xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision c5541aee854b0d32586182b733a9ea4d4c92168b)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
34  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
35  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.11 2004/04/05 17:47:01 dillon Exp $
36  */
37 
38 #include "opt_ipsec.h"
39 #include "opt_inet6.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/globaldata.h>
47 #include <sys/thread.h>
48 
49 #include <sys/mbuf.h>
50 #ifdef INET6
51 #include <sys/domain.h>
52 #endif /* INET6 */
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 
57 #include <net/if.h>
58 #include <net/route.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #ifdef INET6
63 #include <netinet/ip6.h>
64 #endif
65 #include <netinet/in_pcb.h>
66 #ifdef INET6
67 #include <netinet6/in6_pcb.h>
68 #endif
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #ifdef INET6
72 #include <netinet6/ip6_var.h>
73 #endif
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #include <netinet/tcpip.h>
80 #ifdef TCPDEBUG
81 #include <netinet/tcp_debug.h>
82 #endif
83 
84 #ifdef IPSEC
85 #include <netinet6/ipsec.h>
86 #endif /*IPSEC*/
87 
88 /*
89  * TCP protocol interface to socket abstraction.
90  */
91 extern	char *tcpstates[];	/* XXX ??? */
92 
93 static int	tcp_attach (struct socket *, struct pru_attach_info *);
94 static int	tcp_connect (struct tcpcb *, struct sockaddr *,
95 				 struct thread *);
96 #ifdef INET6
97 static int	tcp6_connect (struct tcpcb *, struct sockaddr *,
98 				 struct thread *);
99 #endif /* INET6 */
100 static struct tcpcb *
101 		tcp_disconnect (struct tcpcb *);
102 static struct tcpcb *
103 		tcp_usrclosed (struct tcpcb *);
104 
105 #ifdef TCPDEBUG
106 #define	TCPDEBUG0	int ostate = 0
107 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
108 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
109 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
110 #else
111 #define	TCPDEBUG0
112 #define	TCPDEBUG1()
113 #define	TCPDEBUG2(req)
114 #endif
115 
116 /*
117  * TCP attaches to socket via pru_attach(), reserving space,
118  * and an internet control block.
119  */
120 static int
121 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
122 {
123 	int s = splnet();
124 	int error;
125 	struct inpcb *inp = sotoinpcb(so);
126 	struct tcpcb *tp = 0;
127 	TCPDEBUG0;
128 
129 	TCPDEBUG1();
130 	if (inp) {
131 		error = EISCONN;
132 		goto out;
133 	}
134 
135 	error = tcp_attach(so, ai);
136 	if (error)
137 		goto out;
138 
139 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
140 		so->so_linger = TCP_LINGERTIME;
141 	tp = sototcpcb(so);
142 out:
143 	TCPDEBUG2(PRU_ATTACH);
144 	splx(s);
145 	return error;
146 }
147 
148 /*
149  * pru_detach() detaches the TCP protocol from the socket.
150  * If the protocol state is non-embryonic, then can't
151  * do this directly: have to initiate a pru_disconnect(),
152  * which may finish later; embryonic TCB's can just
153  * be discarded here.
154  */
155 static int
156 tcp_usr_detach(struct socket *so)
157 {
158 	int s = splnet();
159 	int error = 0;
160 	struct inpcb *inp = sotoinpcb(so);
161 	struct tcpcb *tp;
162 	TCPDEBUG0;
163 
164 	if (inp == 0) {
165 		splx(s);
166 		return EINVAL;	/* XXX */
167 	}
168 	tp = intotcpcb(inp);
169 	TCPDEBUG1();
170 	tp = tcp_disconnect(tp);
171 
172 	TCPDEBUG2(PRU_DETACH);
173 	splx(s);
174 	return error;
175 }
176 
177 #define	COMMON_START()	TCPDEBUG0; \
178 			do { \
179 				     if (inp == 0) { \
180 					     splx(s); \
181 					     return EINVAL; \
182 				     } \
183 				     tp = intotcpcb(inp); \
184 				     TCPDEBUG1(); \
185 		     } while(0)
186 
187 #define COMMON_END(req)	out: TCPDEBUG2(req); splx(s); return error; goto out
188 
189 
190 /*
191  * Give the socket an address.
