xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 3b435a73967be44dfb4a27315acd72bfacde430c)
1 /*	$NetBSD: tcp_subr.c,v 1.81 1999/09/23 04:02:27 enami 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 #endif
143 
144 #include <netinet/tcp.h>
145 #include <netinet/tcp_fsm.h>
146 #include <netinet/tcp_seq.h>
147 #include <netinet/tcp_timer.h>
148 #include <netinet/tcp_var.h>
149 #include <netinet/tcpip.h>
150 
151 #ifdef IPSEC
152 #include <netinet6/ipsec.h>
153 #endif /*IPSEC*/
154 
155 #ifdef INET6
156 struct in6pcb tcb6;
157 #endif
158 
159 /* patchable/settable parameters for tcp */
160 int 	tcp_mssdflt = TCP_MSS;
161 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
162 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
163 int	tcp_do_sack = 1;	/* selective acknowledgement */
164 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
165 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
166 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
167 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
168 int	tcp_init_win = 1;
169 int	tcp_mss_ifmtu = 0;
170 #ifdef TCP_COMPAT_42
171 int	tcp_compat_42 = 1;
172 #else
173 int	tcp_compat_42 = 0;
174 #endif
175 
176 #ifndef TCBHASHSIZE
177 #define	TCBHASHSIZE	128
178 #endif
179 int	tcbhashsize = TCBHASHSIZE;
180 
181 int	tcp_freeq __P((struct tcpcb *));
182 
183 struct pool tcpcb_pool;
184 
185 /*
186  * Tcp initialization
187  */
188 void
189 tcp_init()
190 {
191 	int hlen;
192 
193 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
194 	    0, NULL, NULL, M_PCB);
195 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
196 #ifdef INET6
197 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
198 #endif
199 	LIST_INIT(&tcp_delacks);
200 
201 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
202 #ifdef INET6
203 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
204 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
205 #endif
206 	if (max_protohdr < hlen)
207 		max_protohdr = hlen;
208 	if (max_linkhdr + hlen > MHLEN)
209 		panic("tcp_init");
210 
211 	/* Initialize the compressed state engine. */
212 	syn_cache_init();
213 }
214 
215 /*
216  * Create template to be used to send tcp packets on a connection.
217  * Call after host entry created, allocates an mbuf and fills
218  * in a skeletal tcp/ip header, minimizing the amount of work
219  * necessary when the connection is used.
220  */
221 struct mbuf *
222 tcp_template(tp)
223 	struct tcpcb *tp;
224 {
225 	register struct inpcb *inp = tp->t_inpcb;
226 #ifdef INET6
227 	register struct in6pcb *in6p = tp->t_in6pcb;
228 #endif
229 	register struct tcphdr *n;
230 	register struct mbuf *m;
231 	int hlen;
232 
233 	switch (tp->t_family) {
234 	case AF_INET:
235 		hlen = sizeof(struct ip);
236 		if (inp)
237 			break;
238 #ifdef INET6
239 		if (in6p) {
240 			/* mapped addr case */
241 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
242 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
243 				break;
244 		}
245 #endif
246 		return NULL;	/*EINVAL*/
247 #ifdef INET6
248 	case AF_INET6:
249 		hlen = sizeof(struct ip6_hdr);
250 		if (in6p) {
251 			/* more sainty check? */
252 			break;
253 		}
254 		return NULL;	/*EINVAL*/
255 #endif
256 	default:
257 		hlen = 0;	/*pacify gcc*/
258 		return NULL;	/*EAFNOSUPPORT*/
259 	}
260 	if ((m = tp->t_template) == 0) {
261 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
262 		if (m) {
263 			MCLGET(m, M_DONTWAIT);
264 			if ((m->m_flags & M_EXT) == 0) {
265 				m_free(m);
266 				m = NULL;
267 			}
268 		}
269 		if (m == NULL)
270 			return NULL;
271 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
272 	}
273 	bzero(mtod(m, caddr_t), m->m_len);
274 	switch (tp->t_family) {
275 	case AF_INET:
276 	    {
277 		struct ipovly *ipov;
278 		mtod(m, struct ip *)->ip_v = 4;
279 		ipov = mtod(m, struct ipovly *);
280 		ipov->ih_pr = IPPROTO_TCP;
281 		ipov->ih_len = htons(sizeof(struct tcphdr));
282 		if (inp) {
283 			ipov->ih_src = inp->inp_laddr;
284 			ipov->ih_dst = inp->inp_faddr;
285 		}
286 #ifdef INET6
287 		else if (in6p) {
288 			/* mapped addr case */
289 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
290 				sizeof(ipov->ih_src));
291 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
292 				sizeof(ipov->ih_dst));
293 		}
294 #endif
295 		break;
296 	    }
297 #ifdef INET6
298 	case AF_INET6:
299 	    {
300 		struct ip6_hdr *ip6;
301 		mtod(m, struct ip *)->ip_v = 6;
302 		ip6 = mtod(m, struct ip6_hdr *);
303 		ip6->ip6_nxt = IPPROTO_TCP;
304 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
305 		ip6->ip6_src = in6p->in6p_laddr;
306 		ip6->ip6_dst = in6p->in6p_faddr;
307 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
308 		if (ip6_auto_flowlabel) {
309 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
310 			ip6->ip6_flow |=
311 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
312 		}
313 		ip6->ip6_vfc = IPV6_VERSION;
314 		break;
315 	    }
316 #endif
317 	}
318 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
319 	if (inp) {
320 		n->th_sport = inp->inp_lport;
321 		n->th_dport = inp->inp_fport;
322 	}
323 #ifdef INET6
324 	else if (in6p) {
325 		n->th_sport = in6p->in6p_lport;
326 		n->th_dport = in6p->in6p_fport;
327 	}
328 #endif
329 	n->th_seq = 0;
330 	n->th_ack = 0;
331 	n->th_x2 = 0;
332 	n->th_off = 5;
333 	n->th_flags = 0;
334 	n->th_win = 0;
335 	n->th_sum = 0;
336 	n->th_urp = 0;
337 	return (m);
338 }
339 
340 /*
341  * Send a single message to the TCP at address specified by
342  * the given TCP/IP header.  If m == 0, then we make a copy
343  * of the tcpiphdr at ti and send directly to the addressed host.
344  * This is used to force keep alive messages out using the TCP
345  * template for a connection tp->t_template.  If flags are given
346  * then we send a message back to the TCP which originated the
347  * segment ti, and discard the mbuf containing it and any other
348  * attached mbufs.
