xref: /netbsd-src/sys/netinet/tcp_subr.c (revision da5f4674a3fc214be3572d358b66af40ab9401e7)
1 /*	$NetBSD: tcp_subr.c,v 1.150 2003/08/22 22:27:07 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. Neither the name of the University nor the names of its contributors
82  *    may be used to endorse or promote products derived from this software
83  *    without specific prior written permission.
84  *
85  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
86  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
87  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
88  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
89  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
90  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
91  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
92  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
93  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
94  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
95  * SUCH DAMAGE.
96  *
97  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
98  */
99 
100 #include <sys/cdefs.h>
101 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.150 2003/08/22 22:27:07 itojun Exp $");
102 
103 #include "opt_inet.h"
104 #include "opt_ipsec.h"
105 #include "opt_tcp_compat_42.h"
106 #include "opt_inet_csum.h"
107 #include "opt_mbuftrace.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include <netinet/tcp.h>
150 #include <netinet/tcp_fsm.h>
151 #include <netinet/tcp_seq.h>
152 #include <netinet/tcp_timer.h>
153 #include <netinet/tcp_var.h>
154 #include <netinet/tcpip.h>
155 
156 #ifdef IPSEC
157 #include <netinet6/ipsec.h>
158 #endif /*IPSEC*/
159 
160 #ifdef FAST_IPSEC
161 #include <netipsec/ipsec.h>
162 #ifdef INET6
163 #include <netipsec/ipsec6.h>
164 #endif
165 #endif	/* FAST_IPSEC*/
166 
167 
168 #ifdef INET6
169 struct in6pcb tcb6;
170 #endif
171 
172 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
173 struct	tcpstat tcpstat;	/* tcp statistics */
174 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
175 
176 /* patchable/settable parameters for tcp */
177 int 	tcp_mssdflt = TCP_MSS;
178 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
179 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
180 #if NRND > 0
181 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
182 #endif
183 int	tcp_do_sack = 1;	/* selective acknowledgement */
184 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
185 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
186 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
187 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
188 #ifndef TCP_INIT_WIN
189 #define	TCP_INIT_WIN	1	/* initial slow start window */
190 #endif
191 #ifndef TCP_INIT_WIN_LOCAL
192 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
193 #endif
194 int	tcp_init_win = TCP_INIT_WIN;
195 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
196 int	tcp_mss_ifmtu = 0;
197 #ifdef TCP_COMPAT_42
198 int	tcp_compat_42 = 1;
199 #else
200 int	tcp_compat_42 = 0;
201 #endif
202 int	tcp_rst_ppslim = 100;	/* 100pps */
203 
204 /* tcb hash */
205 #ifndef TCBHASHSIZE
206 #define	TCBHASHSIZE	128
207 #endif
208 int	tcbhashsize = TCBHASHSIZE;
209 
210 /* syn hash parameters */
211 #define	TCP_SYN_HASH_SIZE	293
212 #define	TCP_SYN_BUCKET_SIZE	35
213 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
214 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
215 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
216 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
217 
218 int	tcp_freeq __P((struct tcpcb *));
219 
220 #ifdef INET
221 void	tcp_mtudisc_callback __P((struct in_addr));
222 #endif
223 #ifdef INET6
224 void	tcp6_mtudisc_callback __P((struct in6_addr *));
225 #endif
226 
227 void	tcp_mtudisc __P((struct inpcb *, int));
228 #ifdef INET6
229 void	tcp6_mtudisc __P((struct in6pcb *, int));
230 #endif
231 
232 struct pool tcpcb_pool;
233 
234 #ifdef TCP_CSUM_COUNTERS
235 #include <sys/device.h>
236 
237 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
238     NULL, "tcp", "hwcsum bad");
239 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "hwcsum ok");
241 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
242     NULL, "tcp", "hwcsum data");
243 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
244     NULL, "tcp", "swcsum");
245 #endif /* TCP_CSUM_COUNTERS */
246 
247 #ifdef TCP_OUTPUT_COUNTERS
248 #include <sys/device.h>
249 
250 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
251     NULL, "tcp", "output big header");
252 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
253     NULL, "tcp", "output copy small");
254 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
255     NULL, "tcp", "output copy big");
256 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
257     NULL, "tcp", "output reference big");
258 #endif /* TCP_OUTPUT_COUNTERS */
259 
260 #ifdef TCP_REASS_COUNTERS
261 #include <sys/device.h>
262 
263 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp_reass", "calls");
265 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     &tcp_reass_, "tcp_reass", "insert into empty queue");
267 struct evcnt tcp_reass_iteration[8] = {
268     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
269     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
270     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
271     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
272     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
273     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
274     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
275     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
276 };
277 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     &tcp_reass_, "tcp_reass", "prepend to first");
279 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     &tcp_reass_, "tcp_reass", "prepend");
281 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     &tcp_reass_, "tcp_reass", "insert");
283 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "insert at tail");
285 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286     &tcp_reass_, "tcp_reass", "append");
287 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
288     &tcp_reass_, "tcp_reass", "append to tail fragment");
289 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
290     &tcp_reass_, "tcp_reass", "overlap at end");
291 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
292     &tcp_reass_, "tcp_reass", "overlap at start");
293 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
294     &tcp_reass_, "tcp_reass", "duplicate segment");
295 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
296     &tcp_reass_, "tcp_reass", "duplicate fragment");
297 
298 #endif /* TCP_REASS_COUNTERS */
299 
300 #ifdef MBUFTRACE
301 struct mowner tcp_mowner = { "tcp" };
302 struct mowner tcp_rx_mowner = { "tcp", "rx" };
303 struct mowner tcp_tx_mowner = { "tcp", "tx" };
304 #endif
305 
306 /*
307  * Tcp initialization
308  */
309 void
310 tcp_init()
311 {
312 	int hlen;
313 
314 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
315 	    NULL);
316 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
317 #ifdef INET6
318 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
319 #endif
320 
321 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
322 #ifdef INET6
323 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
324 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
325 #endif
326 	if (max_protohdr < hlen)
327 		max_protohdr = hlen;
328 	if (max_linkhdr + hlen > MHLEN)
329 		panic("tcp_init");
330 
331 #ifdef INET
332 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
333 #endif
334 #ifdef INET6
335 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
336 #endif
337 
338 	/* Initialize timer state. */
339 	tcp_timer_init();
340 
341 	/* Initialize the compressed state engine. */
342 	syn_cache_init();
343 
344 #ifdef TCP_CSUM_COUNTERS
345 	evcnt_attach_static(&tcp_hwcsum_bad);
346 	evcnt_attach_static(&tcp_hwcsum_ok);
347 	evcnt_attach_static(&tcp_hwcsum_data);
348 	evcnt_attach_static(&tcp_swcsum);
349 #endif /* TCP_CSUM_COUNTERS */
350 
351 #ifdef TCP_OUTPUT_COUNTERS
352 	evcnt_attach_static(&tcp_output_bigheader);
353 	evcnt_attach_static(&tcp_output_copysmall);
354 	evcnt_attach_static(&tcp_output_copybig);
355 	evcnt_attach_static(&tcp_output_refbig);
356 #endif /* TCP_OUTPUT_COUNTERS */
357 
358 #ifdef TCP_REASS_COUNTERS
359 	evcnt_attach_static(&tcp_reass_);
360 	evcnt_attach_static(&tcp_reass_empty);
361 	evcnt_attach_static(&tcp_reass_iteration[0]);
362 	evcnt_attach_static(&tcp_reass_iteration[1]);
363 	evcnt_attach_static(&tcp_reass_iteration[2]);
364 	evcnt_attach_static(&tcp_reass_iteration[3]);
365 	evcnt_attach_static(&tcp_reass_iteration[4]);
366 	evcnt_attach_static(&tcp_reass_iteration[5]);
367 	evcnt_attach_static(&tcp_reass_iteration[6]);
368 	evcnt_attach_static(&tcp_reass_iteration[7]);
369 	evcnt_attach_static(&tcp_reass_prependfirst);
370 	evcnt_attach_static(&tcp_reass_prepend);
371 	evcnt_attach_static(&tcp_reass_insert);
372 	evcnt_attach_static(&tcp_reass_inserttail);
373 	evcnt_attach_static(&tcp_reass_append);
374 	evcnt_attach_static(&tcp_reass_appendtail);
375 	evcnt_attach_static(&tcp_reass_overlaptail);
376 	evcnt_attach_static(&tcp_reass_overlapfront);
377 	evcnt_attach_static(&tcp_reass_segdup);
378 	evcnt_attach_static(&tcp_reass_fragdup);
379 #endif /* TCP_REASS_COUNTERS */
380 
381 	MOWNER_ATTACH(&tcp_tx_mowner);
382 	MOWNER_ATTACH(&tcp_rx_mowner);
383 	MOWNER_ATTACH(&tcp_mowner);
384 }
385 
386 /*
387  * Create template to be used to send tcp packets on a connection.
