xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: tcp_subr.c,v 1.139 2003/03/01 04:40:28 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. All advertising materials mentioning features or use of this software
82  *    must display the following acknowledgement:
83  *	This product includes software developed by the University of
84  *	California, Berkeley and its contributors.
85  * 4. Neither the name of the University nor the names of its contributors
86  *    may be used to endorse or promote products derived from this software
87  *    without specific prior written permission.
88  *
89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
99  * SUCH DAMAGE.
100  *
101  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
102  */
103 
104 #include <sys/cdefs.h>
105 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.139 2003/03/01 04:40:28 thorpej Exp $");
106 
107 #include "opt_inet.h"
108 #include "opt_ipsec.h"
109 #include "opt_tcp_compat_42.h"
110 #include "opt_inet_csum.h"
111 #include "rnd.h"
112 
113 #include <sys/param.h>
114 #include <sys/proc.h>
115 #include <sys/systm.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/socket.h>
119 #include <sys/socketvar.h>
120 #include <sys/protosw.h>
121 #include <sys/errno.h>
122 #include <sys/kernel.h>
123 #include <sys/pool.h>
124 #if NRND > 0
125 #include <sys/md5.h>
126 #include <sys/rnd.h>
127 #endif
128 
129 #include <net/route.h>
130 #include <net/if.h>
131 
132 #include <netinet/in.h>
133 #include <netinet/in_systm.h>
134 #include <netinet/ip.h>
135 #include <netinet/in_pcb.h>
136 #include <netinet/ip_var.h>
137 #include <netinet/ip_icmp.h>
138 
139 #ifdef INET6
140 #ifndef INET
141 #include <netinet/in.h>
142 #endif
143 #include <netinet/ip6.h>
144 #include <netinet6/in6_pcb.h>
145 #include <netinet6/ip6_var.h>
146 #include <netinet6/in6_var.h>
147 #include <netinet6/ip6protosw.h>
148 #include <netinet/icmp6.h>
149 #include <netinet6/nd6.h>
150 #endif
151 
152 #include <netinet/tcp.h>
153 #include <netinet/tcp_fsm.h>
154 #include <netinet/tcp_seq.h>
155 #include <netinet/tcp_timer.h>
156 #include <netinet/tcp_var.h>
157 #include <netinet/tcpip.h>
158 
159 #ifdef IPSEC
160 #include <netinet6/ipsec.h>
161 #endif /*IPSEC*/
162 
163 #ifdef INET6
164 struct in6pcb tcb6;
165 #endif
166 
167 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
168 struct	tcpstat tcpstat;	/* tcp statistics */
169 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
170 
171 /* patchable/settable parameters for tcp */
172 int 	tcp_mssdflt = TCP_MSS;
173 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
174 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
175 #if NRND > 0
176 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
177 #endif
178 int	tcp_do_sack = 1;	/* selective acknowledgement */
179 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
180 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
181 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
182 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
183 int	tcp_init_win = 1;	/* initial slow start window */
184 int	tcp_init_win_local = 4;	/* initial slow start window for local nets */
185 int	tcp_mss_ifmtu = 0;
186 #ifdef TCP_COMPAT_42
187 int	tcp_compat_42 = 1;
188 #else
189 int	tcp_compat_42 = 0;
190 #endif
191 int	tcp_rst_ppslim = 100;	/* 100pps */
192 
193 /* tcb hash */
194 #ifndef TCBHASHSIZE
195 #define	TCBHASHSIZE	128
196 #endif
197 int	tcbhashsize = TCBHASHSIZE;
198 
199 /* syn hash parameters */
200 #define	TCP_SYN_HASH_SIZE	293
201 #define	TCP_SYN_BUCKET_SIZE	35
202 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
203 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
204 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
205 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
206 
207 int	tcp_freeq __P((struct tcpcb *));
208 
209 #ifdef INET
210 void	tcp_mtudisc_callback __P((struct in_addr));
211 #endif
212 #ifdef INET6
213 void	tcp6_mtudisc_callback __P((struct in6_addr *));
214 #endif
215 
216 void	tcp_mtudisc __P((struct inpcb *, int));
217 #ifdef INET6
218 void	tcp6_mtudisc __P((struct in6pcb *, int));
219 #endif
220 
221 struct pool tcpcb_pool;
222 
223 #ifdef TCP_CSUM_COUNTERS
224 #include <sys/device.h>
225 
226 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
227     NULL, "tcp", "hwcsum bad");
228 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
229     NULL, "tcp", "hwcsum ok");
230 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
231     NULL, "tcp", "hwcsum data");
232 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
233     NULL, "tcp", "swcsum");
234 #endif /* TCP_CSUM_COUNTERS */
235 
236 #ifdef TCP_OUTPUT_COUNTERS
237 #include <sys/device.h>
238 
239 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "output big header");
241 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
242     NULL, "tcp", "output copy small");
243 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
244     NULL, "tcp", "output copy big");
245 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
246     NULL, "tcp", "output reference big");
247 #endif /* TCP_OUTPUT_COUNTERS */
248 
249 #ifdef TCP_REASS_COUNTERS
250 #include <sys/device.h>
251 
252 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
253     NULL, "tcp_reass", "calls");
254 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
255     &tcp_reass_, "tcp_reass", "insert into empty queue");
256 struct evcnt tcp_reass_iteration[8] = {
257     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
258     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
259     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
260     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
261     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
262     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
263     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
264     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
265 };
266 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
267     &tcp_reass_, "tcp_reass", "prepend to first");
268 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
269     &tcp_reass_, "tcp_reass", "prepend");
270 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
271     &tcp_reass_, "tcp_reass", "insert");
272 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
273     &tcp_reass_, "tcp_reass", "insert at tail");
274 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
275     &tcp_reass_, "tcp_reass", "append");
276 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
277     &tcp_reass_, "tcp_reass", "append to tail fragment");
278 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
279     &tcp_reass_, "tcp_reass", "overlap at end");
280 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
281     &tcp_reass_, "tcp_reass", "overlap at start");
282 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
283     &tcp_reass_, "tcp_reass", "duplicate segment");
284 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
285     &tcp_reass_, "tcp_reass", "duplicate fragment");
286 
287 #endif /* TCP_REASS_COUNTERS */
288 
289 #ifdef MBUFTRACE
290 struct mowner tcp_mowner = { "tcp" };
291 struct mowner tcp_rx_mowner = { "tcp", "rx" };
292 struct mowner tcp_tx_mowner = { "tcp", "tx" };
293 #endif
294 
295 /*
296  * Tcp initialization
297  */
298 void
299 tcp_init()
300 {
301 	int hlen;
302 
303 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
304 	    NULL);
305 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
306 #ifdef INET6
307 	tcb6.in6p_next = tcb6.in6p_prev = &tcb6;
308 #endif
309 
310 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
311 #ifdef INET6
312 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
313 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
314 #endif
315 	if (max_protohdr < hlen)
316 		max_protohdr = hlen;
317 	if (max_linkhdr + hlen > MHLEN)
318 		panic("tcp_init");
319 
320 #ifdef INET
321 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
322 #endif
323 #ifdef INET6
324 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
325 #endif
326 
327 	/* Initialize timer state. */
328 	tcp_timer_init();
329 
330 	/* Initialize the compressed state engine. */
331 	syn_cache_init();
332 
333 #ifdef TCP_CSUM_COUNTERS
334 	evcnt_attach_static(&tcp_hwcsum_bad);
335 	evcnt_attach_static(&tcp_hwcsum_ok);
336 	evcnt_attach_static(&tcp_hwcsum_data);
337 	evcnt_attach_static(&tcp_swcsum);
338 #endif /* TCP_CSUM_COUNTERS */
339 
340 #ifdef TCP_OUTPUT_COUNTERS
341 	evcnt_attach_static(&tcp_output_bigheader);
342 	evcnt_attach_static(&tcp_output_copysmall);
343 	evcnt_attach_static(&tcp_output_copybig);
344 	evcnt_attach_static(&tcp_output_refbig);
345 #endif /* TCP_OUTPUT_COUNTERS */
346 
347 #ifdef TCP_REASS_COUNTERS
348 	evcnt_attach_static(&tcp_reass_);
349 	evcnt_attach_static(&tcp_reass_empty);
350 	evcnt_attach_static(&tcp_reass_iteration[0]);
351 	evcnt_attach_static(&tcp_reass_iteration[1]);
352 	evcnt_attach_static(&tcp_reass_iteration[2]);
353 	evcnt_attach_static(&tcp_reass_iteration[3]);
354 	evcnt_attach_static(&tcp_reass_iteration[4]);
355 	evcnt_attach_static(&tcp_reass_iteration[5]);
356 	evcnt_attach_static(&tcp_reass_iteration[6]);
357 	evcnt_attach_static(&tcp_reass_iteration[7]);
358 	evcnt_attach_static(&tcp_reass_prependfirst);
359 	evcnt_attach_static(&tcp_reass_prepend);
360 	evcnt_attach_static(&tcp_reass_insert);
361 	evcnt_attach_static(&tcp_reass_inserttail);
362 	evcnt_attach_static(&tcp_reass_append);
363 	evcnt_attach_static(&tcp_reass_appendtail);
364 	evcnt_attach_static(&tcp_reass_overlaptail);
365 	evcnt_attach_static(&tcp_reass_overlapfront);
366 	evcnt_attach_static(&tcp_reass_segdup);
367 	evcnt_attach_static(&tcp_reass_fragdup);
368 #endif /* TCP_REASS_COUNTERS */
369 
370 	MOWNER_ATTACH(&tcp_tx_mowner);
371 	MOWNER_ATTACH(&tcp_rx_mowner);
372 	MOWNER_ATTACH(&tcp_mowner);
373 }
374 
375 /*
376  * Create template to be used to send tcp packets on a connection.
