xref: /dragonfly/sys/netinet/tcp_usrreq.c (revision eca362d0)
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
64  */
65 
66 #include "opt_ipsec.h"
67 #include "opt_inet.h"
68 #include "opt_inet6.h"
69 #include "opt_tcpdebug.h"
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
78 
79 #include <sys/mbuf.h>
80 #ifdef INET6
81 #include <sys/domain.h>
82 #endif /* INET6 */
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/socketops.h>
86 #include <sys/protosw.h>
87 
88 #include <sys/thread2.h>
89 #include <sys/msgport2.h>
90 #include <sys/socketvar2.h>
91 
92 #include <net/if.h>
93 #include <net/netisr.h>
94 #include <net/route.h>
95 
96 #include <net/netmsg2.h>
97 #include <net/netisr2.h>
98 
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #ifdef INET6
102 #include <netinet/ip6.h>
103 #endif
104 #include <netinet/in_pcb.h>
105 #ifdef INET6
106 #include <netinet6/in6_pcb.h>
107 #endif
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
110 #ifdef INET6
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/tcp6_var.h>
113 #endif
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_timer2.h>
119 #include <netinet/tcp_var.h>
120 #include <netinet/tcpip.h>
121 #ifdef TCPDEBUG
122 #include <netinet/tcp_debug.h>
123 #endif
124 
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #endif /*IPSEC*/
128 
129 /*
130  * TCP protocol interface to socket abstraction.
131  */
132 extern	char *tcpstates[];	/* XXX ??? */
133 
134 static int	tcp_attach (struct socket *, struct pru_attach_info *);
135 static void	tcp_connect (netmsg_t msg);
136 #ifdef INET6
137 static void	tcp6_connect (netmsg_t msg);
138 static int	tcp6_connect_oncpu(struct tcpcb *tp, int flags,
139 				struct mbuf **mp,
140 				struct sockaddr_in6 *sin6,
141 				struct in6_addr *addr6);
142 #endif /* INET6 */
143 static struct tcpcb *
144 		tcp_disconnect (struct tcpcb *);
145 static struct tcpcb *
146 		tcp_usrclosed (struct tcpcb *);
147 
148 #ifdef TCPDEBUG
149 #define	TCPDEBUG0	int ostate = 0
150 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
151 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
152 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
153 #else
154 #define	TCPDEBUG0
155 #define	TCPDEBUG1()
156 #define	TCPDEBUG2(req)
157 #endif
158 
159 /*
160  * For some ill optimized programs, which try to use TCP_NOPUSH
161  * to improve performance, will have small amount of data sits
162  * in the sending buffer.  These small amount of data will _not_
163  * be pushed into the network until more data are written into
164  * the socket or the socket write side is shutdown.
165  */
166 static int	tcp_disable_nopush = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
168     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
169 
170 /*
171  * Allocate socket buffer space.
172  */
173 static int
174 tcp_usr_preattach(struct socket *so, int proto __unused,
175     struct pru_attach_info *ai)
176 {
177 	int error;
178 
179 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
180 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
181 				  ai->sb_rlimit);
182 		if (error)
183 			return (error);
184 	}
185 	atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
186 	atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
187 
188 	return 0;
189 }
190 
191 /*
192  * TCP attaches to socket via pru_attach(), reserving space,
193  * and an internet control block.  This socket may move to
194  * other CPU later when we bind/connect.
195  */
196 static void
197 tcp_usr_attach(netmsg_t msg)
198 {
199 	struct socket *so = msg->base.nm_so;
200 	struct pru_attach_info *ai = msg->attach.nm_ai;
201 	int error;
202 	struct inpcb *inp;
203 	struct tcpcb *tp = NULL;
204 	TCPDEBUG0;
205 
206 	inp = so->so_pcb;
207 	KASSERT(inp == NULL, ("tcp socket attached"));
208 	TCPDEBUG1();
209 
210 	error = tcp_attach(so, ai);
211 	if (error)
212 		goto out;
213 
214 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
215 		so->so_linger = TCP_LINGERTIME;
216 	tp = sototcpcb(so);
217 out:
218 	TCPDEBUG2(PRU_ATTACH);
219 	lwkt_replymsg(&msg->lmsg, error);
220 }
221 
222 /*
223  * pru_detach() detaches the TCP protocol from the socket.
224  * If the protocol state is non-embryonic, then can't
225  * do this directly: have to initiate a pru_disconnect(),
226  * which may finish later; embryonic TCB's can just
227  * be discarded here.
228  */
229 static void
230 tcp_usr_detach(netmsg_t msg)
231 {
232 	struct socket *so = msg->base.nm_so;
233 	int error = 0;
234 	struct inpcb *inp;
235 	struct tcpcb *tp;
236 	TCPDEBUG0;
237 
238 	inp = so->so_pcb;
239 
240 	/*
241 	 * If the inp is already detached or never attached, it may have
242 	 * been due to an async close or async attach failure.  Just return
243 	 * as if no error occured.
244 	 */
245 	if (inp) {
246 		tp = intotcpcb(inp);
247 		KASSERT(tp != NULL, ("tcp_usr_detach: tp is NULL"));
248 		TCPDEBUG1();
249 		tp = tcp_disconnect(tp);
250 		TCPDEBUG2(PRU_DETACH);
251 	}
252 	lwkt_replymsg(&msg->lmsg, error);
253 }
254 
255 /*
256  * NOTE: ignore_error is non-zero for certain disconnection races
257  * which we want to silently allow, otherwise close() may return
258  * an unexpected error.
259  *
260  * NOTE: The variables (msg) and (tp) are assumed.
261  */
262 #define	COMMON_START(so, inp, ignore_error)			\
263 	TCPDEBUG0; 						\
264 								\
265 	inp = so->so_pcb; 					\
266 	do {							\
267 		if (inp == NULL) {				\
268 			error = ignore_error ? 0 : EINVAL;	\
269 			tp = NULL;				\
270 			goto out;				\
271 		}						\
272 		tp = intotcpcb(inp);				\
273 		TCPDEBUG1();					\
274 	} while(0)
275 
276 #define COMMON_END1(req, noreply)				\
277 	out: do {						\
278 		TCPDEBUG2(req);					\
279 		if (!(noreply))					\
280 			lwkt_replymsg(&msg->lmsg, error);	\
281 		return;						\
282 	} while(0)
283 
284 #define COMMON_END(req)		COMMON_END1((req), 0)
285 
286 static void
287 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
288 {
289 	sosetport(((struct netmsg_base *)msg)->nm_so, port);
290 }
291 
292 /*
293  * Give the socket an address.
294  */
295 static void
296 tcp_usr_bind(netmsg_t msg)
297 {
298 	struct socket *so = msg->bind.base.nm_so;
299 	struct sockaddr *nam = msg->bind.nm_nam;
300 	struct thread *td = msg->bind.nm_td;
301 	int error = 0;
302 	struct inpcb *inp;
303 	struct tcpcb *tp;
304 	struct sockaddr_in *sinp;
305 	lwkt_port_t port0 = netisr_cpuport(0);
306 
307 	COMMON_START(so, inp, 0);
308 
309 	/*
310 	 * Must check for multicast addresses and disallow binding
311 	 * to them.
312 	 */
313 	sinp = (struct sockaddr_in *)nam;
314 	if (sinp->sin_family == AF_INET &&
315 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
316 		error = EAFNOSUPPORT;
317 		goto out;
318 	}
319 
320 	/*
321 	 * Check "already bound" here (in_pcbbind() does the same check
322 	 * though), so we don't forward a connected socket to netisr0,
323 	 * which would panic in the following in_pcbunlink().
