xref: /netbsd-src/sys/netinet/tcp_subr.c (revision aa73cae19608873cc4d1f712c4a0f8f8435f1ffa)
1 /*	$NetBSD: tcp_subr.c,v 1.182 2005/02/26 22:45:12 perry 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 1.182 2005/02/26 22:45:12 perry 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(struct tcpcb *);
220 
221 #ifdef INET
222 void	tcp_mtudisc_callback(struct in_addr);
223 #endif
224 #ifdef INET6
225 void	tcp6_mtudisc_callback(struct in6_addr *);
226 #endif
227 
228 void	tcp_mtudisc(struct inpcb *, int);
229 #ifdef INET6
230 void	tcp6_mtudisc(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(void)
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(struct tcpcb *tp)
395 {
396 	struct inpcb *inp = tp->t_inpcb;
397 #ifdef INET6
398 	struct in6pcb *in6p = tp->t_in6pcb;
399 #endif
400 	struct tcphdr *n;
401 	struct mbuf *m;
402 	int hlen;
403 
404 	switch (tp->t_family) {
405 	case AF_INET:
406 		hlen = sizeof(struct ip);
407 		if (inp)
408 			break;
409 #ifdef INET6
410 		if (in6p) {
411 			/* mapped addr case */
412 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
413 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
414 				break;
415 		}
416 #endif
417 		return NULL;	/*EINVAL*/
418 #ifdef INET6
419 	case AF_INET6:
420 		hlen = sizeof(struct ip6_hdr);
421 		if (in6p) {
422 			/* more sainty check? */
423 			break;
424 		}
425 		return NULL;	/*EINVAL*/
426 #endif
427 	default:
428 		hlen = 0;	/*pacify gcc*/
429 		return NULL;	/*EAFNOSUPPORT*/
430 	}
431 #ifdef DIAGNOSTIC
432 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
433 		panic("mclbytes too small for t_template");
434 #endif
435 	m = tp->t_template;
436 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
437 		;
438 	else {
439 		if (m)
440 			m_freem(m);
441 		m = tp->t_template = NULL;
442 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
443 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
444 			MCLGET(m, M_DONTWAIT);
445 			if ((m->m_flags & M_EXT) == 0) {
446 				m_free(m);
447 				m = NULL;
448 			}
449 		}
450 		if (m == NULL)
451 			return NULL;
452 		MCLAIM(m, &tcp_mowner);
453 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
454 	}
455 
456 	bzero(mtod(m, caddr_t), m->m_len);
457 
458 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
459 
460 	switch (tp->t_family) {
461 	case AF_INET:
462 	    {
463 		struct ipovly *ipov;
464 		mtod(m, struct ip *)->ip_v = 4;
465 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
466 		ipov = mtod(m, struct ipovly *);
467 		ipov->ih_pr = IPPROTO_TCP;
468 		ipov->ih_len = htons(sizeof(struct tcphdr));
469 		if (inp) {
470 			ipov->ih_src = inp->inp_laddr;
471 			ipov->ih_dst = inp->inp_faddr;
472 		}
473 #ifdef INET6
474 		else if (in6p) {
475 			/* mapped addr case */
476 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
477 				sizeof(ipov->ih_src));
478 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
479 				sizeof(ipov->ih_dst));
480 		}
481 #endif
482 		/*
483 		 * Compute the pseudo-header portion of the checksum
484 		 * now.  We incrementally add in the TCP option and
485 		 * payload lengths later, and then compute the TCP
486 		 * checksum right before the packet is sent off onto
487 		 * the wire.
488 		 */
489 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
490 		    ipov->ih_dst.s_addr,
491 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
492 		break;
493 	    }
494 #ifdef INET6
495 	case AF_INET6:
496 	    {
497 		struct ip6_hdr *ip6;
498 		mtod(m, struct ip *)->ip_v = 6;
499 		ip6 = mtod(m, struct ip6_hdr *);
500 		ip6->ip6_nxt = IPPROTO_TCP;
501 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
502 		ip6->ip6_src = in6p->in6p_laddr;
503 		ip6->ip6_dst = in6p->in6p_faddr;
504 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
505 		if (ip6_auto_flowlabel) {
506 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
507 			ip6->ip6_flow |=
508 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
509 		}
510 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
511 		ip6->ip6_vfc |= IPV6_VERSION;
512 
513 		/*
514 		 * Compute the pseudo-header portion of the checksum
515 		 * now.  We incrementally add in the TCP option and
516 		 * payload lengths later, and then compute the TCP
517 		 * checksum right before the packet is sent off onto
518 		 * the wire.
519 		 */
520 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
521 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
522 		    htonl(IPPROTO_TCP));
523 		break;
524 	    }
525 #endif
526 	}
527 	if (inp) {
528 		n->th_sport = inp->inp_lport;
529 		n->th_dport = inp->inp_fport;
530 	}
531 #ifdef INET6
532 	else if (in6p) {
533 		n->th_sport = in6p->in6p_lport;
534 		n->th_dport = in6p->in6p_fport;
535 	}
536 #endif
537 	n->th_seq = 0;
538 	n->th_ack = 0;
539 	n->th_x2 = 0;
540 	n->th_off = 5;
541 	n->th_flags = 0;
542 	n->th_win = 0;
543 	n->th_urp = 0;
544 	return (m);
545 }
546 
547 /*
548  * Send a single message to the TCP at address specified by
549  * the given TCP/IP header.  If m == 0, then we make a copy
550  * of the tcpiphdr at ti and send directly to the addressed host.
551  * This is used to force keep alive messages out using the TCP
552  * template for a connection tp->t_template.  If flags are given
553  * then we send a message back to the TCP which originated the
554  * segment ti, and discard the mbuf containing it and any other
555  * attached mbufs.
556  *
557  * In any case the ack and sequence number of the transmitted
558  * segment are as specified by the parameters.