377  * Call after host entry created, allocates an mbuf and fills
378  * in a skeletal tcp/ip header, minimizing the amount of work
379  * necessary when the connection is used.
380  */
381 struct mbuf *
382 tcp_template(tp)
383 	struct tcpcb *tp;
384 {
385 	struct inpcb *inp = tp->t_inpcb;
386 #ifdef INET6
387 	struct in6pcb *in6p = tp->t_in6pcb;
388 #endif
389 	struct tcphdr *n;
390 	struct mbuf *m;
391 	int hlen;
392 
393 	switch (tp->t_family) {
394 	case AF_INET:
395 		hlen = sizeof(struct ip);
396 		if (inp)
397 			break;
398 #ifdef INET6
399 		if (in6p) {
400 			/* mapped addr case */
401 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
402 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
403 				break;
404 		}
405 #endif
406 		return NULL;	/*EINVAL*/
407 #ifdef INET6
408 	case AF_INET6:
409 		hlen = sizeof(struct ip6_hdr);
410 		if (in6p) {
411 			/* more sainty check? */
412 			break;
413 		}
414 		return NULL;	/*EINVAL*/
415 #endif
416 	default:
417 		hlen = 0;	/*pacify gcc*/
418 		return NULL;	/*EAFNOSUPPORT*/
419 	}
420 #ifdef DIAGNOSTIC
421 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
422 		panic("mclbytes too small for t_template");
423 #endif
424 	m = tp->t_template;
425 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
426 		;
427 	else {
428 		if (m)
429 			m_freem(m);
430 		m = tp->t_template = NULL;
431 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
432 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
433 			MCLGET(m, M_DONTWAIT);
434 			if ((m->m_flags & M_EXT) == 0) {
435 				m_free(m);
436 				m = NULL;
437 			}
438 		}
439 		if (m == NULL)
440 			return NULL;
441 		MCLAIM(m, &tcp_mowner);
442 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
443 	}
444 
445 	bzero(mtod(m, caddr_t), m->m_len);
446 
447 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
448 
449 	switch (tp->t_family) {
450 	case AF_INET:
451 	    {
452 		struct ipovly *ipov;
453 		mtod(m, struct ip *)->ip_v = 4;
454 		ipov = mtod(m, struct ipovly *);
455 		ipov->ih_pr = IPPROTO_TCP;
456 		ipov->ih_len = htons(sizeof(struct tcphdr));
457 		if (inp) {
458 			ipov->ih_src = inp->inp_laddr;
459 			ipov->ih_dst = inp->inp_faddr;
460 		}
461 #ifdef INET6
462 		else if (in6p) {
463 			/* mapped addr case */
464 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
465 				sizeof(ipov->ih_src));
466 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
467 				sizeof(ipov->ih_dst));
468 		}
469 #endif
470 		/*
471 		 * Compute the pseudo-header portion of the checksum
472 		 * now.  We incrementally add in the TCP option and
473 		 * payload lengths later, and then compute the TCP
474 		 * checksum right before the packet is sent off onto
475 		 * the wire.
476 		 */
477 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
478 		    ipov->ih_dst.s_addr,
479 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
480 		break;
481 	    }
482 #ifdef INET6
483 	case AF_INET6:
484 	    {
485 		struct ip6_hdr *ip6;
486 		mtod(m, struct ip *)->ip_v = 6;
487 		ip6 = mtod(m, struct ip6_hdr *);
488 		ip6->ip6_nxt = IPPROTO_TCP;
489 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
490 		ip6->ip6_src = in6p->in6p_laddr;
491 		ip6->ip6_dst = in6p->in6p_faddr;
492 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
493 		if (ip6_auto_flowlabel) {
494 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
495 			ip6->ip6_flow |=
496 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
497 		}
498 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
499 		ip6->ip6_vfc |= IPV6_VERSION;
500 
501 		/*
502 		 * Compute the pseudo-header portion of the checksum
503 		 * now.  We incrementally add in the TCP option and
504 		 * payload lengths later, and then compute the TCP
505 		 * checksum right before the packet is sent off onto
506 		 * the wire.
507 		 */
508 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
509 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
510 		    htonl(IPPROTO_TCP));
511 		break;
512 	    }
513 #endif
514 	}
515 	if (inp) {
516 		n->th_sport = inp->inp_lport;
517 		n->th_dport = inp->inp_fport;
518 	}
519 #ifdef INET6
520 	else if (in6p) {
521 		n->th_sport = in6p->in6p_lport;
522 		n->th_dport = in6p->in6p_fport;
523 	}
524 #endif
525 	n->th_seq = 0;
526 	n->th_ack = 0;
527 	n->th_x2 = 0;
528 	n->th_off = 5;
529 	n->th_flags = 0;
530 	n->th_win = 0;
531 	n->th_urp = 0;
532 	return (m);
533 }
534 
535 /*
536  * Send a single message to the TCP at address specified by
537  * the given TCP/IP header.  If m == 0, then we make a copy
538  * of the tcpiphdr at ti and send directly to the addressed host.
539  * This is used to force keep alive messages out using the TCP
540  * template for a connection tp->t_template.  If flags are given
541  * then we send a message back to the TCP which originated the
542  * segment ti, and discard the mbuf containing it and any other
543  * attached mbufs.
544  *
545  * In any case the ack and sequence number of the transmitted
546  * segment are as specified by the parameters.