324 	 */
325 	if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY) {
326 		error = EINVAL;	/* already bound */
327 		goto out;
328 	}
329 
330 	/*
331 	 * Use netisr0 to serialize in_pcbbind(), so that pru_detach and
332 	 * pru_bind for different sockets on the same local port could be
333 	 * properly ordered.  The original race is illustrated here for
334 	 * reference.
335 	 *
336 	 * s1 = socket();
337 	 * bind(s1, *.PORT);
338 	 * close(s1);  <----- asynchronous
339 	 * s2 = socket();
340 	 * bind(s2, *.PORT);
341 	 *
342 	 * All will expect bind(s2, *.PORT) to succeed.  However, it will
343 	 * fail, if following sequence happens due to random socket initial
344 	 * msgport and asynchronous close(2):
345 	 *
346 	 *    netisrN                  netisrM
347 	 *       :                        :
348 	 *       :                    pru_bind(s2) [*.PORT is used by s1]
349 	 *  pru_detach(s1)                :
350 	 */
351 	if (&curthread->td_msgport != port0) {
352 		lwkt_msg_t lmsg = &msg->bind.base.lmsg;
353 
354 		KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0,
355 		    ("already asked to relink"));
356 
357 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
358 		msg->bind.nm_flags |= PRUB_RELINK;
359 
360 		TCP_STATE_MIGRATE_START(tp);
361 
362 		/* See the related comment in tcp_connect() */
363 		lwkt_setmsg_receipt(lmsg, tcp_sosetport);
364 		lwkt_forwardmsg(port0, lmsg);
365 		/* msg invalid now */
366 		return;
367 	}
368 	KASSERT(so->so_port == port0, ("so_port is not netisr0"));
369 
370 	if (msg->bind.nm_flags & PRUB_RELINK) {
371 		msg->bind.nm_flags &= ~PRUB_RELINK;
372 		TCP_STATE_MIGRATE_END(tp);
373 		in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
374 	}
375 	KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
376 
377 	error = in_pcbbind(inp, nam, td);
378 	if (error)
379 		goto out;
380 
381 	COMMON_END(PRU_BIND);
382 }
383 
384 #ifdef INET6
385 
386 static void
387 tcp6_usr_bind(netmsg_t msg)
388 {
389 	struct socket *so = msg->bind.base.nm_so;
390 	struct sockaddr *nam = msg->bind.nm_nam;
391 	struct thread *td = msg->bind.nm_td;
392 	int error = 0;
393 	struct inpcb *inp;
394 	struct tcpcb *tp;
395 	struct sockaddr_in6 *sin6p;
396 
397 	COMMON_START(so, inp, 0);
398 
399 	/*
400 	 * Must check for multicast addresses and disallow binding
401 	 * to them.
402 	 */
403 	sin6p = (struct sockaddr_in6 *)nam;
404 	if (sin6p->sin6_family == AF_INET6 &&
405 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
406 		error = EAFNOSUPPORT;
407 		goto out;
408 	}
409 	error = in6_pcbbind(inp, nam, td);
410 	if (error)
411 		goto out;
412 	COMMON_END(PRU_BIND);
413 }
414 #endif /* INET6 */
415 
416 struct netmsg_inswildcard {
417 	struct netmsg_base	base;
418 	struct inpcb		*nm_inp;
419 };
420 
421 static void
422 in_pcbinswildcardhash_handler(netmsg_t msg)
423 {
424 	struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
425 	int cpu = mycpuid, nextcpu;
426 
427 	in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
428 
429 	nextcpu = cpu + 1;
430 	if (nextcpu < netisr_ncpus)
431 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
432 	else
433 		lwkt_replymsg(&nm->base.lmsg, 0);
434 }
435 
436 /*
437  * Prepare to accept connections.
438  */
439 static void
440 tcp_usr_listen(netmsg_t msg)
441 {
442 	struct socket *so = msg->listen.base.nm_so;
443 	struct thread *td = msg->listen.nm_td;
444 	int error = 0;
445 	struct inpcb *inp;
446 	struct tcpcb *tp;
447 	struct netmsg_inswildcard nm;
448 	lwkt_port_t port0 = netisr_cpuport(0);
449 
450 	COMMON_START(so, inp, 0);
451 
452 	if (&curthread->td_msgport != port0) {
453 		lwkt_msg_t lmsg = &msg->listen.base.lmsg;
454 
455 		KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
456 		    ("already asked to relink"));
457 
458 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
459 		msg->listen.nm_flags |= PRUL_RELINK;
460 
461 		TCP_STATE_MIGRATE_START(tp);
462 
463 		/* See the related comment in tcp_connect() */
464 		lwkt_setmsg_receipt(lmsg, tcp_sosetport);
465 		lwkt_forwardmsg(port0, lmsg);
466 		/* msg invalid now */
467 		return;
468 	}
469 	KASSERT(so->so_port == port0, ("so_port is not netisr0"));
470 
471 	if (msg->listen.nm_flags & PRUL_RELINK) {
472 		msg->listen.nm_flags &= ~PRUL_RELINK;
473 		TCP_STATE_MIGRATE_END(tp);
474 		in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
475 	}
476 	KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
477 
478 	if (tp->t_flags & TF_LISTEN)
479 		goto out;
480 
481 	if (inp->inp_lport == 0) {
482 		error = in_pcbbind(inp, NULL, td);
483 		if (error)
484 			goto out;
485 	}
486 
487 	TCP_STATE_CHANGE(tp, TCPS_LISTEN);
488 	tp->t_flags |= TF_LISTEN;
489 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
490 
491 	/*
492 	 * Create tcpcb per-cpu port cache
493 	 *
494 	 * NOTE:
495 	 * This _must_ be done before installing this inpcb into
496 	 * wildcard hash.
497 	 */
498 	tcp_pcbport_create(tp);
499 
500 	if (netisr_ncpus > 1) {
501 		/*
502 		 * Put this inpcb into wildcard hash on other cpus.
503 		 */
504 		ASSERT_INP_NOTINHASH(inp);
505 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
506 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
507 		nm.nm_inp = inp;
508 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
509 	}
510 	in_pcbinswildcardhash(inp);
511 	COMMON_END(PRU_LISTEN);
512 }
513 
514 #ifdef INET6
515 
516 static void
517 tcp6_usr_listen(netmsg_t msg)
518 {
519 	struct socket *so = msg->listen.base.nm_so;
520 	struct thread *td = msg->listen.nm_td;
521 	int error = 0;
522 	struct inpcb *inp;
523 	struct tcpcb *tp;
524 	struct netmsg_inswildcard nm;
525 
526 	COMMON_START(so, inp, 0);
527 
528 	if (tp->t_flags & TF_LISTEN)
529 		goto out;
530 
531 	if (inp->inp_lport == 0) {
532 		error = in6_pcbbind(inp, NULL, td);
533 		if (error)
534 			goto out;
535 	}
536 
537 	TCP_STATE_CHANGE(tp, TCPS_LISTEN);
538 	tp->t_flags |= TF_LISTEN;
539 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
540 
541 	/*
542 	 * Create tcpcb per-cpu port cache
543 	 *
544 	 * NOTE:
545 	 * This _must_ be done before installing this inpcb into
546 	 * wildcard hash.
547 	 */
548 	tcp_pcbport_create(tp);
549 
550 	if (netisr_ncpus > 1) {
551 		/*
552 		 * Put this inpcb into wildcard hash on other cpus.
553 		 */
554 		KKASSERT(so->so_port == netisr_cpuport(0));
555 		ASSERT_NETISR0;
556 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
557 		ASSERT_INP_NOTINHASH(inp);
558 
559 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
560 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
561 		nm.nm_inp = inp;
562 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
563 	}
564 	in_pcbinswildcardhash(inp);
565 	COMMON_END(PRU_LISTEN);
566 }
567 #endif /* INET6 */
568 
569 /*
570  * Initiate connection to peer.
