xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision 895c1f850848cf50a3243e134345f0d4ff33ac59)
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 1.51 2008/09/29 20:52:23 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 
207 	/*
208 	 * If the inp is already detached it may have been due to an async
209 	 * close.  Just return as if no error occured.
210 	 */
211 	if (inp == NULL) {
212 		crit_exit();
213 		return 0;
214 	}
215 
216 	/*
217 	 * It's possible for the tcpcb (tp) to disconnect from the inp due
218 	 * to tcp_drop()->tcp_close() being called.  This may occur *after*
219 	 * the detach message has been queued so we may find a NULL tp here.
220 	 */
221 	if ((tp = intotcpcb(inp)) != NULL) {
222 		TCPDEBUG1();
223 		tp = tcp_disconnect(tp);
224 		TCPDEBUG2(PRU_DETACH);
225 	}
226 	crit_exit();
227 	return error;
228 }
229 
230 /*
231  * Note: ignore_error is non-zero for certain disconnection races
232  * which we want to silently allow, otherwise close() may return
233  * an unexpected error.
234  */
235 #define	COMMON_START(so, inp, ignore_error)			\
236 	TCPDEBUG0; 		\
237 				\
238 	crit_enter();		\
239 	inp = so->so_pcb; 	\
240 	do {			\
241 		 if (inp == NULL) {				\
242 			 crit_exit();				\
243 			 return (ignore_error ? 0 : EINVAL);	\
244 		 }						\
245 		 tp = intotcpcb(inp);				\
246 		 TCPDEBUG1();					\
247 	} while(0)
248 
249 #define COMMON_END(req)	out: TCPDEBUG2(req); crit_exit(); return error; goto out
250 
251 
252 /*
253  * Give the socket an address.
254  */
255 static int
256 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
257 {
258 	int error = 0;
259 	struct inpcb *inp;
260 	struct tcpcb *tp;
261 	struct sockaddr_in *sinp;
262 
263 	COMMON_START(so, inp, 0);
264 
265 	/*
266 	 * Must check for multicast addresses and disallow binding
267 	 * to them.
268 	 */
269 	sinp = (struct sockaddr_in *)nam;
270 	if (sinp->sin_family == AF_INET &&
271 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
272 		error = EAFNOSUPPORT;
273 		goto out;
274 	}
275 	error = in_pcbbind(inp, nam, td);
276 	if (error)
277 		goto out;
278 	COMMON_END(PRU_BIND);
279 
280 }
281 
282 #ifdef INET6
283 static int
284 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
285 {
286 	int error = 0;
287 	struct inpcb *inp;
288 	struct tcpcb *tp;
289 	struct sockaddr_in6 *sin6p;
290 
291 	COMMON_START(so, inp, 0);
292 
293 	/*
294 	 * Must check for multicast addresses and disallow binding
295 	 * to them.
296 	 */
297 	sin6p = (struct sockaddr_in6 *)nam;
298 	if (sin6p->sin6_family == AF_INET6 &&
299 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
300 		error = EAFNOSUPPORT;
301 		goto out;
302 	}
303 	inp->inp_vflag &= ~INP_IPV4;
304 	inp->inp_vflag |= INP_IPV6;
305 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
306 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
307 			inp->inp_vflag |= INP_IPV4;
308 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
309 			struct sockaddr_in sin;
310 
311 			in6_sin6_2_sin(&sin, sin6p);
312 			inp->inp_vflag |= INP_IPV4;
313 			inp->inp_vflag &= ~INP_IPV6;
314 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
315 			goto out;
316 		}
317 	}
318 	error = in6_pcbbind(inp, nam, td);
319 	if (error)
320 		goto out;
321 	COMMON_END(PRU_BIND);
322 }
323 #endif /* INET6 */
324 
325 #ifdef SMP
326 struct netmsg_inswildcard {
327 	struct netmsg		nm_netmsg;
328 	struct inpcb		*nm_inp;
329 	struct inpcbinfo	*nm_pcbinfo;
330 };
331 
332 static void
333 in_pcbinswildcardhash_handler(struct netmsg *msg0)
334 {
335 	struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
336 
337 	in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
338 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
339 }
340 #endif
341 
342 /*
343  * Prepare to accept connections.
344  */
345 static int
346 tcp_usr_listen(struct socket *so, struct thread *td)
347 {
348 	int error = 0;
349 	struct inpcb *inp;
350 	struct tcpcb *tp;
351 #ifdef SMP
352 	int cpu;
353 #endif
354 
355 	COMMON_START(so, inp, 0);
356 	if (inp->inp_lport == 0) {
357 		error = in_pcbbind(inp, NULL, td);
358 		if (error != 0)
359 			goto out;
360 	}
361 
362 	tp->t_state = TCPS_LISTEN;
363 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
364 #ifdef SMP
365 	/*
366 	 * We have to set the flag because we can't have other cpus
367 	 * messing with our inp's flags.
