--- /dev/null
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.gnu.org/licenses/gpl-2.0.html
+ *
+ * GPL HEADER END
+ */
+
+#ifndef __LIBCFS_LINUX_NET_H__
+#define __LIBCFS_LINUX_NET_H__
+
+#ifdef HAVE_KERNEL_SETSOCKOPT
+
+#include <net/tcp.h>
+
+static inline void tcp_sock_set_quickack(struct sock *sk, int opt)
+{
+ struct socket *sock = sk->sk_socket;
+
+ kernel_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
+ (char *)&opt, sizeof(opt));
+}
+
+static inline void tcp_sock_set_nodelay(struct sock *sk)
+{
+ int opt = 1;
+ struct socket *sock = sk->sk_socket;
+
+ kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
+ (char *)&opt, sizeof(opt));
+}
+
+static inline int tcp_sock_set_keepidle(struct sock *sk, int opt)
+{
+ struct socket *sock = sk->sk_socket;
+
+ return kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
+ (char *)&opt, sizeof(opt));
+}
+
+static inline int tcp_sock_set_keepintvl(struct sock *sk, int opt)
+{
+ struct socket *sock = sk->sk_socket;
+
+ return kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
+ (char *)&opt, sizeof(opt));
+}
+
+static inline int tcp_sock_set_keepcnt(struct sock *sk, int opt)
+{
+ struct socket *sock = sk->sk_socket;
+
+ return kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
+ (char *)&opt, sizeof(opt));
+}
+#endif /* HAVE_KERNEL_SETSOCKOPT */
+
+#endif /* __LIBCFS_LINUX_NET_H__ */
void
ksocknal_lib_eager_ack(struct ksock_conn *conn)
{
- int opt = 1;
struct socket *sock = conn->ksnc_sock;
- /* Remind the socket to ACK eagerly. If I don't, the socket might
- * think I'm about to send something it could piggy-back the ACK
- * on, introducing delay in completing zero-copy sends in my
- * peer_ni. */
+ /* Remind the socket to ACK eagerly. If I don't, the socket might
+ * think I'm about to send something it could piggy-back the ACK on,
+ * introducing delay in completing zero-copy sends in my peer_ni.
+ */
- kernel_setsockopt(sock, SOL_TCP, TCP_QUICKACK,
- (char *)&opt, sizeof(opt));
+ tcp_sock_set_quickack(sock->sk, 1);
}
int
int
ksocknal_lib_get_conn_tunables(struct ksock_conn *conn, int *txmem, int *rxmem, int *nagle)
{
- struct socket *sock = conn->ksnc_sock;
- int len;
- int rc;
-
- rc = ksocknal_connsock_addref(conn);
- if (rc != 0) {
- LASSERT (conn->ksnc_closing);
- *txmem = *rxmem = *nagle = 0;
- return (-ESHUTDOWN);
- }
+ struct socket *sock = conn->ksnc_sock;
+ struct tcp_sock *tp = tcp_sk(sock->sk);
+
+ if (ksocknal_connsock_addref(conn) < 0) {
+ LASSERT(conn->ksnc_closing);
+ *txmem = 0;
+ *rxmem = 0;
+ *nagle = 0;
+ return -ESHUTDOWN;
+ }
- rc = lnet_sock_getbuf(sock, txmem, rxmem);
- if (rc == 0) {
- len = sizeof(*nagle);
- rc = kernel_getsockopt(sock, SOL_TCP, TCP_NODELAY,
- (char *)nagle, &len);
- }
+ lnet_sock_getbuf(sock, txmem, rxmem);
+
+ *nagle = !(tp->nonagle & TCP_NAGLE_OFF);
- ksocknal_connsock_decref(conn);
+ ksocknal_connsock_decref(conn);
- if (rc == 0)
- *nagle = !*nagle;
- else
- *txmem = *rxmem = *nagle = 0;
- return (rc);
+ return 0;
}
int
ksocknal_lib_setup_sock (struct socket *sock)
{
- int rc;
- int option;
- int keep_idle;
- int keep_intvl;
- int keep_count;
- int do_keepalive;
- struct linger linger;
-
- sock->sk->sk_allocation = GFP_NOFS;
-
- /* Ensure this socket aborts active sends immediately when we close
- * it. */
-
- linger.l_onoff = 0;
- linger.l_linger = 0;
-
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER,
- (char *)&linger, sizeof(linger));
- if (rc != 0) {
- CERROR ("Can't set SO_LINGER: %d\n", rc);
