xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision 52ac7bd72cefb1c30ca0bd47dd92fbf37486a264)
1 /*
2  * Copyright (c) 2003-2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 1982, 1986, 1988, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by the University of
17  *	California, Berkeley and its contributors.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
35  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
36  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.24 2004/07/02 04:41:01 hsu Exp $
37  */
38 
39 #include "opt_ipsec.h"
40 #include "opt_inet6.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/sysctl.h>
48 #include <sys/globaldata.h>
49 #include <sys/thread.h>
50 
51 #include <sys/mbuf.h>
52 #ifdef INET6
53 #include <sys/domain.h>
54 #endif /* INET6 */
55 #include <sys/socket.h>
56 #include <sys/socketvar.h>
57 #include <sys/protosw.h>
58 
59 #include <sys/msgport2.h>
60 
61 #include <net/if.h>
62 #include <net/netisr.h>
63 #include <net/route.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #endif
79 #include <netinet/tcp.h>
80 #include <netinet/tcp_fsm.h>
81 #include <netinet/tcp_seq.h>
82 #include <netinet/tcp_timer.h>
83 #include <netinet/tcp_var.h>
84 #include <netinet/tcpip.h>
85 #ifdef TCPDEBUG
86 #include <netinet/tcp_debug.h>
87 #endif
88 
89 #ifdef IPSEC
90 #include <netinet6/ipsec.h>
91 #endif /*IPSEC*/
92 
93 /*
94  * TCP protocol interface to socket abstraction.
95  */
96 extern	char *tcpstates[];	/* XXX ??? */
97 
98 static int	tcp_attach (struct socket *, struct pru_attach_info *);
99 static int	tcp_connect (struct tcpcb *, struct sockaddr *,
100 				 struct thread *);
101 #ifdef INET6
102 static int	tcp6_connect (struct tcpcb *, struct sockaddr *,
103 				 struct thread *);
104 #endif /* INET6 */
105 static struct tcpcb *
106 		tcp_disconnect (struct tcpcb *);
107 static struct tcpcb *
108 		tcp_usrclosed (struct tcpcb *);
109 
110 #ifdef TCPDEBUG
111 #define	TCPDEBUG0	int ostate = 0
112 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
113 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
114 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
115 #else
116 #define	TCPDEBUG0
117 #define	TCPDEBUG1()
118 #define	TCPDEBUG2(req)
119 #endif
120 
121 /*
122  * TCP attaches to socket via pru_attach(), reserving space,
123  * and an internet control block.
124  */
125 static int
126 tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
127 {
128 	int s = splnet();
129 	int error;
130 	struct inpcb *inp = sotoinpcb(so);
131 	struct tcpcb *tp = 0;
132 	TCPDEBUG0;
133 
134 	TCPDEBUG1();
135 	if (inp) {
136 		error = EISCONN;
137 		goto out;
138 	}
139 
140 	error = tcp_attach(so, ai);
141 	if (error)
142 		goto out;
143 
144 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
145 		so->so_linger = TCP_LINGERTIME;
146 	tp = sototcpcb(so);
147 out:
148 	TCPDEBUG2(PRU_ATTACH);
149 	splx(s);
150 	return error;
151 }
152 
153 /*
154  * pru_detach() detaches the TCP protocol from the socket.
155  * If the protocol state is non-embryonic, then can't
156  * do this directly: have to initiate a pru_disconnect(),
157  * which may finish later; embryonic TCB's can just
158  * be discarded here.
159  */
160 static int
161 tcp_usr_detach(struct socket *so)
162 {
163 	int s = splnet();
164 	int error = 0;
165 	struct inpcb *inp = sotoinpcb(so);
166 	struct tcpcb *tp;
167 	TCPDEBUG0;
168 
169 	if (inp == 0) {
170 		splx(s);
171 		return EINVAL;	/* XXX */
172 	}
173 	tp = intotcpcb(inp);
174 	TCPDEBUG1();
175 	tp = tcp_disconnect(tp);
176 
177 	TCPDEBUG2(PRU_DETACH);
178 	splx(s);
179 	return error;
180 }
181 
182 #define	COMMON_START()	TCPDEBUG0; \
183 			do { \
184 				     if (inp == 0) { \
185 					     splx(s); \
186 					     return EINVAL; \
187 				     } \
188 				     tp = intotcpcb(inp); \
189 				     TCPDEBUG1(); \
190 		     } while(0)
191 
192 #define COMMON_END(req)	out: TCPDEBUG2(req); splx(s); return error; goto out
193 
194 
195 /*
196  * Give the socket an address.
197  */
198 static int
199 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
200 {
201 	int s = splnet();
202 	int error = 0;
203 	struct inpcb *inp = sotoinpcb(so);
204 	struct tcpcb *tp;
205 	struct sockaddr_in *sinp;
206 
207 	COMMON_START();
208 
209 	/*
210 	 * Must check for multicast addresses and disallow binding
211 	 * to them.
