xref: /netbsd-src/sys/netinet/tcp_subr.c (revision f17b710f3d406bee67aa39c65053114ab78297c5)
1 /*	$NetBSD: tcp_subr.c,v 1.262 2015/05/19 17:33:43 kefren 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, 2008 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  *
49  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59  * POSSIBILITY OF SUCH DAMAGE.
60  */
61 
62 /*
63  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
64  *	The Regents of the University of California.  All rights reserved.
65  *
66  * Redistribution and use in source and binary forms, with or without
67  * modification, are permitted provided that the following conditions
68  * are met:
69  * 1. Redistributions of source code must retain the above copyright
70  *    notice, this list of conditions and the following disclaimer.
71  * 2. Redistributions in binary form must reproduce the above copyright
72  *    notice, this list of conditions and the following disclaimer in the
73  *    documentation and/or other materials provided with the distribution.
74  * 3. Neither the name of the University nor the names of its contributors
75  *    may be used to endorse or promote products derived from this software
76  *    without specific prior written permission.
77  *
78  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88  * SUCH DAMAGE.
89  *
90  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
91  */
92 
93 #include <sys/cdefs.h>
94 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.262 2015/05/19 17:33:43 kefren Exp $");
95 
96 #include "opt_inet.h"
97 #include "opt_ipsec.h"
98 #include "opt_tcp_compat_42.h"
99 #include "opt_inet_csum.h"
100 #include "opt_mbuftrace.h"
101 
102 #include <sys/param.h>
103 #include <sys/atomic.h>
104 #include <sys/proc.h>
105 #include <sys/systm.h>
106 #include <sys/mbuf.h>
107 #include <sys/once.h>
108 #include <sys/socket.h>
109 #include <sys/socketvar.h>
110 #include <sys/protosw.h>
111 #include <sys/errno.h>
112 #include <sys/kernel.h>
113 #include <sys/pool.h>
114 #include <sys/md5.h>
115 #include <sys/cprng.h>
116 
117 #include <net/route.h>
118 #include <net/if.h>
119 
120 #include <netinet/in.h>
121 #include <netinet/in_systm.h>
122 #include <netinet/ip.h>
123 #include <netinet/in_pcb.h>
124 #include <netinet/ip_var.h>
125 #include <netinet/ip_icmp.h>
126 
127 #ifdef INET6
128 #ifndef INET
129 #include <netinet/in.h>
130 #endif
131 #include <netinet/ip6.h>
132 #include <netinet6/in6_pcb.h>
133 #include <netinet6/ip6_var.h>
134 #include <netinet6/in6_var.h>
135 #include <netinet6/ip6protosw.h>
136 #include <netinet/icmp6.h>
137 #include <netinet6/nd6.h>
138 #endif
139 
140 #include <netinet/tcp.h>
141 #include <netinet/tcp_fsm.h>
142 #include <netinet/tcp_seq.h>
143 #include <netinet/tcp_timer.h>
144 #include <netinet/tcp_var.h>
145 #include <netinet/tcp_vtw.h>
146 #include <netinet/tcp_private.h>
147 #include <netinet/tcp_congctl.h>
148 #include <netinet/tcpip.h>
149 
150 #ifdef IPSEC
151 #include <netipsec/ipsec.h>
152 #include <netipsec/xform.h>
153 #ifdef INET6
154 #include <netipsec/ipsec6.h>
155 #endif
156  #include <netipsec/key.h>
157 #endif	/* IPSEC*/
158 
159 
160 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
161 u_int32_t tcp_now;		/* slow ticks, for RFC 1323 timestamps */
162 
163 percpu_t *tcpstat_percpu;
164 
165 /* patchable/settable parameters for tcp */
166 int 	tcp_mssdflt = TCP_MSS;
167 int	tcp_minmss = TCP_MINMSS;
168 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
169 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
170 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
171 int	tcp_do_sack = 1;	/* selective acknowledgement */
172 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
173 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
174 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
175 int	tcp_do_ecn = 0;		/* Explicit Congestion Notification */
176 #ifndef TCP_INIT_WIN
177 #define	TCP_INIT_WIN	4	/* initial slow start window */
178 #endif
179 #ifndef TCP_INIT_WIN_LOCAL
180 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
181 #endif
182 /*
183  * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460.
184  * This is to simulate current behavior for iw == 4
185  */
186 int tcp_init_win_max[] = {
187 	 1 * 1460,
188 	 1 * 1460,
189 	 2 * 1460,
190 	 2 * 1460,
191 	 3 * 1460,
192 	 5 * 1460,
193 	 6 * 1460,
194 	 7 * 1460,
195 	 8 * 1460,
196 	 9 * 1460,
197 	10 * 1460
198 };
199 int	tcp_init_win = TCP_INIT_WIN;
200 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
201 int	tcp_mss_ifmtu = 0;
202 #ifdef TCP_COMPAT_42
203 int	tcp_compat_42 = 1;
204 #else
205 int	tcp_compat_42 = 0;
206 #endif
207 int	tcp_rst_ppslim = 100;	/* 100pps */
208 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
209 int	tcp_do_loopback_cksum = 0;
210 int	tcp_do_abc = 1;		/* RFC3465 Appropriate byte counting. */
211 int	tcp_abc_aggressive = 1;	/* 1: L=2*SMSS  0: L=1*SMSS */
212 int	tcp_sack_tp_maxholes = 32;
213 int	tcp_sack_globalmaxholes = 1024;
214 int	tcp_sack_globalholes = 0;
215 int	tcp_ecn_maxretries = 1;
216 int	tcp_msl_enable = 1;		/* enable TIME_WAIT truncation	*/
217 int	tcp_msl_loop   = PR_SLOWHZ;	/* MSL for loopback		*/
218 int	tcp_msl_local  = 5 * PR_SLOWHZ;	/* MSL for 'local'		*/
219 int	tcp_msl_remote = TCPTV_MSL;	/* MSL otherwise		*/
220 int	tcp_msl_remote_threshold = TCPTV_SRTTDFLT;	/* RTT threshold */
221 int	tcp_rttlocal = 0;		/* Use RTT to decide who's 'local' */
222 
223 int	tcp4_vtw_enable = 0;		/* 1 to enable */
224 int	tcp6_vtw_enable = 0;		/* 1 to enable */
225 int	tcp_vtw_was_enabled = 0;
226 int	tcp_vtw_entries = 1 << 4;	/* 16 vestigial TIME_WAIT entries */
227 
228 /* tcb hash */
229 #ifndef TCBHASHSIZE
230 #define	TCBHASHSIZE	128
231 #endif
232 int	tcbhashsize = TCBHASHSIZE;
233 
234 /* syn hash parameters */
235 #define	TCP_SYN_HASH_SIZE	293
236 #define	TCP_SYN_BUCKET_SIZE	35
237 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
238 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
239 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
240 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
241 
242 int	tcp_freeq(struct tcpcb *);
243 static int	tcp_iss_secret_init(void);
244 
245 #ifdef INET
246 static void	tcp_mtudisc_callback(struct in_addr);
247 #endif
248 
249 #ifdef INET6
250 void	tcp6_mtudisc(struct in6pcb *, int);
251 #endif
252 
253 static struct pool tcpcb_pool;
254 
255 static int tcp_drainwanted;
256 
257 #ifdef TCP_CSUM_COUNTERS
258 #include <sys/device.h>
259 
260 #if defined(INET)
261 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "hwcsum bad");
263 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "hwcsum ok");
265 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     NULL, "tcp", "hwcsum data");
267 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     NULL, "tcp", "swcsum");
269 
270 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
271 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
272 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
273 EVCNT_ATTACH_STATIC(tcp_swcsum);
274 #endif /* defined(INET) */
275 
276 #if defined(INET6)
277 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     NULL, "tcp6", "hwcsum bad");
279 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     NULL, "tcp6", "hwcsum ok");
281 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     NULL, "tcp6", "hwcsum data");
283 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     NULL, "tcp6", "swcsum");
285 
286 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
287 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
288 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
289 EVCNT_ATTACH_STATIC(tcp6_swcsum);
290 #endif /* defined(INET6) */
291 #endif /* TCP_CSUM_COUNTERS */
292 
293 
294 #ifdef TCP_OUTPUT_COUNTERS
295 #include <sys/device.h>
296 
297 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
298     NULL, "tcp", "output big header");
299 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
300     NULL, "tcp", "output predict hit");
301 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
302     NULL, "tcp", "output predict miss");
303 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
304     NULL, "tcp", "output copy small");
305 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
306     NULL, "tcp", "output copy big");
307 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
308     NULL, "tcp", "output reference big");
309 
310 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
311 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
312 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
313 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
314 EVCNT_ATTACH_STATIC(tcp_output_copybig);
315 EVCNT_ATTACH_STATIC(tcp_output_refbig);
316 
317 #endif /* TCP_OUTPUT_COUNTERS */
318 
319 #ifdef TCP_REASS_COUNTERS
320 #include <sys/device.h>
321 
322 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
323     NULL, "tcp_reass", "calls");
324 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
325     &tcp_reass_, "tcp_reass", "insert into empty queue");
326 struct evcnt tcp_reass_iteration[8] = {
327     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
329     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
330     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
331     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
