xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision 90ea502b8c5d21f908cedff6680ee2bc9e74ce74)
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 #include <netinet6/tcp6_var.h>
114 #endif
115 #include <netinet/tcp.h>
116 #include <netinet/tcp_fsm.h>
117 #include <netinet/tcp_seq.h>
118 #include <netinet/tcp_timer.h>
119 #include <netinet/tcp_timer2.h>
120 #include <netinet/tcp_var.h>
121 #include <netinet/tcpip.h>
122 #ifdef TCPDEBUG
123 #include <netinet/tcp_debug.h>
124 #endif
125 
126 #ifdef IPSEC
127 #include <netinet6/ipsec.h>
128 #endif /*IPSEC*/
129 
130 /*
131  * TCP protocol interface to socket abstraction.
132  */
133 extern	char *tcpstates[];	/* XXX ??? */
134 
135 static int	tcp_attach (struct socket *, struct pru_attach_info *);
136 static int	tcp_connect (struct tcpcb *, int flags, struct mbuf *m,
137 				struct sockaddr *, struct thread *);
138 #ifdef INET6
139 static int	tcp6_connect (struct tcpcb *, int flags, struct mbuf *m,
140 				struct sockaddr *, struct thread *);
141 static int	tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
142 				struct sockaddr_in6 *sin6,
143 				struct in6_addr *addr6);
144 #endif /* INET6 */
145 static struct tcpcb *
146 		tcp_disconnect (struct tcpcb *);
147 static struct tcpcb *
148 		tcp_usrclosed (struct tcpcb *);
149 
150 #ifdef TCPDEBUG
151 #define	TCPDEBUG0	int ostate = 0
152 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
153 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
154 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
155 #else
156 #define	TCPDEBUG0
157 #define	TCPDEBUG1()
158 #define	TCPDEBUG2(req)
159 #endif
160 
161 /*
162  * TCP attaches to socket via pru_attach(), reserving space,
163  * and an internet control block.
164  */
165 static int
166 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
167 {
168 	int error;
169 	struct inpcb *inp;
170 	struct tcpcb *tp = 0;
171 	TCPDEBUG0;
172 
173 	crit_enter();
174 	inp = so->so_pcb;
175 	TCPDEBUG1();
176 	if (inp) {
177 		error = EISCONN;
178 		goto out;
179 	}
180 
181 	error = tcp_attach(so, ai);
182 	if (error)
183 		goto out;
184 
185 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
186 		so->so_linger = TCP_LINGERTIME;
187 	tp = sototcpcb(so);
188 out:
189 	TCPDEBUG2(PRU_ATTACH);
190 	crit_exit();
191 	return error;
192 }
193 
194 /*
195  * pru_detach() detaches the TCP protocol from the socket.
196  * If the protocol state is non-embryonic, then can't
197  * do this directly: have to initiate a pru_disconnect(),
198  * which may finish later; embryonic TCB's can just
199  * be discarded here.
200  */
201 static int
202 tcp_usr_detach(struct socket *so)
203 {
204 	int error = 0;
205 	struct inpcb *inp;
206 	struct tcpcb *tp;
207 	TCPDEBUG0;
208 
209 	crit_enter();
210 	inp = so->so_pcb;
211 
212 	/*
213 	 * If the inp is already detached it may have been due to an async
214 	 * close.  Just return as if no error occured.
215 	 */
216 	if (inp == NULL) {
217 		crit_exit();
218 		return 0;
219 	}
220 
221 	/*
222 	 * It's possible for the tcpcb (tp) to disconnect from the inp due
223 	 * to tcp_drop()->tcp_close() being called.  This may occur *after*
224 	 * the detach message has been queued so we may find a NULL tp here.
225 	 */
226 	if ((tp = intotcpcb(inp)) != NULL) {
227 		TCPDEBUG1();
228 		tp = tcp_disconnect(tp);
229 		TCPDEBUG2(PRU_DETACH);
230 	}
231 	crit_exit();
232 	return error;
233 }
234 
235 /*
236  * Note: ignore_error is non-zero for certain disconnection races
237  * which we want to silently allow, otherwise close() may return
238  * an unexpected error.
239  */
240 #define	COMMON_START(so, inp, ignore_error)			\
241 	TCPDEBUG0; 		\
242 				\
243 	crit_enter();		\
244 	inp = so->so_pcb; 	\
245 	do {			\
246 		 if (inp == NULL) {				\
247 			 crit_exit();				\
248 			 return (ignore_error ? 0 : EINVAL);	\
249 		 }						\
250 		 tp = intotcpcb(inp);				\
251 		 TCPDEBUG1();					\
252 	} while(0)
253 
254 #define COMMON_END(req)	out: TCPDEBUG2(req); crit_exit(); return error; goto out
255 
256 
257 /*
258  * Give the socket an address.
259  */
260 static int
261 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
262 {
263 	int error = 0;
264 	struct inpcb *inp;
265 	struct tcpcb *tp;
266 	struct sockaddr_in *sinp;
267 
268 	COMMON_START(so, inp, 0);
269 
270 	/*
271 	 * Must check for multicast addresses and disallow binding
272 	 * to them.