192  */
193 static int
194 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
195 {
196 	int s = splnet();
197 	int error = 0;
198 	struct inpcb *inp = sotoinpcb(so);
199 	struct tcpcb *tp;
200 	struct sockaddr_in *sinp;
201 
202 	COMMON_START();
203 
204 	/*
205 	 * Must check for multicast addresses and disallow binding
206 	 * to them.
207 	 */
208 	sinp = (struct sockaddr_in *)nam;
209 	if (sinp->sin_family == AF_INET &&
210 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
211 		error = EAFNOSUPPORT;
212 		goto out;
213 	}
214 	error = in_pcbbind(inp, nam, td);
215 	if (error)
216 		goto out;
217 	COMMON_END(PRU_BIND);
218 
219 }
220 
221 #ifdef INET6
222 static int
223 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
224 {
225 	int s = splnet();
226 	int error = 0;
227 	struct inpcb *inp = sotoinpcb(so);
228 	struct tcpcb *tp;
229 	struct sockaddr_in6 *sin6p;
230 
231 	COMMON_START();
232 
233 	/*
234 	 * Must check for multicast addresses and disallow binding
235 	 * to them.
236 	 */
237 	sin6p = (struct sockaddr_in6 *)nam;
238 	if (sin6p->sin6_family == AF_INET6 &&
239 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
240 		error = EAFNOSUPPORT;
241 		goto out;
242 	}
243 	inp->inp_vflag &= ~INP_IPV4;
244 	inp->inp_vflag |= INP_IPV6;
245 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
246 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
247 			inp->inp_vflag |= INP_IPV4;
248 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
249 			struct sockaddr_in sin;
250 
251 			in6_sin6_2_sin(&sin, sin6p);
252 			inp->inp_vflag |= INP_IPV4;
253 			inp->inp_vflag &= ~INP_IPV6;
254 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
255 			goto out;
256 		}
257 	}
258 	error = in6_pcbbind(inp, nam, td);
259 	if (error)
260 		goto out;
261 	COMMON_END(PRU_BIND);
262 }
263 #endif /* INET6 */
264 
265 /*
266  * Prepare to accept connections.
267  */
268 static int
269 tcp_usr_listen(struct socket *so, struct thread *td)
270 {
271 	int s = splnet();
272 	int error = 0;
273 	struct inpcb *inp = sotoinpcb(so);
274 	struct tcpcb *tp;
275 
276 	COMMON_START();
277 	if (inp->inp_lport == 0)
278 		error = in_pcbbind(inp, (struct sockaddr *)0, td);
279 	if (error == 0) {
280 		tp->t_state = TCPS_LISTEN;
281 		in_pcbinswildcardhash(inp);
282 	}
283 	COMMON_END(PRU_LISTEN);
284 }
285 
286 #ifdef INET6
287 static int
288 tcp6_usr_listen(struct socket *so, struct thread *td)
289 {
290 	int s = splnet();
291 	int error = 0;
292 	struct inpcb *inp = sotoinpcb(so);
293 	struct tcpcb *tp;
294 
295 	COMMON_START();
296 	if (inp->inp_lport == 0) {
297 		inp->inp_vflag &= ~INP_IPV4;
298 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
299 			inp->inp_vflag |= INP_IPV4;
300 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
301 	}
302 	if (error == 0)
303 		tp->t_state = TCPS_LISTEN;
304 	COMMON_END(PRU_LISTEN);
305 }
306 #endif /* INET6 */
307 
308 /*
309  * Initiate connection to peer.
310  * Create a template for use in transmissions on this connection.
311  * Enter SYN_SENT state, and mark socket as connecting.
312  * Start keep-alive timer, and seed output sequence space.
313  * Send initial segment on connection.
314  */
315 static int
316 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
317 {
318 	int s = splnet();
319 	int error = 0;
320 	struct inpcb *inp = sotoinpcb(so);
321 	struct tcpcb *tp;
322 	struct sockaddr_in *sinp;
323 
324 	COMMON_START();
325 
326 	/*
327 	 * Must disallow TCP ``connections'' to multicast addresses.