332     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
333     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
334     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
335 };
336 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
337     &tcp_reass_, "tcp_reass", "prepend to first");
338 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
339     &tcp_reass_, "tcp_reass", "prepend");
340 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
341     &tcp_reass_, "tcp_reass", "insert");
342 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
343     &tcp_reass_, "tcp_reass", "insert at tail");
344 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
345     &tcp_reass_, "tcp_reass", "append");
346 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
347     &tcp_reass_, "tcp_reass", "append to tail fragment");
348 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
349     &tcp_reass_, "tcp_reass", "overlap at end");
350 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
351     &tcp_reass_, "tcp_reass", "overlap at start");
352 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
353     &tcp_reass_, "tcp_reass", "duplicate segment");
354 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
355     &tcp_reass_, "tcp_reass", "duplicate fragment");
356 
357 EVCNT_ATTACH_STATIC(tcp_reass_);
358 EVCNT_ATTACH_STATIC(tcp_reass_empty);
359 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
366 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
367 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
368 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
369 EVCNT_ATTACH_STATIC(tcp_reass_insert);
370 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
371 EVCNT_ATTACH_STATIC(tcp_reass_append);
372 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
373 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
374 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
375 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
376 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
377 
378 #endif /* TCP_REASS_COUNTERS */
379 
380 #ifdef MBUFTRACE
381 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
382 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
383 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
384 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
385 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
386 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
387 #endif
388 
389 callout_t tcp_slowtimo_ch;
390 
391 static int
392 do_tcpinit(void)
393 {
394 
395 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
396 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
397 	    NULL, IPL_SOFTNET);
398 
399 	tcp_usrreq_init();
400 
401 	/* Initialize timer state. */
402 	tcp_timer_init();
403 
404 	/* Initialize the compressed state engine. */
405 	syn_cache_init();
406 
407 	/* Initialize the congestion control algorithms. */
408 	tcp_congctl_init();
409 
410 	/* Initialize the TCPCB template. */
411 	tcp_tcpcb_template();
412 
413 	/* Initialize reassembly queue */
414 	tcpipqent_init();
415 
416 	/* SACK */
417 	tcp_sack_init();
418 
419 	MOWNER_ATTACH(&tcp_tx_mowner);
420 	MOWNER_ATTACH(&tcp_rx_mowner);
421 	MOWNER_ATTACH(&tcp_reass_mowner);
422 	MOWNER_ATTACH(&tcp_sock_mowner);
423 	MOWNER_ATTACH(&tcp_sock_tx_mowner);
424 	MOWNER_ATTACH(&tcp_sock_rx_mowner);
425 	MOWNER_ATTACH(&tcp_mowner);
426 
427 	tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
428 
429 	vtw_earlyinit();
430 
431 	callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
432 	callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
433 
434 	return 0;
435 }
436 
437 void
438 tcp_init_common(unsigned basehlen)
439 {
440 	static ONCE_DECL(dotcpinit);
441 	unsigned hlen = basehlen + sizeof(struct tcphdr);
442 	unsigned oldhlen;
443 
444 	if (max_linkhdr + hlen > MHLEN)
445 		panic("tcp_init");
446 	while ((oldhlen = max_protohdr) < hlen)
447 		atomic_cas_uint(&max_protohdr, oldhlen, hlen);
448 
449 	RUN_ONCE(&dotcpinit, do_tcpinit);
450 }
451 
452 /*
453  * Tcp initialization
454  */
455 void
456 tcp_init(void)
457 {
458 
459 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
460 
461 	tcp_init_common(sizeof(struct ip));
462 }
463 
464 /*
465  * Create template to be used to send tcp packets on a connection.
466  * Call after host entry created, allocates an mbuf and fills
467  * in a skeletal tcp/ip header, minimizing the amount of work
468  * necessary when the connection is used.
469  */
470 struct mbuf *
471 tcp_template(struct tcpcb *tp)
472 {
473 	struct inpcb *inp = tp->t_inpcb;
474 #ifdef INET6
475 	struct in6pcb *in6p = tp->t_in6pcb;
476 #endif
477 	struct tcphdr *n;
478 	struct mbuf *m;
479 	int hlen;
480 
481 	switch (tp->t_family) {
482 	case AF_INET:
483 		hlen = sizeof(struct ip);
484 		if (inp)
485 			break;
486 #ifdef INET6
487 		if (in6p) {
488 			/* mapped addr case */
489 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
490 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
491 				break;
492 		}
493 #endif
494 		return NULL;	/*EINVAL*/
495 #ifdef INET6
496 	case AF_INET6:
497 		hlen = sizeof(struct ip6_hdr);
498 		if (in6p) {
499 			/* more sainty check? */
500 			break;
501 		}
502 		return NULL;	/*EINVAL*/
503 #endif
504 	default:
505 		hlen = 0;	/*pacify gcc*/
506 		return NULL;	/*EAFNOSUPPORT*/
507 	}
508 #ifdef DIAGNOSTIC
509 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
510 		panic("mclbytes too small for t_template");
511 #endif
512 	m = tp->t_template;
513 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
514 		;
515 	else {
516 		if (m)
517 			m_freem(m);
518 		m = tp->t_template = NULL;
519 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
520 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
521 			MCLGET(m, M_DONTWAIT);
522 			if ((m->m_flags & M_EXT) == 0) {
523 				m_free(m);
524 				m = NULL;
525 			}
526 		}
527 		if (m == NULL)
528 			return NULL;
529 		MCLAIM(m, &tcp_mowner);
530 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
531 	}
532 
533 	memset(mtod(m, void *), 0, m->m_len);
534 
535 	n = (struct tcphdr *)(mtod(m, char *) + hlen);
536 
537 	switch (tp->t_family) {
538 	case AF_INET:
539 	    {
540 		struct ipovly *ipov;
541 		mtod(m, struct ip *)->ip_v = 4;
542 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
543 		ipov = mtod(m, struct ipovly *);
544 		ipov->ih_pr = IPPROTO_TCP;
545 		ipov->ih_len = htons(sizeof(struct tcphdr));
546 		if (inp) {
547 			ipov->ih_src = inp->inp_laddr;
548 			ipov->ih_dst = inp->inp_faddr;
549 		}
550 #ifdef INET6
551 		else if (in6p) {
552 			/* mapped addr case */
553 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
554 				sizeof(ipov->ih_src));
555 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
556 				sizeof(ipov->ih_dst));
557 		}
558 #endif
559 		/*
560 		 * Compute the pseudo-header portion of the checksum
561 		 * now.  We incrementally add in the TCP option and
562 		 * payload lengths later, and then compute the TCP
563 		 * checksum right before the packet is sent off onto
564 		 * the wire.
565 		 */
566 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
567 		    ipov->ih_dst.s_addr,
568 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
569 		break;
570 	    }
571 #ifdef INET6
572 	case AF_INET6:
573 	    {
574 		struct ip6_hdr *ip6;
575 		mtod(m, struct ip *)->ip_v = 6;
576 		ip6 = mtod(m, struct ip6_hdr *);
577 		ip6->ip6_nxt = IPPROTO_TCP;
578 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
579 		ip6->ip6_src = in6p->in6p_laddr;
580 		ip6->ip6_dst = in6p->in6p_faddr;
581 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
582 		if (ip6_auto_flowlabel) {
583 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
584 			ip6->ip6_flow |=
585 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
586 		}
587 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
588 		ip6->ip6_vfc |= IPV6_VERSION;
589 
590 		/*
591 		 * Compute the pseudo-header portion of the checksum
592 		 * now.  We incrementally add in the TCP option and
593 		 * payload lengths later, and then compute the TCP
594 		 * checksum right before the packet is sent off onto
595 		 * the wire.
596 		 */
597 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
598 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
599 		    htonl(IPPROTO_TCP));
600 		break;
601 	    }
602 #endif
603 	}
604 	if (inp) {
605 		n->th_sport = inp->inp_lport;
606 		n->th_dport = inp->inp_fport;
607 	}
608 #ifdef INET6
609 	else if (in6p) {
610 		n->th_sport = in6p->in6p_lport;
611 		n->th_dport = in6p->in6p_fport;
612 	}
613 #endif
614 	n->th_seq = 0;
615 	n->th_ack = 0;
616 	n->th_x2 = 0;
617 	n->th_off = 5;
618 	n->th_flags = 0;
619 	n->th_win = 0;
620 	n->th_urp = 0;
621 	return (m);
622 }
623 
624 /*
625  * Send a single message to the TCP at address specified by
626  * the given TCP/IP header.  If m == 0, then we make a copy
627  * of the tcpiphdr at ti and send directly to the addressed host.
628  * This is used to force keep alive messages out using the TCP
629  * template for a connection tp->t_template.  If flags are given
630  * then we send a message back to the TCP which originated the
631  * segment ti, and discard the mbuf containing it and any other
632  * attached mbufs.
633  *
634  * In any case the ack and sequence number of the transmitted
635  * segment are as specified by the parameters.
