xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 990562bfef3015a6e18a209dde598d5588ea144c)
1 /*	$NetBSD: tcp_subr.c,v 1.137 2002/11/24 10:52:47 scw Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998, 2000, 2001 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38  * Facility, NASA Ames Research Center.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. All advertising materials mentioning features or use of this software
82  *    must display the following acknowledgement:
83  *	This product includes software developed by the University of
84  *	California, Berkeley and its contributors.
85  * 4. Neither the name of the University nor the names of its contributors
86  *    may be used to endorse or promote products derived from this software
87  *    without specific prior written permission.
88  *
89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
99  * SUCH DAMAGE.
100  *
101  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
102  */
103 
104 #include <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.137 2002/11/24 10:52:47 scw Exp $");
106 
107 #include "opt_inet.h"
108 #include "opt_ipsec.h"
109 #include "opt_tcp_compat_42.h"
110 #include "opt_inet_csum.h"
111 #include "rnd.h"
112 
113 #include <sys/param.h>
114 #include <sys/proc.h>
115 #include <sys/systm.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/socket.h>
119 #include <sys/socketvar.h>
120 #include <sys/protosw.h>
121 #include <sys/errno.h>
122 #include <sys/kernel.h>
123 #include <sys/pool.h>
124 #if NRND > 0
125 #include <sys/md5.h>
126 #include <sys/rnd.h>
127 #endif
128 
129 #include <net/route.h>
130 #include <net/if.h>
131 
132 #include <netinet/in.h>
133 #include <netinet/in_systm.h>
134 #include <netinet/ip.h>
135 #include <netinet/in_pcb.h>
136 #include <netinet/ip_var.h>
137 #include <netinet/ip_icmp.h>
138 
139 #ifdef INET6
140 #ifndef INET
141 #include <netinet/in.h>
142 #endif
143 #include <netinet/ip6.h>
144 #include <netinet6/in6_pcb.h>
145 #include <netinet6/ip6_var.h>
146 #include <netinet6/in6_var.h>
147 #include <netinet6/ip6protosw.h>
148 #include <netinet/icmp6.h>
149 #include <netinet6/nd6.h>
150 #endif
151 
152 #include <netinet/tcp.h>
153 #include <netinet/tcp_fsm.h>
154 #include <netinet/tcp_seq.h>
155 #include <netinet/tcp_timer.h>
156 #include <netinet/tcp_var.h>
157 #include <netinet/tcpip.h>
158 
159 #ifdef IPSEC
160 #include <netinet6/ipsec.h>
161 #endif /*IPSEC*/
162 
163 #ifdef INET6
164 struct in6pcb tcb6;
165 #endif
166 
167 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
168 struct	tcpstat tcpstat;	/* tcp statistics */
169 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
170 
171 /* patchable/settable parameters for tcp */
172 int 	tcp_mssdflt = TCP_MSS;
173 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
174 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
175 #if NRND > 0
176 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
177 #endif
178 int	tcp_do_sack = 1;	/* selective acknowledgement */
179 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
180 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
181 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
182 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
183 int	tcp_init_win = 1;
184 int	tcp_mss_ifmtu = 0;
185 #ifdef TCP_COMPAT_42
186 int	tcp_compat_42 = 1;
187 #else
188 int	tcp_compat_42 = 0;
189 #endif
190 int	tcp_rst_ppslim = 100;	/* 100pps */
191 
192 /* tcb hash */
193 #ifndef TCBHASHSIZE
194 #define	TCBHASHSIZE	128
195 #endif
196 int	tcbhashsize = TCBHASHSIZE;
197 
198 /* syn hash parameters */
199 #define	TCP_SYN_HASH_SIZE	293
200 #define	TCP_SYN_BUCKET_SIZE	35
201 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
202 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
203 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
204 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
205 
206 int	tcp_freeq __P((struct tcpcb *));
207 
208 #ifdef INET
209 void	tcp_mtudisc_callback __P((struct in_addr));
210 #endif
211 #ifdef INET6
212 void	tcp6_mtudisc_callback __P((struct in6_addr *));
213 #endif
214 
215 void	tcp_mtudisc __P((struct inpcb *, int));
216 #ifdef INET6
217 void	tcp6_mtudisc __P((struct in6pcb *, int));
218 #endif
219 
220 struct pool tcpcb_pool;
221 
222 #ifdef TCP_CSUM_COUNTERS
223 #include <sys/device.h>
224 
225 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
226     NULL, "tcp", "hwcsum bad");
227 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
228     NULL, "tcp", "hwcsum ok");
229 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
230     NULL, "tcp", "hwcsum data");
231 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
232     NULL, "tcp", "swcsum");
233 #endif /* TCP_CSUM_COUNTERS */
234 
235 #ifdef TCP_OUTPUT_COUNTERS
236 #include <sys/device.h>
237 
238 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
239     NULL, "tcp", "output big header");
240 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
241     NULL, "tcp", "output copy small");
242 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243     NULL, "tcp", "output copy big");
244 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245     NULL, "tcp", "output reference big");
246 #endif /* TCP_OUTPUT_COUNTERS */
247 
248 #ifdef TCP_REASS_COUNTERS
249 #include <sys/device.h>
250 
251 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
252     NULL, "tcp_reass", "calls");
253 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
254     &tcp_reass_, "tcp_reass", "insert into empty queue");
255 struct evcnt tcp_reass_iteration[8] = {
256     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
257     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
258     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
259     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
260     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
261     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
262     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
263     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
264 };
265 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     &tcp_reass_, "tcp_reass", "prepend to first");
267 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     &tcp_reass_, "tcp_reass", "prepend");
269 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
270     &tcp_reass_, "tcp_reass", "insert");
271 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
272     &tcp_reass_, "tcp_reass", "insert at tail");
273 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
274     &tcp_reass_, "tcp_reass", "append");
275 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
276     &tcp_reass_, "tcp_reass", "append to tail fragment");
277 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     &tcp_reass_, "tcp_reass", "overlap at end");
279 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     &tcp_reass_, "tcp_reass", "overlap at start");
281 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     &tcp_reass_, "tcp_reass", "duplicate segment");
283 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "duplicate fragment");
285 
286 #endif /* TCP_REASS_COUNTERS */
287 
288 /*
289  * Tcp initialization
290  */
291 void
292 tcp_init()
293 {
294 	int hlen;
295 
296 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
297 	    NULL);
298 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
299 #ifdef INET6
300 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
301 #endif
302 
303 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
304 #ifdef INET6
305 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
306 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
307 #endif
308 	if (max_protohdr < hlen)
309 		max_protohdr = hlen;
310 	if (max_linkhdr + hlen > MHLEN)
311 		panic("tcp_init");
312 
313 #ifdef INET
314 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
315 #endif
316 #ifdef INET6
317 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
318 #endif
319 
320 	/* Initialize timer state. */
321 	tcp_timer_init();
322 
323 	/* Initialize the compressed state engine. */
324 	syn_cache_init();
325 
326 #ifdef TCP_CSUM_COUNTERS
327 	evcnt_attach_static(&tcp_hwcsum_bad);
328 	evcnt_attach_static(&tcp_hwcsum_ok);
329 	evcnt_attach_static(&tcp_hwcsum_data);
330 	evcnt_attach_static(&tcp_swcsum);
331 #endif /* TCP_CSUM_COUNTERS */
332 
333 #ifdef TCP_OUTPUT_COUNTERS
334 	evcnt_attach_static(&tcp_output_bigheader);
335 	evcnt_attach_static(&tcp_output_copysmall);
336 	evcnt_attach_static(&tcp_output_copybig);
337 	evcnt_attach_static(&tcp_output_refbig);
338 #endif /* TCP_OUTPUT_COUNTERS */
339 
340 #ifdef TCP_REASS_COUNTERS
341 	evcnt_attach_static(&tcp_reass_);
342 	evcnt_attach_static(&tcp_reass_empty);
343 	evcnt_attach_static(&tcp_reass_iteration[0]);
344 	evcnt_attach_static(&tcp_reass_iteration[1]);
345 	evcnt_attach_static(&tcp_reass_iteration[2]);
346 	evcnt_attach_static(&tcp_reass_iteration[3]);
347 	evcnt_attach_static(&tcp_reass_iteration[4]);
348 	evcnt_attach_static(&tcp_reass_iteration[5]);
349 	evcnt_attach_static(&tcp_reass_iteration[6]);
350 	evcnt_attach_static(&tcp_reass_iteration[7]);
351 	evcnt_attach_static(&tcp_reass_prependfirst);
352 	evcnt_attach_static(&tcp_reass_prepend);
353 	evcnt_attach_static(&tcp_reass_insert);
354 	evcnt_attach_static(&tcp_reass_inserttail);
355 	evcnt_attach_static(&tcp_reass_append);
356 	evcnt_attach_static(&tcp_reass_appendtail);
357 	evcnt_attach_static(&tcp_reass_overlaptail);
358 	evcnt_attach_static(&tcp_reass_overlapfront);
359 	evcnt_attach_static(&tcp_reass_segdup);
360 	evcnt_attach_static(&tcp_reass_fragdup);
361 #endif /* TCP_REASS_COUNTERS */
362 }
363 
364 /*
365  * Create template to be used to send tcp packets on a connection.
