xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: tcp_subr.c,v 1.252 2013/11/23 14:20:21 christos 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 1.252 2013/11/23 14:20:21 christos 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/proc.h>
104 #include <sys/systm.h>
105 #include <sys/malloc.h>
106 #include <sys/mbuf.h>
107 #include <sys/socket.h>
108 #include <sys/socketvar.h>
109 #include <sys/protosw.h>
110 #include <sys/errno.h>
111 #include <sys/kernel.h>
112 #include <sys/pool.h>
113 #include <sys/md5.h>
114 #include <sys/cprng.h>
115 
116 #include <net/route.h>
117 #include <net/if.h>
118 
119 #include <netinet/in.h>
120 #include <netinet/in_systm.h>
121 #include <netinet/ip.h>
122 #include <netinet/in_pcb.h>
123 #include <netinet/ip_var.h>
124 #include <netinet/ip_icmp.h>
125 
126 #ifdef INET6
127 #ifndef INET
128 #include <netinet/in.h>
129 #endif
130 #include <netinet/ip6.h>
131 #include <netinet6/in6_pcb.h>
132 #include <netinet6/ip6_var.h>
133 #include <netinet6/in6_var.h>
134 #include <netinet6/ip6protosw.h>
135 #include <netinet/icmp6.h>
136 #include <netinet6/nd6.h>
137 #endif
138 
139 #include <netinet/tcp.h>
140 #include <netinet/tcp_fsm.h>
141 #include <netinet/tcp_seq.h>
142 #include <netinet/tcp_timer.h>
143 #include <netinet/tcp_var.h>
144 #include <netinet/tcp_vtw.h>
145 #include <netinet/tcp_private.h>
146 #include <netinet/tcp_congctl.h>
147 #include <netinet/tcpip.h>
148 
149 #ifdef IPSEC
150 #include <netipsec/ipsec.h>
151 #include <netipsec/xform.h>
152 #ifdef INET6
153 #include <netipsec/ipsec6.h>
154 #endif
155  #include <netipsec/key.h>
156 #endif	/* IPSEC*/
157 
158 
159 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
160 u_int32_t tcp_now;		/* slow ticks, for RFC 1323 timestamps */
161 
162 percpu_t *tcpstat_percpu;
163 
164 /* patchable/settable parameters for tcp */
165 int 	tcp_mssdflt = TCP_MSS;
166 int	tcp_minmss = TCP_MINMSS;
167 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
168 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
169 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
170 int	tcp_do_sack = 1;	/* selective acknowledgement */
171 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
172 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
173 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
174 int	tcp_do_ecn = 0;		/* Explicit Congestion Notification */
175 #ifndef TCP_INIT_WIN
176 #define	TCP_INIT_WIN	4	/* initial slow start window */
177 #endif
178 #ifndef TCP_INIT_WIN_LOCAL
179 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
180 #endif
181 /*
182  * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460.
183  * This is to simulate current behavior for iw == 4
184  */
185 int tcp_init_win_max[] = {
186 	 1 * 1460,
187 	 1 * 1460,
188 	 2 * 1460,
189 	 2 * 1460,
190 	 3 * 1460,
191 	 5 * 1460,
192 	 6 * 1460,
193 	 7 * 1460,
194 	 8 * 1460,
195 	 9 * 1460,
196 	10 * 1460
197 };
198 int	tcp_init_win = TCP_INIT_WIN;
199 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
200 int	tcp_mss_ifmtu = 0;
201 #ifdef TCP_COMPAT_42
202 int	tcp_compat_42 = 1;
203 #else
204 int	tcp_compat_42 = 0;
205 #endif
206 int	tcp_rst_ppslim = 100;	/* 100pps */
207 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
208 int	tcp_do_loopback_cksum = 0;
209 int	tcp_do_abc = 1;		/* RFC3465 Appropriate byte counting. */
210 int	tcp_abc_aggressive = 1;	/* 1: L=2*SMSS  0: L=1*SMSS */
211 int	tcp_sack_tp_maxholes = 32;
212 int	tcp_sack_globalmaxholes = 1024;
213 int	tcp_sack_globalholes = 0;
214 int	tcp_ecn_maxretries = 1;
215 int	tcp_msl_enable = 1;		/* enable TIME_WAIT truncation	*/
216 int	tcp_msl_loop   = PR_SLOWHZ;	/* MSL for loopback		*/
217 int	tcp_msl_local  = 5 * PR_SLOWHZ;	/* MSL for 'local'		*/
218 int	tcp_msl_remote = TCPTV_MSL;	/* MSL otherwise		*/
219 int	tcp_msl_remote_threshold = TCPTV_SRTTDFLT;	/* RTT threshold */
220 int	tcp_rttlocal = 0;		/* Use RTT to decide who's 'local' */
221 
222 int	tcp4_vtw_enable = 0;		/* 1 to enable */
223 int	tcp6_vtw_enable = 0;		/* 1 to enable */
224 int	tcp_vtw_was_enabled = 0;
225 int	tcp_vtw_entries = 1 << 16;	/* 64K vestigial TIME_WAIT entries */
226 
227 /* tcb hash */
228 #ifndef TCBHASHSIZE
229 #define	TCBHASHSIZE	128
230 #endif
231 int	tcbhashsize = TCBHASHSIZE;
232 
233 /* syn hash parameters */
234 #define	TCP_SYN_HASH_SIZE	293
235 #define	TCP_SYN_BUCKET_SIZE	35
236 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
237 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
238 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
239 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
240 
241 int	tcp_freeq(struct tcpcb *);
242 
243 #ifdef INET
244 void	tcp_mtudisc_callback(struct in_addr);
245 #endif
246 #ifdef INET6
247 void	tcp6_mtudisc_callback(struct in6_addr *);
248 #endif
249 
250 #ifdef INET6
251 void	tcp6_mtudisc(struct in6pcb *, int);
252 #endif
253 
254 static struct pool tcpcb_pool;
255 
256 static int tcp_drainwanted;
257 
258 #ifdef TCP_CSUM_COUNTERS
259 #include <sys/device.h>
260 
261 #if defined(INET)
262 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
263     NULL, "tcp", "hwcsum bad");
264 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
265     NULL, "tcp", "hwcsum ok");
266 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
267     NULL, "tcp", "hwcsum data");
268 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
269     NULL, "tcp", "swcsum");
270 
271 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
272 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
273 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
274 EVCNT_ATTACH_STATIC(tcp_swcsum);
275 #endif /* defined(INET) */
276 
277 #if defined(INET6)
278 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
279     NULL, "tcp6", "hwcsum bad");
280 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
281     NULL, "tcp6", "hwcsum ok");
282 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
283     NULL, "tcp6", "hwcsum data");
284 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
285     NULL, "tcp6", "swcsum");
286 
287 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
288 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
289 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
290 EVCNT_ATTACH_STATIC(tcp6_swcsum);
291 #endif /* defined(INET6) */
292 #endif /* TCP_CSUM_COUNTERS */
293 
294 
295 #ifdef TCP_OUTPUT_COUNTERS
296 #include <sys/device.h>
297 
298 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
299     NULL, "tcp", "output big header");
300 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
301     NULL, "tcp", "output predict hit");
302 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303     NULL, "tcp", "output predict miss");
304 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305     NULL, "tcp", "output copy small");
306 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
307     NULL, "tcp", "output copy big");
308 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
309     NULL, "tcp", "output reference big");
310 
311 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
312 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
313 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
314 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
315 EVCNT_ATTACH_STATIC(tcp_output_copybig);
316 EVCNT_ATTACH_STATIC(tcp_output_refbig);
317 
318 #endif /* TCP_OUTPUT_COUNTERS */
319 
320 #ifdef TCP_REASS_COUNTERS
321 #include <sys/device.h>
322 
323 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
324     NULL, "tcp_reass", "calls");
325 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
326     &tcp_reass_, "tcp_reass", "insert into empty queue");
327 struct evcnt tcp_reass_iteration[8] = {
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
329     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
330     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
331     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
332     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
333     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
334     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
335     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
336 };
337 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
338     &tcp_reass_, "tcp_reass", "prepend to first");
339 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
340     &tcp_reass_, "tcp_reass", "prepend");
341 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
342     &tcp_reass_, "tcp_reass", "insert");
343 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
344     &tcp_reass_, "tcp_reass", "insert at tail");
345 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
346     &tcp_reass_, "tcp_reass", "append");
347 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
348     &tcp_reass_, "tcp_reass", "append to tail fragment");
349 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
350     &tcp_reass_, "tcp_reass", "overlap at end");
351 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
352     &tcp_reass_, "tcp_reass", "overlap at start");
353 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
354     &tcp_reass_, "tcp_reass", "duplicate segment");
355 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
356     &tcp_reass_, "tcp_reass", "duplicate fragment");
357 
358 EVCNT_ATTACH_STATIC(tcp_reass_);
359 EVCNT_ATTACH_STATIC(tcp_reass_empty);
360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
366 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
367 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
368 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
369 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
370 EVCNT_ATTACH_STATIC(tcp_reass_insert);
371 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
372 EVCNT_ATTACH_STATIC(tcp_reass_append);
373 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
374 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
375 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
376 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
377 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
378 
379 #endif /* TCP_REASS_COUNTERS */
380 
381 #ifdef MBUFTRACE
382 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
383 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
384 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
385 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
386 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
387 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
388 #endif
389 
390 /*
391  * Tcp initialization
392  */
393 void
394 tcp_init(void)
395 {
396 	int hlen;
397 
398 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
399 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
400 	    NULL, IPL_SOFTNET);
401 
402 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
403 #ifdef INET6
404 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
405 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
406 #endif
407 	if (max_protohdr < hlen)
408 		max_protohdr = hlen;
409 	if (max_linkhdr + hlen > MHLEN)
410 		panic("tcp_init");
411 
412 #ifdef INET
413 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
414 #endif
415 #ifdef INET6
416 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
417 #endif
418 
419 	tcp_usrreq_init();
420 
421 	/* Initialize timer state. */
422 	tcp_timer_init();
423 
424 	/* Initialize the compressed state engine. */
425 	syn_cache_init();
426 
427 	/* Initialize the congestion control algorithms. */
428 	tcp_congctl_init();
429 
430 	/* Initialize the TCPCB template. */
431 	tcp_tcpcb_template();
432 
433 	/* Initialize reassembly queue */
434 	tcpipqent_init();
435 
436 	/* SACK */
437 	tcp_sack_init();
438 
439 	MOWNER_ATTACH(&tcp_tx_mowner);
440 	MOWNER_ATTACH(&tcp_rx_mowner);
441 	MOWNER_ATTACH(&tcp_reass_mowner);
442 	MOWNER_ATTACH(&tcp_sock_mowner);
443 	MOWNER_ATTACH(&tcp_sock_tx_mowner);
444 	MOWNER_ATTACH(&tcp_sock_rx_mowner);
445 	MOWNER_ATTACH(&tcp_mowner);
446 
447 	tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
448 
449 	vtw_earlyinit();
450 }
451 
452 /*
453  * Create template to be used to send tcp packets on a connection.
