xref: /netbsd-src/sys/netinet/tcp_subr.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: tcp_subr.c,v 1.160 2004/01/07 19:15:43 matt 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 1.160 2004/01/07 19:15:43 matt Exp $");
102 
103 #include "opt_inet.h"
104 #include "opt_ipsec.h"
105 #include "opt_tcp_compat_42.h"
106 #include "opt_inet_csum.h"
107 #include "opt_mbuftrace.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include <netinet/tcp.h>
150 #include <netinet/tcp_fsm.h>
151 #include <netinet/tcp_seq.h>
152 #include <netinet/tcp_timer.h>
153 #include <netinet/tcp_var.h>
154 #include <netinet/tcpip.h>
155 
156 #ifdef IPSEC
157 #include <netinet6/ipsec.h>
158 #endif /*IPSEC*/
159 
160 #ifdef FAST_IPSEC
161 #include <netipsec/ipsec.h>
162 #ifdef INET6
163 #include <netipsec/ipsec6.h>
164 #endif
165 #endif	/* FAST_IPSEC*/
166 
167 
168 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
169 struct	tcpstat tcpstat;	/* tcp statistics */
170 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
171 
172 /* patchable/settable parameters for tcp */
173 int 	tcp_mssdflt = TCP_MSS;
174 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
175 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
176 #if NRND > 0
177 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
178 #endif
179 int	tcp_do_sack = 1;	/* selective acknowledgement */
180 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
181 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
182 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
183 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
184 #ifndef TCP_INIT_WIN
185 #define	TCP_INIT_WIN	1	/* initial slow start window */
186 #endif
187 #ifndef TCP_INIT_WIN_LOCAL
188 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
189 #endif
190 int	tcp_init_win = TCP_INIT_WIN;
191 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
192 int	tcp_mss_ifmtu = 0;
193 #ifdef TCP_COMPAT_42
194 int	tcp_compat_42 = 1;
195 #else
196 int	tcp_compat_42 = 0;
197 #endif
198 int	tcp_rst_ppslim = 100;	/* 100pps */
199 
200 /* tcb hash */
201 #ifndef TCBHASHSIZE
202 #define	TCBHASHSIZE	128
203 #endif
204 int	tcbhashsize = TCBHASHSIZE;
205 
206 /* syn hash parameters */
207 #define	TCP_SYN_HASH_SIZE	293
208 #define	TCP_SYN_BUCKET_SIZE	35
209 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
210 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
211 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
212 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
213 
214 int	tcp_freeq __P((struct tcpcb *));
215 
216 #ifdef INET
217 void	tcp_mtudisc_callback __P((struct in_addr));
218 #endif
219 #ifdef INET6
220 void	tcp6_mtudisc_callback __P((struct in6_addr *));
221 #endif
222 
223 void	tcp_mtudisc __P((struct inpcb *, int));
224 #ifdef INET6
225 void	tcp6_mtudisc __P((struct in6pcb *, int));
226 #endif
227 
228 struct pool tcpcb_pool;
229 
230 #ifdef TCP_CSUM_COUNTERS
231 #include <sys/device.h>
232 
233 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
234     NULL, "tcp", "hwcsum bad");
235 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
236     NULL, "tcp", "hwcsum ok");
237 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
238     NULL, "tcp", "hwcsum data");
239 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "swcsum");
241 #endif /* TCP_CSUM_COUNTERS */
242 
243 #ifdef TCP_OUTPUT_COUNTERS
244 #include <sys/device.h>
245 
246 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247     NULL, "tcp", "output big header");
248 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
249     NULL, "tcp", "output predict hit");
250 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
251     NULL, "tcp", "output predict miss");
252 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
253     NULL, "tcp", "output copy small");
254 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
255     NULL, "tcp", "output copy big");
256 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
257     NULL, "tcp", "output reference big");
258 #endif /* TCP_OUTPUT_COUNTERS */
259 
260 #ifdef TCP_REASS_COUNTERS
261 #include <sys/device.h>
262 
263 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp_reass", "calls");
265 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     &tcp_reass_, "tcp_reass", "insert into empty queue");
267 struct evcnt tcp_reass_iteration[8] = {
268     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
269     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
270     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
271     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
272     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
273     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
274     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
275     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
276 };
277 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     &tcp_reass_, "tcp_reass", "prepend to first");
279 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     &tcp_reass_, "tcp_reass", "prepend");
281 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     &tcp_reass_, "tcp_reass", "insert");
283 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "insert at tail");
285 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286     &tcp_reass_, "tcp_reass", "append");
287 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
288     &tcp_reass_, "tcp_reass", "append to tail fragment");
289 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
290     &tcp_reass_, "tcp_reass", "overlap at end");
291 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
292     &tcp_reass_, "tcp_reass", "overlap at start");
293 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
294     &tcp_reass_, "tcp_reass", "duplicate segment");
295 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
296     &tcp_reass_, "tcp_reass", "duplicate fragment");
297 
298 #endif /* TCP_REASS_COUNTERS */
299 
300 #ifdef MBUFTRACE
301 struct mowner tcp_mowner = { "tcp" };
302 struct mowner tcp_rx_mowner = { "tcp", "rx" };
303 struct mowner tcp_tx_mowner = { "tcp", "tx" };
304 #endif
305 
306 /*
307  * Tcp initialization
308  */
309 void
310 tcp_init()
311 {
312 	int hlen;
313 
314 	/* Initialize the TCPCB template. */
315 	tcp_tcpcb_template();
316 
317 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
318 	    NULL);
319 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
320 
321 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
322 #ifdef INET6
323 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
324 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
325 #endif
326 	if (max_protohdr < hlen)
327 		max_protohdr = hlen;
328 	if (max_linkhdr + hlen > MHLEN)
329 		panic("tcp_init");
330 
331 #ifdef INET
332 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
333 #endif
334 #ifdef INET6
335 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
336 #endif
337 
338 	/* Initialize timer state. */
339 	tcp_timer_init();
340 
341 	/* Initialize the compressed state engine. */
342 	syn_cache_init();
343 
344 #ifdef TCP_CSUM_COUNTERS
345 	evcnt_attach_static(&tcp_hwcsum_bad);
346 	evcnt_attach_static(&tcp_hwcsum_ok);
347 	evcnt_attach_static(&tcp_hwcsum_data);
348 	evcnt_attach_static(&tcp_swcsum);
349 #endif /* TCP_CSUM_COUNTERS */
350 
351 #ifdef TCP_OUTPUT_COUNTERS
352 	evcnt_attach_static(&tcp_output_bigheader);
353 	evcnt_attach_static(&tcp_output_predict_hit);
354 	evcnt_attach_static(&tcp_output_predict_miss);
355 	evcnt_attach_static(&tcp_output_copysmall);
356 	evcnt_attach_static(&tcp_output_copybig);
357 	evcnt_attach_static(&tcp_output_refbig);
358 #endif /* TCP_OUTPUT_COUNTERS */
359 
360 #ifdef TCP_REASS_COUNTERS
361 	evcnt_attach_static(&tcp_reass_);
362 	evcnt_attach_static(&tcp_reass_empty);
363 	evcnt_attach_static(&tcp_reass_iteration[0]);
364 	evcnt_attach_static(&tcp_reass_iteration[1]);
365 	evcnt_attach_static(&tcp_reass_iteration[2]);
366 	evcnt_attach_static(&tcp_reass_iteration[3]);
367 	evcnt_attach_static(&tcp_reass_iteration[4]);
368 	evcnt_attach_static(&tcp_reass_iteration[5]);
369 	evcnt_attach_static(&tcp_reass_iteration[6]);
370 	evcnt_attach_static(&tcp_reass_iteration[7]);
371 	evcnt_attach_static(&tcp_reass_prependfirst);
372 	evcnt_attach_static(&tcp_reass_prepend);
373 	evcnt_attach_static(&tcp_reass_insert);
374 	evcnt_attach_static(&tcp_reass_inserttail);
375 	evcnt_attach_static(&tcp_reass_append);
376 	evcnt_attach_static(&tcp_reass_appendtail);
377 	evcnt_attach_static(&tcp_reass_overlaptail);
378 	evcnt_attach_static(&tcp_reass_overlapfront);
379 	evcnt_attach_static(&tcp_reass_segdup);
380 	evcnt_attach_static(&tcp_reass_fragdup);
381 #endif /* TCP_REASS_COUNTERS */
382 
383 	MOWNER_ATTACH(&tcp_tx_mowner);
384 	MOWNER_ATTACH(&tcp_rx_mowner);
385 	MOWNER_ATTACH(&tcp_mowner);
386 }
387 
388 /*
389  * Create template to be used to send tcp packets on a connection.