- return (rc);
- }
+ int rc;
+ int keep_idle;
+ int keep_intvl;
+ int keep_count;
+ int do_keepalive;
+ struct tcp_sock *tp = tcp_sk(sock->sk);
- option = -1;
- rc = kernel_setsockopt(sock, SOL_TCP, TCP_LINGER2,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR ("Can't set SO_LINGER2: %d\n", rc);
- return (rc);
- }
+ sock->sk->sk_allocation = GFP_NOFS;
- if (!*ksocknal_tunables.ksnd_nagle) {
- option = 1;
+ /* Ensure this socket aborts active sends immediately when closed. */
+ sock_reset_flag(sock->sk, SOCK_LINGER);
- rc = kernel_setsockopt(sock, SOL_TCP, TCP_NODELAY,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR ("Can't disable nagle: %d\n", rc);
- return (rc);
- }
- }
+ tp->linger2 = -1;
- rc = lnet_sock_setbuf(sock,
- *ksocknal_tunables.ksnd_tx_buffer_size,
- *ksocknal_tunables.ksnd_rx_buffer_size);
- if (rc != 0) {
- CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
- *ksocknal_tunables.ksnd_tx_buffer_size,
- *ksocknal_tunables.ksnd_rx_buffer_size, rc);
- return (rc);
- }
+ if (!*ksocknal_tunables.ksnd_nagle)
+ tcp_sock_set_nodelay(sock->sk);
+
+ lnet_sock_setbuf(sock,
+ *ksocknal_tunables.ksnd_tx_buffer_size,
+ *ksocknal_tunables.ksnd_rx_buffer_size);
/* TCP_BACKOFF_* sockopt tunables unsupported in stock kernels */
#ifdef SOCKNAL_BACKOFF
- if (*ksocknal_tunables.ksnd_backoff_init > 0) {
- option = *ksocknal_tunables.ksnd_backoff_init;
+ if (*ksocknal_tunables.ksnd_backoff_init > 0) {
+ int option = *ksocknal_tunables.ksnd_backoff_init;
#ifdef SOCKNAL_BACKOFF_MS
- option *= 1000;
+ option *= 1000;
#endif
rc = kernel_setsockopt(sock, SOL_TCP, TCP_BACKOFF_INIT,
(char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR ("Can't set initial tcp backoff %d: %d\n",
- option, rc);
- return (rc);
- }
- }
+ if (rc != 0) {
+ CERROR("Can't set initial tcp backoff %d: %d\n",
+ option, rc);
+ return rc;
+ }
+ }
- if (*ksocknal_tunables.ksnd_backoff_max > 0) {
- option = *ksocknal_tunables.ksnd_backoff_max;
+ if (*ksocknal_tunables.ksnd_backoff_max > 0) {
+ int option = *ksocknal_tunables.ksnd_backoff_max;
#ifdef SOCKNAL_BACKOFF_MS
- option *= 1000;
+ option *= 1000;
#endif
rc = kernel_setsockopt(sock, SOL_TCP, TCP_BACKOFF_MAX,
(char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR ("Can't set maximum tcp backoff %d: %d\n",
- option, rc);
- return (rc);
- }
- }
+ if (rc != 0) {
+ CERROR("Can't set maximum tcp backoff %d: %d\n",
+ option, rc);
+ return rc;
+ }
+ }
#endif
- /* snapshot tunables */
- keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
- keep_count = *ksocknal_tunables.ksnd_keepalive_count;
- keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
-
- do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
+ /* snapshot tunables */
+ keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
+ keep_count = *ksocknal_tunables.ksnd_keepalive_count;
+ keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
- option = (do_keepalive ? 1 : 0);
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
- return (rc);
- }
+ do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
- if (!do_keepalive)
- return (0);
-
- rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPIDLE,
- (char *)&keep_idle, sizeof(keep_idle));
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPIDLE: %d\n", rc);
- return (rc);
- }
+#ifdef HAVE_KERNEL_SETSOCKOPT
+ /* open-coded version doesn't work in all kernels, and
+ * there is no helper function, so call kernel_setsockopt()
+ * directly.
