xref: /dragonfly/sys/netinet/tcp_usrreq.c (revision 3641b7ca)
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 1.48 2008/06/05 18:06:32 swildner 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 #ifdef SMP
432 static void
433 tcp_output_dispatch(struct netmsg *nmsg)
434 {
435 	struct lwkt_msg *msg = &nmsg->nm_lmsg;
436 	struct tcpcb *tp = msg->u.ms_resultp;
437 	int error;
438 
439 	error = tcp_output(tp);
440 	lwkt_replymsg(msg, error);
441 }
442 #endif
443 
444 /*
445  * Initiate connection to peer.
446  * Create a template for use in transmissions on this connection.
447  * Enter SYN_SENT state, and mark socket as connecting.
448  * Start keep-alive timer, and seed output sequence space.
449  * Send initial segment on connection.
450  */
451 static int
452 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
453 {
454 	int error = 0;
455 	struct inpcb *inp;
456 	struct tcpcb *tp;
457 	struct sockaddr_in *sinp;
458 #ifdef SMP
459 	lwkt_port_t port;
460 #endif
461 
462 	COMMON_START(so, inp);
463 
464 	/*
465 	 * Must disallow TCP ``connections'' to multicast addresses.
466 	 */
467 	sinp = (struct sockaddr_in *)nam;
468 	if (sinp->sin_family == AF_INET
469 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
470 		error = EAFNOSUPPORT;
471 		goto out;
472 	}
473 
474 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
475 		error = EAFNOSUPPORT; /* IPv6 only jail */
476 		goto out;
477 	}
478 
479 	if ((error = tcp_connect(tp, nam, td)) != 0)
480 		goto out;
481 
482 #ifdef SMP
483 	port = tcp_addrport(inp->inp_faddr.s_addr, inp->inp_fport,
484 			    inp->inp_laddr.s_addr, inp->inp_lport);
485 	if (port != &curthread->td_msgport) {
486 		struct netmsg nmsg;
487 		struct lwkt_msg *msg;
488 
489 		netmsg_init(&nmsg, &curthread->td_msgport, 0,
490 			    tcp_output_dispatch);
491 		msg = &nmsg.nm_lmsg;
492 		msg->u.ms_resultp = tp;
493 
494 		error = lwkt_domsg(port, msg, 0);
495 	} else
496 #endif
497 		error = tcp_output(tp);
498 	COMMON_END(PRU_CONNECT);
499 }
500 
501 #ifdef INET6
502 static int
503 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
504 {
505 	int error = 0;
506 	struct inpcb *inp;
507 	struct tcpcb *tp;
508 	struct sockaddr_in6 *sin6p;
509 
510 	COMMON_START(so, inp);
511 
512 	/*
513 	 * Must disallow TCP ``connections'' to multicast addresses.
514 	 */
515 	sin6p = (struct sockaddr_in6 *)nam;
516 	if (sin6p->sin6_family == AF_INET6
517 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
518 		error = EAFNOSUPPORT;
519 		goto out;
520 	}
521 
522 	if (!prison_remote_ip(td, nam)) {
523 		error = EAFNOSUPPORT; /* IPv4 only jail */
524 		goto out;
525 	}
526 
527 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
528 		struct sockaddr_in sin;
529 
530 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
531 			error = EINVAL;
532 			goto out;
533 		}
534 
535 		in6_sin6_2_sin(&sin, sin6p);
536 		inp->inp_vflag |= INP_IPV4;
537 		inp->inp_vflag &= ~INP_IPV6;
538 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
539 			goto out;
540 		error = tcp_output(tp);
541 		goto out;
542 	}
543 	inp->inp_vflag &= ~INP_IPV4;
544 	inp->inp_vflag |= INP_IPV6;
545 	inp->inp_inc.inc_isipv6 = 1;
546 	if ((error = tcp6_connect(tp, nam, td)) != 0)
547 		goto out;
548 	error = tcp_output(tp);
549 	COMMON_END(PRU_CONNECT);
550 }
551 #endif /* INET6 */
552 
553 /*
554  * Initiate disconnect from peer.