547  */
548 int
549 tcp_respond(tp, template, m, th0, ack, seq, flags)
550 	struct tcpcb *tp;
551 	struct mbuf *template;
552 	struct mbuf *m;
553 	struct tcphdr *th0;
554 	tcp_seq ack, seq;
555 	int flags;
556 {
557 	struct route *ro;
558 	int error, tlen, win = 0;
559 	int hlen;
560 	struct ip *ip;
561 #ifdef INET6
562 	struct ip6_hdr *ip6;
563 #endif
564 	int family;	/* family on packet, not inpcb/in6pcb! */
565 	struct tcphdr *th;
566 
567 	if (tp != NULL && (flags & TH_RST) == 0) {
568 #ifdef DIAGNOSTIC
569 		if (tp->t_inpcb && tp->t_in6pcb)
570 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
571 #endif
572 #ifdef INET
573 		if (tp->t_inpcb)
574 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
575 #endif
576 #ifdef INET6
577 		if (tp->t_in6pcb)
578 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
579 #endif
580 	}
581 
582 	th = NULL;	/* Quell uninitialized warning */
583 	ip = NULL;
584 #ifdef INET6
585 	ip6 = NULL;
586 #endif
587 	if (m == 0) {
588 		if (!template)
589 			return EINVAL;
590 
591 		/* get family information from template */
592 		switch (mtod(template, struct ip *)->ip_v) {
593 		case 4:
594 			family = AF_INET;
595 			hlen = sizeof(struct ip);
596 			break;
597 #ifdef INET6
598 		case 6:
599 			family = AF_INET6;
600 			hlen = sizeof(struct ip6_hdr);
601 			break;
602 #endif
603 		default:
604 			return EAFNOSUPPORT;
605 		}
606 
607 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
608 		if (m) {
609 			MCLAIM(m, &tcp_tx_mowner);
610 			MCLGET(m, M_DONTWAIT);
611 			if ((m->m_flags & M_EXT) == 0) {
612 				m_free(m);
613 				m = NULL;
614 			}
615 		}
616 		if (m == NULL)
617 			return (ENOBUFS);
618 
619 		if (tcp_compat_42)
620 			tlen = 1;
621 		else
622 			tlen = 0;
623 
624 		m->m_data += max_linkhdr;
625 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
626 			template->m_len);
627 		switch (family) {
628 		case AF_INET:
629 			ip = mtod(m, struct ip *);
630 			th = (struct tcphdr *)(ip + 1);
631 			break;
632 #ifdef INET6
633 		case AF_INET6:
634 			ip6 = mtod(m, struct ip6_hdr *);
635 			th = (struct tcphdr *)(ip6 + 1);
636 			break;
637 #endif
638 #if 0
639 		default:
640 			/* noone will visit here */
641 			m_freem(m);
642 			return EAFNOSUPPORT;
643 #endif
644 		}
645 		flags = TH_ACK;
646 	} else {
647 
648 		if ((m->m_flags & M_PKTHDR) == 0) {
649 #if 0
650 			printf("non PKTHDR to tcp_respond\n");
651 #endif
652 			m_freem(m);
653 			return EINVAL;
654 		}
655 #ifdef DIAGNOSTIC
656 		if (!th0)
657 			panic("th0 == NULL in tcp_respond");
658 #endif
659 
660 		/* get family information from m */
661 		switch (mtod(m, struct ip *)->ip_v) {
662 		case 4:
663 			family = AF_INET;
664 			hlen = sizeof(struct ip);
665 			ip = mtod(m, struct ip *);
666 			break;
667 #ifdef INET6
668 		case 6:
669 			family = AF_INET6;
670 			hlen = sizeof(struct ip6_hdr);
671 			ip6 = mtod(m, struct ip6_hdr *);
672 			break;
673 #endif
674 		default:
675 			m_freem(m);
676 			return EAFNOSUPPORT;
677 		}
678 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
679 			tlen = sizeof(*th0);
680 		else
681 			tlen = th0->th_off << 2;
682 
683 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
684 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
685 			m->m_len = hlen + tlen;
686 			m_freem(m->m_next);
687 			m->m_next = NULL;
688 		} else {
689 			struct mbuf *n;
690 
691 #ifdef DIAGNOSTIC
692 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
693 				m_freem(m);
694 				return EMSGSIZE;
695 			}
696 #endif
697 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
698 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
699 				MCLGET(n, M_DONTWAIT);
700 				if ((n->m_flags & M_EXT) == 0) {
701 					m_freem(n);
702 					n = NULL;
703 				}
704 			}
705 			if (!n) {
706 				m_freem(m);
707 				return ENOBUFS;
708 			}
709 
710 			MCLAIM(n, &tcp_tx_mowner);
711 			n->m_data += max_linkhdr;
712 			n->m_len = hlen + tlen;
713 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
714 			m_copyback(n, hlen, tlen, (caddr_t)th0);
715 
716 			m_freem(m);
717 			m = n;
718 			n = NULL;
719 		}
720 
721 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
722 		switch (family) {
723 		case AF_INET:
724 			ip = mtod(m, struct ip *);
725 			th = (struct tcphdr *)(ip + 1);
726 			ip->ip_p = IPPROTO_TCP;
727 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
728 			ip->ip_p = IPPROTO_TCP;
729 			break;
730 #ifdef INET6
731 		case AF_INET6:
732 			ip6 = mtod(m, struct ip6_hdr *);
733 			th = (struct tcphdr *)(ip6 + 1);
734 			ip6->ip6_nxt = IPPROTO_TCP;
735 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
736 			ip6->ip6_nxt = IPPROTO_TCP;
737 			break;
738 #endif
739 #if 0
740 		default:
741 			/* noone will visit here */
742 			m_freem(m);
743 			return EAFNOSUPPORT;
744 #endif
745 		}
746 		xchg(th->th_dport, th->th_sport, u_int16_t);
747 #undef xchg
748 		tlen = 0;	/*be friendly with the following code*/
749 	}
750 	th->th_seq = htonl(seq);
751 	th->th_ack = htonl(ack);
752 	th->th_x2 = 0;
753 	if ((flags & TH_SYN) == 0) {
754 		if (tp)
755 			win >>= tp->rcv_scale;
756 		if (win > TCP_MAXWIN)
757 			win = TCP_MAXWIN;
758 		th->th_win = htons((u_int16_t)win);
759 		th->th_off = sizeof (struct tcphdr) >> 2;
760 		tlen += sizeof(*th);
761 	} else
762 		tlen += th->th_off << 2;
763 	m->m_len = hlen + tlen;
764 	m->m_pkthdr.len = hlen + tlen;
765 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
766 	th->th_flags = flags;
767 	th->th_urp = 0;
768 
769 	switch (family) {
770 #ifdef INET
771 	case AF_INET:
772 	    {
773 		struct ipovly *ipov = (struct ipovly *)ip;
774 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
775 		ipov->ih_len = htons((u_int16_t)tlen);
776 
777 		th->th_sum = 0;
778 		th->th_sum = in_cksum(m, hlen + tlen);
779 		ip->ip_len = htons(hlen + tlen);
780 		ip->ip_ttl = ip_defttl;
781 		break;
782 	    }
783 #endif
784 #ifdef INET6
785 	case AF_INET6:
786 	    {
787 		th->th_sum = 0;
788 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
789 				tlen);
790 		ip6->ip6_plen = ntohs(tlen);
791 		if (tp && tp->t_in6pcb) {
792 			struct ifnet *oifp;
793 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
794 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
795 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
796 		} else
797 			ip6->ip6_hlim = ip6_defhlim;
798 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
799 		if (ip6_auto_flowlabel) {
800 			ip6->ip6_flow |=
801 				(htonl(ip6_flow_seq++) & IPV6_FLOWLABEL_MASK);
802 		}
803 		break;
804 	    }
805 #endif
806 	}
807 
808 #ifdef IPSEC
809 	(void)ipsec_setsocket(m, NULL);
810 #endif /*IPSEC*/
811 
812 	if (tp != NULL && tp->t_inpcb != NULL) {
813 		ro = &tp->t_inpcb->inp_route;
814 #ifdef IPSEC
815 		if (ipsec_setsocket(m, tp->t_inpcb->inp_socket) != 0) {
816 			m_freem(m);
817 			return ENOBUFS;
818 		}
819 #endif
820 #ifdef DIAGNOSTIC
821 		if (family != AF_INET)
822 			panic("tcp_respond: address family mismatch");
823 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
824 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
825 			    ntohl(ip->ip_dst.s_addr),
826 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
827 		}
828 #endif
829 	}
830 #ifdef INET6
831 	else if (tp != NULL && tp->t_in6pcb != NULL) {
832 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
833 #ifdef IPSEC
834 		if (ipsec_setsocket(m, tp->t_in6pcb->in6p_socket) != 0) {
835 			m_freem(m);
836 			return ENOBUFS;
837 		}
838 #endif
839 #ifdef DIAGNOSTIC
840 		if (family == AF_INET) {
841 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
842 				panic("tcp_respond: not mapped addr");
843 			if (bcmp(&ip->ip_dst,
844 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
845 			    sizeof(ip->ip_dst)) != 0) {
846 				panic("tcp_respond: ip_dst != in6p_faddr");
847 			}
848 		} else if (family == AF_INET6) {
849 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
850 			    &tp->t_in6pcb->in6p_faddr))
851 				panic("tcp_respond: ip6_dst != in6p_faddr");
852 		} else
853 			panic("tcp_respond: address family mismatch");
854 #endif
855 	}
856 #endif
857 	else
858 		ro = NULL;
859 
860 	switch (family) {
861 #ifdef INET
862 	case AF_INET:
863 		error = ip_output(m, NULL, ro,
864 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
865 		    NULL);
866 		break;
867 #endif
868 #ifdef INET6
869 	case AF_INET6:
870 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0, NULL,
871 		    NULL);
872 		break;
873 #endif
874 	default:
875 		error = EAFNOSUPPORT;
876 		break;
877 	}
878 
879 	return (error);
880 }
881 
882 /*
883  * Create a new TCP control block, making an
884  * empty reassembly queue and hooking it to the argument
885  * protocol control block.
