xref: /netbsd-src/sys/netinet/tcp_subr.c (revision db6316d1518382eecd2fdbe55a1205e0620a1b35)
1 /*	$NetBSD: tcp_subr.c,v 1.174 2004/12/15 04:25:20 thorpej 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 1.174 2004/12/15 04:25:20 thorpej Exp $");
102 
103 #include "opt_inet.h"
104 #include "opt_ipsec.h"
105 #include "opt_tcp_compat_42.h"
106 #include "opt_inet_csum.h"
107 #include "opt_mbuftrace.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include <netinet/tcp.h>
150 #include <netinet/tcp_fsm.h>
151 #include <netinet/tcp_seq.h>
152 #include <netinet/tcp_timer.h>
153 #include <netinet/tcp_var.h>
154 #include <netinet/tcpip.h>
155 
156 #ifdef IPSEC
157 #include <netinet6/ipsec.h>
158 #include <netkey/key.h>
159 #endif /*IPSEC*/
160 
161 #ifdef FAST_IPSEC
162 #include <netipsec/ipsec.h>
163 #include <netipsec/xform.h>
164 #ifdef INET6
165 #include <netipsec/ipsec6.h>
166 #endif
167  #include <netipsec/key.h>
168 #endif	/* FAST_IPSEC*/
169 
170 
171 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
172 struct	tcpstat tcpstat;	/* tcp statistics */
173 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
174 
175 /* patchable/settable parameters for tcp */
176 int 	tcp_mssdflt = TCP_MSS;
177 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
178 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
179 #if NRND > 0
180 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
181 #endif
182 int	tcp_do_sack = 1;	/* selective acknowledgement */
183 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
184 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
185 int	tcp_do_newreno = 1;	/* Use the New Reno algorithms */
186 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
187 #ifndef TCP_INIT_WIN
188 #define	TCP_INIT_WIN	0	/* initial slow start window */
189 #endif
190 #ifndef TCP_INIT_WIN_LOCAL
191 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
192 #endif
193 int	tcp_init_win = TCP_INIT_WIN;
194 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
195 int	tcp_mss_ifmtu = 0;
196 #ifdef TCP_COMPAT_42
197 int	tcp_compat_42 = 1;
198 #else
199 int	tcp_compat_42 = 0;
200 #endif
201 int	tcp_rst_ppslim = 100;	/* 100pps */
202 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
203 int	tcp_do_loopback_cksum = 0;
204 
205 /* tcb hash */
206 #ifndef TCBHASHSIZE
207 #define	TCBHASHSIZE	128
208 #endif
209 int	tcbhashsize = TCBHASHSIZE;
210 
211 /* syn hash parameters */
212 #define	TCP_SYN_HASH_SIZE	293
213 #define	TCP_SYN_BUCKET_SIZE	35
214 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
215 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
216 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
217 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
218 
219 int	tcp_freeq __P((struct tcpcb *));
220 
221 #ifdef INET
222 void	tcp_mtudisc_callback __P((struct in_addr));
223 #endif
224 #ifdef INET6
225 void	tcp6_mtudisc_callback __P((struct in6_addr *));
226 #endif
227 
228 void	tcp_mtudisc __P((struct inpcb *, int));
229 #ifdef INET6
230 void	tcp6_mtudisc __P((struct in6pcb *, int));
231 #endif
232 
233 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL);
234 
235 #ifdef TCP_CSUM_COUNTERS
236 #include <sys/device.h>
237 
238 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
239     NULL, "tcp", "hwcsum bad");
240 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
241     NULL, "tcp", "hwcsum ok");
242 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243     NULL, "tcp", "hwcsum data");
244 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245     NULL, "tcp", "swcsum");
246 
247 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
248 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
249 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
250 EVCNT_ATTACH_STATIC(tcp_swcsum);
251 #endif /* TCP_CSUM_COUNTERS */
252 
253 
254 #ifdef TCP_OUTPUT_COUNTERS
255 #include <sys/device.h>
256 
257 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
258     NULL, "tcp", "output big header");
259 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
260     NULL, "tcp", "output predict hit");
261 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "output predict miss");
263 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "output copy small");
265 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     NULL, "tcp", "output copy big");
267 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     NULL, "tcp", "output reference big");
269 
270 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
271 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
272 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
273 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
274 EVCNT_ATTACH_STATIC(tcp_output_copybig);
275 EVCNT_ATTACH_STATIC(tcp_output_refbig);
276 
277 #endif /* TCP_OUTPUT_COUNTERS */
278 
279 #ifdef TCP_REASS_COUNTERS
280 #include <sys/device.h>
281 
282 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
283     NULL, "tcp_reass", "calls");
284 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
285     &tcp_reass_, "tcp_reass", "insert into empty queue");
286 struct evcnt tcp_reass_iteration[8] = {
287     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
288     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
289     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
290     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
291     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
292     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
293     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
294     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
295 };
296 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
297     &tcp_reass_, "tcp_reass", "prepend to first");
298 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
299     &tcp_reass_, "tcp_reass", "prepend");
300 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
301     &tcp_reass_, "tcp_reass", "insert");
302 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303     &tcp_reass_, "tcp_reass", "insert at tail");
304 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305     &tcp_reass_, "tcp_reass", "append");
306 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
307     &tcp_reass_, "tcp_reass", "append to tail fragment");
308 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
309     &tcp_reass_, "tcp_reass", "overlap at end");
310 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
311     &tcp_reass_, "tcp_reass", "overlap at start");
312 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
313     &tcp_reass_, "tcp_reass", "duplicate segment");
314 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
315     &tcp_reass_, "tcp_reass", "duplicate fragment");
316 
317 EVCNT_ATTACH_STATIC(tcp_reass_);
318 EVCNT_ATTACH_STATIC(tcp_reass_empty);
319 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
320 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
321 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
322 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
323 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
324 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
325 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
326 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
327 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
328 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
329 EVCNT_ATTACH_STATIC(tcp_reass_insert);
330 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
331 EVCNT_ATTACH_STATIC(tcp_reass_append);
332 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
333 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
334 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
335 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
336 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
337 
338 #endif /* TCP_REASS_COUNTERS */
339 
340 #ifdef MBUFTRACE
341 struct mowner tcp_mowner = { "tcp" };
342 struct mowner tcp_rx_mowner = { "tcp", "rx" };
343 struct mowner tcp_tx_mowner = { "tcp", "tx" };
344 #endif
345 
346 /*
347  * Tcp initialization
348  */
349 void
350 tcp_init()
351 {
352 	int hlen;
353 
354 	/* Initialize the TCPCB template. */
355 	tcp_tcpcb_template();
356 
357 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
358 
359 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
360 #ifdef INET6
361 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
362 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
363 #endif
364 	if (max_protohdr < hlen)
365 		max_protohdr = hlen;
366 	if (max_linkhdr + hlen > MHLEN)
367 		panic("tcp_init");
368 
369 #ifdef INET
370 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
371 #endif
372 #ifdef INET6
373 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
374 #endif
375 
376 	/* Initialize timer state. */
377 	tcp_timer_init();
378 
379 	/* Initialize the compressed state engine. */
380 	syn_cache_init();
381 
382 	MOWNER_ATTACH(&tcp_tx_mowner);
383 	MOWNER_ATTACH(&tcp_rx_mowner);
384 	MOWNER_ATTACH(&tcp_mowner);
385 }
386 
387 /*
388  * Create template to be used to send tcp packets on a connection.
389  * Call after host entry created, allocates an mbuf and fills
390  * in a skeletal tcp/ip header, minimizing the amount of work
391  * necessary when the connection is used.