+ */
+ {
+ int option = (do_keepalive ? 1 : 0);
+ kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
+ (char *)&option, sizeof(option));
+ }
+#else
+ if (sock->sk->sk_prot->keepalive)
+ sock->sk->sk_prot->keepalive(sock->sk, do_keepalive);
+ if (do_keepalive)
+ sock_set_flag(sock->sk, SOCK_KEEPOPEN);
+ else
+ sock_reset_flag(sock->sk, SOCK_KEEPOPEN);
+#endif /* HAVE_KERNEL_SETSOCKOPT */
+
+ if (!do_keepalive)
+ return (0);
+
+ rc = tcp_sock_set_keepidle(sock->sk, keep_idle);
+ if (rc != 0) {
+ CERROR("Can't set TCP_KEEPIDLE: %d\n", rc);
+ return rc;
+ }
- rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPINTVL,
- (char *)&keep_intvl, sizeof(keep_intvl));
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPINTVL: %d\n", rc);
- return (rc);
- }
+ rc = tcp_sock_set_keepintvl(sock->sk, keep_intvl);
+ if (rc != 0) {
+ CERROR("Can't set TCP_KEEPINTVL: %d\n", rc);
+ return rc;
+ }
- rc = kernel_setsockopt(sock, SOL_TCP, TCP_KEEPCNT,
- (char *)&keep_count, sizeof(keep_count));
- if (rc != 0) {
- CERROR ("Can't set TCP_KEEPCNT: %d\n", rc);
- return (rc);
- }
+ rc = tcp_sock_set_keepcnt(sock->sk, keep_count);
+ if (rc != 0) {
+ CERROR("Can't set TCP_KEEPCNT: %d\n", rc);
+ return rc;
+ }
- return (0);
+ return (0);
}
void
struct sock *sk;
struct tcp_sock *tp;
int nonagle;
- int val = 1;
int rc;
- rc = ksocknal_connsock_addref(conn);
- if (rc != 0) /* being shut down */
- return;
+ rc = ksocknal_connsock_addref(conn);
+ if (rc != 0) /* being shut down */
+ return;
sk = conn->ksnc_sock->sk;
tp = tcp_sk(sk);
- lock_sock (sk);
- nonagle = tp->nonagle;
- tp->nonagle = 1;
- release_sock (sk);
+ lock_sock(sk);
+ nonagle = tp->nonagle;
+ tp->nonagle = TCP_NAGLE_OFF;
+ release_sock(sk);
- rc = kernel_setsockopt(conn->ksnc_sock, SOL_TCP, TCP_NODELAY,
- (char *)&val, sizeof(val));
- LASSERT (rc == 0);
+ tcp_sock_set_nodelay(conn->ksnc_sock->sk);
- lock_sock (sk);
- tp->nonagle = nonagle;
- release_sock (sk);
+ lock_sock(sk);
+ tp->nonagle = nonagle;
+ release_sock(sk);
- ksocknal_connsock_decref(conn);
+ ksocknal_connsock_decref(conn);
}
void ksocknal_read_callback(struct ksock_conn *conn);
#include <linux/inetdevice.h>
#include <libcfs/linux/linux-time.h>
+#include <libcfs/linux/linux-net.h>
#include <libcfs/libcfs.h>
#include <lnet/lib-lnet.h>
{
struct socket *sock;
int rc;
- int option;
int family;
family = AF_INET6;
return ERR_PTR(rc);
}
- option = 1;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
- (char *)&option, sizeof(option));
- if (rc) {
- CERROR("Can't set SO_REUSEADDR for socket: %d\n", rc);
- goto failed;
- }
+ sock->sk->sk_reuseport = 1;
if (interface >= 0 || local_port != 0) {
struct sockaddr_storage locaddr = {};
return ERR_PTR(rc);
}
-int
+void
lnet_sock_setbuf(struct socket *sock, int txbufsize, int rxbufsize)
{
- int option;
- int rc;
+ struct sock *sk = sock->sk;
if (txbufsize != 0) {
- option = txbufsize;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR("Can't set send buffer %d: %d\n",
- option, rc);
- return rc;
- }
+ sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
+ sk->sk_sndbuf = txbufsize;
+ sk->sk_write_space(sk);
}
if (rxbufsize != 0) {
- option = rxbufsize;
- rc = kernel_setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
- (char *)&option, sizeof(option));
- if (rc != 0) {
- CERROR("Can't set receive buffer %d: %d\n",
- option, rc);
- return rc;
- }
+ sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
+ sk->sk_sndbuf = rxbufsize;
}
- return 0;
}
EXPORT_SYMBOL(lnet_sock_setbuf);
}
EXPORT_SYMBOL(lnet_sock_getaddr);
-int
-lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
+void lnet_sock_getbuf(struct socket *sock, int *txbufsize, int *rxbufsize)
{
if (txbufsize != NULL)
*txbufsize = sock->sk->sk_sndbuf;
if (rxbufsize != NULL)
*rxbufsize = sock->sk->sk_rcvbuf;
-
- return 0;
}
EXPORT_SYMBOL(lnet_sock_getbuf);
git://git.whamcloud.com / fs / lustre-release.git / commitdiff