886  */
887 struct tcpcb *
888 tcp_newtcpcb(family, aux)
889 	int family;	/* selects inpcb, or in6pcb */
890 	void *aux;
891 {
892 	struct tcpcb *tp;
893 	int i;
894 
895 	switch (family) {
896 	case PF_INET:
897 		break;
898 #ifdef INET6
899 	case PF_INET6:
900 		break;
901 #endif
902 	default:
903 		return NULL;
904 	}
905 
906 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
907 	if (tp == NULL)
908 		return (NULL);
909 	bzero((caddr_t)tp, sizeof(struct tcpcb));
910 	TAILQ_INIT(&tp->segq);
911 	TAILQ_INIT(&tp->timeq);
912 	tp->t_family = family;		/* may be overridden later on */
913 	tp->t_peermss = tcp_mssdflt;
914 	tp->t_ourmss = tcp_mssdflt;
915 	tp->t_segsz = tcp_mssdflt;
916 	LIST_INIT(&tp->t_sc);
917 
918 	callout_init(&tp->t_delack_ch);
919 	for (i = 0; i < TCPT_NTIMERS; i++)
920 		TCP_TIMER_INIT(tp, i);
921 
922 	tp->t_flags = 0;
923 	if (tcp_do_rfc1323 && tcp_do_win_scale)
924 		tp->t_flags |= TF_REQ_SCALE;
925 	if (tcp_do_rfc1323 && tcp_do_timestamps)
926 		tp->t_flags |= TF_REQ_TSTMP;
927 	if (tcp_do_sack == 2)
928 		tp->t_flags |= TF_WILL_SACK;
929 	else if (tcp_do_sack == 1)
930 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
931 	tp->t_flags |= TF_CANT_TXSACK;
932 	switch (family) {
933 	case PF_INET:
934 		tp->t_inpcb = (struct inpcb *)aux;
935 		tp->t_mtudisc = ip_mtudisc;
936 		break;
937 #ifdef INET6
938 	case PF_INET6:
939 		tp->t_in6pcb = (struct in6pcb *)aux;
940 		/* for IPv6, always try to run path MTU discovery */
941 		tp->t_mtudisc = 1;
942 		break;
943 #endif
944 	}
945 	/*
946 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
947 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
948 	 * reasonable initial retransmit time.
949 	 */
950 	tp->t_srtt = TCPTV_SRTTBASE;
951 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
952 	tp->t_rttmin = TCPTV_MIN;
953 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
954 	    TCPTV_MIN, TCPTV_REXMTMAX);
955 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
956 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
957 	if (family == AF_INET) {
958 		struct inpcb *inp = (struct inpcb *)aux;
959 		inp->inp_ip.ip_ttl = ip_defttl;
960 		inp->inp_ppcb = (caddr_t)tp;
961 	}
962 #ifdef INET6
963 	else if (family == AF_INET6) {
964 		struct in6pcb *in6p = (struct in6pcb *)aux;
965 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
966 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
967 					       : NULL);
968 		in6p->in6p_ppcb = (caddr_t)tp;
969 	}
970 #endif
971 
972 	/*
973 	 * Initialize our timebase.  When we send timestamps, we take
974 	 * the delta from tcp_now -- this means each connection always
975 	 * gets a timebase of 0, which makes it, among other things,
976 	 * more difficult to determine how long a system has been up,
977 	 * and thus how many TCP sequence increments have occurred.
978 	 */
979 	tp->ts_timebase = tcp_now;
980 
981 	return (tp);
982 }
983 
984 /*
985  * Drop a TCP connection, reporting
986  * the specified error.  If connection is synchronized,
987  * then send a RST to peer.
988  */
989 struct tcpcb *
990 tcp_drop(tp, errno)
991 	struct tcpcb *tp;
992 	int errno;
993 {
994 	struct socket *so = NULL;
995 
996 #ifdef DIAGNOSTIC
997 	if (tp->t_inpcb && tp->t_in6pcb)
998 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
999 #endif
1000 #ifdef INET
1001 	if (tp->t_inpcb)
1002 		so = tp->t_inpcb->inp_socket;
1003 #endif
1004 #ifdef INET6
1005 	if (tp->t_in6pcb)
1006 		so = tp->t_in6pcb->in6p_socket;
1007 #endif
1008 	if (!so)
1009 		return NULL;
1010 
1011 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1012 		tp->t_state = TCPS_CLOSED;
1013 		(void) tcp_output(tp);
1014 		tcpstat.tcps_drops++;
1015 	} else
1016 		tcpstat.tcps_conndrops++;
1017 	if (errno == ETIMEDOUT && tp->t_softerror)
1018 		errno = tp->t_softerror;
1019 	so->so_error = errno;
1020 	return (tcp_close(tp));
1021 }
1022 
1023 /*
1024  * Close a TCP control block:
1025  *	discard all space held by the tcp
1026  *	discard internet protocol block
1027  *	wake up any sleepers
1028  */
1029 struct tcpcb *
1030 tcp_close(tp)
1031 	struct tcpcb *tp;
1032 {
1033 	struct inpcb *inp;
1034 #ifdef INET6
1035 	struct in6pcb *in6p;
1036 #endif
1037 	struct socket *so;
1038 #ifdef RTV_RTT
1039 	struct rtentry *rt;
1040 #endif
1041 	struct route *ro;
1042 
1043 	inp = tp->t_inpcb;
1044 #ifdef INET6
1045 	in6p = tp->t_in6pcb;
1046 #endif
1047 	so = NULL;
1048 	ro = NULL;
1049 	if (inp) {
1050 		so = inp->inp_socket;
1051 		ro = &inp->inp_route;
1052 	}
1053 #ifdef INET6
1054 	else if (in6p) {
1055 		so = in6p->in6p_socket;
1056 		ro = (struct route *)&in6p->in6p_route;
1057 	}
1058 #endif
1059 
1060 #ifdef RTV_RTT
1061 	/*
1062 	 * If we sent enough data to get some meaningful characteristics,
1063 	 * save them in the routing entry.  'Enough' is arbitrarily
1064 	 * defined as the sendpipesize (default 4K) * 16.  This would
1065 	 * give us 16 rtt samples assuming we only get one sample per
1066 	 * window (the usual case on a long haul net).  16 samples is
1067 	 * enough for the srtt filter to converge to within 5% of the correct
1068 	 * value; fewer samples and we could save a very bogus rtt.
1069 	 *
1070 	 * Don't update the default route's characteristics and don't
1071 	 * update anything that the user "locked".
1072 	 */
1073 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1074 	    ro && (rt = ro->ro_rt) &&
1075 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1076 		u_long i = 0;
1077 
1078 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1079 			i = tp->t_srtt *
1080 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1081 			if (rt->rt_rmx.rmx_rtt && i)
1082 				/*
1083 				 * filter this update to half the old & half
1084 				 * the new values, converting scale.
1085 				 * See route.h and tcp_var.h for a
1086 				 * description of the scaling constants.
1087 				 */
1088 				rt->rt_rmx.rmx_rtt =
1089 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1090 			else
1091 				rt->rt_rmx.rmx_rtt = i;
1092 		}
1093 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1094 			i = tp->t_rttvar *
1095 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1096 			if (rt->rt_rmx.rmx_rttvar && i)
1097 				rt->rt_rmx.rmx_rttvar =
1098 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1099 			else
1100 				rt->rt_rmx.rmx_rttvar = i;
1101 		}
1102 		/*
1103 		 * update the pipelimit (ssthresh) if it has been updated
1104 		 * already or if a pipesize was specified & the threshhold
1105 		 * got below half the pipesize.  I.e., wait for bad news
1106 		 * before we start updating, then update on both good
1107 		 * and bad news.
1108 		 */
1109 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1110 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1111 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1112 			/*
1113 			 * convert the limit from user data bytes to
1114 			 * packets then to packet data bytes.
