xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 27578b9aac214cc7796ead81dcc5427e79d5f2a0)
1 /*	$NetBSD: tcp_subr.c,v 1.119 2001/09/11 21:03:21 thorpej 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, 2000, 2001 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 "opt_inet_csum.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #endif
147 
148 #include <netinet/tcp.h>
149 #include <netinet/tcp_fsm.h>
150 #include <netinet/tcp_seq.h>
151 #include <netinet/tcp_timer.h>
152 #include <netinet/tcp_var.h>
153 #include <netinet/tcpip.h>
154 
155 #ifdef IPSEC
156 #include <netinet6/ipsec.h>
157 #endif /*IPSEC*/
158 
159 #ifdef INET6
160 struct in6pcb tcb6;
161 #endif
162 
163 /* patchable/settable parameters for tcp */
164 int 	tcp_mssdflt = TCP_MSS;
165 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
166 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
167 #if NRND > 0
168 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
169 #endif
170 int	tcp_do_sack = 1;	/* selective acknowledgement */
171 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
172 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
173 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
174 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
175 int	tcp_init_win = 1;
176 int	tcp_mss_ifmtu = 0;
177 #ifdef TCP_COMPAT_42
178 int	tcp_compat_42 = 1;
179 #else
180 int	tcp_compat_42 = 0;
181 #endif
182 int	tcp_rst_ppslim = 100;	/* 100pps */
183 
184 /* tcb hash */
185 #ifndef TCBHASHSIZE
186 #define	TCBHASHSIZE	128
187 #endif
188 int	tcbhashsize = TCBHASHSIZE;
189 
190 /* syn hash parameters */
191 #define	TCP_SYN_HASH_SIZE	293
192 #define	TCP_SYN_BUCKET_SIZE	35
193 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
194 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
195 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
196 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
197 
198 int	tcp_freeq __P((struct tcpcb *));
199 
200 #ifdef INET
201 void	tcp_mtudisc_callback __P((struct in_addr));
202 #endif
203 #ifdef INET6
204 void	tcp6_mtudisc_callback __P((struct in6_addr *));
205 #endif
206 
207 void	tcp_mtudisc __P((struct inpcb *, int));
208 #ifdef INET6
209 void	tcp6_mtudisc __P((struct in6pcb *, int));
210 #endif
211 
212 struct pool tcpcb_pool;
213 
214 #ifdef TCP_CSUM_COUNTERS
215 #include <sys/device.h>
216 
217 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
218     NULL, "tcp", "hwcsum bad");
219 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
220     NULL, "tcp", "hwcsum ok");
221 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
222     NULL, "tcp", "hwcsum data");
223 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
224     NULL, "tcp", "swcsum");
225 #endif /* TCP_CSUM_COUNTERS */
226 
227 /*
228  * Tcp initialization
229  */
230 void
231 tcp_init()
232 {
233 	int hlen;
234 
235 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
236 	    0, NULL, NULL, M_PCB);
237 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
238 #ifdef INET6
239 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
240 #endif
241 
242 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
243 #ifdef INET6
244 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
245 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
246 #endif
247 	if (max_protohdr < hlen)
248 		max_protohdr = hlen;
249 	if (max_linkhdr + hlen > MHLEN)
250 		panic("tcp_init");
251 
252 #ifdef INET
253 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
254 #endif
255 #ifdef INET6
256 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
257 #endif
258 
259 	/* Initialize timer state. */
260 	tcp_timer_init();
261 
262 	/* Initialize the compressed state engine. */
263 	syn_cache_init();
264 
265 #ifdef TCP_CSUM_COUNTERS
266 	evcnt_attach_static(&tcp_hwcsum_bad);
267 	evcnt_attach_static(&tcp_hwcsum_ok);
268 	evcnt_attach_static(&tcp_hwcsum_data);
269 	evcnt_attach_static(&tcp_swcsum);
270 #endif /* TCP_CSUM_COUNTERS */
271 }
272 
273 /*
274  * Create template to be used to send tcp packets on a connection.
275  * Call after host entry created, allocates an mbuf and fills
276  * in a skeletal tcp/ip header, minimizing the amount of work
277  * necessary when the connection is used.
278  */
279 struct mbuf *
280 tcp_template(tp)
281 	struct tcpcb *tp;
282 {
283 	struct inpcb *inp = tp->t_inpcb;
284 #ifdef INET6
285 	struct in6pcb *in6p = tp->t_in6pcb;
286 #endif
287 	struct tcphdr *n;
288 	struct mbuf *m;
289 	int hlen;
290 
291 	switch (tp->t_family) {
292 	case AF_INET:
293 		hlen = sizeof(struct ip);
294 		if (inp)
295 			break;
296 #ifdef INET6
297 		if (in6p) {
298 			/* mapped addr case */
299 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
300 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
301 				break;
302 		}
303 #endif
304 		return NULL;	/*EINVAL*/
305 #ifdef INET6
306 	case AF_INET6:
307 		hlen = sizeof(struct ip6_hdr);
308 		if (in6p) {
309 			/* more sainty check? */
310 			break;
311 		}
312 		return NULL;	/*EINVAL*/
313 #endif
314 	default:
315 		hlen = 0;	/*pacify gcc*/
316 		return NULL;	/*EAFNOSUPPORT*/
317 	}
318 #ifdef DIAGNOSTIC
319 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
320 		panic("mclbytes too small for t_template");
321 #endif
322 	m = tp->t_template;
323 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
324 		;
325 	else {
326 		if (m)
327 			m_freem(m);
328 		m = tp->t_template = NULL;
329 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
330 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
331 			MCLGET(m, M_DONTWAIT);
332 			if ((m->m_flags & M_EXT) == 0) {
333 				m_free(m);
334 				m = NULL;
335 			}
336 		}
337 		if (m == NULL)
338 			return NULL;
339 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
340 	}
341 
342 	bzero(mtod(m, caddr_t), m->m_len);
343 
344 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
345 
346 	switch (tp->t_family) {
347 	case AF_INET:
348 	    {
349 		struct ipovly *ipov;
350 		mtod(m, struct ip *)->ip_v = 4;
351 		ipov = mtod(m, struct ipovly *);
352 		ipov->ih_pr = IPPROTO_TCP;
353 		ipov->ih_len = htons(sizeof(struct tcphdr));
354 		if (inp) {
355 			ipov->ih_src = inp->inp_laddr;
356 			ipov->ih_dst = inp->inp_faddr;
357 		}
358 #ifdef INET6
359 		else if (in6p) {
360 			/* mapped addr case */
361 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
362 				sizeof(ipov->ih_src));
363 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
364 				sizeof(ipov->ih_dst));
365 		}
366 #endif
367 		/*
368 		 * Compute the pseudo-header portion of the checksum
369 		 * now.  We incrementally add in the TCP option and
370 		 * payload lengths later, and then compute the TCP
371 		 * checksum right before the packet is sent off onto
372 		 * the wire.
