xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: tcp_subr.c,v 1.92 2000/06/30 16:44:34 itojun Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38  * Facility, NASA Ames Research Center.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. All advertising materials mentioning features or use of this software
82  *    must display the following acknowledgement:
83  *	This product includes software developed by the University of
84  *	California, Berkeley and its contributors.
85  * 4. Neither the name of the University nor the names of its contributors
86  *    may be used to endorse or promote products derived from this software
87  *    without specific prior written permission.
88  *
89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
99  * SUCH DAMAGE.
100  *
101  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
102  */
103 
104 #include "opt_inet.h"
105 #include "opt_ipsec.h"
106 #include "opt_tcp_compat_42.h"
107 #include "rnd.h"
108 
109 #include <sys/param.h>
110 #include <sys/proc.h>
111 #include <sys/systm.h>
112 #include <sys/malloc.h>
113 #include <sys/mbuf.h>
114 #include <sys/socket.h>
115 #include <sys/socketvar.h>
116 #include <sys/protosw.h>
117 #include <sys/errno.h>
118 #include <sys/kernel.h>
119 #include <sys/pool.h>
120 #if NRND > 0
121 #include <sys/rnd.h>
122 #endif
123 
124 #include <net/route.h>
125 #include <net/if.h>
126 
127 #include <netinet/in.h>
128 #include <netinet/in_systm.h>
129 #include <netinet/ip.h>
130 #include <netinet/in_pcb.h>
131 #include <netinet/ip_var.h>
132 #include <netinet/ip_icmp.h>
133 
134 #ifdef INET6
135 #ifndef INET
136 #include <netinet/in.h>
137 #endif
138 #include <netinet/ip6.h>
139 #include <netinet6/in6_pcb.h>
140 #include <netinet6/ip6_var.h>
141 #include <netinet6/in6_var.h>
142 #include <netinet6/ip6protosw.h>
143 #endif
144 
145 #include <netinet/tcp.h>
146 #include <netinet/tcp_fsm.h>
147 #include <netinet/tcp_seq.h>
148 #include <netinet/tcp_timer.h>
149 #include <netinet/tcp_var.h>
150 #include <netinet/tcpip.h>
151 
152 #ifdef IPSEC
153 #include <netinet6/ipsec.h>
154 #endif /*IPSEC*/
155 
156 #ifdef INET6
157 struct in6pcb tcb6;
158 #endif
159 
160 /* patchable/settable parameters for tcp */
161 int 	tcp_mssdflt = TCP_MSS;
162 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
163 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
164 int	tcp_do_sack = 1;	/* selective acknowledgement */
165 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
166 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
167 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
168 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
169 int	tcp_init_win = 1;
170 int	tcp_mss_ifmtu = 0;
171 #ifdef TCP_COMPAT_42
172 int	tcp_compat_42 = 1;
173 #else
174 int	tcp_compat_42 = 0;
175 #endif
176 
177 #ifndef TCBHASHSIZE
178 #define	TCBHASHSIZE	128
179 #endif
180 int	tcbhashsize = TCBHASHSIZE;
181 
182 int	tcp_freeq __P((struct tcpcb *));
183 
184 struct pool tcpcb_pool;
185 
186 /*
187  * Tcp initialization
188  */
189 void
190 tcp_init()
191 {
192 	int hlen;
193 
194 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
195 	    0, NULL, NULL, M_PCB);
196 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
197 #ifdef INET6
198 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
199 #endif
200 	LIST_INIT(&tcp_delacks);
201 
202 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
203 #ifdef INET6
204 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
205 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
206 #endif
207 	if (max_protohdr < hlen)
208 		max_protohdr = hlen;
209 	if (max_linkhdr + hlen > MHLEN)
210 		panic("tcp_init");
211 
212 	/* Initialize the compressed state engine. */
213 	syn_cache_init();
214 }
215 
216 /*
217  * Create template to be used to send tcp packets on a connection.
218  * Call after host entry created, allocates an mbuf and fills
219  * in a skeletal tcp/ip header, minimizing the amount of work
220  * necessary when the connection is used.
221  */
222 struct mbuf *
223 tcp_template(tp)
224 	struct tcpcb *tp;
225 {
226 	struct inpcb *inp = tp->t_inpcb;
227 #ifdef INET6
228 	struct in6pcb *in6p = tp->t_in6pcb;
229 #endif
230 	struct tcphdr *n;
231 	struct mbuf *m;
232 	int hlen;
233 
234 	switch (tp->t_family) {
235 	case AF_INET:
236 		hlen = sizeof(struct ip);
237 		if (inp)
238 			break;
239 #ifdef INET6
240 		if (in6p) {
241 			/* mapped addr case */
242 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
243 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
244 				break;
245 		}
246 #endif
247 		return NULL;	/*EINVAL*/
248 #ifdef INET6
249 	case AF_INET6:
250 		hlen = sizeof(struct ip6_hdr);
251 		if (in6p) {
252 			/* more sainty check? */
253 			break;
254 		}
255 		return NULL;	/*EINVAL*/
256 #endif
257 	default:
258 		hlen = 0;	/*pacify gcc*/
259 		return NULL;	/*EAFNOSUPPORT*/
260 	}
261 	if ((m = tp->t_template) == 0) {
262 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
263 		if (m) {
264 			MCLGET(m, M_DONTWAIT);
265 			if ((m->m_flags & M_EXT) == 0) {
266 				m_free(m);
267 				m = NULL;
268 			}
269 		}
270 		if (m == NULL)
271 			return NULL;
272 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
273 	}
274 	bzero(mtod(m, caddr_t), m->m_len);
275 	switch (tp->t_family) {
276 	case AF_INET:
277 	    {
278 		struct ipovly *ipov;
279 		mtod(m, struct ip *)->ip_v = 4;
280 		ipov = mtod(m, struct ipovly *);
281 		ipov->ih_pr = IPPROTO_TCP;
282 		ipov->ih_len = htons(sizeof(struct tcphdr));
283 		if (inp) {
284 			ipov->ih_src = inp->inp_laddr;
285 			ipov->ih_dst = inp->inp_faddr;
286 		}
287 #ifdef INET6
288 		else if (in6p) {
289 			/* mapped addr case */
290 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
291 				sizeof(ipov->ih_src));
292 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
293 				sizeof(ipov->ih_dst));
294 		}
295 #endif
296 		break;
297 	    }
298 #ifdef INET6
299 	case AF_INET6:
300 	    {
301 		struct ip6_hdr *ip6;
302 		mtod(m, struct ip *)->ip_v = 6;
303 		ip6 = mtod(m, struct ip6_hdr *);
304 		ip6->ip6_nxt = IPPROTO_TCP;
305 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
306 		ip6->ip6_src = in6p->in6p_laddr;
307 		ip6->ip6_dst = in6p->in6p_faddr;
308 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
309 		if (ip6_auto_flowlabel) {
310 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
311 			ip6->ip6_flow |=
312 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
313 		}
314 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
315 		ip6->ip6_vfc |= IPV6_VERSION;
316 		break;
317 	    }
318 #endif
319 	}
320 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
321 	if (inp) {
322 		n->th_sport = inp->inp_lport;
323 		n->th_dport = inp->inp_fport;
324 	}
325 #ifdef INET6
326 	else if (in6p) {
327 		n->th_sport = in6p->in6p_lport;
328 		n->th_dport = in6p->in6p_fport;
329 	}
330 #endif
331 	n->th_seq = 0;
332 	n->th_ack = 0;
333 	n->th_x2 = 0;
334 	n->th_off = 5;
335 	n->th_flags = 0;
336 	n->th_win = 0;
337 	n->th_sum = 0;
338 	n->th_urp = 0;
339 	return (m);
340 }
341 
342 /*
343  * Send a single message to the TCP at address specified by
344  * the given TCP/IP header.  If m == 0, then we make a copy
345  * of the tcpiphdr at ti and send directly to the addressed host.
