xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 366e81869c056a2c569cba90e971c65ad4d009e6)
1 /*	$NetBSD: tcp_subr.c,v 1.124 2002/03/15 09:25:41 itojun Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998, 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 <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.124 2002/03/15 09:25:41 itojun Exp $");
106 
107 #include "opt_inet.h"
108 #include "opt_ipsec.h"
109 #include "opt_tcp_compat_42.h"
110 #include "opt_inet_csum.h"
111 #include "rnd.h"
112 
113 #include <sys/param.h>
114 #include <sys/proc.h>
115 #include <sys/systm.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/socket.h>
119 #include <sys/socketvar.h>
120 #include <sys/protosw.h>
121 #include <sys/errno.h>
122 #include <sys/kernel.h>
123 #include <sys/pool.h>
124 #if NRND > 0
125 #include <sys/md5.h>
126 #include <sys/rnd.h>
127 #endif
128 
129 #include <net/route.h>
130 #include <net/if.h>
131 
132 #include <netinet/in.h>
133 #include <netinet/in_systm.h>
134 #include <netinet/ip.h>
135 #include <netinet/in_pcb.h>
136 #include <netinet/ip_var.h>
137 #include <netinet/ip_icmp.h>
138 
139 #ifdef INET6
140 #ifndef INET
141 #include <netinet/in.h>
142 #endif
143 #include <netinet/ip6.h>
144 #include <netinet6/in6_pcb.h>
145 #include <netinet6/ip6_var.h>
146 #include <netinet6/in6_var.h>
147 #include <netinet6/ip6protosw.h>
148 #include <netinet/icmp6.h>
149 #endif
150 
151 #include <netinet/tcp.h>
152 #include <netinet/tcp_fsm.h>
153 #include <netinet/tcp_seq.h>
154 #include <netinet/tcp_timer.h>
155 #include <netinet/tcp_var.h>
156 #include <netinet/tcpip.h>
157 
158 #ifdef IPSEC
159 #include <netinet6/ipsec.h>
160 #endif /*IPSEC*/
161 
162 #ifdef INET6
163 struct in6pcb tcb6;
164 #endif
165 
166 /* patchable/settable parameters for tcp */
167 int 	tcp_mssdflt = TCP_MSS;
168 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
169 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
170 #if NRND > 0
171 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
172 #endif
173 int	tcp_do_sack = 1;	/* selective acknowledgement */
174 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
175 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
176 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
177 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
178 int	tcp_init_win = 1;
179 int	tcp_mss_ifmtu = 0;
180 #ifdef TCP_COMPAT_42
181 int	tcp_compat_42 = 1;
182 #else
183 int	tcp_compat_42 = 0;
184 #endif
185 int	tcp_rst_ppslim = 100;	/* 100pps */
186 
187 /* tcb hash */
188 #ifndef TCBHASHSIZE
189 #define	TCBHASHSIZE	128
190 #endif
191 int	tcbhashsize = TCBHASHSIZE;
192 
193 /* syn hash parameters */
194 #define	TCP_SYN_HASH_SIZE	293
195 #define	TCP_SYN_BUCKET_SIZE	35
196 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
197 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
198 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
199 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
200 
201 int	tcp_freeq __P((struct tcpcb *));
202 
203 #ifdef INET
204 void	tcp_mtudisc_callback __P((struct in_addr));
205 #endif
206 #ifdef INET6
207 void	tcp6_mtudisc_callback __P((struct in6_addr *));
208 #endif
209 
210 void	tcp_mtudisc __P((struct inpcb *, int));
211 #ifdef INET6
212 void	tcp6_mtudisc __P((struct in6pcb *, int));
213 #endif
214 
215 struct pool tcpcb_pool;
216 
217 #ifdef TCP_CSUM_COUNTERS
218 #include <sys/device.h>
219 
220 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
221     NULL, "tcp", "hwcsum bad");
222 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
223     NULL, "tcp", "hwcsum ok");
224 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
225     NULL, "tcp", "hwcsum data");
226 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
227     NULL, "tcp", "swcsum");
228 #endif /* TCP_CSUM_COUNTERS */
229 
230 /*
231  * Tcp initialization
232  */
233 void
234 tcp_init()
235 {
236 	int hlen;
237 
238 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
239 	    NULL);
240 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
241 #ifdef INET6
242 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
243 #endif
244 
245 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
246 #ifdef INET6
247 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
248 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
249 #endif
250 	if (max_protohdr < hlen)
251 		max_protohdr = hlen;
252 	if (max_linkhdr + hlen > MHLEN)
253 		panic("tcp_init");
254 
255 #ifdef INET
256 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
257 #endif
258 #ifdef INET6
259 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
260 #endif
261 
262 	/* Initialize timer state. */
263 	tcp_timer_init();
264 
265 	/* Initialize the compressed state engine. */
266 	syn_cache_init();
267 
268 #ifdef TCP_CSUM_COUNTERS
269 	evcnt_attach_static(&tcp_hwcsum_bad);
270 	evcnt_attach_static(&tcp_hwcsum_ok);
271 	evcnt_attach_static(&tcp_hwcsum_data);
272 	evcnt_attach_static(&tcp_swcsum);
273 #endif /* TCP_CSUM_COUNTERS */
274 }
275 
276 /*
277  * Create template to be used to send tcp packets on a connection.
278  * Call after host entry created, allocates an mbuf and fills
279  * in a skeletal tcp/ip header, minimizing the amount of work
280  * necessary when the connection is used.
281  */
282 struct mbuf *
283 tcp_template(tp)
284 	struct tcpcb *tp;
285 {
286 	struct inpcb *inp = tp->t_inpcb;
287 #ifdef INET6
288 	struct in6pcb *in6p = tp->t_in6pcb;
289 #endif
290 	struct tcphdr *n;
291 	struct mbuf *m;
292 	int hlen;
293 
294 	switch (tp->t_family) {
295 	case AF_INET:
296 		hlen = sizeof(struct ip);
297 		if (inp)
298 			break;
299 #ifdef INET6
300 		if (in6p) {
301 			/* mapped addr case */
302 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
303 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
304 				break;
305 		}
306 #endif
307 		return NULL;	/*EINVAL*/
308 #ifdef INET6
309 	case AF_INET6:
310 		hlen = sizeof(struct ip6_hdr);
311 		if (in6p) {
312 			/* more sainty check? */
313 			break;
314 		}
315 		return NULL;	/*EINVAL*/
316 #endif
317 	default:
318 		hlen = 0;	/*pacify gcc*/
319 		return NULL;	/*EAFNOSUPPORT*/
320 	}
321 #ifdef DIAGNOSTIC
322 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
323 		panic("mclbytes too small for t_template");
324 #endif
325 	m = tp->t_template;
326 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
327 		;
328 	else {
329 		if (m)
330 			m_freem(m);
331 		m = tp->t_template = NULL;
332 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
333 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
334 			MCLGET(m, M_DONTWAIT);
335 			if ((m->m_flags & M_EXT) == 0) {
336 				m_free(m);
337 				m = NULL;
338 			}
339 		}
340 		if (m == NULL)
341 			return NULL;
342 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
343 	}
344 
345 	bzero(mtod(m, caddr_t), m->m_len);
346 
347 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
348 
349 	switch (tp->t_family) {
350 	case AF_INET:
351 	    {
352 		struct ipovly *ipov;
353 		mtod(m, struct ip *)->ip_v = 4;
354 		ipov = mtod(m, struct ipovly *);
355 		ipov->ih_pr = IPPROTO_TCP;
356 		ipov->ih_len = htons(sizeof(struct tcphdr));
357 		if (inp) {
358 			ipov->ih_src = inp->inp_laddr;
359 			ipov->ih_dst = inp->inp_faddr;
360 		}
361 #ifdef INET6
362 		else if (in6p) {
363 			/* mapped addr case */
364 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
365 				sizeof(ipov->ih_src));
366 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
367 				sizeof(ipov->ih_dst));
368 		}
369 #endif
370 		/*
371 		 * Compute the pseudo-header portion of the checksum
372 		 * now.  We incrementally add in the TCP option and
373 		 * payload lengths later, and then compute the TCP
374 		 * checksum right before the packet is sent off onto
375 		 * the wire.
