1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see [sun.com URL with a
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lnet/klnds/socklnd/socklnd_lib-darwin.c
38 * Darwin porting library
39 * Make things easy to port
41 * Author: Phil Schwan <phil@clusterfs.com>
43 #include <mach/mach_types.h>
45 #include <netinet/in.h>
46 #include <netinet/tcp.h>
51 # if CONFIG_SYSCTL && !CFS_SYSFS_MODULE_PARM
55 SYSCTL_NODE (_lnet, OID_AUTO, ksocknal, CTLFLAG_RW,
56 0, "ksocknal_sysctl");
58 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, timeout,
59 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_timeout,
61 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, credits,
62 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_credits,
64 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, peer_credits,
65 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_peercredits,
67 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, nconnds,
68 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_nconnds,
70 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, min_reconnectms,
71 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_min_reconnectms,
72 0, "min_reconnectms");
73 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, max_reconnectms,
74 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_max_reconnectms,
75 0, "max_reconnectms");
76 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, eager_ack,
77 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_eager_ack,
79 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, typed,
80 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_typed_conns,
82 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, min_bulk,
83 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_min_bulk,
85 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, rx_buffer_size,
86 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_rx_buffer_size,
88 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, tx_buffer_size,
89 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_tx_buffer_size,
91 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, nagle,
92 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_nagle,
94 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_idle,
95 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_idle,
97 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_count,
98 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_count,
99 0, "keepalive_count");
100 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_intvl,
101 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_intvl,
102 0, "keepalive_intvl");
104 cfs_sysctl_table_t ksocknal_top_ctl_table [] = {
105 &sysctl__lnet_ksocknal,
106 &sysctl__lnet_ksocknal_timeout,
107 &sysctl__lnet_ksocknal_credits,
108 &sysctl__lnet_ksocknal_peer_credits,
109 &sysctl__lnet_ksocknal_nconnds,
110 &sysctl__lnet_ksocknal_min_reconnectms,
111 &sysctl__lnet_ksocknal_max_reconnectms,
112 &sysctl__lnet_ksocknal_eager_ack,
113 &sysctl__lnet_ksocknal_typed,
114 &sysctl__lnet_ksocknal_min_bulk,
115 &sysctl__lnet_ksocknal_rx_buffer_size,
116 &sysctl__lnet_ksocknal_tx_buffer_size,
117 &sysctl__lnet_ksocknal_nagle,
118 &sysctl__lnet_ksocknal_keepalive_idle,
119 &sysctl__lnet_ksocknal_keepalive_count,
120 &sysctl__lnet_ksocknal_keepalive_intvl,
125 ksocknal_lib_tunables_init ()
127 ksocknal_tunables.ksnd_sysctl =
128 cfs_register_sysctl_table (ksocknal_top_ctl_table, 0);
130 if (ksocknal_tunables.ksnd_sysctl == NULL)
137 ksocknal_lib_tunables_fini ()
139 if (ksocknal_tunables.ksnd_sysctl != NULL)
140 cfs_unregister_sysctl_table (ksocknal_tunables.ksnd_sysctl);
144 ksocknal_lib_tunables_init ()
150 ksocknal_lib_tunables_fini ()
156 * To use bigger buffer for socket:
157 * 1. Increase nmbclusters (Cannot increased by sysctl because it's ready only, so
158 * we must patch kernel).
