xref: /netbsd-src/sys/netinet/tcp_subr.c (revision eb7c1594f145c931049e1fd9eb056a5987e87e59)
1 /*	$NetBSD: tcp_subr.c,v 1.145 2003/08/07 16:33:18 agc 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 1.145 2003/08/07 16:33:18 agc Exp $");
102 
103 #include "opt_inet.h"
104 #include "opt_ipsec.h"
105 #include "opt_tcp_compat_42.h"
106 #include "opt_inet_csum.h"
107 #include "opt_mbuftrace.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include <netinet/tcp.h>
150 #include <netinet/tcp_fsm.h>
151 #include <netinet/tcp_seq.h>
152 #include <netinet/tcp_timer.h>
153 #include <netinet/tcp_var.h>
154 #include <netinet/tcpip.h>
155 
156 #ifdef IPSEC
157 #include <netinet6/ipsec.h>
158 #endif /*IPSEC*/
159 
160 #ifdef INET6
161 struct in6pcb tcb6;
162 #endif
163 
164 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
165 struct	tcpstat tcpstat;	/* tcp statistics */
166 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
167 
168 /* patchable/settable parameters for tcp */
169 int 	tcp_mssdflt = TCP_MSS;
170 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
171 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
172 #if NRND > 0
173 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
174 #endif
175 int	tcp_do_sack = 1;	/* selective acknowledgement */
176 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
177 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
178 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
179 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
180 #ifndef TCP_INIT_WIN
181 #define	TCP_INIT_WIN	1	/* initial slow start window */
182 #endif
183 #ifndef TCP_INIT_WIN_LOCAL
184 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
185 #endif
186 int	tcp_init_win = TCP_INIT_WIN;
187 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
188 int	tcp_mss_ifmtu = 0;
189 #ifdef TCP_COMPAT_42
190 int	tcp_compat_42 = 1;
191 #else
192 int	tcp_compat_42 = 0;
193 #endif
194 int	tcp_rst_ppslim = 100;	/* 100pps */
195 
196 /* tcb hash */
197 #ifndef TCBHASHSIZE
198 #define	TCBHASHSIZE	128
199 #endif
200 int	tcbhashsize = TCBHASHSIZE;
201 
202 /* syn hash parameters */
203 #define	TCP_SYN_HASH_SIZE	293
204 #define	TCP_SYN_BUCKET_SIZE	35
205 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
206 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
207 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
208 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
209 
210 int	tcp_freeq __P((struct tcpcb *));
211 
212 #ifdef INET
213 void	tcp_mtudisc_callback __P((struct in_addr));
214 #endif
215 #ifdef INET6
216 void	tcp6_mtudisc_callback __P((struct in6_addr *));
217 #endif
218 
219 void	tcp_mtudisc __P((struct inpcb *, int));
220 #ifdef INET6
221 void	tcp6_mtudisc __P((struct in6pcb *, int));
222 #endif
223 
224 struct pool tcpcb_pool;
225 
226 #ifdef TCP_CSUM_COUNTERS
227 #include <sys/device.h>
228 
229 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
230     NULL, "tcp", "hwcsum bad");
231 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
232     NULL, "tcp", "hwcsum ok");
233 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
234     NULL, "tcp", "hwcsum data");
235 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
236     NULL, "tcp", "swcsum");
237 #endif /* TCP_CSUM_COUNTERS */
238 
239 #ifdef TCP_OUTPUT_COUNTERS
240 #include <sys/device.h>
241 
242 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243     NULL, "tcp", "output big header");
244 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245     NULL, "tcp", "output copy small");
246 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247     NULL, "tcp", "output copy big");
248 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
249     NULL, "tcp", "output reference big");
250 #endif /* TCP_OUTPUT_COUNTERS */
251 
252 #ifdef TCP_REASS_COUNTERS
253 #include <sys/device.h>
254 
255 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
256     NULL, "tcp_reass", "calls");
257 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
258     &tcp_reass_, "tcp_reass", "insert into empty queue");
259 struct evcnt tcp_reass_iteration[8] = {
260     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
261     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
262     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
263     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
264     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
265     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
266     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
267     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
268 };
269 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
270     &tcp_reass_, "tcp_reass", "prepend to first");
271 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
272     &tcp_reass_, "tcp_reass", "prepend");
273 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
274     &tcp_reass_, "tcp_reass", "insert");
275 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
276     &tcp_reass_, "tcp_reass", "insert at tail");
277 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     &tcp_reass_, "tcp_reass", "append");
279 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     &tcp_reass_, "tcp_reass", "append to tail fragment");
281 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     &tcp_reass_, "tcp_reass", "overlap at end");
283 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "overlap at start");
285 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286     &tcp_reass_, "tcp_reass", "duplicate segment");
287 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
288     &tcp_reass_, "tcp_reass", "duplicate fragment");
289 
290 #endif /* TCP_REASS_COUNTERS */
291 
292 #ifdef MBUFTRACE
293 struct mowner tcp_mowner = { "tcp" };
294 struct mowner tcp_rx_mowner = { "tcp", "rx" };
295 struct mowner tcp_tx_mowner = { "tcp", "tx" };
296 #endif
297 
298 /*
299  * Tcp initialization
300  */
301 void
302 tcp_init()
303 {
304 	int hlen;
305 
306 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
307 	    NULL);
308 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
309 #ifdef INET6
310 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
311 #endif
312 
313 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
314 #ifdef INET6
315 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
316 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
317 #endif
318 	if (max_protohdr < hlen)
319 		max_protohdr = hlen;
320 	if (max_linkhdr + hlen > MHLEN)
321 		panic("tcp_init");
322 
323 #ifdef INET
324 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
325 #endif
326 #ifdef INET6
327 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
328 #endif
329 
330 	/* Initialize timer state. */
331 	tcp_timer_init();
332 
333 	/* Initialize the compressed state engine. */
334 	syn_cache_init();
335 
336 #ifdef TCP_CSUM_COUNTERS
337 	evcnt_attach_static(&tcp_hwcsum_bad);
338 	evcnt_attach_static(&tcp_hwcsum_ok);
339 	evcnt_attach_static(&tcp_hwcsum_data);
340 	evcnt_attach_static(&tcp_swcsum);
341 #endif /* TCP_CSUM_COUNTERS */
342 
343 #ifdef TCP_OUTPUT_COUNTERS
344 	evcnt_attach_static(&tcp_output_bigheader);
345 	evcnt_attach_static(&tcp_output_copysmall);
346 	evcnt_attach_static(&tcp_output_copybig);
347 	evcnt_attach_static(&tcp_output_refbig);
348 #endif /* TCP_OUTPUT_COUNTERS */
349 
350 #ifdef TCP_REASS_COUNTERS
351 	evcnt_attach_static(&tcp_reass_);
352 	evcnt_attach_static(&tcp_reass_empty);
353 	evcnt_attach_static(&tcp_reass_iteration[0]);
354 	evcnt_attach_static(&tcp_reass_iteration[1]);
355 	evcnt_attach_static(&tcp_reass_iteration[2]);
356 	evcnt_attach_static(&tcp_reass_iteration[3]);
357 	evcnt_attach_static(&tcp_reass_iteration[4]);
358 	evcnt_attach_static(&tcp_reass_iteration[5]);
359 	evcnt_attach_static(&tcp_reass_iteration[6]);
360 	evcnt_attach_static(&tcp_reass_iteration[7]);
361 	evcnt_attach_static(&tcp_reass_prependfirst);
362 	evcnt_attach_static(&tcp_reass_prepend);
363 	evcnt_attach_static(&tcp_reass_insert);
364 	evcnt_attach_static(&tcp_reass_inserttail);
365 	evcnt_attach_static(&tcp_reass_append);
366 	evcnt_attach_static(&tcp_reass_appendtail);
367 	evcnt_attach_static(&tcp_reass_overlaptail);
368 	evcnt_attach_static(&tcp_reass_overlapfront);
369 	evcnt_attach_static(&tcp_reass_segdup);
370 	evcnt_attach_static(&tcp_reass_fragdup);
371 #endif /* TCP_REASS_COUNTERS */
372 
373 	MOWNER_ATTACH(&tcp_tx_mowner);
374 	MOWNER_ATTACH(&tcp_rx_mowner);
375 	MOWNER_ATTACH(&tcp_mowner);
376 }
377 
378 /*
379  * Create template to be used to send tcp packets on a connection.