212 	 */
213 	sinp = (struct sockaddr_in *)nam;
214 	if (sinp->sin_family == AF_INET &&
215 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
216 		error = EAFNOSUPPORT;
217 		goto out;
218 	}
219 	error = in_pcbbind(inp, nam, td);
220 	if (error)
221 		goto out;
222 	COMMON_END(PRU_BIND);
223 
224 }
225 
226 #ifdef INET6
227 static int
228 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
229 {
230 	int s = splnet();
231 	int error = 0;
232 	struct inpcb *inp = sotoinpcb(so);
233 	struct tcpcb *tp;
234 	struct sockaddr_in6 *sin6p;
235 
236 	COMMON_START();
237 
238 	/*
239 	 * Must check for multicast addresses and disallow binding
240 	 * to them.
241 	 */
242 	sin6p = (struct sockaddr_in6 *)nam;
243 	if (sin6p->sin6_family == AF_INET6 &&
244 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
245 		error = EAFNOSUPPORT;
246 		goto out;
247 	}
248 	inp->inp_vflag &= ~INP_IPV4;
249 	inp->inp_vflag |= INP_IPV6;
250 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
251 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
252 			inp->inp_vflag |= INP_IPV4;
253 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
254 			struct sockaddr_in sin;
255 
256 			in6_sin6_2_sin(&sin, sin6p);
257 			inp->inp_vflag |= INP_IPV4;
258 			inp->inp_vflag &= ~INP_IPV6;
259 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
260 			goto out;
261 		}
262 	}
263 	error = in6_pcbbind(inp, nam, td);
264 	if (error)
265 		goto out;
266 	COMMON_END(PRU_BIND);
267 }
268 #endif /* INET6 */
269 
270 #ifdef SMP
271 struct netmsg_inswildcard {
272 	struct lwkt_msg		nm_lmsg;
273 	struct inpcb		*nm_inp;
274 	struct inpcbinfo	*nm_pcbinfo;
275 };
276 
277 static int
278 in_pcbinswildcardhash_handler(struct lwkt_msg *msg0)
279 {
280 	struct netmsg_inswildcard *msg = (struct netmsg_inswildcard *)msg0;
281 
282 	in_pcbinswildcardhash_oncpu(msg->nm_inp, msg->nm_pcbinfo);
283 	lwkt_replymsg(&msg->nm_lmsg, 0);
284 	return (EASYNC);
285 }
286 #endif
287 
288 /*
289  * Prepare to accept connections.
290  */
291 static int
292 tcp_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 #ifdef SMP
299 	int cpu;
300 #endif
301 
302 	COMMON_START();
303 	if (inp->inp_lport == 0) {
304 		error = in_pcbbind(inp, NULL, td);
305 		if (error != 0)
306 			goto out;
307 	}
308 
309 	tp->t_state = TCPS_LISTEN;
310 #ifdef SMP
311 	for (cpu = 0; cpu < ncpus2; cpu++) {
312 		struct netmsg_inswildcard *msg;
313 
314 		if (cpu == mycpu->gd_cpuid) {
315 			in_pcbinswildcardhash_oncpu(inp, &tcbinfo[cpu]);
316 			continue;
317 		}
318 
319 		msg = malloc(sizeof(struct netmsg_inswildcard), M_LWKTMSG,
320 		    M_INTWAIT);
321 		lwkt_initmsg(&msg->nm_lmsg, &netisr_afree_rport, 0,
322 		    lwkt_cmd_func(in_pcbinswildcardhash_handler),
323 		    lwkt_cmd_op_none);
324 		msg->nm_inp = inp;
325 		msg->nm_pcbinfo = &tcbinfo[cpu];
326 		lwkt_sendmsg(tcp_cport(cpu), &msg->nm_lmsg);
327 	}
328 	inp->inp_flags |= INP_WILDCARD_MP;
329 #else
330 	in_pcbinswildcardhash(inp);
331 #endif
332 	COMMON_END(PRU_LISTEN);
333 }
334 
335 #ifdef INET6
336 static int
337 tcp6_usr_listen(struct socket *so, struct thread *td)
338 {
339 	int s = splnet();
340 	int error = 0;
341 	struct inpcb *inp = sotoinpcb(so);
342 	struct tcpcb *tp;
343 
344 	COMMON_START();
345 	if (inp->inp_lport == 0) {
346 		inp->inp_vflag &= ~INP_IPV4;
347 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
348 			inp->inp_vflag |= INP_IPV4;
349 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
350 	}
351 	if (error == 0)
352 		tp->t_state = TCPS_LISTEN;
353 	in_pcbinswildcardhash(inp);
354 	COMMON_END(PRU_LISTEN);
355 }
356 #endif /* INET6 */
357 
358 /*
359  * Initiate connection to peer.