559  */
560 int
561 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m,
562     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
563 {
564 	struct route *ro;
565 	int error, tlen, win = 0;
566 	int hlen;
567 	struct ip *ip;
568 #ifdef INET6
569 	struct ip6_hdr *ip6;
570 #endif
571 	int family;	/* family on packet, not inpcb/in6pcb! */
572 	struct tcphdr *th;
573 	struct socket *so;
574 
575 	if (tp != NULL && (flags & TH_RST) == 0) {
576 #ifdef DIAGNOSTIC
577 		if (tp->t_inpcb && tp->t_in6pcb)
578 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
579 #endif
580 #ifdef INET
581 		if (tp->t_inpcb)
582 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
583 #endif
584 #ifdef INET6
585 		if (tp->t_in6pcb)
586 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
587 #endif
588 	}
589 
590 	th = NULL;	/* Quell uninitialized warning */
591 	ip = NULL;
592 #ifdef INET6
593 	ip6 = NULL;
594 #endif
595 	if (m == 0) {
596 		if (!template)
597 			return EINVAL;
598 
599 		/* get family information from template */
600 		switch (mtod(template, struct ip *)->ip_v) {
601 		case 4:
602 			family = AF_INET;
603 			hlen = sizeof(struct ip);
604 			break;
605 #ifdef INET6
606 		case 6:
607 			family = AF_INET6;
608 			hlen = sizeof(struct ip6_hdr);
609 			break;
610 #endif
611 		default:
612 			return EAFNOSUPPORT;
613 		}
614 
615 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
616 		if (m) {
617 			MCLAIM(m, &tcp_tx_mowner);
618 			MCLGET(m, M_DONTWAIT);
619 			if ((m->m_flags & M_EXT) == 0) {
620 				m_free(m);
621 				m = NULL;
622 			}
623 		}
624 		if (m == NULL)
625 			return (ENOBUFS);
626 
627 		if (tcp_compat_42)
628 			tlen = 1;
629 		else
630 			tlen = 0;
631 
632 		m->m_data += max_linkhdr;
633 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
634 			template->m_len);
635 		switch (family) {
636 		case AF_INET:
637 			ip = mtod(m, struct ip *);
638 			th = (struct tcphdr *)(ip + 1);
639 			break;
640 #ifdef INET6
641 		case AF_INET6:
642 			ip6 = mtod(m, struct ip6_hdr *);
643 			th = (struct tcphdr *)(ip6 + 1);
644 			break;
645 #endif
646 #if 0
647 		default:
648 			/* noone will visit here */
649 			m_freem(m);
650 			return EAFNOSUPPORT;
651 #endif
652 		}
653 		flags = TH_ACK;
654 	} else {
655 
656 		if ((m->m_flags & M_PKTHDR) == 0) {
657 #if 0
658 			printf("non PKTHDR to tcp_respond\n");
659 #endif
660 			m_freem(m);
661 			return EINVAL;
662 		}
663 #ifdef DIAGNOSTIC
664 		if (!th0)
665 			panic("th0 == NULL in tcp_respond");
666 #endif
667 
668 		/* get family information from m */
669 		switch (mtod(m, struct ip *)->ip_v) {
670 		case 4:
671 			family = AF_INET;
672 			hlen = sizeof(struct ip);
673 			ip = mtod(m, struct ip *);
674 			break;
675 #ifdef INET6
676 		case 6:
677 			family = AF_INET6;
678 			hlen = sizeof(struct ip6_hdr);
679 			ip6 = mtod(m, struct ip6_hdr *);
680 			break;
681 #endif
682 		default:
683 			m_freem(m);
684 			return EAFNOSUPPORT;
685 		}
686 		/* clear h/w csum flags inherited from rx packet */
687 		m->m_pkthdr.csum_flags = 0;
688 
689 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
690 			tlen = sizeof(*th0);
691 		else
692 			tlen = th0->th_off << 2;
693 
694 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
695 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
696 			m->m_len = hlen + tlen;
697 			m_freem(m->m_next);
698 			m->m_next = NULL;
699 		} else {
700 			struct mbuf *n;
701 
702 #ifdef DIAGNOSTIC
703 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
704 				m_freem(m);
705 				return EMSGSIZE;
706 			}
707 #endif
708 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
709 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
710 				MCLGET(n, M_DONTWAIT);
711 				if ((n->m_flags & M_EXT) == 0) {
712 					m_freem(n);
713 					n = NULL;
714 				}
715 			}
716 			if (!n) {
717 				m_freem(m);
718 				return ENOBUFS;
719 			}
720 
721 			MCLAIM(n, &tcp_tx_mowner);
722 			n->m_data += max_linkhdr;
723 			n->m_len = hlen + tlen;
724 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
725 			m_copyback(n, hlen, tlen, (caddr_t)th0);
726 
727 			m_freem(m);
728 			m = n;
729 			n = NULL;
730 		}
731 
732 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
733 		switch (family) {
734 		case AF_INET:
735 			ip = mtod(m, struct ip *);
736 			th = (struct tcphdr *)(ip + 1);
737 			ip->ip_p = IPPROTO_TCP;
738 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
739 			ip->ip_p = IPPROTO_TCP;
740 			break;
741 #ifdef INET6
742 		case AF_INET6:
743 			ip6 = mtod(m, struct ip6_hdr *);
744 			th = (struct tcphdr *)(ip6 + 1);
745 			ip6->ip6_nxt = IPPROTO_TCP;
746 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
747 			ip6->ip6_nxt = IPPROTO_TCP;
748 			break;
749 #endif
750 #if 0
751 		default:
752 			/* noone will visit here */
753 			m_freem(m);
754 			return EAFNOSUPPORT;
755 #endif
756 		}
757 		xchg(th->th_dport, th->th_sport, u_int16_t);
758 #undef xchg
759 		tlen = 0;	/*be friendly with the following code*/
760 	}
761 	th->th_seq = htonl(seq);
762 	th->th_ack = htonl(ack);
763 	th->th_x2 = 0;
764 	if ((flags & TH_SYN) == 0) {
765 		if (tp)
766 			win >>= tp->rcv_scale;
767 		if (win > TCP_MAXWIN)
768 			win = TCP_MAXWIN;
769 		th->th_win = htons((u_int16_t)win);
770 		th->th_off = sizeof (struct tcphdr) >> 2;
771 		tlen += sizeof(*th);
772 	} else
773 		tlen += th->th_off << 2;
774 	m->m_len = hlen + tlen;
775 	m->m_pkthdr.len = hlen + tlen;
776 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
777 	th->th_flags = flags;
778 	th->th_urp = 0;
779 
780 	switch (family) {
781 #ifdef INET
782 	case AF_INET:
783 	    {
784 		struct ipovly *ipov = (struct ipovly *)ip;
785 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
786 		ipov->ih_len = htons((u_int16_t)tlen);
787 
788 		th->th_sum = 0;
789 		th->th_sum = in_cksum(m, hlen + tlen);
790 		ip->ip_len = htons(hlen + tlen);
791 		ip->ip_ttl = ip_defttl;
792 		break;
793 	    }
794 #endif
795 #ifdef INET6
796 	case AF_INET6:
797 	    {
798 		th->th_sum = 0;
799 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
800 				tlen);
801 		ip6->ip6_plen = htons(tlen);
802 		if (tp && tp->t_in6pcb) {
803 			struct ifnet *oifp;
804 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
805 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
806 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
807 		} else
808 			ip6->ip6_hlim = ip6_defhlim;
809 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
810 		if (ip6_auto_flowlabel) {
811 			ip6->ip6_flow |=
812 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
813 		}
814 		break;
815 	    }
816 #endif
817 	}
818 
819 	if (tp && tp->t_inpcb)
820 		so = tp->t_inpcb->inp_socket;
821 #ifdef INET6
822 	else if (tp && tp->t_in6pcb)
823 		so = tp->t_in6pcb->in6p_socket;
824 #endif
825 	else
826 		so = NULL;
827 
828 	if (tp != NULL && tp->t_inpcb != NULL) {
829 		ro = &tp->t_inpcb->inp_route;
830 #ifdef DIAGNOSTIC
831 		if (family != AF_INET)
832 			panic("tcp_respond: address family mismatch");
833 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
834 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
835 			    ntohl(ip->ip_dst.s_addr),
836 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
837 		}
838 #endif
839 	}
840 #ifdef INET6
841 	else if (tp != NULL && tp->t_in6pcb != NULL) {
842 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
843 #ifdef DIAGNOSTIC
844 		if (family == AF_INET) {
845 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
846 				panic("tcp_respond: not mapped addr");
847 			if (bcmp(&ip->ip_dst,
848 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
849 			    sizeof(ip->ip_dst)) != 0) {
850 				panic("tcp_respond: ip_dst != in6p_faddr");
851 			}
852 		} else if (family == AF_INET6) {
853 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
854 			    &tp->t_in6pcb->in6p_faddr))
855 				panic("tcp_respond: ip6_dst != in6p_faddr");
856 		} else
857 			panic("tcp_respond: address family mismatch");
858 #endif
859 	}
860 #endif
861 	else
862 		ro = NULL;
863 
864 	switch (family) {
865 #ifdef INET
866 	case AF_INET:
867 		error = ip_output(m, NULL, ro,
868 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
869 		    (struct ip_moptions *)0, so);
870 		break;
871 #endif
872 #ifdef INET6
873 	case AF_INET6:
874 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
875 		    (struct ip6_moptions *)0, so, NULL);
876 		break;
877 #endif
878 	default:
879 		error = EAFNOSUPPORT;
880 		break;
881 	}
882 
883 	return (error);
884 }
885 
886 /*
887  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
888  * a bunch of members individually, we maintain this template for the
889  * static and mostly-static components of the TCPCB, and copy it into
890  * the new TCPCB instead.