273 	 */
274 	sinp = (struct sockaddr_in *)nam;
275 	if (sinp->sin_family == AF_INET &&
276 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
277 		error = EAFNOSUPPORT;
278 		goto out;
279 	}
280 	error = in_pcbbind(inp, nam, td);
281 	if (error)
282 		goto out;
283 	COMMON_END(PRU_BIND);
284 
285 }
286 
287 #ifdef INET6
288 static int
289 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
290 {
291 	int error = 0;
292 	struct inpcb *inp;
293 	struct tcpcb *tp;
294 	struct sockaddr_in6 *sin6p;
295 
296 	COMMON_START(so, inp, 0);
297 
298 	/*
299 	 * Must check for multicast addresses and disallow binding
300 	 * to them.
301 	 */
302 	sin6p = (struct sockaddr_in6 *)nam;
303 	if (sin6p->sin6_family == AF_INET6 &&
304 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
305 		error = EAFNOSUPPORT;
306 		goto out;
307 	}
308 	inp->inp_vflag &= ~INP_IPV4;
309 	inp->inp_vflag |= INP_IPV6;
310 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
311 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
312 			inp->inp_vflag |= INP_IPV4;
313 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
314 			struct sockaddr_in sin;
315 
316 			in6_sin6_2_sin(&sin, sin6p);
317 			inp->inp_vflag |= INP_IPV4;
318 			inp->inp_vflag &= ~INP_IPV6;
319 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
320 			goto out;
321 		}
322 	}
323 	error = in6_pcbbind(inp, nam, td);
324 	if (error)
325 		goto out;
326 	COMMON_END(PRU_BIND);
327 }
328 #endif /* INET6 */
329 
330 #ifdef SMP
331 struct netmsg_inswildcard {
332 	struct netmsg		nm_netmsg;
333 	struct inpcb		*nm_inp;
334 	struct inpcbinfo	*nm_pcbinfo;
335 };
336 
337 static void
338 in_pcbinswildcardhash_handler(struct netmsg *msg0)
339 {
340 	struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
341 
342 	in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
343 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 0);
344 }
345 #endif
346 
347 /*
348  * Prepare to accept connections.
349  */
350 static int
351 tcp_usr_listen(struct socket *so, struct thread *td)
352 {
353 	int error = 0;
354 	struct inpcb *inp;
355 	struct tcpcb *tp;
356 #ifdef SMP
357 	int cpu;
358 #endif
359 
360 	COMMON_START(so, inp, 0);
361 	if (inp->inp_lport == 0) {
362 		error = in_pcbbind(inp, NULL, td);
363 		if (error != 0)
364 			goto out;
365 	}
366 
367 	tp->t_state = TCPS_LISTEN;
368 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
369 #ifdef SMP
370 	/*
371 	 * We have to set the flag because we can't have other cpus
372 	 * messing with our inp's flags.
373 	 */
374 	inp->inp_flags |= INP_WILDCARD_MP;
375 	for (cpu = 0; cpu < ncpus2; cpu++) {
376 		struct netmsg_inswildcard *msg;
377 
378 		if (cpu == mycpu->gd_cpuid) {
379 			in_pcbinswildcardhash(inp);
380 			continue;
381 		}
382 
383 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
384 			      M_INTWAIT);
385 		netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
386 			    0, in_pcbinswildcardhash_handler);
387 		msg->nm_inp = inp;
388 		msg->nm_pcbinfo = &tcbinfo[cpu];
389 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
390 	}
391 #else
392 	in_pcbinswildcardhash(inp);
393 #endif
394 	COMMON_END(PRU_LISTEN);
395 }
396 
397 #ifdef INET6
398 static int
399 tcp6_usr_listen(struct socket *so, struct thread *td)
400 {
401 	int error = 0;
402 	struct inpcb *inp;
403 	struct tcpcb *tp;
404 #ifdef SMP
405 	int cpu;
406 #endif
407 
408 	COMMON_START(so, inp, 0);
409 	if (inp->inp_lport == 0) {
410 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
411 			inp->inp_vflag |= INP_IPV4;
412 		else
413 			inp->inp_vflag &= ~INP_IPV4;
414 		error = in6_pcbbind(inp, NULL, td);
415 	}
416 	if (error == 0)
417 		tp->t_state = TCPS_LISTEN;
418 #ifdef SMP
419 	/*
420 	 * We have to set the flag because we can't have other cpus
421 	 * messing with our inp's flags.
422 	 */
423 	inp->inp_flags |= INP_WILDCARD_MP;
424 	for (cpu = 0; cpu < ncpus2; cpu++) {
425 		struct netmsg_inswildcard *msg;
426 
427 		if (cpu == mycpu->gd_cpuid) {
428 			in_pcbinswildcardhash(inp);
429 			continue;
430 		}
431 
432 		msg = kmalloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
433 			      M_INTWAIT);
434 		netmsg_init(&msg->nm_netmsg, NULL, &netisr_afree_rport,
435 			    0, in_pcbinswildcardhash_handler);
436 		msg->nm_inp = inp;
437 		msg->nm_pcbinfo = &tcbinfo[cpu];
438 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_netmsg.nm_lmsg);
439 	}
440 #else
441 	in_pcbinswildcardhash(inp);
442 #endif
443 	COMMON_END(PRU_LISTEN);
444 }
445 #endif /* INET6 */
446 
447 /*
448  * Initiate connection to peer.
449  * Create a template for use in transmissions on this connection.
450  * Enter SYN_SENT state, and mark socket as connecting.
451  * Start keep-alive timer, and seed output sequence space.
452  * Send initial segment on connection.