349  *
350  * In any case the ack and sequence number of the transmitted
351  * segment are as specified by the parameters.
352  */
353 int
354 tcp_respond(tp, template, m, th0, ack, seq, flags)
355 	struct tcpcb *tp;
356 	struct mbuf *template;
357 	register struct mbuf *m;
358 	struct tcphdr *th0;
359 	tcp_seq ack, seq;
360 	int flags;
361 {
362 #ifndef INET6
363 	struct route iproute;
364 #else
365 	struct route_in6 iproute;	/* sizeof(route_in6) > sizeof(route) */
366 #endif
367 	struct route *ro;
368 	struct rtentry *rt;
369 	int error, tlen, win = 0;
370 	int hlen;
371 	struct ip *ip;
372 #ifdef INET6
373 	struct ip6_hdr *ip6;
374 #endif
375 	int family;	/* family on packet, not inpcb/in6pcb! */
376 	struct tcphdr *th;
377 
378 	if (tp != NULL && (flags & TH_RST) == 0) {
379 		if (tp->t_inpcb)
380 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
381 #ifdef INET6
382 		else if (tp->t_in6pcb)
383 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
384 #endif
385 	}
386 
387 	ip = NULL;
388 #ifdef INET6
389 	ip6 = NULL;
390 #endif
391 	if (m == 0) {
392 		if (!template)
393 			return EINVAL;
394 
395 		/* get family information from template */
396 		switch (mtod(template, struct ip *)->ip_v) {
397 		case 4:
398 			family = AF_INET;
399 			hlen = sizeof(struct ip);
400 			break;
401 #ifdef INET6
402 		case 6:
403 			family = AF_INET6;
404 			hlen = sizeof(struct ip6_hdr);
405 			break;
406 #endif
407 		default:
408 			return EAFNOSUPPORT;
409 		}
410 
411 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
412 		if (m) {
413 			MCLGET(m, M_DONTWAIT);
414 			if ((m->m_flags & M_EXT) == 0) {
415 				m_free(m);
416 				m = NULL;
417 			}
418 		}
419 		if (m == NULL)
420 			return (ENOBUFS);
421 
422 		if (tcp_compat_42)
423 			tlen = 1;
424 		else
425 			tlen = 0;
426 
427 		m->m_data += max_linkhdr;
428 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
429 			template->m_len);
430 		switch (family) {
431 		case AF_INET:
432 			ip = mtod(m, struct ip *);
433 			th = (struct tcphdr *)(ip + 1);
434 			break;
435 #ifdef INET6
436 		case AF_INET6:
437 			ip6 = mtod(m, struct ip6_hdr *);
438 			th = (struct tcphdr *)(ip6 + 1);
439 			break;
440 #endif
441 		default:	/*pacify gcc*/
442 			ip = NULL;
443 #ifdef INET6
444 			ip6 = NULL;
445 #endif
446 			th = NULL;
447 			break;
448 		}
449 		flags = TH_ACK;
450 	} else {
451 		/* get family information from m */
452 		switch (mtod(m, struct ip *)->ip_v) {
453 		case 4:
454 			family = AF_INET;
455 			hlen = sizeof(struct ip);
456 			break;
457 #ifdef INET6
458 		case 6:
459 			family = AF_INET6;
460 			hlen = sizeof(struct ip6_hdr);
461 			break;
462 #endif
463 		default:
464 			if (m)
465 				m_freem(m);
466 			return EAFNOSUPPORT;
467 		}
468 
469 		/* template pointer almost has no meaning */
470 		m_freem(m->m_next);
471 		m->m_next = 0;
472 		m->m_len = hlen + sizeof(struct tcphdr);
473 		if ((m->m_flags & M_PKTHDR) == 0) {
474 			printf("non PKTHDR to tcp_respond\n");
475 			m_freem(m);
476 			return EINVAL;
477 		}
478 
479 		tlen = 0;
480 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
481 		switch (family) {
482 		case AF_INET:
483 			ip = mtod(m, struct ip *);
484 			th = (struct tcphdr *)(ip + 1);
485 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
486 			ip->ip_p = IPPROTO_TCP;
487 			break;
488 #ifdef INET6
489 		case AF_INET6:
490 			ip6 = mtod(m, struct ip6_hdr *);
491 			th = (struct tcphdr *)(ip6 + 1);
492 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
493 			ip6->ip6_nxt = IPPROTO_TCP;
494 			break;
495 #endif
496 		}
497 		*th = *th0;
498 		xchg(th->th_dport, th->th_sport, u_int16_t);
499 #undef xchg
500 	}
501 	th->th_seq = htonl(seq);
502 	th->th_ack = htonl(ack);
503 	th->th_x2 = 0;
504 	if ((flags & TH_SYN) == 0) {
505 		if (tp)
506 			th->th_win = htons((u_int16_t) (win >> tp->rcv_scale));
507 		else
508 			th->th_win = htons((u_int16_t)win);
509 		th->th_off = sizeof (struct tcphdr) >> 2;
510 		tlen += sizeof (struct tcphdr);
511 	} else
512 		tlen += th->th_off << 2;
513 	m->m_len = hlen + tlen;
514 	m->m_pkthdr.len = hlen + tlen;
515 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
516 	th->th_flags = flags;
517 	th->th_urp = 0;
518 
519 	switch (family) {
520 	case AF_INET:
521 	    {
522 		struct ipovly *ipov = (struct ipovly *)ip;
523 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
524 		ipov->ih_len = htons((u_int16_t)tlen);
525 
526 		th->th_sum = 0;
527 		th->th_sum = in_cksum(m, hlen + tlen);
528 		ip->ip_len = hlen + tlen;	/*will be flipped on output*/
529 		ip->ip_ttl = ip_defttl;
530 		break;
531 	    }
532 #ifdef INET6
533 	case AF_INET6:
534 	    {
535 		th->th_sum = 0;
536 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
537 				tlen);
538 		ip6->ip6_plen = ntohs(tlen);
539 		ip6->ip6_hlim = ip6_defhlim;
540 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
541 		if (ip6_auto_flowlabel) {
542 			ip6->ip6_flow |=
543 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
544 		}
545 		break;
546 	    }
547 #endif
548 	}
549 
550 #ifdef IPSEC
551 	m->m_pkthdr.rcvif = NULL;
552 #endif /*IPSEC*/
553 
554 	/*
555 	 * If we're doing Path MTU discovery, we need to set DF unless
556 	 * the route's MTU is locked.  If we lack a route, we need to
557 	 * look it up now.