380  * Call after host entry created, allocates an mbuf and fills
381  * in a skeletal tcp/ip header, minimizing the amount of work
382  * necessary when the connection is used.
383  */
384 struct mbuf *
385 tcp_template(tp)
386 	struct tcpcb *tp;
387 {
388 	struct inpcb *inp = tp->t_inpcb;
389 #ifdef INET6
390 	struct in6pcb *in6p = tp->t_in6pcb;
391 #endif
392 	struct tcphdr *n;
393 	struct mbuf *m;
394 	int hlen;
395 
396 	switch (tp->t_family) {
397 	case AF_INET:
398 		hlen = sizeof(struct ip);
399 		if (inp)
400 			break;
401 #ifdef INET6
402 		if (in6p) {
403 			/* mapped addr case */
404 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
405 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
406 				break;
407 		}
408 #endif
409 		return NULL;	/*EINVAL*/
410 #ifdef INET6
411 	case AF_INET6:
412 		hlen = sizeof(struct ip6_hdr);
413 		if (in6p) {
414 			/* more sainty check? */
415 			break;
416 		}
417 		return NULL;	/*EINVAL*/
418 #endif
419 	default:
420 		hlen = 0;	/*pacify gcc*/
421 		return NULL;	/*EAFNOSUPPORT*/
422 	}
423 #ifdef DIAGNOSTIC
424 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
425 		panic("mclbytes too small for t_template");
426 #endif
427 	m = tp->t_template;
428 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
429 		;
430 	else {
431 		if (m)
432 			m_freem(m);
433 		m = tp->t_template = NULL;
434 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
435 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
436 			MCLGET(m, M_DONTWAIT);
437 			if ((m->m_flags & M_EXT) == 0) {
438 				m_free(m);
439 				m = NULL;
440 			}
441 		}
442 		if (m == NULL)
443 			return NULL;
444 		MCLAIM(m, &tcp_mowner);
445 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
446 	}
447 
448 	bzero(mtod(m, caddr_t), m->m_len);
449 
450 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
451 
452 	switch (tp->t_family) {
453 	case AF_INET:
454 	    {
455 		struct ipovly *ipov;
456 		mtod(m, struct ip *)->ip_v = 4;
457 		ipov = mtod(m, struct ipovly *);
458 		ipov->ih_pr = IPPROTO_TCP;
459 		ipov->ih_len = htons(sizeof(struct tcphdr));
460 		if (inp) {
461 			ipov->ih_src = inp->inp_laddr;
462 			ipov->ih_dst = inp->inp_faddr;
463 		}
464 #ifdef INET6
465 		else if (in6p) {
466 			/* mapped addr case */
467 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
468 				sizeof(ipov->ih_src));
469 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
470 				sizeof(ipov->ih_dst));
471 		}
472 #endif
473 		/*
474 		 * Compute the pseudo-header portion of the checksum
475 		 * now.  We incrementally add in the TCP option and
476 		 * payload lengths later, and then compute the TCP
477 		 * checksum right before the packet is sent off onto
478 		 * the wire.
479 		 */
480 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
481 		    ipov->ih_dst.s_addr,
482 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
483 		break;
484 	    }
485 #ifdef INET6
486 	case AF_INET6:
487 	    {
488 		struct ip6_hdr *ip6;
489 		mtod(m, struct ip *)->ip_v = 6;
490 		ip6 = mtod(m, struct ip6_hdr *);
491 		ip6->ip6_nxt = IPPROTO_TCP;
492 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
493 		ip6->ip6_src = in6p->in6p_laddr;
494 		ip6->ip6_dst = in6p->in6p_faddr;
495 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
496 		if (ip6_auto_flowlabel) {
497 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
498 			ip6->ip6_flow |=
499 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
500 		}
501 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
502 		ip6->ip6_vfc |= IPV6_VERSION;
503 
504 		/*
505 		 * Compute the pseudo-header portion of the checksum
506 		 * now.  We incrementally add in the TCP option and
507 		 * payload lengths later, and then compute the TCP
508 		 * checksum right before the packet is sent off onto
509 		 * the wire.
510 		 */
511 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
512 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
513 		    htonl(IPPROTO_TCP));
514 		break;
515 	    }
516 #endif
517 	}
518 	if (inp) {
519 		n->th_sport = inp->inp_lport;
520 		n->th_dport = inp->inp_fport;
521 	}
522 #ifdef INET6
523 	else if (in6p) {
524 		n->th_sport = in6p->in6p_lport;
525 		n->th_dport = in6p->in6p_fport;
526 	}
527 #endif
528 	n->th_seq = 0;
529 	n->th_ack = 0;
530 	n->th_x2 = 0;
531 	n->th_off = 5;
532 	n->th_flags = 0;
533 	n->th_win = 0;
534 	n->th_urp = 0;
535 	return (m);
536 }
537 
538 /*
539  * Send a single message to the TCP at address specified by
540  * the given TCP/IP header.  If m == 0, then we make a copy
541  * of the tcpiphdr at ti and send directly to the addressed host.
542  * This is used to force keep alive messages out using the TCP
543  * template for a connection tp->t_template.  If flags are given
544  * then we send a message back to the TCP which originated the
545  * segment ti, and discard the mbuf containing it and any other
546  * attached mbufs.
547  *
548  * In any case the ack and sequence number of the transmitted
549  * segment are as specified by the parameters.
550  */
551 int
552 tcp_respond(tp, template, m, th0, ack, seq, flags)
553 	struct tcpcb *tp;
554 	struct mbuf *template;
555 	struct mbuf *m;
556 	struct tcphdr *th0;
557 	tcp_seq ack, seq;
558 	int flags;
559 {
560 	struct route *ro;
561 	int error, tlen, win = 0;
562 	int hlen;
563 	struct ip *ip;
564 #ifdef INET6
565 	struct ip6_hdr *ip6;
566 #endif
567 	int family;	/* family on packet, not inpcb/in6pcb! */
568 	struct tcphdr *th;
569 
570 	if (tp != NULL && (flags & TH_RST) == 0) {
571 #ifdef DIAGNOSTIC
572 		if (tp->t_inpcb && tp->t_in6pcb)
573 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
574 #endif
575 #ifdef INET
576 		if (tp->t_inpcb)
577 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
578 #endif
579 #ifdef INET6
580 		if (tp->t_in6pcb)
581 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
582 #endif
583 	}
584 
585 	th = NULL;	/* Quell uninitialized warning */
586 	ip = NULL;
587 #ifdef INET6
588 	ip6 = NULL;
589 #endif
590 	if (m == 0) {
591 		if (!template)
592 			return EINVAL;
593 
594 		/* get family information from template */
595 		switch (mtod(template, struct ip *)->ip_v) {
596 		case 4:
597 			family = AF_INET;
598 			hlen = sizeof(struct ip);
599 			break;
600 #ifdef INET6
601 		case 6:
602 			family = AF_INET6;
603 			hlen = sizeof(struct ip6_hdr);
604 			break;
605 #endif
606 		default:
607 			return EAFNOSUPPORT;
608 		}
609 
610 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
611 		if (m) {
612 			MCLAIM(m, &tcp_tx_mowner);
613 			MCLGET(m, M_DONTWAIT);
614 			if ((m->m_flags & M_EXT) == 0) {
615 				m_free(m);
616 				m = NULL;
617 			}
618 		}
619 		if (m == NULL)
620 			return (ENOBUFS);
621 
622 		if (tcp_compat_42)
623 			tlen = 1;
624 		else
625 			tlen = 0;
626 
627 		m->m_data += max_linkhdr;
628 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
