2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2004 The FreeBSD Foundation. All rights reserved.
7 * Copyright (c) 2004-2008 Robert N. M. Watson. All rights reserved.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * Excerpts taken from tcp_subr.c, tcp_usrreq.c, uipc_socket.c
38 * Copyright (c) 2010 Isilon Systems, Inc.
39 * Copyright (c) 2010 iX Systems, Inc.
40 * Copyright (c) 2010 Panasas, Inc.
41 * All rights reserved.
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
46 * 1. Redistributions of source code must retain the above copyright
47 * notice unmodified, this list of conditions, and the following
49 * 2. Redistributions in binary form must reproduce the above copyright
50 * notice, this list of conditions and the following disclaimer in the
51 * documentation and/or other materials provided with the distribution.
53 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
54 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
55 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
56 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
57 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
58 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
59 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
60 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
61 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
62 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD$");
68 #include <sys/param.h>
69 #include <sys/eventhandler.h>
70 #include <sys/kernel.h>
71 #include <sys/malloc.h>
76 #include <net/route.h>
78 #include <sys/sysctl.h>
81 struct rwlock sdp_lock;
82 LIST_HEAD(, sdp_sock) sdp_list;
84 struct workqueue_struct *rx_comp_wq;
86 RW_SYSINIT(sdplockinit, &sdp_lock, "SDP lock");
87 #define SDP_LIST_WLOCK() rw_wlock(&sdp_lock)
88 #define SDP_LIST_RLOCK() rw_rlock(&sdp_lock)
89 #define SDP_LIST_WUNLOCK() rw_wunlock(&sdp_lock)
90 #define SDP_LIST_RUNLOCK() rw_runlock(&sdp_lock)
91 #define SDP_LIST_WLOCK_ASSERT() rw_assert(&sdp_lock, RW_WLOCKED)
92 #define SDP_LIST_RLOCK_ASSERT() rw_assert(&sdp_lock, RW_RLOCKED)
93 #define SDP_LIST_LOCK_ASSERT() rw_assert(&sdp_lock, RW_LOCKED)
95 MALLOC_DEFINE(M_SDP, "sdp", "Sockets Direct Protocol");
97 static void sdp_stop_keepalive_timer(struct socket *so);
100 * SDP protocol interface to socket abstraction.
103 * sdp_sendspace and sdp_recvspace are the default send and receive window
104 * sizes, respectively.
106 u_long sdp_sendspace = 1024*32;
107 u_long sdp_recvspace = 1024*64;
109 static int sdp_count;
112 * Disable async. CMA events for sockets which are being torn down.
115 sdp_destroy_cma(struct sdp_sock *ssk)
120 rdma_destroy_id(ssk->id);
125 sdp_pcbbind(struct sdp_sock *ssk, struct sockaddr *nam, struct ucred *cred)
127 struct sockaddr_in *sin;
128 struct sockaddr_in null;
131 SDP_WLOCK_ASSERT(ssk);
133 if (ssk->lport != 0 || ssk->laddr != INADDR_ANY)
135 /* rdma_bind_addr handles bind races. */
138 ssk->id = rdma_create_id(&init_net, sdp_cma_handler, ssk, RDMA_PS_SDP, IB_QPT_RC);
139 if (ssk->id == NULL) {
144 null.sin_family = AF_INET;
145 null.sin_len = sizeof(null);
146 null.sin_addr.s_addr = INADDR_ANY;
148 bzero(&null.sin_zero, sizeof(null.sin_zero));
149 nam = (struct sockaddr *)&null;
151 error = -rdma_bind_addr(ssk->id, nam);
154 sin = (struct sockaddr_in *)&ssk->id->route.addr.src_addr;
155 ssk->laddr = sin->sin_addr.s_addr;
156 ssk->lport = sin->sin_port;
158 sdp_destroy_cma(ssk);
163 sdp_pcbfree(struct sdp_sock *ssk)
166 KASSERT(ssk->socket == NULL, ("ssk %p socket still attached", ssk));
167 KASSERT((ssk->flags & SDP_DESTROY) == 0,
168 ("ssk %p already destroyed", ssk));
170 sdp_dbg(ssk->socket, "Freeing pcb");
171 SDP_WLOCK_ASSERT(ssk);
172 ssk->flags |= SDP_DESTROY;
176 LIST_REMOVE(ssk, list);
181 ib_destroy_qp(ssk->qp);
184 sdp_tx_ring_destroy(ssk);
185 sdp_rx_ring_destroy(ssk);
186 sdp_destroy_cma(ssk);
187 rw_destroy(&ssk->rx_ring.destroyed_lock);
188 rw_destroy(&ssk->lock);
189 uma_zfree(sdp_zone, ssk);
193 * Common routines to return a socket address.
195 static struct sockaddr *
196 sdp_sockaddr(in_port_t port, struct in_addr *addr_p)
198 struct sockaddr_in *sin;
200 sin = malloc(sizeof *sin, M_SONAME,
202 sin->sin_family = AF_INET;
203 sin->sin_len = sizeof(*sin);
204 sin->sin_addr = *addr_p;
205 sin->sin_port = port;
207 return (struct sockaddr *)sin;
211 sdp_getsockaddr(struct socket *so, struct sockaddr **nam)
213 struct sdp_sock *ssk;
220 addr.s_addr = ssk->laddr;
223 *nam = sdp_sockaddr(port, &addr);
228 sdp_getpeeraddr(struct socket *so, struct sockaddr **nam)
230 struct sdp_sock *ssk;
237 addr.s_addr = ssk->faddr;
240 *nam = sdp_sockaddr(port, &addr);
245 sdp_pcbnotifyall(struct in_addr faddr, int errno,
246 struct sdp_sock *(*notify)(struct sdp_sock *, int))
248 struct sdp_sock *ssk, *ssk_temp;
251 LIST_FOREACH_SAFE(ssk, &sdp_list, list, ssk_temp) {
253 if (ssk->faddr != faddr.s_addr || ssk->socket == NULL) {
257 if ((ssk->flags & SDP_DESTROY) == 0)
258 if ((*notify)(ssk, errno))
266 sdp_apply_all(void (*func)(struct sdp_sock *, void *), void *arg)
268 struct sdp_sock *ssk;
271 LIST_FOREACH(ssk, &sdp_list, list) {
281 sdp_output_reset(struct sdp_sock *ssk)
283 struct rdma_cm_id *id;
285 SDP_WLOCK_ASSERT(ssk);
293 ssk->state = TCPS_CLOSED;
297 * Attempt to close a SDP socket, marking it as dropped, and freeing
298 * the socket if we hold the only reference.
