xref: /netbsd-src/sys/netinet/tcp_subr.c (revision ed8e534c7e8d32417264bdd5a364dee751abead3)
1 /*	$NetBSD: tcp_subr.c,v 1.125 2002/04/27 01:47:58 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 <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.125 2002/04/27 01:47:58 thorpej 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 #ifdef TCP_OUTPUT_COUNTERS
231 #include <sys/device.h>
232 
233 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
234     NULL, "tcp", "output big header");
235 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
236     NULL, "tcp", "output copy small");
237 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
238     NULL, "tcp", "output copy big");
239 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "output reference big");
241 #endif /* TCP_OUTPUT_COUNTERS */
242 
243 /*
244  * Tcp initialization
245  */
246 void
247 tcp_init()
248 {
249 	int hlen;
250 
251 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
252 	    NULL);
253 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
254 #ifdef INET6
255 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
256 #endif
257 
258 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
259 #ifdef INET6
260 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
261 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
262 #endif
263 	if (max_protohdr < hlen)
264 		max_protohdr = hlen;
265 	if (max_linkhdr + hlen > MHLEN)
266 		panic("tcp_init");
267 
268 #ifdef INET
269 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
270 #endif
271 #ifdef INET6
272 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
273 #endif
274 
275 	/* Initialize timer state. */
276 	tcp_timer_init();
277 
278 	/* Initialize the compressed state engine. */
279 	syn_cache_init();
280 
281 #ifdef TCP_CSUM_COUNTERS
282 	evcnt_attach_static(&tcp_hwcsum_bad);
283 	evcnt_attach_static(&tcp_hwcsum_ok);
284 	evcnt_attach_static(&tcp_hwcsum_data);
285 	evcnt_attach_static(&tcp_swcsum);
286 #endif /* TCP_CSUM_COUNTERS */
287 
288 #ifdef TCP_OUTPUT_COUNTERS
289 	evcnt_attach_static(&tcp_output_bigheader);
290 	evcnt_attach_static(&tcp_output_copysmall);
291 	evcnt_attach_static(&tcp_output_copybig);
292 	evcnt_attach_static(&tcp_output_refbig);
293 #endif /* TCP_OUTPUT_COUNTERS */
294 }
295 
296 /*
297  * Create template to be used to send tcp packets on a connection.
298  * Call after host entry created, allocates an mbuf and fills
299  * in a skeletal tcp/ip header, minimizing the amount of work
300  * necessary when the connection is used.
301  */
302 struct mbuf *
303 tcp_template(tp)
304 	struct tcpcb *tp;
305 {
306 	struct inpcb *inp = tp->t_inpcb;
307 #ifdef INET6
308 	struct in6pcb *in6p = tp->t_in6pcb;
309 #endif
310 	struct tcphdr *n;
311 	struct mbuf *m;
312 	int hlen;
313 
314 	switch (tp->t_family) {
315 	case AF_INET:
316 		hlen = sizeof(struct ip);
317 		if (inp)
318 			break;
319 #ifdef INET6
320 		if (in6p) {
321 			/* mapped addr case */
322 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
323 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
324 				break;
325 		}
326 #endif
327 		return NULL;	/*EINVAL*/
328 #ifdef INET6
329 	case AF_INET6:
330 		hlen = sizeof(struct ip6_hdr);
331 		if (in6p) {
332 			/* more sainty check? */
333 			break;
334 		}
335 		return NULL;	/*EINVAL*/
336 #endif
337 	default:
338 		hlen = 0;	/*pacify gcc*/
339 		return NULL;	/*EAFNOSUPPORT*/
340 	}
341 #ifdef DIAGNOSTIC
342 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
343 		panic("mclbytes too small for t_template");
344 #endif
345 	m = tp->t_template;
346 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
347 		;
348 	else {
349 		if (m)
350 			m_freem(m);
351 		m = tp->t_template = NULL;
352 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
353 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
354 			MCLGET(m, M_DONTWAIT);
355 			if ((m->m_flags & M_EXT) == 0) {
356 				m_free(m);
357 				m = NULL;
358 			}
359 		}
360 		if (m == NULL)
361 			return NULL;
362 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
363 	}
364 
365 	bzero(mtod(m, caddr_t), m->m_len);
366 
367 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
368 
369 	switch (tp->t_family) {
370 	case AF_INET:
371 	    {
372 		struct ipovly *ipov;
373 		mtod(m, struct ip *)->ip_v = 4;
374 		ipov = mtod(m, struct ipovly *);
375 		ipov->ih_pr = IPPROTO_TCP;
376 		ipov->ih_len = htons(sizeof(struct tcphdr));
377 		if (inp) {
378 			ipov->ih_src = inp->inp_laddr;
379 			ipov->ih_dst = inp->inp_faddr;
380 		}
381 #ifdef INET6
382 		else if (in6p) {
383 			/* mapped addr case */
384 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
385 				sizeof(ipov->ih_src));
386 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
387 				sizeof(ipov->ih_dst));
388 		}
389 #endif
390 		/*
391 		 * Compute the pseudo-header portion of the checksum
392 		 * now.  We incrementally add in the TCP option and
393 		 * payload lengths later, and then compute the TCP
394 		 * checksum right before the packet is sent off onto
395 		 * the wire.
396 		 */
397 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
398 		    ipov->ih_dst.s_addr,
399 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
400 		break;
401 	    }
402 #ifdef INET6
403 	case AF_INET6:
404 	    {
405 		struct ip6_hdr *ip6;
406 		mtod(m, struct ip *)->ip_v = 6;
407 		ip6 = mtod(m, struct ip6_hdr *);
408 		ip6->ip6_nxt = IPPROTO_TCP;
409 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
410 		ip6->ip6_src = in6p->in6p_laddr;
411 		ip6->ip6_dst = in6p->in6p_faddr;
412 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
413 		if (ip6_auto_flowlabel) {
414 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
415 			ip6->ip6_flow |=
416 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
417 		}
418 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
419 		ip6->ip6_vfc |= IPV6_VERSION;
420 
421 		/*
422 		 * Compute the pseudo-header portion of the checksum
423 		 * now.  We incrementally add in the TCP option and
424 		 * payload lengths later, and then compute the TCP
425 		 * checksum right before the packet is sent off onto
426 		 * the wire.
427 		 */
428 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
429 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
430 		    htonl(IPPROTO_TCP));
431 		break;
432 	    }
433 #endif
434 	}
435 	if (inp) {
436 		n->th_sport = inp->inp_lport;
437 		n->th_dport = inp->inp_fport;
438 	}
439 #ifdef INET6
440 	else if (in6p) {
441 		n->th_sport = in6p->in6p_lport;
442 		n->th_dport = in6p->in6p_fport;
443 	}
444 #endif
445 	n->th_seq = 0;
446 	n->th_ack = 0;
447 	n->th_x2 = 0;
448 	n->th_off = 5;
449 	n->th_flags = 0;
450 	n->th_win = 0;
451 	n->th_urp = 0;
452 	return (m);
453 }
454 
455 /*
456  * Send a single message to the TCP at address specified by
457  * the given TCP/IP header.  If m == 0, then we make a copy
458  * of the tcpiphdr at ti and send directly to the addressed host.