388  * Call after host entry created, allocates an mbuf and fills
389  * in a skeletal tcp/ip header, minimizing the amount of work
390  * necessary when the connection is used.
391  */
392 struct mbuf *
393 tcp_template(tp)
394 	struct tcpcb *tp;
395 {
396 	struct inpcb *inp = tp->t_inpcb;
397 #ifdef INET6
398 	struct in6pcb *in6p = tp->t_in6pcb;
399 #endif
400 	struct tcphdr *n;
401 	struct mbuf *m;
402 	int hlen;
403 
404 	switch (tp->t_family) {
405 	case AF_INET:
406 		hlen = sizeof(struct ip);
407 		if (inp)
408 			break;
409 #ifdef INET6
410 		if (in6p) {
411 			/* mapped addr case */
412 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
413 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
414 				break;
415 		}
416 #endif
417 		return NULL;	/*EINVAL*/
418 #ifdef INET6
419 	case AF_INET6:
420 		hlen = sizeof(struct ip6_hdr);
421 		if (in6p) {
422 			/* more sainty check? */
423 			break;
424 		}
425 		return NULL;	/*EINVAL*/
426 #endif
427 	default:
428 		hlen = 0;	/*pacify gcc*/
429 		return NULL;	/*EAFNOSUPPORT*/
430 	}
431 #ifdef DIAGNOSTIC
432 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
433 		panic("mclbytes too small for t_template");
434 #endif
435 	m = tp->t_template;
436 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
437 		;
438 	else {
439 		if (m)
440 			m_freem(m);
441 		m = tp->t_template = NULL;
442 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
443 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
444 			MCLGET(m, M_DONTWAIT);
445 			if ((m->m_flags & M_EXT) == 0) {
446 				m_free(m);
447 				m = NULL;
448 			}
449 		}
450 		if (m == NULL)
451 			return NULL;
452 		MCLAIM(m, &tcp_mowner);
453 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
454 	}
455 
456 	bzero(mtod(m, caddr_t), m->m_len);
457 
458 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
459 
460 	switch (tp->t_family) {
461 	case AF_INET:
462 	    {
463 		struct ipovly *ipov;
464 		mtod(m, struct ip *)->ip_v = 4;
465 		ipov = mtod(m, struct ipovly *);
466 		ipov->ih_pr = IPPROTO_TCP;
467 		ipov->ih_len = htons(sizeof(struct tcphdr));
468 		if (inp) {
469 			ipov->ih_src = inp->inp_laddr;
470 			ipov->ih_dst = inp->inp_faddr;
471 		}
472 #ifdef INET6
473 		else if (in6p) {
474 			/* mapped addr case */
475 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
476 				sizeof(ipov->ih_src));
477 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
478 				sizeof(ipov->ih_dst));
479 		}
480 #endif
481 		/*
482 		 * Compute the pseudo-header portion of the checksum
483 		 * now.  We incrementally add in the TCP option and
484 		 * payload lengths later, and then compute the TCP
485 		 * checksum right before the packet is sent off onto
486 		 * the wire.
487 		 */
488 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
489 		    ipov->ih_dst.s_addr,
490 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
491 		break;
492 	    }
493 #ifdef INET6
494 	case AF_INET6:
495 	    {
496 		struct ip6_hdr *ip6;
497 		mtod(m, struct ip *)->ip_v = 6;
498 		ip6 = mtod(m, struct ip6_hdr *);
499 		ip6->ip6_nxt = IPPROTO_TCP;
500 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
501 		ip6->ip6_src = in6p->in6p_laddr;
502 		ip6->ip6_dst = in6p->in6p_faddr;
503 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
504 		if (ip6_auto_flowlabel) {
505 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
506 			ip6->ip6_flow |=
507 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
508 		}
509 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
510 		ip6->ip6_vfc |= IPV6_VERSION;
511 
512 		/*
513 		 * Compute the pseudo-header portion of the checksum
514 		 * now.  We incrementally add in the TCP option and
515 		 * payload lengths later, and then compute the TCP
516 		 * checksum right before the packet is sent off onto
517 		 * the wire.
518 		 */
519 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
520 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
521 		    htonl(IPPROTO_TCP));
522 		break;
523 	    }
524 #endif
525 	}
526 	if (inp) {
527 		n->th_sport = inp->inp_lport;
528 		n->th_dport = inp->inp_fport;
529 	}
530 #ifdef INET6
531 	else if (in6p) {
532 		n->th_sport = in6p->in6p_lport;
533 		n->th_dport = in6p->in6p_fport;
534 	}
535 #endif
536 	n->th_seq = 0;
537 	n->th_ack = 0;
538 	n->th_x2 = 0;
539 	n->th_off = 5;
540 	n->th_flags = 0;
541 	n->th_win = 0;
542 	n->th_urp = 0;
543 	return (m);
544 }
545 
546 /*
547  * Send a single message to the TCP at address specified by
548  * the given TCP/IP header.  If m == 0, then we make a copy
549  * of the tcpiphdr at ti and send directly to the addressed host.
550  * This is used to force keep alive messages out using the TCP
551  * template for a connection tp->t_template.  If flags are given
552  * then we send a message back to the TCP which originated the
553  * segment ti, and discard the mbuf containing it and any other
554  * attached mbufs.
555  *
556  * In any case the ack and sequence number of the transmitted
557  * segment are as specified by the parameters.