376 		 */
377 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
378 		    ipov->ih_dst.s_addr,
379 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
380 		break;
381 	    }
382 #ifdef INET6
383 	case AF_INET6:
384 	    {
385 		struct ip6_hdr *ip6;
386 		mtod(m, struct ip *)->ip_v = 6;
387 		ip6 = mtod(m, struct ip6_hdr *);
388 		ip6->ip6_nxt = IPPROTO_TCP;
389 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
390 		ip6->ip6_src = in6p->in6p_laddr;
391 		ip6->ip6_dst = in6p->in6p_faddr;
392 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
393 		if (ip6_auto_flowlabel) {
394 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
395 			ip6->ip6_flow |=
396 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
397 		}
398 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
399 		ip6->ip6_vfc |= IPV6_VERSION;
400 
401 		/*
402 		 * Compute the pseudo-header portion of the checksum
403 		 * now.  We incrementally add in the TCP option and
404 		 * payload lengths later, and then compute the TCP
405 		 * checksum right before the packet is sent off onto
406 		 * the wire.
407 		 */
408 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
409 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
410 		    htonl(IPPROTO_TCP));
411 		break;
412 	    }
413 #endif
414 	}
415 	if (inp) {
416 		n->th_sport = inp->inp_lport;
417 		n->th_dport = inp->inp_fport;
418 	}
419 #ifdef INET6
420 	else if (in6p) {
421 		n->th_sport = in6p->in6p_lport;
422 		n->th_dport = in6p->in6p_fport;
423 	}
424 #endif
425 	n->th_seq = 0;
426 	n->th_ack = 0;
427 	n->th_x2 = 0;
428 	n->th_off = 5;
429 	n->th_flags = 0;
430 	n->th_win = 0;
431 	n->th_urp = 0;
432 	return (m);
433 }
434 
435 /*
436  * Send a single message to the TCP at address specified by
437  * the given TCP/IP header.  If m == 0, then we make a copy
438  * of the tcpiphdr at ti and send directly to the addressed host.
439  * This is used to force keep alive messages out using the TCP
440  * template for a connection tp->t_template.  If flags are given
441  * then we send a message back to the TCP which originated the
442  * segment ti, and discard the mbuf containing it and any other
443  * attached mbufs.
444  *
445  * In any case the ack and sequence number of the transmitted
446  * segment are as specified by the parameters.
447  */
448 int
449 tcp_respond(tp, template, m, th0, ack, seq, flags)
450 	struct tcpcb *tp;
451 	struct mbuf *template;
452 	struct mbuf *m;
453 	struct tcphdr *th0;
454 	tcp_seq ack, seq;
455 	int flags;
456 {
457 	struct route *ro;
458 	int error, tlen, win = 0;
459 	int hlen;
460 	struct ip *ip;
461 #ifdef INET6
462 	struct ip6_hdr *ip6;
463 #endif
464 	int family;	/* family on packet, not inpcb/in6pcb! */
465 	struct tcphdr *th;
466 
467 	if (tp != NULL && (flags & TH_RST) == 0) {
468 #ifdef DIAGNOSTIC
469 		if (tp->t_inpcb && tp->t_in6pcb)
470 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
471 #endif
472 #ifdef INET
473 		if (tp->t_inpcb)
474 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
475 #endif
476 #ifdef INET6
477 		if (tp->t_in6pcb)
478 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
479 #endif
480 	}
481 
482 	ip = NULL;
483 #ifdef INET6
484 	ip6 = NULL;
485 #endif
486 	if (m == 0) {
487 		if (!template)
488 			return EINVAL;
489 
490 		/* get family information from template */
491 		switch (mtod(template, struct ip *)->ip_v) {
492 		case 4:
493 			family = AF_INET;
494 			hlen = sizeof(struct ip);
495 			break;
496 #ifdef INET6
497 		case 6:
498 			family = AF_INET6;
499 			hlen = sizeof(struct ip6_hdr);
500 			break;
501 #endif
502 		default:
503 			return EAFNOSUPPORT;
504 		}
505 
506 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
507 		if (m) {
508 			MCLGET(m, M_DONTWAIT);
509 			if ((m->m_flags & M_EXT) == 0) {
510 				m_free(m);
511 				m = NULL;
512 			}
513 		}
514 		if (m == NULL)
515 			return (ENOBUFS);
516 
517 		if (tcp_compat_42)
518 			tlen = 1;
519 		else
520 			tlen = 0;
521 
522 		m->m_data += max_linkhdr;
523 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
524 			template->m_len);
525 		switch (family) {
526 		case AF_INET:
527 			ip = mtod(m, struct ip *);
528 			th = (struct tcphdr *)(ip + 1);
529 			break;
530 #ifdef INET6
531 		case AF_INET6:
532 			ip6 = mtod(m, struct ip6_hdr *);
533 			th = (struct tcphdr *)(ip6 + 1);
534 			break;
535 #endif
536 #if 0
537 		default:
538 			/* noone will visit here */
539 			m_freem(m);
540 			return EAFNOSUPPORT;
541 #endif
542 		}
543 		flags = TH_ACK;
544 	} else {
545 
546 		if ((m->m_flags & M_PKTHDR) == 0) {
547 #if 0
548 			printf("non PKTHDR to tcp_respond\n");
549 #endif
550 			m_freem(m);
551 			return EINVAL;
552 		}
553 #ifdef DIAGNOSTIC
554 		if (!th0)
555 			panic("th0 == NULL in tcp_respond");
556 #endif
557 
558 		/* get family information from m */
559 		switch (mtod(m, struct ip *)->ip_v) {
560 		case 4:
561 			family = AF_INET;
562 			hlen = sizeof(struct ip);
563 			ip = mtod(m, struct ip *);
564 			break;
565 #ifdef INET6
566 		case 6:
567 			family = AF_INET6;
568 			hlen = sizeof(struct ip6_hdr);
569 			ip6 = mtod(m, struct ip6_hdr *);
570 			break;
571 #endif
572 		default:
573 			m_freem(m);
574 			return EAFNOSUPPORT;
575 		}
576 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
577 			tlen = sizeof(*th0);
578 		else
579 			tlen = th0->th_off << 2;
580 
581 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
582 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
583 			m->m_len = hlen + tlen;
584 			m_freem(m->m_next);
585 			m->m_next = NULL;
586 		} else {
587 			struct mbuf *n;
588 
589 #ifdef DIAGNOSTIC
590 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
591 				m_freem(m);
592 				return EMSGSIZE;
593 			}
594 #endif
595 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
596 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
597 				MCLGET(n, M_DONTWAIT);
598 				if ((n->m_flags & M_EXT) == 0) {
599 					m_freem(n);
600 					n = NULL;
601 				}
602 			}
603 			if (!n) {
604 				m_freem(m);
605 				return ENOBUFS;
606 			}
607 
608 			n->m_data += max_linkhdr;
609 			n->m_len = hlen + tlen;
610 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
611 			m_copyback(n, hlen, tlen, (caddr_t)th0);