390  * Call after host entry created, allocates an mbuf and fills
391  * in a skeletal tcp/ip header, minimizing the amount of work
392  * necessary when the connection is used.
393  */
394 struct mbuf *
395 tcp_template(tp)
396 	struct tcpcb *tp;
397 {
398 	struct inpcb *inp = tp->t_inpcb;
399 #ifdef INET6
400 	struct in6pcb *in6p = tp->t_in6pcb;
401 #endif
402 	struct tcphdr *n;
403 	struct mbuf *m;
404 	int hlen;
405 
406 	switch (tp->t_family) {
407 	case AF_INET:
408 		hlen = sizeof(struct ip);
409 		if (inp)
410 			break;
411 #ifdef INET6
412 		if (in6p) {
413 			/* mapped addr case */
414 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
415 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
416 				break;
417 		}
418 #endif
419 		return NULL;	/*EINVAL*/
420 #ifdef INET6
421 	case AF_INET6:
422 		hlen = sizeof(struct ip6_hdr);
423 		if (in6p) {
424 			/* more sainty check? */
425 			break;
426 		}
427 		return NULL;	/*EINVAL*/
428 #endif
429 	default:
430 		hlen = 0;	/*pacify gcc*/
431 		return NULL;	/*EAFNOSUPPORT*/
432 	}
433 #ifdef DIAGNOSTIC
434 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
435 		panic("mclbytes too small for t_template");
436 #endif
437 	m = tp->t_template;
438 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
439 		;
440 	else {
441 		if (m)
442 			m_freem(m);
443 		m = tp->t_template = NULL;
444 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
445 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
446 			MCLGET(m, M_DONTWAIT);
447 			if ((m->m_flags & M_EXT) == 0) {
448 				m_free(m);
449 				m = NULL;
450 			}
451 		}
452 		if (m == NULL)
453 			return NULL;
454 		MCLAIM(m, &tcp_mowner);
455 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
456 	}
457 
458 	bzero(mtod(m, caddr_t), m->m_len);
459 
460 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
461 
462 	switch (tp->t_family) {
463 	case AF_INET:
464 	    {
465 		struct ipovly *ipov;
466 		mtod(m, struct ip *)->ip_v = 4;
467 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
468 		ipov = mtod(m, struct ipovly *);
469 		ipov->ih_pr = IPPROTO_TCP;
470 		ipov->ih_len = htons(sizeof(struct tcphdr));
471 		if (inp) {
472 			ipov->ih_src = inp->inp_laddr;
473 			ipov->ih_dst = inp->inp_faddr;
474 		}
475 #ifdef INET6
476 		else if (in6p) {
477 			/* mapped addr case */
478 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
479 				sizeof(ipov->ih_src));
480 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
481 				sizeof(ipov->ih_dst));
482 		}
483 #endif
484 		/*
485 		 * Compute the pseudo-header portion of the checksum
486 		 * now.  We incrementally add in the TCP option and
487 		 * payload lengths later, and then compute the TCP
488 		 * checksum right before the packet is sent off onto
489 		 * the wire.
490 		 */
491 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
492 		    ipov->ih_dst.s_addr,
493 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
494 		break;
495 	    }
496 #ifdef INET6
497 	case AF_INET6:
498 	    {
499 		struct ip6_hdr *ip6;
500 		mtod(m, struct ip *)->ip_v = 6;
501 		ip6 = mtod(m, struct ip6_hdr *);
502 		ip6->ip6_nxt = IPPROTO_TCP;
503 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
504 		ip6->ip6_src = in6p->in6p_laddr;
505 		ip6->ip6_dst = in6p->in6p_faddr;
506 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
507 		if (ip6_auto_flowlabel) {
508 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
509 			ip6->ip6_flow |=
510 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
511 		}
512 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
513 		ip6->ip6_vfc |= IPV6_VERSION;
514 
515 		/*
516 		 * Compute the pseudo-header portion of the checksum
517 		 * now.  We incrementally add in the TCP option and
518 		 * payload lengths later, and then compute the TCP
519 		 * checksum right before the packet is sent off onto
520 		 * the wire.
521 		 */
522 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
523 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
524 		    htonl(IPPROTO_TCP));
525 		break;
526 	    }
527 #endif
528 	}
529 	if (inp) {
530 		n->th_sport = inp->inp_lport;
531 		n->th_dport = inp->inp_fport;
532 	}
533 #ifdef INET6
534 	else if (in6p) {
535 		n->th_sport = in6p->in6p_lport;
536 		n->th_dport = in6p->in6p_fport;
537 	}
538 #endif
539 	n->th_seq = 0;
540 	n->th_ack = 0;
541 	n->th_x2 = 0;
542 	n->th_off = 5;
543 	n->th_flags = 0;
544 	n->th_win = 0;
545 	n->th_urp = 0;
546 	return (m);
547 }
548 
549 /*
550  * Send a single message to the TCP at address specified by
551  * the given TCP/IP header.  If m == 0, then we make a copy
552  * of the tcpiphdr at ti and send directly to the addressed host.
553  * This is used to force keep alive messages out using the TCP
554  * template for a connection tp->t_template.  If flags are given
555  * then we send a message back to the TCP which originated the
556  * segment ti, and discard the mbuf containing it and any other
557  * attached mbufs.
558  *
559  * In any case the ack and sequence number of the transmitted
560  * segment are as specified by the parameters.
561  */
562 int
563 tcp_respond(tp, template, m, th0, ack, seq, flags)
564 	struct tcpcb *tp;
565 	struct mbuf *template;
566 	struct mbuf *m;
567 	struct tcphdr *th0;
568 	tcp_seq ack, seq;
569 	int flags;
570 {
571 	struct route *ro;
572 	int error, tlen, win = 0;
573 	int hlen;
574 	struct ip *ip;
575 #ifdef INET6
576 	struct ip6_hdr *ip6;
577 #endif
578 	int family;	/* family on packet, not inpcb/in6pcb! */
579 	struct tcphdr *th;
580 	struct socket *so;
581 
582 	if (tp != NULL && (flags & TH_RST) == 0) {
583 #ifdef DIAGNOSTIC
584 		if (tp->t_inpcb && tp->t_in6pcb)
585 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
586 #endif
587 #ifdef INET
588 		if (tp->t_inpcb)
589 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
590 #endif
591 #ifdef INET6
592 		if (tp->t_in6pcb)
593 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
594 #endif
595 	}
596 
597 	th = NULL;	/* Quell uninitialized warning */
598 	ip = NULL;
599 #ifdef INET6
600 	ip6 = NULL;
601 #endif
602 	if (m == 0) {
603 		if (!template)
604 			return EINVAL;
605 
606 		/* get family information from template */
607 		switch (mtod(template, struct ip *)->ip_v) {
608 		case 4:
609 			family = AF_INET;
610 			hlen = sizeof(struct ip);
611 			break;
612 #ifdef INET6
613 		case 6:
614 			family = AF_INET6;
615 			hlen = sizeof(struct ip6_hdr);
616 			break;
617 #endif
618 		default:
619 			return EAFNOSUPPORT;
620 		}
621 
622 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
623 		if (m) {
624 			MCLAIM(m, &tcp_tx_mowner);
625 			MCLGET(m, M_DONTWAIT);
626 			if ((m->m_flags & M_EXT) == 0) {
627 				m_free(m);
628 				m = NULL;
629 			}
630 		}
631 		if (m == NULL)
632 			return (ENOBUFS);
633 
634 		if (tcp_compat_42)
635 			tlen = 1;
636 		else
637 			tlen = 0;
638 
639 		m->m_data += max_linkhdr;
640 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
641 			template->m_len);
642 		switch (family) {
643 		case AF_INET:
644 			ip = mtod(m, struct ip *);
645 			th = (struct tcphdr *)(ip + 1);
646 			break;
647 #ifdef INET6
648 		case AF_INET6:
649 			ip6 = mtod(m, struct ip6_hdr *);
650 			th = (struct tcphdr *)(ip6 + 1);