459  * This is used to force keep alive messages out using the TCP
460  * template for a connection tp->t_template.  If flags are given
461  * then we send a message back to the TCP which originated the
462  * segment ti, and discard the mbuf containing it and any other
463  * attached mbufs.
464  *
465  * In any case the ack and sequence number of the transmitted
466  * segment are as specified by the parameters.
467  */
468 int
469 tcp_respond(tp, template, m, th0, ack, seq, flags)
470 	struct tcpcb *tp;
471 	struct mbuf *template;
472 	struct mbuf *m;
473 	struct tcphdr *th0;
474 	tcp_seq ack, seq;
475 	int flags;
476 {
477 	struct route *ro;
478 	int error, tlen, win = 0;
479 	int hlen;
480 	struct ip *ip;
481 #ifdef INET6
482 	struct ip6_hdr *ip6;
483 #endif
484 	int family;	/* family on packet, not inpcb/in6pcb! */
485 	struct tcphdr *th;
486 
487 	if (tp != NULL && (flags & TH_RST) == 0) {
488 #ifdef DIAGNOSTIC
489 		if (tp->t_inpcb && tp->t_in6pcb)
490 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
491 #endif
492 #ifdef INET
493 		if (tp->t_inpcb)
494 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
495 #endif
496 #ifdef INET6
497 		if (tp->t_in6pcb)
498 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
499 #endif
500 	}
501 
502 	ip = NULL;
503 #ifdef INET6
504 	ip6 = NULL;
505 #endif
506 	if (m == 0) {
507 		if (!template)
508 			return EINVAL;
509 
510 		/* get family information from template */
511 		switch (mtod(template, struct ip *)->ip_v) {
512 		case 4:
513 			family = AF_INET;
514 			hlen = sizeof(struct ip);
515 			break;
516 #ifdef INET6
517 		case 6:
518 			family = AF_INET6;
519 			hlen = sizeof(struct ip6_hdr);
520 			break;
521 #endif
522 		default:
523 			return EAFNOSUPPORT;
524 		}
525 
526 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
527 		if (m) {
528 			MCLGET(m, M_DONTWAIT);
529 			if ((m->m_flags & M_EXT) == 0) {
530 				m_free(m);
531 				m = NULL;
532 			}
533 		}
534 		if (m == NULL)
535 			return (ENOBUFS);
536 
537 		if (tcp_compat_42)
538 			tlen = 1;
539 		else
540 			tlen = 0;
541 
542 		m->m_data += max_linkhdr;
543 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
544 			template->m_len);
545 		switch (family) {
546 		case AF_INET:
547 			ip = mtod(m, struct ip *);
548 			th = (struct tcphdr *)(ip + 1);
549 			break;
550 #ifdef INET6
551 		case AF_INET6:
552 			ip6 = mtod(m, struct ip6_hdr *);
553 			th = (struct tcphdr *)(ip6 + 1);
554 			break;
555 #endif
556 #if 0
557 		default:
558 			/* noone will visit here */
559 			m_freem(m);
560 			return EAFNOSUPPORT;
561 #endif
562 		}
563 		flags = TH_ACK;
564 	} else {
565 
566 		if ((m->m_flags & M_PKTHDR) == 0) {
567 #if 0
568 			printf("non PKTHDR to tcp_respond\n");
569 #endif
570 			m_freem(m);
571 			return EINVAL;
572 		}
573 #ifdef DIAGNOSTIC
574 		if (!th0)
575 			panic("th0 == NULL in tcp_respond");
576 #endif
577 
578 		/* get family information from m */
579 		switch (mtod(m, struct ip *)->ip_v) {
580 		case 4:
581 			family = AF_INET;
582 			hlen = sizeof(struct ip);
583 			ip = mtod(m, struct ip *);
584 			break;
585 #ifdef INET6
586 		case 6:
587 			family = AF_INET6;
588 			hlen = sizeof(struct ip6_hdr);
589 			ip6 = mtod(m, struct ip6_hdr *);
590 			break;
591 #endif
592 		default:
593 			m_freem(m);
594 			return EAFNOSUPPORT;
595 		}
596 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
597 			tlen = sizeof(*th0);
598 		else
599 			tlen = th0->th_off << 2;
600 
601 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
602 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
603 			m->m_len = hlen + tlen;
604 			m_freem(m->m_next);
605 			m->m_next = NULL;
606 		} else {
607 			struct mbuf *n;
608 
609 #ifdef DIAGNOSTIC
610 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
611 				m_freem(m);
612 				return EMSGSIZE;
613 			}
614 #endif
615 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
616 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
617 				MCLGET(n, M_DONTWAIT);
618 				if ((n->m_flags & M_EXT) == 0) {
619 					m_freem(n);
620 					n = NULL;
621 				}
622 			}
623 			if (!n) {
624 				m_freem(m);
625 				return ENOBUFS;
626 			}
627 
628 			n->m_data += max_linkhdr;
629 			n->m_len = hlen + tlen;
630 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
631 			m_copyback(n, hlen, tlen, (caddr_t)th0);
632 
633 			m_freem(m);
634 			m = n;
635 			n = NULL;
636 		}
637 
638 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
639 		switch (family) {
640 		case AF_INET:
641 			ip = mtod(m, struct ip *);
642 			th = (struct tcphdr *)(ip + 1);
643 			ip->ip_p = IPPROTO_TCP;
644 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
645 			ip->ip_p = IPPROTO_TCP;
646 			break;
647 #ifdef INET6
648 		case AF_INET6:
649 			ip6 = mtod(m, struct ip6_hdr *);
650 			th = (struct tcphdr *)(ip6 + 1);
651 			ip6->ip6_nxt = IPPROTO_TCP;
652 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
653 			ip6->ip6_nxt = IPPROTO_TCP;
654 			break;
655 #endif
656 #if 0
657 		default:
658 			/* noone will visit here */
659 			m_freem(m);
660 			return EAFNOSUPPORT;
661 #endif
662 		}
663 		xchg(th->th_dport, th->th_sport, u_int16_t);
664 #undef xchg
665 		tlen = 0;	/*be friendly with the following code*/
666 	}
667 	th->th_seq = htonl(seq);
668 	th->th_ack = htonl(ack);
669 	th->th_x2 = 0;
670 	if ((flags & TH_SYN) == 0) {
671 		if (tp)
672 			win >>= tp->rcv_scale;
673 		if (win > TCP_MAXWIN)
674 			win = TCP_MAXWIN;
675 		th->th_win = htons((u_int16_t)win);
676 		th->th_off = sizeof (struct tcphdr) >> 2;
677 		tlen += sizeof(*th);
678 	} else
679 		tlen += th->th_off << 2;
680 	m->m_len = hlen + tlen;
681 	m->m_pkthdr.len = hlen + tlen;
682 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
683 	th->th_flags = flags;
684 	th->th_urp = 0;
685 
686 	switch (family) {
687 #ifdef INET
688 	case AF_INET:
689 	    {
690 		struct ipovly *ipov = (struct ipovly *)ip;
691 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