159 * 2. Increase net.inet.tcp.reass.maxsegments
160 * 3. Increase net.inet.tcp.sendspace
161 * 4. Increase net.inet.tcp.recvspace
162 * 5. Increase kern.ipc.maxsockbuf
164 #define KSOCKNAL_MAX_BUFFER (1152*1024)
167 ksocknal_lib_bind_irq (unsigned int irq)
173 ksocknal_lib_sock_irq (cfs_socket_t *sock)
179 ksocknal_lib_get_conn_addrs (ksock_conn_t *conn)
181 int rc = libcfs_sock_getaddr(conn->ksnc_sock, 1,
185 /* Didn't need the {get,put}connsock dance to deref ksnc_sock... */
186 LASSERT (!conn->ksnc_closing);
189 CERROR ("Error %d getting sock peer IP\n", rc);
193 rc = libcfs_sock_getaddr(conn->ksnc_sock, 0,
194 &conn->ksnc_myipaddr, NULL);
196 CERROR ("Error %d getting sock local IP\n", rc);
206 ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
208 socket_t sock = C2B_SOCK(conn->ksnc_sock);
213 #if SOCKNAL_SINGLE_FRAG_TX
214 struct iovec scratch;
215 struct iovec *scratchiov = &scratch;
216 unsigned int niov = 1;
218 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
219 unsigned int niov = tx->tx_niov;
221 struct msghdr msg = {
224 .msg_iov = scratchiov,
228 .msg_flags = MSG_DONTWAIT
233 for (nob = i = 0; i < niov; i++) {
234 scratchiov[i] = tx->tx_iov[i];
235 nob += scratchiov[i].iov_len;
240 * Linux has MSG_MORE, do we have anything to
241 * reduce number of partial TCP segments sent?
243 rc = -sock_send(sock, &msg, MSG_DONTWAIT, &sndlen);
250 ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
252 socket_t sock = C2B_SOCK(conn->ksnc_sock);
253 lnet_kiov_t *kiov = tx->tx_kiov;
258 #if SOCKNAL_SINGLE_FRAG_TX
259 struct iovec scratch;
260 struct iovec *scratchiov = &scratch;
261 unsigned int niov = 1;
263 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
264 unsigned int niov = tx->tx_nkiov;
266 struct msghdr msg = {
269 .msg_iov = scratchiov,
273 .msg_flags = MSG_DONTWAIT
278 for (nob = i = 0; i < niov; i++) {
279 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) +
281 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
286 * Linux has MSG_MORE, do wen have anyting to
287 * reduce number of partial TCP segments sent?
289 rc = -sock_send(sock, &msg, MSG_DONTWAIT, &sndlen);
290 for (i = 0; i < niov; i++)
291 cfs_kunmap(kiov[i].kiov_page);
298 ksocknal_lib_recv_iov (ksock_conn_t *conn)
300 #if SOCKNAL_SINGLE_FRAG_RX
301 struct iovec scratch;
302 struct iovec *scratchiov = &scratch;
303 unsigned int niov = 1;
305 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
306 unsigned int niov = conn->ksnc_rx_niov;
308 struct iovec *iov = conn->ksnc_rx_iov;
309 struct msghdr msg = {
312 .msg_iov = scratchiov,
325 for (nob = i = 0; i < niov; i++) {
326 scratchiov[i] = iov[i];
327 nob += scratchiov[i].iov_len;
329 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
330 rc = -sock_receive (C2B_SOCK(conn->ksnc_sock), &msg, MSG_DONTWAIT, &rcvlen);
338 ksocknal_lib_recv_kiov (ksock_conn_t *conn)
340 #if SOCKNAL_SINGLE_FRAG_RX
341 struct iovec scratch;
342 struct iovec *scratchiov = &scratch;
343 unsigned int niov = 1;
345 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
346 unsigned int niov = conn->ksnc_rx_nkiov;
348 lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
349 struct msghdr msg = {
352 .msg_iov = scratchiov,
363 /* NB we can't trust socket ops to either consume our iovs