346  * This is used to force keep alive messages out using the TCP
347  * template for a connection tp->t_template.  If flags are given
348  * then we send a message back to the TCP which originated the
349  * segment ti, and discard the mbuf containing it and any other
350  * attached mbufs.
351  *
352  * In any case the ack and sequence number of the transmitted
353  * segment are as specified by the parameters.
354  */
355 int
356 tcp_respond(tp, template, m, th0, ack, seq, flags)
357 	struct tcpcb *tp;
358 	struct mbuf *template;
359 	struct mbuf *m;
360 	struct tcphdr *th0;
361 	tcp_seq ack, seq;
362 	int flags;
363 {
364 #ifndef INET6
365 	struct route iproute;
366 #else
367 	struct route_in6 iproute;	/* sizeof(route_in6) > sizeof(route) */
368 #endif
369 	struct route *ro;
370 	struct rtentry *rt;
371 	int error, tlen, win = 0;
372 	int hlen;
373 	struct ip *ip;
374 #ifdef INET6
375 	struct ip6_hdr *ip6;
376 #endif
377 	int family;	/* family on packet, not inpcb/in6pcb! */
378 	struct tcphdr *th;
379 
380 	if (tp != NULL && (flags & TH_RST) == 0) {
381 		if (tp->t_inpcb)
382 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
383 #ifdef INET6
384 		else if (tp->t_in6pcb)
385 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
386 #endif
387 	}
388 
389 	ip = NULL;
390 #ifdef INET6
391 	ip6 = NULL;
392 #endif
393 	if (m == 0) {
394 		if (!template)
395 			return EINVAL;
396 
397 		/* get family information from template */
398 		switch (mtod(template, struct ip *)->ip_v) {
399 		case 4:
400 			family = AF_INET;
401 			hlen = sizeof(struct ip);
402 			break;
403 #ifdef INET6
404 		case 6:
405 			family = AF_INET6;
406 			hlen = sizeof(struct ip6_hdr);
407 			break;
408 #endif
409 		default:
410 			return EAFNOSUPPORT;
411 		}
412 
413 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
414 		if (m) {
415 			MCLGET(m, M_DONTWAIT);
416 			if ((m->m_flags & M_EXT) == 0) {
417 				m_free(m);
418 				m = NULL;
419 			}
420 		}
421 		if (m == NULL)
422 			return (ENOBUFS);
423 
424 		if (tcp_compat_42)
425 			tlen = 1;
426 		else
427 			tlen = 0;
428 
429 		m->m_data += max_linkhdr;
430 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
431 			template->m_len);
432 		switch (family) {
433 		case AF_INET:
434 			ip = mtod(m, struct ip *);
435 			th = (struct tcphdr *)(ip + 1);
436 			break;
437 #ifdef INET6
438 		case AF_INET6:
439 			ip6 = mtod(m, struct ip6_hdr *);
440 			th = (struct tcphdr *)(ip6 + 1);
441 			break;
442 #endif
443 #if 0
444 		default:
445 			/* noone will visit here */
446 			m_freem(m);
447 			return EAFNOSUPPORT;
448 #endif
449 		}
450 		flags = TH_ACK;
451 	} else {
452 
453 		if ((m->m_flags & M_PKTHDR) == 0) {
454 #if 0
455 			printf("non PKTHDR to tcp_respond\n");
456 #endif
457 			m_freem(m);
458 			return EINVAL;
459 		}
460 #ifdef DIAGNOSTIC
461 		if (!th0)
462 			panic("th0 == NULL in tcp_respond");
463 #endif
464 
465 		/* get family information from m */
466 		switch (mtod(m, struct ip *)->ip_v) {
467 		case 4:
468 			family = AF_INET;
469 			hlen = sizeof(struct ip);
470 			ip = mtod(m, struct ip *);
471 			break;
472 #ifdef INET6
473 		case 6:
474 			family = AF_INET6;
475 			hlen = sizeof(struct ip6_hdr);
476 			ip6 = mtod(m, struct ip6_hdr *);
477 			break;
478 #endif
479 		default:
480 			m_freem(m);
481 			return EAFNOSUPPORT;
482 		}
483 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
484 			tlen = sizeof(*th0);
485 		else
486 			tlen = th0->th_off << 2;
487 
488 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
489 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
490 			m->m_len = hlen + tlen;
491 			m_freem(m->m_next);
492 			m->m_next = NULL;
493 		} else {
494 			struct mbuf *n;
495 
496 #ifdef DIAGNOSTIC
497 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
498 				m_freem(m);
499 				return EMSGSIZE;
500 			}
501 #endif
502 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
503 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
504 				MCLGET(n, M_DONTWAIT);
505 				if ((n->m_flags & M_EXT) == 0) {
506 					m_freem(n);
507 					n = NULL;
508 				}
509 			}
510 			if (!n) {
511 				m_freem(m);
512 				return ENOBUFS;
513 			}
514 
515 			n->m_data += max_linkhdr;
516 			n->m_len = hlen + tlen;
517 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
518 			m_copyback(n, hlen, tlen, (caddr_t)th0);
519 
520 			m_freem(m);
521 			m = n;
522 			n = NULL;
523 		}
524 
525 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
526 		switch (family) {
527 		case AF_INET:
528 			ip = mtod(m, struct ip *);
529 			th = (struct tcphdr *)(ip + 1);
530 			ip->ip_p = IPPROTO_TCP;
531 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
532 			ip->ip_p = IPPROTO_TCP;
533 			break;
534 #ifdef INET6
535 		case AF_INET6:
536 			ip6 = mtod(m, struct ip6_hdr *);
537 			th = (struct tcphdr *)(ip6 + 1);
538 			ip6->ip6_nxt = IPPROTO_TCP;
539 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
540 			ip6->ip6_nxt = IPPROTO_TCP;
541 			break;
542 #endif
543 #if 0
544 		default:
545 			/* noone will visit here */
546 			m_freem(m);
547 			return EAFNOSUPPORT;
548 #endif
549 		}
550 		xchg(th->th_dport, th->th_sport, u_int16_t);