555  * If connection never passed embryonic stage, just drop;
556  * else if don't need to let data drain, then can just drop anyways,
557  * else have to begin TCP shutdown process: mark socket disconnecting,
558  * drain unread data, state switch to reflect user close, and
559  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
560  * when peer sends FIN and acks ours.
561  *
562  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
563  */
564 static int
565 tcp_usr_disconnect(struct socket *so)
566 {
567 	int error = 0;
568 	struct inpcb *inp;
569 	struct tcpcb *tp;
570 
571 	COMMON_START(so, inp);
572 	tp = tcp_disconnect(tp);
573 	COMMON_END(PRU_DISCONNECT);
574 }
575 
576 /*
577  * Accept a connection.  Essentially all the work is
578  * done at higher levels; just return the address
579  * of the peer, storing through addr.
580  */
581 static int
582 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
583 {
584 	int error = 0;
585 	struct inpcb *inp;
586 	struct tcpcb *tp = NULL;
587 	TCPDEBUG0;
588 
589 	crit_enter();
590 	inp = so->so_pcb;
591 	if (so->so_state & SS_ISDISCONNECTED) {
592 		error = ECONNABORTED;
593 		goto out;
594 	}
595 	if (inp == 0) {
596 		crit_exit();
597 		return (EINVAL);
598 	}
599 	tp = intotcpcb(inp);
600 	TCPDEBUG1();
601 	in_setpeeraddr(so, nam);
602 	COMMON_END(PRU_ACCEPT);
603 }
604 
605 #ifdef INET6
606 static int
607 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
608 {
609 	int error = 0;
610 	struct inpcb *inp;
611 	struct tcpcb *tp = NULL;
612 	TCPDEBUG0;
613 
614 	crit_enter();
615 	inp = so->so_pcb;
616 
617 	if (so->so_state & SS_ISDISCONNECTED) {
618 		error = ECONNABORTED;
619 		goto out;
620 	}
621 	if (inp == 0) {
622 		crit_exit();
623 		return (EINVAL);
624 	}
625 	tp = intotcpcb(inp);
626 	TCPDEBUG1();
627 	in6_mapped_peeraddr(so, nam);
628 	COMMON_END(PRU_ACCEPT);
629 }
630 #endif /* INET6 */
631 /*
632  * Mark the connection as being incapable of further output.
633  */
634 static int
635 tcp_usr_shutdown(struct socket *so)
636 {
637 	int error = 0;
638 	struct inpcb *inp;
639 	struct tcpcb *tp;
640 
641 	COMMON_START(so, inp);
642 	socantsendmore(so);
643 	tp = tcp_usrclosed(tp);
644 	if (tp)
645 		error = tcp_output(tp);
646 	COMMON_END(PRU_SHUTDOWN);
647 }
648 
649 /*
650  * After a receive, possibly send window update to peer.
651  */
652 static int
653 tcp_usr_rcvd(struct socket *so, int flags)
654 {
655 	int error = 0;
656 	struct inpcb *inp;
657 	struct tcpcb *tp;
658 
659 	COMMON_START(so, inp);
660 	tcp_output(tp);
661 	COMMON_END(PRU_RCVD);
662 }
663 
664 /*
665  * Do a send by putting data in output queue and updating urgent
666  * marker if URG set.  Possibly send more data.  Unlike the other
667  * pru_*() routines, the mbuf chains are our responsibility.  We
668  * must either enqueue them or free them.  The other pru_* routines
669  * generally are caller-frees.
670  */
671 static int
672 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
673 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
674 {
675 	int error = 0;
676 	struct inpcb *inp;
677 	struct tcpcb *tp;
678 #ifdef INET6
679 	int isipv6;
680 #endif
681 	TCPDEBUG0;
682 
683 	crit_enter();
684 	inp = so->so_pcb;
685 
686 	if (inp == NULL) {
687 		/*
688 		 * OOPS! we lost a race, the TCP session got reset after
689 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
690 		 * network interrupt in the non-critical section of sosend().