558 	 *
559 	 * ip_output() could do this for us, but it's convenient to just
560 	 * do it here unconditionally.
561 	 */
562 	if (tp != NULL && tp->t_inpcb != NULL) {
563 		ro = &tp->t_inpcb->inp_route;
564 #ifdef IPSEC
565 		m->m_pkthdr.rcvif = (struct ifnet *)tp->t_inpcb->inp_socket;
566 #endif
567 #ifdef DIAGNOSTIC
568 		if (family != AF_INET)
569 			panic("tcp_respond: address family mismatch");
570 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
571 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
572 			    ntohl(ip->ip_dst.s_addr),
573 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
574 		}
575 #endif
576 	}
577 #ifdef INET6
578 	else if (tp != NULL && tp->t_in6pcb != NULL) {
579 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
580 #ifdef IPSEC
581 		m->m_pkthdr.rcvif = (struct ifnet *)tp->t_in6pcb->in6p_socket;
582 #endif
583 #ifdef DIAGNOSTIC
584 		if (family == AF_INET) {
585 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
586 				panic("tcp_respond: not mapped addr");
587 			if (bcmp(&ip->ip_dst,
588 					&tp->t_in6pcb->in6p_faddr.s6_addr32[3],
589 					sizeof(ip->ip_dst)) != 0) {
590 				panic("tcp_respond: ip_dst != in6p_faddr");
591 			}
592 		} else if (family == AF_INET6) {
593 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr))
594 				panic("tcp_respond: ip6_dst != in6p_faddr");
595 		} else
596 			panic("tcp_respond: address family mismatch");
597 #endif
598 	}
599 #endif
600 	else {
601 		ro = (struct route *)&iproute;
602 		bzero(ro, sizeof(iproute));
603 	}
604 	if ((rt = ro->ro_rt) == NULL || (rt->rt_flags & RTF_UP) == 0) {
605 		if (ro->ro_rt != NULL) {
606 			RTFREE(ro->ro_rt);
607 			ro->ro_rt = NULL;
608 		}
609 		switch (family) {
610 		case AF_INET:
611 		    {
612 			struct sockaddr_in *dst;
613 			dst = satosin(&ro->ro_dst);
614 			dst->sin_family = AF_INET;
615 			dst->sin_len = sizeof(*dst);
616 			dst->sin_addr = ip->ip_dst;
617 			break;
618 		    }
619 #ifdef INET6
620 		case AF_INET6:
621 		    {
622 			struct sockaddr_in6 *dst;
623 			dst = satosin6(&ro->ro_dst);
624 			bzero(dst, sizeof(*dst));
625 			dst->sin6_family = AF_INET6;
626 			dst->sin6_len = sizeof(*dst);
627 			dst->sin6_addr = ip6->ip6_dst;
628 			break;
629 		    }
630 #endif
631 		}
632 		rtalloc(ro);
633 		if ((rt = ro->ro_rt) == NULL) {
634 			m_freem(m);
635 			switch (family) {
636 			case AF_INET:
637 				ipstat.ips_noroute++;
638 				break;
639 #ifdef INET6
640 			case AF_INET6:
641 				ip6stat.ip6s_noroute++;
642 				break;
643 #endif
644 			}
645 			return (EHOSTUNREACH);
646 		}
647 	}
648 	switch (family) {
649 	case AF_INET:
650 		if (ip_mtudisc != 0 && (rt->rt_rmx.rmx_locks & RTV_MTU) == 0)
651 			ip->ip_off |= IP_DF;
652 
653 		error = ip_output(m, NULL, ro, 0, NULL);
654 		break;
655 #ifdef INET6
656 	case AF_INET6:
657 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL);
658 		break;
659 #endif
660 	default:
661 		error = EAFNOSUPPORT;
662 		break;
663 	}
664 
665 	if (ro == (struct route *)&iproute) {
666 		RTFREE(ro->ro_rt);
667 		ro->ro_rt = NULL;
668 	}
669 
670 	return (error);
671 }
672 
673 /*
674  * Create a new TCP control block, making an
675  * empty reassembly queue and hooking it to the argument
676  * protocol control block.
677  */
678 struct tcpcb *
679 tcp_newtcpcb(family, aux)
680 	int family;	/* selects inpcb, or in6pcb */
681 	void *aux;
682 {
683 	register struct tcpcb *tp;
684 
685 	switch (family) {
686 	case PF_INET:
687 		break;
688 #ifdef INET6
689 	case PF_INET6:
690 		break;
691 #endif
692 	default:
693 		return NULL;
694 	}
695 
696 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
697 	if (tp == NULL)
698 		return (NULL);
699 	bzero((caddr_t)tp, sizeof(struct tcpcb));
700 	LIST_INIT(&tp->segq);
701 	LIST_INIT(&tp->timeq);
702 	tp->t_family = family;		/* may be overridden later on */
703 	tp->t_peermss = tcp_mssdflt;
704 	tp->t_ourmss = tcp_mssdflt;
705 	tp->t_segsz = tcp_mssdflt;
706 	LIST_INIT(&tp->t_sc);
707 
708 	tp->t_flags = 0;
709 	if (tcp_do_rfc1323 && tcp_do_win_scale)
710 		tp->t_flags |= TF_REQ_SCALE;
711 	if (tcp_do_rfc1323 && tcp_do_timestamps)
712 		tp->t_flags |= TF_REQ_TSTMP;
713 	if (tcp_do_sack == 2)
714 		tp->t_flags |= TF_WILL_SACK;
715 	else if (tcp_do_sack == 1)
716 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
717 	tp->t_flags |= TF_CANT_TXSACK;
718 	switch (family) {
719 	case PF_INET:
720 		tp->t_inpcb = (struct inpcb *)aux;
721 		break;
722 #ifdef INET6
723 	case PF_INET6:
724 		tp->t_in6pcb = (struct in6pcb *)aux;
725 		break;
726 #endif
727 	}
728 	/*
729 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
730 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
731 	 * reasonable initial retransmit time.