392  */
393 struct mbuf *
394 tcp_template(tp)
395 	struct tcpcb *tp;
396 {
397 	struct inpcb *inp = tp->t_inpcb;
398 #ifdef INET6
399 	struct in6pcb *in6p = tp->t_in6pcb;
400 #endif
401 	struct tcphdr *n;
402 	struct mbuf *m;
403 	int hlen;
404 
405 	switch (tp->t_family) {
406 	case AF_INET:
407 		hlen = sizeof(struct ip);
408 		if (inp)
409 			break;
410 #ifdef INET6
411 		if (in6p) {
412 			/* mapped addr case */
413 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
414 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
415 				break;
416 		}
417 #endif
418 		return NULL;	/*EINVAL*/
419 #ifdef INET6
420 	case AF_INET6:
421 		hlen = sizeof(struct ip6_hdr);
422 		if (in6p) {
423 			/* more sainty check? */
424 			break;
425 		}
426 		return NULL;	/*EINVAL*/
427 #endif
428 	default:
429 		hlen = 0;	/*pacify gcc*/
430 		return NULL;	/*EAFNOSUPPORT*/
431 	}
432 #ifdef DIAGNOSTIC
433 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
434 		panic("mclbytes too small for t_template");
435 #endif
436 	m = tp->t_template;
437 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
438 		;
439 	else {
440 		if (m)
441 			m_freem(m);
442 		m = tp->t_template = NULL;
443 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
444 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
445 			MCLGET(m, M_DONTWAIT);
446 			if ((m->m_flags & M_EXT) == 0) {
447 				m_free(m);
448 				m = NULL;
449 			}
450 		}
451 		if (m == NULL)
452 			return NULL;
453 		MCLAIM(m, &tcp_mowner);
454 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
455 	}
456 
457 	bzero(mtod(m, caddr_t), m->m_len);
458 
459 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
460 
461 	switch (tp->t_family) {
462 	case AF_INET:
463 	    {
464 		struct ipovly *ipov;
465 		mtod(m, struct ip *)->ip_v = 4;
466 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
467 		ipov = mtod(m, struct ipovly *);
468 		ipov->ih_pr = IPPROTO_TCP;
469 		ipov->ih_len = htons(sizeof(struct tcphdr));
470 		if (inp) {
471 			ipov->ih_src = inp->inp_laddr;
472 			ipov->ih_dst = inp->inp_faddr;
473 		}
474 #ifdef INET6
475 		else if (in6p) {
476 			/* mapped addr case */
477 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
478 				sizeof(ipov->ih_src));
479 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
480 				sizeof(ipov->ih_dst));
481 		}
482 #endif
483 		/*
484 		 * Compute the pseudo-header portion of the checksum
485 		 * now.  We incrementally add in the TCP option and
486 		 * payload lengths later, and then compute the TCP
487 		 * checksum right before the packet is sent off onto
488 		 * the wire.
489 		 */
490 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
491 		    ipov->ih_dst.s_addr,
492 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
493 		break;
494 	    }
495 #ifdef INET6
496 	case AF_INET6:
497 	    {
498 		struct ip6_hdr *ip6;
499 		mtod(m, struct ip *)->ip_v = 6;
500 		ip6 = mtod(m, struct ip6_hdr *);
501 		ip6->ip6_nxt = IPPROTO_TCP;
502 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
503 		ip6->ip6_src = in6p->in6p_laddr;
504 		ip6->ip6_dst = in6p->in6p_faddr;
505 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
506 		if (ip6_auto_flowlabel) {
507 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
508 			ip6->ip6_flow |=
509 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
510 		}
511 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
512 		ip6->ip6_vfc |= IPV6_VERSION;
513 
514 		/*
515 		 * Compute the pseudo-header portion of the checksum
516 		 * now.  We incrementally add in the TCP option and
517 		 * payload lengths later, and then compute the TCP
518 		 * checksum right before the packet is sent off onto
519 		 * the wire.
520 		 */
521 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
522 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
523 		    htonl(IPPROTO_TCP));
524 		break;
525 	    }
526 #endif
527 	}
528 	if (inp) {
529 		n->th_sport = inp->inp_lport;
530 		n->th_dport = inp->inp_fport;
531 	}
532 #ifdef INET6
533 	else if (in6p) {
534 		n->th_sport = in6p->in6p_lport;
535 		n->th_dport = in6p->in6p_fport;
536 	}
537 #endif
538 	n->th_seq = 0;
539 	n->th_ack = 0;
540 	n->th_x2 = 0;
541 	n->th_off = 5;
542 	n->th_flags = 0;
543 	n->th_win = 0;
544 	n->th_urp = 0;
545 	return (m);
546 }
547 
548 /*
549  * Send a single message to the TCP at address specified by
550  * the given TCP/IP header.  If m == 0, then we make a copy
551  * of the tcpiphdr at ti and send directly to the addressed host.
552  * This is used to force keep alive messages out using the TCP
553  * template for a connection tp->t_template.  If flags are given
554  * then we send a message back to the TCP which originated the
555  * segment ti, and discard the mbuf containing it and any other
556  * attached mbufs.
557  *
558  * In any case the ack and sequence number of the transmitted
559  * segment are as specified by the parameters.
560  */
561 int
562 tcp_respond(tp, template, m, th0, ack, seq, flags)
563 	struct tcpcb *tp;
564 	struct mbuf *template;
565 	struct mbuf *m;
566 	struct tcphdr *th0;
567 	tcp_seq ack, seq;
568 	int flags;
569 {
570 	struct route *ro;
571 	int error, tlen, win = 0;
572 	int hlen;
573 	struct ip *ip;
574 #ifdef INET6
575 	struct ip6_hdr *ip6;
576 #endif
577 	int family;	/* family on packet, not inpcb/in6pcb! */
578 	struct tcphdr *th;
579 	struct socket *so;
580 
581 	if (tp != NULL && (flags & TH_RST) == 0) {
582 #ifdef DIAGNOSTIC
583 		if (tp->t_inpcb && tp->t_in6pcb)
584 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
585 #endif
586 #ifdef INET
587 		if (tp->t_inpcb)
588 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
589 #endif
590 #ifdef INET6
591 		if (tp->t_in6pcb)
592 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
593 #endif
594 	}
595 
596 	th = NULL;	/* Quell uninitialized warning */
597 	ip = NULL;
598 #ifdef INET6
599 	ip6 = NULL;
600 #endif
601 	if (m == 0) {
602 		if (!template)
603 			return EINVAL;
604 
605 		/* get family information from template */
606 		switch (mtod(template, struct ip *)->ip_v) {
607 		case 4:
608 			family = AF_INET;
609 			hlen = sizeof(struct ip);
610 			break;
611 #ifdef INET6
612 		case 6:
613 			family = AF_INET6;
614 			hlen = sizeof(struct ip6_hdr);
615 			break;
616 #endif
617 		default:
618 			return EAFNOSUPPORT;
619 		}
620 
621 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
622 		if (m) {
623 			MCLAIM(m, &tcp_tx_mowner);
624 			MCLGET(m, M_DONTWAIT);
625 			if ((m->m_flags & M_EXT) == 0) {
626 				m_free(m);
627 				m = NULL;
628 			}
629 		}
630 		if (m == NULL)
631 			return (ENOBUFS);
632 
633 		if (tcp_compat_42)
634 			tlen = 1;
635 		else
636 			tlen = 0;
637 
638 		m->m_data += max_linkhdr;
639 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
640 			template->m_len);
641 		switch (family) {
642 		case AF_INET:
643 			ip = mtod(m, struct ip *);
644 			th = (struct tcphdr *)(ip + 1);
645 			break;
646 #ifdef INET6
647 		case AF_INET6:
648 			ip6 = mtod(m, struct ip6_hdr *);
649 			th = (struct tcphdr *)(ip6 + 1);
650 			break;
651 #endif
652 #if 0
653 		default:
654 			/* noone will visit here */
655 			m_freem(m);
656 			return EAFNOSUPPORT;
657 #endif
658 		}
659 		flags = TH_ACK;
660 	} else {
661 
662 		if ((m->m_flags & M_PKTHDR) == 0) {
663 #if 0
664 			printf("non PKTHDR to tcp_respond\n");
665 #endif
666 			m_freem(m);
667 			return EINVAL;
668 		}
669 #ifdef DIAGNOSTIC
670 		if (!th0)
671 			panic("th0 == NULL in tcp_respond");
672 #endif
673 
674 		/* get family information from m */
675 		switch (mtod(m, struct ip *)->ip_v) {
676 		case 4:
677 			family = AF_INET;
678 			hlen = sizeof(struct ip);
679 			ip = mtod(m, struct ip *);