571  * Create a template for use in transmissions on this connection.
572  * Enter SYN_SENT state, and mark socket as connecting.
573  * Start keep-alive timer, and seed output sequence space.
574  * Send initial segment on connection.
575  */
576 static void
577 tcp_usr_connect(netmsg_t msg)
578 {
579 	struct socket *so = msg->connect.base.nm_so;
580 	struct sockaddr *nam = msg->connect.nm_nam;
581 	struct thread *td = msg->connect.nm_td;
582 	int error = 0;
583 	struct inpcb *inp;
584 	struct tcpcb *tp;
585 	struct sockaddr_in *sinp;
586 
587 	ASSERT_NETISR_NCPUS(mycpuid);
588 
589 	COMMON_START(so, inp, 0);
590 
591 	/*
592 	 * Must disallow TCP ``connections'' to multicast addresses.
593 	 */
594 	sinp = (struct sockaddr_in *)nam;
595 	if (sinp->sin_family == AF_INET
596 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
597 		error = EAFNOSUPPORT;
598 		goto out;
599 	}
600 
601 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
602 		error = EAFNOSUPPORT; /* IPv6 only jail */
603 		goto out;
604 	}
605 
606 	tcp_connect(msg);
607 	/* msg is invalid now */
608 	return;
609 out:
610 	if (msg->connect.nm_m) {
611 		m_freem(msg->connect.nm_m);
612 		msg->connect.nm_m = NULL;
613 	}
614 	if (msg->connect.nm_flags & PRUC_HELDTD)
615 		lwkt_rele(td);
616 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
617 		so->so_error = error;
618 		soisdisconnected(so);
619 	}
620 	lwkt_replymsg(&msg->lmsg, error);
621 }
622 
623 #ifdef INET6
624 
625 static void
626 tcp6_usr_connect(netmsg_t msg)
627 {
628 	struct socket *so = msg->connect.base.nm_so;
629 	struct sockaddr *nam = msg->connect.nm_nam;
630 	struct thread *td = msg->connect.nm_td;
631 	int error = 0;
632 	struct inpcb *inp;
633 	struct tcpcb *tp;
634 	struct sockaddr_in6 *sin6p;
635 
636 	ASSERT_NETISR_NCPUS(mycpuid);
637 
638 	COMMON_START(so, inp, 0);
639 
640 	/*
641 	 * Must disallow TCP ``connections'' to multicast addresses.
642 	 */
643 	sin6p = (struct sockaddr_in6 *)nam;
644 	if (sin6p->sin6_family == AF_INET6
645 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
646 		error = EAFNOSUPPORT;
647 		goto out;
648 	}
649 
650 	if (!prison_remote_ip(td, nam)) {
651 		error = EAFNOSUPPORT; /* IPv4 only jail */
652 		goto out;
653 	}
654 
655 	/* Reject v4-mapped address */
656 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
657 		error = EADDRNOTAVAIL;
658 		goto out;
659 	}
660 
661 	inp->inp_inc.inc_isipv6 = 1;
662 	tcp6_connect(msg);
663 	/* msg is invalid now */
664 	return;
665 out:
666 	if (msg->connect.nm_m) {
667 		m_freem(msg->connect.nm_m);
668 		msg->connect.nm_m = NULL;
669 	}
670 	lwkt_replymsg(&msg->lmsg, error);
671 }
672 
673 #endif /* INET6 */
674 
675 /*
676  * Initiate disconnect from peer.
677  * If connection never passed embryonic stage, just drop;
678  * else if don't need to let data drain, then can just drop anyways,
679  * else have to begin TCP shutdown process: mark socket disconnecting,
680  * drain unread data, state switch to reflect user close, and
681  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
682  * when peer sends FIN and acks ours.
683  *
684  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
685  */
686 static void
687 tcp_usr_disconnect(netmsg_t msg)
688 {
689 	struct socket *so = msg->disconnect.base.nm_so;
690 	int error = 0;
691 	struct inpcb *inp;
692 	struct tcpcb *tp;
693 
694 	COMMON_START(so, inp, 1);
695 	tp = tcp_disconnect(tp);
696 	COMMON_END(PRU_DISCONNECT);
697 }
698 
699 /*
700  * Accept a connection.  Essentially all the work is
701  * done at higher levels; just return the address
702  * of the peer, storing through addr.
703  */
704 static void
705 tcp_usr_accept(netmsg_t msg)
706 {
707 	struct socket *so = msg->accept.base.nm_so;
708 	struct sockaddr **nam = msg->accept.nm_nam;
709 	int error = 0;
710 	struct inpcb *inp;
711 	struct tcpcb *tp = NULL;
712 	TCPDEBUG0;
713 
714 	inp = so->so_pcb;
715 	if (so->so_state & SS_ISDISCONNECTED) {
716 		error = ECONNABORTED;
717 		goto out;
718 	}
719 	if (inp == NULL) {
720 		error = EINVAL;
721 		goto out;
722 	}
723 
724 	tp = intotcpcb(inp);
725 	TCPDEBUG1();
726 	in_setpeeraddr(so, nam);
727 	COMMON_END(PRU_ACCEPT);
728 }
729 
730 #ifdef INET6
731 static void
732 tcp6_usr_accept(netmsg_t msg)
733 {
734 	struct socket *so = msg->accept.base.nm_so;
735 	struct sockaddr **nam = msg->accept.nm_nam;
736 	int error = 0;
737 	struct inpcb *inp;
738 	struct tcpcb *tp = NULL;
739 	TCPDEBUG0;
740 
741 	inp = so->so_pcb;
742 
743 	if (so->so_state & SS_ISDISCONNECTED) {
744 		error = ECONNABORTED;
745 		goto out;
746 	}
747 	if (inp == NULL) {
748 		error = EINVAL;
749 		goto out;
750 	}
751 	tp = intotcpcb(inp);
752 	TCPDEBUG1();
753 	in6_setpeeraddr(so, nam);
754 	COMMON_END(PRU_ACCEPT);
755 }
756 #endif /* INET6 */
757 
758 /*
759  * Mark the connection as being incapable of further output.
760  */
761 static void
762 tcp_usr_shutdown(netmsg_t msg)
763 {
764 	struct socket *so = msg->shutdown.base.nm_so;
765 	int error = 0;
766 	struct inpcb *inp;
767 	struct tcpcb *tp;
768 
769 	COMMON_START(so, inp, 0);
770 	socantsendmore(so);
771 	tp = tcp_usrclosed(tp);
772 	if (tp)
773 		error = tcp_output(tp);
774 	COMMON_END(PRU_SHUTDOWN);
775 }
776 
777 /*
778  * After a receive, possibly send window update to peer.
779  */
780 static void
781 tcp_usr_rcvd(netmsg_t msg)
782 {
783 	struct socket *so = msg->rcvd.base.nm_so;
784 	int error = 0, noreply = 0;
785 	struct inpcb *inp;
786 	struct tcpcb *tp;
787 
788 	COMMON_START(so, inp, 0);
789 
790 	if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
791 		noreply = 1;
792 		so_async_rcvd_reply(so);
793 	}
794 	tcp_output(tp);
795 
796 	COMMON_END1(PRU_RCVD, noreply);
797 }
798 
799 /*
800  * Do a send by putting data in output queue and updating urgent
801  * marker if URG set.  Possibly send more data.  Unlike the other
802  * pru_*() routines, the mbuf chains are our responsibility.  We
803  * must either enqueue them or free them.  The other pru_* routines
804  * generally are caller-frees.