551 #undef xchg
552 		tlen = 0;	/*be friendly with the following code*/
553 	}
554 	th->th_seq = htonl(seq);
555 	th->th_ack = htonl(ack);
556 	th->th_x2 = 0;
557 	if ((flags & TH_SYN) == 0) {
558 		if (tp)
559 			win >>= tp->rcv_scale;
560 		if (win > TCP_MAXWIN)
561 			win = TCP_MAXWIN;
562 		th->th_win = htons((u_int16_t)win);
563 		th->th_off = sizeof (struct tcphdr) >> 2;
564 		tlen += sizeof(*th);
565 	} else
566 		tlen += th->th_off << 2;
567 	m->m_len = hlen + tlen;
568 	m->m_pkthdr.len = hlen + tlen;
569 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
570 	th->th_flags = flags;
571 	th->th_urp = 0;
572 
573 	switch (family) {
574 	case AF_INET:
575 	    {
576 		struct ipovly *ipov = (struct ipovly *)ip;
577 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
578 		ipov->ih_len = htons((u_int16_t)tlen);
579 
580 		th->th_sum = 0;
581 		th->th_sum = in_cksum(m, hlen + tlen);
582 		ip->ip_len = hlen + tlen;	/*will be flipped on output*/
583 		ip->ip_ttl = ip_defttl;
584 		break;
585 	    }
586 #ifdef INET6
587 	case AF_INET6:
588 	    {
589 		th->th_sum = 0;
590 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
591 				tlen);
592 		ip6->ip6_plen = ntohs(tlen);
593 		if (tp && tp->t_in6pcb) {
594 			struct ifnet *oifp;
595 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
596 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
597 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
598 		} else
599 			ip6->ip6_hlim = ip6_defhlim;
600 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
601 		if (ip6_auto_flowlabel) {
602 			ip6->ip6_flow |=
603 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
604 		}
605 		break;
606 	    }
607 #endif
608 	}
609 
610 #ifdef IPSEC
611 	ipsec_setsocket(m, NULL);
612 #endif /*IPSEC*/
613 
614 	/*
615 	 * If we're doing Path MTU discovery, we need to set DF unless
616 	 * the route's MTU is locked.  If we lack a route, we need to
617 	 * look it up now.
618 	 *
619 	 * ip_output() could do this for us, but it's convenient to just
620 	 * do it here unconditionally.
621 	 */
622 	if (tp != NULL && tp->t_inpcb != NULL) {
623 		ro = &tp->t_inpcb->inp_route;
624 #ifdef IPSEC
625 		ipsec_setsocket(m, tp->t_inpcb->inp_socket);
626 #endif
627 #ifdef DIAGNOSTIC
628 		if (family != AF_INET)
629 			panic("tcp_respond: address family mismatch");
630 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
631 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
632 			    ntohl(ip->ip_dst.s_addr),
633 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
634 		}
635 #endif
636 	}
637 #ifdef INET6
638 	else if (tp != NULL && tp->t_in6pcb != NULL) {
639 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
640 #ifdef IPSEC
641 		ipsec_setsocket(m, tp->t_in6pcb->in6p_socket);
642 #endif
643 #ifdef DIAGNOSTIC
644 		if (family == AF_INET) {
645 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
646 				panic("tcp_respond: not mapped addr");
647 			if (bcmp(&ip->ip_dst,
648 					&tp->t_in6pcb->in6p_faddr.s6_addr32[3],
649 					sizeof(ip->ip_dst)) != 0) {
650 				panic("tcp_respond: ip_dst != in6p_faddr");
651 			}
652 		} else if (family == AF_INET6) {
653 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr))
654 				panic("tcp_respond: ip6_dst != in6p_faddr");
655 		} else
656 			panic("tcp_respond: address family mismatch");
657 #endif
658 	}
659 #endif
660 	else {
661 		ro = (struct route *)&iproute;
662 		bzero(ro, sizeof(iproute));
663 	}
664 	if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) {
665 		if (ro->ro_rt != NULL) {
666 			RTFREE(ro->ro_rt);
667 			ro->ro_rt = NULL;
668 		}
669 		switch (family) {
670 		case AF_INET:
671 		    {
672 			struct sockaddr_in *dst;
673 			dst = satosin(&ro->ro_dst);
674 			dst->sin_family = AF_INET;
675 			dst->sin_len = sizeof(*dst);
676 			dst->sin_addr = ip->ip_dst;
677 			break;
678 		    }
679 #ifdef INET6
680 		case AF_INET6:
681 		    {
682 			struct sockaddr_in6 *dst;
683 			dst = satosin6(&ro->ro_dst);
684 			bzero(dst, sizeof(*dst));
685 			dst->sin6_family = AF_INET6;
686 			dst->sin6_len = sizeof(*dst);
687 			dst->sin6_addr = ip6->ip6_dst;
688 			break;
689 		    }
690 #endif
691 		}
692 		rtalloc(ro);
693 		if ((rt = ro->ro_rt) == NULL) {
694 			m_freem(m);
695 			switch (family) {
696 			case AF_INET:
697 				ipstat.ips_noroute++;
698 				break;
699 #ifdef INET6
700 			case AF_INET6:
701 				ip6stat.ip6s_noroute++;
702 				break;
703 #endif
704 			}
705 			return (EHOSTUNREACH);
706 		}
707 	}
708 	switch (family) {
709 	case AF_INET:
710 		if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
711 			ip->ip_off |= IP_DF;
712 
713 		error = ip_output(m, NULL, ro, 0, NULL);
714 		break;
715 #ifdef INET6
716 	case AF_INET6:
717 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
718 			NULL);
719 		break;
720 #endif
721 	default:
722 		error = EAFNOSUPPORT;
723 		break;
724 	}
725 
726 	if (ro == (struct route *)&iproute) {
727 		RTFREE(ro->ro_rt);
728 		ro->ro_rt = NULL;
729 	}
730 
731 	return (error);
732 }
733 
734 /*
735  * Create a new TCP control block, making an
736  * empty reassembly queue and hooking it to the argument
737  * protocol control block.