1115 			 */
1116 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1117 			if (i < 2)
1118 				i = 2;
1119 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1120 			if (rt->rt_rmx.rmx_ssthresh)
1121 				rt->rt_rmx.rmx_ssthresh =
1122 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1123 			else
1124 				rt->rt_rmx.rmx_ssthresh = i;
1125 		}
1126 	}
1127 #endif /* RTV_RTT */
1128 	/* free the reassembly queue, if any */
1129 	TCP_REASS_LOCK(tp);
1130 	(void) tcp_freeq(tp);
1131 	TCP_REASS_UNLOCK(tp);
1132 
1133 	tcp_canceltimers(tp);
1134 	TCP_CLEAR_DELACK(tp);
1135 	syn_cache_cleanup(tp);
1136 
1137 	if (tp->t_template) {
1138 		m_free(tp->t_template);
1139 		tp->t_template = NULL;
1140 	}
1141 	pool_put(&tcpcb_pool, tp);
1142 	if (inp) {
1143 		inp->inp_ppcb = 0;
1144 		soisdisconnected(so);
1145 		in_pcbdetach(inp);
1146 	}
1147 #ifdef INET6
1148 	else if (in6p) {
1149 		in6p->in6p_ppcb = 0;
1150 		soisdisconnected(so);
1151 		in6_pcbdetach(in6p);
1152 	}
1153 #endif
1154 	tcpstat.tcps_closed++;
1155 	return ((struct tcpcb *)0);
1156 }
1157 
1158 int
1159 tcp_freeq(tp)
1160 	struct tcpcb *tp;
1161 {
1162 	struct ipqent *qe;
1163 	int rv = 0;
1164 #ifdef TCPREASS_DEBUG
1165 	int i = 0;
1166 #endif
1167 
1168 	TCP_REASS_LOCK_CHECK(tp);
1169 
1170 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1171 #ifdef TCPREASS_DEBUG
1172 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1173 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1174 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1175 #endif
1176 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1177 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1178 		m_freem(qe->ipqe_m);
1179 		pool_put(&ipqent_pool, qe);
1180 		rv = 1;
1181 	}
1182 	return (rv);
1183 }
1184 
1185 /*
1186  * Protocol drain routine.  Called when memory is in short supply.
1187  */
1188 void
1189 tcp_drain()
1190 {
1191 	struct inpcb *inp;
1192 	struct tcpcb *tp;
1193 
1194 	/*
1195 	 * Free the sequence queue of all TCP connections.
1196 	 */
1197 	inp = CIRCLEQ_FIRST(&tcbtable.inpt_queue);
1198 	if (inp)						/* XXX */
1199 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1200 		if ((tp = intotcpcb(inp)) != NULL) {
1201 			/*
1202 			 * We may be called from a device's interrupt
1203 			 * context.  If the tcpcb is already busy,
1204 			 * just bail out now.
1205 			 */
1206 			if (tcp_reass_lock_try(tp) == 0)
1207 				continue;
1208 			if (tcp_freeq(tp))
1209 				tcpstat.tcps_connsdrained++;
1210 			TCP_REASS_UNLOCK(tp);
1211 		}
1212 	}
1213 }
1214 
1215 #ifdef INET6
1216 void
1217 tcp6_drain()
1218 {
1219 	struct in6pcb *in6p;
1220 	struct tcpcb *tp;
1221 	struct in6pcb *head = &tcb6;
1222 
1223 	/*
1224 	 * Free the sequence queue of all TCP connections.
1225 	 */
1226 	for (in6p = head->in6p_next; in6p != head; in6p = in6p->in6p_next) {
1227 		if ((tp = in6totcpcb(in6p)) != NULL) {
1228 			/*
1229 			 * We may be called from a device's interrupt
1230 			 * context.  If the tcpcb is already busy,
1231 			 * just bail out now.
1232 			 */
1233 			if (tcp_reass_lock_try(tp) == 0)
1234 				continue;
1235 			if (tcp_freeq(tp))
1236 				tcpstat.tcps_connsdrained++;
1237 			TCP_REASS_UNLOCK(tp);
1238 		}
1239 	}
1240 }
1241 #endif
1242 
1243 /*
1244  * Notify a tcp user of an asynchronous error;
1245  * store error as soft error, but wake up user
1246  * (for now, won't do anything until can select for soft error).
1247  */
1248 void
1249 tcp_notify(inp, error)
1250 	struct inpcb *inp;
1251 	int error;
1252 {
1253 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1254 	struct socket *so = inp->inp_socket;
1255 
1256 	/*
1257 	 * Ignore some errors if we are hooked up.
1258 	 * If connection hasn't completed, has retransmitted several times,
1259 	 * and receives a second error, give up now.  This is better
1260 	 * than waiting a long time to establish a connection that
1261 	 * can never complete.
1262 	 */
1263 	if (tp->t_state == TCPS_ESTABLISHED &&
1264 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1265 	      error == EHOSTDOWN)) {
1266 		return;
1267 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1268 	    tp->t_rxtshift > 3 && tp->t_softerror)
1269 		so->so_error = error;
1270 	else
1271 		tp->t_softerror = error;
1272 	wakeup((caddr_t) &so->so_timeo);
1273 	sorwakeup(so);
1274 	sowwakeup(so);
1275 }
1276 
1277 #ifdef INET6
1278 void
1279 tcp6_notify(in6p, error)
1280 	struct in6pcb *in6p;
1281 	int error;
1282 {
1283 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1284 	struct socket *so = in6p->in6p_socket;
1285 
1286 	/*
1287 	 * Ignore some errors if we are hooked up.
1288 	 * If connection hasn't completed, has retransmitted several times,
1289 	 * and receives a second error, give up now.  This is better
1290 	 * than waiting a long time to establish a connection that
1291 	 * can never complete.
1292 	 */
1293 	if (tp->t_state == TCPS_ESTABLISHED &&
1294 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1295 	      error == EHOSTDOWN)) {
1296 		return;
1297 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1298 	    tp->t_rxtshift > 3 && tp->t_softerror)
1299 		so->so_error = error;
1300 	else
1301 		tp->t_softerror = error;
1302 	wakeup((caddr_t) &so->so_timeo);
1303 	sorwakeup(so);
1304 	sowwakeup(so);
1305 }
1306 #endif
1307 
1308 #ifdef INET6
1309 void
1310 tcp6_ctlinput(cmd, sa, d)
1311 	int cmd;
1312 	struct sockaddr *sa;
1313 	void *d;
1314 {
1315 	struct tcphdr th;
1316 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1317 	int nmatch;
1318 	struct ip6_hdr *ip6;
1319 	const struct sockaddr_in6 *sa6_src = NULL;
1320 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1321 	struct mbuf *m;
1322 	int off;
1323 
1324 	if (sa->sa_family != AF_INET6 ||
1325 	    sa->sa_len != sizeof(struct sockaddr_in6))
1326 		return;
1327 	if ((unsigned)cmd >= PRC_NCMDS)
1328 		return;
1329 	else if (cmd == PRC_QUENCH) {
1330 		/* XXX there's no PRC_QUENCH in IPv6 */
1331 		notify = tcp6_quench;
1332 	} else if (PRC_IS_REDIRECT(cmd))
1333 		notify = in6_rtchange, d = NULL;
1334 	else if (cmd == PRC_MSGSIZE)
1335 		; /* special code is present, see below */
1336 	else if (cmd == PRC_HOSTDEAD)
1337 		d = NULL;
1338 	else if (inet6ctlerrmap[cmd] == 0)
1339 		return;
1340 
1341 	/* if the parameter is from icmp6, decode it. */
1342 	if (d != NULL) {
1343 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1344 		m = ip6cp->ip6c_m;
1345 		ip6 = ip6cp->ip6c_ip6;
1346 		off = ip6cp->ip6c_off;
1347 		sa6_src = ip6cp->ip6c_src;
1348 	} else {
1349 		m = NULL;
1350 		ip6 = NULL;
1351 		sa6_src = &sa6_any;
1352 	}
1353 
1354 	if (ip6) {
1355 		/*
1356 		 * XXX: We assume that when ip6 is non NULL,
1357 		 * M and OFF are valid.
1358 		 */
1359 
1360 		/* check if we can safely examine src and dst ports */
1361 		if (m->m_pkthdr.len < off + sizeof(th)) {
1362 			if (cmd == PRC_MSGSIZE)
1363 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1364 			return;
1365 		}
1366 
1367 		bzero(&th, sizeof(th));
1368 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1369 
1370 		if (cmd == PRC_MSGSIZE) {
1371 			int valid = 0;
1372 
1373 			/*
1374 			 * Check to see if we have a valid TCP connection
1375 			 * corresponding to the address in the ICMPv6 message
1376 			 * payload.
