1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
5 * Author: Phil Schwan <phil@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 * Darwin porting library
23 * Make things easy to port
25 #include <mach/mach_types.h>
27 #include <netinet/in.h>
28 #include <netinet/tcp.h>
33 # if CONFIG_SYSCTL && !CFS_SYSFS_MODULE_PARM
37 SYSCTL_NODE (_lnet, OID_AUTO, ksocknal, CTLFLAG_RW,
38 0, "ksocknal_sysctl");
40 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, timeout,
41 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_timeout,
43 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, credits,
44 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_credits,
46 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, peer_credits,
47 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_peercredits,
49 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, nconnds,
50 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_nconnds,
52 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, min_reconnectms,
53 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_min_reconnectms,
54 0, "min_reconnectms");
55 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, max_reconnectms,
56 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_max_reconnectms,
57 0, "max_reconnectms");
58 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, eager_ack,
59 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_eager_ack,
61 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, typed,
62 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_typed_conns,
64 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, min_bulk,
65 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_min_bulk,
67 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, rx_buffer_size,
68 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_rx_buffer_size,
70 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, tx_buffer_size,
71 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_tx_buffer_size,
73 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, nagle,
74 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_nagle,
76 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_idle,
77 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_idle,
79 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_count,
80 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_count,
81 0, "keepalive_count");
82 SYSCTL_INT(_lnet_ksocknal, OID_AUTO, keepalive_intvl,
83 CTLTYPE_INT | CTLFLAG_RW , &ksocknal_tunables.ksnd_keepalive_intvl,
84 0, "keepalive_intvl");
86 cfs_sysctl_table_t ksocknal_top_ctl_table [] = {
87 &sysctl__lnet_ksocknal,
88 &sysctl__lnet_ksocknal_timeout,
89 &sysctl__lnet_ksocknal_credits,
90 &sysctl__lnet_ksocknal_peer_credits,
91 &sysctl__lnet_ksocknal_nconnds,
92 &sysctl__lnet_ksocknal_min_reconnectms,
93 &sysctl__lnet_ksocknal_max_reconnectms,
94 &sysctl__lnet_ksocknal_eager_ack,
95 &sysctl__lnet_ksocknal_typed,
96 &sysctl__lnet_ksocknal_min_bulk,
97 &sysctl__lnet_ksocknal_rx_buffer_size,
98 &sysctl__lnet_ksocknal_tx_buffer_size,
99 &sysctl__lnet_ksocknal_nagle,
100 &sysctl__lnet_ksocknal_keepalive_idle,
101 &sysctl__lnet_ksocknal_keepalive_count,
102 &sysctl__lnet_ksocknal_keepalive_intvl,
107 ksocknal_lib_tunables_init ()
109 ksocknal_tunables.ksnd_sysctl =
110 cfs_register_sysctl_table (ksocknal_top_ctl_table, 0);
112 if (ksocknal_tunables.ksnd_sysctl == NULL)
119 ksocknal_lib_tunables_fini ()
121 if (ksocknal_tunables.ksnd_sysctl != NULL)
122 cfs_unregister_sysctl_table (ksocknal_tunables.ksnd_sysctl);
126 ksocknal_lib_tunables_init ()
132 ksocknal_lib_tunables_fini ()
138 * To use bigger buffer for socket:
139 * 1. Increase nmbclusters (Cannot increased by sysctl because it's ready only, so
140 * we must patch kernel).