732 	 */
733 	tp->t_srtt = TCPTV_SRTTBASE;
734 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
735 	tp->t_rttmin = TCPTV_MIN;
736 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
737 	    TCPTV_MIN, TCPTV_REXMTMAX);
738 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
739 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
740 	if (family == AF_INET) {
741 		struct inpcb *inp = (struct inpcb *)aux;
742 		inp->inp_ip.ip_ttl = ip_defttl;
743 		inp->inp_ppcb = (caddr_t)tp;
744 	}
745 #ifdef INET6
746 	else if (family == AF_INET6) {
747 		struct in6pcb *in6p = (struct in6pcb *)aux;
748 		in6p->in6p_ip6.ip6_hlim = ip6_defhlim;
749 		in6p->in6p_ppcb = (caddr_t)tp;
750 	}
751 #endif
752 	return (tp);
753 }
754 
755 /*
756  * Drop a TCP connection, reporting
757  * the specified error.  If connection is synchronized,
758  * then send a RST to peer.
759  */
760 struct tcpcb *
761 tcp_drop(tp, errno)
762 	register struct tcpcb *tp;
763 	int errno;
764 {
765 	struct socket *so;
766 
767 	if (tp->t_inpcb)
768 		so = tp->t_inpcb->inp_socket;
769 #ifdef INET6
770 	else if (tp->t_in6pcb)
771 		so = tp->t_in6pcb->in6p_socket;
772 #endif
773 	else
774 		return NULL;
775 
776 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
777 		tp->t_state = TCPS_CLOSED;
778 		(void) tcp_output(tp);
779 		tcpstat.tcps_drops++;
780 	} else
781 		tcpstat.tcps_conndrops++;
782 	if (errno == ETIMEDOUT && tp->t_softerror)
783 		errno = tp->t_softerror;
784 	so->so_error = errno;
785 	return (tcp_close(tp));
786 }
787 
788 /*
789  * Close a TCP control block:
790  *	discard all space held by the tcp
791  *	discard internet protocol block
792  *	wake up any sleepers
793  */
794 struct tcpcb *
795 tcp_close(tp)
796 	register struct tcpcb *tp;
797 {
798 	struct inpcb *inp;
799 #ifdef INET6
800 	struct in6pcb *in6p;
801 #endif
802 	struct socket *so;
803 #ifdef RTV_RTT
804 	register struct rtentry *rt;
805 #endif
806 	struct route *ro;
807 
808 	inp = tp->t_inpcb;
809 #ifdef INET6
810 	in6p = tp->t_in6pcb;
811 #endif
812 	so = NULL;
813 	ro = NULL;
814 	if (inp) {
815 		so = inp->inp_socket;
816 		ro = &inp->inp_route;
817 	}
818 #ifdef INET6
819 	else if (in6p) {
820 		so = in6p->in6p_socket;
821 		ro = (struct route *)&in6p->in6p_route;
822 	}
823 #endif
824 
825 #ifdef RTV_RTT
826 	/*
827 	 * If we sent enough data to get some meaningful characteristics,
828 	 * save them in the routing entry.  'Enough' is arbitrarily
829 	 * defined as the sendpipesize (default 4K) * 16.  This would
830 	 * give us 16 rtt samples assuming we only get one sample per
831 	 * window (the usual case on a long haul net).  16 samples is
832 	 * enough for the srtt filter to converge to within 5% of the correct
833 	 * value; fewer samples and we could save a very bogus rtt.
834 	 *
835 	 * Don't update the default route's characteristics and don't
836 	 * update anything that the user "locked".
837 	 */
838 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
839 	    ro && (rt = ro->ro_rt) &&
840 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
841 		register u_long i = 0;
842 
843 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
844 			i = tp->t_srtt *
845 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
846 			if (rt->rt_rmx.rmx_rtt && i)
847 				/*
848 				 * filter this update to half the old & half
849 				 * the new values, converting scale.
850 				 * See route.h and tcp_var.h for a
851 				 * description of the scaling constants.
852 				 */
853 				rt->rt_rmx.rmx_rtt =
854 				    (rt->rt_rmx.rmx_rtt + i) / 2;
855 			else
856 				rt->rt_rmx.rmx_rtt = i;
857 		}
858 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
859 			i = tp->t_rttvar *
860 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
861 			if (rt->rt_rmx.rmx_rttvar && i)
862 				rt->rt_rmx.rmx_rttvar =
863 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
864 			else
865 				rt->rt_rmx.rmx_rttvar = i;
866 		}
867 		/*
868 		 * update the pipelimit (ssthresh) if it has been updated
869 		 * already or if a pipesize was specified & the threshhold
870 		 * got below half the pipesize.  I.e., wait for bad news
871 		 * before we start updating, then update on both good
872 		 * and bad news.
873 		 */
874 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
875 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
876 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
877 			/*
878 			 * convert the limit from user data bytes to
879 			 * packets then to packet data bytes.
880 			 */
881 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
882 			if (i < 2)
883 				i = 2;
884 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
885 			if (rt->rt_rmx.rmx_ssthresh)
886 				rt->rt_rmx.rmx_ssthresh =
887 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
888 			else
889 				rt->rt_rmx.rmx_ssthresh = i;
890 		}
891 	}
892 #endif /* RTV_RTT */
893 	/* free the reassembly queue, if any */
894 	TCP_REASS_LOCK(tp);
895 	(void) tcp_freeq(tp);
896 	TCP_REASS_UNLOCK(tp);
897 
898 	TCP_CLEAR_DELACK(tp);
899 	syn_cache_cleanup(tp);
900 
901 	if (tp->t_template) {
902 		m_free(tp->t_template);
903 		tp->t_template = NULL;
904 	}
905 	pool_put(&tcpcb_pool, tp);
906 	if (inp) {
907 		inp->inp_ppcb = 0;
908 		soisdisconnected(so);
909 		in_pcbdetach(inp);
910 	}
911 #ifdef INET6
912 	else if (in6p) {
913 		in6p->in6p_ppcb = 0;
914 		soisdisconnected(so);
915 		in6_pcbdetach(in6p);
916 	}
917 #endif
918 	tcpstat.tcps_closed++;
919 	return ((struct tcpcb *)0);
920 }
921 
922 int
923 tcp_freeq(tp)
924 	struct tcpcb *tp;
925 {
926 	register struct ipqent *qe;
927 	int rv = 0;
928 #ifdef TCPREASS_DEBUG
929 	int i = 0;
930 #endif
931 
932 	TCP_REASS_LOCK_CHECK(tp);
933 
934 	while ((qe = tp->segq.lh_first) != NULL) {
935 #ifdef TCPREASS_DEBUG
936 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
937 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
938 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
939 #endif
940 		LIST_REMOVE(qe, ipqe_q);
941 		LIST_REMOVE(qe, ipqe_timeq);
942 		m_freem(qe->ipqe_m);
943 		pool_put(&ipqent_pool, qe);
944 		rv = 1;
945 	}
946 	return (rv);
947 }
948 
949 /*
950  * Protocol drain routine.  Called when memory is in short supply.