805  */
806 static void
807 tcp_usr_send(netmsg_t msg)
808 {
809 	struct socket *so = msg->send.base.nm_so;
810 	int flags = msg->send.nm_flags;
811 	struct mbuf *m = msg->send.nm_m;
812 	int error = 0;
813 	struct inpcb *inp;
814 	struct tcpcb *tp;
815 	TCPDEBUG0;
816 
817 	KKASSERT(msg->send.nm_control == NULL);
818 	KKASSERT(msg->send.nm_addr == NULL);
819 	KKASSERT((flags & PRUS_FREEADDR) == 0);
820 
821 	inp = so->so_pcb;
822 
823 	if (inp == NULL) {
824 		/*
825 		 * OOPS! we lost a race, the TCP session got reset after
826 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
827 		 * network interrupt in the non-critical section of sosend().
828 		 */
829 		m_freem(m);
830 		error = ECONNRESET;	/* XXX EPIPE? */
831 		tp = NULL;
832 		TCPDEBUG1();
833 		goto out;
834 	}
835 	tp = intotcpcb(inp);
836 	TCPDEBUG1();
837 
838 #ifdef foo
839 	/*
840 	 * This is no longer necessary, since:
841 	 * - sosendtcp() has already checked it for us
842 	 * - It does not work with asynchronized send
843 	 */
844 
845 	/*
846 	 * Don't let too much OOB data build up
847 	 */
848 	if (flags & PRUS_OOB) {
849 		if (ssb_space(&so->so_snd) < -512) {
850 			m_freem(m);
851 			error = ENOBUFS;
852 			goto out;
853 		}
854 	}
855 #endif
856 
857 	/*
858 	 * Pump the data into the socket.
859 	 */
860 	if (m) {
861 		ssb_appendstream(&so->so_snd, m);
862 		sowwakeup(so);
863 	}
864 	if (flags & PRUS_OOB) {
865 		/*
866 		 * According to RFC961 (Assigned Protocols),
867 		 * the urgent pointer points to the last octet
868 		 * of urgent data.  We continue, however,
869 		 * to consider it to indicate the first octet
870 		 * of data past the urgent section.
871 		 * Otherwise, snd_up should be one lower.
872 		 */
873 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
874 		tp->t_flags |= TF_FORCE;
875 		error = tcp_output(tp);
876 		tp->t_flags &= ~TF_FORCE;
877 	} else {
878 		if (flags & PRUS_EOF) {
879 			/*
880 			 * Close the send side of the connection after
881 			 * the data is sent.
882 			 */
883 			socantsendmore(so);
884 			tp = tcp_usrclosed(tp);
885 		}
886 		if (tp != NULL && !tcp_output_pending(tp)) {
887 			if (flags & PRUS_MORETOCOME)
888 				tp->t_flags |= TF_MORETOCOME;
889 			error = tcp_output_fair(tp);
890 			if (flags & PRUS_MORETOCOME)
891 				tp->t_flags &= ~TF_MORETOCOME;
892 		}
893 	}
894 	COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
895 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
896 		   (flags & PRUS_NOREPLY));
897 }
898 
899 /*
900  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
901  *	 will sofree() it when we return.
902  */
903 static void
904 tcp_usr_abort(netmsg_t msg)
905 {
906 	struct socket *so = msg->abort.base.nm_so;
907 	int error = 0;
908 	struct inpcb *inp;
909 	struct tcpcb *tp;
910 
911 	COMMON_START(so, inp, 1);
912 	tp = tcp_drop(tp, ECONNABORTED);
913 	COMMON_END(PRU_ABORT);
914 }
915 
916 /*
917  * Receive out-of-band data.
918  */
919 static void
920 tcp_usr_rcvoob(netmsg_t msg)
921 {
922 	struct socket *so = msg->rcvoob.base.nm_so;
923 	struct mbuf *m = msg->rcvoob.nm_m;
924 	int flags = msg->rcvoob.nm_flags;
925 	int error = 0;
926 	struct inpcb *inp;
927 	struct tcpcb *tp;
928 
929 	COMMON_START(so, inp, 0);
930 	if ((so->so_oobmark == 0 &&
931 	     (so->so_state & SS_RCVATMARK) == 0) ||
932 	    so->so_options & SO_OOBINLINE ||
933 	    tp->t_oobflags & TCPOOB_HADDATA) {
934 		error = EINVAL;
935 		goto out;
936 	}
937 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
938 		error = EWOULDBLOCK;
939 		goto out;
940 	}
941 	m->m_len = 1;
942 	*mtod(m, caddr_t) = tp->t_iobc;
943 	if ((flags & MSG_PEEK) == 0)
944 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
945 	COMMON_END(PRU_RCVOOB);
946 }
947 
948 static void
949 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
950 {
951 	in_savefaddr(so, faddr);
952 }
953 
954 #ifdef INET6
955 static void
956 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
957 {
958 	in6_savefaddr(so, faddr);
959 }
960 #endif
961 
962 static int
963 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
964     struct thread *td __unused)
965 {
966 	const struct sockaddr_in *sinp;
967 
968 	sinp = (const struct sockaddr_in *)nam;
969 	if (sinp->sin_family == AF_INET &&
970 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
971 		return EAFNOSUPPORT;
972 
973 	soisconnecting(so);
974 	return 0;
975 }
976 
977 /* xxx - should be const */
978 struct pr_usrreqs tcp_usrreqs = {
979 	.pru_abort = tcp_usr_abort,
980 	.pru_accept = tcp_usr_accept,
981 	.pru_attach = tcp_usr_attach,
982 	.pru_bind = tcp_usr_bind,
983 	.pru_connect = tcp_usr_connect,
984 	.pru_connect2 = pr_generic_notsupp,
985 	.pru_control = in_control_dispatch,
986 	.pru_detach = tcp_usr_detach,
987 	.pru_disconnect = tcp_usr_disconnect,
988 	.pru_listen = tcp_usr_listen,
989 	.pru_peeraddr = in_setpeeraddr_dispatch,
990 	.pru_rcvd = tcp_usr_rcvd,
991 	.pru_rcvoob = tcp_usr_rcvoob,
992 	.pru_send = tcp_usr_send,
993 	.pru_sense = pru_sense_null,
994 	.pru_shutdown = tcp_usr_shutdown,
995 	.pru_sockaddr = in_setsockaddr_dispatch,
996 	.pru_sosend = sosendtcp,
997 	.pru_soreceive = sorecvtcp,
998 	.pru_savefaddr = tcp_usr_savefaddr,
999 	.pru_preconnect = tcp_usr_preconnect,
1000 	.pru_preattach = tcp_usr_preattach
1001 };
1002 
1003 #ifdef INET6
1004 struct pr_usrreqs tcp6_usrreqs = {
1005 	.pru_abort = tcp_usr_abort,
1006 	.pru_accept = tcp6_usr_accept,
1007 	.pru_attach = tcp_usr_attach,
1008 	.pru_bind = tcp6_usr_bind,
1009 	.pru_connect = tcp6_usr_connect,
1010 	.pru_connect2 = pr_generic_notsupp,
1011 	.pru_control = in6_control_dispatch,
1012 	.pru_detach = tcp_usr_detach,
1013 	.pru_disconnect = tcp_usr_disconnect,
1014 	.pru_listen = tcp6_usr_listen,
1015 	.pru_peeraddr = in6_setpeeraddr_dispatch,
1016 	.pru_rcvd = tcp_usr_rcvd,
1017 	.pru_rcvoob = tcp_usr_rcvoob,
1018 	.pru_send = tcp_usr_send,
1019 	.pru_sense = pru_sense_null,
1020 	.pru_shutdown = tcp_usr_shutdown,
1021 	.pru_sockaddr = in6_setsockaddr_dispatch,
1022 	.pru_sosend = sosendtcp,
1023 	.pru_soreceive = sorecvtcp,
1024 	.pru_savefaddr = tcp6_usr_savefaddr
1025 };
1026 #endif /* INET6 */
1027 
1028 static int
1029 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1030 		  const struct sockaddr_in *sin, struct sockaddr_in *if_sin,
1031 		  uint16_t hash)
1032 {
1033 	struct inpcb *inp = tp->t_inpcb, *oinp;
1034 	struct socket *so = inp->inp_socket;
1035 	struct route *ro = &inp->inp_route;
1036 
1037 	KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
1038 	    ("pcbinfo mismatch"));
1039 
1040 	oinp = in_pcblookup_hash(inp->inp_pcbinfo,
1041 				 sin->sin_addr, sin->sin_port,
1042 				 (inp->inp_laddr.s_addr != INADDR_ANY ?
