xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 69b6d498973bb4d7230c2d3c12bd9a032738ec8e)
1 /*	$NetBSD: tcp_subr.c,v 1.193 2005/07/20 08:05:43 he 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 1.193 2005/07/20 08:05:43 he 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 #include <netkey/key.h>
159 #endif /*IPSEC*/
160 
161 #ifdef FAST_IPSEC
162 #include <netipsec/ipsec.h>
163 #include <netipsec/xform.h>
164 #ifdef INET6
165 #include <netipsec/ipsec6.h>
166 #endif
167  #include <netipsec/key.h>
168 #endif	/* FAST_IPSEC*/
169 
170 
171 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
172 struct	tcpstat tcpstat;	/* tcp statistics */
173 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
174 
175 /* patchable/settable parameters for tcp */
176 int 	tcp_mssdflt = TCP_MSS;
177 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
178 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
179 #if NRND > 0
180 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
181 #endif
182 int	tcp_do_sack = 1;	/* selective acknowledgement */
183 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
184 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
185 int	tcp_do_newreno = 1;	/* Use the New Reno algorithms */
186 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
187 #ifndef TCP_INIT_WIN
188 #define	TCP_INIT_WIN	0	/* initial slow start window */
189 #endif
190 #ifndef TCP_INIT_WIN_LOCAL
191 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
192 #endif
193 int	tcp_init_win = TCP_INIT_WIN;
194 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
195 int	tcp_mss_ifmtu = 0;
196 #ifdef TCP_COMPAT_42
197 int	tcp_compat_42 = 1;
198 #else
199 int	tcp_compat_42 = 0;
200 #endif
201 int	tcp_rst_ppslim = 100;	/* 100pps */
202 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
203 int	tcp_sack_tp_maxholes = 32;
204 int	tcp_sack_globalmaxholes = 1024;
205 int	tcp_sack_globalholes = 0;
206 
207 
208 /* tcb hash */
209 #ifndef TCBHASHSIZE
210 #define	TCBHASHSIZE	128
211 #endif
212 int	tcbhashsize = TCBHASHSIZE;
213 
214 /* syn hash parameters */
215 #define	TCP_SYN_HASH_SIZE	293
216 #define	TCP_SYN_BUCKET_SIZE	35
217 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
218 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
219 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
220 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
221 
222 int	tcp_freeq(struct tcpcb *);
223 
224 #ifdef INET
225 void	tcp_mtudisc_callback(struct in_addr);
226 #endif
227 #ifdef INET6
228 void	tcp6_mtudisc_callback(struct in6_addr *);
229 #endif
230 
231 #ifdef INET6
232 void	tcp6_mtudisc(struct in6pcb *, int);
233 #endif
234 
235 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL);
236 
237 #ifdef TCP_CSUM_COUNTERS
238 #include <sys/device.h>
239 
240 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
241     NULL, "tcp", "hwcsum bad");
242 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243     NULL, "tcp", "hwcsum ok");
244 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245     NULL, "tcp", "hwcsum data");
246 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247     NULL, "tcp", "swcsum");
248 
249 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
250 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
251 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
252 EVCNT_ATTACH_STATIC(tcp_swcsum);
253 #endif /* TCP_CSUM_COUNTERS */
254 
255 
256 #ifdef TCP_OUTPUT_COUNTERS
257 #include <sys/device.h>
258 
259 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
260     NULL, "tcp", "output big header");
261 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "output predict hit");
263 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "output predict miss");
265 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     NULL, "tcp", "output copy small");
267 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     NULL, "tcp", "output copy big");
269 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
270     NULL, "tcp", "output reference big");
271 
272 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
273 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
274 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
275 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
276 EVCNT_ATTACH_STATIC(tcp_output_copybig);
277 EVCNT_ATTACH_STATIC(tcp_output_refbig);
278 
279 #endif /* TCP_OUTPUT_COUNTERS */
280 
281 #ifdef TCP_REASS_COUNTERS
282 #include <sys/device.h>
283 
284 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
285     NULL, "tcp_reass", "calls");
286 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
287     &tcp_reass_, "tcp_reass", "insert into empty queue");
288 struct evcnt tcp_reass_iteration[8] = {
289     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
290     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
291     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
292     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
293     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
294     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
295     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
296     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
297 };
298 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
299     &tcp_reass_, "tcp_reass", "prepend to first");
300 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
301     &tcp_reass_, "tcp_reass", "prepend");
302 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303     &tcp_reass_, "tcp_reass", "insert");
304 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305     &tcp_reass_, "tcp_reass", "insert at tail");
306 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
307     &tcp_reass_, "tcp_reass", "append");
308 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
309     &tcp_reass_, "tcp_reass", "append to tail fragment");
310 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
311     &tcp_reass_, "tcp_reass", "overlap at end");
312 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
313     &tcp_reass_, "tcp_reass", "overlap at start");
314 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
315     &tcp_reass_, "tcp_reass", "duplicate segment");
316 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
317     &tcp_reass_, "tcp_reass", "duplicate fragment");
318 
319 EVCNT_ATTACH_STATIC(tcp_reass_);
320 EVCNT_ATTACH_STATIC(tcp_reass_empty);
321 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
322 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
323 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
324 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
325 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
326 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
327 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
328 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
329 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
330 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
331 EVCNT_ATTACH_STATIC(tcp_reass_insert);
332 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
333 EVCNT_ATTACH_STATIC(tcp_reass_append);
334 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
335 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
336 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
337 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
338 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
339 
340 #endif /* TCP_REASS_COUNTERS */
341 
342 #ifdef MBUFTRACE
343 struct mowner tcp_mowner = { "tcp" };
344 struct mowner tcp_rx_mowner = { "tcp", "rx" };
345 struct mowner tcp_tx_mowner = { "tcp", "tx" };
346 #endif
347 
348 /*
349  * Tcp initialization
350  */
351 void
352 tcp_init(void)
353 {
354 	int hlen;
355 
356 	/* Initialize the TCPCB template. */
357 	tcp_tcpcb_template();
358 
359 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
360 
361 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
362 #ifdef INET6
363 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
364 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
365 #endif
366 	if (max_protohdr < hlen)
367 		max_protohdr = hlen;
368 	if (max_linkhdr + hlen > MHLEN)
369 		panic("tcp_init");
370 
371 #ifdef INET
372 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
373 #endif
374 #ifdef INET6
375 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
376 #endif
377 
378 	/* Initialize timer state. */
379 	tcp_timer_init();
380 
381 	/* Initialize the compressed state engine. */
382 	syn_cache_init();
383 
384 	MOWNER_ATTACH(&tcp_tx_mowner);
385 	MOWNER_ATTACH(&tcp_rx_mowner);
386 	MOWNER_ATTACH(&tcp_mowner);
387 }
388 
389 /*
390  * Create template to be used to send tcp packets on a connection.
391  * Call after host entry created, allocates an mbuf and fills
392  * in a skeletal tcp/ip header, minimizing the amount of work
393  * necessary when the connection is used.