360  * Create a template for use in transmissions on this connection.
361  * Enter SYN_SENT state, and mark socket as connecting.
362  * Start keep-alive timer, and seed output sequence space.
363  * Send initial segment on connection.
364  */
365 static int
366 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
367 {
368 	int s = splnet();
369 	int error = 0;
370 	struct inpcb *inp = sotoinpcb(so);
371 	struct tcpcb *tp;
372 	struct sockaddr_in *sinp;
373 
374 	COMMON_START();
375 
376 	/*
377 	 * Must disallow TCP ``connections'' to multicast addresses.
378 	 */
379 	sinp = (struct sockaddr_in *)nam;
380 	if (sinp->sin_family == AF_INET
381 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
382 		error = EAFNOSUPPORT;
383 		goto out;
384 	}
385 
386 	prison_remote_ip(td, 0, &sinp->sin_addr.s_addr);
387 
388 	if ((error = tcp_connect(tp, nam, td)) != 0)
389 		goto out;
390 	error = tcp_output(tp);
391 	COMMON_END(PRU_CONNECT);
392 }
393 
394 #ifdef INET6
395 static int
396 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
397 {
398 	int s = splnet();
399 	int error = 0;
400 	struct inpcb *inp = sotoinpcb(so);
401 	struct tcpcb *tp;
402 	struct sockaddr_in6 *sin6p;
403 
404 	COMMON_START();
405 
406 	/*
407 	 * Must disallow TCP ``connections'' to multicast addresses.
408 	 */
409 	sin6p = (struct sockaddr_in6 *)nam;
410 	if (sin6p->sin6_family == AF_INET6
411 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
412 		error = EAFNOSUPPORT;
413 		goto out;
414 	}
415 
416 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
417 		struct sockaddr_in sin;
418 
419 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
420 			error = EINVAL;
421 			goto out;
422 		}
423 
424 		in6_sin6_2_sin(&sin, sin6p);
425 		inp->inp_vflag |= INP_IPV4;
426 		inp->inp_vflag &= ~INP_IPV6;
427 		if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
428 			goto out;
429 		error = tcp_output(tp);
430 		goto out;
431 	}
432 	inp->inp_vflag &= ~INP_IPV4;
433 	inp->inp_vflag |= INP_IPV6;
434 	inp->inp_inc.inc_isipv6 = 1;
435 	if ((error = tcp6_connect(tp, nam, td)) != 0)
436 		goto out;
437 	error = tcp_output(tp);
438 	COMMON_END(PRU_CONNECT);
439 }
440 #endif /* INET6 */
441 
442 /*
443  * Initiate disconnect from peer.
444  * If connection never passed embryonic stage, just drop;
445  * else if don't need to let data drain, then can just drop anyways,
446  * else have to begin TCP shutdown process: mark socket disconnecting,
447  * drain unread data, state switch to reflect user close, and
448  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
449  * when peer sends FIN and acks ours.
450  *
451  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
452  */
453 static int
454 tcp_usr_disconnect(struct socket *so)
455 {
456 	int s = splnet();
457 	int error = 0;
458 	struct inpcb *inp = sotoinpcb(so);
459 	struct tcpcb *tp;
460 
461 	COMMON_START();
462 	tp = tcp_disconnect(tp);
463 	COMMON_END(PRU_DISCONNECT);
464 }
465 
466 /*
467  * Accept a connection.  Essentially all the work is
468  * done at higher levels; just return the address
469  * of the peer, storing through addr.
470  */
471 static int
472 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
473 {
474 	int s = splnet();
475 	int error = 0;
476 	struct inpcb *inp = sotoinpcb(so);
477 	struct tcpcb *tp = NULL;
478 	TCPDEBUG0;
479 
480 	if (so->so_state & SS_ISDISCONNECTED) {
481 		error = ECONNABORTED;
482 		goto out;
483 	}
484 	if (inp == 0) {
485 		splx(s);
486 		return (EINVAL);
487 	}
488 	tp = intotcpcb(inp);
489 	TCPDEBUG1();
490 	in_setpeeraddr(so, nam);
491 	COMMON_END(PRU_ACCEPT);
492 }
493 
494 #ifdef INET6
495 static int
496 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
497 {
498 	int s = splnet();
499 	int error = 0;
500 	struct inpcb *inp = sotoinpcb(so);
501 	struct tcpcb *tp = NULL;
502 	TCPDEBUG0;
503 
504 	if (so->so_state & SS_ISDISCONNECTED) {
505 		error = ECONNABORTED;
506 		goto out;
507 	}
508 	if (inp == 0) {
509 		splx(s);
510 		return (EINVAL);
511 	}
512 	tp = intotcpcb(inp);
513 	TCPDEBUG1();
514 	in6_mapped_peeraddr(so, nam);
515 	COMMON_END(PRU_ACCEPT);
516 }
517 #endif /* INET6 */
518 /*
519  * Mark the connection as being incapable of further output.