454  * Call after host entry created, allocates an mbuf and fills
455  * in a skeletal tcp/ip header, minimizing the amount of work
456  * necessary when the connection is used.
457  */
458 struct mbuf *
459 tcp_template(struct tcpcb *tp)
460 {
461 	struct inpcb *inp = tp->t_inpcb;
462 #ifdef INET6
463 	struct in6pcb *in6p = tp->t_in6pcb;
464 #endif
465 	struct tcphdr *n;
466 	struct mbuf *m;
467 	int hlen;
468 
469 	switch (tp->t_family) {
470 	case AF_INET:
471 		hlen = sizeof(struct ip);
472 		if (inp)
473 			break;
474 #ifdef INET6
475 		if (in6p) {
476 			/* mapped addr case */
477 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
478 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
479 				break;
480 		}
481 #endif
482 		return NULL;	/*EINVAL*/
483 #ifdef INET6
484 	case AF_INET6:
485 		hlen = sizeof(struct ip6_hdr);
486 		if (in6p) {
487 			/* more sainty check? */
488 			break;
489 		}
490 		return NULL;	/*EINVAL*/
491 #endif
492 	default:
493 		hlen = 0;	/*pacify gcc*/
494 		return NULL;	/*EAFNOSUPPORT*/
495 	}
496 #ifdef DIAGNOSTIC
497 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
498 		panic("mclbytes too small for t_template");
499 #endif
500 	m = tp->t_template;
501 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
502 		;
503 	else {
504 		if (m)
505 			m_freem(m);
506 		m = tp->t_template = NULL;
507 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
508 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
509 			MCLGET(m, M_DONTWAIT);
510 			if ((m->m_flags & M_EXT) == 0) {
511 				m_free(m);
512 				m = NULL;
513 			}
514 		}
515 		if (m == NULL)
516 			return NULL;
517 		MCLAIM(m, &tcp_mowner);
518 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
519 	}
520 
521 	memset(mtod(m, void *), 0, m->m_len);
522 
523 	n = (struct tcphdr *)(mtod(m, char *) + hlen);
524 
525 	switch (tp->t_family) {
526 	case AF_INET:
527 	    {
528 		struct ipovly *ipov;
529 		mtod(m, struct ip *)->ip_v = 4;
530 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
531 		ipov = mtod(m, struct ipovly *);
532 		ipov->ih_pr = IPPROTO_TCP;
533 		ipov->ih_len = htons(sizeof(struct tcphdr));
534 		if (inp) {
535 			ipov->ih_src = inp->inp_laddr;
536 			ipov->ih_dst = inp->inp_faddr;
537 		}
538 #ifdef INET6
539 		else if (in6p) {
540 			/* mapped addr case */
541 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
542 				sizeof(ipov->ih_src));
543 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
544 				sizeof(ipov->ih_dst));
545 		}
546 #endif
547 		/*
548 		 * Compute the pseudo-header portion of the checksum
549 		 * now.  We incrementally add in the TCP option and
550 		 * payload lengths later, and then compute the TCP
551 		 * checksum right before the packet is sent off onto
552 		 * the wire.
553 		 */
554 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
555 		    ipov->ih_dst.s_addr,
556 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
557 		break;
558 	    }
559 #ifdef INET6
560 	case AF_INET6:
561 	    {
562 		struct ip6_hdr *ip6;
563 		mtod(m, struct ip *)->ip_v = 6;
564 		ip6 = mtod(m, struct ip6_hdr *);
565 		ip6->ip6_nxt = IPPROTO_TCP;
566 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
567 		ip6->ip6_src = in6p->in6p_laddr;
568 		ip6->ip6_dst = in6p->in6p_faddr;
569 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
570 		if (ip6_auto_flowlabel) {
571 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
572 			ip6->ip6_flow |=
573 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
574 		}
575 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
576 		ip6->ip6_vfc |= IPV6_VERSION;
577 
578 		/*
579 		 * Compute the pseudo-header portion of the checksum
580 		 * now.  We incrementally add in the TCP option and
581 		 * payload lengths later, and then compute the TCP
582 		 * checksum right before the packet is sent off onto
583 		 * the wire.
584 		 */
585 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
586 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
587 		    htonl(IPPROTO_TCP));
588 		break;
589 	    }
590 #endif
591 	}
592 	if (inp) {
593 		n->th_sport = inp->inp_lport;
594 		n->th_dport = inp->inp_fport;
595 	}
596 #ifdef INET6
597 	else if (in6p) {
598 		n->th_sport = in6p->in6p_lport;
599 		n->th_dport = in6p->in6p_fport;
600 	}
601 #endif
602 	n->th_seq = 0;
603 	n->th_ack = 0;
604 	n->th_x2 = 0;
605 	n->th_off = 5;
606 	n->th_flags = 0;
607 	n->th_win = 0;
608 	n->th_urp = 0;
609 	return (m);
610 }
611 
612 /*
613  * Send a single message to the TCP at address specified by
614  * the given TCP/IP header.  If m == 0, then we make a copy
615  * of the tcpiphdr at ti and send directly to the addressed host.
616  * This is used to force keep alive messages out using the TCP
617  * template for a connection tp->t_template.  If flags are given
618  * then we send a message back to the TCP which originated the
619  * segment ti, and discard the mbuf containing it and any other
620  * attached mbufs.
621  *
622  * In any case the ack and sequence number of the transmitted
623  * segment are as specified by the parameters.
