xref: /dragonfly/sys/netinet/tcp_usrreq.c (revision 9b5a9965)
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. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
67  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
68  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.45 2007/05/24 20:51:22 dillon Exp $
69  */
70 
71 #include "opt_ipsec.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/sysctl.h>
80 #include <sys/globaldata.h>
81 #include <sys/thread.h>
82 
83 #include <sys/mbuf.h>
84 #ifdef INET6
85 #include <sys/domain.h>
86 #endif /* INET6 */
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/protosw.h>
90 
91 #include <sys/thread2.h>
92 #include <sys/msgport2.h>
93 
94 #include <net/if.h>
95 #include <net/netisr.h>
96 #include <net/route.h>
97 
98 #include <net/netmsg2.h>
99 
100 #include <netinet/in.h>
101 #include <netinet/in_systm.h>
102 #ifdef INET6
103 #include <netinet/ip6.h>
104 #endif
105 #include <netinet/in_pcb.h>
106 #ifdef INET6
107 #include <netinet6/in6_pcb.h>
108 #endif
109 #include <netinet/in_var.h>
110 #include <netinet/ip_var.h>
111 #ifdef INET6
112 #include <netinet6/ip6_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_var.h>
119 #include <netinet/tcpip.h>
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif
123 
124 #ifdef IPSEC
125 #include <netinet6/ipsec.h>
126 #endif /*IPSEC*/
127 
128 /*
129  * TCP protocol interface to socket abstraction.
130  */
131 extern	char *tcpstates[];	/* XXX ??? */
132 
133 static int	tcp_attach (struct socket *, struct pru_attach_info *);
134 static int	tcp_connect (struct tcpcb *, struct sockaddr *,
135 				 struct thread *);
136 #ifdef INET6
137 static int	tcp6_connect (struct tcpcb *, struct sockaddr *,
138 				 struct thread *);
139 #endif /* INET6 */
140 static struct tcpcb *
141 		tcp_disconnect (struct tcpcb *);
142 static struct tcpcb *
143 		tcp_usrclosed (struct tcpcb *);
144 
145 #ifdef TCPDEBUG
146 #define	TCPDEBUG0	int ostate = 0
147 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
148 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
149 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
150 #else
151 #define	TCPDEBUG0
152 #define	TCPDEBUG1()
153 #define	TCPDEBUG2(req)
154 #endif
155 
156 /*
157  * TCP attaches to socket via pru_attach(), reserving space,
158  * and an internet control block.
159  */
160 static int
161 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
162 {
163 	int error;
164 	struct inpcb *inp;
165 	struct tcpcb *tp = 0;
166 	TCPDEBUG0;
167 
168 	crit_enter();
169 	inp = so->so_pcb;
170 	TCPDEBUG1();
171 	if (inp) {
172 		error = EISCONN;
173 		goto out;
174 	}
175 
176 	error = tcp_attach(so, ai);
177 	if (error)
178 		goto out;
179 
180 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
181 		so->so_linger = TCP_LINGERTIME;
182 	tp = sototcpcb(so);
183 out:
184 	TCPDEBUG2(PRU_ATTACH);
185 	crit_exit();
186 	return error;
187 }
188 
189 /*
190  * pru_detach() detaches the TCP protocol from the socket.
191  * If the protocol state is non-embryonic, then can't
192  * do this directly: have to initiate a pru_disconnect(),
193  * which may finish later; embryonic TCB's can just
194  * be discarded here.
195  */
196 static int
197 tcp_usr_detach(struct socket *so)
198 {
199 	int error = 0;
200 	struct inpcb *inp;
201 	struct tcpcb *tp;
202 	TCPDEBUG0;
203 
204 	crit_enter();
205 	inp = so->so_pcb;
206 	if (inp == NULL) {
207 		crit_exit();
208 		return EINVAL;	/* XXX */
209 	}
210 
211 	/*
212 	 * It's possible for the tcpcb (tp) to disconnect from the inp due
213 	 * to tcp_drop()->tcp_close() being called.  This may occur *after*
214 	 * the detach message has been queued so we may find a NULL tp here.
215 	 */
216 	if ((tp = intotcpcb(inp)) != NULL) {
217 		TCPDEBUG1();
218 		tp = tcp_disconnect(tp);
219 		TCPDEBUG2(PRU_DETACH);
220 	}
221 	crit_exit();
222 	return error;
223 }
224 
225 #define	COMMON_START(so, inp)		\
226 			TCPDEBUG0; 	\
227 					\
228 			crit_enter();	\
229 			inp = so->so_pcb; \
230 			do { \
231 				     if (inp == 0) { \
232 					     crit_exit(); \
233 					     return EINVAL; \
234 				     } \
235 				     tp = intotcpcb(inp); \
236 				     TCPDEBUG1(); \
237 		     } while(0)
238 
239 #define COMMON_END(req)	out: TCPDEBUG2(req); crit_exit(); return error; goto out
240 
241 
242 /*
243  * Give the socket an address.
