xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision bc76a771df54af7e361532b257cecc26227736b4)
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 1.15 2004/04/13 07:10:34 hsu 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/malloc.h>
46 #include <sys/sysctl.h>
47 #include <sys/globaldata.h>
48 #include <sys/thread.h>
49 
50 #include <sys/mbuf.h>
51 #ifdef INET6
52 #include <sys/domain.h>
53 #endif /* INET6 */
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 
58 #include <sys/msgport2.h>
59 
60 #include <net/if.h>
61 #include <net/netisr.h>
62 #include <net/route.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69 #include <netinet/in_pcb.h>
70 #ifdef INET6
71 #include <netinet6/in6_pcb.h>
72 #endif
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #endif
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet/tcpip.h>
84 #ifdef TCPDEBUG
85 #include <netinet/tcp_debug.h>
86 #endif
87 
88 #ifdef IPSEC
89 #include <netinet6/ipsec.h>
90 #endif /*IPSEC*/
91 
92 /*
93  * TCP protocol interface to socket abstraction.
94  */
95 extern	char *tcpstates[];	/* XXX ??? */
96 
97 static int	tcp_attach (struct socket *, struct pru_attach_info *);
98 static int	tcp_connect (struct tcpcb *, struct sockaddr *,
99 				 struct thread *);
100 #ifdef INET6
101 static int	tcp6_connect (struct tcpcb *, struct sockaddr *,
102 				 struct thread *);
103 #endif /* INET6 */
104 static struct tcpcb *
105 		tcp_disconnect (struct tcpcb *);
106 static struct tcpcb *
107 		tcp_usrclosed (struct tcpcb *);
108 
109 #ifdef TCPDEBUG
110 #define	TCPDEBUG0	int ostate = 0
111 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
112 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
113 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
114 #else
115 #define	TCPDEBUG0
116 #define	TCPDEBUG1()
117 #define	TCPDEBUG2(req)
118 #endif
119 
120 /*
121  * TCP attaches to socket via pru_attach(), reserving space,
122  * and an internet control block.
123  */
124 static int
125 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
126 {
127 	int s = splnet();
128 	int error;
129 	struct inpcb *inp = sotoinpcb(so);
130 	struct tcpcb *tp = 0;
131 	TCPDEBUG0;
132 
133 	TCPDEBUG1();
134 	if (inp) {
135 		error = EISCONN;
136 		goto out;
137 	}
138 
139 	error = tcp_attach(so, ai);
140 	if (error)
141 		goto out;
142 
143 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
144 		so->so_linger = TCP_LINGERTIME;
145 	tp = sototcpcb(so);
146 out:
147 	TCPDEBUG2(PRU_ATTACH);
148 	splx(s);
149 	return error;
150 }
151 
152 /*
153  * pru_detach() detaches the TCP protocol from the socket.
154  * If the protocol state is non-embryonic, then can't
155  * do this directly: have to initiate a pru_disconnect(),
156  * which may finish later; embryonic TCB's can just
157  * be discarded here.
158  */
159 static int
160 tcp_usr_detach(struct socket *so)
161 {
162 	int s = splnet();
163 	int error = 0;
164 	struct inpcb *inp = sotoinpcb(so);
165 	struct tcpcb *tp;
166 	TCPDEBUG0;
167 
168 	if (inp == 0) {
169 		splx(s);
170 		return EINVAL;	/* XXX */
171 	}
172 	tp = intotcpcb(inp);
173 	TCPDEBUG1();
174 	tp = tcp_disconnect(tp);
175 
176 	TCPDEBUG2(PRU_DETACH);
177 	splx(s);
178 	return error;
179 }
180 
181 #define	COMMON_START()	TCPDEBUG0; \
182 			do { \
183 				     if (inp == 0) { \
184 					     splx(s); \
185 					     return EINVAL; \
186 				     } \
187 				     tp = intotcpcb(inp); \
188 				     TCPDEBUG1(); \
189 		     } while(0)
190 
191 #define COMMON_END(req)	out: TCPDEBUG2(req); splx(s); return error; goto out
192 
193 
194 /*
195  * Give the socket an address.
196  */
197 static int
198 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
199 {
200 	int s = splnet();
201 	int error = 0;
202 	struct inpcb *inp = sotoinpcb(so);
203 	struct tcpcb *tp;
204 	struct sockaddr_in *sinp;
205 
206 	COMMON_START();
207 
208 	/*
209 	 * Must check for multicast addresses and disallow binding
210 	 * to them.