373 		 */
374 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
375 		    ipov->ih_dst.s_addr,
376 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
377 		break;
378 	    }
379 #ifdef INET6
380 	case AF_INET6:
381 	    {
382 		struct ip6_hdr *ip6;
383 		mtod(m, struct ip *)->ip_v = 6;
384 		ip6 = mtod(m, struct ip6_hdr *);
385 		ip6->ip6_nxt = IPPROTO_TCP;
386 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
387 		ip6->ip6_src = in6p->in6p_laddr;
388 		ip6->ip6_dst = in6p->in6p_faddr;
389 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
390 		if (ip6_auto_flowlabel) {
391 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
392 			ip6->ip6_flow |=
393 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
394 		}
395 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
396 		ip6->ip6_vfc |= IPV6_VERSION;
397 
398 		/*
399 		 * Compute the pseudo-header portion of the checksum
400 		 * now.  We incrementally add in the TCP option and
401 		 * payload lengths later, and then compute the TCP
402 		 * checksum right before the packet is sent off onto
403 		 * the wire.
404 		 */
405 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
406 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
407 		    htonl(IPPROTO_TCP));
408 		break;
409 	    }
410 #endif
411 	}
412 	if (inp) {
413 		n->th_sport = inp->inp_lport;
414 		n->th_dport = inp->inp_fport;
415 	}
416 #ifdef INET6
417 	else if (in6p) {
418 		n->th_sport = in6p->in6p_lport;
419 		n->th_dport = in6p->in6p_fport;
420 	}
421 #endif
422 	n->th_seq = 0;
423 	n->th_ack = 0;
424 	n->th_x2 = 0;
425 	n->th_off = 5;
426 	n->th_flags = 0;
427 	n->th_win = 0;
428 	n->th_urp = 0;
429 	return (m);
430 }
431 
432 /*
433  * Send a single message to the TCP at address specified by
434  * the given TCP/IP header.  If m == 0, then we make a copy
435  * of the tcpiphdr at ti and send directly to the addressed host.
436  * This is used to force keep alive messages out using the TCP
437  * template for a connection tp->t_template.  If flags are given
438  * then we send a message back to the TCP which originated the
439  * segment ti, and discard the mbuf containing it and any other
440  * attached mbufs.
441  *
442  * In any case the ack and sequence number of the transmitted
443  * segment are as specified by the parameters.
444  */
445 int
446 tcp_respond(tp, template, m, th0, ack, seq, flags)
447 	struct tcpcb *tp;
448 	struct mbuf *template;
449 	struct mbuf *m;
450 	struct tcphdr *th0;
451 	tcp_seq ack, seq;
452 	int flags;
453 {
454 	struct route *ro;
455 	int error, tlen, win = 0;
456 	int hlen;
457 	struct ip *ip;
458 #ifdef INET6
459 	struct ip6_hdr *ip6;
460 #endif
461 	int family;	/* family on packet, not inpcb/in6pcb! */
462 	struct tcphdr *th;
463 
464 	if (tp != NULL && (flags & TH_RST) == 0) {
465 #ifdef DIAGNOSTIC
466 		if (tp->t_inpcb && tp->t_in6pcb)
467 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
468 #endif
469 #ifdef INET
470 		if (tp->t_inpcb)
471 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
472 #endif
473 #ifdef INET6
474 		if (tp->t_in6pcb)
475 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
476 #endif
477 	}
478 
479 	ip = NULL;
480 #ifdef INET6
481 	ip6 = NULL;
482 #endif
483 	if (m == 0) {
484 		if (!template)
485 			return EINVAL;
486 
487 		/* get family information from template */
488 		switch (mtod(template, struct ip *)->ip_v) {
489 		case 4:
490 			family = AF_INET;
491 			hlen = sizeof(struct ip);
492 			break;
493 #ifdef INET6
494 		case 6:
495 			family = AF_INET6;
496 			hlen = sizeof(struct ip6_hdr);
497 			break;
498 #endif
499 		default:
500 			return EAFNOSUPPORT;
501 		}
502 
503 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
504 		if (m) {
505 			MCLGET(m, M_DONTWAIT);
506 			if ((m->m_flags & M_EXT) == 0) {
507 				m_free(m);
508 				m = NULL;
509 			}
510 		}
511 		if (m == NULL)
512 			return (ENOBUFS);
513 
514 		if (tcp_compat_42)
515 			tlen = 1;
516 		else
517 			tlen = 0;
518 
519 		m->m_data += max_linkhdr;
520 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
521 			template->m_len);
522 		switch (family) {
523 		case AF_INET:
524 			ip = mtod(m, struct ip *);
525 			th = (struct tcphdr *)(ip + 1);
526 			break;
527 #ifdef INET6
528 		case AF_INET6:
529 			ip6 = mtod(m, struct ip6_hdr *);
530 			th = (struct tcphdr *)(ip6 + 1);
531 			break;
532 #endif
533 #if 0
534 		default:
535 			/* noone will visit here */
536 			m_freem(m);
537 			return EAFNOSUPPORT;
538 #endif
539 		}
540 		flags = TH_ACK;
541 	} else {
542 
543 		if ((m->m_flags & M_PKTHDR) == 0) {
544 #if 0
545 			printf("non PKTHDR to tcp_respond\n");
546 #endif
547 			m_freem(m);
548 			return EINVAL;
549 		}
550 #ifdef DIAGNOSTIC
551 		if (!th0)
552 			panic("th0 == NULL in tcp_respond");
553 #endif
554 
555 		/* get family information from m */
556 		switch (mtod(m, struct ip *)->ip_v) {
557 		case 4:
558 			family = AF_INET;
559 			hlen = sizeof(struct ip);
560 			ip = mtod(m, struct ip *);
561 			break;
562 #ifdef INET6
563 		case 6:
564 			family = AF_INET6;
565 			hlen = sizeof(struct ip6_hdr);
566 			ip6 = mtod(m, struct ip6_hdr *);
567 			break;
568 #endif
569 		default:
570 			m_freem(m);
571 			return EAFNOSUPPORT;
572 		}
573 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
574 			tlen = sizeof(*th0);
575 		else
576 			tlen = th0->th_off << 2;
577 
578 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
579 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
580 			m->m_len = hlen + tlen;
581 			m_freem(m->m_next);
582 			m->m_next = NULL;
583 		} else {
584 			struct mbuf *n;
585 
586 #ifdef DIAGNOSTIC
587 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
588 				m_freem(m);
589 				return EMSGSIZE;
590 			}
591 #endif
592 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
593 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
594 				MCLGET(n, M_DONTWAIT);
595 				if ((n->m_flags & M_EXT) == 0) {
596 					m_freem(n);
597 					n = NULL;
598 				}
599 			}
600 			if (!n) {
601 				m_freem(m);
602 				return ENOBUFS;
603 			}
604 
605 			n->m_data += max_linkhdr;
606 			n->m_len = hlen + tlen;
607 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
608 			m_copyback(n, hlen, tlen, (caddr_t)th0);
609 
610 			m_freem(m);
611 			m = n;
612 			n = NULL;
613 		}
614 
615 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
616 		switch (family) {
617 		case AF_INET:
618 			ip = mtod(m, struct ip *);
619 			th = (struct tcphdr *)(ip + 1);
620 			ip->ip_p = IPPROTO_TCP;
621 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
622 			ip->ip_p = IPPROTO_TCP;
623 			break;
624 #ifdef INET6
625 		case AF_INET6:
626 			ip6 = mtod(m, struct ip6_hdr *);
627 			th = (struct tcphdr *)(ip6 + 1);
628 			ip6->ip6_nxt = IPPROTO_TCP;