368 	 */
369 	inp->inp_flags |= INP_WILDCARD_MP;
370 	for (cpu = 0; cpu < ncpus2; cpu++) {
371 		struct netmsg_inswildcard *msg;
372 
373 		if (cpu == mycpu->gd_cpuid) {
374 			in_pcbinswildcardhash(inp);
375 			continue;
376 		}
377 
378 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
379 			      M_INTWAIT);
380 		netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
381 			    in_pcbinswildcardhash_handler);
382 		msg->nm_inp = inp;
383 		msg->nm_pcbinfo = &tcbinfo[cpu];
384 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
385 	}
386 #else
387 	in_pcbinswildcardhash(inp);
388 #endif
389 	COMMON_END(PRU_LISTEN);
390 }
391 
392 #ifdef INET6
393 static int
394 tcp6_usr_listen(struct socket *so, struct thread *td)
395 {
396 	int error = 0;
397 	struct inpcb *inp;
398 	struct tcpcb *tp;
399 #ifdef SMP
400 	int cpu;
401 #endif
402 
403 	COMMON_START(so, inp, 0);
404 	if (inp->inp_lport == 0) {
405 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
406 			inp->inp_vflag |= INP_IPV4;
407 		else
408 			inp->inp_vflag &= ~INP_IPV4;
409 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
410 	}
411 	if (error == 0)
412 		tp->t_state = TCPS_LISTEN;
413 #ifdef SMP
414 	/*
415 	 * We have to set the flag because we can't have other cpus
416 	 * messing with our inp's flags.
417 	 */
418 	inp->inp_flags |= INP_WILDCARD_MP;
419 	for (cpu = 0; cpu < ncpus2; cpu++) {
420 		struct netmsg_inswildcard *msg;
421 
422 		if (cpu == mycpu->gd_cpuid) {
423 			in_pcbinswildcardhash(inp);
424 			continue;
425 		}
426 
427 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
428 			      M_INTWAIT);
429 		netmsg_init(&msg->nm_netmsg, &netisr_afree_rport, 0,
430 			    in_pcbinswildcardhash_handler);
431 		msg->nm_inp = inp;
432 		msg->nm_pcbinfo = &tcbinfo[cpu];
433 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
434 	}
435 #else
436 	in_pcbinswildcardhash(inp);
437 #endif
438 	COMMON_END(PRU_LISTEN);
439 }
440 #endif /* INET6 */
441 
442 #ifdef SMP
443 static void
444 tcp_output_dispatch(struct netmsg *nmsg)
445 {
446 	struct lwkt_msg *msg = &nmsg->nm_lmsg;
447 	struct tcpcb *tp = msg->u.ms_resultp;
448 	int error;
449 
450 	error = tcp_output(tp);
451 	lwkt_replymsg(msg, error);
452 }
453 #endif
454 
455 /*
456  * Initiate connection to peer.
457  * Create a template for use in transmissions on this connection.
458  * Enter SYN_SENT state, and mark socket as connecting.
459  * Start keep-alive timer, and seed output sequence space.
460  * Send initial segment on connection.
461  */
462 static int
463 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
464 {
465 	int error = 0;
466 	struct inpcb *inp;
467 	struct tcpcb *tp;
468 	struct sockaddr_in *sinp;
469 #ifdef SMP
470 	lwkt_port_t port;
471 #endif
472 
473 	COMMON_START(so, inp, 0);
474 
475 	/*
476 	 * Must disallow TCP ``connections'' to multicast addresses.
477 	 */
478 	sinp = (struct sockaddr_in *)nam;
479 	if (sinp->sin_family == AF_INET
480 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
481 		error = EAFNOSUPPORT;
482 		goto out;
483 	}
484 
485 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
486 		error = EAFNOSUPPORT; /* IPv6 only jail */
487 		goto out;
488 	}
489 
490 	if ((error = tcp_connect(tp, nam, td)) != 0)
491 		goto out;
492 
493 #ifdef SMP
494 	port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
495 			    inp->inp_laddr.s_addr, inp->inp_lport);
496 	if (port != &curthread->td_msgport) {
497 		struct netmsg nmsg;
498 		struct lwkt_msg *msg;
499 
500 		netmsg_init(&nmsg, &curthread->td_msgport, 0,
501 			    tcp_output_dispatch);
502 		msg = &nmsg.nm_lmsg;
503 		msg->u.ms_resultp = tp;
504 
505 		error = lwkt_domsg(port, msg, 0);
506 	} else
507 #endif
508 		error = tcp_output(tp);
509 	COMMON_END(PRU_CONNECT);
510 }
511 
512 #ifdef INET6
513 static int
514 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
515 {
516 	int error = 0;
517 	struct inpcb *inp;
518 	struct tcpcb *tp;
519 	struct sockaddr_in6 *sin6p;
520 
521 	COMMON_START(so, inp, 0);
522 
523 	/*
524 	 * Must disallow TCP ``connections'' to multicast addresses.