453  */
454 static int
455 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
456 {
457 	int error = 0;
458 	struct inpcb *inp;
459 	struct tcpcb *tp;
460 	struct sockaddr_in *sinp;
461 
462 	COMMON_START(so, inp, 0);
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, 0, NULL, nam, td)) != 0)
480 		goto out;
481 	COMMON_END(PRU_CONNECT);
482 }
483 
484 #ifdef INET6
485 static int
486 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
487 {
488 	int error = 0;
489 	struct inpcb *inp;
490 	struct tcpcb *tp;
491 	struct sockaddr_in6 *sin6p;
492 
493 	COMMON_START(so, inp, 0);
494 
495 	/*
496 	 * Must disallow TCP ``connections'' to multicast addresses.
497 	 */
498 	sin6p = (struct sockaddr_in6 *)nam;
499 	if (sin6p->sin6_family == AF_INET6
500 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
501 		error = EAFNOSUPPORT;
502 		goto out;
503 	}
504 
505 	if (!prison_remote_ip(td, nam)) {
506 		error = EAFNOSUPPORT; /* IPv4 only jail */
507 		goto out;
508 	}
509 
510 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
511 		struct sockaddr_in sin;
512 
513 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
514 			error = EINVAL;
515 			goto out;
516 		}
517 
518 		in6_sin6_2_sin(&sin, sin6p);
519 		inp->inp_vflag |= INP_IPV4;
520 		inp->inp_vflag &= ~INP_IPV6;
521 		error = tcp_connect(tp, 0, NULL, (struct sockaddr *)&sin, td);
522 		if (error)
523 			goto out;
524 		goto out;
525 	}
526 	inp->inp_vflag &= ~INP_IPV4;
527 	inp->inp_vflag |= INP_IPV6;
528 	inp->inp_inc.inc_isipv6 = 1;
529 	if ((error = tcp6_connect(tp, 0, NULL, nam, td)) != 0)
530 		goto out;
531 	error = tcp_output(tp);
532 	COMMON_END(PRU_CONNECT);
533 }
534 #endif /* INET6 */
535 
536 /*
537  * Initiate disconnect from peer.
538  * If connection never passed embryonic stage, just drop;
539  * else if don't need to let data drain, then can just drop anyways,
540  * else have to begin TCP shutdown process: mark socket disconnecting,
541  * drain unread data, state switch to reflect user close, and
542  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
543  * when peer sends FIN and acks ours.
544  *
545  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
546  */
547 static int
548 tcp_usr_disconnect(struct socket *so)
549 {
550 	int error = 0;
551 	struct inpcb *inp;
552 	struct tcpcb *tp;
553 
554 	COMMON_START(so, inp, 1);
555 	tp = tcp_disconnect(tp);
556 	COMMON_END(PRU_DISCONNECT);
557 }
558 
559 /*
560  * Accept a connection.  Essentially all the work is
561  * done at higher levels; just return the address
562  * of the peer, storing through addr.
563  */
564 static int
565 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
566 {
567 	int error = 0;
568 	struct inpcb *inp;
569 	struct tcpcb *tp = NULL;
570 	TCPDEBUG0;
571 
572 	crit_enter();
573 	inp = so->so_pcb;
574 	if (so->so_state & SS_ISDISCONNECTED) {
575 		error = ECONNABORTED;
576 		goto out;
577 	}
578 	if (inp == 0) {
579 		crit_exit();
580 		return (EINVAL);
581 	}
582 	tp = intotcpcb(inp);
583 	TCPDEBUG1();
584 	in_setpeeraddr(so, nam);
585 	COMMON_END(PRU_ACCEPT);
586 }
587 
588 #ifdef INET6
589 static int
590 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
591 {
592 	int error = 0;
593 	struct inpcb *inp;
594 	struct tcpcb *tp = NULL;
595 	TCPDEBUG0;
596 
597 	crit_enter();
598 	inp = so->so_pcb;
599 
600 	if (so->so_state & SS_ISDISCONNECTED) {
601 		error = ECONNABORTED;
602 		goto out;
603 	}
604 	if (inp == 0) {
605 		crit_exit();
606 		return (EINVAL);
607 	}
608 	tp = intotcpcb(inp);
609 	TCPDEBUG1();
610 	in6_mapped_peeraddr(so, nam);
611 	COMMON_END(PRU_ACCEPT);
612 }
613 #endif /* INET6 */
614 /*
615  * Mark the connection as being incapable of further output.
616  */
617 static int
618 tcp_usr_shutdown(struct socket *so)
619 {
620 	int error = 0;
621 	struct inpcb *inp;
622 	struct tcpcb *tp;
623 
624 	COMMON_START(so, inp, 0);
625 	socantsendmore(so);
626 	tp = tcp_usrclosed(tp);
627 	if (tp)
628 		error = tcp_output(tp);
629 	COMMON_END(PRU_SHUTDOWN);
630 }
631 
632 /*
633  * After a receive, possibly send window update to peer.
634  */
635 static int
636 tcp_usr_rcvd(struct socket *so, int flags)
637 {
638 	int error = 0;
639 	struct inpcb *inp;
640 	struct tcpcb *tp;
641 
642 	COMMON_START(so, inp, 0);
643 	tcp_output(tp);
644 	COMMON_END(PRU_RCVD);
645 }
646 
647 /*
648  * Do a send by putting data in output queue and updating urgent
649  * marker if URG set.  Possibly send more data.  Unlike the other
650  * pru_*() routines, the mbuf chains are our responsibility.  We
651  * must either enqueue them or free them.  The other pru_* routines
652  * generally are caller-frees.