629 			template->m_len);
630 		switch (family) {
631 		case AF_INET:
632 			ip = mtod(m, struct ip *);
633 			th = (struct tcphdr *)(ip + 1);
634 			break;
635 #ifdef INET6
636 		case AF_INET6:
637 			ip6 = mtod(m, struct ip6_hdr *);
638 			th = (struct tcphdr *)(ip6 + 1);
639 			break;
640 #endif
641 #if 0
642 		default:
643 			/* noone will visit here */
644 			m_freem(m);
645 			return EAFNOSUPPORT;
646 #endif
647 		}
648 		flags = TH_ACK;
649 	} else {
650 
651 		if ((m->m_flags & M_PKTHDR) == 0) {
652 #if 0
653 			printf("non PKTHDR to tcp_respond\n");
654 #endif
655 			m_freem(m);
656 			return EINVAL;
657 		}
658 #ifdef DIAGNOSTIC
659 		if (!th0)
660 			panic("th0 == NULL in tcp_respond");
661 #endif
662 
663 		/* get family information from m */
664 		switch (mtod(m, struct ip *)->ip_v) {
665 		case 4:
666 			family = AF_INET;
667 			hlen = sizeof(struct ip);
668 			ip = mtod(m, struct ip *);
669 			break;
670 #ifdef INET6
671 		case 6:
672 			family = AF_INET6;
673 			hlen = sizeof(struct ip6_hdr);
674 			ip6 = mtod(m, struct ip6_hdr *);
675 			break;
676 #endif
677 		default:
678 			m_freem(m);
679 			return EAFNOSUPPORT;
680 		}
681 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
682 			tlen = sizeof(*th0);
683 		else
684 			tlen = th0->th_off << 2;
685 
686 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
687 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
688 			m->m_len = hlen + tlen;
689 			m_freem(m->m_next);
690 			m->m_next = NULL;
691 		} else {
692 			struct mbuf *n;
693 
694 #ifdef DIAGNOSTIC
695 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
696 				m_freem(m);
697 				return EMSGSIZE;
698 			}
699 #endif
700 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
701 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
702 				MCLGET(n, M_DONTWAIT);
703 				if ((n->m_flags & M_EXT) == 0) {
704 					m_freem(n);
705 					n = NULL;
706 				}
707 			}
708 			if (!n) {
709 				m_freem(m);
710 				return ENOBUFS;
711 			}
712 
713 			MCLAIM(n, &tcp_tx_mowner);
714 			n->m_data += max_linkhdr;
715 			n->m_len = hlen + tlen;
716 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
717 			m_copyback(n, hlen, tlen, (caddr_t)th0);
718 
719 			m_freem(m);
720 			m = n;
721 			n = NULL;
722 		}
723 
724 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
725 		switch (family) {
726 		case AF_INET:
727 			ip = mtod(m, struct ip *);
728 			th = (struct tcphdr *)(ip + 1);
729 			ip->ip_p = IPPROTO_TCP;
730 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
731 			ip->ip_p = IPPROTO_TCP;
732 			break;
733 #ifdef INET6
734 		case AF_INET6:
735 			ip6 = mtod(m, struct ip6_hdr *);
736 			th = (struct tcphdr *)(ip6 + 1);
737 			ip6->ip6_nxt = IPPROTO_TCP;
738 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
739 			ip6->ip6_nxt = IPPROTO_TCP;
740 			break;
741 #endif
742 #if 0
743 		default:
744 			/* noone will visit here */
745 			m_freem(m);
746 			return EAFNOSUPPORT;
747 #endif
748 		}
749 		xchg(th->th_dport, th->th_sport, u_int16_t);
750 #undef xchg
751 		tlen = 0;	/*be friendly with the following code*/
752 	}
753 	th->th_seq = htonl(seq);
754 	th->th_ack = htonl(ack);
755 	th->th_x2 = 0;
756 	if ((flags & TH_SYN) == 0) {
757 		if (tp)
758 			win >>= tp->rcv_scale;
759 		if (win > TCP_MAXWIN)
760 			win = TCP_MAXWIN;
761 		th->th_win = htons((u_int16_t)win);
762 		th->th_off = sizeof (struct tcphdr) >> 2;
763 		tlen += sizeof(*th);
764 	} else
765 		tlen += th->th_off << 2;
766 	m->m_len = hlen + tlen;
767 	m->m_pkthdr.len = hlen + tlen;
768 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
769 	th->th_flags = flags;
770 	th->th_urp = 0;
771 
772 	switch (family) {
773 #ifdef INET
774 	case AF_INET:
775 	    {
776 		struct ipovly *ipov = (struct ipovly *)ip;
777 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
778 		ipov->ih_len = htons((u_int16_t)tlen);
779 
780 		th->th_sum = 0;
781 		th->th_sum = in_cksum(m, hlen + tlen);
782 		ip->ip_len = htons(hlen + tlen);
783 		ip->ip_ttl = ip_defttl;
784 		break;
785 	    }
786 #endif
787 #ifdef INET6
788 	case AF_INET6:
789 	    {
790 		th->th_sum = 0;
791 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
792 				tlen);
793 		ip6->ip6_plen = ntohs(tlen);
794 		if (tp && tp->t_in6pcb) {
795 			struct ifnet *oifp;
796 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
797 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
798 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
799 		} else
800 			ip6->ip6_hlim = ip6_defhlim;
801 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
802 		if (ip6_auto_flowlabel) {
803 			ip6->ip6_flow |=
804 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
805 		}
806 		break;
807 	    }
808 #endif
809 	}
810 
811 #ifdef IPSEC
812 	(void)ipsec_setsocket(m, NULL);
813 #endif /*IPSEC*/
814 
815 	if (tp != NULL && tp->t_inpcb != NULL) {
816 		ro = &tp->t_inpcb->inp_route;
817 #ifdef IPSEC
818 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
819 			m_freem(m);
820 			return ENOBUFS;
821 		}
822 #endif
823 #ifdef DIAGNOSTIC
824 		if (family != AF_INET)
825 			panic("tcp_respond: address family mismatch");
826 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
827 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
828 			    ntohl(ip->ip_dst.s_addr),
829 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
830 		}
831 #endif
832 	}
833 #ifdef INET6
834 	else if (tp != NULL && tp->t_in6pcb != NULL) {
835 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
836 #ifdef IPSEC
837 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
838 			m_freem(m);
839 			return ENOBUFS;
840 		}
841 #endif
842 #ifdef DIAGNOSTIC
843 		if (family == AF_INET) {
844 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
845 				panic("tcp_respond: not mapped addr");
846 			if (bcmp(&ip->ip_dst,
847 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
848 			    sizeof(ip->ip_dst)) != 0) {
849 				panic("tcp_respond: ip_dst != in6p_faddr");
850 			}
851 		} else if (family == AF_INET6) {
852 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
853 			    &tp->t_in6pcb->in6p_faddr))
854 				panic("tcp_respond: ip6_dst != in6p_faddr");
855 		} else
856 			panic("tcp_respond: address family mismatch");
857 #endif
858 	}
859 #endif
860 	else
861 		ro = NULL;
862 
863 	switch (family) {
864 #ifdef INET
865 	case AF_INET:
866 		error = ip_output(m, NULL, ro,
867 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
868 		    NULL);
869 		break;
870 #endif
871 #ifdef INET6
872 	case AF_INET6:
873 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
874 		    NULL);
875 		break;
876 #endif
877 	default:
878 		error = EAFNOSUPPORT;
879 		break;
880 	}
881 
882 	return (error);
883 }
884 
885 /*
886  * Create a new TCP control block, making an
887  * empty reassembly queue and hooking it to the argument
888  * protocol control block.