244  */
245 static int
246 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
247 {
248 	int error = 0;
249 	struct inpcb *inp;
250 	struct tcpcb *tp;
251 	struct sockaddr_in *sinp;
252 
253 	COMMON_START(so, inp);
254 
255 	/*
256 	 * Must check for multicast addresses and disallow binding
257 	 * to them.
258 	 */
259 	sinp = (struct sockaddr_in *)nam;
260 	if (sinp->sin_family == AF_INET &&
261 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
262 		error = EAFNOSUPPORT;
263 		goto out;
264 	}
265 	error = in_pcbbind(inp, nam, td);
266 	if (error)
267 		goto out;
268 	COMMON_END(PRU_BIND);
269 
270 }
271 
272 #ifdef INET6
273 static int
274 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
275 {
276 	int error = 0;
277 	struct inpcb *inp;
278 	struct tcpcb *tp;
279 	struct sockaddr_in6 *sin6p;
280 
281 	COMMON_START(so, inp);
282 
283 	/*
284 	 * Must check for multicast addresses and disallow binding
285 	 * to them.
286 	 */
287 	sin6p = (struct sockaddr_in6 *)nam;
288 	if (sin6p->sin6_family == AF_INET6 &&
289 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
290 		error = EAFNOSUPPORT;
291 		goto out;
292 	}
293 	inp->inp_vflag &= ~INP_IPV4;
294 	inp->inp_vflag |= INP_IPV6;
295 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
296 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
297 			inp->inp_vflag |= INP_IPV4;
298 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
299 			struct sockaddr_in sin;
300 
301 			in6_sin6_2_sin(&sin, sin6p);
302 			inp->inp_vflag |= INP_IPV4;
303 			inp->inp_vflag &= ~INP_IPV6;
304 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
305 			goto out;
306 		}
307 	}
308 	error = in6_pcbbind(inp, nam, td);
309 	if (error)
310 		goto out;
311 	COMMON_END(PRU_BIND);
312 }
313 #endif /* INET6 */
314 
315 #ifdef SMP
316 struct netmsg_inswildcard {
317 	struct netmsg		nm_netmsg;
318 	struct inpcb		*nm_inp;
319 	struct inpcbinfo	*nm_pcbinfo;
320 };
321 
322 static void
323 in_pcbinswildcardhash_handler(struct netmsg *msg0)
324 {
325 	struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
326 
327 	in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
328 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
329 }
330 #endif
331 
332 /*
333  * Prepare to accept connections.
334  */
335 static int
336 tcp_usr_listen(struct socket *so, struct thread *td)
337 {
338 	int error = 0;
339 	struct inpcb *inp;
340 	struct tcpcb *tp;
341 #ifdef SMP
342 	int cpu;
343 #endif
344 
345 	COMMON_START(so, inp);
346 	if (inp->inp_lport == 0) {
347 		error = in_pcbbind(inp, NULL, td);
348 		if (error != 0)
349 			goto out;
350 	}
351 
352 	tp->t_state = TCPS_LISTEN;
353 #ifdef SMP
354 	/*
355 	 * We have to set the flag because we can't have other cpus
356 	 * messing with our inp's flags.
357 	 */
358 	inp->inp_flags |= INP_WILDCARD_MP;
359 	for (cpu = 0; cpu < ncpus2; cpu++) {
360 		struct netmsg_inswildcard *msg;
361 
362 		if (cpu == mycpu->gd_cpuid) {
363 			in_pcbinswildcardhash(inp);
364 			continue;
365 		}
366 
367 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
368 			      M_INTWAIT);
369 		netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
370 			    in_pcbinswildcardhash_handler);
371 		msg->nm_inp = inp;
372 		msg->nm_pcbinfo = &tcbinfo[cpu];
373 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
374 	}
375 #else
376 	in_pcbinswildcardhash(inp);
377 #endif
378 	COMMON_END(PRU_LISTEN);
379 }
380 
381 #ifdef INET6
382 static int
383 tcp6_usr_listen(struct socket *so, struct thread *td)
384 {
385 	int error = 0;
386 	struct inpcb *inp;
387 	struct tcpcb *tp;
388 #ifdef SMP
389 	int cpu;
390 #endif
391 
392 	COMMON_START(so, inp);
393 	if (inp->inp_lport == 0) {
394 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
395 			inp->inp_vflag |= INP_IPV4;
396 		else
397 			inp->inp_vflag &= ~INP_IPV4;
398 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
399 	}
400 	if (error == 0)
401 		tp->t_state = TCPS_LISTEN;
402 #ifdef SMP
403 	/*
404 	 * We have to set the flag because we can't have other cpus
405 	 * messing with our inp's flags.