364 * or leave them alone. */
365 for (nob = i = 0; i < niov; i++) {
366 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) + \
368 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
370 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
371 rc = -sock_receive(C2B_SOCK(conn->ksnc_sock), &msg, MSG_DONTWAIT, &rcvlen);
372 for (i = 0; i < niov; i++)
373 cfs_kunmap(kiov[i].kiov_page);
380 ksocknal_lib_eager_ack (ksock_conn_t *conn)
386 ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
388 socket_t sock = C2B_SOCK(conn->ksnc_sock);
392 rc = ksocknal_connsock_addref(conn);
394 LASSERT (conn->ksnc_closing);
395 *txmem = *rxmem = *nagle = 0;
398 rc = libcfs_sock_getbuf(conn->ksnc_sock, txmem, rxmem);
400 len = sizeof(*nagle);
401 rc = -sock_getsockopt(sock, IPPROTO_TCP, TCP_NODELAY,
404 ksocknal_connsock_decref(conn);
409 *txmem = *rxmem = *nagle = 0;
415 ksocknal_lib_setup_sock (cfs_socket_t *sock)
423 socket_t so = C2B_SOCK(sock);
424 struct linger linger;
426 /* Ensure this socket aborts active sends immediately when we close
430 rc = -sock_setsockopt(so, SOL_SOCKET, SO_LINGER, &linger, sizeof(linger));
432 CERROR ("Can't set SO_LINGER: %d\n", rc);
436 if (!*ksocknal_tunables.ksnd_nagle) {
438 rc = -sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &option, sizeof(option));
440 CERROR ("Can't disable nagle: %d\n", rc);
445 rc = libcfs_sock_setbuf(sock,
446 *ksocknal_tunables.ksnd_tx_buffer_size,
447 *ksocknal_tunables.ksnd_rx_buffer_size);
449 CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
450 *ksocknal_tunables.ksnd_tx_buffer_size,
451 *ksocknal_tunables.ksnd_rx_buffer_size, rc);
455 /* snapshot tunables */
456 keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
457 keep_count = *ksocknal_tunables.ksnd_keepalive_count;
458 keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
460 do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
461 option = (do_keepalive ? 1 : 0);
463 rc = -sock_setsockopt(so, SOL_SOCKET, SO_KEEPALIVE, &option, sizeof(option));
465 CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
471 rc = -sock_setsockopt(so, IPPROTO_TCP, TCP_KEEPALIVE,
472 &keep_idle, sizeof(keep_idle));
478 ksocknal_lib_push_conn(ksock_conn_t *conn)
484 rc = ksocknal_connsock_addref(conn);
485 if (rc != 0) /* being shut down */
487 sock = C2B_SOCK(conn->ksnc_sock);
489 rc = -sock_setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, &val, sizeof(val));
492 ksocknal_connsock_decref(conn);
496 extern void ksocknal_read_callback (ksock_conn_t *conn);
497 extern void ksocknal_write_callback (ksock_conn_t *conn);
500 ksocknal_upcall(socket_t so, void *arg, int waitf)
502 ksock_conn_t *conn = (ksock_conn_t *)arg;
505 read_lock (&ksocknal_data.ksnd_global_lock);
509 ksocknal_read_callback (conn);
511 ksocknal_write_callback (conn);
513 read_unlock (&ksocknal_data.ksnd_global_lock);
518 ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn)
520 /* No callback need to save in osx */
525 ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn)
527 libcfs_sock_set_cb(sock, ksocknal_upcall, (void *)conn);
532 ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn)
534 libcfs_sock_reset_cb(sock);
537 #else /* !__DARWIN8__ */