624  */
625 int
626 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m,
627     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
628 {
629 #ifdef INET6
630 	struct rtentry *rt;
631 #endif
632 	struct route *ro;
633 	int error, tlen, win = 0;
634 	int hlen;
635 	struct ip *ip;
636 #ifdef INET6
637 	struct ip6_hdr *ip6;
638 #endif
639 	int family;	/* family on packet, not inpcb/in6pcb! */
640 	struct tcphdr *th;
641 	struct socket *so;
642 
643 	if (tp != NULL && (flags & TH_RST) == 0) {
644 #ifdef DIAGNOSTIC
645 		if (tp->t_inpcb && tp->t_in6pcb)
646 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
647 #endif
648 #ifdef INET
649 		if (tp->t_inpcb)
650 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
651 #endif
652 #ifdef INET6
653 		if (tp->t_in6pcb)
654 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
655 #endif
656 	}
657 
658 	th = NULL;	/* Quell uninitialized warning */
659 	ip = NULL;
660 #ifdef INET6
661 	ip6 = NULL;
662 #endif
663 	if (m == 0) {
664 		if (!template)
665 			return EINVAL;
666 
667 		/* get family information from template */
668 		switch (mtod(template, struct ip *)->ip_v) {
669 		case 4:
670 			family = AF_INET;
671 			hlen = sizeof(struct ip);
672 			break;
673 #ifdef INET6
674 		case 6:
675 			family = AF_INET6;
676 			hlen = sizeof(struct ip6_hdr);
677 			break;
678 #endif
679 		default:
680 			return EAFNOSUPPORT;
681 		}
682 
683 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
684 		if (m) {
685 			MCLAIM(m, &tcp_tx_mowner);
686 			MCLGET(m, M_DONTWAIT);
687 			if ((m->m_flags & M_EXT) == 0) {
688 				m_free(m);
689 				m = NULL;
690 			}
691 		}
692 		if (m == NULL)
693 			return (ENOBUFS);
694 
695 		if (tcp_compat_42)
696 			tlen = 1;
697 		else
698 			tlen = 0;
699 
700 		m->m_data += max_linkhdr;
701 		bcopy(mtod(template, void *), mtod(m, void *),
702 			template->m_len);
703 		switch (family) {
704 		case AF_INET:
705 			ip = mtod(m, struct ip *);
706 			th = (struct tcphdr *)(ip + 1);
707 			break;
708 #ifdef INET6
709 		case AF_INET6:
710 			ip6 = mtod(m, struct ip6_hdr *);
711 			th = (struct tcphdr *)(ip6 + 1);
712 			break;
713 #endif
714 #if 0
715 		default:
716 			/* noone will visit here */
717 			m_freem(m);
718 			return EAFNOSUPPORT;
719 #endif
720 		}
721 		flags = TH_ACK;
722 	} else {
723 
724 		if ((m->m_flags & M_PKTHDR) == 0) {
725 #if 0
726 			printf("non PKTHDR to tcp_respond\n");
727 #endif
728 			m_freem(m);
729 			return EINVAL;
730 		}
731 #ifdef DIAGNOSTIC
732 		if (!th0)
733 			panic("th0 == NULL in tcp_respond");
734 #endif
735 
736 		/* get family information from m */
737 		switch (mtod(m, struct ip *)->ip_v) {
738 		case 4:
739 			family = AF_INET;
740 			hlen = sizeof(struct ip);
741 			ip = mtod(m, struct ip *);
742 			break;
743 #ifdef INET6
744 		case 6:
745 			family = AF_INET6;
746 			hlen = sizeof(struct ip6_hdr);
747 			ip6 = mtod(m, struct ip6_hdr *);
748 			break;
749 #endif
750 		default:
751 			m_freem(m);
752 			return EAFNOSUPPORT;
753 		}
754 		/* clear h/w csum flags inherited from rx packet */
755 		m->m_pkthdr.csum_flags = 0;
756 
757 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
758 			tlen = sizeof(*th0);
759 		else
760 			tlen = th0->th_off << 2;
761 
762 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
763 		    mtod(m, char *) + hlen == (char *)th0) {
764 			m->m_len = hlen + tlen;
765 			m_freem(m->m_next);
766 			m->m_next = NULL;
767 		} else {
768 			struct mbuf *n;
769 
770 #ifdef DIAGNOSTIC
771 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
772 				m_freem(m);
773 				return EMSGSIZE;
774 			}
775 #endif
776 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
777 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
778 				MCLGET(n, M_DONTWAIT);
779 				if ((n->m_flags & M_EXT) == 0) {
780 					m_freem(n);
781 					n = NULL;
782 				}
783 			}
784 			if (!n) {
785 				m_freem(m);
786 				return ENOBUFS;
787 			}
788 
789 			MCLAIM(n, &tcp_tx_mowner);
790 			n->m_data += max_linkhdr;
791 			n->m_len = hlen + tlen;
792 			m_copyback(n, 0, hlen, mtod(m, void *));
793 			m_copyback(n, hlen, tlen, (void *)th0);
794 
795 			m_freem(m);
796 			m = n;
797 			n = NULL;
798 		}
799 
800 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
801 		switch (family) {
802 		case AF_INET:
803 			ip = mtod(m, struct ip *);
804 			th = (struct tcphdr *)(ip + 1);
805 			ip->ip_p = IPPROTO_TCP;
806 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
807 			ip->ip_p = IPPROTO_TCP;
808 			break;
809 #ifdef INET6
810 		case AF_INET6:
811 			ip6 = mtod(m, struct ip6_hdr *);
812 			th = (struct tcphdr *)(ip6 + 1);
813 			ip6->ip6_nxt = IPPROTO_TCP;
814 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
815 			ip6->ip6_nxt = IPPROTO_TCP;
816 			break;
817 #endif
818 #if 0
819 		default:
820 			/* noone will visit here */
821 			m_freem(m);
822 			return EAFNOSUPPORT;
823 #endif
824 		}
825 		xchg(th->th_dport, th->th_sport, u_int16_t);
826 #undef xchg
827 		tlen = 0;	/*be friendly with the following code*/
828 	}
829 	th->th_seq = htonl(seq);
830 	th->th_ack = htonl(ack);
831 	th->th_x2 = 0;
832 	if ((flags & TH_SYN) == 0) {
833 		if (tp)
834 			win >>= tp->rcv_scale;
835 		if (win > TCP_MAXWIN)
836 			win = TCP_MAXWIN;
837 		th->th_win = htons((u_int16_t)win);
838 		th->th_off = sizeof (struct tcphdr) >> 2;
839 		tlen += sizeof(*th);
840 	} else
841 		tlen += th->th_off << 2;
842 	m->m_len = hlen + tlen;
843 	m->m_pkthdr.len = hlen + tlen;
844 	m->m_pkthdr.rcvif = NULL;
845 	th->th_flags = flags;
846 	th->th_urp = 0;
847 
848 	switch (family) {
849 #ifdef INET
850 	case AF_INET:
851 	    {
852 		struct ipovly *ipov = (struct ipovly *)ip;
853 		memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
854 		ipov->ih_len = htons((u_int16_t)tlen);
855 
856 		th->th_sum = 0;
857 		th->th_sum = in_cksum(m, hlen + tlen);
858 		ip->ip_len = htons(hlen + tlen);
859 		ip->ip_ttl = ip_defttl;
860 		break;
861 	    }
862 #endif
863 #ifdef INET6
864 	case AF_INET6:
865 	    {
866 		th->th_sum = 0;
867 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
868 				tlen);
869 		ip6->ip6_plen = htons(tlen);
870 		if (tp && tp->t_in6pcb) {
871 			struct ifnet *oifp;
872 			ro = &tp->t_in6pcb->in6p_route;
873 			oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
874 			                                           : NULL;
875 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
876 		} else
877 			ip6->ip6_hlim = ip6_defhlim;
878 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
879 		if (ip6_auto_flowlabel) {
880 			ip6->ip6_flow |=
881 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
882 		}
883 		break;
884 	    }
885 #endif
886 	}
887 
888 	if (tp && tp->t_inpcb)
889 		so = tp->t_inpcb->inp_socket;
890 #ifdef INET6
891 	else if (tp && tp->t_in6pcb)
892 		so = tp->t_in6pcb->in6p_socket;
893 #endif
894 	else
895 		so = NULL;
896 
897 	if (tp != NULL && tp->t_inpcb != NULL) {
898 		ro = &tp->t_inpcb->inp_route;
899 #ifdef DIAGNOSTIC
900 		if (family != AF_INET)
901 			panic("tcp_respond: address family mismatch");
902 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
903 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
904 			    ntohl(ip->ip_dst.s_addr),
905 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
906 		}
907 #endif
908 	}
909 #ifdef INET6
910 	else if (tp != NULL && tp->t_in6pcb != NULL) {
911 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
912 #ifdef DIAGNOSTIC
913 		if (family == AF_INET) {
914 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
915 				panic("tcp_respond: not mapped addr");
916 			if (memcmp(&ip->ip_dst,
917 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
918 			    sizeof(ip->ip_dst)) != 0) {
919 				panic("tcp_respond: ip_dst != in6p_faddr");
920 			}
921 		} else if (family == AF_INET6) {
922 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
923 			    &tp->t_in6pcb->in6p_faddr))
924 				panic("tcp_respond: ip6_dst != in6p_faddr");
925 		} else
926 			panic("tcp_respond: address family mismatch");
927 #endif
928 	}
929 #endif
930 	else
931 		ro = NULL;
932 
933 	switch (family) {
934 #ifdef INET
935 	case AF_INET:
936 		error = ip_output(m, NULL, ro,
937 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so);
938 		break;
939 #endif
940 #ifdef INET6
941 	case AF_INET6:
942 		error = ip6_output(m, NULL, ro, 0, NULL, so, NULL);
943 		break;
944 #endif
945 	default:
946 		error = EAFNOSUPPORT;
947 		break;
948 	}
949 
950 	return (error);
951 }
952 
953 /*
954  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
955  * a bunch of members individually, we maintain this template for the
956  * static and mostly-static components of the TCPCB, and copy it into
957  * the new TCPCB instead.
958  */
959 static struct tcpcb tcpcb_template = {
960 	.t_srtt = TCPTV_SRTTBASE,
961 	.t_rttmin = TCPTV_MIN,
962 
963 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
964 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
965 	.snd_numholes = 0,
966 	.snd_cubic_wmax = 0,
967 	.snd_cubic_wmax_last = 0,
968 	.snd_cubic_ctime = 0,
969 
970 	.t_partialacks = -1,
971 	.t_bytes_acked = 0,
972 };
973 
974 /*
975  * Updates the TCPCB template whenever a parameter that would affect
976  * the template is changed.
977  */
978 void
979 tcp_tcpcb_template(void)
980 {
981 	struct tcpcb *tp = &tcpcb_template;
982 	int flags;
983 
984 	tp->t_peermss = tcp_mssdflt;
985 	tp->t_ourmss = tcp_mssdflt;
986 	tp->t_segsz = tcp_mssdflt;
987 
988 	flags = 0;
989 	if (tcp_do_rfc1323 && tcp_do_win_scale)
990 		flags |= TF_REQ_SCALE;
991 	if (tcp_do_rfc1323 && tcp_do_timestamps)
992 		flags |= TF_REQ_TSTMP;
993 	tp->t_flags = flags;
994 
995 	/*
996 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
997 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
998 	 * reasonable initial retransmit time.