300 static struct sdp_sock *
301 sdp_closed(struct sdp_sock *ssk)
305 SDP_WLOCK_ASSERT(ssk);
307 ssk->flags |= SDP_DROPPED;
309 soisdisconnected(so);
310 if (ssk->flags & SDP_SOCKREF) {
311 KASSERT(so->so_state & SS_PROTOREF,
312 ("sdp_closed: !SS_PROTOREF"));
313 ssk->flags &= ~SDP_SOCKREF;
316 so->so_state &= ~SS_PROTOREF;
324 * Perform timer based shutdowns which can not operate in
328 sdp_shutdown_task(void *data, int pending)
330 struct sdp_sock *ssk;
335 * I don't think this can race with another call to pcbfree()
336 * because SDP_TIMEWAIT protects it. SDP_DESTROY may be redundant.
338 if (ssk->flags & SDP_DESTROY)
339 panic("sdp_shutdown_task: Racing with pcbfree for ssk %p",
341 if (ssk->flags & SDP_DISCON)
342 sdp_output_reset(ssk);
343 /* We have to clear this so sdp_detach() will call pcbfree(). */
344 ssk->flags &= ~(SDP_TIMEWAIT | SDP_DREQWAIT);
345 if ((ssk->flags & SDP_DROPPED) == 0 &&
346 sdp_closed(ssk) == NULL)
348 if (ssk->socket == NULL) {
356 * 2msl has expired, schedule the shutdown task.
359 sdp_2msl_timeout(void *data)
361 struct sdp_sock *ssk;
364 /* Callout canceled. */
365 if (!callout_active(&ssk->keep2msl))
367 callout_deactivate(&ssk->keep2msl);
368 /* Should be impossible, defensive programming. */
369 if ((ssk->flags & SDP_TIMEWAIT) == 0)
371 taskqueue_enqueue(taskqueue_thread, &ssk->shutdown_task);
378 * Schedule the 2msl wait timer.
381 sdp_2msl_wait(struct sdp_sock *ssk)
384 SDP_WLOCK_ASSERT(ssk);
385 ssk->flags |= SDP_TIMEWAIT;
386 ssk->state = TCPS_TIME_WAIT;
387 soisdisconnected(ssk->socket);
388 callout_reset(&ssk->keep2msl, TCPTV_MSL, sdp_2msl_timeout, ssk);
392 * Timed out waiting for the final fin/ack from rdma_disconnect().
395 sdp_dreq_timeout(void *data)
397 struct sdp_sock *ssk;
400 /* Callout canceled. */
401 if (!callout_active(&ssk->keep2msl))
403 /* Callout rescheduled, probably as a different timer. */
404 if (callout_pending(&ssk->keep2msl))
406 callout_deactivate(&ssk->keep2msl);
407 if (ssk->state != TCPS_FIN_WAIT_1 && ssk->state != TCPS_LAST_ACK)
409 if ((ssk->flags & SDP_DREQWAIT) == 0)
411 ssk->flags &= ~SDP_DREQWAIT;
412 ssk->flags |= SDP_DISCON;
420 * Received the final fin/ack. Cancel the 2msl.
423 sdp_cancel_dreq_wait_timeout(struct sdp_sock *ssk)
425 sdp_dbg(ssk->socket, "cancelling dreq wait timeout\n");
426 ssk->flags &= ~SDP_DREQWAIT;
431 sdp_init_sock(struct socket *sk)
433 struct sdp_sock *ssk = sdp_sk(sk);
435 sdp_dbg(sk, "%s\n", __func__);
437 callout_init_rw(&ssk->keep2msl, &ssk->lock, CALLOUT_RETURNUNLOCKED);
438 TASK_INIT(&ssk->shutdown_task, 0, sdp_shutdown_task, ssk);
440 INIT_DELAYED_WORK(&ssk->srcavail_cancel_work, srcavail_cancel_timeout);
441 ssk->zcopy_thresh = -1; /* use global sdp_zcopy_thresh */
442 ssk->tx_ring.rdma_inflight = NULL;
444 atomic_set(&ssk->mseq_ack, 0);
445 sdp_rx_ring_init(ssk);
446 ssk->tx_ring.buffer = NULL;
452 * Allocate an sdp_sock for the socket and reserve socket buffer space.
455 sdp_attach(struct socket *so, int proto, struct thread *td)
457 struct sdp_sock *ssk;
461 KASSERT(ssk == NULL, ("sdp_attach: ssk already set on so %p", so));
462 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
463 error = soreserve(so, sdp_sendspace, sdp_recvspace);
467 so->so_rcv.sb_flags |= SB_AUTOSIZE;
468 so->so_snd.sb_flags |= SB_AUTOSIZE;
469 ssk = uma_zalloc(sdp_zone, M_NOWAIT | M_ZERO);
472 rw_init(&ssk->lock, "sdpsock");
474 ssk->cred = crhold(so->so_cred);
475 so->so_pcb = (caddr_t)ssk;
479 ssk->state = TCPS_CLOSED;
480 mbufq_init(&ssk->rxctlq, INT_MAX);
482 LIST_INSERT_HEAD(&sdp_list, ssk, list);
485 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
486 so->so_linger = TCP_LINGERTIME;
492 * Detach SDP from the socket, potentially leaving it around for the
493 * timewait to expire.