1043 				  inp->inp_laddr : if_sin->sin_addr),
1044 				inp->inp_lport, 0, NULL);
1045 	if (oinp != NULL) {
1046 		m_freem(m);
1047 		return (EADDRINUSE);
1048 	}
1049 	if (inp->inp_laddr.s_addr == INADDR_ANY)
1050 		inp->inp_laddr = if_sin->sin_addr;
1051 	KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1052 	    ("faddr mismatch for reconnect"));
1053 	KASSERT(inp->inp_fport == sin->sin_port,
1054 	    ("fport mismatch for reconnect"));
1055 	in_pcbinsconnhash(inp);
1056 
1057 	inp->inp_flags |= INP_HASH;
1058 	inp->inp_hashval = hash;
1059 
1060 	/*
1061 	 * We are now on the inpcb's owner CPU, if the cached route was
1062 	 * freed because the rtentry's owner CPU is not the current CPU
1063 	 * (e.g. in tcp_connect()), then we try to reallocate it here with
1064 	 * the hope that a rtentry may be cloned from a RTF_PRCLONING
1065 	 * rtentry.
1066 	 */
1067 	if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1068 	    ro->ro_rt == NULL) {
1069 		bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1070 		ro->ro_dst.sa_family = AF_INET;
1071 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1072 		((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1073 			sin->sin_addr;
1074 		rtalloc(ro);
1075 	}
1076 
1077 	/*
1078 	 * Now that no more errors can occur, change the protocol processing
1079 	 * port to the current thread (which is the correct thread).
1080 	 *
1081 	 * Create TCP timer message now; we are on the tcpcb's owner
1082 	 * CPU/thread.
1083 	 */
1084 	tcp_create_timermsg(tp, &curthread->td_msgport);
1085 
1086 	/*
1087 	 * Compute window scaling to request.  Use a larger scaling then
1088 	 * needed for the initial receive buffer in case the receive buffer
1089 	 * gets expanded.
1090 	 */
1091 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1092 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1093 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1094 	       (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1095 	) {
1096 		tp->request_r_scale++;
1097 	}
1098 
1099 	soisconnecting(so);
1100 	tcpstat.tcps_connattempt++;
1101 	TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1102 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1103 	tp->iss = tcp_new_isn(tp);
1104 	tcp_sendseqinit(tp);
1105 	if (m) {
1106 		ssb_appendstream(&so->so_snd, m);
1107 		m = NULL;
1108 		if (flags & PRUS_OOB)
1109 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1110 	}
1111 
1112 	/*
1113 	 * Close the send side of the connection after
1114 	 * the data is sent if flagged.
1115 	 */
1116 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1117 		socantsendmore(so);
1118 		tp = tcp_usrclosed(tp);
1119 	}
1120 	return (tcp_output(tp));
1121 }
1122 
1123 /*
1124  * Common subroutine to open a TCP connection to remote host specified
1125  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1126  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1127  * a local host address (interface).
1128  * Initialize connection parameters and enter SYN-SENT state.
1129  */
1130 static void
1131 tcp_connect(netmsg_t msg)
1132 {
1133 	struct socket *so = msg->connect.base.nm_so;
1134 	struct sockaddr *nam = msg->connect.nm_nam;
1135 	struct thread *td = msg->connect.nm_td;
1136 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1137 	struct sockaddr_in *if_sin = NULL;
1138 	struct inpcb *inp;
1139 	struct tcpcb *tp;
1140 	int error;
1141 	uint16_t hash;
1142 	lwkt_port_t port;
1143 
1144 	COMMON_START(so, inp, 0);
1145 
1146 	/*
1147 	 * Reconnect our pcb if we have to
1148 	 */
1149 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1150 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1151 		TCP_STATE_MIGRATE_END(tp);
1152 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1153 	} else {
1154 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1155 			kprintf("inpcb %p, double-connect race\n", inp);
1156 			error = EISCONN;
1157 			if (so->so_state & SS_ISCONNECTING)
1158 				error = EALREADY;
1159 			goto out;
1160 		}
1161 		KASSERT(inp->inp_fport == 0, ("invalid fport"));
1162 	}
1163 
1164 	/*
1165 	 * Select local port, if it is not yet selected.
1166 	 */
1167 	if (inp->inp_lport == 0) {
1168 		KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1169 
1170 		error = in_pcbladdr(inp, nam, &if_sin, td);
1171 		if (error)
1172 			goto out;
1173 		inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1174 		msg->connect.nm_flags |= PRUC_HASLADDR;
1175 
1176 		/*
1177 		 * Install faddr/fport earlier, so that when this
1178 		 * inpcb is installed on to the lport hash, the
1179 		 * 4-tuple contains correct value.
1180 		 *
1181 		 * NOTE: The faddr/fport will have to be installed
1182 		 * after the in_pcbladdr(), which may change them.
1183 		 */
1184 		inp->inp_faddr = sin->sin_addr;
1185 		inp->inp_fport = sin->sin_port;
1186 
1187 		error = in_pcbbind_remote(inp, nam, td);
1188 		if (error)
1189 			goto out;
1190 	}
1191 
1192 	if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1193 		/*
1194 		 * Rarely used path:
1195 		 * This inpcb was bound before this connect.
1196 		 */
1197 		error = in_pcbladdr(inp, nam, &if_sin, td);
1198 		if (error)
1199 			goto out;
1200 
1201 		/*
1202 		 * Save or refresh the faddr/fport, since they may
1203 		 * be changed by in_pcbladdr().
1204 		 */
1205 		inp->inp_faddr = sin->sin_addr;
1206 		inp->inp_fport = sin->sin_port;
1207 	}
1208 #ifdef INVARIANTS
1209 	else {
1210 		KASSERT(inp->inp_faddr.s_addr == sin->sin_addr.s_addr,
1211 		    ("faddr mismatch for reconnect"));
1212 		KASSERT(inp->inp_fport == sin->sin_port,
1213 		    ("fport mismatch for reconnect"));
1214 	}
1215 #endif
1216 	KKASSERT(inp->inp_socket == so);
1217 
1218 	hash = tcp_addrhash(sin->sin_addr.s_addr, sin->sin_port,
1219 			    (inp->inp_laddr.s_addr != INADDR_ANY ?
1220 			     inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1221 			    inp->inp_lport);
1222 	port = netisr_hashport(hash);
1223 
1224 	if (port != &curthread->td_msgport) {
1225 		lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1226 
1227 		/*
1228 		 * in_pcbladdr() may have allocated a route entry for us
1229 		 * on the current CPU, but we need a route entry on the
1230 		 * inpcb's owner CPU, so free it here.
1231 		 */
1232 		in_pcbresetroute(inp);
1233 
1234 		/*
1235 		 * We are moving the protocol processing port the socket
1236 		 * is on, we have to unlink here and re-link on the
1237 		 * target cpu.