612 
613 			m_freem(m);
614 			m = n;
615 			n = NULL;
616 		}
617 
618 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
619 		switch (family) {
620 		case AF_INET:
621 			ip = mtod(m, struct ip *);
622 			th = (struct tcphdr *)(ip + 1);
623 			ip->ip_p = IPPROTO_TCP;
624 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
625 			ip->ip_p = IPPROTO_TCP;
626 			break;
627 #ifdef INET6
628 		case AF_INET6:
629 			ip6 = mtod(m, struct ip6_hdr *);
630 			th = (struct tcphdr *)(ip6 + 1);
631 			ip6->ip6_nxt = IPPROTO_TCP;
632 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
633 			ip6->ip6_nxt = IPPROTO_TCP;
634 			break;
635 #endif
636 #if 0
637 		default:
638 			/* noone will visit here */
639 			m_freem(m);
640 			return EAFNOSUPPORT;
641 #endif
642 		}
643 		xchg(th->th_dport, th->th_sport, u_int16_t);
644 #undef xchg
645 		tlen = 0;	/*be friendly with the following code*/
646 	}
647 	th->th_seq = htonl(seq);
648 	th->th_ack = htonl(ack);
649 	th->th_x2 = 0;
650 	if ((flags & TH_SYN) == 0) {
651 		if (tp)
652 			win >>= tp->rcv_scale;
653 		if (win > TCP_MAXWIN)
654 			win = TCP_MAXWIN;
655 		th->th_win = htons((u_int16_t)win);
656 		th->th_off = sizeof (struct tcphdr) >> 2;
657 		tlen += sizeof(*th);
658 	} else
659 		tlen += th->th_off << 2;
660 	m->m_len = hlen + tlen;
661 	m->m_pkthdr.len = hlen + tlen;
662 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
663 	th->th_flags = flags;
664 	th->th_urp = 0;
665 
666 	switch (family) {
667 #ifdef INET
668 	case AF_INET:
669 	    {
670 		struct ipovly *ipov = (struct ipovly *)ip;
671 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
672 		ipov->ih_len = htons((u_int16_t)tlen);
673 
674 		th->th_sum = 0;
675 		th->th_sum = in_cksum(m, hlen + tlen);
676 		ip->ip_len = hlen + tlen;	/*will be flipped on output*/
677 		ip->ip_ttl = ip_defttl;
678 		break;
679 	    }
680 #endif
681 #ifdef INET6
682 	case AF_INET6:
683 	    {
684 		th->th_sum = 0;
685 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
686 				tlen);
687 		ip6->ip6_plen = ntohs(tlen);
688 		if (tp && tp->t_in6pcb) {
689 			struct ifnet *oifp;
690 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
691 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
692 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
693 		} else
694 			ip6->ip6_hlim = ip6_defhlim;
695 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
696 		if (ip6_auto_flowlabel) {
697 			ip6->ip6_flow |=
698 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
699 		}
700 		break;
701 	    }
702 #endif
703 	}
704 
705 #ifdef IPSEC
706 	(void)ipsec_setsocket(m, NULL);
707 #endif /*IPSEC*/
708 
709 	if (tp != NULL && tp->t_inpcb != NULL) {
710 		ro = &tp->t_inpcb->inp_route;
711 #ifdef IPSEC
712 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
713 			m_freem(m);
714 			return ENOBUFS;
715 		}
716 #endif
717 #ifdef DIAGNOSTIC
718 		if (family != AF_INET)
719 			panic("tcp_respond: address family mismatch");
720 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
721 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
722 			    ntohl(ip->ip_dst.s_addr),
723 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
724 		}
725 #endif
726 	}
727 #ifdef INET6
728 	else if (tp != NULL && tp->t_in6pcb != NULL) {
729 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
730 #ifdef IPSEC
731 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
732 			m_freem(m);
733 			return ENOBUFS;
734 		}
735 #endif
736 #ifdef DIAGNOSTIC
737 		if (family == AF_INET) {
738 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
739 				panic("tcp_respond: not mapped addr");
740 			if (bcmp(&ip->ip_dst,
741 					&tp->t_in6pcb->in6p_faddr.s6_addr32[3],
742 					sizeof(ip->ip_dst)) != 0) {
743 				panic("tcp_respond: ip_dst != in6p_faddr");
744 			}
745 		} else if (family == AF_INET6) {
746 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr))
747 				panic("tcp_respond: ip6_dst != in6p_faddr");
748 		} else
749 			panic("tcp_respond: address family mismatch");
750 #endif
751 	}
752 #endif
753 	else
754 		ro = NULL;
755 
756 	switch (family) {
757 #ifdef INET
758 	case AF_INET:
759 		error = ip_output(m, NULL, ro,
760 		    (ip_mtudisc ? IP_MTUDISC : 0),
761 		    NULL);
762 		break;
763 #endif
764 #ifdef INET6
765 	case AF_INET6:
766 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
767 			NULL);
768 		break;
769 #endif
770 	default:
771 		error = EAFNOSUPPORT;
772 		break;
773 	}
774 
775 	return (error);
776 }
777 
778 /*
779  * Create a new TCP control block, making an
780  * empty reassembly queue and hooking it to the argument
781  * protocol control block.
782  */
783 struct tcpcb *
784 tcp_newtcpcb(family, aux)
785 	int family;	/* selects inpcb, or in6pcb */
786 	void *aux;
787 {
788 	struct tcpcb *tp;
789 	int i;
790 
791 	switch (family) {
792 	case PF_INET:
793 		break;
794 #ifdef INET6
795 	case PF_INET6:
796 		break;
797 #endif
798 	default:
799 		return NULL;
800 	}
801 
802 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
803 	if (tp == NULL)
804 		return (NULL);
805 	bzero((caddr_t)tp, sizeof(struct tcpcb));
806 	LIST_INIT(&tp->segq);
807 	LIST_INIT(&tp->timeq);
808 	tp->t_family = family;		/* may be overridden later on */
809 	tp->t_peermss = tcp_mssdflt;
810 	tp->t_ourmss = tcp_mssdflt;
811 	tp->t_segsz = tcp_mssdflt;
812 	LIST_INIT(&tp->t_sc);
813 
814 	callout_init(&tp->t_delack_ch);
815 	for (i = 0; i < TCPT_NTIMERS; i++)
816 		TCP_TIMER_INIT(tp, i);
817 
818 	tp->t_flags = 0;
819 	if (tcp_do_rfc1323 && tcp_do_win_scale)
820 		tp->t_flags |= TF_REQ_SCALE;
821 	if (tcp_do_rfc1323 && tcp_do_timestamps)
822 		tp->t_flags |= TF_REQ_TSTMP;
823 	if (tcp_do_sack == 2)
824 		tp->t_flags |= TF_WILL_SACK;
825 	else if (tcp_do_sack == 1)
826 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
827 	tp->t_flags |= TF_CANT_TXSACK;
828 	switch (family) {
829 	case PF_INET:
830 		tp->t_inpcb = (struct inpcb *)aux;
831 		break;
832 #ifdef INET6
833 	case PF_INET6:
834 		tp->t_in6pcb = (struct in6pcb *)aux;
835 		break;
836 #endif
837 	}
838 	/*
839 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
840 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
841 	 * reasonable initial retransmit time.