525 	 */
526 	sin6p = (struct sockaddr_in6 *)nam;
527 	if (sin6p->sin6_family == AF_INET6
528 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
529 		error = EAFNOSUPPORT;
530 		goto out;
531 	}
532 
533 	if (!prison_remote_ip(td, nam)) {
534 		error = EAFNOSUPPORT; /* IPv4 only jail */
535 		goto out;
536 	}
537 
538 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
539 		struct sockaddr_in sin;
540 
541 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
542 			error = EINVAL;
543 			goto out;
544 		}
545 
546 		in6_sin6_2_sin(&sin, sin6p);
547 		inp->inp_vflag |= INP_IPV4;
548 		inp->inp_vflag &= ~INP_IPV6;
549 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
550 			goto out;
551 		error = tcp_output(tp);
552 		goto out;
553 	}
554 	inp->inp_vflag &= ~INP_IPV4;
555 	inp->inp_vflag |= INP_IPV6;
556 	inp->inp_inc.inc_isipv6 = 1;
557 	if ((error = tcp6_connect(tp, nam, td)) != 0)
558 		goto out;
559 	error = tcp_output(tp);
560 	COMMON_END(PRU_CONNECT);
561 }
562 #endif /* INET6 */
563 
564 /*
565  * Initiate disconnect from peer.
566  * If connection never passed embryonic stage, just drop;
567  * else if don't need to let data drain, then can just drop anyways,
568  * else have to begin TCP shutdown process: mark socket disconnecting,
569  * drain unread data, state switch to reflect user close, and
570  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
571  * when peer sends FIN and acks ours.
572  *
573  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
574  */
575 static int
576 tcp_usr_disconnect(struct socket *so)
577 {
578 	int error = 0;
579 	struct inpcb *inp;
580 	struct tcpcb *tp;
581 
582 	COMMON_START(so, inp, 1);
583 	tp = tcp_disconnect(tp);
584 	COMMON_END(PRU_DISCONNECT);
585 }
586 
587 /*
588  * Accept a connection.  Essentially all the work is
589  * done at higher levels; just return the address
590  * of the peer, storing through addr.
591  */
592 static int
593 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
594 {
595 	int error = 0;
596 	struct inpcb *inp;
597 	struct tcpcb *tp = NULL;
598 	TCPDEBUG0;
599 
600 	crit_enter();
601 	inp = so->so_pcb;
602 	if (so->so_state & SS_ISDISCONNECTED) {
603 		error = ECONNABORTED;
604 		goto out;
605 	}
606 	if (inp == 0) {
607 		crit_exit();
608 		return (EINVAL);
609 	}
610 	tp = intotcpcb(inp);
611 	TCPDEBUG1();
612 	in_setpeeraddr(so, nam);
613 	COMMON_END(PRU_ACCEPT);
614 }
615 
616 #ifdef INET6
617 static int
618 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
619 {
620 	int error = 0;
621 	struct inpcb *inp;
622 	struct tcpcb *tp = NULL;
623 	TCPDEBUG0;
624 
625 	crit_enter();
626 	inp = so->so_pcb;
627 
628 	if (so->so_state & SS_ISDISCONNECTED) {
629 		error = ECONNABORTED;
630 		goto out;
631 	}
632 	if (inp == 0) {
633 		crit_exit();
634 		return (EINVAL);
635 	}
636 	tp = intotcpcb(inp);
637 	TCPDEBUG1();
638 	in6_mapped_peeraddr(so, nam);
639 	COMMON_END(PRU_ACCEPT);
640 }
641 #endif /* INET6 */
642 /*
643  * Mark the connection as being incapable of further output.
644  */
645 static int
646 tcp_usr_shutdown(struct socket *so)
647 {
648 	int error = 0;
649 	struct inpcb *inp;
650 	struct tcpcb *tp;
651 
652 	COMMON_START(so, inp, 0);
653 	socantsendmore(so);
654 	tp = tcp_usrclosed(tp);
655 	if (tp)
656 		error = tcp_output(tp);
657 	COMMON_END(PRU_SHUTDOWN);
658 }
659 
660 /*
661  * After a receive, possibly send window update to peer.
662  */
663 static int
664 tcp_usr_rcvd(struct socket *so, int flags)
665 {
666 	int error = 0;
667 	struct inpcb *inp;
668 	struct tcpcb *tp;
669 
670 	COMMON_START(so, inp, 0);
671 	tcp_output(tp);
672 	COMMON_END(PRU_RCVD);
673 }
674 
675 /*
676  * Do a send by putting data in output queue and updating urgent
677  * marker if URG set.  Possibly send more data.  Unlike the other
678  * pru_*() routines, the mbuf chains are our responsibility.  We
679  * must either enqueue them or free them.  The other pru_* routines
680  * generally are caller-frees.