629 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
630 			ip6->ip6_nxt = IPPROTO_TCP;
631 			break;
632 #endif
633 #if 0
634 		default:
635 			/* noone will visit here */
636 			m_freem(m);
637 			return EAFNOSUPPORT;
638 #endif
639 		}
640 		xchg(th->th_dport, th->th_sport, u_int16_t);
641 #undef xchg
642 		tlen = 0;	/*be friendly with the following code*/
643 	}
644 	th->th_seq = htonl(seq);
645 	th->th_ack = htonl(ack);
646 	th->th_x2 = 0;
647 	if ((flags & TH_SYN) == 0) {
648 		if (tp)
649 			win >>= tp->rcv_scale;
650 		if (win > TCP_MAXWIN)
651 			win = TCP_MAXWIN;
652 		th->th_win = htons((u_int16_t)win);
653 		th->th_off = sizeof (struct tcphdr) >> 2;
654 		tlen += sizeof(*th);
655 	} else
656 		tlen += th->th_off << 2;
657 	m->m_len = hlen + tlen;
658 	m->m_pkthdr.len = hlen + tlen;
659 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
660 	th->th_flags = flags;
661 	th->th_urp = 0;
662 
663 	switch (family) {
664 #ifdef INET
665 	case AF_INET:
666 	    {
667 		struct ipovly *ipov = (struct ipovly *)ip;
668 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
669 		ipov->ih_len = htons((u_int16_t)tlen);
670 
671 		th->th_sum = 0;
672 		th->th_sum = in_cksum(m, hlen + tlen);
673 		ip->ip_len = hlen + tlen;	/*will be flipped on output*/
674 		ip->ip_ttl = ip_defttl;
675 		break;
676 	    }
677 #endif
678 #ifdef INET6
679 	case AF_INET6:
680 	    {
681 		th->th_sum = 0;
682 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
683 				tlen);
684 		ip6->ip6_plen = ntohs(tlen);
685 		if (tp && tp->t_in6pcb) {
686 			struct ifnet *oifp;
687 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
688 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
689 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
690 		} else
691 			ip6->ip6_hlim = ip6_defhlim;
692 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
693 		if (ip6_auto_flowlabel) {
694 			ip6->ip6_flow |=
695 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
696 		}
697 		break;
698 	    }
699 #endif
700 	}
701 
702 #ifdef IPSEC
703 	(void)ipsec_setsocket(m, NULL);
704 #endif /*IPSEC*/
705 
706 	if (tp != NULL && tp->t_inpcb != NULL) {
707 		ro = &tp->t_inpcb->inp_route;
708 #ifdef IPSEC
709 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
710 			m_freem(m);
711 			return ENOBUFS;
712 		}
713 #endif
714 #ifdef DIAGNOSTIC
715 		if (family != AF_INET)
716 			panic("tcp_respond: address family mismatch");
717 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
718 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
719 			    ntohl(ip->ip_dst.s_addr),
720 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
721 		}
722 #endif
723 	}
724 #ifdef INET6
725 	else if (tp != NULL && tp->t_in6pcb != NULL) {
726 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
727 #ifdef IPSEC
728 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
729 			m_freem(m);
730 			return ENOBUFS;
731 		}
732 #endif
733 #ifdef DIAGNOSTIC
734 		if (family == AF_INET) {
735 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
736 				panic("tcp_respond: not mapped addr");
737 			if (bcmp(&ip->ip_dst,
738 					&tp->t_in6pcb->in6p_faddr.s6_addr32[3],
739 					sizeof(ip->ip_dst)) != 0) {
740 				panic("tcp_respond: ip_dst != in6p_faddr");
741 			}
742 		} else if (family == AF_INET6) {
743 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr))
744 				panic("tcp_respond: ip6_dst != in6p_faddr");
745 		} else
746 			panic("tcp_respond: address family mismatch");
747 #endif
748 	}
749 #endif
750 	else
751 		ro = NULL;
752 
753 	switch (family) {
754 #ifdef INET
755 	case AF_INET:
756 		error = ip_output(m, NULL, ro,
757 		    (ip_mtudisc ? IP_MTUDISC : 0),
758 		    NULL);
759 		break;
760 #endif
761 #ifdef INET6
762 	case AF_INET6:
763 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
764 			NULL);
765 		break;
766 #endif
767 	default:
768 		error = EAFNOSUPPORT;
769 		break;
770 	}
771 
772 	return (error);
773 }
774 
775 /*
776  * Create a new TCP control block, making an
777  * empty reassembly queue and hooking it to the argument
778  * protocol control block.
779  */
780 struct tcpcb *
781 tcp_newtcpcb(family, aux)
782 	int family;	/* selects inpcb, or in6pcb */
783 	void *aux;
784 {
785 	struct tcpcb *tp;
786 	int i;
787 
788 	switch (family) {
789 	case PF_INET:
790 		break;
791 #ifdef INET6
792 	case PF_INET6:
793 		break;
794 #endif
795 	default:
796 		return NULL;
797 	}
798 
799 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
800 	if (tp == NULL)
801 		return (NULL);
802 	bzero((caddr_t)tp, sizeof(struct tcpcb));
803 	LIST_INIT(&tp->segq);
804 	LIST_INIT(&tp->timeq);
805 	tp->t_family = family;		/* may be overridden later on */
806 	tp->t_peermss = tcp_mssdflt;
807 	tp->t_ourmss = tcp_mssdflt;
808 	tp->t_segsz = tcp_mssdflt;
809 	LIST_INIT(&tp->t_sc);
810 
811 	callout_init(&tp->t_delack_ch);
812 	for (i = 0; i < TCPT_NTIMERS; i++)
813 		TCP_TIMER_INIT(tp, i);
814 
815 	tp->t_flags = 0;
816 	if (tcp_do_rfc1323 && tcp_do_win_scale)
817 		tp->t_flags |= TF_REQ_SCALE;
818 	if (tcp_do_rfc1323 && tcp_do_timestamps)
819 		tp->t_flags |= TF_REQ_TSTMP;
820 	if (tcp_do_sack == 2)
821 		tp->t_flags |= TF_WILL_SACK;
822 	else if (tcp_do_sack == 1)
823 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
824 	tp->t_flags |= TF_CANT_TXSACK;
825 	switch (family) {
826 	case PF_INET:
827 		tp->t_inpcb = (struct inpcb *)aux;
828 		break;
829 #ifdef INET6
830 	case PF_INET6:
831 		tp->t_in6pcb = (struct in6pcb *)aux;
832 		break;
833 #endif
834 	}
835 	/*
836 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
837 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
838 	 * reasonable initial retransmit time.
839 	 */
840 	tp->t_srtt = TCPTV_SRTTBASE;
841 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
842 	tp->t_rttmin = TCPTV_MIN;
843 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
844 	    TCPTV_MIN, TCPTV_REXMTMAX);
845 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
846 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
847 	if (family == AF_INET) {
848 		struct inpcb *inp = (struct inpcb *)aux;
849 		inp->inp_ip.ip_ttl = ip_defttl;
850 		inp->inp_ppcb = (caddr_t)tp;
851 	}
852 #ifdef INET6
853 	else if (family == AF_INET6) {
854 		struct in6pcb *in6p = (struct in6pcb *)aux;
855 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
856 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
857 					       : NULL);
858 		in6p->in6p_ppcb = (caddr_t)tp;
859 	}
860 #endif
861 
862 	/*
863 	 * Initialize our timebase.  When we send timestamps, we take
864 	 * the delta from tcp_now -- this means each connection always
865 	 * gets a timebase of 0, which makes it, among other things,
866 	 * more difficult to determine how long a system has been up,
867 	 * and thus how many TCP sequence increments have occurred.