406 	 */
407 	inp->inp_flags |= INP_WILDCARD_MP;
408 	for (cpu = 0; cpu < ncpus2; cpu++) {
409 		struct netmsg_inswildcard *msg;
410 
411 		if (cpu == mycpu->gd_cpuid) {
412 			in_pcbinswildcardhash(inp);
413 			continue;
414 		}
415 
416 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
417 			      M_INTWAIT);
418 		netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
419 			    in_pcbinswildcardhash_handler);
420 		msg->nm_inp = inp;
421 		msg->nm_pcbinfo = &tcbinfo[cpu];
422 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
423 	}
424 #else
425 	in_pcbinswildcardhash(inp);
426 #endif
427 	COMMON_END(PRU_LISTEN);
428 }
429 #endif /* INET6 */
430 
431 /*
432  * Initiate connection to peer.
433  * Create a template for use in transmissions on this connection.
434  * Enter SYN_SENT state, and mark socket as connecting.
435  * Start keep-alive timer, and seed output sequence space.
436  * Send initial segment on connection.
437  */
438 static int
439 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
440 {
441 	int error = 0;
442 	struct inpcb *inp;
443 	struct tcpcb *tp;
444 	struct sockaddr_in *sinp;
445 
446 	COMMON_START(so, inp);
447 
448 	/*
449 	 * Must disallow TCP ``connections'' to multicast addresses.
450 	 */
451 	sinp = (struct sockaddr_in *)nam;
452 	if (sinp->sin_family == AF_INET
453 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
454 		error = EAFNOSUPPORT;
455 		goto out;
456 	}
457 
458 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
459 		error = EAFNOSUPPORT; /* IPv6 only jail */
460 		goto out;
461 	}
462 
463 	if ((error = tcp_connect(tp, nam, td)) != 0)
464 		goto out;
465 	error = tcp_output(tp);
466 	COMMON_END(PRU_CONNECT);
467 }
468 
469 #ifdef INET6
470 static int
471 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
472 {
473 	int error = 0;
474 	struct inpcb *inp;
475 	struct tcpcb *tp;
476 	struct sockaddr_in6 *sin6p;
477 
478 	COMMON_START(so, inp);
479 
480 	/*
481 	 * Must disallow TCP ``connections'' to multicast addresses.
482 	 */
483 	sin6p = (struct sockaddr_in6 *)nam;
484 	if (sin6p->sin6_family == AF_INET6
485 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
486 		error = EAFNOSUPPORT;
487 		goto out;
488 	}
489 
490 	if (!prison_remote_ip(td, nam)) {
491 		error = EAFNOSUPPORT; /* IPv4 only jail */
492 		goto out;
493 	}
494 
495 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
496 		struct sockaddr_in sin;
497 
498 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
499 			error = EINVAL;
500 			goto out;
501 		}
502 
503 		in6_sin6_2_sin(&sin, sin6p);
504 		inp->inp_vflag |= INP_IPV4;
505 		inp->inp_vflag &= ~INP_IPV6;
506 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
507 			goto out;
508 		error = tcp_output(tp);
509 		goto out;
510 	}
511 	inp->inp_vflag &= ~INP_IPV4;
512 	inp->inp_vflag |= INP_IPV6;
513 	inp->inp_inc.inc_isipv6 = 1;
514 	if ((error = tcp6_connect(tp, nam, td)) != 0)
515 		goto out;
516 	error = tcp_output(tp);
517 	COMMON_END(PRU_CONNECT);
518 }
519 #endif /* INET6 */
520 
521 /*
522  * Initiate disconnect from peer.
523  * If connection never passed embryonic stage, just drop;
524  * else if don't need to let data drain, then can just drop anyways,
525  * else have to begin TCP shutdown process: mark socket disconnecting,
526  * drain unread data, state switch to reflect user close, and
527  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
528  * when peer sends FIN and acks ours.
529  *
530  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
531  */
532 static int
533 tcp_usr_disconnect(struct socket *so)
534 {
535 	int error = 0;
536 	struct inpcb *inp;
537 	struct tcpcb *tp;
538 
539 	COMMON_START(so, inp);
540 	tp = tcp_disconnect(tp);
541 	COMMON_END(PRU_DISCONNECT);
542 }
543 
544 /*
545  * Accept a connection.  Essentially all the work is
546  * done at higher levels; just return the address
547  * of the peer, storing through addr.
548  */
549 static int
550 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
551 {
552 	int error = 0;
553 	struct inpcb *inp;
554 	struct tcpcb *tp = NULL;
555 	TCPDEBUG0;
556 
557 	crit_enter();
558 	inp = so->so_pcb;
559 	if (so->so_state & SS_ISDISCONNECTED) {
560 		error = ECONNABORTED;
561 		goto out;
562 	}
563 	if (inp == 0) {
564 		crit_exit();
565 		return (EINVAL);
566 	}
567 	tp = intotcpcb(inp);
568 	TCPDEBUG1();
569 	in_setpeeraddr(so, nam);
570 	COMMON_END(PRU_ACCEPT);
571 }
572 
573 #ifdef INET6
574 static int
575 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
576 {
577 	int error = 0;
578 	struct inpcb *inp;
579 	struct tcpcb *tp = NULL;
580 	TCPDEBUG0;
581 
582 	crit_enter();
583 	inp = so->so_pcb;
584 
585 	if (so->so_state & SS_ISDISCONNECTED) {
586 		error = ECONNABORTED;
587 		goto out;
588 	}
589 	if (inp == 0) {
590 		crit_exit();
591 		return (EINVAL);
592 	}
593 	tp = intotcpcb(inp);
594 	TCPDEBUG1();
595 	in6_mapped_peeraddr(so, nam);
596 	COMMON_END(PRU_ACCEPT);
597 }
598 #endif /* INET6 */
599 /*
600  * Mark the connection as being incapable of further output.
