xref: /netbsd-src/sys/netinet/tcp_subr.c (revision f89f6560d453f5e37386cc7938c072d2f528b9fa)
1 /*	$NetBSD: tcp_subr.c,v 1.259 2015/04/13 15:51:00 riastradh 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 1.259 2015/04/13 15:51:00 riastradh 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 
244 #ifdef INET
245 static void	tcp_mtudisc_callback(struct in_addr);
246 #endif
247 
248 #ifdef INET6
249 void	tcp6_mtudisc(struct in6pcb *, int);
250 #endif
251 
252 static struct pool tcpcb_pool;
253 
254 static int tcp_drainwanted;
255 
256 #ifdef TCP_CSUM_COUNTERS
257 #include <sys/device.h>
258 
259 #if defined(INET)
260 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
261     NULL, "tcp", "hwcsum bad");
262 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
263     NULL, "tcp", "hwcsum ok");
264 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
265     NULL, "tcp", "hwcsum data");
266 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
267     NULL, "tcp", "swcsum");
268 
269 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
270 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
271 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
272 EVCNT_ATTACH_STATIC(tcp_swcsum);
273 #endif /* defined(INET) */
274 
275 #if defined(INET6)
276 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
277     NULL, "tcp6", "hwcsum bad");
278 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
279     NULL, "tcp6", "hwcsum ok");
280 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
281     NULL, "tcp6", "hwcsum data");
282 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
283     NULL, "tcp6", "swcsum");
284 
285 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
286 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
287 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
288 EVCNT_ATTACH_STATIC(tcp6_swcsum);
289 #endif /* defined(INET6) */
290 #endif /* TCP_CSUM_COUNTERS */
291 
292 
293 #ifdef TCP_OUTPUT_COUNTERS
294 #include <sys/device.h>
295 
296 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
297     NULL, "tcp", "output big header");
298 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
299     NULL, "tcp", "output predict hit");
300 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
301     NULL, "tcp", "output predict miss");
302 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303     NULL, "tcp", "output copy small");
304 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305     NULL, "tcp", "output copy big");
306 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
307     NULL, "tcp", "output reference big");
308 
309 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
310 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
311 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
312 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
313 EVCNT_ATTACH_STATIC(tcp_output_copybig);
314 EVCNT_ATTACH_STATIC(tcp_output_refbig);
315 
316 #endif /* TCP_OUTPUT_COUNTERS */
317 
318 #ifdef TCP_REASS_COUNTERS
319 #include <sys/device.h>
320 
321 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
322     NULL, "tcp_reass", "calls");
323 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
324     &tcp_reass_, "tcp_reass", "insert into empty queue");
325 struct evcnt tcp_reass_iteration[8] = {
326     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
327     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
329     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
330     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
331     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
332     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
333     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
334 };
335 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
336     &tcp_reass_, "tcp_reass", "prepend to first");
337 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
338     &tcp_reass_, "tcp_reass", "prepend");
339 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
340     &tcp_reass_, "tcp_reass", "insert");
341 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
342     &tcp_reass_, "tcp_reass", "insert at tail");
343 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
344     &tcp_reass_, "tcp_reass", "append");
345 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
346     &tcp_reass_, "tcp_reass", "append to tail fragment");
347 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
348     &tcp_reass_, "tcp_reass", "overlap at end");
349 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
350     &tcp_reass_, "tcp_reass", "overlap at start");
351 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
352     &tcp_reass_, "tcp_reass", "duplicate segment");
353 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
354     &tcp_reass_, "tcp_reass", "duplicate fragment");
355 
356 EVCNT_ATTACH_STATIC(tcp_reass_);
357 EVCNT_ATTACH_STATIC(tcp_reass_empty);
358 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
359 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
366 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
367 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
368 EVCNT_ATTACH_STATIC(tcp_reass_insert);
369 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
370 EVCNT_ATTACH_STATIC(tcp_reass_append);