651 			break;
652 #endif
653 #if 0
654 		default:
655 			/* noone will visit here */
656 			m_freem(m);
657 			return EAFNOSUPPORT;
658 #endif
659 		}
660 		flags = TH_ACK;
661 	} else {
662 
663 		if ((m->m_flags & M_PKTHDR) == 0) {
664 #if 0
665 			printf("non PKTHDR to tcp_respond\n");
666 #endif
667 			m_freem(m);
668 			return EINVAL;
669 		}
670 #ifdef DIAGNOSTIC
671 		if (!th0)
672 			panic("th0 == NULL in tcp_respond");
673 #endif
674 
675 		/* get family information from m */
676 		switch (mtod(m, struct ip *)->ip_v) {
677 		case 4:
678 			family = AF_INET;
679 			hlen = sizeof(struct ip);
680 			ip = mtod(m, struct ip *);
681 			break;
682 #ifdef INET6
683 		case 6:
684 			family = AF_INET6;
685 			hlen = sizeof(struct ip6_hdr);
686 			ip6 = mtod(m, struct ip6_hdr *);
687 			break;
688 #endif
689 		default:
690 			m_freem(m);
691 			return EAFNOSUPPORT;
692 		}
693 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
694 			tlen = sizeof(*th0);
695 		else
696 			tlen = th0->th_off << 2;
697 
698 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
699 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
700 			m->m_len = hlen + tlen;
701 			m_freem(m->m_next);
702 			m->m_next = NULL;
703 		} else {
704 			struct mbuf *n;
705 
706 #ifdef DIAGNOSTIC
707 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
708 				m_freem(m);
709 				return EMSGSIZE;
710 			}
711 #endif
712 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
713 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
714 				MCLGET(n, M_DONTWAIT);
715 				if ((n->m_flags & M_EXT) == 0) {
716 					m_freem(n);
717 					n = NULL;
718 				}
719 			}
720 			if (!n) {
721 				m_freem(m);
722 				return ENOBUFS;
723 			}
724 
725 			MCLAIM(n, &tcp_tx_mowner);
726 			n->m_data += max_linkhdr;
727 			n->m_len = hlen + tlen;
728 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
729 			m_copyback(n, hlen, tlen, (caddr_t)th0);
730 
731 			m_freem(m);
732 			m = n;
733 			n = NULL;
734 		}
735 
736 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
737 		switch (family) {
738 		case AF_INET:
739 			ip = mtod(m, struct ip *);
740 			th = (struct tcphdr *)(ip + 1);
741 			ip->ip_p = IPPROTO_TCP;
742 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
743 			ip->ip_p = IPPROTO_TCP;
744 			break;
745 #ifdef INET6
746 		case AF_INET6:
747 			ip6 = mtod(m, struct ip6_hdr *);
748 			th = (struct tcphdr *)(ip6 + 1);
749 			ip6->ip6_nxt = IPPROTO_TCP;
750 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
751 			ip6->ip6_nxt = IPPROTO_TCP;
752 			break;
753 #endif
754 #if 0
755 		default:
756 			/* noone will visit here */
757 			m_freem(m);
758 			return EAFNOSUPPORT;
759 #endif
760 		}
761 		xchg(th->th_dport, th->th_sport, u_int16_t);
762 #undef xchg
763 		tlen = 0;	/*be friendly with the following code*/
764 	}
765 	th->th_seq = htonl(seq);
766 	th->th_ack = htonl(ack);
767 	th->th_x2 = 0;
768 	if ((flags & TH_SYN) == 0) {
769 		if (tp)
770 			win >>= tp->rcv_scale;
771 		if (win > TCP_MAXWIN)
772 			win = TCP_MAXWIN;
773 		th->th_win = htons((u_int16_t)win);
774 		th->th_off = sizeof (struct tcphdr) >> 2;
775 		tlen += sizeof(*th);
776 	} else
777 		tlen += th->th_off << 2;
778 	m->m_len = hlen + tlen;
779 	m->m_pkthdr.len = hlen + tlen;
780 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
781 	th->th_flags = flags;
782 	th->th_urp = 0;
783 
784 	switch (family) {
785 #ifdef INET
786 	case AF_INET:
787 	    {
788 		struct ipovly *ipov = (struct ipovly *)ip;
789 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
790 		ipov->ih_len = htons((u_int16_t)tlen);
791 
792 		th->th_sum = 0;
793 		th->th_sum = in_cksum(m, hlen + tlen);
794 		ip->ip_len = htons(hlen + tlen);
795 		ip->ip_ttl = ip_defttl;
796 		break;
797 	    }
798 #endif
799 #ifdef INET6
800 	case AF_INET6:
801 	    {
802 		th->th_sum = 0;
803 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
804 				tlen);
805 		ip6->ip6_plen = ntohs(tlen);
806 		if (tp && tp->t_in6pcb) {
807 			struct ifnet *oifp;
808 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
809 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
810 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
811 		} else
812 			ip6->ip6_hlim = ip6_defhlim;
813 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
814 		if (ip6_auto_flowlabel) {
815 			ip6->ip6_flow |=
816 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
817 		}
818 		break;
819 	    }
820 #endif
821 	}
822 
823 	if (tp && tp->t_inpcb)
824 		so = tp->t_inpcb->inp_socket;
825 #ifdef INET6
826 	else if (tp && tp->t_in6pcb)
827 		so = tp->t_in6pcb->in6p_socket;
828 #endif
829 	else
830 		so = NULL;
831 
832 	if (tp != NULL && tp->t_inpcb != NULL) {
833 		ro = &tp->t_inpcb->inp_route;
834 #ifdef DIAGNOSTIC
835 		if (family != AF_INET)
836 			panic("tcp_respond: address family mismatch");
837 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
838 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
839 			    ntohl(ip->ip_dst.s_addr),
840 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
841 		}
842 #endif
843 	}
844 #ifdef INET6
845 	else if (tp != NULL && tp->t_in6pcb != NULL) {
846 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
847 #ifdef DIAGNOSTIC
848 		if (family == AF_INET) {
849 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
850 				panic("tcp_respond: not mapped addr");
851 			if (bcmp(&ip->ip_dst,
852 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
853 			    sizeof(ip->ip_dst)) != 0) {
854 				panic("tcp_respond: ip_dst != in6p_faddr");
855 			}
856 		} else if (family == AF_INET6) {
857 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
858 			    &tp->t_in6pcb->in6p_faddr))
859 				panic("tcp_respond: ip6_dst != in6p_faddr");
860 		} else
861 			panic("tcp_respond: address family mismatch");
862 #endif
863 	}
864 #endif
865 	else
866 		ro = NULL;
867 
868 	switch (family) {
869 #ifdef INET
870 	case AF_INET:
871 		error = ip_output(m, NULL, ro,
872 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
873 		    (struct ip_moptions *)0, so);
874 		break;
875 #endif
876 #ifdef INET6
877 	case AF_INET6:
878 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
879 		    (struct ip6_moptions *)0, so, NULL);
880 		break;
881 #endif
882 	default:
883 		error = EAFNOSUPPORT;
884 		break;
885 	}
886 
887 	return (error);
888 }
889 
890 /*
891  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
892  * a bunch of members individually, we maintain this template for the
893  * static and mostly-static components of the TCPCB, and copy it into
894  * the new TCPCB instead.
895  */
896 static struct tcpcb tcpcb_template = {
897 	/*
898 	 * If TCP_NTIMERS ever changes, we'll need to update this
899 	 * initializer.
900 	 */
901 	.t_timer = {
902 		CALLOUT_INITIALIZER,
903 		CALLOUT_INITIALIZER,
904 		CALLOUT_INITIALIZER,
905 		CALLOUT_INITIALIZER,
906 	},
907 	.t_delack_ch = CALLOUT_INITIALIZER,
908 
909 	.t_srtt = TCPTV_SRTTBASE,
910 	.t_rttmin = TCPTV_MIN,
911 
912 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
913 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
914 };
915 
916 /*
917  * Updates the TCPCB template whenever a parameter that would affect
918  * the template is changed.