636  */
637 int
638 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
639     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
640 {
641 	struct route *ro;
642 	int error, tlen, win = 0;
643 	int hlen;
644 	struct ip *ip;
645 #ifdef INET6
646 	struct ip6_hdr *ip6;
647 #endif
648 	int family;	/* family on packet, not inpcb/in6pcb! */
649 	struct tcphdr *th;
650 	struct socket *so;
651 
652 	if (tp != NULL && (flags & TH_RST) == 0) {
653 #ifdef DIAGNOSTIC
654 		if (tp->t_inpcb && tp->t_in6pcb)
655 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
656 #endif
657 #ifdef INET
658 		if (tp->t_inpcb)
659 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
660 #endif
661 #ifdef INET6
662 		if (tp->t_in6pcb)
663 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
664 #endif
665 	}
666 
667 	th = NULL;	/* Quell uninitialized warning */
668 	ip = NULL;
669 #ifdef INET6
670 	ip6 = NULL;
671 #endif
672 	if (m == 0) {
673 		if (!mtemplate)
674 			return EINVAL;
675 
676 		/* get family information from template */
677 		switch (mtod(mtemplate, struct ip *)->ip_v) {
678 		case 4:
679 			family = AF_INET;
680 			hlen = sizeof(struct ip);
681 			break;
682 #ifdef INET6
683 		case 6:
684 			family = AF_INET6;
685 			hlen = sizeof(struct ip6_hdr);
686 			break;
687 #endif
688 		default:
689 			return EAFNOSUPPORT;
690 		}
691 
692 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
693 		if (m) {
694 			MCLAIM(m, &tcp_tx_mowner);
695 			MCLGET(m, M_DONTWAIT);
696 			if ((m->m_flags & M_EXT) == 0) {
697 				m_free(m);
698 				m = NULL;
699 			}
700 		}
701 		if (m == NULL)
702 			return (ENOBUFS);
703 
704 		if (tcp_compat_42)
705 			tlen = 1;
706 		else
707 			tlen = 0;
708 
709 		m->m_data += max_linkhdr;
710 		bcopy(mtod(mtemplate, void *), mtod(m, void *),
711 			mtemplate->m_len);
712 		switch (family) {
713 		case AF_INET:
714 			ip = mtod(m, struct ip *);
715 			th = (struct tcphdr *)(ip + 1);
716 			break;
717 #ifdef INET6
718 		case AF_INET6:
719 			ip6 = mtod(m, struct ip6_hdr *);
720 			th = (struct tcphdr *)(ip6 + 1);
721 			break;
722 #endif
723 #if 0
724 		default:
725 			/* noone will visit here */
726 			m_freem(m);
727 			return EAFNOSUPPORT;
728 #endif
729 		}
730 		flags = TH_ACK;
731 	} else {
732 
733 		if ((m->m_flags & M_PKTHDR) == 0) {
734 #if 0
735 			printf("non PKTHDR to tcp_respond\n");
736 #endif
737 			m_freem(m);
738 			return EINVAL;
739 		}
740 #ifdef DIAGNOSTIC
741 		if (!th0)
742 			panic("th0 == NULL in tcp_respond");
743 #endif
744 
745 		/* get family information from m */
746 		switch (mtod(m, struct ip *)->ip_v) {
747 		case 4:
748 			family = AF_INET;
749 			hlen = sizeof(struct ip);
750 			ip = mtod(m, struct ip *);
751 			break;
752 #ifdef INET6
753 		case 6:
754 			family = AF_INET6;
755 			hlen = sizeof(struct ip6_hdr);
756 			ip6 = mtod(m, struct ip6_hdr *);
757 			break;
758 #endif
759 		default:
760 			m_freem(m);
761 			return EAFNOSUPPORT;
762 		}
763 		/* clear h/w csum flags inherited from rx packet */
764 		m->m_pkthdr.csum_flags = 0;
765 
766 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
767 			tlen = sizeof(*th0);
768 		else
769 			tlen = th0->th_off << 2;
770 
771 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
772 		    mtod(m, char *) + hlen == (char *)th0) {
773 			m->m_len = hlen + tlen;
774 			m_freem(m->m_next);
775 			m->m_next = NULL;
776 		} else {
777 			struct mbuf *n;
778 
779 #ifdef DIAGNOSTIC
780 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
781 				m_freem(m);
782 				return EMSGSIZE;
783 			}
784 #endif
785 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
786 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
787 				MCLGET(n, M_DONTWAIT);
788 				if ((n->m_flags & M_EXT) == 0) {
789 					m_freem(n);
790 					n = NULL;
791 				}
792 			}
793 			if (!n) {
794 				m_freem(m);
795 				return ENOBUFS;
796 			}
797 
798 			MCLAIM(n, &tcp_tx_mowner);
799 			n->m_data += max_linkhdr;
800 			n->m_len = hlen + tlen;
801 			m_copyback(n, 0, hlen, mtod(m, void *));
802 			m_copyback(n, hlen, tlen, (void *)th0);
803 
804 			m_freem(m);
805 			m = n;
806 			n = NULL;
807 		}
808 
809 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
810 		switch (family) {
811 		case AF_INET:
812 			ip = mtod(m, struct ip *);
813 			th = (struct tcphdr *)(ip + 1);
814 			ip->ip_p = IPPROTO_TCP;
815 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
816 			ip->ip_p = IPPROTO_TCP;
817 			break;
818 #ifdef INET6
819 		case AF_INET6:
820 			ip6 = mtod(m, struct ip6_hdr *);
821 			th = (struct tcphdr *)(ip6 + 1);
822 			ip6->ip6_nxt = IPPROTO_TCP;
823 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
824 			ip6->ip6_nxt = IPPROTO_TCP;
825 			break;
826 #endif
827 #if 0
828 		default:
829 			/* noone will visit here */
830 			m_freem(m);
831 			return EAFNOSUPPORT;
832 #endif
833 		}
834 		xchg(th->th_dport, th->th_sport, u_int16_t);
835 #undef xchg
836 		tlen = 0;	/*be friendly with the following code*/
837 	}
838 	th->th_seq = htonl(seq);
839 	th->th_ack = htonl(ack);
840 	th->th_x2 = 0;
841 	if ((flags & TH_SYN) == 0) {
842 		if (tp)
843 			win >>= tp->rcv_scale;
844 		if (win > TCP_MAXWIN)
845 			win = TCP_MAXWIN;
846 		th->th_win = htons((u_int16_t)win);
847 		th->th_off = sizeof (struct tcphdr) >> 2;
848 		tlen += sizeof(*th);
849 	} else
850 		tlen += th->th_off << 2;
851 	m->m_len = hlen + tlen;
852 	m->m_pkthdr.len = hlen + tlen;
853 	m->m_pkthdr.rcvif = NULL;
854 	th->th_flags = flags;
855 	th->th_urp = 0;
856 
857 	switch (family) {
858 #ifdef INET
859 	case AF_INET:
860 	    {
861 		struct ipovly *ipov = (struct ipovly *)ip;
862 		memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
863 		ipov->ih_len = htons((u_int16_t)tlen);
864 
865 		th->th_sum = 0;
866 		th->th_sum = in_cksum(m, hlen + tlen);
867 		ip->ip_len = htons(hlen + tlen);
868 		ip->ip_ttl = ip_defttl;
869 		break;
870 	    }
871 #endif
872 #ifdef INET6
873 	case AF_INET6:
874 	    {
875 		th->th_sum = 0;
876 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
877 				tlen);
878 		ip6->ip6_plen = htons(tlen);
879 		if (tp && tp->t_in6pcb)
880 			ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb);
881 		else
882 			ip6->ip6_hlim = ip6_defhlim;
883 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
884 		if (ip6_auto_flowlabel) {
885 			ip6->ip6_flow |=
886 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
887 		}
888 		break;
889 	    }
890 #endif
891 	}
892 
893 	if (tp && tp->t_inpcb)
894 		so = tp->t_inpcb->inp_socket;
895 #ifdef INET6
896 	else if (tp && tp->t_in6pcb)
897 		so = tp->t_in6pcb->in6p_socket;
898 #endif
899 	else
900 		so = NULL;
901 
902 	if (tp != NULL && tp->t_inpcb != NULL) {
903 		ro = &tp->t_inpcb->inp_route;
904 #ifdef DIAGNOSTIC
905 		if (family != AF_INET)
906 			panic("tcp_respond: address family mismatch");
907 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
908 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
909 			    ntohl(ip->ip_dst.s_addr),
910 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
911 		}
912 #endif
913 	}
914 #ifdef INET6
915 	else if (tp != NULL && tp->t_in6pcb != NULL) {
916 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
917 #ifdef DIAGNOSTIC
918 		if (family == AF_INET) {
919 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
920 				panic("tcp_respond: not mapped addr");
921 			if (memcmp(&ip->ip_dst,
922 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
923 			    sizeof(ip->ip_dst)) != 0) {
924 				panic("tcp_respond: ip_dst != in6p_faddr");
925 			}
926 		} else if (family == AF_INET6) {
927 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
928 			    &tp->t_in6pcb->in6p_faddr))
929 				panic("tcp_respond: ip6_dst != in6p_faddr");
930 		} else
931 			panic("tcp_respond: address family mismatch");
932 #endif
933 	}
934 #endif
935 	else
936 		ro = NULL;
937 
938 	switch (family) {
939 #ifdef INET
940 	case AF_INET:
941 		error = ip_output(m, NULL, ro,
942 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so);
943 		break;
944 #endif
945 #ifdef INET6
946 	case AF_INET6:
947 		error = ip6_output(m, NULL, ro, 0, NULL, so, NULL);
948 		break;
949 #endif
950 	default:
951 		error = EAFNOSUPPORT;
952 		break;
953 	}
954 
955 	return (error);
956 }
957 
958 /*
959  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
960  * a bunch of members individually, we maintain this template for the
961  * static and mostly-static components of the TCPCB, and copy it into
962  * the new TCPCB instead.
963  */
964 static struct tcpcb tcpcb_template = {
965 	.t_srtt = TCPTV_SRTTBASE,
966 	.t_rttmin = TCPTV_MIN,
967 
968 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
969 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
970 	.snd_numholes = 0,
971 	.snd_cubic_wmax = 0,
972 	.snd_cubic_wmax_last = 0,
973 	.snd_cubic_ctime = 0,
974 
975 	.t_partialacks = -1,
976 	.t_bytes_acked = 0,
977 	.t_sndrexmitpack = 0,
978 	.t_rcvoopack = 0,
979 	.t_sndzerowin = 0,
980 };
981 
982 /*
983  * Updates the TCPCB template whenever a parameter that would affect
984  * the template is changed.