680 			break;
681 #ifdef INET6
682 		case 6:
683 			family = AF_INET6;
684 			hlen = sizeof(struct ip6_hdr);
685 			ip6 = mtod(m, struct ip6_hdr *);
686 			break;
687 #endif
688 		default:
689 			m_freem(m);
690 			return EAFNOSUPPORT;
691 		}
692 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
693 			tlen = sizeof(*th0);
694 		else
695 			tlen = th0->th_off << 2;
696 
697 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
698 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
699 			m->m_len = hlen + tlen;
700 			m_freem(m->m_next);
701 			m->m_next = NULL;
702 		} else {
703 			struct mbuf *n;
704 
705 #ifdef DIAGNOSTIC
706 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
707 				m_freem(m);
708 				return EMSGSIZE;
709 			}
710 #endif
711 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
712 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
713 				MCLGET(n, M_DONTWAIT);
714 				if ((n->m_flags & M_EXT) == 0) {
715 					m_freem(n);
716 					n = NULL;
717 				}
718 			}
719 			if (!n) {
720 				m_freem(m);
721 				return ENOBUFS;
722 			}
723 
724 			MCLAIM(n, &tcp_tx_mowner);
725 			n->m_data += max_linkhdr;
726 			n->m_len = hlen + tlen;
727 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
728 			m_copyback(n, hlen, tlen, (caddr_t)th0);
729 
730 			m_freem(m);
731 			m = n;
732 			n = NULL;
733 		}
734 
735 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
736 		switch (family) {
737 		case AF_INET:
738 			ip = mtod(m, struct ip *);
739 			th = (struct tcphdr *)(ip + 1);
740 			ip->ip_p = IPPROTO_TCP;
741 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
742 			ip->ip_p = IPPROTO_TCP;
743 			break;
744 #ifdef INET6
745 		case AF_INET6:
746 			ip6 = mtod(m, struct ip6_hdr *);
747 			th = (struct tcphdr *)(ip6 + 1);
748 			ip6->ip6_nxt = IPPROTO_TCP;
749 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
750 			ip6->ip6_nxt = IPPROTO_TCP;
751 			break;
752 #endif
753 #if 0
754 		default:
755 			/* noone will visit here */
756 			m_freem(m);
757 			return EAFNOSUPPORT;
758 #endif
759 		}
760 		xchg(th->th_dport, th->th_sport, u_int16_t);
761 #undef xchg
762 		tlen = 0;	/*be friendly with the following code*/
763 	}
764 	th->th_seq = htonl(seq);
765 	th->th_ack = htonl(ack);
766 	th->th_x2 = 0;
767 	if ((flags & TH_SYN) == 0) {
768 		if (tp)
769 			win >>= tp->rcv_scale;
770 		if (win > TCP_MAXWIN)
771 			win = TCP_MAXWIN;
772 		th->th_win = htons((u_int16_t)win);
773 		th->th_off = sizeof (struct tcphdr) >> 2;
774 		tlen += sizeof(*th);
775 	} else
776 		tlen += th->th_off << 2;
777 	m->m_len = hlen + tlen;
778 	m->m_pkthdr.len = hlen + tlen;
779 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
780 	th->th_flags = flags;
781 	th->th_urp = 0;
782 
783 	switch (family) {
784 #ifdef INET
785 	case AF_INET:
786 	    {
787 		struct ipovly *ipov = (struct ipovly *)ip;
788 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
789 		ipov->ih_len = htons((u_int16_t)tlen);
790 
791 		th->th_sum = 0;
792 		th->th_sum = in_cksum(m, hlen + tlen);
793 		ip->ip_len = htons(hlen + tlen);
794 		ip->ip_ttl = ip_defttl;
795 		break;
796 	    }
797 #endif
798 #ifdef INET6
799 	case AF_INET6:
800 	    {
801 		th->th_sum = 0;
802 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
803 				tlen);
804 		ip6->ip6_plen = ntohs(tlen);
805 		if (tp && tp->t_in6pcb) {
806 			struct ifnet *oifp;
807 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
808 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
809 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
810 		} else
811 			ip6->ip6_hlim = ip6_defhlim;
812 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
813 		if (ip6_auto_flowlabel) {
814 			ip6->ip6_flow |=
815 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
816 		}
817 		break;
818 	    }
819 #endif
820 	}
821 
822 	if (tp && tp->t_inpcb)
823 		so = tp->t_inpcb->inp_socket;
824 #ifdef INET6
825 	else if (tp && tp->t_in6pcb)
826 		so = tp->t_in6pcb->in6p_socket;
827 #endif
828 	else
829 		so = NULL;
830 
831 	if (tp != NULL && tp->t_inpcb != NULL) {
832 		ro = &tp->t_inpcb->inp_route;
833 #ifdef DIAGNOSTIC
834 		if (family != AF_INET)
835 			panic("tcp_respond: address family mismatch");
836 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
837 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
838 			    ntohl(ip->ip_dst.s_addr),
839 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
840 		}
841 #endif
842 	}
843 #ifdef INET6
844 	else if (tp != NULL && tp->t_in6pcb != NULL) {
845 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
846 #ifdef DIAGNOSTIC
847 		if (family == AF_INET) {
848 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
849 				panic("tcp_respond: not mapped addr");
850 			if (bcmp(&ip->ip_dst,
851 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
852 			    sizeof(ip->ip_dst)) != 0) {
853 				panic("tcp_respond: ip_dst != in6p_faddr");
854 			}
855 		} else if (family == AF_INET6) {
856 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
857 			    &tp->t_in6pcb->in6p_faddr))
858 				panic("tcp_respond: ip6_dst != in6p_faddr");
859 		} else
860 			panic("tcp_respond: address family mismatch");
861 #endif
862 	}
863 #endif
864 	else
865 		ro = NULL;
866 
867 	switch (family) {
868 #ifdef INET
869 	case AF_INET:
870 		error = ip_output(m, NULL, ro,
871 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
872 		    (struct ip_moptions *)0, so);
873 		break;
874 #endif
875 #ifdef INET6
876 	case AF_INET6:
877 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
878 		    (struct ip6_moptions *)0, so, NULL);
879 		break;
880 #endif
881 	default:
882 		error = EAFNOSUPPORT;
883 		break;
884 	}
885 
886 	return (error);
887 }
888 
889 /*
890  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
891  * a bunch of members individually, we maintain this template for the
892  * static and mostly-static components of the TCPCB, and copy it into
893  * the new TCPCB instead.
894  */
895 static struct tcpcb tcpcb_template = {
896 	/*
897 	 * If TCP_NTIMERS ever changes, we'll need to update this
898 	 * initializer.
899 	 */
900 	.t_timer = {
901 		CALLOUT_INITIALIZER,
902 		CALLOUT_INITIALIZER,
903 		CALLOUT_INITIALIZER,
904 		CALLOUT_INITIALIZER,
905 	},
906 	.t_delack_ch = CALLOUT_INITIALIZER,
907 
908 	.t_srtt = TCPTV_SRTTBASE,
909 	.t_rttmin = TCPTV_MIN,
910 
911 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
912 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
913 };
914 
915 /*
916  * Updates the TCPCB template whenever a parameter that would affect
917  * the template is changed.
918  */
919 void
920 tcp_tcpcb_template(void)
921 {
922 	struct tcpcb *tp = &tcpcb_template;
923 	int flags;
924 
925 	tp->t_peermss = tcp_mssdflt;
926 	tp->t_ourmss = tcp_mssdflt;
927 	tp->t_segsz = tcp_mssdflt;
928 
929 	flags = 0;
930 	if (tcp_do_rfc1323 && tcp_do_win_scale)
931 		flags |= TF_REQ_SCALE;
932 	if (tcp_do_rfc1323 && tcp_do_timestamps)
933 		flags |= TF_REQ_TSTMP;
934 	if (tcp_do_sack == 2)
935 		flags |= TF_WILL_SACK;
936 	else if (tcp_do_sack == 1)
937 		flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
938 	flags |= TF_CANT_TXSACK;
939 	tp->t_flags = flags;
940 
941 	/*
942 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
943 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
944 	 * reasonable initial retransmit time.
945 	 */
946 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
947 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
948 	    TCPTV_MIN, TCPTV_REXMTMAX);
949 }
950 
951 /*
952  * Create a new TCP control block, making an
953  * empty reassembly queue and hooking it to the argument
954  * protocol control block.