1238 		 */
1239 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1240 		msg->connect.nm_flags |= PRUC_RECONNECT;
1241 		msg->connect.base.nm_dispatch = tcp_connect;
1242 
1243 		TCP_STATE_MIGRATE_START(tp);
1244 
1245 		/*
1246 		 * Use message put done receipt to change this socket's
1247 		 * so_port, i.e. _after_ this message was put onto the
1248 		 * target netisr's msgport but _before_ the message could
1249 		 * be pulled from the target netisr's msgport, so that:
1250 		 * - The upper half (socket code) will not see the new
1251 		 *   msgport before this message reaches the new msgport
1252 		 *   and messages for this socket will be ordered.
1253 		 * - This message will see the new msgport, when its
1254 		 *   handler is called in the target netisr.
1255 		 *
1256 		 * NOTE:
1257 		 * We MUST use messege put done receipt to change this
1258 		 * socket's so_port:
1259 		 * If we changed the so_port in this netisr after the
1260 		 * lwkt_forwardmsg (so messages for this socket will be
1261 		 * ordered) and changed the so_port in the target netisr
1262 		 * at the very beginning of this message's handler, we
1263 		 * would suffer so_port overwritten race, given this
1264 		 * message might be forwarded again.
1265 		 *
1266 		 * NOTE:
1267 		 * This mechanism depends on that the netisr's msgport
1268 		 * is spin msgport (currently it is :).
1269 		 *
1270 		 * If the upper half saw the new msgport before this
1271 		 * message reached the target netisr's msgport, the
1272 		 * messages sent from the upper half could reach the new
1273 		 * msgport before this message, thus there would be
1274 		 * message reordering.  The worst case could be soclose()
1275 		 * saw the new msgport and the detach message could reach
1276 		 * the new msgport before this message, i.e. the inpcb
1277 		 * could have been destroyed when this message was still
1278 		 * pending on or on its way to the new msgport.  Other
1279 		 * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1280 		 * since we have unlinked this inpcb from the current
1281 		 * pcbinfo first.
1282 		 */
1283 		lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1284 		lwkt_forwardmsg(port, lmsg);
1285 		/* msg invalid now */
1286 		return;
1287 	} else if (msg->connect.nm_flags & PRUC_HELDTD) {
1288 		/*
1289 		 * The original thread is no longer needed; release it.
1290 		 */
1291 		lwkt_rele(td);
1292 		msg->connect.nm_flags &= ~PRUC_HELDTD;
1293 	}
1294 	error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1295 				  msg->connect.nm_m, sin, if_sin, hash);
1296 	msg->connect.nm_m = NULL;
1297 out:
1298 	if (msg->connect.nm_m) {
1299 		m_freem(msg->connect.nm_m);
1300 		msg->connect.nm_m = NULL;
1301 	}
1302 	if (msg->connect.nm_flags & PRUC_HELDTD)
1303 		lwkt_rele(td);
1304 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1305 		so->so_error = error;
1306 		soisdisconnected(so);
1307 	}
1308 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1309 	/* msg invalid now */
1310 }
1311 
1312 #ifdef INET6
1313 
1314 static void
1315 tcp6_connect(netmsg_t msg)
1316 {
1317 	struct tcpcb *tp;
1318 	struct socket *so = msg->connect.base.nm_so;
1319 	struct sockaddr *nam = msg->connect.nm_nam;
1320 	struct thread *td = msg->connect.nm_td;
1321 	struct inpcb *inp;
1322 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1323 	struct in6_addr *addr6;
1324 	lwkt_port_t port;
1325 	int error;
1326 
1327 	COMMON_START(so, inp, 0);
1328 
1329 	/*
1330 	 * Reconnect our pcb if we have to
1331 	 */
1332 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1333 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1334 		TCP_STATE_MIGRATE_END(tp);
1335 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1336 	}
1337 
1338 	/*
1339 	 * Bind if we have to
1340 	 */
1341 	if (inp->inp_lport == 0) {
1342 		error = in6_pcbbind(inp, NULL, td);
1343 		if (error)
1344 			goto out;
1345 	}
1346 
1347 	/*
1348 	 * Cannot simply call in_pcbconnect, because there might be an
1349 	 * earlier incarnation of this same connection still in
1350 	 * TIME_WAIT state, creating an ADDRINUSE error.
1351 	 */
1352 	error = in6_pcbladdr(inp, nam, &addr6, td);
1353 	if (error)
1354 		goto out;
1355 
1356 	port = tcp6_addrport();	/* XXX hack for now, always cpu0 */
1357 
1358 	if (port != &curthread->td_msgport) {
1359 		lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1360 
1361 		/*
1362 		 * in_pcbladdr() may have allocated a route entry for us
1363 		 * on the current CPU, but we need a route entry on the
1364 		 * inpcb's owner CPU, so free it here.
1365 		 */
1366 		in_pcbresetroute(inp);
1367 
1368 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1369 		msg->connect.nm_flags |= PRUC_RECONNECT;
1370 		msg->connect.base.nm_dispatch = tcp6_connect;
1371 
1372 		TCP_STATE_MIGRATE_START(tp);
1373 
1374 		/* See the related comment in tcp_connect() */
1375 		lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1376 		lwkt_forwardmsg(port, lmsg);
1377 		/* msg invalid now */
1378 		return;
1379 	}
1380 	error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1381 				   &msg->connect.nm_m, sin6, addr6);
1382 	/* nm_m may still be intact */
1383 out:
1384 	if (msg->connect.nm_m) {
1385 		m_freem(msg->connect.nm_m);
1386 		msg->connect.nm_m = NULL;
1387 	}
1388 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1389 	/* msg invalid now */
1390 }
1391 
1392 static int
1393 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1394 		   struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1395 {
1396 	struct mbuf *m = *mp;
1397 	struct inpcb *inp = tp->t_inpcb;
1398 	struct socket *so = inp->inp_socket;
1399 	struct inpcb *oinp;
1400 
1401 	/*
1402 	 * Cannot simply call in_pcbconnect, because there might be an
1403 	 * earlier incarnation of this same connection still in
1404 	 * TIME_WAIT state, creating an ADDRINUSE error.
1405 	 */
1406 	oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1407 				  &sin6->sin6_addr, sin6->sin6_port,
1408 				  (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1409 				      addr6 : &inp->in6p_laddr),
1410 				  inp->inp_lport,  0, NULL);
1411 	if (oinp)
1412 		return (EADDRINUSE);
1413 
1414 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1415 		inp->in6p_laddr = *addr6;
1416 	inp->in6p_faddr = sin6->sin6_addr;
1417 	inp->inp_fport = sin6->sin6_port;
1418 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1419 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1420 	in_pcbinsconnhash(inp);
1421 
1422 	/*
1423 	 * Now that no more errors can occur, change the protocol processing
1424 	 * port to the current thread (which is the correct thread).
1425 	 *
1426 	 * Create TCP timer message now; we are on the tcpcb's owner
1427 	 * CPU/thread.
1428 	 */
1429 	tcp_create_timermsg(tp, &curthread->td_msgport);
1430 
1431 	/* Compute window scaling to request.  */
1432 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1433 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1434 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1435 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1436 		tp->request_r_scale++;
1437 	}
1438 
1439 	soisconnecting(so);
1440 	tcpstat.tcps_connattempt++;
1441 	TCP_STATE_CHANGE(tp, TCPS_SYN_SENT);
1442 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1443 	tp->iss = tcp_new_isn(tp);
1444 	tcp_sendseqinit(tp);
1445 	if (m) {
1446 		ssb_appendstream(&so->so_snd, m);
1447 		*mp = NULL;
1448 		if (flags & PRUS_OOB)
1449 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1450 	}
1451 
1452 	/*
1453 	 * Close the send side of the connection after
1454 	 * the data is sent if flagged.