141 * 2. Increase net.inet.tcp.reass.maxsegments
142 * 3. Increase net.inet.tcp.sendspace
143 * 4. Increase net.inet.tcp.recvspace
144 * 5. Increase kern.ipc.maxsockbuf
146 #define KSOCKNAL_MAX_BUFFER (1152*1024)
149 ksocknal_lib_bind_irq (unsigned int irq)
155 ksocknal_lib_sock_irq (cfs_socket_t *sock)
161 ksocknal_lib_get_conn_addrs (ksock_conn_t *conn)
163 int rc = libcfs_sock_getaddr(conn->ksnc_sock, 1,
167 /* Didn't need the {get,put}connsock dance to deref ksnc_sock... */
168 LASSERT (!conn->ksnc_closing);
171 CERROR ("Error %d getting sock peer IP\n", rc);
175 rc = libcfs_sock_getaddr(conn->ksnc_sock, 0,
176 &conn->ksnc_myipaddr, NULL);
178 CERROR ("Error %d getting sock local IP\n", rc);
188 ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
190 socket_t sock = C2B_SOCK(conn->ksnc_sock);
195 #if SOCKNAL_SINGLE_FRAG_TX
196 struct iovec scratch;
197 struct iovec *scratchiov = &scratch;
198 unsigned int niov = 1;
200 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
201 unsigned int niov = tx->tx_niov;
203 struct msghdr msg = {
206 .msg_iov = scratchiov,
210 .msg_flags = MSG_DONTWAIT
215 for (nob = i = 0; i < niov; i++) {
216 scratchiov[i] = tx->tx_iov[i];
217 nob += scratchiov[i].iov_len;
222 * Linux has MSG_MORE, do we have anything to
223 * reduce number of partial TCP segments sent?
225 rc = -sock_send(sock, &msg, MSG_DONTWAIT, &sndlen);
232 ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
234 socket_t sock = C2B_SOCK(conn->ksnc_sock);
235 lnet_kiov_t *kiov = tx->tx_kiov;
240 #if SOCKNAL_SINGLE_FRAG_TX
241 struct iovec scratch;
242 struct iovec *scratchiov = &scratch;
243 unsigned int niov = 1;
245 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
246 unsigned int niov = tx->tx_nkiov;
248 struct msghdr msg = {
251 .msg_iov = scratchiov,
255 .msg_flags = MSG_DONTWAIT
260 for (nob = i = 0; i < niov; i++) {
261 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) +
263 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
268 * Linux has MSG_MORE, do wen have anyting to
269 * reduce number of partial TCP segments sent?
271 rc = -sock_send(sock, &msg, MSG_DONTWAIT, &sndlen);
272 for (i = 0; i < niov; i++)
273 cfs_kunmap(kiov[i].kiov_page);
280 ksocknal_lib_recv_iov (ksock_conn_t *conn)
282 #if SOCKNAL_SINGLE_FRAG_RX
283 struct iovec scratch;
284 struct iovec *scratchiov = &scratch;
285 unsigned int niov = 1;
287 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
288 unsigned int niov = conn->ksnc_rx_niov;
290 struct iovec *iov = conn->ksnc_rx_iov;
291 struct msghdr msg = {
294 .msg_iov = scratchiov,
307 for (nob = i = 0; i < niov; i++) {
308 scratchiov[i] = iov[i];
309 nob += scratchiov[i].iov_len;
311 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
312 rc = -sock_receive (C2B_SOCK(conn->ksnc_sock), &msg, MSG_DONTWAIT, &rcvlen);
320 ksocknal_lib_recv_kiov (ksock_conn_t *conn)
322 #if SOCKNAL_SINGLE_FRAG_RX
323 struct iovec scratch;
324 struct iovec *scratchiov = &scratch;
325 unsigned int niov = 1;
327 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
328 unsigned int niov = conn->ksnc_rx_nkiov;
330 lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
331 struct msghdr msg = {
334 .msg_iov = scratchiov,
345 /* NB we can't trust socket ops to either consume our iovs
346 * or leave them alone. */