394  */
395 struct mbuf *
396 tcp_template(struct tcpcb *tp)
397 {
398 	struct inpcb *inp = tp->t_inpcb;
399 #ifdef INET6
400 	struct in6pcb *in6p = tp->t_in6pcb;
401 #endif
402 	struct tcphdr *n;
403 	struct mbuf *m;
404 	int hlen;
405 
406 	switch (tp->t_family) {
407 	case AF_INET:
408 		hlen = sizeof(struct ip);
409 		if (inp)
410 			break;
411 #ifdef INET6
412 		if (in6p) {
413 			/* mapped addr case */
414 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
415 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
416 				break;
417 		}
418 #endif
419 		return NULL;	/*EINVAL*/
420 #ifdef INET6
421 	case AF_INET6:
422 		hlen = sizeof(struct ip6_hdr);
423 		if (in6p) {
424 			/* more sainty check? */
425 			break;
426 		}
427 		return NULL;	/*EINVAL*/
428 #endif
429 	default:
430 		hlen = 0;	/*pacify gcc*/
431 		return NULL;	/*EAFNOSUPPORT*/
432 	}
433 #ifdef DIAGNOSTIC
434 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
435 		panic("mclbytes too small for t_template");
436 #endif
437 	m = tp->t_template;
438 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
439 		;
440 	else {
441 		if (m)
442 			m_freem(m);
443 		m = tp->t_template = NULL;
444 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
445 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
446 			MCLGET(m, M_DONTWAIT);
447 			if ((m->m_flags & M_EXT) == 0) {
448 				m_free(m);
449 				m = NULL;
450 			}
451 		}
452 		if (m == NULL)
453 			return NULL;
454 		MCLAIM(m, &tcp_mowner);
455 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
456 	}
457 
458 	bzero(mtod(m, caddr_t), m->m_len);
459 
460 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
461 
462 	switch (tp->t_family) {
463 	case AF_INET:
464 	    {
465 		struct ipovly *ipov;
466 		mtod(m, struct ip *)->ip_v = 4;
467 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
468 		ipov = mtod(m, struct ipovly *);
469 		ipov->ih_pr = IPPROTO_TCP;
470 		ipov->ih_len = htons(sizeof(struct tcphdr));
471 		if (inp) {
472 			ipov->ih_src = inp->inp_laddr;
473 			ipov->ih_dst = inp->inp_faddr;
474 		}
475 #ifdef INET6
476 		else if (in6p) {
477 			/* mapped addr case */
478 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
479 				sizeof(ipov->ih_src));
480 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
481 				sizeof(ipov->ih_dst));
482 		}
483 #endif
484 		/*
485 		 * Compute the pseudo-header portion of the checksum
486 		 * now.  We incrementally add in the TCP option and
487 		 * payload lengths later, and then compute the TCP
488 		 * checksum right before the packet is sent off onto
489 		 * the wire.
490 		 */
491 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
492 		    ipov->ih_dst.s_addr,
493 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
494 		break;
495 	    }
496 #ifdef INET6
497 	case AF_INET6:
498 	    {
499 		struct ip6_hdr *ip6;
500 		mtod(m, struct ip *)->ip_v = 6;
501 		ip6 = mtod(m, struct ip6_hdr *);
502 		ip6->ip6_nxt = IPPROTO_TCP;
503 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
504 		ip6->ip6_src = in6p->in6p_laddr;
505 		ip6->ip6_dst = in6p->in6p_faddr;
506 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
507 		if (ip6_auto_flowlabel) {
508 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
509 			ip6->ip6_flow |=
510 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
511 		}
512 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
513 		ip6->ip6_vfc |= IPV6_VERSION;
514 
515 		/*
516 		 * Compute the pseudo-header portion of the checksum
517 		 * now.  We incrementally add in the TCP option and
518 		 * payload lengths later, and then compute the TCP
519 		 * checksum right before the packet is sent off onto
520 		 * the wire.
521 		 */
522 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
523 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
524 		    htonl(IPPROTO_TCP));
525 		break;
526 	    }
527 #endif
528 	}
529 	if (inp) {
530 		n->th_sport = inp->inp_lport;
531 		n->th_dport = inp->inp_fport;
532 	}
533 #ifdef INET6
534 	else if (in6p) {
535 		n->th_sport = in6p->in6p_lport;
536 		n->th_dport = in6p->in6p_fport;
537 	}
538 #endif
539 	n->th_seq = 0;
540 	n->th_ack = 0;
541 	n->th_x2 = 0;
542 	n->th_off = 5;
543 	n->th_flags = 0;
544 	n->th_win = 0;
545 	n->th_urp = 0;
546 	return (m);
547 }
548 
549 /*
550  * Send a single message to the TCP at address specified by
551  * the given TCP/IP header.  If m == 0, then we make a copy
552  * of the tcpiphdr at ti and send directly to the addressed host.
553  * This is used to force keep alive messages out using the TCP
554  * template for a connection tp->t_template.  If flags are given
555  * then we send a message back to the TCP which originated the
556  * segment ti, and discard the mbuf containing it and any other
557  * attached mbufs.
558  *
559  * In any case the ack and sequence number of the transmitted
560  * segment are as specified by the parameters.
561  */
562 int
563 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m,
564     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
565 {
566 	struct route *ro;
567 	int error, tlen, win = 0;
568 	int hlen;
569 	struct ip *ip;
570 #ifdef INET6
571 	struct ip6_hdr *ip6;
572 #endif
573 	int family;	/* family on packet, not inpcb/in6pcb! */
574 	struct tcphdr *th;
575 	struct socket *so;
576 
577 	if (tp != NULL && (flags & TH_RST) == 0) {
578 #ifdef DIAGNOSTIC
579 		if (tp->t_inpcb && tp->t_in6pcb)
580 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
581 #endif
582 #ifdef INET
583 		if (tp->t_inpcb)
584 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
585 #endif
586 #ifdef INET6
587 		if (tp->t_in6pcb)
588 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
589 #endif
590 	}
591 
592 	th = NULL;	/* Quell uninitialized warning */
593 	ip = NULL;
594 #ifdef INET6
595 	ip6 = NULL;
596 #endif
597 	if (m == 0) {
598 		if (!template)
599 			return EINVAL;
600 
601 		/* get family information from template */
602 		switch (mtod(template, struct ip *)->ip_v) {
603 		case 4:
604 			family = AF_INET;
605 			hlen = sizeof(struct ip);
606 			break;
607 #ifdef INET6
608 		case 6:
609 			family = AF_INET6;
610 			hlen = sizeof(struct ip6_hdr);
611 			break;
612 #endif
613 		default:
614 			return EAFNOSUPPORT;
615 		}
616 
617 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
618 		if (m) {
619 			MCLAIM(m, &tcp_tx_mowner);
620 			MCLGET(m, M_DONTWAIT);
621 			if ((m->m_flags & M_EXT) == 0) {
622 				m_free(m);
623 				m = NULL;
624 			}
625 		}
626 		if (m == NULL)
627 			return (ENOBUFS);
628 
629 		if (tcp_compat_42)
630 			tlen = 1;
631 		else
632 			tlen = 0;
633 
634 		m->m_data += max_linkhdr;
635 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
636 			template->m_len);
637 		switch (family) {
638 		case AF_INET:
639 			ip = mtod(m, struct ip *);
640 			th = (struct tcphdr *)(ip + 1);
641 			break;
642 #ifdef INET6
643 		case AF_INET6:
644 			ip6 = mtod(m, struct ip6_hdr *);
645 			th = (struct tcphdr *)(ip6 + 1);
646 			break;
647 #endif
648 #if 0
649 		default:
650 			/* noone will visit here */
651 			m_freem(m);
652 			return EAFNOSUPPORT;
653 #endif
654 		}
655 		flags = TH_ACK;
656 	} else {
657 
658 		if ((m->m_flags & M_PKTHDR) == 0) {
659 #if 0
660 			printf("non PKTHDR to tcp_respond\n");
661 #endif
662 			m_freem(m);
663 			return EINVAL;
664 		}
665 #ifdef DIAGNOSTIC
666 		if (!th0)
667 			panic("th0 == NULL in tcp_respond");
668 #endif
669 
670 		/* get family information from m */
671 		switch (mtod(m, struct ip *)->ip_v) {
672 		case 4:
673 			family = AF_INET;
674 			hlen = sizeof(struct ip);
675 			ip = mtod(m, struct ip *);
676 			break;
677 #ifdef INET6
678 		case 6:
679 			family = AF_INET6;
680 			hlen = sizeof(struct ip6_hdr);
681 			ip6 = mtod(m, struct ip6_hdr *);
682 			break;
683 #endif
684 		default:
685 			m_freem(m);
686 			return EAFNOSUPPORT;
687 		}
688 		/* clear h/w csum flags inherited from rx packet */
689 		m->m_pkthdr.csum_flags = 0;
690 
691 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
692 			tlen = sizeof(*th0);
693 		else
694 			tlen = th0->th_off << 2;
695 
696 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
697 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
698 			m->m_len = hlen + tlen;
699 			m_freem(m->m_next);
700 			m->m_next = NULL;
701 		} else {
702 			struct mbuf *n;
703 
704 #ifdef DIAGNOSTIC
705 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
706 				m_freem(m);
707 				return EMSGSIZE;
708 			}
709 #endif
710 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