328 	 */
329 	sinp = (struct sockaddr_in *)nam;
330 	if (sinp->sin_family == AF_INET
331 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
332 		error = EAFNOSUPPORT;
333 		goto out;
334 	}
335 
336 	prison_remote_ip(td, 0, &sinp->sin_addr.s_addr);
337 
338 	if ((error = tcp_connect(tp, nam, td)) != 0)
339 		goto out;
340 	error = tcp_output(tp);
341 	COMMON_END(PRU_CONNECT);
342 }
343 
344 #ifdef INET6
345 static int
346 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
347 {
348 	int s = splnet();
349 	int error = 0;
350 	struct inpcb *inp = sotoinpcb(so);
351 	struct tcpcb *tp;
352 	struct sockaddr_in6 *sin6p;
353 
354 	COMMON_START();
355 
356 	/*
357 	 * Must disallow TCP ``connections'' to multicast addresses.
358 	 */
359 	sin6p = (struct sockaddr_in6 *)nam;
360 	if (sin6p->sin6_family == AF_INET6
361 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
362 		error = EAFNOSUPPORT;
363 		goto out;
364 	}
365 
366 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
367 		struct sockaddr_in sin;
368 
369 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
370 			error = EINVAL;
371 			goto out;
372 		}
373 
374 		in6_sin6_2_sin(&sin, sin6p);
375 		inp->inp_vflag |= INP_IPV4;
376 		inp->inp_vflag &= ~INP_IPV6;
377 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
378 			goto out;
379 		error = tcp_output(tp);
380 		goto out;
381 	}
382 	inp->inp_vflag &= ~INP_IPV4;
383 	inp->inp_vflag |= INP_IPV6;
384 	inp->inp_inc.inc_isipv6 = 1;
385 	if ((error = tcp6_connect(tp, nam, td)) != 0)
386 		goto out;
387 	error = tcp_output(tp);
388 	COMMON_END(PRU_CONNECT);
389 }
390 #endif /* INET6 */
391 
392 /*
393  * Initiate disconnect from peer.
394  * If connection never passed embryonic stage, just drop;
395  * else if don't need to let data drain, then can just drop anyways,
396  * else have to begin TCP shutdown process: mark socket disconnecting,
397  * drain unread data, state switch to reflect user close, and
398  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
399  * when peer sends FIN and acks ours.
400  *
401  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
402  */
403 static int
404 tcp_usr_disconnect(struct socket *so)
405 {
406 	int s = splnet();
407 	int error = 0;
408 	struct inpcb *inp = sotoinpcb(so);
409 	struct tcpcb *tp;
410 
411 	COMMON_START();
412 	tp = tcp_disconnect(tp);
413 	COMMON_END(PRU_DISCONNECT);
414 }
415 
416 /*
417  * Accept a connection.  Essentially all the work is
418  * done at higher levels; just return the address
419  * of the peer, storing through addr.
420  */
421 static int
422 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
423 {
424 	int s = splnet();
425 	int error = 0;
426 	struct inpcb *inp = sotoinpcb(so);
427 	struct tcpcb *tp = NULL;
428 	TCPDEBUG0;
429 
430 	if (so->so_state & SS_ISDISCONNECTED) {
431 		error = ECONNABORTED;
432 		goto out;
433 	}
434 	if (inp == 0) {
435 		splx(s);
436 		return (EINVAL);
437 	}
438 	tp = intotcpcb(inp);
439 	TCPDEBUG1();
440 	in_setpeeraddr(so, nam);
441 	COMMON_END(PRU_ACCEPT);
442 }
443 
444 #ifdef INET6
445 static int
446 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
447 {
448 	int s = splnet();
449 	int error = 0;
450 	struct inpcb *inp = sotoinpcb(so);
451 	struct tcpcb *tp = NULL;
452 	TCPDEBUG0;
453 
454 	if (so->so_state & SS_ISDISCONNECTED) {
455 		error = ECONNABORTED;
456 		goto out;
457 	}
458 	if (inp == 0) {
459 		splx(s);
460 		return (EINVAL);
461 	}
462 	tp = intotcpcb(inp);
463 	TCPDEBUG1();
464 	in6_mapped_peeraddr(so, nam);
465 	COMMON_END(PRU_ACCEPT);
466 }
467 #endif /* INET6 */
468 /*
469  * Mark the connection as being incapable of further output.
470  */
471 static int
472 tcp_usr_shutdown(struct socket *so)
473 {
474 	int s = splnet();
475 	int error = 0;
476 	struct inpcb *inp = sotoinpcb(so);
477 	struct tcpcb *tp;
478 
479 	COMMON_START();
480 	socantsendmore(so);
481 	tp = tcp_usrclosed(tp);
482 	if (tp)
483 		error = tcp_output(tp);
484 	COMMON_END(PRU_SHUTDOWN);
485 }
486 
487 /*
488  * After a receive, possibly send window update to peer.