347 for (nob = i = 0; i < niov; i++) {
348 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) + \
350 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
352 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
353 rc = -sock_receive(C2B_SOCK(conn->ksnc_sock), &msg, MSG_DONTWAIT, &rcvlen);
354 for (i = 0; i < niov; i++)
355 cfs_kunmap(kiov[i].kiov_page);
362 ksocknal_lib_eager_ack (ksock_conn_t *conn)
368 ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
370 socket_t sock = C2B_SOCK(conn->ksnc_sock);
374 rc = ksocknal_connsock_addref(conn);
376 LASSERT (conn->ksnc_closing);
377 *txmem = *rxmem = *nagle = 0;
380 rc = libcfs_sock_getbuf(conn->ksnc_sock, txmem, rxmem);
382 len = sizeof(*nagle);
383 rc = -sock_getsockopt(sock, IPPROTO_TCP, TCP_NODELAY,
386 ksocknal_connsock_decref(conn);
391 *txmem = *rxmem = *nagle = 0;
397 ksocknal_lib_setup_sock (cfs_socket_t *sock)
405 socket_t so = C2B_SOCK(sock);
406 struct linger linger;
408 /* Ensure this socket aborts active sends immediately when we close
412 rc = -sock_setsockopt(so, SOL_SOCKET, SO_LINGER, &linger, sizeof(linger));
414 CERROR ("Can't set SO_LINGER: %d\n", rc);
418 if (!*ksocknal_tunables.ksnd_nagle) {
420 rc = -sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &option, sizeof(option));
422 CERROR ("Can't disable nagle: %d\n", rc);
427 rc = libcfs_sock_setbuf(sock,
428 *ksocknal_tunables.ksnd_tx_buffer_size,
429 *ksocknal_tunables.ksnd_rx_buffer_size);
431 CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
432 *ksocknal_tunables.ksnd_tx_buffer_size,
433 *ksocknal_tunables.ksnd_rx_buffer_size, rc);
437 /* snapshot tunables */
438 keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
439 keep_count = *ksocknal_tunables.ksnd_keepalive_count;
440 keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
442 do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
443 option = (do_keepalive ? 1 : 0);
445 rc = -sock_setsockopt(so, SOL_SOCKET, SO_KEEPALIVE, &option, sizeof(option));
447 CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
453 rc = -sock_setsockopt(so, IPPROTO_TCP, TCP_KEEPALIVE,
454 &keep_idle, sizeof(keep_idle));
460 ksocknal_lib_push_conn(ksock_conn_t *conn)
466 rc = ksocknal_connsock_addref(conn);
467 if (rc != 0) /* being shut down */
469 sock = C2B_SOCK(conn->ksnc_sock);
471 rc = -sock_setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, &val, sizeof(val));
474 ksocknal_connsock_decref(conn);
478 extern void ksocknal_read_callback (ksock_conn_t *conn);
479 extern void ksocknal_write_callback (ksock_conn_t *conn);
482 ksocknal_upcall(socket_t so, void *arg, int waitf)
484 ksock_conn_t *conn = (ksock_conn_t *)arg;
487 read_lock (&ksocknal_data.ksnd_global_lock);
491 ksocknal_read_callback (conn);
493 ksocknal_write_callback (conn);
495 read_unlock (&ksocknal_data.ksnd_global_lock);
500 ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn)
502 /* No callback need to save in osx */
507 ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn)
509 libcfs_sock_set_cb(sock, ksocknal_upcall, (void *)conn);
514 ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn)
516 libcfs_sock_reset_cb(sock);
519 #else /* !__DARWIN8__ */
522 ksocknal_lib_send_iov (ksock_conn_t *conn, ksock_tx_t *tx)
524 #if SOCKNAL_SINGLE_FRAG_TX
525 struct iovec scratch;
526 struct iovec *scratchiov = &scratch;