891  */
892 static struct tcpcb tcpcb_template = {
893 	/*
894 	 * If TCP_NTIMERS ever changes, we'll need to update this
895 	 * initializer.
896 	 */
897 	.t_timer = {
898 		CALLOUT_INITIALIZER,
899 		CALLOUT_INITIALIZER,
900 		CALLOUT_INITIALIZER,
901 		CALLOUT_INITIALIZER,
902 	},
903 	.t_delack_ch = CALLOUT_INITIALIZER,
904 
905 	.t_srtt = TCPTV_SRTTBASE,
906 	.t_rttmin = TCPTV_MIN,
907 
908 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
909 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
910 
911 	.t_partialacks = -1,
912 };
913 
914 /*
915  * Updates the TCPCB template whenever a parameter that would affect
916  * the template is changed.
917  */
918 void
919 tcp_tcpcb_template(void)
920 {
921 	struct tcpcb *tp = &tcpcb_template;
922 	int flags;
923 
924 	tp->t_peermss = tcp_mssdflt;
925 	tp->t_ourmss = tcp_mssdflt;
926 	tp->t_segsz = tcp_mssdflt;
927 
928 	flags = 0;
929 	if (tcp_do_rfc1323 && tcp_do_win_scale)
930 		flags |= TF_REQ_SCALE;
931 	if (tcp_do_rfc1323 && tcp_do_timestamps)
932 		flags |= TF_REQ_TSTMP;
933 	if (tcp_do_sack == 2)
934 		flags |= TF_WILL_SACK;
935 	else if (tcp_do_sack == 1)
936 		flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
937 	flags |= TF_CANT_TXSACK;
938 	tp->t_flags = flags;
939 
940 	/*
941 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
942 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
943 	 * reasonable initial retransmit time.
944 	 */
945 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
946 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
947 	    TCPTV_MIN, TCPTV_REXMTMAX);
948 }
949 
950 /*
951  * Create a new TCP control block, making an
952  * empty reassembly queue and hooking it to the argument
953  * protocol control block.
954  */
955 /* family selects inpcb, or in6pcb */
956 struct tcpcb *
957 tcp_newtcpcb(int family, void *aux)
958 {
959 	struct tcpcb *tp;
960 	int i;
961 
962 	/* XXX Consider using a pool_cache for speed. */
963 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
964 	if (tp == NULL)
965 		return (NULL);
966 	memcpy(tp, &tcpcb_template, sizeof(*tp));
967 	TAILQ_INIT(&tp->segq);
968 	TAILQ_INIT(&tp->timeq);
969 	tp->t_family = family;		/* may be overridden later on */
970 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
971 
972 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
973 	for (i = 0; i < TCPT_NTIMERS; i++)
974 		TCP_TIMER_INIT(tp, i);
975 
976 	switch (family) {
977 	case AF_INET:
978 	    {
979 		struct inpcb *inp = (struct inpcb *)aux;
980 
981 		inp->inp_ip.ip_ttl = ip_defttl;
982 		inp->inp_ppcb = (caddr_t)tp;
983 
984 		tp->t_inpcb = inp;
985 		tp->t_mtudisc = ip_mtudisc;
986 		break;
987 	    }
988 #ifdef INET6
989 	case AF_INET6:
990 	    {
991 		struct in6pcb *in6p = (struct in6pcb *)aux;
992 
993 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
994 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
995 					       : NULL);
996 		in6p->in6p_ppcb = (caddr_t)tp;
997 
998 		tp->t_in6pcb = in6p;
999 		/* for IPv6, always try to run path MTU discovery */
1000 		tp->t_mtudisc = 1;
1001 		break;
1002 	    }
1003 #endif /* INET6 */
1004 	default:
1005 		pool_put(&tcpcb_pool, tp);
1006 		return (NULL);
1007 	}
1008 
1009 	/*
1010 	 * Initialize our timebase.  When we send timestamps, we take
1011 	 * the delta from tcp_now -- this means each connection always
1012 	 * gets a timebase of 0, which makes it, among other things,
1013 	 * more difficult to determine how long a system has been up,
1014 	 * and thus how many TCP sequence increments have occurred.
1015 	 */
1016 	tp->ts_timebase = tcp_now;
1017 
1018 	return (tp);
1019 }
1020 
1021 /*
1022  * Drop a TCP connection, reporting
1023  * the specified error.  If connection is synchronized,
1024  * then send a RST to peer.
1025  */
1026 struct tcpcb *
1027 tcp_drop(struct tcpcb *tp, int errno)
1028 {
1029 	struct socket *so = NULL;
1030 
1031 #ifdef DIAGNOSTIC
1032 	if (tp->t_inpcb && tp->t_in6pcb)
1033 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1034 #endif
1035 #ifdef INET
1036 	if (tp->t_inpcb)
1037 		so = tp->t_inpcb->inp_socket;
1038 #endif
1039 #ifdef INET6
1040 	if (tp->t_in6pcb)
1041 		so = tp->t_in6pcb->in6p_socket;
1042 #endif
1043 	if (!so)
1044 		return NULL;
1045 
1046 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1047 		tp->t_state = TCPS_CLOSED;
1048 		(void) tcp_output(tp);
1049 		tcpstat.tcps_drops++;
1050 	} else
1051 		tcpstat.tcps_conndrops++;
1052 	if (errno == ETIMEDOUT && tp->t_softerror)
1053 		errno = tp->t_softerror;
1054 	so->so_error = errno;
1055 	return (tcp_close(tp));
1056 }
1057 
1058 /*
1059  * Return whether this tcpcb is marked as dead, indicating
1060  * to the calling timer function that no further action should
1061  * be taken, as we are about to release this tcpcb.  The release
1062  * of the storage will be done if this is the last timer running.
1063  *
1064  * This should be called from the callout handler function after
1065  * callout_ack() is done, so that the number of invoking timer
1066  * functions is 0.