520  */
521 static int
522 tcp_usr_shutdown(struct socket *so)
523 {
524 	int s = splnet();
525 	int error = 0;
526 	struct inpcb *inp = sotoinpcb(so);
527 	struct tcpcb *tp;
528 
529 	COMMON_START();
530 	socantsendmore(so);
531 	tp = tcp_usrclosed(tp);
532 	if (tp)
533 		error = tcp_output(tp);
534 	COMMON_END(PRU_SHUTDOWN);
535 }
536 
537 /*
538  * After a receive, possibly send window update to peer.
539  */
540 static int
541 tcp_usr_rcvd(struct socket *so, int flags)
542 {
543 	int s = splnet();
544 	int error = 0;
545 	struct inpcb *inp = sotoinpcb(so);
546 	struct tcpcb *tp;
547 
548 	COMMON_START();
549 	tcp_output(tp);
550 	COMMON_END(PRU_RCVD);
551 }
552 
553 /*
554  * Do a send by putting data in output queue and updating urgent
555  * marker if URG set.  Possibly send more data.  Unlike the other
556  * pru_*() routines, the mbuf chains are our responsibility.  We
557  * must either enqueue them or free them.  The other pru_* routines
558  * generally are caller-frees.
559  */
560 static int
561 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
562 	     struct sockaddr *nam, struct mbuf *control, struct thread *td)
563 {
564 	int s = splnet();
565 	int error = 0;
566 	struct inpcb *inp = sotoinpcb(so);
567 	struct tcpcb *tp;
568 #ifdef INET6
569 	int isipv6;
570 #endif
571 	TCPDEBUG0;
572 
573 	if (inp == NULL) {
574 		/*
575 		 * OOPS! we lost a race, the TCP session got reset after
576 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
577 		 * network interrupt in the non-splnet() section of sosend().
578 		 */
579 		if (m)
580 			m_freem(m);
581 		if (control)
582 			m_freem(control);
583 		error = ECONNRESET;	/* XXX EPIPE? */
584 		tp = NULL;
585 		TCPDEBUG1();
586 		goto out;
587 	}
588 #ifdef INET6
589 	isipv6 = nam && nam->sa_family == AF_INET6;
590 #endif /* INET6 */
591 	tp = intotcpcb(inp);
592 	TCPDEBUG1();
593 	if (control) {
594 		/* TCP doesn't do control messages (rights, creds, etc) */
595 		if (control->m_len) {
596 			m_freem(control);
597 			if (m)
598 				m_freem(m);
599 			error = EINVAL;
600 			goto out;
601 		}
602 		m_freem(control);	/* empty control, just free it */
603 	}
604 	if(!(flags & PRUS_OOB)) {
605 		sbappend(&so->so_snd, m);
606 		if (nam && tp->t_state < TCPS_SYN_SENT) {
607 			/*
608 			 * Do implied connect if not yet connected,
609 			 * initialize window to default value, and
610 			 * initialize maxseg/maxopd using peer's cached
611 			 * MSS.
612 			 */
613 #ifdef INET6
614 			if (isipv6)
615 				error = tcp6_connect(tp, nam, td);
616 			else
617 #endif /* INET6 */
618 			error = tcp_connect(tp, nam, td);
619 			if (error)
620 				goto out;
621 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
622 			tcp_mss(tp, -1);
623 		}
624 
625 		if (flags & PRUS_EOF) {
626 			/*
627 			 * Close the send side of the connection after
628 			 * the data is sent.
629 			 */
630 			socantsendmore(so);
631 			tp = tcp_usrclosed(tp);
632 		}
633 		if (tp != NULL) {
634 			if (flags & PRUS_MORETOCOME)
635 				tp->t_flags |= TF_MORETOCOME;
636 			error = tcp_output(tp);
637 			if (flags & PRUS_MORETOCOME)
638 				tp->t_flags &= ~TF_MORETOCOME;
639 		}
640 	} else {
641 		if (sbspace(&so->so_snd) < -512) {
642 			m_freem(m);
643 			error = ENOBUFS;
644 			goto out;
645 		}
646 		/*
647 		 * According to RFC961 (Assigned Protocols),
648 		 * the urgent pointer points to the last octet
649 		 * of urgent data.  We continue, however,
650 		 * to consider it to indicate the first octet
651 		 * of data past the urgent section.
652 		 * Otherwise, snd_up should be one lower.
653 		 */
654 		sbappend(&so->so_snd, m);
655 		if (nam && tp->t_state < TCPS_SYN_SENT) {
656 			/*
657 			 * Do implied connect if not yet connected,
658 			 * initialize window to default value, and
659 			 * initialize maxseg/maxopd using peer's cached
660 			 * MSS.