691 		 */
692 		if (m)
693 			m_freem(m);
694 		if (control)
695 			m_freem(control);
696 		error = ECONNRESET;	/* XXX EPIPE? */
697 		tp = NULL;
698 		TCPDEBUG1();
699 		goto out;
700 	}
701 #ifdef INET6
702 	isipv6 = nam && nam->sa_family == AF_INET6;
703 #endif /* INET6 */
704 	tp = intotcpcb(inp);
705 	TCPDEBUG1();
706 	if (control) {
707 		/* TCP doesn't do control messages (rights, creds, etc) */
708 		if (control->m_len) {
709 			m_freem(control);
710 			if (m)
711 				m_freem(m);
712 			error = EINVAL;
713 			goto out;
714 		}
715 		m_freem(control);	/* empty control, just free it */
716 	}
717 	if(!(flags & PRUS_OOB)) {
718 		ssb_appendstream(&so->so_snd, m);
719 		if (nam && tp->t_state < TCPS_SYN_SENT) {
720 			/*
721 			 * Do implied connect if not yet connected,
722 			 * initialize window to default value, and
723 			 * initialize maxseg/maxopd using peer's cached
724 			 * MSS.
725 			 */
726 #ifdef INET6
727 			if (isipv6)
728 				error = tcp6_connect(tp, nam, td);
729 			else
730 #endif /* INET6 */
731 			error = tcp_connect(tp, nam, td);
732 			if (error)
733 				goto out;
734 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
735 			tcp_mss(tp, -1);
736 		}
737 
738 		if (flags & PRUS_EOF) {
739 			/*
740 			 * Close the send side of the connection after
741 			 * the data is sent.
742 			 */
743 			socantsendmore(so);
744 			tp = tcp_usrclosed(tp);
745 		}
746 		if (tp != NULL) {
747 			if (flags & PRUS_MORETOCOME)
748 				tp->t_flags |= TF_MORETOCOME;
749 			error = tcp_output(tp);
750 			if (flags & PRUS_MORETOCOME)
751 				tp->t_flags &= ~TF_MORETOCOME;
752 		}
753 	} else {
754 		if (ssb_space(&so->so_snd) < -512) {
755 			m_freem(m);
756 			error = ENOBUFS;
757 			goto out;
758 		}
759 		/*
760 		 * According to RFC961 (Assigned Protocols),
761 		 * the urgent pointer points to the last octet
762 		 * of urgent data.  We continue, however,
763 		 * to consider it to indicate the first octet
764 		 * of data past the urgent section.
765 		 * Otherwise, snd_up should be one lower.
766 		 */
767 		ssb_appendstream(&so->so_snd, m);
768 		if (nam && tp->t_state < TCPS_SYN_SENT) {
769 			/*
770 			 * Do implied connect if not yet connected,
771 			 * initialize window to default value, and
772 			 * initialize maxseg/maxopd using peer's cached
773 			 * MSS.
774 			 */
775 #ifdef INET6
776 			if (isipv6)
777 				error = tcp6_connect(tp, nam, td);
778 			else
779 #endif /* INET6 */
780 			error = tcp_connect(tp, nam, td);
781 			if (error)
782 				goto out;
783 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
784 			tcp_mss(tp, -1);
785 		}
786 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
787 		tp->t_flags |= TF_FORCE;
788 		error = tcp_output(tp);
789 		tp->t_flags &= ~TF_FORCE;
790 	}
791 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
792 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
793 }
794 
795 /*
796  * Abort the TCP.
797  */
798 static int
799 tcp_usr_abort(struct socket *so)
800 {
801 	int error = 0;
802 	struct inpcb *inp;
803 	struct tcpcb *tp;
804 
805 	COMMON_START(so, inp);
806 	tp = tcp_drop(tp, ECONNABORTED);
807 	COMMON_END(PRU_ABORT);
808 }
809 
810 /*
811  * Receive out-of-band data.