738  */
739 struct tcpcb *
740 tcp_newtcpcb(family, aux)
741 	int family;	/* selects inpcb, or in6pcb */
742 	void *aux;
743 {
744 	struct tcpcb *tp;
745 
746 	switch (family) {
747 	case PF_INET:
748 		break;
749 #ifdef INET6
750 	case PF_INET6:
751 		break;
752 #endif
753 	default:
754 		return NULL;
755 	}
756 
757 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
758 	if (tp == NULL)
759 		return (NULL);
760 	bzero((caddr_t)tp, sizeof(struct tcpcb));
761 	LIST_INIT(&tp->segq);
762 	LIST_INIT(&tp->timeq);
763 	tp->t_family = family;		/* may be overridden later on */
764 	tp->t_peermss = tcp_mssdflt;
765 	tp->t_ourmss = tcp_mssdflt;
766 	tp->t_segsz = tcp_mssdflt;
767 	LIST_INIT(&tp->t_sc);
768 
769 	tp->t_flags = 0;
770 	if (tcp_do_rfc1323 && tcp_do_win_scale)
771 		tp->t_flags |= TF_REQ_SCALE;
772 	if (tcp_do_rfc1323 && tcp_do_timestamps)
773 		tp->t_flags |= TF_REQ_TSTMP;
774 	if (tcp_do_sack == 2)
775 		tp->t_flags |= TF_WILL_SACK;
776 	else if (tcp_do_sack == 1)
777 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
778 	tp->t_flags |= TF_CANT_TXSACK;
779 	switch (family) {
780 	case PF_INET:
781 		tp->t_inpcb = (struct inpcb *)aux;
782 		break;
783 #ifdef INET6
784 	case PF_INET6:
785 		tp->t_in6pcb = (struct in6pcb *)aux;
786 		break;
787 #endif
788 	}
789 	/*
790 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
791 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
792 	 * reasonable initial retransmit time.
793 	 */
794 	tp->t_srtt = TCPTV_SRTTBASE;
795 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
796 	tp->t_rttmin = TCPTV_MIN;
797 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
798 	    TCPTV_MIN, TCPTV_REXMTMAX);
799 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
800 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
801 	if (family == AF_INET) {
802 		struct inpcb *inp = (struct inpcb *)aux;
803 		inp->inp_ip.ip_ttl = ip_defttl;
804 		inp->inp_ppcb = (caddr_t)tp;
805 	}
806 #ifdef INET6
807 	else if (family == AF_INET6) {
808 		struct in6pcb *in6p = (struct in6pcb *)aux;
809 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
810 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
811 					       : NULL);
812 		in6p->in6p_ppcb = (caddr_t)tp;
813 	}
814 #endif
815 	return (tp);
816 }
817 
818 /*
819  * Drop a TCP connection, reporting
820  * the specified error.  If connection is synchronized,
821  * then send a RST to peer.
822  */
823 struct tcpcb *
824 tcp_drop(tp, errno)
825 	struct tcpcb *tp;
826 	int errno;
827 {
828 	struct socket *so;
829 
830 	if (tp->t_inpcb)
831 		so = tp->t_inpcb->inp_socket;
832 #ifdef INET6
833 	else if (tp->t_in6pcb)
834 		so = tp->t_in6pcb->in6p_socket;
835 #endif
836 	else
837 		return NULL;
838 
839 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
840 		tp->t_state = TCPS_CLOSED;
841 		(void) tcp_output(tp);
842 		tcpstat.tcps_drops++;
843 	} else
844 		tcpstat.tcps_conndrops++;
845 	if (errno == ETIMEDOUT && tp->t_softerror)
846 		errno = tp->t_softerror;
847 	so->so_error = errno;
848 	return (tcp_close(tp));
849 }
850 
851 /*
852  * Close a TCP control block:
853  *	discard all space held by the tcp
854  *	discard internet protocol block
855  *	wake up any sleepers
856  */
857 struct tcpcb *
858 tcp_close(tp)
859 	struct tcpcb *tp;
860 {
861 	struct inpcb *inp;
862 #ifdef INET6
863 	struct in6pcb *in6p;
864 #endif
865 	struct socket *so;
866 #ifdef RTV_RTT
867 	struct rtentry *rt;
868 #endif
869 	struct route *ro;
870 
871 	inp = tp->t_inpcb;
872 #ifdef INET6
873 	in6p = tp->t_in6pcb;
874 #endif
875 	so = NULL;
876 	ro = NULL;
877 	if (inp) {
878 		so = inp->inp_socket;
879 		ro = &inp->inp_route;
880 	}
881 #ifdef INET6
882 	else if (in6p) {
883 		so = in6p->in6p_socket;
884 		ro = (struct route *)&in6p->in6p_route;
885 	}
886 #endif
887 
888 #ifdef RTV_RTT
889 	/*
890 	 * If we sent enough data to get some meaningful characteristics,
891 	 * save them in the routing entry.  'Enough' is arbitrarily
892 	 * defined as the sendpipesize (default 4K) * 16.  This would
893 	 * give us 16 rtt samples assuming we only get one sample per
894 	 * window (the usual case on a long haul net).  16 samples is
895 	 * enough for the srtt filter to converge to within 5% of the correct
896 	 * value; fewer samples and we could save a very bogus rtt.
897 	 *
898 	 * Don't update the default route's characteristics and don't
899 	 * update anything that the user "locked".
900 	 */
901 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
902 	    ro && (rt = ro->ro_rt) &&
903 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
904 		u_long i = 0;
905 
906 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
907 			i = tp->t_srtt *
908 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
909 			if (rt->rt_rmx.rmx_rtt && i)
910 				/*
911 				 * filter this update to half the old & half
912 				 * the new values, converting scale.
913 				 * See route.h and tcp_var.h for a
914 				 * description of the scaling constants.
915 				 */
916 				rt->rt_rmx.rmx_rtt =
917 				    (rt->rt_rmx.rmx_rtt + i) / 2;
918 			else
919 				rt->rt_rmx.rmx_rtt = i;
920 		}
921 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
922 			i = tp->t_rttvar *
923 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
924 			if (rt->rt_rmx.rmx_rttvar && i)
925 				rt->rt_rmx.rmx_rttvar =
926 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
927 			else
928 				rt->rt_rmx.rmx_rttvar = i;
929 		}
930 		/*
931 		 * update the pipelimit (ssthresh) if it has been updated
932 		 * already or if a pipesize was specified & the threshhold
933 		 * got below half the pipesize.  I.e., wait for bad news
934 		 * before we start updating, then update on both good
935 		 * and bad news.
936 		 */
937 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
938 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
939 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
940 			/*
941 			 * convert the limit from user data bytes to
942 			 * packets then to packet data bytes.