496 sdp_detach(struct socket *so)
498 struct sdp_sock *ssk;
502 KASSERT(ssk->socket != NULL, ("sdp_detach: socket is NULL"));
503 ssk->socket->so_pcb = NULL;
505 if (ssk->flags & (SDP_TIMEWAIT | SDP_DREQWAIT))
507 else if (ssk->flags & SDP_DROPPED || ssk->state < TCPS_SYN_SENT)
510 panic("sdp_detach: Unexpected state, ssk %p.\n", ssk);
514 * Allocate a local address for the socket.
517 sdp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
520 struct sdp_sock *ssk;
521 struct sockaddr_in *sin;
523 sin = (struct sockaddr_in *)nam;
524 if (nam->sa_len != sizeof (*sin))
526 if (sin->sin_family != AF_INET)
528 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
529 return (EAFNOSUPPORT);
533 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
537 error = sdp_pcbbind(ssk, nam, td->td_ucred);
545 * Prepare to accept connections.
548 sdp_listen(struct socket *so, int backlog, struct thread *td)
551 struct sdp_sock *ssk;
555 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
559 if (error == 0 && ssk->lport == 0)
560 error = sdp_pcbbind(ssk, (struct sockaddr *)0, td->td_ucred);
563 error = solisten_proto_check(so);
565 solisten_proto(so, backlog);
566 ssk->state = TCPS_LISTEN;
573 error = -rdma_listen(ssk->id, backlog);
578 * Initiate a SDP connection to nam.
581 sdp_start_connect(struct sdp_sock *ssk, struct sockaddr *nam, struct thread *td)
583 struct sockaddr_in src;
589 SDP_WLOCK_ASSERT(ssk);
590 if (ssk->lport == 0) {
591 error = sdp_pcbbind(ssk, (struct sockaddr *)0, td->td_ucred);
595 src.sin_family = AF_INET;
596 src.sin_len = sizeof(src);
597 bzero(&src.sin_zero, sizeof(src.sin_zero));
598 src.sin_port = ssk->lport;
599 src.sin_addr.s_addr = ssk->laddr;
602 error = -rdma_resolve_addr(ssk->id, (struct sockaddr *)&src, nam,
603 SDP_RESOLVE_TIMEOUT);
606 ssk->state = TCPS_SYN_SENT;
612 * Initiate SDP connection.
615 sdp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
618 struct sdp_sock *ssk;
619 struct sockaddr_in *sin;
621 sin = (struct sockaddr_in *)nam;
622 if (nam->sa_len != sizeof (*sin))
624 if (sin->sin_family != AF_INET)
626 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
627 return (EAFNOSUPPORT);
628 if ((error = prison_remote_ip4(td->td_ucred, &sin->sin_addr)) != 0)
632 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED))
635 error = sdp_start_connect(ssk, nam, td);
641 * Drop a SDP socket, reporting
642 * the specified error. If connection is synchronized,
643 * then send a RST to peer.
645 static struct sdp_sock *
646 sdp_drop(struct sdp_sock *ssk, int errno)
650 SDP_WLOCK_ASSERT(ssk);
652 if (TCPS_HAVERCVDSYN(ssk->state))
653 sdp_output_reset(ssk);
654 if (errno == ETIMEDOUT && ssk->softerror)
655 errno = ssk->softerror;
656 so->so_error = errno;
657 return (sdp_closed(ssk));
661 * User issued close, and wish to trail through shutdown states:
662 * if never received SYN, just forget it. If got a SYN from peer,
663 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
664 * If already got a FIN from peer, then almost done; go to LAST_ACK
665 * state. In all other cases, have already sent FIN to peer (e.g.
666 * after PRU_SHUTDOWN), and just have to play tedious game waiting
667 * for peer to send FIN or not respond to keep-alives, etc.
668 * We can let the user exit from the close as soon as the FIN is acked.
671 sdp_usrclosed(struct sdp_sock *ssk)
674 SDP_WLOCK_ASSERT(ssk);
676 switch (ssk->state) {
678 ssk->state = TCPS_CLOSED;
680 sdp_destroy_cma(ssk);
684 ssk = sdp_closed(ssk);
686 * sdp_closed() should never return NULL here as the socket is
690 ("sdp_usrclosed: sdp_closed() returned NULL"));
695 case TCPS_SYN_RECEIVED:
696 ssk->flags |= SDP_NEEDFIN;
699 case TCPS_ESTABLISHED:
700 ssk->flags |= SDP_NEEDFIN;
701 ssk->state = TCPS_FIN_WAIT_1;
704 case TCPS_CLOSE_WAIT:
705 ssk->state = TCPS_LAST_ACK;
708 if (ssk->state >= TCPS_FIN_WAIT_2) {
709 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
710 if (ssk->state == TCPS_FIN_WAIT_2)
713 soisdisconnected(ssk->socket);
718 sdp_output_disconnect(struct sdp_sock *ssk)
721 SDP_WLOCK_ASSERT(ssk);
722 callout_reset(&ssk->keep2msl, SDP_FIN_WAIT_TIMEOUT,
723 sdp_dreq_timeout, ssk);
724 ssk->flags |= SDP_NEEDFIN | SDP_DREQWAIT;
725 sdp_post_sends(ssk, M_NOWAIT);
729 * Initiate or continue a disconnect.
730 * If embryonic state, just send reset (once).
731 * If in ``let data drain'' option and linger null, just drop.