999 	 */
1000 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
1001 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1002 	    TCPTV_MIN, TCPTV_REXMTMAX);
1003 
1004 	/* Keep Alive */
1005 	tp->t_keepinit = tcp_keepinit;
1006 	tp->t_keepidle = tcp_keepidle;
1007 	tp->t_keepintvl = tcp_keepintvl;
1008 	tp->t_keepcnt = tcp_keepcnt;
1009 	tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1010 
1011 	/* MSL */
1012 	tp->t_msl = TCPTV_MSL;
1013 }
1014 
1015 /*
1016  * Create a new TCP control block, making an
1017  * empty reassembly queue and hooking it to the argument
1018  * protocol control block.
1019  */
1020 /* family selects inpcb, or in6pcb */
1021 struct tcpcb *
1022 tcp_newtcpcb(int family, void *aux)
1023 {
1024 #ifdef INET6
1025 	struct rtentry *rt;
1026 #endif
1027 	struct tcpcb *tp;
1028 	int i;
1029 
1030 	/* XXX Consider using a pool_cache for speed. */
1031 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);	/* splsoftnet via tcp_usrreq */
1032 	if (tp == NULL)
1033 		return (NULL);
1034 	memcpy(tp, &tcpcb_template, sizeof(*tp));
1035 	TAILQ_INIT(&tp->segq);
1036 	TAILQ_INIT(&tp->timeq);
1037 	tp->t_family = family;		/* may be overridden later on */
1038 	TAILQ_INIT(&tp->snd_holes);
1039 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
1040 
1041 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
1042 	for (i = 0; i < TCPT_NTIMERS; i++) {
1043 		callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1044 		TCP_TIMER_INIT(tp, i);
1045 	}
1046 	callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1047 
1048 	switch (family) {
1049 	case AF_INET:
1050 	    {
1051 		struct inpcb *inp = (struct inpcb *)aux;
1052 
1053 		inp->inp_ip.ip_ttl = ip_defttl;
1054 		inp->inp_ppcb = (void *)tp;
1055 
1056 		tp->t_inpcb = inp;
1057 		tp->t_mtudisc = ip_mtudisc;
1058 		break;
1059 	    }
1060 #ifdef INET6
1061 	case AF_INET6:
1062 	    {
1063 		struct in6pcb *in6p = (struct in6pcb *)aux;
1064 
1065 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
1066 			(rt = rtcache_validate(&in6p->in6p_route)) != NULL
1067 			    ? rt->rt_ifp
1068 			    : NULL);
1069 		in6p->in6p_ppcb = (void *)tp;
1070 
1071 		tp->t_in6pcb = in6p;
1072 		/* for IPv6, always try to run path MTU discovery */
1073 		tp->t_mtudisc = 1;
1074 		break;
1075 	    }
1076 #endif /* INET6 */
1077 	default:
1078 		for (i = 0; i < TCPT_NTIMERS; i++)
1079 			callout_destroy(&tp->t_timer[i]);
1080 		callout_destroy(&tp->t_delack_ch);
1081 		pool_put(&tcpcb_pool, tp);	/* splsoftnet via tcp_usrreq */
1082 		return (NULL);
1083 	}
1084 
1085 	/*
1086 	 * Initialize our timebase.  When we send timestamps, we take
1087 	 * the delta from tcp_now -- this means each connection always
1088 	 * gets a timebase of 1, which makes it, among other things,
1089 	 * more difficult to determine how long a system has been up,
1090 	 * and thus how many TCP sequence increments have occurred.
1091 	 *
1092 	 * We start with 1, because 0 doesn't work with linux, which
1093 	 * considers timestamp 0 in a SYN packet as a bug and disables
1094 	 * timestamps.
1095 	 */
1096 	tp->ts_timebase = tcp_now - 1;
1097 
1098 	tcp_congctl_select(tp, tcp_congctl_global_name);
1099 
1100 	return (tp);
1101 }
1102 
1103 /*
1104  * Drop a TCP connection, reporting
1105  * the specified error.  If connection is synchronized,
1106  * then send a RST to peer.
1107  */
1108 struct tcpcb *
1109 tcp_drop(struct tcpcb *tp, int errno)
1110 {
1111 	struct socket *so = NULL;
1112 
1113 #ifdef DIAGNOSTIC
1114 	if (tp->t_inpcb && tp->t_in6pcb)
1115 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1116 #endif
1117 #ifdef INET
1118 	if (tp->t_inpcb)
1119 		so = tp->t_inpcb->inp_socket;
1120 #endif
1121 #ifdef INET6
1122 	if (tp->t_in6pcb)
1123 		so = tp->t_in6pcb->in6p_socket;
1124 #endif
1125 	if (!so)
1126 		return NULL;
1127 
1128 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1129 		tp->t_state = TCPS_CLOSED;
1130 		(void) tcp_output(tp);
1131 		TCP_STATINC(TCP_STAT_DROPS);
1132 	} else
1133 		TCP_STATINC(TCP_STAT_CONNDROPS);
1134 	if (errno == ETIMEDOUT && tp->t_softerror)
1135 		errno = tp->t_softerror;
1136 	so->so_error = errno;
1137 	return (tcp_close(tp));
1138 }
1139 
1140 /*
1141  * Close a TCP control block:
1142  *	discard all space held by the tcp
1143  *	discard internet protocol block
1144  *	wake up any sleepers
1145  */
1146 struct tcpcb *
1147 tcp_close(struct tcpcb *tp)
1148 {
1149 	struct inpcb *inp;
1150 #ifdef INET6
1151 	struct in6pcb *in6p;
1152 #endif
1153 	struct socket *so;
1154 #ifdef RTV_RTT
1155 	struct rtentry *rt;
1156 #endif
1157 	struct route *ro;
1158 	int j;
1159 
1160 	inp = tp->t_inpcb;
1161 #ifdef INET6
1162 	in6p = tp->t_in6pcb;
1163 #endif
1164 	so = NULL;
1165 	ro = NULL;
1166 	if (inp) {
1167 		so = inp->inp_socket;
1168 		ro = &inp->inp_route;
1169 	}
1170 #ifdef INET6
1171 	else if (in6p) {
1172 		so = in6p->in6p_socket;
1173 		ro = (struct route *)&in6p->in6p_route;
1174 	}
1175 #endif
1176 
1177 #ifdef RTV_RTT
1178 	/*
1179 	 * If we sent enough data to get some meaningful characteristics,
1180 	 * save them in the routing entry.  'Enough' is arbitrarily
1181 	 * defined as the sendpipesize (default 4K) * 16.  This would
1182 	 * give us 16 rtt samples assuming we only get one sample per
1183 	 * window (the usual case on a long haul net).  16 samples is
1184 	 * enough for the srtt filter to converge to within 5% of the correct
1185 	 * value; fewer samples and we could save a very bogus rtt.
1186 	 *
1187 	 * Don't update the default route's characteristics and don't
1188 	 * update anything that the user "locked".
1189 	 */
1190 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1191 	    ro && (rt = rtcache_validate(ro)) != NULL &&
1192 	    !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1193 		u_long i = 0;
1194 
1195 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1196 			i = tp->t_srtt *
1197 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1198 			if (rt->rt_rmx.rmx_rtt && i)
1199 				/*
1200 				 * filter this update to half the old & half
1201 				 * the new values, converting scale.
1202 				 * See route.h and tcp_var.h for a
1203 				 * description of the scaling constants.
1204 				 */
1205 				rt->rt_rmx.rmx_rtt =
1206 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1207 			else
1208 				rt->rt_rmx.rmx_rtt = i;
1209 		}
1210 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1211 			i = tp->t_rttvar *
1212 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1213 			if (rt->rt_rmx.rmx_rttvar && i)
1214 				rt->rt_rmx.rmx_rttvar =
1215 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1216 			else
1217 				rt->rt_rmx.rmx_rttvar = i;
1218 		}
1219 		/*
1220 		 * update the pipelimit (ssthresh) if it has been updated
1221 		 * already or if a pipesize was specified & the threshhold
1222 		 * got below half the pipesize.  I.e., wait for bad news
1223 		 * before we start updating, then update on both good
1224 		 * and bad news.
1225 		 */
1226 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1227 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1228 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1229 			/*
1230 			 * convert the limit from user data bytes to
1231 			 * packets then to packet data bytes.
1232 			 */
1233 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1234 			if (i < 2)
1235 				i = 2;
1236 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1237 			if (rt->rt_rmx.rmx_ssthresh)
1238 				rt->rt_rmx.rmx_ssthresh =
1239 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1240 			else
1241 				rt->rt_rmx.rmx_ssthresh = i;
1242 		}
1243 	}
1244 #endif /* RTV_RTT */
1245 	/* free the reassembly queue, if any */
1246 	TCP_REASS_LOCK(tp);
1247 	(void) tcp_freeq(tp);
1248 	TCP_REASS_UNLOCK(tp);
1249 
1250 	/* free the SACK holes list. */
1251 	tcp_free_sackholes(tp);
1252 	tcp_congctl_release(tp);
1253 	syn_cache_cleanup(tp);
1254 
1255 	if (tp->t_template) {
1256 		m_free(tp->t_template);
1257 		tp->t_template = NULL;
1258 	}
1259 
1260 	/*
1261 	 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1262 	 * the timers can also drop the lock.  We need to prevent access
1263 	 * to the tcpcb as it's half torn down.  Flag the pcb as dead
1264 	 * (prevents access by timers) and only then detach it.