366  * Call after host entry created, allocates an mbuf and fills
367  * in a skeletal tcp/ip header, minimizing the amount of work
368  * necessary when the connection is used.
369  */
370 struct mbuf *
371 tcp_template(tp)
372 	struct tcpcb *tp;
373 {
374 	struct inpcb *inp = tp->t_inpcb;
375 #ifdef INET6
376 	struct in6pcb *in6p = tp->t_in6pcb;
377 #endif
378 	struct tcphdr *n;
379 	struct mbuf *m;
380 	int hlen;
381 
382 	switch (tp->t_family) {
383 	case AF_INET:
384 		hlen = sizeof(struct ip);
385 		if (inp)
386 			break;
387 #ifdef INET6
388 		if (in6p) {
389 			/* mapped addr case */
390 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
391 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
392 				break;
393 		}
394 #endif
395 		return NULL;	/*EINVAL*/
396 #ifdef INET6
397 	case AF_INET6:
398 		hlen = sizeof(struct ip6_hdr);
399 		if (in6p) {
400 			/* more sainty check? */
401 			break;
402 		}
403 		return NULL;	/*EINVAL*/
404 #endif
405 	default:
406 		hlen = 0;	/*pacify gcc*/
407 		return NULL;	/*EAFNOSUPPORT*/
408 	}
409 #ifdef DIAGNOSTIC
410 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
411 		panic("mclbytes too small for t_template");
412 #endif
413 	m = tp->t_template;
414 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
415 		;
416 	else {
417 		if (m)
418 			m_freem(m);
419 		m = tp->t_template = NULL;
420 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
421 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
422 			MCLGET(m, M_DONTWAIT);
423 			if ((m->m_flags & M_EXT) == 0) {
424 				m_free(m);
425 				m = NULL;
426 			}
427 		}
428 		if (m == NULL)
429 			return NULL;
430 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
431 	}
432 
433 	bzero(mtod(m, caddr_t), m->m_len);
434 
435 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
436 
437 	switch (tp->t_family) {
438 	case AF_INET:
439 	    {
440 		struct ipovly *ipov;
441 		mtod(m, struct ip *)->ip_v = 4;
442 		ipov = mtod(m, struct ipovly *);
443 		ipov->ih_pr = IPPROTO_TCP;
444 		ipov->ih_len = htons(sizeof(struct tcphdr));
445 		if (inp) {
446 			ipov->ih_src = inp->inp_laddr;
447 			ipov->ih_dst = inp->inp_faddr;
448 		}
449 #ifdef INET6
450 		else if (in6p) {
451 			/* mapped addr case */
452 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
453 				sizeof(ipov->ih_src));
454 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
455 				sizeof(ipov->ih_dst));
456 		}
457 #endif
458 		/*
459 		 * Compute the pseudo-header portion of the checksum
460 		 * now.  We incrementally add in the TCP option and
461 		 * payload lengths later, and then compute the TCP
462 		 * checksum right before the packet is sent off onto
463 		 * the wire.
464 		 */
465 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
466 		    ipov->ih_dst.s_addr,
467 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
468 		break;
469 	    }
470 #ifdef INET6
471 	case AF_INET6:
472 	    {
473 		struct ip6_hdr *ip6;
474 		mtod(m, struct ip *)->ip_v = 6;
475 		ip6 = mtod(m, struct ip6_hdr *);
476 		ip6->ip6_nxt = IPPROTO_TCP;
477 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
478 		ip6->ip6_src = in6p->in6p_laddr;
479 		ip6->ip6_dst = in6p->in6p_faddr;
480 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
481 		if (ip6_auto_flowlabel) {
482 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
483 			ip6->ip6_flow |=
484 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
485 		}
486 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
487 		ip6->ip6_vfc |= IPV6_VERSION;
488 
489 		/*
490 		 * Compute the pseudo-header portion of the checksum
491 		 * now.  We incrementally add in the TCP option and
492 		 * payload lengths later, and then compute the TCP
493 		 * checksum right before the packet is sent off onto
494 		 * the wire.
495 		 */
496 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
497 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
498 		    htonl(IPPROTO_TCP));
499 		break;
500 	    }
501 #endif
502 	}
503 	if (inp) {
504 		n->th_sport = inp->inp_lport;
505 		n->th_dport = inp->inp_fport;
506 	}
507 #ifdef INET6
508 	else if (in6p) {
509 		n->th_sport = in6p->in6p_lport;
510 		n->th_dport = in6p->in6p_fport;
511 	}
512 #endif
513 	n->th_seq = 0;
514 	n->th_ack = 0;
515 	n->th_x2 = 0;
516 	n->th_off = 5;
517 	n->th_flags = 0;
518 	n->th_win = 0;
519 	n->th_urp = 0;
520 	return (m);
521 }
522 
523 /*
524  * Send a single message to the TCP at address specified by
525  * the given TCP/IP header.  If m == 0, then we make a copy
526  * of the tcpiphdr at ti and send directly to the addressed host.
527  * This is used to force keep alive messages out using the TCP
528  * template for a connection tp->t_template.  If flags are given
529  * then we send a message back to the TCP which originated the
530  * segment ti, and discard the mbuf containing it and any other
531  * attached mbufs.
532  *
533  * In any case the ack and sequence number of the transmitted
534  * segment are as specified by the parameters.