681  */
682 static int
683 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
684 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
685 {
686 	int error = 0;
687 	struct inpcb *inp;
688 	struct tcpcb *tp;
689 #ifdef INET6
690 	int isipv6;
691 #endif
692 	TCPDEBUG0;
693 
694 	crit_enter();
695 	inp = so->so_pcb;
696 
697 	if (inp == NULL) {
698 		/*
699 		 * OOPS! we lost a race, the TCP session got reset after
700 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
701 		 * network interrupt in the non-critical section of sosend().
702 		 */
703 		if (m)
704 			m_freem(m);
705 		if (control)
706 			m_freem(control);
707 		error = ECONNRESET;	/* XXX EPIPE? */
708 		tp = NULL;
709 		TCPDEBUG1();
710 		goto out;
711 	}
712 #ifdef INET6
713 	isipv6 = nam && nam->sa_family == AF_INET6;
714 #endif /* INET6 */
715 	tp = intotcpcb(inp);
716 	TCPDEBUG1();
717 	if (control) {
718 		/* TCP doesn't do control messages (rights, creds, etc) */
719 		if (control->m_len) {
720 			m_freem(control);
721 			if (m)
722 				m_freem(m);
723 			error = EINVAL;
724 			goto out;
725 		}
726 		m_freem(control);	/* empty control, just free it */
727 	}
728 	if(!(flags & PRUS_OOB)) {
729 		ssb_appendstream(&so->so_snd, m);
730 		if (nam && tp->t_state < TCPS_SYN_SENT) {
731 			/*
732 			 * Do implied connect if not yet connected,
733 			 * initialize window to default value, and
734 			 * initialize maxseg/maxopd using peer's cached
735 			 * MSS.
736 			 */
737 #ifdef INET6
738 			if (isipv6)
739 				error = tcp6_connect(tp, nam, td);
740 			else
741 #endif /* INET6 */
742 			error = tcp_connect(tp, nam, td);
743 			if (error)
744 				goto out;
745 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
746 			tcp_mss(tp, -1);
747 		}
748 
749 		if (flags & PRUS_EOF) {
750 			/*
751 			 * Close the send side of the connection after
752 			 * the data is sent.
753 			 */
754 			socantsendmore(so);
755 			tp = tcp_usrclosed(tp);
756 		}
757 		if (tp != NULL) {
758 			if (flags & PRUS_MORETOCOME)
759 				tp->t_flags |= TF_MORETOCOME;
760 			error = tcp_output(tp);
761 			if (flags & PRUS_MORETOCOME)
762 				tp->t_flags &= ~TF_MORETOCOME;
763 		}
764 	} else {
765 		if (ssb_space(&so->so_snd) < -512) {
766 			m_freem(m);
767 			error = ENOBUFS;
768 			goto out;
769 		}
770 		/*
771 		 * According to RFC961 (Assigned Protocols),
772 		 * the urgent pointer points to the last octet
773 		 * of urgent data.  We continue, however,
774 		 * to consider it to indicate the first octet
775 		 * of data past the urgent section.
776 		 * Otherwise, snd_up should be one lower.
777 		 */
778 		ssb_appendstream(&so->so_snd, m);
779 		if (nam && tp->t_state < TCPS_SYN_SENT) {
780 			/*
781 			 * Do implied connect if not yet connected,
782 			 * initialize window to default value, and
783 			 * initialize maxseg/maxopd using peer's cached
784 			 * MSS.
785 			 */
786 #ifdef INET6
787 			if (isipv6)
788 				error = tcp6_connect(tp, nam, td);
789 			else
790 #endif /* INET6 */
791 			error = tcp_connect(tp, nam, td);
792 			if (error)
793 				goto out;
794 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
795 			tcp_mss(tp, -1);
796 		}
797 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
798 		tp->t_flags |= TF_FORCE;
799 		error = tcp_output(tp);
800 		tp->t_flags &= ~TF_FORCE;
801 	}
802 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
803 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
804 }
805 
806 /*
807  * Abort the TCP.
808  */
809 static int
810 tcp_usr_abort(struct socket *so)
811 {
812 	int error = 0;
813 	struct inpcb *inp;
814 	struct tcpcb *tp;
815 
816 	COMMON_START(so, inp, 1);
817 	tp = tcp_drop(tp, ECONNABORTED);
818 	COMMON_END(PRU_ABORT);
819 }
820 
821 /*
822  * Receive out-of-band data.