711 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
712 				MCLGET(n, M_DONTWAIT);
713 				if ((n->m_flags & M_EXT) == 0) {
714 					m_freem(n);
715 					n = NULL;
716 				}
717 			}
718 			if (!n) {
719 				m_freem(m);
720 				return ENOBUFS;
721 			}
722 
723 			MCLAIM(n, &tcp_tx_mowner);
724 			n->m_data += max_linkhdr;
725 			n->m_len = hlen + tlen;
726 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
727 			m_copyback(n, hlen, tlen, (caddr_t)th0);
728 
729 			m_freem(m);
730 			m = n;
731 			n = NULL;
732 		}
733 
734 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
735 		switch (family) {
736 		case AF_INET:
737 			ip = mtod(m, struct ip *);
738 			th = (struct tcphdr *)(ip + 1);
739 			ip->ip_p = IPPROTO_TCP;
740 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
741 			ip->ip_p = IPPROTO_TCP;
742 			break;
743 #ifdef INET6
744 		case AF_INET6:
745 			ip6 = mtod(m, struct ip6_hdr *);
746 			th = (struct tcphdr *)(ip6 + 1);
747 			ip6->ip6_nxt = IPPROTO_TCP;
748 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
749 			ip6->ip6_nxt = IPPROTO_TCP;
750 			break;
751 #endif
752 #if 0
753 		default:
754 			/* noone will visit here */
755 			m_freem(m);
756 			return EAFNOSUPPORT;
757 #endif
758 		}
759 		xchg(th->th_dport, th->th_sport, u_int16_t);
760 #undef xchg
761 		tlen = 0;	/*be friendly with the following code*/
762 	}
763 	th->th_seq = htonl(seq);
764 	th->th_ack = htonl(ack);
765 	th->th_x2 = 0;
766 	if ((flags & TH_SYN) == 0) {
767 		if (tp)
768 			win >>= tp->rcv_scale;
769 		if (win > TCP_MAXWIN)
770 			win = TCP_MAXWIN;
771 		th->th_win = htons((u_int16_t)win);
772 		th->th_off = sizeof (struct tcphdr) >> 2;
773 		tlen += sizeof(*th);
774 	} else
775 		tlen += th->th_off << 2;
776 	m->m_len = hlen + tlen;
777 	m->m_pkthdr.len = hlen + tlen;
778 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
779 	th->th_flags = flags;
780 	th->th_urp = 0;
781 
782 	switch (family) {
783 #ifdef INET
784 	case AF_INET:
785 	    {
786 		struct ipovly *ipov = (struct ipovly *)ip;
787 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
788 		ipov->ih_len = htons((u_int16_t)tlen);
789 
790 		th->th_sum = 0;
791 		th->th_sum = in_cksum(m, hlen + tlen);
792 		ip->ip_len = htons(hlen + tlen);
793 		ip->ip_ttl = ip_defttl;
794 		break;
795 	    }
796 #endif
797 #ifdef INET6
798 	case AF_INET6:
799 	    {
800 		th->th_sum = 0;
801 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
802 				tlen);
803 		ip6->ip6_plen = htons(tlen);
804 		if (tp && tp->t_in6pcb) {
805 			struct ifnet *oifp;
806 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
807 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
808 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
809 		} else
810 			ip6->ip6_hlim = ip6_defhlim;
811 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
812 		if (ip6_auto_flowlabel) {
813 			ip6->ip6_flow |=
814 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
815 		}
816 		break;
817 	    }
818 #endif
819 	}
820 
821 	if (tp && tp->t_inpcb)
822 		so = tp->t_inpcb->inp_socket;
823 #ifdef INET6
824 	else if (tp && tp->t_in6pcb)
825 		so = tp->t_in6pcb->in6p_socket;
826 #endif
827 	else
828 		so = NULL;
829 
830 	if (tp != NULL && tp->t_inpcb != NULL) {
831 		ro = &tp->t_inpcb->inp_route;
832 #ifdef DIAGNOSTIC
833 		if (family != AF_INET)
834 			panic("tcp_respond: address family mismatch");
835 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
836 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
837 			    ntohl(ip->ip_dst.s_addr),
838 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
839 		}
840 #endif
841 	}
842 #ifdef INET6
843 	else if (tp != NULL && tp->t_in6pcb != NULL) {
844 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
845 #ifdef DIAGNOSTIC
846 		if (family == AF_INET) {
847 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
848 				panic("tcp_respond: not mapped addr");
849 			if (bcmp(&ip->ip_dst,
850 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
851 			    sizeof(ip->ip_dst)) != 0) {
852 				panic("tcp_respond: ip_dst != in6p_faddr");
853 			}
854 		} else if (family == AF_INET6) {
855 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
856 			    &tp->t_in6pcb->in6p_faddr))
857 				panic("tcp_respond: ip6_dst != in6p_faddr");
858 		} else
859 			panic("tcp_respond: address family mismatch");
860 #endif
861 	}
862 #endif
863 	else
864 		ro = NULL;
865 
866 	switch (family) {
867 #ifdef INET
868 	case AF_INET:
869 		error = ip_output(m, NULL, ro,
870 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
871 		    (struct ip_moptions *)0, so);
872 		break;
873 #endif
874 #ifdef INET6
875 	case AF_INET6:
876 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
877 		    (struct ip6_moptions *)0, so, NULL);
878 		break;
879 #endif
880 	default:
881 		error = EAFNOSUPPORT;
882 		break;
883 	}
884 
885 	return (error);
886 }
887 
888 /*
889  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
890  * a bunch of members individually, we maintain this template for the
891  * static and mostly-static components of the TCPCB, and copy it into
892  * the new TCPCB instead.
893  */
894 static struct tcpcb tcpcb_template = {
895 	/*
896 	 * If TCP_NTIMERS ever changes, we'll need to update this
897 	 * initializer.
898 	 */
899 	.t_timer = {
900 		CALLOUT_INITIALIZER,
901 		CALLOUT_INITIALIZER,
902 		CALLOUT_INITIALIZER,
903 		CALLOUT_INITIALIZER,
904 	},
905 	.t_delack_ch = CALLOUT_INITIALIZER,
906 
907 	.t_srtt = TCPTV_SRTTBASE,
908 	.t_rttmin = TCPTV_MIN,
909 
910 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
911 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
912 	.snd_numholes = 0,
913 
914 	.t_partialacks = -1,
915 };
916 
917 /*
918  * Updates the TCPCB template whenever a parameter that would affect
919  * the template is changed.
920  */
921 void
922 tcp_tcpcb_template(void)
923 {
924 	struct tcpcb *tp = &tcpcb_template;
925 	int flags;
926 
927 	tp->t_peermss = tcp_mssdflt;
928 	tp->t_ourmss = tcp_mssdflt;
929 	tp->t_segsz = tcp_mssdflt;
930 
931 	flags = 0;
932 	if (tcp_do_rfc1323 && tcp_do_win_scale)
933 		flags |= TF_REQ_SCALE;
934 	if (tcp_do_rfc1323 && tcp_do_timestamps)
935 		flags |= TF_REQ_TSTMP;
936 	tp->t_flags = flags;
937 
938 	/*
939 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
940 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
941 	 * reasonable initial retransmit time.
942 	 */
943 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
944 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
945 	    TCPTV_MIN, TCPTV_REXMTMAX);
946 }
947 
948 /*
949  * Create a new TCP control block, making an
950  * empty reassembly queue and hooking it to the argument
951  * protocol control block.
952  */
953 /* family selects inpcb, or in6pcb */
954 struct tcpcb *
955 tcp_newtcpcb(int family, void *aux)
956 {
957 	struct tcpcb *tp;
958 	int i;
959 
960 	/* XXX Consider using a pool_cache for speed. */
961 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
962 	if (tp == NULL)
963 		return (NULL);
964 	memcpy(tp, &tcpcb_template, sizeof(*tp));
965 	TAILQ_INIT(&tp->segq);
966 	TAILQ_INIT(&tp->timeq);
967 	tp->t_family = family;		/* may be overridden later on */
968 	TAILQ_INIT(&tp->snd_holes);
969 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
970 
971 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
972 	for (i = 0; i < TCPT_NTIMERS; i++)
973 		TCP_TIMER_INIT(tp, i);
974 
975 	switch (family) {
976 	case AF_INET:
977 	    {
978 		struct inpcb *inp = (struct inpcb *)aux;
979 
980 		inp->inp_ip.ip_ttl = ip_defttl;
981 		inp->inp_ppcb = (caddr_t)tp;
982 
983 		tp->t_inpcb = inp;
984 		tp->t_mtudisc = ip_mtudisc;
985 		break;
986 	    }
987 #ifdef INET6
988 	case AF_INET6:
989 	    {
990 		struct in6pcb *in6p = (struct in6pcb *)aux;
991 
992 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
993 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
994 					       : NULL);
995 		in6p->in6p_ppcb = (caddr_t)tp;
996 
997 		tp->t_in6pcb = in6p;
998 		/* for IPv6, always try to run path MTU discovery */
999 		tp->t_mtudisc = 1;
1000 		break;
1001 	    }
1002 #endif /* INET6 */
1003 	default:
1004 		pool_put(&tcpcb_pool, tp);
1005 		return (NULL);
1006 	}
1007 
1008 	/*
1009 	 * Initialize our timebase.  When we send timestamps, we take
1010 	 * the delta from tcp_now -- this means each connection always
1011 	 * gets a timebase of 0, which makes it, among other things,
1012 	 * more difficult to determine how long a system has been up,
1013 	 * and thus how many TCP sequence increments have occurred.