535  */
536 int
537 tcp_respond(tp, template, m, th0, ack, seq, flags)
538 	struct tcpcb *tp;
539 	struct mbuf *template;
540 	struct mbuf *m;
541 	struct tcphdr *th0;
542 	tcp_seq ack, seq;
543 	int flags;
544 {
545 	struct route *ro;
546 	int error, tlen, win = 0;
547 	int hlen;
548 	struct ip *ip;
549 #ifdef INET6
550 	struct ip6_hdr *ip6;
551 #endif
552 	int family;	/* family on packet, not inpcb/in6pcb! */
553 	struct tcphdr *th;
554 
555 	if (tp != NULL && (flags & TH_RST) == 0) {
556 #ifdef DIAGNOSTIC
557 		if (tp->t_inpcb && tp->t_in6pcb)
558 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
559 #endif
560 #ifdef INET
561 		if (tp->t_inpcb)
562 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
563 #endif
564 #ifdef INET6
565 		if (tp->t_in6pcb)
566 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
567 #endif
568 	}
569 
570 	th = NULL;	/* Quell uninitialized warning */
571 	ip = NULL;
572 #ifdef INET6
573 	ip6 = NULL;
574 #endif
575 	if (m == 0) {
576 		if (!template)
577 			return EINVAL;
578 
579 		/* get family information from template */
580 		switch (mtod(template, struct ip *)->ip_v) {
581 		case 4:
582 			family = AF_INET;
583 			hlen = sizeof(struct ip);
584 			break;
585 #ifdef INET6
586 		case 6:
587 			family = AF_INET6;
588 			hlen = sizeof(struct ip6_hdr);
589 			break;
590 #endif
591 		default:
592 			return EAFNOSUPPORT;
593 		}
594 
595 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
596 		if (m) {
597 			MCLGET(m, M_DONTWAIT);
598 			if ((m->m_flags & M_EXT) == 0) {
599 				m_free(m);
600 				m = NULL;
601 			}
602 		}
603 		if (m == NULL)
604 			return (ENOBUFS);
605 
606 		if (tcp_compat_42)
607 			tlen = 1;
608 		else
609 			tlen = 0;
610 
611 		m->m_data += max_linkhdr;
612 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
613 			template->m_len);
614 		switch (family) {
615 		case AF_INET:
616 			ip = mtod(m, struct ip *);
617 			th = (struct tcphdr *)(ip + 1);
618 			break;
619 #ifdef INET6
620 		case AF_INET6:
621 			ip6 = mtod(m, struct ip6_hdr *);
622 			th = (struct tcphdr *)(ip6 + 1);
623 			break;
624 #endif
625 #if 0
626 		default:
627 			/* noone will visit here */
628 			m_freem(m);
629 			return EAFNOSUPPORT;
630 #endif
631 		}
632 		flags = TH_ACK;
633 	} else {
634 
635 		if ((m->m_flags & M_PKTHDR) == 0) {
636 #if 0
637 			printf("non PKTHDR to tcp_respond\n");
638 #endif
639 			m_freem(m);
640 			return EINVAL;
641 		}
642 #ifdef DIAGNOSTIC
643 		if (!th0)
644 			panic("th0 == NULL in tcp_respond");
645 #endif
646 
647 		/* get family information from m */
648 		switch (mtod(m, struct ip *)->ip_v) {
649 		case 4:
650 			family = AF_INET;
651 			hlen = sizeof(struct ip);
652 			ip = mtod(m, struct ip *);
653 			break;
654 #ifdef INET6
655 		case 6:
656 			family = AF_INET6;
657 			hlen = sizeof(struct ip6_hdr);
658 			ip6 = mtod(m, struct ip6_hdr *);
659 			break;
660 #endif
661 		default:
662 			m_freem(m);
663 			return EAFNOSUPPORT;
664 		}
665 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
666 			tlen = sizeof(*th0);
667 		else
668 			tlen = th0->th_off << 2;
669 
670 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
671 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
672 			m->m_len = hlen + tlen;
673 			m_freem(m->m_next);
674 			m->m_next = NULL;
675 		} else {
676 			struct mbuf *n;
677 
678 #ifdef DIAGNOSTIC
679 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
680 				m_freem(m);
681 				return EMSGSIZE;
682 			}
683 #endif
684 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
685 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
686 				MCLGET(n, M_DONTWAIT);
687 				if ((n->m_flags & M_EXT) == 0) {
688 					m_freem(n);
689 					n = NULL;
690 				}
691 			}
692 			if (!n) {
693 				m_freem(m);
694 				return ENOBUFS;
695 			}
696 
697 			n->m_data += max_linkhdr;
698 			n->m_len = hlen + tlen;
699 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
700 			m_copyback(n, hlen, tlen, (caddr_t)th0);
701 
702 			m_freem(m);
703 			m = n;
704 			n = NULL;
705 		}
706 
707 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
708 		switch (family) {
709 		case AF_INET:
710 			ip = mtod(m, struct ip *);
711 			th = (struct tcphdr *)(ip + 1);
712 			ip->ip_p = IPPROTO_TCP;
713 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
714 			ip->ip_p = IPPROTO_TCP;
715 			break;
716 #ifdef INET6
717 		case AF_INET6:
718 			ip6 = mtod(m, struct ip6_hdr *);
719 			th = (struct tcphdr *)(ip6 + 1);
720 			ip6->ip6_nxt = IPPROTO_TCP;
721 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
722 			ip6->ip6_nxt = IPPROTO_TCP;
723 			break;
724 #endif
725 #if 0
726 		default:
727 			/* noone will visit here */
728 			m_freem(m);
729 			return EAFNOSUPPORT;
730 #endif
731 		}
732 		xchg(th->th_dport, th->th_sport, u_int16_t);
733 #undef xchg
734 		tlen = 0;	/*be friendly with the following code*/
735 	}
736 	th->th_seq = htonl(seq);
737 	th->th_ack = htonl(ack);
738 	th->th_x2 = 0;
739 	if ((flags & TH_SYN) == 0) {
740 		if (tp)
741 			win >>= tp->rcv_scale;
742 		if (win > TCP_MAXWIN)
743 			win = TCP_MAXWIN;
744 		th->th_win = htons((u_int16_t)win);
745 		th->th_off = sizeof (struct tcphdr) >> 2;
746 		tlen += sizeof(*th);
747 	} else
748 		tlen += th->th_off << 2;
749 	m->m_len = hlen + tlen;
750 	m->m_pkthdr.len = hlen + tlen;
751 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
752 	th->th_flags = flags;
753 	th->th_urp = 0;
754 
755 	switch (family) {
756 #ifdef INET
757 	case AF_INET:
758 	    {
759 		struct ipovly *ipov = (struct ipovly *)ip;
760 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
761 		ipov->ih_len = htons((u_int16_t)tlen);
762 
763 		th->th_sum = 0;
764 		th->th_sum = in_cksum(m, hlen + tlen);
765 		ip->ip_len = htons(hlen + tlen);
766 		ip->ip_ttl = ip_defttl;
767 		break;
768 	    }
769 #endif
770 #ifdef INET6
771 	case AF_INET6:
772 	    {
773 		th->th_sum = 0;
774 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
775 				tlen);
776 		ip6->ip6_plen = ntohs(tlen);
777 		if (tp && tp->t_in6pcb) {
778 			struct ifnet *oifp;
779 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
780 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
781 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
782 		} else
783 			ip6->ip6_hlim = ip6_defhlim;
784 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
785 		if (ip6_auto_flowlabel) {
786 			ip6->ip6_flow |=
787 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
788 		}
789 		break;
790 	    }
791 #endif
792 	}
793 
794 #ifdef IPSEC
795 	(void)ipsec_setsocket(m, NULL);
796 #endif /*IPSEC*/
797 
798 	if (tp != NULL && tp->t_inpcb != NULL) {
799 		ro = &tp->t_inpcb->inp_route;
800 #ifdef IPSEC
801 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
802 			m_freem(m);
803 			return ENOBUFS;
804 		}
805 #endif
806 #ifdef DIAGNOSTIC
807 		if (family != AF_INET)
808 			panic("tcp_respond: address family mismatch");
809 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
810 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
811 			    ntohl(ip->ip_dst.s_addr),
812 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
813 		}
814 #endif
815 	}
816 #ifdef INET6
817 	else if (tp != NULL && tp->t_in6pcb != NULL) {
818 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
819 #ifdef IPSEC
820 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
821 			m_freem(m);
822 			return ENOBUFS;
823 		}
824 #endif
825 #ifdef DIAGNOSTIC
826 		if (family == AF_INET) {
827 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
828 				panic("tcp_respond: not mapped addr");
829 			if (bcmp(&ip->ip_dst,
830 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
831 			    sizeof(ip->ip_dst)) != 0) {
832 				panic("tcp_respond: ip_dst != in6p_faddr");
833 			}
834 		} else if (family == AF_INET6) {
835 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
836 			    &tp->t_in6pcb->in6p_faddr))
837 				panic("tcp_respond: ip6_dst != in6p_faddr");
838 		} else
839 			panic("tcp_respond: address family mismatch");
840 #endif
841 	}
842 #endif
843 	else
844 		ro = NULL;
845 
846 	switch (family) {
847 #ifdef INET
848 	case AF_INET:
849 		error = ip_output(m, NULL, ro,
850 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
851 		    NULL);
852 		break;
853 #endif
854 #ifdef INET6
855 	case AF_INET6:
856 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
857 		    NULL);
858 		break;
859 #endif
860 	default:
861 		error = EAFNOSUPPORT;
862 		break;
863 	}
864 
865 	return (error);
866 }
867 
868 /*
869  * Create a new TCP control block, making an
870  * empty reassembly queue and hooking it to the argument
871  * protocol control block.