1455 	 */
1456 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1457 		socantsendmore(so);
1458 		tp = tcp_usrclosed(tp);
1459 	}
1460 	return (tcp_output(tp));
1461 }
1462 
1463 #endif /* INET6 */
1464 
1465 /*
1466  * The new sockopt interface makes it possible for us to block in the
1467  * copyin/out step (if we take a page fault).  Taking a page fault while
1468  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1469  * both now use TSM, there probably isn't any need for this function to
1470  * run in a critical section any more.  This needs more examination.)
1471  */
1472 void
1473 tcp_ctloutput(netmsg_t msg)
1474 {
1475 	struct socket *so = msg->base.nm_so;
1476 	struct sockopt *sopt = msg->ctloutput.nm_sopt;
1477 	struct thread *td = NULL;
1478 	int	error, opt, optval, opthz;
1479 	struct	inpcb *inp;
1480 	struct	tcpcb *tp;
1481 
1482 	if (msg->ctloutput.nm_flags & PRCO_HELDTD)
1483 		td = sopt->sopt_td;
1484 
1485 	error = 0;
1486 	inp = so->so_pcb;
1487 	if (inp == NULL) {
1488 		error = ECONNRESET;
1489 		goto done;
1490 	}
1491 	tp = intotcpcb(inp);
1492 
1493 	/* Get socket's owner cpuid hint */
1494 	if (sopt->sopt_level == SOL_SOCKET &&
1495 	    sopt->sopt_dir == SOPT_GET &&
1496 	    sopt->sopt_name == SO_CPUHINT) {
1497 		if (tp->t_flags & TF_LISTEN) {
1498 			/*
1499 			 * Listen sockets owner cpuid is always 0,
1500 			 * which does not make sense if SO_REUSEPORT
1501 			 * is not set.
1502 			 *
1503 			 * NOTE: inp_lgrpindex is _not_ assigned in jail.
1504 			 */
1505 			if ((so->so_options & SO_REUSEPORT) &&
1506 			    inp->inp_lgrpindex >= 0)
1507 				optval = inp->inp_lgrpindex % netisr_ncpus;
1508 			else
1509 				optval = -1; /* no hint */
1510 		} else {
1511 			optval = mycpuid;
1512 		}
1513 		soopt_from_kbuf(sopt, &optval, sizeof(optval));
1514 		goto done;
1515 	}
1516 
1517 	if (sopt->sopt_level != IPPROTO_TCP) {
1518 		if (sopt->sopt_level == IPPROTO_IP) {
1519 			switch (sopt->sopt_name) {
1520 			case IP_MULTICAST_IF:
1521 			case IP_MULTICAST_VIF:
1522 			case IP_MULTICAST_TTL:
1523 			case IP_MULTICAST_LOOP:
1524 			case IP_ADD_MEMBERSHIP:
1525 			case IP_DROP_MEMBERSHIP:
1526 				/*
1527 				 * Multicast does not make sense on
1528 				 * TCP sockets.
1529 				 */
1530 				error = EOPNOTSUPP;
1531 				goto done;
1532 			}
1533 		}
1534 #ifdef INET6
1535 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1536 			ip6_ctloutput_dispatch(msg);
1537 		else
1538 #endif /* INET6 */
1539 		ip_ctloutput(msg);
1540 		/* msg invalid now */
1541 		if (td != NULL)
1542 			lwkt_rele(td);
1543 		return;
1544 	}
1545 
1546 	switch (sopt->sopt_dir) {
1547 	case SOPT_SET:
1548 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1549 				      sizeof optval);
1550 		if (error)
1551 			break;
1552 		switch (sopt->sopt_name) {
1553 		case TCP_FASTKEEP:
1554 			if (optval > 0)
1555 				tp->t_keepidle = tp->t_keepintvl;
1556 			else
1557 				tp->t_keepidle = tcp_keepidle;
1558 			tcp_timer_keep_activity(tp, 0);
1559 			break;
1560 #ifdef TCP_SIGNATURE
1561 		case TCP_SIGNATURE_ENABLE:
1562 			if (tp->t_state == TCPS_CLOSED) {
1563 				/*
1564 				 * This is the only safe state that this
1565 				 * option could be changed.  Some segments
1566 				 * could already have been sent in other
1567 				 * states.
1568 				 */
1569 				if (optval > 0)
1570 					tp->t_flags |= TF_SIGNATURE;
1571 				else
1572 					tp->t_flags &= ~TF_SIGNATURE;
1573 			} else {
1574 				error = EOPNOTSUPP;
1575 			}
1576 			break;
1577 #endif /* TCP_SIGNATURE */
1578 		case TCP_NODELAY:
1579 		case TCP_NOOPT:
1580 			switch (sopt->sopt_name) {
1581 			case TCP_NODELAY:
1582 				opt = TF_NODELAY;
1583 				break;
1584 			case TCP_NOOPT:
1585 				opt = TF_NOOPT;
1586 				break;
1587 			default:
1588 				opt = 0; /* dead code to fool gcc */
1589 				break;
1590 			}
1591 
1592 			if (optval)
1593 				tp->t_flags |= opt;
1594 			else
1595 				tp->t_flags &= ~opt;
1596 			break;
1597 
1598 		case TCP_NOPUSH:
1599 			if (tcp_disable_nopush)
1600 				break;
1601 			if (optval)
1602 				tp->t_flags |= TF_NOPUSH;
1603 			else {
1604 				tp->t_flags &= ~TF_NOPUSH;
1605 				error = tcp_output(tp);
1606 			}
1607 			break;
1608 
1609 		case TCP_MAXSEG:
1610 			/*
1611 			 * Must be between 0 and maxseg.  If the requested
1612 			 * maxseg is too small to satisfy the desired minmss,
1613 			 * pump it up (silently so sysctl modifications of
1614 			 * minmss do not create unexpected program failures).
1615 			 * Handle degenerate cases.
1616 			 */
1617 			if (optval > 0 && optval <= tp->t_maxseg) {
1618 				if (optval + 40 < tcp_minmss) {
1619 					optval = tcp_minmss - 40;
1620 					if (optval < 0)
1621 						optval = 1;
1622 				}
1623 				tp->t_maxseg = optval;
1624 			} else {
1625 				error = EINVAL;
1626 			}
1627 			break;
1628 
1629 		case TCP_KEEPINIT:
1630 			opthz = ((int64_t)optval * hz) / 1000;
1631 			if (opthz >= 1)
1632 				tp->t_keepinit = opthz;
1633 			else
1634 				error = EINVAL;
1635 			break;
1636 
1637 		case TCP_KEEPIDLE:
1638 			opthz = ((int64_t)optval * hz) / 1000;
1639 			if (opthz >= 1) {
1640 				tp->t_keepidle = opthz;
1641 				tcp_timer_keep_activity(tp, 0);
1642 			} else {
1643 				error = EINVAL;
1644 			}
1645 			break;
1646 
1647 		case TCP_KEEPINTVL:
1648 			opthz = ((int64_t)optval * hz) / 1000;
1649 			if (opthz >= 1) {
1650 				tp->t_keepintvl = opthz;
1651 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1652 			} else {
1653 				error = EINVAL;
1654 			}
1655 			break;
1656 
1657 		case TCP_KEEPCNT:
1658 			if (optval > 0) {
1659 				tp->t_keepcnt = optval;
1660 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1661 			} else {
1662 				error = EINVAL;
1663 			}
1664 			break;
1665 
1666 		default:
1667 			error = ENOPROTOOPT;
1668 			break;
1669 		}
1670 		break;
1671 
1672 	case SOPT_GET:
1673 		switch (sopt->sopt_name) {
1674 #ifdef TCP_SIGNATURE
1675 		case TCP_SIGNATURE_ENABLE:
1676 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1677 			break;
1678 #endif /* TCP_SIGNATURE */
1679 		case TCP_NODELAY:
1680 			optval = tp->t_flags & TF_NODELAY;
1681 			break;
1682 		case TCP_MAXSEG:
1683 			optval = tp->t_maxseg;
1684 			break;
1685 		case TCP_NOOPT:
1686 			optval = tp->t_flags & TF_NOOPT;
1687 			break;
1688 		case TCP_NOPUSH:
1689 			optval = tp->t_flags & TF_NOPUSH;
1690 			break;
1691 		case TCP_KEEPINIT:
1692 			optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1693 			break;
1694 		case TCP_KEEPIDLE:
1695 			optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1696 			break;
1697 		case TCP_KEEPINTVL:
1698 			optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1699 			break;
1700 		case TCP_KEEPCNT:
1701 			optval = tp->t_keepcnt;
1702 			break;
1703 		default:
1704 			error = ENOPROTOOPT;
1705 			break;
1706 		}
1707 		if (error == 0)
1708 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1709 		break;
1710 	}
1711 done:
1712 	if (td != NULL)
1713 		lwkt_rele(td);
1714 	lwkt_replymsg(&msg->lmsg, error);
1715 }
1716 
1717 struct netmsg_tcp_ctloutput {
1718 	struct netmsg_pr_ctloutput ctloutput;
1719 	struct sockopt		sopt;
1720 	int			sopt_val;
1721 };
1722 
1723 /*
1724  * Allocate netmsg_pr_ctloutput for asynchronous tcp_ctloutput.