955  */
956 struct tcpcb *
957 tcp_newtcpcb(family, aux)
958 	int family;	/* selects inpcb, or in6pcb */
959 	void *aux;
960 {
961 	struct tcpcb *tp;
962 	int i;
963 
964 	/* XXX Consider using a pool_cache for speed. */
965 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
966 	if (tp == NULL)
967 		return (NULL);
968 	memcpy(tp, &tcpcb_template, sizeof(*tp));
969 	TAILQ_INIT(&tp->segq);
970 	TAILQ_INIT(&tp->timeq);
971 	tp->t_family = family;		/* may be overridden later on */
972 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
973 
974 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
975 	for (i = 0; i < TCPT_NTIMERS; i++)
976 		TCP_TIMER_INIT(tp, i);
977 
978 	switch (family) {
979 	case AF_INET:
980 	    {
981 		struct inpcb *inp = (struct inpcb *)aux;
982 
983 		inp->inp_ip.ip_ttl = ip_defttl;
984 		inp->inp_ppcb = (caddr_t)tp;
985 
986 		tp->t_inpcb = inp;
987 		tp->t_mtudisc = ip_mtudisc;
988 		break;
989 	    }
990 #ifdef INET6
991 	case AF_INET6:
992 	    {
993 		struct in6pcb *in6p = (struct in6pcb *)aux;
994 
995 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
996 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
997 					       : NULL);
998 		in6p->in6p_ppcb = (caddr_t)tp;
999 
1000 		tp->t_in6pcb = in6p;
1001 		/* for IPv6, always try to run path MTU discovery */
1002 		tp->t_mtudisc = 1;
1003 		break;
1004 	    }
1005 #endif /* INET6 */
1006 	default:
1007 		pool_put(&tcpcb_pool, tp);
1008 		return (NULL);
1009 	}
1010 
1011 	/*
1012 	 * Initialize our timebase.  When we send timestamps, we take
1013 	 * the delta from tcp_now -- this means each connection always
1014 	 * gets a timebase of 0, which makes it, among other things,
1015 	 * more difficult to determine how long a system has been up,
1016 	 * and thus how many TCP sequence increments have occurred.
1017 	 */
1018 	tp->ts_timebase = tcp_now;
1019 
1020 	return (tp);
1021 }
1022 
1023 /*
1024  * Drop a TCP connection, reporting
1025  * the specified error.  If connection is synchronized,
1026  * then send a RST to peer.
1027  */
1028 struct tcpcb *
1029 tcp_drop(tp, errno)
1030 	struct tcpcb *tp;
1031 	int errno;
1032 {
1033 	struct socket *so = NULL;
1034 
1035 #ifdef DIAGNOSTIC
1036 	if (tp->t_inpcb && tp->t_in6pcb)
1037 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1038 #endif
1039 #ifdef INET
1040 	if (tp->t_inpcb)
1041 		so = tp->t_inpcb->inp_socket;
1042 #endif
1043 #ifdef INET6
1044 	if (tp->t_in6pcb)
1045 		so = tp->t_in6pcb->in6p_socket;
1046 #endif
1047 	if (!so)
1048 		return NULL;
1049 
1050 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1051 		tp->t_state = TCPS_CLOSED;
1052 		(void) tcp_output(tp);
1053 		tcpstat.tcps_drops++;
1054 	} else
1055 		tcpstat.tcps_conndrops++;
1056 	if (errno == ETIMEDOUT && tp->t_softerror)
1057 		errno = tp->t_softerror;
1058 	so->so_error = errno;
1059 	return (tcp_close(tp));
1060 }
1061 
1062 /*
1063  * Return whether this tcpcb is marked as dead, indicating
1064  * to the calling timer function that no further action should
1065  * be taken, as we are about to release this tcpcb.  The release
1066  * of the storage will be done if this is the last timer running.
1067  *
1068  * This should be called from the callout handler function after
1069  * callout_ack() is done, so that the number of invoking timer
1070  * functions is 0.
1071  */
1072 int
1073 tcp_isdead(tp)
1074 	struct tcpcb *tp;
1075 {
1076 	int dead = (tp->t_flags & TF_DEAD);
1077 
1078 	if (__predict_false(dead)) {
1079 		if (tcp_timers_invoking(tp) > 0)
1080 				/* not quite there yet -- count separately? */
1081 			return dead;
1082 		tcpstat.tcps_delayed_free++;
1083 		pool_put(&tcpcb_pool, tp);
1084 	}
1085 	return dead;
1086 }
1087 
1088 /*
1089  * Close a TCP control block:
1090  *	discard all space held by the tcp
1091  *	discard internet protocol block
1092  *	wake up any sleepers
1093  */
1094 struct tcpcb *
1095 tcp_close(tp)
1096 	struct tcpcb *tp;
1097 {
1098 	struct inpcb *inp;
1099 #ifdef INET6
1100 	struct in6pcb *in6p;
1101 #endif
1102 	struct socket *so;
1103 #ifdef RTV_RTT
1104 	struct rtentry *rt;
1105 #endif
1106 	struct route *ro;
1107 
1108 	inp = tp->t_inpcb;
1109 #ifdef INET6
1110 	in6p = tp->t_in6pcb;
1111 #endif
1112 	so = NULL;
1113 	ro = NULL;
1114 	if (inp) {
1115 		so = inp->inp_socket;
1116 		ro = &inp->inp_route;
1117 	}
1118 #ifdef INET6
1119 	else if (in6p) {
1120 		so = in6p->in6p_socket;
1121 		ro = (struct route *)&in6p->in6p_route;
1122 	}
1123 #endif
1124 
1125 #ifdef RTV_RTT
1126 	/*
1127 	 * If we sent enough data to get some meaningful characteristics,
1128 	 * save them in the routing entry.  'Enough' is arbitrarily
1129 	 * defined as the sendpipesize (default 4K) * 16.  This would
1130 	 * give us 16 rtt samples assuming we only get one sample per
1131 	 * window (the usual case on a long haul net).  16 samples is
1132 	 * enough for the srtt filter to converge to within 5% of the correct
1133 	 * value; fewer samples and we could save a very bogus rtt.
1134 	 *
1135 	 * Don't update the default route's characteristics and don't
1136 	 * update anything that the user "locked".
1137 	 */
1138 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1139 	    ro && (rt = ro->ro_rt) &&
1140 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1141 		u_long i = 0;
1142 
1143 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1144 			i = tp->t_srtt *
1145 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1146 			if (rt->rt_rmx.rmx_rtt && i)
1147 				/*
1148 				 * filter this update to half the old & half
1149 				 * the new values, converting scale.
1150 				 * See route.h and tcp_var.h for a
1151 				 * description of the scaling constants.
1152 				 */
1153 				rt->rt_rmx.rmx_rtt =
1154 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1155 			else
1156 				rt->rt_rmx.rmx_rtt = i;
1157 		}
1158 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1159 			i = tp->t_rttvar *
1160 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1161 			if (rt->rt_rmx.rmx_rttvar && i)
1162 				rt->rt_rmx.rmx_rttvar =
1163 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1164 			else
1165 				rt->rt_rmx.rmx_rttvar = i;
1166 		}
1167 		/*
1168 		 * update the pipelimit (ssthresh) if it has been updated
1169 		 * already or if a pipesize was specified & the threshhold
1170 		 * got below half the pipesize.  I.e., wait for bad news
1171 		 * before we start updating, then update on both good
1172 		 * and bad news.
1173 		 */
1174 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1175 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1176 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1177 			/*
1178 			 * convert the limit from user data bytes to
1179 			 * packets then to packet data bytes.