558  */
559 int
560 tcp_respond(tp, template, m, th0, ack, seq, flags)
561 	struct tcpcb *tp;
562 	struct mbuf *template;
563 	struct mbuf *m;
564 	struct tcphdr *th0;
565 	tcp_seq ack, seq;
566 	int flags;
567 {
568 	struct route *ro;
569 	int error, tlen, win = 0;
570 	int hlen;
571 	struct ip *ip;
572 #ifdef INET6
573 	struct ip6_hdr *ip6;
574 #endif
575 	int family;	/* family on packet, not inpcb/in6pcb! */
576 	struct tcphdr *th;
577 	struct socket *so;
578 
579 	if (tp != NULL && (flags & TH_RST) == 0) {
580 #ifdef DIAGNOSTIC
581 		if (tp->t_inpcb && tp->t_in6pcb)
582 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
583 #endif
584 #ifdef INET
585 		if (tp->t_inpcb)
586 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
587 #endif
588 #ifdef INET6
589 		if (tp->t_in6pcb)
590 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
591 #endif
592 	}
593 
594 	th = NULL;	/* Quell uninitialized warning */
595 	ip = NULL;
596 #ifdef INET6
597 	ip6 = NULL;
598 #endif
599 	if (m == 0) {
600 		if (!template)
601 			return EINVAL;
602 
603 		/* get family information from template */
604 		switch (mtod(template, struct ip *)->ip_v) {
605 		case 4:
606 			family = AF_INET;
607 			hlen = sizeof(struct ip);
608 			break;
609 #ifdef INET6
610 		case 6:
611 			family = AF_INET6;
612 			hlen = sizeof(struct ip6_hdr);
613 			break;
614 #endif
615 		default:
616 			return EAFNOSUPPORT;
617 		}
618 
619 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
620 		if (m) {
621 			MCLAIM(m, &tcp_tx_mowner);
622 			MCLGET(m, M_DONTWAIT);
623 			if ((m->m_flags & M_EXT) == 0) {
624 				m_free(m);
625 				m = NULL;
626 			}
627 		}
628 		if (m == NULL)
629 			return (ENOBUFS);
630 
631 		if (tcp_compat_42)
632 			tlen = 1;
633 		else
634 			tlen = 0;
635 
636 		m->m_data += max_linkhdr;
637 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
638 			template->m_len);
639 		switch (family) {
640 		case AF_INET:
641 			ip = mtod(m, struct ip *);
642 			th = (struct tcphdr *)(ip + 1);
643 			break;
644 #ifdef INET6
645 		case AF_INET6:
646 			ip6 = mtod(m, struct ip6_hdr *);
647 			th = (struct tcphdr *)(ip6 + 1);
648 			break;
649 #endif
650 #if 0
651 		default:
652 			/* noone will visit here */
653 			m_freem(m);
654 			return EAFNOSUPPORT;
655 #endif
656 		}
657 		flags = TH_ACK;
658 	} else {
659 
660 		if ((m->m_flags & M_PKTHDR) == 0) {
661 #if 0
662 			printf("non PKTHDR to tcp_respond\n");
663 #endif
664 			m_freem(m);
665 			return EINVAL;
666 		}
667 #ifdef DIAGNOSTIC
668 		if (!th0)
669 			panic("th0 == NULL in tcp_respond");
670 #endif
671 
672 		/* get family information from m */
673 		switch (mtod(m, struct ip *)->ip_v) {
674 		case 4:
675 			family = AF_INET;
676 			hlen = sizeof(struct ip);
677 			ip = mtod(m, struct ip *);
678 			break;
679 #ifdef INET6
680 		case 6:
681 			family = AF_INET6;
682 			hlen = sizeof(struct ip6_hdr);
683 			ip6 = mtod(m, struct ip6_hdr *);
684 			break;
685 #endif
686 		default:
687 			m_freem(m);
688 			return EAFNOSUPPORT;
689 		}
690 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
691 			tlen = sizeof(*th0);
692 		else
693 			tlen = th0->th_off << 2;
694 
695 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
696 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
697 			m->m_len = hlen + tlen;
698 			m_freem(m->m_next);
699 			m->m_next = NULL;
700 		} else {
701 			struct mbuf *n;
702 
703 #ifdef DIAGNOSTIC
704 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
705 				m_freem(m);
706 				return EMSGSIZE;
707 			}
708 #endif
709 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
710 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
711 				MCLGET(n, M_DONTWAIT);
712 				if ((n->m_flags & M_EXT) == 0) {
713 					m_freem(n);
714 					n = NULL;
715 				}
716 			}
717 			if (!n) {
718 				m_freem(m);
719 				return ENOBUFS;
720 			}
721 
722 			MCLAIM(n, &tcp_tx_mowner);
723 			n->m_data += max_linkhdr;
724 			n->m_len = hlen + tlen;
725 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
726 			m_copyback(n, hlen, tlen, (caddr_t)th0);
727 
728 			m_freem(m);
729 			m = n;
730 			n = NULL;
731 		}
732 
733 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
734 		switch (family) {
735 		case AF_INET:
736 			ip = mtod(m, struct ip *);
737 			th = (struct tcphdr *)(ip + 1);
738 			ip->ip_p = IPPROTO_TCP;
739 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
740 			ip->ip_p = IPPROTO_TCP;
741 			break;
742 #ifdef INET6
743 		case AF_INET6:
744 			ip6 = mtod(m, struct ip6_hdr *);
745 			th = (struct tcphdr *)(ip6 + 1);
746 			ip6->ip6_nxt = IPPROTO_TCP;
747 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
748 			ip6->ip6_nxt = IPPROTO_TCP;
749 			break;
750 #endif
751 #if 0
752 		default:
753 			/* noone will visit here */
754 			m_freem(m);
755 			return EAFNOSUPPORT;
756 #endif
757 		}
758 		xchg(th->th_dport, th->th_sport, u_int16_t);
759 #undef xchg
760 		tlen = 0;	/*be friendly with the following code*/
761 	}
762 	th->th_seq = htonl(seq);
763 	th->th_ack = htonl(ack);
764 	th->th_x2 = 0;
765 	if ((flags & TH_SYN) == 0) {
766 		if (tp)
767 			win >>= tp->rcv_scale;
768 		if (win > TCP_MAXWIN)
769 			win = TCP_MAXWIN;
770 		th->th_win = htons((u_int16_t)win);
771 		th->th_off = sizeof (struct tcphdr) >> 2;
772 		tlen += sizeof(*th);
773 	} else
774 		tlen += th->th_off << 2;
775 	m->m_len = hlen + tlen;
776 	m->m_pkthdr.len = hlen + tlen;
777 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
778 	th->th_flags = flags;
779 	th->th_urp = 0;
780 
781 	switch (family) {
782 #ifdef INET
783 	case AF_INET:
784 	    {
785 		struct ipovly *ipov = (struct ipovly *)ip;
786 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
787 		ipov->ih_len = htons((u_int16_t)tlen);
788 
789 		th->th_sum = 0;
790 		th->th_sum = in_cksum(m, hlen + tlen);
791 		ip->ip_len = htons(hlen + tlen);
792 		ip->ip_ttl = ip_defttl;
793 		break;
794 	    }
795 #endif
796 #ifdef INET6
797 	case AF_INET6:
798 	    {
799 		th->th_sum = 0;
800 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
801 				tlen);
802 		ip6->ip6_plen = ntohs(tlen);
803 		if (tp && tp->t_in6pcb) {
804 			struct ifnet *oifp;
805 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
806 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
807 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
808 		} else
809 			ip6->ip6_hlim = ip6_defhlim;
810 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
811 		if (ip6_auto_flowlabel) {
812 			ip6->ip6_flow |=
813 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
814 		}
815 		break;
816 	    }
817 #endif
818 	}
819 
820 	if (tp && tp->t_inpcb)
821 		so = tp->t_inpcb->inp_socket;
822 #ifdef INET6
823 	else if (tp && tp->t_in6pcb)
824 		so = tp->t_in6pcb->in6p_socket;
825 #endif
826 	else
827 		so = NULL;
828 
829 	if (tp != NULL && tp->t_inpcb != NULL) {
830 		ro = &tp->t_inpcb->inp_route;
831 #ifdef DIAGNOSTIC
832 		if (family != AF_INET)
833 			panic("tcp_respond: address family mismatch");
834 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
835 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
836 			    ntohl(ip->ip_dst.s_addr),
837 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
838 		}
839 #endif
840 	}
841 #ifdef INET6
842 	else if (tp != NULL && tp->t_in6pcb != NULL) {
843 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
844 #ifdef DIAGNOSTIC
845 		if (family == AF_INET) {
846 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
847 				panic("tcp_respond: not mapped addr");
848 			if (bcmp(&ip->ip_dst,
849 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
850 			    sizeof(ip->ip_dst)) != 0) {
851 				panic("tcp_respond: ip_dst != in6p_faddr");
852 			}
853 		} else if (family == AF_INET6) {
854 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
855 			    &tp->t_in6pcb->in6p_faddr))
856 				panic("tcp_respond: ip6_dst != in6p_faddr");
857 		} else
858 			panic("tcp_respond: address family mismatch");
859 #endif
860 	}
861 #endif
862 	else
863 		ro = NULL;
864 
865 	switch (family) {
866 #ifdef INET
867 	case AF_INET:
868 		error = ip_output(m, NULL, ro,
869 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
870 		    (struct ip_moptions *)0, so);
871 		break;
872 #endif
873 #ifdef INET6
874 	case AF_INET6:
875 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
876 		    (struct ip6_moptions *)0, so, NULL);
877 		break;
878 #endif
879 	default:
880 		error = EAFNOSUPPORT;
881 		break;
882 	}
883 
884 	return (error);
885 }
886 
887 /*
888  * Create a new TCP control block, making an
889  * empty reassembly queue and hooking it to the argument
890  * protocol control block.