653  */
654 static int
655 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
656 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
657 {
658 	int error = 0;
659 	struct inpcb *inp;
660 	struct tcpcb *tp;
661 #ifdef INET6
662 	int isipv6;
663 #endif
664 	TCPDEBUG0;
665 
666 	crit_enter();
667 	inp = so->so_pcb;
668 
669 	if (inp == NULL) {
670 		/*
671 		 * OOPS! we lost a race, the TCP session got reset after
672 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
673 		 * network interrupt in the non-critical section of sosend().
674 		 */
675 		m_freem(m);
676 		if (control)
677 			m_freem(control);
678 		error = ECONNRESET;	/* XXX EPIPE? */
679 		tp = NULL;
680 		TCPDEBUG1();
681 		goto out;
682 	}
683 #ifdef INET6
684 	isipv6 = nam && nam->sa_family == AF_INET6;
685 #endif /* INET6 */
686 	tp = intotcpcb(inp);
687 	TCPDEBUG1();
688 	if (control) {
689 		/* TCP doesn't do control messages (rights, creds, etc) */
690 		if (control->m_len) {
691 			m_freem(control);
692 			m_freem(m);
693 			error = EINVAL;
694 			goto out;
695 		}
696 		m_freem(control);	/* empty control, just free it */
697 	}
698 
699 	/*
700 	 * Don't let too much OOB data build up
701 	 */
702 	if (flags & PRUS_OOB) {
703 		if (ssb_space(&so->so_snd) < -512) {
704 			m_freem(m);
705 			error = ENOBUFS;
706 			goto out;
707 		}
708 	}
709 
710 	/*
711 	 * Do implied connect if not yet connected.  Any data sent
712 	 * with the connect is handled by tcp_connect() and friends.
713 	 *
714 	 * NOTE!  PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
715 	 */
716 	if (nam && tp->t_state < TCPS_SYN_SENT) {
717 #ifdef INET6
718 		if (isipv6)
719 			error = tcp6_connect(tp, flags, m, nam, td);
720 		else
721 #endif /* INET6 */
722 		error = tcp_connect(tp, flags, m, nam, td);
723 #if 0
724 		/* WTF is this doing here? */
725 		tp->snd_wnd = TTCP_CLIENT_SND_WND;
726 		tcp_mss(tp, -1);
727 #endif
728 		goto out;
729 	}
730 
731 	/*
732 	 * Pump the data into the socket.
733 	 */
734 	if (m)
735 		ssb_appendstream(&so->so_snd, m);
736 	if (flags & PRUS_OOB) {
737 		/*
738 		 * According to RFC961 (Assigned Protocols),
739 		 * the urgent pointer points to the last octet
740 		 * of urgent data.  We continue, however,
741 		 * to consider it to indicate the first octet
742 		 * of data past the urgent section.
743 		 * Otherwise, snd_up should be one lower.
744 		 */
745 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
746 		tp->t_flags |= TF_FORCE;
747 		error = tcp_output(tp);
748 		tp->t_flags &= ~TF_FORCE;
749 	} else {
750 		if (flags & PRUS_EOF) {
751 			/*
752 			 * Close the send side of the connection after
753 			 * the data is sent.
754 			 */
755 			socantsendmore(so);
756 			tp = tcp_usrclosed(tp);
757 		}
758 		if (tp != NULL) {
759 			if (flags & PRUS_MORETOCOME)
760 				tp->t_flags |= TF_MORETOCOME;
761 			error = tcp_output(tp);
762 			if (flags & PRUS_MORETOCOME)
763 				tp->t_flags &= ~TF_MORETOCOME;
764 		}
765 	}
766 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
767 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
768 }
769 
770 /*
771  * Abort the TCP.
772  */
773 static int
774 tcp_usr_abort(struct socket *so)
775 {
776 	int error = 0;
777 	struct inpcb *inp;
778 	struct tcpcb *tp;
779 
780 	COMMON_START(so, inp, 1);
781 	tp = tcp_drop(tp, ECONNABORTED);
782 	COMMON_END(PRU_ABORT);
783 }
784 
785 /*
786  * Receive out-of-band data.