661 			 */
662 #ifdef INET6
663 			if (isipv6)
664 				error = tcp6_connect(tp, nam, td);
665 			else
666 #endif /* INET6 */
667 			error = tcp_connect(tp, nam, td);
668 			if (error)
669 				goto out;
670 			tp->snd_wnd = TTCP_CLIENT_SND_WND;
671 			tcp_mss(tp, -1);
672 		}
673 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
674 		tp->t_flags |= TF_FORCE;
675 		error = tcp_output(tp);
676 		tp->t_flags &= ~TF_FORCE;
677 	}
678 	COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
679 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
680 }
681 
682 /*
683  * Abort the TCP.
684  */
685 static int
686 tcp_usr_abort(struct socket *so)
687 {
688 	int s = splnet();
689 	int error = 0;
690 	struct inpcb *inp = sotoinpcb(so);
691 	struct tcpcb *tp;
692 
693 	COMMON_START();
694 	tp = tcp_drop(tp, ECONNABORTED);
695 	COMMON_END(PRU_ABORT);
696 }
697 
698 /*
699  * Receive out-of-band data.
700  */
701 static int
702 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
703 {
704 	int s = splnet();
705 	int error = 0;
706 	struct inpcb *inp = sotoinpcb(so);
707 	struct tcpcb *tp;
708 
709 	COMMON_START();
710 	if ((so->so_oobmark == 0 &&
711 	     (so->so_state & SS_RCVATMARK) == 0) ||
712 	    so->so_options & SO_OOBINLINE ||
713 	    tp->t_oobflags & TCPOOB_HADDATA) {
714 		error = EINVAL;
715 		goto out;
716 	}
717 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
718 		error = EWOULDBLOCK;
719 		goto out;
720 	}
721 	m->m_len = 1;
722 	*mtod(m, caddr_t) = tp->t_iobc;
723 	if ((flags & MSG_PEEK) == 0)
724 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
725 	COMMON_END(PRU_RCVOOB);
726 }
727 
728 /* xxx - should be const */
729 struct pr_usrreqs tcp_usrreqs = {
730 	tcp_usr_abort, tcp_usr_accept, tcp_usr_attach, tcp_usr_bind,
731 	tcp_usr_connect, pru_connect2_notsupp, in_control, tcp_usr_detach,
732 	tcp_usr_disconnect, tcp_usr_listen, in_setpeeraddr, tcp_usr_rcvd,
733 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
734 	in_setsockaddr, sosend, soreceive, sopoll
735 };
736 
737 #ifdef INET6
738 struct pr_usrreqs tcp6_usrreqs = {
739 	tcp_usr_abort, tcp6_usr_accept, tcp_usr_attach, tcp6_usr_bind,
740 	tcp6_usr_connect, pru_connect2_notsupp, in6_control, tcp_usr_detach,
741 	tcp_usr_disconnect, tcp6_usr_listen, in6_mapped_peeraddr, tcp_usr_rcvd,
742 	tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
743 	in6_mapped_sockaddr, sosend, soreceive, sopoll
744 };
745 #endif /* INET6 */
746 
747 static int
748 tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
749 		  struct sockaddr_in *if_sin)
750 {
751 	struct inpcb *inp = tp->t_inpcb, *oinp;
752 	struct socket *so = inp->inp_socket;
753 	struct tcpcb *otp;
754 	struct rmxp_tao *taop;
755 	struct rmxp_tao tao_noncached;
756 
757 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
758 	    sin->sin_addr, sin->sin_port,
759 	    inp->inp_laddr.s_addr != INADDR_ANY ?
760 	        inp->inp_laddr : if_sin->sin_addr,
761 	    inp->inp_lport, 0, NULL);
762 	if (oinp != NULL) {
763 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
764 		    otp->t_state == TCPS_TIME_WAIT &&
765 		    (ticks - otp->t_starttime) < tcp_msl &&
766 		    (otp->t_flags & TF_RCVD_CC))
767 			(void) tcp_close(otp);
768 		else
769 			return (EADDRINUSE);
770 	}
771 	if (inp->inp_laddr.s_addr == INADDR_ANY)
772 		inp->inp_laddr = if_sin->sin_addr;
773 	inp->inp_faddr = sin->sin_addr;
774 	inp->inp_fport = sin->sin_port;
775 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
776 	in_pcbinsconnhash(inp);
777 
778 	/* Compute window scaling to request.  */
779 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
780 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
781 		tp->request_r_scale++;
782 
783 	soisconnecting(so);
784 	tcpstat.tcps_connattempt++;
785 	tp->t_state = TCPS_SYN_SENT;
786 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
787 	tp->iss = tcp_new_isn(tp);
788 	tp->t_bw_rtseq = tp->iss;
789 	tcp_sendseqinit(tp);
790 
791 	/*
792 	 * Generate a CC value for this connection and
793 	 * check whether CC or CCnew should be used.