1067  */
1068 int
1069 tcp_isdead(struct tcpcb *tp)
1070 {
1071 	int dead = (tp->t_flags & TF_DEAD);
1072 
1073 	if (__predict_false(dead)) {
1074 		if (tcp_timers_invoking(tp) > 0)
1075 				/* not quite there yet -- count separately? */
1076 			return dead;
1077 		tcpstat.tcps_delayed_free++;
1078 		pool_put(&tcpcb_pool, tp);
1079 	}
1080 	return dead;
1081 }
1082 
1083 /*
1084  * Close a TCP control block:
1085  *	discard all space held by the tcp
1086  *	discard internet protocol block
1087  *	wake up any sleepers
1088  */
1089 struct tcpcb *
1090 tcp_close(struct tcpcb *tp)
1091 {
1092 	struct inpcb *inp;
1093 #ifdef INET6
1094 	struct in6pcb *in6p;
1095 #endif
1096 	struct socket *so;
1097 #ifdef RTV_RTT
1098 	struct rtentry *rt;
1099 #endif
1100 	struct route *ro;
1101 
1102 	inp = tp->t_inpcb;
1103 #ifdef INET6
1104 	in6p = tp->t_in6pcb;
1105 #endif
1106 	so = NULL;
1107 	ro = NULL;
1108 	if (inp) {
1109 		so = inp->inp_socket;
1110 		ro = &inp->inp_route;
1111 	}
1112 #ifdef INET6
1113 	else if (in6p) {
1114 		so = in6p->in6p_socket;
1115 		ro = (struct route *)&in6p->in6p_route;
1116 	}
1117 #endif
1118 
1119 #ifdef RTV_RTT
1120 	/*
1121 	 * If we sent enough data to get some meaningful characteristics,
1122 	 * save them in the routing entry.  'Enough' is arbitrarily
1123 	 * defined as the sendpipesize (default 4K) * 16.  This would
1124 	 * give us 16 rtt samples assuming we only get one sample per
1125 	 * window (the usual case on a long haul net).  16 samples is
1126 	 * enough for the srtt filter to converge to within 5% of the correct
1127 	 * value; fewer samples and we could save a very bogus rtt.
1128 	 *
1129 	 * Don't update the default route's characteristics and don't
1130 	 * update anything that the user "locked".
1131 	 */
1132 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1133 	    ro && (rt = ro->ro_rt) &&
1134 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1135 		u_long i = 0;
1136 
1137 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1138 			i = tp->t_srtt *
1139 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1140 			if (rt->rt_rmx.rmx_rtt && i)
1141 				/*
1142 				 * filter this update to half the old & half
1143 				 * the new values, converting scale.
1144 				 * See route.h and tcp_var.h for a
1145 				 * description of the scaling constants.
1146 				 */
1147 				rt->rt_rmx.rmx_rtt =
1148 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1149 			else
1150 				rt->rt_rmx.rmx_rtt = i;
1151 		}
1152 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1153 			i = tp->t_rttvar *
1154 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1155 			if (rt->rt_rmx.rmx_rttvar && i)
1156 				rt->rt_rmx.rmx_rttvar =
1157 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1158 			else
1159 				rt->rt_rmx.rmx_rttvar = i;
1160 		}
1161 		/*
1162 		 * update the pipelimit (ssthresh) if it has been updated
1163 		 * already or if a pipesize was specified & the threshhold
1164 		 * got below half the pipesize.  I.e., wait for bad news
1165 		 * before we start updating, then update on both good
1166 		 * and bad news.
1167 		 */
1168 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1169 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1170 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1171 			/*
1172 			 * convert the limit from user data bytes to
1173 			 * packets then to packet data bytes.
1174 			 */
1175 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1176 			if (i < 2)
1177 				i = 2;
1178 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1179 			if (rt->rt_rmx.rmx_ssthresh)
1180 				rt->rt_rmx.rmx_ssthresh =
1181 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1182 			else
1183 				rt->rt_rmx.rmx_ssthresh = i;
1184 		}
1185 	}
1186 #endif /* RTV_RTT */
1187 	/* free the reassembly queue, if any */
1188 	TCP_REASS_LOCK(tp);
1189 	(void) tcp_freeq(tp);
1190 	TCP_REASS_UNLOCK(tp);
1191 
1192 	tcp_canceltimers(tp);
1193 	TCP_CLEAR_DELACK(tp);
1194 	syn_cache_cleanup(tp);
1195 
1196 	if (tp->t_template) {
1197 		m_free(tp->t_template);
1198 		tp->t_template = NULL;
1199 	}
1200 	if (tcp_timers_invoking(tp))
1201 		tp->t_flags |= TF_DEAD;
1202 	else
1203 		pool_put(&tcpcb_pool, tp);
1204 
1205 	if (inp) {
1206 		inp->inp_ppcb = 0;
1207 		soisdisconnected(so);
1208 		in_pcbdetach(inp);
1209 	}
1210 #ifdef INET6
1211 	else if (in6p) {
1212 		in6p->in6p_ppcb = 0;
1213 		soisdisconnected(so);
1214 		in6_pcbdetach(in6p);
1215 	}
1216 #endif
1217 	tcpstat.tcps_closed++;
1218 	return ((struct tcpcb *)0);
1219 }
1220 
1221 int
1222 tcp_freeq(tp)
1223 	struct tcpcb *tp;
1224 {
1225 	struct ipqent *qe;
1226 	int rv = 0;
1227 #ifdef TCPREASS_DEBUG
1228 	int i = 0;
1229 #endif
1230 
1231 	TCP_REASS_LOCK_CHECK(tp);
1232 
1233 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1234 #ifdef TCPREASS_DEBUG
1235 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1236 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1237 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1238 #endif
1239 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1240 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1241 		m_freem(qe->ipqe_m);
1242 		pool_put(&tcpipqent_pool, qe);
1243 		rv = 1;
1244 	}
1245 	return (rv);
1246 }
1247 
1248 /*
1249  * Protocol drain routine.  Called when memory is in short supply.
1250  */
1251 void
1252 tcp_drain(void)
1253 {
1254 	struct inpcb_hdr *inph;
1255 	struct tcpcb *tp;
1256 
1257 	/*
1258 	 * Free the sequence queue of all TCP connections.
1259 	 */
1260 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1261 		switch (inph->inph_af) {
1262 		case AF_INET:
1263 			tp = intotcpcb((struct inpcb *)inph);
1264 			break;
1265 #ifdef INET6
1266 		case AF_INET6:
1267 			tp = in6totcpcb((struct in6pcb *)inph);
1268 			break;
1269 #endif
1270 		default:
1271 			tp = NULL;
1272 			break;
1273 		}
1274 		if (tp != NULL) {
1275 			/*
1276 			 * We may be called from a device's interrupt
1277 			 * context.  If the tcpcb is already busy,
1278 			 * just bail out now.
1279 			 */
1280 			if (tcp_reass_lock_try(tp) == 0)
1281 				continue;
1282 			if (tcp_freeq(tp))
1283 				tcpstat.tcps_connsdrained++;
1284 			TCP_REASS_UNLOCK(tp);
1285 		}
1286 	}
1287 }
1288 
1289 /*
1290  * Notify a tcp user of an asynchronous error;
1291  * store error as soft error, but wake up user
1292  * (for now, won't do anything until can select for soft error).