823  */
824 static int
825 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
826 {
827 	int error = 0;
828 	struct inpcb *inp;
829 	struct tcpcb *tp;
830 
831 	COMMON_START(so, inp, 0);
832 	if ((so->so_oobmark == 0 &&
833 	     (so->so_state & SS_RCVATMARK) == 0) ||
834 	    so->so_options & SO_OOBINLINE ||
835 	    tp->t_oobflags & TCPOOB_HADDATA) {
836 		error = EINVAL;
837 		goto out;
838 	}
839 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
840 		error = EWOULDBLOCK;
841 		goto out;
842 	}
843 	m->m_len = 1;
844 	*mtod(m, caddr_t) = tp->t_iobc;
845 	if ((flags & MSG_PEEK) == 0)
846 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
847 	COMMON_END(PRU_RCVOOB);
848 }
849 
850 /* xxx - should be const */
851 struct pr_usrreqs tcp_usrreqs = {
852 	.pru_abort = tcp_usr_abort,
853 	.pru_accept = tcp_usr_accept,
854 	.pru_attach = tcp_usr_attach,
855 	.pru_bind = tcp_usr_bind,
856 	.pru_connect = tcp_usr_connect,
857 	.pru_connect2 = pru_connect2_notsupp,
858 	.pru_control = in_control,
859 	.pru_detach = tcp_usr_detach,
860 	.pru_disconnect = tcp_usr_disconnect,
861 	.pru_listen = tcp_usr_listen,
862 	.pru_peeraddr = in_setpeeraddr,
863 	.pru_rcvd = tcp_usr_rcvd,
864 	.pru_rcvoob = tcp_usr_rcvoob,
865 	.pru_send = tcp_usr_send,
866 	.pru_sense = pru_sense_null,
867 	.pru_shutdown = tcp_usr_shutdown,
868 	.pru_sockaddr = in_setsockaddr,
869 	.pru_sosend = sosend,
870 	.pru_soreceive = soreceive,
871 	.pru_sopoll = sopoll
872 };
873 
874 #ifdef INET6
875 struct pr_usrreqs tcp6_usrreqs = {
876 	.pru_abort = tcp_usr_abort,
877 	.pru_accept = tcp6_usr_accept,
878 	.pru_attach = tcp_usr_attach,
879 	.pru_bind = tcp6_usr_bind,
880 	.pru_connect = tcp6_usr_connect,
881 	.pru_connect2 = pru_connect2_notsupp,
882 	.pru_control = in6_control,
883 	.pru_detach = tcp_usr_detach,
884 	.pru_disconnect = tcp_usr_disconnect,
885 	.pru_listen = tcp6_usr_listen,
886 	.pru_peeraddr = in6_mapped_peeraddr,
887 	.pru_rcvd = tcp_usr_rcvd,
888 	.pru_rcvoob = tcp_usr_rcvoob,
889 	.pru_send = tcp_usr_send,
890 	.pru_sense = pru_sense_null,
891 	.pru_shutdown = tcp_usr_shutdown,
892 	.pru_sockaddr = in6_mapped_sockaddr,
893 	.pru_sosend = sosend,
894 	.pru_soreceive = soreceive,
895 	.pru_sopoll = sopoll
896 };
897 #endif /* INET6 */
898 
899 static int
900 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
901 		  struct sockaddr_in *if_sin)
902 {
903 	struct inpcb *inp = tp->t_inpcb, *oinp;
904 	struct socket *so = inp->inp_socket;
905 	struct tcpcb *otp;
906 	struct rmxp_tao *taop;
907 	struct rmxp_tao tao_noncached;
908 
909 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
910 	    sin->sin_addr, sin->sin_port,
911 	    inp->inp_laddr.s_addr != INADDR_ANY ?
912 		inp->inp_laddr : if_sin->sin_addr,
913 	    inp->inp_lport, 0, NULL);
914 	if (oinp != NULL) {
915 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
916 		    otp->t_state == TCPS_TIME_WAIT &&
917 		    (ticks - otp->t_starttime) < tcp_msl &&
918 		    (otp->t_flags & TF_RCVD_CC))
919 			tcp_close(otp);
920 		else
921 			return (EADDRINUSE);
922 	}
923 	if (inp->inp_laddr.s_addr == INADDR_ANY)
924 		inp->inp_laddr = if_sin->sin_addr;
925 	inp->inp_faddr = sin->sin_addr;
926 	inp->inp_fport = sin->sin_port;
927 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
928 	in_pcbinsconnhash(inp);
929 
930 	tcp_create_timermsg(tp);
931 
932 	/* Compute window scaling to request.  */
933 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
934 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
935 		tp->request_r_scale++;
936 
937 	soisconnecting(so);
938 	tcpstat.tcps_connattempt++;
939 	tp->t_state = TCPS_SYN_SENT;
940 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
941 	tp->iss = tcp_new_isn(tp);
942 	tcp_sendseqinit(tp);
943 
944 	/*
945 	 * Generate a CC value for this connection and
946 	 * check whether CC or CCnew should be used.
