From: Christoph Hellwig <hch@lst.de>
Date: Thu, 28 May 2020 07:12:19 +0200
Subject: tcp: add tcp_sock_set_nodelay
Patch-mainline: v5.8-rc1
Git-commit: 12abc5ee7873a085cc280240822b8ac53c86fecd
References: bsc#1177028
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>
Acked-by: Gary Lin <glin@suse.com>
---
drivers/block/drbd/drbd_int.h | 7 -----
drivers/block/drbd/drbd_main.c | 2 -
drivers/block/drbd/drbd_receiver.c | 4 +--
drivers/infiniband/sw/siw/siw_cm.c | 24 +++---------------
drivers/nvme/host/tcp.c | 9 ------
drivers/nvme/target/tcp.c | 12 +--------
drivers/target/iscsi/iscsi_target_login.c | 15 ++---------
fs/cifs/connect.c | 10 +------
fs/dlm/lowcomms.c | 8 +-----
fs/ocfs2/cluster/tcp.c | 20 +--------------
include/linux/tcp.h | 1
net/ceph/messenger.c | 11 +-------
net/ipv4/tcp.c | 39 +++++++++++++++++++-----------
net/rds/tcp.c | 11 --------
net/rds/tcp.h | 1
net/rds/tcp_listen.c | 2 -
16 files changed, 49 insertions(+), 127 deletions(-)
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1570,13 +1570,6 @@ extern void drbd_set_recv_tcq(struct drb
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;
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -660,7 +660,7 @@ static int __send_command(struct drbd_co
/* 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;
}
--- a/drivers/block/drbd/drbd_receiver.c
+++ b/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);
- drbd_tcp_nodelay(msock.socket);
+ tcp_sock_set_nodelay(sock.socket->sk);
+ tcp_sock_set_nodelay(msock.socket->sk);
connection->data.socket = sock.socket;
connection->meta.socket = msock.socket;
--- a/drivers/infiniband/sw/siw/siw_cm.c
+++ b/drivers/infiniband/sw/siw/siw_cm.c
@@ -947,16 +947,8 @@ static void siw_accept_newconn(struct si
siw_cep_get(new_cep);
new_s->sk->sk_user_data = new_cep;
- if (siw_tcp_nagle == false) {
- int val = 1;
-
- 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;
- }
- }
+ if (siw_tcp_nagle == false)
+ tcp_sock_set_nodelay(new_s->sk);
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, str
siw_dbg_qp(qp, "kernel_bindconnect: error %d\n", rv);
goto error;
}
- if (siw_tcp_nagle == false) {
- int val = 1;
-
- 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;
- }
- }
+ if (siw_tcp_nagle == false)
+ tcp_sock_set_nodelay(s->sk);
cep = siw_cep_alloc(sdev);
if (!cep) {
rv = -ENOMEM;
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -1373,14 +1373,7 @@ static int nvme_tcp_alloc_queue(struct n
}
/* Set TCP no delay */
- opt = 1;
- 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;
- }
+ tcp_sock_set_nodelay(queue->sock->sk);
/*
* Cleanup whatever is sitting in the TCP transmit queue on socket
--- a/drivers/nvme/target/tcp.c
+++ b/drivers/nvme/target/tcp.c
@@ -1587,7 +1587,7 @@ static int nvmet_tcp_add_port(struct nvm
{
struct nvmet_tcp_port *port;
__kernel_sa_family_t af;
- int opt, ret;
+ int ret;
port = kzalloc(sizeof(*port), GFP_KERNEL);
if (!port)
@@ -1632,15 +1632,7 @@ static int nvmet_tcp_add_port(struct nvm
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);
-
- 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;
- }
-
+ tcp_sock_set_nodelay(port->sock->sk);
if (so_priority > 0)
sock_set_priority(port->sock->sk, so_priority);
--- a/drivers/target/iscsi/iscsi_target_login.c
+++ b/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 */
- opt = 1;
- 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;
- }
- }
-
+ if (np->np_network_transport == ISCSI_TCP)
+ tcp_sock_set_nodelay(sock->sk);
sock_set_reuseaddr(sock->sk);
+ opt = 1;
ret = kernel_setsockopt(sock, IPPROTO_IP, IP_FREEBIND,
(char *)&opt, sizeof(opt));
if (ret < 0) {
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -3855,14 +3855,8 @@ generic_ip_connect(struct TCP_Server_Inf