527 unsigned int niov = 1;
529 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
530 unsigned int niov = tx->tx_niov;
532 struct socket *sock = conn->ksnc_sock;
537 .uio_iov = scratchiov,
540 .uio_resid = 0, /* This will be valued after a while */
541 .uio_segflg = UIO_SYSSPACE,
545 int flags = MSG_DONTWAIT;
548 for (nob = i = 0; i < niov; i++) {
549 scratchiov[i] = tx->tx_iov[i];
550 nob += scratchiov[i].iov_len;
552 suio.uio_resid = nob;
555 rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
558 /* NB there is no return value can indicate how many
559 * have been sent and how many resid, we have to get
560 * sent bytes from suio. */
562 if (suio.uio_resid != nob &&\
563 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
564 /* We have sent something */
565 rc = nob - suio.uio_resid;
566 else if ( rc == EWOULDBLOCK )
567 /* Actually, EAGAIN and EWOULDBLOCK have same value in OSX */
572 rc = nob - suio.uio_resid;
578 ksocknal_lib_send_kiov (ksock_conn_t *conn, ksock_tx_t *tx)
580 #if SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_RISK_KMAP_DEADLOCK
581 struct iovec scratch;
582 struct iovec *scratchiov = &scratch;
583 unsigned int niov = 1;
585 struct iovec *scratchiov = conn->ksnc_tx_scratch_iov;
586 unsigned int niov = tx->tx_nkiov;
588 struct socket *sock = conn->ksnc_sock;
589 lnet_kiov_t *kiov = tx->tx_kiov;
594 .uio_iov = scratchiov,
597 .uio_resid = 0, /* It should be valued after a while */
598 .uio_segflg = UIO_SYSSPACE,
602 int flags = MSG_DONTWAIT;
605 for (nob = i = 0; i < niov; i++) {
606 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) +
608 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
610 suio.uio_resid = nob;
613 rc = sosend(sock, NULL, &suio, (struct mbuf *)0, (struct mbuf *)0, flags);
616 for (i = 0; i < niov; i++)
617 cfs_kunmap(kiov[i].kiov_page);
620 if (suio.uio_resid != nob &&\
621 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
622 /* We have sent something */
623 rc = nob - suio.uio_resid;
624 else if ( rc == EWOULDBLOCK )
625 /* EAGAIN and EWOULD BLOCK have same value in OSX */
630 rc = nob - suio.uio_resid;
636 * liang: Hack of inpcb and tcpcb.
637 * To get tcpcb of a socket, and call tcp_output
648 LIST_HEAD(ks_tsegqe_head, ks_tseg_qent);
651 struct ks_tsegqe_head t_segq;
653 struct ks_tcptemp *unused;
655 struct inpcb *t_inpcb;
659 * There are more fields but we dont need
664 #define TF_ACKNOW 0x00001
665 #define TF_DELACK 0x00002
668 LIST_ENTRY(ks_inpcb) inp_hash;
669 struct in_addr reserved1;
670 struct in_addr reserved2;
673 LIST_ENTRY(inpcb) inp_list;
676 * There are more fields but we dont need
681 #define ks_sotoinpcb(so) ((struct ks_inpcb *)(so)->so_pcb)
682 #define ks_intotcpcb(ip) ((struct ks_tcpcb *)(ip)->inp_ppcb)
683 #define ks_sototcpcb(so) (intotcpcb(sotoinpcb(so)))
686 ksocknal_lib_eager_ack (ksock_conn_t *conn)
688 struct socket *sock = conn->ksnc_sock;
689 struct ks_inpcb *inp = ks_sotoinpcb(sock);
690 struct ks_tcpcb *tp = ks_intotcpcb(inp);
694 extern int tcp_output(register struct ks_tcpcb *tp);
700 * No TCP_QUICKACK supported in BSD, so I have to call tcp_fasttimo
701 * to send immediate ACK.
703 if (tp && tp->t_flags & TF_DELACK){
704 tp->t_flags &= ~TF_DELACK;
705 tp->t_flags |= TF_ACKNOW;
706 (void) tcp_output(tp);