1377 			 */
1378 			if (in6_pcblookup_connect(&tcb6, &sa6->sin6_addr,
1379 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1380 			    th.th_sport, 0))
1381 				valid++;
1382 
1383 			/*
1384 			 * Depending on the value of "valid" and routing table
1385 			 * size (mtudisc_{hi,lo}wat), we will:
1386 			 * - recalcurate the new MTU and create the
1387 			 *   corresponding routing entry, or
1388 			 * - ignore the MTU change notification.
1389 			 */
1390 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1391 
1392 			/*
1393 			 * no need to call in6_pcbnotify, it should have been
1394 			 * called via callback if necessary
1395 			 */
1396 			return;
1397 		}
1398 
1399 		nmatch = in6_pcbnotify(&tcb6, sa, th.th_dport,
1400 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1401 		if (nmatch == 0 && syn_cache_count &&
1402 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1403 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1404 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1405 			syn_cache_unreach((struct sockaddr *)sa6_src,
1406 					  sa, &th);
1407 	} else {
1408 		(void) in6_pcbnotify(&tcb6, sa, 0, (struct sockaddr *)sa6_src,
1409 		    0, cmd, NULL, notify);
1410 	}
1411 }
1412 #endif
1413 
1414 #ifdef INET
1415 /* assumes that ip header and tcp header are contiguous on mbuf */
1416 void *
1417 tcp_ctlinput(cmd, sa, v)
1418 	int cmd;
1419 	struct sockaddr *sa;
1420 	void *v;
1421 {
1422 	struct ip *ip = v;
1423 	struct tcphdr *th;
1424 	struct icmp *icp;
1425 	extern const int inetctlerrmap[];
1426 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1427 	int errno;
1428 	int nmatch;
1429 #ifdef INET6
1430 	struct in6_addr src6, dst6;
1431 #endif
1432 
1433 	if (sa->sa_family != AF_INET ||
1434 	    sa->sa_len != sizeof(struct sockaddr_in))
1435 		return NULL;
1436 	if ((unsigned)cmd >= PRC_NCMDS)
1437 		return NULL;
1438 	errno = inetctlerrmap[cmd];
1439 	if (cmd == PRC_QUENCH)
1440 		notify = tcp_quench;
1441 	else if (PRC_IS_REDIRECT(cmd))
1442 		notify = in_rtchange, ip = 0;
1443 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1444 		/*
1445 		 * Check to see if we have a valid TCP connection
1446 		 * corresponding to the address in the ICMP message
1447 		 * payload.
1448 		 *
1449 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1450 		 */
1451 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1452 #ifdef INET6
1453 		memset(&src6, 0, sizeof(src6));
1454 		memset(&dst6, 0, sizeof(dst6));
1455 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1456 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1457 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1458 #endif
1459 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1460 		    ip->ip_src, th->th_sport) != NULL)
1461 			;
1462 #ifdef INET6
1463 		else if (in6_pcblookup_connect(&tcb6, &dst6,
1464 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1465 			;
1466 #endif
1467 		else
1468 			return NULL;
1469 
1470 		/*
1471 		 * Now that we've validated that we are actually communicating
1472 		 * with the host indicated in the ICMP message, locate the
1473 		 * ICMP header, recalculate the new MTU, and create the
1474 		 * corresponding routing entry.
1475 		 */
1476 		icp = (struct icmp *)((caddr_t)ip -
1477 		    offsetof(struct icmp, icmp_ip));
1478 		icmp_mtudisc(icp, ip->ip_dst);
1479 
1480 		return NULL;
1481 	} else if (cmd == PRC_HOSTDEAD)
1482 		ip = 0;
1483 	else if (errno == 0)
1484 		return NULL;
1485 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1486 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1487 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1488 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1489 		if (nmatch == 0 && syn_cache_count &&
1490 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1491 		    inetctlerrmap[cmd] == ENETUNREACH ||
1492 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1493 			struct sockaddr_in sin;
1494 			bzero(&sin, sizeof(sin));
1495 			sin.sin_len = sizeof(sin);
1496 			sin.sin_family = AF_INET;
1497 			sin.sin_port = th->th_sport;
1498 			sin.sin_addr = ip->ip_src;
1499 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1500 		}
1501 
1502 		/* XXX mapped address case */
1503 	} else
1504 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1505 		    notify);
1506 	return NULL;
1507 }
1508 
1509 /*
1510  * When a source quence is received, we are being notifed of congestion.
1511  * Close the congestion window down to the Loss Window (one segment).
1512  * We will gradually open it again as we proceed.
1513  */
1514 void
1515 tcp_quench(inp, errno)
1516 	struct inpcb *inp;
1517 	int errno;
1518 {
1519 	struct tcpcb *tp = intotcpcb(inp);
1520 
1521 	if (tp)
1522 		tp->snd_cwnd = tp->t_segsz;
1523 }
1524 #endif
1525 
1526 #ifdef INET6
1527 void
1528 tcp6_quench(in6p, errno)
1529 	struct in6pcb *in6p;
1530 	int errno;
1531 {
1532 	struct tcpcb *tp = in6totcpcb(in6p);
1533 
1534 	if (tp)
1535 		tp->snd_cwnd = tp->t_segsz;
1536 }
1537 #endif
1538 
1539 #ifdef INET
1540 /*
1541  * Path MTU Discovery handlers.
1542  */
1543 void
1544 tcp_mtudisc_callback(faddr)
1545 	struct in_addr faddr;
1546 {
1547 #ifdef INET6
1548 	struct in6_addr in6;
1549 #endif
1550 
1551 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1552 #ifdef INET6
1553 	memset(&in6, 0, sizeof(in6));
1554 	in6.s6_addr16[5] = 0xffff;
1555 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1556 	tcp6_mtudisc_callback(&in6);
1557 #endif
1558 }
1559 
1560 /*
1561  * On receipt of path MTU corrections, flush old route and replace it
1562  * with the new one.  Retransmit all unacknowledged packets, to ensure
1563  * that all packets will be received.
1564  */
1565 void
1566 tcp_mtudisc(inp, errno)
1567 	struct inpcb *inp;
1568 	int errno;
1569 {
1570 	struct tcpcb *tp = intotcpcb(inp);
1571 	struct rtentry *rt = in_pcbrtentry(inp);
1572 
1573 	if (tp != 0) {
1574 		if (rt != 0) {
1575 			/*
1576 			 * If this was not a host route, remove and realloc.
1577 			 */
1578 			if ((rt->rt_flags & RTF_HOST) == 0) {
1579 				in_rtchange(inp, errno);
1580 				if ((rt = in_pcbrtentry(inp)) == 0)
1581 					return;
1582 			}
1583 
1584 			/*
1585 			 * Slow start out of the error condition.  We
1586 			 * use the MTU because we know it's smaller
1587 			 * than the previously transmitted segment.
1588 			 *
1589 			 * Note: This is more conservative than the
1590 			 * suggestion in draft-floyd-incr-init-win-03.
1591 			 */
1592 			if (rt->rt_rmx.rmx_mtu != 0)
1593 				tp->snd_cwnd =
1594 				    TCP_INITIAL_WINDOW(tcp_init_win,
1595 				    rt->rt_rmx.rmx_mtu);
1596 		}
1597 
1598 		/*
1599 		 * Resend unacknowledged packets.
1600 		 */
1601 		tp->snd_nxt = tp->snd_una;
1602 		tcp_output(tp);
1603 	}
1604 }
1605 #endif
1606 
1607 #ifdef INET6
1608 /*
1609  * Path MTU Discovery handlers.
1610  */
1611 void
1612 tcp6_mtudisc_callback(faddr)
1613 	struct in6_addr *faddr;
1614 {
1615 	struct sockaddr_in6 sin6;
1616 
1617 	bzero(&sin6, sizeof(sin6));
1618 	sin6.sin6_family = AF_INET6;
1619 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1620 	sin6.sin6_addr = *faddr;
1621 	(void) in6_pcbnotify(&tcb6, (struct sockaddr *)&sin6, 0,
1622 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1623 }
1624 
1625 void
1626 tcp6_mtudisc(in6p, errno)
1627 	struct in6pcb *in6p;
1628 	int errno;
1629 {
1630 	struct tcpcb *tp = in6totcpcb(in6p);
1631 	struct rtentry *rt = in6_pcbrtentry(in6p);
1632 
1633 	if (tp != 0) {
1634 		if (rt != 0) {
1635 			/*
1636 			 * If this was not a host route, remove and realloc.