1180 			 */
1181 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1182 			if (i < 2)
1183 				i = 2;
1184 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1185 			if (rt->rt_rmx.rmx_ssthresh)
1186 				rt->rt_rmx.rmx_ssthresh =
1187 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1188 			else
1189 				rt->rt_rmx.rmx_ssthresh = i;
1190 		}
1191 	}
1192 #endif /* RTV_RTT */
1193 	/* free the reassembly queue, if any */
1194 	TCP_REASS_LOCK(tp);
1195 	(void) tcp_freeq(tp);
1196 	TCP_REASS_UNLOCK(tp);
1197 
1198 	tcp_canceltimers(tp);
1199 	TCP_CLEAR_DELACK(tp);
1200 	syn_cache_cleanup(tp);
1201 
1202 	if (tp->t_template) {
1203 		m_free(tp->t_template);
1204 		tp->t_template = NULL;
1205 	}
1206 	if (tcp_timers_invoking(tp))
1207 		tp->t_flags |= TF_DEAD;
1208 	else
1209 		pool_put(&tcpcb_pool, tp);
1210 
1211 	if (inp) {
1212 		inp->inp_ppcb = 0;
1213 		soisdisconnected(so);
1214 		in_pcbdetach(inp);
1215 	}
1216 #ifdef INET6
1217 	else if (in6p) {
1218 		in6p->in6p_ppcb = 0;
1219 		soisdisconnected(so);
1220 		in6_pcbdetach(in6p);
1221 	}
1222 #endif
1223 	tcpstat.tcps_closed++;
1224 	return ((struct tcpcb *)0);
1225 }
1226 
1227 int
1228 tcp_freeq(tp)
1229 	struct tcpcb *tp;
1230 {
1231 	struct ipqent *qe;
1232 	int rv = 0;
1233 #ifdef TCPREASS_DEBUG
1234 	int i = 0;
1235 #endif
1236 
1237 	TCP_REASS_LOCK_CHECK(tp);
1238 
1239 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1240 #ifdef TCPREASS_DEBUG
1241 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1242 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1243 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1244 #endif
1245 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1246 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1247 		m_freem(qe->ipqe_m);
1248 		pool_put(&tcpipqent_pool, qe);
1249 		rv = 1;
1250 	}
1251 	return (rv);
1252 }
1253 
1254 /*
1255  * Protocol drain routine.  Called when memory is in short supply.
1256  */
1257 void
1258 tcp_drain()
1259 {
1260 	struct inpcb_hdr *inph;
1261 	struct tcpcb *tp;
1262 
1263 	/*
1264 	 * Free the sequence queue of all TCP connections.
1265 	 */
1266 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1267 		switch (inph->inph_af) {
1268 		case AF_INET:
1269 			tp = intotcpcb((struct inpcb *)inph);
1270 			break;
1271 #ifdef INET6
1272 		case AF_INET6:
1273 			tp = in6totcpcb((struct in6pcb *)inph);
1274 			break;
1275 #endif
1276 		default:
1277 			tp = NULL;
1278 			break;
1279 		}
1280 		if (tp != NULL) {
1281 			/*
1282 			 * We may be called from a device's interrupt
1283 			 * context.  If the tcpcb is already busy,
1284 			 * just bail out now.
1285 			 */
1286 			if (tcp_reass_lock_try(tp) == 0)
1287 				continue;
1288 			if (tcp_freeq(tp))
1289 				tcpstat.tcps_connsdrained++;
1290 			TCP_REASS_UNLOCK(tp);
1291 		}
1292 	}
1293 }
1294 
1295 /*
1296  * Notify a tcp user of an asynchronous error;
1297  * store error as soft error, but wake up user
1298  * (for now, won't do anything until can select for soft error).
1299  */
1300 void
1301 tcp_notify(inp, error)
1302 	struct inpcb *inp;
1303 	int error;
1304 {
1305 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1306 	struct socket *so = inp->inp_socket;
1307 
1308 	/*
1309 	 * Ignore some errors if we are hooked up.
1310 	 * If connection hasn't completed, has retransmitted several times,
1311 	 * and receives a second error, give up now.  This is better
1312 	 * than waiting a long time to establish a connection that
1313 	 * can never complete.
1314 	 */
1315 	if (tp->t_state == TCPS_ESTABLISHED &&
1316 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1317 	      error == EHOSTDOWN)) {
1318 		return;
1319 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1320 	    tp->t_rxtshift > 3 && tp->t_softerror)
1321 		so->so_error = error;
1322 	else
1323 		tp->t_softerror = error;
1324 	wakeup((caddr_t) &so->so_timeo);
1325 	sorwakeup(so);
1326 	sowwakeup(so);
1327 }
1328 
1329 #ifdef INET6
1330 void
1331 tcp6_notify(in6p, error)
1332 	struct in6pcb *in6p;
1333 	int error;
1334 {
1335 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1336 	struct socket *so = in6p->in6p_socket;
1337 
1338 	/*
1339 	 * Ignore some errors if we are hooked up.
1340 	 * If connection hasn't completed, has retransmitted several times,
1341 	 * and receives a second error, give up now.  This is better
1342 	 * than waiting a long time to establish a connection that
1343 	 * can never complete.
1344 	 */
1345 	if (tp->t_state == TCPS_ESTABLISHED &&
1346 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1347 	      error == EHOSTDOWN)) {
1348 		return;
1349 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1350 	    tp->t_rxtshift > 3 && tp->t_softerror)
1351 		so->so_error = error;
1352 	else
1353 		tp->t_softerror = error;
1354 	wakeup((caddr_t) &so->so_timeo);
1355 	sorwakeup(so);
1356 	sowwakeup(so);
1357 }
1358 #endif
1359 
1360 #ifdef INET6
1361 void
1362 tcp6_ctlinput(cmd, sa, d)
1363 	int cmd;
1364 	struct sockaddr *sa;
1365 	void *d;
1366 {
1367 	struct tcphdr th;
1368 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1369 	int nmatch;
1370 	struct ip6_hdr *ip6;
1371 	const struct sockaddr_in6 *sa6_src = NULL;
1372 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1373 	struct mbuf *m;
1374 	int off;
1375 
1376 	if (sa->sa_family != AF_INET6 ||
1377 	    sa->sa_len != sizeof(struct sockaddr_in6))
1378 		return;
1379 	if ((unsigned)cmd >= PRC_NCMDS)
1380 		return;
1381 	else if (cmd == PRC_QUENCH) {
1382 		/* XXX there's no PRC_QUENCH in IPv6 */
1383 		notify = tcp6_quench;
1384 	} else if (PRC_IS_REDIRECT(cmd))
1385 		notify = in6_rtchange, d = NULL;
1386 	else if (cmd == PRC_MSGSIZE)
1387 		; /* special code is present, see below */
1388 	else if (cmd == PRC_HOSTDEAD)
1389 		d = NULL;
1390 	else if (inet6ctlerrmap[cmd] == 0)
1391 		return;
1392 
1393 	/* if the parameter is from icmp6, decode it. */
1394 	if (d != NULL) {
1395 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1396 		m = ip6cp->ip6c_m;
1397 		ip6 = ip6cp->ip6c_ip6;
1398 		off = ip6cp->ip6c_off;
1399 		sa6_src = ip6cp->ip6c_src;
1400 	} else {
1401 		m = NULL;
1402 		ip6 = NULL;
1403 		sa6_src = &sa6_any;
1404 		off = 0;
1405 	}
1406 
1407 	if (ip6) {
1408 		/*
1409 		 * XXX: We assume that when ip6 is non NULL,
1410 		 * M and OFF are valid.
1411 		 */
1412 
1413 		/* check if we can safely examine src and dst ports */
1414 		if (m->m_pkthdr.len < off + sizeof(th)) {
1415 			if (cmd == PRC_MSGSIZE)
1416 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1417 			return;
1418 		}
1419 
1420 		bzero(&th, sizeof(th));
1421 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1422 
1423 		if (cmd == PRC_MSGSIZE) {
1424 			int valid = 0;
1425 
1426 			/*
1427 			 * Check to see if we have a valid TCP connection
1428 			 * corresponding to the address in the ICMPv6 message
1429 			 * payload.
1430 			 */
1431 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1432 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1433 			    th.th_sport, 0))
1434 				valid++;
1435 
1436 			/*
1437 			 * Depending on the value of "valid" and routing table
1438 			 * size (mtudisc_{hi,lo}wat), we will:
1439 			 * - recalcurate the new MTU and create the
1440 			 *   corresponding routing entry, or
1441 			 * - ignore the MTU change notification.