951  */
952 void
953 tcp_drain()
954 {
955 	register struct inpcb *inp;
956 	register struct tcpcb *tp;
957 
958 	/*
959 	 * Free the sequence queue of all TCP connections.
960 	 */
961 	inp = tcbtable.inpt_queue.cqh_first;
962 	if (inp)						/* XXX */
963 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
964 	    inp = inp->inp_queue.cqe_next) {
965 		if ((tp = intotcpcb(inp)) != NULL) {
966 			/*
967 			 * We may be called from a device's interrupt
968 			 * context.  If the tcpcb is already busy,
969 			 * just bail out now.
970 			 */
971 			if (tcp_reass_lock_try(tp) == 0)
972 				continue;
973 			if (tcp_freeq(tp))
974 				tcpstat.tcps_connsdrained++;
975 			TCP_REASS_UNLOCK(tp);
976 		}
977 	}
978 }
979 
980 /*
981  * Notify a tcp user of an asynchronous error;
982  * store error as soft error, but wake up user
983  * (for now, won't do anything until can select for soft error).
984  */
985 void
986 tcp_notify(inp, error)
987 	struct inpcb *inp;
988 	int error;
989 {
990 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
991 	register struct socket *so = inp->inp_socket;
992 
993 	/*
994 	 * Ignore some errors if we are hooked up.
995 	 * If connection hasn't completed, has retransmitted several times,
996 	 * and receives a second error, give up now.  This is better
997 	 * than waiting a long time to establish a connection that
998 	 * can never complete.
999 	 */
1000 	if (tp->t_state == TCPS_ESTABLISHED &&
1001 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1002 	      error == EHOSTDOWN)) {
1003 		return;
1004 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1005 	    tp->t_rxtshift > 3 && tp->t_softerror)
1006 		so->so_error = error;
1007 	else
1008 		tp->t_softerror = error;
1009 	wakeup((caddr_t) &so->so_timeo);
1010 	sorwakeup(so);
1011 	sowwakeup(so);
1012 }
1013 
1014 #if defined(INET6) && !defined(TCP6)
1015 void
1016 tcp6_notify(in6p, error)
1017 	struct in6pcb *in6p;
1018 	int error;
1019 {
1020 	register struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1021 	register struct socket *so = in6p->in6p_socket;
1022 
1023 	/*
1024 	 * Ignore some errors if we are hooked up.
1025 	 * If connection hasn't completed, has retransmitted several times,
1026 	 * and receives a second error, give up now.  This is better
1027 	 * than waiting a long time to establish a connection that
1028 	 * can never complete.
1029 	 */
1030 	if (tp->t_state == TCPS_ESTABLISHED &&
1031 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1032 	      error == EHOSTDOWN)) {
1033 		return;
1034 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1035 	    tp->t_rxtshift > 3 && tp->t_softerror)
1036 		so->so_error = error;
1037 	else
1038 		tp->t_softerror = error;
1039 	wakeup((caddr_t) &so->so_timeo);
1040 	sorwakeup(so);
1041 	sowwakeup(so);
1042 }
1043 #endif
1044 
1045 #if defined(INET6) && !defined(TCP6)
1046 void
1047 tcp6_ctlinput(cmd, sa, ip6, m, off)
1048 	int cmd;
1049 	struct sockaddr *sa;
1050 	register struct ip6_hdr *ip6;
1051 	struct mbuf *m;
1052 	int off;
1053 {
1054 	register struct tcphdr *thp;
1055 	struct tcphdr th;
1056 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1057 	int nmatch;
1058 	extern struct in6_addr zeroin6_addr;	/* netinet6/in6_pcb.c */
1059 	struct sockaddr_in6 sa6;
1060 
1061 	if (sa->sa_family != AF_INET6 ||
1062 	    sa->sa_len != sizeof(struct sockaddr_in6))
1063 		return;
1064 	if (cmd == PRC_QUENCH)
1065 		notify = tcp6_quench;
1066 	else if (cmd == PRC_MSGSIZE)
1067 		notify = tcp6_mtudisc;
1068 	else if (!PRC_IS_REDIRECT(cmd) &&
1069 		 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1070 		return;
1071 
1072 	/* translate addresses into internal form */
1073 	sa6 = *(struct sockaddr_in6 *)sa;
1074 	if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr))
1075 		sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1076 
1077 	if (ip6) {
1078 		/*
1079 		 * XXX: We assume that when ip6 is non NULL,
1080 		 * M and OFF are valid.
1081 		 */
1082 		struct in6_addr s;
1083 
1084 		/* translate addresses into internal form */
1085 		memcpy(&s, &ip6->ip6_src, sizeof(s));
1086 		if (IN6_IS_ADDR_LINKLOCAL(&s))
1087 			s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1088 
1089 		if (m->m_len < off + sizeof(th)) {
1090 			/*
1091 			 * this should be rare case,
1092 			 * so we compromise on this copy...