1637 			 */
1638 			if ((rt->rt_flags & RTF_HOST) == 0) {
1639 				in6_rtchange(in6p, errno);
1640 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1641 					return;
1642 			}
1643 
1644 			/*
1645 			 * Slow start out of the error condition.  We
1646 			 * use the MTU because we know it's smaller
1647 			 * than the previously transmitted segment.
1648 			 *
1649 			 * Note: This is more conservative than the
1650 			 * suggestion in draft-floyd-incr-init-win-03.
1651 			 */
1652 			if (rt->rt_rmx.rmx_mtu != 0)
1653 				tp->snd_cwnd =
1654 				    TCP_INITIAL_WINDOW(tcp_init_win,
1655 				    rt->rt_rmx.rmx_mtu);
1656 		}
1657 
1658 		/*
1659 		 * Resend unacknowledged packets.
1660 		 */
1661 		tp->snd_nxt = tp->snd_una;
1662 		tcp_output(tp);
1663 	}
1664 }
1665 #endif /* INET6 */
1666 
1667 /*
1668  * Compute the MSS to advertise to the peer.  Called only during
1669  * the 3-way handshake.  If we are the server (peer initiated
1670  * connection), we are called with a pointer to the interface
1671  * on which the SYN packet arrived.  If we are the client (we
1672  * initiated connection), we are called with a pointer to the
1673  * interface out which this connection should go.
1674  *
1675  * NOTE: Do not subtract IP option/extension header size nor IPsec
1676  * header size from MSS advertisement.  MSS option must hold the maximum
1677  * segment size we can accept, so it must always be:
1678  *	 max(if mtu) - ip header - tcp header
1679  */
1680 u_long
1681 tcp_mss_to_advertise(ifp, af)
1682 	const struct ifnet *ifp;
1683 	int af;
1684 {
1685 	extern u_long in_maxmtu;
1686 	u_long mss = 0;
1687 	u_long hdrsiz;
1688 
1689 	/*
1690 	 * In order to avoid defeating path MTU discovery on the peer,
1691 	 * we advertise the max MTU of all attached networks as our MSS,
1692 	 * per RFC 1191, section 3.1.
1693 	 *
1694 	 * We provide the option to advertise just the MTU of
1695 	 * the interface on which we hope this connection will
1696 	 * be receiving.  If we are responding to a SYN, we
1697 	 * will have a pretty good idea about this, but when
1698 	 * initiating a connection there is a bit more doubt.
1699 	 *
1700 	 * We also need to ensure that loopback has a large enough
1701 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1702 	 */
1703 
1704 	if (ifp != NULL)
1705 		switch (af) {
1706 		case AF_INET:
1707 			mss = ifp->if_mtu;
1708 			break;
1709 #ifdef INET6
1710 		case AF_INET6:
1711 			mss = IN6_LINKMTU(ifp);
1712 			break;
1713 #endif
1714 		}
1715 
1716 	if (tcp_mss_ifmtu == 0)
1717 		switch (af) {
1718 		case AF_INET:
1719 			mss = max(in_maxmtu, mss);
1720 			break;
1721 #ifdef INET6
1722 		case AF_INET6:
1723 			mss = max(in6_maxmtu, mss);
1724 			break;
1725 #endif
1726 		}
1727 
1728 	switch (af) {
1729 	case AF_INET:
1730 		hdrsiz = sizeof(struct ip);
1731 		break;
1732 #ifdef INET6
1733 	case AF_INET6:
1734 		hdrsiz = sizeof(struct ip6_hdr);
1735 		break;
1736 #endif
1737 	default:
1738 		hdrsiz = 0;
1739 		break;
1740 	}
1741 	hdrsiz += sizeof(struct tcphdr);
1742 	if (mss > hdrsiz)
1743 		mss -= hdrsiz;
1744 
1745 	mss = max(tcp_mssdflt, mss);
1746 	return (mss);
1747 }
1748 
1749 /*
1750  * Set connection variables based on the peer's advertised MSS.
1751  * We are passed the TCPCB for the actual connection.  If we
1752  * are the server, we are called by the compressed state engine
1753  * when the 3-way handshake is complete.  If we are the client,
1754  * we are called when we receive the SYN,ACK from the server.
1755  *
1756  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1757  * before this routine is called!
1758  */
1759 void
1760 tcp_mss_from_peer(tp, offer)
1761 	struct tcpcb *tp;
1762 	int offer;
1763 {
1764 	struct socket *so;
1765 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1766 	struct rtentry *rt;
1767 #endif
1768 	u_long bufsize;
1769 	int mss;
1770 
1771 #ifdef DIAGNOSTIC
1772 	if (tp->t_inpcb && tp->t_in6pcb)
1773 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1774 #endif
1775 	so = NULL;
1776 	rt = NULL;
1777 #ifdef INET
1778 	if (tp->t_inpcb) {
1779 		so = tp->t_inpcb->inp_socket;
1780 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1781 		rt = in_pcbrtentry(tp->t_inpcb);
1782 #endif
1783 	}
1784 #endif
1785 #ifdef INET6
1786 	if (tp->t_in6pcb) {
1787 		so = tp->t_in6pcb->in6p_socket;
1788 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1789 		rt = in6_pcbrtentry(tp->t_in6pcb);
1790 #endif
1791 	}
1792 #endif
1793 
1794 	/*
1795 	 * As per RFC1122, use the default MSS value, unless they
1796 	 * sent us an offer.  Do not accept offers less than 32 bytes.
1797 	 */
1798 	mss = tcp_mssdflt;
1799 	if (offer)
1800 		mss = offer;
1801 	mss = max(mss, 32);		/* sanity */
1802 	tp->t_peermss = mss;
1803 	mss -= tcp_optlen(tp);
1804 #ifdef INET
1805 	if (tp->t_inpcb)
1806 		mss -= ip_optlen(tp->t_inpcb);
1807 #endif
1808 #ifdef INET6
1809 	if (tp->t_in6pcb)
1810 		mss -= ip6_optlen(tp->t_in6pcb);
1811 #endif
1812 
1813 	/*
1814 	 * If there's a pipesize, change the socket buffer to that size.
1815 	 * Make the socket buffer an integral number of MSS units.  If
1816 	 * the MSS is larger than the socket buffer, artificially decrease
1817 	 * the MSS.
1818 	 */
1819 #ifdef RTV_SPIPE
1820 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1821 		bufsize = rt->rt_rmx.rmx_sendpipe;
1822 	else
1823 #endif
1824 		bufsize = so->so_snd.sb_hiwat;
1825 	if (bufsize < mss)
1826 		mss = bufsize;
1827 	else {
1828 		bufsize = roundup(bufsize, mss);
1829 		if (bufsize > sb_max)
1830 			bufsize = sb_max;
1831 		(void) sbreserve(&so->so_snd, bufsize);
1832 	}
1833 	tp->t_segsz = mss;
1834 
1835 #ifdef RTV_SSTHRESH
1836 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1837 		/*
1838 		 * There's some sort of gateway or interface buffer
1839 		 * limit on the path.  Use this to set the slow
1840 		 * start threshold, but set the threshold to no less
1841 		 * than 2 * MSS.
1842 		 */
1843 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1844 	}
1845 #endif
1846 }
1847 
1848 /*
1849  * Processing necessary when a TCP connection is established.
1850  */
1851 void
1852 tcp_established(tp)
1853 	struct tcpcb *tp;
1854 {
1855 	struct socket *so;
1856 #ifdef RTV_RPIPE
1857 	struct rtentry *rt;
1858 #endif
1859 	u_long bufsize;
1860 
1861 #ifdef DIAGNOSTIC
1862 	if (tp->t_inpcb && tp->t_in6pcb)
1863 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1864 #endif
1865 	so = NULL;
1866 	rt = NULL;
1867 #ifdef INET
1868 	if (tp->t_inpcb) {
1869 		so = tp->t_inpcb->inp_socket;
1870 #if defined(RTV_RPIPE)
1871 		rt = in_pcbrtentry(tp->t_inpcb);
1872 #endif
1873 	}
1874 #endif
1875 #ifdef INET6
1876 	if (tp->t_in6pcb) {
1877 		so = tp->t_in6pcb->in6p_socket;
1878 #if defined(RTV_RPIPE)
1879 		rt = in6_pcbrtentry(tp->t_in6pcb);
1880 #endif
1881 	}
1882 #endif
1883 
1884 	tp->t_state = TCPS_ESTABLISHED;
1885 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1886 
1887 #ifdef RTV_RPIPE
1888 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1889 		bufsize = rt->rt_rmx.rmx_recvpipe;
1890 	else
1891 #endif
1892 		bufsize = so->so_rcv.sb_hiwat;
1893 	if (bufsize > tp->t_ourmss) {
1894 		bufsize = roundup(bufsize, tp->t_ourmss);
1895 		if (bufsize > sb_max)
1896 			bufsize = sb_max;
1897 		(void) sbreserve(&so->so_rcv, bufsize);
1898 	}
1899 }
1900 
1901 /*
1902  * Check if there's an initial rtt or rttvar.  Convert from the
1903  * route-table units to scaled multiples of the slow timeout timer.