1442 			 */
1443 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1444 
1445 			/*
1446 			 * no need to call in6_pcbnotify, it should have been
1447 			 * called via callback if necessary
1448 			 */
1449 			return;
1450 		}
1451 
1452 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1453 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1454 		if (nmatch == 0 && syn_cache_count &&
1455 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1456 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1457 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1458 			syn_cache_unreach((struct sockaddr *)sa6_src,
1459 					  sa, &th);
1460 	} else {
1461 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1462 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1463 	}
1464 }
1465 #endif
1466 
1467 #ifdef INET
1468 /* assumes that ip header and tcp header are contiguous on mbuf */
1469 void *
1470 tcp_ctlinput(cmd, sa, v)
1471 	int cmd;
1472 	struct sockaddr *sa;
1473 	void *v;
1474 {
1475 	struct ip *ip = v;
1476 	struct tcphdr *th;
1477 	struct icmp *icp;
1478 	extern const int inetctlerrmap[];
1479 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1480 	int errno;
1481 	int nmatch;
1482 #ifdef INET6
1483 	struct in6_addr src6, dst6;
1484 #endif
1485 
1486 	if (sa->sa_family != AF_INET ||
1487 	    sa->sa_len != sizeof(struct sockaddr_in))
1488 		return NULL;
1489 	if ((unsigned)cmd >= PRC_NCMDS)
1490 		return NULL;
1491 	errno = inetctlerrmap[cmd];
1492 	if (cmd == PRC_QUENCH)
1493 		notify = tcp_quench;
1494 	else if (PRC_IS_REDIRECT(cmd))
1495 		notify = in_rtchange, ip = 0;
1496 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1497 		/*
1498 		 * Check to see if we have a valid TCP connection
1499 		 * corresponding to the address in the ICMP message
1500 		 * payload.
1501 		 *
1502 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1503 		 */
1504 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1505 #ifdef INET6
1506 		memset(&src6, 0, sizeof(src6));
1507 		memset(&dst6, 0, sizeof(dst6));
1508 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1509 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1510 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1511 #endif
1512 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1513 		    ip->ip_src, th->th_sport) != NULL)
1514 			;
1515 #ifdef INET6
1516 		else if (in6_pcblookup_connect(&tcbtable, &dst6,
1517 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1518 			;
1519 #endif
1520 		else
1521 			return NULL;
1522 
1523 		/*
1524 		 * Now that we've validated that we are actually communicating
1525 		 * with the host indicated in the ICMP message, locate the
1526 		 * ICMP header, recalculate the new MTU, and create the
1527 		 * corresponding routing entry.
1528 		 */
1529 		icp = (struct icmp *)((caddr_t)ip -
1530 		    offsetof(struct icmp, icmp_ip));
1531 		icmp_mtudisc(icp, ip->ip_dst);
1532 
1533 		return NULL;
1534 	} else if (cmd == PRC_HOSTDEAD)
1535 		ip = 0;
1536 	else if (errno == 0)
1537 		return NULL;
1538 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1539 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1540 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1541 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1542 		if (nmatch == 0 && syn_cache_count &&
1543 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1544 		    inetctlerrmap[cmd] == ENETUNREACH ||
1545 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1546 			struct sockaddr_in sin;
1547 			bzero(&sin, sizeof(sin));
1548 			sin.sin_len = sizeof(sin);
1549 			sin.sin_family = AF_INET;
1550 			sin.sin_port = th->th_sport;
1551 			sin.sin_addr = ip->ip_src;
1552 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1553 		}
1554 
1555 		/* XXX mapped address case */
1556 	} else
1557 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1558 		    notify);
1559 	return NULL;
1560 }
1561 
1562 /*
1563  * When a source quence is received, we are being notifed of congestion.
1564  * Close the congestion window down to the Loss Window (one segment).
1565  * We will gradually open it again as we proceed.
1566  */
1567 void
1568 tcp_quench(inp, errno)
1569 	struct inpcb *inp;
1570 	int errno;
1571 {
1572 	struct tcpcb *tp = intotcpcb(inp);
1573 
1574 	if (tp)
1575 		tp->snd_cwnd = tp->t_segsz;
1576 }
1577 #endif
1578 
1579 #ifdef INET6
1580 void
1581 tcp6_quench(in6p, errno)
1582 	struct in6pcb *in6p;
1583 	int errno;
1584 {
1585 	struct tcpcb *tp = in6totcpcb(in6p);
1586 
1587 	if (tp)
1588 		tp->snd_cwnd = tp->t_segsz;
1589 }
1590 #endif
1591 
1592 #ifdef INET
1593 /*
1594  * Path MTU Discovery handlers.
1595  */
1596 void
1597 tcp_mtudisc_callback(faddr)
1598 	struct in_addr faddr;
1599 {
1600 #ifdef INET6
1601 	struct in6_addr in6;
1602 #endif
1603 
1604 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1605 #ifdef INET6
1606 	memset(&in6, 0, sizeof(in6));
1607 	in6.s6_addr16[5] = 0xffff;
1608 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1609 	tcp6_mtudisc_callback(&in6);
1610 #endif
1611 }
1612 
1613 /*
1614  * On receipt of path MTU corrections, flush old route and replace it
1615  * with the new one.  Retransmit all unacknowledged packets, to ensure
1616  * that all packets will be received.
1617  */
1618 void
1619 tcp_mtudisc(inp, errno)
1620 	struct inpcb *inp;
1621 	int errno;
1622 {
1623 	struct tcpcb *tp = intotcpcb(inp);
1624 	struct rtentry *rt = in_pcbrtentry(inp);
1625 
1626 	if (tp != 0) {
1627 		if (rt != 0) {
1628 			/*
1629 			 * If this was not a host route, remove and realloc.
1630 			 */
1631 			if ((rt->rt_flags & RTF_HOST) == 0) {
1632 				in_rtchange(inp, errno);
1633 				if ((rt = in_pcbrtentry(inp)) == 0)
1634 					return;
1635 			}
1636 
1637 			/*
1638 			 * Slow start out of the error condition.  We
1639 			 * use the MTU because we know it's smaller
1640 			 * than the previously transmitted segment.
1641 			 *
1642 			 * Note: This is more conservative than the
1643 			 * suggestion in draft-floyd-incr-init-win-03.
1644 			 */
1645 			if (rt->rt_rmx.rmx_mtu != 0)
1646 				tp->snd_cwnd =
1647 				    TCP_INITIAL_WINDOW(tcp_init_win,
1648 				    rt->rt_rmx.rmx_mtu);
1649 		}
1650 
1651 		/*
1652 		 * Resend unacknowledged packets.
1653 		 */
1654 		tp->snd_nxt = tp->snd_una;
1655 		tcp_output(tp);
1656 	}
1657 }
1658 #endif
1659 
1660 #ifdef INET6
1661 /*
1662  * Path MTU Discovery handlers.
1663  */
1664 void
1665 tcp6_mtudisc_callback(faddr)
1666 	struct in6_addr *faddr;
1667 {
1668 	struct sockaddr_in6 sin6;
1669 
1670 	bzero(&sin6, sizeof(sin6));
1671 	sin6.sin6_family = AF_INET6;
1672 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1673 	sin6.sin6_addr = *faddr;
1674 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1675 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1676 }
1677 
1678 void
1679 tcp6_mtudisc(in6p, errno)
1680 	struct in6pcb *in6p;
1681 	int errno;
1682 {
1683 	struct tcpcb *tp = in6totcpcb(in6p);
1684 	struct rtentry *rt = in6_pcbrtentry(in6p);
1685 
1686 	if (tp != 0) {
1687 		if (rt != 0) {
1688 			/*
1689 			 * If this was not a host route, remove and realloc.
1690 			 */
1691 			if ((rt->rt_flags & RTF_HOST) == 0) {
1692 				in6_rtchange(in6p, errno);
1693 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1694 					return;
1695 			}
1696 
1697 			/*
1698 			 * Slow start out of the error condition.  We
1699 			 * use the MTU because we know it's smaller
1700 			 * than the previously transmitted segment.
1701 			 *
1702 			 * Note: This is more conservative than the
1703 			 * suggestion in draft-floyd-incr-init-win-03.
1704 			 */
1705 			if (rt->rt_rmx.rmx_mtu != 0)
1706 				tp->snd_cwnd =
1707 				    TCP_INITIAL_WINDOW(tcp_init_win,
1708 				    rt->rt_rmx.rmx_mtu);
1709 		}
1710 
1711 		/*
1712 		 * Resend unacknowledged packets.