842 	 */
843 	tp->t_srtt = TCPTV_SRTTBASE;
844 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
845 	tp->t_rttmin = TCPTV_MIN;
846 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
847 	    TCPTV_MIN, TCPTV_REXMTMAX);
848 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
849 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
850 	if (family == AF_INET) {
851 		struct inpcb *inp = (struct inpcb *)aux;
852 		inp->inp_ip.ip_ttl = ip_defttl;
853 		inp->inp_ppcb = (caddr_t)tp;
854 	}
855 #ifdef INET6
856 	else if (family == AF_INET6) {
857 		struct in6pcb *in6p = (struct in6pcb *)aux;
858 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
859 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
860 					       : NULL);
861 		in6p->in6p_ppcb = (caddr_t)tp;
862 	}
863 #endif
864 
865 	/*
866 	 * Initialize our timebase.  When we send timestamps, we take
867 	 * the delta from tcp_now -- this means each connection always
868 	 * gets a timebase of 0, which makes it, among other things,
869 	 * more difficult to determine how long a system has been up,
870 	 * and thus how many TCP sequence increments have occurred.
871 	 */
872 	tp->ts_timebase = tcp_now;
873 
874 	return (tp);
875 }
876 
877 /*
878  * Drop a TCP connection, reporting
879  * the specified error.  If connection is synchronized,
880  * then send a RST to peer.
881  */
882 struct tcpcb *
883 tcp_drop(tp, errno)
884 	struct tcpcb *tp;
885 	int errno;
886 {
887 	struct socket *so = NULL;
888 
889 #ifdef DIAGNOSTIC
890 	if (tp->t_inpcb && tp->t_in6pcb)
891 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
892 #endif
893 #ifdef INET
894 	if (tp->t_inpcb)
895 		so = tp->t_inpcb->inp_socket;
896 #endif
897 #ifdef INET6
898 	if (tp->t_in6pcb)
899 		so = tp->t_in6pcb->in6p_socket;
900 #endif
901 	if (!so)
902 		return NULL;
903 
904 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
905 		tp->t_state = TCPS_CLOSED;
906 		(void) tcp_output(tp);
907 		tcpstat.tcps_drops++;
908 	} else
909 		tcpstat.tcps_conndrops++;
910 	if (errno == ETIMEDOUT && tp->t_softerror)
911 		errno = tp->t_softerror;
912 	so->so_error = errno;
913 	return (tcp_close(tp));
914 }
915 
916 /*
917  * Close a TCP control block:
918  *	discard all space held by the tcp
919  *	discard internet protocol block
920  *	wake up any sleepers
921  */
922 struct tcpcb *
923 tcp_close(tp)
924 	struct tcpcb *tp;
925 {
926 	struct inpcb *inp;
927 #ifdef INET6
928 	struct in6pcb *in6p;
929 #endif
930 	struct socket *so;
931 #ifdef RTV_RTT
932 	struct rtentry *rt;
933 #endif
934 	struct route *ro;
935 
936 	inp = tp->t_inpcb;
937 #ifdef INET6
938 	in6p = tp->t_in6pcb;
939 #endif
940 	so = NULL;
941 	ro = NULL;
942 	if (inp) {
943 		so = inp->inp_socket;
944 		ro = &inp->inp_route;
945 	}
946 #ifdef INET6
947 	else if (in6p) {
948 		so = in6p->in6p_socket;
949 		ro = (struct route *)&in6p->in6p_route;
950 	}
951 #endif
952 
953 #ifdef RTV_RTT
954 	/*
955 	 * If we sent enough data to get some meaningful characteristics,
956 	 * save them in the routing entry.  'Enough' is arbitrarily
957 	 * defined as the sendpipesize (default 4K) * 16.  This would
958 	 * give us 16 rtt samples assuming we only get one sample per
959 	 * window (the usual case on a long haul net).  16 samples is
960 	 * enough for the srtt filter to converge to within 5% of the correct
961 	 * value; fewer samples and we could save a very bogus rtt.
962 	 *
963 	 * Don't update the default route's characteristics and don't
964 	 * update anything that the user "locked".
965 	 */
966 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
967 	    ro && (rt = ro->ro_rt) &&
968 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
969 		u_long i = 0;
970 
971 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
972 			i = tp->t_srtt *
973 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
974 			if (rt->rt_rmx.rmx_rtt && i)
975 				/*
976 				 * filter this update to half the old & half
977 				 * the new values, converting scale.
978 				 * See route.h and tcp_var.h for a
979 				 * description of the scaling constants.
980 				 */
981 				rt->rt_rmx.rmx_rtt =
982 				    (rt->rt_rmx.rmx_rtt + i) / 2;
983 			else
984 				rt->rt_rmx.rmx_rtt = i;
985 		}
986 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
987 			i = tp->t_rttvar *
988 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
989 			if (rt->rt_rmx.rmx_rttvar && i)
990 				rt->rt_rmx.rmx_rttvar =
991 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
992 			else
993 				rt->rt_rmx.rmx_rttvar = i;
994 		}
995 		/*
996 		 * update the pipelimit (ssthresh) if it has been updated
997 		 * already or if a pipesize was specified & the threshhold
998 		 * got below half the pipesize.  I.e., wait for bad news
999 		 * before we start updating, then update on both good
1000 		 * and bad news.
1001 		 */
1002 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1003 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1004 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1005 			/*
1006 			 * convert the limit from user data bytes to
1007 			 * packets then to packet data bytes.
1008 			 */
1009 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1010 			if (i < 2)
1011 				i = 2;
1012 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1013 			if (rt->rt_rmx.rmx_ssthresh)
1014 				rt->rt_rmx.rmx_ssthresh =
1015 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1016 			else
1017 				rt->rt_rmx.rmx_ssthresh = i;
1018 		}
1019 	}
1020 #endif /* RTV_RTT */
1021 	/* free the reassembly queue, if any */
1022 	TCP_REASS_LOCK(tp);
1023 	(void) tcp_freeq(tp);
1024 	TCP_REASS_UNLOCK(tp);
1025 
1026 	tcp_canceltimers(tp);
1027 	TCP_CLEAR_DELACK(tp);
1028 	syn_cache_cleanup(tp);
1029 
1030 	if (tp->t_template) {
1031 		m_free(tp->t_template);
1032 		tp->t_template = NULL;
1033 	}
1034 	pool_put(&tcpcb_pool, tp);
1035 	if (inp) {
1036 		inp->inp_ppcb = 0;
1037 		soisdisconnected(so);
1038 		in_pcbdetach(inp);
1039 	}
1040 #ifdef INET6
1041 	else if (in6p) {
1042 		in6p->in6p_ppcb = 0;
1043 		soisdisconnected(so);
1044 		in6_pcbdetach(in6p);
1045 	}
1046 #endif
1047 	tcpstat.tcps_closed++;
1048 	return ((struct tcpcb *)0);
1049 }
1050 
1051 int
1052 tcp_freeq(tp)
1053 	struct tcpcb *tp;
1054 {
1055 	struct ipqent *qe;
1056 	int rv = 0;
1057 #ifdef TCPREASS_DEBUG
1058 	int i = 0;
1059 #endif
1060 
1061 	TCP_REASS_LOCK_CHECK(tp);
1062 
1063 	while ((qe = LIST_FIRST(&tp->segq)) != NULL) {
1064 #ifdef TCPREASS_DEBUG
1065 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1066 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1067 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1068 #endif
1069 		LIST_REMOVE(qe, ipqe_q);
1070 		LIST_REMOVE(qe, ipqe_timeq);
1071 		m_freem(qe->ipqe_m);
1072 		pool_put(&ipqent_pool, qe);
1073 		rv = 1;
1074 	}
1075 	return (rv);
1076 }
1077 
1078 /*
1079  * Protocol drain routine.  Called when memory is in short supply.