891  */
892 struct tcpcb *
893 tcp_newtcpcb(family, aux)
894 	int family;	/* selects inpcb, or in6pcb */
895 	void *aux;
896 {
897 	struct tcpcb *tp;
898 	int i;
899 
900 	switch (family) {
901 	case PF_INET:
902 		break;
903 #ifdef INET6
904 	case PF_INET6:
905 		break;
906 #endif
907 	default:
908 		return NULL;
909 	}
910 
911 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
912 	if (tp == NULL)
913 		return (NULL);
914 	bzero((caddr_t)tp, sizeof(struct tcpcb));
915 	TAILQ_INIT(&tp->segq);
916 	TAILQ_INIT(&tp->timeq);
917 	tp->t_family = family;		/* may be overridden later on */
918 	tp->t_peermss = tcp_mssdflt;
919 	tp->t_ourmss = tcp_mssdflt;
920 	tp->t_segsz = tcp_mssdflt;
921 	LIST_INIT(&tp->t_sc);
922 
923 	tp->t_lastm = NULL;
924 	tp->t_lastoff = 0;
925 
926 	callout_init(&tp->t_delack_ch);
927 	for (i = 0; i < TCPT_NTIMERS; i++)
928 		TCP_TIMER_INIT(tp, i);
929 
930 	tp->t_flags = 0;
931 	if (tcp_do_rfc1323 && tcp_do_win_scale)
932 		tp->t_flags |= TF_REQ_SCALE;
933 	if (tcp_do_rfc1323 && tcp_do_timestamps)
934 		tp->t_flags |= TF_REQ_TSTMP;
935 	if (tcp_do_sack == 2)
936 		tp->t_flags |= TF_WILL_SACK;
937 	else if (tcp_do_sack == 1)
938 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
939 	tp->t_flags |= TF_CANT_TXSACK;
940 	switch (family) {
941 	case PF_INET:
942 		tp->t_inpcb = (struct inpcb *)aux;
943 		tp->t_mtudisc = ip_mtudisc;
944 		break;
945 #ifdef INET6
946 	case PF_INET6:
947 		tp->t_in6pcb = (struct in6pcb *)aux;
948 		/* for IPv6, always try to run path MTU discovery */
949 		tp->t_mtudisc = 1;
950 		break;
951 #endif
952 	}
953 	/*
954 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
955 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
956 	 * reasonable initial retransmit time.
957 	 */
958 	tp->t_srtt = TCPTV_SRTTBASE;
959 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
960 	tp->t_rttmin = TCPTV_MIN;
961 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
962 	    TCPTV_MIN, TCPTV_REXMTMAX);
963 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
964 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
965 	if (family == AF_INET) {
966 		struct inpcb *inp = (struct inpcb *)aux;
967 		inp->inp_ip.ip_ttl = ip_defttl;
968 		inp->inp_ppcb = (caddr_t)tp;
969 	}
970 #ifdef INET6
971 	else if (family == AF_INET6) {
972 		struct in6pcb *in6p = (struct in6pcb *)aux;
973 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
974 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
975 					       : NULL);
976 		in6p->in6p_ppcb = (caddr_t)tp;
977 	}
978 #endif
979 
980 	/*
981 	 * Initialize our timebase.  When we send timestamps, we take
982 	 * the delta from tcp_now -- this means each connection always
983 	 * gets a timebase of 0, which makes it, among other things,
984 	 * more difficult to determine how long a system has been up,
985 	 * and thus how many TCP sequence increments have occurred.
986 	 */
987 	tp->ts_timebase = tcp_now;
988 
989 	return (tp);
990 }
991 
992 /*
993  * Drop a TCP connection, reporting
994  * the specified error.  If connection is synchronized,
995  * then send a RST to peer.
996  */
997 struct tcpcb *
998 tcp_drop(tp, errno)
999 	struct tcpcb *tp;
1000 	int errno;
1001 {
1002 	struct socket *so = NULL;
1003 
1004 #ifdef DIAGNOSTIC
1005 	if (tp->t_inpcb && tp->t_in6pcb)
1006 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1007 #endif
1008 #ifdef INET
1009 	if (tp->t_inpcb)
1010 		so = tp->t_inpcb->inp_socket;
1011 #endif
1012 #ifdef INET6
1013 	if (tp->t_in6pcb)
1014 		so = tp->t_in6pcb->in6p_socket;
1015 #endif
1016 	if (!so)
1017 		return NULL;
1018 
1019 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1020 		tp->t_state = TCPS_CLOSED;
1021 		(void) tcp_output(tp);
1022 		tcpstat.tcps_drops++;
1023 	} else
1024 		tcpstat.tcps_conndrops++;
1025 	if (errno == ETIMEDOUT && tp->t_softerror)
1026 		errno = tp->t_softerror;
1027 	so->so_error = errno;
1028 	return (tcp_close(tp));
1029 }
1030 
1031 /*
1032  * Return whether this tcpcb is marked as dead, indicating
1033  * to the calling timer function that no further action should
1034  * be taken, as we are about to release this tcpcb.  The release
1035  * of the storage will be done if this is the last timer running.
1036  *
1037  * This is typically called from the callout handler function before
1038  * callout_ack() is done, therefore we need to test the number of
1039  * running timer functions against 1 below, not 0.
1040  */
1041 int
1042 tcp_isdead(tp)
1043 	struct tcpcb *tp;
1044 {
1045 	int dead = (tp->t_flags & TF_DEAD);
1046 
1047 	if (__predict_false(dead)) {
1048 		if (tcp_timers_invoking(tp) > 1)
1049 				/* not quite there yet -- count separately? */
1050 			return dead;
1051 		tcpstat.tcps_delayed_free++;
1052 		pool_put(&tcpcb_pool, tp);
1053 	}
1054 	return dead;
1055 }
1056 
1057 /*
1058  * Close a TCP control block:
1059  *	discard all space held by the tcp
1060  *	discard internet protocol block
1061  *	wake up any sleepers
1062  */
1063 struct tcpcb *
1064 tcp_close(tp)
1065 	struct tcpcb *tp;
1066 {
1067 	struct inpcb *inp;
1068 #ifdef INET6
1069 	struct in6pcb *in6p;
1070 #endif
1071 	struct socket *so;
1072 #ifdef RTV_RTT
1073 	struct rtentry *rt;
1074 #endif
1075 	struct route *ro;
1076 
1077 	inp = tp->t_inpcb;
1078 #ifdef INET6
1079 	in6p = tp->t_in6pcb;
1080 #endif
1081 	so = NULL;
1082 	ro = NULL;
1083 	if (inp) {
1084 		so = inp->inp_socket;
1085 		ro = &inp->inp_route;
1086 	}
1087 #ifdef INET6
1088 	else if (in6p) {
1089 		so = in6p->in6p_socket;
1090 		ro = (struct route *)&in6p->in6p_route;
1091 	}
1092 #endif
1093 
1094 #ifdef RTV_RTT
1095 	/*
1096 	 * If we sent enough data to get some meaningful characteristics,
1097 	 * save them in the routing entry.  'Enough' is arbitrarily
1098 	 * defined as the sendpipesize (default 4K) * 16.  This would
1099 	 * give us 16 rtt samples assuming we only get one sample per
1100 	 * window (the usual case on a long haul net).  16 samples is
1101 	 * enough for the srtt filter to converge to within 5% of the correct
1102 	 * value; fewer samples and we could save a very bogus rtt.
1103 	 *
1104 	 * Don't update the default route's characteristics and don't
1105 	 * update anything that the user "locked".
1106 	 */
1107 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1108 	    ro && (rt = ro->ro_rt) &&
1109 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1110 		u_long i = 0;
1111 
1112 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1113 			i = tp->t_srtt *
1114 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1115 			if (rt->rt_rmx.rmx_rtt && i)
1116 				/*
1117 				 * filter this update to half the old & half
1118 				 * the new values, converting scale.
1119 				 * See route.h and tcp_var.h for a
1120 				 * description of the scaling constants.
1121 				 */
1122 				rt->rt_rmx.rmx_rtt =
1123 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1124 			else
1125 				rt->rt_rmx.rmx_rtt = i;
1126 		}
1127 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1128 			i = tp->t_rttvar *
1129 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1130 			if (rt->rt_rmx.rmx_rttvar && i)
1131 				rt->rt_rmx.rmx_rttvar =
1132 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1133 			else
1134 				rt->rt_rmx.rmx_rttvar = i;
1135 		}
1136 		/*
1137 		 * update the pipelimit (ssthresh) if it has been updated
1138 		 * already or if a pipesize was specified & the threshhold
1139 		 * got below half the pipesize.  I.e., wait for bad news
1140 		 * before we start updating, then update on both good
1141 		 * and bad news.
1142 		 */
1143 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1144 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1145 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1146 			/*
1147 			 * convert the limit from user data bytes to
1148 			 * packets then to packet data bytes.