872  */
873 struct tcpcb *
874 tcp_newtcpcb(family, aux)
875 	int family;	/* selects inpcb, or in6pcb */
876 	void *aux;
877 {
878 	struct tcpcb *tp;
879 	int i;
880 
881 	switch (family) {
882 	case PF_INET:
883 		break;
884 #ifdef INET6
885 	case PF_INET6:
886 		break;
887 #endif
888 	default:
889 		return NULL;
890 	}
891 
892 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
893 	if (tp == NULL)
894 		return (NULL);
895 	bzero((caddr_t)tp, sizeof(struct tcpcb));
896 	TAILQ_INIT(&tp->segq);
897 	TAILQ_INIT(&tp->timeq);
898 	tp->t_family = family;		/* may be overridden later on */
899 	tp->t_peermss = tcp_mssdflt;
900 	tp->t_ourmss = tcp_mssdflt;
901 	tp->t_segsz = tcp_mssdflt;
902 	LIST_INIT(&tp->t_sc);
903 
904 	callout_init(&tp->t_delack_ch);
905 	for (i = 0; i < TCPT_NTIMERS; i++)
906 		TCP_TIMER_INIT(tp, i);
907 
908 	tp->t_flags = 0;
909 	if (tcp_do_rfc1323 && tcp_do_win_scale)
910 		tp->t_flags |= TF_REQ_SCALE;
911 	if (tcp_do_rfc1323 && tcp_do_timestamps)
912 		tp->t_flags |= TF_REQ_TSTMP;
913 	if (tcp_do_sack == 2)
914 		tp->t_flags |= TF_WILL_SACK;
915 	else if (tcp_do_sack == 1)
916 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
917 	tp->t_flags |= TF_CANT_TXSACK;
918 	switch (family) {
919 	case PF_INET:
920 		tp->t_inpcb = (struct inpcb *)aux;
921 		tp->t_mtudisc = ip_mtudisc;
922 		break;
923 #ifdef INET6
924 	case PF_INET6:
925 		tp->t_in6pcb = (struct in6pcb *)aux;
926 		/* for IPv6, always try to run path MTU discovery */
927 		tp->t_mtudisc = 1;
928 		break;
929 #endif
930 	}
931 	/*
932 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
933 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
934 	 * reasonable initial retransmit time.
935 	 */
936 	tp->t_srtt = TCPTV_SRTTBASE;
937 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
938 	tp->t_rttmin = TCPTV_MIN;
939 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
940 	    TCPTV_MIN, TCPTV_REXMTMAX);
941 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
942 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
943 	if (family == AF_INET) {
944 		struct inpcb *inp = (struct inpcb *)aux;
945 		inp->inp_ip.ip_ttl = ip_defttl;
946 		inp->inp_ppcb = (caddr_t)tp;
947 	}
948 #ifdef INET6
949 	else if (family == AF_INET6) {
950 		struct in6pcb *in6p = (struct in6pcb *)aux;
951 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
952 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
953 					       : NULL);
954 		in6p->in6p_ppcb = (caddr_t)tp;
955 	}
956 #endif
957 
958 	/*
959 	 * Initialize our timebase.  When we send timestamps, we take
960 	 * the delta from tcp_now -- this means each connection always
961 	 * gets a timebase of 0, which makes it, among other things,
962 	 * more difficult to determine how long a system has been up,
963 	 * and thus how many TCP sequence increments have occurred.
964 	 */
965 	tp->ts_timebase = tcp_now;
966 
967 	return (tp);
968 }
969 
970 /*
971  * Drop a TCP connection, reporting
972  * the specified error.  If connection is synchronized,
973  * then send a RST to peer.
974  */
975 struct tcpcb *
976 tcp_drop(tp, errno)
977 	struct tcpcb *tp;
978 	int errno;
979 {
980 	struct socket *so = NULL;
981 
982 #ifdef DIAGNOSTIC
983 	if (tp->t_inpcb && tp->t_in6pcb)
984 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
985 #endif
986 #ifdef INET
987 	if (tp->t_inpcb)
988 		so = tp->t_inpcb->inp_socket;
989 #endif
990 #ifdef INET6
991 	if (tp->t_in6pcb)
992 		so = tp->t_in6pcb->in6p_socket;
993 #endif
994 	if (!so)
995 		return NULL;
996 
997 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
998 		tp->t_state = TCPS_CLOSED;
999 		(void) tcp_output(tp);
1000 		tcpstat.tcps_drops++;
1001 	} else
1002 		tcpstat.tcps_conndrops++;
1003 	if (errno == ETIMEDOUT && tp->t_softerror)
1004 		errno = tp->t_softerror;
1005 	so->so_error = errno;
1006 	return (tcp_close(tp));
1007 }
1008 
1009 /*
1010  * Close a TCP control block:
1011  *	discard all space held by the tcp
1012  *	discard internet protocol block
1013  *	wake up any sleepers
1014  */
1015 struct tcpcb *
1016 tcp_close(tp)
1017 	struct tcpcb *tp;
1018 {
1019 	struct inpcb *inp;
1020 #ifdef INET6
1021 	struct in6pcb *in6p;
1022 #endif
1023 	struct socket *so;
1024 #ifdef RTV_RTT
1025 	struct rtentry *rt;
1026 #endif
1027 	struct route *ro;
1028 
1029 	inp = tp->t_inpcb;
1030 #ifdef INET6
1031 	in6p = tp->t_in6pcb;
1032 #endif
1033 	so = NULL;
1034 	ro = NULL;
1035 	if (inp) {
1036 		so = inp->inp_socket;
1037 		ro = &inp->inp_route;
1038 	}
1039 #ifdef INET6
1040 	else if (in6p) {
1041 		so = in6p->in6p_socket;
1042 		ro = (struct route *)&in6p->in6p_route;
1043 	}
1044 #endif
1045 
1046 #ifdef RTV_RTT
1047 	/*
1048 	 * If we sent enough data to get some meaningful characteristics,
1049 	 * save them in the routing entry.  'Enough' is arbitrarily
1050 	 * defined as the sendpipesize (default 4K) * 16.  This would
1051 	 * give us 16 rtt samples assuming we only get one sample per
1052 	 * window (the usual case on a long haul net).  16 samples is
1053 	 * enough for the srtt filter to converge to within 5% of the correct
1054 	 * value; fewer samples and we could save a very bogus rtt.
1055 	 *
1056 	 * Don't update the default route's characteristics and don't
1057 	 * update anything that the user "locked".
1058 	 */
1059 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1060 	    ro && (rt = ro->ro_rt) &&
1061 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1062 		u_long i = 0;
1063 
1064 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1065 			i = tp->t_srtt *
1066 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1067 			if (rt->rt_rmx.rmx_rtt && i)
1068 				/*
1069 				 * filter this update to half the old & half
1070 				 * the new values, converting scale.
1071 				 * See route.h and tcp_var.h for a
1072 				 * description of the scaling constants.
1073 				 */
1074 				rt->rt_rmx.rmx_rtt =
1075 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1076 			else
1077 				rt->rt_rmx.rmx_rtt = i;
1078 		}
1079 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1080 			i = tp->t_rttvar *
1081 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1082 			if (rt->rt_rmx.rmx_rttvar && i)
1083 				rt->rt_rmx.rmx_rttvar =
1084 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1085 			else
1086 				rt->rt_rmx.rmx_rttvar = i;
1087 		}
1088 		/*
1089 		 * update the pipelimit (ssthresh) if it has been updated
1090 		 * already or if a pipesize was specified & the threshhold
1091 		 * got below half the pipesize.  I.e., wait for bad news
1092 		 * before we start updating, then update on both good
1093 		 * and bad news.
1094 		 */
1095 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1096 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1097 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1098 			/*
1099 			 * convert the limit from user data bytes to
1100 			 * packets then to packet data bytes.