732 * Otherwise (hard), mark socket disconnecting and drop
733 * current input data; switch states based on user close, and
734 * send segment to peer (with FIN).
737 sdp_start_disconnect(struct sdp_sock *ssk)
743 SDP_WLOCK_ASSERT(ssk);
744 sdp_stop_keepalive_timer(so);
746 * Neither sdp_closed() nor sdp_drop() should return NULL, as the
747 * socket is still open.
749 if (ssk->state < TCPS_ESTABLISHED) {
750 ssk = sdp_closed(ssk);
752 ("sdp_start_disconnect: sdp_close() returned NULL"));
753 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
754 ssk = sdp_drop(ssk, 0);
756 ("sdp_start_disconnect: sdp_drop() returned NULL"));
758 soisdisconnecting(so);
759 unread = sbused(&so->so_rcv);
760 sbflush(&so->so_rcv);
762 if (!(ssk->flags & SDP_DROPPED)) {
764 sdp_output_reset(ssk);
766 sdp_output_disconnect(ssk);
772 * User initiated disconnect.
775 sdp_disconnect(struct socket *so)
777 struct sdp_sock *ssk;
782 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
786 sdp_start_disconnect(ssk);
793 * Accept a connection. Essentially all the work is done at higher levels;
794 * just return the address of the peer, storing through addr.
797 * XXX This is broken XXX
799 * The rationale for acquiring the sdp lock here is somewhat complicated,
800 * and is described in detail in the commit log entry for r175612. Acquiring
801 * it delays an accept(2) racing with sonewconn(), which inserts the socket
802 * before the address/port fields are initialized. A better fix would
803 * prevent the socket from being placed in the listen queue until all fields
804 * are fully initialized.
807 sdp_accept(struct socket *so, struct sockaddr **nam)
809 struct sdp_sock *ssk = NULL;
814 if (so->so_state & SS_ISDISCONNECTED)
815 return (ECONNABORTED);
822 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
823 error = ECONNABORTED;
827 addr.s_addr = ssk->faddr;
831 *nam = sdp_sockaddr(port, &addr);
836 * Mark the connection as being incapable of further output.
839 sdp_shutdown(struct socket *so)
842 struct sdp_sock *ssk;
846 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
852 if (!(ssk->flags & SDP_DROPPED))
853 sdp_output_disconnect(ssk);
862 sdp_append(struct sdp_sock *ssk, struct sockbuf *sb, struct mbuf *mb, int cnt)
867 SOCKBUF_LOCK_ASSERT(sb);
869 KASSERT(mb->m_flags & M_PKTHDR,
870 ("sdp_append: %p Missing packet header.\n", mb));
871 n = sb->sb_lastrecord;
873 * If the queue is empty just set all pointers and proceed.
876 sb->sb_lastrecord = sb->sb_mb = sb->sb_sndptr = mb;
877 for (; mb; mb = mb->m_next) {
884 * Count the number of mbufs in the current tail.
886 for (ncnt = 0; n->m_next; n = n->m_next)
888 n = sb->sb_lastrecord;
890 * If the two chains can fit in a single sdp packet and
891 * the last record has not been sent yet (WRITABLE) coalesce
892 * them. The lastrecord remains the same but we must strip the
893 * packet header and then let sbcompress do the hard part.
895 if (M_WRITABLE(n) && ncnt + cnt < SDP_MAX_SEND_SGES &&
896 n->m_pkthdr.len + mb->m_pkthdr.len - SDP_HEAD_SIZE <
897 ssk->xmit_size_goal) {
898 m_adj(mb, SDP_HEAD_SIZE);
899 n->m_pkthdr.len += mb->m_pkthdr.len;
900 n->m_flags |= mb->m_flags & (M_PUSH | M_URG);
902 sbcompress(sb, mb, sb->sb_mbtail);
906 * Not compressible, just append to the end and adjust counters.
908 sb->sb_lastrecord->m_flags |= M_PUSH;
909 sb->sb_lastrecord->m_nextpkt = mb;
910 sb->sb_lastrecord = mb;
911 if (sb->sb_sndptr == NULL)
913 for (; mb; mb = mb->m_next) {
920 * Do a send by putting data in output queue and updating urgent
921 * marker if URG set. Possibly send more data. Unlike the other
922 * pru_*() routines, the mbuf chains are our responsibility. We
923 * must either enqueue them or free them. The other pru_* routines
924 * generally are caller-frees.
926 * This comes from sendfile, normal sends will come from sdp_sosend().
929 sdp_send(struct socket *so, int flags, struct mbuf *m,
930 struct sockaddr *nam, struct mbuf *control, struct thread *td)
932 struct sdp_sock *ssk;
939 KASSERT(m->m_flags & M_PKTHDR,
940 ("sdp_send: %p no packet header", m));
941 M_PREPEND(m, SDP_HEAD_SIZE, M_WAITOK);
942 mtod(m, struct sdp_bsdh *)->mid = SDP_MID_DATA;
943 for (n = m, cnt = 0; n->m_next; n = n->m_next)
945 if (cnt > SDP_MAX_SEND_SGES) {
946 n = m_collapse(m, M_WAITOK, SDP_MAX_SEND_SGES);
952 for (cnt = 0; n->m_next; n = n->m_next)
956 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
965 /* SDP doesn't support control messages. */
966 if (control->m_len) {
973 m_freem(control); /* empty control, just free it */
975 if (!(flags & PRUS_OOB)) {
976 SOCKBUF_LOCK(&so->so_snd);
977 sdp_append(ssk, &so->so_snd, m, cnt);
978 SOCKBUF_UNLOCK(&so->so_snd);
979 if (nam && ssk->state < TCPS_SYN_SENT) {
981 * Do implied connect if not yet connected.