692 		ipov->ih_len = htons((u_int16_t)tlen);
693 
694 		th->th_sum = 0;
695 		th->th_sum = in_cksum(m, hlen + tlen);
696 		ip->ip_len = hlen + tlen;	/*will be flipped on output*/
697 		ip->ip_ttl = ip_defttl;
698 		break;
699 	    }
700 #endif
701 #ifdef INET6
702 	case AF_INET6:
703 	    {
704 		th->th_sum = 0;
705 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
706 				tlen);
707 		ip6->ip6_plen = ntohs(tlen);
708 		if (tp && tp->t_in6pcb) {
709 			struct ifnet *oifp;
710 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
711 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
712 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
713 		} else
714 			ip6->ip6_hlim = ip6_defhlim;
715 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
716 		if (ip6_auto_flowlabel) {
717 			ip6->ip6_flow |=
718 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
719 		}
720 		break;
721 	    }
722 #endif
723 	}
724 
725 #ifdef IPSEC
726 	(void)ipsec_setsocket(m, NULL);
727 #endif /*IPSEC*/
728 
729 	if (tp != NULL && tp->t_inpcb != NULL) {
730 		ro = &tp->t_inpcb->inp_route;
731 #ifdef IPSEC
732 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
733 			m_freem(m);
734 			return ENOBUFS;
735 		}
736 #endif
737 #ifdef DIAGNOSTIC
738 		if (family != AF_INET)
739 			panic("tcp_respond: address family mismatch");
740 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
741 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
742 			    ntohl(ip->ip_dst.s_addr),
743 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
744 		}
745 #endif
746 	}
747 #ifdef INET6
748 	else if (tp != NULL && tp->t_in6pcb != NULL) {
749 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
750 #ifdef IPSEC
751 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
752 			m_freem(m);
753 			return ENOBUFS;
754 		}
755 #endif
756 #ifdef DIAGNOSTIC
757 		if (family == AF_INET) {
758 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
759 				panic("tcp_respond: not mapped addr");
760 			if (bcmp(&ip->ip_dst,
761 					&tp->t_in6pcb->in6p_faddr.s6_addr32[3],
762 					sizeof(ip->ip_dst)) != 0) {
763 				panic("tcp_respond: ip_dst != in6p_faddr");
764 			}
765 		} else if (family == AF_INET6) {
766 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &tp->t_in6pcb->in6p_faddr))
767 				panic("tcp_respond: ip6_dst != in6p_faddr");
768 		} else
769 			panic("tcp_respond: address family mismatch");
770 #endif
771 	}
772 #endif
773 	else
774 		ro = NULL;
775 
776 	switch (family) {
777 #ifdef INET
778 	case AF_INET:
779 		error = ip_output(m, NULL, ro,
780 		    (ip_mtudisc ? IP_MTUDISC : 0),
781 		    NULL);
782 		break;
783 #endif
784 #ifdef INET6
785 	case AF_INET6:
786 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
787 			NULL);
788 		break;
789 #endif
790 	default:
791 		error = EAFNOSUPPORT;
792 		break;
793 	}
794 
795 	return (error);
796 }
797 
798 /*
799  * Create a new TCP control block, making an
800  * empty reassembly queue and hooking it to the argument
801  * protocol control block.
802  */
803 struct tcpcb *
804 tcp_newtcpcb(family, aux)
805 	int family;	/* selects inpcb, or in6pcb */
806 	void *aux;
807 {
808 	struct tcpcb *tp;
809 	int i;
810 
811 	switch (family) {
812 	case PF_INET:
813 		break;
814 #ifdef INET6
815 	case PF_INET6:
816 		break;
817 #endif
818 	default:
819 		return NULL;
820 	}
821 
822 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
823 	if (tp == NULL)
824 		return (NULL);
825 	bzero((caddr_t)tp, sizeof(struct tcpcb));
826 	LIST_INIT(&tp->segq);
827 	LIST_INIT(&tp->timeq);
828 	tp->t_family = family;		/* may be overridden later on */
829 	tp->t_peermss = tcp_mssdflt;
830 	tp->t_ourmss = tcp_mssdflt;
831 	tp->t_segsz = tcp_mssdflt;
832 	LIST_INIT(&tp->t_sc);
833 
834 	callout_init(&tp->t_delack_ch);
835 	for (i = 0; i < TCPT_NTIMERS; i++)
836 		TCP_TIMER_INIT(tp, i);
837 
838 	tp->t_flags = 0;
839 	if (tcp_do_rfc1323 && tcp_do_win_scale)
840 		tp->t_flags |= TF_REQ_SCALE;
841 	if (tcp_do_rfc1323 && tcp_do_timestamps)
842 		tp->t_flags |= TF_REQ_TSTMP;
843 	if (tcp_do_sack == 2)
844 		tp->t_flags |= TF_WILL_SACK;
845 	else if (tcp_do_sack == 1)
846 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
847 	tp->t_flags |= TF_CANT_TXSACK;
848 	switch (family) {
849 	case PF_INET:
850 		tp->t_inpcb = (struct inpcb *)aux;
851 		break;
852 #ifdef INET6
853 	case PF_INET6:
854 		tp->t_in6pcb = (struct in6pcb *)aux;
855 		break;
856 #endif
857 	}
858 	/*
859 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
860 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
861 	 * reasonable initial retransmit time.
862 	 */
863 	tp->t_srtt = TCPTV_SRTTBASE;
864 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
865 	tp->t_rttmin = TCPTV_MIN;
866 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
867 	    TCPTV_MIN, TCPTV_REXMTMAX);
868 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
869 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
870 	if (family == AF_INET) {
871 		struct inpcb *inp = (struct inpcb *)aux;
872 		inp->inp_ip.ip_ttl = ip_defttl;
873 		inp->inp_ppcb = (caddr_t)tp;
874 	}
875 #ifdef INET6
876 	else if (family == AF_INET6) {
877 		struct in6pcb *in6p = (struct in6pcb *)aux;
878 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
879 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
880 					       : NULL);
881 		in6p->in6p_ppcb = (caddr_t)tp;
882 	}
883 #endif
884 
885 	/*
886 	 * Initialize our timebase.  When we send timestamps, we take
887 	 * the delta from tcp_now -- this means each connection always
888 	 * gets a timebase of 0, which makes it, among other things,
889 	 * more difficult to determine how long a system has been up,
890 	 * and thus how many TCP sequence increments have occurred.
891 	 */
892 	tp->ts_timebase = tcp_now;
893 
894 	return (tp);
895 }
896 
897 /*
898  * Drop a TCP connection, reporting
899  * the specified error.  If connection is synchronized,
900  * then send a RST to peer.