601  */
602 static int
603 tcp_usr_shutdown(struct socket *so)
604 {
605 	int error = 0;
606 	struct inpcb *inp;
607 	struct tcpcb *tp;
608 
609 	COMMON_START(so, inp);
610 	socantsendmore(so);
611 	tp = tcp_usrclosed(tp);
612 	if (tp)
613 		error = tcp_output(tp);
614 	COMMON_END(PRU_SHUTDOWN);
615 }
616 
617 /*
618  * After a receive, possibly send window update to peer.
619  */
620 static int
621 tcp_usr_rcvd(struct socket *so, int flags)
622 {
623 	int error = 0;
624 	struct inpcb *inp;
625 	struct tcpcb *tp;
626 
627 	COMMON_START(so, inp);
628 	tcp_output(tp);
629 	COMMON_END(PRU_RCVD);
630 }
631 
632 /*
633  * Do a send by putting data in output queue and updating urgent
634  * marker if URG set.  Possibly send more data.  Unlike the other
635  * pru_*() routines, the mbuf chains are our responsibility.  We
636  * must either enqueue them or free them.  The other pru_* routines
637  * generally are caller-frees.
638  */
639 static int
640 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
641 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
642 {
643 	int error = 0;
644 	struct inpcb *inp;
645 	struct tcpcb *tp;
646 #ifdef INET6
647 	int isipv6;
648 #endif
649 	TCPDEBUG0;
650 
651 	crit_enter();
652 	inp = so->so_pcb;
653 
654 	if (inp == NULL) {
655 		/*
656 		 * OOPS! we lost a race, the TCP session got reset after
657 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
658 		 * network interrupt in the non-critical section of sosend().
659 		 */
660 		if (m)
661 			m_freem(m);
662 		if (control)
663 			m_freem(control);
664 		error = ECONNRESET;	/* XXX EPIPE? */
665 		tp = NULL;
666 		TCPDEBUG1();
667 		goto out;
668 	}
669 #ifdef INET6
670 	isipv6 = nam && nam->sa_family == AF_INET6;
671 #endif /* INET6 */
672 	tp = intotcpcb(inp);
673 	TCPDEBUG1();
674 	if (control) {
675 		/* TCP doesn't do control messages (rights, creds, etc) */
676 		if (control->m_len) {
677 			m_freem(control);
678 			if (m)
679 				m_freem(m);
680 			error = EINVAL;
681 			goto out;
682 		}
683 		m_freem(control);	/* empty control, just free it */
684 	}
685 	if(!(flags & PRUS_OOB)) {
686 		ssb_appendstream(&so->so_snd, m);
687 		if (nam && tp->t_state < TCPS_SYN_SENT) {
688 			/*
689 			 * Do implied connect if not yet connected,
690 			 * initialize window to default value, and
691 			 * initialize maxseg/maxopd using peer's cached
692 			 * MSS.
693 			 */
694 #ifdef INET6
695 			if (isipv6)
696 				error = tcp6_connect(tp, nam, td);
697 			else
698 #endif /* INET6 */
699 			error = tcp_connect(tp, nam, td);
700 			if (error)
701 				goto out;
702 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
703 			tcp_mss(tp, -1);
704 		}
705 
706 		if (flags & PRUS_EOF) {
707 			/*
708 			 * Close the send side of the connection after
709 			 * the data is sent.
710 			 */
711 			socantsendmore(so);
712 			tp = tcp_usrclosed(tp);
713 		}
714 		if (tp != NULL) {
715 			if (flags & PRUS_MORETOCOME)
716 				tp->t_flags |= TF_MORETOCOME;
717 			error = tcp_output(tp);
718 			if (flags & PRUS_MORETOCOME)
719 				tp->t_flags &= ~TF_MORETOCOME;
720 		}
721 	} else {
722 		if (ssb_space(&so->so_snd) < -512) {
723 			m_freem(m);
724 			error = ENOBUFS;
725 			goto out;
726 		}
727 		/*
728 		 * According to RFC961 (Assigned Protocols),
729 		 * the urgent pointer points to the last octet
730 		 * of urgent data.  We continue, however,
731 		 * to consider it to indicate the first octet
732 		 * of data past the urgent section.
733 		 * Otherwise, snd_up should be one lower.
734 		 */
735 		ssb_appendstream(&so->so_snd, m);
736 		if (nam && tp->t_state < TCPS_SYN_SENT) {
737 			/*
738 			 * Do implied connect if not yet connected,
739 			 * initialize window to default value, and
740 			 * initialize maxseg/maxopd using peer's cached
741 			 * MSS.