1093 			 */
1094 			m_copydata(m, off, sizeof(th), (caddr_t)&th);
1095 			thp = &th;
1096 		} else
1097 			thp = (struct tcphdr *)(mtod(m, caddr_t) + off);
1098 		nmatch = in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6,
1099 		    thp->th_dport, &s, thp->th_sport, cmd, notify);
1100 		if (nmatch == 0 && syn_cache_count &&
1101 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1102 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1103 		     inet6ctlerrmap[cmd] == EHOSTDOWN)) {
1104 			struct sockaddr_in6 sin6;
1105 			bzero(&sin6, sizeof(sin6));
1106 			sin6.sin6_len = sizeof(sin6);
1107 			sin6.sin6_family = AF_INET6;
1108 			sin6.sin6_port = thp->th_sport;
1109 			sin6.sin6_addr = s;
1110 			syn_cache_unreach((struct sockaddr *)&sin6, sa, thp);
1111 		}
1112 	} else {
1113 		(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sa6, 0,
1114 				     &zeroin6_addr, 0, cmd, notify);
1115 	}
1116 }
1117 #endif
1118 
1119 /* assumes that ip header and tcp header are contiguous on mbuf */
1120 void *
1121 tcp_ctlinput(cmd, sa, v)
1122 	int cmd;
1123 	struct sockaddr *sa;
1124 	register void *v;
1125 {
1126 	register struct ip *ip = v;
1127 	register struct tcphdr *th;
1128 	extern int inetctlerrmap[];
1129 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1130 	int errno;
1131 	int nmatch;
1132 
1133 	if (sa->sa_family != AF_INET ||
1134 	    sa->sa_len != sizeof(struct sockaddr_in))
1135 		return NULL;
1136 	if ((unsigned)cmd >= PRC_NCMDS)
1137 		return NULL;
1138 	errno = inetctlerrmap[cmd];
1139 	if (cmd == PRC_QUENCH)
1140 		notify = tcp_quench;
1141 	else if (PRC_IS_REDIRECT(cmd))
1142 		notify = in_rtchange, ip = 0;
1143 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
1144 		notify = tcp_mtudisc, ip = 0;
1145 	else if (cmd == PRC_HOSTDEAD)
1146 		ip = 0;
1147 	else if (errno == 0)
1148 		return NULL;
1149 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1150 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1151 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1152 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1153 		if (nmatch == 0 && syn_cache_count &&
1154 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1155 		    inetctlerrmap[cmd] == ENETUNREACH ||
1156 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1157 			struct sockaddr_in sin;
1158 			bzero(&sin, sizeof(sin));
1159 			sin.sin_len = sizeof(sin);
1160 			sin.sin_family = AF_INET;
1161 			sin.sin_port = th->th_sport;
1162 			sin.sin_addr = ip->ip_src;
1163 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1164 		}
1165 
1166 		/* XXX mapped address case */
1167 	}
1168 	else {
1169 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1170 		    notify);
1171 	}
1172 	return NULL;
1173 }
1174 
1175 /*
1176  * When a source quence is received, we are being notifed of congestion.
1177  * Close the congestion window down to the Loss Window (one segment).
1178  * We will gradually open it again as we proceed.
1179  */
1180 void
1181 tcp_quench(inp, errno)
1182 	struct inpcb *inp;
1183 	int errno;
1184 {
1185 	struct tcpcb *tp = intotcpcb(inp);
1186 
1187 	if (tp)
1188 		tp->snd_cwnd = tp->t_segsz;
1189 }
1190 
1191 #if defined(INET6) && !defined(TCP6)
1192 void
1193 tcp6_quench(in6p, errno)
1194 	struct in6pcb *in6p;
1195 	int errno;
1196 {
1197 	struct tcpcb *tp = in6totcpcb(in6p);
1198 
1199 	if (tp)
1200 		tp->snd_cwnd = tp->t_segsz;
1201 }
1202 #endif
1203 
1204 /*
1205  * On receipt of path MTU corrections, flush old route and replace it
1206  * with the new one.  Retransmit all unacknowledged packets, to ensure
1207  * that all packets will be received.
1208  */
1209 void
1210 tcp_mtudisc(inp, errno)
1211 	struct inpcb *inp;
1212 	int errno;
1213 {
1214 	struct tcpcb *tp = intotcpcb(inp);
1215 	struct rtentry *rt = in_pcbrtentry(inp);
1216 
1217 	if (tp != 0) {
1218 		if (rt != 0) {
1219 			/*
1220 			 * If this was not a host route, remove and realloc.
1221 			 */
1222 			if ((rt->rt_flags & RTF_HOST) == 0) {
1223 				in_rtchange(inp, errno);
1224 				if ((rt = in_pcbrtentry(inp)) == 0)
1225 					return;
1226 			}
1227 
1228 			/*
1229 			 * Slow start out of the error condition.  We
1230 			 * use the MTU because we know it's smaller
1231 			 * than the previously transmitted segment.
1232 			 *
1233 			 * Note: This is more conservative than the
1234 			 * suggestion in draft-floyd-incr-init-win-03.
1235 			 */
1236 			if (rt->rt_rmx.rmx_mtu != 0)
1237 				tp->snd_cwnd =
1238 				    TCP_INITIAL_WINDOW(tcp_init_win,
1239 				    rt->rt_rmx.rmx_mtu);
1240 		}
1241 
1242 		/*
1243 		 * Resend unacknowledged packets.
1244 		 */
1245 		tp->snd_nxt = tp->snd_una;
1246 		tcp_output(tp);
1247 	}
1248 }
1249 
1250 #if defined(INET6) && !defined(TCP6)
1251 void
1252 tcp6_mtudisc(in6p, errno)
1253 	struct in6pcb *in6p;
1254 	int errno;
1255 {
1256 	struct tcpcb *tp = in6totcpcb(in6p);
1257 	struct rtentry *rt = in6_pcbrtentry(in6p);
1258 
1259 	if (tp != 0) {
1260 		if (rt != 0) {
1261 			/*
1262 			 * If this was not a host route, remove and realloc.
1263 			 */
1264 			if ((rt->rt_flags & RTF_HOST) == 0) {
1265 				in6_rtchange(in6p, errno);
1266 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1267 					return;
1268 			}
1269 
1270 			/*
1271 			 * Slow start out of the error condition.  We
1272 			 * use the MTU because we know it's smaller
1273 			 * than the previously transmitted segment.
1274 			 *
1275 			 * Note: This is more conservative than the
1276 			 * suggestion in draft-floyd-incr-init-win-03.
1277 			 */
1278 			if (rt->rt_rmx.rmx_mtu != 0)
1279 				tp->snd_cwnd =
1280 				    TCP_INITIAL_WINDOW(tcp_init_win,
1281 				    rt->rt_rmx.rmx_mtu);
1282 		}
1283 
1284 		/*
1285 		 * Resend unacknowledged packets.
1286 		 */
1287 		tp->snd_nxt = tp->snd_una;
1288 		tcp_output(tp);
1289 	}
1290 }
1291 #endif
1292 
1293 /*
1294  * Compute the MSS to advertise to the peer.  Called only during
1295  * the 3-way handshake.  If we are the server (peer initiated
1296  * connection), we are called with a pointer to the interface
1297  * on which the SYN packet arrived.  If we are the client (we
1298  * initiated connection), we are called with a pointer to the
1299  * interface out which this connection should go.