1080  */
1081 void
1082 tcp_drain()
1083 {
1084 	struct inpcb *inp;
1085 	struct tcpcb *tp;
1086 
1087 	/*
1088 	 * Free the sequence queue of all TCP connections.
1089 	 */
1090 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1091 	if (inp)						/* XXX */
1092 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1093 		if ((tp = intotcpcb(inp)) != NULL) {
1094 			/*
1095 			 * We may be called from a device's interrupt
1096 			 * context.  If the tcpcb is already busy,
1097 			 * just bail out now.
1098 			 */
1099 			if (tcp_reass_lock_try(tp) == 0)
1100 				continue;
1101 			if (tcp_freeq(tp))
1102 				tcpstat.tcps_connsdrained++;
1103 			TCP_REASS_UNLOCK(tp);
1104 		}
1105 	}
1106 }
1107 
1108 #ifdef INET6
1109 void
1110 tcp6_drain()
1111 {
1112 	struct in6pcb *in6p;
1113 	struct tcpcb *tp;
1114 	struct in6pcb *head = &tcb6;
1115 
1116 	/*
1117 	 * Free the sequence queue of all TCP connections.
1118 	 */
1119 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1120 		if ((tp = in6totcpcb(in6p)) != NULL) {
1121 			/*
1122 			 * We may be called from a device's interrupt
1123 			 * context.  If the tcpcb is already busy,
1124 			 * just bail out now.
1125 			 */
1126 			if (tcp_reass_lock_try(tp) == 0)
1127 				continue;
1128 			if (tcp_freeq(tp))
1129 				tcpstat.tcps_connsdrained++;
1130 			TCP_REASS_UNLOCK(tp);
1131 		}
1132 	}
1133 }
1134 #endif
1135 
1136 /*
1137  * Notify a tcp user of an asynchronous error;
1138  * store error as soft error, but wake up user
1139  * (for now, won't do anything until can select for soft error).
1140  */
1141 void
1142 tcp_notify(inp, error)
1143 	struct inpcb *inp;
1144 	int error;
1145 {
1146 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1147 	struct socket *so = inp->inp_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 
1170 #ifdef INET6
1171 void
1172 tcp6_notify(in6p, error)
1173 	struct in6pcb *in6p;
1174 	int error;
1175 {
1176 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1177 	struct socket *so = in6p->in6p_socket;
1178 
1179 	/*
1180 	 * Ignore some errors if we are hooked up.
1181 	 * If connection hasn't completed, has retransmitted several times,
1182 	 * and receives a second error, give up now.  This is better
1183 	 * than waiting a long time to establish a connection that
1184 	 * can never complete.
1185 	 */
1186 	if (tp->t_state == TCPS_ESTABLISHED &&
1187 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1188 	      error == EHOSTDOWN)) {
1189 		return;
1190 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1191 	    tp->t_rxtshift > 3 && tp->t_softerror)
1192 		so->so_error = error;
1193 	else
1194 		tp->t_softerror = error;
1195 	wakeup((caddr_t) &so->so_timeo);
1196 	sorwakeup(so);
1197 	sowwakeup(so);
1198 }
1199 #endif
1200 
1201 #ifdef INET6
1202 void
1203 tcp6_ctlinput(cmd, sa, d)
1204 	int cmd;
1205 	struct sockaddr *sa;
1206 	void *d;
1207 {
1208 	struct tcphdr th;
1209 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1210 	int nmatch;
1211 	struct ip6_hdr *ip6;
1212 	const struct sockaddr_in6 *sa6_src = NULL;
1213 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1214 	struct mbuf *m;
1215 	int off;
1216 
1217 	if (sa->sa_family != AF_INET6 ||
1218 	    sa->sa_len != sizeof(struct sockaddr_in6))
1219 		return;
1220 	if ((unsigned)cmd >= PRC_NCMDS)
1221 		return;
1222 	else if (cmd == PRC_QUENCH) {
1223 		/* XXX there's no PRC_QUENCH in IPv6 */
1224 		notify = tcp6_quench;
1225 	} else if (PRC_IS_REDIRECT(cmd))
1226 		notify = in6_rtchange, d = NULL;
1227 	else if (cmd == PRC_MSGSIZE)
1228 		; /* special code is present, see below */
1229 	else if (cmd == PRC_HOSTDEAD)
1230 		d = NULL;
1231 	else if (inet6ctlerrmap[cmd] == 0)
1232 		return;
1233 
1234 	/* if the parameter is from icmp6, decode it. */
1235 	if (d != NULL) {
1236 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1237 		m = ip6cp->ip6c_m;
1238 		ip6 = ip6cp->ip6c_ip6;
1239 		off = ip6cp->ip6c_off;
1240 		sa6_src = ip6cp->ip6c_src;
1241 	} else {
1242 		m = NULL;
1243 		ip6 = NULL;
1244 		sa6_src = &sa6_any;
1245 	}
1246 
1247 	if (ip6) {
1248 		/*
1249 		 * XXX: We assume that when ip6 is non NULL,
1250 		 * M and OFF are valid.
1251 		 */
1252 
1253 		/* check if we can safely examine src and dst ports */
1254 		if (m->m_pkthdr.len < off + sizeof(th)) {
1255 			if (cmd == PRC_MSGSIZE)
1256 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1257 			return;
1258 		}
1259 
1260 		bzero(&th, sizeof(th));
1261 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1262 
1263 		if (cmd == PRC_MSGSIZE) {
1264 			int valid = 0;
1265 
1266 			/*
1267 			 * Check to see if we have a valid TCP connection
1268 			 * corresponding to the address in the ICMPv6 message
1269 			 * payload.
1270 			 */
1271 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1272 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1273 			    th.th_sport, 0))
1274 				valid++;
1275 
1276 			/*
1277 			 * Depending on the value of "valid" and routing table
1278 			 * size (mtudisc_{hi,lo}wat), we will:
1279 			 * - recalcurate the new MTU and create the
1280 			 *   corresponding routing entry, or
1281 			 * - ignore the MTU change notification.
1282 			 */
1283 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1284 
1285 			/*
1286 			 * no need to call in6_pcbnotify, it should have been
1287 			 * called via callback if necessary
1288 			 */
1289 			return;
1290 		}
1291 
1292 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1293 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1294 		if (nmatch == 0 && syn_cache_count &&
1295 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1296 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1297 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1298 			syn_cache_unreach((struct sockaddr *)sa6_src,
1299 					  sa, &th);
1300 	} else {
1301 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1302 		    0, cmd, NULL, notify);
1303 	}
1304 }
1305 #endif
1306 
1307 #ifdef INET
1308 /* assumes that ip header and tcp header are contiguous on mbuf */
1309 void *
1310 tcp_ctlinput(cmd, sa, v)
1311 	int cmd;
1312 	struct sockaddr *sa;
1313 	void *v;
1314 {
1315 	struct ip *ip = v;
1316 	struct tcphdr *th;
1317 	struct icmp *icp;
1318 	extern const int inetctlerrmap[];
1319 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1320 	int errno;
1321 	int nmatch;
1322 
1323 	if (sa->sa_family != AF_INET ||
1324 	    sa->sa_len != sizeof(struct sockaddr_in))
1325 		return NULL;
1326 	if ((unsigned)cmd >= PRC_NCMDS)
1327 		return NULL;
1328 	errno = inetctlerrmap[cmd];
1329 	if (cmd == PRC_QUENCH)
1330 		notify = tcp_quench;
1331 	else if (PRC_IS_REDIRECT(cmd))
1332 		notify = in_rtchange, ip = 0;
1333 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) {
1334 		/*
1335 		 * Check to see if we have a valid TCP connection
1336 		 * corresponding to the address in the ICMP message
1337 		 * payload.