211 	 */
212 	sinp = (struct sockaddr_in *)nam;
213 	if (sinp->sin_family == AF_INET &&
214 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
215 		error = EAFNOSUPPORT;
216 		goto out;
217 	}
218 	error = in_pcbbind(inp, nam, td);
219 	if (error)
220 		goto out;
221 	COMMON_END(PRU_BIND);
222 
223 }
224 
225 #ifdef INET6
226 static int
227 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
228 {
229 	int s = splnet();
230 	int error = 0;
231 	struct inpcb *inp = sotoinpcb(so);
232 	struct tcpcb *tp;
233 	struct sockaddr_in6 *sin6p;
234 
235 	COMMON_START();
236 
237 	/*
238 	 * Must check for multicast addresses and disallow binding
239 	 * to them.
240 	 */
241 	sin6p = (struct sockaddr_in6 *)nam;
242 	if (sin6p->sin6_family == AF_INET6 &&
243 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
244 		error = EAFNOSUPPORT;
245 		goto out;
246 	}
247 	inp->inp_vflag &= ~INP_IPV4;
248 	inp->inp_vflag |= INP_IPV6;
249 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
250 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
251 			inp->inp_vflag |= INP_IPV4;
252 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
253 			struct sockaddr_in sin;
254 
255 			in6_sin6_2_sin(&sin, sin6p);
256 			inp->inp_vflag |= INP_IPV4;
257 			inp->inp_vflag &= ~INP_IPV6;
258 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
259 			goto out;
260 		}
261 	}
262 	error = in6_pcbbind(inp, nam, td);
263 	if (error)
264 		goto out;
265 	COMMON_END(PRU_BIND);
266 }
267 #endif /* INET6 */
268 
269 /*
270  * Prepare to accept connections.
271  */
272 static int
273 tcp_usr_listen(struct socket *so, struct thread *td)
274 {
275 	int s = splnet();
276 	int error = 0;
277 	struct inpcb *inp = sotoinpcb(so);
278 	struct tcpcb *tp;
279 
280 	COMMON_START();
281 	if (inp->inp_lport == 0)
282 		error = in_pcbbind(inp, (struct sockaddr *)0, td);
283 	if (error == 0) {
284 		tp->t_state = TCPS_LISTEN;
285 		in_pcbinswildcardhash(inp);
286 	}
287 	COMMON_END(PRU_LISTEN);
288 }
289 
290 #ifdef INET6
291 static int
292 tcp6_usr_listen(struct socket *so, struct thread *td)
293 {
294 	int s = splnet();
295 	int error = 0;
296 	struct inpcb *inp = sotoinpcb(so);
297 	struct tcpcb *tp;
298 
299 	COMMON_START();
300 	if (inp->inp_lport == 0) {
301 		inp->inp_vflag &= ~INP_IPV4;
302 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
303 			inp->inp_vflag |= INP_IPV4;
304 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
305 	}
306 	if (error == 0)
307 		tp->t_state = TCPS_LISTEN;
308 	COMMON_END(PRU_LISTEN);
309 }
310 #endif /* INET6 */
311 
312 /*
313  * Initiate connection to peer.
314  * Create a template for use in transmissions on this connection.
315  * Enter SYN_SENT state, and mark socket as connecting.
316  * Start keep-alive timer, and seed output sequence space.
317  * Send initial segment on connection.
318  */
319 static int
320 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
321 {
322 	int s = splnet();
323 	int error = 0;
324 	struct inpcb *inp = sotoinpcb(so);
325 	struct tcpcb *tp;
326 	struct sockaddr_in *sinp;
327 
328 	COMMON_START();
329 
330 	/*
331 	 * Must disallow TCP ``connections'' to multicast addresses.
332 	 */
333 	sinp = (struct sockaddr_in *)nam;
334 	if (sinp->sin_family == AF_INET
335 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
336 		error = EAFNOSUPPORT;
337 		goto out;
338 	}
339 
340 	prison_remote_ip(td, 0, &sinp->sin_addr.s_addr);
341 
342 	if ((error = tcp_connect(tp, nam, td)) != 0)
343 		goto out;
344 	error = tcp_output(tp);
345 	COMMON_END(PRU_CONNECT);
346 }
347 
348 #ifdef INET6
349 static int
350 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
351 {
352 	int s = splnet();
353 	int error = 0;
354 	struct inpcb *inp = sotoinpcb(so);
355 	struct tcpcb *tp;
356 	struct sockaddr_in6 *sin6p;
357 
358 	COMMON_START();
359 
360 	/*
361 	 * Must disallow TCP ``connections'' to multicast addresses.