540 ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
542 #if SOCKNAL_SINGLE_FRAG_TX
543 struct iovec scratch;
544 struct iovec *scratchiov = &scratch;
545 unsigned int niov = 1;
547 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
548 unsigned int niov = tx->tx_niov;
550 struct socket *sock = conn->ksnc_sock;
555 .uio_iov = scratchiov,
558 .uio_resid = 0, /* This will be valued after a while */
559 .uio_segflg = UIO_SYSSPACE,
563 int flags = MSG_DONTWAIT;
566 for (nob = i = 0; i < niov; i++) {
567 scratchiov[i] = tx->tx_iov[i];
568 nob += scratchiov[i].iov_len;
570 suio.uio_resid = nob;
573 rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
576 /* NB there is no return value can indicate how many
577 * have been sent and how many resid, we have to get
578 * sent bytes from suio. */
580 if (suio.uio_resid != nob &&\
581 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
582 /* We have sent something */
583 rc = nob - suio.uio_resid;
584 else if ( rc == EWOULDBLOCK )
585 /* Actually, EAGAIN and EWOULDBLOCK have same value in OSX */
590 rc = nob - suio.uio_resid;
596 ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
598 #if SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_RISK_KMAP_DEADLOCK
599 struct iovec scratch;
600 struct iovec *scratchiov = &scratch;
601 unsigned int niov = 1;
603 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
604 unsigned int niov = tx->tx_nkiov;
606 struct socket *sock = conn->ksnc_sock;
607 lnet_kiov_t *kiov = tx->tx_kiov;
612 .uio_iov = scratchiov,
615 .uio_resid = 0, /* It should be valued after a while */
616 .uio_segflg = UIO_SYSSPACE,
620 int flags = MSG_DONTWAIT;
623 for (nob = i = 0; i < niov; i++) {
624 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) +
626 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
628 suio.uio_resid = nob;
631 rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
634 for (i = 0; i < niov; i++)
635 cfs_kunmap(kiov[i].kiov_page);
638 if (suio.uio_resid != nob &&\
639 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
640 /* We have sent something */
641 rc = nob - suio.uio_resid;
642 else if ( rc == EWOULDBLOCK )
643 /* EAGAIN and EWOULD BLOCK have same value in OSX */
648 rc = nob - suio.uio_resid;
654 * liang: Hack of inpcb and tcpcb.
655 * To get tcpcb of a socket, and call tcp_output
666 LIST_HEAD(ks_tsegqe_head, ks_tseg_qent);
669 struct ks_tsegqe_head t_segq;
671 struct ks_tcptemp *unused;
673 struct inpcb *t_inpcb;
677 * There are more fields but we dont need
682 #define TF_ACKNOW 0x00001
683 #define TF_DELACK 0x00002
686 LIST_ENTRY(ks_inpcb) inp_hash;
687 struct in_addr reserved1;
688 struct in_addr reserved2;
691 LIST_ENTRY(inpcb) inp_list;
694 * There are more fields but we dont need
699 #define ks_sotoinpcb(so) ((struct ks_inpcb *)(so)->so_pcb)
700 #define ks_intotcpcb(ip) ((struct ks_tcpcb *)(ip)->inp_ppcb)
701 #define ks_sototcpcb(so) (intotcpcb(sotoinpcb(so)))
704 ksocknal_lib_eager_ack (ksock_conn_t *conn)
706 struct socket *sock = conn->ksnc_sock;
707 struct ks_inpcb *inp = ks_sotoinpcb(sock);
708 struct ks_tcpcb *tp = ks_intotcpcb(inp);
712 extern int tcp_output(register struct ks_tcpcb *tp);
718 * No TCP_QUICKACK supported in BSD, so I have to call tcp_fasttimo
719 * to send immediate ACK.