868 	 */
869 	tp->ts_timebase = tcp_now;
870 
871 	return (tp);
872 }
873 
874 /*
875  * Drop a TCP connection, reporting
876  * the specified error.  If connection is synchronized,
877  * then send a RST to peer.
878  */
879 struct tcpcb *
880 tcp_drop(tp, errno)
881 	struct tcpcb *tp;
882 	int errno;
883 {
884 	struct socket *so = NULL;
885 
886 #ifdef DIAGNOSTIC
887 	if (tp->t_inpcb && tp->t_in6pcb)
888 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
889 #endif
890 #ifdef INET
891 	if (tp->t_inpcb)
892 		so = tp->t_inpcb->inp_socket;
893 #endif
894 #ifdef INET6
895 	if (tp->t_in6pcb)
896 		so = tp->t_in6pcb->in6p_socket;
897 #endif
898 	if (!so)
899 		return NULL;
900 
901 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
902 		tp->t_state = TCPS_CLOSED;
903 		(void) tcp_output(tp);
904 		tcpstat.tcps_drops++;
905 	} else
906 		tcpstat.tcps_conndrops++;
907 	if (errno == ETIMEDOUT && tp->t_softerror)
908 		errno = tp->t_softerror;
909 	so->so_error = errno;
910 	return (tcp_close(tp));
911 }
912 
913 /*
914  * Close a TCP control block:
915  *	discard all space held by the tcp
916  *	discard internet protocol block
917  *	wake up any sleepers
918  */
919 struct tcpcb *
920 tcp_close(tp)
921 	struct tcpcb *tp;
922 {
923 	struct inpcb *inp;
924 #ifdef INET6
925 	struct in6pcb *in6p;
926 #endif
927 	struct socket *so;
928 #ifdef RTV_RTT
929 	struct rtentry *rt;
930 #endif
931 	struct route *ro;
932 
933 	inp = tp->t_inpcb;
934 #ifdef INET6
935 	in6p = tp->t_in6pcb;
936 #endif
937 	so = NULL;
938 	ro = NULL;
939 	if (inp) {
940 		so = inp->inp_socket;
941 		ro = &inp->inp_route;
942 	}
943 #ifdef INET6
944 	else if (in6p) {
945 		so = in6p->in6p_socket;
946 		ro = (struct route *)&in6p->in6p_route;
947 	}
948 #endif
949 
950 #ifdef RTV_RTT
951 	/*
952 	 * If we sent enough data to get some meaningful characteristics,
953 	 * save them in the routing entry.  'Enough' is arbitrarily
954 	 * defined as the sendpipesize (default 4K) * 16.  This would
955 	 * give us 16 rtt samples assuming we only get one sample per
956 	 * window (the usual case on a long haul net).  16 samples is
957 	 * enough for the srtt filter to converge to within 5% of the correct
958 	 * value; fewer samples and we could save a very bogus rtt.
959 	 *
960 	 * Don't update the default route's characteristics and don't
961 	 * update anything that the user "locked".
962 	 */
963 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
964 	    ro && (rt = ro->ro_rt) &&
965 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
966 		u_long i = 0;
967 
968 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
969 			i = tp->t_srtt *
970 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
971 			if (rt->rt_rmx.rmx_rtt && i)
972 				/*
973 				 * filter this update to half the old & half
974 				 * the new values, converting scale.
975 				 * See route.h and tcp_var.h for a
976 				 * description of the scaling constants.
977 				 */
978 				rt->rt_rmx.rmx_rtt =
979 				    (rt->rt_rmx.rmx_rtt + i) / 2;
980 			else
981 				rt->rt_rmx.rmx_rtt = i;
982 		}
983 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
984 			i = tp->t_rttvar *
985 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
986 			if (rt->rt_rmx.rmx_rttvar && i)
987 				rt->rt_rmx.rmx_rttvar =
988 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
989 			else
990 				rt->rt_rmx.rmx_rttvar = i;
991 		}
992 		/*
993 		 * update the pipelimit (ssthresh) if it has been updated
994 		 * already or if a pipesize was specified & the threshhold
995 		 * got below half the pipesize.  I.e., wait for bad news
996 		 * before we start updating, then update on both good
997 		 * and bad news.
998 		 */
999 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1000 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1001 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1002 			/*
1003 			 * convert the limit from user data bytes to
1004 			 * packets then to packet data bytes.
1005 			 */
1006 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1007 			if (i < 2)
1008 				i = 2;
1009 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1010 			if (rt->rt_rmx.rmx_ssthresh)
1011 				rt->rt_rmx.rmx_ssthresh =
1012 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1013 			else
1014 				rt->rt_rmx.rmx_ssthresh = i;
1015 		}
1016 	}
1017 #endif /* RTV_RTT */
1018 	/* free the reassembly queue, if any */
1019 	TCP_REASS_LOCK(tp);
1020 	(void) tcp_freeq(tp);
1021 	TCP_REASS_UNLOCK(tp);
1022 
1023 	tcp_canceltimers(tp);
1024 	TCP_CLEAR_DELACK(tp);
1025 	syn_cache_cleanup(tp);
1026 
1027 	if (tp->t_template) {
1028 		m_free(tp->t_template);
1029 		tp->t_template = NULL;
1030 	}
1031 	pool_put(&tcpcb_pool, tp);
1032 	if (inp) {
1033 		inp->inp_ppcb = 0;
1034 		soisdisconnected(so);
1035 		in_pcbdetach(inp);
1036 	}
1037 #ifdef INET6
1038 	else if (in6p) {
1039 		in6p->in6p_ppcb = 0;
1040 		soisdisconnected(so);
1041 		in6_pcbdetach(in6p);
1042 	}
1043 #endif
1044 	tcpstat.tcps_closed++;
1045 	return ((struct tcpcb *)0);
1046 }
1047 
1048 int
1049 tcp_freeq(tp)
1050 	struct tcpcb *tp;
1051 {
1052 	struct ipqent *qe;
1053 	int rv = 0;
1054 #ifdef TCPREASS_DEBUG
1055 	int i = 0;
1056 #endif
1057 
1058 	TCP_REASS_LOCK_CHECK(tp);
1059 
1060 	while ((qe = tp->segq.lh_first) != NULL) {
1061 #ifdef TCPREASS_DEBUG
1062 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1063 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1064 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1065 #endif
1066 		LIST_REMOVE(qe, ipqe_q);
1067 		LIST_REMOVE(qe, ipqe_timeq);
1068 		m_freem(qe->ipqe_m);
1069 		pool_put(&ipqent_pool, qe);
1070 		rv = 1;
1071 	}
1072 	return (rv);
1073 }
1074 
1075 /*
1076  * Protocol drain routine.  Called when memory is in short supply.
1077  */
1078 void
1079 tcp_drain()
1080 {
1081 	struct inpcb *inp;
1082 	struct tcpcb *tp;
1083 
1084 	/*
1085 	 * Free the sequence queue of all TCP connections.
1086 	 */
1087 	inp = tcbtable.inpt_queue.cqh_first;
1088 	if (inp)						/* XXX */
1089 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
1090 	    inp = inp->inp_queue.cqe_next) {
1091 		if ((tp = intotcpcb(inp)) != NULL) {
1092 			/*
1093 			 * We may be called from a device's interrupt
1094 			 * context.  If the tcpcb is already busy,
1095 			 * just bail out now.