371 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
372 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
373 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
374 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
375 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
376 
377 #endif /* TCP_REASS_COUNTERS */
378 
379 #ifdef MBUFTRACE
380 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
381 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
382 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
383 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
384 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
385 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
386 #endif
387 
388 callout_t tcp_slowtimo_ch;
389 
390 static int
391 do_tcpinit(void)
392 {
393 
394 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
395 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
396 	    NULL, IPL_SOFTNET);
397 
398 	tcp_usrreq_init();
399 
400 	/* Initialize timer state. */
401 	tcp_timer_init();
402 
403 	/* Initialize the compressed state engine. */
404 	syn_cache_init();
405 
406 	/* Initialize the congestion control algorithms. */
407 	tcp_congctl_init();
408 
409 	/* Initialize the TCPCB template. */
410 	tcp_tcpcb_template();
411 
412 	/* Initialize reassembly queue */
413 	tcpipqent_init();
414 
415 	/* SACK */
416 	tcp_sack_init();
417 
418 	MOWNER_ATTACH(&tcp_tx_mowner);
419 	MOWNER_ATTACH(&tcp_rx_mowner);
420 	MOWNER_ATTACH(&tcp_reass_mowner);
421 	MOWNER_ATTACH(&tcp_sock_mowner);
422 	MOWNER_ATTACH(&tcp_sock_tx_mowner);
423 	MOWNER_ATTACH(&tcp_sock_rx_mowner);
424 	MOWNER_ATTACH(&tcp_mowner);
425 
426 	tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
427 
428 	vtw_earlyinit();
429 
430 	callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
431 	callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
432 
433 	return 0;
434 }
435 
436 void
437 tcp_init_common(unsigned basehlen)
438 {
439 	static ONCE_DECL(dotcpinit);
440 	unsigned hlen = basehlen + sizeof(struct tcphdr);
441 	unsigned oldhlen;
442 
443 	if (max_linkhdr + hlen > MHLEN)
444 		panic("tcp_init");
445 	while ((oldhlen = max_protohdr) < hlen)
446 		atomic_cas_uint(&max_protohdr, oldhlen, hlen);
447 
448 	RUN_ONCE(&dotcpinit, do_tcpinit);
449 }
450 
451 /*
452  * Tcp initialization
453  */
454 void
455 tcp_init(void)
456 {
457 
458 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
459 
460 	tcp_init_common(sizeof(struct ip));
461 }
462 
463 /*
464  * Create template to be used to send tcp packets on a connection.
465  * Call after host entry created, allocates an mbuf and fills
466  * in a skeletal tcp/ip header, minimizing the amount of work
467  * necessary when the connection is used.
468  */
469 struct mbuf *
470 tcp_template(struct tcpcb *tp)
471 {
472 	struct inpcb *inp = tp->t_inpcb;
473 #ifdef INET6
474 	struct in6pcb *in6p = tp->t_in6pcb;
475 #endif
476 	struct tcphdr *n;
477 	struct mbuf *m;
478 	int hlen;
479 
480 	switch (tp->t_family) {
481 	case AF_INET:
482 		hlen = sizeof(struct ip);
483 		if (inp)
484 			break;
485 #ifdef INET6
486 		if (in6p) {
487 			/* mapped addr case */
488 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
489 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
490 				break;
491 		}
492 #endif
493 		return NULL;	/*EINVAL*/
494 #ifdef INET6
495 	case AF_INET6:
496 		hlen = sizeof(struct ip6_hdr);
497 		if (in6p) {
498 			/* more sainty check? */
499 			break;
500 		}
501 		return NULL;	/*EINVAL*/
502 #endif
503 	default:
504 		hlen = 0;	/*pacify gcc*/
505 		return NULL;	/*EAFNOSUPPORT*/
506 	}
507 #ifdef DIAGNOSTIC
508 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
509 		panic("mclbytes too small for t_template");
510 #endif
511 	m = tp->t_template;
512 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
513 		;
514 	else {
515 		if (m)
516 			m_freem(m);
517 		m = tp->t_template = NULL;
518 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
519 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
520 			MCLGET(m, M_DONTWAIT);
521 			if ((m->m_flags & M_EXT) == 0) {
522 				m_free(m);
523 				m = NULL;
524 			}
525 		}
526 		if (m == NULL)
527 			return NULL;
528 		MCLAIM(m, &tcp_mowner);
529 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
530 	}
531 
532 	memset(mtod(m, void *), 0, m->m_len);
533 
534 	n = (struct tcphdr *)(mtod(m, char *) + hlen);
535 
536 	switch (tp->t_family) {
537 	case AF_INET:
538 	    {
539 		struct ipovly *ipov;
540 		mtod(m, struct ip *)->ip_v = 4;
541 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
542 		ipov = mtod(m, struct ipovly *);
543 		ipov->ih_pr = IPPROTO_TCP;
544 		ipov->ih_len = htons(sizeof(struct tcphdr));
545 		if (inp) {
546 			ipov->ih_src = inp->inp_laddr;
547 			ipov->ih_dst = inp->inp_faddr;
548 		}
549 #ifdef INET6
550 		else if (in6p) {
551 			/* mapped addr case */
552 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
553 				sizeof(ipov->ih_src));
554 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
555 				sizeof(ipov->ih_dst));
556 		}
557 #endif
558 		/*
559 		 * Compute the pseudo-header portion of the checksum
560 		 * now.  We incrementally add in the TCP option and
561 		 * payload lengths later, and then compute the TCP
562 		 * checksum right before the packet is sent off onto
563 		 * the wire.