1149 			 */
1150 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1151 			if (i < 2)
1152 				i = 2;
1153 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1154 			if (rt->rt_rmx.rmx_ssthresh)
1155 				rt->rt_rmx.rmx_ssthresh =
1156 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1157 			else
1158 				rt->rt_rmx.rmx_ssthresh = i;
1159 		}
1160 	}
1161 #endif /* RTV_RTT */
1162 	/* free the reassembly queue, if any */
1163 	TCP_REASS_LOCK(tp);
1164 	(void) tcp_freeq(tp);
1165 	TCP_REASS_UNLOCK(tp);
1166 
1167 	tcp_canceltimers(tp);
1168 	TCP_CLEAR_DELACK(tp);
1169 	syn_cache_cleanup(tp);
1170 
1171 	if (tp->t_template) {
1172 		m_free(tp->t_template);
1173 		tp->t_template = NULL;
1174 	}
1175 	if (tcp_timers_invoking(tp))
1176 		tp->t_flags |= TF_DEAD;
1177 	else
1178 		pool_put(&tcpcb_pool, tp);
1179 
1180 	if (inp) {
1181 		inp->inp_ppcb = 0;
1182 		soisdisconnected(so);
1183 		in_pcbdetach(inp);
1184 	}
1185 #ifdef INET6
1186 	else if (in6p) {
1187 		in6p->in6p_ppcb = 0;
1188 		soisdisconnected(so);
1189 		in6_pcbdetach(in6p);
1190 	}
1191 #endif
1192 	tcpstat.tcps_closed++;
1193 	return ((struct tcpcb *)0);
1194 }
1195 
1196 int
1197 tcp_freeq(tp)
1198 	struct tcpcb *tp;
1199 {
1200 	struct ipqent *qe;
1201 	int rv = 0;
1202 #ifdef TCPREASS_DEBUG
1203 	int i = 0;
1204 #endif
1205 
1206 	TCP_REASS_LOCK_CHECK(tp);
1207 
1208 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1209 #ifdef TCPREASS_DEBUG
1210 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1211 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1212 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1213 #endif
1214 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1215 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1216 		m_freem(qe->ipqe_m);
1217 		pool_put(&ipqent_pool, qe);
1218 		rv = 1;
1219 	}
1220 	return (rv);
1221 }
1222 
1223 /*
1224  * Protocol drain routine.  Called when memory is in short supply.
1225  */
1226 void
1227 tcp_drain()
1228 {
1229 	struct inpcb *inp;
1230 	struct tcpcb *tp;
1231 
1232 	/*
1233 	 * Free the sequence queue of all TCP connections.
1234 	 */
1235 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1236 	if (inp)						/* XXX */
1237 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1238 		if ((tp = intotcpcb(inp)) != NULL) {
1239 			/*
1240 			 * We may be called from a device's interrupt
1241 			 * context.  If the tcpcb is already busy,
1242 			 * just bail out now.
1243 			 */
1244 			if (tcp_reass_lock_try(tp) == 0)
1245 				continue;
1246 			if (tcp_freeq(tp))
1247 				tcpstat.tcps_connsdrained++;
1248 			TCP_REASS_UNLOCK(tp);
1249 		}
1250 	}
1251 }
1252 
1253 #ifdef INET6
1254 void
1255 tcp6_drain()
1256 {
1257 	struct in6pcb *in6p;
1258 	struct tcpcb *tp;
1259 	struct in6pcb *head = &tcb6;
1260 
1261 	/*
1262 	 * Free the sequence queue of all TCP connections.
1263 	 */
1264 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1265 		if ((tp = in6totcpcb(in6p)) != NULL) {
1266 			/*
1267 			 * We may be called from a device's interrupt
1268 			 * context.  If the tcpcb is already busy,
1269 			 * just bail out now.
1270 			 */
1271 			if (tcp_reass_lock_try(tp) == 0)
1272 				continue;
1273 			if (tcp_freeq(tp))
1274 				tcpstat.tcps_connsdrained++;
1275 			TCP_REASS_UNLOCK(tp);
1276 		}
1277 	}
1278 }
1279 #endif
1280 
1281 /*
1282  * Notify a tcp user of an asynchronous error;
1283  * store error as soft error, but wake up user
1284  * (for now, won't do anything until can select for soft error).
1285  */
1286 void
1287 tcp_notify(inp, error)
1288 	struct inpcb *inp;
1289 	int error;
1290 {
1291 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1292 	struct socket *so = inp->inp_socket;
1293 
1294 	/*
1295 	 * Ignore some errors if we are hooked up.
1296 	 * If connection hasn't completed, has retransmitted several times,
1297 	 * and receives a second error, give up now.  This is better
1298 	 * than waiting a long time to establish a connection that
1299 	 * can never complete.
1300 	 */
1301 	if (tp->t_state == TCPS_ESTABLISHED &&
1302 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1303 	      error == EHOSTDOWN)) {
1304 		return;
1305 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1306 	    tp->t_rxtshift > 3 && tp->t_softerror)
1307 		so->so_error = error;
1308 	else
1309 		tp->t_softerror = error;
1310 	wakeup((caddr_t) &so->so_timeo);
1311 	sorwakeup(so);
1312 	sowwakeup(so);
1313 }
1314 
1315 #ifdef INET6
1316 void
1317 tcp6_notify(in6p, error)
1318 	struct in6pcb *in6p;
1319 	int error;
1320 {
1321 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1322 	struct socket *so = in6p->in6p_socket;
1323 
1324 	/*
1325 	 * Ignore some errors if we are hooked up.
1326 	 * If connection hasn't completed, has retransmitted several times,
1327 	 * and receives a second error, give up now.  This is better
1328 	 * than waiting a long time to establish a connection that
1329 	 * can never complete.
1330 	 */
1331 	if (tp->t_state == TCPS_ESTABLISHED &&
1332 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1333 	      error == EHOSTDOWN)) {
1334 		return;
1335 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1336 	    tp->t_rxtshift > 3 && tp->t_softerror)
1337 		so->so_error = error;
1338 	else
1339 		tp->t_softerror = error;
1340 	wakeup((caddr_t) &so->so_timeo);
1341 	sorwakeup(so);
1342 	sowwakeup(so);
1343 }
1344 #endif
1345 
1346 #ifdef INET6
1347 void
1348 tcp6_ctlinput(cmd, sa, d)
1349 	int cmd;
1350 	struct sockaddr *sa;
1351 	void *d;
1352 {
1353 	struct tcphdr th;
1354 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1355 	int nmatch;
1356 	struct ip6_hdr *ip6;
1357 	const struct sockaddr_in6 *sa6_src = NULL;
1358 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1359 	struct mbuf *m;
1360 	int off;
1361 
1362 	if (sa->sa_family != AF_INET6 ||
1363 	    sa->sa_len != sizeof(struct sockaddr_in6))
1364 		return;
1365 	if ((unsigned)cmd >= PRC_NCMDS)
1366 		return;
1367 	else if (cmd == PRC_QUENCH) {
1368 		/* XXX there's no PRC_QUENCH in IPv6 */
1369 		notify = tcp6_quench;
1370 	} else if (PRC_IS_REDIRECT(cmd))
1371 		notify = in6_rtchange, d = NULL;
1372 	else if (cmd == PRC_MSGSIZE)
1373 		; /* special code is present, see below */
1374 	else if (cmd == PRC_HOSTDEAD)
1375 		d = NULL;
1376 	else if (inet6ctlerrmap[cmd] == 0)
1377 		return;
1378 
1379 	/* if the parameter is from icmp6, decode it. */
1380 	if (d != NULL) {
1381 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1382 		m = ip6cp->ip6c_m;
1383 		ip6 = ip6cp->ip6c_ip6;
1384 		off = ip6cp->ip6c_off;
1385 		sa6_src = ip6cp->ip6c_src;
1386 	} else {
1387 		m = NULL;
1388 		ip6 = NULL;
1389 		sa6_src = &sa6_any;
1390 	}
1391 
1392 	if (ip6) {
1393 		/*
1394 		 * XXX: We assume that when ip6 is non NULL,
1395 		 * M and OFF are valid.
1396 		 */
1397 
1398 		/* check if we can safely examine src and dst ports */
1399 		if (m->m_pkthdr.len < off + sizeof(th)) {
1400 			if (cmd == PRC_MSGSIZE)
1401 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1402 			return;
1403 		}
1404 
1405 		bzero(&th, sizeof(th));
1406 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1407 
1408 		if (cmd == PRC_MSGSIZE) {
1409 			int valid = 0;
1410 
1411 			/*
1412 			 * Check to see if we have a valid TCP connection
1413 			 * corresponding to the address in the ICMPv6 message
1414 			 * payload.