943 			 */
944 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
945 			if (i < 2)
946 				i = 2;
947 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
948 			if (rt->rt_rmx.rmx_ssthresh)
949 				rt->rt_rmx.rmx_ssthresh =
950 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
951 			else
952 				rt->rt_rmx.rmx_ssthresh = i;
953 		}
954 	}
955 #endif /* RTV_RTT */
956 	/* free the reassembly queue, if any */
957 	TCP_REASS_LOCK(tp);
958 	(void) tcp_freeq(tp);
959 	TCP_REASS_UNLOCK(tp);
960 
961 	TCP_CLEAR_DELACK(tp);
962 	syn_cache_cleanup(tp);
963 
964 	if (tp->t_template) {
965 		m_free(tp->t_template);
966 		tp->t_template = NULL;
967 	}
968 	pool_put(&tcpcb_pool, tp);
969 	if (inp) {
970 		inp->inp_ppcb = 0;
971 		soisdisconnected(so);
972 		in_pcbdetach(inp);
973 	}
974 #ifdef INET6
975 	else if (in6p) {
976 		in6p->in6p_ppcb = 0;
977 		soisdisconnected(so);
978 		in6_pcbdetach(in6p);
979 	}
980 #endif
981 	tcpstat.tcps_closed++;
982 	return ((struct tcpcb *)0);
983 }
984 
985 int
986 tcp_freeq(tp)
987 	struct tcpcb *tp;
988 {
989 	struct ipqent *qe;
990 	int rv = 0;
991 #ifdef TCPREASS_DEBUG
992 	int i = 0;
993 #endif
994 
995 	TCP_REASS_LOCK_CHECK(tp);
996 
997 	while ((qe = tp->segq.lh_first) != NULL) {
998 #ifdef TCPREASS_DEBUG
999 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1000 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1001 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1002 #endif
1003 		LIST_REMOVE(qe, ipqe_q);
1004 		LIST_REMOVE(qe, ipqe_timeq);
1005 		m_freem(qe->ipqe_m);
1006 		pool_put(&ipqent_pool, qe);
1007 		rv = 1;
1008 	}
1009 	return (rv);
1010 }
1011 
1012 /*
1013  * Protocol drain routine.  Called when memory is in short supply.
1014  */
1015 void
1016 tcp_drain()
1017 {
1018 	struct inpcb *inp;
1019 	struct tcpcb *tp;
1020 
1021 	/*
1022 	 * Free the sequence queue of all TCP connections.
1023 	 */
1024 	inp = tcbtable.inpt_queue.cqh_first;
1025 	if (inp)						/* XXX */
1026 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
1027 	    inp = inp->inp_queue.cqe_next) {
1028 		if ((tp = intotcpcb(inp)) != NULL) {
1029 			/*
1030 			 * We may be called from a device's interrupt
1031 			 * context.  If the tcpcb is already busy,
1032 			 * just bail out now.
1033 			 */
1034 			if (tcp_reass_lock_try(tp) == 0)
1035 				continue;
1036 			if (tcp_freeq(tp))
1037 				tcpstat.tcps_connsdrained++;
1038 			TCP_REASS_UNLOCK(tp);
1039 		}
1040 	}
1041 }
1042 
1043 /*
1044  * Notify a tcp user of an asynchronous error;
1045  * store error as soft error, but wake up user
1046  * (for now, won't do anything until can select for soft error).
1047  */
1048 void
1049 tcp_notify(inp, error)
1050 	struct inpcb *inp;
1051 	int error;
1052 {
1053 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1054 	struct socket *so = inp->inp_socket;
1055 
1056 	/*
1057 	 * Ignore some errors if we are hooked up.
1058 	 * If connection hasn't completed, has retransmitted several times,
1059 	 * and receives a second error, give up now.  This is better
1060 	 * than waiting a long time to establish a connection that
1061 	 * can never complete.
1062 	 */
1063 	if (tp->t_state == TCPS_ESTABLISHED &&
1064 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1065 	      error == EHOSTDOWN)) {
1066 		return;
1067 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1068 	    tp->t_rxtshift > 3 && tp->t_softerror)
1069 		so->so_error = error;
1070 	else
1071 		tp->t_softerror = error;
1072 	wakeup((caddr_t) &so->so_timeo);
1073 	sorwakeup(so);
1074 	sowwakeup(so);
1075 }
1076 
1077 #if defined(INET6) && !defined(TCP6)
1078 void
1079 tcp6_notify(in6p, error)
1080 	struct in6pcb *in6p;
1081 	int error;
1082 {
1083 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1084 	struct socket *so = in6p->in6p_socket;
1085 
1086 	/*
1087 	 * Ignore some errors if we are hooked up.
1088 	 * If connection hasn't completed, has retransmitted several times,
1089 	 * and receives a second error, give up now.  This is better
1090 	 * than waiting a long time to establish a connection that
1091 	 * can never complete.
1092 	 */
1093 	if (tp->t_state == TCPS_ESTABLISHED &&
1094 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1095 	      error == EHOSTDOWN)) {
1096 		return;
1097 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1098 	    tp->t_rxtshift > 3 && tp->t_softerror)
1099 		so->so_error = error;
1100 	else
1101 		tp->t_softerror = error;
1102 	wakeup((caddr_t) &so->so_timeo);
1103 	sorwakeup(so);
1104 	sowwakeup(so);
1105 }
1106 #endif
1107 
1108 #if defined(INET6) && !defined(TCP6)
1109 void
1110 tcp6_ctlinput(cmd, sa, d)
1111 	int cmd;
1112 	struct sockaddr *sa;
1113 	void *d;
1114 {
1115 	struct tcphdr *thp;
1116 	struct tcphdr th;
1117 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1118 	int nmatch;
1119 	struct sockaddr_in6 sa6;
1120 	struct ip6_hdr *ip6;
1121 	struct mbuf *m;
1122 	int off;
1123 
1124 	if (sa->sa_family != AF_INET6 ||
1125 	    sa->sa_len != sizeof(struct sockaddr_in6))
1126 		return;
1127 	if ((unsigned)cmd >= PRC_NCMDS)
1128 		return;
1129 	else if (cmd == PRC_QUENCH) {
1130 		/* XXX there's no PRC_QUENCH in IPv6 */
1131 		notify = tcp6_quench;
1132 	} else if (PRC_IS_REDIRECT(cmd))
1133 		notify = in6_rtchange, d = NULL;
1134 	else if (cmd == PRC_MSGSIZE)
1135 		notify = tcp6_mtudisc, d = NULL;
1136 	else if (cmd == PRC_HOSTDEAD)
1137 		d = NULL;
1138 	else if (inet6ctlerrmap[cmd] == 0)
1139 		return;
1140 
1141 	/* if the parameter is from icmp6, decode it. */
1142 	if (d != NULL) {
1143 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1144 		m = ip6cp->ip6c_m;
1145 		ip6 = ip6cp->ip6c_ip6;
1146 		off = ip6cp->ip6c_off;
1147 	} else {
1148 		m = NULL;
1149 		ip6 = NULL;
1150 	}
1151 
1152 	/* translate addresses into internal form */
1153 	sa6 = *(struct sockaddr_in6 *)sa;
1154 	if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) && m && m->m_pkthdr.rcvif)
1155 		sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1156 
1157 	if (ip6) {
1158 		/*
1159 		 * XXX: We assume that when ip6 is non NULL,
1160 		 * M and OFF are valid.