1265 	 */
1266 	tp->t_flags |= TF_DEAD;
1267 	if (inp) {
1268 		inp->inp_ppcb = 0;
1269 		soisdisconnected(so);
1270 		in_pcbdetach(inp);
1271 	}
1272 #ifdef INET6
1273 	else if (in6p) {
1274 		in6p->in6p_ppcb = 0;
1275 		soisdisconnected(so);
1276 		in6_pcbdetach(in6p);
1277 	}
1278 #endif
1279 	/*
1280 	 * pcb is no longer visble elsewhere, so we can safely release
1281 	 * the lock in callout_halt() if needed.
1282 	 */
1283 	TCP_STATINC(TCP_STAT_CLOSED);
1284 	for (j = 0; j < TCPT_NTIMERS; j++) {
1285 		callout_halt(&tp->t_timer[j], softnet_lock);
1286 		callout_destroy(&tp->t_timer[j]);
1287 	}
1288 	callout_halt(&tp->t_delack_ch, softnet_lock);
1289 	callout_destroy(&tp->t_delack_ch);
1290 	pool_put(&tcpcb_pool, tp);
1291 
1292 	return NULL;
1293 }
1294 
1295 int
1296 tcp_freeq(struct tcpcb *tp)
1297 {
1298 	struct ipqent *qe;
1299 	int rv = 0;
1300 #ifdef TCPREASS_DEBUG
1301 	int i = 0;
1302 #endif
1303 
1304 	TCP_REASS_LOCK_CHECK(tp);
1305 
1306 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1307 #ifdef TCPREASS_DEBUG
1308 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1309 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1310 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1311 #endif
1312 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1313 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1314 		m_freem(qe->ipqe_m);
1315 		tcpipqent_free(qe);
1316 		rv = 1;
1317 	}
1318 	tp->t_segqlen = 0;
1319 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1320 	return (rv);
1321 }
1322 
1323 void
1324 tcp_fasttimo(void)
1325 {
1326 	if (tcp_drainwanted) {
1327 		tcp_drain();
1328 		tcp_drainwanted = 0;
1329 	}
1330 }
1331 
1332 void
1333 tcp_drainstub(void)
1334 {
1335 	tcp_drainwanted = 1;
1336 }
1337 
1338 /*
1339  * Protocol drain routine.  Called when memory is in short supply.
1340  * Called from pr_fasttimo thus a callout context.
1341  */
1342 void
1343 tcp_drain(void)
1344 {
1345 	struct inpcb_hdr *inph;
1346 	struct tcpcb *tp;
1347 
1348 	mutex_enter(softnet_lock);
1349 	KERNEL_LOCK(1, NULL);
1350 
1351 	/*
1352 	 * Free the sequence queue of all TCP connections.
1353 	 */
1354 	TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1355 		switch (inph->inph_af) {
1356 		case AF_INET:
1357 			tp = intotcpcb((struct inpcb *)inph);
1358 			break;
1359 #ifdef INET6
1360 		case AF_INET6:
1361 			tp = in6totcpcb((struct in6pcb *)inph);
1362 			break;
1363 #endif
1364 		default:
1365 			tp = NULL;
1366 			break;
1367 		}
1368 		if (tp != NULL) {
1369 			/*
1370 			 * We may be called from a device's interrupt
1371 			 * context.  If the tcpcb is already busy,
1372 			 * just bail out now.
1373 			 */
1374 			if (tcp_reass_lock_try(tp) == 0)
1375 				continue;
1376 			if (tcp_freeq(tp))
1377 				TCP_STATINC(TCP_STAT_CONNSDRAINED);
1378 			TCP_REASS_UNLOCK(tp);
1379 		}
1380 	}
1381 
1382 	KERNEL_UNLOCK_ONE(NULL);
1383 	mutex_exit(softnet_lock);
1384 }
1385 
1386 /*
1387  * Notify a tcp user of an asynchronous error;
1388  * store error as soft error, but wake up user
1389  * (for now, won't do anything until can select for soft error).
1390  */
1391 void
1392 tcp_notify(struct inpcb *inp, int error)
1393 {
1394 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1395 	struct socket *so = inp->inp_socket;
1396 
1397 	/*
1398 	 * Ignore some errors if we are hooked up.
1399 	 * If connection hasn't completed, has retransmitted several times,
1400 	 * and receives a second error, give up now.  This is better
1401 	 * than waiting a long time to establish a connection that
1402 	 * can never complete.
1403 	 */
1404 	if (tp->t_state == TCPS_ESTABLISHED &&
1405 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1406 	      error == EHOSTDOWN)) {
1407 		return;
1408 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1409 	    tp->t_rxtshift > 3 && tp->t_softerror)
1410 		so->so_error = error;
1411 	else
1412 		tp->t_softerror = error;
1413 	cv_broadcast(&so->so_cv);
1414 	sorwakeup(so);
1415 	sowwakeup(so);
1416 }
1417 
1418 #ifdef INET6
1419 void
1420 tcp6_notify(struct in6pcb *in6p, int error)
1421 {
1422 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1423 	struct socket *so = in6p->in6p_socket;
1424 
1425 	/*
1426 	 * Ignore some errors if we are hooked up.
1427 	 * If connection hasn't completed, has retransmitted several times,
1428 	 * and receives a second error, give up now.  This is better
1429 	 * than waiting a long time to establish a connection that
1430 	 * can never complete.
1431 	 */
1432 	if (tp->t_state == TCPS_ESTABLISHED &&
1433 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1434 	      error == EHOSTDOWN)) {
1435 		return;
1436 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1437 	    tp->t_rxtshift > 3 && tp->t_softerror)
1438 		so->so_error = error;
1439 	else
1440 		tp->t_softerror = error;
1441 	cv_broadcast(&so->so_cv);
1442 	sorwakeup(so);
1443 	sowwakeup(so);
1444 }
1445 #endif
1446 
1447 #ifdef INET6
1448 void *
1449 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1450 {
1451 	struct tcphdr th;
1452 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1453 	int nmatch;
1454 	struct ip6_hdr *ip6;
1455 	const struct sockaddr_in6 *sa6_src = NULL;
1456 	const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1457 	struct mbuf *m;
1458 	int off;
1459 
1460 	if (sa->sa_family != AF_INET6 ||
1461 	    sa->sa_len != sizeof(struct sockaddr_in6))
1462 		return NULL;
1463 	if ((unsigned)cmd >= PRC_NCMDS)
1464 		return NULL;
1465 	else if (cmd == PRC_QUENCH) {
1466 		/*
1467 		 * Don't honor ICMP Source Quench messages meant for
1468 		 * TCP connections.
1469 		 */
1470 		return NULL;
1471 	} else if (PRC_IS_REDIRECT(cmd))
1472 		notify = in6_rtchange, d = NULL;
1473 	else if (cmd == PRC_MSGSIZE)
1474 		; /* special code is present, see below */
1475 	else if (cmd == PRC_HOSTDEAD)
1476 		d = NULL;
1477 	else if (inet6ctlerrmap[cmd] == 0)
1478 		return NULL;
1479 
1480 	/* if the parameter is from icmp6, decode it. */
1481 	if (d != NULL) {
1482 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1483 		m = ip6cp->ip6c_m;
1484 		ip6 = ip6cp->ip6c_ip6;
1485 		off = ip6cp->ip6c_off;
1486 		sa6_src = ip6cp->ip6c_src;
1487 	} else {
1488 		m = NULL;
1489 		ip6 = NULL;
1490 		sa6_src = &sa6_any;
1491 		off = 0;
1492 	}
1493 
1494 	if (ip6) {
1495 		/*
1496 		 * XXX: We assume that when ip6 is non NULL,
1497 		 * M and OFF are valid.
1498 		 */
1499 
1500 		/* check if we can safely examine src and dst ports */
1501 		if (m->m_pkthdr.len < off + sizeof(th)) {
1502 			if (cmd == PRC_MSGSIZE)
1503 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1504 			return NULL;
1505 		}
1506 
1507 		memset(&th, 0, sizeof(th));
1508 		m_copydata(m, off, sizeof(th), (void *)&th);
1509 
1510 		if (cmd == PRC_MSGSIZE) {
1511 			int valid = 0;
1512 
1513 			/*
1514 			 * Check to see if we have a valid TCP connection
1515 			 * corresponding to the address in the ICMPv6 message
1516 			 * payload.
1517 			 */
1518 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1519 			    th.th_dport,
1520 			    (const struct in6_addr *)&sa6_src->sin6_addr,
1521 						  th.th_sport, 0, 0))
1522 				valid++;
1523 
1524 			/*
1525 			 * Depending on the value of "valid" and routing table
1526 			 * size (mtudisc_{hi,lo}wat), we will:
1527 			 * - recalcurate the new MTU and create the
1528 			 *   corresponding routing entry, or
1529 			 * - ignore the MTU change notification.