362 	 */
363 	sin6p = (struct sockaddr_in6 *)nam;
364 	if (sin6p->sin6_family == AF_INET6
365 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
366 		error = EAFNOSUPPORT;
367 		goto out;
368 	}
369 
370 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
371 		struct sockaddr_in sin;
372 
373 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
374 			error = EINVAL;
375 			goto out;
376 		}
377 
378 		in6_sin6_2_sin(&sin, sin6p);
379 		inp->inp_vflag |= INP_IPV4;
380 		inp->inp_vflag &= ~INP_IPV6;
381 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
382 			goto out;
383 		error = tcp_output(tp);
384 		goto out;
385 	}
386 	inp->inp_vflag &= ~INP_IPV4;
387 	inp->inp_vflag |= INP_IPV6;
388 	inp->inp_inc.inc_isipv6 = 1;
389 	if ((error = tcp6_connect(tp, nam, td)) != 0)
390 		goto out;
391 	error = tcp_output(tp);
392 	COMMON_END(PRU_CONNECT);
393 }
394 #endif /* INET6 */
395 
396 /*
397  * Initiate disconnect from peer.
398  * If connection never passed embryonic stage, just drop;
399  * else if don't need to let data drain, then can just drop anyways,
400  * else have to begin TCP shutdown process: mark socket disconnecting,
401  * drain unread data, state switch to reflect user close, and
402  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
403  * when peer sends FIN and acks ours.
404  *
405  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
406  */
407 static int
408 tcp_usr_disconnect(struct socket *so)
409 {
410 	int s = splnet();
411 	int error = 0;
412 	struct inpcb *inp = sotoinpcb(so);
413 	struct tcpcb *tp;
414 
415 	COMMON_START();
416 	tp = tcp_disconnect(tp);
417 	COMMON_END(PRU_DISCONNECT);
418 }
419 
420 /*
421  * Accept a connection.  Essentially all the work is
422  * done at higher levels; just return the address
423  * of the peer, storing through addr.
424  */
425 static int
426 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
427 {
428 	int s = splnet();
429 	int error = 0;
430 	struct inpcb *inp = sotoinpcb(so);
431 	struct tcpcb *tp = NULL;
432 	TCPDEBUG0;
433 
434 	if (so->so_state & SS_ISDISCONNECTED) {
435 		error = ECONNABORTED;
436 		goto out;
437 	}
438 	if (inp == 0) {
439 		splx(s);
440 		return (EINVAL);
441 	}
442 	tp = intotcpcb(inp);
443 	TCPDEBUG1();
444 	in_setpeeraddr(so, nam);
445 	COMMON_END(PRU_ACCEPT);
446 }
447 
448 #ifdef INET6
449 static int
450 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
451 {
452 	int s = splnet();
453 	int error = 0;
454 	struct inpcb *inp = sotoinpcb(so);
455 	struct tcpcb *tp = NULL;
456 	TCPDEBUG0;
457 
458 	if (so->so_state & SS_ISDISCONNECTED) {
459 		error = ECONNABORTED;
460 		goto out;
461 	}
462 	if (inp == 0) {
463 		splx(s);
464 		return (EINVAL);
465 	}
466 	tp = intotcpcb(inp);
467 	TCPDEBUG1();
468 	in6_mapped_peeraddr(so, nam);
469 	COMMON_END(PRU_ACCEPT);
470 }
471 #endif /* INET6 */
472 /*
473  * Mark the connection as being incapable of further output.
474  */
475 static int
476 tcp_usr_shutdown(struct socket *so)
477 {
478 	int s = splnet();
479 	int error = 0;
480 	struct inpcb *inp = sotoinpcb(so);
481 	struct tcpcb *tp;
482 
483 	COMMON_START();
484 	socantsendmore(so);
485 	tp = tcp_usrclosed(tp);
486 	if (tp)
487 		error = tcp_output(tp);
488 	COMMON_END(PRU_SHUTDOWN);
489 }
490 
491 /*
492  * After a receive, possibly send window update to peer.
493  */
494 static int
495 tcp_usr_rcvd(struct socket *so, int flags)
496 {
497 	int s = splnet();
498 	int error = 0;
499 	struct inpcb *inp = sotoinpcb(so);
500 	struct tcpcb *tp;
501 
502 	COMMON_START();
503 	tcp_output(tp);
504 	COMMON_END(PRU_RCVD);
505 }
506 
507 /*
508  * Do a send by putting data in output queue and updating urgent
509  * marker if URG set.  Possibly send more data.  Unlike the other
510  * pru_*() routines, the mbuf chains are our responsibility.  We
511  * must either enqueue them or free them.  The other pru_* routines
512  * generally are caller-frees.