787  */
788 static int
789 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
790 {
791 	int error = 0;
792 	struct inpcb *inp;
793 	struct tcpcb *tp;
794 
795 	COMMON_START(so, inp, 0);
796 	if ((so->so_oobmark == 0 &&
797 	     (so->so_state & SS_RCVATMARK) == 0) ||
798 	    so->so_options & SO_OOBINLINE ||
799 	    tp->t_oobflags & TCPOOB_HADDATA) {
800 		error = EINVAL;
801 		goto out;
802 	}
803 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
804 		error = EWOULDBLOCK;
805 		goto out;
806 	}
807 	m->m_len = 1;
808 	*mtod(m, caddr_t) = tp->t_iobc;
809 	if ((flags & MSG_PEEK) == 0)
810 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
811 	COMMON_END(PRU_RCVOOB);
812 }
813 
814 /* xxx - should be const */
815 struct pr_usrreqs tcp_usrreqs = {
816 	.pru_abort = tcp_usr_abort,
817 	.pru_accept = tcp_usr_accept,
818 	.pru_attach = tcp_usr_attach,
819 	.pru_bind = tcp_usr_bind,
820 	.pru_connect = tcp_usr_connect,
821 	.pru_connect2 = pru_connect2_notsupp,
822 	.pru_control = in_control,
823 	.pru_detach = tcp_usr_detach,
824 	.pru_disconnect = tcp_usr_disconnect,
825 	.pru_listen = tcp_usr_listen,
826 	.pru_peeraddr = in_setpeeraddr,
827 	.pru_rcvd = tcp_usr_rcvd,
828 	.pru_rcvoob = tcp_usr_rcvoob,
829 	.pru_send = tcp_usr_send,
830 	.pru_sense = pru_sense_null,
831 	.pru_shutdown = tcp_usr_shutdown,
832 	.pru_sockaddr = in_setsockaddr,
833 	.pru_sosend = sosend,
834 	.pru_soreceive = soreceive,
835 	.pru_sopoll = sopoll
836 };
837 
838 #ifdef INET6
839 struct pr_usrreqs tcp6_usrreqs = {
840 	.pru_abort = tcp_usr_abort,
841 	.pru_accept = tcp6_usr_accept,
842 	.pru_attach = tcp_usr_attach,
843 	.pru_bind = tcp6_usr_bind,
844 	.pru_connect = tcp6_usr_connect,
845 	.pru_connect2 = pru_connect2_notsupp,
846 	.pru_control = in6_control,
847 	.pru_detach = tcp_usr_detach,
848 	.pru_disconnect = tcp_usr_disconnect,
849 	.pru_listen = tcp6_usr_listen,
850 	.pru_peeraddr = in6_mapped_peeraddr,
851 	.pru_rcvd = tcp_usr_rcvd,
852 	.pru_rcvoob = tcp_usr_rcvoob,
853 	.pru_send = tcp_usr_send,
854 	.pru_sense = pru_sense_null,
855 	.pru_shutdown = tcp_usr_shutdown,
856 	.pru_sockaddr = in6_mapped_sockaddr,
857 	.pru_sosend = sosend,
858 	.pru_soreceive = soreceive,
859 	.pru_sopoll = sopoll
860 };
861 #endif /* INET6 */
862 
863 static int
864 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
865 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
866 {
867 	struct inpcb *inp = tp->t_inpcb, *oinp;
868 	struct socket *so = inp->inp_socket;
869 	struct route *ro = &inp->inp_route;
870 
871 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
872 	    sin->sin_addr, sin->sin_port,
873 	    inp->inp_laddr.s_addr != INADDR_ANY ?
874 		inp->inp_laddr : if_sin->sin_addr,
875 	    inp->inp_lport, 0, NULL);
876 	if (oinp != NULL) {
877 		m_freem(m);
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 	/*
888 	 * We are now on the inpcb's owner CPU, if the cached route was
889 	 * freed because the rtentry's owner CPU is not the current CPU
890 	 * (e.g. in tcp_connect()), then we try to reallocate it here with
891 	 * the hope that a rtentry may be cloned from a RTF_PRCLONING
892 	 * rtentry.
893 	 */
894 	if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
895 	    ro->ro_rt == NULL) {
896 		bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
897 		ro->ro_dst.sa_family = AF_INET;
898 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
899 		((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
900 			sin->sin_addr;
901 		rtalloc(ro);
902 	}
903 
904 	/*
905 	 * Now that no more errors can occur, change the protocol processing
906 	 * port to the current thread (which is the correct thread).
907 	 *
908 	 * Create TCP timer message now; we are on the tcpcb's owner
909 	 * CPU/thread.
910 	 */
911 	sosetport(so, &curthread->td_msgport);
912 	tcp_create_timermsg(tp, &curthread->td_msgport);
913 
914 	/*
915 	 * Compute window scaling to request.  Use a larger scaling then
916 	 * needed for the initial receive buffer in case the receive buffer
917 	 * gets expanded.
918 	 */
919 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
920 		tp->request_r_scale = TCP_MIN_WINSHIFT;
921 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
922 	       (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
923 	) {
924 		tp->request_r_scale++;
925 	}
926 
927 	soisconnecting(so);
928 	tcpstat.tcps_connattempt++;
929 	tp->t_state = TCPS_SYN_SENT;
930 	tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
931 	tp->iss = tcp_new_isn(tp);
932 	tcp_sendseqinit(tp);
933 	if (m) {
934 		ssb_appendstream(&so->so_snd, m);
935 		m = NULL;
936 		if (flags & PRUS_OOB)
937 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
938 	}
939 
940 	/*
941 	 * Close the send side of the connection after
942 	 * the data is sent if flagged.
943 	 */
944 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
945 		socantsendmore(so);
946 		tp = tcp_usrclosed(tp);
947 	}
948 	return (tcp_output(tp));
949 }
950 
951 #ifdef SMP
952 
953 struct netmsg_tcp_connect {
954 	struct netmsg		nm_netmsg;
955 	struct tcpcb		*nm_tp;
956 	struct sockaddr_in	*nm_sin;
957 	struct sockaddr_in	*nm_ifsin;
958 	int			nm_flags;
959 	struct mbuf		*nm_m;
960 };
961 
962 static void
963 tcp_connect_handler(netmsg_t netmsg)
964 {
965 	struct netmsg_tcp_connect *msg = (void *)netmsg;
966 	int error;
967 
968 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
969 				  msg->nm_sin, msg->nm_ifsin);
970 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
971 }
972 
973 struct netmsg_tcp6_connect {
974 	struct netmsg		nm_netmsg;
975 	struct tcpcb		*nm_tp;
976 	struct sockaddr_in6	*nm_sin6;
977 	struct in6_addr		*nm_addr6;
978 	int			nm_flags;
979 	struct mbuf		*nm_m;
980 };
981 
982 static void
983 tcp6_connect_handler(netmsg_t netmsg)
984 {
985 	struct netmsg_tcp6_connect *msg = (void *)netmsg;
986 	int error;
987 
988 	error = tcp6_connect_oncpu(msg->nm_tp, msg->nm_flags, msg->nm_m,
989 				   msg->nm_sin6, msg->nm_addr6);
990 	lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, error);
991 }
992 
993 #endif
994 
995 /*
996  * Common subroutine to open a TCP connection to remote host specified
997  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
998  * port number if needed.  Call in_pcbladdr to do the routing and to choose
999  * a local host address (interface).