1014 	 */
1015 	tp->ts_timebase = tcp_now;
1016 
1017 	return (tp);
1018 }
1019 
1020 /*
1021  * Drop a TCP connection, reporting
1022  * the specified error.  If connection is synchronized,
1023  * then send a RST to peer.
1024  */
1025 struct tcpcb *
1026 tcp_drop(struct tcpcb *tp, int errno)
1027 {
1028 	struct socket *so = NULL;
1029 
1030 #ifdef DIAGNOSTIC
1031 	if (tp->t_inpcb && tp->t_in6pcb)
1032 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1033 #endif
1034 #ifdef INET
1035 	if (tp->t_inpcb)
1036 		so = tp->t_inpcb->inp_socket;
1037 #endif
1038 #ifdef INET6
1039 	if (tp->t_in6pcb)
1040 		so = tp->t_in6pcb->in6p_socket;
1041 #endif
1042 	if (!so)
1043 		return NULL;
1044 
1045 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1046 		tp->t_state = TCPS_CLOSED;
1047 		(void) tcp_output(tp);
1048 		tcpstat.tcps_drops++;
1049 	} else
1050 		tcpstat.tcps_conndrops++;
1051 	if (errno == ETIMEDOUT && tp->t_softerror)
1052 		errno = tp->t_softerror;
1053 	so->so_error = errno;
1054 	return (tcp_close(tp));
1055 }
1056 
1057 /*
1058  * Return whether this tcpcb is marked as dead, indicating
1059  * to the calling timer function that no further action should
1060  * be taken, as we are about to release this tcpcb.  The release
1061  * of the storage will be done if this is the last timer running.
1062  *
1063  * This should be called from the callout handler function after
1064  * callout_ack() is done, so that the number of invoking timer
1065  * functions is 0.
1066  */
1067 int
1068 tcp_isdead(struct tcpcb *tp)
1069 {
1070 	int dead = (tp->t_flags & TF_DEAD);
1071 
1072 	if (__predict_false(dead)) {
1073 		if (tcp_timers_invoking(tp) > 0)
1074 				/* not quite there yet -- count separately? */
1075 			return dead;
1076 		tcpstat.tcps_delayed_free++;
1077 		pool_put(&tcpcb_pool, tp);
1078 	}
1079 	return dead;
1080 }
1081 
1082 /*
1083  * Close a TCP control block:
1084  *	discard all space held by the tcp
1085  *	discard internet protocol block
1086  *	wake up any sleepers
1087  */
1088 struct tcpcb *
1089 tcp_close(struct tcpcb *tp)
1090 {
1091 	struct inpcb *inp;
1092 #ifdef INET6
1093 	struct in6pcb *in6p;
1094 #endif
1095 	struct socket *so;
1096 #ifdef RTV_RTT
1097 	struct rtentry *rt;
1098 #endif
1099 	struct route *ro;
1100 
1101 	inp = tp->t_inpcb;
1102 #ifdef INET6
1103 	in6p = tp->t_in6pcb;
1104 #endif
1105 	so = NULL;
1106 	ro = NULL;
1107 	if (inp) {
1108 		so = inp->inp_socket;
1109 		ro = &inp->inp_route;
1110 	}
1111 #ifdef INET6
1112 	else if (in6p) {
1113 		so = in6p->in6p_socket;
1114 		ro = (struct route *)&in6p->in6p_route;
1115 	}
1116 #endif
1117 
1118 #ifdef RTV_RTT
1119 	/*
1120 	 * If we sent enough data to get some meaningful characteristics,
1121 	 * save them in the routing entry.  'Enough' is arbitrarily
1122 	 * defined as the sendpipesize (default 4K) * 16.  This would
1123 	 * give us 16 rtt samples assuming we only get one sample per
1124 	 * window (the usual case on a long haul net).  16 samples is
1125 	 * enough for the srtt filter to converge to within 5% of the correct
1126 	 * value; fewer samples and we could save a very bogus rtt.
1127 	 *
1128 	 * Don't update the default route's characteristics and don't
1129 	 * update anything that the user "locked".
1130 	 */
1131 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1132 	    ro && (rt = ro->ro_rt) &&
1133 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1134 		u_long i = 0;
1135 
1136 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1137 			i = tp->t_srtt *
1138 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1139 			if (rt->rt_rmx.rmx_rtt && i)
1140 				/*
1141 				 * filter this update to half the old & half
1142 				 * the new values, converting scale.
1143 				 * See route.h and tcp_var.h for a
1144 				 * description of the scaling constants.
1145 				 */
1146 				rt->rt_rmx.rmx_rtt =
1147 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1148 			else
1149 				rt->rt_rmx.rmx_rtt = i;
1150 		}
1151 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1152 			i = tp->t_rttvar *
1153 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1154 			if (rt->rt_rmx.rmx_rttvar && i)
1155 				rt->rt_rmx.rmx_rttvar =
1156 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1157 			else
1158 				rt->rt_rmx.rmx_rttvar = i;
1159 		}
1160 		/*
1161 		 * update the pipelimit (ssthresh) if it has been updated
1162 		 * already or if a pipesize was specified & the threshhold
1163 		 * got below half the pipesize.  I.e., wait for bad news
1164 		 * before we start updating, then update on both good
1165 		 * and bad news.
1166 		 */
1167 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1168 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1169 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1170 			/*
1171 			 * convert the limit from user data bytes to
1172 			 * packets then to packet data bytes.
1173 			 */
1174 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1175 			if (i < 2)
1176 				i = 2;
1177 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1178 			if (rt->rt_rmx.rmx_ssthresh)
1179 				rt->rt_rmx.rmx_ssthresh =
1180 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1181 			else
1182 				rt->rt_rmx.rmx_ssthresh = i;
1183 		}
1184 	}
1185 #endif /* RTV_RTT */
1186 	/* free the reassembly queue, if any */
1187 	TCP_REASS_LOCK(tp);
1188 	(void) tcp_freeq(tp);
1189 	TCP_REASS_UNLOCK(tp);
1190 
1191 	/* free the SACK holes list. */
1192 	tcp_free_sackholes(tp);
1193 
1194 	tcp_canceltimers(tp);
1195 	TCP_CLEAR_DELACK(tp);
1196 	syn_cache_cleanup(tp);
1197 
1198 	if (tp->t_template) {
1199 		m_free(tp->t_template);
1200 		tp->t_template = NULL;
1201 	}
1202 	if (tcp_timers_invoking(tp))
1203 		tp->t_flags |= TF_DEAD;
1204 	else
1205 		pool_put(&tcpcb_pool, tp);
1206 
1207 	if (inp) {
1208 		inp->inp_ppcb = 0;
1209 		soisdisconnected(so);
1210 		in_pcbdetach(inp);
1211 	}
1212 #ifdef INET6
1213 	else if (in6p) {
1214 		in6p->in6p_ppcb = 0;
1215 		soisdisconnected(so);
1216 		in6_pcbdetach(in6p);
1217 	}
1218 #endif
1219 	tcpstat.tcps_closed++;
1220 	return ((struct tcpcb *)0);
1221 }
1222 
1223 int
1224 tcp_freeq(tp)
1225 	struct tcpcb *tp;
1226 {
1227 	struct ipqent *qe;
1228 	int rv = 0;
1229 #ifdef TCPREASS_DEBUG
1230 	int i = 0;
1231 #endif
1232 
1233 	TCP_REASS_LOCK_CHECK(tp);
1234 
1235 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1236 #ifdef TCPREASS_DEBUG
1237 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1238 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1239 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1240 #endif
1241 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1242 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1243 		m_freem(qe->ipqe_m);
1244 		tcpipqent_free(qe);
1245 		rv = 1;
1246 	}
1247 	tp->t_segqlen = 0;
1248 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1249 	return (rv);
1250 }
1251 
1252 /*
1253  * Protocol drain routine.  Called when memory is in short supply.