1096 			 */
1097 			if (tcp_reass_lock_try(tp) == 0)
1098 				continue;
1099 			if (tcp_freeq(tp))
1100 				tcpstat.tcps_connsdrained++;
1101 			TCP_REASS_UNLOCK(tp);
1102 		}
1103 	}
1104 }
1105 
1106 /*
1107  * Notify a tcp user of an asynchronous error;
1108  * store error as soft error, but wake up user
1109  * (for now, won't do anything until can select for soft error).
1110  */
1111 void
1112 tcp_notify(inp, error)
1113 	struct inpcb *inp;
1114 	int error;
1115 {
1116 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1117 	struct socket *so = inp->inp_socket;
1118 
1119 	/*
1120 	 * Ignore some errors if we are hooked up.
1121 	 * If connection hasn't completed, has retransmitted several times,
1122 	 * and receives a second error, give up now.  This is better
1123 	 * than waiting a long time to establish a connection that
1124 	 * can never complete.
1125 	 */
1126 	if (tp->t_state == TCPS_ESTABLISHED &&
1127 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1128 	      error == EHOSTDOWN)) {
1129 		return;
1130 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1131 	    tp->t_rxtshift > 3 && tp->t_softerror)
1132 		so->so_error = error;
1133 	else
1134 		tp->t_softerror = error;
1135 	wakeup((caddr_t) &so->so_timeo);
1136 	sorwakeup(so);
1137 	sowwakeup(so);
1138 }
1139 
1140 #ifdef INET6
1141 void
1142 tcp6_notify(in6p, error)
1143 	struct in6pcb *in6p;
1144 	int error;
1145 {
1146 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1147 	struct socket *so = in6p->in6p_socket;
1148 
1149 	/*
1150 	 * Ignore some errors if we are hooked up.
1151 	 * If connection hasn't completed, has retransmitted several times,
1152 	 * and receives a second error, give up now.  This is better
1153 	 * than waiting a long time to establish a connection that
1154 	 * can never complete.
1155 	 */
1156 	if (tp->t_state == TCPS_ESTABLISHED &&
1157 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1158 	      error == EHOSTDOWN)) {
1159 		return;
1160 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1161 	    tp->t_rxtshift > 3 && tp->t_softerror)
1162 		so->so_error = error;
1163 	else
1164 		tp->t_softerror = error;
1165 	wakeup((caddr_t) &so->so_timeo);
1166 	sorwakeup(so);
1167 	sowwakeup(so);
1168 }
1169 #endif
1170 
1171 #ifdef INET6
1172 void
1173 tcp6_ctlinput(cmd, sa, d)
1174 	int cmd;
1175 	struct sockaddr *sa;
1176 	void *d;
1177 {
1178 	struct tcphdr th;
1179 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1180 	int nmatch;
1181 	struct ip6_hdr *ip6;
1182 	const struct sockaddr_in6 *sa6_src = NULL;
1183 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1184 	struct mbuf *m;
1185 	int off;
1186 
1187 	if (sa->sa_family != AF_INET6 ||
1188 	    sa->sa_len != sizeof(struct sockaddr_in6))
1189 		return;
1190 	if ((unsigned)cmd >= PRC_NCMDS)
1191 		return;
1192 	else if (cmd == PRC_QUENCH) {
1193 		/* XXX there's no PRC_QUENCH in IPv6 */
1194 		notify = tcp6_quench;
1195 	} else if (PRC_IS_REDIRECT(cmd))
1196 		notify = in6_rtchange, d = NULL;
1197 	else if (cmd == PRC_MSGSIZE)
1198 		; /* special code is present, see below */
1199 	else if (cmd == PRC_HOSTDEAD)
1200 		d = NULL;
1201 	else if (inet6ctlerrmap[cmd] == 0)
1202 		return;
1203 
1204 	/* if the parameter is from icmp6, decode it. */
1205 	if (d != NULL) {
1206 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1207 		m = ip6cp->ip6c_m;
1208 		ip6 = ip6cp->ip6c_ip6;
1209 		off = ip6cp->ip6c_off;
1210 		sa6_src = ip6cp->ip6c_src;
1211 	} else {
1212 		m = NULL;
1213 		ip6 = NULL;
1214 		sa6_src = &sa6_any;
1215 	}
1216 
1217 	if (ip6) {
1218 		/*
1219 		 * XXX: We assume that when ip6 is non NULL,
1220 		 * M and OFF are valid.
1221 		 */
1222 
1223 		/* check if we can safely examine src and dst ports */
1224 		if (m->m_pkthdr.len < off + sizeof(th)) {
1225 			if (cmd == PRC_MSGSIZE)
1226 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1227 			return;
1228 		}
1229 
1230 		bzero(&th, sizeof(th));
1231 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1232 
1233 		if (cmd == PRC_MSGSIZE) {
1234 			int valid = 0;
1235 
1236 			/*
1237 			 * Check to see if we have a valid TCP connection
1238 			 * corresponding to the address in the ICMPv6 message
1239 			 * payload.
1240 			 */
1241 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1242 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1243 			    th.th_sport, 0))
1244 				valid++;
1245 
1246 			/*
1247 			 * Depending on the value of "valid" and routing table
1248 			 * size (mtudisc_{hi,lo}wat), we will:
1249 			 * - recalcurate the new MTU and create the
1250 			 *   corresponding routing entry, or
1251 			 * - ignore the MTU change notification.
1252 			 */
1253 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1254 
1255 			/*
1256 			 * no need to call in6_pcbnotify, it should have been
1257 			 * called via callback if necessary
1258 			 */
1259 			return;
1260 		}
1261 
1262 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1263 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1264 		if (nmatch == 0 && syn_cache_count &&
1265 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1266 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1267 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1268 			syn_cache_unreach((struct sockaddr *)sa6_src,
1269 					  sa, &th);
1270 	} else {
1271 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1272 		    0, cmd, NULL, notify);
1273 	}
1274 }
1275 #endif
1276 
1277 #ifdef INET
1278 /* assumes that ip header and tcp header are contiguous on mbuf */
1279 void *
1280 tcp_ctlinput(cmd, sa, v)
1281 	int cmd;
1282 	struct sockaddr *sa;
1283 	void *v;
1284 {
1285 	struct ip *ip = v;
1286 	struct tcphdr *th;
1287 	struct icmp *icp;
1288 	extern int inetctlerrmap[];
1289 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1290 	int errno;
1291 	int nmatch;
1292 
1293 	if (sa->sa_family != AF_INET ||
1294 	    sa->sa_len != sizeof(struct sockaddr_in))
1295 		return NULL;
1296 	if ((unsigned)cmd >= PRC_NCMDS)
1297 		return NULL;
1298 	errno = inetctlerrmap[cmd];
1299 	if (cmd == PRC_QUENCH)
1300 		notify = tcp_quench;
1301 	else if (PRC_IS_REDIRECT(cmd))
1302 		notify = in_rtchange, ip = 0;
1303 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) {
1304 		/*
1305 		 * Check to see if we have a valid TCP connection
1306 		 * corresponding to the address in the ICMP message
1307 		 * payload.