1000  * Initialize connection parameters and enter SYN-SENT state.
1001  */
1002 static int
1003 tcp_connect(struct tcpcb *tp, int flags, struct mbuf *m,
1004 	    struct sockaddr *nam, struct thread *td)
1005 {
1006 	struct inpcb *inp = tp->t_inpcb;
1007 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1008 	struct sockaddr_in *if_sin;
1009 	int error;
1010 #ifdef SMP
1011 	lwkt_port_t port;
1012 #endif
1013 
1014 	/*
1015 	 * Bind if we have to
1016 	 */
1017 	if (inp->inp_lport == 0) {
1018 		error = in_pcbbind(inp, NULL, td);
1019 		if (error) {
1020 			m_freem(m);
1021 			return (error);
1022 		}
1023 	}
1024 
1025 	/*
1026 	 * Calculate the correct protocol processing thread.  The connect
1027 	 * operation must run there.
1028 	 */
1029 	error = in_pcbladdr(inp, nam, &if_sin, td);
1030 	if (error) {
1031 		m_freem(m);
1032 		return (error);
1033 	}
1034 
1035 #ifdef SMP
1036 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1037 	    inp->inp_laddr.s_addr ?
1038 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
1039 	    inp->inp_lport);
1040 
1041 	if (port != &curthread->td_msgport) {
1042 		struct netmsg_tcp_connect msg;
1043 		struct route *ro = &inp->inp_route;
1044 
1045 		/*
1046 		 * in_pcbladdr() may have allocated a route entry for us
1047 		 * on the current CPU, but we need a route entry on the
1048 		 * inpcb's owner CPU, so free it here.
1049 		 */
1050 		if (ro->ro_rt != NULL)
1051 			RTFREE(ro->ro_rt);
1052 		bzero(ro, sizeof(*ro));
1053 
1054 		/*
1055 		 * NOTE: We haven't set so->so_port yet do not pass so
1056 		 *	 to netmsg_init() or it will be improperly forwarded.
1057 		 */
1058 		netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1059 			    0, tcp_connect_handler);
1060 		msg.nm_tp = tp;
1061 		msg.nm_sin = sin;
1062 		msg.nm_ifsin = if_sin;
1063 		msg.nm_flags = flags;
1064 		msg.nm_m = m;
1065 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1066 	} else {
1067 		error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1068 	}
1069 #else
1070 	error = tcp_connect_oncpu(tp, flags, m, sin, if_sin);
1071 #endif
1072 	return (error);
1073 }
1074 
1075 #ifdef INET6
1076 
1077 static int
1078 tcp6_connect(struct tcpcb *tp, int flags, struct mbuf *m,
1079 	     struct sockaddr *nam, struct thread *td)
1080 {
1081 	struct inpcb *inp = tp->t_inpcb;
1082 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1083 	struct in6_addr *addr6;
1084 #ifdef SMP
1085 	lwkt_port_t port;
1086 #endif
1087 	int error;
1088 
1089 	if (inp->inp_lport == 0) {
1090 		error = in6_pcbbind(inp, NULL, td);
1091 		if (error) {
1092 			m_freem(m);
1093 			return (error);
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * Cannot simply call in_pcbconnect, because there might be an
1099 	 * earlier incarnation of this same connection still in
1100 	 * TIME_WAIT state, creating an ADDRINUSE error.
1101 	 */
1102 	error = in6_pcbladdr(inp, nam, &addr6, td);
1103 	if (error) {
1104 		m_freem(m);
1105 		return (error);
1106 	}
1107 
1108 #ifdef SMP
1109 	port = tcp6_addrport();	/* XXX hack for now, always cpu0 */
1110 
1111 	if (port != &curthread->td_msgport) {
1112 		struct netmsg_tcp6_connect msg;
1113 		struct route *ro = &inp->inp_route;
1114 
1115 		/*
1116 		 * in_pcbladdr() may have allocated a route entry for us
1117 		 * on the current CPU, but we need a route entry on the
1118 		 * inpcb's owner CPU, so free it here.
1119 		 */
1120 		if (ro->ro_rt != NULL)
1121 			RTFREE(ro->ro_rt);
1122 		bzero(ro, sizeof(*ro));
1123 
1124 		netmsg_init(&msg.nm_netmsg, NULL, &curthread->td_msgport,
1125 			    0, tcp6_connect_handler);
1126 		msg.nm_tp = tp;
1127 		msg.nm_sin6 = sin6;
1128 		msg.nm_addr6 = addr6;
1129 		msg.nm_flags = flags;
1130 		msg.nm_m = m;
1131 		error = lwkt_domsg(port, &msg.nm_netmsg.nm_lmsg, 0);
1132 	} else {
1133 		error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1134 	}
1135 #else
1136 	error = tcp6_connect_oncpu(tp, flags, m, sin6, addr6);
1137 #endif
1138 	return (error);
1139 }
1140 
1141 static int
1142 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
1143 		   struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1144 {
1145 	struct inpcb *inp = tp->t_inpcb;
1146 	struct socket *so = inp->inp_socket;
1147 	struct inpcb *oinp;
1148 
1149 	/*
1150 	 * Cannot simply call in_pcbconnect, because there might be an
1151 	 * earlier incarnation of this same connection still in
1152 	 * TIME_WAIT state, creating an ADDRINUSE error.