983 error = sdp_start_connect(ssk, nam, td);
987 if (flags & PRUS_EOF) {
989 * Close the send side of the connection after
994 if (!(ssk->flags & SDP_DROPPED))
995 sdp_output_disconnect(ssk);
996 } else if (!(ssk->flags & SDP_DROPPED) &&
997 !(flags & PRUS_MORETOCOME))
998 sdp_post_sends(ssk, M_NOWAIT);
1002 SOCKBUF_LOCK(&so->so_snd);
1003 if (sbspace(&so->so_snd) < -512) {
1004 SOCKBUF_UNLOCK(&so->so_snd);
1010 * According to RFC961 (Assigned Protocols),
1011 * the urgent pointer points to the last octet
1012 * of urgent data. We continue, however,
1013 * to consider it to indicate the first octet
1014 * of data past the urgent section.
1015 * Otherwise, snd_up should be one lower.
1017 m->m_flags |= M_URG | M_PUSH;
1018 sdp_append(ssk, &so->so_snd, m, cnt);
1019 SOCKBUF_UNLOCK(&so->so_snd);
1020 if (nam && ssk->state < TCPS_SYN_SENT) {
1022 * Do implied connect if not yet connected.
1024 error = sdp_start_connect(ssk, nam, td);
1028 sdp_post_sends(ssk, M_NOWAIT);
1037 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1040 * Send on a socket. If send must go all at once and message is larger than
1041 * send buffering, then hard error. Lock against other senders. If must go
1042 * all at once and not enough room now, then inform user that this would
1043 * block and do nothing. Otherwise, if nonblocking, send as much as
1044 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1045 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1046 * in mbuf chain must be small enough to send all at once.
1048 * Returns nonzero on error, timeout or signal; callers must check for short
1049 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1053 sdp_sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1056 struct sdp_sock *ssk;
1063 resid = uio->uio_resid;
1065 resid = top->m_pkthdr.len;
1066 atomic = top != NULL;
1067 if (control != NULL) {
1068 if (control->m_len) {
1078 * In theory resid should be unsigned. However, space must be
1079 * signed, as it might be less than 0 if we over-committed, and we
1080 * must use a signed comparison of space and resid. On the other
1081 * hand, a negative resid causes us to loop sending 0-length
1082 * segments to the protocol.
1084 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1085 * type sockets since that's an error.
1087 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1092 td->td_ru.ru_msgsnd++;
1095 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1101 SOCKBUF_LOCK(&so->so_snd);
1102 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1103 SOCKBUF_UNLOCK(&so->so_snd);
1108 error = so->so_error;
1110 SOCKBUF_UNLOCK(&so->so_snd);
1113 if ((so->so_state & SS_ISCONNECTED) == 0 && addr == NULL) {
1114 SOCKBUF_UNLOCK(&so->so_snd);
1118 space = sbspace(&so->so_snd);
1119 if (flags & MSG_OOB)
1121 if (atomic && resid > ssk->xmit_size_goal - SDP_HEAD_SIZE) {
1122 SOCKBUF_UNLOCK(&so->so_snd);
1126 if (space < resid &&
1127 (atomic || space < so->so_snd.sb_lowat)) {
1128 if ((so->so_state & SS_NBIO) ||
1129 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1130 SOCKBUF_UNLOCK(&so->so_snd);
1131 error = EWOULDBLOCK;
1134 error = sbwait(&so->so_snd);
1135 SOCKBUF_UNLOCK(&so->so_snd);
1140 SOCKBUF_UNLOCK(&so->so_snd);
1144 if (flags & MSG_EOR)
1145 top->m_flags |= M_EOR;
1148 * Copy the data from userland into a mbuf
1149 * chain. If no data is to be copied in,
1150 * a single empty mbuf is returned.
1153 ssk->xmit_size_goal - SDP_HEAD_SIZE);
1154 top = m_uiotombuf(uio, M_WAITOK, copy,
1156 ((flags & MSG_EOR) ? M_EOR : 0));
1158 /* only possible error */
1162 space -= resid - uio->uio_resid;
1163 resid = uio->uio_resid;
1166 * XXX all the SBS_CANTSENDMORE checks previously
1167 * done could be out of date after dropping the
1170 error = sdp_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1172 * Set EOF on the last send if the user specified
1175 ((flags & MSG_EOF) && (resid <= 0)) ? PRUS_EOF :
1176 /* If there is more to send set PRUS_MORETOCOME. */
1177 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1178 top, addr, NULL, td);
1182 } while (resid && space > 0);
1186 sbunlock(&so->so_snd);
1194 * The part of soreceive() that implements reading non-inline out-of-band
1195 * data from a socket. For more complete comments, see soreceive(), from
1196 * which this code originated.
1198 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1199 * unable to return an mbuf chain to the caller.
1202 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1204 struct protosw *pr = so->so_proto;
1208 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1210 m = m_get(M_WAITOK, MT_DATA);
1211 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1215 error = uiomove(mtod(m, void *),
1216 (int) min(uio->uio_resid, m->m_len), uio);
1218 } while (uio->uio_resid && error == 0 && m);
1226 * Optimized version of soreceive() for stream (TCP) sockets.