889  */
890 struct tcpcb *
891 tcp_newtcpcb(family, aux)
892 	int family;	/* selects inpcb, or in6pcb */
893 	void *aux;
894 {
895 	struct tcpcb *tp;
896 	int i;
897 
898 	switch (family) {
899 	case PF_INET:
900 		break;
901 #ifdef INET6
902 	case PF_INET6:
903 		break;
904 #endif
905 	default:
906 		return NULL;
907 	}
908 
909 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
910 	if (tp == NULL)
911 		return (NULL);
912 	bzero((caddr_t)tp, sizeof(struct tcpcb));
913 	TAILQ_INIT(&tp->segq);
914 	TAILQ_INIT(&tp->timeq);
915 	tp->t_family = family;		/* may be overridden later on */
916 	tp->t_peermss = tcp_mssdflt;
917 	tp->t_ourmss = tcp_mssdflt;
918 	tp->t_segsz = tcp_mssdflt;
919 	LIST_INIT(&tp->t_sc);
920 
921 	tp->t_lastm = NULL;
922 	tp->t_lastoff = 0;
923 
924 	callout_init(&tp->t_delack_ch);
925 	for (i = 0; i < TCPT_NTIMERS; i++)
926 		TCP_TIMER_INIT(tp, i);
927 
928 	tp->t_flags = 0;
929 	if (tcp_do_rfc1323 && tcp_do_win_scale)
930 		tp->t_flags |= TF_REQ_SCALE;
931 	if (tcp_do_rfc1323 && tcp_do_timestamps)
932 		tp->t_flags |= TF_REQ_TSTMP;
933 	if (tcp_do_sack == 2)
934 		tp->t_flags |= TF_WILL_SACK;
935 	else if (tcp_do_sack == 1)
936 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
937 	tp->t_flags |= TF_CANT_TXSACK;
938 	switch (family) {
939 	case PF_INET:
940 		tp->t_inpcb = (struct inpcb *)aux;
941 		tp->t_mtudisc = ip_mtudisc;
942 		break;
943 #ifdef INET6
944 	case PF_INET6:
945 		tp->t_in6pcb = (struct in6pcb *)aux;
946 		/* for IPv6, always try to run path MTU discovery */
947 		tp->t_mtudisc = 1;
948 		break;
949 #endif
950 	}
951 	/*
952 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
953 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
954 	 * reasonable initial retransmit time.
955 	 */
956 	tp->t_srtt = TCPTV_SRTTBASE;
957 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
958 	tp->t_rttmin = TCPTV_MIN;
959 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
960 	    TCPTV_MIN, TCPTV_REXMTMAX);
961 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
962 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
963 	if (family == AF_INET) {
964 		struct inpcb *inp = (struct inpcb *)aux;
965 		inp->inp_ip.ip_ttl = ip_defttl;
966 		inp->inp_ppcb = (caddr_t)tp;
967 	}
968 #ifdef INET6
969 	else if (family == AF_INET6) {
970 		struct in6pcb *in6p = (struct in6pcb *)aux;
971 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
972 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
973 					       : NULL);
974 		in6p->in6p_ppcb = (caddr_t)tp;
975 	}
976 #endif
977 
978 	/*
979 	 * Initialize our timebase.  When we send timestamps, we take
980 	 * the delta from tcp_now -- this means each connection always
981 	 * gets a timebase of 0, which makes it, among other things,
982 	 * more difficult to determine how long a system has been up,
983 	 * and thus how many TCP sequence increments have occurred.
984 	 */
985 	tp->ts_timebase = tcp_now;
986 
987 	return (tp);
988 }
989 
990 /*
991  * Drop a TCP connection, reporting
992  * the specified error.  If connection is synchronized,
993  * then send a RST to peer.
994  */
995 struct tcpcb *
996 tcp_drop(tp, errno)
997 	struct tcpcb *tp;
998 	int errno;
999 {
1000 	struct socket *so = NULL;
1001 
1002 #ifdef DIAGNOSTIC
1003 	if (tp->t_inpcb && tp->t_in6pcb)
1004 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1005 #endif
1006 #ifdef INET
1007 	if (tp->t_inpcb)
1008 		so = tp->t_inpcb->inp_socket;
1009 #endif
1010 #ifdef INET6
1011 	if (tp->t_in6pcb)
1012 		so = tp->t_in6pcb->in6p_socket;
1013 #endif
1014 	if (!so)
1015 		return NULL;
1016 
1017 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1018 		tp->t_state = TCPS_CLOSED;
1019 		(void) tcp_output(tp);
1020 		tcpstat.tcps_drops++;
1021 	} else
1022 		tcpstat.tcps_conndrops++;
1023 	if (errno == ETIMEDOUT && tp->t_softerror)
1024 		errno = tp->t_softerror;
1025 	so->so_error = errno;
1026 	return (tcp_close(tp));
1027 }
1028 
1029 /*
1030  * Return whether this tcpcb is marked as dead, indicating
1031  * to the calling timer function that no further action should
1032  * be taken, as we are about to release this tcpcb.  The release
1033  * of the storage will be done if this is the last timer running.
1034  *
1035  * This is typically called from the callout handler function before
1036  * callout_ack() is done, therefore we need to test the number of
1037  * running timer functions against 1 below, not 0.
1038  */
1039 int
1040 tcp_isdead(tp)
1041 	struct tcpcb *tp;
1042 {
1043 	int dead = (tp->t_flags & TF_DEAD);
1044 
1045 	if (__predict_false(dead)) {
1046 		if (tcp_timers_invoking(tp) > 1)
1047 				/* not quite there yet -- count separately? */
1048 			return dead;
1049 		tcpstat.tcps_delayed_free++;
1050 		pool_put(&tcpcb_pool, tp);
1051 	}
1052 	return dead;
1053 }
1054 
1055 /*
1056  * Close a TCP control block:
1057  *	discard all space held by the tcp
1058  *	discard internet protocol block
1059  *	wake up any sleepers
1060  */
1061 struct tcpcb *
1062 tcp_close(tp)
1063 	struct tcpcb *tp;
1064 {
1065 	struct inpcb *inp;
1066 #ifdef INET6
1067 	struct in6pcb *in6p;
1068 #endif
1069 	struct socket *so;
1070 #ifdef RTV_RTT
1071 	struct rtentry *rt;
1072 #endif
1073 	struct route *ro;
1074 
1075 	inp = tp->t_inpcb;
1076 #ifdef INET6
1077 	in6p = tp->t_in6pcb;
1078 #endif
1079 	so = NULL;
1080 	ro = NULL;
1081 	if (inp) {
1082 		so = inp->inp_socket;
1083 		ro = &inp->inp_route;
1084 	}
1085 #ifdef INET6
1086 	else if (in6p) {
1087 		so = in6p->in6p_socket;
1088 		ro = (struct route *)&in6p->in6p_route;
1089 	}
1090 #endif
1091 
1092 #ifdef RTV_RTT
1093 	/*
1094 	 * If we sent enough data to get some meaningful characteristics,
1095 	 * save them in the routing entry.  'Enough' is arbitrarily
1096 	 * defined as the sendpipesize (default 4K) * 16.  This would
1097 	 * give us 16 rtt samples assuming we only get one sample per
1098 	 * window (the usual case on a long haul net).  16 samples is
1099 	 * enough for the srtt filter to converge to within 5% of the correct
1100 	 * value; fewer samples and we could save a very bogus rtt.
1101 	 *
1102 	 * Don't update the default route's characteristics and don't
1103 	 * update anything that the user "locked".
1104 	 */
1105 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1106 	    ro && (rt = ro->ro_rt) &&
1107 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1108 		u_long i = 0;
1109 
1110 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1111 			i = tp->t_srtt *
1112 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1113 			if (rt->rt_rmx.rmx_rtt && i)
1114 				/*
1115 				 * filter this update to half the old & half
1116 				 * the new values, converting scale.
1117 				 * See route.h and tcp_var.h for a
1118 				 * description of the scaling constants.
1119 				 */
1120 				rt->rt_rmx.rmx_rtt =
1121 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1122 			else
1123 				rt->rt_rmx.rmx_rtt = i;
1124 		}
1125 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1126 			i = tp->t_rttvar *
1127 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1128 			if (rt->rt_rmx.rmx_rttvar && i)
1129 				rt->rt_rmx.rmx_rttvar =
1130 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1131 			else
1132 				rt->rt_rmx.rmx_rttvar = i;
1133 		}
1134 		/*
1135 		 * update the pipelimit (ssthresh) if it has been updated
1136 		 * already or if a pipesize was specified & the threshhold
1137 		 * got below half the pipesize.  I.e., wait for bad news
1138 		 * before we start updating, then update on both good
1139 		 * and bad news.
1140 		 */
1141 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1142 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1143 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1144 			/*
1145 			 * convert the limit from user data bytes to
1146 			 * packets then to packet data bytes.