1153 	 */
1154 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1155 				  &sin6->sin6_addr, sin6->sin6_port,
1156 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1157 				      addr6 : &inp->in6p_laddr,
1158 				  inp->inp_lport,  0, NULL);
1159 	if (oinp) {
1160 		m_freem(m);
1161 		return (EADDRINUSE);
1162 	}
1163 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1164 		inp->in6p_laddr = *addr6;
1165 	inp->in6p_faddr = sin6->sin6_addr;
1166 	inp->inp_fport = sin6->sin6_port;
1167 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1168 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1169 	in_pcbinsconnhash(inp);
1170 
1171 	/*
1172 	 * Now that no more errors can occur, change the protocol processing
1173 	 * port to the current thread (which is the correct thread).
1174 	 *
1175 	 * Create TCP timer message now; we are on the tcpcb's owner
1176 	 * CPU/thread.
1177 	 */
1178 	sosetport(so, &curthread->td_msgport);
1179 	tcp_create_timermsg(tp, &curthread->td_msgport);
1180 
1181 	/* Compute window scaling to request.  */
1182 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1183 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1184 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1185 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1186 		tp->request_r_scale++;
1187 	}
1188 
1189 	soisconnecting(so);
1190 	tcpstat.tcps_connattempt++;
1191 	tp->t_state = TCPS_SYN_SENT;
1192 	tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1193 	tp->iss = tcp_new_isn(tp);
1194 	tcp_sendseqinit(tp);
1195 	if (m) {
1196 		ssb_appendstream(&so->so_snd, m);
1197 		m = NULL;
1198 		if (flags & PRUS_OOB)
1199 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1200 	}
1201 
1202 	/*
1203 	 * Close the send side of the connection after
1204 	 * the data is sent if flagged.
1205 	 */
1206 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1207 		socantsendmore(so);
1208 		tp = tcp_usrclosed(tp);
1209 	}
1210 	return (tcp_output(tp));
1211 }
1212 
1213 #endif /* INET6 */
1214 
1215 /*
1216  * The new sockopt interface makes it possible for us to block in the
1217  * copyin/out step (if we take a page fault).  Taking a page fault while
1218  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1219  * both now use TSM, there probably isn't any need for this function to
1220  * run in a critical section any more.  This needs more examination.)
1221  */
1222 int
1223 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1224 {
1225 	int	error, opt, optval;
1226 	struct	inpcb *inp;
1227 	struct	tcpcb *tp;
1228 
1229 	error = 0;
1230 	crit_enter();		/* XXX */
1231 	inp = so->so_pcb;
1232 	if (inp == NULL) {
1233 		crit_exit();
1234 		return (ECONNRESET);
1235 	}
1236 	if (sopt->sopt_level != IPPROTO_TCP) {
1237 #ifdef INET6
1238 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1239 			error = ip6_ctloutput(so, sopt);
1240 		else
1241 #endif /* INET6 */
1242 		error = ip_ctloutput(so, sopt);
1243 		crit_exit();
1244 		return (error);
1245 	}
1246 	tp = intotcpcb(inp);
1247 
1248 	switch (sopt->sopt_dir) {
1249 	case SOPT_SET:
1250 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1251 				      sizeof optval);
1252 		if (error)
1253 			break;
1254 		switch (sopt->sopt_name) {
1255 		case TCP_NODELAY:
1256 		case TCP_NOOPT:
1257 			switch (sopt->sopt_name) {
1258 			case TCP_NODELAY:
1259 				opt = TF_NODELAY;
1260 				break;
1261 			case TCP_NOOPT:
1262 				opt = TF_NOOPT;
1263 				break;
1264 			default:
1265 				opt = 0; /* dead code to fool gcc */
1266 				break;
1267 			}
1268 
1269 			if (optval)
1270 				tp->t_flags |= opt;
1271 			else
1272 				tp->t_flags &= ~opt;
1273 			break;
1274 
1275 		case TCP_NOPUSH:
1276 			if (optval)
1277 				tp->t_flags |= TF_NOPUSH;
1278 			else {
1279 				tp->t_flags &= ~TF_NOPUSH;
1280 				error = tcp_output(tp);
1281 			}
1282 			break;
1283 
1284 		case TCP_MAXSEG:
1285 			/*
1286 			 * Must be between 0 and maxseg.  If the requested
1287 			 * maxseg is too small to satisfy the desired minmss,
1288 			 * pump it up (silently so sysctl modifications of
1289 			 * minmss do not create unexpected program failures).
1290 			 * Handle degenerate cases.