985  */
986 void
987 tcp_tcpcb_template(void)
988 {
989 	struct tcpcb *tp = &tcpcb_template;
990 	int flags;
991 
992 	tp->t_peermss = tcp_mssdflt;
993 	tp->t_ourmss = tcp_mssdflt;
994 	tp->t_segsz = tcp_mssdflt;
995 
996 	flags = 0;
997 	if (tcp_do_rfc1323 && tcp_do_win_scale)
998 		flags |= TF_REQ_SCALE;
999 	if (tcp_do_rfc1323 && tcp_do_timestamps)
1000 		flags |= TF_REQ_TSTMP;
1001 	tp->t_flags = flags;
1002 
1003 	/*
1004 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1005 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
1006 	 * reasonable initial retransmit time.
1007 	 */
1008 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
1009 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1010 	    TCPTV_MIN, TCPTV_REXMTMAX);
1011 
1012 	/* Keep Alive */
1013 	tp->t_keepinit = tcp_keepinit;
1014 	tp->t_keepidle = tcp_keepidle;
1015 	tp->t_keepintvl = tcp_keepintvl;
1016 	tp->t_keepcnt = tcp_keepcnt;
1017 	tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1018 
1019 	/* MSL */
1020 	tp->t_msl = TCPTV_MSL;
1021 }
1022 
1023 /*
1024  * Create a new TCP control block, making an
1025  * empty reassembly queue and hooking it to the argument
1026  * protocol control block.
1027  */
1028 /* family selects inpcb, or in6pcb */
1029 struct tcpcb *
1030 tcp_newtcpcb(int family, void *aux)
1031 {
1032 	struct tcpcb *tp;
1033 	int i;
1034 
1035 	/* XXX Consider using a pool_cache for speed. */
1036 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);	/* splsoftnet via tcp_usrreq */
1037 	if (tp == NULL)
1038 		return (NULL);
1039 	memcpy(tp, &tcpcb_template, sizeof(*tp));
1040 	TAILQ_INIT(&tp->segq);
1041 	TAILQ_INIT(&tp->timeq);
1042 	tp->t_family = family;		/* may be overridden later on */
1043 	TAILQ_INIT(&tp->snd_holes);
1044 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
1045 
1046 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
1047 	for (i = 0; i < TCPT_NTIMERS; i++) {
1048 		callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1049 		TCP_TIMER_INIT(tp, i);
1050 	}
1051 	callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1052 
1053 	switch (family) {
1054 	case AF_INET:
1055 	    {
1056 		struct inpcb *inp = (struct inpcb *)aux;
1057 
1058 		inp->inp_ip.ip_ttl = ip_defttl;
1059 		inp->inp_ppcb = (void *)tp;
1060 
1061 		tp->t_inpcb = inp;
1062 		tp->t_mtudisc = ip_mtudisc;
1063 		break;
1064 	    }
1065 #ifdef INET6
1066 	case AF_INET6:
1067 	    {
1068 		struct in6pcb *in6p = (struct in6pcb *)aux;
1069 
1070 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p);
1071 		in6p->in6p_ppcb = (void *)tp;
1072 
1073 		tp->t_in6pcb = in6p;
1074 		/* for IPv6, always try to run path MTU discovery */
1075 		tp->t_mtudisc = 1;
1076 		break;
1077 	    }
1078 #endif /* INET6 */
1079 	default:
1080 		for (i = 0; i < TCPT_NTIMERS; i++)
1081 			callout_destroy(&tp->t_timer[i]);
1082 		callout_destroy(&tp->t_delack_ch);
1083 		pool_put(&tcpcb_pool, tp);	/* splsoftnet via tcp_usrreq */
1084 		return (NULL);
1085 	}
1086 
1087 	/*
1088 	 * Initialize our timebase.  When we send timestamps, we take
1089 	 * the delta from tcp_now -- this means each connection always
1090 	 * gets a timebase of 1, which makes it, among other things,
1091 	 * more difficult to determine how long a system has been up,
1092 	 * and thus how many TCP sequence increments have occurred.
1093 	 *
1094 	 * We start with 1, because 0 doesn't work with linux, which
1095 	 * considers timestamp 0 in a SYN packet as a bug and disables
1096 	 * timestamps.
1097 	 */
1098 	tp->ts_timebase = tcp_now - 1;
1099 
1100 	tcp_congctl_select(tp, tcp_congctl_global_name);
1101 
1102 	return (tp);
1103 }
1104 
1105 /*
1106  * Drop a TCP connection, reporting
1107  * the specified error.  If connection is synchronized,
1108  * then send a RST to peer.
1109  */
1110 struct tcpcb *
1111 tcp_drop(struct tcpcb *tp, int errno)
1112 {
1113 	struct socket *so = NULL;
1114 
1115 #ifdef DIAGNOSTIC
1116 	if (tp->t_inpcb && tp->t_in6pcb)
1117 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1118 #endif
1119 #ifdef INET
1120 	if (tp->t_inpcb)
1121 		so = tp->t_inpcb->inp_socket;
1122 #endif
1123 #ifdef INET6
1124 	if (tp->t_in6pcb)
1125 		so = tp->t_in6pcb->in6p_socket;
1126 #endif
1127 	if (!so)
1128 		return NULL;
1129 
1130 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1131 		tp->t_state = TCPS_CLOSED;
1132 		(void) tcp_output(tp);
1133 		TCP_STATINC(TCP_STAT_DROPS);
1134 	} else
1135 		TCP_STATINC(TCP_STAT_CONNDROPS);
1136 	if (errno == ETIMEDOUT && tp->t_softerror)
1137 		errno = tp->t_softerror;
1138 	so->so_error = errno;
1139 	return (tcp_close(tp));
1140 }
1141 
1142 /*
1143  * Close a TCP control block:
1144  *	discard all space held by the tcp
1145  *	discard internet protocol block
1146  *	wake up any sleepers
1147  */
1148 struct tcpcb *
1149 tcp_close(struct tcpcb *tp)
1150 {
1151 	struct inpcb *inp;
1152 #ifdef INET6
1153 	struct in6pcb *in6p;
1154 #endif
1155 	struct socket *so;
1156 #ifdef RTV_RTT
1157 	struct rtentry *rt;
1158 #endif
1159 	struct route *ro;
1160 	int j;
1161 
1162 	inp = tp->t_inpcb;
1163 #ifdef INET6
1164 	in6p = tp->t_in6pcb;
1165 #endif
1166 	so = NULL;
1167 	ro = NULL;
1168 	if (inp) {
1169 		so = inp->inp_socket;
1170 		ro = &inp->inp_route;
1171 	}
1172 #ifdef INET6
1173 	else if (in6p) {
1174 		so = in6p->in6p_socket;
1175 		ro = (struct route *)&in6p->in6p_route;
1176 	}
1177 #endif
1178 
1179 #ifdef RTV_RTT
1180 	/*
1181 	 * If we sent enough data to get some meaningful characteristics,
1182 	 * save them in the routing entry.  'Enough' is arbitrarily
1183 	 * defined as the sendpipesize (default 4K) * 16.  This would
1184 	 * give us 16 rtt samples assuming we only get one sample per
1185 	 * window (the usual case on a long haul net).  16 samples is
1186 	 * enough for the srtt filter to converge to within 5% of the correct
1187 	 * value; fewer samples and we could save a very bogus rtt.
1188 	 *
1189 	 * Don't update the default route's characteristics and don't
1190 	 * update anything that the user "locked".
1191 	 */
1192 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1193 	    ro && (rt = rtcache_validate(ro)) != NULL &&
1194 	    !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1195 		u_long i = 0;
1196 
1197 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1198 			i = tp->t_srtt *
1199 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1200 			if (rt->rt_rmx.rmx_rtt && i)
1201 				/*
1202 				 * filter this update to half the old & half
1203 				 * the new values, converting scale.
1204 				 * See route.h and tcp_var.h for a
1205 				 * description of the scaling constants.
1206 				 */
1207 				rt->rt_rmx.rmx_rtt =
1208 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1209 			else
1210 				rt->rt_rmx.rmx_rtt = i;
1211 		}
1212 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1213 			i = tp->t_rttvar *
1214 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1215 			if (rt->rt_rmx.rmx_rttvar && i)
1216 				rt->rt_rmx.rmx_rttvar =
1217 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1218 			else
1219 				rt->rt_rmx.rmx_rttvar = i;
1220 		}
1221 		/*
1222 		 * update the pipelimit (ssthresh) if it has been updated
1223 		 * already or if a pipesize was specified & the threshhold
1224 		 * got below half the pipesize.  I.e., wait for bad news
1225 		 * before we start updating, then update on both good
1226 		 * and bad news.
1227 		 */
1228 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1229 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1230 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1231 			/*
1232 			 * convert the limit from user data bytes to
1233 			 * packets then to packet data bytes.