564 		 */
565 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
566 		    ipov->ih_dst.s_addr,
567 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
568 		break;
569 	    }
570 #ifdef INET6
571 	case AF_INET6:
572 	    {
573 		struct ip6_hdr *ip6;
574 		mtod(m, struct ip *)->ip_v = 6;
575 		ip6 = mtod(m, struct ip6_hdr *);
576 		ip6->ip6_nxt = IPPROTO_TCP;
577 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
578 		ip6->ip6_src = in6p->in6p_laddr;
579 		ip6->ip6_dst = in6p->in6p_faddr;
580 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
581 		if (ip6_auto_flowlabel) {
582 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
583 			ip6->ip6_flow |=
584 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
585 		}
586 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
587 		ip6->ip6_vfc |= IPV6_VERSION;
588 
589 		/*
590 		 * Compute the pseudo-header portion of the checksum
591 		 * now.  We incrementally add in the TCP option and
592 		 * payload lengths later, and then compute the TCP
593 		 * checksum right before the packet is sent off onto
594 		 * the wire.
595 		 */
596 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
597 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
598 		    htonl(IPPROTO_TCP));
599 		break;
600 	    }
601 #endif
602 	}
603 	if (inp) {
604 		n->th_sport = inp->inp_lport;
605 		n->th_dport = inp->inp_fport;
606 	}
607 #ifdef INET6
608 	else if (in6p) {
609 		n->th_sport = in6p->in6p_lport;
610 		n->th_dport = in6p->in6p_fport;
611 	}
612 #endif
613 	n->th_seq = 0;
614 	n->th_ack = 0;
615 	n->th_x2 = 0;
616 	n->th_off = 5;
617 	n->th_flags = 0;
618 	n->th_win = 0;
619 	n->th_urp = 0;
620 	return (m);
621 }
622 
623 /*
624  * Send a single message to the TCP at address specified by
625  * the given TCP/IP header.  If m == 0, then we make a copy
626  * of the tcpiphdr at ti and send directly to the addressed host.
627  * This is used to force keep alive messages out using the TCP
628  * template for a connection tp->t_template.  If flags are given
629  * then we send a message back to the TCP which originated the
630  * segment ti, and discard the mbuf containing it and any other
631  * attached mbufs.
632  *
633  * In any case the ack and sequence number of the transmitted
634  * segment are as specified by the parameters.
635  */
636 int
637 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
638     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
639 {
640 #ifdef INET6
641 	struct rtentry *rt;
642 #endif
643 	struct route *ro;
644 	int error, tlen, win = 0;
645 	int hlen;
646 	struct ip *ip;
647 #ifdef INET6
648 	struct ip6_hdr *ip6;
649 #endif
650 	int family;	/* family on packet, not inpcb/in6pcb! */
651 	struct tcphdr *th;
652 	struct socket *so;
653 
654 	if (tp != NULL && (flags & TH_RST) == 0) {
655 #ifdef DIAGNOSTIC
656 		if (tp->t_inpcb && tp->t_in6pcb)
657 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
658 #endif
659 #ifdef INET
660 		if (tp->t_inpcb)
661 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
662 #endif
663 #ifdef INET6
664 		if (tp->t_in6pcb)
665 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
666 #endif
667 	}
668 
669 	th = NULL;	/* Quell uninitialized warning */
670 	ip = NULL;
671 #ifdef INET6
672 	ip6 = NULL;
673 #endif
674 	if (m == 0) {
675 		if (!mtemplate)
676 			return EINVAL;
677 
678 		/* get family information from template */
679 		switch (mtod(mtemplate, struct ip *)->ip_v) {
680 		case 4:
681 			family = AF_INET;
682 			hlen = sizeof(struct ip);
683 			break;
684 #ifdef INET6
685 		case 6:
686 			family = AF_INET6;
687 			hlen = sizeof(struct ip6_hdr);
688 			break;
689 #endif
690 		default:
691 			return EAFNOSUPPORT;
692 		}
693 
694 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
695 		if (m) {
696 			MCLAIM(m, &tcp_tx_mowner);
697 			MCLGET(m, M_DONTWAIT);
698 			if ((m->m_flags & M_EXT) == 0) {
699 				m_free(m);
700 				m = NULL;
701 			}
702 		}
703 		if (m == NULL)
704 			return (ENOBUFS);
705 
706 		if (tcp_compat_42)
707 			tlen = 1;
708 		else
709 			tlen = 0;
710 