489  */
490 static int
491 tcp_usr_rcvd(struct socket *so, int flags)
492 {
493 	int s = splnet();
494 	int error = 0;
495 	struct inpcb *inp = sotoinpcb(so);
496 	struct tcpcb *tp;
497 
498 	COMMON_START();
499 	tcp_output(tp);
500 	COMMON_END(PRU_RCVD);
501 }
502 
503 /*
504  * Do a send by putting data in output queue and updating urgent
505  * marker if URG set.  Possibly send more data.  Unlike the other
506  * pru_*() routines, the mbuf chains are our responsibility.  We
507  * must either enqueue them or free them.  The other pru_* routines
508  * generally are caller-frees.
509  */
510 static int
511 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
512 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
513 {
514 	int s = splnet();
515 	int error = 0;
516 	struct inpcb *inp = sotoinpcb(so);
517 	struct tcpcb *tp;
518 #ifdef INET6
519 	int isipv6;
520 #endif
521 	TCPDEBUG0;
522 
523 	if (inp == NULL) {
524 		/*
525 		 * OOPS! we lost a race, the TCP session got reset after
526 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
527 		 * network interrupt in the non-splnet() section of sosend().
528 		 */
529 		if (m)
530 			m_freem(m);
531 		if (control)
532 			m_freem(control);
533 		error = ECONNRESET;	/* XXX EPIPE? */
534 		tp = NULL;
535 		TCPDEBUG1();
536 		goto out;
537 	}
538 #ifdef INET6
539 	isipv6 = nam && nam->sa_family == AF_INET6;
540 #endif /* INET6 */
541 	tp = intotcpcb(inp);
542 	TCPDEBUG1();
543 	if (control) {
544 		/* TCP doesn't do control messages (rights, creds, etc) */
545 		if (control->m_len) {
546 			m_freem(control);
547 			if (m)
548 				m_freem(m);
549 			error = EINVAL;
550 			goto out;
551 		}
552 		m_freem(control);	/* empty control, just free it */
553 	}
554 	if(!(flags & PRUS_OOB)) {
555 		sbappend(&so->so_snd, m);
556 		if (nam && tp->t_state < TCPS_SYN_SENT) {
557 			/*
558 			 * Do implied connect if not yet connected,
559 			 * initialize window to default value, and
560 			 * initialize maxseg/maxopd using peer's cached
561 			 * MSS.
562 			 */
563 #ifdef INET6
564 			if (isipv6)
565 				error = tcp6_connect(tp, nam, td);
566 			else
567 #endif /* INET6 */
568 			error = tcp_connect(tp, nam, td);
569 			if (error)
570 				goto out;
571 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
572 			tcp_mss(tp, -1);
573 		}
574 
575 		if (flags & PRUS_EOF) {
576 			/*
577 			 * Close the send side of the connection after
578 			 * the data is sent.
579 			 */
580 			socantsendmore(so);
581 			tp = tcp_usrclosed(tp);
582 		}
583 		if (tp != NULL) {
584 			if (flags & PRUS_MORETOCOME)
585 				tp->t_flags |= TF_MORETOCOME;
586 			error = tcp_output(tp);
587 			if (flags & PRUS_MORETOCOME)
588 				tp->t_flags &= ~TF_MORETOCOME;
589 		}
590 	} else {
591 		if (sbspace(&so->so_snd) < -512) {
592 			m_freem(m);
593 			error = ENOBUFS;
594 			goto out;
595 		}
596 		/*
597 		 * According to RFC961 (Assigned Protocols),
598 		 * the urgent pointer points to the last octet
599 		 * of urgent data.  We continue, however,
600 		 * to consider it to indicate the first octet
601 		 * of data past the urgent section.
602 		 * Otherwise, snd_up should be one lower.
603 		 */
604 		sbappend(&so->so_snd, m);
605 		if (nam && tp->t_state < TCPS_SYN_SENT) {
606 			/*
607 			 * Do implied connect if not yet connected,
608 			 * initialize window to default value, and
609 			 * initialize maxseg/maxopd using peer's cached
610 			 * MSS.