1338 		 *
1339 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1340 		 */
1341 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1342 		if (in_pcblookup_connect(&tcbtable,
1343 					 ip->ip_dst, th->th_dport,
1344 					 ip->ip_src, th->th_sport) == NULL)
1345 			return NULL;
1346 
1347 		/*
1348 		 * Now that we've validated that we are actually communicating
1349 		 * with the host indicated in the ICMP message, locate the
1350 		 * ICMP header, recalculate the new MTU, and create the
1351 		 * corresponding routing entry.
1352 		 */
1353 		icp = (struct icmp *)((caddr_t)ip -
1354 		    offsetof(struct icmp, icmp_ip));
1355 		icmp_mtudisc(icp, ip->ip_dst);
1356 
1357 		return NULL;
1358 	} else if (cmd == PRC_HOSTDEAD)
1359 		ip = 0;
1360 	else if (errno == 0)
1361 		return NULL;
1362 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1363 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1364 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1365 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1366 		if (nmatch == 0 && syn_cache_count &&
1367 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1368 		    inetctlerrmap[cmd] == ENETUNREACH ||
1369 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1370 			struct sockaddr_in sin;
1371 			bzero(&sin, sizeof(sin));
1372 			sin.sin_len = sizeof(sin);
1373 			sin.sin_family = AF_INET;
1374 			sin.sin_port = th->th_sport;
1375 			sin.sin_addr = ip->ip_src;
1376 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1377 		}
1378 
1379 		/* XXX mapped address case */
1380 	} else
1381 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1382 		    notify);
1383 	return NULL;
1384 }
1385 
1386 /*
1387  * When a source quence is received, we are being notifed of congestion.
1388  * Close the congestion window down to the Loss Window (one segment).
1389  * We will gradually open it again as we proceed.
1390  */
1391 void
1392 tcp_quench(inp, errno)
1393 	struct inpcb *inp;
1394 	int errno;
1395 {
1396 	struct tcpcb *tp = intotcpcb(inp);
1397 
1398 	if (tp)
1399 		tp->snd_cwnd = tp->t_segsz;
1400 }
1401 #endif
1402 
1403 #ifdef INET6
1404 void
1405 tcp6_quench(in6p, errno)
1406 	struct in6pcb *in6p;
1407 	int errno;
1408 {
1409 	struct tcpcb *tp = in6totcpcb(in6p);
1410 
1411 	if (tp)
1412 		tp->snd_cwnd = tp->t_segsz;
1413 }
1414 #endif
1415 
1416 #ifdef INET
1417 /*
1418  * Path MTU Discovery handlers.
1419  */
1420 void
1421 tcp_mtudisc_callback(faddr)
1422 	struct in_addr faddr;
1423 {
1424 
1425 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1426 }
1427 
1428 /*
1429  * On receipt of path MTU corrections, flush old route and replace it
1430  * with the new one.  Retransmit all unacknowledged packets, to ensure
1431  * that all packets will be received.
1432  */
1433 void
1434 tcp_mtudisc(inp, errno)
1435 	struct inpcb *inp;
1436 	int errno;
1437 {
1438 	struct tcpcb *tp = intotcpcb(inp);
1439 	struct rtentry *rt = in_pcbrtentry(inp);
1440 
1441 	if (tp != 0) {
1442 		if (rt != 0) {
1443 			/*
1444 			 * If this was not a host route, remove and realloc.
1445 			 */
1446 			if ((rt->rt_flags & RTF_HOST) == 0) {
1447 				in_rtchange(inp, errno);
1448 				if ((rt = in_pcbrtentry(inp)) == 0)
1449 					return;
1450 			}
1451 
1452 			/*
1453 			 * Slow start out of the error condition.  We
1454 			 * use the MTU because we know it's smaller
1455 			 * than the previously transmitted segment.
1456 			 *
1457 			 * Note: This is more conservative than the
1458 			 * suggestion in draft-floyd-incr-init-win-03.
1459 			 */
1460 			if (rt->rt_rmx.rmx_mtu != 0)
1461 				tp->snd_cwnd =
1462 				    TCP_INITIAL_WINDOW(tcp_init_win,
1463 				    rt->rt_rmx.rmx_mtu);
1464 		}
1465 
1466 		/*
1467 		 * Resend unacknowledged packets.
1468 		 */
1469 		tp->snd_nxt = tp->snd_una;
1470 		tcp_output(tp);
1471 	}
1472 }
1473 #endif
1474 
1475 #ifdef INET6
1476 /*
1477  * Path MTU Discovery handlers.
1478  */
1479 void
1480 tcp6_mtudisc_callback(faddr)
1481 	struct in6_addr *faddr;
1482 {
1483 	struct sockaddr_in6 sin6;
1484 
1485 	bzero(&sin6, sizeof(sin6));
1486 	sin6.sin6_family = AF_INET6;
1487 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1488 	sin6.sin6_addr = *faddr;
1489 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1490 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1491 }
1492 
1493 void
1494 tcp6_mtudisc(in6p, errno)
1495 	struct in6pcb *in6p;
1496 	int errno;
1497 {
1498 	struct tcpcb *tp = in6totcpcb(in6p);
1499 	struct rtentry *rt = in6_pcbrtentry(in6p);
1500 
1501 	if (tp != 0) {
1502 		if (rt != 0) {
1503 			/*
1504 			 * If this was not a host route, remove and realloc.
1505 			 */
1506 			if ((rt->rt_flags & RTF_HOST) == 0) {
1507 				in6_rtchange(in6p, errno);
1508 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1509 					return;
1510 			}
1511 
1512 			/*
1513 			 * Slow start out of the error condition.  We
1514 			 * use the MTU because we know it's smaller
1515 			 * than the previously transmitted segment.
1516 			 *
1517 			 * Note: This is more conservative than the
1518 			 * suggestion in draft-floyd-incr-init-win-03.
1519 			 */
1520 			if (rt->rt_rmx.rmx_mtu != 0)
1521 				tp->snd_cwnd =
1522 				    TCP_INITIAL_WINDOW(tcp_init_win,
1523 				    rt->rt_rmx.rmx_mtu);
1524 		}
1525 
1526 		/*
1527 		 * Resend unacknowledged packets.
1528 		 */
1529 		tp->snd_nxt = tp->snd_una;
1530 		tcp_output(tp);
1531 	}
1532 }
1533 #endif /* INET6 */
1534 
1535 /*
1536  * Compute the MSS to advertise to the peer.  Called only during
1537  * the 3-way handshake.  If we are the server (peer initiated
1538  * connection), we are called with a pointer to the interface
1539  * on which the SYN packet arrived.  If we are the client (we
1540  * initiated connection), we are called with a pointer to the
1541  * interface out which this connection should go.
1542  *
1543  * NOTE: Do not subtract IP option/extension header size nor IPsec
1544  * header size from MSS advertisement.  MSS option must hold the maximum
1545  * segment size we can accept, so it must always be:
1546  *	 max(if mtu) - ip header - tcp header
1547  */
1548 u_long
1549 tcp_mss_to_advertise(ifp, af)
1550 	const struct ifnet *ifp;
1551 	int af;
1552 {
1553 	extern u_long in_maxmtu;
1554 	u_long mss = 0;
1555 	u_long hdrsiz;
1556 
1557 	/*
1558 	 * In order to avoid defeating path MTU discovery on the peer,
1559 	 * we advertise the max MTU of all attached networks as our MSS,
1560 	 * per RFC 1191, section 3.1.