812  */
813 static int
814 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
815 {
816 	int error = 0;
817 	struct inpcb *inp;
818 	struct tcpcb *tp;
819 
820 	COMMON_START(so, inp);
821 	if ((so->so_oobmark == 0 &&
822 	     (so->so_state & SS_RCVATMARK) == 0) ||
823 	    so->so_options & SO_OOBINLINE ||
824 	    tp->t_oobflags & TCPOOB_HADDATA) {
825 		error = EINVAL;
826 		goto out;
827 	}
828 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
829 		error = EWOULDBLOCK;
830 		goto out;
831 	}
832 	m->m_len = 1;
833 	*mtod(m, caddr_t) = tp->t_iobc;
834 	if ((flags & MSG_PEEK) == 0)
835 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
836 	COMMON_END(PRU_RCVOOB);
837 }
838 
839 /* xxx - should be const */
840 struct pr_usrreqs tcp_usrreqs = {
841 	.pru_abort = tcp_usr_abort,
842 	.pru_accept = tcp_usr_accept,
843 	.pru_attach = tcp_usr_attach,
844 	.pru_bind = tcp_usr_bind,
845 	.pru_connect = tcp_usr_connect,
846 	.pru_connect2 = pru_connect2_notsupp,
847 	.pru_control = in_control,
848 	.pru_detach = tcp_usr_detach,
849 	.pru_disconnect = tcp_usr_disconnect,
850 	.pru_listen = tcp_usr_listen,
851 	.pru_peeraddr = in_setpeeraddr,
852 	.pru_rcvd = tcp_usr_rcvd,
853 	.pru_rcvoob = tcp_usr_rcvoob,
854 	.pru_send = tcp_usr_send,
855 	.pru_sense = pru_sense_null,
856 	.pru_shutdown = tcp_usr_shutdown,
857 	.pru_sockaddr = in_setsockaddr,
858 	.pru_sosend = sosend,
859 	.pru_soreceive = soreceive,
860 	.pru_sopoll = sopoll
861 };
862 
863 #ifdef INET6
864 struct pr_usrreqs tcp6_usrreqs = {
865 	.pru_abort = tcp_usr_abort,
866 	.pru_accept = tcp6_usr_accept,
867 	.pru_attach = tcp_usr_attach,
868 	.pru_bind = tcp6_usr_bind,
869 	.pru_connect = tcp6_usr_connect,
870 	.pru_connect2 = pru_connect2_notsupp,
871 	.pru_control = in6_control,
872 	.pru_detach = tcp_usr_detach,
873 	.pru_disconnect = tcp_usr_disconnect,
874 	.pru_listen = tcp6_usr_listen,
875 	.pru_peeraddr = in6_mapped_peeraddr,
876 	.pru_rcvd = tcp_usr_rcvd,
877 	.pru_rcvoob = tcp_usr_rcvoob,
878 	.pru_send = tcp_usr_send,
879 	.pru_sense = pru_sense_null,
880 	.pru_shutdown = tcp_usr_shutdown,
881 	.pru_sockaddr = in6_mapped_sockaddr,
882 	.pru_sosend = sosend,
883 	.pru_soreceive = soreceive,
884 	.pru_sopoll = sopoll
885 };
886 #endif /* INET6 */
887 
888 static int
889 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
890 		  struct sockaddr_in *if_sin)
891 {
892 	struct inpcb *inp = tp->t_inpcb, *oinp;
893 	struct socket *so = inp->inp_socket;
894 	struct tcpcb *otp;
895 	struct rmxp_tao *taop;
896 	struct rmxp_tao tao_noncached;
897 
898 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
899 	    sin->sin_addr, sin->sin_port,
900 	    inp->inp_laddr.s_addr != INADDR_ANY ?