1229 sdp_sorecv(struct socket *so, struct sockaddr **psa, struct uio *uio,
1230 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1232 int len = 0, error = 0, flags, oresid;
1234 struct mbuf *m, *n = NULL;
1235 struct sdp_sock *ssk;
1237 /* We only do stream sockets. */
1238 if (so->so_type != SOCK_STREAM)
1242 if (controlp != NULL)
1245 flags = *flagsp &~ MSG_EOR;
1248 if (flags & MSG_OOB)
1249 return (soreceive_rcvoob(so, uio, flags));
1256 /* Prevent other readers from entering the socket. */
1257 error = sblock(sb, SBLOCKWAIT(flags));
1262 /* Easy one, no space to copyout anything. */
1263 if (uio->uio_resid == 0) {
1267 oresid = uio->uio_resid;
1269 /* We will never ever get anything unless we are connected. */
1270 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1271 /* When disconnecting there may be still some data left. */
1274 if (!(so->so_state & SS_ISDISCONNECTED))
1279 /* Socket buffer is empty and we shall not block. */
1280 if (sbavail(sb) == 0 &&
1281 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1287 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1289 /* Abort if socket has reported problems. */
1293 if (oresid > uio->uio_resid)
1295 error = so->so_error;
1296 if (!(flags & MSG_PEEK))
1301 /* Door is closed. Deliver what is left, if any. */
1302 if (sb->sb_state & SBS_CANTRCVMORE) {
1309 /* Socket buffer got some data that we shall deliver now. */
1310 if (sbavail(sb) && !(flags & MSG_WAITALL) &&
1311 ((so->so_state & SS_NBIO) ||
1312 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1313 sbavail(sb) >= sb->sb_lowat ||
1314 sbavail(sb) >= uio->uio_resid ||
1315 sbavail(sb) >= sb->sb_hiwat) ) {
1319 /* On MSG_WAITALL we must wait until all data or error arrives. */
1320 if ((flags & MSG_WAITALL) &&
1321 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_lowat))
1325 * Wait and block until (more) data comes in.
1326 * NB: Drops the sockbuf lock during wait.
1334 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1335 KASSERT(sbavail(sb), ("%s: sockbuf empty", __func__));
1336 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1340 uio->uio_td->td_ru.ru_msgrcv++;
1342 /* Fill uio until full or current end of socket buffer is reached. */
1343 len = min(uio->uio_resid, sbavail(sb));
1345 /* Dequeue as many mbufs as possible. */
1346 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1347 for (*mp0 = m = sb->sb_mb;
1348 m != NULL && m->m_len <= len;
1351 uio->uio_resid -= m->m_len;
1356 if (sb->sb_mb == NULL)
1360 /* Copy the remainder. */
1362 KASSERT(sb->sb_mb != NULL,
1363 ("%s: len > 0 && sb->sb_mb empty", __func__));
1365 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
1367 len = 0; /* Don't flush data from sockbuf. */
1369 uio->uio_resid -= m->m_len;
1380 /* NB: Must unlock socket buffer as uiomove may sleep. */
1382 error = m_mbuftouio(uio, sb->sb_mb, len);
1387 SBLASTRECORDCHK(sb);
1391 * Remove the delivered data from the socket buffer unless we
1392 * were only peeking.
1394 if (!(flags & MSG_PEEK)) {
1396 sbdrop_locked(sb, len);
1398 /* Notify protocol that we drained some data. */
1407 * For MSG_WAITALL we may have to loop again and wait for
1408 * more data to come in.
1410 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
1413 SOCKBUF_LOCK_ASSERT(sb);
1414 SBLASTRECORDCHK(sb);
1422 * Abort is used to teardown a connection typically while sitting in
1426 sdp_abort(struct socket *so)
1428 struct sdp_sock *ssk;
1433 * If we have not yet dropped, do it now.
1435 if (!(ssk->flags & SDP_TIMEWAIT) &&
1436 !(ssk->flags & SDP_DROPPED))
1437 sdp_drop(ssk, ECONNABORTED);
1438 KASSERT(ssk->flags & SDP_DROPPED, ("sdp_abort: %p not dropped 0x%X",
1444 * Close a SDP socket and initiate a friendly disconnect.
1447 sdp_close(struct socket *so)
1449 struct sdp_sock *ssk;
1454 * If we have not yet dropped, do it now.
1456 if (!(ssk->flags & SDP_TIMEWAIT) &&
1457 !(ssk->flags & SDP_DROPPED))
1458 sdp_start_disconnect(ssk);
1461 * If we've still not dropped let the socket layer know we're
1462 * holding on to the socket and pcb for a while.
1464 if (!(ssk->flags & SDP_DROPPED)) {
1466 so->so_state |= SS_PROTOREF;
1468 ssk->flags |= SDP_SOCKREF;
1474 * User requests out-of-band data.
1477 sdp_rcvoob(struct socket *so, struct mbuf *m, int flags)
1480 struct sdp_sock *ssk;
1484 if (!rx_ring_trylock(&ssk->rx_ring)) {
1486 return (ECONNRESET);
1488 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
1492 if ((so->so_oobmark == 0 &&
1493 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1494 so->so_options & SO_OOBINLINE ||
1495 ssk->oobflags & SDP_HADOOB) {
1499 if ((ssk->oobflags & SDP_HAVEOOB) == 0) {
1500 error = EWOULDBLOCK;
1504 *mtod(m, caddr_t) = ssk->iobc;
1505 if ((flags & MSG_PEEK) == 0)
1506 ssk->oobflags ^= (SDP_HAVEOOB | SDP_HADOOB);
1508 rx_ring_unlock(&ssk->rx_ring);
1514 sdp_urg(struct sdp_sock *ssk, struct mbuf *mb)
1523 so->so_oobmark = sbused(&so->so_rcv) + mb->m_pkthdr.len - 1;
1525 ssk->oobflags &= ~(SDP_HAVEOOB | SDP_HADOOB);
1526 if (!(so->so_options & SO_OOBINLINE)) {
1527 for (m = mb; m->m_next != NULL; m = m->m_next);
1528 ssk->iobc = *(mtod(m, char *) + m->m_len - 1);
1529 ssk->oobflags |= SDP_HAVEOOB;
1536 * Notify a sdp socket of an asynchronous error.
1538 * Do not wake up user since there currently is no mechanism for
1539 * reporting soft errors (yet - a kqueue filter may be added).