513  */
514 static int
515 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
516 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
517 {
518 	int s = splnet();
519 	int error = 0;
520 	struct inpcb *inp = sotoinpcb(so);
521 	struct tcpcb *tp;
522 #ifdef INET6
523 	int isipv6;
524 #endif
525 	TCPDEBUG0;
526 
527 	if (inp == NULL) {
528 		/*
529 		 * OOPS! we lost a race, the TCP session got reset after
530 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
531 		 * network interrupt in the non-splnet() section of sosend().
532 		 */
533 		if (m)
534 			m_freem(m);
535 		if (control)
536 			m_freem(control);
537 		error = ECONNRESET;	/* XXX EPIPE? */
538 		tp = NULL;
539 		TCPDEBUG1();
540 		goto out;
541 	}
542 #ifdef INET6
543 	isipv6 = nam && nam->sa_family == AF_INET6;
544 #endif /* INET6 */
545 	tp = intotcpcb(inp);
546 	TCPDEBUG1();
547 	if (control) {
548 		/* TCP doesn't do control messages (rights, creds, etc) */
549 		if (control->m_len) {
550 			m_freem(control);
551 			if (m)
552 				m_freem(m);
553 			error = EINVAL;
554 			goto out;
555 		}
556 		m_freem(control);	/* empty control, just free it */
557 	}
558 	if(!(flags & PRUS_OOB)) {
559 		sbappend(&so->so_snd, m);
560 		if (nam && tp->t_state < TCPS_SYN_SENT) {
561 			/*
562 			 * Do implied connect if not yet connected,
563 			 * initialize window to default value, and
564 			 * initialize maxseg/maxopd using peer's cached
565 			 * MSS.
566 			 */
567 #ifdef INET6
568 			if (isipv6)
569 				error = tcp6_connect(tp, nam, td);
570 			else
571 #endif /* INET6 */
572 			error = tcp_connect(tp, nam, td);
573 			if (error)
574 				goto out;
575 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
576 			tcp_mss(tp, -1);
577 		}
578 
579 		if (flags & PRUS_EOF) {
580 			/*
581 			 * Close the send side of the connection after
582 			 * the data is sent.
583 			 */
584 			socantsendmore(so);
585 			tp = tcp_usrclosed(tp);
586 		}
587 		if (tp != NULL) {
588 			if (flags & PRUS_MORETOCOME)
589 				tp->t_flags |= TF_MORETOCOME;
590 			error = tcp_output(tp);
591 			if (flags & PRUS_MORETOCOME)
592 				tp->t_flags &= ~TF_MORETOCOME;
593 		}
594 	} else {
595 		if (sbspace(&so->so_snd) < -512) {
596 			m_freem(m);
597 			error = ENOBUFS;
598 			goto out;
599 		}
600 		/*
601 		 * According to RFC961 (Assigned Protocols),
602 		 * the urgent pointer points to the last octet
603 		 * of urgent data.  We continue, however,
604 		 * to consider it to indicate the first octet
605 		 * of data past the urgent section.
606 		 * Otherwise, snd_up should be one lower.
607 		 */
608 		sbappend(&so->so_snd, m);
609 		if (nam && tp->t_state < TCPS_SYN_SENT) {
610 			/*
611 			 * Do implied connect if not yet connected,
612 			 * initialize window to default value, and
613 			 * initialize maxseg/maxopd using peer's cached
614 			 * MSS.
615 			 */
616 #ifdef INET6
617 			if (isipv6)
618 				error = tcp6_connect(tp, nam, td);
619 			else
620 #endif /* INET6 */
621 			error = tcp_connect(tp, nam, td);
622 			if (error)
623 				goto out;
624 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
625 			tcp_mss(tp, -1);
626 		}
627 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
628 		tp->t_force = 1;
629 		error = tcp_output(tp);
630 		tp->t_force = 0;
631 	}
632 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
633 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
634 }
635 
636 /*
637  * Abort the TCP.
638  */
639 static int
640 tcp_usr_abort(struct socket *so)
641 {
642 	int s = splnet();
643 	int error = 0;
644 	struct inpcb *inp = sotoinpcb(so);
645 	struct tcpcb *tp;
646 
647 	COMMON_START();
648 	tp = tcp_drop(tp, ECONNABORTED);
649 	COMMON_END(PRU_ABORT);
650 }
651 
652 /*
653  * Receive out-of-band data.