1161 		 */
1162 		struct in6_addr s;
1163 
1164 		/* translate addresses into internal form */
1165 		memcpy(&s, &ip6->ip6_src, sizeof(s));
1166 		if (IN6_IS_ADDR_LINKLOCAL(&s))
1167 			s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1168 
1169 		if (m->m_len < off + sizeof(th)) {
1170 			/*
1171 			 * this should be rare case,
1172 			 * so we compromise on this copy...
1173 			 */
1174 			m_copydata(m, off, sizeof(th), (caddr_t)&th);
1175 			thp = &th;
1176 		} else
1177 			thp = (struct tcphdr *)(mtod(m, caddr_t) + off);
1178 		nmatch = in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6,
1179 		    thp->th_dport, &s, thp->th_sport, cmd, notify);
1180 		if (nmatch == 0 && syn_cache_count &&
1181 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1182 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1183 		     inet6ctlerrmap[cmd] == EHOSTDOWN)) {
1184 			struct sockaddr_in6 sin6;
1185 			bzero(&sin6, sizeof(sin6));
1186 			sin6.sin6_len = sizeof(sin6);
1187 			sin6.sin6_family = AF_INET6;
1188 			sin6.sin6_port = thp->th_sport;
1189 			sin6.sin6_addr = s;
1190 			syn_cache_unreach((struct sockaddr *)&sin6, sa, thp);
1191 		}
1192 	} else {
1193 		(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 0,
1194 				     &zeroin6_addr, 0, cmd, notify);
1195 	}
1196 }
1197 #endif
1198 
1199 /* assumes that ip header and tcp header are contiguous on mbuf */
1200 void *
1201 tcp_ctlinput(cmd, sa, v)
1202 	int cmd;
1203 	struct sockaddr *sa;
1204 	void *v;
1205 {
1206 	struct ip *ip = v;
1207 	struct tcphdr *th;
1208 	extern int inetctlerrmap[];
1209 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1210 	int errno;
1211 	int nmatch;
1212 
1213 	if (sa->sa_family != AF_INET ||
1214 	    sa->sa_len != sizeof(struct sockaddr_in))
1215 		return NULL;
1216 	if ((unsigned)cmd >= PRC_NCMDS)
1217 		return NULL;
1218 	errno = inetctlerrmap[cmd];
1219 	if (cmd == PRC_QUENCH)
1220 		notify = tcp_quench;
1221 	else if (PRC_IS_REDIRECT(cmd))
1222 		notify = in_rtchange, ip = 0;
1223 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
1224 		notify = tcp_mtudisc, ip = 0;
1225 	else if (cmd == PRC_HOSTDEAD)
1226 		ip = 0;
1227 	else if (errno == 0)
1228 		return NULL;
1229 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1230 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1231 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1232 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1233 		if (nmatch == 0 && syn_cache_count &&
1234 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1235 		    inetctlerrmap[cmd] == ENETUNREACH ||
1236 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1237 			struct sockaddr_in sin;
1238 			bzero(&sin, sizeof(sin));
1239 			sin.sin_len = sizeof(sin);
1240 			sin.sin_family = AF_INET;
1241 			sin.sin_port = th->th_sport;
1242 			sin.sin_addr = ip->ip_src;
1243 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1244 		}
1245 
1246 		/* XXX mapped address case */
1247 	}
1248 	else {
1249 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1250 		    notify);
1251 	}
1252 	return NULL;
1253 }
1254 
1255 /*
1256  * When a source quence is received, we are being notifed of congestion.
1257  * Close the congestion window down to the Loss Window (one segment).
1258  * We will gradually open it again as we proceed.
1259  */
1260 void
1261 tcp_quench(inp, errno)
1262 	struct inpcb *inp;
1263 	int errno;
1264 {
1265 	struct tcpcb *tp = intotcpcb(inp);
1266 
1267 	if (tp)
1268 		tp->snd_cwnd = tp->t_segsz;
1269 }
1270 
1271 #if defined(INET6) && !defined(TCP6)
1272 void
1273 tcp6_quench(in6p, errno)
1274 	struct in6pcb *in6p;
1275 	int errno;
1276 {
1277 	struct tcpcb *tp = in6totcpcb(in6p);
1278 
1279 	if (tp)
1280 		tp->snd_cwnd = tp->t_segsz;
1281 }
1282 #endif
1283 
1284 /*
1285  * On receipt of path MTU corrections, flush old route and replace it
1286  * with the new one.  Retransmit all unacknowledged packets, to ensure
1287  * that all packets will be received.
1288  */
1289 void
1290 tcp_mtudisc(inp, errno)
1291 	struct inpcb *inp;
1292 	int errno;
1293 {
1294 	struct tcpcb *tp = intotcpcb(inp);
1295 	struct rtentry *rt = in_pcbrtentry(inp);
1296 
1297 	if (tp != 0) {
1298 		if (rt != 0) {
1299 			/*
1300 			 * If this was not a host route, remove and realloc.
1301 			 */
1302 			if ((rt->rt_flags & RTF_HOST) == 0) {
1303 				in_rtchange(inp, errno);
1304 				if ((rt = in_pcbrtentry(inp)) == 0)
1305 					return;
1306 			}
1307 
1308 			/*
1309 			 * Slow start out of the error condition.  We
1310 			 * use the MTU because we know it's smaller
1311 			 * than the previously transmitted segment.
1312 			 *
1313 			 * Note: This is more conservative than the
1314 			 * suggestion in draft-floyd-incr-init-win-03.
1315 			 */
1316 			if (rt->rt_rmx.rmx_mtu != 0)
1317 				tp->snd_cwnd =
1318 				    TCP_INITIAL_WINDOW(tcp_init_win,
1319 				    rt->rt_rmx.rmx_mtu);
1320 		}
1321 
1322 		/*
1323 		 * Resend unacknowledged packets.
1324 		 */
1325 		tp->snd_nxt = tp->snd_una;
1326 		tcp_output(tp);
1327 	}
1328 }
1329 
1330 #if defined(INET6) && !defined(TCP6)
1331 void
1332 tcp6_mtudisc(in6p, errno)
1333 	struct in6pcb *in6p;
1334 	int errno;
1335 {
1336 	struct tcpcb *tp = in6totcpcb(in6p);
1337 	struct rtentry *rt = in6_pcbrtentry(in6p);
1338 
1339 	if (tp != 0) {
1340 		if (rt != 0) {
1341 			/*
1342 			 * If this was not a host route, remove and realloc.
1343 			 */
1344 			if ((rt->rt_flags & RTF_HOST) == 0) {
1345 				in6_rtchange(in6p, errno);
1346 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1347 					return;
1348 			}
1349 
1350 			/*
1351 			 * Slow start out of the error condition.  We
1352 			 * use the MTU because we know it's smaller
1353 			 * than the previously transmitted segment.
1354 			 *
1355 			 * Note: This is more conservative than the
1356 			 * suggestion in draft-floyd-incr-init-win-03.