611 			 */
612 #ifdef INET6
613 			if (isipv6)
614 				error = tcp6_connect(tp, nam, td);
615 			else
616 #endif /* INET6 */
617 			error = tcp_connect(tp, nam, td);
618 			if (error)
619 				goto out;
620 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
621 			tcp_mss(tp, -1);
622 		}
623 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
624 		tp->t_force = 1;
625 		error = tcp_output(tp);
626 		tp->t_force = 0;
627 	}
628 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
629 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
630 }
631 
632 /*
633  * Abort the TCP.
634  */
635 static int
636 tcp_usr_abort(struct socket *so)
637 {
638 	int s = splnet();
639 	int error = 0;
640 	struct inpcb *inp = sotoinpcb(so);
641 	struct tcpcb *tp;
642 
643 	COMMON_START();
644 	tp = tcp_drop(tp, ECONNABORTED);
645 	COMMON_END(PRU_ABORT);
646 }
647 
648 /*
649  * Receive out-of-band data.
650  */
651 static int
652 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
653 {
654 	int s = splnet();
655 	int error = 0;
656 	struct inpcb *inp = sotoinpcb(so);
657 	struct tcpcb *tp;
658 
659 	COMMON_START();
660 	if ((so->so_oobmark == 0 &&
661 	     (so->so_state & SS_RCVATMARK) == 0) ||
662 	    so->so_options & SO_OOBINLINE ||
663 	    tp->t_oobflags & TCPOOB_HADDATA) {
664 		error = EINVAL;
665 		goto out;
666 	}
667 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
668 		error = EWOULDBLOCK;
669 		goto out;
670 	}
671 	m->m_len = 1;
672 	*mtod(m, caddr_t) = tp->t_iobc;
673 	if ((flags & MSG_PEEK) == 0)
674 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
675 	COMMON_END(PRU_RCVOOB);
676 }
677 
678 /* xxx - should be const */
679 struct pr_usrreqs tcp_usrreqs = {
680 	tcp_usr_abort, tcp_usr_accept, tcp_usr_attach, tcp_usr_bind,
681 	tcp_usr_connect, pru_connect2_notsupp, in_control, tcp_usr_detach,
682 	tcp_usr_disconnect, tcp_usr_listen, in_setpeeraddr, tcp_usr_rcvd,
683 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
684 	in_setsockaddr, sosend, soreceive, sopoll
685 };
686 
687 #ifdef INET6
688 struct pr_usrreqs tcp6_usrreqs = {
689 	tcp_usr_abort, tcp6_usr_accept, tcp_usr_attach, tcp6_usr_bind,
690 	tcp6_usr_connect, pru_connect2_notsupp, in6_control, tcp_usr_detach,
691 	tcp_usr_disconnect, tcp6_usr_listen, in6_mapped_peeraddr, tcp_usr_rcvd,
692 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
693 	in6_mapped_sockaddr, sosend, soreceive, sopoll
694 };
695 #endif /* INET6 */
696 
697 /*
698  * Common subroutine to open a TCP connection to remote host specified
699  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
700  * port number if needed.  Call in_pcbladdr to do the routing and to choose
701  * a local host address (interface).  If there is an existing incarnation
702  * of the same connection in TIME-WAIT state and if the remote host was
703  * sending CC options and if the connection duration was < MSL, then
704  * truncate the previous TIME-WAIT state and proceed.
705  * Initialize connection parameters and enter SYN-SENT state.
706  */
707 static int
708 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
709 {
710 	struct inpcb *inp = tp->t_inpcb, *oinp;
711 	struct socket *so = inp->inp_socket;
712 	struct tcpcb *otp;
713 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
714 	struct sockaddr_in *if_sin;
715 	struct rmxp_tao *taop;
716 	struct rmxp_tao tao_noncached;
717 	int error;
718 
719 	if (inp->inp_lport == 0) {
720 		error = in_pcbbind(inp, (struct sockaddr *)0, td);
721 		if (error)
722 			return (error);
723 	}
724 
725 	/*
726 	 * Cannot simply call in_pcbconnect, because there might be an
727 	 * earlier incarnation of this same connection still in
728 	 * TIME_WAIT state, creating an ADDRINUSE error.
729 	 */
730 	error = in_pcbladdr(inp, nam, &if_sin);
731 	if (error)
732 		return (error);
733 	oinp = in_pcblookup_hash(inp->inp_pcbinfo,
734 	    sin->sin_addr, sin->sin_port,
735 	    inp->inp_laddr.s_addr != INADDR_ANY ?