654  */
655 static int
656 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
657 {
658 	int s = splnet();
659 	int error = 0;
660 	struct inpcb *inp = sotoinpcb(so);
661 	struct tcpcb *tp;
662 
663 	COMMON_START();
664 	if ((so->so_oobmark == 0 &&
665 	     (so->so_state & SS_RCVATMARK) == 0) ||
666 	    so->so_options & SO_OOBINLINE ||
667 	    tp->t_oobflags & TCPOOB_HADDATA) {
668 		error = EINVAL;
669 		goto out;
670 	}
671 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
672 		error = EWOULDBLOCK;
673 		goto out;
674 	}
675 	m->m_len = 1;
676 	*mtod(m, caddr_t) = tp->t_iobc;
677 	if ((flags & MSG_PEEK) == 0)
678 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
679 	COMMON_END(PRU_RCVOOB);
680 }
681 
682 /* xxx - should be const */
683 struct pr_usrreqs tcp_usrreqs = {
684 	tcp_usr_abort, tcp_usr_accept, tcp_usr_attach, tcp_usr_bind,
685 	tcp_usr_connect, pru_connect2_notsupp, in_control, tcp_usr_detach,
686 	tcp_usr_disconnect, tcp_usr_listen, in_setpeeraddr, tcp_usr_rcvd,
687 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
688 	in_setsockaddr, sosend, soreceive, sopoll
689 };
690 
691 #ifdef INET6
692 struct pr_usrreqs tcp6_usrreqs = {
693 	tcp_usr_abort, tcp6_usr_accept, tcp_usr_attach, tcp6_usr_bind,
694 	tcp6_usr_connect, pru_connect2_notsupp, in6_control, tcp_usr_detach,
695 	tcp_usr_disconnect, tcp6_usr_listen, in6_mapped_peeraddr, tcp_usr_rcvd,
696 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
697 	in6_mapped_sockaddr, sosend, soreceive, sopoll
698 };
699 #endif /* INET6 */
700 
701 struct netmsg_tcp_connect {
702 	struct lwkt_msg		nm_lmsg;
703 	netisr_fn_t		nm_handler;
704 	struct tcpcb		*nm_tp;
705 	struct sockaddr_in	*nm_sin;
706 	struct sockaddr_in	*nm_ifsin;
707 };
708 
709 static int
710 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
711 		  struct sockaddr_in *if_sin)
712 {
713 	struct inpcb *inp = tp->t_inpcb, *oinp;
714 	struct socket *so = inp->inp_socket;
715 	struct tcpcb *otp;
716 	struct rmxp_tao *taop;
717 	struct rmxp_tao tao_noncached;
718 
719 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
720 	    sin->sin_addr, sin->sin_port,
721 	    inp->inp_laddr.s_addr != INADDR_ANY ?
722 	        inp->inp_laddr : if_sin->sin_addr,
723 	    inp->inp_lport, 0, NULL);
724 	if (oinp != NULL) {
725 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
726 		    otp->t_state == TCPS_TIME_WAIT &&
727 		    (ticks - otp->t_starttime) < tcp_msl &&
728 		    (otp->t_flags & TF_RCVD_CC))
729 			(void) tcp_close(otp);
730 		else
731 			return (EADDRINUSE);
732 	}
733 	if (inp->inp_laddr.s_addr == INADDR_ANY)
734 		inp->inp_laddr = if_sin->sin_addr;
735 	inp->inp_faddr = sin->sin_addr;
736 	inp->inp_fport = sin->sin_port;
737 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
738 	in_pcbinsconnhash(inp);
739 
740 	/* Compute window scaling to request.  */
741 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
742 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
743 		tp->request_r_scale++;
744 
745 	soisconnecting(so);
746 	tcpstat.tcps_connattempt++;
747 	tp->t_state = TCPS_SYN_SENT;
748 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
749 	tp->iss = tcp_new_isn(tp);
750 	tp->t_bw_rtseq = tp->iss;
751 	tcp_sendseqinit(tp);
752 
753 	/*
754 	 * Generate a CC value for this connection and
755 	 * check whether CC or CCnew should be used.
756 	 */
757 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
758 		taop = &tao_noncached;
759 		bzero(taop, sizeof(*taop));
760 	}
761 
762 	tp->cc_send = CC_INC(tcp_ccgen);
763 	if (taop->tao_ccsent != 0 &&
764 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
765 		taop->tao_ccsent = tp->cc_send;
766 	} else {
767 		taop->tao_ccsent = 0;
768 		tp->t_flags |= TF_SENDCCNEW;
769 	}
770 
771 	return (0);
772 }
773 
774 static void
775 tcp_connect_handler(struct netmsg *msg0)
776 {
777 	struct netmsg_tcp_connect *msg = (struct netmsg_tcp_connect *)msg0;
778 	int error;
779 
780 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
781 	lwkt_replymsg(&msg0->nm_lmsg, error);
782 }
783 
784 /*
785  * Common subroutine to open a TCP connection to remote host specified
786  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
787  * port number if needed.  Call in_pcbladdr to do the routing and to choose
788  * a local host address (interface).  If there is an existing incarnation
789  * of the same connection in TIME-WAIT state and if the remote host was
790  * sending CC options and if the connection duration was < MSL, then
791  * truncate the previous TIME-WAIT state and proceed.
