sizeof(val));
-		if (rc)
-			cifs_dbg(FYI, "set TCP_NODELAY socket option error %d\n",
-				 rc);
-	}
 
 	cifs_dbg(FYI, "sndbuf %d rcvbuf %d rcvtimeo 0x%lx\n",
 		 socket->sk->sk_sndbuf,

fs/dlm/lowcomms.c

@@ -1011,7 +1011,6 @@ static void tcp_connect_to_sock(struct connection *con)
 	struct sockaddr_storage saddr, src_addr;
 	int addr_len;
 	struct socket *sock = NULL;
-	int one = 1;
 	int result;
 
 	if (con->nodeid == 0) {
@@ -1060,8 +1059,7 @@ static void tcp_connect_to_sock(struct connection *con)
 	log_print("connecting to %d", con->nodeid);
 
 	/* Turn off Nagle's algorithm */
-	kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
+	tcp_sock_set_nodelay(sock->sk);
-			  sizeof(one));
 
 	result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
 				   O_NONBLOCK);
@@ -1103,7 +1101,6 @@ static struct socket *tcp_create_listen_sock(struct connection *con,
 {
 	struct socket *sock = NULL;
 	int result = 0;
-	int one = 1;
 	int addr_len;
 
 	if (dlm_local_addr[0]->ss_family == AF_INET)
@@ -1120,8 +1117,7 @@ static struct socket *tcp_create_listen_sock(struct connection *con,
 	}
 
 	/* Turn off Nagle's algorithm */
-	kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
+	tcp_sock_set_nodelay(sock->sk);
-			  sizeof(one));
 
 	sock_set_reuseaddr(sock->sk);
 

fs/ocfs2/cluster/tcp.c

@@ -1441,14 +1441,6 @@ static void o2net_rx_until_empty(struct work_struct *work)
 	sc_put(sc);
 }
 
-static int o2net_set_nodelay(struct socket *sock)
-{
-	int val = 1;
-
-	return kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
-				    (void *)&val, sizeof(val));
-}
-
 static int o2net_set_usertimeout(struct socket *sock)
 {
 	int user_timeout = O2NET_TCP_USER_TIMEOUT;
@@ -1636,11 +1628,7 @@ static void o2net_start_connect(struct work_struct *work)
 		goto out;
 	}
 
-	ret = o2net_set_nodelay(sc->sc_sock);
+	tcp_sock_set_nodelay(sc->sc_sock->sk);
-	if (ret) {
-		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
-		goto out;
-	}
 
 	ret = o2net_set_usertimeout(sock);
 	if (ret) {
@@ -1832,11 +1820,7 @@ static int o2net_accept_one(struct socket *sock, int *more)
 	*more = 1;
 	new_sock->sk->sk_allocation = GFP_ATOMIC;
 
-	ret = o2net_set_nodelay(new_sock);
+	tcp_sock_set_nodelay(new_sock->sk);
-	if (ret) {
-		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
-		goto out;
-	}
 
 	ret = o2net_set_usertimeout(new_sock);
 	if (ret) {

include/linux/tcp.h

@@ -498,5 +498,6 @@ int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
 		  int shiftlen);
 
 void tcp_sock_set_cork(struct sock *sk, bool on);
+void tcp_sock_set_nodelay(struct sock *sk);
 
 #endif	/* _LINUX_TCP_H */

net/ceph/messenger.c

@@ -490,15 +490,8 @@ static int ceph_tcp_connect(struct ceph_connection *con)
 		return ret;
 	}
 
-	if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
+	if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
-		int optval = 1;
+		tcp_sock_set_nodelay(sock->sk);
-
-		ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
-					(char *)&optval, sizeof(optval));
-		if (ret)
-			pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
-			       ret);
-	}
 
 	con->sock = sock;
 	return 0;

net/ipv4/tcp.c

@@ -2832,6 +2832,30 @@ void tcp_sock_set_cork(struct sock *sk, bool on)
 }
 EXPORT_SYMBOL(tcp_sock_set_cork);
 
+/* TCP_NODELAY is weaker than TCP_CORK, so that this option on corked socket is
+ * remembered, but it is not activated until cork is cleared.
+ *
+ * However, when TCP_NODELAY is set we make an explicit push, which overrides
+ * even TCP_CORK for currently queued segments.
+ */
+static void __tcp_sock_set_nodelay(struct sock *sk, bool on)
+{
+	if (on) {
+		tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
+		tcp_push_pending_frames(sk);
+	} else {
+		tcp_sk(sk)->nonagle &= ~TCP_NAGLE_OFF;
+	}
+}
+
+void tcp_sock_set_nodelay(struct sock *sk)
+{
+	lock_sock(sk);
+	__tcp_sock_set_nodelay(sk, true);
+	release_sock(sk);
+}
+EXPORT_SYMBOL(tcp_sock_set_nodelay);
+
 /*
  *	Socket option code for TCP.
  */
@@ -2929,20 +2953,7 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
 		break;
 
 	case TCP_NODELAY:
-		if (val) {
+		__tcp_sock_set_nodelay(sk, val);
-			/* TCP_NODELAY is weaker than TCP_CORK, so that
-			 * this option on corked socket is remembered, but
-			 * it is not activated until cork is cleared.
-			 *
-			 * However, when TCP_NODELAY is set we make
-			 * an explicit push, which overrides even TCP_CORK
-			 * for currently queued segments.
-			 */
-			tp->nonagle |= TCP_NAGLE_OFF|TCP_NAGLE_PUSH;
-			tcp_push_pending_frames(sk);
-		} else {
-			tp->nonagle &= ~TCP_NAGLE_OFF;
-		}
 		break;
 
 	case TCP_THIN_LINEAR_TIMEOUTS:

net/rds/tcp.c

@@ -89,15 +89,6 @@ static struct ctl_table rds_tcp_sysctl_table[] = {
 	{ }
 };
 
-/* doing it this way avoids calling tcp_sk() */
-void rds_tcp_nonagle(struct socket *sock)
-{
-	int val = 1;
-
-	kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (void *)&val,
-			      sizeof(val));
-}
-
 u32 rds_tcp_write_seq(struct rds_tcp_connection *tc)
 {
 	/* seq# of the last byte of data in tcp send buffer */
@@ -502,7 +493,7 @@ void rds_tcp_tune(struct socket *sock)
 	struct net *net = sock_net(sk);
 	struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
 
-	rds_tcp_nonagle(sock);
+	tcp_sock_set_nodelay(sock->sk);
 	lock_sock(sk);
 	if (rtn->sndbuf_size > 0) {
 		sk->sk_sndbuf = rtn->sndbuf_size;

net/rds/tcp.h

@@ -50,7 +50,6 @@ struct rds_tcp_statistics {
 
 /* tcp.c */
 void rds_tcp_tune(struct socket *sock);
-void rds_tcp_nonagle(struct socket *sock);
 void rds_tcp_set_callbacks(struct socket *sock, struct rds_conn_path *cp);
 void rds_tcp_reset_callbacks(struct socket *sock, struct rds_conn_path *cp);
 void rds_tcp_restore_callbacks(struct socket *sock,

net/rds/tcp_listen.c

@@ -288,7 +288,7 @@ struct socket *rds_tcp_listen_init(struct net *net, bool isv6)
 	}
 
 	sock->sk->sk_reuse = SK_CAN_REUSE;
-	rds_tcp_nonagle(sock);
+	tcp_sock_set_nodelay(sock->sk);
 
 	write_lock_bh(&sock->sk->sk_callback_lock);
 	sock->sk->sk_user_data = sock->sk->sk_data_ready;