1147 			 */
1148 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1149 			if (i < 2)
1150 				i = 2;
1151 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1152 			if (rt->rt_rmx.rmx_ssthresh)
1153 				rt->rt_rmx.rmx_ssthresh =
1154 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1155 			else
1156 				rt->rt_rmx.rmx_ssthresh = i;
1157 		}
1158 	}
1159 #endif /* RTV_RTT */
1160 	/* free the reassembly queue, if any */
1161 	TCP_REASS_LOCK(tp);
1162 	(void) tcp_freeq(tp);
1163 	TCP_REASS_UNLOCK(tp);
1164 
1165 	tcp_canceltimers(tp);
1166 	TCP_CLEAR_DELACK(tp);
1167 	syn_cache_cleanup(tp);
1168 
1169 	if (tp->t_template) {
1170 		m_free(tp->t_template);
1171 		tp->t_template = NULL;
1172 	}
1173 	if (tcp_timers_invoking(tp))
1174 		tp->t_flags |= TF_DEAD;
1175 	else
1176 		pool_put(&tcpcb_pool, tp);
1177 
1178 	if (inp) {
1179 		inp->inp_ppcb = 0;
1180 		soisdisconnected(so);
1181 		in_pcbdetach(inp);
1182 	}
1183 #ifdef INET6
1184 	else if (in6p) {
1185 		in6p->in6p_ppcb = 0;
1186 		soisdisconnected(so);
1187 		in6_pcbdetach(in6p);
1188 	}
1189 #endif
1190 	tcpstat.tcps_closed++;
1191 	return ((struct tcpcb *)0);
1192 }
1193 
1194 int
1195 tcp_freeq(tp)
1196 	struct tcpcb *tp;
1197 {
1198 	struct ipqent *qe;
1199 	int rv = 0;
1200 #ifdef TCPREASS_DEBUG
1201 	int i = 0;
1202 #endif
1203 
1204 	TCP_REASS_LOCK_CHECK(tp);
1205 
1206 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1207 #ifdef TCPREASS_DEBUG
1208 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1209 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1210 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1211 #endif
1212 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1213 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1214 		m_freem(qe->ipqe_m);
1215 		pool_put(&ipqent_pool, qe);
1216 		rv = 1;
1217 	}
1218 	return (rv);
1219 }
1220 
1221 /*
1222  * Protocol drain routine.  Called when memory is in short supply.
1223  */
1224 void
1225 tcp_drain()
1226 {
1227 	struct inpcb *inp;
1228 	struct tcpcb *tp;
1229 
1230 	/*
1231 	 * Free the sequence queue of all TCP connections.
1232 	 */
1233 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1234 	if (inp)						/* XXX */
1235 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1236 		if ((tp = intotcpcb(inp)) != NULL) {
1237 			/*
1238 			 * We may be called from a device's interrupt
1239 			 * context.  If the tcpcb is already busy,
1240 			 * just bail out now.
1241 			 */
1242 			if (tcp_reass_lock_try(tp) == 0)
1243 				continue;
1244 			if (tcp_freeq(tp))
1245 				tcpstat.tcps_connsdrained++;
1246 			TCP_REASS_UNLOCK(tp);
1247 		}
1248 	}
1249 }
1250 
1251 #ifdef INET6
1252 void
1253 tcp6_drain()
1254 {
1255 	struct in6pcb *in6p;
1256 	struct tcpcb *tp;
1257 	struct in6pcb *head = &tcb6;
1258 
1259 	/*
1260 	 * Free the sequence queue of all TCP connections.
1261 	 */
1262 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1263 		if ((tp = in6totcpcb(in6p)) != NULL) {
1264 			/*
1265 			 * We may be called from a device's interrupt
1266 			 * context.  If the tcpcb is already busy,
1267 			 * just bail out now.
1268 			 */
1269 			if (tcp_reass_lock_try(tp) == 0)
1270 				continue;
1271 			if (tcp_freeq(tp))
1272 				tcpstat.tcps_connsdrained++;
1273 			TCP_REASS_UNLOCK(tp);
1274 		}
1275 	}
1276 }
1277 #endif
1278 
1279 /*
1280  * Notify a tcp user of an asynchronous error;
1281  * store error as soft error, but wake up user
1282  * (for now, won't do anything until can select for soft error).
1283  */
1284 void
1285 tcp_notify(inp, error)
1286 	struct inpcb *inp;
1287 	int error;
1288 {
1289 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1290 	struct socket *so = inp->inp_socket;
1291 
1292 	/*
1293 	 * Ignore some errors if we are hooked up.
1294 	 * If connection hasn't completed, has retransmitted several times,
1295 	 * and receives a second error, give up now.  This is better
1296 	 * than waiting a long time to establish a connection that
1297 	 * can never complete.
1298 	 */
1299 	if (tp->t_state == TCPS_ESTABLISHED &&
1300 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1301 	      error == EHOSTDOWN)) {
1302 		return;
1303 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1304 	    tp->t_rxtshift > 3 && tp->t_softerror)
1305 		so->so_error = error;
1306 	else
1307 		tp->t_softerror = error;
1308 	wakeup((caddr_t) &so->so_timeo);
1309 	sorwakeup(so);
1310 	sowwakeup(so);
1311 }
1312 
1313 #ifdef INET6
1314 void
1315 tcp6_notify(in6p, error)
1316 	struct in6pcb *in6p;
1317 	int error;
1318 {
1319 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1320 	struct socket *so = in6p->in6p_socket;
1321 
1322 	/*
1323 	 * Ignore some errors if we are hooked up.
1324 	 * If connection hasn't completed, has retransmitted several times,
1325 	 * and receives a second error, give up now.  This is better
1326 	 * than waiting a long time to establish a connection that
1327 	 * can never complete.
1328 	 */
1329 	if (tp->t_state == TCPS_ESTABLISHED &&
1330 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1331 	      error == EHOSTDOWN)) {
1332 		return;
1333 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1334 	    tp->t_rxtshift > 3 && tp->t_softerror)
1335 		so->so_error = error;
1336 	else
1337 		tp->t_softerror = error;
1338 	wakeup((caddr_t) &so->so_timeo);
1339 	sorwakeup(so);
1340 	sowwakeup(so);
1341 }
1342 #endif
1343 
1344 #ifdef INET6
1345 void
1346 tcp6_ctlinput(cmd, sa, d)
1347 	int cmd;
1348 	struct sockaddr *sa;
1349 	void *d;
1350 {
1351 	struct tcphdr th;
1352 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1353 	int nmatch;
1354 	struct ip6_hdr *ip6;
1355 	const struct sockaddr_in6 *sa6_src = NULL;
1356 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1357 	struct mbuf *m;
1358 	int off;
1359 
1360 	if (sa->sa_family != AF_INET6 ||
1361 	    sa->sa_len != sizeof(struct sockaddr_in6))
1362 		return;
1363 	if ((unsigned)cmd >= PRC_NCMDS)
1364 		return;
1365 	else if (cmd == PRC_QUENCH) {
1366 		/* XXX there's no PRC_QUENCH in IPv6 */
1367 		notify = tcp6_quench;
1368 	} else if (PRC_IS_REDIRECT(cmd))
1369 		notify = in6_rtchange, d = NULL;
1370 	else if (cmd == PRC_MSGSIZE)
1371 		; /* special code is present, see below */
1372 	else if (cmd == PRC_HOSTDEAD)
1373 		d = NULL;
1374 	else if (inet6ctlerrmap[cmd] == 0)
1375 		return;
1376 
1377 	/* if the parameter is from icmp6, decode it. */
1378 	if (d != NULL) {
1379 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1380 		m = ip6cp->ip6c_m;
1381 		ip6 = ip6cp->ip6c_ip6;
1382 		off = ip6cp->ip6c_off;
1383 		sa6_src = ip6cp->ip6c_src;
1384 	} else {
1385 		m = NULL;
1386 		ip6 = NULL;
1387 		sa6_src = &sa6_any;
1388 	}
1389 
1390 	if (ip6) {
1391 		/*
1392 		 * XXX: We assume that when ip6 is non NULL,
1393 		 * M and OFF are valid.