792  * Initialize connection parameters and enter SYN-SENT state.
793  */
794 static int
795 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
796 {
797 	struct inpcb *inp = tp->t_inpcb;
798 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
799 	struct sockaddr_in *if_sin;
800 	int error;
801 	boolean_t didbind = FALSE;
802 #ifdef SMP
803 	lwkt_port_t port;
804 #endif
805 
806 	if (inp->inp_lport == 0) {
807 		error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
808 		if (error)
809 			return (error);
810 		didbind = TRUE;
811 	}
812 
813 	/*
814 	 * Cannot simply call in_pcbconnect, because there might be an
815 	 * earlier incarnation of this same connection still in
816 	 * TIME_WAIT state, creating an ADDRINUSE error.
817 	 */
818 	error = in_pcbladdr(inp, nam, &if_sin);
819 	if (error)
820 		return (error);
821 
822 #ifdef SMP
823 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
824 	    inp->inp_laddr.s_addr ?
825 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
826 	    inp->inp_lport);
827 
828 	if (port->mp_td != curthread) {
829 		struct netmsg_tcp_connect *msg;
830 
831 		msg = malloc(sizeof(struct netmsg_tcp_connect), M_LWKTMSG,
832 		    M_NOWAIT);
833 		if (msg == NULL) {
834 			if (didbind) {	/* need to unwind bind */
835 				inp->inp_lport = 0;
836 				inp->inp_laddr.s_addr = INADDR_ANY;
837 				in_pcbremwildcardhash(inp);
838 			}
839 			return (ENOMEM);
840 		}
841 		lwkt_initmsg(&msg->nm_lmsg, CMD_NETMSG_ONCPU);
842 		msg->nm_handler = tcp_connect_handler;
843 		msg->nm_tp = tp;
844 		msg->nm_sin = sin;
845 		msg->nm_ifsin = if_sin;
846 		error = lwkt_domsg(port, &msg->nm_lmsg);
847 	} else
848 #endif
849 		error = tcp_connect_oncpu(tp, sin, if_sin);
850 
851 	return (error);
852 }
853 
854 #ifdef INET6
855 static int
856 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
857 {
858 	struct inpcb *inp = tp->t_inpcb, *oinp;
859 	struct socket *so = inp->inp_socket;
860 	struct tcpcb *otp;
861 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
862 	struct in6_addr *addr6;
863 	struct rmxp_tao *taop;
864 	struct rmxp_tao tao_noncached;
865 	int error;
866 
867 	if (inp->inp_lport == 0) {
868 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
869 		if (error)
870 			return error;
871 	}
872 
873 	/*
874 	 * Cannot simply call in_pcbconnect, because there might be an
875 	 * earlier incarnation of this same connection still in
876 	 * TIME_WAIT state, creating an ADDRINUSE error.
877 	 */
878 	error = in6_pcbladdr(inp, nam, &addr6);
879 	if (error)
880 		return error;
881 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
882 				  &sin6->sin6_addr, sin6->sin6_port,
883 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
884 				      addr6 : &inp->in6p_laddr,
885 				  inp->inp_lport,  0, NULL);
886 	if (oinp) {
887 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
888 		    otp->t_state == TCPS_TIME_WAIT &&
889 		    (ticks - otp->t_starttime) < tcp_msl &&
890 		    (otp->t_flags & TF_RCVD_CC))
891 			otp = tcp_close(otp);
892 		else
893 			return (EADDRINUSE);
894 	}
895 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
896 		inp->in6p_laddr = *addr6;
897 	inp->in6p_faddr = sin6->sin6_addr;
898 	inp->inp_fport = sin6->sin6_port;
899 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
900 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
901 	in_pcbinsconnhash(inp);
902 
903 	/* Compute window scaling to request.  */
904 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
905 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
906 		tp->request_r_scale++;
907 
908 	soisconnecting(so);
909 	tcpstat.tcps_connattempt++;
910 	tp->t_state = TCPS_SYN_SENT;
911 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
912 	tp->iss = tcp_new_isn(tp);
913 	tp->t_bw_rtseq = tp->iss;
914 	tcp_sendseqinit(tp);
915 
916 	/*
917 	 * Generate a CC value for this connection and
918 	 * check whether CC or CCnew should be used.