711 		m->m_data += max_linkhdr;
712 		bcopy(mtod(mtemplate, void *), mtod(m, void *),
713 			mtemplate->m_len);
714 		switch (family) {
715 		case AF_INET:
716 			ip = mtod(m, struct ip *);
717 			th = (struct tcphdr *)(ip + 1);
718 			break;
719 #ifdef INET6
720 		case AF_INET6:
721 			ip6 = mtod(m, struct ip6_hdr *);
722 			th = (struct tcphdr *)(ip6 + 1);
723 			break;
724 #endif
725 #if 0
726 		default:
727 			/* noone will visit here */
728 			m_freem(m);
729 			return EAFNOSUPPORT;
730 #endif
731 		}
732 		flags = TH_ACK;
733 	} else {
734 
735 		if ((m->m_flags & M_PKTHDR) == 0) {
736 #if 0
737 			printf("non PKTHDR to tcp_respond\n");
738 #endif
739 			m_freem(m);
740 			return EINVAL;
741 		}
742 #ifdef DIAGNOSTIC
743 		if (!th0)
744 			panic("th0 == NULL in tcp_respond");
745 #endif
746 
747 		/* get family information from m */
748 		switch (mtod(m, struct ip *)->ip_v) {
749 		case 4:
750 			family = AF_INET;
751 			hlen = sizeof(struct ip);
752 			ip = mtod(m, struct ip *);
753 			break;
754 #ifdef INET6
755 		case 6:
756 			family = AF_INET6;
757 			hlen = sizeof(struct ip6_hdr);
758 			ip6 = mtod(m, struct ip6_hdr *);
759 			break;
760 #endif
761 		default:
762 			m_freem(m);
763 			return EAFNOSUPPORT;
764 		}
765 		/* clear h/w csum flags inherited from rx packet */
766 		m->m_pkthdr.csum_flags = 0;
767 
768 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
769 			tlen = sizeof(*th0);
770 		else
771 			tlen = th0->th_off << 2;
772 
773 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
774 		    mtod(m, char *) + hlen == (char *)th0) {
775 			m->m_len = hlen + tlen;
776 			m_freem(m->m_next);
777 			m->m_next = NULL;
778 		} else {
779 			struct mbuf *n;
780 
781 #ifdef DIAGNOSTIC
782 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
783 				m_freem(m);
784 				return EMSGSIZE;
785 			}
786 #endif
787 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
788 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
789 				MCLGET(n, M_DONTWAIT);
790 				if ((n->m_flags & M_EXT) == 0) {
791 					m_freem(n);
792 					n = NULL;
793 				}
794 			}
795 			if (!n) {
796 				m_freem(m);
797 				return ENOBUFS;
798 			}
799 
800 			MCLAIM(n, &tcp_tx_mowner);
801 			n->m_data += max_linkhdr;
802 			n->m_len = hlen + tlen;
803 			m_copyback(n, 0, hlen, mtod(m, void *));
804 			m_copyback(n, hlen, tlen, (void *)th0);
805 
806 			m_freem(m);
807 			m = n;
808 			n = NULL;
809 		}
810 
811 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
812 		switch (family) {
813 		case AF_INET:
814 			ip = mtod(m, struct ip *);
815 			th = (struct tcphdr *)(ip + 1);
816 			ip->ip_p = IPPROTO_TCP;
817 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
818 			ip->ip_p = IPPROTO_TCP;
819 			break;
820 #ifdef INET6
821 		case AF_INET6:
822 			ip6 = mtod(m, struct ip6_hdr *);
823 			th = (struct tcphdr *)(ip6 + 1);
824 			ip6->ip6_nxt = IPPROTO_TCP;
825 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
826 			ip6->ip6_nxt = IPPROTO_TCP;
827 			break;
828 #endif
829 #if 0
830 		default:
831 			/* noone will visit here */
832 			m_freem(m);
833 			return EAFNOSUPPORT;
834 #endif
835 		}
836 		xchg(th->th_dport, th->th_sport, u_int16_t);
837 #undef xchg
838 		tlen = 0;	/*be friendly with the following code*/
839 	}
840 	th->th_seq = htonl(seq);
841 	th->th_ack = htonl(ack);
842 	th->th_x2 = 0;
843 	if ((flags & TH_SYN) == 0) {
844 		if (tp)
845 			win >>= tp->rcv_scale;
846 		if (win > TCP_MAXWIN)
847 			win = TCP_MAXWIN;
848 		th->th_win = htons((u_int16_t)win);
849 		th->th_off = sizeof (struct tcphdr) >> 2;
850 		tlen += sizeof(*th);
851 	} else
852 		tlen += th->th_off << 2;
853 	m->m_len = hlen + tlen;
854 	m->m_pkthdr.len = hlen + tlen;
855 	m->m_pkthdr.rcvif = NULL;
856 	th->th_flags = flags;
857 	th->th_urp = 0;
858 
859 	switch (family) {
860 #ifdef INET
861 	case AF_INET:
862 	    {
863 		struct ipovly *ipov = (struct ipovly *)ip;
864 		memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
865 		ipov->ih_len = htons((u_int16_t)tlen);
866 
867 		th->th_sum = 0;
868 		th->th_sum = in_cksum(m, hlen + tlen);