742 			 */
743 #ifdef INET6
744 			if (isipv6)
745 				error = tcp6_connect(tp, nam, td);
746 			else
747 #endif /* INET6 */
748 			error = tcp_connect(tp, nam, td);
749 			if (error)
750 				goto out;
751 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
752 			tcp_mss(tp, -1);
753 		}
754 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
755 		tp->t_flags |= TF_FORCE;
756 		error = tcp_output(tp);
757 		tp->t_flags &= ~TF_FORCE;
758 	}
759 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
760 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
761 }
762 
763 /*
764  * Abort the TCP.
765  */
766 static int
767 tcp_usr_abort(struct socket *so)
768 {
769 	int error = 0;
770 	struct inpcb *inp;
771 	struct tcpcb *tp;
772 
773 	COMMON_START(so, inp);
774 	tp = tcp_drop(tp, ECONNABORTED);
775 	COMMON_END(PRU_ABORT);
776 }
777 
778 /*
779  * Receive out-of-band data.
780  */
781 static int
782 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
783 {
784 	int error = 0;
785 	struct inpcb *inp;
786 	struct tcpcb *tp;
787 
788 	COMMON_START(so, inp);
789 	if ((so->so_oobmark == 0 &&
790 	     (so->so_state & SS_RCVATMARK) == 0) ||
791 	    so->so_options & SO_OOBINLINE ||
792 	    tp->t_oobflags & TCPOOB_HADDATA) {
793 		error = EINVAL;
794 		goto out;
795 	}
796 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
797 		error = EWOULDBLOCK;
798 		goto out;
799 	}
800 	m->m_len = 1;
801 	*mtod(m, caddr_t) = tp->t_iobc;
802 	if ((flags & MSG_PEEK) == 0)
803 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
804 	COMMON_END(PRU_RCVOOB);
805 }
806 
807 /* xxx - should be const */
808 struct pr_usrreqs tcp_usrreqs = {
809 	.pru_abort = tcp_usr_abort,
810 	.pru_accept = tcp_usr_accept,
811 	.pru_attach = tcp_usr_attach,
812 	.pru_bind = tcp_usr_bind,
813 	.pru_connect = tcp_usr_connect,
814 	.pru_connect2 = pru_connect2_notsupp,
815 	.pru_control = in_control,
816 	.pru_detach = tcp_usr_detach,
817 	.pru_disconnect = tcp_usr_disconnect,
818 	.pru_listen = tcp_usr_listen,
819 	.pru_peeraddr = in_setpeeraddr,
820 	.pru_rcvd = tcp_usr_rcvd,
821 	.pru_rcvoob = tcp_usr_rcvoob,
822 	.pru_send = tcp_usr_send,
823 	.pru_sense = pru_sense_null,
824 	.pru_shutdown = tcp_usr_shutdown,
825 	.pru_sockaddr = in_setsockaddr,
826 	.pru_sosend = sosend,
827 	.pru_soreceive = soreceive,
828 	.pru_sopoll = sopoll
829 };
830 
831 #ifdef INET6
832 struct pr_usrreqs tcp6_usrreqs = {
833 	.pru_abort = tcp_usr_abort,
834 	.pru_accept = tcp6_usr_accept,
835 	.pru_attach = tcp_usr_attach,
836 	.pru_bind = tcp6_usr_bind,
837 	.pru_connect = tcp6_usr_connect,
838 	.pru_connect2 = pru_connect2_notsupp,
839 	.pru_control = in6_control,
840 	.pru_detach = tcp_usr_detach,
841 	.pru_disconnect = tcp_usr_disconnect,
842 	.pru_listen = tcp6_usr_listen,
843 	.pru_peeraddr = in6_mapped_peeraddr,
844 	.pru_rcvd = tcp_usr_rcvd,
845 	.pru_rcvoob = tcp_usr_rcvoob,
846 	.pru_send = tcp_usr_send,
847 	.pru_sense = pru_sense_null,
848 	.pru_shutdown = tcp_usr_shutdown,
849 	.pru_sockaddr = in6_mapped_sockaddr,
850 	.pru_sosend = sosend,
851 	.pru_soreceive = soreceive,
852 	.pru_sopoll = sopoll
853 };
854 #endif /* INET6 */
855 
856 static int
857 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
858 		  struct sockaddr_in *if_sin)
859 {
860 	struct inpcb *inp = tp->t_inpcb, *oinp;
861 	struct socket *so = inp->inp_socket;
862 	struct tcpcb *otp;
863 	struct rmxp_tao *taop;
864 	struct rmxp_tao tao_noncached;
865 
866 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
867 	    sin->sin_addr, sin->sin_port,
868 	    inp->inp_laddr.s_addr != INADDR_ANY ?