1234 			 */
1235 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1236 			if (i < 2)
1237 				i = 2;
1238 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1239 			if (rt->rt_rmx.rmx_ssthresh)
1240 				rt->rt_rmx.rmx_ssthresh =
1241 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1242 			else
1243 				rt->rt_rmx.rmx_ssthresh = i;
1244 		}
1245 	}
1246 #endif /* RTV_RTT */
1247 	/* free the reassembly queue, if any */
1248 	TCP_REASS_LOCK(tp);
1249 	(void) tcp_freeq(tp);
1250 	TCP_REASS_UNLOCK(tp);
1251 
1252 	/* free the SACK holes list. */
1253 	tcp_free_sackholes(tp);
1254 	tcp_congctl_release(tp);
1255 	syn_cache_cleanup(tp);
1256 
1257 	if (tp->t_template) {
1258 		m_free(tp->t_template);
1259 		tp->t_template = NULL;
1260 	}
1261 
1262 	/*
1263 	 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1264 	 * the timers can also drop the lock.  We need to prevent access
1265 	 * to the tcpcb as it's half torn down.  Flag the pcb as dead
1266 	 * (prevents access by timers) and only then detach it.
1267 	 */
1268 	tp->t_flags |= TF_DEAD;
1269 	if (inp) {
1270 		inp->inp_ppcb = 0;
1271 		soisdisconnected(so);
1272 		in_pcbdetach(inp);
1273 	}
1274 #ifdef INET6
1275 	else if (in6p) {
1276 		in6p->in6p_ppcb = 0;
1277 		soisdisconnected(so);
1278 		in6_pcbdetach(in6p);
1279 	}
1280 #endif
1281 	/*
1282 	 * pcb is no longer visble elsewhere, so we can safely release
1283 	 * the lock in callout_halt() if needed.
1284 	 */
1285 	TCP_STATINC(TCP_STAT_CLOSED);
1286 	for (j = 0; j < TCPT_NTIMERS; j++) {
1287 		callout_halt(&tp->t_timer[j], softnet_lock);
1288 		callout_destroy(&tp->t_timer[j]);
1289 	}
1290 	callout_halt(&tp->t_delack_ch, softnet_lock);
1291 	callout_destroy(&tp->t_delack_ch);
1292 	pool_put(&tcpcb_pool, tp);
1293 
1294 	return NULL;
1295 }
1296 
1297 int
1298 tcp_freeq(struct tcpcb *tp)
1299 {
1300 	struct ipqent *qe;
1301 	int rv = 0;
1302 #ifdef TCPREASS_DEBUG
1303 	int i = 0;
1304 #endif
1305 
1306 	TCP_REASS_LOCK_CHECK(tp);
1307 
1308 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1309 #ifdef TCPREASS_DEBUG
1310 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1311 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1312 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1313 #endif
1314 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1315 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1316 		m_freem(qe->ipqe_m);
1317 		tcpipqent_free(qe);
1318 		rv = 1;
1319 	}
1320 	tp->t_segqlen = 0;
1321 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1322 	return (rv);
1323 }
1324 
1325 void
1326 tcp_fasttimo(void)
1327 {
1328 	if (tcp_drainwanted) {
1329 		tcp_drain();
1330 		tcp_drainwanted = 0;
1331 	}
1332 }
1333 
1334 void
1335 tcp_drainstub(void)
1336 {
1337 	tcp_drainwanted = 1;
1338 }
1339 
1340 /*
1341  * Protocol drain routine.  Called when memory is in short supply.
1342  * Called from pr_fasttimo thus a callout context.
1343  */
1344 void
1345 tcp_drain(void)
1346 {
1347 	struct inpcb_hdr *inph;
1348 	struct tcpcb *tp;
1349 
1350 	mutex_enter(softnet_lock);
1351 	KERNEL_LOCK(1, NULL);
1352 
1353 	/*
1354 	 * Free the sequence queue of all TCP connections.
1355 	 */
1356 	TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1357 		switch (inph->inph_af) {
1358 		case AF_INET:
1359 			tp = intotcpcb((struct inpcb *)inph);
1360 			break;
1361 #ifdef INET6
1362 		case AF_INET6:
1363 			tp = in6totcpcb((struct in6pcb *)inph);
1364 			break;
1365 #endif
1366 		default:
1367 			tp = NULL;
1368 			break;
1369 		}
1370 		if (tp != NULL) {
1371 			/*
1372 			 * We may be called from a device's interrupt
1373 			 * context.  If the tcpcb is already busy,
1374 			 * just bail out now.
1375 			 */
1376 			if (tcp_reass_lock_try(tp) == 0)
1377 				continue;
1378 			if (tcp_freeq(tp))
1379 				TCP_STATINC(TCP_STAT_CONNSDRAINED);
1380 			TCP_REASS_UNLOCK(tp);
1381 		}
1382 	}
1383 
1384 	KERNEL_UNLOCK_ONE(NULL);
1385 	mutex_exit(softnet_lock);
1386 }
1387 
1388 /*
1389  * Notify a tcp user of an asynchronous error;
1390  * store error as soft error, but wake up user
1391  * (for now, won't do anything until can select for soft error).
1392  */
1393 void
1394 tcp_notify(struct inpcb *inp, int error)
1395 {
1396 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1397 	struct socket *so = inp->inp_socket;
1398 
1399 	/*
1400 	 * Ignore some errors if we are hooked up.
1401 	 * If connection hasn't completed, has retransmitted several times,
1402 	 * and receives a second error, give up now.  This is better
1403 	 * than waiting a long time to establish a connection that
1404 	 * can never complete.
1405 	 */
1406 	if (tp->t_state == TCPS_ESTABLISHED &&
1407 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1408 	      error == EHOSTDOWN)) {
1409 		return;
1410 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1411 	    tp->t_rxtshift > 3 && tp->t_softerror)
1412 		so->so_error = error;
1413 	else
1414 		tp->t_softerror = error;
1415 	cv_broadcast(&so->so_cv);
1416 	sorwakeup(so);
1417 	sowwakeup(so);
1418 }
1419 
1420 #ifdef INET6
1421 void
1422 tcp6_notify(struct in6pcb *in6p, int error)
1423 {
1424 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1425 	struct socket *so = in6p->in6p_socket;
1426 
1427 	/*
1428 	 * Ignore some errors if we are hooked up.
1429 	 * If connection hasn't completed, has retransmitted several times,
1430 	 * and receives a second error, give up now.  This is better
1431 	 * than waiting a long time to establish a connection that
1432 	 * can never complete.
1433 	 */
1434 	if (tp->t_state == TCPS_ESTABLISHED &&
1435 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1436 	      error == EHOSTDOWN)) {
1437 		return;
1438 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1439 	    tp->t_rxtshift > 3 && tp->t_softerror)
1440 		so->so_error = error;
1441 	else
1442 		tp->t_softerror = error;
1443 	cv_broadcast(&so->so_cv);
1444 	sorwakeup(so);
1445 	sowwakeup(so);
1446 }
1447 #endif
1448 
1449 #ifdef INET6
1450 void *
1451 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1452 {
1453 	struct tcphdr th;
1454 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1455 	int nmatch;
1456 	struct ip6_hdr *ip6;
1457 	const struct sockaddr_in6 *sa6_src = NULL;
1458 	const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1459 	struct mbuf *m;
1460 	int off;
1461 
1462 	if (sa->sa_family != AF_INET6 ||
1463 	    sa->sa_len != sizeof(struct sockaddr_in6))
1464 		return NULL;
1465 	if ((unsigned)cmd >= PRC_NCMDS)
1466 		return NULL;
1467 	else if (cmd == PRC_QUENCH) {
1468 		/*
1469 		 * Don't honor ICMP Source Quench messages meant for
1470 		 * TCP connections.
1471 		 */
1472 		return NULL;
1473 	} else if (PRC_IS_REDIRECT(cmd))
1474 		notify = in6_rtchange, d = NULL;
1475 	else if (cmd == PRC_MSGSIZE)
1476 		; /* special code is present, see below */
1477 	else if (cmd == PRC_HOSTDEAD)
1478 		d = NULL;
1479 	else if (inet6ctlerrmap[cmd] == 0)
1480 		return NULL;
1481 
1482 	/* if the parameter is from icmp6, decode it. */
1483 	if (d != NULL) {
1484 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1485 		m = ip6cp->ip6c_m;
1486 		ip6 = ip6cp->ip6c_ip6;
1487 		off = ip6cp->ip6c_off;
1488 		sa6_src = ip6cp->ip6c_src;
1489 	} else {
1490 		m = NULL;
1491 		ip6 = NULL;
1492 		sa6_src = &sa6_any;
1493 		off = 0;
1494 	}
1495 
1496 	if (ip6) {
1497 		/*
1498 		 * XXX: We assume that when ip6 is non NULL,
1499 		 * M and OFF are valid.
1500 		 */
1501 
1502 		/* check if we can safely examine src and dst ports */
1503 		if (m->m_pkthdr.len < off + sizeof(th)) {
1504 			if (cmd == PRC_MSGSIZE)
1505 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1506 			return NULL;
1507 		}
1508 
1509 		memset(&th, 0, sizeof(th));
1510 		m_copydata(m, off, sizeof(th), (void *)&th);
1511 
1512 		if (cmd == PRC_MSGSIZE) {
1513 			int valid = 0;
1514 
1515 			/*
1516 			 * Check to see if we have a valid TCP connection
1517 			 * corresponding to the address in the ICMPv6 message
1518 			 * payload.
1519 			 */
1520 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1521 			    th.th_dport,
1522 			    (const struct in6_addr *)&sa6_src->sin6_addr,
1523 						  th.th_sport, 0, 0))
1524 				valid++;
1525 
1526 			/*
1527 			 * Depending on the value of "valid" and routing table
1528 			 * size (mtudisc_{hi,lo}wat), we will:
1529 			 * - recalcurate the new MTU and create the
1530 			 *   corresponding routing entry, or
1531 			 * - ignore the MTU change notification.