869 		ip->ip_len = htons(hlen + tlen);
870 		ip->ip_ttl = ip_defttl;
871 		break;
872 	    }
873 #endif
874 #ifdef INET6
875 	case AF_INET6:
876 	    {
877 		th->th_sum = 0;
878 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
879 				tlen);
880 		ip6->ip6_plen = htons(tlen);
881 		if (tp && tp->t_in6pcb) {
882 			struct ifnet *oifp;
883 			ro = &tp->t_in6pcb->in6p_route;
884 			oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
885 			                                           : NULL;
886 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
887 		} else
888 			ip6->ip6_hlim = ip6_defhlim;
889 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
890 		if (ip6_auto_flowlabel) {
891 			ip6->ip6_flow |=
892 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
893 		}
894 		break;
895 	    }
896 #endif
897 	}
898 
899 	if (tp && tp->t_inpcb)
900 		so = tp->t_inpcb->inp_socket;
901 #ifdef INET6
902 	else if (tp && tp->t_in6pcb)
903 		so = tp->t_in6pcb->in6p_socket;
904 #endif
905 	else
906 		so = NULL;
907 
908 	if (tp != NULL && tp->t_inpcb != NULL) {
909 		ro = &tp->t_inpcb->inp_route;
910 #ifdef DIAGNOSTIC
911 		if (family != AF_INET)
912 			panic("tcp_respond: address family mismatch");
913 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
914 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
915 			    ntohl(ip->ip_dst.s_addr),
916 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
917 		}
918 #endif
919 	}
920 #ifdef INET6
921 	else if (tp != NULL && tp->t_in6pcb != NULL) {
922 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
923 #ifdef DIAGNOSTIC
924 		if (family == AF_INET) {
925 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
926 				panic("tcp_respond: not mapped addr");
927 			if (memcmp(&ip->ip_dst,
928 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
929 			    sizeof(ip->ip_dst)) != 0) {
930 				panic("tcp_respond: ip_dst != in6p_faddr");
931 			}
932 		} else if (family == AF_INET6) {
933 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
934 			    &tp->t_in6pcb->in6p_faddr))
935 				panic("tcp_respond: ip6_dst != in6p_faddr");
936 		} else
937 			panic("tcp_respond: address family mismatch");
938 #endif
939 	}
940 #endif
941 	else
942 		ro = NULL;
943 
944 	switch (family) {
945 #ifdef INET
946 	case AF_INET:
947 		error = ip_output(m, NULL, ro,
948 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so);
949 		break;
950 #endif
951 #ifdef INET6
952 	case AF_INET6:
953 		error = ip6_output(m, NULL, ro, 0, NULL, so, NULL);
954 		break;
955 #endif
956 	default:
957 		error = EAFNOSUPPORT;
958 		break;
959 	}
960 
961 	return (error);
962 }
963 
964 /*
965  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
966  * a bunch of members individually, we maintain this template for the
967  * static and mostly-static components of the TCPCB, and copy it into
968  * the new TCPCB instead.
969  */
970 static struct tcpcb tcpcb_template = {
971 	.t_srtt = TCPTV_SRTTBASE,
972 	.t_rttmin = TCPTV_MIN,
973 
974 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
975 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
976 	.snd_numholes = 0,
977 	.snd_cubic_wmax = 0,
978 	.snd_cubic_wmax_last = 0,
979 	.snd_cubic_ctime = 0,
980 
981 	.t_partialacks = -1,
982 	.t_bytes_acked = 0,
983 	.t_sndrexmitpack = 0,
984 	.t_rcvoopack = 0,
985 	.t_sndzerowin = 0,
986 };
987 
988 /*
989  * Updates the TCPCB template whenever a parameter that would affect
990  * the template is changed.
991  */
992 void
993 tcp_tcpcb_template(void)
994 {
995 	struct tcpcb *tp = &tcpcb_template;
996 	int flags;
997 
998 	tp->t_peermss = tcp_mssdflt;
999 	tp->t_ourmss = tcp_mssdflt;
1000 	tp->t_segsz = tcp_mssdflt;
1001 
1002 	flags = 0;
1003 	if (tcp_do_rfc1323 && tcp_do_win_scale)
1004 		flags |= TF_REQ_SCALE;
1005 	if (tcp_do_rfc1323 && tcp_do_timestamps)
1006 		flags |= TF_REQ_TSTMP;
1007 	tp->t_flags = flags;
1008 
1009 	/*
1010 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1011 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
1012 	 * reasonable initial retransmit time.