1713 		 */
1714 		tp->snd_nxt = tp->snd_una;
1715 		tcp_output(tp);
1716 	}
1717 }
1718 #endif /* INET6 */
1719 
1720 /*
1721  * Compute the MSS to advertise to the peer.  Called only during
1722  * the 3-way handshake.  If we are the server (peer initiated
1723  * connection), we are called with a pointer to the interface
1724  * on which the SYN packet arrived.  If we are the client (we
1725  * initiated connection), we are called with a pointer to the
1726  * interface out which this connection should go.
1727  *
1728  * NOTE: Do not subtract IP option/extension header size nor IPsec
1729  * header size from MSS advertisement.  MSS option must hold the maximum
1730  * segment size we can accept, so it must always be:
1731  *	 max(if mtu) - ip header - tcp header
1732  */
1733 u_long
1734 tcp_mss_to_advertise(ifp, af)
1735 	const struct ifnet *ifp;
1736 	int af;
1737 {
1738 	extern u_long in_maxmtu;
1739 	u_long mss = 0;
1740 	u_long hdrsiz;
1741 
1742 	/*
1743 	 * In order to avoid defeating path MTU discovery on the peer,
1744 	 * we advertise the max MTU of all attached networks as our MSS,
1745 	 * per RFC 1191, section 3.1.
1746 	 *
1747 	 * We provide the option to advertise just the MTU of
1748 	 * the interface on which we hope this connection will
1749 	 * be receiving.  If we are responding to a SYN, we
1750 	 * will have a pretty good idea about this, but when
1751 	 * initiating a connection there is a bit more doubt.
1752 	 *
1753 	 * We also need to ensure that loopback has a large enough
1754 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1755 	 */
1756 
1757 	if (ifp != NULL)
1758 		switch (af) {
1759 		case AF_INET:
1760 			mss = ifp->if_mtu;
1761 			break;
1762 #ifdef INET6
1763 		case AF_INET6:
1764 			mss = IN6_LINKMTU(ifp);
1765 			break;
1766 #endif
1767 		}
1768 
1769 	if (tcp_mss_ifmtu == 0)
1770 		switch (af) {
1771 		case AF_INET:
1772 			mss = max(in_maxmtu, mss);
1773 			break;
1774 #ifdef INET6
1775 		case AF_INET6:
1776 			mss = max(in6_maxmtu, mss);
1777 			break;
1778 #endif
1779 		}
1780 
1781 	switch (af) {
1782 	case AF_INET:
1783 		hdrsiz = sizeof(struct ip);
1784 		break;
1785 #ifdef INET6
1786 	case AF_INET6:
1787 		hdrsiz = sizeof(struct ip6_hdr);
1788 		break;
1789 #endif
1790 	default:
1791 		hdrsiz = 0;
1792 		break;
1793 	}
1794 	hdrsiz += sizeof(struct tcphdr);
1795 	if (mss > hdrsiz)
1796 		mss -= hdrsiz;
1797 
1798 	mss = max(tcp_mssdflt, mss);
1799 	return (mss);
1800 }
1801 
1802 /*
1803  * Set connection variables based on the peer's advertised MSS.
1804  * We are passed the TCPCB for the actual connection.  If we
1805  * are the server, we are called by the compressed state engine
1806  * when the 3-way handshake is complete.  If we are the client,
1807  * we are called when we receive the SYN,ACK from the server.
1808  *
1809  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1810  * before this routine is called!
1811  */
1812 void
1813 tcp_mss_from_peer(tp, offer)
1814 	struct tcpcb *tp;
1815 	int offer;
1816 {
1817 	struct socket *so;
1818 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1819 	struct rtentry *rt;
1820 #endif
1821 	u_long bufsize;
1822 	int mss;
1823 
1824 #ifdef DIAGNOSTIC
1825 	if (tp->t_inpcb && tp->t_in6pcb)
1826 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1827 #endif
1828 	so = NULL;
1829 	rt = NULL;
1830 #ifdef INET
1831 	if (tp->t_inpcb) {
1832 		so = tp->t_inpcb->inp_socket;
1833 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1834 		rt = in_pcbrtentry(tp->t_inpcb);
1835 #endif
1836 	}
1837 #endif
1838 #ifdef INET6
1839 	if (tp->t_in6pcb) {
1840 		so = tp->t_in6pcb->in6p_socket;
1841 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1842 		rt = in6_pcbrtentry(tp->t_in6pcb);
1843 #endif
1844 	}
1845 #endif
1846 
1847 	/*
1848 	 * As per RFC1122, use the default MSS value, unless they
1849 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1850 	 */
1851 	mss = tcp_mssdflt;
1852 	if (offer)
1853 		mss = offer;
1854 	mss = max(mss, 256);		/* sanity */
1855 	tp->t_peermss = mss;
1856 	mss -= tcp_optlen(tp);
1857 #ifdef INET
1858 	if (tp->t_inpcb)
1859 		mss -= ip_optlen(tp->t_inpcb);
1860 #endif
1861 #ifdef INET6
1862 	if (tp->t_in6pcb)
1863 		mss -= ip6_optlen(tp->t_in6pcb);
1864 #endif
1865 
1866 	/*
1867 	 * If there's a pipesize, change the socket buffer to that size.
1868 	 * Make the socket buffer an integral number of MSS units.  If
1869 	 * the MSS is larger than the socket buffer, artificially decrease
1870 	 * the MSS.
1871 	 */
1872 #ifdef RTV_SPIPE
1873 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1874 		bufsize = rt->rt_rmx.rmx_sendpipe;
1875 	else
1876 #endif
1877 		bufsize = so->so_snd.sb_hiwat;
1878 	if (bufsize < mss)
1879 		mss = bufsize;
1880 	else {
1881 		bufsize = roundup(bufsize, mss);
1882 		if (bufsize > sb_max)
1883 			bufsize = sb_max;
1884 		(void) sbreserve(&so->so_snd, bufsize, so);
1885 	}
1886 	tp->t_segsz = mss;
1887 
1888 #ifdef RTV_SSTHRESH
1889 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1890 		/*
1891 		 * There's some sort of gateway or interface buffer
1892 		 * limit on the path.  Use this to set the slow
1893 		 * start threshold, but set the threshold to no less
1894 		 * than 2 * MSS.
1895 		 */
1896 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1897 	}
1898 #endif
1899 }
1900 
1901 /*
1902  * Processing necessary when a TCP connection is established.
1903  */
1904 void
1905 tcp_established(tp)
1906 	struct tcpcb *tp;
1907 {
1908 	struct socket *so;
1909 #ifdef RTV_RPIPE
1910 	struct rtentry *rt;
1911 #endif
1912 	u_long bufsize;
1913 
1914 #ifdef DIAGNOSTIC
1915 	if (tp->t_inpcb && tp->t_in6pcb)
1916 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1917 #endif
1918 	so = NULL;
1919 	rt = NULL;
1920 #ifdef INET
1921 	if (tp->t_inpcb) {
1922 		so = tp->t_inpcb->inp_socket;
1923 #if defined(RTV_RPIPE)
1924 		rt = in_pcbrtentry(tp->t_inpcb);
1925 #endif
1926 	}
1927 #endif
1928 #ifdef INET6
1929 	if (tp->t_in6pcb) {
1930 		so = tp->t_in6pcb->in6p_socket;
1931 #if defined(RTV_RPIPE)
1932 		rt = in6_pcbrtentry(tp->t_in6pcb);
1933 #endif
1934 	}
1935 #endif
1936 
1937 	tp->t_state = TCPS_ESTABLISHED;
1938 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1939 
1940 #ifdef RTV_RPIPE
1941 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1942 		bufsize = rt->rt_rmx.rmx_recvpipe;
1943 	else
1944 #endif
1945 		bufsize = so->so_rcv.sb_hiwat;
1946 	if (bufsize > tp->t_ourmss) {
1947 		bufsize = roundup(bufsize, tp->t_ourmss);
1948 		if (bufsize > sb_max)
1949 			bufsize = sb_max;
1950 		(void) sbreserve(&so->so_rcv, bufsize, so);
1951 	}
1952 }
1953 
1954 /*
1955  * Check if there's an initial rtt or rttvar.  Convert from the
1956  * route-table units to scaled multiples of the slow timeout timer.
1957  * Called only during the 3-way handshake.