919  */
920 void
921 tcp_tcpcb_template(void)
922 {
923 	struct tcpcb *tp = &tcpcb_template;
924 	int flags;
925 
926 	tp->t_peermss = tcp_mssdflt;
927 	tp->t_ourmss = tcp_mssdflt;
928 	tp->t_segsz = tcp_mssdflt;
929 
930 	flags = 0;
931 	if (tcp_do_rfc1323 && tcp_do_win_scale)
932 		flags |= TF_REQ_SCALE;
933 	if (tcp_do_rfc1323 && tcp_do_timestamps)
934 		flags |= TF_REQ_TSTMP;
935 	if (tcp_do_sack == 2)
936 		flags |= TF_WILL_SACK;
937 	else if (tcp_do_sack == 1)
938 		flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
939 	flags |= TF_CANT_TXSACK;
940 	tp->t_flags = flags;
941 
942 	/*
943 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
944 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
945 	 * reasonable initial retransmit time.
946 	 */
947 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
948 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
949 	    TCPTV_MIN, TCPTV_REXMTMAX);
950 }
951 
952 /*
953  * Create a new TCP control block, making an
954  * empty reassembly queue and hooking it to the argument
955  * protocol control block.
956  */
957 struct tcpcb *
958 tcp_newtcpcb(family, aux)
959 	int family;	/* selects inpcb, or in6pcb */
960 	void *aux;
961 {
962 	struct tcpcb *tp;
963 	int i;
964 
965 	/* XXX Consider using a pool_cache for speed. */
966 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
967 	if (tp == NULL)
968 		return (NULL);
969 	memcpy(tp, &tcpcb_template, sizeof(*tp));
970 	TAILQ_INIT(&tp->segq);
971 	TAILQ_INIT(&tp->timeq);
972 	tp->t_family = family;		/* may be overridden later on */
973 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
974 
975 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
976 	for (i = 0; i < TCPT_NTIMERS; i++)
977 		TCP_TIMER_INIT(tp, i);
978 
979 	switch (family) {
980 	case AF_INET:
981 	    {
982 		struct inpcb *inp = (struct inpcb *)aux;
983 
984 		inp->inp_ip.ip_ttl = ip_defttl;
985 		inp->inp_ppcb = (caddr_t)tp;
986 
987 		tp->t_inpcb = inp;
988 		tp->t_mtudisc = ip_mtudisc;
989 		break;
990 	    }
991 #ifdef INET6
992 	case AF_INET6:
993 	    {
994 		struct in6pcb *in6p = (struct in6pcb *)aux;
995 
996 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
997 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
998 					       : NULL);
999 		in6p->in6p_ppcb = (caddr_t)tp;
1000 
1001 		tp->t_in6pcb = in6p;
1002 		/* for IPv6, always try to run path MTU discovery */
1003 		tp->t_mtudisc = 1;
1004 		break;
1005 	    }
1006 #endif /* INET6 */
1007 	default:
1008 		pool_put(&tcpcb_pool, tp);
1009 		return (NULL);
1010 	}
1011 
1012 	/*
1013 	 * Initialize our timebase.  When we send timestamps, we take
1014 	 * the delta from tcp_now -- this means each connection always
1015 	 * gets a timebase of 0, which makes it, among other things,
1016 	 * more difficult to determine how long a system has been up,
1017 	 * and thus how many TCP sequence increments have occurred.
1018 	 */
1019 	tp->ts_timebase = tcp_now;
1020 
1021 	return (tp);
1022 }
1023 
1024 /*
1025  * Drop a TCP connection, reporting
1026  * the specified error.  If connection is synchronized,
1027  * then send a RST to peer.
1028  */
1029 struct tcpcb *
1030 tcp_drop(tp, errno)
1031 	struct tcpcb *tp;
1032 	int errno;
1033 {
1034 	struct socket *so = NULL;
1035 
1036 #ifdef DIAGNOSTIC
1037 	if (tp->t_inpcb && tp->t_in6pcb)
1038 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1039 #endif
1040 #ifdef INET
1041 	if (tp->t_inpcb)
1042 		so = tp->t_inpcb->inp_socket;
1043 #endif
1044 #ifdef INET6
1045 	if (tp->t_in6pcb)
1046 		so = tp->t_in6pcb->in6p_socket;
1047 #endif
1048 	if (!so)
1049 		return NULL;
1050 
1051 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1052 		tp->t_state = TCPS_CLOSED;
1053 		(void) tcp_output(tp);
1054 		tcpstat.tcps_drops++;
1055 	} else
1056 		tcpstat.tcps_conndrops++;
1057 	if (errno == ETIMEDOUT && tp->t_softerror)
1058 		errno = tp->t_softerror;
1059 	so->so_error = errno;
1060 	return (tcp_close(tp));
1061 }
1062 
1063 /*
1064  * Return whether this tcpcb is marked as dead, indicating
1065  * to the calling timer function that no further action should
1066  * be taken, as we are about to release this tcpcb.  The release
1067  * of the storage will be done if this is the last timer running.
1068  *
1069  * This is typically called from the callout handler function before
1070  * callout_ack() is done, therefore we need to test the number of
1071  * running timer functions against 1 below, not 0.
1072  */
1073 int
1074 tcp_isdead(tp)
1075 	struct tcpcb *tp;
1076 {
1077 	int dead = (tp->t_flags & TF_DEAD);
1078 
1079 	if (__predict_false(dead)) {
1080 		if (tcp_timers_invoking(tp) > 1)
1081 				/* not quite there yet -- count separately? */
1082 			return dead;
1083 		tcpstat.tcps_delayed_free++;
1084 		pool_put(&tcpcb_pool, tp);
1085 	}
1086 	return dead;
1087 }
1088 
1089 /*
1090  * Close a TCP control block:
1091  *	discard all space held by the tcp
1092  *	discard internet protocol block
1093  *	wake up any sleepers
1094  */
1095 struct tcpcb *
1096 tcp_close(tp)
1097 	struct tcpcb *tp;
1098 {
1099 	struct inpcb *inp;
1100 #ifdef INET6
1101 	struct in6pcb *in6p;
1102 #endif
1103 	struct socket *so;
1104 #ifdef RTV_RTT
1105 	struct rtentry *rt;
1106 #endif
1107 	struct route *ro;
1108 
1109 	inp = tp->t_inpcb;
1110 #ifdef INET6
1111 	in6p = tp->t_in6pcb;
1112 #endif
1113 	so = NULL;
1114 	ro = NULL;
1115 	if (inp) {
1116 		so = inp->inp_socket;
1117 		ro = &inp->inp_route;
1118 	}
1119 #ifdef INET6
1120 	else if (in6p) {
1121 		so = in6p->in6p_socket;
1122 		ro = (struct route *)&in6p->in6p_route;
1123 	}
1124 #endif
1125 
1126 #ifdef RTV_RTT
1127 	/*
1128 	 * If we sent enough data to get some meaningful characteristics,
1129 	 * save them in the routing entry.  'Enough' is arbitrarily
1130 	 * defined as the sendpipesize (default 4K) * 16.  This would
1131 	 * give us 16 rtt samples assuming we only get one sample per
1132 	 * window (the usual case on a long haul net).  16 samples is
1133 	 * enough for the srtt filter to converge to within 5% of the correct
1134 	 * value; fewer samples and we could save a very bogus rtt.
1135 	 *
1136 	 * Don't update the default route's characteristics and don't
1137 	 * update anything that the user "locked".
1138 	 */
1139 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1140 	    ro && (rt = ro->ro_rt) &&
1141 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1142 		u_long i = 0;
1143 
1144 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1145 			i = tp->t_srtt *
1146 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1147 			if (rt->rt_rmx.rmx_rtt && i)
1148 				/*
1149 				 * filter this update to half the old & half
1150 				 * the new values, converting scale.
1151 				 * See route.h and tcp_var.h for a
1152 				 * description of the scaling constants.
1153 				 */
1154 				rt->rt_rmx.rmx_rtt =
1155 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1156 			else
1157 				rt->rt_rmx.rmx_rtt = i;
1158 		}
1159 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1160 			i = tp->t_rttvar *
1161 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1162 			if (rt->rt_rmx.rmx_rttvar && i)
1163 				rt->rt_rmx.rmx_rttvar =
1164 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1165 			else
1166 				rt->rt_rmx.rmx_rttvar = i;
1167 		}
1168 		/*
1169 		 * update the pipelimit (ssthresh) if it has been updated
1170 		 * already or if a pipesize was specified & the threshhold
1171 		 * got below half the pipesize.  I.e., wait for bad news
1172 		 * before we start updating, then update on both good
1173 		 * and bad news.