1415 			 */
1416 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1417 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1418 			    th.th_sport, 0))
1419 				valid++;
1420 
1421 			/*
1422 			 * Depending on the value of "valid" and routing table
1423 			 * size (mtudisc_{hi,lo}wat), we will:
1424 			 * - recalcurate the new MTU and create the
1425 			 *   corresponding routing entry, or
1426 			 * - ignore the MTU change notification.
1427 			 */
1428 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1429 
1430 			/*
1431 			 * no need to call in6_pcbnotify, it should have been
1432 			 * called via callback if necessary
1433 			 */
1434 			return;
1435 		}
1436 
1437 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1438 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1439 		if (nmatch == 0 && syn_cache_count &&
1440 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1441 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1442 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1443 			syn_cache_unreach((struct sockaddr *)sa6_src,
1444 					  sa, &th);
1445 	} else {
1446 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1447 		    0, cmd, NULL, notify);
1448 	}
1449 }
1450 #endif
1451 
1452 #ifdef INET
1453 /* assumes that ip header and tcp header are contiguous on mbuf */
1454 void *
1455 tcp_ctlinput(cmd, sa, v)
1456 	int cmd;
1457 	struct sockaddr *sa;
1458 	void *v;
1459 {
1460 	struct ip *ip = v;
1461 	struct tcphdr *th;
1462 	struct icmp *icp;
1463 	extern const int inetctlerrmap[];
1464 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1465 	int errno;
1466 	int nmatch;
1467 #ifdef INET6
1468 	struct in6_addr src6, dst6;
1469 #endif
1470 
1471 	if (sa->sa_family != AF_INET ||
1472 	    sa->sa_len != sizeof(struct sockaddr_in))
1473 		return NULL;
1474 	if ((unsigned)cmd >= PRC_NCMDS)
1475 		return NULL;
1476 	errno = inetctlerrmap[cmd];
1477 	if (cmd == PRC_QUENCH)
1478 		notify = tcp_quench;
1479 	else if (PRC_IS_REDIRECT(cmd))
1480 		notify = in_rtchange, ip = 0;
1481 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1482 		/*
1483 		 * Check to see if we have a valid TCP connection
1484 		 * corresponding to the address in the ICMP message
1485 		 * payload.
1486 		 *
1487 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1488 		 */
1489 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1490 #ifdef INET6
1491 		memset(&src6, 0, sizeof(src6));
1492 		memset(&dst6, 0, sizeof(dst6));
1493 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1494 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1495 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1496 #endif
1497 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1498 		    ip->ip_src, th->th_sport) != NULL)
1499 			;
1500 #ifdef INET6
1501 		else if (in6_pcblookup_connect(&tcb6, &dst6,
1502 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1503 			;
1504 #endif
1505 		else
1506 			return NULL;
1507 
1508 		/*
1509 		 * Now that we've validated that we are actually communicating
1510 		 * with the host indicated in the ICMP message, locate the
1511 		 * ICMP header, recalculate the new MTU, and create the
1512 		 * corresponding routing entry.
1513 		 */
1514 		icp = (struct icmp *)((caddr_t)ip -
1515 		    offsetof(struct icmp, icmp_ip));
1516 		icmp_mtudisc(icp, ip->ip_dst);
1517 
1518 		return NULL;
1519 	} else if (cmd == PRC_HOSTDEAD)
1520 		ip = 0;
1521 	else if (errno == 0)
1522 		return NULL;
1523 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1524 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1525 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1526 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1527 		if (nmatch == 0 && syn_cache_count &&
1528 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1529 		    inetctlerrmap[cmd] == ENETUNREACH ||
1530 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1531 			struct sockaddr_in sin;
1532 			bzero(&sin, sizeof(sin));
1533 			sin.sin_len = sizeof(sin);
1534 			sin.sin_family = AF_INET;
1535 			sin.sin_port = th->th_sport;
1536 			sin.sin_addr = ip->ip_src;
1537 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1538 		}
1539 
1540 		/* XXX mapped address case */
1541 	} else
1542 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1543 		    notify);
1544 	return NULL;
1545 }
1546 
1547 /*
1548  * When a source quence is received, we are being notifed of congestion.
1549  * Close the congestion window down to the Loss Window (one segment).
1550  * We will gradually open it again as we proceed.
1551  */
1552 void
1553 tcp_quench(inp, errno)
1554 	struct inpcb *inp;
1555 	int errno;
1556 {
1557 	struct tcpcb *tp = intotcpcb(inp);
1558 
1559 	if (tp)
1560 		tp->snd_cwnd = tp->t_segsz;
1561 }
1562 #endif
1563 
1564 #ifdef INET6
1565 void
1566 tcp6_quench(in6p, errno)
1567 	struct in6pcb *in6p;
1568 	int errno;
1569 {
1570 	struct tcpcb *tp = in6totcpcb(in6p);
1571 
1572 	if (tp)
1573 		tp->snd_cwnd = tp->t_segsz;
1574 }
1575 #endif
1576 
1577 #ifdef INET
1578 /*
1579  * Path MTU Discovery handlers.
1580  */
1581 void
1582 tcp_mtudisc_callback(faddr)
1583 	struct in_addr faddr;
1584 {
1585 #ifdef INET6
1586 	struct in6_addr in6;
1587 #endif
1588 
1589 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1590 #ifdef INET6
1591 	memset(&in6, 0, sizeof(in6));
1592 	in6.s6_addr16[5] = 0xffff;
1593 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1594 	tcp6_mtudisc_callback(&in6);
1595 #endif
1596 }
1597 
1598 /*
1599  * On receipt of path MTU corrections, flush old route and replace it
1600  * with the new one.  Retransmit all unacknowledged packets, to ensure
1601  * that all packets will be received.
1602  */
1603 void
1604 tcp_mtudisc(inp, errno)
1605 	struct inpcb *inp;
1606 	int errno;
1607 {
1608 	struct tcpcb *tp = intotcpcb(inp);
1609 	struct rtentry *rt = in_pcbrtentry(inp);
1610 
1611 	if (tp != 0) {
1612 		if (rt != 0) {
1613 			/*
1614 			 * If this was not a host route, remove and realloc.
1615 			 */
1616 			if ((rt->rt_flags & RTF_HOST) == 0) {
1617 				in_rtchange(inp, errno);
1618 				if ((rt = in_pcbrtentry(inp)) == 0)
1619 					return;
1620 			}
1621 
1622 			/*
1623 			 * Slow start out of the error condition.  We
1624 			 * use the MTU because we know it's smaller
1625 			 * than the previously transmitted segment.
1626 			 *
1627 			 * Note: This is more conservative than the
1628 			 * suggestion in draft-floyd-incr-init-win-03.
1629 			 */
1630 			if (rt->rt_rmx.rmx_mtu != 0)
1631 				tp->snd_cwnd =
1632 				    TCP_INITIAL_WINDOW(tcp_init_win,
1633 				    rt->rt_rmx.rmx_mtu);
1634 		}
1635 
1636 		/*
1637 		 * Resend unacknowledged packets.
1638 		 */
1639 		tp->snd_nxt = tp->snd_una;
1640 		tcp_output(tp);
1641 	}
1642 }
1643 #endif
1644 
1645 #ifdef INET6
1646 /*
1647  * Path MTU Discovery handlers.
1648  */
1649 void
1650 tcp6_mtudisc_callback(faddr)
1651 	struct in6_addr *faddr;
1652 {
1653 	struct sockaddr_in6 sin6;
1654 
1655 	bzero(&sin6, sizeof(sin6));
1656 	sin6.sin6_family = AF_INET6;
1657 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1658 	sin6.sin6_addr = *faddr;
1659 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1660 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1661 }
1662 
1663 void
1664 tcp6_mtudisc(in6p, errno)
1665 	struct in6pcb *in6p;
1666 	int errno;
1667 {
1668 	struct tcpcb *tp = in6totcpcb(in6p);
1669 	struct rtentry *rt = in6_pcbrtentry(in6p);
1670 
1671 	if (tp != 0) {
1672 		if (rt != 0) {
1673 			/*
1674 			 * If this was not a host route, remove and realloc.
1675 			 */
1676 			if ((rt->rt_flags & RTF_HOST) == 0) {
1677 				in6_rtchange(in6p, errno);
1678 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1679 					return;
1680 			}
1681 
1682 			/*
1683 			 * Slow start out of the error condition.  We
1684 			 * use the MTU because we know it's smaller
1685 			 * than the previously transmitted segment.