869 		inp->inp_laddr : if_sin->sin_addr,
870 	    inp->inp_lport, 0, NULL);
871 	if (oinp != NULL) {
872 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
873 		    otp->t_state == TCPS_TIME_WAIT &&
874 		    (ticks - otp->t_starttime) < tcp_msl &&
875 		    (otp->t_flags & TF_RCVD_CC))
876 			tcp_close(otp);
877 		else
878 			return (EADDRINUSE);
879 	}
880 	if (inp->inp_laddr.s_addr == INADDR_ANY)
881 		inp->inp_laddr = if_sin->sin_addr;
882 	inp->inp_faddr = sin->sin_addr;
883 	inp->inp_fport = sin->sin_port;
884 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
885 	in_pcbinsconnhash(inp);
886 
887 	/* Compute window scaling to request.  */
888 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
889 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
890 		tp->request_r_scale++;
891 
892 	soisconnecting(so);
893 	tcpstat.tcps_connattempt++;
894 	tp->t_state = TCPS_SYN_SENT;
895 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
896 	tp->iss = tcp_new_isn(tp);
897 	tcp_sendseqinit(tp);
898 
899 	/*
900 	 * Generate a CC value for this connection and
901 	 * check whether CC or CCnew should be used.
902 	 */
903 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
904 		taop = &tao_noncached;
905 		bzero(taop, sizeof *taop);
906 	}
907 
908 	tp->cc_send = CC_INC(tcp_ccgen);
909 	if (taop->tao_ccsent != 0 &&
910 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
911 		taop->tao_ccsent = tp->cc_send;
912 	} else {
913 		taop->tao_ccsent = 0;
914 		tp->t_flags |= TF_SENDCCNEW;
915 	}
916 
917 	return (0);
918 }
919 
920 #ifdef SMP
921 
922 struct netmsg_tcp_connect {
923 	struct netmsg		nm_netmsg;
924 	struct tcpcb		*nm_tp;
925 	struct sockaddr_in	*nm_sin;
926 	struct sockaddr_in	*nm_ifsin;
927 };
928 
929 static void
930 tcp_connect_handler(netmsg_t netmsg)
931 {
932 	struct netmsg_tcp_connect *msg = (void *)netmsg;
933 	int error;
934 
935 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
936 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
937 }
938 
939 #endif
940 
941 /*
942  * Common subroutine to open a TCP connection to remote host specified
943  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
944  * port number if needed.  Call in_pcbladdr to do the routing and to choose
945  * a local host address (interface).  If there is an existing incarnation
946  * of the same connection in TIME-WAIT state and if the remote host was
947  * sending CC options and if the connection duration was < MSL, then
948  * truncate the previous TIME-WAIT state and proceed.
949  * Initialize connection parameters and enter SYN-SENT state.
950  */
951 static int
952 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
953 {
954 	struct inpcb *inp = tp->t_inpcb;
955 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
956 	struct sockaddr_in *if_sin;
957 	int error;
958 #ifdef SMP
959 	lwkt_port_t port;
960 #endif
961 
962 	if (inp->inp_lport == 0) {
963 		error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
964 		if (error)
965 			return (error);
966 	}
967 
968 	/*
969 	 * Cannot simply call in_pcbconnect, because there might be an
970 	 * earlier incarnation of this same connection still in
971 	 * TIME_WAIT state, creating an ADDRINUSE error.
972 	 */
973 	error = in_pcbladdr(inp, nam, &if_sin, td);
974 	if (error)
975 		return (error);
976 
977 #ifdef SMP
978 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
979 	    inp->inp_laddr.s_addr ?
980 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
981 	    inp->inp_lport);
982 
983 	if (port != &curthread->td_msgport) {
984 		struct netmsg_tcp_connect msg;
985 
986 		netmsg_init(&msg.nm_netmsg, &curthread->td_msgport, 0,
987 			    tcp_connect_handler);
988 		msg.nm_tp = tp;
989 		msg.nm_sin = sin;
990 		msg.nm_ifsin = if_sin;
991 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
992 	} else
993 #endif
994 		error = tcp_connect_oncpu(tp, sin, if_sin);
995 
996 	return (error);
997 }
998 
999 #ifdef INET6
1000 static int
1001 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1002 {
1003 	struct inpcb *inp = tp->t_inpcb, *oinp;
1004 	struct socket *so = inp->inp_socket;
1005 	struct tcpcb *otp;
1006 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1007 	struct in6_addr *addr6;
1008 	struct rmxp_tao *taop;
1009 	struct rmxp_tao tao_noncached;
1010 	int error;
1011 
1012 	if (inp->inp_lport == 0) {
1013 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
1014 		if (error)
1015 			return error;
1016 	}
1017 
1018 	/*
1019 	 * Cannot simply call in_pcbconnect, because there might be an
1020 	 * earlier incarnation of this same connection still in
1021 	 * TIME_WAIT state, creating an ADDRINUSE error.