1101 			 */
1102 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1103 			if (i < 2)
1104 				i = 2;
1105 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1106 			if (rt->rt_rmx.rmx_ssthresh)
1107 				rt->rt_rmx.rmx_ssthresh =
1108 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1109 			else
1110 				rt->rt_rmx.rmx_ssthresh = i;
1111 		}
1112 	}
1113 #endif /* RTV_RTT */
1114 	/* free the reassembly queue, if any */
1115 	TCP_REASS_LOCK(tp);
1116 	(void) tcp_freeq(tp);
1117 	TCP_REASS_UNLOCK(tp);
1118 
1119 	tcp_canceltimers(tp);
1120 	TCP_CLEAR_DELACK(tp);
1121 	syn_cache_cleanup(tp);
1122 
1123 	if (tp->t_template) {
1124 		m_free(tp->t_template);
1125 		tp->t_template = NULL;
1126 	}
1127 	pool_put(&tcpcb_pool, tp);
1128 	if (inp) {
1129 		inp->inp_ppcb = 0;
1130 		soisdisconnected(so);
1131 		in_pcbdetach(inp);
1132 	}
1133 #ifdef INET6
1134 	else if (in6p) {
1135 		in6p->in6p_ppcb = 0;
1136 		soisdisconnected(so);
1137 		in6_pcbdetach(in6p);
1138 	}
1139 #endif
1140 	tcpstat.tcps_closed++;
1141 	return ((struct tcpcb *)0);
1142 }
1143 
1144 int
1145 tcp_freeq(tp)
1146 	struct tcpcb *tp;
1147 {
1148 	struct ipqent *qe;
1149 	int rv = 0;
1150 #ifdef TCPREASS_DEBUG
1151 	int i = 0;
1152 #endif
1153 
1154 	TCP_REASS_LOCK_CHECK(tp);
1155 
1156 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1157 #ifdef TCPREASS_DEBUG
1158 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1159 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1160 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1161 #endif
1162 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1163 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1164 		m_freem(qe->ipqe_m);
1165 		pool_put(&ipqent_pool, qe);
1166 		rv = 1;
1167 	}
1168 	return (rv);
1169 }
1170 
1171 /*
1172  * Protocol drain routine.  Called when memory is in short supply.
1173  */
1174 void
1175 tcp_drain()
1176 {
1177 	struct inpcb *inp;
1178 	struct tcpcb *tp;
1179 
1180 	/*
1181 	 * Free the sequence queue of all TCP connections.
1182 	 */
1183 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1184 	if (inp)						/* XXX */
1185 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1186 		if ((tp = intotcpcb(inp)) != NULL) {
1187 			/*
1188 			 * We may be called from a device's interrupt
1189 			 * context.  If the tcpcb is already busy,
1190 			 * just bail out now.
1191 			 */
1192 			if (tcp_reass_lock_try(tp) == 0)
1193 				continue;
1194 			if (tcp_freeq(tp))
1195 				tcpstat.tcps_connsdrained++;
1196 			TCP_REASS_UNLOCK(tp);
1197 		}
1198 	}
1199 }
1200 
1201 #ifdef INET6
1202 void
1203 tcp6_drain()
1204 {
1205 	struct in6pcb *in6p;
1206 	struct tcpcb *tp;
1207 	struct in6pcb *head = &tcb6;
1208 
1209 	/*
1210 	 * Free the sequence queue of all TCP connections.
1211 	 */
1212 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1213 		if ((tp = in6totcpcb(in6p)) != NULL) {
1214 			/*
1215 			 * We may be called from a device's interrupt
1216 			 * context.  If the tcpcb is already busy,
1217 			 * just bail out now.
1218 			 */
1219 			if (tcp_reass_lock_try(tp) == 0)
1220 				continue;
1221 			if (tcp_freeq(tp))
1222 				tcpstat.tcps_connsdrained++;
1223 			TCP_REASS_UNLOCK(tp);
1224 		}
1225 	}
1226 }
1227 #endif
1228 
1229 /*
1230  * Notify a tcp user of an asynchronous error;
1231  * store error as soft error, but wake up user
1232  * (for now, won't do anything until can select for soft error).
1233  */
1234 void
1235 tcp_notify(inp, error)
1236 	struct inpcb *inp;
1237 	int error;
1238 {
1239 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1240 	struct socket *so = inp->inp_socket;
1241 
1242 	/*
1243 	 * Ignore some errors if we are hooked up.
1244 	 * If connection hasn't completed, has retransmitted several times,
1245 	 * and receives a second error, give up now.  This is better
1246 	 * than waiting a long time to establish a connection that
1247 	 * can never complete.
1248 	 */
1249 	if (tp->t_state == TCPS_ESTABLISHED &&
1250 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1251 	      error == EHOSTDOWN)) {
1252 		return;
1253 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1254 	    tp->t_rxtshift > 3 && tp->t_softerror)
1255 		so->so_error = error;
1256 	else
1257 		tp->t_softerror = error;
1258 	wakeup((caddr_t) &so->so_timeo);
1259 	sorwakeup(so);
1260 	sowwakeup(so);
1261 }
1262 
1263 #ifdef INET6
1264 void
1265 tcp6_notify(in6p, error)
1266 	struct in6pcb *in6p;
1267 	int error;
1268 {
1269 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1270 	struct socket *so = in6p->in6p_socket;
1271 
1272 	/*
1273 	 * Ignore some errors if we are hooked up.
1274 	 * If connection hasn't completed, has retransmitted several times,
1275 	 * and receives a second error, give up now.  This is better
1276 	 * than waiting a long time to establish a connection that
1277 	 * can never complete.
1278 	 */
1279 	if (tp->t_state == TCPS_ESTABLISHED &&
1280 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1281 	      error == EHOSTDOWN)) {
1282 		return;
1283 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1284 	    tp->t_rxtshift > 3 && tp->t_softerror)
1285 		so->so_error = error;
1286 	else
1287 		tp->t_softerror = error;
1288 	wakeup((caddr_t) &so->so_timeo);
1289 	sorwakeup(so);
1290 	sowwakeup(so);
1291 }
1292 #endif
1293 
1294 #ifdef INET6
1295 void
1296 tcp6_ctlinput(cmd, sa, d)
1297 	int cmd;
1298 	struct sockaddr *sa;
1299 	void *d;
1300 {
1301 	struct tcphdr th;
1302 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1303 	int nmatch;
1304 	struct ip6_hdr *ip6;
1305 	const struct sockaddr_in6 *sa6_src = NULL;
1306 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1307 	struct mbuf *m;
1308 	int off;
1309 
1310 	if (sa->sa_family != AF_INET6 ||
1311 	    sa->sa_len != sizeof(struct sockaddr_in6))
1312 		return;
1313 	if ((unsigned)cmd >= PRC_NCMDS)
1314 		return;
1315 	else if (cmd == PRC_QUENCH) {
1316 		/* XXX there's no PRC_QUENCH in IPv6 */
1317 		notify = tcp6_quench;
1318 	} else if (PRC_IS_REDIRECT(cmd))
1319 		notify = in6_rtchange, d = NULL;
1320 	else if (cmd == PRC_MSGSIZE)
1321 		; /* special code is present, see below */
1322 	else if (cmd == PRC_HOSTDEAD)
1323 		d = NULL;
1324 	else if (inet6ctlerrmap[cmd] == 0)
1325 		return;
1326 
1327 	/* if the parameter is from icmp6, decode it. */
1328 	if (d != NULL) {
1329 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1330 		m = ip6cp->ip6c_m;
1331 		ip6 = ip6cp->ip6c_ip6;
1332 		off = ip6cp->ip6c_off;
1333 		sa6_src = ip6cp->ip6c_src;
1334 	} else {
1335 		m = NULL;
1336 		ip6 = NULL;
1337 		sa6_src = &sa6_any;
1338 	}
1339 
1340 	if (ip6) {
1341 		/*
1342 		 * XXX: We assume that when ip6 is non NULL,
1343 		 * M and OFF are valid.
1344 		 */
1345 
1346 		/* check if we can safely examine src and dst ports */
1347 		if (m->m_pkthdr.len < off + sizeof(th)) {
1348 			if (cmd == PRC_MSGSIZE)
1349 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1350 			return;
1351 		}
1352 
1353 		bzero(&th, sizeof(th));
1354 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1355 
1356 		if (cmd == PRC_MSGSIZE) {
1357 			int valid = 0;
1358 
1359 			/*
1360 			 * Check to see if we have a valid TCP connection
1361 			 * corresponding to the address in the ICMPv6 message
1362 			 * payload.
1363 			 */
1364 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1365 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1366 			    th.th_sport, 0))
1367 				valid++;
1368 
1369 			/*
1370 			 * Depending on the value of "valid" and routing table
1371 			 * size (mtudisc_{hi,lo}wat), we will:
1372 			 * - recalcurate the new MTU and create the
1373 			 *   corresponding routing entry, or
1374 			 * - ignore the MTU change notification.