716 ksocknal_lib_recv_iov (ksock_conn_t *conn)
718 #if SOCKNAL_SINGLE_FRAG_RX
719 struct iovec scratch;
720 struct iovec *scratchiov = &scratch;
721 unsigned int niov = 1;
723 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
724 unsigned int niov = conn->ksnc_rx_niov;
726 struct iovec *iov = conn->ksnc_rx_iov;
731 .uio_iov = scratchiov,
734 .uio_resid = 0, /* It should be valued after a while */
735 .uio_segflg = UIO_SYSSPACE,
739 int flags = MSG_DONTWAIT;
742 for (nob = i = 0; i < niov; i++) {
743 scratchiov[i] = iov[i];
744 nob += scratchiov[i].iov_len;
746 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
748 ruio.uio_resid = nob;
751 rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, (struct mbuf **)0, &flags);
754 if (ruio.uio_resid != nob && \
755 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK || rc == EAGAIN))
756 /* data particially received */
757 rc = nob - ruio.uio_resid;
758 else if (rc == EWOULDBLOCK)
759 /* EAGAIN and EWOULD BLOCK have same value in OSX */
764 rc = nob - ruio.uio_resid;
770 ksocknal_lib_recv_kiov (ksock_conn_t *conn)
772 #if SOCKNAL_SINGLE_FRAG_RX || !SOCKNAL_RISK_KMAP_DEADLOCK
773 struct iovec scratch;
774 struct iovec *scratchiov = &scratch;
775 unsigned int niov = 1;
777 struct iovec *scratchiov = conn->ksnc_rx_scratch_iov;
778 unsigned int niov = conn->ksnc_rx_nkiov;
780 lnet_kiov_t *kiov = conn->ksnc_rx_kiov;
785 .uio_iov = scratchiov,
789 .uio_segflg = UIO_SYSSPACE,
793 int flags = MSG_DONTWAIT;
796 for (nob = i = 0; i < niov; i++) {
797 scratchiov[i].iov_base = cfs_kmap(kiov[i].kiov_page) + kiov[i].kiov_offset;
798 nob += scratchiov[i].iov_len = kiov[i].kiov_len;
800 LASSERT (nob <= conn->ksnc_rx_nob_wanted);
802 ruio.uio_resid = nob;
805 rc = soreceive(conn->ksnc_sock, (struct sockaddr **)0, &ruio, (struct mbuf **)0, NULL, &flags);
808 for (i = 0; i < niov; i++)
809 cfs_kunmap(kiov[i].kiov_page);
812 if (ruio.uio_resid != nob && \
813 (rc == ERESTART || rc == EINTR || rc == EWOULDBLOCK))
814 /* data particially received */
815 rc = nob - ruio.uio_resid;
816 else if (rc == EWOULDBLOCK)
817 /* receive blocked, EWOULDBLOCK == EAGAIN */
822 rc = nob - ruio.uio_resid;
828 ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem, int *rxmem, int *nagle)
830 struct socket *sock = conn->ksnc_sock;
833 rc = ksocknal_connsock_addref(conn);
835 LASSERT (conn->ksnc_closing);
836 *txmem = *rxmem = *nagle = 0;
839 rc = libcfs_sock_getbuf(sock, txmem, rxmem);
845 len = sizeof(*nagle);
846 bzero(&sopt, sizeof sopt);
847 sopt.sopt_dir = SOPT_GET;
848 sopt.sopt_level = IPPROTO_TCP;
849 sopt.sopt_name = TCP_NODELAY;
850 sopt.sopt_val = nagle;
851 sopt.sopt_valsize = len;
854 rc = -sogetopt(sock, &sopt);
858 ksocknal_connsock_decref(conn);
863 *txmem = *rxmem = *nagle = 0;
868 ksocknal_lib_setup_sock (struct socket *so)
877 struct linger linger;
880 rc = libcfs_sock_setbuf(so,
881 *ksocknal_tunables.ksnd_tx_buffer_size,
882 *ksocknal_tunables.ksnd_rx_buffer_size);
884 CERROR ("Can't set buffer tx %d, rx %d buffers: %d\n",
885 *ksocknal_tunables.ksnd_tx_buffer_size,
886 *ksocknal_tunables.ksnd_rx_buffer_size, rc);
890 /* Ensure this socket aborts active sends immediately when we close
892 bzero(&sopt, sizeof sopt);
896 sopt.sopt_dir = SOPT_SET;
897 sopt.sopt_level = SOL_SOCKET;
898 sopt.sopt_name = SO_LINGER;
899 sopt.sopt_val = &linger;