1022 	 */
1023 	error = in6_pcbladdr(inp, nam, &addr6, td);
1024 	if (error)
1025 		return error;
1026 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1027 				  &sin6->sin6_addr, sin6->sin6_port,
1028 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1029 				      addr6 : &inp->in6p_laddr,
1030 				  inp->inp_lport,  0, NULL);
1031 	if (oinp) {
1032 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
1033 		    otp->t_state == TCPS_TIME_WAIT &&
1034 		    (ticks - otp->t_starttime) < tcp_msl &&
1035 		    (otp->t_flags & TF_RCVD_CC))
1036 			otp = tcp_close(otp);
1037 		else
1038 			return (EADDRINUSE);
1039 	}
1040 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1041 		inp->in6p_laddr = *addr6;
1042 	inp->in6p_faddr = sin6->sin6_addr;
1043 	inp->inp_fport = sin6->sin6_port;
1044 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
1045 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1046 	in_pcbinsconnhash(inp);
1047 
1048 	/* Compute window scaling to request.  */
1049 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1050 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
1051 		tp->request_r_scale++;
1052 
1053 	soisconnecting(so);
1054 	tcpstat.tcps_connattempt++;
1055 	tp->t_state = TCPS_SYN_SENT;
1056 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1057 	tp->iss = tcp_new_isn(tp);
1058 	tcp_sendseqinit(tp);
1059 
1060 	/*
1061 	 * Generate a CC value for this connection and
1062 	 * check whether CC or CCnew should be used.
1063 	 */
1064 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
1065 		taop = &tao_noncached;
1066 		bzero(taop, sizeof *taop);
1067 	}
1068 
1069 	tp->cc_send = CC_INC(tcp_ccgen);
1070 	if (taop->tao_ccsent != 0 &&
1071 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
1072 		taop->tao_ccsent = tp->cc_send;
1073 	} else {
1074 		taop->tao_ccsent = 0;
1075 		tp->t_flags |= TF_SENDCCNEW;
1076 	}
1077 
1078 	return (0);
1079 }
1080 #endif /* INET6 */
1081 
1082 /*
1083  * The new sockopt interface makes it possible for us to block in the
1084  * copyin/out step (if we take a page fault).  Taking a page fault while
1085  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1086  * both now use TSM, there probably isn't any need for this function to
1087  * run in a critical section any more.  This needs more examination.)
1088  */
1089 int
1090 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1091 {
1092 	int	error, opt, optval;
1093 	struct	inpcb *inp;
1094 	struct	tcpcb *tp;
1095 
1096 	error = 0;
1097 	crit_enter();		/* XXX */
1098 	inp = so->so_pcb;
1099 	if (inp == NULL) {
1100 		crit_exit();
1101 		return (ECONNRESET);
1102 	}
1103 	if (sopt->sopt_level != IPPROTO_TCP) {
1104 #ifdef INET6
1105 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1106 			error = ip6_ctloutput(so, sopt);
1107 		else
1108 #endif /* INET6 */
1109 		error = ip_ctloutput(so, sopt);
1110 		crit_exit();
1111 		return (error);
1112 	}
1113 	tp = intotcpcb(inp);
1114 
1115 	switch (sopt->sopt_dir) {
1116 	case SOPT_SET:
1117 		switch (sopt->sopt_name) {
1118 		case TCP_NODELAY:
1119 		case TCP_NOOPT:
1120 			error = sooptcopyin(sopt, &optval, sizeof optval,
1121 					    sizeof optval);
1122 			if (error)
1123 				break;
1124 
1125 			switch (sopt->sopt_name) {
1126 			case TCP_NODELAY:
1127 				opt = TF_NODELAY;
1128 				break;
1129 			case TCP_NOOPT:
1130 				opt = TF_NOOPT;
1131 				break;
1132 			default:
1133 				opt = 0; /* dead code to fool gcc */
1134 				break;
1135 			}
1136 
1137 			if (optval)
1138 				tp->t_flags |= opt;
1139 			else
1140 				tp->t_flags &= ~opt;
1141 			break;
1142 
1143 		case TCP_NOPUSH:
1144 			error = sooptcopyin(sopt, &optval, sizeof optval,
1145 					    sizeof optval);
1146 			if (error)
1147 				break;
1148 
1149 			if (optval)
1150 				tp->t_flags |= TF_NOPUSH;
1151 			else {
1152 				tp->t_flags &= ~TF_NOPUSH;
1153 				error = tcp_output(tp);
1154 			}
1155 			break;
1156 
1157 		case TCP_MAXSEG:
1158 			error = sooptcopyin(sopt, &optval, sizeof optval,
1159 					    sizeof optval);
1160 			if (error)
1161 				break;
1162 
1163 			if (optval > 0 && optval <= tp->t_maxseg)
1164 				tp->t_maxseg = optval;
1165 			else
1166 				error = EINVAL;
1167 			break;
1168 
1169 		default:
1170 			error = ENOPROTOOPT;
1171 			break;
1172 		}
1173 		break;
1174 
1175 	case SOPT_GET:
1176 		switch (sopt->sopt_name) {
1177 		case TCP_NODELAY:
1178 			optval = tp->t_flags & TF_NODELAY;
1179 			break;
1180 		case TCP_MAXSEG:
1181 			optval = tp->t_maxseg;
1182 			break;
1183 		case TCP_NOOPT:
1184 			optval = tp->t_flags & TF_NOOPT;
1185 			break;
1186 		case TCP_NOPUSH:
1187 			optval = tp->t_flags & TF_NOPUSH;
1188 			break;
1189 		default:
1190 			error = ENOPROTOOPT;
1191 			break;
1192 		}
1193 		if (error == 0)
1194 			error = sooptcopyout(sopt, &optval, sizeof optval);
1195 		break;
1196 	}
1197 	crit_exit();
1198 	return (error);
1199 }
1200 
1201 /*
1202  * tcp_sendspace and tcp_recvspace are the default send and receive window
1203  * sizes, respectively.  These are obsolescent (this information should
1204  * be set by the route).