901  */
902 struct tcpcb *
903 tcp_drop(tp, errno)
904 	struct tcpcb *tp;
905 	int errno;
906 {
907 	struct socket *so = NULL;
908 
909 #ifdef DIAGNOSTIC
910 	if (tp->t_inpcb && tp->t_in6pcb)
911 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
912 #endif
913 #ifdef INET
914 	if (tp->t_inpcb)
915 		so = tp->t_inpcb->inp_socket;
916 #endif
917 #ifdef INET6
918 	if (tp->t_in6pcb)
919 		so = tp->t_in6pcb->in6p_socket;
920 #endif
921 	if (!so)
922 		return NULL;
923 
924 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
925 		tp->t_state = TCPS_CLOSED;
926 		(void) tcp_output(tp);
927 		tcpstat.tcps_drops++;
928 	} else
929 		tcpstat.tcps_conndrops++;
930 	if (errno == ETIMEDOUT && tp->t_softerror)
931 		errno = tp->t_softerror;
932 	so->so_error = errno;
933 	return (tcp_close(tp));
934 }
935 
936 /*
937  * Close a TCP control block:
938  *	discard all space held by the tcp
939  *	discard internet protocol block
940  *	wake up any sleepers
941  */
942 struct tcpcb *
943 tcp_close(tp)
944 	struct tcpcb *tp;
945 {
946 	struct inpcb *inp;
947 #ifdef INET6
948 	struct in6pcb *in6p;
949 #endif
950 	struct socket *so;
951 #ifdef RTV_RTT
952 	struct rtentry *rt;
953 #endif
954 	struct route *ro;
955 
956 	inp = tp->t_inpcb;
957 #ifdef INET6
958 	in6p = tp->t_in6pcb;
959 #endif
960 	so = NULL;
961 	ro = NULL;
962 	if (inp) {
963 		so = inp->inp_socket;
964 		ro = &inp->inp_route;
965 	}
966 #ifdef INET6
967 	else if (in6p) {
968 		so = in6p->in6p_socket;
969 		ro = (struct route *)&in6p->in6p_route;
970 	}
971 #endif
972 
973 #ifdef RTV_RTT
974 	/*
975 	 * If we sent enough data to get some meaningful characteristics,
976 	 * save them in the routing entry.  'Enough' is arbitrarily
977 	 * defined as the sendpipesize (default 4K) * 16.  This would
978 	 * give us 16 rtt samples assuming we only get one sample per
979 	 * window (the usual case on a long haul net).  16 samples is
980 	 * enough for the srtt filter to converge to within 5% of the correct
981 	 * value; fewer samples and we could save a very bogus rtt.
982 	 *
983 	 * Don't update the default route's characteristics and don't
984 	 * update anything that the user "locked".
985 	 */
986 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
987 	    ro && (rt = ro->ro_rt) &&
988 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
989 		u_long i = 0;
990 
991 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
992 			i = tp->t_srtt *
993 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
994 			if (rt->rt_rmx.rmx_rtt && i)
995 				/*
996 				 * filter this update to half the old & half
997 				 * the new values, converting scale.
998 				 * See route.h and tcp_var.h for a
999 				 * description of the scaling constants.
1000 				 */
1001 				rt->rt_rmx.rmx_rtt =
1002 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1003 			else
1004 				rt->rt_rmx.rmx_rtt = i;
1005 		}
1006 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1007 			i = tp->t_rttvar *
1008 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1009 			if (rt->rt_rmx.rmx_rttvar && i)
1010 				rt->rt_rmx.rmx_rttvar =
1011 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1012 			else
1013 				rt->rt_rmx.rmx_rttvar = i;
1014 		}
1015 		/*
1016 		 * update the pipelimit (ssthresh) if it has been updated
1017 		 * already or if a pipesize was specified & the threshhold
1018 		 * got below half the pipesize.  I.e., wait for bad news
1019 		 * before we start updating, then update on both good
1020 		 * and bad news.
1021 		 */
1022 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1023 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1024 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1025 			/*
1026 			 * convert the limit from user data bytes to
1027 			 * packets then to packet data bytes.
1028 			 */
1029 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1030 			if (i < 2)
1031 				i = 2;
1032 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1033 			if (rt->rt_rmx.rmx_ssthresh)
1034 				rt->rt_rmx.rmx_ssthresh =
1035 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1036 			else
1037 				rt->rt_rmx.rmx_ssthresh = i;
1038 		}
1039 	}
1040 #endif /* RTV_RTT */
1041 	/* free the reassembly queue, if any */
1042 	TCP_REASS_LOCK(tp);
1043 	(void) tcp_freeq(tp);
1044 	TCP_REASS_UNLOCK(tp);
1045 
1046 	tcp_canceltimers(tp);
1047 	TCP_CLEAR_DELACK(tp);
1048 	syn_cache_cleanup(tp);
1049 
1050 	if (tp->t_template) {
1051 		m_free(tp->t_template);
1052 		tp->t_template = NULL;
1053 	}
1054 	pool_put(&tcpcb_pool, tp);
1055 	if (inp) {
1056 		inp->inp_ppcb = 0;
1057 		soisdisconnected(so);
1058 		in_pcbdetach(inp);
1059 	}
1060 #ifdef INET6
1061 	else if (in6p) {
1062 		in6p->in6p_ppcb = 0;
1063 		soisdisconnected(so);
1064 		in6_pcbdetach(in6p);
1065 	}
1066 #endif
1067 	tcpstat.tcps_closed++;
1068 	return ((struct tcpcb *)0);
1069 }
1070 
1071 int
1072 tcp_freeq(tp)
1073 	struct tcpcb *tp;
1074 {
1075 	struct ipqent *qe;
1076 	int rv = 0;
1077 #ifdef TCPREASS_DEBUG
1078 	int i = 0;
1079 #endif
1080 
1081 	TCP_REASS_LOCK_CHECK(tp);
1082 
1083 	while ((qe = LIST_FIRST(&tp->segq)) != NULL) {
1084 #ifdef TCPREASS_DEBUG
1085 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1086 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1087 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1088 #endif
1089 		LIST_REMOVE(qe, ipqe_q);
1090 		LIST_REMOVE(qe, ipqe_timeq);
1091 		m_freem(qe->ipqe_m);
1092 		pool_put(&ipqent_pool, qe);
1093 		rv = 1;
1094 	}
1095 	return (rv);
1096 }
1097 
1098 /*
1099  * Protocol drain routine.  Called when memory is in short supply.
1100  */
1101 void
1102 tcp_drain()
1103 {
1104 	struct inpcb *inp;
1105 	struct tcpcb *tp;
1106 
1107 	/*
1108 	 * Free the sequence queue of all TCP connections.
1109 	 */
1110 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1111 	if (inp)						/* XXX */
1112 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1113 		if ((tp = intotcpcb(inp)) != NULL) {
1114 			/*
1115 			 * We may be called from a device's interrupt
1116 			 * context.  If the tcpcb is already busy,
1117 			 * just bail out now.
1118 			 */
1119 			if (tcp_reass_lock_try(tp) == 0)
1120 				continue;
1121 			if (tcp_freeq(tp))
1122 				tcpstat.tcps_connsdrained++;
1123 			TCP_REASS_UNLOCK(tp);
1124 		}
1125 	}
1126 }
1127 
1128 #ifdef INET6
1129 void
1130 tcp6_drain()
1131 {
1132 	struct in6pcb *in6p;
1133 	struct tcpcb *tp;
1134 	struct in6pcb *head = &tcb6;
1135 
1136 	/*
1137 	 * Free the sequence queue of all TCP connections.
1138 	 */
1139 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1140 		if ((tp = in6totcpcb(in6p)) != NULL) {
1141 			/*
1142 			 * We may be called from a device's interrupt
1143 			 * context.  If the tcpcb is already busy,
1144 			 * just bail out now.
1145 			 */
1146 			if (tcp_reass_lock_try(tp) == 0)
1147 				continue;
1148 			if (tcp_freeq(tp))
1149 				tcpstat.tcps_connsdrained++;
1150 			TCP_REASS_UNLOCK(tp);
1151 		}
1152 	}
1153 }
1154 #endif
1155 
1156 /*
1157  * Notify a tcp user of an asynchronous error;
1158  * store error as soft error, but wake up user
1159  * (for now, won't do anything until can select for soft error).