1686 			 *
1687 			 * Note: This is more conservative than the
1688 			 * suggestion in draft-floyd-incr-init-win-03.
1689 			 */
1690 			if (rt->rt_rmx.rmx_mtu != 0)
1691 				tp->snd_cwnd =
1692 				    TCP_INITIAL_WINDOW(tcp_init_win,
1693 				    rt->rt_rmx.rmx_mtu);
1694 		}
1695 
1696 		/*
1697 		 * Resend unacknowledged packets.
1698 		 */
1699 		tp->snd_nxt = tp->snd_una;
1700 		tcp_output(tp);
1701 	}
1702 }
1703 #endif /* INET6 */
1704 
1705 /*
1706  * Compute the MSS to advertise to the peer.  Called only during
1707  * the 3-way handshake.  If we are the server (peer initiated
1708  * connection), we are called with a pointer to the interface
1709  * on which the SYN packet arrived.  If we are the client (we
1710  * initiated connection), we are called with a pointer to the
1711  * interface out which this connection should go.
1712  *
1713  * NOTE: Do not subtract IP option/extension header size nor IPsec
1714  * header size from MSS advertisement.  MSS option must hold the maximum
1715  * segment size we can accept, so it must always be:
1716  *	 max(if mtu) - ip header - tcp header
1717  */
1718 u_long
1719 tcp_mss_to_advertise(ifp, af)
1720 	const struct ifnet *ifp;
1721 	int af;
1722 {
1723 	extern u_long in_maxmtu;
1724 	u_long mss = 0;
1725 	u_long hdrsiz;
1726 
1727 	/*
1728 	 * In order to avoid defeating path MTU discovery on the peer,
1729 	 * we advertise the max MTU of all attached networks as our MSS,
1730 	 * per RFC 1191, section 3.1.
1731 	 *
1732 	 * We provide the option to advertise just the MTU of
1733 	 * the interface on which we hope this connection will
1734 	 * be receiving.  If we are responding to a SYN, we
1735 	 * will have a pretty good idea about this, but when
1736 	 * initiating a connection there is a bit more doubt.
1737 	 *
1738 	 * We also need to ensure that loopback has a large enough
1739 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1740 	 */
1741 
1742 	if (ifp != NULL)
1743 		switch (af) {
1744 		case AF_INET:
1745 			mss = ifp->if_mtu;
1746 			break;
1747 #ifdef INET6
1748 		case AF_INET6:
1749 			mss = IN6_LINKMTU(ifp);
1750 			break;
1751 #endif
1752 		}
1753 
1754 	if (tcp_mss_ifmtu == 0)
1755 		switch (af) {
1756 		case AF_INET:
1757 			mss = max(in_maxmtu, mss);
1758 			break;
1759 #ifdef INET6
1760 		case AF_INET6:
1761 			mss = max(in6_maxmtu, mss);
1762 			break;
1763 #endif
1764 		}
1765 
1766 	switch (af) {
1767 	case AF_INET:
1768 		hdrsiz = sizeof(struct ip);
1769 		break;
1770 #ifdef INET6
1771 	case AF_INET6:
1772 		hdrsiz = sizeof(struct ip6_hdr);
1773 		break;
1774 #endif
1775 	default:
1776 		hdrsiz = 0;
1777 		break;
1778 	}
1779 	hdrsiz += sizeof(struct tcphdr);
1780 	if (mss > hdrsiz)
1781 		mss -= hdrsiz;
1782 
1783 	mss = max(tcp_mssdflt, mss);
1784 	return (mss);
1785 }
1786 
1787 /*
1788  * Set connection variables based on the peer's advertised MSS.
1789  * We are passed the TCPCB for the actual connection.  If we
1790  * are the server, we are called by the compressed state engine
1791  * when the 3-way handshake is complete.  If we are the client,
1792  * we are called when we receive the SYN,ACK from the server.
1793  *
1794  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1795  * before this routine is called!
1796  */
1797 void
1798 tcp_mss_from_peer(tp, offer)
1799 	struct tcpcb *tp;
1800 	int offer;
1801 {
1802 	struct socket *so;
1803 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1804 	struct rtentry *rt;
1805 #endif
1806 	u_long bufsize;
1807 	int mss;
1808 
1809 #ifdef DIAGNOSTIC
1810 	if (tp->t_inpcb && tp->t_in6pcb)
1811 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1812 #endif
1813 	so = NULL;
1814 	rt = NULL;
1815 #ifdef INET
1816 	if (tp->t_inpcb) {
1817 		so = tp->t_inpcb->inp_socket;
1818 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1819 		rt = in_pcbrtentry(tp->t_inpcb);
1820 #endif
1821 	}
1822 #endif
1823 #ifdef INET6
1824 	if (tp->t_in6pcb) {
1825 		so = tp->t_in6pcb->in6p_socket;
1826 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1827 		rt = in6_pcbrtentry(tp->t_in6pcb);
1828 #endif
1829 	}
1830 #endif
1831 
1832 	/*
1833 	 * As per RFC1122, use the default MSS value, unless they
1834 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1835 	 */
1836 	mss = tcp_mssdflt;
1837 	if (offer)
1838 		mss = offer;
1839 	mss = max(mss, 32);		/* sanity */
1840 	tp->t_peermss = mss;
1841 	mss -= tcp_optlen(tp);
1842 #ifdef INET
1843 	if (tp->t_inpcb)
1844 		mss -= ip_optlen(tp->t_inpcb);
1845 #endif
1846 #ifdef INET6
1847 	if (tp->t_in6pcb)
1848 		mss -= ip6_optlen(tp->t_in6pcb);
1849 #endif
1850 
1851 	/*
1852 	 * If there's a pipesize, change the socket buffer to that size.
1853 	 * Make the socket buffer an integral number of MSS units.  If
1854 	 * the MSS is larger than the socket buffer, artificially decrease
1855 	 * the MSS.
1856 	 */
1857 #ifdef RTV_SPIPE
1858 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1859 		bufsize = rt->rt_rmx.rmx_sendpipe;
1860 	else
1861 #endif
1862 		bufsize = so->so_snd.sb_hiwat;
1863 	if (bufsize < mss)
1864 		mss = bufsize;
1865 	else {
1866 		bufsize = roundup(bufsize, mss);
1867 		if (bufsize > sb_max)
1868 			bufsize = sb_max;
1869 		(void) sbreserve(&so->so_snd, bufsize);
1870 	}
1871 	tp->t_segsz = mss;
1872 
1873 #ifdef RTV_SSTHRESH
1874 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1875 		/*
1876 		 * There's some sort of gateway or interface buffer
1877 		 * limit on the path.  Use this to set the slow
1878 		 * start threshold, but set the threshold to no less
1879 		 * than 2 * MSS.
1880 		 */
1881 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1882 	}
1883 #endif
1884 }
1885 
1886 /*
1887  * Processing necessary when a TCP connection is established.
1888  */
1889 void
1890 tcp_established(tp)
1891 	struct tcpcb *tp;
1892 {
1893 	struct socket *so;
1894 #ifdef RTV_RPIPE
1895 	struct rtentry *rt;
1896 #endif
1897 	u_long bufsize;
1898 
1899 #ifdef DIAGNOSTIC
1900 	if (tp->t_inpcb && tp->t_in6pcb)
1901 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1902 #endif
1903 	so = NULL;
1904 	rt = NULL;
1905 #ifdef INET
1906 	if (tp->t_inpcb) {
1907 		so = tp->t_inpcb->inp_socket;
1908 #if defined(RTV_RPIPE)
1909 		rt = in_pcbrtentry(tp->t_inpcb);
1910 #endif
1911 	}
1912 #endif
1913 #ifdef INET6
1914 	if (tp->t_in6pcb) {
1915 		so = tp->t_in6pcb->in6p_socket;
1916 #if defined(RTV_RPIPE)
1917 		rt = in6_pcbrtentry(tp->t_in6pcb);
1918 #endif
1919 	}
1920 #endif
1921 
1922 	tp->t_state = TCPS_ESTABLISHED;
1923 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1924 
1925 #ifdef RTV_RPIPE
1926 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1927 		bufsize = rt->rt_rmx.rmx_recvpipe;
1928 	else
1929 #endif
1930 		bufsize = so->so_rcv.sb_hiwat;
1931 	if (bufsize > tp->t_ourmss) {
1932 		bufsize = roundup(bufsize, tp->t_ourmss);
1933 		if (bufsize > sb_max)
1934 			bufsize = sb_max;
1935 		(void) sbreserve(&so->so_rcv, bufsize);
1936 	}
1937 }
1938 
1939 /*
1940  * Check if there's an initial rtt or rttvar.  Convert from the
1941  * route-table units to scaled multiples of the slow timeout timer.