721 if (tp && tp->t_flags & TF_DELACK){
722 tp->t_flags &= ~TF_DELACK;
723 tp->t_flags |= TF_ACKNOW;
724 (void) tcp_output(tp);
734 ksocknal_lib_recv_iov (ksock_conn_t *conn)
736 #if SOCKNAL_SINGLE_FRAG_RX
737 struct iovec scratch;
738 struct iovec *scratchiov = &scratch;
739 unsigned int niov = 1;
741 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
742 unsigned int niov = conn->ksnc_rx_niov;
744 struct iovec *iov = conn->ksnc_rx_iov;
749 .uio_iov = scratchiov,
752 .uio_resid = 0, /* It should be valued after a while */
753 .uio_segflg = UIO_SYSSPACE,
757 int flags = MSG_DONTWAIT;
760 for (nob = i = 0; i < niov; i++) {
761 scratchiov[i] = iov[i];
762 nob += scratchiov[i].iov_len;
764 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
766 ruio.uio_resid = nob;
769 rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, (struct mbuf **)0, &flags);
772 if (ruio.uio_resid != nob && \
773 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK || rc == EAGAIN))
774 /* data particially received */
775 rc = nob - ruio.uio_resid;
776 else if (rc == EWOULDBLOCK)
777 /* EAGAIN and EWOULD BLOCK have same value in OSX */
782 rc = nob - ruio.uio_resid;
788 ksocknal_lib_recv_kiov (ksock_conn_t *conn)
790 #if SOCKNAL_SINGLE_FRAG_RX || !SOCKNAL_RISK_KMAP_DEADLOCK
791 struct iovec scratch;
792 struct iovec *scratchiov = &scratch;
793 unsigned int niov = 1;
795 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
796 unsigned int niov = conn->ksnc_rx_nkiov;
798 lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
803 .uio_iov = scratchiov,
807 .uio_segflg = UIO_SYSSPACE,
811 int flags = MSG_DONTWAIT;
814 for (nob = i = 0; i < niov; i++) {
815 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) + kiov[i].kiov_offset;
816 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
818 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
820 ruio.uio_resid = nob;
823 rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, NULL, &flags);
826 for (i = 0; i < niov; i++)
827 cfs_kunmap(kiov[i].kiov_page);
830 if (ruio.uio_resid != nob && \
831 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
832 /* data particially received */
833 rc = nob - ruio.uio_resid;
834 else if (rc == EWOULDBLOCK)
835 /* receive blocked, EWOULDBLOCK == EAGAIN */
840 rc = nob - ruio.uio_resid;
846 ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
848 struct socket *sock = conn->ksnc_sock;
851 rc = ksocknal_connsock_addref(conn);
853 LASSERT (conn->ksnc_closing);
854 *txmem = *rxmem = *nagle = 0;
857 rc = libcfs_sock_getbuf(sock, txmem, rxmem);
863 len = sizeof(*nagle);
864 bzero(&sopt, sizeof sopt);
865 sopt.sopt_dir = SOPT_GET;
866 sopt.sopt_level = IPPROTO_TCP;
867 sopt.sopt_name = TCP_NODELAY;
868 sopt.sopt_val = nagle;
869 sopt.sopt_valsize = len;
872 rc = -sogetopt(sock, &sopt);
876 ksocknal_connsock_decref(conn);
881 *txmem = *rxmem = *nagle = 0;
886 ksocknal_lib_setup_sock (struct socket *so)
895 struct linger linger;
898 rc = libcfs_sock_setbuf(so,
899 *ksocknal_tunables.ksnd_tx_buffer_size,
900 *ksocknal_tunables.ksnd_rx_buffer_size);
902 CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
903 *ksocknal_tunables.ksnd_tx_buffer_size,
904 *ksocknal_tunables.ksnd_rx_buffer_size, rc);
908 /* Ensure this socket aborts active sends immediately when we close
910 bzero(&sopt, sizeof sopt);
914 sopt.sopt_dir = SOPT_SET;