1394 		 */
1395 
1396 		/* check if we can safely examine src and dst ports */
1397 		if (m->m_pkthdr.len < off + sizeof(th)) {
1398 			if (cmd == PRC_MSGSIZE)
1399 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1400 			return;
1401 		}
1402 
1403 		bzero(&th, sizeof(th));
1404 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1405 
1406 		if (cmd == PRC_MSGSIZE) {
1407 			int valid = 0;
1408 
1409 			/*
1410 			 * Check to see if we have a valid TCP connection
1411 			 * corresponding to the address in the ICMPv6 message
1412 			 * payload.
1413 			 */
1414 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1415 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1416 			    th.th_sport, 0))
1417 				valid++;
1418 
1419 			/*
1420 			 * Depending on the value of "valid" and routing table
1421 			 * size (mtudisc_{hi,lo}wat), we will:
1422 			 * - recalcurate the new MTU and create the
1423 			 *   corresponding routing entry, or
1424 			 * - ignore the MTU change notification.
1425 			 */
1426 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1427 
1428 			/*
1429 			 * no need to call in6_pcbnotify, it should have been
1430 			 * called via callback if necessary
1431 			 */
1432 			return;
1433 		}
1434 
1435 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1436 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1437 		if (nmatch == 0 && syn_cache_count &&
1438 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1439 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1440 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1441 			syn_cache_unreach((struct sockaddr *)sa6_src,
1442 					  sa, &th);
1443 	} else {
1444 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1445 		    0, cmd, NULL, notify);
1446 	}
1447 }
1448 #endif
1449 
1450 #ifdef INET
1451 /* assumes that ip header and tcp header are contiguous on mbuf */
1452 void *
1453 tcp_ctlinput(cmd, sa, v)
1454 	int cmd;
1455 	struct sockaddr *sa;
1456 	void *v;
1457 {
1458 	struct ip *ip = v;
1459 	struct tcphdr *th;
1460 	struct icmp *icp;
1461 	extern const int inetctlerrmap[];
1462 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1463 	int errno;
1464 	int nmatch;
1465 #ifdef INET6
1466 	struct in6_addr src6, dst6;
1467 #endif
1468 
1469 	if (sa->sa_family != AF_INET ||
1470 	    sa->sa_len != sizeof(struct sockaddr_in))
1471 		return NULL;
1472 	if ((unsigned)cmd >= PRC_NCMDS)
1473 		return NULL;
1474 	errno = inetctlerrmap[cmd];
1475 	if (cmd == PRC_QUENCH)
1476 		notify = tcp_quench;
1477 	else if (PRC_IS_REDIRECT(cmd))
1478 		notify = in_rtchange, ip = 0;
1479 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1480 		/*
1481 		 * Check to see if we have a valid TCP connection
1482 		 * corresponding to the address in the ICMP message
1483 		 * payload.
1484 		 *
1485 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1486 		 */
1487 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1488 #ifdef INET6
1489 		memset(&src6, 0, sizeof(src6));
1490 		memset(&dst6, 0, sizeof(dst6));
1491 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1492 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1493 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1494 #endif
1495 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1496 		    ip->ip_src, th->th_sport) != NULL)
1497 			;
1498 #ifdef INET6
1499 		else if (in6_pcblookup_connect(&tcb6, &dst6,
1500 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1501 			;
1502 #endif
1503 		else
1504 			return NULL;
1505 
1506 		/*
1507 		 * Now that we've validated that we are actually communicating
1508 		 * with the host indicated in the ICMP message, locate the
1509 		 * ICMP header, recalculate the new MTU, and create the
1510 		 * corresponding routing entry.
1511 		 */
1512 		icp = (struct icmp *)((caddr_t)ip -
1513 		    offsetof(struct icmp, icmp_ip));
1514 		icmp_mtudisc(icp, ip->ip_dst);
1515 
1516 		return NULL;
1517 	} else if (cmd == PRC_HOSTDEAD)
1518 		ip = 0;
1519 	else if (errno == 0)
1520 		return NULL;
1521 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1522 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1523 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1524 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1525 		if (nmatch == 0 && syn_cache_count &&
1526 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1527 		    inetctlerrmap[cmd] == ENETUNREACH ||
1528 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1529 			struct sockaddr_in sin;
1530 			bzero(&sin, sizeof(sin));
1531 			sin.sin_len = sizeof(sin);
1532 			sin.sin_family = AF_INET;
1533 			sin.sin_port = th->th_sport;
1534 			sin.sin_addr = ip->ip_src;
1535 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1536 		}
1537 
1538 		/* XXX mapped address case */
1539 	} else
1540 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1541 		    notify);
1542 	return NULL;
1543 }
1544 
1545 /*
1546  * When a source quence is received, we are being notifed of congestion.
1547  * Close the congestion window down to the Loss Window (one segment).
1548  * We will gradually open it again as we proceed.
1549  */
1550 void
1551 tcp_quench(inp, errno)
1552 	struct inpcb *inp;
1553 	int errno;
1554 {
1555 	struct tcpcb *tp = intotcpcb(inp);
1556 
1557 	if (tp)
1558 		tp->snd_cwnd = tp->t_segsz;
1559 }
1560 #endif
1561 
1562 #ifdef INET6
1563 void
1564 tcp6_quench(in6p, errno)
1565 	struct in6pcb *in6p;
1566 	int errno;
1567 {
1568 	struct tcpcb *tp = in6totcpcb(in6p);
1569 
1570 	if (tp)
1571 		tp->snd_cwnd = tp->t_segsz;
1572 }
1573 #endif
1574 
1575 #ifdef INET
1576 /*
1577  * Path MTU Discovery handlers.
1578  */
1579 void
1580 tcp_mtudisc_callback(faddr)
1581 	struct in_addr faddr;
1582 {
1583 #ifdef INET6
1584 	struct in6_addr in6;
1585 #endif
1586 
1587 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1588 #ifdef INET6
1589 	memset(&in6, 0, sizeof(in6));
1590 	in6.s6_addr16[5] = 0xffff;
1591 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1592 	tcp6_mtudisc_callback(&in6);
1593 #endif
1594 }
1595 
1596 /*
1597  * On receipt of path MTU corrections, flush old route and replace it
1598  * with the new one.  Retransmit all unacknowledged packets, to ensure
1599  * that all packets will be received.
1600  */
1601 void
1602 tcp_mtudisc(inp, errno)
1603 	struct inpcb *inp;
1604 	int errno;
1605 {
1606 	struct tcpcb *tp = intotcpcb(inp);
1607 	struct rtentry *rt = in_pcbrtentry(inp);
1608 
1609 	if (tp != 0) {
1610 		if (rt != 0) {
1611 			/*
1612 			 * If this was not a host route, remove and realloc.
1613 			 */
1614 			if ((rt->rt_flags & RTF_HOST) == 0) {
1615 				in_rtchange(inp, errno);
1616 				if ((rt = in_pcbrtentry(inp)) == 0)
1617 					return;
1618 			}
1619 
1620 			/*
1621 			 * Slow start out of the error condition.  We
1622 			 * use the MTU because we know it's smaller
1623 			 * than the previously transmitted segment.
1624 			 *
1625 			 * Note: This is more conservative than the
1626 			 * suggestion in draft-floyd-incr-init-win-03.
1627 			 */
1628 			if (rt->rt_rmx.rmx_mtu != 0)
1629 				tp->snd_cwnd =
1630 				    TCP_INITIAL_WINDOW(tcp_init_win,
1631 				    rt->rt_rmx.rmx_mtu);
1632 		}
1633 
1634 		/*
1635 		 * Resend unacknowledged packets.
1636 		 */
1637 		tp->snd_nxt = tp->snd_una;
1638 		tcp_output(tp);
1639 	}
1640 }
1641 #endif
1642 
1643 #ifdef INET6
1644 /*
1645  * Path MTU Discovery handlers.