919 	 */
920 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
921 		taop = &tao_noncached;
922 		bzero(taop, sizeof(*taop));
923 	}
924 
925 	tp->cc_send = CC_INC(tcp_ccgen);
926 	if (taop->tao_ccsent != 0 &&
927 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
928 		taop->tao_ccsent = tp->cc_send;
929 	} else {
930 		taop->tao_ccsent = 0;
931 		tp->t_flags |= TF_SENDCCNEW;
932 	}
933 
934 	return (0);
935 }
936 #endif /* INET6 */
937 
938 /*
939  * The new sockopt interface makes it possible for us to block in the
940  * copyin/out step (if we take a page fault).  Taking a page fault at
941  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
942  * use TSM, there probably isn't any need for this function to run at
943  * splnet() any more.  This needs more examination.)
944  */
945 int
946 tcp_ctloutput(so, sopt)
947 	struct socket *so;
948 	struct sockopt *sopt;
949 {
950 	int	error, opt, optval, s;
951 	struct	inpcb *inp;
952 	struct	tcpcb *tp;
953 
954 	error = 0;
955 	s = splnet();		/* XXX */
956 	inp = sotoinpcb(so);
957 	if (inp == NULL) {
958 		splx(s);
959 		return (ECONNRESET);
960 	}
961 	if (sopt->sopt_level != IPPROTO_TCP) {
962 #ifdef INET6
963 		if (INP_CHECK_SOCKAF(so, AF_INET6))
964 			error = ip6_ctloutput(so, sopt);
965 		else
966 #endif /* INET6 */
967 		error = ip_ctloutput(so, sopt);
968 		splx(s);
969 		return (error);
970 	}
971 	tp = intotcpcb(inp);
972 
973 	switch (sopt->sopt_dir) {
974 	case SOPT_SET:
975 		switch (sopt->sopt_name) {
976 		case TCP_NODELAY:
977 		case TCP_NOOPT:
978 			error = sooptcopyin(sopt, &optval, sizeof optval,
979 					    sizeof optval);
980 			if (error)
981 				break;
982 
983 			switch (sopt->sopt_name) {
984 			case TCP_NODELAY:
985 				opt = TF_NODELAY;
986 				break;
987 			case TCP_NOOPT:
988 				opt = TF_NOOPT;
989 				break;
990 			default:
991 				opt = 0; /* dead code to fool gcc */
992 				break;
993 			}
994 
995 			if (optval)
996 				tp->t_flags |= opt;
997 			else
998 				tp->t_flags &= ~opt;
999 			break;
1000 
1001 		case TCP_NOPUSH:
1002 			error = sooptcopyin(sopt, &optval, sizeof optval,
1003 					    sizeof optval);
1004 			if (error)
1005 				break;
1006 
1007 			if (optval)
1008 				tp->t_flags |= TF_NOPUSH;
1009 			else {
1010 				tp->t_flags &= ~TF_NOPUSH;
1011 				error = tcp_output(tp);
1012 			}
1013 			break;
1014 
1015 		case TCP_MAXSEG:
1016 			error = sooptcopyin(sopt, &optval, sizeof optval,
1017 					    sizeof optval);
1018 			if (error)
1019 				break;
1020 
1021 			if (optval > 0 && optval <= tp->t_maxseg)
1022 				tp->t_maxseg = optval;
1023 			else
1024 				error = EINVAL;
1025 			break;
1026 
1027 		default:
1028 			error = ENOPROTOOPT;
1029 			break;
1030 		}
1031 		break;
1032 
1033 	case SOPT_GET:
1034 		switch (sopt->sopt_name) {
1035 		case TCP_NODELAY:
1036 			optval = tp->t_flags & TF_NODELAY;
1037 			break;
1038 		case TCP_MAXSEG:
1039 			optval = tp->t_maxseg;
1040 			break;
1041 		case TCP_NOOPT:
1042 			optval = tp->t_flags & TF_NOOPT;
1043 			break;
1044 		case TCP_NOPUSH:
1045 			optval = tp->t_flags & TF_NOPUSH;
1046 			break;
1047 		default:
1048 			error = ENOPROTOOPT;
1049 			break;
1050 		}
1051 		if (error == 0)
1052 			error = sooptcopyout(sopt, &optval, sizeof optval);
1053 		break;
1054 	}
1055 	splx(s);
1056 	return (error);
1057 }
1058 
1059 /*
1060  * tcp_sendspace and tcp_recvspace are the default send and receive window
1061  * sizes, respectively.  These are obsolescent (this information should
1062  * be set by the route).