947 	 */
948 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
949 		taop = &tao_noncached;
950 		bzero(taop, sizeof *taop);
951 	}
952 
953 	tp->cc_send = CC_INC(tcp_ccgen);
954 	if (taop->tao_ccsent != 0 &&
955 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
956 		taop->tao_ccsent = tp->cc_send;
957 	} else {
958 		taop->tao_ccsent = 0;
959 		tp->t_flags |= TF_SENDCCNEW;
960 	}
961 
962 	return (0);
963 }
964 
965 #ifdef SMP
966 
967 struct netmsg_tcp_connect {
968 	struct netmsg		nm_netmsg;
969 	struct tcpcb		*nm_tp;
970 	struct sockaddr_in	*nm_sin;
971 	struct sockaddr_in	*nm_ifsin;
972 };
973 
974 static void
975 tcp_connect_handler(netmsg_t netmsg)
976 {
977 	struct netmsg_tcp_connect *msg = (void *)netmsg;
978 	int error;
979 
980 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
981 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
982 }
983 
984 #endif
985 
986 /*
987  * Common subroutine to open a TCP connection to remote host specified
988  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
989  * port number if needed.  Call in_pcbladdr to do the routing and to choose
990  * a local host address (interface).  If there is an existing incarnation
991  * of the same connection in TIME-WAIT state and if the remote host was
992  * sending CC options and if the connection duration was < MSL, then
993  * truncate the previous TIME-WAIT state and proceed.
994  * Initialize connection parameters and enter SYN-SENT state.
995  */
996 static int
997 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
998 {
999 	struct inpcb *inp = tp->t_inpcb;
1000 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1001 	struct sockaddr_in *if_sin;
1002 	int error;
1003 #ifdef SMP
1004 	lwkt_port_t port;
1005 #endif
1006 
1007 	if (inp->inp_lport == 0) {
1008 		error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
1009 		if (error)
1010 			return (error);
1011 	}
1012 
1013 	/*
1014 	 * Cannot simply call in_pcbconnect, because there might be an
1015 	 * earlier incarnation of this same connection still in
1016 	 * TIME_WAIT state, creating an ADDRINUSE error.
1017 	 */
1018 	error = in_pcbladdr(inp, nam, &if_sin, td);
1019 	if (error)
1020 		return (error);
1021 
1022 #ifdef SMP
1023 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1024 	    inp->inp_laddr.s_addr ?
1025 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
1026 	    inp->inp_lport);
1027 
1028 	if (port != &curthread->td_msgport) {
1029 		struct netmsg_tcp_connect msg;
1030 		struct route *ro = &inp->inp_route;
1031 
1032 		/*
1033 		 * in_pcbladdr() may have allocated a route entry for us
1034 		 * on the current CPU, but we need a route entry on the
1035 		 * target CPU, so free it here.
1036 		 */
1037 		if (ro->ro_rt != NULL)
1038 			RTFREE(ro->ro_rt);
1039 		bzero(ro, sizeof(*ro));
1040 
1041 		netmsg_init(&msg.nm_netmsg, &curthread->td_msgport, 0,
1042 			    tcp_connect_handler);
1043 		msg.nm_tp = tp;
1044 		msg.nm_sin = sin;
1045 		msg.nm_ifsin = if_sin;
1046 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1047 	} else
1048 #endif
1049 		error = tcp_connect_oncpu(tp, sin, if_sin);
1050 
1051 	return (error);
1052 }
1053 
1054 #ifdef INET6
1055 static int
1056 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1057 {
1058 	struct inpcb *inp = tp->t_inpcb, *oinp;
1059 	struct socket *so = inp->inp_socket;
1060 	struct tcpcb *otp;
1061 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1062 	struct in6_addr *addr6;
1063 	struct rmxp_tao *taop;
1064 	struct rmxp_tao tao_noncached;
1065 	int error;
1066 
1067 	if (inp->inp_lport == 0) {
1068 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
1069 		if (error)
1070 			return error;
1071 	}
1072 
1073 	/*
1074 	 * Cannot simply call in_pcbconnect, because there might be an
1075 	 * earlier incarnation of this same connection still in
1076 	 * TIME_WAIT state, creating an ADDRINUSE error.