915 sopt.sopt_level = SOL_SOCKET;
916 sopt.sopt_name = SO_LINGER;
917 sopt.sopt_val = &linger;
918 sopt.sopt_valsize = sizeof(linger);
921 rc = -sosetopt(so, &sopt);
923 CERROR ("Can't set SO_LINGER: %d\n", rc);
927 if (!*ksocknal_tunables.ksnd_nagle) {
929 bzero(&sopt, sizeof sopt);
930 sopt.sopt_dir = SOPT_SET;
931 sopt.sopt_level = IPPROTO_TCP;
932 sopt.sopt_name = TCP_NODELAY;
933 sopt.sopt_val = &option;
934 sopt.sopt_valsize = sizeof(option);
935 rc = -sosetopt(so, &sopt);
937 CERROR ("Can't disable nagle: %d\n", rc);
942 /* snapshot tunables */
943 keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
944 keep_count = *ksocknal_tunables.ksnd_keepalive_count;
945 keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
947 do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
948 option = (do_keepalive ? 1 : 0);
949 bzero(&sopt, sizeof sopt);
950 sopt.sopt_dir = SOPT_SET;
951 sopt.sopt_level = SOL_SOCKET;
952 sopt.sopt_name = SO_KEEPALIVE;
953 sopt.sopt_val = &option;
954 sopt.sopt_valsize = sizeof(option);
955 rc = -sosetopt(so, &sopt);
957 CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
962 /* no more setting, just return */
967 bzero(&sopt, sizeof sopt);
968 sopt.sopt_dir = SOPT_SET;
969 sopt.sopt_level = IPPROTO_TCP;
970 sopt.sopt_name = TCP_KEEPALIVE;
971 sopt.sopt_val = &keep_idle;
972 sopt.sopt_valsize = sizeof(keep_idle);
973 rc = -sosetopt(so, &sopt);
975 CERROR ("Can't set TCP_KEEPALIVE : %d\n", rc);
984 ksocknal_lib_push_conn(ksock_conn_t *conn)
992 rc = ksocknal_connsock_addref(conn);
993 if (rc != 0) /* being shut down */
995 sock = conn->ksnc_sock;
996 bzero(&sopt, sizeof sopt);
997 sopt.sopt_dir = SOPT_SET;
998 sopt.sopt_level = IPPROTO_TCP;
999 sopt.sopt_name = TCP_NODELAY;
1000 sopt.sopt_val = &val;
1001 sopt.sopt_valsize = sizeof val;
1004 sosetopt(sock, &sopt);
1007 ksocknal_connsock_decref(conn);
1012 extern void ksocknal_read_callback (ksock_conn_t *conn);
1013 extern void ksocknal_write_callback (ksock_conn_t *conn);
1016 ksocknal_upcall(struct socket *so, caddr_t arg, int waitf)
1018 ksock_conn_t *conn = (ksock_conn_t *)arg;
1021 read_lock (&ksocknal_data.ksnd_global_lock);
1025 if (so->so_rcv.sb_flags & SB_UPCALL) {
1026 extern int soreadable(struct socket *so);
1027 if (conn->ksnc_rx_nob_wanted && soreadable(so))
1028 /* To verify whether the upcall is for receive */
1029 ksocknal_read_callback (conn);
1032 if (so->so_snd.sb_flags & SB_UPCALL){
1033 extern int sowriteable(struct socket *so);
1034 if (sowriteable(so))
1035 /* socket is writable */
1036 ksocknal_write_callback(conn);
1039 read_unlock (&ksocknal_data.ksnd_global_lock);
1045 ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn)
1047 /* No callback need to save in osx */
1052 ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn)
1057 sock->so_upcallarg = (void *)conn;
1058 sock->so_upcall = ksocknal_upcall;
1059 sock->so_snd.sb_timeo = 0;
1060 sock->so_rcv.sb_timeo = cfs_time_seconds(2);
1061 sock->so_rcv.sb_flags |= SB_UPCALL;
1062 sock->so_snd.sb_flags |= SB_UPCALL;
1068 ksocknal_lib_act_callback(struct socket *sock, ksock_conn_t *conn)
1073 ksocknal_upcall (sock, (void *)conn, 0);
1078 ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn)
1083 sock->so_rcv.sb_flags &= ~SB_UPCALL;
1084 sock->so_snd.sb_flags &= ~SB_UPCALL;
1085 sock->so_upcall = NULL;
1086 sock->so_upcallarg = NULL;
1090 #endif /* !__DARWIN8__ */