1725  */
1726 struct netmsg_pr_ctloutput *
1727 tcp_ctloutmsg(struct sockopt *sopt)
1728 {
1729 	struct netmsg_tcp_ctloutput *msg;
1730 	int flags = 0, error;
1731 
1732 	KASSERT(sopt->sopt_dir == SOPT_SET, ("not from ctloutput"));
1733 
1734 	/* Only small set of options allows asynchronous setting. */
1735 	if (sopt->sopt_level != IPPROTO_TCP)
1736 		return NULL;
1737 	switch (sopt->sopt_name) {
1738 	case TCP_NODELAY:
1739 	case TCP_NOOPT:
1740 	case TCP_NOPUSH:
1741 	case TCP_FASTKEEP:
1742 		break;
1743 	default:
1744 		return NULL;
1745 	}
1746 
1747 	msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_NULLOK);
1748 	if (msg == NULL) {
1749 		/* Fallback to synchronous tcp_ctloutput */
1750 		return NULL;
1751 	}
1752 
1753 	/* Save the sockopt */
1754 	msg->sopt = *sopt;
1755 
1756 	/* Fixup the sopt.sopt_val ptr */
1757 	error = sooptcopyin(sopt, &msg->sopt_val,
1758 	    sizeof(msg->sopt_val), sizeof(msg->sopt_val));
1759 	if (error) {
1760 		kfree(msg, M_LWKTMSG);
1761 		return NULL;
1762 	}
1763 	msg->sopt.sopt_val = &msg->sopt_val;
1764 
1765 	/* Hold the current thread */
1766 	if (msg->sopt.sopt_td != NULL) {
1767 		flags |= PRCO_HELDTD;
1768 		lwkt_hold(msg->sopt.sopt_td);
1769 	}
1770 
1771 	msg->ctloutput.nm_flags = flags;
1772 	msg->ctloutput.nm_sopt = &msg->sopt;
1773 
1774 	return &msg->ctloutput;
1775 }
1776 
1777 /*
1778  * tcp_sendspace and tcp_recvspace are the default send and receive window
1779  * sizes, respectively.  These are obsolescent (this information should
1780  * be set by the route).
1781  *
1782  * Use a default that does not require tcp window scaling to be turned
1783  * on.  Individual programs or the administrator can increase the default.
1784  */
1785 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1786 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1787     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1788 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1789 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1790     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1791 
1792 /*
1793  * Attach TCP protocol to socket, allocating internet protocol control
1794  * block, tcp control block, buffer space, and entering CLOSED state.
1795  */
1796 static int
1797 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1798 {
1799 	struct inpcb *inp;
1800 	int error;
1801 	int cpu;
1802 #ifdef INET6
1803 	boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1804 #endif
1805 
1806 	if (ai != NULL) {
1807 		error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1808 		if (error)
1809 			return (error);
1810 	} else {
1811 		/* Post attach; do nothing */
1812 	}
1813 
1814 	cpu = mycpu->gd_cpuid;
1815 
1816 	/*
1817 	 * Set the default pcbinfo.  This will likely change when we
1818 	 * bind/connect.
1819 	 */
1820 	error = in_pcballoc(so, &tcbinfo[cpu]);
1821 	if (error)
1822 		return (error);
1823 	inp = so->so_pcb;
1824 #ifdef INET6
1825 	if (isipv6)
1826 		inp->in6p_hops = -1;	/* use kernel default */
1827 #endif
1828 	tcp_newtcpcb(inp);
1829 	/* Keep a reference for asynchronized pru_rcvd */
1830 	soreference(so);
1831 	return (0);
1832 }
1833 
1834 /*
1835  * Initiate (or continue) disconnect.
1836  * If embryonic state, just send reset (once).
1837  * If in ``let data drain'' option and linger null, just drop.
1838  * Otherwise (hard), mark socket disconnecting and drop
1839  * current input data; switch states based on user close, and
1840  * send segment to peer (with FIN).
1841  */
1842 static struct tcpcb *
1843 tcp_disconnect(struct tcpcb *tp)
1844 {
1845 	struct socket *so = tp->t_inpcb->inp_socket;
1846 
1847 	if (tp->t_state < TCPS_ESTABLISHED) {
1848 		tp = tcp_close(tp);
1849 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1850 		tp = tcp_drop(tp, 0);
1851 	} else {
1852 		lwkt_gettoken(&so->so_rcv.ssb_token);
1853 		soisdisconnecting(so);
1854 		sbflush(&so->so_rcv.sb);
1855 		tp = tcp_usrclosed(tp);
1856 		if (tp)
1857 			tcp_output(tp);
1858 		lwkt_reltoken(&so->so_rcv.ssb_token);
1859 	}
1860 	return (tp);
1861 }
1862 
1863 /*
1864  * User issued close, and wish to trail through shutdown states:
1865  * if never received SYN, just forget it.  If got a SYN from peer,
1866  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1867  * If already got a FIN from peer, then almost done; go to LAST_ACK
1868  * state.  In all other cases, have already sent FIN to peer (e.g.
1869  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1870  * for peer to send FIN or not respond to keep-alives, etc.
1871  * We can let the user exit from the close as soon as the FIN is acked.
1872  */
1873 static struct tcpcb *
1874 tcp_usrclosed(struct tcpcb *tp)
1875 {
1876 
1877 	switch (tp->t_state) {
1878 
1879 	case TCPS_CLOSED:
1880 	case TCPS_LISTEN:
1881 		TCP_STATE_CHANGE(tp, TCPS_CLOSED);
1882 		tp = tcp_close(tp);
1883 		break;
1884 
1885 	case TCPS_SYN_SENT:
1886 	case TCPS_SYN_RECEIVED:
1887 		tp->t_flags |= TF_NEEDFIN;
1888 		break;
1889 
1890 	case TCPS_ESTABLISHED:
1891 		TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1892 		break;
1893 
1894 	case TCPS_CLOSE_WAIT:
1895 		TCP_STATE_CHANGE(tp, TCPS_LAST_ACK);
1896 		break;
1897 	}
1898 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1899 		soisdisconnected(tp->t_inpcb->inp_socket);
1900 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1901 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1902 			tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1903 			    tcp_timer_2msl);
1904 		}
1905 	}
1906 	return (tp);
1907 }
1908