1205  */
1206 u_long	tcp_sendspace = 1024*32;
1207 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1208     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1209 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1211     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1212 
1213 /*
1214  * Attach TCP protocol to socket, allocating
1215  * internet protocol control block, tcp control block,
1216  * bufer space, and entering LISTEN state if to accept connections.
1217  */
1218 static int
1219 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1220 {
1221 	struct tcpcb *tp;
1222 	struct inpcb *inp;
1223 	int error;
1224 	int cpu;
1225 #ifdef INET6
1226 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
1227 #endif
1228 
1229 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1230 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1231 				  ai->sb_rlimit);
1232 		if (error)
1233 			return (error);
1234 	}
1235 	cpu = mycpu->gd_cpuid;
1236 	error = in_pcballoc(so, &tcbinfo[cpu]);
1237 	if (error)
1238 		return (error);
1239 	inp = so->so_pcb;
1240 #ifdef INET6
1241 	if (isipv6) {
1242 		inp->inp_vflag |= INP_IPV6;
1243 		inp->in6p_hops = -1;	/* use kernel default */
1244 	}
1245 	else
1246 #endif
1247 	inp->inp_vflag |= INP_IPV4;
1248 	tp = tcp_newtcpcb(inp);
1249 	if (tp == 0) {
1250 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1251 
1252 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1253 #ifdef INET6
1254 		if (isipv6)
1255 			in6_pcbdetach(inp);
1256 		else
1257 #endif
1258 		in_pcbdetach(inp);
1259 		so->so_state |= nofd;
1260 		return (ENOBUFS);
1261 	}
1262 	tp->t_state = TCPS_CLOSED;
1263 	return (0);
1264 }
1265 
1266 /*
1267  * Initiate (or continue) disconnect.
1268  * If embryonic state, just send reset (once).
1269  * If in ``let data drain'' option and linger null, just drop.
1270  * Otherwise (hard), mark socket disconnecting and drop
1271  * current input data; switch states based on user close, and
1272  * send segment to peer (with FIN).
1273  */
1274 static struct tcpcb *
1275 tcp_disconnect(struct tcpcb *tp)
1276 {
1277 	struct socket *so = tp->t_inpcb->inp_socket;
1278 
1279 	if (tp->t_state < TCPS_ESTABLISHED)
1280 		tp = tcp_close(tp);
1281 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1282 		tp = tcp_drop(tp, 0);
1283 	else {
1284 		soisdisconnecting(so);
1285 		sbflush(&so->so_rcv.sb);
1286 		tp = tcp_usrclosed(tp);
1287 		if (tp)
1288 			tcp_output(tp);
1289 	}
1290 	return (tp);
1291 }
1292 
1293 /*
1294  * User issued close, and wish to trail through shutdown states:
1295  * if never received SYN, just forget it.  If got a SYN from peer,
1296  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1297  * If already got a FIN from peer, then almost done; go to LAST_ACK
1298  * state.  In all other cases, have already sent FIN to peer (e.g.
1299  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1300  * for peer to send FIN or not respond to keep-alives, etc.
1301  * We can let the user exit from the close as soon as the FIN is acked.
1302  */
1303 static struct tcpcb *
1304 tcp_usrclosed(struct tcpcb *tp)
1305 {
1306 
1307 	switch (tp->t_state) {
1308 
1309 	case TCPS_CLOSED:
1310 	case TCPS_LISTEN:
1311 		tp->t_state = TCPS_CLOSED;
1312 		tp = tcp_close(tp);
1313 		break;
1314 
1315 	case TCPS_SYN_SENT:
1316 	case TCPS_SYN_RECEIVED:
1317 		tp->t_flags |= TF_NEEDFIN;
1318 		break;
1319 
1320 	case TCPS_ESTABLISHED:
1321 		tp->t_state = TCPS_FIN_WAIT_1;
1322 		break;
1323 
1324 	case TCPS_CLOSE_WAIT:
1325 		tp->t_state = TCPS_LAST_ACK;
1326 		break;
1327 	}
1328 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1329 		soisdisconnected(tp->t_inpcb->inp_socket);
1330 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1331 		if (tp->t_state == TCPS_FIN_WAIT_2)
1332 			callout_reset(tp->tt_2msl, tcp_maxidle,
1333 				      tcp_timer_2msl, tp);
1334 	}
1335 	return (tp);
1336 }
1337