1646  */
1647 void
1648 tcp6_mtudisc_callback(faddr)
1649 	struct in6_addr *faddr;
1650 {
1651 	struct sockaddr_in6 sin6;
1652 
1653 	bzero(&sin6, sizeof(sin6));
1654 	sin6.sin6_family = AF_INET6;
1655 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1656 	sin6.sin6_addr = *faddr;
1657 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1658 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1659 }
1660 
1661 void
1662 tcp6_mtudisc(in6p, errno)
1663 	struct in6pcb *in6p;
1664 	int errno;
1665 {
1666 	struct tcpcb *tp = in6totcpcb(in6p);
1667 	struct rtentry *rt = in6_pcbrtentry(in6p);
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 				in6_rtchange(in6p, errno);
1676 				if ((rt = in6_pcbrtentry(in6p)) == 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 /* INET6 */
1702 
1703 /*
1704  * Compute the MSS to advertise to the peer.  Called only during
1705  * the 3-way handshake.  If we are the server (peer initiated
1706  * connection), we are called with a pointer to the interface
1707  * on which the SYN packet arrived.  If we are the client (we
1708  * initiated connection), we are called with a pointer to the
1709  * interface out which this connection should go.
1710  *
1711  * NOTE: Do not subtract IP option/extension header size nor IPsec
1712  * header size from MSS advertisement.  MSS option must hold the maximum
1713  * segment size we can accept, so it must always be:
1714  *	 max(if mtu) - ip header - tcp header
1715  */
1716 u_long
1717 tcp_mss_to_advertise(ifp, af)
1718 	const struct ifnet *ifp;
1719 	int af;
1720 {
1721 	extern u_long in_maxmtu;
1722 	u_long mss = 0;
1723 	u_long hdrsiz;
1724 
1725 	/*
1726 	 * In order to avoid defeating path MTU discovery on the peer,
1727 	 * we advertise the max MTU of all attached networks as our MSS,
1728 	 * per RFC 1191, section 3.1.
1729 	 *
1730 	 * We provide the option to advertise just the MTU of
1731 	 * the interface on which we hope this connection will
1732 	 * be receiving.  If we are responding to a SYN, we
1733 	 * will have a pretty good idea about this, but when
1734 	 * initiating a connection there is a bit more doubt.
1735 	 *
1736 	 * We also need to ensure that loopback has a large enough
1737 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1738 	 */
1739 
1740 	if (ifp != NULL)
1741 		switch (af) {
1742 		case AF_INET:
1743 			mss = ifp->if_mtu;
1744 			break;
1745 #ifdef INET6
1746 		case AF_INET6:
1747 			mss = IN6_LINKMTU(ifp);
1748 			break;
1749 #endif
1750 		}
1751 
1752 	if (tcp_mss_ifmtu == 0)
1753 		switch (af) {
1754 		case AF_INET:
1755 			mss = max(in_maxmtu, mss);
1756 			break;
1757 #ifdef INET6
1758 		case AF_INET6:
1759 			mss = max(in6_maxmtu, mss);
1760 			break;
1761 #endif
1762 		}
1763 
1764 	switch (af) {
1765 	case AF_INET:
1766 		hdrsiz = sizeof(struct ip);
1767 		break;
1768 #ifdef INET6
1769 	case AF_INET6:
1770 		hdrsiz = sizeof(struct ip6_hdr);
1771 		break;
1772 #endif
1773 	default:
1774 		hdrsiz = 0;
1775 		break;
1776 	}
1777 	hdrsiz += sizeof(struct tcphdr);
1778 	if (mss > hdrsiz)
1779 		mss -= hdrsiz;
1780 
1781 	mss = max(tcp_mssdflt, mss);
1782 	return (mss);
1783 }
1784 
1785 /*
1786  * Set connection variables based on the peer's advertised MSS.
1787  * We are passed the TCPCB for the actual connection.  If we
1788  * are the server, we are called by the compressed state engine
1789  * when the 3-way handshake is complete.  If we are the client,
1790  * we are called when we receive the SYN,ACK from the server.
1791  *
1792  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1793  * before this routine is called!
1794  */
1795 void
1796 tcp_mss_from_peer(tp, offer)
1797 	struct tcpcb *tp;
1798 	int offer;
1799 {
1800 	struct socket *so;
1801 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1802 	struct rtentry *rt;
1803 #endif
1804 	u_long bufsize;
1805 	int mss;
1806 
1807 #ifdef DIAGNOSTIC
1808 	if (tp->t_inpcb && tp->t_in6pcb)
1809 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1810 #endif
1811 	so = NULL;
1812 	rt = NULL;
1813 #ifdef INET
1814 	if (tp->t_inpcb) {
1815 		so = tp->t_inpcb->inp_socket;
1816 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1817 		rt = in_pcbrtentry(tp->t_inpcb);
1818 #endif
1819 	}
1820 #endif
1821 #ifdef INET6
1822 	if (tp->t_in6pcb) {
1823 		so = tp->t_in6pcb->in6p_socket;
1824 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1825 		rt = in6_pcbrtentry(tp->t_in6pcb);
1826 #endif
1827 	}
1828 #endif
1829 
1830 	/*
1831 	 * As per RFC1122, use the default MSS value, unless they
1832 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1833 	 */
1834 	mss = tcp_mssdflt;
1835 	if (offer)
1836 		mss = offer;
1837 	mss = max(mss, 32);		/* sanity */
1838 	tp->t_peermss = mss;
1839 	mss -= tcp_optlen(tp);
1840 #ifdef INET
1841 	if (tp->t_inpcb)
1842 		mss -= ip_optlen(tp->t_inpcb);
1843 #endif
1844 #ifdef INET6
1845 	if (tp->t_in6pcb)
1846 		mss -= ip6_optlen(tp->t_in6pcb);
1847 #endif
1848 
1849 	/*
1850 	 * If there's a pipesize, change the socket buffer to that size.
1851 	 * Make the socket buffer an integral number of MSS units.  If
1852 	 * the MSS is larger than the socket buffer, artificially decrease
1853 	 * the MSS.
1854 	 */
1855 #ifdef RTV_SPIPE
1856 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1857 		bufsize = rt->rt_rmx.rmx_sendpipe;
1858 	else
1859 #endif
1860 		bufsize = so->so_snd.sb_hiwat;
1861 	if (bufsize < mss)
1862 		mss = bufsize;
1863 	else {
1864 		bufsize = roundup(bufsize, mss);
1865 		if (bufsize > sb_max)
1866 			bufsize = sb_max;
1867 		(void) sbreserve(&so->so_snd, bufsize);
1868 	}
1869 	tp->t_segsz = mss;
1870 
1871 #ifdef RTV_SSTHRESH
1872 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1873 		/*
1874 		 * There's some sort of gateway or interface buffer
1875 		 * limit on the path.  Use this to set the slow
1876 		 * start threshold, but set the threshold to no less
1877 		 * than 2 * MSS.
1878 		 */
1879 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1880 	}
1881 #endif
1882 }
1883 
1884 /*
1885  * Processing necessary when a TCP connection is established.
1886  */
1887 void
1888 tcp_established(tp)
1889 	struct tcpcb *tp;
1890 {
1891 	struct socket *so;
1892 #ifdef RTV_RPIPE
1893 	struct rtentry *rt;
1894 #endif
1895 	u_long bufsize;
1896 
1897 #ifdef DIAGNOSTIC
1898 	if (tp->t_inpcb && tp->t_in6pcb)
1899 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1900 #endif
1901 	so = NULL;
1902 	rt = NULL;
1903 #ifdef INET
1904 	if (tp->t_inpcb) {
1905 		so = tp->t_inpcb->inp_socket;
1906 #if defined(RTV_RPIPE)
1907 		rt = in_pcbrtentry(tp->t_inpcb);
1908 #endif
1909 	}
1910 #endif
1911 #ifdef INET6
1912 	if (tp->t_in6pcb) {
1913 		so = tp->t_in6pcb->in6p_socket;
1914 #if defined(RTV_RPIPE)
1915 		rt = in6_pcbrtentry(tp->t_in6pcb);
1916 #endif
1917 	}
1918 #endif
1919 
1920 	tp->t_state = TCPS_ESTABLISHED;
1921 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1922 
1923 #ifdef RTV_RPIPE
1924 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1925 		bufsize = rt->rt_rmx.rmx_recvpipe;
1926 	else
1927 #endif
1928 		bufsize = so->so_rcv.sb_hiwat;
1929 	if (bufsize > tp->t_ourmss) {
1930 		bufsize = roundup(bufsize, tp->t_ourmss);
1931 		if (bufsize > sb_max)
1932 			bufsize = sb_max;
1933 		(void) sbreserve(&so->so_rcv, bufsize);
1934 	}
1935 }
1936 
1937 /*
1938  * Check if there's an initial rtt or rttvar.  Convert from the
1939  * route-table units to scaled multiples of the slow timeout timer.