901 		inp->inp_laddr : if_sin->sin_addr,
902 	    inp->inp_lport, 0, NULL);
903 	if (oinp != NULL) {
904 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
905 		    otp->t_state == TCPS_TIME_WAIT &&
906 		    (ticks - otp->t_starttime) < tcp_msl &&
907 		    (otp->t_flags & TF_RCVD_CC))
908 			tcp_close(otp);
909 		else
910 			return (EADDRINUSE);
911 	}
912 	if (inp->inp_laddr.s_addr == INADDR_ANY)
913 		inp->inp_laddr = if_sin->sin_addr;
914 	inp->inp_faddr = sin->sin_addr;
915 	inp->inp_fport = sin->sin_port;
916 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
917 	in_pcbinsconnhash(inp);
918 
919 	/* Compute window scaling to request.  */
920 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
921 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
922 		tp->request_r_scale++;
923 
924 	soisconnecting(so);
925 	tcpstat.tcps_connattempt++;
926 	tp->t_state = TCPS_SYN_SENT;
927 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
928 	tp->iss = tcp_new_isn(tp);
929 	tcp_sendseqinit(tp);
930 
931 	/*
932 	 * Generate a CC value for this connection and
933 	 * check whether CC or CCnew should be used.
934 	 */
935 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
936 		taop = &tao_noncached;
937 		bzero(taop, sizeof *taop);
938 	}
939 
940 	tp->cc_send = CC_INC(tcp_ccgen);
941 	if (taop->tao_ccsent != 0 &&
942 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
943 		taop->tao_ccsent = tp->cc_send;
944 	} else {
945 		taop->tao_ccsent = 0;
946 		tp->t_flags |= TF_SENDCCNEW;
947 	}
948 
949 	return (0);
950 }
951 
952 #ifdef SMP
953 
954 struct netmsg_tcp_connect {
955 	struct netmsg		nm_netmsg;
956 	struct tcpcb		*nm_tp;
957 	struct sockaddr_in	*nm_sin;
958 	struct sockaddr_in	*nm_ifsin;
959 };
960 
961 static void
962 tcp_connect_handler(netmsg_t netmsg)
963 {
964 	struct netmsg_tcp_connect *msg = (void *)netmsg;
965 	int error;
966 
967 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
968 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
969 }
970 
971 #endif
972 
973 /*
974  * Common subroutine to open a TCP connection to remote host specified
975  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
976  * port number if needed.  Call in_pcbladdr to do the routing and to choose
977  * a local host address (interface).  If there is an existing incarnation
978  * of the same connection in TIME-WAIT state and if the remote host was
979  * sending CC options and if the connection duration was < MSL, then
980  * truncate the previous TIME-WAIT state and proceed.
981  * Initialize connection parameters and enter SYN-SENT state.
982  */
983 static int
984 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
985 {
986 	struct inpcb *inp = tp->t_inpcb;
987 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
988 	struct sockaddr_in *if_sin;
989 	int error;
990 #ifdef SMP
991 	lwkt_port_t port;
992 #endif
993 
994 	if (inp->inp_lport == 0) {
995 		error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
996 		if (error)
997 			return (error);
998 	}
999 
1000 	/*
1001 	 * Cannot simply call in_pcbconnect, because there might be an
1002 	 * earlier incarnation of this same connection still in
1003 	 * TIME_WAIT state, creating an ADDRINUSE error.
1004 	 */
1005 	error = in_pcbladdr(inp, nam, &if_sin, td);
1006 	if (error)
1007 		return (error);
1008 
1009 #ifdef SMP
1010 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1011 	    inp->inp_laddr.s_addr ?
1012 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
1013 	    inp->inp_lport);
1014 
1015 	if (port != &curthread->td_msgport) {
1016 		struct netmsg_tcp_connect msg;
1017 		struct route *ro = &inp->inp_route;
1018 
1019 		/*
1020 		 * in_pcbladdr() may have allocated a route entry for us
1021 		 * on the current CPU, but we need a route entry on the
1022 		 * target CPU, so free it here.