1293  */
1294 void
1295 tcp_notify(struct inpcb *inp, int error)
1296 {
1297 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1298 	struct socket *so = inp->inp_socket;
1299 
1300 	/*
1301 	 * Ignore some errors if we are hooked up.
1302 	 * If connection hasn't completed, has retransmitted several times,
1303 	 * and receives a second error, give up now.  This is better
1304 	 * than waiting a long time to establish a connection that
1305 	 * can never complete.
1306 	 */
1307 	if (tp->t_state == TCPS_ESTABLISHED &&
1308 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1309 	      error == EHOSTDOWN)) {
1310 		return;
1311 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1312 	    tp->t_rxtshift > 3 && tp->t_softerror)
1313 		so->so_error = error;
1314 	else
1315 		tp->t_softerror = error;
1316 	wakeup((caddr_t) &so->so_timeo);
1317 	sorwakeup(so);
1318 	sowwakeup(so);
1319 }
1320 
1321 #ifdef INET6
1322 void
1323 tcp6_notify(struct in6pcb *in6p, int error)
1324 {
1325 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1326 	struct socket *so = in6p->in6p_socket;
1327 
1328 	/*
1329 	 * Ignore some errors if we are hooked up.
1330 	 * If connection hasn't completed, has retransmitted several times,
1331 	 * and receives a second error, give up now.  This is better
1332 	 * than waiting a long time to establish a connection that
1333 	 * can never complete.
1334 	 */
1335 	if (tp->t_state == TCPS_ESTABLISHED &&
1336 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1337 	      error == EHOSTDOWN)) {
1338 		return;
1339 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1340 	    tp->t_rxtshift > 3 && tp->t_softerror)
1341 		so->so_error = error;
1342 	else
1343 		tp->t_softerror = error;
1344 	wakeup((caddr_t) &so->so_timeo);
1345 	sorwakeup(so);
1346 	sowwakeup(so);
1347 }
1348 #endif
1349 
1350 #ifdef INET6
1351 void
1352 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1353 {
1354 	struct tcphdr th;
1355 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1356 	int nmatch;
1357 	struct ip6_hdr *ip6;
1358 	const struct sockaddr_in6 *sa6_src = NULL;
1359 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1360 	struct mbuf *m;
1361 	int off;
1362 
1363 	if (sa->sa_family != AF_INET6 ||
1364 	    sa->sa_len != sizeof(struct sockaddr_in6))
1365 		return;
1366 	if ((unsigned)cmd >= PRC_NCMDS)
1367 		return;
1368 	else if (cmd == PRC_QUENCH) {
1369 		/* XXX there's no PRC_QUENCH in IPv6 */
1370 		notify = tcp6_quench;
1371 	} else if (PRC_IS_REDIRECT(cmd))
1372 		notify = in6_rtchange, d = NULL;
1373 	else if (cmd == PRC_MSGSIZE)
1374 		; /* special code is present, see below */
1375 	else if (cmd == PRC_HOSTDEAD)
1376 		d = NULL;
1377 	else if (inet6ctlerrmap[cmd] == 0)
1378 		return;
1379 
1380 	/* if the parameter is from icmp6, decode it. */
1381 	if (d != NULL) {
1382 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1383 		m = ip6cp->ip6c_m;
1384 		ip6 = ip6cp->ip6c_ip6;
1385 		off = ip6cp->ip6c_off;
1386 		sa6_src = ip6cp->ip6c_src;
1387 	} else {
1388 		m = NULL;
1389 		ip6 = NULL;
1390 		sa6_src = &sa6_any;
1391 		off = 0;
1392 	}
1393 
1394 	if (ip6) {
1395 		/*
1396 		 * XXX: We assume that when ip6 is non NULL,
1397 		 * M and OFF are valid.
1398 		 */
1399 
1400 		/* check if we can safely examine src and dst ports */
1401 		if (m->m_pkthdr.len < off + sizeof(th)) {
1402 			if (cmd == PRC_MSGSIZE)
1403 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1404 			return;
1405 		}
1406 
1407 		bzero(&th, sizeof(th));
1408 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1409 
1410 		if (cmd == PRC_MSGSIZE) {
1411 			int valid = 0;
1412 
1413 			/*
1414 			 * Check to see if we have a valid TCP connection
1415 			 * corresponding to the address in the ICMPv6 message
1416 			 * payload.
1417 			 */
1418 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1419 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1420 			    th.th_sport, 0))
1421 				valid++;
1422 
1423 			/*
1424 			 * Depending on the value of "valid" and routing table
1425 			 * size (mtudisc_{hi,lo}wat), we will:
1426 			 * - recalcurate the new MTU and create the
1427 			 *   corresponding routing entry, or
1428 			 * - ignore the MTU change notification.
1429 			 */
1430 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1431 
1432 			/*
1433 			 * no need to call in6_pcbnotify, it should have been
1434 			 * called via callback if necessary
1435 			 */
1436 			return;
1437 		}
1438 
1439 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1440 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1441 		if (nmatch == 0 && syn_cache_count &&
1442 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1443 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1444 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1445 			syn_cache_unreach((struct sockaddr *)sa6_src,
1446 					  sa, &th);
1447 	} else {
1448 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1449 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1450 	}
1451 }
1452 #endif
1453 
1454 #ifdef INET
1455 /* assumes that ip header and tcp header are contiguous on mbuf */
1456 void *
1457 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v)
1458 {
1459 	struct ip *ip = v;
1460 	struct tcphdr *th;
1461 	struct icmp *icp;
1462 	extern const int inetctlerrmap[];
1463 	void (*notify)(struct inpcb *, int) = tcp_notify;
1464 	int errno;
1465 	int nmatch;
1466 #ifdef INET6
1467 	struct in6_addr src6, dst6;
1468 #endif
1469 
1470 	if (sa->sa_family != AF_INET ||
1471 	    sa->sa_len != sizeof(struct sockaddr_in))
1472 		return NULL;
1473 	if ((unsigned)cmd >= PRC_NCMDS)
1474 		return NULL;
1475 	errno = inetctlerrmap[cmd];
1476 	if (cmd == PRC_QUENCH)
1477 		notify = tcp_quench;
1478 	else if (PRC_IS_REDIRECT(cmd))
1479 		notify = in_rtchange, ip = 0;
1480 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1481 		/*
1482 		 * Check to see if we have a valid TCP connection
1483 		 * corresponding to the address in the ICMP message
1484 		 * payload.
1485 		 *
1486 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1487 		 */
1488 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1489 #ifdef INET6
1490 		memset(&src6, 0, sizeof(src6));
1491 		memset(&dst6, 0, sizeof(dst6));
1492 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1493 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1494 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1495 #endif
1496 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1497 		    ip->ip_src, th->th_sport) != NULL)
1498 			;
1499 #ifdef INET6
1500 		else if (in6_pcblookup_connect(&tcbtable, &dst6,
1501 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1502 			;
1503 #endif
1504 		else
1505 			return NULL;
1506 
1507 		/*
1508 		 * Now that we've validated that we are actually communicating
1509 		 * with the host indicated in the ICMP message, locate the
1510 		 * ICMP header, recalculate the new MTU, and create the
1511 		 * corresponding routing entry.