1291 			 */
1292 			if (optval > 0 && optval <= tp->t_maxseg) {
1293 				if (optval + 40 < tcp_minmss) {
1294 					optval = tcp_minmss - 40;
1295 					if (optval < 0)
1296 						optval = 1;
1297 				}
1298 				tp->t_maxseg = optval;
1299 			} else {
1300 				error = EINVAL;
1301 			}
1302 			break;
1303 
1304 		default:
1305 			error = ENOPROTOOPT;
1306 			break;
1307 		}
1308 		break;
1309 
1310 	case SOPT_GET:
1311 		switch (sopt->sopt_name) {
1312 		case TCP_NODELAY:
1313 			optval = tp->t_flags & TF_NODELAY;
1314 			break;
1315 		case TCP_MAXSEG:
1316 			optval = tp->t_maxseg;
1317 			break;
1318 		case TCP_NOOPT:
1319 			optval = tp->t_flags & TF_NOOPT;
1320 			break;
1321 		case TCP_NOPUSH:
1322 			optval = tp->t_flags & TF_NOPUSH;
1323 			break;
1324 		default:
1325 			error = ENOPROTOOPT;
1326 			break;
1327 		}
1328 		if (error == 0)
1329 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1330 		break;
1331 	}
1332 	crit_exit();
1333 	return (error);
1334 }
1335 
1336 /*
1337  * tcp_sendspace and tcp_recvspace are the default send and receive window
1338  * sizes, respectively.  These are obsolescent (this information should
1339  * be set by the route).
1340  *
1341  * Use a default that does not require tcp window scaling to be turned
1342  * on.  Individual programs or the administrator can increase the default.
1343  */
1344 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1345 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1346     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1347 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1348 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1349     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1350 
1351 /*
1352  * Attach TCP protocol to socket, allocating
1353  * internet protocol control block, tcp control block,
1354  * bufer space, and entering LISTEN state if to accept connections.
1355  */
1356 static int
1357 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1358 {
1359 	struct tcpcb *tp;
1360 	struct inpcb *inp;
1361 	int error;
1362 	int cpu;
1363 #ifdef INET6
1364 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1365 #endif
1366 
1367 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1368 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1369 				  ai->sb_rlimit);
1370 		if (error)
1371 			return (error);
1372 	}
1373 	so->so_rcv.ssb_flags |= SSB_AUTOSIZE;
1374 	so->so_snd.ssb_flags |= SSB_AUTOSIZE;
1375 	cpu = mycpu->gd_cpuid;
1376 	error = in_pcballoc(so, &tcbinfo[cpu]);
1377 	if (error)
1378 		return (error);
1379 	inp = so->so_pcb;
1380 #ifdef INET6
1381 	if (isipv6) {
1382 		inp->inp_vflag |= INP_IPV6;
1383 		inp->in6p_hops = -1;	/* use kernel default */
1384 	}
1385 	else
1386 #endif
1387 	inp->inp_vflag |= INP_IPV4;
1388 	tp = tcp_newtcpcb(inp);
1389 	if (tp == 0) {
1390 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1391 
1392 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1393 #ifdef INET6
1394 		if (isipv6)
1395 			in6_pcbdetach(inp);
1396 		else
1397 #endif
1398 		in_pcbdetach(inp);
1399 		so->so_state |= nofd;
1400 		return (ENOBUFS);
1401 	}
1402 	tp->t_state = TCPS_CLOSED;
1403 	so->so_port = tcp_soport_attach(so);
1404 	return (0);
1405 }
1406 
1407 /*
1408  * Initiate (or continue) disconnect.
1409  * If embryonic state, just send reset (once).
1410  * If in ``let data drain'' option and linger null, just drop.
1411  * Otherwise (hard), mark socket disconnecting and drop
1412  * current input data; switch states based on user close, and
1413  * send segment to peer (with FIN).
1414  */
1415 static struct tcpcb *
1416 tcp_disconnect(struct tcpcb *tp)
1417 {
1418 	struct socket *so = tp->t_inpcb->inp_socket;
1419 
1420 	if (tp->t_state < TCPS_ESTABLISHED)
1421 		tp = tcp_close(tp);
1422 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1423 		tp = tcp_drop(tp, 0);
1424 	else {
1425 		soisdisconnecting(so);
1426 		sbflush(&so->so_rcv.sb);
1427 		tp = tcp_usrclosed(tp);
1428 		if (tp)
1429 			tcp_output(tp);
1430 	}
1431 	return (tp);
1432 }
1433 
1434 /*
1435  * User issued close, and wish to trail through shutdown states:
1436  * if never received SYN, just forget it.  If got a SYN from peer,
1437  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1438  * If already got a FIN from peer, then almost done; go to LAST_ACK
1439  * state.  In all other cases, have already sent FIN to peer (e.g.
1440  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1441  * for peer to send FIN or not respond to keep-alives, etc.
1442  * We can let the user exit from the close as soon as the FIN is acked.
1443  */
1444 static struct tcpcb *
1445 tcp_usrclosed(struct tcpcb *tp)
1446 {
1447 
1448 	switch (tp->t_state) {
1449 
1450 	case TCPS_CLOSED:
1451 	case TCPS_LISTEN:
1452 		tp->t_state = TCPS_CLOSED;
1453 		tp = tcp_close(tp);
1454 		break;
1455 
1456 	case TCPS_SYN_SENT:
1457 	case TCPS_SYN_RECEIVED:
1458 		tp->t_flags |= TF_NEEDFIN;
1459 		break;
1460 
1461 	case TCPS_ESTABLISHED:
1462 		tp->t_state = TCPS_FIN_WAIT_1;
1463 		break;
1464 
1465 	case TCPS_CLOSE_WAIT:
1466 		tp->t_state = TCPS_LAST_ACK;
1467 		break;
1468 	}
1469 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1470 		soisdisconnected(tp->t_inpcb->inp_socket);
1471 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1472 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1473 			tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,
1474 			    tcp_timer_2msl);
1475 		}
1476 	}
1477 	return (tp);
1478 }
1479