1375 			 */
1376 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1377 
1378 			/*
1379 			 * no need to call in6_pcbnotify, it should have been
1380 			 * called via callback if necessary
1381 			 */
1382 			return;
1383 		}
1384 
1385 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1386 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1387 		if (nmatch == 0 && syn_cache_count &&
1388 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1389 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1390 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1391 			syn_cache_unreach((struct sockaddr *)sa6_src,
1392 					  sa, &th);
1393 	} else {
1394 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1395 		    0, cmd, NULL, notify);
1396 	}
1397 }
1398 #endif
1399 
1400 #ifdef INET
1401 /* assumes that ip header and tcp header are contiguous on mbuf */
1402 void *
1403 tcp_ctlinput(cmd, sa, v)
1404 	int cmd;
1405 	struct sockaddr *sa;
1406 	void *v;
1407 {
1408 	struct ip *ip = v;
1409 	struct tcphdr *th;
1410 	struct icmp *icp;
1411 	extern const int inetctlerrmap[];
1412 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1413 	int errno;
1414 	int nmatch;
1415 #ifdef INET6
1416 	struct in6_addr src6, dst6;
1417 #endif
1418 
1419 	if (sa->sa_family != AF_INET ||
1420 	    sa->sa_len != sizeof(struct sockaddr_in))
1421 		return NULL;
1422 	if ((unsigned)cmd >= PRC_NCMDS)
1423 		return NULL;
1424 	errno = inetctlerrmap[cmd];
1425 	if (cmd == PRC_QUENCH)
1426 		notify = tcp_quench;
1427 	else if (PRC_IS_REDIRECT(cmd))
1428 		notify = in_rtchange, ip = 0;
1429 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1430 		/*
1431 		 * Check to see if we have a valid TCP connection
1432 		 * corresponding to the address in the ICMP message
1433 		 * payload.
1434 		 *
1435 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1436 		 */
1437 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1438 #ifdef INET6
1439 		memset(&src6, 0, sizeof(src6));
1440 		memset(&dst6, 0, sizeof(dst6));
1441 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1442 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1443 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1444 #endif
1445 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1446 		    ip->ip_src, th->th_sport) != NULL)
1447 			;
1448 #ifdef INET6
1449 		else if (in6_pcblookup_connect(&tcb6, &dst6,
1450 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1451 			;
1452 #endif
1453 		else
1454 			return NULL;
1455 
1456 		/*
1457 		 * Now that we've validated that we are actually communicating
1458 		 * with the host indicated in the ICMP message, locate the
1459 		 * ICMP header, recalculate the new MTU, and create the
1460 		 * corresponding routing entry.
1461 		 */
1462 		icp = (struct icmp *)((caddr_t)ip -
1463 		    offsetof(struct icmp, icmp_ip));
1464 		icmp_mtudisc(icp, ip->ip_dst);
1465 
1466 		return NULL;
1467 	} else if (cmd == PRC_HOSTDEAD)
1468 		ip = 0;
1469 	else if (errno == 0)
1470 		return NULL;
1471 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1472 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1473 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1474 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1475 		if (nmatch == 0 && syn_cache_count &&
1476 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1477 		    inetctlerrmap[cmd] == ENETUNREACH ||
1478 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1479 			struct sockaddr_in sin;
1480 			bzero(&sin, sizeof(sin));
1481 			sin.sin_len = sizeof(sin);
1482 			sin.sin_family = AF_INET;
1483 			sin.sin_port = th->th_sport;
1484 			sin.sin_addr = ip->ip_src;
1485 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1486 		}
1487 
1488 		/* XXX mapped address case */
1489 	} else
1490 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1491 		    notify);
1492 	return NULL;
1493 }
1494 
1495 /*
1496  * When a source quence is received, we are being notifed of congestion.
1497  * Close the congestion window down to the Loss Window (one segment).
1498  * We will gradually open it again as we proceed.
1499  */
1500 void
1501 tcp_quench(inp, errno)
1502 	struct inpcb *inp;
1503 	int errno;
1504 {
1505 	struct tcpcb *tp = intotcpcb(inp);
1506 
1507 	if (tp)
1508 		tp->snd_cwnd = tp->t_segsz;
1509 }
1510 #endif
1511 
1512 #ifdef INET6
1513 void
1514 tcp6_quench(in6p, errno)
1515 	struct in6pcb *in6p;
1516 	int errno;
1517 {
1518 	struct tcpcb *tp = in6totcpcb(in6p);
1519 
1520 	if (tp)
1521 		tp->snd_cwnd = tp->t_segsz;
1522 }
1523 #endif
1524 
1525 #ifdef INET
1526 /*
1527  * Path MTU Discovery handlers.
1528  */
1529 void
1530 tcp_mtudisc_callback(faddr)
1531 	struct in_addr faddr;
1532 {
1533 #ifdef INET6
1534 	struct in6_addr in6;
1535 #endif
1536 
1537 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1538 #ifdef INET6
1539 	memset(&in6, 0, sizeof(in6));
1540 	in6.s6_addr16[5] = 0xffff;
1541 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1542 	tcp6_mtudisc_callback(&in6);
1543 #endif
1544 }
1545 
1546 /*
1547  * On receipt of path MTU corrections, flush old route and replace it
1548  * with the new one.  Retransmit all unacknowledged packets, to ensure
1549  * that all packets will be received.
1550  */
1551 void
1552 tcp_mtudisc(inp, errno)
1553 	struct inpcb *inp;
1554 	int errno;
1555 {
1556 	struct tcpcb *tp = intotcpcb(inp);
1557 	struct rtentry *rt = in_pcbrtentry(inp);
1558 
1559 	if (tp != 0) {
1560 		if (rt != 0) {
1561 			/*
1562 			 * If this was not a host route, remove and realloc.
1563 			 */
1564 			if ((rt->rt_flags & RTF_HOST) == 0) {
1565 				in_rtchange(inp, errno);
1566 				if ((rt = in_pcbrtentry(inp)) == 0)
1567 					return;
1568 			}
1569 
1570 			/*
1571 			 * Slow start out of the error condition.  We
1572 			 * use the MTU because we know it's smaller
1573 			 * than the previously transmitted segment.
1574 			 *
1575 			 * Note: This is more conservative than the
1576 			 * suggestion in draft-floyd-incr-init-win-03.
1577 			 */
1578 			if (rt->rt_rmx.rmx_mtu != 0)
1579 				tp->snd_cwnd =
1580 				    TCP_INITIAL_WINDOW(tcp_init_win,
1581 				    rt->rt_rmx.rmx_mtu);
1582 		}
1583 
1584 		/*
1585 		 * Resend unacknowledged packets.
1586 		 */
1587 		tp->snd_nxt = tp->snd_una;
1588 		tcp_output(tp);
1589 	}
1590 }
1591 #endif
1592 
1593 #ifdef INET6
1594 /*
1595  * Path MTU Discovery handlers.
1596  */
1597 void
1598 tcp6_mtudisc_callback(faddr)
1599 	struct in6_addr *faddr;
1600 {
1601 	struct sockaddr_in6 sin6;
1602 
1603 	bzero(&sin6, sizeof(sin6));
1604 	sin6.sin6_family = AF_INET6;
1605 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1606 	sin6.sin6_addr = *faddr;
1607 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1608 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1609 }
1610 
1611 void
1612 tcp6_mtudisc(in6p, errno)
1613 	struct in6pcb *in6p;
1614 	int errno;
1615 {
1616 	struct tcpcb *tp = in6totcpcb(in6p);
1617 	struct rtentry *rt = in6_pcbrtentry(in6p);
1618 
1619 	if (tp != 0) {
1620 		if (rt != 0) {
1621 			/*
1622 			 * If this was not a host route, remove and realloc.
1623 			 */
1624 			if ((rt->rt_flags & RTF_HOST) == 0) {
1625 				in6_rtchange(in6p, errno);
1626 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1627 					return;
1628 			}
1629 
1630 			/*
1631 			 * Slow start out of the error condition.  We
1632 			 * use the MTU because we know it's smaller
1633 			 * than the previously transmitted segment.
1634 			 *
1635 			 * Note: This is more conservative than the
1636 			 * suggestion in draft-floyd-incr-init-win-03.