1357 			 */
1358 			if (rt->rt_rmx.rmx_mtu != 0)
1359 				tp->snd_cwnd =
1360 				    TCP_INITIAL_WINDOW(tcp_init_win,
1361 				    rt->rt_rmx.rmx_mtu);
1362 		}
1363 
1364 		/*
1365 		 * Resend unacknowledged packets.
1366 		 */
1367 		tp->snd_nxt = tp->snd_una;
1368 		tcp_output(tp);
1369 	}
1370 }
1371 #endif
1372 
1373 /*
1374  * Compute the MSS to advertise to the peer.  Called only during
1375  * the 3-way handshake.  If we are the server (peer initiated
1376  * connection), we are called with a pointer to the interface
1377  * on which the SYN packet arrived.  If we are the client (we
1378  * initiated connection), we are called with a pointer to the
1379  * interface out which this connection should go.
1380  *
1381  * NOTE: Do not subtract IP option/extension header size nor IPsec
1382  * header size from MSS advertisement.  MSS option must hold the maximum
1383  * segment size we can accept, so it must always be:
1384  *	 max(if mtu) - ip header - tcp header
1385  */
1386 u_long
1387 tcp_mss_to_advertise(ifp, af)
1388 	const struct ifnet *ifp;
1389 	int af;
1390 {
1391 	extern u_long in_maxmtu;
1392 	u_long mss = 0;
1393 	u_long hdrsiz;
1394 
1395 	/*
1396 	 * In order to avoid defeating path MTU discovery on the peer,
1397 	 * we advertise the max MTU of all attached networks as our MSS,
1398 	 * per RFC 1191, section 3.1.
1399 	 *
1400 	 * We provide the option to advertise just the MTU of
1401 	 * the interface on which we hope this connection will
1402 	 * be receiving.  If we are responding to a SYN, we
1403 	 * will have a pretty good idea about this, but when
1404 	 * initiating a connection there is a bit more doubt.
1405 	 *
1406 	 * We also need to ensure that loopback has a large enough
1407 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1408 	 */
1409 
1410 	if (ifp != NULL)
1411 		mss = ifp->if_mtu;
1412 
1413 	if (tcp_mss_ifmtu == 0)
1414 		mss = max(in_maxmtu, mss);
1415 
1416 	switch (af) {
1417 	case AF_INET:
1418 		hdrsiz = sizeof(struct ip);
1419 		break;
1420 #ifdef INET6
1421 	case AF_INET6:
1422 		hdrsiz = sizeof(struct ip6_hdr);
1423 		break;
1424 #endif
1425 	default:
1426 		hdrsiz = 0;
1427 		break;
1428 	}
1429 	hdrsiz += sizeof(struct tcphdr);
1430 	if (mss > hdrsiz)
1431 		mss -= hdrsiz;
1432 
1433 	mss = max(tcp_mssdflt, mss);
1434 	return (mss);
1435 }
1436 
1437 /*
1438  * Set connection variables based on the peer's advertised MSS.
1439  * We are passed the TCPCB for the actual connection.  If we
1440  * are the server, we are called by the compressed state engine
1441  * when the 3-way handshake is complete.  If we are the client,
1442  * we are called when we recieve the SYN,ACK from the server.
1443  *
1444  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1445  * before this routine is called!
1446  */
1447 void
1448 tcp_mss_from_peer(tp, offer)
1449 	struct tcpcb *tp;
1450 	int offer;
1451 {
1452 	struct socket *so;
1453 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1454 	struct rtentry *rt;
1455 #endif
1456 	u_long bufsize;
1457 	int mss;
1458 
1459 	so = NULL;
1460 	rt = NULL;
1461 	if (tp->t_inpcb) {
1462 		so = tp->t_inpcb->inp_socket;
1463 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1464 		rt = in_pcbrtentry(tp->t_inpcb);
1465 #endif
1466 	}
1467 #ifdef INET6
1468 	else if (tp->t_in6pcb) {
1469 		so = tp->t_in6pcb->in6p_socket;
1470 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1471 #ifdef TCP6
1472 		rt = NULL;
1473 #else
1474 		rt = in6_pcbrtentry(tp->t_in6pcb);
1475 #endif
1476 #endif
1477 	}
1478 #endif
1479 
1480 	/*
1481 	 * As per RFC1122, use the default MSS value, unless they
1482 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1483 	 */
1484 	mss = tcp_mssdflt;
1485 	if (offer)
1486 		mss = offer;
1487 	mss = max(mss, 32);		/* sanity */
1488 	tp->t_peermss = mss;
1489 	mss -= tcp_optlen(tp);
1490 	if (tp->t_inpcb)
1491 		mss -= ip_optlen(tp->t_inpcb);
1492 #ifdef INET6
1493 	else if (tp->t_in6pcb)
1494 		mss -= ip6_optlen(tp->t_in6pcb);
1495 #endif
1496 
1497 	/*
1498 	 * If there's a pipesize, change the socket buffer to that size.
1499 	 * Make the socket buffer an integral number of MSS units.  If
1500 	 * the MSS is larger than the socket buffer, artificially decrease
1501 	 * the MSS.
1502 	 */
1503 #ifdef RTV_SPIPE
1504 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1505 		bufsize = rt->rt_rmx.rmx_sendpipe;
1506 	else
1507 #endif
1508 		bufsize = so->so_snd.sb_hiwat;
1509 	if (bufsize < mss)
1510 		mss = bufsize;
1511 	else {
1512 		bufsize = roundup(bufsize, mss);
1513 		if (bufsize > sb_max)
1514 			bufsize = sb_max;
1515 		(void) sbreserve(&so->so_snd, bufsize);
1516 	}
1517 	tp->t_segsz = mss;
1518 
1519 #ifdef RTV_SSTHRESH
1520 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1521 		/*
1522 		 * There's some sort of gateway or interface buffer
1523 		 * limit on the path.  Use this to set the slow
1524 		 * start threshold, but set the threshold to no less
1525 		 * than 2 * MSS.
1526 		 */
1527 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1528 	}
1529 #endif
1530 }
1531 
1532 /*
1533  * Processing necessary when a TCP connection is established.