1174 		 */
1175 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1176 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1177 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1178 			/*
1179 			 * convert the limit from user data bytes to
1180 			 * packets then to packet data bytes.
1181 			 */
1182 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1183 			if (i < 2)
1184 				i = 2;
1185 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1186 			if (rt->rt_rmx.rmx_ssthresh)
1187 				rt->rt_rmx.rmx_ssthresh =
1188 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1189 			else
1190 				rt->rt_rmx.rmx_ssthresh = i;
1191 		}
1192 	}
1193 #endif /* RTV_RTT */
1194 	/* free the reassembly queue, if any */
1195 	TCP_REASS_LOCK(tp);
1196 	(void) tcp_freeq(tp);
1197 	TCP_REASS_UNLOCK(tp);
1198 
1199 	tcp_canceltimers(tp);
1200 	TCP_CLEAR_DELACK(tp);
1201 	syn_cache_cleanup(tp);
1202 
1203 	if (tp->t_template) {
1204 		m_free(tp->t_template);
1205 		tp->t_template = NULL;
1206 	}
1207 	if (tcp_timers_invoking(tp))
1208 		tp->t_flags |= TF_DEAD;
1209 	else
1210 		pool_put(&tcpcb_pool, tp);
1211 
1212 	if (inp) {
1213 		inp->inp_ppcb = 0;
1214 		soisdisconnected(so);
1215 		in_pcbdetach(inp);
1216 	}
1217 #ifdef INET6
1218 	else if (in6p) {
1219 		in6p->in6p_ppcb = 0;
1220 		soisdisconnected(so);
1221 		in6_pcbdetach(in6p);
1222 	}
1223 #endif
1224 	tcpstat.tcps_closed++;
1225 	return ((struct tcpcb *)0);
1226 }
1227 
1228 int
1229 tcp_freeq(tp)
1230 	struct tcpcb *tp;
1231 {
1232 	struct ipqent *qe;
1233 	int rv = 0;
1234 #ifdef TCPREASS_DEBUG
1235 	int i = 0;
1236 #endif
1237 
1238 	TCP_REASS_LOCK_CHECK(tp);
1239 
1240 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1241 #ifdef TCPREASS_DEBUG
1242 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1243 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1244 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1245 #endif
1246 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1247 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1248 		m_freem(qe->ipqe_m);
1249 		pool_put(&ipqent_pool, qe);
1250 		rv = 1;
1251 	}
1252 	return (rv);
1253 }
1254 
1255 /*
1256  * Protocol drain routine.  Called when memory is in short supply.
1257  */
1258 void
1259 tcp_drain()
1260 {
1261 	struct inpcb_hdr *inph;
1262 	struct tcpcb *tp;
1263 
1264 	/*
1265 	 * Free the sequence queue of all TCP connections.
1266 	 */
1267 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1268 		switch (inph->inph_af) {
1269 		case AF_INET:
1270 			tp = intotcpcb((struct inpcb *)inph);
1271 			break;
1272 #ifdef INET6
1273 		case AF_INET6:
1274 			tp = in6totcpcb((struct in6pcb *)inph);
1275 			break;
1276 #endif
1277 		default:
1278 			tp = NULL;
1279 			break;
1280 		}
1281 		if (tp != NULL) {
1282 			/*
1283 			 * We may be called from a device's interrupt
1284 			 * context.  If the tcpcb is already busy,
1285 			 * just bail out now.
1286 			 */
1287 			if (tcp_reass_lock_try(tp) == 0)
1288 				continue;
1289 			if (tcp_freeq(tp))
1290 				tcpstat.tcps_connsdrained++;
1291 			TCP_REASS_UNLOCK(tp);
1292 		}
1293 	}
1294 }
1295 
1296 /*
1297  * Notify a tcp user of an asynchronous error;
1298  * store error as soft error, but wake up user
1299  * (for now, won't do anything until can select for soft error).
1300  */
1301 void
1302 tcp_notify(inp, error)
1303 	struct inpcb *inp;
1304 	int error;
1305 {
1306 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1307 	struct socket *so = inp->inp_socket;
1308 
1309 	/*
1310 	 * Ignore some errors if we are hooked up.
1311 	 * If connection hasn't completed, has retransmitted several times,
1312 	 * and receives a second error, give up now.  This is better
1313 	 * than waiting a long time to establish a connection that
1314 	 * can never complete.
1315 	 */
1316 	if (tp->t_state == TCPS_ESTABLISHED &&
1317 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1318 	      error == EHOSTDOWN)) {
1319 		return;
1320 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1321 	    tp->t_rxtshift > 3 && tp->t_softerror)
1322 		so->so_error = error;
1323 	else
1324 		tp->t_softerror = error;
1325 	wakeup((caddr_t) &so->so_timeo);
1326 	sorwakeup(so);
1327 	sowwakeup(so);
1328 }
1329 
1330 #ifdef INET6
1331 void
1332 tcp6_notify(in6p, error)
1333 	struct in6pcb *in6p;
1334 	int error;
1335 {
1336 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1337 	struct socket *so = in6p->in6p_socket;
1338 
1339 	/*
1340 	 * Ignore some errors if we are hooked up.
1341 	 * If connection hasn't completed, has retransmitted several times,
1342 	 * and receives a second error, give up now.  This is better
1343 	 * than waiting a long time to establish a connection that
1344 	 * can never complete.
1345 	 */
1346 	if (tp->t_state == TCPS_ESTABLISHED &&
1347 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1348 	      error == EHOSTDOWN)) {
1349 		return;
1350 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1351 	    tp->t_rxtshift > 3 && tp->t_softerror)
1352 		so->so_error = error;
1353 	else
1354 		tp->t_softerror = error;
1355 	wakeup((caddr_t) &so->so_timeo);
1356 	sorwakeup(so);
1357 	sowwakeup(so);
1358 }
1359 #endif
1360 
1361 #ifdef INET6
1362 void
1363 tcp6_ctlinput(cmd, sa, d)
1364 	int cmd;
1365 	struct sockaddr *sa;
1366 	void *d;
1367 {
1368 	struct tcphdr th;
1369 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1370 	int nmatch;
1371 	struct ip6_hdr *ip6;
1372 	const struct sockaddr_in6 *sa6_src = NULL;
1373 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1374 	struct mbuf *m;
1375 	int off;
1376 
1377 	if (sa->sa_family != AF_INET6 ||
1378 	    sa->sa_len != sizeof(struct sockaddr_in6))
1379 		return;
1380 	if ((unsigned)cmd >= PRC_NCMDS)
1381 		return;
1382 	else if (cmd == PRC_QUENCH) {
1383 		/* XXX there's no PRC_QUENCH in IPv6 */
1384 		notify = tcp6_quench;
1385 	} else if (PRC_IS_REDIRECT(cmd))
1386 		notify = in6_rtchange, d = NULL;
1387 	else if (cmd == PRC_MSGSIZE)
1388 		; /* special code is present, see below */
1389 	else if (cmd == PRC_HOSTDEAD)
1390 		d = NULL;
1391 	else if (inet6ctlerrmap[cmd] == 0)
1392 		return;
1393 
1394 	/* if the parameter is from icmp6, decode it. */
1395 	if (d != NULL) {
1396 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1397 		m = ip6cp->ip6c_m;
1398 		ip6 = ip6cp->ip6c_ip6;
1399 		off = ip6cp->ip6c_off;
1400 		sa6_src = ip6cp->ip6c_src;
1401 	} else {
1402 		m = NULL;
1403 		ip6 = NULL;
1404 		sa6_src = &sa6_any;
1405 		off = 0;
1406 	}
1407 
1408 	if (ip6) {
1409 		/*
1410 		 * XXX: We assume that when ip6 is non NULL,
1411 		 * M and OFF are valid.
1412 		 */
1413 
1414 		/* check if we can safely examine src and dst ports */
1415 		if (m->m_pkthdr.len < off + sizeof(th)) {
1416 			if (cmd == PRC_MSGSIZE)
1417 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1418 			return;
1419 		}
1420 
1421 		bzero(&th, sizeof(th));
1422 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1423 
1424 		if (cmd == PRC_MSGSIZE) {
1425 			int valid = 0;
1426 
1427 			/*
1428 			 * Check to see if we have a valid TCP connection
1429 			 * corresponding to the address in the ICMPv6 message
1430 			 * payload.