1637 			 */
1638 			if (rt->rt_rmx.rmx_mtu != 0)
1639 				tp->snd_cwnd =
1640 				    TCP_INITIAL_WINDOW(tcp_init_win,
1641 				    rt->rt_rmx.rmx_mtu);
1642 		}
1643 
1644 		/*
1645 		 * Resend unacknowledged packets.
1646 		 */
1647 		tp->snd_nxt = tp->snd_una;
1648 		tcp_output(tp);
1649 	}
1650 }
1651 #endif /* INET6 */
1652 
1653 /*
1654  * Compute the MSS to advertise to the peer.  Called only during
1655  * the 3-way handshake.  If we are the server (peer initiated
1656  * connection), we are called with a pointer to the interface
1657  * on which the SYN packet arrived.  If we are the client (we
1658  * initiated connection), we are called with a pointer to the
1659  * interface out which this connection should go.
1660  *
1661  * NOTE: Do not subtract IP option/extension header size nor IPsec
1662  * header size from MSS advertisement.  MSS option must hold the maximum
1663  * segment size we can accept, so it must always be:
1664  *	 max(if mtu) - ip header - tcp header
1665  */
1666 u_long
1667 tcp_mss_to_advertise(ifp, af)
1668 	const struct ifnet *ifp;
1669 	int af;
1670 {
1671 	extern u_long in_maxmtu;
1672 	u_long mss = 0;
1673 	u_long hdrsiz;
1674 
1675 	/*
1676 	 * In order to avoid defeating path MTU discovery on the peer,
1677 	 * we advertise the max MTU of all attached networks as our MSS,
1678 	 * per RFC 1191, section 3.1.
1679 	 *
1680 	 * We provide the option to advertise just the MTU of
1681 	 * the interface on which we hope this connection will
1682 	 * be receiving.  If we are responding to a SYN, we
1683 	 * will have a pretty good idea about this, but when
1684 	 * initiating a connection there is a bit more doubt.
1685 	 *
1686 	 * We also need to ensure that loopback has a large enough
1687 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1688 	 */
1689 
1690 	if (ifp != NULL)
1691 		switch (af) {
1692 		case AF_INET:
1693 			mss = ifp->if_mtu;
1694 			break;
1695 #ifdef INET6
1696 		case AF_INET6:
1697 			mss = IN6_LINKMTU(ifp);
1698 			break;
1699 #endif
1700 		}
1701 
1702 	if (tcp_mss_ifmtu == 0)
1703 		switch (af) {
1704 		case AF_INET:
1705 			mss = max(in_maxmtu, mss);
1706 			break;
1707 #ifdef INET6
1708 		case AF_INET6:
1709 			mss = max(in6_maxmtu, mss);
1710 			break;
1711 #endif
1712 		}
1713 
1714 	switch (af) {
1715 	case AF_INET:
1716 		hdrsiz = sizeof(struct ip);
1717 		break;
1718 #ifdef INET6
1719 	case AF_INET6:
1720 		hdrsiz = sizeof(struct ip6_hdr);
1721 		break;
1722 #endif
1723 	default:
1724 		hdrsiz = 0;
1725 		break;
1726 	}
1727 	hdrsiz += sizeof(struct tcphdr);
1728 	if (mss > hdrsiz)
1729 		mss -= hdrsiz;
1730 
1731 	mss = max(tcp_mssdflt, mss);
1732 	return (mss);
1733 }
1734 
1735 /*
1736  * Set connection variables based on the peer's advertised MSS.
1737  * We are passed the TCPCB for the actual connection.  If we
1738  * are the server, we are called by the compressed state engine
1739  * when the 3-way handshake is complete.  If we are the client,
1740  * we are called when we receive the SYN,ACK from the server.
1741  *
1742  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1743  * before this routine is called!
1744  */
1745 void
1746 tcp_mss_from_peer(tp, offer)
1747 	struct tcpcb *tp;
1748 	int offer;
1749 {
1750 	struct socket *so;
1751 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1752 	struct rtentry *rt;
1753 #endif
1754 	u_long bufsize;
1755 	int mss;
1756 
1757 #ifdef DIAGNOSTIC
1758 	if (tp->t_inpcb && tp->t_in6pcb)
1759 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1760 #endif
1761 	so = NULL;
1762 	rt = NULL;
1763 #ifdef INET
1764 	if (tp->t_inpcb) {
1765 		so = tp->t_inpcb->inp_socket;
1766 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1767 		rt = in_pcbrtentry(tp->t_inpcb);
1768 #endif
1769 	}
1770 #endif
1771 #ifdef INET6
1772 	if (tp->t_in6pcb) {
1773 		so = tp->t_in6pcb->in6p_socket;
1774 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1775 		rt = in6_pcbrtentry(tp->t_in6pcb);
1776 #endif
1777 	}
1778 #endif
1779 
1780 	/*
1781 	 * As per RFC1122, use the default MSS value, unless they
1782 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1783 	 */
1784 	mss = tcp_mssdflt;
1785 	if (offer)
1786 		mss = offer;
1787 	mss = max(mss, 32);		/* sanity */
1788 	tp->t_peermss = mss;
1789 	mss -= tcp_optlen(tp);
1790 #ifdef INET
1791 	if (tp->t_inpcb)
1792 		mss -= ip_optlen(tp->t_inpcb);
1793 #endif
1794 #ifdef INET6
1795 	if (tp->t_in6pcb)
1796 		mss -= ip6_optlen(tp->t_in6pcb);
1797 #endif
1798 
1799 	/*
1800 	 * If there's a pipesize, change the socket buffer to that size.
1801 	 * Make the socket buffer an integral number of MSS units.  If
1802 	 * the MSS is larger than the socket buffer, artificially decrease
1803 	 * the MSS.
1804 	 */
1805 #ifdef RTV_SPIPE
1806 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1807 		bufsize = rt->rt_rmx.rmx_sendpipe;
1808 	else
1809 #endif
1810 		bufsize = so->so_snd.sb_hiwat;
1811 	if (bufsize < mss)
1812 		mss = bufsize;
1813 	else {
1814 		bufsize = roundup(bufsize, mss);
1815 		if (bufsize > sb_max)
1816 			bufsize = sb_max;
1817 		(void) sbreserve(&so->so_snd, bufsize);
1818 	}
1819 	tp->t_segsz = mss;
1820 
1821 #ifdef RTV_SSTHRESH
1822 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1823 		/*
1824 		 * There's some sort of gateway or interface buffer
1825 		 * limit on the path.  Use this to set the slow
1826 		 * start threshold, but set the threshold to no less
1827 		 * than 2 * MSS.
1828 		 */
1829 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1830 	}
1831 #endif
1832 }
1833 
1834 /*
1835  * Processing necessary when a TCP connection is established.
1836  */
1837 void
1838 tcp_established(tp)
1839 	struct tcpcb *tp;
1840 {
1841 	struct socket *so;
1842 #ifdef RTV_RPIPE
1843 	struct rtentry *rt;
1844 #endif
1845 	u_long bufsize;
1846 
1847 #ifdef DIAGNOSTIC
1848 	if (tp->t_inpcb && tp->t_in6pcb)
1849 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1850 #endif
1851 	so = NULL;
1852 	rt = NULL;
1853 #ifdef INET
1854 	if (tp->t_inpcb) {
1855 		so = tp->t_inpcb->inp_socket;
1856 #if defined(RTV_RPIPE)
1857 		rt = in_pcbrtentry(tp->t_inpcb);
1858 #endif
1859 	}
1860 #endif
1861 #ifdef INET6
1862 	if (tp->t_in6pcb) {
1863 		so = tp->t_in6pcb->in6p_socket;
1864 #if defined(RTV_RPIPE)
1865 		rt = in6_pcbrtentry(tp->t_in6pcb);
1866 #endif
1867 	}
1868 #endif
1869 
1870 	tp->t_state = TCPS_ESTABLISHED;
1871 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1872 
1873 #ifdef RTV_RPIPE
1874 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1875 		bufsize = rt->rt_rmx.rmx_recvpipe;
1876 	else
1877 #endif
1878 		bufsize = so->so_rcv.sb_hiwat;
1879 	if (bufsize > tp->t_ourmss) {
1880 		bufsize = roundup(bufsize, tp->t_ourmss);
1881 		if (bufsize > sb_max)
1882 			bufsize = sb_max;
1883 		(void) sbreserve(&so->so_rcv, bufsize);
1884 	}
1885 }
1886 
1887 /*
1888  * Check if there's an initial rtt or rttvar.  Convert from the
1889  * route-table units to scaled multiples of the slow timeout timer.