1308 		 *
1309 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1310 		 */
1311 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1312 		if (in_pcblookup_connect(&tcbtable,
1313 					 ip->ip_dst, th->th_dport,
1314 					 ip->ip_src, th->th_sport) == NULL)
1315 			return NULL;
1316 
1317 		/*
1318 		 * Now that we've validated that we are actually communicating
1319 		 * with the host indicated in the ICMP message, locate the
1320 		 * ICMP header, recalculate the new MTU, and create the
1321 		 * corresponding routing entry.
1322 		 */
1323 		icp = (struct icmp *)((caddr_t)ip -
1324 		    offsetof(struct icmp, icmp_ip));
1325 		icmp_mtudisc(icp, ip->ip_dst);
1326 
1327 		return NULL;
1328 	} else if (cmd == PRC_HOSTDEAD)
1329 		ip = 0;
1330 	else if (errno == 0)
1331 		return NULL;
1332 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1333 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1334 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1335 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1336 		if (nmatch == 0 && syn_cache_count &&
1337 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1338 		    inetctlerrmap[cmd] == ENETUNREACH ||
1339 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1340 			struct sockaddr_in sin;
1341 			bzero(&sin, sizeof(sin));
1342 			sin.sin_len = sizeof(sin);
1343 			sin.sin_family = AF_INET;
1344 			sin.sin_port = th->th_sport;
1345 			sin.sin_addr = ip->ip_src;
1346 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1347 		}
1348 
1349 		/* XXX mapped address case */
1350 	} else
1351 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1352 		    notify);
1353 	return NULL;
1354 }
1355 
1356 /*
1357  * When a source quence is received, we are being notifed of congestion.
1358  * Close the congestion window down to the Loss Window (one segment).
1359  * We will gradually open it again as we proceed.
1360  */
1361 void
1362 tcp_quench(inp, errno)
1363 	struct inpcb *inp;
1364 	int errno;
1365 {
1366 	struct tcpcb *tp = intotcpcb(inp);
1367 
1368 	if (tp)
1369 		tp->snd_cwnd = tp->t_segsz;
1370 }
1371 #endif
1372 
1373 #ifdef INET6
1374 void
1375 tcp6_quench(in6p, errno)
1376 	struct in6pcb *in6p;
1377 	int errno;
1378 {
1379 	struct tcpcb *tp = in6totcpcb(in6p);
1380 
1381 	if (tp)
1382 		tp->snd_cwnd = tp->t_segsz;
1383 }
1384 #endif
1385 
1386 #ifdef INET
1387 /*
1388  * Path MTU Discovery handlers.
1389  */
1390 void
1391 tcp_mtudisc_callback(faddr)
1392 	struct in_addr faddr;
1393 {
1394 
1395 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1396 }
1397 
1398 /*
1399  * On receipt of path MTU corrections, flush old route and replace it
1400  * with the new one.  Retransmit all unacknowledged packets, to ensure
1401  * that all packets will be received.
1402  */
1403 void
1404 tcp_mtudisc(inp, errno)
1405 	struct inpcb *inp;
1406 	int errno;
1407 {
1408 	struct tcpcb *tp = intotcpcb(inp);
1409 	struct rtentry *rt = in_pcbrtentry(inp);
1410 
1411 	if (tp != 0) {
1412 		if (rt != 0) {
1413 			/*
1414 			 * If this was not a host route, remove and realloc.
1415 			 */
1416 			if ((rt->rt_flags & RTF_HOST) == 0) {
1417 				in_rtchange(inp, errno);
1418 				if ((rt = in_pcbrtentry(inp)) == 0)
1419 					return;
1420 			}
1421 
1422 			/*
1423 			 * Slow start out of the error condition.  We
1424 			 * use the MTU because we know it's smaller
1425 			 * than the previously transmitted segment.
1426 			 *
1427 			 * Note: This is more conservative than the
1428 			 * suggestion in draft-floyd-incr-init-win-03.
1429 			 */
1430 			if (rt->rt_rmx.rmx_mtu != 0)
1431 				tp->snd_cwnd =
1432 				    TCP_INITIAL_WINDOW(tcp_init_win,
1433 				    rt->rt_rmx.rmx_mtu);
1434 		}
1435 
1436 		/*
1437 		 * Resend unacknowledged packets.
1438 		 */
1439 		tp->snd_nxt = tp->snd_una;
1440 		tcp_output(tp);
1441 	}
1442 }
1443 #endif
1444 
1445 #ifdef INET6
1446 /*
1447  * Path MTU Discovery handlers.
1448  */
1449 void
1450 tcp6_mtudisc_callback(faddr)
1451 	struct in6_addr *faddr;
1452 {
1453 	struct sockaddr_in6 sin6;
1454 
1455 	bzero(&sin6, sizeof(sin6));
1456 	sin6.sin6_family = AF_INET6;
1457 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1458 	sin6.sin6_addr = *faddr;
1459 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1460 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1461 }
1462 
1463 void
1464 tcp6_mtudisc(in6p, errno)
1465 	struct in6pcb *in6p;
1466 	int errno;
1467 {
1468 	struct tcpcb *tp = in6totcpcb(in6p);
1469 	struct rtentry *rt = in6_pcbrtentry(in6p);
1470 
1471 	if (tp != 0) {
1472 		if (rt != 0) {
1473 			/*
1474 			 * If this was not a host route, remove and realloc.
1475 			 */
1476 			if ((rt->rt_flags & RTF_HOST) == 0) {
1477 				in6_rtchange(in6p, errno);
1478 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1479 					return;
1480 			}
1481 
1482 			/*
1483 			 * Slow start out of the error condition.  We
1484 			 * use the MTU because we know it's smaller
1485 			 * than the previously transmitted segment.
1486 			 *
1487 			 * Note: This is more conservative than the
1488 			 * suggestion in draft-floyd-incr-init-win-03.
1489 			 */
1490 			if (rt->rt_rmx.rmx_mtu != 0)
1491 				tp->snd_cwnd =
1492 				    TCP_INITIAL_WINDOW(tcp_init_win,
1493 				    rt->rt_rmx.rmx_mtu);
1494 		}
1495 
1496 		/*
1497 		 * Resend unacknowledged packets.
1498 		 */
1499 		tp->snd_nxt = tp->snd_una;
1500 		tcp_output(tp);
1501 	}
1502 }
1503 #endif /* INET6 */
1504 
1505 /*
1506  * Compute the MSS to advertise to the peer.  Called only during
1507  * the 3-way handshake.  If we are the server (peer initiated
1508  * connection), we are called with a pointer to the interface
1509  * on which the SYN packet arrived.  If we are the client (we
1510  * initiated connection), we are called with a pointer to the
1511  * interface out which this connection should go.
1512  *
1513  * NOTE: Do not subtract IP option/extension header size nor IPsec
1514  * header size from MSS advertisement.  MSS option must hold the maximum
1515  * segment size we can accept, so it must always be:
1516  *	 max(if mtu) - ip header - tcp header
1517  */
1518 u_long
1519 tcp_mss_to_advertise(ifp, af)
1520 	const struct ifnet *ifp;
1521 	int af;
1522 {
1523 	extern u_long in_maxmtu;
1524 	u_long mss = 0;
1525 	u_long hdrsiz;
1526 
1527 	/*
1528 	 * In order to avoid defeating path MTU discovery on the peer,
1529 	 * we advertise the max MTU of all attached networks as our MSS,
1530 	 * per RFC 1191, section 3.1.