1512 		 */
1513 		icp = (struct icmp *)((caddr_t)ip -
1514 		    offsetof(struct icmp, icmp_ip));
1515 		icmp_mtudisc(icp, ip->ip_dst);
1516 
1517 		return NULL;
1518 	} else if (cmd == PRC_HOSTDEAD)
1519 		ip = 0;
1520 	else if (errno == 0)
1521 		return NULL;
1522 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1523 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1524 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1525 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1526 		if (nmatch == 0 && syn_cache_count &&
1527 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1528 		    inetctlerrmap[cmd] == ENETUNREACH ||
1529 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1530 			struct sockaddr_in sin;
1531 			bzero(&sin, sizeof(sin));
1532 			sin.sin_len = sizeof(sin);
1533 			sin.sin_family = AF_INET;
1534 			sin.sin_port = th->th_sport;
1535 			sin.sin_addr = ip->ip_src;
1536 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1537 		}
1538 
1539 		/* XXX mapped address case */
1540 	} else
1541 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1542 		    notify);
1543 	return NULL;
1544 }
1545 
1546 /*
1547  * When a source quence is received, we are being notifed of congestion.
1548  * Close the congestion window down to the Loss Window (one segment).
1549  * We will gradually open it again as we proceed.
1550  */
1551 void
1552 tcp_quench(struct inpcb *inp, int errno)
1553 {
1554 	struct tcpcb *tp = intotcpcb(inp);
1555 
1556 	if (tp)
1557 		tp->snd_cwnd = tp->t_segsz;
1558 }
1559 #endif
1560 
1561 #ifdef INET6
1562 void
1563 tcp6_quench(struct in6pcb *in6p, int errno)
1564 {
1565 	struct tcpcb *tp = in6totcpcb(in6p);
1566 
1567 	if (tp)
1568 		tp->snd_cwnd = tp->t_segsz;
1569 }
1570 #endif
1571 
1572 #ifdef INET
1573 /*
1574  * Path MTU Discovery handlers.
1575  */
1576 void
1577 tcp_mtudisc_callback(struct in_addr faddr)
1578 {
1579 #ifdef INET6
1580 	struct in6_addr in6;
1581 #endif
1582 
1583 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1584 #ifdef INET6
1585 	memset(&in6, 0, sizeof(in6));
1586 	in6.s6_addr16[5] = 0xffff;
1587 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1588 	tcp6_mtudisc_callback(&in6);
1589 #endif
1590 }
1591 
1592 /*
1593  * On receipt of path MTU corrections, flush old route and replace it
1594  * with the new one.  Retransmit all unacknowledged packets, to ensure
1595  * that all packets will be received.
1596  */
1597 void
1598 tcp_mtudisc(struct inpcb *inp, int errno)
1599 {
1600 	struct tcpcb *tp = intotcpcb(inp);
1601 	struct rtentry *rt = in_pcbrtentry(inp);
1602 
1603 	if (tp != 0) {
1604 		if (rt != 0) {
1605 			/*
1606 			 * If this was not a host route, remove and realloc.
1607 			 */
1608 			if ((rt->rt_flags & RTF_HOST) == 0) {
1609 				in_rtchange(inp, errno);
1610 				if ((rt = in_pcbrtentry(inp)) == 0)
1611 					return;
1612 			}
1613 
1614 			/*
1615 			 * Slow start out of the error condition.  We
1616 			 * use the MTU because we know it's smaller
1617 			 * than the previously transmitted segment.
1618 			 *
1619 			 * Note: This is more conservative than the
1620 			 * suggestion in draft-floyd-incr-init-win-03.
1621 			 */
1622 			if (rt->rt_rmx.rmx_mtu != 0)
1623 				tp->snd_cwnd =
1624 				    TCP_INITIAL_WINDOW(tcp_init_win,
1625 				    rt->rt_rmx.rmx_mtu);
1626 		}
1627 
1628 		/*
1629 		 * Resend unacknowledged packets.
1630 		 */
1631 		tp->snd_nxt = tp->snd_una;
1632 		tcp_output(tp);
1633 	}
1634 }
1635 #endif
1636 
1637 #ifdef INET6
1638 /*
1639  * Path MTU Discovery handlers.
1640  */
1641 void
1642 tcp6_mtudisc_callback(struct in6_addr *faddr)
1643 {
1644 	struct sockaddr_in6 sin6;
1645 
1646 	bzero(&sin6, sizeof(sin6));
1647 	sin6.sin6_family = AF_INET6;
1648 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1649 	sin6.sin6_addr = *faddr;
1650 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1651 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1652 }
1653 
1654 void
1655 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1656 {
1657 	struct tcpcb *tp = in6totcpcb(in6p);
1658 	struct rtentry *rt = in6_pcbrtentry(in6p);
1659 
1660 	if (tp != 0) {
1661 		if (rt != 0) {
1662 			/*
1663 			 * If this was not a host route, remove and realloc.
1664 			 */
1665 			if ((rt->rt_flags & RTF_HOST) == 0) {
1666 				in6_rtchange(in6p, errno);
1667 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1668 					return;
1669 			}
1670 
1671 			/*
1672 			 * Slow start out of the error condition.  We
1673 			 * use the MTU because we know it's smaller
1674 			 * than the previously transmitted segment.
1675 			 *
1676 			 * Note: This is more conservative than the
1677 			 * suggestion in draft-floyd-incr-init-win-03.
1678 			 */
1679 			if (rt->rt_rmx.rmx_mtu != 0)
1680 				tp->snd_cwnd =
1681 				    TCP_INITIAL_WINDOW(tcp_init_win,
1682 				    rt->rt_rmx.rmx_mtu);
1683 		}
1684 
1685 		/*
1686 		 * Resend unacknowledged packets.
1687 		 */
1688 		tp->snd_nxt = tp->snd_una;
1689 		tcp_output(tp);
1690 	}
1691 }
1692 #endif /* INET6 */
1693 
1694 /*
1695  * Compute the MSS to advertise to the peer.  Called only during
1696  * the 3-way handshake.  If we are the server (peer initiated
1697  * connection), we are called with a pointer to the interface
1698  * on which the SYN packet arrived.  If we are the client (we
1699  * initiated connection), we are called with a pointer to the
1700  * interface out which this connection should go.
1701  *
1702  * NOTE: Do not subtract IP option/extension header size nor IPsec
1703  * header size from MSS advertisement.  MSS option must hold the maximum
1704  * segment size we can accept, so it must always be:
1705  *	 max(if mtu) - ip header - tcp header
1706  */
1707 u_long
1708 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1709 {
1710 	extern u_long in_maxmtu;
1711 	u_long mss = 0;
1712 	u_long hdrsiz;
1713 
1714 	/*
1715 	 * In order to avoid defeating path MTU discovery on the peer,
1716 	 * we advertise the max MTU of all attached networks as our MSS,
1717 	 * per RFC 1191, section 3.1.
1718 	 *
1719 	 * We provide the option to advertise just the MTU of
1720 	 * the interface on which we hope this connection will
1721 	 * be receiving.  If we are responding to a SYN, we
1722 	 * will have a pretty good idea about this, but when
1723 	 * initiating a connection there is a bit more doubt.