794 	 */
795 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
796 		taop = &tao_noncached;
797 		bzero(taop, sizeof(*taop));
798 	}
799 
800 	tp->cc_send = CC_INC(tcp_ccgen);
801 	if (taop->tao_ccsent != 0 &&
802 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
803 		taop->tao_ccsent = tp->cc_send;
804 	} else {
805 		taop->tao_ccsent = 0;
806 		tp->t_flags |= TF_SENDCCNEW;
807 	}
808 
809 	return (0);
810 }
811 
812 #ifdef SMP
813 
814 struct netmsg_tcp_connect {
815 	struct lwkt_msg		nm_lmsg;
816 	struct tcpcb		*nm_tp;
817 	struct sockaddr_in	*nm_sin;
818 	struct sockaddr_in	*nm_ifsin;
819 };
820 
821 static int
822 tcp_connect_handler(lwkt_msg_t lmsg)
823 {
824 	struct netmsg_tcp_connect *msg = (void *)lmsg;
825 	int error;
826 
827 	error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
828 	lwkt_replymsg(lmsg, error);
829 	return(EASYNC);
830 }
831 
832 #endif
833 
834 /*
835  * Common subroutine to open a TCP connection to remote host specified
836  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
837  * port number if needed.  Call in_pcbladdr to do the routing and to choose
838  * a local host address (interface).  If there is an existing incarnation
839  * of the same connection in TIME-WAIT state and if the remote host was
840  * sending CC options and if the connection duration was < MSL, then
841  * truncate the previous TIME-WAIT state and proceed.
842  * Initialize connection parameters and enter SYN-SENT state.
843  */
844 static int
845 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
846 {
847 	struct inpcb *inp = tp->t_inpcb;
848 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
849 	struct sockaddr_in *if_sin;
850 	int error;
851 #ifdef SMP
852 	lwkt_port_t port;
853 #endif
854 
855 	if (inp->inp_lport == 0) {
856 		error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
857 		if (error)
858 			return (error);
859 	}
860 
861 	/*
862 	 * Cannot simply call in_pcbconnect, because there might be an
863 	 * earlier incarnation of this same connection still in
864 	 * TIME_WAIT state, creating an ADDRINUSE error.
865 	 */
866 	error = in_pcbladdr(inp, nam, &if_sin);
867 	if (error)
868 		return (error);
869 
870 #ifdef SMP
871 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
872 	    inp->inp_laddr.s_addr ?
873 		inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
874 	    inp->inp_lport);
875 
876 	if (port->mp_td != curthread) {
877 		struct netmsg_tcp_connect msg;
878 
879 		lwkt_initmsg(&msg.nm_lmsg, &curthread->td_msgport, 0,
880 		    lwkt_cmd_func(tcp_connect_handler), lwkt_cmd_op_none);
881 		msg.nm_tp = tp;
882 		msg.nm_sin = sin;
883 		msg.nm_ifsin = if_sin;
884 		error = lwkt_domsg(port, &msg.nm_lmsg);
885 	} else
886 #endif
887 		error = tcp_connect_oncpu(tp, sin, if_sin);
888 
889 	return (error);
890 }
891 
892 #ifdef INET6
893 static int
894 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
895 {
896 	struct inpcb *inp = tp->t_inpcb, *oinp;
897 	struct socket *so = inp->inp_socket;
898 	struct tcpcb *otp;
899 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
900 	struct in6_addr *addr6;
901 	struct rmxp_tao *taop;
902 	struct rmxp_tao tao_noncached;
903 	int error;
904 
905 	if (inp->inp_lport == 0) {
906 		error = in6_pcbbind(inp, (struct sockaddr *)0, td);
907 		if (error)
908 			return error;
909 	}
910 
911 	/*
912 	 * Cannot simply call in_pcbconnect, because there might be an
913 	 * earlier incarnation of this same connection still in
914 	 * TIME_WAIT state, creating an ADDRINUSE error.
915 	 */
916 	error = in6_pcbladdr(inp, nam, &addr6);
917 	if (error)
918 		return error;
919 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
920 				  &sin6->sin6_addr, sin6->sin6_port,
921 				  IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
922 				      addr6 : &inp->in6p_laddr,
923 				  inp->inp_lport,  0, NULL);
924 	if (oinp) {
925 		if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
926 		    otp->t_state == TCPS_TIME_WAIT &&
927 		    (ticks - otp->t_starttime) < tcp_msl &&
928 		    (otp->t_flags & TF_RCVD_CC))
929 			otp = tcp_close(otp);
930 		else
931 			return (EADDRINUSE);
932 	}
933 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
934 		inp->in6p_laddr = *addr6;
935 	inp->in6p_faddr = sin6->sin6_addr;
936 	inp->inp_fport = sin6->sin6_port;
937 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
938 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
939 	in_pcbinsconnhash(inp);
940 
941 	/* Compute window scaling to request.  */
942 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
943 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
944 		tp->request_r_scale++;
945 
946 	soisconnecting(so);
947 	tcpstat.tcps_connattempt++;
948 	tp->t_state = TCPS_SYN_SENT;
949 	callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
950 	tp->iss = tcp_new_isn(tp);
951 	tp->t_bw_rtseq = tp->iss;
952 	tcp_sendseqinit(tp);
953 
954 	/*
955 	 * Generate a CC value for this connection and
956 	 * check whether CC or CCnew should be used.