1063  */
1064 u_long	tcp_sendspace = 1024*32;
1065 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1066     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1067 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1068 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1069     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1070 
1071 /*
1072  * Attach TCP protocol to socket, allocating
1073  * internet protocol control block, tcp control block,
1074  * bufer space, and entering LISTEN state if to accept connections.
1075  */
1076 static int
1077 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1078 {
1079 	struct tcpcb *tp;
1080 	struct inpcb *inp;
1081 	int error;
1082 	int cpu;
1083 #ifdef INET6
1084 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
1085 #endif
1086 
1087 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1088 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1089 				  ai->sb_rlimit);
1090 		if (error)
1091 			return (error);
1092 	}
1093 	cpu = mycpu->gd_cpuid;
1094 	error = in_pcballoc(so, &tcbinfo[cpu]);
1095 	if (error)
1096 		return (error);
1097 	inp = sotoinpcb(so);
1098 #ifdef INET6
1099 	if (isipv6) {
1100 		inp->inp_vflag |= INP_IPV6;
1101 		inp->in6p_hops = -1;	/* use kernel default */
1102 	}
1103 	else
1104 #endif
1105 	inp->inp_vflag |= INP_IPV4;
1106 	tp = tcp_newtcpcb(inp);
1107 	if (tp == 0) {
1108 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1109 
1110 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1111 #ifdef INET6
1112 		if (isipv6)
1113 			in6_pcbdetach(inp);
1114 		else
1115 #endif
1116 		in_pcbdetach(inp);
1117 		so->so_state |= nofd;
1118 		return (ENOBUFS);
1119 	}
1120 	tp->t_state = TCPS_CLOSED;
1121 	return (0);
1122 }
1123 
1124 /*
1125  * Initiate (or continue) disconnect.
1126  * If embryonic state, just send reset (once).
1127  * If in ``let data drain'' option and linger null, just drop.
1128  * Otherwise (hard), mark socket disconnecting and drop
1129  * current input data; switch states based on user close, and
1130  * send segment to peer (with FIN).
1131  */
1132 static struct tcpcb *
1133 tcp_disconnect(tp)
1134 	struct tcpcb *tp;
1135 {
1136 	struct socket *so = tp->t_inpcb->inp_socket;
1137 
1138 	if (tp->t_state < TCPS_ESTABLISHED)
1139 		tp = tcp_close(tp);
1140 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1141 		tp = tcp_drop(tp, 0);
1142 	else {
1143 		soisdisconnecting(so);
1144 		sbflush(&so->so_rcv);
1145 		tp = tcp_usrclosed(tp);
1146 		if (tp)
1147 			(void) tcp_output(tp);
1148 	}
1149 	return (tp);
1150 }
1151 
1152 /*
1153  * User issued close, and wish to trail through shutdown states:
1154  * if never received SYN, just forget it.  If got a SYN from peer,
1155  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1156  * If already got a FIN from peer, then almost done; go to LAST_ACK
1157  * state.  In all other cases, have already sent FIN to peer (e.g.
1158  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1159  * for peer to send FIN or not respond to keep-alives, etc.
1160  * We can let the user exit from the close as soon as the FIN is acked.
1161  */
1162 static struct tcpcb *
1163 tcp_usrclosed(tp)
1164 	struct tcpcb *tp;
1165 {
1166 
1167 	switch (tp->t_state) {
1168 
1169 	case TCPS_CLOSED:
1170 	case TCPS_LISTEN:
1171 		tp->t_state = TCPS_CLOSED;
1172 		tp = tcp_close(tp);
1173 		break;
1174 
1175 	case TCPS_SYN_SENT:
1176 	case TCPS_SYN_RECEIVED:
1177 		tp->t_flags |= TF_NEEDFIN;
1178 		break;
1179 
1180 	case TCPS_ESTABLISHED:
1181 		tp->t_state = TCPS_FIN_WAIT_1;
1182 		break;
1183 
1184 	case TCPS_CLOSE_WAIT:
1185 		tp->t_state = TCPS_LAST_ACK;
1186 		break;
1187 	}
1188 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1189 		soisdisconnected(tp->t_inpcb->inp_socket);
1190 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1191 		if (tp->t_state == TCPS_FIN_WAIT_2)
1192 			callout_reset(tp->tt_2msl, tcp_maxidle,
1193 				      tcp_timer_2msl, tp);
1194 	}
1195 	return (tp);
1196 }
1197 
1198