1013 	 */
1014 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
1015 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1016 	    TCPTV_MIN, TCPTV_REXMTMAX);
1017 
1018 	/* Keep Alive */
1019 	tp->t_keepinit = tcp_keepinit;
1020 	tp->t_keepidle = tcp_keepidle;
1021 	tp->t_keepintvl = tcp_keepintvl;
1022 	tp->t_keepcnt = tcp_keepcnt;
1023 	tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1024 
1025 	/* MSL */
1026 	tp->t_msl = TCPTV_MSL;
1027 }
1028 
1029 /*
1030  * Create a new TCP control block, making an
1031  * empty reassembly queue and hooking it to the argument
1032  * protocol control block.
1033  */
1034 /* family selects inpcb, or in6pcb */
1035 struct tcpcb *
1036 tcp_newtcpcb(int family, void *aux)
1037 {
1038 #ifdef INET6
1039 	struct rtentry *rt;
1040 #endif
1041 	struct tcpcb *tp;
1042 	int i;
1043 
1044 	/* XXX Consider using a pool_cache for speed. */
1045 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);	/* splsoftnet via tcp_usrreq */
1046 	if (tp == NULL)
1047 		return (NULL);
1048 	memcpy(tp, &tcpcb_template, sizeof(*tp));
1049 	TAILQ_INIT(&tp->segq);
1050 	TAILQ_INIT(&tp->timeq);
1051 	tp->t_family = family;		/* may be overridden later on */
1052 	TAILQ_INIT(&tp->snd_holes);
1053 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
1054 
1055 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
1056 	for (i = 0; i < TCPT_NTIMERS; i++) {
1057 		callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1058 		TCP_TIMER_INIT(tp, i);
1059 	}
1060 	callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1061 
1062 	switch (family) {
1063 	case AF_INET:
1064 	    {
1065 		struct inpcb *inp = (struct inpcb *)aux;
1066 
1067 		inp->inp_ip.ip_ttl = ip_defttl;
1068 		inp->inp_ppcb = (void *)tp;
1069 
1070 		tp->t_inpcb = inp;
1071 		tp->t_mtudisc = ip_mtudisc;
1072 		break;
1073 	    }
1074 #ifdef INET6
1075 	case AF_INET6:
1076 	    {
1077 		struct in6pcb *in6p = (struct in6pcb *)aux;
1078 
1079 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
1080 			(rt = rtcache_validate(&in6p->in6p_route)) != NULL
1081 			    ? rt->rt_ifp
1082 			    : NULL);
1083 		in6p->in6p_ppcb = (void *)tp;
1084 
1085 		tp->t_in6pcb = in6p;
1086 		/* for IPv6, always try to run path MTU discovery */
1087 		tp->t_mtudisc = 1;
1088 		break;
1089 	    }
1090 #endif /* INET6 */
1091 	default:
1092 		for (i = 0; i < TCPT_NTIMERS; i++)
1093 			callout_destroy(&tp->t_timer[i]);
1094 		callout_destroy(&tp->t_delack_ch);
1095 		pool_put(&tcpcb_pool, tp);	/* splsoftnet via tcp_usrreq */
1096 		return (NULL);
1097 	}
1098 
1099 	/*
1100 	 * Initialize our timebase.  When we send timestamps, we take
1101 	 * the delta from tcp_now -- this means each connection always
1102 	 * gets a timebase of 1, which makes it, among other things,
1103 	 * more difficult to determine how long a system has been up,
1104 	 * and thus how many TCP sequence increments have occurred.
1105 	 *
1106 	 * We start with 1, because 0 doesn't work with linux, which
1107 	 * considers timestamp 0 in a SYN packet as a bug and disables
1108 	 * timestamps.
1109 	 */
1110 	tp->ts_timebase = tcp_now - 1;
1111 
1112 	tcp_congctl_select(tp, tcp_congctl_global_name);
1113 
1114 	return (tp);
1115 }
1116 
1117 /*
1118  * Drop a TCP connection, reporting
1119  * the specified error.  If connection is synchronized,
1120  * then send a RST to peer.