1532 			 */
1533 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1534 
1535 			/*
1536 			 * no need to call in6_pcbnotify, it should have been
1537 			 * called via callback if necessary
1538 			 */
1539 			return NULL;
1540 		}
1541 
1542 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1543 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1544 		if (nmatch == 0 && syn_cache_count &&
1545 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1546 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1547 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1548 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1549 					  sa, &th);
1550 	} else {
1551 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1552 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1553 	}
1554 
1555 	return NULL;
1556 }
1557 #endif
1558 
1559 #ifdef INET
1560 /* assumes that ip header and tcp header are contiguous on mbuf */
1561 void *
1562 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1563 {
1564 	struct ip *ip = v;
1565 	struct tcphdr *th;
1566 	struct icmp *icp;
1567 	extern const int inetctlerrmap[];
1568 	void (*notify)(struct inpcb *, int) = tcp_notify;
1569 	int errno;
1570 	int nmatch;
1571 	struct tcpcb *tp;
1572 	u_int mtu;
1573 	tcp_seq seq;
1574 	struct inpcb *inp;
1575 #ifdef INET6
1576 	struct in6pcb *in6p;
1577 	struct in6_addr src6, dst6;
1578 #endif
1579 
1580 	if (sa->sa_family != AF_INET ||
1581 	    sa->sa_len != sizeof(struct sockaddr_in))
1582 		return NULL;
1583 	if ((unsigned)cmd >= PRC_NCMDS)
1584 		return NULL;
1585 	errno = inetctlerrmap[cmd];
1586 	if (cmd == PRC_QUENCH)
1587 		/*
1588 		 * Don't honor ICMP Source Quench messages meant for
1589 		 * TCP connections.
1590 		 */
1591 		return NULL;
1592 	else if (PRC_IS_REDIRECT(cmd))
1593 		notify = in_rtchange, ip = 0;
1594 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1595 		/*
1596 		 * Check to see if we have a valid TCP connection
1597 		 * corresponding to the address in the ICMP message
1598 		 * payload.
1599 		 *
1600 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1601 		 */
1602 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1603 #ifdef INET6
1604 		memset(&src6, 0, sizeof(src6));
1605 		memset(&dst6, 0, sizeof(dst6));
1606 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1607 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1608 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1609 #endif
1610 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1611 						th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1612 #ifdef INET6
1613 			in6p = NULL;
1614 #else
1615 			;
1616 #endif
1617 #ifdef INET6
1618 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1619 						       th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1620 			;
1621 #endif
1622 		else
1623 			return NULL;
1624 
1625 		/*
1626 		 * Now that we've validated that we are actually communicating
1627 		 * with the host indicated in the ICMP message, locate the
1628 		 * ICMP header, recalculate the new MTU, and create the
1629 		 * corresponding routing entry.
1630 		 */
1631 		icp = (struct icmp *)((char *)ip -
1632 		    offsetof(struct icmp, icmp_ip));
1633 		if (inp) {
1634 			if ((tp = intotcpcb(inp)) == NULL)
1635 				return NULL;
1636 		}
1637 #ifdef INET6
1638 		else if (in6p) {
1639 			if ((tp = in6totcpcb(in6p)) == NULL)
1640 				return NULL;
1641 		}
1642 #endif
1643 		else
1644 			return NULL;
1645 		seq = ntohl(th->th_seq);
1646 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1647 			return NULL;
1648 		/*
1649 		 * If the ICMP message advertises a Next-Hop MTU
1650 		 * equal or larger than the maximum packet size we have
1651 		 * ever sent, drop the message.
1652 		 */
1653 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1654 		if (mtu >= tp->t_pmtud_mtu_sent)
1655 			return NULL;
1656 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1657 			/*
1658 			 * Calculate new MTU, and create corresponding
1659 			 * route (traditional PMTUD).
1660 			 */
1661 			tp->t_flags &= ~TF_PMTUD_PEND;
1662 			icmp_mtudisc(icp, ip->ip_dst);
1663 		} else {
1664 			/*
1665 			 * Record the information got in the ICMP
1666 			 * message; act on it later.
1667 			 * If we had already recorded an ICMP message,
1668 			 * replace the old one only if the new message
1669 			 * refers to an older TCP segment
1670 			 */
1671 			if (tp->t_flags & TF_PMTUD_PEND) {
1672 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1673 					return NULL;
1674 			} else
1675 				tp->t_flags |= TF_PMTUD_PEND;
1676 			tp->t_pmtud_th_seq = seq;
1677 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1678 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1679 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1680 		}
1681 		return NULL;
1682 	} else if (cmd == PRC_HOSTDEAD)
1683 		ip = 0;
1684 	else if (errno == 0)
1685 		return NULL;
1686 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1687 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1688 		nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1689 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1690 		if (nmatch == 0 && syn_cache_count &&
1691 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1692 		    inetctlerrmap[cmd] == ENETUNREACH ||
1693 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1694 			struct sockaddr_in sin;
1695 			memset(&sin, 0, sizeof(sin));
1696 			sin.sin_len = sizeof(sin);
1697 			sin.sin_family = AF_INET;
1698 			sin.sin_port = th->th_sport;
1699 			sin.sin_addr = ip->ip_src;
1700 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1701 		}
1702 
1703 		/* XXX mapped address case */
1704 	} else
1705 		in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1706 		    notify);
1707 	return NULL;
1708 }
1709 
1710 /*
1711  * When a source quench is received, we are being notified of congestion.
1712  * Close the congestion window down to the Loss Window (one segment).
1713  * We will gradually open it again as we proceed.
1714  */
1715 void
1716 tcp_quench(struct inpcb *inp, int errno)
1717 {
1718 	struct tcpcb *tp = intotcpcb(inp);
1719 
1720 	if (tp) {
1721 		tp->snd_cwnd = tp->t_segsz;
1722 		tp->t_bytes_acked = 0;
1723 	}
1724 }
1725 #endif
1726 
1727 #ifdef INET6
1728 void
1729 tcp6_quench(struct in6pcb *in6p, int errno)
1730 {
1731 	struct tcpcb *tp = in6totcpcb(in6p);
1732 
1733 	if (tp) {
1734 		tp->snd_cwnd = tp->t_segsz;
1735 		tp->t_bytes_acked = 0;
1736 	}
1737 }
1738 #endif
1739 
1740 #ifdef INET
1741 /*
1742  * Path MTU Discovery handlers.
1743  */
1744 void
1745 tcp_mtudisc_callback(struct in_addr faddr)
1746 {
1747 #ifdef INET6
1748 	struct in6_addr in6;
1749 #endif
1750 
1751 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1752 #ifdef INET6
1753 	memset(&in6, 0, sizeof(in6));
1754 	in6.s6_addr16[5] = 0xffff;
1755 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1756 	tcp6_mtudisc_callback(&in6);
1757 #endif
1758 }
1759 
1760 /*
1761  * On receipt of path MTU corrections, flush old route and replace it
1762  * with the new one.  Retransmit all unacknowledged packets, to ensure
1763  * that all packets will be received.
1764  */
1765 void
1766 tcp_mtudisc(struct inpcb *inp, int errno)
1767 {
1768 	struct tcpcb *tp = intotcpcb(inp);
1769 	struct rtentry *rt = in_pcbrtentry(inp);
1770 
1771 	if (tp != 0) {
1772 		if (rt != 0) {
1773 			/*
1774 			 * If this was not a host route, remove and realloc.
1775 			 */
1776 			if ((rt->rt_flags & RTF_HOST) == 0) {
1777 				in_rtchange(inp, errno);
1778 				if ((rt = in_pcbrtentry(inp)) == 0)
1779 					return;
1780 			}
1781 
1782 			/*
1783 			 * Slow start out of the error condition.  We
1784 			 * use the MTU because we know it's smaller
1785 			 * than the previously transmitted segment.
1786 			 *
1787 			 * Note: This is more conservative than the
1788 			 * suggestion in draft-floyd-incr-init-win-03.
1789 			 */
1790 			if (rt->rt_rmx.rmx_mtu != 0)
1791 				tp->snd_cwnd =
1792 				    TCP_INITIAL_WINDOW(tcp_init_win,
1793 				    rt->rt_rmx.rmx_mtu);
1794 		}
1795 
1796 		/*
1797 		 * Resend unacknowledged packets.
1798 		 */
1799 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1800 		tcp_output(tp);
1801 	}
1802 }
1803 #endif
1804 
1805 #ifdef INET6
1806 /*
1807  * Path MTU Discovery handlers.
1808  */
1809 void
1810 tcp6_mtudisc_callback(struct in6_addr *faddr)
1811 {
1812 	struct sockaddr_in6 sin6;
1813 
1814 	memset(&sin6, 0, sizeof(sin6));
1815 	sin6.sin6_family = AF_INET6;
1816 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1817 	sin6.sin6_addr = *faddr;
1818 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1819 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1820 }
1821 
1822 void
1823 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1824 {
1825 	struct tcpcb *tp = in6totcpcb(in6p);
1826 	struct rtentry *rt = in6_pcbrtentry(in6p);
1827 
1828 	if (tp != 0) {
1829 		if (rt != 0) {
1830 			/*
1831 			 * If this was not a host route, remove and realloc.