1077 	 */
1078 	error = in6_pcbladdr(inp, nam, &addr6, td);
1079 	if (error)
1080 		return error;
1081 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1082 				  &sin6->sin6_addr, sin6->sin6_port,
1083 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1084 				      addr6 : &inp->in6p_laddr,
1085 				  inp->inp_lport,  0, NULL);
1086 	if (oinp) {
1087 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
1088 		    otp->t_state == TCPS_TIME_WAIT &&
1089 		    (ticks - otp->t_starttime) < tcp_msl &&
1090 		    (otp->t_flags & TF_RCVD_CC))
1091 			otp = tcp_close(otp);
1092 		else
1093 			return (EADDRINUSE);
1094 	}
1095 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1096 		inp->in6p_laddr = *addr6;
1097 	inp->in6p_faddr = sin6->sin6_addr;
1098 	inp->inp_fport = sin6->sin6_port;
1099 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1100 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1101 	in_pcbinsconnhash(inp);
1102 
1103 	/* Compute window scaling to request.  */
1104 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1105 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
1106 		tp->request_r_scale++;
1107 
1108 	soisconnecting(so);
1109 	tcpstat.tcps_connattempt++;
1110 	tp->t_state = TCPS_SYN_SENT;
1111 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1112 	tp->iss = tcp_new_isn(tp);
1113 	tcp_sendseqinit(tp);
1114 
1115 	/*
1116 	 * Generate a CC value for this connection and
1117 	 * check whether CC or CCnew should be used.
1118 	 */
1119 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
1120 		taop = &tao_noncached;
1121 		bzero(taop, sizeof *taop);
1122 	}
1123 
1124 	tp->cc_send = CC_INC(tcp_ccgen);
1125 	if (taop->tao_ccsent != 0 &&
1126 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
1127 		taop->tao_ccsent = tp->cc_send;
1128 	} else {
1129 		taop->tao_ccsent = 0;
1130 		tp->t_flags |= TF_SENDCCNEW;
1131 	}
1132 
1133 	return (0);
1134 }
1135 #endif /* INET6 */
1136 
1137 /*
1138  * The new sockopt interface makes it possible for us to block in the
1139  * copyin/out step (if we take a page fault).  Taking a page fault while
1140  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1141  * both now use TSM, there probably isn't any need for this function to
1142  * run in a critical section any more.  This needs more examination.)
1143  */
1144 int
1145 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1146 {
1147 	int	error, opt, optval;
1148 	struct	inpcb *inp;
1149 	struct	tcpcb *tp;
1150 
1151 	error = 0;
1152 	crit_enter();		/* XXX */
1153 	inp = so->so_pcb;
1154 	if (inp == NULL) {
1155 		crit_exit();
1156 		return (ECONNRESET);
1157 	}
1158 	if (sopt->sopt_level != IPPROTO_TCP) {
1159 #ifdef INET6
1160 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1161 			error = ip6_ctloutput(so, sopt);
1162 		else
1163 #endif /* INET6 */
1164 		error = ip_ctloutput(so, sopt);
1165 		crit_exit();
1166 		return (error);
1167 	}
1168 	tp = intotcpcb(inp);
1169 
1170 	switch (sopt->sopt_dir) {
1171 	case SOPT_SET:
1172 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1173 				      sizeof optval);
1174 		if (error)
1175 			break;
1176 		switch (sopt->sopt_name) {
1177 		case TCP_NODELAY:
1178 		case TCP_NOOPT:
1179 			switch (sopt->sopt_name) {
1180 			case TCP_NODELAY:
1181 				opt = TF_NODELAY;
1182 				break;
1183 			case TCP_NOOPT:
1184 				opt = TF_NOOPT;
1185 				break;
1186 			default:
1187 				opt = 0; /* dead code to fool gcc */
1188 				break;
1189 			}
1190 
1191 			if (optval)
1192 				tp->t_flags |= opt;
1193 			else
1194 				tp->t_flags &= ~opt;
1195 			break;
1196 
1197 		case TCP_NOPUSH:
1198 			if (optval)
1199 				tp->t_flags |= TF_NOPUSH;
1200 			else {
1201 				tp->t_flags &= ~TF_NOPUSH;
1202 				error = tcp_output(tp);
1203 			}
1204 			break;
1205 
1206 		case TCP_MAXSEG:
1207 			if (optval > 0 && optval <= tp->t_maxseg)
1208 				tp->t_maxseg = optval;
1209 			else
1210 				error = EINVAL;
1211 			break;
1212 
1213 		default:
1214 			error = ENOPROTOOPT;
1215 			break;
1216 		}
1217 		break;
1218 
1219 	case SOPT_GET:
1220 		switch (sopt->sopt_name) {
1221 		case TCP_NODELAY:
1222 			optval = tp->t_flags & TF_NODELAY;
1223 			break;
1224 		case TCP_MAXSEG:
1225 			optval = tp->t_maxseg;
1226 			break;
1227 		case TCP_NOOPT:
1228 			optval = tp->t_flags & TF_NOOPT;
1229 			break;
1230 		case TCP_NOPUSH:
1231 			optval = tp->t_flags & TF_NOPUSH;
1232 			break;
1233 		default:
1234 			error = ENOPROTOOPT;
1235 			break;
1236 		}
1237 		if (error == 0)
1238 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1239 		break;
1240 	}
1241 	crit_exit();
1242 	return (error);
1243 }
1244 
1245 /*
1246  * tcp_sendspace and tcp_recvspace are the default send and receive window
1247  * sizes, respectively.  These are obsolescent (this information should
1248  * be set by the route).