1160  */
1161 void
1162 tcp_notify(inp, error)
1163 	struct inpcb *inp;
1164 	int error;
1165 {
1166 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1167 	struct socket *so = inp->inp_socket;
1168 
1169 	/*
1170 	 * Ignore some errors if we are hooked up.
1171 	 * If connection hasn't completed, has retransmitted several times,
1172 	 * and receives a second error, give up now.  This is better
1173 	 * than waiting a long time to establish a connection that
1174 	 * can never complete.
1175 	 */
1176 	if (tp->t_state == TCPS_ESTABLISHED &&
1177 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1178 	      error == EHOSTDOWN)) {
1179 		return;
1180 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1181 	    tp->t_rxtshift > 3 && tp->t_softerror)
1182 		so->so_error = error;
1183 	else
1184 		tp->t_softerror = error;
1185 	wakeup((caddr_t) &so->so_timeo);
1186 	sorwakeup(so);
1187 	sowwakeup(so);
1188 }
1189 
1190 #ifdef INET6
1191 void
1192 tcp6_notify(in6p, error)
1193 	struct in6pcb *in6p;
1194 	int error;
1195 {
1196 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1197 	struct socket *so = in6p->in6p_socket;
1198 
1199 	/*
1200 	 * Ignore some errors if we are hooked up.
1201 	 * If connection hasn't completed, has retransmitted several times,
1202 	 * and receives a second error, give up now.  This is better
1203 	 * than waiting a long time to establish a connection that
1204 	 * can never complete.
1205 	 */
1206 	if (tp->t_state == TCPS_ESTABLISHED &&
1207 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1208 	      error == EHOSTDOWN)) {
1209 		return;
1210 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1211 	    tp->t_rxtshift > 3 && tp->t_softerror)
1212 		so->so_error = error;
1213 	else
1214 		tp->t_softerror = error;
1215 	wakeup((caddr_t) &so->so_timeo);
1216 	sorwakeup(so);
1217 	sowwakeup(so);
1218 }
1219 #endif
1220 
1221 #ifdef INET6
1222 void
1223 tcp6_ctlinput(cmd, sa, d)
1224 	int cmd;
1225 	struct sockaddr *sa;
1226 	void *d;
1227 {
1228 	struct tcphdr th;
1229 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1230 	int nmatch;
1231 	struct ip6_hdr *ip6;
1232 	const struct sockaddr_in6 *sa6_src = NULL;
1233 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1234 	struct mbuf *m;
1235 	int off;
1236 
1237 	if (sa->sa_family != AF_INET6 ||
1238 	    sa->sa_len != sizeof(struct sockaddr_in6))
1239 		return;
1240 	if ((unsigned)cmd >= PRC_NCMDS)
1241 		return;
1242 	else if (cmd == PRC_QUENCH) {
1243 		/* XXX there's no PRC_QUENCH in IPv6 */
1244 		notify = tcp6_quench;
1245 	} else if (PRC_IS_REDIRECT(cmd))
1246 		notify = in6_rtchange, d = NULL;
1247 	else if (cmd == PRC_MSGSIZE)
1248 		; /* special code is present, see below */
1249 	else if (cmd == PRC_HOSTDEAD)
1250 		d = NULL;
1251 	else if (inet6ctlerrmap[cmd] == 0)
1252 		return;
1253 
1254 	/* if the parameter is from icmp6, decode it. */
1255 	if (d != NULL) {
1256 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1257 		m = ip6cp->ip6c_m;
1258 		ip6 = ip6cp->ip6c_ip6;
1259 		off = ip6cp->ip6c_off;
1260 		sa6_src = ip6cp->ip6c_src;
1261 	} else {
1262 		m = NULL;
1263 		ip6 = NULL;
1264 		sa6_src = &sa6_any;
1265 	}
1266 
1267 	if (ip6) {
1268 		/*
1269 		 * XXX: We assume that when ip6 is non NULL,
1270 		 * M and OFF are valid.
1271 		 */
1272 
1273 		/* check if we can safely examine src and dst ports */
1274 		if (m->m_pkthdr.len < off + sizeof(th)) {
1275 			if (cmd == PRC_MSGSIZE)
1276 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1277 			return;
1278 		}
1279 
1280 		bzero(&th, sizeof(th));
1281 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1282 
1283 		if (cmd == PRC_MSGSIZE) {
1284 			int valid = 0;
1285 
1286 			/*
1287 			 * Check to see if we have a valid TCP connection
1288 			 * corresponding to the address in the ICMPv6 message
1289 			 * payload.
1290 			 */
1291 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1292 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1293 			    th.th_sport, 0))
1294 				valid++;
1295 
1296 			/*
1297 			 * Depending on the value of "valid" and routing table
1298 			 * size (mtudisc_{hi,lo}wat), we will:
1299 			 * - recalcurate the new MTU and create the
1300 			 *   corresponding routing entry, or
1301 			 * - ignore the MTU change notification.
1302 			 */
1303 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1304 
1305 			/*
1306 			 * no need to call in6_pcbnotify, it should have been
1307 			 * called via callback if necessary
1308 			 */
1309 			return;
1310 		}
1311 
1312 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1313 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1314 		if (nmatch == 0 && syn_cache_count &&
1315 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1316 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1317 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1318 			syn_cache_unreach((struct sockaddr *)sa6_src,
1319 					  sa, &th);
1320 	} else {
1321 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1322 		    0, cmd, NULL, notify);
1323 	}
1324 }
1325 #endif
1326 
1327 #ifdef INET
1328 /* assumes that ip header and tcp header are contiguous on mbuf */
1329 void *
1330 tcp_ctlinput(cmd, sa, v)
1331 	int cmd;
1332 	struct sockaddr *sa;
1333 	void *v;
1334 {
1335 	struct ip *ip = v;
1336 	struct tcphdr *th;
1337 	struct icmp *icp;
1338 	extern const int inetctlerrmap[];
1339 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1340 	int errno;
1341 	int nmatch;
1342 
1343 	if (sa->sa_family != AF_INET ||
1344 	    sa->sa_len != sizeof(struct sockaddr_in))
1345 		return NULL;
1346 	if ((unsigned)cmd >= PRC_NCMDS)
1347 		return NULL;
1348 	errno = inetctlerrmap[cmd];
1349 	if (cmd == PRC_QUENCH)
1350 		notify = tcp_quench;
1351 	else if (PRC_IS_REDIRECT(cmd))
1352 		notify = in_rtchange, ip = 0;
1353 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip && ip->ip_v == 4) {
1354 		/*
1355 		 * Check to see if we have a valid TCP connection
1356 		 * corresponding to the address in the ICMP message
1357 		 * payload.
1358 		 *
1359 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1360 		 */
1361 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1362 		if (in_pcblookup_connect(&tcbtable,
1363 					 ip->ip_dst, th->th_dport,
1364 					 ip->ip_src, th->th_sport) == NULL)
1365 			return NULL;
1366 
1367 		/*
1368 		 * Now that we've validated that we are actually communicating
1369 		 * with the host indicated in the ICMP message, locate the
1370 		 * ICMP header, recalculate the new MTU, and create the
1371 		 * corresponding routing entry.