socket->sk->sk_rcvbuf = 140 * 1024;
}
- if (server->tcp_nodelay) {
- int val = 1;
- 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);
- }
+ if (server->tcp_nodelay)
+ tcp_sock_set_nodelay(socket->sk);
cifs_dbg(FYI, "sndbuf %d rcvbuf %d rcvtimeo 0x%lx\n",
socket->sk->sk_sndbuf,
--- a/fs/dlm/lowcomms.c
+++ b/fs/dlm/lowcomms.c
@@ -1132,7 +1132,6 @@ static void tcp_connect_to_sock(struct c
struct sockaddr_storage saddr, src_addr;
int addr_len;
struct socket *sock = NULL;
- int one = 1;
int result;
if (con->nodeid == 0) {
@@ -1181,8 +1180,7 @@ static void tcp_connect_to_sock(struct c
log_print("connecting to %d", con->nodeid);
/* Turn off Nagle's algorithm */
- kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
- sizeof(one));
+ tcp_sock_set_nodelay(sock->sk);
result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
O_NONBLOCK);
@@ -1224,7 +1222,6 @@ static struct socket *tcp_create_listen_
{
struct socket *sock = NULL;
int result = 0;
- int one = 1;
int addr_len;
if (dlm_local_addr[0]->ss_family == AF_INET)
@@ -1241,8 +1238,7 @@ static struct socket *tcp_create_listen_
}
/* Turn off Nagle's algorithm */
- kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY, (char *)&one,
- sizeof(one));
+ tcp_sock_set_nodelay(sock->sk);
sock_set_reuseaddr(sock->sk);
--- a/fs/ocfs2/cluster/tcp.c
+++ b/fs/ocfs2/cluster/tcp.c
@@ -1441,14 +1441,6 @@ static void o2net_rx_until_empty(struct
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;
@@ -1638,11 +1630,7 @@ static void o2net_start_connect(struct w
goto out;
}
- ret = o2net_set_nodelay(sc->sc_sock);
- if (ret) {
- mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
- goto out;
- }
+ tcp_sock_set_nodelay(sc->sc_sock->sk);
ret = o2net_set_usertimeout(sock);
if (ret) {
@@ -1836,11 +1824,7 @@ static int o2net_accept_one(struct socke
*more = 1;
new_sock->sk->sk_allocation = GFP_ATOMIC;
- ret = o2net_set_nodelay(new_sock);
- if (ret) {
- mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
- goto out;
- }
+ tcp_sock_set_nodelay(new_sock->sk);
ret = o2net_set_usertimeout(new_sock);
if (ret) {
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -488,5 +488,6 @@ int tcp_skb_shift(struct sk_buff *to, st
int shiftlen);
void tcp_sock_set_cork(struct sock *sk, bool on);
+void tcp_sock_set_nodelay(struct sock *sk);
#endif /* _LINUX_TCP_H */
--- a/net/ceph/messenger.c
+++ b/net/ceph/messenger.c
@@ -490,15 +490,8 @@ static int ceph_tcp_connect(struct ceph_
return ret;
}
- if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY)) {
- int optval = 1;
-
- ret = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
- (char *)&optval, sizeof(optval));
- if (ret)
- pr_err("kernel_setsockopt(TCP_NODELAY) failed: %d",
- ret);
- }
+ if (ceph_test_opt(from_msgr(con->msgr), TCP_NODELAY))
+ tcp_sock_set_nodelay(sock->sk);
con->sock = sock;
return 0;
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2826,6 +2826,30 @@ void tcp_sock_set_cork(struct sock *sk,
}
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.
*/
@@ -2923,20 +2947,7 @@ static int do_tcp_setsockopt(struct sock
break;
case TCP_NODELAY:
- if (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;
- }
+ __tcp_sock_set_nodelay(sk, val);
break;
case TCP_THIN_LINEAR_TIMEOUTS:
--- a/net/rds/tcp.c
+++ b/net/rds/tcp.c
@@ -90,15 +90,6 @@ static struct ctl_table rds_tcp_sysctl_t
{ }
};
-/* 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 */
@@ -503,7 +494,7 @@ 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;
--- a/net/rds/tcp.h
+++ b/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,
--- a/net/rds/tcp_listen.c
+++ b/net/rds/tcp_listen.c
@@ -288,7 +288,7 @@ struct socket *rds_tcp_listen_init(struc
}
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;