1300  *
1301  * NOTE: Do not subtract IP option/extension header size nor IPsec
1302  * header size from MSS advertisement.  MSS option must hold the maximum
1303  * segment size we can accept, so it must always be:
1304  *	 max(if mtu) - ip header - tcp header
1305  */
1306 u_long
1307 tcp_mss_to_advertise(ifp, af)
1308 	const struct ifnet *ifp;
1309 	int af;
1310 {
1311 	extern u_long in_maxmtu;
1312 	u_long mss = 0;
1313 	u_long hdrsiz;
1314 
1315 	/*
1316 	 * In order to avoid defeating path MTU discovery on the peer,
1317 	 * we advertise the max MTU of all attached networks as our MSS,
1318 	 * per RFC 1191, section 3.1.
1319 	 *
1320 	 * We provide the option to advertise just the MTU of
1321 	 * the interface on which we hope this connection will
1322 	 * be receiving.  If we are responding to a SYN, we
1323 	 * will have a pretty good idea about this, but when
1324 	 * initiating a connection there is a bit more doubt.
1325 	 *
1326 	 * We also need to ensure that loopback has a large enough
1327 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1328 	 */
1329 
1330 	if (ifp != NULL)
1331 		mss = ifp->if_mtu;
1332 
1333 	if (tcp_mss_ifmtu == 0)
1334 		mss = max(in_maxmtu, mss);
1335 
1336 	switch (af) {
1337 	case AF_INET:
1338 		hdrsiz = sizeof(struct ip);
1339 		break;
1340 #ifdef INET6
1341 	case AF_INET6:
1342 		hdrsiz = sizeof(struct ip6_hdr);
1343 		break;
1344 #endif
1345 	default:
1346 		hdrsiz = 0;
1347 		break;
1348 	}
1349 	hdrsiz += sizeof(struct tcphdr);
1350 	if (mss > hdrsiz)
1351 		mss -= hdrsiz;
1352 
1353 	mss = max(tcp_mssdflt, mss);
1354 	return (mss);
1355 }
1356 
1357 /*
1358  * Set connection variables based on the peer's advertised MSS.
1359  * We are passed the TCPCB for the actual connection.  If we
1360  * are the server, we are called by the compressed state engine
1361  * when the 3-way handshake is complete.  If we are the client,
1362  * we are called when we recieve the SYN,ACK from the server.
1363  *
1364  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1365  * before this routine is called!
1366  */
1367 void
1368 tcp_mss_from_peer(tp, offer)
1369 	struct tcpcb *tp;
1370 	int offer;
1371 {
1372 	struct socket *so;
1373 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1374 	struct rtentry *rt;
1375 #endif
1376 	u_long bufsize;
1377 	int mss;
1378 
1379 	so = NULL;
1380 	rt = NULL;
1381 	if (tp->t_inpcb) {
1382 		so = tp->t_inpcb->inp_socket;
1383 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1384 		rt = in_pcbrtentry(tp->t_inpcb);
1385 #endif
1386 	}
1387 #ifdef INET6
1388 	else if (tp->t_in6pcb) {
1389 		so = tp->t_in6pcb->in6p_socket;
1390 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1391 #ifdef TCP6
1392 		rt = NULL;
1393 #else
1394 		rt = in6_pcbrtentry(tp->t_in6pcb);
1395 #endif
1396 #endif
1397 	}
1398 #endif
1399 
1400 	/*
1401 	 * As per RFC1122, use the default MSS value, unless they
1402 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1403 	 */
1404 	mss = tcp_mssdflt;
1405 	if (offer)
1406 		mss = offer;
1407 	mss = max(mss, 32);		/* sanity */
1408 	tp->t_peermss = mss;
1409 	mss -= tcp_optlen(tp);
1410 	if (tp->t_inpcb)
1411 		mss -= ip_optlen(tp->t_inpcb);
1412 #ifdef INET6
1413 	else if (tp->t_in6pcb)
1414 		mss -= ip6_optlen(tp->t_in6pcb);
1415 #endif
1416 
1417 	/*
1418 	 * If there's a pipesize, change the socket buffer to that size.
1419 	 * Make the socket buffer an integral number of MSS units.  If
1420 	 * the MSS is larger than the socket buffer, artificially decrease
1421 	 * the MSS.
1422 	 */
1423 #ifdef RTV_SPIPE
1424 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1425 		bufsize = rt->rt_rmx.rmx_sendpipe;
1426 	else
1427 #endif
1428 		bufsize = so->so_snd.sb_hiwat;
1429 	if (bufsize < mss)
1430 		mss = bufsize;
1431 	else {
1432 		bufsize = roundup(bufsize, mss);
1433 		if (bufsize > sb_max)
1434 			bufsize = sb_max;
1435 		(void) sbreserve(&so->so_snd, bufsize);
1436 	}
1437 	tp->t_segsz = mss;
1438 
1439 #ifdef RTV_SSTHRESH
1440 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1441 		/*
1442 		 * There's some sort of gateway or interface buffer
1443 		 * limit on the path.  Use this to set the slow
1444 		 * start threshold, but set the threshold to no less
1445 		 * than 2 * MSS.
1446 		 */
1447 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1448 	}
1449 #endif
1450 }
1451 
1452 /*
1453  * Processing necessary when a TCP connection is established.
1454  */
1455 void
1456 tcp_established(tp)
1457 	struct tcpcb *tp;
1458 {
1459 	struct socket *so;
1460 #ifdef RTV_RPIPE
1461 	struct rtentry *rt;
1462 #endif
1463 	u_long bufsize;
1464 
1465 	so = NULL;
1466 	rt = NULL;
1467 	if (tp->t_inpcb) {
1468 		so = tp->t_inpcb->inp_socket;
1469 #if defined(RTV_RPIPE)
1470 		rt = in_pcbrtentry(tp->t_inpcb);
1471 #endif
1472 	}
1473 #ifdef INET6
1474 	else if (tp->t_in6pcb) {
1475 		so = tp->t_in6pcb->in6p_socket;
1476 #if defined(RTV_RPIPE)
1477 #ifdef TCP6
1478 		rt = NULL;
1479 #else
1480 		rt = in6_pcbrtentry(tp->t_in6pcb);
1481 #endif
1482 #endif
1483 	}
1484 #endif
1485 
1486 	tp->t_state = TCPS_ESTABLISHED;
1487 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1488 
1489 #ifdef RTV_RPIPE
1490 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1491 		bufsize = rt->rt_rmx.rmx_recvpipe;
1492 	else
1493 #endif
1494 		bufsize = so->so_rcv.sb_hiwat;
1495 	if (bufsize > tp->t_ourmss) {
1496 		bufsize = roundup(bufsize, tp->t_ourmss);
1497 		if (bufsize > sb_max)
1498 			bufsize = sb_max;
1499 		(void) sbreserve(&so->so_rcv, bufsize);
1500 	}
1501 }
1502 
1503 /*
1504  * Check if there's an initial rtt or rttvar.  Convert from the
1505  * route-table units to scaled multiples of the slow timeout timer.