1372 		 */
1373 		icp = (struct icmp *)((caddr_t)ip -
1374 		    offsetof(struct icmp, icmp_ip));
1375 		icmp_mtudisc(icp, ip->ip_dst);
1376 
1377 		return NULL;
1378 	} else if (cmd == PRC_HOSTDEAD)
1379 		ip = 0;
1380 	else if (errno == 0)
1381 		return NULL;
1382 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1383 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1384 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1385 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1386 		if (nmatch == 0 && syn_cache_count &&
1387 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1388 		    inetctlerrmap[cmd] == ENETUNREACH ||
1389 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1390 			struct sockaddr_in sin;
1391 			bzero(&sin, sizeof(sin));
1392 			sin.sin_len = sizeof(sin);
1393 			sin.sin_family = AF_INET;
1394 			sin.sin_port = th->th_sport;
1395 			sin.sin_addr = ip->ip_src;
1396 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1397 		}
1398 
1399 		/* XXX mapped address case */
1400 	} else
1401 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1402 		    notify);
1403 	return NULL;
1404 }
1405 
1406 /*
1407  * When a source quence is received, we are being notifed of congestion.
1408  * Close the congestion window down to the Loss Window (one segment).
1409  * We will gradually open it again as we proceed.
1410  */
1411 void
1412 tcp_quench(inp, errno)
1413 	struct inpcb *inp;
1414 	int errno;
1415 {
1416 	struct tcpcb *tp = intotcpcb(inp);
1417 
1418 	if (tp)
1419 		tp->snd_cwnd = tp->t_segsz;
1420 }
1421 #endif
1422 
1423 #ifdef INET6
1424 void
1425 tcp6_quench(in6p, errno)
1426 	struct in6pcb *in6p;
1427 	int errno;
1428 {
1429 	struct tcpcb *tp = in6totcpcb(in6p);
1430 
1431 	if (tp)
1432 		tp->snd_cwnd = tp->t_segsz;
1433 }
1434 #endif
1435 
1436 #ifdef INET
1437 /*
1438  * Path MTU Discovery handlers.
1439  */
1440 void
1441 tcp_mtudisc_callback(faddr)
1442 	struct in_addr faddr;
1443 {
1444 
1445 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1446 }
1447 
1448 /*
1449  * On receipt of path MTU corrections, flush old route and replace it
1450  * with the new one.  Retransmit all unacknowledged packets, to ensure
1451  * that all packets will be received.
1452  */
1453 void
1454 tcp_mtudisc(inp, errno)
1455 	struct inpcb *inp;
1456 	int errno;
1457 {
1458 	struct tcpcb *tp = intotcpcb(inp);
1459 	struct rtentry *rt = in_pcbrtentry(inp);
1460 
1461 	if (tp != 0) {
1462 		if (rt != 0) {
1463 			/*
1464 			 * If this was not a host route, remove and realloc.
1465 			 */
1466 			if ((rt->rt_flags & RTF_HOST) == 0) {
1467 				in_rtchange(inp, errno);
1468 				if ((rt = in_pcbrtentry(inp)) == 0)
1469 					return;
1470 			}
1471 
1472 			/*
1473 			 * Slow start out of the error condition.  We
1474 			 * use the MTU because we know it's smaller
1475 			 * than the previously transmitted segment.
1476 			 *
1477 			 * Note: This is more conservative than the
1478 			 * suggestion in draft-floyd-incr-init-win-03.
1479 			 */
1480 			if (rt->rt_rmx.rmx_mtu != 0)
1481 				tp->snd_cwnd =
1482 				    TCP_INITIAL_WINDOW(tcp_init_win,
1483 				    rt->rt_rmx.rmx_mtu);
1484 		}
1485 
1486 		/*
1487 		 * Resend unacknowledged packets.
1488 		 */
1489 		tp->snd_nxt = tp->snd_una;
1490 		tcp_output(tp);
1491 	}
1492 }
1493 #endif
1494 
1495 #ifdef INET6
1496 /*
1497  * Path MTU Discovery handlers.
1498  */
1499 void
1500 tcp6_mtudisc_callback(faddr)
1501 	struct in6_addr *faddr;
1502 {
1503 	struct sockaddr_in6 sin6;
1504 
1505 	bzero(&sin6, sizeof(sin6));
1506 	sin6.sin6_family = AF_INET6;
1507 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1508 	sin6.sin6_addr = *faddr;
1509 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1510 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1511 }
1512 
1513 void
1514 tcp6_mtudisc(in6p, errno)
1515 	struct in6pcb *in6p;
1516 	int errno;
1517 {
1518 	struct tcpcb *tp = in6totcpcb(in6p);
1519 	struct rtentry *rt = in6_pcbrtentry(in6p);
1520 
1521 	if (tp != 0) {
1522 		if (rt != 0) {
1523 			/*
1524 			 * If this was not a host route, remove and realloc.
1525 			 */
1526 			if ((rt->rt_flags & RTF_HOST) == 0) {
1527 				in6_rtchange(in6p, errno);
1528 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1529 					return;
1530 			}
1531 
1532 			/*
1533 			 * Slow start out of the error condition.  We
1534 			 * use the MTU because we know it's smaller
1535 			 * than the previously transmitted segment.
1536 			 *
1537 			 * Note: This is more conservative than the
1538 			 * suggestion in draft-floyd-incr-init-win-03.
1539 			 */
1540 			if (rt->rt_rmx.rmx_mtu != 0)
1541 				tp->snd_cwnd =
1542 				    TCP_INITIAL_WINDOW(tcp_init_win,
1543 				    rt->rt_rmx.rmx_mtu);
1544 		}
1545 
1546 		/*
1547 		 * Resend unacknowledged packets.
1548 		 */
1549 		tp->snd_nxt = tp->snd_una;
1550 		tcp_output(tp);
1551 	}
1552 }
1553 #endif /* INET6 */
1554 
1555 /*
1556  * Compute the MSS to advertise to the peer.  Called only during
1557  * the 3-way handshake.  If we are the server (peer initiated
1558  * connection), we are called with a pointer to the interface
1559  * on which the SYN packet arrived.  If we are the client (we
1560  * initiated connection), we are called with a pointer to the
1561  * interface out which this connection should go.
1562  *
1563  * NOTE: Do not subtract IP option/extension header size nor IPsec
1564  * header size from MSS advertisement.  MSS option must hold the maximum
1565  * segment size we can accept, so it must always be:
1566  *	 max(if mtu) - ip header - tcp header
1567  */
1568 u_long
1569 tcp_mss_to_advertise(ifp, af)
1570 	const struct ifnet *ifp;
1571 	int af;
1572 {
1573 	extern u_long in_maxmtu;
1574 	u_long mss = 0;
1575 	u_long hdrsiz;
1576 
1577 	/*
1578 	 * In order to avoid defeating path MTU discovery on the peer,
1579 	 * we advertise the max MTU of all attached networks as our MSS,
1580 	 * per RFC 1191, section 3.1.
1581 	 *
1582 	 * We provide the option to advertise just the MTU of
1583 	 * the interface on which we hope this connection will
1584 	 * be receiving.  If we are responding to a SYN, we
1585 	 * will have a pretty good idea about this, but when
1586 	 * initiating a connection there is a bit more doubt.