736 	        inp->inp_laddr : if_sin->sin_addr,
737 	    inp->inp_lport,  0, NULL);
738 	if (oinp != NULL) {
739 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
740 		    otp->t_state == TCPS_TIME_WAIT &&
741 		    (ticks - otp->t_starttime) < tcp_msl &&
742 		    (otp->t_flags & TF_RCVD_CC))
743 			(void) tcp_close(otp);
744 		else
745 			return (EADDRINUSE);
746 	}
747 	if (inp->inp_laddr.s_addr == INADDR_ANY)
748 		inp->inp_laddr = if_sin->sin_addr;
749 	inp->inp_faddr = sin->sin_addr;
750 	inp->inp_fport = sin->sin_port;
751 	in_pcbinsconnhash(inp);
752 
753 	/* Compute window scaling to request.  */
754 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
755 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
756 		tp->request_r_scale++;
757 
758 	soisconnecting(so);
759 	tcpstat.tcps_connattempt++;
760 	tp->t_state = TCPS_SYN_SENT;
761 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
762 	tp->iss = tcp_new_isn(tp);
763 	tp->t_bw_rtseq = tp->iss;
764 	tcp_sendseqinit(tp);
765 
766 	/*
767 	 * Generate a CC value for this connection and
768 	 * check whether CC or CCnew should be used.
769 	 */
770 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
771 		taop = &tao_noncached;
772 		bzero(taop, sizeof(*taop));
773 	}
774 
775 	tp->cc_send = CC_INC(tcp_ccgen);
776 	if (taop->tao_ccsent != 0 &&
777 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
778 		taop->tao_ccsent = tp->cc_send;
779 	} else {
780 		taop->tao_ccsent = 0;
781 		tp->t_flags |= TF_SENDCCNEW;
782 	}
783 
784 	return (0);
785 }
786 
787 #ifdef INET6
788 static int
789 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
790 {
791 	struct inpcb *inp = tp->t_inpcb, *oinp;
792 	struct socket *so = inp->inp_socket;
793 	struct tcpcb *otp;
794 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
795 	struct in6_addr *addr6;
796 	struct rmxp_tao *taop;
797 	struct rmxp_tao tao_noncached;
798 	int error;
799 
800 	if (inp->inp_lport == 0) {
801 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
802 		if (error)
803 			return error;
804 	}
805 
806 	/*
807 	 * Cannot simply call in_pcbconnect, because there might be an
808 	 * earlier incarnation of this same connection still in
809 	 * TIME_WAIT state, creating an ADDRINUSE error.
810 	 */
811 	error = in6_pcbladdr(inp, nam, &addr6);
812 	if (error)
813 		return error;
814 	oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
815 				  &sin6->sin6_addr, sin6->sin6_port,
816 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
817 				      addr6 : &inp->in6p_laddr,
818 				  inp->inp_lport,  0, NULL);
819 	if (oinp) {
820 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
821 		    otp->t_state == TCPS_TIME_WAIT &&
822 		    (ticks - otp->t_starttime) < tcp_msl &&
823 		    (otp->t_flags & TF_RCVD_CC))
824 			otp = tcp_close(otp);
825 		else
826 			return (EADDRINUSE);
827 	}
828 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
829 		inp->in6p_laddr = *addr6;
830 	inp->in6p_faddr = sin6->sin6_addr;
831 	inp->inp_fport = sin6->sin6_port;
832 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
833 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
834 	in_pcbinsconnhash(inp);
835 
836 	/* Compute window scaling to request.  */
837 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
838 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
839 		tp->request_r_scale++;
840 
841 	soisconnecting(so);
842 	tcpstat.tcps_connattempt++;
843 	tp->t_state = TCPS_SYN_SENT;
844 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
845 	tp->iss = tcp_new_isn(tp);
846 	tp->t_bw_rtseq = tp->iss;
847 	tcp_sendseqinit(tp);
848 
849 	/*
850 	 * Generate a CC value for this connection and
851 	 * check whether CC or CCnew should be used.