1542 sdp_notify(struct sdp_sock *ssk, int error)
1545 SDP_WLOCK_ASSERT(ssk);
1547 if ((ssk->flags & SDP_TIMEWAIT) ||
1548 (ssk->flags & SDP_DROPPED))
1552 * Ignore some errors if we are hooked up.
1554 if (ssk->state == TCPS_ESTABLISHED &&
1555 (error == EHOSTUNREACH || error == ENETUNREACH ||
1556 error == EHOSTDOWN))
1558 ssk->softerror = error;
1559 return sdp_drop(ssk, error);
1563 sdp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1565 struct in_addr faddr;
1567 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1568 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1571 sdp_pcbnotifyall(faddr, inetctlerrmap[cmd], sdp_notify);
1575 sdp_control(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp,
1578 return (EOPNOTSUPP);
1582 sdp_keepalive_timeout(void *data)
1584 struct sdp_sock *ssk;
1587 /* Callout canceled. */
1588 if (!callout_active(&ssk->keep2msl))
1590 /* Callout rescheduled as a different kind of timer. */
1591 if (callout_pending(&ssk->keep2msl))
1593 callout_deactivate(&ssk->keep2msl);
1594 if (ssk->flags & SDP_DROPPED ||
1595 (ssk->socket->so_options & SO_KEEPALIVE) == 0)
1597 sdp_post_keepalive(ssk);
1598 callout_reset(&ssk->keep2msl, SDP_KEEPALIVE_TIME,
1599 sdp_keepalive_timeout, ssk);
1606 sdp_start_keepalive_timer(struct socket *so)
1608 struct sdp_sock *ssk;
1611 if (!callout_pending(&ssk->keep2msl))
1612 callout_reset(&ssk->keep2msl, SDP_KEEPALIVE_TIME,
1613 sdp_keepalive_timeout, ssk);
1617 sdp_stop_keepalive_timer(struct socket *so)
1619 struct sdp_sock *ssk;
1622 callout_stop(&ssk->keep2msl);
1626 * sdp_ctloutput() must drop the inpcb lock before performing copyin on
1627 * socket option arguments. When it re-acquires the lock after the copy, it
1628 * has to revalidate that the connection is still valid for the socket
1631 #define SDP_WLOCK_RECHECK(inp) do { \
1633 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) { \
1635 return (ECONNRESET); \
1640 sdp_ctloutput(struct socket *so, struct sockopt *sopt)
1642 int error, opt, optval;
1643 struct sdp_sock *ssk;
1647 if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_KEEPALIVE) {
1649 if (so->so_options & SO_KEEPALIVE)
1650 sdp_start_keepalive_timer(so);
1652 sdp_stop_keepalive_timer(so);
1655 if (sopt->sopt_level != IPPROTO_TCP)
1659 if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
1661 return (ECONNRESET);
1664 switch (sopt->sopt_dir) {
1666 switch (sopt->sopt_name) {
1669 error = sooptcopyin(sopt, &optval, sizeof optval,
1674 SDP_WLOCK_RECHECK(ssk);
1686 error = ENOPROTOOPT;
1692 switch (sopt->sopt_name) {
1694 optval = ssk->flags & SDP_NODELAY;
1696 error = sooptcopyout(sopt, &optval, sizeof optval);
1700 error = ENOPROTOOPT;
1707 #undef SDP_WLOCK_RECHECK
1709 int sdp_mod_count = 0;
1710 int sdp_mod_usec = 0;
1713 sdp_set_default_moderation(struct sdp_sock *ssk)
1715 if (sdp_mod_count <= 0 || sdp_mod_usec <= 0)
1717 ib_modify_cq(ssk->rx_ring.cq, sdp_mod_count, sdp_mod_usec);
1721 sdp_dev_add(struct ib_device *device)
1723 struct ib_fmr_pool_param param;
1724 struct sdp_device *sdp_dev;
1726 sdp_dev = malloc(sizeof(*sdp_dev), M_SDP, M_WAITOK | M_ZERO);
1727 sdp_dev->pd = ib_alloc_pd(device, 0);
1728 if (IS_ERR(sdp_dev->pd))
1730 memset(¶m, 0, sizeof param);
1731 param.max_pages_per_fmr = SDP_FMR_SIZE;
1732 param.page_shift = PAGE_SHIFT;
1733 param.access = (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_READ);
1734 param.pool_size = SDP_FMR_POOL_SIZE;
1735 param.dirty_watermark = SDP_FMR_DIRTY_SIZE;
1737 sdp_dev->fmr_pool = ib_create_fmr_pool(sdp_dev->pd, ¶m);
1738 if (IS_ERR(sdp_dev->fmr_pool))
1740 ib_set_client_data(device, &sdp_client, sdp_dev);
1744 ib_dealloc_pd(sdp_dev->pd);
1746 free(sdp_dev, M_SDP);
1750 sdp_dev_rem(struct ib_device *device, void *client_data)
1752 struct sdp_device *sdp_dev;
1753 struct sdp_sock *ssk;
1756 LIST_FOREACH(ssk, &sdp_list, list) {
1757 if (ssk->ib_device != device)
1760 if ((ssk->flags & SDP_DESTROY) == 0)
1761 ssk = sdp_notify(ssk, ECONNRESET);
1767 * XXX Do I need to wait between these two?
1769 sdp_dev = ib_get_client_data(device, &sdp_client);
1772 ib_flush_fmr_pool(sdp_dev->fmr_pool);
1773 ib_destroy_fmr_pool(sdp_dev->fmr_pool);
1774 ib_dealloc_pd(sdp_dev->pd);
1775 free(sdp_dev, M_SDP);
1778 struct ib_client sdp_client =
1779 { .name = "sdp", .add = sdp_dev_add, .remove = sdp_dev_rem };
1783 sdp_pcblist(SYSCTL_HANDLER_ARGS)
1786 struct sdp_sock *ssk;
1790 * The process of preparing the TCB list is too time-consuming and
1791 * resource-intensive to repeat twice on every request.