1561 	 *
1562 	 * We provide the option to advertise just the MTU of
1563 	 * the interface on which we hope this connection will
1564 	 * be receiving.  If we are responding to a SYN, we
1565 	 * will have a pretty good idea about this, but when
1566 	 * initiating a connection there is a bit more doubt.
1567 	 *
1568 	 * We also need to ensure that loopback has a large enough
1569 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1570 	 */
1571 
1572 	if (ifp != NULL)
1573 		mss = ifp->if_mtu;
1574 
1575 	if (tcp_mss_ifmtu == 0)
1576 		switch (af) {
1577 		case AF_INET:
1578 			mss = max(in_maxmtu, mss);
1579 			break;
1580 #ifdef INET6
1581 		case AF_INET6:
1582 			mss = max(in6_maxmtu, mss);
1583 			break;
1584 #endif
1585 		}
1586 
1587 	switch (af) {
1588 	case AF_INET:
1589 		hdrsiz = sizeof(struct ip);
1590 		break;
1591 #ifdef INET6
1592 	case AF_INET6:
1593 		hdrsiz = sizeof(struct ip6_hdr);
1594 		break;
1595 #endif
1596 	default:
1597 		hdrsiz = 0;
1598 		break;
1599 	}
1600 	hdrsiz += sizeof(struct tcphdr);
1601 	if (mss > hdrsiz)
1602 		mss -= hdrsiz;
1603 
1604 	mss = max(tcp_mssdflt, mss);
1605 	return (mss);
1606 }
1607 
1608 /*
1609  * Set connection variables based on the peer's advertised MSS.
1610  * We are passed the TCPCB for the actual connection.  If we
1611  * are the server, we are called by the compressed state engine
1612  * when the 3-way handshake is complete.  If we are the client,
1613  * we are called when we receive the SYN,ACK from the server.
1614  *
1615  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1616  * before this routine is called!
1617  */
1618 void
1619 tcp_mss_from_peer(tp, offer)
1620 	struct tcpcb *tp;
1621 	int offer;
1622 {
1623 	struct socket *so;
1624 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1625 	struct rtentry *rt;
1626 #endif
1627 	u_long bufsize;
1628 	int mss;
1629 
1630 #ifdef DIAGNOSTIC
1631 	if (tp->t_inpcb && tp->t_in6pcb)
1632 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1633 #endif
1634 	so = NULL;
1635 	rt = NULL;
1636 #ifdef INET
1637 	if (tp->t_inpcb) {
1638 		so = tp->t_inpcb->inp_socket;
1639 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1640 		rt = in_pcbrtentry(tp->t_inpcb);
1641 #endif
1642 	}
1643 #endif
1644 #ifdef INET6
1645 	if (tp->t_in6pcb) {
1646 		so = tp->t_in6pcb->in6p_socket;
1647 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1648 		rt = in6_pcbrtentry(tp->t_in6pcb);
1649 #endif
1650 	}
1651 #endif
1652 
1653 	/*
1654 	 * As per RFC1122, use the default MSS value, unless they
1655 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1656 	 */
1657 	mss = tcp_mssdflt;
1658 	if (offer)
1659 		mss = offer;
1660 	mss = max(mss, 32);		/* sanity */
1661 	tp->t_peermss = mss;
1662 	mss -= tcp_optlen(tp);
1663 #ifdef INET
1664 	if (tp->t_inpcb)
1665 		mss -= ip_optlen(tp->t_inpcb);
1666 #endif
1667 #ifdef INET6
1668 	if (tp->t_in6pcb)
1669 		mss -= ip6_optlen(tp->t_in6pcb);
1670 #endif
1671 
1672 	/*
1673 	 * If there's a pipesize, change the socket buffer to that size.
1674 	 * Make the socket buffer an integral number of MSS units.  If
1675 	 * the MSS is larger than the socket buffer, artificially decrease
1676 	 * the MSS.
1677 	 */
1678 #ifdef RTV_SPIPE
1679 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1680 		bufsize = rt->rt_rmx.rmx_sendpipe;
1681 	else
1682 #endif
1683 		bufsize = so->so_snd.sb_hiwat;
1684 	if (bufsize < mss)
1685 		mss = bufsize;
1686 	else {
1687 		bufsize = roundup(bufsize, mss);
1688 		if (bufsize > sb_max)
1689 			bufsize = sb_max;
1690 		(void) sbreserve(&so->so_snd, bufsize);
1691 	}
1692 	tp->t_segsz = mss;
1693 
1694 #ifdef RTV_SSTHRESH
1695 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1696 		/*
1697 		 * There's some sort of gateway or interface buffer
1698 		 * limit on the path.  Use this to set the slow
1699 		 * start threshold, but set the threshold to no less
1700 		 * than 2 * MSS.
1701 		 */
1702 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1703 	}
1704 #endif
1705 }
1706 
1707 /*
1708  * Processing necessary when a TCP connection is established.
1709  */
1710 void
1711 tcp_established(tp)
1712 	struct tcpcb *tp;
1713 {
1714 	struct socket *so;
1715 #ifdef RTV_RPIPE
1716 	struct rtentry *rt;
1717 #endif
1718 	u_long bufsize;
1719 
1720 #ifdef DIAGNOSTIC
1721 	if (tp->t_inpcb && tp->t_in6pcb)
1722 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1723 #endif
1724 	so = NULL;
1725 	rt = NULL;
1726 #ifdef INET
1727 	if (tp->t_inpcb) {
1728 		so = tp->t_inpcb->inp_socket;
1729 #if defined(RTV_RPIPE)
1730 		rt = in_pcbrtentry(tp->t_inpcb);
1731 #endif
1732 	}
1733 #endif
1734 #ifdef INET6
1735 	if (tp->t_in6pcb) {
1736 		so = tp->t_in6pcb->in6p_socket;
1737 #if defined(RTV_RPIPE)
1738 		rt = in6_pcbrtentry(tp->t_in6pcb);
1739 #endif
1740 	}
1741 #endif
1742 
1743 	tp->t_state = TCPS_ESTABLISHED;
1744 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1745 
1746 #ifdef RTV_RPIPE
1747 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1748 		bufsize = rt->rt_rmx.rmx_recvpipe;
1749 	else
1750 #endif
1751 		bufsize = so->so_rcv.sb_hiwat;
1752 	if (bufsize > tp->t_ourmss) {
1753 		bufsize = roundup(bufsize, tp->t_ourmss);
1754 		if (bufsize > sb_max)
1755 			bufsize = sb_max;
1756 		(void) sbreserve(&so->so_rcv, bufsize);
1757 	}
1758 }
1759 
1760 /*
1761  * Check if there's an initial rtt or rttvar.  Convert from the
1762  * route-table units to scaled multiples of the slow timeout timer.
1763  * Called only during the 3-way handshake.