1587 	 *
1588 	 * We also need to ensure that loopback has a large enough
1589 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1590 	 */
1591 
1592 	if (ifp != NULL)
1593 		mss = ifp->if_mtu;
1594 
1595 	if (tcp_mss_ifmtu == 0)
1596 		switch (af) {
1597 		case AF_INET:
1598 			mss = max(in_maxmtu, mss);
1599 			break;
1600 #ifdef INET6
1601 		case AF_INET6:
1602 			mss = max(in6_maxmtu, mss);
1603 			break;
1604 #endif
1605 		}
1606 
1607 	switch (af) {
1608 	case AF_INET:
1609 		hdrsiz = sizeof(struct ip);
1610 		break;
1611 #ifdef INET6
1612 	case AF_INET6:
1613 		hdrsiz = sizeof(struct ip6_hdr);
1614 		break;
1615 #endif
1616 	default:
1617 		hdrsiz = 0;
1618 		break;
1619 	}
1620 	hdrsiz += sizeof(struct tcphdr);
1621 	if (mss > hdrsiz)
1622 		mss -= hdrsiz;
1623 
1624 	mss = max(tcp_mssdflt, mss);
1625 	return (mss);
1626 }
1627 
1628 /*
1629  * Set connection variables based on the peer's advertised MSS.
1630  * We are passed the TCPCB for the actual connection.  If we
1631  * are the server, we are called by the compressed state engine
1632  * when the 3-way handshake is complete.  If we are the client,
1633  * we are called when we receive the SYN,ACK from the server.
1634  *
1635  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1636  * before this routine is called!
1637  */
1638 void
1639 tcp_mss_from_peer(tp, offer)
1640 	struct tcpcb *tp;
1641 	int offer;
1642 {
1643 	struct socket *so;
1644 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1645 	struct rtentry *rt;
1646 #endif
1647 	u_long bufsize;
1648 	int mss;
1649 
1650 #ifdef DIAGNOSTIC
1651 	if (tp->t_inpcb && tp->t_in6pcb)
1652 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1653 #endif
1654 	so = NULL;
1655 	rt = NULL;
1656 #ifdef INET
1657 	if (tp->t_inpcb) {
1658 		so = tp->t_inpcb->inp_socket;
1659 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1660 		rt = in_pcbrtentry(tp->t_inpcb);
1661 #endif
1662 	}
1663 #endif
1664 #ifdef INET6
1665 	if (tp->t_in6pcb) {
1666 		so = tp->t_in6pcb->in6p_socket;
1667 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1668 		rt = in6_pcbrtentry(tp->t_in6pcb);
1669 #endif
1670 	}
1671 #endif
1672 
1673 	/*
1674 	 * As per RFC1122, use the default MSS value, unless they
1675 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1676 	 */
1677 	mss = tcp_mssdflt;
1678 	if (offer)
1679 		mss = offer;
1680 	mss = max(mss, 32);		/* sanity */
1681 	tp->t_peermss = mss;
1682 	mss -= tcp_optlen(tp);
1683 #ifdef INET
1684 	if (tp->t_inpcb)
1685 		mss -= ip_optlen(tp->t_inpcb);
1686 #endif
1687 #ifdef INET6
1688 	if (tp->t_in6pcb)
1689 		mss -= ip6_optlen(tp->t_in6pcb);
1690 #endif
1691 
1692 	/*
1693 	 * If there's a pipesize, change the socket buffer to that size.
1694 	 * Make the socket buffer an integral number of MSS units.  If
1695 	 * the MSS is larger than the socket buffer, artificially decrease
1696 	 * the MSS.
1697 	 */
1698 #ifdef RTV_SPIPE
1699 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1700 		bufsize = rt->rt_rmx.rmx_sendpipe;
1701 	else
1702 #endif
1703 		bufsize = so->so_snd.sb_hiwat;
1704 	if (bufsize < mss)
1705 		mss = bufsize;
1706 	else {
1707 		bufsize = roundup(bufsize, mss);
1708 		if (bufsize > sb_max)
1709 			bufsize = sb_max;
1710 		(void) sbreserve(&so->so_snd, bufsize);
1711 	}
1712 	tp->t_segsz = mss;
1713 
1714 #ifdef RTV_SSTHRESH
1715 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1716 		/*
1717 		 * There's some sort of gateway or interface buffer
1718 		 * limit on the path.  Use this to set the slow
1719 		 * start threshold, but set the threshold to no less
1720 		 * than 2 * MSS.
1721 		 */
1722 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1723 	}
1724 #endif
1725 }
1726 
1727 /*
1728  * Processing necessary when a TCP connection is established.
1729  */
1730 void
1731 tcp_established(tp)
1732 	struct tcpcb *tp;
1733 {
1734 	struct socket *so;
1735 #ifdef RTV_RPIPE
1736 	struct rtentry *rt;
1737 #endif
1738 	u_long bufsize;
1739 
1740 #ifdef DIAGNOSTIC
1741 	if (tp->t_inpcb && tp->t_in6pcb)
1742 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1743 #endif
1744 	so = NULL;
1745 	rt = NULL;
1746 #ifdef INET
1747 	if (tp->t_inpcb) {
1748 		so = tp->t_inpcb->inp_socket;
1749 #if defined(RTV_RPIPE)
1750 		rt = in_pcbrtentry(tp->t_inpcb);
1751 #endif
1752 	}
1753 #endif
1754 #ifdef INET6
1755 	if (tp->t_in6pcb) {
1756 		so = tp->t_in6pcb->in6p_socket;
1757 #if defined(RTV_RPIPE)
1758 		rt = in6_pcbrtentry(tp->t_in6pcb);
1759 #endif
1760 	}
1761 #endif
1762 
1763 	tp->t_state = TCPS_ESTABLISHED;
1764 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1765 
1766 #ifdef RTV_RPIPE
1767 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1768 		bufsize = rt->rt_rmx.rmx_recvpipe;
1769 	else
1770 #endif
1771 		bufsize = so->so_rcv.sb_hiwat;
1772 	if (bufsize > tp->t_ourmss) {
1773 		bufsize = roundup(bufsize, tp->t_ourmss);
1774 		if (bufsize > sb_max)
1775 			bufsize = sb_max;
1776 		(void) sbreserve(&so->so_rcv, bufsize);
1777 	}
1778 }
1779 
1780 /*
1781  * Check if there's an initial rtt or rttvar.  Convert from the
1782  * route-table units to scaled multiples of the slow timeout timer.
1783  * Called only during the 3-way handshake.
1784  */
1785 void
1786 tcp_rmx_rtt(tp)
1787 	struct tcpcb *tp;
1788 {
1789 #ifdef RTV_RTT
1790 	struct rtentry *rt = NULL;
1791 	int rtt;
1792 
1793 #ifdef DIAGNOSTIC
1794 	if (tp->t_inpcb && tp->t_in6pcb)
1795 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1796 #endif
1797 #ifdef INET
1798 	if (tp->t_inpcb)
1799 		rt = in_pcbrtentry(tp->t_inpcb);
1800 #endif
1801 #ifdef INET6
1802 	if (tp->t_in6pcb)
1803 		rt = in6_pcbrtentry(tp->t_in6pcb);
1804 #endif
1805 	if (rt == NULL)
1806 		return;
1807 
1808 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1809 		/*
1810 		 * XXX The lock bit for MTU indicates that the value
1811 		 * is also a minimum value; this is subject to time.