1254  */
1255 void
1256 tcp_drain(void)
1257 {
1258 	struct inpcb_hdr *inph;
1259 	struct tcpcb *tp;
1260 
1261 	/*
1262 	 * Free the sequence queue of all TCP connections.
1263 	 */
1264 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1265 		switch (inph->inph_af) {
1266 		case AF_INET:
1267 			tp = intotcpcb((struct inpcb *)inph);
1268 			break;
1269 #ifdef INET6
1270 		case AF_INET6:
1271 			tp = in6totcpcb((struct in6pcb *)inph);
1272 			break;
1273 #endif
1274 		default:
1275 			tp = NULL;
1276 			break;
1277 		}
1278 		if (tp != NULL) {
1279 			/*
1280 			 * We may be called from a device's interrupt
1281 			 * context.  If the tcpcb is already busy,
1282 			 * just bail out now.
1283 			 */
1284 			if (tcp_reass_lock_try(tp) == 0)
1285 				continue;
1286 			if (tcp_freeq(tp))
1287 				tcpstat.tcps_connsdrained++;
1288 			TCP_REASS_UNLOCK(tp);
1289 		}
1290 	}
1291 }
1292 
1293 /*
1294  * Notify a tcp user of an asynchronous error;
1295  * store error as soft error, but wake up user
1296  * (for now, won't do anything until can select for soft error).
1297  */
1298 void
1299 tcp_notify(struct inpcb *inp, int error)
1300 {
1301 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1302 	struct socket *so = inp->inp_socket;
1303 
1304 	/*
1305 	 * Ignore some errors if we are hooked up.
1306 	 * If connection hasn't completed, has retransmitted several times,
1307 	 * and receives a second error, give up now.  This is better
1308 	 * than waiting a long time to establish a connection that
1309 	 * can never complete.
1310 	 */
1311 	if (tp->t_state == TCPS_ESTABLISHED &&
1312 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1313 	      error == EHOSTDOWN)) {
1314 		return;
1315 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1316 	    tp->t_rxtshift > 3 && tp->t_softerror)
1317 		so->so_error = error;
1318 	else
1319 		tp->t_softerror = error;
1320 	wakeup((caddr_t) &so->so_timeo);
1321 	sorwakeup(so);
1322 	sowwakeup(so);
1323 }
1324 
1325 #ifdef INET6
1326 void
1327 tcp6_notify(struct in6pcb *in6p, int error)
1328 {
1329 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1330 	struct socket *so = in6p->in6p_socket;
1331 
1332 	/*
1333 	 * Ignore some errors if we are hooked up.
1334 	 * If connection hasn't completed, has retransmitted several times,
1335 	 * and receives a second error, give up now.  This is better
1336 	 * than waiting a long time to establish a connection that
1337 	 * can never complete.
1338 	 */
1339 	if (tp->t_state == TCPS_ESTABLISHED &&
1340 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1341 	      error == EHOSTDOWN)) {
1342 		return;
1343 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1344 	    tp->t_rxtshift > 3 && tp->t_softerror)
1345 		so->so_error = error;
1346 	else
1347 		tp->t_softerror = error;
1348 	wakeup((caddr_t) &so->so_timeo);
1349 	sorwakeup(so);
1350 	sowwakeup(so);
1351 }
1352 #endif
1353 
1354 #ifdef INET6
1355 void
1356 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1357 {
1358 	struct tcphdr th;
1359 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1360 	int nmatch;
1361 	struct ip6_hdr *ip6;
1362 	const struct sockaddr_in6 *sa6_src = NULL;
1363 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1364 	struct mbuf *m;
1365 	int off;
1366 
1367 	if (sa->sa_family != AF_INET6 ||
1368 	    sa->sa_len != sizeof(struct sockaddr_in6))
1369 		return;
1370 	if ((unsigned)cmd >= PRC_NCMDS)
1371 		return;
1372 	else if (cmd == PRC_QUENCH) {
1373 		/*
1374 		 * Don't honor ICMP Source Quench messages meant for
1375 		 * TCP connections.
1376 		 */
1377 		return;
1378 	} else if (PRC_IS_REDIRECT(cmd))
1379 		notify = in6_rtchange, d = NULL;
1380 	else if (cmd == PRC_MSGSIZE)
1381 		; /* special code is present, see below */
1382 	else if (cmd == PRC_HOSTDEAD)
1383 		d = NULL;
1384 	else if (inet6ctlerrmap[cmd] == 0)
1385 		return;
1386 
1387 	/* if the parameter is from icmp6, decode it. */
1388 	if (d != NULL) {
1389 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1390 		m = ip6cp->ip6c_m;
1391 		ip6 = ip6cp->ip6c_ip6;
1392 		off = ip6cp->ip6c_off;
1393 		sa6_src = ip6cp->ip6c_src;
1394 	} else {
1395 		m = NULL;
1396 		ip6 = NULL;
1397 		sa6_src = &sa6_any;
1398 		off = 0;
1399 	}
1400 
1401 	if (ip6) {
1402 		/*
1403 		 * XXX: We assume that when ip6 is non NULL,
1404 		 * M and OFF are valid.
1405 		 */
1406 
1407 		/* check if we can safely examine src and dst ports */
1408 		if (m->m_pkthdr.len < off + sizeof(th)) {
1409 			if (cmd == PRC_MSGSIZE)
1410 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1411 			return;
1412 		}
1413 
1414 		bzero(&th, sizeof(th));
1415 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1416 
1417 		if (cmd == PRC_MSGSIZE) {
1418 			int valid = 0;
1419 
1420 			/*
1421 			 * Check to see if we have a valid TCP connection
1422 			 * corresponding to the address in the ICMPv6 message
1423 			 * payload.
1424 			 */
1425 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1426 			    th.th_dport, (const struct in6_addr *)&sa6_src->sin6_addr,
1427 			    th.th_sport, 0))
1428 				valid++;
1429 
1430 			/*
1431 			 * Depending on the value of "valid" and routing table
1432 			 * size (mtudisc_{hi,lo}wat), we will:
1433 			 * - recalcurate the new MTU and create the
1434 			 *   corresponding routing entry, or
1435 			 * - ignore the MTU change notification.
1436 			 */
1437 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1438 
1439 			/*
1440 			 * no need to call in6_pcbnotify, it should have been
1441 			 * called via callback if necessary
1442 			 */
1443 			return;
1444 		}
1445 
1446 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1447 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1448 		if (nmatch == 0 && syn_cache_count &&
1449 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1450 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1451 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1452 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1453 					  sa, &th);
1454 	} else {
1455 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1456 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1457 	}
1458 }
1459 #endif
1460 
1461 #ifdef INET
1462 /* assumes that ip header and tcp header are contiguous on mbuf */
1463 void *
1464 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v)
1465 {
1466 	struct ip *ip = v;
1467 	struct tcphdr *th;
1468 	struct icmp *icp;
1469 	extern const int inetctlerrmap[];
1470 	void (*notify)(struct inpcb *, int) = tcp_notify;
1471 	int errno;
1472 	int nmatch;
1473 	struct tcpcb *tp;
1474 	u_int mtu;
1475 	tcp_seq seq;
1476 	struct inpcb *inp;
1477 #ifdef INET6
1478 	struct in6pcb *in6p;
1479 	struct in6_addr src6, dst6;
1480 #endif
1481 
1482 	if (sa->sa_family != AF_INET ||
1483 	    sa->sa_len != sizeof(struct sockaddr_in))
1484 		return NULL;
1485 	if ((unsigned)cmd >= PRC_NCMDS)
1486 		return NULL;
1487 	errno = inetctlerrmap[cmd];
1488 	if (cmd == PRC_QUENCH)
1489 		/*
1490 		 * Don't honor ICMP Source Quench messages meant for
1491 		 * TCP connections.
1492 		 */
1493 		return NULL;
1494 	else if (PRC_IS_REDIRECT(cmd))
1495 		notify = in_rtchange, ip = 0;
1496 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1497 		/*
1498 		 * Check to see if we have a valid TCP connection
1499 		 * corresponding to the address in the ICMP message
1500 		 * payload.