957 	 */
958 	if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
959 		taop = &tao_noncached;
960 		bzero(taop, sizeof(*taop));
961 	}
962 
963 	tp->cc_send = CC_INC(tcp_ccgen);
964 	if (taop->tao_ccsent != 0 &&
965 	    CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
966 		taop->tao_ccsent = tp->cc_send;
967 	} else {
968 		taop->tao_ccsent = 0;
969 		tp->t_flags |= TF_SENDCCNEW;
970 	}
971 
972 	return (0);
973 }
974 #endif /* INET6 */
975 
976 /*
977  * The new sockopt interface makes it possible for us to block in the
978  * copyin/out step (if we take a page fault).  Taking a page fault at
979  * splnet() is probably a Bad Thing.  (Since sockets and pcbs both now
980  * use TSM, there probably isn't any need for this function to run at
981  * splnet() any more.  This needs more examination.)
982  */
983 int
984 tcp_ctloutput(so, sopt)
985 	struct socket *so;
986 	struct sockopt *sopt;
987 {
988 	int	error, opt, optval, s;
989 	struct	inpcb *inp;
990 	struct	tcpcb *tp;
991 
992 	error = 0;
993 	s = splnet();		/* XXX */
994 	inp = sotoinpcb(so);
995 	if (inp == NULL) {
996 		splx(s);
997 		return (ECONNRESET);
998 	}
999 	if (sopt->sopt_level != IPPROTO_TCP) {
1000 #ifdef INET6
1001 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1002 			error = ip6_ctloutput(so, sopt);
1003 		else
1004 #endif /* INET6 */
1005 		error = ip_ctloutput(so, sopt);
1006 		splx(s);
1007 		return (error);
1008 	}
1009 	tp = intotcpcb(inp);
1010 
1011 	switch (sopt->sopt_dir) {
1012 	case SOPT_SET:
1013 		switch (sopt->sopt_name) {
1014 		case TCP_NODELAY:
1015 		case TCP_NOOPT:
1016 			error = sooptcopyin(sopt, &optval, sizeof optval,
1017 					    sizeof optval);
1018 			if (error)
1019 				break;
1020 
1021 			switch (sopt->sopt_name) {
1022 			case TCP_NODELAY:
1023 				opt = TF_NODELAY;
1024 				break;
1025 			case TCP_NOOPT:
1026 				opt = TF_NOOPT;
1027 				break;
1028 			default:
1029 				opt = 0; /* dead code to fool gcc */
1030 				break;
1031 			}
1032 
1033 			if (optval)
1034 				tp->t_flags |= opt;
1035 			else
1036 				tp->t_flags &= ~opt;
1037 			break;
1038 
1039 		case TCP_NOPUSH:
1040 			error = sooptcopyin(sopt, &optval, sizeof optval,
1041 					    sizeof optval);
1042 			if (error)
1043 				break;
1044 
1045 			if (optval)
1046 				tp->t_flags |= TF_NOPUSH;
1047 			else {
1048 				tp->t_flags &= ~TF_NOPUSH;
1049 				error = tcp_output(tp);
1050 			}
1051 			break;
1052 
1053 		case TCP_MAXSEG:
1054 			error = sooptcopyin(sopt, &optval, sizeof optval,
1055 					    sizeof optval);
1056 			if (error)
1057 				break;
1058 
1059 			if (optval > 0 && optval <= tp->t_maxseg)
1060 				tp->t_maxseg = optval;
1061 			else
1062 				error = EINVAL;
1063 			break;
1064 
1065 		default:
1066 			error = ENOPROTOOPT;
1067 			break;
1068 		}
1069 		break;
1070 
1071 	case SOPT_GET:
1072 		switch (sopt->sopt_name) {
1073 		case TCP_NODELAY:
1074 			optval = tp->t_flags & TF_NODELAY;
1075 			break;
1076 		case TCP_MAXSEG:
1077 			optval = tp->t_maxseg;
1078 			break;
1079 		case TCP_NOOPT:
1080 			optval = tp->t_flags & TF_NOOPT;
1081 			break;
1082 		case TCP_NOPUSH:
1083 			optval = tp->t_flags & TF_NOPUSH;
1084 			break;
1085 		default:
1086 			error = ENOPROTOOPT;
1087 			break;
1088 		}
1089 		if (error == 0)
1090 			error = sooptcopyout(sopt, &optval, sizeof optval);
1091 		break;
1092 	}
1093 	splx(s);
1094 	return (error);
1095 }
1096 
1097 /*
1098  * tcp_sendspace and tcp_recvspace are the default send and receive window
1099  * sizes, respectively.  These are obsolescent (this information should
1100  * be set by the route).