1793 if (req->oldptr == NULL) {
1795 n += imax(n / 8, 10);
1796 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1800 if (req->newptr != NULL)
1804 * OK, now we're committed to doing something.
1810 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1811 + n * sizeof(struct xtcpcb));
1815 bzero(&xig, sizeof(xig));
1816 xig.xig_len = sizeof xig;
1819 xig.xig_sogen = so_gencnt;
1820 error = SYSCTL_OUT(req, &xig, sizeof xig);
1825 for (ssk = LIST_FIRST(&sdp_list), i = 0;
1826 ssk != NULL && i < n; ssk = LIST_NEXT(ssk, list)) {
1830 if (ssk->flags & SDP_TIMEWAIT) {
1831 if (ssk->cred != NULL)
1832 error = cr_cansee(req->td->td_ucred,
1835 error = EINVAL; /* Skip this inp. */
1836 } else if (ssk->socket)
1837 error = cr_canseesocket(req->td->td_ucred,
1846 bzero(&xt, sizeof(xt));
1847 xt.xt_len = sizeof xt;
1848 xt.xt_inp.inp_gencnt = 0;
1849 xt.xt_inp.inp_vflag = INP_IPV4;
1850 memcpy(&xt.xt_inp.inp_laddr, &ssk->laddr, sizeof(ssk->laddr));
1851 xt.xt_inp.inp_lport = ssk->lport;
1852 memcpy(&xt.xt_inp.inp_faddr, &ssk->faddr, sizeof(ssk->faddr));
1853 xt.xt_inp.inp_fport = ssk->fport;
1854 xt.t_state = ssk->state;
1855 if (ssk->socket != NULL)
1856 sotoxsocket(ssk->socket, &xt.xt_inp.xi_socket);
1857 xt.xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
1859 error = SYSCTL_OUT(req, &xt, sizeof xt);
1869 * Give the user an updated idea of our state.
1870 * If the generation differs from what we told
1871 * her before, she knows that something happened
1872 * while we were processing this request, and it
1873 * might be necessary to retry.
1876 xig.xig_sogen = so_gencnt;
1877 xig.xig_count = sdp_count;
1878 error = SYSCTL_OUT(req, &xig, sizeof xig);
1884 static SYSCTL_NODE(_net_inet, -1, sdp, CTLFLAG_RW, 0, "SDP");
1886 SYSCTL_PROC(_net_inet_sdp, TCPCTL_PCBLIST, pcblist,
1887 CTLFLAG_RD | CTLTYPE_STRUCT, 0, 0, sdp_pcblist, "S,xtcpcb",
1888 "List of active SDP connections");
1891 sdp_zone_change(void *tag)
1894 uma_zone_set_max(sdp_zone, maxsockets);
1901 LIST_INIT(&sdp_list);
1902 sdp_zone = uma_zcreate("sdp_sock", sizeof(struct sdp_sock),
1903 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1904 uma_zone_set_max(sdp_zone, maxsockets);
1905 EVENTHANDLER_REGISTER(maxsockets_change, sdp_zone_change, NULL,
1906 EVENTHANDLER_PRI_ANY);
1907 rx_comp_wq = create_singlethread_workqueue("rx_comp_wq");
1908 ib_register_client(&sdp_client);
1911 extern struct domain sdpdomain;
1913 struct pr_usrreqs sdp_usrreqs = {
1914 .pru_abort = sdp_abort,
1915 .pru_accept = sdp_accept,
1916 .pru_attach = sdp_attach,
1917 .pru_bind = sdp_bind,
1918 .pru_connect = sdp_connect,
1919 .pru_control = sdp_control,
1920 .pru_detach = sdp_detach,
1921 .pru_disconnect = sdp_disconnect,
1922 .pru_listen = sdp_listen,
1923 .pru_peeraddr = sdp_getpeeraddr,
1924 .pru_rcvoob = sdp_rcvoob,
1925 .pru_send = sdp_send,
1926 .pru_sosend = sdp_sosend,
1927 .pru_soreceive = sdp_sorecv,
1928 .pru_shutdown = sdp_shutdown,
1929 .pru_sockaddr = sdp_getsockaddr,
1930 .pru_close = sdp_close,
1933 struct protosw sdpsw[] = {
1935 .pr_type = SOCK_STREAM,
1936 .pr_domain = &sdpdomain,
1937 .pr_protocol = IPPROTO_IP,
1938 .pr_flags = PR_CONNREQUIRED|PR_IMPLOPCL|PR_WANTRCVD,
1939 .pr_ctlinput = sdp_ctlinput,
1940 .pr_ctloutput = sdp_ctloutput,
1941 .pr_usrreqs = &sdp_usrreqs
1944 .pr_type = SOCK_STREAM,
1945 .pr_domain = &sdpdomain,
1946 .pr_protocol = IPPROTO_TCP,
1947 .pr_flags = PR_CONNREQUIRED|PR_IMPLOPCL|PR_WANTRCVD,
1948 .pr_ctlinput = sdp_ctlinput,
1949 .pr_ctloutput = sdp_ctloutput,
1950 .pr_usrreqs = &sdp_usrreqs
1954 struct domain sdpdomain = {
1955 .dom_family = AF_INET_SDP,
1957 .dom_init = sdp_init,
1958 .dom_protosw = sdpsw,
1959 .dom_protoswNPROTOSW = &sdpsw[sizeof(sdpsw)/sizeof(sdpsw[0])],
1964 int sdp_debug_level = 1;
1965 int sdp_data_debug_level = 0;