1904  * Called only during the 3-way handshake.
1905  */
1906 void
1907 tcp_rmx_rtt(tp)
1908 	struct tcpcb *tp;
1909 {
1910 #ifdef RTV_RTT
1911 	struct rtentry *rt = NULL;
1912 	int rtt;
1913 
1914 #ifdef DIAGNOSTIC
1915 	if (tp->t_inpcb && tp->t_in6pcb)
1916 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1917 #endif
1918 #ifdef INET
1919 	if (tp->t_inpcb)
1920 		rt = in_pcbrtentry(tp->t_inpcb);
1921 #endif
1922 #ifdef INET6
1923 	if (tp->t_in6pcb)
1924 		rt = in6_pcbrtentry(tp->t_in6pcb);
1925 #endif
1926 	if (rt == NULL)
1927 		return;
1928 
1929 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1930 		/*
1931 		 * XXX The lock bit for MTU indicates that the value
1932 		 * is also a minimum value; this is subject to time.
1933 		 */
1934 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1935 			TCPT_RANGESET(tp->t_rttmin,
1936 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1937 			    TCPTV_MIN, TCPTV_REXMTMAX);
1938 		tp->t_srtt = rtt /
1939 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1940 		if (rt->rt_rmx.rmx_rttvar) {
1941 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1942 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1943 				(TCP_RTTVAR_SHIFT + 2));
1944 		} else {
1945 			/* Default variation is +- 1 rtt */
1946 			tp->t_rttvar =
1947 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1948 		}
1949 		TCPT_RANGESET(tp->t_rxtcur,
1950 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1951 		    tp->t_rttmin, TCPTV_REXMTMAX);
1952 	}
1953 #endif
1954 }
1955 
1956 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1957 #if NRND > 0
1958 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1959 #endif
1960 
1961 /*
1962  * Get a new sequence value given a tcp control block
1963  */
1964 tcp_seq
1965 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1966 {
1967 
1968 #ifdef INET
1969 	if (tp->t_inpcb != NULL) {
1970 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1971 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1972 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1973 		    addin));
1974 	}
1975 #endif
1976 #ifdef INET6
1977 	if (tp->t_in6pcb != NULL) {
1978 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
1979 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
1980 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
1981 		    addin));
1982 	}
1983 #endif
1984 	/* Not possible. */
1985 	panic("tcp_new_iss");
1986 }
1987 
1988 /*
1989  * This routine actually generates a new TCP initial sequence number.
1990  */
1991 tcp_seq
1992 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
1993     size_t addrsz, tcp_seq addin)
1994 {
1995 	tcp_seq tcp_iss;
1996 
1997 #if NRND > 0
1998 	static int beenhere;
1999 
2000 	/*
2001 	 * If we haven't been here before, initialize our cryptographic
2002 	 * hash secret.
2003 	 */
2004 	if (beenhere == 0) {
2005 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2006 		    RND_EXTRACT_ANY);
2007 		beenhere = 1;
2008 	}
2009 
2010 	if (tcp_do_rfc1948) {
2011 		MD5_CTX ctx;
2012 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2013 
2014 		/*
2015 		 * Compute the base value of the ISS.  It is a hash
2016 		 * of (saddr, sport, daddr, dport, secret).
2017 		 */
2018 		MD5Init(&ctx);
2019 
2020 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2021 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2022 
2023 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2024 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2025 
2026 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2027 
2028 		MD5Final(hash, &ctx);
2029 
2030 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2031 
2032 		/*
2033 		 * Now increment our "timer", and add it in to
2034 		 * the computed value.
2035 		 *
2036 		 * XXX Use `addin'?
2037 		 * XXX TCP_ISSINCR too large to use?
2038 		 */
2039 		tcp_iss_seq += TCP_ISSINCR;
2040 #ifdef TCPISS_DEBUG
2041 		printf("ISS hash 0x%08x, ", tcp_iss);
2042 #endif
2043 		tcp_iss += tcp_iss_seq + addin;
2044 #ifdef TCPISS_DEBUG
2045 		printf("new ISS 0x%08x\n", tcp_iss);
2046 #endif
2047 	} else
2048 #endif /* NRND > 0 */
2049 	{
2050 		/*
2051 		 * Randomize.
2052 		 */
2053 #if NRND > 0
2054 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2055 #else
2056 		tcp_iss = arc4random();
2057 #endif
2058 
2059 		/*
2060 		 * If we were asked to add some amount to a known value,
2061 		 * we will take a random value obtained above, mask off
2062 		 * the upper bits, and add in the known value.  We also
2063 		 * add in a constant to ensure that we are at least a
2064 		 * certain distance from the original value.
2065 		 *
2066 		 * This is used when an old connection is in timed wait
2067 		 * and we have a new one coming in, for instance.
2068 		 */
2069 		if (addin != 0) {
2070 #ifdef TCPISS_DEBUG
2071 			printf("Random %08x, ", tcp_iss);
2072 #endif
2073 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2074 			tcp_iss += addin + TCP_ISSINCR;
2075 #ifdef TCPISS_DEBUG
2076 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2077 #endif
2078 		} else {
2079 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2080 			tcp_iss += tcp_iss_seq;
2081 			tcp_iss_seq += TCP_ISSINCR;
2082 #ifdef TCPISS_DEBUG
2083 			printf("ISS %08x\n", tcp_iss);
2084 #endif
2085 		}
2086 	}
2087 
2088 	if (tcp_compat_42) {
2089 		/*
2090 		 * Limit it to the positive range for really old TCP
2091 		 * implementations.
2092 		 * Just AND off the top bit instead of checking if
2093 		 * is set first - saves a branch 50% of the time.
2094 		 */
2095 		tcp_iss &= 0x7fffffff;		/* XXX */
2096 	}
2097 
2098 	return (tcp_iss);
2099 }
2100 
2101 #ifdef IPSEC
2102 /* compute ESP/AH header size for TCP, including outer IP header. */
2103 size_t
2104 ipsec4_hdrsiz_tcp(tp)
2105 	struct tcpcb *tp;
2106 {
2107 	struct inpcb *inp;
2108 	size_t hdrsiz;
2109 
2110 	/* XXX mapped addr case (tp->t_in6pcb) */
2111 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2112 		return 0;
2113 	switch (tp->t_family) {
2114 	case AF_INET:
2115 		/* XXX: should use currect direction. */
2116 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2117 		break;
2118 	default:
2119 		hdrsiz = 0;
2120 		break;
2121 	}
2122 
2123 	return hdrsiz;
2124 }
2125 
2126 #ifdef INET6
2127 size_t
2128 ipsec6_hdrsiz_tcp(tp)
2129 	struct tcpcb *tp;
2130 {
2131 	struct in6pcb *in6p;
2132 	size_t hdrsiz;
2133 
2134 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2135 		return 0;
2136 	switch (tp->t_family) {
2137 	case AF_INET6:
2138 		/* XXX: should use currect direction. */
2139 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2140 		break;
2141 	case AF_INET:
2142 		/* mapped address case - tricky */
2143 	default:
2144 		hdrsiz = 0;
2145 		break;
2146 	}
2147 
2148 	return hdrsiz;
2149 }
2150 #endif
2151 #endif /*IPSEC*/
2152 
2153 /*
2154  * Determine the length of the TCP options for this connection.
2155  *
2156  * XXX:  What do we do for SACK, when we add that?  Just reserve
2157  *       all of the space?  Otherwise we can't exactly be incrementing
2158  *       cwnd by an amount that varies depending on the amount we last
2159  *       had to SACK!
2160  */
2161 
2162 u_int
2163 tcp_optlen(tp)
2164 	struct tcpcb *tp;
2165 {
2166 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2167 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2168 		return TCPOLEN_TSTAMP_APPA;
2169 	else
2170 		return 0;
2171 }
2172