1530 			 */
1531 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1532 
1533 			/*
1534 			 * no need to call in6_pcbnotify, it should have been
1535 			 * called via callback if necessary
1536 			 */
1537 			return NULL;
1538 		}
1539 
1540 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1541 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1542 		if (nmatch == 0 && syn_cache_count &&
1543 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1544 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1545 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1546 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1547 					  sa, &th);
1548 	} else {
1549 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1550 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1551 	}
1552 
1553 	return NULL;
1554 }
1555 #endif
1556 
1557 #ifdef INET
1558 /* assumes that ip header and tcp header are contiguous on mbuf */
1559 void *
1560 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1561 {
1562 	struct ip *ip = v;
1563 	struct tcphdr *th;
1564 	struct icmp *icp;
1565 	extern const int inetctlerrmap[];
1566 	void (*notify)(struct inpcb *, int) = tcp_notify;
1567 	int errno;
1568 	int nmatch;
1569 	struct tcpcb *tp;
1570 	u_int mtu;
1571 	tcp_seq seq;
1572 	struct inpcb *inp;
1573 #ifdef INET6
1574 	struct in6pcb *in6p;
1575 	struct in6_addr src6, dst6;
1576 #endif
1577 
1578 	if (sa->sa_family != AF_INET ||
1579 	    sa->sa_len != sizeof(struct sockaddr_in))
1580 		return NULL;
1581 	if ((unsigned)cmd >= PRC_NCMDS)
1582 		return NULL;
1583 	errno = inetctlerrmap[cmd];
1584 	if (cmd == PRC_QUENCH)
1585 		/*
1586 		 * Don't honor ICMP Source Quench messages meant for
1587 		 * TCP connections.
1588 		 */
1589 		return NULL;
1590 	else if (PRC_IS_REDIRECT(cmd))
1591 		notify = in_rtchange, ip = 0;
1592 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1593 		/*
1594 		 * Check to see if we have a valid TCP connection
1595 		 * corresponding to the address in the ICMP message
1596 		 * payload.
1597 		 *
1598 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1599 		 */
1600 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1601 #ifdef INET6
1602 		memset(&src6, 0, sizeof(src6));
1603 		memset(&dst6, 0, sizeof(dst6));
1604 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1605 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1606 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1607 #endif
1608 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1609 						th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1610 #ifdef INET6
1611 			in6p = NULL;
1612 #else
1613 			;
1614 #endif
1615 #ifdef INET6
1616 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1617 						       th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1618 			;
1619 #endif
1620 		else
1621 			return NULL;
1622 
1623 		/*
1624 		 * Now that we've validated that we are actually communicating
1625 		 * with the host indicated in the ICMP message, locate the
1626 		 * ICMP header, recalculate the new MTU, and create the
1627 		 * corresponding routing entry.
1628 		 */
1629 		icp = (struct icmp *)((char *)ip -
1630 		    offsetof(struct icmp, icmp_ip));
1631 		if (inp) {
1632 			if ((tp = intotcpcb(inp)) == NULL)
1633 				return NULL;
1634 		}
1635 #ifdef INET6
1636 		else if (in6p) {
1637 			if ((tp = in6totcpcb(in6p)) == NULL)
1638 				return NULL;
1639 		}
1640 #endif
1641 		else
1642 			return NULL;
1643 		seq = ntohl(th->th_seq);
1644 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1645 			return NULL;
1646 		/*
1647 		 * If the ICMP message advertises a Next-Hop MTU
1648 		 * equal or larger than the maximum packet size we have
1649 		 * ever sent, drop the message.
1650 		 */
1651 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1652 		if (mtu >= tp->t_pmtud_mtu_sent)
1653 			return NULL;
1654 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1655 			/*
1656 			 * Calculate new MTU, and create corresponding
1657 			 * route (traditional PMTUD).
1658 			 */
1659 			tp->t_flags &= ~TF_PMTUD_PEND;
1660 			icmp_mtudisc(icp, ip->ip_dst);
1661 		} else {
1662 			/*
1663 			 * Record the information got in the ICMP
1664 			 * message; act on it later.
1665 			 * If we had already recorded an ICMP message,
1666 			 * replace the old one only if the new message
1667 			 * refers to an older TCP segment
1668 			 */
1669 			if (tp->t_flags & TF_PMTUD_PEND) {
1670 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1671 					return NULL;
1672 			} else
1673 				tp->t_flags |= TF_PMTUD_PEND;
1674 			tp->t_pmtud_th_seq = seq;
1675 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1676 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1677 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1678 		}
1679 		return NULL;
1680 	} else if (cmd == PRC_HOSTDEAD)
1681 		ip = 0;
1682 	else if (errno == 0)
1683 		return NULL;
1684 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1685 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1686 		nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1687 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1688 		if (nmatch == 0 && syn_cache_count &&
1689 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1690 		    inetctlerrmap[cmd] == ENETUNREACH ||
1691 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1692 			struct sockaddr_in sin;
1693 			memset(&sin, 0, sizeof(sin));
1694 			sin.sin_len = sizeof(sin);
1695 			sin.sin_family = AF_INET;
1696 			sin.sin_port = th->th_sport;
1697 			sin.sin_addr = ip->ip_src;
1698 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1699 		}
1700 
1701 		/* XXX mapped address case */
1702 	} else
1703 		in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1704 		    notify);
1705 	return NULL;
1706 }
1707 
1708 /*
1709  * When a source quench is received, we are being notified of congestion.
1710  * Close the congestion window down to the Loss Window (one segment).
1711  * We will gradually open it again as we proceed.
1712  */
1713 void
1714 tcp_quench(struct inpcb *inp, int errno)
1715 {
1716 	struct tcpcb *tp = intotcpcb(inp);
1717 
1718 	if (tp) {
1719 		tp->snd_cwnd = tp->t_segsz;
1720 		tp->t_bytes_acked = 0;
1721 	}
1722 }
1723 #endif
1724 
1725 #ifdef INET6
1726 void
1727 tcp6_quench(struct in6pcb *in6p, int errno)
1728 {
1729 	struct tcpcb *tp = in6totcpcb(in6p);
1730 
1731 	if (tp) {
1732 		tp->snd_cwnd = tp->t_segsz;
1733 		tp->t_bytes_acked = 0;
1734 	}
1735 }
1736 #endif
1737 
1738 #ifdef INET
1739 /*
1740  * Path MTU Discovery handlers.
1741  */
1742 void
1743 tcp_mtudisc_callback(struct in_addr faddr)
1744 {
1745 #ifdef INET6
1746 	struct in6_addr in6;
1747 #endif
1748 
1749 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1750 #ifdef INET6
1751 	memset(&in6, 0, sizeof(in6));
1752 	in6.s6_addr16[5] = 0xffff;
1753 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1754 	tcp6_mtudisc_callback(&in6);
1755 #endif
1756 }
1757 
1758 /*
1759  * On receipt of path MTU corrections, flush old route and replace it
1760  * with the new one.  Retransmit all unacknowledged packets, to ensure
1761  * that all packets will be received.
1762  */
1763 void
1764 tcp_mtudisc(struct inpcb *inp, int errno)
1765 {
1766 	struct tcpcb *tp = intotcpcb(inp);
1767 	struct rtentry *rt = in_pcbrtentry(inp);
1768 
1769 	if (tp != 0) {
1770 		if (rt != 0) {
1771 			/*
1772 			 * If this was not a host route, remove and realloc.
1773 			 */
1774 			if ((rt->rt_flags & RTF_HOST) == 0) {
1775 				in_rtchange(inp, errno);
1776 				if ((rt = in_pcbrtentry(inp)) == 0)
1777 					return;
1778 			}
1779 
1780 			/*
1781 			 * Slow start out of the error condition.  We
1782 			 * use the MTU because we know it's smaller
1783 			 * than the previously transmitted segment.
1784 			 *
1785 			 * Note: This is more conservative than the
1786 			 * suggestion in draft-floyd-incr-init-win-03.
1787 			 */
1788 			if (rt->rt_rmx.rmx_mtu != 0)
1789 				tp->snd_cwnd =
1790 				    TCP_INITIAL_WINDOW(tcp_init_win,
1791 				    rt->rt_rmx.rmx_mtu);
1792 		}
1793 
1794 		/*
1795 		 * Resend unacknowledged packets.
1796 		 */
1797 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1798 		tcp_output(tp);
1799 	}
1800 }
1801 #endif
1802 
1803 #ifdef INET6
1804 /*
1805  * Path MTU Discovery handlers.
1806  */
1807 void
1808 tcp6_mtudisc_callback(struct in6_addr *faddr)
1809 {
1810 	struct sockaddr_in6 sin6;
1811 
1812 	memset(&sin6, 0, sizeof(sin6));
1813 	sin6.sin6_family = AF_INET6;
1814 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1815 	sin6.sin6_addr = *faddr;
1816 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1817 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1818 }
1819 
1820 void
1821 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1822 {
1823 	struct tcpcb *tp = in6totcpcb(in6p);
1824 	struct rtentry *rt = in6_pcbrtentry(in6p);
1825 
1826 	if (tp != 0) {
1827 		if (rt != 0) {
1828 			/*
1829 			 * If this was not a host route, remove and realloc.
1830 			 */
1831 			if ((rt->rt_flags & RTF_HOST) == 0) {
1832 				in6_rtchange(in6p, errno);
1833 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1834 					return;
1835 			}
1836 
1837 			/*
1838 			 * Slow start out of the error condition.  We
1839 			 * use the MTU because we know it's smaller
1840 			 * than the previously transmitted segment.
1841 			 *
1842 			 * Note: This is more conservative than the
1843 			 * suggestion in draft-floyd-incr-init-win-03.