1501 		 *
1502 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1503 		 */
1504 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1505 #ifdef INET6
1506 		memset(&src6, 0, sizeof(src6));
1507 		memset(&dst6, 0, sizeof(dst6));
1508 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1509 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1510 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1511 #endif
1512 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1513 		    th->th_dport, ip->ip_src, th->th_sport)) != NULL)
1514 #ifdef INET6
1515 			in6p = NULL;
1516 #else
1517 			;
1518 #endif
1519 #ifdef INET6
1520 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1521 		    th->th_dport, &src6, th->th_sport, 0)) != NULL)
1522 			;
1523 #endif
1524 		else
1525 			return NULL;
1526 
1527 		/*
1528 		 * Now that we've validated that we are actually communicating
1529 		 * with the host indicated in the ICMP message, locate the
1530 		 * ICMP header, recalculate the new MTU, and create the
1531 		 * corresponding routing entry.
1532 		 */
1533 		icp = (struct icmp *)((caddr_t)ip -
1534 		    offsetof(struct icmp, icmp_ip));
1535 		if (inp) {
1536 			if ((tp = intotcpcb(inp)) == NULL)
1537 				return NULL;
1538 		}
1539 #ifdef INET6
1540 		else if (in6p) {
1541 			if ((tp = in6totcpcb(in6p)) == NULL)
1542 				return NULL;
1543 		}
1544 #endif
1545 		else
1546 			return NULL;
1547 		seq = ntohl(th->th_seq);
1548 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1549 			return NULL;
1550 		/*
1551 		 * If the ICMP message advertises a Next-Hop MTU
1552 		 * equal or larger than the maximum packet size we have
1553 		 * ever sent, drop the message.
1554 		 */
1555 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1556 		if (mtu >= tp->t_pmtud_mtu_sent)
1557 			return NULL;
1558 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1559 			/*
1560 			 * Calculate new MTU, and create corresponding
1561 			 * route (traditional PMTUD).
1562 			 */
1563 			tp->t_flags &= ~TF_PMTUD_PEND;
1564 			icmp_mtudisc(icp, ip->ip_dst);
1565 		} else {
1566 			/*
1567 			 * Record the information got in the ICMP
1568 			 * message; act on it later.
1569 			 * If we had already recorded an ICMP message,
1570 			 * replace the old one only if the new message
1571 			 * refers to an older TCP segment
1572 			 */
1573 			if (tp->t_flags & TF_PMTUD_PEND) {
1574 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1575 					return NULL;
1576 			} else
1577 				tp->t_flags |= TF_PMTUD_PEND;
1578 			tp->t_pmtud_th_seq = seq;
1579 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1580 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1581 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1582 		}
1583 		return NULL;
1584 	} else if (cmd == PRC_HOSTDEAD)
1585 		ip = 0;
1586 	else if (errno == 0)
1587 		return NULL;
1588 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1589 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1590 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1591 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1592 		if (nmatch == 0 && syn_cache_count &&
1593 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1594 		    inetctlerrmap[cmd] == ENETUNREACH ||
1595 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1596 			struct sockaddr_in sin;
1597 			bzero(&sin, sizeof(sin));
1598 			sin.sin_len = sizeof(sin);
1599 			sin.sin_family = AF_INET;
1600 			sin.sin_port = th->th_sport;
1601 			sin.sin_addr = ip->ip_src;
1602 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1603 		}
1604 
1605 		/* XXX mapped address case */
1606 	} else
1607 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1608 		    notify);
1609 	return NULL;
1610 }
1611 
1612 /*
1613  * When a source quench is received, we are being notified of congestion.
1614  * Close the congestion window down to the Loss Window (one segment).
1615  * We will gradually open it again as we proceed.
1616  */
1617 void
1618 tcp_quench(struct inpcb *inp, int errno)
1619 {
1620 	struct tcpcb *tp = intotcpcb(inp);
1621 
1622 	if (tp)
1623 		tp->snd_cwnd = tp->t_segsz;
1624 }
1625 #endif
1626 
1627 #ifdef INET6
1628 void
1629 tcp6_quench(struct in6pcb *in6p, int errno)
1630 {
1631 	struct tcpcb *tp = in6totcpcb(in6p);
1632 
1633 	if (tp)
1634 		tp->snd_cwnd = tp->t_segsz;
1635 }
1636 #endif
1637 
1638 #ifdef INET
1639 /*
1640  * Path MTU Discovery handlers.
1641  */
1642 void
1643 tcp_mtudisc_callback(struct in_addr faddr)
1644 {
1645 #ifdef INET6
1646 	struct in6_addr in6;
1647 #endif
1648 
1649 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1650 #ifdef INET6
1651 	memset(&in6, 0, sizeof(in6));
1652 	in6.s6_addr16[5] = 0xffff;
1653 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1654 	tcp6_mtudisc_callback(&in6);
1655 #endif
1656 }
1657 
1658 /*
1659  * On receipt of path MTU corrections, flush old route and replace it
1660  * with the new one.  Retransmit all unacknowledged packets, to ensure
1661  * that all packets will be received.
1662  */
1663 void
1664 tcp_mtudisc(struct inpcb *inp, int errno)
1665 {
1666 	struct tcpcb *tp = intotcpcb(inp);
1667 	struct rtentry *rt = in_pcbrtentry(inp);
1668 
1669 	if (tp != 0) {
1670 		if (rt != 0) {
1671 			/*
1672 			 * If this was not a host route, remove and realloc.
1673 			 */
1674 			if ((rt->rt_flags & RTF_HOST) == 0) {
1675 				in_rtchange(inp, errno);
1676 				if ((rt = in_pcbrtentry(inp)) == 0)
1677 					return;
1678 			}
1679 
1680 			/*
1681 			 * Slow start out of the error condition.  We
1682 			 * use the MTU because we know it's smaller
1683 			 * than the previously transmitted segment.
1684 			 *
1685 			 * Note: This is more conservative than the
1686 			 * suggestion in draft-floyd-incr-init-win-03.
1687 			 */
1688 			if (rt->rt_rmx.rmx_mtu != 0)
1689 				tp->snd_cwnd =
1690 				    TCP_INITIAL_WINDOW(tcp_init_win,
1691 				    rt->rt_rmx.rmx_mtu);
1692 		}
1693 
1694 		/*
1695 		 * Resend unacknowledged packets.
1696 		 */
1697 		tp->snd_nxt = tp->snd_una;
1698 		tcp_output(tp);
1699 	}
1700 }
1701 #endif
1702 
1703 #ifdef INET6
1704 /*
1705  * Path MTU Discovery handlers.
1706  */
1707 void
1708 tcp6_mtudisc_callback(struct in6_addr *faddr)
1709 {
1710 	struct sockaddr_in6 sin6;
1711 
1712 	bzero(&sin6, sizeof(sin6));
1713 	sin6.sin6_family = AF_INET6;
1714 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1715 	sin6.sin6_addr = *faddr;
1716 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1717 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1718 }
1719 
1720 void
1721 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1722 {
1723 	struct tcpcb *tp = in6totcpcb(in6p);
1724 	struct rtentry *rt = in6_pcbrtentry(in6p);
1725 
1726 	if (tp != 0) {
1727 		if (rt != 0) {
1728 			/*
1729 			 * If this was not a host route, remove and realloc.
1730 			 */
1731 			if ((rt->rt_flags & RTF_HOST) == 0) {
1732 				in6_rtchange(in6p, errno);
1733 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1734 					return;
1735 			}
1736 
1737 			/*
1738 			 * Slow start out of the error condition.  We
1739 			 * use the MTU because we know it's smaller
1740 			 * than the previously transmitted segment.
1741 			 *
1742 			 * Note: This is more conservative than the
1743 			 * suggestion in draft-floyd-incr-init-win-03.
1744 			 */
1745 			if (rt->rt_rmx.rmx_mtu != 0)
1746 				tp->snd_cwnd =
1747 				    TCP_INITIAL_WINDOW(tcp_init_win,
1748 				    rt->rt_rmx.rmx_mtu);
1749 		}
1750 
1751 		/*
1752 		 * Resend unacknowledged packets.