1023 		 */
1024 		if (ro->ro_rt != NULL)
1025 			RTFREE(ro->ro_rt);
1026 		bzero(ro, sizeof(*ro));
1027 
1028 		netmsg_init(&msg.nm_netmsg, &curthread->td_msgport, 0,
1029 			    tcp_connect_handler);
1030 		msg.nm_tp = tp;
1031 		msg.nm_sin = sin;
1032 		msg.nm_ifsin = if_sin;
1033 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1034 	} else
1035 #endif
1036 		error = tcp_connect_oncpu(tp, sin, if_sin);
1037 
1038 	return (error);
1039 }
1040 
1041 #ifdef INET6
1042 static int
1043 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1044 {
1045 	struct inpcb *inp = tp->t_inpcb, *oinp;
1046 	struct socket *so = inp->inp_socket;
1047 	struct tcpcb *otp;
1048 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1049 	struct in6_addr *addr6;
1050 	struct rmxp_tao *taop;
1051 	struct rmxp_tao tao_noncached;
1052 	int error;
1053 
1054 	if (inp->inp_lport == 0) {
1055 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
1056 		if (error)
1057 			return error;
1058 	}
1059 
1060 	/*
1061 	 * Cannot simply call in_pcbconnect, because there might be an
1062 	 * earlier incarnation of this same connection still in
1063 	 * TIME_WAIT state, creating an ADDRINUSE error.
1064 	 */
1065 	error = in6_pcbladdr(inp, nam, &addr6, td);
1066 	if (error)
1067 		return error;
1068 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1069 				  &sin6->sin6_addr, sin6->sin6_port,
1070 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1071 				      addr6 : &inp->in6p_laddr,
1072 				  inp->inp_lport,  0, NULL);
1073 	if (oinp) {
1074 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
1075 		    otp->t_state == TCPS_TIME_WAIT &&
1076 		    (ticks - otp->t_starttime) < tcp_msl &&
1077 		    (otp->t_flags & TF_RCVD_CC))
1078 			otp = tcp_close(otp);
1079 		else
1080 			return (EADDRINUSE);
1081 	}
1082 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1083 		inp->in6p_laddr = *addr6;
1084 	inp->in6p_faddr = sin6->sin6_addr;
1085 	inp->inp_fport = sin6->sin6_port;
1086 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1087 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1088 	in_pcbinsconnhash(inp);
1089 
1090 	/* Compute window scaling to request.  */
1091 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1092 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat)
1093 		tp->request_r_scale++;
1094 
1095 	soisconnecting(so);
1096 	tcpstat.tcps_connattempt++;
1097 	tp->t_state = TCPS_SYN_SENT;
1098 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
1099 	tp->iss = tcp_new_isn(tp);
1100 	tcp_sendseqinit(tp);
1101 
1102 	/*
1103 	 * Generate a CC value for this connection and
1104 	 * check whether CC or CCnew should be used.
1105 	 */
1106 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
1107 		taop = &tao_noncached;
1108 		bzero(taop, sizeof *taop);
1109 	}
1110 
1111 	tp->cc_send = CC_INC(tcp_ccgen);
1112 	if (taop->tao_ccsent != 0 &&
1113 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
1114 		taop->tao_ccsent = tp->cc_send;
1115 	} else {
1116 		taop->tao_ccsent = 0;
1117 		tp->t_flags |= TF_SENDCCNEW;
1118 	}
1119 
1120 	return (0);
1121 }
1122 #endif /* INET6 */
1123 
1124 /*
1125  * The new sockopt interface makes it possible for us to block in the
1126  * copyin/out step (if we take a page fault).  Taking a page fault while
1127  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1128  * both now use TSM, there probably isn't any need for this function to
1129  * run in a critical section any more.  This needs more examination.)