1832 			 */
1833 			if ((rt->rt_flags & RTF_HOST) == 0) {
1834 				in6_rtchange(in6p, errno);
1835 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1836 					return;
1837 			}
1838 
1839 			/*
1840 			 * Slow start out of the error condition.  We
1841 			 * use the MTU because we know it's smaller
1842 			 * than the previously transmitted segment.
1843 			 *
1844 			 * Note: This is more conservative than the
1845 			 * suggestion in draft-floyd-incr-init-win-03.
1846 			 */
1847 			if (rt->rt_rmx.rmx_mtu != 0)
1848 				tp->snd_cwnd =
1849 				    TCP_INITIAL_WINDOW(tcp_init_win,
1850 				    rt->rt_rmx.rmx_mtu);
1851 		}
1852 
1853 		/*
1854 		 * Resend unacknowledged packets.
1855 		 */
1856 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1857 		tcp_output(tp);
1858 	}
1859 }
1860 #endif /* INET6 */
1861 
1862 /*
1863  * Compute the MSS to advertise to the peer.  Called only during
1864  * the 3-way handshake.  If we are the server (peer initiated
1865  * connection), we are called with a pointer to the interface
1866  * on which the SYN packet arrived.  If we are the client (we
1867  * initiated connection), we are called with a pointer to the
1868  * interface out which this connection should go.
1869  *
1870  * NOTE: Do not subtract IP option/extension header size nor IPsec
1871  * header size from MSS advertisement.  MSS option must hold the maximum
1872  * segment size we can accept, so it must always be:
1873  *	 max(if mtu) - ip header - tcp header
1874  */
1875 u_long
1876 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1877 {
1878 	extern u_long in_maxmtu;
1879 	u_long mss = 0;
1880 	u_long hdrsiz;
1881 
1882 	/*
1883 	 * In order to avoid defeating path MTU discovery on the peer,
1884 	 * we advertise the max MTU of all attached networks as our MSS,
1885 	 * per RFC 1191, section 3.1.
1886 	 *
1887 	 * We provide the option to advertise just the MTU of
1888 	 * the interface on which we hope this connection will
1889 	 * be receiving.  If we are responding to a SYN, we
1890 	 * will have a pretty good idea about this, but when
1891 	 * initiating a connection there is a bit more doubt.
1892 	 *
1893 	 * We also need to ensure that loopback has a large enough
1894 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1895 	 */
1896 
1897 	if (ifp != NULL)
1898 		switch (af) {
1899 		case AF_INET:
1900 			mss = ifp->if_mtu;
1901 			break;
1902 #ifdef INET6
1903 		case AF_INET6:
1904 			mss = IN6_LINKMTU(ifp);
1905 			break;
1906 #endif
1907 		}
1908 
1909 	if (tcp_mss_ifmtu == 0)
1910 		switch (af) {
1911 		case AF_INET:
1912 			mss = max(in_maxmtu, mss);
1913 			break;
1914 #ifdef INET6
1915 		case AF_INET6:
1916 			mss = max(in6_maxmtu, mss);
1917 			break;
1918 #endif
1919 		}
1920 
1921 	switch (af) {
1922 	case AF_INET:
1923 		hdrsiz = sizeof(struct ip);
1924 		break;
1925 #ifdef INET6
1926 	case AF_INET6:
1927 		hdrsiz = sizeof(struct ip6_hdr);
1928 		break;
1929 #endif
1930 	default:
1931 		hdrsiz = 0;
1932 		break;
1933 	}
1934 	hdrsiz += sizeof(struct tcphdr);
1935 	if (mss > hdrsiz)
1936 		mss -= hdrsiz;
1937 
1938 	mss = max(tcp_mssdflt, mss);
1939 	return (mss);
1940 }
1941 
1942 /*
1943  * Set connection variables based on the peer's advertised MSS.
1944  * We are passed the TCPCB for the actual connection.  If we
1945  * are the server, we are called by the compressed state engine
1946  * when the 3-way handshake is complete.  If we are the client,
1947  * we are called when we receive the SYN,ACK from the server.
1948  *
1949  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1950  * before this routine is called!
1951  */
1952 void
1953 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1954 {
1955 	struct socket *so;
1956 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1957 	struct rtentry *rt;
1958 #endif
1959 	u_long bufsize;
1960 	int mss;
1961 
1962 #ifdef DIAGNOSTIC
1963 	if (tp->t_inpcb && tp->t_in6pcb)
1964 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1965 #endif
1966 	so = NULL;
1967 	rt = NULL;
1968 #ifdef INET
1969 	if (tp->t_inpcb) {
1970 		so = tp->t_inpcb->inp_socket;
1971 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1972 		rt = in_pcbrtentry(tp->t_inpcb);
1973 #endif
1974 	}
1975 #endif
1976 #ifdef INET6
1977 	if (tp->t_in6pcb) {
1978 		so = tp->t_in6pcb->in6p_socket;
1979 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1980 		rt = in6_pcbrtentry(tp->t_in6pcb);
1981 #endif
1982 	}
1983 #endif
1984 
1985 	/*
1986 	 * As per RFC1122, use the default MSS value, unless they
1987 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1988 	 */
1989 	mss = tcp_mssdflt;
1990 	if (offer)
1991 		mss = offer;
1992 	mss = max(mss, 256);		/* sanity */
1993 	tp->t_peermss = mss;
1994 	mss -= tcp_optlen(tp);
1995 #ifdef INET
1996 	if (tp->t_inpcb)
1997 		mss -= ip_optlen(tp->t_inpcb);
1998 #endif
1999 #ifdef INET6
2000 	if (tp->t_in6pcb)
2001 		mss -= ip6_optlen(tp->t_in6pcb);
2002 #endif
2003 
2004 	/*
2005 	 * If there's a pipesize, change the socket buffer to that size.
2006 	 * Make the socket buffer an integral number of MSS units.  If
2007 	 * the MSS is larger than the socket buffer, artificially decrease
2008 	 * the MSS.
2009 	 */
2010 #ifdef RTV_SPIPE
2011 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
2012 		bufsize = rt->rt_rmx.rmx_sendpipe;
2013 	else
2014 #endif
2015 	{
2016 		KASSERT(so != NULL);
2017 		bufsize = so->so_snd.sb_hiwat;
2018 	}
2019 	if (bufsize < mss)
2020 		mss = bufsize;
2021 	else {
2022 		bufsize = roundup(bufsize, mss);
2023 		if (bufsize > sb_max)
2024 			bufsize = sb_max;
2025 		(void) sbreserve(&so->so_snd, bufsize, so);
2026 	}
2027 	tp->t_segsz = mss;
2028 
2029 #ifdef RTV_SSTHRESH
2030 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2031 		/*
2032 		 * There's some sort of gateway or interface buffer
2033 		 * limit on the path.  Use this to set the slow
2034 		 * start threshold, but set the threshold to no less
2035 		 * than 2 * MSS.
2036 		 */
2037 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2038 	}
2039 #endif
2040 }
2041 
2042 /*
2043  * Processing necessary when a TCP connection is established.
2044  */
2045 void
2046 tcp_established(struct tcpcb *tp)
2047 {
2048 	struct socket *so;
2049 #ifdef RTV_RPIPE
2050 	struct rtentry *rt;
2051 #endif
2052 	u_long bufsize;
2053 
2054 #ifdef DIAGNOSTIC
2055 	if (tp->t_inpcb && tp->t_in6pcb)
2056 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
2057 #endif
2058 	so = NULL;
2059 	rt = NULL;
2060 #ifdef INET
2061 	/* This is a while() to reduce the dreadful stairstepping below */
2062 	while (tp->t_inpcb) {
2063 		so = tp->t_inpcb->inp_socket;
2064 #if defined(RTV_RPIPE)
2065 		rt = in_pcbrtentry(tp->t_inpcb);
2066 #endif
2067 		if (__predict_true(tcp_msl_enable)) {
2068 			if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2069 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2070 				break;
2071 			}
2072 
2073 			if (__predict_false(tcp_rttlocal)) {
2074 				/* This may be adjusted by tcp_input */
2075 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2076 				break;
2077 			}
2078 			if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2079 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2080 				break;
2081 			}
2082 		}
2083 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2084 		break;
2085 	}
2086 #endif
2087 #ifdef INET6
2088 	/* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2089 	while (!tp->t_inpcb && tp->t_in6pcb) {
2090 		so = tp->t_in6pcb->in6p_socket;
2091 #if defined(RTV_RPIPE)
2092 		rt = in6_pcbrtentry(tp->t_in6pcb);
2093 #endif
2094 		if (__predict_true(tcp_msl_enable)) {
2095 			extern const struct in6_addr in6addr_loopback;
2096 
2097 			if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2098 					       &in6addr_loopback)) {
2099 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2100 				break;
2101 			}
2102 
2103 			if (__predict_false(tcp_rttlocal)) {
2104 				/* This may be adjusted by tcp_input */
2105 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2106 				break;
2107 			}
2108 			if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2109 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2110 				break;
2111 			}
2112 		}
2113 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2114 		break;
2115 	}
2116 #endif
2117 
2118 	tp->t_state = TCPS_ESTABLISHED;
2119 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2120 
2121 #ifdef RTV_RPIPE
2122 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2123 		bufsize = rt->rt_rmx.rmx_recvpipe;
2124 	else
2125 #endif
2126 	{
2127 		KASSERT(so != NULL);
2128 		bufsize = so->so_rcv.sb_hiwat;
2129 	}
2130 	if (bufsize > tp->t_ourmss) {
2131 		bufsize = roundup(bufsize, tp->t_ourmss);
2132 		if (bufsize > sb_max)
2133 			bufsize = sb_max;
2134 		(void) sbreserve(&so->so_rcv, bufsize, so);
2135 	}
2136 }
2137 
2138 /*
2139  * Check if there's an initial rtt or rttvar.  Convert from the
2140  * route-table units to scaled multiples of the slow timeout timer.