1121  */
1122 struct tcpcb *
1123 tcp_drop(struct tcpcb *tp, int errno)
1124 {
1125 	struct socket *so = NULL;
1126 
1127 #ifdef DIAGNOSTIC
1128 	if (tp->t_inpcb && tp->t_in6pcb)
1129 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1130 #endif
1131 #ifdef INET
1132 	if (tp->t_inpcb)
1133 		so = tp->t_inpcb->inp_socket;
1134 #endif
1135 #ifdef INET6
1136 	if (tp->t_in6pcb)
1137 		so = tp->t_in6pcb->in6p_socket;
1138 #endif
1139 	if (!so)
1140 		return NULL;
1141 
1142 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1143 		tp->t_state = TCPS_CLOSED;
1144 		(void) tcp_output(tp);
1145 		TCP_STATINC(TCP_STAT_DROPS);
1146 	} else
1147 		TCP_STATINC(TCP_STAT_CONNDROPS);
1148 	if (errno == ETIMEDOUT && tp->t_softerror)
1149 		errno = tp->t_softerror;
1150 	so->so_error = errno;
1151 	return (tcp_close(tp));
1152 }
1153 
1154 /*
1155  * Close a TCP control block:
1156  *	discard all space held by the tcp
1157  *	discard internet protocol block
1158  *	wake up any sleepers
1159  */
1160 struct tcpcb *
1161 tcp_close(struct tcpcb *tp)
1162 {
1163 	struct inpcb *inp;
1164 #ifdef INET6
1165 	struct in6pcb *in6p;
1166 #endif
1167 	struct socket *so;
1168 #ifdef RTV_RTT
1169 	struct rtentry *rt;
1170 #endif
1171 	struct route *ro;
1172 	int j;
1173 
1174 	inp = tp->t_inpcb;
1175 #ifdef INET6
1176 	in6p = tp->t_in6pcb;
1177 #endif
1178 	so = NULL;
1179 	ro = NULL;
1180 	if (inp) {
1181 		so = inp->inp_socket;
1182 		ro = &inp->inp_route;
1183 	}
1184 #ifdef INET6
1185 	else if (in6p) {
1186 		so = in6p->in6p_socket;
1187 		ro = (struct route *)&in6p->in6p_route;
1188 	}
1189 #endif
1190 
1191 #ifdef RTV_RTT
1192 	/*
1193 	 * If we sent enough data to get some meaningful characteristics,
1194 	 * save them in the routing entry.  'Enough' is arbitrarily
1195 	 * defined as the sendpipesize (default 4K) * 16.  This would
1196 	 * give us 16 rtt samples assuming we only get one sample per
1197 	 * window (the usual case on a long haul net).  16 samples is
1198 	 * enough for the srtt filter to converge to within 5% of the correct
1199 	 * value; fewer samples and we could save a very bogus rtt.
1200 	 *
1201 	 * Don't update the default route's characteristics and don't
1202 	 * update anything that the user "locked".
1203 	 */
1204 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1205 	    ro && (rt = rtcache_validate(ro)) != NULL &&
1206 	    !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1207 		u_long i = 0;
1208 
1209 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1210 			i = tp->t_srtt *
1211 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1212 			if (rt->rt_rmx.rmx_rtt && i)
1213 				/*
1214 				 * filter this update to half the old & half
1215 				 * the new values, converting scale.
1216 				 * See route.h and tcp_var.h for a
1217 				 * description of the scaling constants.
1218 				 */
1219 				rt->rt_rmx.rmx_rtt =
1220 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1221 			else
1222 				rt->rt_rmx.rmx_rtt = i;
1223 		}
1224 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1225 			i = tp->t_rttvar *
1226 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1227 			if (rt->rt_rmx.rmx_rttvar && i)
1228 				rt->rt_rmx.rmx_rttvar =
1229 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1230 			else
1231 				rt->rt_rmx.rmx_rttvar = i;
1232 		}
1233 		/*
1234 		 * update the pipelimit (ssthresh) if it has been updated
1235 		 * already or if a pipesize was specified & the threshhold
1236 		 * got below half the pipesize.  I.e., wait for bad news
1237 		 * before we start updating, then update on both good
1238 		 * and bad news.
1239 		 */
1240 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1241 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1242 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1243 			/*
1244 			 * convert the limit from user data bytes to
1245 			 * packets then to packet data bytes.