1940  * Called only during the 3-way handshake.
1941  */
1942 void
1943 tcp_rmx_rtt(tp)
1944 	struct tcpcb *tp;
1945 {
1946 #ifdef RTV_RTT
1947 	struct rtentry *rt = NULL;
1948 	int rtt;
1949 
1950 #ifdef DIAGNOSTIC
1951 	if (tp->t_inpcb && tp->t_in6pcb)
1952 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1953 #endif
1954 #ifdef INET
1955 	if (tp->t_inpcb)
1956 		rt = in_pcbrtentry(tp->t_inpcb);
1957 #endif
1958 #ifdef INET6
1959 	if (tp->t_in6pcb)
1960 		rt = in6_pcbrtentry(tp->t_in6pcb);
1961 #endif
1962 	if (rt == NULL)
1963 		return;
1964 
1965 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1966 		/*
1967 		 * XXX The lock bit for MTU indicates that the value
1968 		 * is also a minimum value; this is subject to time.
1969 		 */
1970 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1971 			TCPT_RANGESET(tp->t_rttmin,
1972 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1973 			    TCPTV_MIN, TCPTV_REXMTMAX);
1974 		tp->t_srtt = rtt /
1975 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1976 		if (rt->rt_rmx.rmx_rttvar) {
1977 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1978 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1979 				(TCP_RTTVAR_SHIFT + 2));
1980 		} else {
1981 			/* Default variation is +- 1 rtt */
1982 			tp->t_rttvar =
1983 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1984 		}
1985 		TCPT_RANGESET(tp->t_rxtcur,
1986 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1987 		    tp->t_rttmin, TCPTV_REXMTMAX);
1988 	}
1989 #endif
1990 }
1991 
1992 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1993 #if NRND > 0
1994 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1995 #endif
1996 
1997 /*
1998  * Get a new sequence value given a tcp control block
1999  */
2000 tcp_seq
2001 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2002 {
2003 
2004 #ifdef INET
2005 	if (tp->t_inpcb != NULL) {
2006 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2007 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2008 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2009 		    addin));
2010 	}
2011 #endif
2012 #ifdef INET6
2013 	if (tp->t_in6pcb != NULL) {
2014 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2015 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2016 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2017 		    addin));
2018 	}
2019 #endif
2020 	/* Not possible. */
2021 	panic("tcp_new_iss");
2022 }
2023 
2024 /*
2025  * This routine actually generates a new TCP initial sequence number.
2026  */
2027 tcp_seq
2028 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2029     size_t addrsz, tcp_seq addin)
2030 {
2031 	tcp_seq tcp_iss;
2032 
2033 #if NRND > 0
2034 	static int beenhere;
2035 
2036 	/*
2037 	 * If we haven't been here before, initialize our cryptographic
2038 	 * hash secret.
2039 	 */
2040 	if (beenhere == 0) {
2041 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2042 		    RND_EXTRACT_ANY);
2043 		beenhere = 1;
2044 	}
2045 
2046 	if (tcp_do_rfc1948) {
2047 		MD5_CTX ctx;
2048 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2049 
2050 		/*
2051 		 * Compute the base value of the ISS.  It is a hash
2052 		 * of (saddr, sport, daddr, dport, secret).
2053 		 */
2054 		MD5Init(&ctx);
2055 
2056 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2057 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2058 
2059 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2060 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2061 
2062 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2063 
2064 		MD5Final(hash, &ctx);
2065 
2066 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2067 
2068 		/*
2069 		 * Now increment our "timer", and add it in to
2070 		 * the computed value.
2071 		 *
2072 		 * XXX Use `addin'?
2073 		 * XXX TCP_ISSINCR too large to use?
2074 		 */
2075 		tcp_iss_seq += TCP_ISSINCR;
2076 #ifdef TCPISS_DEBUG
2077 		printf("ISS hash 0x%08x, ", tcp_iss);
2078 #endif
2079 		tcp_iss += tcp_iss_seq + addin;
2080 #ifdef TCPISS_DEBUG
2081 		printf("new ISS 0x%08x\n", tcp_iss);
2082 #endif
2083 	} else
2084 #endif /* NRND > 0 */
2085 	{
2086 		/*
2087 		 * Randomize.
2088 		 */
2089 #if NRND > 0
2090 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2091 #else
2092 		tcp_iss = arc4random();
2093 #endif
2094 
2095 		/*
2096 		 * If we were asked to add some amount to a known value,
2097 		 * we will take a random value obtained above, mask off
2098 		 * the upper bits, and add in the known value.  We also
2099 		 * add in a constant to ensure that we are at least a
2100 		 * certain distance from the original value.
2101 		 *
2102 		 * This is used when an old connection is in timed wait
2103 		 * and we have a new one coming in, for instance.
2104 		 */
2105 		if (addin != 0) {
2106 #ifdef TCPISS_DEBUG
2107 			printf("Random %08x, ", tcp_iss);
2108 #endif
2109 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2110 			tcp_iss += addin + TCP_ISSINCR;
2111 #ifdef TCPISS_DEBUG
2112 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2113 #endif
2114 		} else {
2115 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2116 			tcp_iss += tcp_iss_seq;
2117 			tcp_iss_seq += TCP_ISSINCR;
2118 #ifdef TCPISS_DEBUG
2119 			printf("ISS %08x\n", tcp_iss);
2120 #endif
2121 		}
2122 	}
2123 
2124 	if (tcp_compat_42) {
2125 		/*
2126 		 * Limit it to the positive range for really old TCP
2127 		 * implementations.
2128 		 * Just AND off the top bit instead of checking if
2129 		 * is set first - saves a branch 50% of the time.
2130 		 */
2131 		tcp_iss &= 0x7fffffff;		/* XXX */
2132 	}
2133 
2134 	return (tcp_iss);
2135 }
2136 
2137 #ifdef IPSEC
2138 /* compute ESP/AH header size for TCP, including outer IP header. */
2139 size_t
2140 ipsec4_hdrsiz_tcp(tp)
2141 	struct tcpcb *tp;
2142 {
2143 	struct inpcb *inp;
2144 	size_t hdrsiz;
2145 
2146 	/* XXX mapped addr case (tp->t_in6pcb) */
2147 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2148 		return 0;
2149 	switch (tp->t_family) {
2150 	case AF_INET:
2151 		/* XXX: should use currect direction. */
2152 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2153 		break;
2154 	default:
2155 		hdrsiz = 0;
2156 		break;
2157 	}
2158 
2159 	return hdrsiz;
2160 }
2161 
2162 #ifdef INET6
2163 size_t
2164 ipsec6_hdrsiz_tcp(tp)
2165 	struct tcpcb *tp;
2166 {
2167 	struct in6pcb *in6p;
2168 	size_t hdrsiz;
2169 
2170 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2171 		return 0;
2172 	switch (tp->t_family) {
2173 	case AF_INET6:
2174 		/* XXX: should use currect direction. */
2175 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2176 		break;
2177 	case AF_INET:
2178 		/* mapped address case - tricky */
2179 	default:
2180 		hdrsiz = 0;
2181 		break;
2182 	}
2183 
2184 	return hdrsiz;
2185 }
2186 #endif
2187 #endif /*IPSEC*/
2188 
2189 /*
2190  * Determine the length of the TCP options for this connection.
2191  *
2192  * XXX:  What do we do for SACK, when we add that?  Just reserve
2193  *       all of the space?  Otherwise we can't exactly be incrementing
2194  *       cwnd by an amount that varies depending on the amount we last
2195  *       had to SACK!
2196  */
2197 
2198 u_int
2199 tcp_optlen(tp)
2200 	struct tcpcb *tp;
2201 {
2202 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2203 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2204 		return TCPOLEN_TSTAMP_APPA;
2205 	else
2206 		return 0;
2207 }
2208