852 	 */
853 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
854 		taop = &tao_noncached;
855 		bzero(taop, sizeof(*taop));
856 	}
857 
858 	tp->cc_send = CC_INC(tcp_ccgen);
859 	if (taop->tao_ccsent != 0 &&
860 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
861 		taop->tao_ccsent = tp->cc_send;
862 	} else {
863 		taop->tao_ccsent = 0;
864 		tp->t_flags |= TF_SENDCCNEW;
865 	}
866 
867 	return (0);
868 }
869 #endif /* INET6 */
870 
871 /*
872  * The new sockopt interface makes it possible for us to block in the
873  * copyin/out step (if we take a page fault).  Taking a page fault at
874  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
875  * use TSM, there probably isn't any need for this function to run at
876  * splnet() any more.  This needs more examination.)
877  */
878 int
879 tcp_ctloutput(so, sopt)
880 	struct socket *so;
881 	struct sockopt *sopt;
882 {
883 	int	error, opt, optval, s;
884 	struct	inpcb *inp;
885 	struct	tcpcb *tp;
886 
887 	error = 0;
888 	s = splnet();		/* XXX */
889 	inp = sotoinpcb(so);
890 	if (inp == NULL) {
891 		splx(s);
892 		return (ECONNRESET);
893 	}
894 	if (sopt->sopt_level != IPPROTO_TCP) {
895 #ifdef INET6
896 		if (INP_CHECK_SOCKAF(so, AF_INET6))
897 			error = ip6_ctloutput(so, sopt);
898 		else
899 #endif /* INET6 */
900 		error = ip_ctloutput(so, sopt);
901 		splx(s);
902 		return (error);
903 	}
904 	tp = intotcpcb(inp);
905 
906 	switch (sopt->sopt_dir) {
907 	case SOPT_SET:
908 		switch (sopt->sopt_name) {
909 		case TCP_NODELAY:
910 		case TCP_NOOPT:
911 			error = sooptcopyin(sopt, &optval, sizeof optval,
912 					    sizeof optval);
913 			if (error)
914 				break;
915 
916 			switch (sopt->sopt_name) {
917 			case TCP_NODELAY:
918 				opt = TF_NODELAY;
919 				break;
920 			case TCP_NOOPT:
921 				opt = TF_NOOPT;
922 				break;
923 			default:
924 				opt = 0; /* dead code to fool gcc */
925 				break;
926 			}
927 
928 			if (optval)
929 				tp->t_flags |= opt;
930 			else
931 				tp->t_flags &= ~opt;
932 			break;
933 
934 		case TCP_NOPUSH:
935 			error = sooptcopyin(sopt, &optval, sizeof optval,
936 					    sizeof optval);
937 			if (error)
938 				break;
939 
940 			if (optval)
941 				tp->t_flags |= TF_NOPUSH;
942 			else {
943 				tp->t_flags &= ~TF_NOPUSH;
944 				error = tcp_output(tp);
945 			}
946 			break;
947 
948 		case TCP_MAXSEG:
949 			error = sooptcopyin(sopt, &optval, sizeof optval,
950 					    sizeof optval);
951 			if (error)
952 				break;
953 
954 			if (optval > 0 && optval <= tp->t_maxseg)
955 				tp->t_maxseg = optval;
956 			else
957 				error = EINVAL;
958 			break;
959 
960 		default:
961 			error = ENOPROTOOPT;
962 			break;
963 		}
964 		break;
965 
966 	case SOPT_GET:
967 		switch (sopt->sopt_name) {
968 		case TCP_NODELAY:
969 			optval = tp->t_flags & TF_NODELAY;
970 			break;
971 		case TCP_MAXSEG:
972 			optval = tp->t_maxseg;
973 			break;
974 		case TCP_NOOPT:
975 			optval = tp->t_flags & TF_NOOPT;
976 			break;
977 		case TCP_NOPUSH:
978 			optval = tp->t_flags & TF_NOPUSH;
979 			break;
980 		default:
981 			error = ENOPROTOOPT;
982 			break;
983 		}
984 		if (error == 0)
985 			error = sooptcopyout(sopt, &optval, sizeof optval);
986 		break;
987 	}
988 	splx(s);
989 	return (error);
990 }
991 
992 /*
993  * tcp_sendspace and tcp_recvspace are the default send and receive window
994  * sizes, respectively.  These are obsolescent (this information should
995  * be set by the route).
996  */
997 u_long	tcp_sendspace = 1024*32;
998 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
999     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1000 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1001 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1002     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1003 
1004 /*
1005  * Attach TCP protocol to socket, allocating
1006  * internet protocol control block, tcp control block,
1007  * bufer space, and entering LISTEN state if to accept connections.