1958  */
1959 void
1960 tcp_rmx_rtt(tp)
1961 	struct tcpcb *tp;
1962 {
1963 #ifdef RTV_RTT
1964 	struct rtentry *rt = NULL;
1965 	int rtt;
1966 
1967 #ifdef DIAGNOSTIC
1968 	if (tp->t_inpcb && tp->t_in6pcb)
1969 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1970 #endif
1971 #ifdef INET
1972 	if (tp->t_inpcb)
1973 		rt = in_pcbrtentry(tp->t_inpcb);
1974 #endif
1975 #ifdef INET6
1976 	if (tp->t_in6pcb)
1977 		rt = in6_pcbrtentry(tp->t_in6pcb);
1978 #endif
1979 	if (rt == NULL)
1980 		return;
1981 
1982 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1983 		/*
1984 		 * XXX The lock bit for MTU indicates that the value
1985 		 * is also a minimum value; this is subject to time.
1986 		 */
1987 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1988 			TCPT_RANGESET(tp->t_rttmin,
1989 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1990 			    TCPTV_MIN, TCPTV_REXMTMAX);
1991 		tp->t_srtt = rtt /
1992 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1993 		if (rt->rt_rmx.rmx_rttvar) {
1994 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1995 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1996 				(TCP_RTTVAR_SHIFT + 2));
1997 		} else {
1998 			/* Default variation is +- 1 rtt */
1999 			tp->t_rttvar =
2000 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2001 		}
2002 		TCPT_RANGESET(tp->t_rxtcur,
2003 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2004 		    tp->t_rttmin, TCPTV_REXMTMAX);
2005 	}
2006 #endif
2007 }
2008 
2009 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2010 #if NRND > 0
2011 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2012 #endif
2013 
2014 /*
2015  * Get a new sequence value given a tcp control block
2016  */
2017 tcp_seq
2018 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2019 {
2020 
2021 #ifdef INET
2022 	if (tp->t_inpcb != NULL) {
2023 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2024 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2025 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2026 		    addin));
2027 	}
2028 #endif
2029 #ifdef INET6
2030 	if (tp->t_in6pcb != NULL) {
2031 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2032 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2033 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2034 		    addin));
2035 	}
2036 #endif
2037 	/* Not possible. */
2038 	panic("tcp_new_iss");
2039 }
2040 
2041 /*
2042  * This routine actually generates a new TCP initial sequence number.
2043  */
2044 tcp_seq
2045 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2046     size_t addrsz, tcp_seq addin)
2047 {
2048 	tcp_seq tcp_iss;
2049 
2050 #if NRND > 0
2051 	static int beenhere;
2052 
2053 	/*
2054 	 * If we haven't been here before, initialize our cryptographic
2055 	 * hash secret.
2056 	 */
2057 	if (beenhere == 0) {
2058 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2059 		    RND_EXTRACT_ANY);
2060 		beenhere = 1;
2061 	}
2062 
2063 	if (tcp_do_rfc1948) {
2064 		MD5_CTX ctx;
2065 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2066 
2067 		/*
2068 		 * Compute the base value of the ISS.  It is a hash
2069 		 * of (saddr, sport, daddr, dport, secret).
2070 		 */
2071 		MD5Init(&ctx);
2072 
2073 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2074 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2075 
2076 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2077 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2078 
2079 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2080 
2081 		MD5Final(hash, &ctx);
2082 
2083 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2084 
2085 		/*
2086 		 * Now increment our "timer", and add it in to
2087 		 * the computed value.
2088 		 *
2089 		 * XXX Use `addin'?
2090 		 * XXX TCP_ISSINCR too large to use?
2091 		 */
2092 		tcp_iss_seq += TCP_ISSINCR;
2093 #ifdef TCPISS_DEBUG
2094 		printf("ISS hash 0x%08x, ", tcp_iss);
2095 #endif
2096 		tcp_iss += tcp_iss_seq + addin;
2097 #ifdef TCPISS_DEBUG
2098 		printf("new ISS 0x%08x\n", tcp_iss);
2099 #endif
2100 	} else
2101 #endif /* NRND > 0 */
2102 	{
2103 		/*
2104 		 * Randomize.
2105 		 */
2106 #if NRND > 0
2107 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2108 #else
2109 		tcp_iss = arc4random();
2110 #endif
2111 
2112 		/*
2113 		 * If we were asked to add some amount to a known value,
2114 		 * we will take a random value obtained above, mask off
2115 		 * the upper bits, and add in the known value.  We also
2116 		 * add in a constant to ensure that we are at least a
2117 		 * certain distance from the original value.
2118 		 *
2119 		 * This is used when an old connection is in timed wait
2120 		 * and we have a new one coming in, for instance.
2121 		 */
2122 		if (addin != 0) {
2123 #ifdef TCPISS_DEBUG
2124 			printf("Random %08x, ", tcp_iss);
2125 #endif
2126 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2127 			tcp_iss += addin + TCP_ISSINCR;
2128 #ifdef TCPISS_DEBUG
2129 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2130 #endif
2131 		} else {
2132 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2133 			tcp_iss += tcp_iss_seq;
2134 			tcp_iss_seq += TCP_ISSINCR;
2135 #ifdef TCPISS_DEBUG
2136 			printf("ISS %08x\n", tcp_iss);
2137 #endif
2138 		}
2139 	}
2140 
2141 	if (tcp_compat_42) {
2142 		/*
2143 		 * Limit it to the positive range for really old TCP
2144 		 * implementations.
2145 		 * Just AND off the top bit instead of checking if
2146 		 * is set first - saves a branch 50% of the time.
2147 		 */
2148 		tcp_iss &= 0x7fffffff;		/* XXX */
2149 	}
2150 
2151 	return (tcp_iss);
2152 }
2153 
2154 #if defined(IPSEC) || defined(FAST_IPSEC)
2155 /* compute ESP/AH header size for TCP, including outer IP header. */
2156 size_t
2157 ipsec4_hdrsiz_tcp(tp)
2158 	struct tcpcb *tp;
2159 {
2160 	struct inpcb *inp;
2161 	size_t hdrsiz;
2162 
2163 	/* XXX mapped addr case (tp->t_in6pcb) */
2164 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2165 		return 0;
2166 	switch (tp->t_family) {
2167 	case AF_INET:
2168 		/* XXX: should use currect direction. */
2169 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2170 		break;
2171 	default:
2172 		hdrsiz = 0;
2173 		break;
2174 	}
2175 
2176 	return hdrsiz;
2177 }
2178 
2179 #ifdef INET6
2180 size_t
2181 ipsec6_hdrsiz_tcp(tp)
2182 	struct tcpcb *tp;
2183 {
2184 	struct in6pcb *in6p;
2185 	size_t hdrsiz;
2186 
2187 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2188 		return 0;
2189 	switch (tp->t_family) {
2190 	case AF_INET6:
2191 		/* XXX: should use currect direction. */
2192 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2193 		break;
2194 	case AF_INET:
2195 		/* mapped address case - tricky */
2196 	default:
2197 		hdrsiz = 0;
2198 		break;
2199 	}
2200 
2201 	return hdrsiz;
2202 }
2203 #endif
2204 #endif /*IPSEC*/
2205 
2206 /*
2207  * Determine the length of the TCP options for this connection.
2208  *
2209  * XXX:  What do we do for SACK, when we add that?  Just reserve
2210  *       all of the space?  Otherwise we can't exactly be incrementing
2211  *       cwnd by an amount that varies depending on the amount we last
2212  *       had to SACK!
2213  */
2214 
2215 u_int
2216 tcp_optlen(tp)
2217 	struct tcpcb *tp;
2218 {
2219 	u_int optlen;
2220 
2221 	optlen = 0;
2222 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2223 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2224 		optlen += TCPOLEN_TSTAMP_APPA;
2225 
2226 #ifdef TCP_SIGNATURE
2227 #if defined(INET6) && defined(FAST_IPSEC)
2228 	if (tp->t_family == AF_INET)
2229 #endif
2230 	if (tp->t_flags & TF_SIGNATURE)
2231 		optlen += TCPOLEN_SIGNATURE + 2;
2232 #endif /* TCP_SIGNATURE */
2233 
2234 	return optlen;
2235 }
2236