2141  * Called only during the 3-way handshake.
2142  */
2143 void
2144 tcp_rmx_rtt(struct tcpcb *tp)
2145 {
2146 #ifdef RTV_RTT
2147 	struct rtentry *rt = NULL;
2148 	int rtt;
2149 
2150 #ifdef DIAGNOSTIC
2151 	if (tp->t_inpcb && tp->t_in6pcb)
2152 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2153 #endif
2154 #ifdef INET
2155 	if (tp->t_inpcb)
2156 		rt = in_pcbrtentry(tp->t_inpcb);
2157 #endif
2158 #ifdef INET6
2159 	if (tp->t_in6pcb)
2160 		rt = in6_pcbrtentry(tp->t_in6pcb);
2161 #endif
2162 	if (rt == NULL)
2163 		return;
2164 
2165 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2166 		/*
2167 		 * XXX The lock bit for MTU indicates that the value
2168 		 * is also a minimum value; this is subject to time.
2169 		 */
2170 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2171 			TCPT_RANGESET(tp->t_rttmin,
2172 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2173 			    TCPTV_MIN, TCPTV_REXMTMAX);
2174 		tp->t_srtt = rtt /
2175 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2176 		if (rt->rt_rmx.rmx_rttvar) {
2177 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2178 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2179 				(TCP_RTTVAR_SHIFT + 2));
2180 		} else {
2181 			/* Default variation is +- 1 rtt */
2182 			tp->t_rttvar =
2183 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2184 		}
2185 		TCPT_RANGESET(tp->t_rxtcur,
2186 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2187 		    tp->t_rttmin, TCPTV_REXMTMAX);
2188 	}
2189 #endif
2190 }
2191 
2192 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2193 
2194 /*
2195  * Get a new sequence value given a tcp control block
2196  */
2197 tcp_seq
2198 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2199 {
2200 
2201 #ifdef INET
2202 	if (tp->t_inpcb != NULL) {
2203 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2204 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2205 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2206 		    addin));
2207 	}
2208 #endif
2209 #ifdef INET6
2210 	if (tp->t_in6pcb != NULL) {
2211 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2212 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2213 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2214 		    addin));
2215 	}
2216 #endif
2217 	/* Not possible. */
2218 	panic("tcp_new_iss");
2219 }
2220 
2221 static u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2222 
2223 /*
2224  * Initialize RFC 1948 ISS Secret
2225  */
2226 static int
2227 tcp_iss_secret_init(void)
2228 {
2229 	cprng_strong(kern_cprng,
2230 	    tcp_iss_secret, sizeof(tcp_iss_secret), 0);
2231 
2232 	return 0;
2233 }
2234 
2235 /*
2236  * This routine actually generates a new TCP initial sequence number.
2237  */
2238 tcp_seq
2239 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2240     size_t addrsz, tcp_seq addin)
2241 {
2242 	tcp_seq tcp_iss;
2243 
2244 	if (tcp_do_rfc1948) {
2245 		MD5_CTX ctx;
2246 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2247 		static ONCE_DECL(tcp_iss_secret_control);
2248 
2249 		/*
2250 		 * If we haven't been here before, initialize our cryptographic
2251 		 * hash secret.
2252 		 */
2253 		RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init);
2254 
2255 		/*
2256 		 * Compute the base value of the ISS.  It is a hash
2257 		 * of (saddr, sport, daddr, dport, secret).
2258 		 */
2259 		MD5Init(&ctx);
2260 
2261 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2262 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2263 
2264 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2265 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2266 
2267 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2268 
2269 		MD5Final(hash, &ctx);
2270 
2271 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2272 
2273 		/*
2274 		 * Now increment our "timer", and add it in to
2275 		 * the computed value.
2276 		 *
2277 		 * XXX Use `addin'?
2278 		 * XXX TCP_ISSINCR too large to use?
2279 		 */
2280 		tcp_iss_seq += TCP_ISSINCR;
2281 #ifdef TCPISS_DEBUG
2282 		printf("ISS hash 0x%08x, ", tcp_iss);
2283 #endif
2284 		tcp_iss += tcp_iss_seq + addin;
2285 #ifdef TCPISS_DEBUG
2286 		printf("new ISS 0x%08x\n", tcp_iss);
2287 #endif
2288 	} else {
2289 		/*
2290 		 * Randomize.
2291 		 */
2292 		tcp_iss = cprng_fast32();
2293 
2294 		/*
2295 		 * If we were asked to add some amount to a known value,
2296 		 * we will take a random value obtained above, mask off
2297 		 * the upper bits, and add in the known value.  We also
2298 		 * add in a constant to ensure that we are at least a
2299 		 * certain distance from the original value.
2300 		 *
2301 		 * This is used when an old connection is in timed wait
2302 		 * and we have a new one coming in, for instance.
2303 		 */
2304 		if (addin != 0) {
2305 #ifdef TCPISS_DEBUG
2306 			printf("Random %08x, ", tcp_iss);
2307 #endif
2308 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2309 			tcp_iss += addin + TCP_ISSINCR;
2310 #ifdef TCPISS_DEBUG
2311 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2312 #endif
2313 		} else {
2314 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2315 			tcp_iss += tcp_iss_seq;
2316 			tcp_iss_seq += TCP_ISSINCR;
2317 #ifdef TCPISS_DEBUG
2318 			printf("ISS %08x\n", tcp_iss);
2319 #endif
2320 		}
2321 	}
2322 
2323 	if (tcp_compat_42) {
2324 		/*
2325 		 * Limit it to the positive range for really old TCP
2326 		 * implementations.
2327 		 * Just AND off the top bit instead of checking if
2328 		 * is set first - saves a branch 50% of the time.
2329 		 */
2330 		tcp_iss &= 0x7fffffff;		/* XXX */
2331 	}
2332 
2333 	return (tcp_iss);
2334 }
2335 
2336 #if defined(IPSEC)
2337 /* compute ESP/AH header size for TCP, including outer IP header. */
2338 size_t
2339 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2340 {
2341 	struct inpcb *inp;
2342 	size_t hdrsiz;
2343 
2344 	/* XXX mapped addr case (tp->t_in6pcb) */
2345 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2346 		return 0;
2347 	switch (tp->t_family) {
2348 	case AF_INET:
2349 		/* XXX: should use currect direction. */
2350 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2351 		break;
2352 	default:
2353 		hdrsiz = 0;
2354 		break;
2355 	}
2356 
2357 	return hdrsiz;
2358 }
2359 
2360 #ifdef INET6
2361 size_t
2362 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2363 {
2364 	struct in6pcb *in6p;
2365 	size_t hdrsiz;
2366 
2367 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2368 		return 0;
2369 	switch (tp->t_family) {
2370 	case AF_INET6:
2371 		/* XXX: should use currect direction. */
2372 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2373 		break;
2374 	case AF_INET:
2375 		/* mapped address case - tricky */
2376 	default:
2377 		hdrsiz = 0;
2378 		break;
2379 	}
2380 
2381 	return hdrsiz;
2382 }
2383 #endif
2384 #endif /*IPSEC*/
2385 
2386 /*
2387  * Determine the length of the TCP options for this connection.
2388  *
2389  * XXX:  What do we do for SACK, when we add that?  Just reserve
2390  *       all of the space?  Otherwise we can't exactly be incrementing
2391  *       cwnd by an amount that varies depending on the amount we last
2392  *       had to SACK!
2393  */
2394 
2395 u_int
2396 tcp_optlen(struct tcpcb *tp)
2397 {
2398 	u_int optlen;
2399 
2400 	optlen = 0;
2401 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2402 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2403 		optlen += TCPOLEN_TSTAMP_APPA;
2404 
2405 #ifdef TCP_SIGNATURE
2406 	if (tp->t_flags & TF_SIGNATURE)
2407 		optlen += TCPOLEN_SIGNATURE + 2;
2408 #endif /* TCP_SIGNATURE */
2409 
2410 	return optlen;
2411 }
2412 
2413 u_int
2414 tcp_hdrsz(struct tcpcb *tp)
2415 {
2416 	u_int hlen;
2417 
2418 	switch (tp->t_family) {
2419 #ifdef INET6
2420 	case AF_INET6:
2421 		hlen = sizeof(struct ip6_hdr);
2422 		break;
2423 #endif
2424 	case AF_INET:
2425 		hlen = sizeof(struct ip);
2426 		break;
2427 	default:
2428 		hlen = 0;
2429 		break;
2430 	}
2431 	hlen += sizeof(struct tcphdr);
2432 
2433 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2434 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2435 		hlen += TCPOLEN_TSTAMP_APPA;
2436 #ifdef TCP_SIGNATURE
2437 	if (tp->t_flags & TF_SIGNATURE)
2438 		hlen += TCPOLEN_SIGLEN;
2439 #endif
2440 	return hlen;
2441 }
2442 
2443 void
2444 tcp_statinc(u_int stat)
2445 {
2446 
2447 	KASSERT(stat < TCP_NSTATS);
2448 	TCP_STATINC(stat);
2449 }
2450 
2451 void
2452 tcp_statadd(u_int stat, uint64_t val)
2453 {
2454 
2455 	KASSERT(stat < TCP_NSTATS);
2456 	TCP_STATADD(stat, val);
2457 }
2458