1249  *
1250  * Use a default that does not require tcp window scaling to be turned
1251  * on.  Individual programs or the administrator can increase the default.
1252  */
1253 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1254 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1255     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1256 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1257 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1258     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1259 
1260 /*
1261  * Attach TCP protocol to socket, allocating
1262  * internet protocol control block, tcp control block,
1263  * bufer space, and entering LISTEN state if to accept connections.
1264  */
1265 static int
1266 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1267 {
1268 	struct tcpcb *tp;
1269 	struct inpcb *inp;
1270 	int error;
1271 	int cpu;
1272 #ifdef INET6
1273 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1274 #endif
1275 
1276 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1277 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1278 				  ai->sb_rlimit);
1279 		if (error)
1280 			return (error);
1281 	}
1282 	cpu = mycpu->gd_cpuid;
1283 	error = in_pcballoc(so, &tcbinfo[cpu]);
1284 	if (error)
1285 		return (error);
1286 	inp = so->so_pcb;
1287 #ifdef INET6
1288 	if (isipv6) {
1289 		inp->inp_vflag |= INP_IPV6;
1290 		inp->in6p_hops = -1;	/* use kernel default */
1291 	}
1292 	else
1293 #endif
1294 	inp->inp_vflag |= INP_IPV4;
1295 	tp = tcp_newtcpcb(inp);
1296 	if (tp == 0) {
1297 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1298 
1299 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1300 #ifdef INET6
1301 		if (isipv6)
1302 			in6_pcbdetach(inp);
1303 		else
1304 #endif
1305 		in_pcbdetach(inp);
1306 		so->so_state |= nofd;
1307 		return (ENOBUFS);
1308 	}
1309 	tp->t_state = TCPS_CLOSED;
1310 	return (0);
1311 }
1312 
1313 /*
1314  * Initiate (or continue) disconnect.
1315  * If embryonic state, just send reset (once).
1316  * If in ``let data drain'' option and linger null, just drop.
1317  * Otherwise (hard), mark socket disconnecting and drop
1318  * current input data; switch states based on user close, and
1319  * send segment to peer (with FIN).
1320  */
1321 static struct tcpcb *
1322 tcp_disconnect(struct tcpcb *tp)
1323 {
1324 	struct socket *so = tp->t_inpcb->inp_socket;
1325 
1326 	if (tp->t_state < TCPS_ESTABLISHED)
1327 		tp = tcp_close(tp);
1328 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1329 		tp = tcp_drop(tp, 0);
1330 	else {
1331 		soisdisconnecting(so);
1332 		sbflush(&so->so_rcv.sb);
1333 		tp = tcp_usrclosed(tp);
1334 		if (tp)
1335 			tcp_output(tp);
1336 	}
1337 	return (tp);
1338 }
1339 
1340 /*
1341  * User issued close, and wish to trail through shutdown states:
1342  * if never received SYN, just forget it.  If got a SYN from peer,
1343  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1344  * If already got a FIN from peer, then almost done; go to LAST_ACK
1345  * state.  In all other cases, have already sent FIN to peer (e.g.
1346  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1347  * for peer to send FIN or not respond to keep-alives, etc.
1348  * We can let the user exit from the close as soon as the FIN is acked.
1349  */
1350 static struct tcpcb *
1351 tcp_usrclosed(struct tcpcb *tp)
1352 {
1353 
1354 	switch (tp->t_state) {
1355 
1356 	case TCPS_CLOSED:
1357 	case TCPS_LISTEN:
1358 		tp->t_state = TCPS_CLOSED;
1359 		tp = tcp_close(tp);
1360 		break;
1361 
1362 	case TCPS_SYN_SENT:
1363 	case TCPS_SYN_RECEIVED:
1364 		tp->t_flags |= TF_NEEDFIN;
1365 		break;
1366 
1367 	case TCPS_ESTABLISHED:
1368 		tp->t_state = TCPS_FIN_WAIT_1;
1369 		break;
1370 
1371 	case TCPS_CLOSE_WAIT:
1372 		tp->t_state = TCPS_LAST_ACK;
1373 		break;
1374 	}
1375 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1376 		soisdisconnected(tp->t_inpcb->inp_socket);
1377 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1378 		if (tp->t_state == TCPS_FIN_WAIT_2)
1379 			callout_reset(tp->tt_2msl, tcp_maxidle,
1380 				      tcp_timer_2msl, tp);
1381 	}
1382 	return (tp);
1383 }
1384