1724 	 *
1725 	 * We also need to ensure that loopback has a large enough
1726 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1727 	 */
1728 
1729 	if (ifp != NULL)
1730 		switch (af) {
1731 		case AF_INET:
1732 			mss = ifp->if_mtu;
1733 			break;
1734 #ifdef INET6
1735 		case AF_INET6:
1736 			mss = IN6_LINKMTU(ifp);
1737 			break;
1738 #endif
1739 		}
1740 
1741 	if (tcp_mss_ifmtu == 0)
1742 		switch (af) {
1743 		case AF_INET:
1744 			mss = max(in_maxmtu, mss);
1745 			break;
1746 #ifdef INET6
1747 		case AF_INET6:
1748 			mss = max(in6_maxmtu, mss);
1749 			break;
1750 #endif
1751 		}
1752 
1753 	switch (af) {
1754 	case AF_INET:
1755 		hdrsiz = sizeof(struct ip);
1756 		break;
1757 #ifdef INET6
1758 	case AF_INET6:
1759 		hdrsiz = sizeof(struct ip6_hdr);
1760 		break;
1761 #endif
1762 	default:
1763 		hdrsiz = 0;
1764 		break;
1765 	}
1766 	hdrsiz += sizeof(struct tcphdr);
1767 	if (mss > hdrsiz)
1768 		mss -= hdrsiz;
1769 
1770 	mss = max(tcp_mssdflt, mss);
1771 	return (mss);
1772 }
1773 
1774 /*
1775  * Set connection variables based on the peer's advertised MSS.
1776  * We are passed the TCPCB for the actual connection.  If we
1777  * are the server, we are called by the compressed state engine
1778  * when the 3-way handshake is complete.  If we are the client,
1779  * we are called when we receive the SYN,ACK from the server.
1780  *
1781  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1782  * before this routine is called!
1783  */
1784 void
1785 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1786 {
1787 	struct socket *so;
1788 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1789 	struct rtentry *rt;
1790 #endif
1791 	u_long bufsize;
1792 	int mss;
1793 
1794 #ifdef DIAGNOSTIC
1795 	if (tp->t_inpcb && tp->t_in6pcb)
1796 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1797 #endif
1798 	so = NULL;
1799 	rt = NULL;
1800 #ifdef INET
1801 	if (tp->t_inpcb) {
1802 		so = tp->t_inpcb->inp_socket;
1803 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1804 		rt = in_pcbrtentry(tp->t_inpcb);
1805 #endif
1806 	}
1807 #endif
1808 #ifdef INET6
1809 	if (tp->t_in6pcb) {
1810 		so = tp->t_in6pcb->in6p_socket;
1811 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1812 		rt = in6_pcbrtentry(tp->t_in6pcb);
1813 #endif
1814 	}
1815 #endif
1816 
1817 	/*
1818 	 * As per RFC1122, use the default MSS value, unless they
1819 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1820 	 */
1821 	mss = tcp_mssdflt;
1822 	if (offer)
1823 		mss = offer;
1824 	mss = max(mss, 256);		/* sanity */
1825 	tp->t_peermss = mss;
1826 	mss -= tcp_optlen(tp);
1827 #ifdef INET
1828 	if (tp->t_inpcb)
1829 		mss -= ip_optlen(tp->t_inpcb);
1830 #endif
1831 #ifdef INET6
1832 	if (tp->t_in6pcb)
1833 		mss -= ip6_optlen(tp->t_in6pcb);
1834 #endif
1835 
1836 	/*
1837 	 * If there's a pipesize, change the socket buffer to that size.
1838 	 * Make the socket buffer an integral number of MSS units.  If
1839 	 * the MSS is larger than the socket buffer, artificially decrease
1840 	 * the MSS.
1841 	 */
1842 #ifdef RTV_SPIPE
1843 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1844 		bufsize = rt->rt_rmx.rmx_sendpipe;
1845 	else
1846 #endif
1847 		bufsize = so->so_snd.sb_hiwat;
1848 	if (bufsize < mss)
1849 		mss = bufsize;
1850 	else {
1851 		bufsize = roundup(bufsize, mss);
1852 		if (bufsize > sb_max)
1853 			bufsize = sb_max;
1854 		(void) sbreserve(&so->so_snd, bufsize, so);
1855 	}
1856 	tp->t_segsz = mss;
1857 
1858 #ifdef RTV_SSTHRESH
1859 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1860 		/*
1861 		 * There's some sort of gateway or interface buffer
1862 		 * limit on the path.  Use this to set the slow
1863 		 * start threshold, but set the threshold to no less
1864 		 * than 2 * MSS.
1865 		 */
1866 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1867 	}
1868 #endif
1869 }
1870 
1871 /*
1872  * Processing necessary when a TCP connection is established.
1873  */
1874 void
1875 tcp_established(struct tcpcb *tp)
1876 {
1877 	struct socket *so;
1878 #ifdef RTV_RPIPE
1879 	struct rtentry *rt;
1880 #endif
1881 	u_long bufsize;
1882 
1883 #ifdef DIAGNOSTIC
1884 	if (tp->t_inpcb && tp->t_in6pcb)
1885 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1886 #endif
1887 	so = NULL;
1888 	rt = NULL;
1889 #ifdef INET
1890 	if (tp->t_inpcb) {
1891 		so = tp->t_inpcb->inp_socket;
1892 #if defined(RTV_RPIPE)
1893 		rt = in_pcbrtentry(tp->t_inpcb);
1894 #endif
1895 	}
1896 #endif
1897 #ifdef INET6
1898 	if (tp->t_in6pcb) {
1899 		so = tp->t_in6pcb->in6p_socket;
1900 #if defined(RTV_RPIPE)
1901 		rt = in6_pcbrtentry(tp->t_in6pcb);
1902 #endif
1903 	}
1904 #endif
1905 
1906 	tp->t_state = TCPS_ESTABLISHED;
1907 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1908 
1909 #ifdef RTV_RPIPE
1910 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1911 		bufsize = rt->rt_rmx.rmx_recvpipe;
1912 	else
1913 #endif
1914 		bufsize = so->so_rcv.sb_hiwat;
1915 	if (bufsize > tp->t_ourmss) {
1916 		bufsize = roundup(bufsize, tp->t_ourmss);
1917 		if (bufsize > sb_max)
1918 			bufsize = sb_max;
1919 		(void) sbreserve(&so->so_rcv, bufsize, so);
1920 	}
1921 }
1922 
1923 /*
1924  * Check if there's an initial rtt or rttvar.  Convert from the
1925  * route-table units to scaled multiples of the slow timeout timer.
1926  * Called only during the 3-way handshake.
1927  */
1928 void
1929 tcp_rmx_rtt(struct tcpcb *tp)
1930 {
1931 #ifdef RTV_RTT
1932 	struct rtentry *rt = NULL;
1933 	int rtt;
1934 
1935 #ifdef DIAGNOSTIC
1936 	if (tp->t_inpcb && tp->t_in6pcb)
1937 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1938 #endif
1939 #ifdef INET
1940 	if (tp->t_inpcb)
1941 		rt = in_pcbrtentry(tp->t_inpcb);
1942 #endif
1943 #ifdef INET6
1944 	if (tp->t_in6pcb)
1945 		rt = in6_pcbrtentry(tp->t_in6pcb);
1946 #endif
1947 	if (rt == NULL)
1948 		return;
1949 
1950 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1951 		/*
1952 		 * XXX The lock bit for MTU indicates that the value
1953 		 * is also a minimum value; this is subject to time.