1130  */
1131 int
1132 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1133 {
1134 	int	error, opt, optval;
1135 	struct	inpcb *inp;
1136 	struct	tcpcb *tp;
1137 
1138 	error = 0;
1139 	crit_enter();		/* XXX */
1140 	inp = so->so_pcb;
1141 	if (inp == NULL) {
1142 		crit_exit();
1143 		return (ECONNRESET);
1144 	}
1145 	if (sopt->sopt_level != IPPROTO_TCP) {
1146 #ifdef INET6
1147 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1148 			error = ip6_ctloutput(so, sopt);
1149 		else
1150 #endif /* INET6 */
1151 		error = ip_ctloutput(so, sopt);
1152 		crit_exit();
1153 		return (error);
1154 	}
1155 	tp = intotcpcb(inp);
1156 
1157 	switch (sopt->sopt_dir) {
1158 	case SOPT_SET:
1159 		switch (sopt->sopt_name) {
1160 		case TCP_NODELAY:
1161 		case TCP_NOOPT:
1162 			error = sooptcopyin(sopt, &optval, sizeof optval,
1163 					    sizeof optval);
1164 			if (error)
1165 				break;
1166 
1167 			switch (sopt->sopt_name) {
1168 			case TCP_NODELAY:
1169 				opt = TF_NODELAY;
1170 				break;
1171 			case TCP_NOOPT:
1172 				opt = TF_NOOPT;
1173 				break;
1174 			default:
1175 				opt = 0; /* dead code to fool gcc */
1176 				break;
1177 			}
1178 
1179 			if (optval)
1180 				tp->t_flags |= opt;
1181 			else
1182 				tp->t_flags &= ~opt;
1183 			break;
1184 
1185 		case TCP_NOPUSH:
1186 			error = sooptcopyin(sopt, &optval, sizeof optval,
1187 					    sizeof optval);
1188 			if (error)
1189 				break;
1190 
1191 			if (optval)
1192 				tp->t_flags |= TF_NOPUSH;
1193 			else {
1194 				tp->t_flags &= ~TF_NOPUSH;
1195 				error = tcp_output(tp);
1196 			}
1197 			break;
1198 
1199 		case TCP_MAXSEG:
1200 			error = sooptcopyin(sopt, &optval, sizeof optval,
1201 					    sizeof optval);
1202 			if (error)
1203 				break;
1204 
1205 			if (optval > 0 && optval <= tp->t_maxseg)
1206 				tp->t_maxseg = optval;
1207 			else
1208 				error = EINVAL;
1209 			break;
1210 
1211 		default:
1212 			error = ENOPROTOOPT;
1213 			break;
1214 		}
1215 		break;
1216 
1217 	case SOPT_GET:
1218 		switch (sopt->sopt_name) {
1219 		case TCP_NODELAY:
1220 			optval = tp->t_flags & TF_NODELAY;
1221 			break;
1222 		case TCP_MAXSEG:
1223 			optval = tp->t_maxseg;
1224 			break;
1225 		case TCP_NOOPT:
1226 			optval = tp->t_flags & TF_NOOPT;
1227 			break;
1228 		case TCP_NOPUSH:
1229 			optval = tp->t_flags & TF_NOPUSH;
1230 			break;
1231 		default:
1232 			error = ENOPROTOOPT;
1233 			break;
1234 		}
1235 		if (error == 0)
1236 			error = sooptcopyout(sopt, &optval, sizeof optval);
1237 		break;
1238 	}
1239 	crit_exit();
1240 	return (error);
1241 }
1242 
1243 /*
1244  * tcp_sendspace and tcp_recvspace are the default send and receive window
1245  * sizes, respectively.  These are obsolescent (this information should
1246  * be set by the route).
1247  */
1248 u_long	tcp_sendspace = 1024*32;
1249 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1250     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1251 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1252 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1253     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1254 
1255 /*
1256  * Attach TCP protocol to socket, allocating
1257  * internet protocol control block, tcp control block,
1258  * bufer space, and entering LISTEN state if to accept connections.