1753 		 */
1754 		tp->snd_nxt = tp->snd_una;
1755 		tcp_output(tp);
1756 	}
1757 }
1758 #endif /* INET6 */
1759 
1760 /*
1761  * Compute the MSS to advertise to the peer.  Called only during
1762  * the 3-way handshake.  If we are the server (peer initiated
1763  * connection), we are called with a pointer to the interface
1764  * on which the SYN packet arrived.  If we are the client (we
1765  * initiated connection), we are called with a pointer to the
1766  * interface out which this connection should go.
1767  *
1768  * NOTE: Do not subtract IP option/extension header size nor IPsec
1769  * header size from MSS advertisement.  MSS option must hold the maximum
1770  * segment size we can accept, so it must always be:
1771  *	 max(if mtu) - ip header - tcp header
1772  */
1773 u_long
1774 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1775 {
1776 	extern u_long in_maxmtu;
1777 	u_long mss = 0;
1778 	u_long hdrsiz;
1779 
1780 	/*
1781 	 * In order to avoid defeating path MTU discovery on the peer,
1782 	 * we advertise the max MTU of all attached networks as our MSS,
1783 	 * per RFC 1191, section 3.1.
1784 	 *
1785 	 * We provide the option to advertise just the MTU of
1786 	 * the interface on which we hope this connection will
1787 	 * be receiving.  If we are responding to a SYN, we
1788 	 * will have a pretty good idea about this, but when
1789 	 * initiating a connection there is a bit more doubt.
1790 	 *
1791 	 * We also need to ensure that loopback has a large enough
1792 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1793 	 */
1794 
1795 	if (ifp != NULL)
1796 		switch (af) {
1797 		case AF_INET:
1798 			mss = ifp->if_mtu;
1799 			break;
1800 #ifdef INET6
1801 		case AF_INET6:
1802 			mss = IN6_LINKMTU(ifp);
1803 			break;
1804 #endif
1805 		}
1806 
1807 	if (tcp_mss_ifmtu == 0)
1808 		switch (af) {
1809 		case AF_INET:
1810 			mss = max(in_maxmtu, mss);
1811 			break;
1812 #ifdef INET6
1813 		case AF_INET6:
1814 			mss = max(in6_maxmtu, mss);
1815 			break;
1816 #endif
1817 		}
1818 
1819 	switch (af) {
1820 	case AF_INET:
1821 		hdrsiz = sizeof(struct ip);
1822 		break;
1823 #ifdef INET6
1824 	case AF_INET6:
1825 		hdrsiz = sizeof(struct ip6_hdr);
1826 		break;
1827 #endif
1828 	default:
1829 		hdrsiz = 0;
1830 		break;
1831 	}
1832 	hdrsiz += sizeof(struct tcphdr);
1833 	if (mss > hdrsiz)
1834 		mss -= hdrsiz;
1835 
1836 	mss = max(tcp_mssdflt, mss);
1837 	return (mss);
1838 }
1839 
1840 /*
1841  * Set connection variables based on the peer's advertised MSS.
1842  * We are passed the TCPCB for the actual connection.  If we
1843  * are the server, we are called by the compressed state engine
1844  * when the 3-way handshake is complete.  If we are the client,
1845  * we are called when we receive the SYN,ACK from the server.
1846  *
1847  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1848  * before this routine is called!
1849  */
1850 void
1851 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1852 {
1853 	struct socket *so;
1854 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1855 	struct rtentry *rt;
1856 #endif
1857 	u_long bufsize;
1858 	int mss;
1859 
1860 #ifdef DIAGNOSTIC
1861 	if (tp->t_inpcb && tp->t_in6pcb)
1862 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1863 #endif
1864 	so = NULL;
1865 	rt = NULL;
1866 #ifdef INET
1867 	if (tp->t_inpcb) {
1868 		so = tp->t_inpcb->inp_socket;
1869 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1870 		rt = in_pcbrtentry(tp->t_inpcb);
1871 #endif
1872 	}
1873 #endif
1874 #ifdef INET6
1875 	if (tp->t_in6pcb) {
1876 		so = tp->t_in6pcb->in6p_socket;
1877 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1878 		rt = in6_pcbrtentry(tp->t_in6pcb);
1879 #endif
1880 	}
1881 #endif
1882 
1883 	/*
1884 	 * As per RFC1122, use the default MSS value, unless they
1885 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1886 	 */
1887 	mss = tcp_mssdflt;
1888 	if (offer)
1889 		mss = offer;
1890 	mss = max(mss, 256);		/* sanity */
1891 	tp->t_peermss = mss;
1892 	mss -= tcp_optlen(tp);
1893 #ifdef INET
1894 	if (tp->t_inpcb)
1895 		mss -= ip_optlen(tp->t_inpcb);
1896 #endif
1897 #ifdef INET6
1898 	if (tp->t_in6pcb)
1899 		mss -= ip6_optlen(tp->t_in6pcb);
1900 #endif
1901 
1902 	/*
1903 	 * If there's a pipesize, change the socket buffer to that size.
1904 	 * Make the socket buffer an integral number of MSS units.  If
1905 	 * the MSS is larger than the socket buffer, artificially decrease
1906 	 * the MSS.
1907 	 */
1908 #ifdef RTV_SPIPE
1909 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1910 		bufsize = rt->rt_rmx.rmx_sendpipe;
1911 	else
1912 #endif
1913 		bufsize = so->so_snd.sb_hiwat;
1914 	if (bufsize < mss)
1915 		mss = bufsize;
1916 	else {
1917 		bufsize = roundup(bufsize, mss);
1918 		if (bufsize > sb_max)
1919 			bufsize = sb_max;
1920 		(void) sbreserve(&so->so_snd, bufsize, so);
1921 	}
1922 	tp->t_segsz = mss;
1923 
1924 #ifdef RTV_SSTHRESH
1925 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1926 		/*
1927 		 * There's some sort of gateway or interface buffer
1928 		 * limit on the path.  Use this to set the slow
1929 		 * start threshold, but set the threshold to no less
1930 		 * than 2 * MSS.
1931 		 */
1932 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1933 	}
1934 #endif
1935 }
1936 
1937 /*
1938  * Processing necessary when a TCP connection is established.
1939  */
1940 void
1941 tcp_established(struct tcpcb *tp)
1942 {
1943 	struct socket *so;
1944 #ifdef RTV_RPIPE
1945 	struct rtentry *rt;
1946 #endif
1947 	u_long bufsize;
1948 
1949 #ifdef DIAGNOSTIC
1950 	if (tp->t_inpcb && tp->t_in6pcb)
1951 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1952 #endif
1953 	so = NULL;
1954 	rt = NULL;
1955 #ifdef INET
1956 	if (tp->t_inpcb) {
1957 		so = tp->t_inpcb->inp_socket;
1958 #if defined(RTV_RPIPE)
1959 		rt = in_pcbrtentry(tp->t_inpcb);
1960 #endif
1961 	}
1962 #endif
1963 #ifdef INET6
1964 	if (tp->t_in6pcb) {
1965 		so = tp->t_in6pcb->in6p_socket;
1966 #if defined(RTV_RPIPE)
1967 		rt = in6_pcbrtentry(tp->t_in6pcb);
1968 #endif
1969 	}
1970 #endif
1971 
1972 	tp->t_state = TCPS_ESTABLISHED;
1973 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1974 
1975 #ifdef RTV_RPIPE
1976 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1977 		bufsize = rt->rt_rmx.rmx_recvpipe;
1978 	else
1979 #endif
1980 		bufsize = so->so_rcv.sb_hiwat;
1981 	if (bufsize > tp->t_ourmss) {
1982 		bufsize = roundup(bufsize, tp->t_ourmss);
1983 		if (bufsize > sb_max)
1984 			bufsize = sb_max;
1985 		(void) sbreserve(&so->so_rcv, bufsize, so);
1986 	}
1987 }
1988 
1989 /*
1990  * Check if there's an initial rtt or rttvar.  Convert from the
1991  * route-table units to scaled multiples of the slow timeout timer.
1992  * Called only during the 3-way handshake.
1993  */
1994 void
1995 tcp_rmx_rtt(struct tcpcb *tp)
1996 {
1997 #ifdef RTV_RTT
1998 	struct rtentry *rt = NULL;
1999 	int rtt;
2000 
2001 #ifdef DIAGNOSTIC
2002 	if (tp->t_inpcb && tp->t_in6pcb)
2003 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2004 #endif
2005 #ifdef INET
2006 	if (tp->t_inpcb)
2007 		rt = in_pcbrtentry(tp->t_inpcb);
2008 #endif
2009 #ifdef INET6
2010 	if (tp->t_in6pcb)
2011 		rt = in6_pcbrtentry(tp->t_in6pcb);
2012 #endif
2013 	if (rt == NULL)
2014 		return;
2015 
2016 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2017 		/*
2018 		 * XXX The lock bit for MTU indicates that the value
2019 		 * is also a minimum value; this is subject to time.