1954 		 */
1955 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1956 			TCPT_RANGESET(tp->t_rttmin,
1957 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1958 			    TCPTV_MIN, TCPTV_REXMTMAX);
1959 		tp->t_srtt = rtt /
1960 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1961 		if (rt->rt_rmx.rmx_rttvar) {
1962 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1963 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1964 				(TCP_RTTVAR_SHIFT + 2));
1965 		} else {
1966 			/* Default variation is +- 1 rtt */
1967 			tp->t_rttvar =
1968 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1969 		}
1970 		TCPT_RANGESET(tp->t_rxtcur,
1971 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1972 		    tp->t_rttmin, TCPTV_REXMTMAX);
1973 	}
1974 #endif
1975 }
1976 
1977 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1978 #if NRND > 0
1979 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1980 #endif
1981 
1982 /*
1983  * Get a new sequence value given a tcp control block
1984  */
1985 tcp_seq
1986 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1987 {
1988 
1989 #ifdef INET
1990 	if (tp->t_inpcb != NULL) {
1991 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1992 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1993 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
1994 		    addin));
1995 	}
1996 #endif
1997 #ifdef INET6
1998 	if (tp->t_in6pcb != NULL) {
1999 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2000 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2001 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2002 		    addin));
2003 	}
2004 #endif
2005 	/* Not possible. */
2006 	panic("tcp_new_iss");
2007 }
2008 
2009 /*
2010  * This routine actually generates a new TCP initial sequence number.
2011  */
2012 tcp_seq
2013 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2014     size_t addrsz, tcp_seq addin)
2015 {
2016 	tcp_seq tcp_iss;
2017 
2018 #if NRND > 0
2019 	static int beenhere;
2020 
2021 	/*
2022 	 * If we haven't been here before, initialize our cryptographic
2023 	 * hash secret.
2024 	 */
2025 	if (beenhere == 0) {
2026 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2027 		    RND_EXTRACT_ANY);
2028 		beenhere = 1;
2029 	}
2030 
2031 	if (tcp_do_rfc1948) {
2032 		MD5_CTX ctx;
2033 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2034 
2035 		/*
2036 		 * Compute the base value of the ISS.  It is a hash
2037 		 * of (saddr, sport, daddr, dport, secret).
2038 		 */
2039 		MD5Init(&ctx);
2040 
2041 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2042 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2043 
2044 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2045 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2046 
2047 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2048 
2049 		MD5Final(hash, &ctx);
2050 
2051 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2052 
2053 		/*
2054 		 * Now increment our "timer", and add it in to
2055 		 * the computed value.
2056 		 *
2057 		 * XXX Use `addin'?
2058 		 * XXX TCP_ISSINCR too large to use?
2059 		 */
2060 		tcp_iss_seq += TCP_ISSINCR;
2061 #ifdef TCPISS_DEBUG
2062 		printf("ISS hash 0x%08x, ", tcp_iss);
2063 #endif
2064 		tcp_iss += tcp_iss_seq + addin;
2065 #ifdef TCPISS_DEBUG
2066 		printf("new ISS 0x%08x\n", tcp_iss);
2067 #endif
2068 	} else
2069 #endif /* NRND > 0 */
2070 	{
2071 		/*
2072 		 * Randomize.
2073 		 */
2074 #if NRND > 0
2075 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2076 #else
2077 		tcp_iss = arc4random();
2078 #endif
2079 
2080 		/*
2081 		 * If we were asked to add some amount to a known value,
2082 		 * we will take a random value obtained above, mask off
2083 		 * the upper bits, and add in the known value.  We also
2084 		 * add in a constant to ensure that we are at least a
2085 		 * certain distance from the original value.
2086 		 *
2087 		 * This is used when an old connection is in timed wait
2088 		 * and we have a new one coming in, for instance.
2089 		 */
2090 		if (addin != 0) {
2091 #ifdef TCPISS_DEBUG
2092 			printf("Random %08x, ", tcp_iss);
2093 #endif
2094 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2095 			tcp_iss += addin + TCP_ISSINCR;
2096 #ifdef TCPISS_DEBUG
2097 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2098 #endif
2099 		} else {
2100 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2101 			tcp_iss += tcp_iss_seq;
2102 			tcp_iss_seq += TCP_ISSINCR;
2103 #ifdef TCPISS_DEBUG
2104 			printf("ISS %08x\n", tcp_iss);
2105 #endif
2106 		}
2107 	}
2108 
2109 	if (tcp_compat_42) {
2110 		/*
2111 		 * Limit it to the positive range for really old TCP
2112 		 * implementations.
2113 		 * Just AND off the top bit instead of checking if
2114 		 * is set first - saves a branch 50% of the time.
2115 		 */
2116 		tcp_iss &= 0x7fffffff;		/* XXX */
2117 	}
2118 
2119 	return (tcp_iss);
2120 }
2121 
2122 #if defined(IPSEC) || defined(FAST_IPSEC)
2123 /* compute ESP/AH header size for TCP, including outer IP header. */
2124 size_t
2125 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2126 {
2127 	struct inpcb *inp;
2128 	size_t hdrsiz;
2129 
2130 	/* XXX mapped addr case (tp->t_in6pcb) */
2131 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2132 		return 0;
2133 	switch (tp->t_family) {
2134 	case AF_INET:
2135 		/* XXX: should use currect direction. */
2136 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2137 		break;
2138 	default:
2139 		hdrsiz = 0;
2140 		break;
2141 	}
2142 
2143 	return hdrsiz;
2144 }
2145 
2146 #ifdef INET6
2147 size_t
2148 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2149 {
2150 	struct in6pcb *in6p;
2151 	size_t hdrsiz;
2152 
2153 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2154 		return 0;
2155 	switch (tp->t_family) {
2156 	case AF_INET6:
2157 		/* XXX: should use currect direction. */
2158 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2159 		break;
2160 	case AF_INET:
2161 		/* mapped address case - tricky */
2162 	default:
2163 		hdrsiz = 0;
2164 		break;
2165 	}
2166 
2167 	return hdrsiz;
2168 }
2169 #endif
2170 #endif /*IPSEC*/
2171 
2172 /*
2173  * Determine the length of the TCP options for this connection.
2174  *
2175  * XXX:  What do we do for SACK, when we add that?  Just reserve
2176  *       all of the space?  Otherwise we can't exactly be incrementing
2177  *       cwnd by an amount that varies depending on the amount we last
2178  *       had to SACK!
2179  */
2180 
2181 u_int
2182 tcp_optlen(struct tcpcb *tp)
2183 {
2184 	u_int optlen;
2185 
2186 	optlen = 0;
2187 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2188 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2189 		optlen += TCPOLEN_TSTAMP_APPA;
2190 
2191 #ifdef TCP_SIGNATURE
2192 #if defined(INET6) && defined(FAST_IPSEC)
2193 	if (tp->t_family == AF_INET)
2194 #endif
2195 	if (tp->t_flags & TF_SIGNATURE)
2196 		optlen += TCPOLEN_SIGNATURE + 2;
2197 #endif /* TCP_SIGNATURE */
2198 
2199 	return optlen;
2200 }
2201