1764  */
1765 void
1766 tcp_rmx_rtt(tp)
1767 	struct tcpcb *tp;
1768 {
1769 #ifdef RTV_RTT
1770 	struct rtentry *rt = NULL;
1771 	int rtt;
1772 
1773 #ifdef DIAGNOSTIC
1774 	if (tp->t_inpcb && tp->t_in6pcb)
1775 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1776 #endif
1777 #ifdef INET
1778 	if (tp->t_inpcb)
1779 		rt = in_pcbrtentry(tp->t_inpcb);
1780 #endif
1781 #ifdef INET6
1782 	if (tp->t_in6pcb)
1783 		rt = in6_pcbrtentry(tp->t_in6pcb);
1784 #endif
1785 	if (rt == NULL)
1786 		return;
1787 
1788 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1789 		/*
1790 		 * XXX The lock bit for MTU indicates that the value
1791 		 * is also a minimum value; this is subject to time.
1792 		 */
1793 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1794 			TCPT_RANGESET(tp->t_rttmin,
1795 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1796 			    TCPTV_MIN, TCPTV_REXMTMAX);
1797 		tp->t_srtt = rtt /
1798 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1799 		if (rt->rt_rmx.rmx_rttvar) {
1800 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1801 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1802 				(TCP_RTTVAR_SHIFT + 2));
1803 		} else {
1804 			/* Default variation is +- 1 rtt */
1805 			tp->t_rttvar =
1806 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1807 		}
1808 		TCPT_RANGESET(tp->t_rxtcur,
1809 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1810 		    tp->t_rttmin, TCPTV_REXMTMAX);
1811 	}
1812 #endif
1813 }
1814 
1815 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1816 #if NRND > 0
1817 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1818 #endif
1819 
1820 /*
1821  * Get a new sequence value given a tcp control block
1822  */
1823 tcp_seq
1824 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1825 {
1826 
1827 #ifdef INET
1828 	if (tp->t_inpcb != NULL) {
1829 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1830 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1831 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1832 		    addin));
1833 	}
1834 #endif
1835 #ifdef INET6
1836 	if (tp->t_in6pcb != NULL) {
1837 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
1838 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
1839 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
1840 		    addin));
1841 	}
1842 #endif
1843 	/* Not possible. */
1844 	panic("tcp_new_iss");
1845 }
1846 
1847 /*
1848  * This routine actually generates a new TCP initial sequence number.
1849  */
1850 tcp_seq
1851 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
1852     size_t addrsz, tcp_seq addin)
1853 {
1854 	tcp_seq tcp_iss;
1855 
1856 #if NRND > 0
1857 	static int beenhere;
1858 
1859 	/*
1860 	 * If we haven't been here before, initialize our cryptographic
1861 	 * hash secret.
1862 	 */
1863 	if (beenhere == 0) {
1864 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
1865 		    RND_EXTRACT_ANY);
1866 		beenhere = 1;
1867 	}
1868 
1869 	if (tcp_do_rfc1948) {
1870 		MD5_CTX ctx;
1871 		u_int8_t hash[16];	/* XXX MD5 knowledge */
1872 
1873 		/*
1874 		 * Compute the base value of the ISS.  It is a hash
1875 		 * of (saddr, sport, daddr, dport, secret).
1876 		 */
1877 		MD5Init(&ctx);
1878 
1879 		MD5Update(&ctx, (u_char *) laddr, addrsz);
1880 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
1881 
1882 		MD5Update(&ctx, (u_char *) faddr, addrsz);
1883 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
1884 
1885 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
1886 
1887 		MD5Final(hash, &ctx);
1888 
1889 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
1890 
1891 		/*
1892 		 * Now increment our "timer", and add it in to
1893 		 * the computed value.
1894 		 *
1895 		 * XXX Use `addin'?
1896 		 * XXX TCP_ISSINCR too large to use?
1897 		 */
1898 		tcp_iss_seq += TCP_ISSINCR;
1899 #ifdef TCPISS_DEBUG
1900 		printf("ISS hash 0x%08x, ", tcp_iss);
1901 #endif
1902 		tcp_iss += tcp_iss_seq + addin;
1903 #ifdef TCPISS_DEBUG
1904 		printf("new ISS 0x%08x\n", tcp_iss);
1905 #endif
1906 	} else
1907 #endif /* NRND > 0 */
1908 	{
1909 		/*
1910 		 * Randomize.
1911 		 */
1912 #if NRND > 0
1913 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
1914 #else
1915 		tcp_iss = random();
1916 #endif
1917 
1918 		/*
1919 		 * If we were asked to add some amount to a known value,
1920 		 * we will take a random value obtained above, mask off
1921 		 * the upper bits, and add in the known value.  We also
1922 		 * add in a constant to ensure that we are at least a
1923 		 * certain distance from the original value.
1924 		 *
1925 		 * This is used when an old connection is in timed wait
1926 		 * and we have a new one coming in, for instance.
1927 		 */
1928 		if (addin != 0) {
1929 #ifdef TCPISS_DEBUG
1930 			printf("Random %08x, ", tcp_iss);
1931 #endif
1932 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1933 			tcp_iss += addin + TCP_ISSINCR;
1934 #ifdef TCPISS_DEBUG
1935 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
1936 #endif
1937 		} else {
1938 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1939 			tcp_iss += tcp_iss_seq;
1940 			tcp_iss_seq += TCP_ISSINCR;
1941 #ifdef TCPISS_DEBUG
1942 			printf("ISS %08x\n", tcp_iss);
1943 #endif
1944 		}
1945 	}
1946 
1947 	if (tcp_compat_42) {
1948 		/*
1949 		 * Limit it to the positive range for really old TCP
1950 		 * implementations.
1951 		 */
1952 		if (tcp_iss >= 0x80000000)
1953 			tcp_iss &= 0x7fffffff;		/* XXX */
1954 	}
1955 
1956 	return (tcp_iss);
1957 }
1958 
1959 #ifdef IPSEC
1960 /* compute ESP/AH header size for TCP, including outer IP header. */
1961 size_t
1962 ipsec4_hdrsiz_tcp(tp)
1963 	struct tcpcb *tp;
1964 {
1965 	struct inpcb *inp;
1966 	size_t hdrsiz;
1967 
1968 	/* XXX mapped addr case (tp->t_in6pcb) */
1969 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1970 		return 0;
1971 	switch (tp->t_family) {
1972 	case AF_INET:
1973 		/* XXX: should use currect direction. */
1974 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
1975 		break;
1976 	default:
1977 		hdrsiz = 0;
1978 		break;
1979 	}
1980 
1981 	return hdrsiz;
1982 }
1983 
1984 #ifdef INET6
1985 size_t
1986 ipsec6_hdrsiz_tcp(tp)
1987 	struct tcpcb *tp;
1988 {
1989 	struct in6pcb *in6p;
1990 	size_t hdrsiz;
1991 
1992 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
1993 		return 0;
1994 	switch (tp->t_family) {
1995 	case AF_INET6:
1996 		/* XXX: should use currect direction. */
1997 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
1998 		break;
1999 	case AF_INET:
2000 		/* mapped address case - tricky */
2001 	default:
2002 		hdrsiz = 0;
2003 		break;
2004 	}
2005 
2006 	return hdrsiz;
2007 }
2008 #endif
2009 #endif /*IPSEC*/
2010 
2011 /*
2012  * Determine the length of the TCP options for this connection.
2013  *
2014  * XXX:  What do we do for SACK, when we add that?  Just reserve
2015  *       all of the space?  Otherwise we can't exactly be incrementing
2016  *       cwnd by an amount that varies depending on the amount we last
2017  *       had to SACK!
2018  */
2019 
2020 u_int
2021 tcp_optlen(tp)
2022 	struct tcpcb *tp;
2023 {
2024 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2025 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2026 		return TCPOLEN_TSTAMP_APPA;
2027 	else
2028 		return 0;
2029 }
2030