1101  */
1102 u_long	tcp_sendspace = 1024*32;
1103 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1104     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1105 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1106 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1107     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1108 
1109 /*
1110  * Attach TCP protocol to socket, allocating
1111  * internet protocol control block, tcp control block,
1112  * bufer space, and entering LISTEN state if to accept connections.
1113  */
1114 static int
1115 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1116 {
1117 	struct tcpcb *tp;
1118 	struct inpcb *inp;
1119 	int error;
1120 	int cpu;
1121 #ifdef INET6
1122 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
1123 #endif
1124 
1125 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1126 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1127 				  ai->sb_rlimit);
1128 		if (error)
1129 			return (error);
1130 	}
1131 	cpu = mycpu->gd_cpuid;
1132 	error = in_pcballoc(so, &tcbinfo[cpu]);
1133 	if (error)
1134 		return (error);
1135 	inp = sotoinpcb(so);
1136 #ifdef INET6
1137 	if (isipv6) {
1138 		inp->inp_vflag |= INP_IPV6;
1139 		inp->in6p_hops = -1;	/* use kernel default */
1140 	}
1141 	else
1142 #endif
1143 	inp->inp_vflag |= INP_IPV4;
1144 	tp = tcp_newtcpcb(inp);
1145 	if (tp == 0) {
1146 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
1147 
1148 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
1149 #ifdef INET6
1150 		if (isipv6)
1151 			in6_pcbdetach(inp);
1152 		else
1153 #endif
1154 		in_pcbdetach(inp);
1155 		so->so_state |= nofd;
1156 		return (ENOBUFS);
1157 	}
1158 	tp->t_state = TCPS_CLOSED;
1159 	return (0);
1160 }
1161 
1162 /*
1163  * Initiate (or continue) disconnect.
1164  * If embryonic state, just send reset (once).
1165  * If in ``let data drain'' option and linger null, just drop.
1166  * Otherwise (hard), mark socket disconnecting and drop
1167  * current input data; switch states based on user close, and
1168  * send segment to peer (with FIN).
1169  */
1170 static struct tcpcb *
1171 tcp_disconnect(tp)
1172 	struct tcpcb *tp;
1173 {
1174 	struct socket *so = tp->t_inpcb->inp_socket;
1175 
1176 	if (tp->t_state < TCPS_ESTABLISHED)
1177 		tp = tcp_close(tp);
1178 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1179 		tp = tcp_drop(tp, 0);
1180 	else {
1181 		soisdisconnecting(so);
1182 		sbflush(&so->so_rcv);
1183 		tp = tcp_usrclosed(tp);
1184 		if (tp)
1185 			(void) tcp_output(tp);
1186 	}
1187 	return (tp);
1188 }
1189 
1190 /*
1191  * User issued close, and wish to trail through shutdown states:
1192  * if never received SYN, just forget it.  If got a SYN from peer,
1193  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1194  * If already got a FIN from peer, then almost done; go to LAST_ACK
1195  * state.  In all other cases, have already sent FIN to peer (e.g.
1196  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1197  * for peer to send FIN or not respond to keep-alives, etc.
1198  * We can let the user exit from the close as soon as the FIN is acked.
1199  */
1200 static struct tcpcb *
1201 tcp_usrclosed(tp)
1202 	struct tcpcb *tp;
1203 {
1204 
1205 	switch (tp->t_state) {
1206 
1207 	case TCPS_CLOSED:
1208 	case TCPS_LISTEN:
1209 		tp->t_state = TCPS_CLOSED;
1210 		tp = tcp_close(tp);
1211 		break;
1212 
1213 	case TCPS_SYN_SENT:
1214 	case TCPS_SYN_RECEIVED:
1215 		tp->t_flags |= TF_NEEDFIN;
1216 		break;
1217 
1218 	case TCPS_ESTABLISHED:
1219 		tp->t_state = TCPS_FIN_WAIT_1;
1220 		break;
1221 
1222 	case TCPS_CLOSE_WAIT:
1223 		tp->t_state = TCPS_LAST_ACK;
1224 		break;
1225 	}
1226 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1227 		soisdisconnected(tp->t_inpcb->inp_socket);
1228 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1229 		if (tp->t_state == TCPS_FIN_WAIT_2)
1230 			callout_reset(tp->tt_2msl, tcp_maxidle,
1231 				      tcp_timer_2msl, tp);
1232 	}
1233 	return (tp);
1234 }
1235