1942  * Called only during the 3-way handshake.
1943  */
1944 void
1945 tcp_rmx_rtt(tp)
1946 	struct tcpcb *tp;
1947 {
1948 #ifdef RTV_RTT
1949 	struct rtentry *rt = NULL;
1950 	int rtt;
1951 
1952 #ifdef DIAGNOSTIC
1953 	if (tp->t_inpcb && tp->t_in6pcb)
1954 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1955 #endif
1956 #ifdef INET
1957 	if (tp->t_inpcb)
1958 		rt = in_pcbrtentry(tp->t_inpcb);
1959 #endif
1960 #ifdef INET6
1961 	if (tp->t_in6pcb)
1962 		rt = in6_pcbrtentry(tp->t_in6pcb);
1963 #endif
1964 	if (rt == NULL)
1965 		return;
1966 
1967 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1968 		/*
1969 		 * XXX The lock bit for MTU indicates that the value
1970 		 * is also a minimum value; this is subject to time.
1971 		 */
1972 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1973 			TCPT_RANGESET(tp->t_rttmin,
1974 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1975 			    TCPTV_MIN, TCPTV_REXMTMAX);
1976 		tp->t_srtt = rtt /
1977 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1978 		if (rt->rt_rmx.rmx_rttvar) {
1979 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1980 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1981 				(TCP_RTTVAR_SHIFT + 2));
1982 		} else {
1983 			/* Default variation is +- 1 rtt */
1984 			tp->t_rttvar =
1985 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1986 		}
1987 		TCPT_RANGESET(tp->t_rxtcur,
1988 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1989 		    tp->t_rttmin, TCPTV_REXMTMAX);
1990 	}
1991 #endif
1992 }
1993 
1994 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1995 #if NRND > 0
1996 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1997 #endif
1998 
1999 /*
2000  * Get a new sequence value given a tcp control block
2001  */
2002 tcp_seq
2003 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2004 {
2005 
2006 #ifdef INET
2007 	if (tp->t_inpcb != NULL) {
2008 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2009 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2010 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2011 		    addin));
2012 	}
2013 #endif
2014 #ifdef INET6
2015 	if (tp->t_in6pcb != NULL) {
2016 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2017 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2018 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2019 		    addin));
2020 	}
2021 #endif
2022 	/* Not possible. */
2023 	panic("tcp_new_iss");
2024 }
2025 
2026 /*
2027  * This routine actually generates a new TCP initial sequence number.
2028  */
2029 tcp_seq
2030 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2031     size_t addrsz, tcp_seq addin)
2032 {
2033 	tcp_seq tcp_iss;
2034 
2035 #if NRND > 0
2036 	static int beenhere;
2037 
2038 	/*
2039 	 * If we haven't been here before, initialize our cryptographic
2040 	 * hash secret.
2041 	 */
2042 	if (beenhere == 0) {
2043 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2044 		    RND_EXTRACT_ANY);
2045 		beenhere = 1;
2046 	}
2047 
2048 	if (tcp_do_rfc1948) {
2049 		MD5_CTX ctx;
2050 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2051 
2052 		/*
2053 		 * Compute the base value of the ISS.  It is a hash
2054 		 * of (saddr, sport, daddr, dport, secret).
2055 		 */
2056 		MD5Init(&ctx);
2057 
2058 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2059 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2060 
2061 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2062 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2063 
2064 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2065 
2066 		MD5Final(hash, &ctx);
2067 
2068 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2069 
2070 		/*
2071 		 * Now increment our "timer", and add it in to
2072 		 * the computed value.
2073 		 *
2074 		 * XXX Use `addin'?
2075 		 * XXX TCP_ISSINCR too large to use?
2076 		 */
2077 		tcp_iss_seq += TCP_ISSINCR;
2078 #ifdef TCPISS_DEBUG
2079 		printf("ISS hash 0x%08x, ", tcp_iss);
2080 #endif
2081 		tcp_iss += tcp_iss_seq + addin;
2082 #ifdef TCPISS_DEBUG
2083 		printf("new ISS 0x%08x\n", tcp_iss);
2084 #endif
2085 	} else
2086 #endif /* NRND > 0 */
2087 	{
2088 		/*
2089 		 * Randomize.
2090 		 */
2091 #if NRND > 0
2092 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2093 #else
2094 		tcp_iss = arc4random();
2095 #endif
2096 
2097 		/*
2098 		 * If we were asked to add some amount to a known value,
2099 		 * we will take a random value obtained above, mask off
2100 		 * the upper bits, and add in the known value.  We also
2101 		 * add in a constant to ensure that we are at least a
2102 		 * certain distance from the original value.
2103 		 *
2104 		 * This is used when an old connection is in timed wait
2105 		 * and we have a new one coming in, for instance.
2106 		 */
2107 		if (addin != 0) {
2108 #ifdef TCPISS_DEBUG
2109 			printf("Random %08x, ", tcp_iss);
2110 #endif
2111 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2112 			tcp_iss += addin + TCP_ISSINCR;
2113 #ifdef TCPISS_DEBUG
2114 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2115 #endif
2116 		} else {
2117 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2118 			tcp_iss += tcp_iss_seq;
2119 			tcp_iss_seq += TCP_ISSINCR;
2120 #ifdef TCPISS_DEBUG
2121 			printf("ISS %08x\n", tcp_iss);
2122 #endif
2123 		}
2124 	}
2125 
2126 	if (tcp_compat_42) {
2127 		/*
2128 		 * Limit it to the positive range for really old TCP
2129 		 * implementations.
2130 		 * Just AND off the top bit instead of checking if
2131 		 * is set first - saves a branch 50% of the time.
2132 		 */
2133 		tcp_iss &= 0x7fffffff;		/* XXX */
2134 	}
2135 
2136 	return (tcp_iss);
2137 }
2138 
2139 #if defined(IPSEC) || defined(FAST_IPSEC)
2140 /* compute ESP/AH header size for TCP, including outer IP header. */
2141 size_t
2142 ipsec4_hdrsiz_tcp(tp)
2143 	struct tcpcb *tp;
2144 {
2145 	struct inpcb *inp;
2146 	size_t hdrsiz;
2147 
2148 	/* XXX mapped addr case (tp->t_in6pcb) */
2149 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2150 		return 0;
2151 	switch (tp->t_family) {
2152 	case AF_INET:
2153 		/* XXX: should use currect direction. */
2154 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2155 		break;
2156 	default:
2157 		hdrsiz = 0;
2158 		break;
2159 	}
2160 
2161 	return hdrsiz;
2162 }
2163 
2164 #ifdef INET6
2165 size_t
2166 ipsec6_hdrsiz_tcp(tp)
2167 	struct tcpcb *tp;
2168 {
2169 	struct in6pcb *in6p;
2170 	size_t hdrsiz;
2171 
2172 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2173 		return 0;
2174 	switch (tp->t_family) {
2175 	case AF_INET6:
2176 		/* XXX: should use currect direction. */
2177 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2178 		break;
2179 	case AF_INET:
2180 		/* mapped address case - tricky */
2181 	default:
2182 		hdrsiz = 0;
2183 		break;
2184 	}
2185 
2186 	return hdrsiz;
2187 }
2188 #endif
2189 #endif /*IPSEC*/
2190 
2191 /*
2192  * Determine the length of the TCP options for this connection.
2193  *
2194  * XXX:  What do we do for SACK, when we add that?  Just reserve
2195  *       all of the space?  Otherwise we can't exactly be incrementing
2196  *       cwnd by an amount that varies depending on the amount we last
2197  *       had to SACK!
2198  */
2199 
2200 u_int
2201 tcp_optlen(tp)
2202 	struct tcpcb *tp;
2203 {
2204 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2205 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2206 		return TCPOLEN_TSTAMP_APPA;
2207 	else
2208 		return 0;
2209 }
2210