900 sopt.sopt_valsize = sizeof(linger);
903 rc = -sosetopt(so, &sopt);
905 CERROR ("Can't set SO_LINGER: %d\n", rc);
909 if (!*ksocknal_tunables.ksnd_nagle) {
911 bzero(&sopt, sizeof sopt);
912 sopt.sopt_dir = SOPT_SET;
913 sopt.sopt_level = IPPROTO_TCP;
914 sopt.sopt_name = TCP_NODELAY;
915 sopt.sopt_val = &option;
916 sopt.sopt_valsize = sizeof(option);
917 rc = -sosetopt(so, &sopt);
919 CERROR ("Can't disable nagle: %d\n", rc);
924 /* snapshot tunables */
925 keep_idle = *ksocknal_tunables.ksnd_keepalive_idle;
926 keep_count = *ksocknal_tunables.ksnd_keepalive_count;
927 keep_intvl = *ksocknal_tunables.ksnd_keepalive_intvl;
929 do_keepalive = (keep_idle > 0 && keep_count > 0 && keep_intvl > 0);
930 option = (do_keepalive ? 1 : 0);
931 bzero(&sopt, sizeof sopt);
932 sopt.sopt_dir = SOPT_SET;
933 sopt.sopt_level = SOL_SOCKET;
934 sopt.sopt_name = SO_KEEPALIVE;
935 sopt.sopt_val = &option;
936 sopt.sopt_valsize = sizeof(option);
937 rc = -sosetopt(so, &sopt);
939 CERROR ("Can't set SO_KEEPALIVE: %d\n", rc);
944 /* no more setting, just return */
949 bzero(&sopt, sizeof sopt);
950 sopt.sopt_dir = SOPT_SET;
951 sopt.sopt_level = IPPROTO_TCP;
952 sopt.sopt_name = TCP_KEEPALIVE;
953 sopt.sopt_val = &keep_idle;
954 sopt.sopt_valsize = sizeof(keep_idle);
955 rc = -sosetopt(so, &sopt);
957 CERROR ("Can't set TCP_KEEPALIVE : %d\n", rc);
966 ksocknal_lib_push_conn(ksock_conn_t *conn)
974 rc = ksocknal_connsock_addref(conn);
975 if (rc != 0) /* being shut down */
977 sock = conn->ksnc_sock;
978 bzero(&sopt, sizeof sopt);
979 sopt.sopt_dir = SOPT_SET;
980 sopt.sopt_level = IPPROTO_TCP;
981 sopt.sopt_name = TCP_NODELAY;
982 sopt.sopt_val = &val;
983 sopt.sopt_valsize = sizeof val;
986 sosetopt(sock, &sopt);
989 ksocknal_connsock_decref(conn);
994 extern void ksocknal_read_callback (ksock_conn_t *conn);
995 extern void ksocknal_write_callback (ksock_conn_t *conn);
998 ksocknal_upcall(struct socket *so, caddr_t arg, int waitf)
1000 ksock_conn_t *conn = (ksock_conn_t *)arg;
1003 read_lock (&ksocknal_data.ksnd_global_lock);
1007 if (so->so_rcv.sb_flags & SB_UPCALL) {
1008 extern int soreadable(struct socket *so);
1009 if (conn->ksnc_rx_nob_wanted && soreadable(so))
1010 /* To verify whether the upcall is for receive */
1011 ksocknal_read_callback (conn);
1014 if (so->so_snd.sb_flags & SB_UPCALL){
1015 extern int sowriteable(struct socket *so);
1016 if (sowriteable(so))
1017 /* socket is writable */
1018 ksocknal_write_callback(conn);
1021 read_unlock (&ksocknal_data.ksnd_global_lock);
1027 ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn)
1029 /* No callback need to save in osx */
1034 ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn)
1039 sock->so_upcallarg = (void *)conn;
1040 sock->so_upcall = ksocknal_upcall;
1041 sock->so_snd.sb_timeo = 0;
1042 sock->so_rcv.sb_timeo = cfs_time_seconds(2);
1043 sock->so_rcv.sb_flags |= SB_UPCALL;
1044 sock->so_snd.sb_flags |= SB_UPCALL;
1050 ksocknal_lib_act_callback(struct socket *sock, ksock_conn_t *conn)
1055 ksocknal_upcall (sock, (void *)conn, 0);
1060 ksocknal_lib_reset_callback(struct socket *sock, ksock_conn_t *conn)
1065 sock->so_rcv.sb_flags &= ~SB_UPCALL;
1066 sock->so_snd.sb_flags &= ~SB_UPCALL;
1067 sock->so_upcall = NULL;
1068 sock->so_upcallarg = NULL;
1072 #endif /* !__DARWIN8__ */