1531 	 *
1532 	 * We provide the option to advertise just the MTU of
1533 	 * the interface on which we hope this connection will
1534 	 * be receiving.  If we are responding to a SYN, we
1535 	 * will have a pretty good idea about this, but when
1536 	 * initiating a connection there is a bit more doubt.
1537 	 *
1538 	 * We also need to ensure that loopback has a large enough
1539 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1540 	 */
1541 
1542 	if (ifp != NULL)
1543 		mss = ifp->if_mtu;
1544 
1545 	if (tcp_mss_ifmtu == 0)
1546 		switch (af) {
1547 		case AF_INET:
1548 			mss = max(in_maxmtu, mss);
1549 			break;
1550 #ifdef INET6
1551 		case AF_INET6:
1552 			mss = max(in6_maxmtu, mss);
1553 			break;
1554 #endif
1555 		}
1556 
1557 	switch (af) {
1558 	case AF_INET:
1559 		hdrsiz = sizeof(struct ip);
1560 		break;
1561 #ifdef INET6
1562 	case AF_INET6:
1563 		hdrsiz = sizeof(struct ip6_hdr);
1564 		break;
1565 #endif
1566 	default:
1567 		hdrsiz = 0;
1568 		break;
1569 	}
1570 	hdrsiz += sizeof(struct tcphdr);
1571 	if (mss > hdrsiz)
1572 		mss -= hdrsiz;
1573 
1574 	mss = max(tcp_mssdflt, mss);
1575 	return (mss);
1576 }
1577 
1578 /*
1579  * Set connection variables based on the peer's advertised MSS.
1580  * We are passed the TCPCB for the actual connection.  If we
1581  * are the server, we are called by the compressed state engine
1582  * when the 3-way handshake is complete.  If we are the client,
1583  * we are called when we receive the SYN,ACK from the server.
1584  *
1585  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1586  * before this routine is called!
1587  */
1588 void
1589 tcp_mss_from_peer(tp, offer)
1590 	struct tcpcb *tp;
1591 	int offer;
1592 {
1593 	struct socket *so;
1594 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1595 	struct rtentry *rt;
1596 #endif
1597 	u_long bufsize;
1598 	int mss;
1599 
1600 #ifdef DIAGNOSTIC
1601 	if (tp->t_inpcb && tp->t_in6pcb)
1602 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1603 #endif
1604 	so = NULL;
1605 	rt = NULL;
1606 #ifdef INET
1607 	if (tp->t_inpcb) {
1608 		so = tp->t_inpcb->inp_socket;
1609 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1610 		rt = in_pcbrtentry(tp->t_inpcb);
1611 #endif
1612 	}
1613 #endif
1614 #ifdef INET6
1615 	if (tp->t_in6pcb) {
1616 		so = tp->t_in6pcb->in6p_socket;
1617 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1618 		rt = in6_pcbrtentry(tp->t_in6pcb);
1619 #endif
1620 	}
1621 #endif
1622 
1623 	/*
1624 	 * As per RFC1122, use the default MSS value, unless they
1625 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1626 	 */
1627 	mss = tcp_mssdflt;
1628 	if (offer)
1629 		mss = offer;
1630 	mss = max(mss, 32);		/* sanity */
1631 	tp->t_peermss = mss;
1632 	mss -= tcp_optlen(tp);
1633 #ifdef INET
1634 	if (tp->t_inpcb)
1635 		mss -= ip_optlen(tp->t_inpcb);
1636 #endif
1637 #ifdef INET6
1638 	if (tp->t_in6pcb)
1639 		mss -= ip6_optlen(tp->t_in6pcb);
1640 #endif
1641 
1642 	/*
1643 	 * If there's a pipesize, change the socket buffer to that size.
1644 	 * Make the socket buffer an integral number of MSS units.  If
1645 	 * the MSS is larger than the socket buffer, artificially decrease
1646 	 * the MSS.
1647 	 */
1648 #ifdef RTV_SPIPE
1649 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1650 		bufsize = rt->rt_rmx.rmx_sendpipe;
1651 	else
1652 #endif
1653 		bufsize = so->so_snd.sb_hiwat;
1654 	if (bufsize < mss)
1655 		mss = bufsize;
1656 	else {
1657 		bufsize = roundup(bufsize, mss);
1658 		if (bufsize > sb_max)
1659 			bufsize = sb_max;
1660 		(void) sbreserve(&so->so_snd, bufsize);
1661 	}
1662 	tp->t_segsz = mss;
1663 
1664 #ifdef RTV_SSTHRESH
1665 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1666 		/*
1667 		 * There's some sort of gateway or interface buffer
1668 		 * limit on the path.  Use this to set the slow
1669 		 * start threshold, but set the threshold to no less
1670 		 * than 2 * MSS.
1671 		 */
1672 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1673 	}
1674 #endif
1675 }
1676 
1677 /*
1678  * Processing necessary when a TCP connection is established.
1679  */
1680 void
1681 tcp_established(tp)
1682 	struct tcpcb *tp;
1683 {
1684 	struct socket *so;
1685 #ifdef RTV_RPIPE
1686 	struct rtentry *rt;
1687 #endif
1688 	u_long bufsize;
1689 
1690 #ifdef DIAGNOSTIC
1691 	if (tp->t_inpcb && tp->t_in6pcb)
1692 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1693 #endif
1694 	so = NULL;
1695 	rt = NULL;
1696 #ifdef INET
1697 	if (tp->t_inpcb) {
1698 		so = tp->t_inpcb->inp_socket;
1699 #if defined(RTV_RPIPE)
1700 		rt = in_pcbrtentry(tp->t_inpcb);
1701 #endif
1702 	}
1703 #endif
1704 #ifdef INET6
1705 	if (tp->t_in6pcb) {
1706 		so = tp->t_in6pcb->in6p_socket;
1707 #if defined(RTV_RPIPE)
1708 		rt = in6_pcbrtentry(tp->t_in6pcb);
1709 #endif
1710 	}
1711 #endif
1712 
1713 	tp->t_state = TCPS_ESTABLISHED;
1714 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1715 
1716 #ifdef RTV_RPIPE
1717 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1718 		bufsize = rt->rt_rmx.rmx_recvpipe;
1719 	else
1720 #endif
1721 		bufsize = so->so_rcv.sb_hiwat;
1722 	if (bufsize > tp->t_ourmss) {
1723 		bufsize = roundup(bufsize, tp->t_ourmss);
1724 		if (bufsize > sb_max)
1725 			bufsize = sb_max;
1726 		(void) sbreserve(&so->so_rcv, bufsize);
1727 	}
1728 }
1729 
1730 /*
1731  * Check if there's an initial rtt or rttvar.  Convert from the
1732  * route-table units to scaled multiples of the slow timeout timer.
1733  * Called only during the 3-way handshake.
1734  */
1735 void
1736 tcp_rmx_rtt(tp)
1737 	struct tcpcb *tp;
1738 {
1739 #ifdef RTV_RTT
1740 	struct rtentry *rt = NULL;
1741 	int rtt;
1742 
1743 #ifdef DIAGNOSTIC
1744 	if (tp->t_inpcb && tp->t_in6pcb)
1745 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1746 #endif
1747 #ifdef INET
1748 	if (tp->t_inpcb)
1749 		rt = in_pcbrtentry(tp->t_inpcb);
1750 #endif
1751 #ifdef INET6
1752 	if (tp->t_in6pcb)
1753 		rt = in6_pcbrtentry(tp->t_in6pcb);
1754 #endif
1755 	if (rt == NULL)
1756 		return;
1757 
1758 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1759 		/*
1760 		 * XXX The lock bit for MTU indicates that the value
1761 		 * is also a minimum value; this is subject to time.