1506  * Called only during the 3-way handshake.
1507  */
1508 void
1509 tcp_rmx_rtt(tp)
1510 	struct tcpcb *tp;
1511 {
1512 #ifdef RTV_RTT
1513 	struct rtentry *rt = NULL;
1514 	int rtt;
1515 
1516 	if (tp->t_inpcb)
1517 		rt = in_pcbrtentry(tp->t_inpcb);
1518 #ifdef INET6
1519 	else if (tp->t_in6pcb) {
1520 #ifdef TCP6
1521 		rt = NULL;
1522 #else
1523 		rt = in6_pcbrtentry(tp->t_in6pcb);
1524 #endif
1525 	}
1526 #endif
1527 	if (rt == NULL)
1528 		return;
1529 
1530 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1531 		/*
1532 		 * XXX The lock bit for MTU indicates that the value
1533 		 * is also a minimum value; this is subject to time.
1534 		 */
1535 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1536 			TCPT_RANGESET(tp->t_rttmin,
1537 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1538 			    TCPTV_MIN, TCPTV_REXMTMAX);
1539 		tp->t_srtt = rtt /
1540 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1541 		if (rt->rt_rmx.rmx_rttvar) {
1542 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1543 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1544 				(TCP_RTTVAR_SHIFT + 2));
1545 		} else {
1546 			/* Default variation is +- 1 rtt */
1547 			tp->t_rttvar =
1548 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1549 		}
1550 		TCPT_RANGESET(tp->t_rxtcur,
1551 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1552 		    tp->t_rttmin, TCPTV_REXMTMAX);
1553 	}
1554 #endif
1555 }
1556 
1557 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1558 
1559 /*
1560  * Get a new sequence value given a tcp control block
1561  */
1562 tcp_seq
1563 tcp_new_iss(tp, len, addin)
1564 	void            *tp;
1565 	u_long           len;
1566 	tcp_seq		 addin;
1567 {
1568 	tcp_seq          tcp_iss;
1569 
1570 	/*
1571 	 * Randomize.
1572 	 */
1573 #if NRND > 0
1574 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
1575 #else
1576 	tcp_iss = random();
1577 #endif
1578 
1579 	/*
1580 	 * If we were asked to add some amount to a known value,
1581 	 * we will take a random value obtained above, mask off the upper
1582 	 * bits, and add in the known value.  We also add in a constant to
1583 	 * ensure that we are at least a certain distance from the original
1584 	 * value.
1585 	 *
1586 	 * This is used when an old connection is in timed wait
1587 	 * and we have a new one coming in, for instance.
1588 	 */
1589 	if (addin != 0) {
1590 #ifdef TCPISS_DEBUG
1591 		printf("Random %08x, ", tcp_iss);
1592 #endif
1593 		tcp_iss &= TCP_ISS_RANDOM_MASK;
1594 		tcp_iss += addin + TCP_ISSINCR;
1595 #ifdef TCPISS_DEBUG
1596 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
1597 #endif
1598 	} else {
1599 		tcp_iss &= TCP_ISS_RANDOM_MASK;
1600 		tcp_iss += tcp_iss_seq;
1601 		tcp_iss_seq += TCP_ISSINCR;
1602 #ifdef TCPISS_DEBUG
1603 		printf("ISS %08x\n", tcp_iss);
1604 #endif
1605 	}
1606 
1607 	if (tcp_compat_42) {
1608 		/*
1609 		 * Limit it to the positive range for really old TCP
1610 		 * implementations.
1611 		 */
1612 		if (tcp_iss >= 0x80000000)
1613 			tcp_iss &= 0x7fffffff;		/* XXX */
1614 	}
1615 
1616 	return tcp_iss;
1617 }
1618 
1619 #ifdef IPSEC
1620 /* compute ESP/AH header size for TCP, including outer IP header. */
1621 size_t
1622 ipsec4_hdrsiz_tcp(tp)
1623 	struct tcpcb *tp;
1624 {
1625 	struct inpcb *inp;
1626 	size_t hdrsiz;
1627 
1628 	/* XXX mapped addr case (tp->t_in6pcb) */
1629 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1630 		return 0;
1631 	switch (tp->t_family) {
1632 	case AF_INET:
1633 		hdrsiz = ipsec4_hdrsiz(tp->t_template, inp);
1634 		break;
1635 	default:
1636 		hdrsiz = 0;
1637 		break;
1638 	}
1639 
1640 	return hdrsiz;
1641 }
1642 
1643 #if defined(INET6) && !defined(TCP6)
1644 size_t
1645 ipsec6_hdrsiz_tcp(tp)
1646 	struct tcpcb *tp;
1647 {
1648 	struct in6pcb *in6p;
1649 	size_t hdrsiz;
1650 
1651 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
1652 		return 0;
1653 	switch (tp->t_family) {
1654 	case AF_INET6:
1655 		hdrsiz = ipsec6_hdrsiz(tp->t_template, in6p);
1656 		break;
1657 	case AF_INET:
1658 		/* mapped address case - tricky */
1659 	default:
1660 		hdrsiz = 0;
1661 		break;
1662 	}
1663 
1664 	return hdrsiz;
1665 }
1666 #endif
1667 #endif /*IPSEC*/
1668 
1669 /*
1670  * Determine the length of the TCP options for this connection.
1671  *
1672  * XXX:  What do we do for SACK, when we add that?  Just reserve
1673  *       all of the space?  Otherwise we can't exactly be incrementing
1674  *       cwnd by an amount that varies depending on the amount we last
1675  *       had to SACK!
1676  */
1677 
1678 u_int
1679 tcp_optlen(tp)
1680 	struct tcpcb *tp;
1681 {
1682 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1683 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
1684 		return TCPOLEN_TSTAMP_APPA;
1685 	else
1686 		return 0;
1687 }
1688