1008  */
1009 static int
1010 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1011 {
1012 	struct tcpcb *tp;
1013 	struct inpcb *inp;
1014 	int error;
1015 	int cpu;
1016 #ifdef INET6
1017 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
1018 #endif
1019 
1020 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1021 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1022 				  ai->sb_rlimit);
1023 		if (error)
1024 			return (error);
1025 	}
1026 #ifdef TCP_DISTRIBUTED_TCBINFO
1027 	cpu = mycpu->gd_cpuid;
1028 #else
1029 	cpu = 0;
1030 #endif
1031 	error = in_pcballoc(so, &tcbinfo[cpu]);
1032 	if (error)
1033 		return (error);
1034 	inp = sotoinpcb(so);
1035 #ifdef INET6
1036 	if (isipv6) {
1037 		inp->inp_vflag |= INP_IPV6;
1038 		inp->in6p_hops = -1;	/* use kernel default */
1039 	}
1040 	else
1041 #endif
1042 	inp->inp_vflag |= INP_IPV4;
1043 	tp = tcp_newtcpcb(inp);
1044 	if (tp == 0) {
1045 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1046 
1047 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1048 #ifdef INET6
1049 		if (isipv6)
1050 			in6_pcbdetach(inp);
1051 		else
1052 #endif
1053 		in_pcbdetach(inp);
1054 		so->so_state |= nofd;
1055 		return (ENOBUFS);
1056 	}
1057 	tp->t_state = TCPS_CLOSED;
1058 	return (0);
1059 }
1060 
1061 /*
1062  * Initiate (or continue) disconnect.
1063  * If embryonic state, just send reset (once).
1064  * If in ``let data drain'' option and linger null, just drop.
1065  * Otherwise (hard), mark socket disconnecting and drop
1066  * current input data; switch states based on user close, and
1067  * send segment to peer (with FIN).
1068  */
1069 static struct tcpcb *
1070 tcp_disconnect(tp)
1071 	struct tcpcb *tp;
1072 {
1073 	struct socket *so = tp->t_inpcb->inp_socket;
1074 
1075 	if (tp->t_state < TCPS_ESTABLISHED)
1076 		tp = tcp_close(tp);
1077 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1078 		tp = tcp_drop(tp, 0);
1079 	else {
1080 		soisdisconnecting(so);
1081 		sbflush(&so->so_rcv);
1082 		tp = tcp_usrclosed(tp);
1083 		if (tp)
1084 			(void) tcp_output(tp);
1085 	}
1086 	return (tp);
1087 }
1088 
1089 /*
1090  * User issued close, and wish to trail through shutdown states:
1091  * if never received SYN, just forget it.  If got a SYN from peer,
1092  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1093  * If already got a FIN from peer, then almost done; go to LAST_ACK
1094  * state.  In all other cases, have already sent FIN to peer (e.g.
1095  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1096  * for peer to send FIN or not respond to keep-alives, etc.
1097  * We can let the user exit from the close as soon as the FIN is acked.
1098  */
1099 static struct tcpcb *
1100 tcp_usrclosed(tp)
1101 	struct tcpcb *tp;
1102 {
1103 
1104 	switch (tp->t_state) {
1105 
1106 	case TCPS_CLOSED:
1107 	case TCPS_LISTEN:
1108 		tp->t_state = TCPS_CLOSED;
1109 		tp = tcp_close(tp);
1110 		break;
1111 
1112 	case TCPS_SYN_SENT:
1113 	case TCPS_SYN_RECEIVED:
1114 		tp->t_flags |= TF_NEEDFIN;
1115 		break;
1116 
1117 	case TCPS_ESTABLISHED:
1118 		tp->t_state = TCPS_FIN_WAIT_1;
1119 		break;
1120 
1121 	case TCPS_CLOSE_WAIT:
1122 		tp->t_state = TCPS_LAST_ACK;
1123 		break;
1124 	}
1125 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1126 		soisdisconnected(tp->t_inpcb->inp_socket);
1127 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1128 		if (tp->t_state == TCPS_FIN_WAIT_2)
1129 			callout_reset(tp->tt_2msl, tcp_maxidle,
1130 				      tcp_timer_2msl, tp);
1131 	}
1132 	return (tp);
1133 }
1134 
1135