1431 			 */
1432 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1433 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1434 			    th.th_sport, 0))
1435 				valid++;
1436 
1437 			/*
1438 			 * Depending on the value of "valid" and routing table
1439 			 * size (mtudisc_{hi,lo}wat), we will:
1440 			 * - recalcurate the new MTU and create the
1441 			 *   corresponding routing entry, or
1442 			 * - ignore the MTU change notification.
1443 			 */
1444 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1445 
1446 			/*
1447 			 * no need to call in6_pcbnotify, it should have been
1448 			 * called via callback if necessary
1449 			 */
1450 			return;
1451 		}
1452 
1453 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1454 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1455 		if (nmatch == 0 && syn_cache_count &&
1456 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1457 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1458 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1459 			syn_cache_unreach((struct sockaddr *)sa6_src,
1460 					  sa, &th);
1461 	} else {
1462 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1463 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1464 	}
1465 }
1466 #endif
1467 
1468 #ifdef INET
1469 /* assumes that ip header and tcp header are contiguous on mbuf */
1470 void *
1471 tcp_ctlinput(cmd, sa, v)
1472 	int cmd;
1473 	struct sockaddr *sa;
1474 	void *v;
1475 {
1476 	struct ip *ip = v;
1477 	struct tcphdr *th;
1478 	struct icmp *icp;
1479 	extern const int inetctlerrmap[];
1480 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1481 	int errno;
1482 	int nmatch;
1483 #ifdef INET6
1484 	struct in6_addr src6, dst6;
1485 #endif
1486 
1487 	if (sa->sa_family != AF_INET ||
1488 	    sa->sa_len != sizeof(struct sockaddr_in))
1489 		return NULL;
1490 	if ((unsigned)cmd >= PRC_NCMDS)
1491 		return NULL;
1492 	errno = inetctlerrmap[cmd];
1493 	if (cmd == PRC_QUENCH)
1494 		notify = tcp_quench;
1495 	else if (PRC_IS_REDIRECT(cmd))
1496 		notify = in_rtchange, ip = 0;
1497 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1498 		/*
1499 		 * Check to see if we have a valid TCP connection
1500 		 * corresponding to the address in the ICMP message
1501 		 * payload.
1502 		 *
1503 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1504 		 */
1505 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1506 #ifdef INET6
1507 		memset(&src6, 0, sizeof(src6));
1508 		memset(&dst6, 0, sizeof(dst6));
1509 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1510 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1511 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1512 #endif
1513 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1514 		    ip->ip_src, th->th_sport) != NULL)
1515 			;
1516 #ifdef INET6
1517 		else if (in6_pcblookup_connect(&tcbtable, &dst6,
1518 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1519 			;
1520 #endif
1521 		else
1522 			return NULL;
1523 
1524 		/*
1525 		 * Now that we've validated that we are actually communicating
1526 		 * with the host indicated in the ICMP message, locate the
1527 		 * ICMP header, recalculate the new MTU, and create the
1528 		 * corresponding routing entry.
1529 		 */
1530 		icp = (struct icmp *)((caddr_t)ip -
1531 		    offsetof(struct icmp, icmp_ip));
1532 		icmp_mtudisc(icp, ip->ip_dst);
1533 
1534 		return NULL;
1535 	} else if (cmd == PRC_HOSTDEAD)
1536 		ip = 0;
1537 	else if (errno == 0)
1538 		return NULL;
1539 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1540 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1541 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1542 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1543 		if (nmatch == 0 && syn_cache_count &&
1544 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1545 		    inetctlerrmap[cmd] == ENETUNREACH ||
1546 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1547 			struct sockaddr_in sin;
1548 			bzero(&sin, sizeof(sin));
1549 			sin.sin_len = sizeof(sin);
1550 			sin.sin_family = AF_INET;
1551 			sin.sin_port = th->th_sport;
1552 			sin.sin_addr = ip->ip_src;
1553 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1554 		}
1555 
1556 		/* XXX mapped address case */
1557 	} else
1558 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1559 		    notify);
1560 	return NULL;
1561 }
1562 
1563 /*
1564  * When a source quence is received, we are being notifed of congestion.
1565  * Close the congestion window down to the Loss Window (one segment).
1566  * We will gradually open it again as we proceed.
1567  */
1568 void
1569 tcp_quench(inp, errno)
1570 	struct inpcb *inp;
1571 	int errno;
1572 {
1573 	struct tcpcb *tp = intotcpcb(inp);
1574 
1575 	if (tp)
1576 		tp->snd_cwnd = tp->t_segsz;
1577 }
1578 #endif
1579 
1580 #ifdef INET6
1581 void
1582 tcp6_quench(in6p, errno)
1583 	struct in6pcb *in6p;
1584 	int errno;
1585 {
1586 	struct tcpcb *tp = in6totcpcb(in6p);
1587 
1588 	if (tp)
1589 		tp->snd_cwnd = tp->t_segsz;
1590 }
1591 #endif
1592 
1593 #ifdef INET
1594 /*
1595  * Path MTU Discovery handlers.
1596  */
1597 void
1598 tcp_mtudisc_callback(faddr)
1599 	struct in_addr faddr;
1600 {
1601 #ifdef INET6
1602 	struct in6_addr in6;
1603 #endif
1604 
1605 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1606 #ifdef INET6
1607 	memset(&in6, 0, sizeof(in6));
1608 	in6.s6_addr16[5] = 0xffff;
1609 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1610 	tcp6_mtudisc_callback(&in6);
1611 #endif
1612 }
1613 
1614 /*
1615  * On receipt of path MTU corrections, flush old route and replace it
1616  * with the new one.  Retransmit all unacknowledged packets, to ensure
1617  * that all packets will be received.
1618  */
1619 void
1620 tcp_mtudisc(inp, errno)
1621 	struct inpcb *inp;
1622 	int errno;
1623 {
1624 	struct tcpcb *tp = intotcpcb(inp);
1625 	struct rtentry *rt = in_pcbrtentry(inp);
1626 
1627 	if (tp != 0) {
1628 		if (rt != 0) {
1629 			/*
1630 			 * If this was not a host route, remove and realloc.
1631 			 */
1632 			if ((rt->rt_flags & RTF_HOST) == 0) {
1633 				in_rtchange(inp, errno);
1634 				if ((rt = in_pcbrtentry(inp)) == 0)
1635 					return;
1636 			}
1637 
1638 			/*
1639 			 * Slow start out of the error condition.  We
1640 			 * use the MTU because we know it's smaller
1641 			 * than the previously transmitted segment.
1642 			 *
1643 			 * Note: This is more conservative than the
1644 			 * suggestion in draft-floyd-incr-init-win-03.
1645 			 */
1646 			if (rt->rt_rmx.rmx_mtu != 0)
1647 				tp->snd_cwnd =
1648 				    TCP_INITIAL_WINDOW(tcp_init_win,
1649 				    rt->rt_rmx.rmx_mtu);
1650 		}
1651 
1652 		/*
1653 		 * Resend unacknowledged packets.
1654 		 */
1655 		tp->snd_nxt = tp->snd_una;
1656 		tcp_output(tp);
1657 	}
1658 }
1659 #endif
1660 
1661 #ifdef INET6
1662 /*
1663  * Path MTU Discovery handlers.
1664  */
1665 void
1666 tcp6_mtudisc_callback(faddr)
1667 	struct in6_addr *faddr;
1668 {
1669 	struct sockaddr_in6 sin6;
1670 
1671 	bzero(&sin6, sizeof(sin6));
1672 	sin6.sin6_family = AF_INET6;
1673 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1674 	sin6.sin6_addr = *faddr;
1675 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1676 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1677 }
1678 
1679 void
1680 tcp6_mtudisc(in6p, errno)
1681 	struct in6pcb *in6p;
1682 	int errno;
1683 {
1684 	struct tcpcb *tp = in6totcpcb(in6p);
1685 	struct rtentry *rt = in6_pcbrtentry(in6p);
1686 
1687 	if (tp != 0) {
1688 		if (rt != 0) {
1689 			/*
1690 			 * If this was not a host route, remove and realloc.
1691 			 */
1692 			if ((rt->rt_flags & RTF_HOST) == 0) {
1693 				in6_rtchange(in6p, errno);
1694 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1695 					return;
1696 			}
1697 
1698 			/*
1699 			 * Slow start out of the error condition.  We
1700 			 * use the MTU because we know it's smaller
1701 			 * than the previously transmitted segment.