1890  * Called only during the 3-way handshake.
1891  */
1892 void
1893 tcp_rmx_rtt(tp)
1894 	struct tcpcb *tp;
1895 {
1896 #ifdef RTV_RTT
1897 	struct rtentry *rt = NULL;
1898 	int rtt;
1899 
1900 #ifdef DIAGNOSTIC
1901 	if (tp->t_inpcb && tp->t_in6pcb)
1902 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1903 #endif
1904 #ifdef INET
1905 	if (tp->t_inpcb)
1906 		rt = in_pcbrtentry(tp->t_inpcb);
1907 #endif
1908 #ifdef INET6
1909 	if (tp->t_in6pcb)
1910 		rt = in6_pcbrtentry(tp->t_in6pcb);
1911 #endif
1912 	if (rt == NULL)
1913 		return;
1914 
1915 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1916 		/*
1917 		 * XXX The lock bit for MTU indicates that the value
1918 		 * is also a minimum value; this is subject to time.
1919 		 */
1920 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1921 			TCPT_RANGESET(tp->t_rttmin,
1922 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1923 			    TCPTV_MIN, TCPTV_REXMTMAX);
1924 		tp->t_srtt = rtt /
1925 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1926 		if (rt->rt_rmx.rmx_rttvar) {
1927 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1928 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1929 				(TCP_RTTVAR_SHIFT + 2));
1930 		} else {
1931 			/* Default variation is +- 1 rtt */
1932 			tp->t_rttvar =
1933 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1934 		}
1935 		TCPT_RANGESET(tp->t_rxtcur,
1936 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1937 		    tp->t_rttmin, TCPTV_REXMTMAX);
1938 	}
1939 #endif
1940 }
1941 
1942 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1943 #if NRND > 0
1944 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1945 #endif
1946 
1947 /*
1948  * Get a new sequence value given a tcp control block
1949  */
1950 tcp_seq
1951 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1952 {
1953 
1954 #ifdef INET
1955 	if (tp->t_inpcb != NULL) {
1956 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1957 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1958 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1959 		    addin));
1960 	}
1961 #endif
1962 #ifdef INET6
1963 	if (tp->t_in6pcb != NULL) {
1964 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
1965 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
1966 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
1967 		    addin));
1968 	}
1969 #endif
1970 	/* Not possible. */
1971 	panic("tcp_new_iss");
1972 }
1973 
1974 /*
1975  * This routine actually generates a new TCP initial sequence number.
1976  */
1977 tcp_seq
1978 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
1979     size_t addrsz, tcp_seq addin)
1980 {
1981 	tcp_seq tcp_iss;
1982 
1983 #if NRND > 0
1984 	static int beenhere;
1985 
1986 	/*
1987 	 * If we haven't been here before, initialize our cryptographic
1988 	 * hash secret.
1989 	 */
1990 	if (beenhere == 0) {
1991 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
1992 		    RND_EXTRACT_ANY);
1993 		beenhere = 1;
1994 	}
1995 
1996 	if (tcp_do_rfc1948) {
1997 		MD5_CTX ctx;
1998 		u_int8_t hash[16];	/* XXX MD5 knowledge */
1999 
2000 		/*
2001 		 * Compute the base value of the ISS.  It is a hash
2002 		 * of (saddr, sport, daddr, dport, secret).
2003 		 */
2004 		MD5Init(&ctx);
2005 
2006 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2007 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2008 
2009 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2010 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2011 
2012 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2013 
2014 		MD5Final(hash, &ctx);
2015 
2016 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2017 
2018 		/*
2019 		 * Now increment our "timer", and add it in to
2020 		 * the computed value.
2021 		 *
2022 		 * XXX Use `addin'?
2023 		 * XXX TCP_ISSINCR too large to use?
2024 		 */
2025 		tcp_iss_seq += TCP_ISSINCR;
2026 #ifdef TCPISS_DEBUG
2027 		printf("ISS hash 0x%08x, ", tcp_iss);
2028 #endif
2029 		tcp_iss += tcp_iss_seq + addin;
2030 #ifdef TCPISS_DEBUG
2031 		printf("new ISS 0x%08x\n", tcp_iss);
2032 #endif
2033 	} else
2034 #endif /* NRND > 0 */
2035 	{
2036 		/*
2037 		 * Randomize.
2038 		 */
2039 #if NRND > 0
2040 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2041 #else
2042 		tcp_iss = arc4random();
2043 #endif
2044 
2045 		/*
2046 		 * If we were asked to add some amount to a known value,
2047 		 * we will take a random value obtained above, mask off
2048 		 * the upper bits, and add in the known value.  We also
2049 		 * add in a constant to ensure that we are at least a
2050 		 * certain distance from the original value.
2051 		 *
2052 		 * This is used when an old connection is in timed wait
2053 		 * and we have a new one coming in, for instance.
2054 		 */
2055 		if (addin != 0) {
2056 #ifdef TCPISS_DEBUG
2057 			printf("Random %08x, ", tcp_iss);
2058 #endif
2059 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2060 			tcp_iss += addin + TCP_ISSINCR;
2061 #ifdef TCPISS_DEBUG
2062 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2063 #endif
2064 		} else {
2065 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2066 			tcp_iss += tcp_iss_seq;
2067 			tcp_iss_seq += TCP_ISSINCR;
2068 #ifdef TCPISS_DEBUG
2069 			printf("ISS %08x\n", tcp_iss);
2070 #endif
2071 		}
2072 	}
2073 
2074 	if (tcp_compat_42) {
2075 		/*
2076 		 * Limit it to the positive range for really old TCP
2077 		 * implementations.
2078 		 * Just AND off the top bit instead of checking if
2079 		 * is set first - saves a branch 50% of the time.
2080 		 */
2081 		tcp_iss &= 0x7fffffff;		/* XXX */
2082 	}
2083 
2084 	return (tcp_iss);
2085 }
2086 
2087 #ifdef IPSEC
2088 /* compute ESP/AH header size for TCP, including outer IP header. */
2089 size_t
2090 ipsec4_hdrsiz_tcp(tp)
2091 	struct tcpcb *tp;
2092 {
2093 	struct inpcb *inp;
2094 	size_t hdrsiz;
2095 
2096 	/* XXX mapped addr case (tp->t_in6pcb) */
2097 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2098 		return 0;
2099 	switch (tp->t_family) {
2100 	case AF_INET:
2101 		/* XXX: should use currect direction. */
2102 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2103 		break;
2104 	default:
2105 		hdrsiz = 0;
2106 		break;
2107 	}
2108 
2109 	return hdrsiz;
2110 }
2111 
2112 #ifdef INET6
2113 size_t
2114 ipsec6_hdrsiz_tcp(tp)
2115 	struct tcpcb *tp;
2116 {
2117 	struct in6pcb *in6p;
2118 	size_t hdrsiz;
2119 
2120 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2121 		return 0;
2122 	switch (tp->t_family) {
2123 	case AF_INET6:
2124 		/* XXX: should use currect direction. */
2125 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2126 		break;
2127 	case AF_INET:
2128 		/* mapped address case - tricky */
2129 	default:
2130 		hdrsiz = 0;
2131 		break;
2132 	}
2133 
2134 	return hdrsiz;
2135 }
2136 #endif
2137 #endif /*IPSEC*/
2138 
2139 /*
2140  * Determine the length of the TCP options for this connection.
2141  *
2142  * XXX:  What do we do for SACK, when we add that?  Just reserve
2143  *       all of the space?  Otherwise we can't exactly be incrementing
2144  *       cwnd by an amount that varies depending on the amount we last
2145  *       had to SACK!
2146  */
2147 
2148 u_int
2149 tcp_optlen(tp)
2150 	struct tcpcb *tp;
2151 {
2152 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2153 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2154 		return TCPOLEN_TSTAMP_APPA;
2155 	else
2156 		return 0;
2157 }
2158