1812 		 */
1813 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1814 			TCPT_RANGESET(tp->t_rttmin,
1815 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1816 			    TCPTV_MIN, TCPTV_REXMTMAX);
1817 		tp->t_srtt = rtt /
1818 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1819 		if (rt->rt_rmx.rmx_rttvar) {
1820 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1821 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1822 				(TCP_RTTVAR_SHIFT + 2));
1823 		} else {
1824 			/* Default variation is +- 1 rtt */
1825 			tp->t_rttvar =
1826 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1827 		}
1828 		TCPT_RANGESET(tp->t_rxtcur,
1829 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1830 		    tp->t_rttmin, TCPTV_REXMTMAX);
1831 	}
1832 #endif
1833 }
1834 
1835 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1836 #if NRND > 0
1837 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1838 #endif
1839 
1840 /*
1841  * Get a new sequence value given a tcp control block
1842  */
1843 tcp_seq
1844 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1845 {
1846 
1847 #ifdef INET
1848 	if (tp->t_inpcb != NULL) {
1849 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1850 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1851 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1852 		    addin));
1853 	}
1854 #endif
1855 #ifdef INET6
1856 	if (tp->t_in6pcb != NULL) {
1857 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
1858 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
1859 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
1860 		    addin));
1861 	}
1862 #endif
1863 	/* Not possible. */
1864 	panic("tcp_new_iss");
1865 }
1866 
1867 /*
1868  * This routine actually generates a new TCP initial sequence number.
1869  */
1870 tcp_seq
1871 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
1872     size_t addrsz, tcp_seq addin)
1873 {
1874 	tcp_seq tcp_iss;
1875 
1876 #if NRND > 0
1877 	static int beenhere;
1878 
1879 	/*
1880 	 * If we haven't been here before, initialize our cryptographic
1881 	 * hash secret.
1882 	 */
1883 	if (beenhere == 0) {
1884 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
1885 		    RND_EXTRACT_ANY);
1886 		beenhere = 1;
1887 	}
1888 
1889 	if (tcp_do_rfc1948) {
1890 		MD5_CTX ctx;
1891 		u_int8_t hash[16];	/* XXX MD5 knowledge */
1892 
1893 		/*
1894 		 * Compute the base value of the ISS.  It is a hash
1895 		 * of (saddr, sport, daddr, dport, secret).
1896 		 */
1897 		MD5Init(&ctx);
1898 
1899 		MD5Update(&ctx, (u_char *) laddr, addrsz);
1900 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
1901 
1902 		MD5Update(&ctx, (u_char *) faddr, addrsz);
1903 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
1904 
1905 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
1906 
1907 		MD5Final(hash, &ctx);
1908 
1909 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
1910 
1911 		/*
1912 		 * Now increment our "timer", and add it in to
1913 		 * the computed value.
1914 		 *
1915 		 * XXX Use `addin'?
1916 		 * XXX TCP_ISSINCR too large to use?
1917 		 */
1918 		tcp_iss_seq += TCP_ISSINCR;
1919 #ifdef TCPISS_DEBUG
1920 		printf("ISS hash 0x%08x, ", tcp_iss);
1921 #endif
1922 		tcp_iss += tcp_iss_seq + addin;
1923 #ifdef TCPISS_DEBUG
1924 		printf("new ISS 0x%08x\n", tcp_iss);
1925 #endif
1926 	} else
1927 #endif /* NRND > 0 */
1928 	{
1929 		/*
1930 		 * Randomize.
1931 		 */
1932 #if NRND > 0
1933 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
1934 #else
1935 		tcp_iss = random();
1936 #endif
1937 
1938 		/*
1939 		 * If we were asked to add some amount to a known value,
1940 		 * we will take a random value obtained above, mask off
1941 		 * the upper bits, and add in the known value.  We also
1942 		 * add in a constant to ensure that we are at least a
1943 		 * certain distance from the original value.
1944 		 *
1945 		 * This is used when an old connection is in timed wait
1946 		 * and we have a new one coming in, for instance.
1947 		 */
1948 		if (addin != 0) {
1949 #ifdef TCPISS_DEBUG
1950 			printf("Random %08x, ", tcp_iss);
1951 #endif
1952 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1953 			tcp_iss += addin + TCP_ISSINCR;
1954 #ifdef TCPISS_DEBUG
1955 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
1956 #endif
1957 		} else {
1958 			tcp_iss &= TCP_ISS_RANDOM_MASK;
1959 			tcp_iss += tcp_iss_seq;
1960 			tcp_iss_seq += TCP_ISSINCR;
1961 #ifdef TCPISS_DEBUG
1962 			printf("ISS %08x\n", tcp_iss);
1963 #endif
1964 		}
1965 	}
1966 
1967 	if (tcp_compat_42) {
1968 		/*
1969 		 * Limit it to the positive range for really old TCP
1970 		 * implementations.
1971 		 */
1972 		if (tcp_iss >= 0x80000000)
1973 			tcp_iss &= 0x7fffffff;		/* XXX */
1974 	}
1975 
1976 	return (tcp_iss);
1977 }
1978 
1979 #ifdef IPSEC
1980 /* compute ESP/AH header size for TCP, including outer IP header. */
1981 size_t
1982 ipsec4_hdrsiz_tcp(tp)
1983 	struct tcpcb *tp;
1984 {
1985 	struct inpcb *inp;
1986 	size_t hdrsiz;
1987 
1988 	/* XXX mapped addr case (tp->t_in6pcb) */
1989 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1990 		return 0;
1991 	switch (tp->t_family) {
1992 	case AF_INET:
1993 		/* XXX: should use currect direction. */
1994 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
1995 		break;
1996 	default:
1997 		hdrsiz = 0;
1998 		break;
1999 	}
2000 
2001 	return hdrsiz;
2002 }
2003 
2004 #ifdef INET6
2005 size_t
2006 ipsec6_hdrsiz_tcp(tp)
2007 	struct tcpcb *tp;
2008 {
2009 	struct in6pcb *in6p;
2010 	size_t hdrsiz;
2011 
2012 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2013 		return 0;
2014 	switch (tp->t_family) {
2015 	case AF_INET6:
2016 		/* XXX: should use currect direction. */
2017 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2018 		break;
2019 	case AF_INET:
2020 		/* mapped address case - tricky */
2021 	default:
2022 		hdrsiz = 0;
2023 		break;
2024 	}
2025 
2026 	return hdrsiz;
2027 }
2028 #endif
2029 #endif /*IPSEC*/
2030 
2031 /*
2032  * Determine the length of the TCP options for this connection.
2033  *
2034  * XXX:  What do we do for SACK, when we add that?  Just reserve
2035  *       all of the space?  Otherwise we can't exactly be incrementing
2036  *       cwnd by an amount that varies depending on the amount we last
2037  *       had to SACK!
2038  */
2039 
2040 u_int
2041 tcp_optlen(tp)
2042 	struct tcpcb *tp;
2043 {
2044 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2045 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2046 		return TCPOLEN_TSTAMP_APPA;
2047 	else
2048 		return 0;
2049 }
2050