2020 		 */
2021 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2022 			TCPT_RANGESET(tp->t_rttmin,
2023 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2024 			    TCPTV_MIN, TCPTV_REXMTMAX);
2025 		tp->t_srtt = rtt /
2026 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2027 		if (rt->rt_rmx.rmx_rttvar) {
2028 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2029 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2030 				(TCP_RTTVAR_SHIFT + 2));
2031 		} else {
2032 			/* Default variation is +- 1 rtt */
2033 			tp->t_rttvar =
2034 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2035 		}
2036 		TCPT_RANGESET(tp->t_rxtcur,
2037 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2038 		    tp->t_rttmin, TCPTV_REXMTMAX);
2039 	}
2040 #endif
2041 }
2042 
2043 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2044 #if NRND > 0
2045 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2046 #endif
2047 
2048 /*
2049  * Get a new sequence value given a tcp control block
2050  */
2051 tcp_seq
2052 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2053 {
2054 
2055 #ifdef INET
2056 	if (tp->t_inpcb != NULL) {
2057 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2058 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2059 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2060 		    addin));
2061 	}
2062 #endif
2063 #ifdef INET6
2064 	if (tp->t_in6pcb != NULL) {
2065 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2066 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2067 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2068 		    addin));
2069 	}
2070 #endif
2071 	/* Not possible. */
2072 	panic("tcp_new_iss");
2073 }
2074 
2075 /*
2076  * This routine actually generates a new TCP initial sequence number.
2077  */
2078 tcp_seq
2079 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2080     size_t addrsz, tcp_seq addin)
2081 {
2082 	tcp_seq tcp_iss;
2083 
2084 #if NRND > 0
2085 	static int beenhere;
2086 
2087 	/*
2088 	 * If we haven't been here before, initialize our cryptographic
2089 	 * hash secret.
2090 	 */
2091 	if (beenhere == 0) {
2092 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2093 		    RND_EXTRACT_ANY);
2094 		beenhere = 1;
2095 	}
2096 
2097 	if (tcp_do_rfc1948) {
2098 		MD5_CTX ctx;
2099 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2100 
2101 		/*
2102 		 * Compute the base value of the ISS.  It is a hash
2103 		 * of (saddr, sport, daddr, dport, secret).
2104 		 */
2105 		MD5Init(&ctx);
2106 
2107 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2108 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2109 
2110 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2111 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2112 
2113 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2114 
2115 		MD5Final(hash, &ctx);
2116 
2117 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2118 
2119 		/*
2120 		 * Now increment our "timer", and add it in to
2121 		 * the computed value.
2122 		 *
2123 		 * XXX Use `addin'?
2124 		 * XXX TCP_ISSINCR too large to use?
2125 		 */
2126 		tcp_iss_seq += TCP_ISSINCR;
2127 #ifdef TCPISS_DEBUG
2128 		printf("ISS hash 0x%08x, ", tcp_iss);
2129 #endif
2130 		tcp_iss += tcp_iss_seq + addin;
2131 #ifdef TCPISS_DEBUG
2132 		printf("new ISS 0x%08x\n", tcp_iss);
2133 #endif
2134 	} else
2135 #endif /* NRND > 0 */
2136 	{
2137 		/*
2138 		 * Randomize.
2139 		 */
2140 #if NRND > 0
2141 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2142 #else
2143 		tcp_iss = arc4random();
2144 #endif
2145 
2146 		/*
2147 		 * If we were asked to add some amount to a known value,
2148 		 * we will take a random value obtained above, mask off
2149 		 * the upper bits, and add in the known value.  We also
2150 		 * add in a constant to ensure that we are at least a
2151 		 * certain distance from the original value.
2152 		 *
2153 		 * This is used when an old connection is in timed wait
2154 		 * and we have a new one coming in, for instance.
2155 		 */
2156 		if (addin != 0) {
2157 #ifdef TCPISS_DEBUG
2158 			printf("Random %08x, ", tcp_iss);
2159 #endif
2160 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2161 			tcp_iss += addin + TCP_ISSINCR;
2162 #ifdef TCPISS_DEBUG
2163 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2164 #endif
2165 		} else {
2166 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2167 			tcp_iss += tcp_iss_seq;
2168 			tcp_iss_seq += TCP_ISSINCR;
2169 #ifdef TCPISS_DEBUG
2170 			printf("ISS %08x\n", tcp_iss);
2171 #endif
2172 		}
2173 	}
2174 
2175 	if (tcp_compat_42) {
2176 		/*
2177 		 * Limit it to the positive range for really old TCP
2178 		 * implementations.
2179 		 * Just AND off the top bit instead of checking if
2180 		 * is set first - saves a branch 50% of the time.
2181 		 */
2182 		tcp_iss &= 0x7fffffff;		/* XXX */
2183 	}
2184 
2185 	return (tcp_iss);
2186 }
2187 
2188 #if defined(IPSEC) || defined(FAST_IPSEC)
2189 /* compute ESP/AH header size for TCP, including outer IP header. */
2190 size_t
2191 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2192 {
2193 	struct inpcb *inp;
2194 	size_t hdrsiz;
2195 
2196 	/* XXX mapped addr case (tp->t_in6pcb) */
2197 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2198 		return 0;
2199 	switch (tp->t_family) {
2200 	case AF_INET:
2201 		/* XXX: should use currect direction. */
2202 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2203 		break;
2204 	default:
2205 		hdrsiz = 0;
2206 		break;
2207 	}
2208 
2209 	return hdrsiz;
2210 }
2211 
2212 #ifdef INET6
2213 size_t
2214 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2215 {
2216 	struct in6pcb *in6p;
2217 	size_t hdrsiz;
2218 
2219 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2220 		return 0;
2221 	switch (tp->t_family) {
2222 	case AF_INET6:
2223 		/* XXX: should use currect direction. */
2224 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2225 		break;
2226 	case AF_INET:
2227 		/* mapped address case - tricky */
2228 	default:
2229 		hdrsiz = 0;
2230 		break;
2231 	}
2232 
2233 	return hdrsiz;
2234 }
2235 #endif
2236 #endif /*IPSEC*/
2237 
2238 /*
2239  * Determine the length of the TCP options for this connection.
2240  *
2241  * XXX:  What do we do for SACK, when we add that?  Just reserve
2242  *       all of the space?  Otherwise we can't exactly be incrementing
2243  *       cwnd by an amount that varies depending on the amount we last
2244  *       had to SACK!
2245  */
2246 
2247 u_int
2248 tcp_optlen(struct tcpcb *tp)
2249 {
2250 	u_int optlen;
2251 
2252 	optlen = 0;
2253 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2254 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2255 		optlen += TCPOLEN_TSTAMP_APPA;
2256 
2257 #ifdef TCP_SIGNATURE
2258 #if defined(INET6) && defined(FAST_IPSEC)
2259 	if (tp->t_family == AF_INET)
2260 #endif
2261 	if (tp->t_flags & TF_SIGNATURE)
2262 		optlen += TCPOLEN_SIGNATURE + 2;
2263 #endif /* TCP_SIGNATURE */
2264 
2265 	return optlen;
2266 }
2267 
2268 u_int
2269 tcp_hdrsz(struct tcpcb *tp)
2270 {
2271 	u_int hlen;
2272 
2273 	switch (tp->t_family) {
2274 #ifdef INET6
2275 	case AF_INET6:
2276 		hlen = sizeof(struct ip6_hdr);
2277 		break;
2278 #endif
2279 	case AF_INET:
2280 		hlen = sizeof(struct ip);
2281 		break;
2282 	default:
2283 		hlen = 0;
2284 		break;
2285 	}
2286 	hlen += sizeof(struct tcphdr);
2287 
2288 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2289 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2290 		hlen += TCPOLEN_TSTAMP_APPA;
2291 #ifdef TCP_SIGNATURE
2292 	if (tp->t_flags & TF_SIGNATURE)
2293 		hlen += TCPOLEN_SIGLEN;
2294 #endif
2295 	return hlen;
2296 }
2297