1702 			 *
1703 			 * Note: This is more conservative than the
1704 			 * suggestion in draft-floyd-incr-init-win-03.
1705 			 */
1706 			if (rt->rt_rmx.rmx_mtu != 0)
1707 				tp->snd_cwnd =
1708 				    TCP_INITIAL_WINDOW(tcp_init_win,
1709 				    rt->rt_rmx.rmx_mtu);
1710 		}
1711 
1712 		/*
1713 		 * Resend unacknowledged packets.
1714 		 */
1715 		tp->snd_nxt = tp->snd_una;
1716 		tcp_output(tp);
1717 	}
1718 }
1719 #endif /* INET6 */
1720 
1721 /*
1722  * Compute the MSS to advertise to the peer.  Called only during
1723  * the 3-way handshake.  If we are the server (peer initiated
1724  * connection), we are called with a pointer to the interface
1725  * on which the SYN packet arrived.  If we are the client (we
1726  * initiated connection), we are called with a pointer to the
1727  * interface out which this connection should go.
1728  *
1729  * NOTE: Do not subtract IP option/extension header size nor IPsec
1730  * header size from MSS advertisement.  MSS option must hold the maximum
1731  * segment size we can accept, so it must always be:
1732  *	 max(if mtu) - ip header - tcp header
1733  */
1734 u_long
1735 tcp_mss_to_advertise(ifp, af)
1736 	const struct ifnet *ifp;
1737 	int af;
1738 {
1739 	extern u_long in_maxmtu;
1740 	u_long mss = 0;
1741 	u_long hdrsiz;
1742 
1743 	/*
1744 	 * In order to avoid defeating path MTU discovery on the peer,
1745 	 * we advertise the max MTU of all attached networks as our MSS,
1746 	 * per RFC 1191, section 3.1.
1747 	 *
1748 	 * We provide the option to advertise just the MTU of
1749 	 * the interface on which we hope this connection will
1750 	 * be receiving.  If we are responding to a SYN, we
1751 	 * will have a pretty good idea about this, but when
1752 	 * initiating a connection there is a bit more doubt.
1753 	 *
1754 	 * We also need to ensure that loopback has a large enough
1755 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1756 	 */
1757 
1758 	if (ifp != NULL)
1759 		switch (af) {
1760 		case AF_INET:
1761 			mss = ifp->if_mtu;
1762 			break;
1763 #ifdef INET6
1764 		case AF_INET6:
1765 			mss = IN6_LINKMTU(ifp);
1766 			break;
1767 #endif
1768 		}
1769 
1770 	if (tcp_mss_ifmtu == 0)
1771 		switch (af) {
1772 		case AF_INET:
1773 			mss = max(in_maxmtu, mss);
1774 			break;
1775 #ifdef INET6
1776 		case AF_INET6:
1777 			mss = max(in6_maxmtu, mss);
1778 			break;
1779 #endif
1780 		}
1781 
1782 	switch (af) {
1783 	case AF_INET:
1784 		hdrsiz = sizeof(struct ip);
1785 		break;
1786 #ifdef INET6
1787 	case AF_INET6:
1788 		hdrsiz = sizeof(struct ip6_hdr);
1789 		break;
1790 #endif
1791 	default:
1792 		hdrsiz = 0;
1793 		break;
1794 	}
1795 	hdrsiz += sizeof(struct tcphdr);
1796 	if (mss > hdrsiz)
1797 		mss -= hdrsiz;
1798 
1799 	mss = max(tcp_mssdflt, mss);
1800 	return (mss);
1801 }
1802 
1803 /*
1804  * Set connection variables based on the peer's advertised MSS.
1805  * We are passed the TCPCB for the actual connection.  If we
1806  * are the server, we are called by the compressed state engine
1807  * when the 3-way handshake is complete.  If we are the client,
1808  * we are called when we receive the SYN,ACK from the server.
1809  *
1810  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1811  * before this routine is called!
1812  */
1813 void
1814 tcp_mss_from_peer(tp, offer)
1815 	struct tcpcb *tp;
1816 	int offer;
1817 {
1818 	struct socket *so;
1819 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1820 	struct rtentry *rt;
1821 #endif
1822 	u_long bufsize;
1823 	int mss;
1824 
1825 #ifdef DIAGNOSTIC
1826 	if (tp->t_inpcb && tp->t_in6pcb)
1827 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1828 #endif
1829 	so = NULL;
1830 	rt = NULL;
1831 #ifdef INET
1832 	if (tp->t_inpcb) {
1833 		so = tp->t_inpcb->inp_socket;
1834 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1835 		rt = in_pcbrtentry(tp->t_inpcb);
1836 #endif
1837 	}
1838 #endif
1839 #ifdef INET6
1840 	if (tp->t_in6pcb) {
1841 		so = tp->t_in6pcb->in6p_socket;
1842 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1843 		rt = in6_pcbrtentry(tp->t_in6pcb);
1844 #endif
1845 	}
1846 #endif
1847 
1848 	/*
1849 	 * As per RFC1122, use the default MSS value, unless they
1850 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1851 	 */
1852 	mss = tcp_mssdflt;
1853 	if (offer)
1854 		mss = offer;
1855 	mss = max(mss, 256);		/* sanity */
1856 	tp->t_peermss = mss;
1857 	mss -= tcp_optlen(tp);
1858 #ifdef INET
1859 	if (tp->t_inpcb)
1860 		mss -= ip_optlen(tp->t_inpcb);
1861 #endif
1862 #ifdef INET6
1863 	if (tp->t_in6pcb)
1864 		mss -= ip6_optlen(tp->t_in6pcb);
1865 #endif
1866 
1867 	/*
1868 	 * If there's a pipesize, change the socket buffer to that size.
1869 	 * Make the socket buffer an integral number of MSS units.  If
1870 	 * the MSS is larger than the socket buffer, artificially decrease
1871 	 * the MSS.
1872 	 */
1873 #ifdef RTV_SPIPE
1874 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1875 		bufsize = rt->rt_rmx.rmx_sendpipe;
1876 	else
1877 #endif
1878 		bufsize = so->so_snd.sb_hiwat;
1879 	if (bufsize < mss)
1880 		mss = bufsize;
1881 	else {
1882 		bufsize = roundup(bufsize, mss);
1883 		if (bufsize > sb_max)
1884 			bufsize = sb_max;
1885 		(void) sbreserve(&so->so_snd, bufsize);
1886 	}
1887 	tp->t_segsz = mss;
1888 
1889 #ifdef RTV_SSTHRESH
1890 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1891 		/*
1892 		 * There's some sort of gateway or interface buffer
1893 		 * limit on the path.  Use this to set the slow
1894 		 * start threshold, but set the threshold to no less
1895 		 * than 2 * MSS.
1896 		 */
1897 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1898 	}
1899 #endif
1900 }
1901 
1902 /*
1903  * Processing necessary when a TCP connection is established.
1904  */
1905 void
1906 tcp_established(tp)
1907 	struct tcpcb *tp;
1908 {
1909 	struct socket *so;
1910 #ifdef RTV_RPIPE
1911 	struct rtentry *rt;
1912 #endif
1913 	u_long bufsize;
1914 
1915 #ifdef DIAGNOSTIC
1916 	if (tp->t_inpcb && tp->t_in6pcb)
1917 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1918 #endif
1919 	so = NULL;
1920 	rt = NULL;
1921 #ifdef INET
1922 	if (tp->t_inpcb) {
1923 		so = tp->t_inpcb->inp_socket;
1924 #if defined(RTV_RPIPE)
1925 		rt = in_pcbrtentry(tp->t_inpcb);
1926 #endif
1927 	}
1928 #endif
1929 #ifdef INET6
1930 	if (tp->t_in6pcb) {
1931 		so = tp->t_in6pcb->in6p_socket;
1932 #if defined(RTV_RPIPE)
1933 		rt = in6_pcbrtentry(tp->t_in6pcb);
1934 #endif
1935 	}
1936 #endif
1937 
1938 	tp->t_state = TCPS_ESTABLISHED;
1939 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1940 
1941 #ifdef RTV_RPIPE
1942 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1943 		bufsize = rt->rt_rmx.rmx_recvpipe;
1944 	else
1945 #endif
1946 		bufsize = so->so_rcv.sb_hiwat;
1947 	if (bufsize > tp->t_ourmss) {
1948 		bufsize = roundup(bufsize, tp->t_ourmss);
1949 		if (bufsize > sb_max)
1950 			bufsize = sb_max;
1951 		(void) sbreserve(&so->so_rcv, bufsize);
1952 	}
1953 }
1954 
1955 /*
1956  * Check if there's an initial rtt or rttvar.  Convert from the
1957  * route-table units to scaled multiples of the slow timeout timer.