1844 			 */
1845 			if (rt->rt_rmx.rmx_mtu != 0)
1846 				tp->snd_cwnd =
1847 				    TCP_INITIAL_WINDOW(tcp_init_win,
1848 				    rt->rt_rmx.rmx_mtu);
1849 		}
1850 
1851 		/*
1852 		 * Resend unacknowledged packets.
1853 		 */
1854 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1855 		tcp_output(tp);
1856 	}
1857 }
1858 #endif /* INET6 */
1859 
1860 /*
1861  * Compute the MSS to advertise to the peer.  Called only during
1862  * the 3-way handshake.  If we are the server (peer initiated
1863  * connection), we are called with a pointer to the interface
1864  * on which the SYN packet arrived.  If we are the client (we
1865  * initiated connection), we are called with a pointer to the
1866  * interface out which this connection should go.
1867  *
1868  * NOTE: Do not subtract IP option/extension header size nor IPsec
1869  * header size from MSS advertisement.  MSS option must hold the maximum
1870  * segment size we can accept, so it must always be:
1871  *	 max(if mtu) - ip header - tcp header
1872  */
1873 u_long
1874 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1875 {
1876 	extern u_long in_maxmtu;
1877 	u_long mss = 0;
1878 	u_long hdrsiz;
1879 
1880 	/*
1881 	 * In order to avoid defeating path MTU discovery on the peer,
1882 	 * we advertise the max MTU of all attached networks as our MSS,
1883 	 * per RFC 1191, section 3.1.
1884 	 *
1885 	 * We provide the option to advertise just the MTU of
1886 	 * the interface on which we hope this connection will
1887 	 * be receiving.  If we are responding to a SYN, we
1888 	 * will have a pretty good idea about this, but when
1889 	 * initiating a connection there is a bit more doubt.
1890 	 *
1891 	 * We also need to ensure that loopback has a large enough
1892 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1893 	 */
1894 
1895 	if (ifp != NULL)
1896 		switch (af) {
1897 		case AF_INET:
1898 			mss = ifp->if_mtu;
1899 			break;
1900 #ifdef INET6
1901 		case AF_INET6:
1902 			mss = IN6_LINKMTU(ifp);
1903 			break;
1904 #endif
1905 		}
1906 
1907 	if (tcp_mss_ifmtu == 0)
1908 		switch (af) {
1909 		case AF_INET:
1910 			mss = max(in_maxmtu, mss);
1911 			break;
1912 #ifdef INET6
1913 		case AF_INET6:
1914 			mss = max(in6_maxmtu, mss);
1915 			break;
1916 #endif
1917 		}
1918 
1919 	switch (af) {
1920 	case AF_INET:
1921 		hdrsiz = sizeof(struct ip);
1922 		break;
1923 #ifdef INET6
1924 	case AF_INET6:
1925 		hdrsiz = sizeof(struct ip6_hdr);
1926 		break;
1927 #endif
1928 	default:
1929 		hdrsiz = 0;
1930 		break;
1931 	}
1932 	hdrsiz += sizeof(struct tcphdr);
1933 	if (mss > hdrsiz)
1934 		mss -= hdrsiz;
1935 
1936 	mss = max(tcp_mssdflt, mss);
1937 	return (mss);
1938 }
1939 
1940 /*
1941  * Set connection variables based on the peer's advertised MSS.
1942  * We are passed the TCPCB for the actual connection.  If we
1943  * are the server, we are called by the compressed state engine
1944  * when the 3-way handshake is complete.  If we are the client,
1945  * we are called when we receive the SYN,ACK from the server.
1946  *
1947  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1948  * before this routine is called!
1949  */
1950 void
1951 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1952 {
1953 	struct socket *so;
1954 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1955 	struct rtentry *rt;
1956 #endif
1957 	u_long bufsize;
1958 	int mss;
1959 
1960 #ifdef DIAGNOSTIC
1961 	if (tp->t_inpcb && tp->t_in6pcb)
1962 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1963 #endif
1964 	so = NULL;
1965 	rt = NULL;
1966 #ifdef INET
1967 	if (tp->t_inpcb) {
1968 		so = tp->t_inpcb->inp_socket;
1969 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1970 		rt = in_pcbrtentry(tp->t_inpcb);
1971 #endif
1972 	}
1973 #endif
1974 #ifdef INET6
1975 	if (tp->t_in6pcb) {
1976 		so = tp->t_in6pcb->in6p_socket;
1977 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1978 		rt = in6_pcbrtentry(tp->t_in6pcb);
1979 #endif
1980 	}
1981 #endif
1982 
1983 	/*
1984 	 * As per RFC1122, use the default MSS value, unless they
1985 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1986 	 */
1987 	mss = tcp_mssdflt;
1988 	if (offer)
1989 		mss = offer;
1990 	mss = max(mss, 256);		/* sanity */
1991 	tp->t_peermss = mss;
1992 	mss -= tcp_optlen(tp);
1993 #ifdef INET
1994 	if (tp->t_inpcb)
1995 		mss -= ip_optlen(tp->t_inpcb);
1996 #endif
1997 #ifdef INET6
1998 	if (tp->t_in6pcb)
1999 		mss -= ip6_optlen(tp->t_in6pcb);
2000 #endif
2001 
2002 	/*
2003 	 * If there's a pipesize, change the socket buffer to that size.
2004 	 * Make the socket buffer an integral number of MSS units.  If
2005 	 * the MSS is larger than the socket buffer, artificially decrease
2006 	 * the MSS.
2007 	 */
2008 #ifdef RTV_SPIPE
2009 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
2010 		bufsize = rt->rt_rmx.rmx_sendpipe;
2011 	else
2012 #endif
2013 	{
2014 		KASSERT(so != NULL);
2015 		bufsize = so->so_snd.sb_hiwat;
2016 	}
2017 	if (bufsize < mss)
2018 		mss = bufsize;
2019 	else {
2020 		bufsize = roundup(bufsize, mss);
2021 		if (bufsize > sb_max)
2022 			bufsize = sb_max;
2023 		(void) sbreserve(&so->so_snd, bufsize, so);
2024 	}
2025 	tp->t_segsz = mss;
2026 
2027 #ifdef RTV_SSTHRESH
2028 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2029 		/*
2030 		 * There's some sort of gateway or interface buffer
2031 		 * limit on the path.  Use this to set the slow
2032 		 * start threshold, but set the threshold to no less
2033 		 * than 2 * MSS.
2034 		 */
2035 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2036 	}
2037 #endif
2038 }
2039 
2040 /*
2041  * Processing necessary when a TCP connection is established.
2042  */
2043 void
2044 tcp_established(struct tcpcb *tp)
2045 {
2046 	struct socket *so;
2047 #ifdef RTV_RPIPE
2048 	struct rtentry *rt;
2049 #endif
2050 	u_long bufsize;
2051 
2052 #ifdef DIAGNOSTIC
2053 	if (tp->t_inpcb && tp->t_in6pcb)
2054 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
2055 #endif
2056 	so = NULL;
2057 	rt = NULL;
2058 #ifdef INET
2059 	/* This is a while() to reduce the dreadful stairstepping below */
2060 	while (tp->t_inpcb) {
2061 		so = tp->t_inpcb->inp_socket;
2062 #if defined(RTV_RPIPE)
2063 		rt = in_pcbrtentry(tp->t_inpcb);
2064 #endif
2065 		if (__predict_true(tcp_msl_enable)) {
2066 			if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2067 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2068 				break;
2069 			}
2070 
2071 			if (__predict_false(tcp_rttlocal)) {
2072 				/* This may be adjusted by tcp_input */
2073 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2074 				break;
2075 			}
2076 			if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2077 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2078 				break;
2079 			}
2080 		}
2081 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2082 		break;
2083 	}
2084 #endif
2085 #ifdef INET6
2086 	/* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2087 	while (!tp->t_inpcb && tp->t_in6pcb) {
2088 		so = tp->t_in6pcb->in6p_socket;
2089 #if defined(RTV_RPIPE)
2090 		rt = in6_pcbrtentry(tp->t_in6pcb);
2091 #endif
2092 		if (__predict_true(tcp_msl_enable)) {
2093 			extern const struct in6_addr in6addr_loopback;
2094 
2095 			if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2096 					       &in6addr_loopback)) {
2097 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2098 				break;
2099 			}
2100 
2101 			if (__predict_false(tcp_rttlocal)) {
2102 				/* This may be adjusted by tcp_input */
2103 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2104 				break;
2105 			}
2106 			if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2107 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2108 				break;
2109 			}
2110 		}
2111 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2112 		break;
2113 	}
2114 #endif
2115 
2116 	tp->t_state = TCPS_ESTABLISHED;
2117 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2118 
2119 #ifdef RTV_RPIPE
2120 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2121 		bufsize = rt->rt_rmx.rmx_recvpipe;
2122 	else
2123 #endif
2124 	{
2125 		KASSERT(so != NULL);
2126 		bufsize = so->so_rcv.sb_hiwat;
2127 	}
2128 	if (bufsize > tp->t_ourmss) {
2129 		bufsize = roundup(bufsize, tp->t_ourmss);
2130 		if (bufsize > sb_max)
2131 			bufsize = sb_max;
2132 		(void) sbreserve(&so->so_rcv, bufsize, so);
2133 	}
2134 }
2135 
2136 /*
2137  * Check if there's an initial rtt or rttvar.  Convert from the
2138  * route-table units to scaled multiples of the slow timeout timer.