1534  */
1535 void
1536 tcp_established(tp)
1537 	struct tcpcb *tp;
1538 {
1539 	struct socket *so;
1540 #ifdef RTV_RPIPE
1541 	struct rtentry *rt;
1542 #endif
1543 	u_long bufsize;
1544 
1545 	so = NULL;
1546 	rt = NULL;
1547 	if (tp->t_inpcb) {
1548 		so = tp->t_inpcb->inp_socket;
1549 #if defined(RTV_RPIPE)
1550 		rt = in_pcbrtentry(tp->t_inpcb);
1551 #endif
1552 	}
1553 #ifdef INET6
1554 	else if (tp->t_in6pcb) {
1555 		so = tp->t_in6pcb->in6p_socket;
1556 #if defined(RTV_RPIPE)
1557 #ifdef TCP6
1558 		rt = NULL;
1559 #else
1560 		rt = in6_pcbrtentry(tp->t_in6pcb);
1561 #endif
1562 #endif
1563 	}
1564 #endif
1565 
1566 	tp->t_state = TCPS_ESTABLISHED;
1567 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1568 
1569 #ifdef RTV_RPIPE
1570 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1571 		bufsize = rt->rt_rmx.rmx_recvpipe;
1572 	else
1573 #endif
1574 		bufsize = so->so_rcv.sb_hiwat;
1575 	if (bufsize > tp->t_ourmss) {
1576 		bufsize = roundup(bufsize, tp->t_ourmss);
1577 		if (bufsize > sb_max)
1578 			bufsize = sb_max;
1579 		(void) sbreserve(&so->so_rcv, bufsize);
1580 	}
1581 }
1582 
1583 /*
1584  * Check if there's an initial rtt or rttvar.  Convert from the
1585  * route-table units to scaled multiples of the slow timeout timer.
1586  * Called only during the 3-way handshake.
1587  */
1588 void
1589 tcp_rmx_rtt(tp)
1590 	struct tcpcb *tp;
1591 {
1592 #ifdef RTV_RTT
1593 	struct rtentry *rt = NULL;
1594 	int rtt;
1595 
1596 	if (tp->t_inpcb)
1597 		rt = in_pcbrtentry(tp->t_inpcb);
1598 #ifdef INET6
1599 	else if (tp->t_in6pcb) {
1600 #ifdef TCP6
1601 		rt = NULL;
1602 #else
1603 		rt = in6_pcbrtentry(tp->t_in6pcb);
1604 #endif
1605 	}
1606 #endif
1607 	if (rt == NULL)
1608 		return;
1609 
1610 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1611 		/*
1612 		 * XXX The lock bit for MTU indicates that the value
1613 		 * is also a minimum value; this is subject to time.
1614 		 */
1615 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1616 			TCPT_RANGESET(tp->t_rttmin,
1617 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1618 			    TCPTV_MIN, TCPTV_REXMTMAX);
1619 		tp->t_srtt = rtt /
1620 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1621 		if (rt->rt_rmx.rmx_rttvar) {
1622 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1623 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1624 				(TCP_RTTVAR_SHIFT + 2));
1625 		} else {
1626 			/* Default variation is +- 1 rtt */
1627 			tp->t_rttvar =
1628 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1629 		}
1630 		TCPT_RANGESET(tp->t_rxtcur,
1631 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1632 		    tp->t_rttmin, TCPTV_REXMTMAX);
1633 	}
1634 #endif
1635 }
1636 
1637 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1638 
1639 /*
1640  * Get a new sequence value given a tcp control block
1641  */
1642 tcp_seq
1643 tcp_new_iss(tp, len, addin)
1644 	void            *tp;
1645 	u_long           len;
1646 	tcp_seq		 addin;
1647 {
1648 	tcp_seq          tcp_iss;
1649 
1650 	/*
1651 	 * Randomize.
1652 	 */
1653 #if NRND > 0
1654 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
1655 #else
1656 	tcp_iss = random();
1657 #endif
1658 
1659 	/*
1660 	 * If we were asked to add some amount to a known value,
1661 	 * we will take a random value obtained above, mask off the upper
1662 	 * bits, and add in the known value.  We also add in a constant to
1663 	 * ensure that we are at least a certain distance from the original
1664 	 * value.
1665 	 *
1666 	 * This is used when an old connection is in timed wait
1667 	 * and we have a new one coming in, for instance.
1668 	 */
1669 	if (addin != 0) {
1670 #ifdef TCPISS_DEBUG
1671 		printf("Random %08x, ", tcp_iss);
1672 #endif
1673 		tcp_iss &= TCP_ISS_RANDOM_MASK;
1674 		tcp_iss += addin + TCP_ISSINCR;
1675 #ifdef TCPISS_DEBUG
1676 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
1677 #endif
1678 	} else {
1679 		tcp_iss &= TCP_ISS_RANDOM_MASK;
1680 		tcp_iss += tcp_iss_seq;
1681 		tcp_iss_seq += TCP_ISSINCR;
1682 #ifdef TCPISS_DEBUG
1683 		printf("ISS %08x\n", tcp_iss);
1684 #endif
1685 	}
1686 
1687 	if (tcp_compat_42) {
1688 		/*
1689 		 * Limit it to the positive range for really old TCP
1690 		 * implementations.
1691 		 */
1692 		if (tcp_iss >= 0x80000000)
1693 			tcp_iss &= 0x7fffffff;		/* XXX */
1694 	}
1695 
1696 	return tcp_iss;
1697 }
1698 
1699 #ifdef IPSEC
1700 /* compute ESP/AH header size for TCP, including outer IP header. */
1701 size_t
1702 ipsec4_hdrsiz_tcp(tp)
1703 	struct tcpcb *tp;
1704 {
1705 	struct inpcb *inp;
1706 	size_t hdrsiz;
1707 
1708 	/* XXX mapped addr case (tp->t_in6pcb) */
1709 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1710 		return 0;
1711 	switch (tp->t_family) {
1712 	case AF_INET:
1713 		/* XXX: should use currect direction. */
1714 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
1715 		break;
1716 	default:
1717 		hdrsiz = 0;
1718 		break;
1719 	}
1720 
1721 	return hdrsiz;
1722 }
1723 
1724 #if defined(INET6) && !defined(TCP6)
1725 size_t
1726 ipsec6_hdrsiz_tcp(tp)
1727 	struct tcpcb *tp;
1728 {
1729 	struct in6pcb *in6p;
1730 	size_t hdrsiz;
1731 
1732 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
1733 		return 0;
1734 	switch (tp->t_family) {
1735 	case AF_INET6:
1736 		/* XXX: should use currect direction. */
1737 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
1738 		break;
1739 	case AF_INET:
1740 		/* mapped address case - tricky */
1741 	default:
1742 		hdrsiz = 0;
1743 		break;
1744 	}
1745 
1746 	return hdrsiz;
1747 }
1748 #endif
1749 #endif /*IPSEC*/
1750 
1751 /*
1752  * Determine the length of the TCP options for this connection.
1753  *
1754  * XXX:  What do we do for SACK, when we add that?  Just reserve
1755  *       all of the space?  Otherwise we can't exactly be incrementing
1756  *       cwnd by an amount that varies depending on the amount we last
1757  *       had to SACK!
1758  */
1759 
1760 u_int
1761 tcp_optlen(tp)
1762 	struct tcpcb *tp;
1763 {
1764 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1765 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
1766 		return TCPOLEN_TSTAMP_APPA;
1767 	else
1768 		return 0;
1769 }
1770