1958  * Called only during the 3-way handshake.
1959  */
1960 void
1961 tcp_rmx_rtt(tp)
1962 	struct tcpcb *tp;
1963 {
1964 #ifdef RTV_RTT
1965 	struct rtentry *rt = NULL;
1966 	int rtt;
1967 
1968 #ifdef DIAGNOSTIC
1969 	if (tp->t_inpcb && tp->t_in6pcb)
1970 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1971 #endif
1972 #ifdef INET
1973 	if (tp->t_inpcb)
1974 		rt = in_pcbrtentry(tp->t_inpcb);
1975 #endif
1976 #ifdef INET6
1977 	if (tp->t_in6pcb)
1978 		rt = in6_pcbrtentry(tp->t_in6pcb);
1979 #endif
1980 	if (rt == NULL)
1981 		return;
1982 
1983 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1984 		/*
1985 		 * XXX The lock bit for MTU indicates that the value
1986 		 * is also a minimum value; this is subject to time.
1987 		 */
1988 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1989 			TCPT_RANGESET(tp->t_rttmin,
1990 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1991 			    TCPTV_MIN, TCPTV_REXMTMAX);
1992 		tp->t_srtt = rtt /
1993 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1994 		if (rt->rt_rmx.rmx_rttvar) {
1995 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1996 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1997 				(TCP_RTTVAR_SHIFT + 2));
1998 		} else {
1999 			/* Default variation is +- 1 rtt */
2000 			tp->t_rttvar =
2001 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2002 		}
2003 		TCPT_RANGESET(tp->t_rxtcur,
2004 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2005 		    tp->t_rttmin, TCPTV_REXMTMAX);
2006 	}
2007 #endif
2008 }
2009 
2010 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2011 #if NRND > 0
2012 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2013 #endif
2014 
2015 /*
2016  * Get a new sequence value given a tcp control block
2017  */
2018 tcp_seq
2019 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2020 {
2021 
2022 #ifdef INET
2023 	if (tp->t_inpcb != NULL) {
2024 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2025 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2026 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2027 		    addin));
2028 	}
2029 #endif
2030 #ifdef INET6
2031 	if (tp->t_in6pcb != NULL) {
2032 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2033 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2034 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2035 		    addin));
2036 	}
2037 #endif
2038 	/* Not possible. */
2039 	panic("tcp_new_iss");
2040 }
2041 
2042 /*
2043  * This routine actually generates a new TCP initial sequence number.
2044  */
2045 tcp_seq
2046 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2047     size_t addrsz, tcp_seq addin)
2048 {
2049 	tcp_seq tcp_iss;
2050 
2051 #if NRND > 0
2052 	static int beenhere;
2053 
2054 	/*
2055 	 * If we haven't been here before, initialize our cryptographic
2056 	 * hash secret.
2057 	 */
2058 	if (beenhere == 0) {
2059 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2060 		    RND_EXTRACT_ANY);
2061 		beenhere = 1;
2062 	}
2063 
2064 	if (tcp_do_rfc1948) {
2065 		MD5_CTX ctx;
2066 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2067 
2068 		/*
2069 		 * Compute the base value of the ISS.  It is a hash
2070 		 * of (saddr, sport, daddr, dport, secret).
2071 		 */
2072 		MD5Init(&ctx);
2073 
2074 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2075 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2076 
2077 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2078 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2079 
2080 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2081 
2082 		MD5Final(hash, &ctx);
2083 
2084 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2085 
2086 		/*
2087 		 * Now increment our "timer", and add it in to
2088 		 * the computed value.
2089 		 *
2090 		 * XXX Use `addin'?
2091 		 * XXX TCP_ISSINCR too large to use?
2092 		 */
2093 		tcp_iss_seq += TCP_ISSINCR;
2094 #ifdef TCPISS_DEBUG
2095 		printf("ISS hash 0x%08x, ", tcp_iss);
2096 #endif
2097 		tcp_iss += tcp_iss_seq + addin;
2098 #ifdef TCPISS_DEBUG
2099 		printf("new ISS 0x%08x\n", tcp_iss);
2100 #endif
2101 	} else
2102 #endif /* NRND > 0 */
2103 	{
2104 		/*
2105 		 * Randomize.
2106 		 */
2107 #if NRND > 0
2108 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2109 #else
2110 		tcp_iss = arc4random();
2111 #endif
2112 
2113 		/*
2114 		 * If we were asked to add some amount to a known value,
2115 		 * we will take a random value obtained above, mask off
2116 		 * the upper bits, and add in the known value.  We also
2117 		 * add in a constant to ensure that we are at least a
2118 		 * certain distance from the original value.
2119 		 *
2120 		 * This is used when an old connection is in timed wait
2121 		 * and we have a new one coming in, for instance.
2122 		 */
2123 		if (addin != 0) {
2124 #ifdef TCPISS_DEBUG
2125 			printf("Random %08x, ", tcp_iss);
2126 #endif
2127 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2128 			tcp_iss += addin + TCP_ISSINCR;
2129 #ifdef TCPISS_DEBUG
2130 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2131 #endif
2132 		} else {
2133 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2134 			tcp_iss += tcp_iss_seq;
2135 			tcp_iss_seq += TCP_ISSINCR;
2136 #ifdef TCPISS_DEBUG
2137 			printf("ISS %08x\n", tcp_iss);
2138 #endif
2139 		}
2140 	}
2141 
2142 	if (tcp_compat_42) {
2143 		/*
2144 		 * Limit it to the positive range for really old TCP
2145 		 * implementations.
2146 		 * Just AND off the top bit instead of checking if
2147 		 * is set first - saves a branch 50% of the time.
2148 		 */
2149 		tcp_iss &= 0x7fffffff;		/* XXX */
2150 	}
2151 
2152 	return (tcp_iss);
2153 }
2154 
2155 #if defined(IPSEC) || defined(FAST_IPSEC)
2156 /* compute ESP/AH header size for TCP, including outer IP header. */
2157 size_t
2158 ipsec4_hdrsiz_tcp(tp)
2159 	struct tcpcb *tp;
2160 {
2161 	struct inpcb *inp;
2162 	size_t hdrsiz;
2163 
2164 	/* XXX mapped addr case (tp->t_in6pcb) */
2165 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2166 		return 0;
2167 	switch (tp->t_family) {
2168 	case AF_INET:
2169 		/* XXX: should use currect direction. */
2170 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2171 		break;
2172 	default:
2173 		hdrsiz = 0;
2174 		break;
2175 	}
2176 
2177 	return hdrsiz;
2178 }
2179 
2180 #ifdef INET6
2181 size_t
2182 ipsec6_hdrsiz_tcp(tp)
2183 	struct tcpcb *tp;
2184 {
2185 	struct in6pcb *in6p;
2186 	size_t hdrsiz;
2187 
2188 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2189 		return 0;
2190 	switch (tp->t_family) {
2191 	case AF_INET6:
2192 		/* XXX: should use currect direction. */
2193 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2194 		break;
2195 	case AF_INET:
2196 		/* mapped address case - tricky */
2197 	default:
2198 		hdrsiz = 0;
2199 		break;
2200 	}
2201 
2202 	return hdrsiz;
2203 }
2204 #endif
2205 #endif /*IPSEC*/
2206 
2207 /*
2208  * Determine the length of the TCP options for this connection.
2209  *
2210  * XXX:  What do we do for SACK, when we add that?  Just reserve
2211  *       all of the space?  Otherwise we can't exactly be incrementing
2212  *       cwnd by an amount that varies depending on the amount we last
2213  *       had to SACK!
2214  */
2215 
2216 u_int
2217 tcp_optlen(tp)
2218 	struct tcpcb *tp;
2219 {
2220 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2221 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2222 		return TCPOLEN_TSTAMP_APPA;
2223 	else
2224 		return 0;
2225 }
2226