1259  */
1260 static int
1261 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1262 {
1263 	struct tcpcb *tp;
1264 	struct inpcb *inp;
1265 	int error;
1266 	int cpu;
1267 #ifdef INET6
1268 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1269 #endif
1270 
1271 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1272 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1273 				  ai->sb_rlimit);
1274 		if (error)
1275 			return (error);
1276 	}
1277 	cpu = mycpu->gd_cpuid;
1278 	error = in_pcballoc(so, &tcbinfo[cpu]);
1279 	if (error)
1280 		return (error);
1281 	inp = so->so_pcb;
1282 #ifdef INET6
1283 	if (isipv6) {
1284 		inp->inp_vflag |= INP_IPV6;
1285 		inp->in6p_hops = -1;	/* use kernel default */
1286 	}
1287 	else
1288 #endif
1289 	inp->inp_vflag |= INP_IPV4;
1290 	tp = tcp_newtcpcb(inp);
1291 	if (tp == 0) {
1292 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1293 
1294 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1295 #ifdef INET6
1296 		if (isipv6)
1297 			in6_pcbdetach(inp);
1298 		else
1299 #endif
1300 		in_pcbdetach(inp);
1301 		so->so_state |= nofd;
1302 		return (ENOBUFS);
1303 	}
1304 	tp->t_state = TCPS_CLOSED;
1305 	return (0);
1306 }
1307 
1308 /*
1309  * Initiate (or continue) disconnect.
1310  * If embryonic state, just send reset (once).
1311  * If in ``let data drain'' option and linger null, just drop.
1312  * Otherwise (hard), mark socket disconnecting and drop
1313  * current input data; switch states based on user close, and
1314  * send segment to peer (with FIN).
1315  */
1316 static struct tcpcb *
1317 tcp_disconnect(struct tcpcb *tp)
1318 {
1319 	struct socket *so = tp->t_inpcb->inp_socket;
1320 
1321 	if (tp->t_state < TCPS_ESTABLISHED)
1322 		tp = tcp_close(tp);
1323 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1324 		tp = tcp_drop(tp, 0);
1325 	else {
1326 		soisdisconnecting(so);
1327 		sbflush(&so->so_rcv.sb);
1328 		tp = tcp_usrclosed(tp);
1329 		if (tp)
1330 			tcp_output(tp);
1331 	}
1332 	return (tp);
1333 }
1334 
1335 /*
1336  * User issued close, and wish to trail through shutdown states:
1337  * if never received SYN, just forget it.  If got a SYN from peer,
1338  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1339  * If already got a FIN from peer, then almost done; go to LAST_ACK
1340  * state.  In all other cases, have already sent FIN to peer (e.g.
1341  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1342  * for peer to send FIN or not respond to keep-alives, etc.
1343  * We can let the user exit from the close as soon as the FIN is acked.
1344  */
1345 static struct tcpcb *
1346 tcp_usrclosed(struct tcpcb *tp)
1347 {
1348 
1349 	switch (tp->t_state) {
1350 
1351 	case TCPS_CLOSED:
1352 	case TCPS_LISTEN:
1353 		tp->t_state = TCPS_CLOSED;
1354 		tp = tcp_close(tp);
1355 		break;
1356 
1357 	case TCPS_SYN_SENT:
1358 	case TCPS_SYN_RECEIVED:
1359 		tp->t_flags |= TF_NEEDFIN;
1360 		break;
1361 
1362 	case TCPS_ESTABLISHED:
1363 		tp->t_state = TCPS_FIN_WAIT_1;
1364 		break;
1365 
1366 	case TCPS_CLOSE_WAIT:
1367 		tp->t_state = TCPS_LAST_ACK;
1368 		break;
1369 	}
1370 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1371 		soisdisconnected(tp->t_inpcb->inp_socket);
1372 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1373 		if (tp->t_state == TCPS_FIN_WAIT_2)
1374 			callout_reset(tp->tt_2msl, tcp_maxidle,
1375 				      tcp_timer_2msl, tp);
1376 	}
1377 	return (tp);
1378 }
1379