2139  * Called only during the 3-way handshake.
2140  */
2141 void
2142 tcp_rmx_rtt(struct tcpcb *tp)
2143 {
2144 #ifdef RTV_RTT
2145 	struct rtentry *rt = NULL;
2146 	int rtt;
2147 
2148 #ifdef DIAGNOSTIC
2149 	if (tp->t_inpcb && tp->t_in6pcb)
2150 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2151 #endif
2152 #ifdef INET
2153 	if (tp->t_inpcb)
2154 		rt = in_pcbrtentry(tp->t_inpcb);
2155 #endif
2156 #ifdef INET6
2157 	if (tp->t_in6pcb)
2158 		rt = in6_pcbrtentry(tp->t_in6pcb);
2159 #endif
2160 	if (rt == NULL)
2161 		return;
2162 
2163 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2164 		/*
2165 		 * XXX The lock bit for MTU indicates that the value
2166 		 * is also a minimum value; this is subject to time.
2167 		 */
2168 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2169 			TCPT_RANGESET(tp->t_rttmin,
2170 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2171 			    TCPTV_MIN, TCPTV_REXMTMAX);
2172 		tp->t_srtt = rtt /
2173 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2174 		if (rt->rt_rmx.rmx_rttvar) {
2175 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2176 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2177 				(TCP_RTTVAR_SHIFT + 2));
2178 		} else {
2179 			/* Default variation is +- 1 rtt */
2180 			tp->t_rttvar =
2181 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2182 		}
2183 		TCPT_RANGESET(tp->t_rxtcur,
2184 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2185 		    tp->t_rttmin, TCPTV_REXMTMAX);
2186 	}
2187 #endif
2188 }
2189 
2190 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2191 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2192 
2193 /*
2194  * Get a new sequence value given a tcp control block
2195  */
2196 tcp_seq
2197 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2198 {
2199 
2200 #ifdef INET
2201 	if (tp->t_inpcb != NULL) {
2202 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2203 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2204 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2205 		    addin));
2206 	}
2207 #endif
2208 #ifdef INET6
2209 	if (tp->t_in6pcb != NULL) {
2210 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2211 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2212 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2213 		    addin));
2214 	}
2215 #endif
2216 	/* Not possible. */
2217 	panic("tcp_new_iss");
2218 }
2219 
2220 /*
2221  * This routine actually generates a new TCP initial sequence number.
2222  */
2223 tcp_seq
2224 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2225     size_t addrsz, tcp_seq addin)
2226 {
2227 	tcp_seq tcp_iss;
2228 
2229 	static bool tcp_iss_gotten_secret;
2230 
2231 	/*
2232 	 * If we haven't been here before, initialize our cryptographic
2233 	 * hash secret.
2234 	 */
2235 	if (tcp_iss_gotten_secret == false) {
2236 		cprng_strong(kern_cprng,
2237 			     tcp_iss_secret, sizeof(tcp_iss_secret), FASYNC);
2238 		tcp_iss_gotten_secret = true;
2239 	}
2240 
2241 	if (tcp_do_rfc1948) {
2242 		MD5_CTX ctx;
2243 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2244 
2245 		/*
2246 		 * Compute the base value of the ISS.  It is a hash
2247 		 * of (saddr, sport, daddr, dport, secret).
2248 		 */
2249 		MD5Init(&ctx);
2250 
2251 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2252 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2253 
2254 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2255 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2256 
2257 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2258 
2259 		MD5Final(hash, &ctx);
2260 
2261 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2262 
2263 		/*
2264 		 * Now increment our "timer", and add it in to
2265 		 * the computed value.
2266 		 *
2267 		 * XXX Use `addin'?
2268 		 * XXX TCP_ISSINCR too large to use?
2269 		 */
2270 		tcp_iss_seq += TCP_ISSINCR;
2271 #ifdef TCPISS_DEBUG
2272 		printf("ISS hash 0x%08x, ", tcp_iss);
2273 #endif
2274 		tcp_iss += tcp_iss_seq + addin;
2275 #ifdef TCPISS_DEBUG
2276 		printf("new ISS 0x%08x\n", tcp_iss);
2277 #endif
2278 	} else {
2279 		/*
2280 		 * Randomize.
2281 		 */
2282 		tcp_iss = cprng_fast32();
2283 
2284 		/*
2285 		 * If we were asked to add some amount to a known value,
2286 		 * we will take a random value obtained above, mask off
2287 		 * the upper bits, and add in the known value.  We also
2288 		 * add in a constant to ensure that we are at least a
2289 		 * certain distance from the original value.
2290 		 *
2291 		 * This is used when an old connection is in timed wait
2292 		 * and we have a new one coming in, for instance.
2293 		 */
2294 		if (addin != 0) {
2295 #ifdef TCPISS_DEBUG
2296 			printf("Random %08x, ", tcp_iss);
2297 #endif
2298 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2299 			tcp_iss += addin + TCP_ISSINCR;
2300 #ifdef TCPISS_DEBUG
2301 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2302 #endif
2303 		} else {
2304 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2305 			tcp_iss += tcp_iss_seq;
2306 			tcp_iss_seq += TCP_ISSINCR;
2307 #ifdef TCPISS_DEBUG
2308 			printf("ISS %08x\n", tcp_iss);
2309 #endif
2310 		}
2311 	}
2312 
2313 	if (tcp_compat_42) {
2314 		/*
2315 		 * Limit it to the positive range for really old TCP
2316 		 * implementations.
2317 		 * Just AND off the top bit instead of checking if
2318 		 * is set first - saves a branch 50% of the time.
2319 		 */
2320 		tcp_iss &= 0x7fffffff;		/* XXX */
2321 	}
2322 
2323 	return (tcp_iss);
2324 }
2325 
2326 #if defined(IPSEC)
2327 /* compute ESP/AH header size for TCP, including outer IP header. */
2328 size_t
2329 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2330 {
2331 	struct inpcb *inp;
2332 	size_t hdrsiz;
2333 
2334 	/* XXX mapped addr case (tp->t_in6pcb) */
2335 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2336 		return 0;
2337 	switch (tp->t_family) {
2338 	case AF_INET:
2339 		/* XXX: should use currect direction. */
2340 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2341 		break;
2342 	default:
2343 		hdrsiz = 0;
2344 		break;
2345 	}
2346 
2347 	return hdrsiz;
2348 }
2349 
2350 #ifdef INET6
2351 size_t
2352 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2353 {
2354 	struct in6pcb *in6p;
2355 	size_t hdrsiz;
2356 
2357 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2358 		return 0;
2359 	switch (tp->t_family) {
2360 	case AF_INET6:
2361 		/* XXX: should use currect direction. */
2362 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2363 		break;
2364 	case AF_INET:
2365 		/* mapped address case - tricky */
2366 	default:
2367 		hdrsiz = 0;
2368 		break;
2369 	}
2370 
2371 	return hdrsiz;
2372 }
2373 #endif
2374 #endif /*IPSEC*/
2375 
2376 /*
2377  * Determine the length of the TCP options for this connection.
2378  *
2379  * XXX:  What do we do for SACK, when we add that?  Just reserve
2380  *       all of the space?  Otherwise we can't exactly be incrementing
2381  *       cwnd by an amount that varies depending on the amount we last
2382  *       had to SACK!
2383  */
2384 
2385 u_int
2386 tcp_optlen(struct tcpcb *tp)
2387 {
2388 	u_int optlen;
2389 
2390 	optlen = 0;
2391 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2392 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2393 		optlen += TCPOLEN_TSTAMP_APPA;
2394 
2395 #ifdef TCP_SIGNATURE
2396 	if (tp->t_flags & TF_SIGNATURE)
2397 		optlen += TCPOLEN_SIGNATURE + 2;
2398 #endif /* TCP_SIGNATURE */
2399 
2400 	return optlen;
2401 }
2402 
2403 u_int
2404 tcp_hdrsz(struct tcpcb *tp)
2405 {
2406 	u_int hlen;
2407 
2408 	switch (tp->t_family) {
2409 #ifdef INET6
2410 	case AF_INET6:
2411 		hlen = sizeof(struct ip6_hdr);
2412 		break;
2413 #endif
2414 	case AF_INET:
2415 		hlen = sizeof(struct ip);
2416 		break;
2417 	default:
2418 		hlen = 0;
2419 		break;
2420 	}
2421 	hlen += sizeof(struct tcphdr);
2422 
2423 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2424 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2425 		hlen += TCPOLEN_TSTAMP_APPA;
2426 #ifdef TCP_SIGNATURE
2427 	if (tp->t_flags & TF_SIGNATURE)
2428 		hlen += TCPOLEN_SIGLEN;
2429 #endif
2430 	return hlen;
2431 }
2432 
2433 void
2434 tcp_statinc(u_int stat)
2435 {
2436 
2437 	KASSERT(stat < TCP_NSTATS);
2438 	TCP_STATINC(stat);
2439 }
2440 
2441 void
2442 tcp_statadd(u_int stat, uint64_t val)
2443 {
2444 
2445 	KASSERT(stat < TCP_NSTATS);
2446 	TCP_STATADD(stat, val);
2447 }
2448