1762 		 */
1763 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1764 			TCPT_RANGESET(tp->t_rttmin,
1765 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1766 			    TCPTV_MIN, TCPTV_REXMTMAX);
1767 		tp->t_srtt = rtt /
1768 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1769 		if (rt->rt_rmx.rmx_rttvar) {
1770 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1771 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1772 				(TCP_RTTVAR_SHIFT + 2));
1773 		} else {
1774 			/* Default variation is +- 1 rtt */
1775 			tp->t_rttvar =
1776 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1777 		}
1778 		TCPT_RANGESET(tp->t_rxtcur,
1779 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1780 		    tp->t_rttmin, TCPTV_REXMTMAX);
1781 	}
1782 #endif
1783 }
1784 
1785 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1786 #if NRND > 0
1787 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1788 #endif
1789 
1790 /*
1791  * Get a new sequence value given a tcp control block
1792  */
1793 tcp_seq
1794 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1795 {
1796 
1797 #ifdef INET
1798 	if (tp->t_inpcb != NULL) {
1799 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1800 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1801 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1802 		    addin));
1803 	}
1804 #endif
1805 #ifdef INET6
1806 	if (tp->t_in6pcb != NULL) {
1807 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
1808 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
1809 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
1810 		    addin));
1811 	}
1812 #endif
1813 	/* Not possible. */
1814 	panic("tcp_new_iss");
1815 }
1816 
1817 /*
1818  * This routine actually generates a new TCP initial sequence number.
1819  */
1820 tcp_seq
1821 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
1822     size_t addrsz, tcp_seq addin)
1823 {
1824 	tcp_seq tcp_iss;
1825 
1826 #if NRND > 0
1827 	static int beenhere;
1828 
1829 	/*
1830 	 * If we haven't been here before, initialize our cryptographic
1831 	 * hash secret.
1832 	 */
1833 	if (beenhere == 0) {
1834 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
1835 		    RND_EXTRACT_ANY);
1836 		beenhere = 1;
1837 	}
1838 
1839 	if (tcp_do_rfc1948) {
1840 		MD5_CTX ctx;
1841 		u_int8_t hash[16];	/* XXX MD5 knowledge */
1842 
1843 		/*
1844 		 * Compute the base value of the ISS.  It is a hash
1845 		 * of (saddr, sport, daddr, dport, secret).
1846 		 */
1847 		MD5Init(&ctx);
1848 
1849 		MD5Update(&ctx, (u_char *) laddr, addrsz);
1850 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
1851 
1852 		MD5Update(&ctx, (u_char *) faddr, addrsz);
1853 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
1854 
1855 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
1856 
1857 		MD5Final(hash, &ctx);
1858 
1859 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
1860 
1861 		/*
1862 		 * Now increment our "timer", and add it in to
1863 		 * the computed value.
1864 		 *
1865 		 * XXX Use `addin'?
1866 		 * XXX TCP_ISSINCR too large to use?
1867 		 */
1868 		tcp_iss_seq += TCP_ISSINCR;
1869 #ifdef TCPISS_DEBUG
1870 		printf("ISS hash 0x%08x, ", tcp_iss);
1871 #endif
1872 		tcp_iss += tcp_iss_seq + addin;
1873 #ifdef TCPISS_DEBUG
1874 		printf("new ISS 0x%08x\n", tcp_iss);
1875 #endif
1876 	} else
1877 #endif /* NRND > 0 */
1878 	{
1879 		/*
1880 		 * Randomize.
1881 		 */
1882 #if NRND > 0
1883 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
1884 #else
1885 		tcp_iss = random();
1886 #endif
1887 
1888 		/*
1889 		 * If we were asked to add some amount to a known value,
1890 		 * we will take a random value obtained above, mask off
1891 		 * the upper bits, and add in the known value.  We also
1892 		 * add in a constant to ensure that we are at least a
1893 		 * certain distance from the original value.
1894 		 *
1895 		 * This is used when an old connection is in timed wait
1896 		 * and we have a new one coming in, for instance.
1897 		 */
1898 		if (addin != 0) {
1899 #ifdef TCPISS_DEBUG
1900 			printf("Random %08x, ", tcp_iss);
1901 #endif
1902 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1903 			tcp_iss += addin + TCP_ISSINCR;
1904 #ifdef TCPISS_DEBUG
1905 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
1906 #endif
1907 		} else {
1908 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1909 			tcp_iss += tcp_iss_seq;
1910 			tcp_iss_seq += TCP_ISSINCR;
1911 #ifdef TCPISS_DEBUG
1912 			printf("ISS %08x\n", tcp_iss);
1913 #endif
1914 		}
1915 	}
1916 
1917 	if (tcp_compat_42) {
1918 		/*
1919 		 * Limit it to the positive range for really old TCP
1920 		 * implementations.
1921 		 */
1922 		if (tcp_iss >= 0x80000000)
1923 			tcp_iss &= 0x7fffffff;		/* XXX */
1924 	}
1925 
1926 	return (tcp_iss);
1927 }
1928 
1929 #ifdef IPSEC
1930 /* compute ESP/AH header size for TCP, including outer IP header. */
1931 size_t
1932 ipsec4_hdrsiz_tcp(tp)
1933 	struct tcpcb *tp;
1934 {
1935 	struct inpcb *inp;
1936 	size_t hdrsiz;
1937 
1938 	/* XXX mapped addr case (tp->t_in6pcb) */
1939 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1940 		return 0;
1941 	switch (tp->t_family) {
1942 	case AF_INET:
1943 		/* XXX: should use currect direction. */
1944 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
1945 		break;
1946 	default:
1947 		hdrsiz = 0;
1948 		break;
1949 	}
1950 
1951 	return hdrsiz;
1952 }
1953 
1954 #ifdef INET6
1955 size_t
1956 ipsec6_hdrsiz_tcp(tp)
1957 	struct tcpcb *tp;
1958 {
1959 	struct in6pcb *in6p;
1960 	size_t hdrsiz;
1961 
1962 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
1963 		return 0;
1964 	switch (tp->t_family) {
1965 	case AF_INET6:
1966 		/* XXX: should use currect direction. */
1967 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
1968 		break;
1969 	case AF_INET:
1970 		/* mapped address case - tricky */
1971 	default:
1972 		hdrsiz = 0;
1973 		break;
1974 	}
1975 
1976 	return hdrsiz;
1977 }
1978 #endif
1979 #endif /*IPSEC*/
1980 
1981 /*
1982  * Determine the length of the TCP options for this connection.
1983  *
1984  * XXX:  What do we do for SACK, when we add that?  Just reserve
1985  *       all of the space?  Otherwise we can't exactly be incrementing
1986  *       cwnd by an amount that varies depending on the amount we last
1987  *       had to SACK!
1988  */
1989 
1990 u_int
1991 tcp_optlen(tp)
1992 	struct tcpcb *tp;
1993 {
1994 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1995 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
1996 		return TCPOLEN_TSTAMP_APPA;
1997 	else
1998 		return 0;
1999 }
2000