1246 			 */
1247 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1248 			if (i < 2)
1249 				i = 2;
1250 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1251 			if (rt->rt_rmx.rmx_ssthresh)
1252 				rt->rt_rmx.rmx_ssthresh =
1253 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1254 			else
1255 				rt->rt_rmx.rmx_ssthresh = i;
1256 		}
1257 	}
1258 #endif /* RTV_RTT */
1259 	/* free the reassembly queue, if any */
1260 	TCP_REASS_LOCK(tp);
1261 	(void) tcp_freeq(tp);
1262 	TCP_REASS_UNLOCK(tp);
1263 
1264 	/* free the SACK holes list. */
1265 	tcp_free_sackholes(tp);
1266 	tcp_congctl_release(tp);
1267 	syn_cache_cleanup(tp);
1268 
1269 	if (tp->t_template) {
1270 		m_free(tp->t_template);
1271 		tp->t_template = NULL;
1272 	}
1273 
1274 	/*
1275 	 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1276 	 * the timers can also drop the lock.  We need to prevent access
1277 	 * to the tcpcb as it's half torn down.  Flag the pcb as dead
1278 	 * (prevents access by timers) and only then detach it.
1279 	 */
1280 	tp->t_flags |= TF_DEAD;
1281 	if (inp) {
1282 		inp->inp_ppcb = 0;
1283 		soisdisconnected(so);
1284 		in_pcbdetach(inp);
1285 	}
1286 #ifdef INET6
1287 	else if (in6p) {
1288 		in6p->in6p_ppcb = 0;
1289 		soisdisconnected(so);
1290 		in6_pcbdetach(in6p);
1291 	}
1292 #endif
1293 	/*
1294 	 * pcb is no longer visble elsewhere, so we can safely release
1295 	 * the lock in callout_halt() if needed.
1296 	 */
1297 	TCP_STATINC(TCP_STAT_CLOSED);
1298 	for (j = 0; j < TCPT_NTIMERS; j++) {
1299 		callout_halt(&tp->t_timer[j], softnet_lock);
1300 		callout_destroy(&tp->t_timer[j]);
1301 	}
1302 	callout_halt(&tp->t_delack_ch, softnet_lock);
1303 	callout_destroy(&tp->t_delack_ch);
1304 	pool_put(&tcpcb_pool, tp);
1305 
1306 	return NULL;
1307 }
1308 
1309 int
1310 tcp_freeq(struct tcpcb *tp)
1311 {
1312 	struct ipqent *qe;
1313 	int rv = 0;
1314 #ifdef TCPREASS_DEBUG
1315 	int i = 0;
1316 #endif
1317 
1318 	TCP_REASS_LOCK_CHECK(tp);
1319 
1320 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1321 #ifdef TCPREASS_DEBUG
1322 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1323 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1324 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1325 #endif
1326 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1327 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1328 		m_freem(qe->ipqe_m);
1329 		tcpipqent_free(qe);
1330 		rv = 1;
1331 	}
1332 	tp->t_segqlen = 0;
1333 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1334 	return (rv);
1335 }
1336 
1337 void
1338 tcp_fasttimo(void)
1339 {
1340 	if (tcp_drainwanted) {
1341 		tcp_drain();
1342 		tcp_drainwanted = 0;
1343 	}
1344 }
1345 
1346 void
1347 tcp_drainstub(void)
1348 {
1349 	tcp_drainwanted = 1;
1350 }
1351 
1352 /*
1353  * Protocol drain routine.  Called when memory is in short supply.
1354  * Called from pr_fasttimo thus a callout context.
1355  */
1356 void
1357 tcp_drain(void)
1358 {
1359 	struct inpcb_hdr *inph;
1360 	struct tcpcb *tp;
1361 
1362 	mutex_enter(softnet_lock);
1363 	KERNEL_LOCK(1, NULL);
1364 
1365 	/*
1366 	 * Free the sequence queue of all TCP connections.
1367 	 */
1368 	TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1369 		switch (inph->inph_af) {
1370 		case AF_INET:
1371 			tp = intotcpcb((struct inpcb *)inph);
1372 			break;
1373 #ifdef INET6
1374 		case AF_INET6:
1375 			tp = in6totcpcb((struct in6pcb *)inph);
1376 			break;
1377 #endif
1378 		default:
1379 			tp = NULL;
1380 			break;
1381 		}
1382 		if (tp != NULL) {
1383 			/*
1384 			 * We may be called from a device's interrupt
1385 			 * context.  If the tcpcb is already busy,
1386 			 * just bail out now.
1387 			 */
1388 			if (tcp_reass_lock_try(tp) == 0)
1389 				continue;
1390 			if (tcp_freeq(tp))
1391 				TCP_STATINC(TCP_STAT_CONNSDRAINED);
1392 			TCP_REASS_UNLOCK(tp);
1393 		}
1394 	}
1395 
1396 	KERNEL_UNLOCK_ONE(NULL);
1397 	mutex_exit(softnet_lock);
1398 }
1399 
1400 /*
1401  * Notify a tcp user of an asynchronous error;
1402  * store error as soft error, but wake up user
1403  * (for now, won't do anything until can select for soft error).
1404  */
1405 void
1406 tcp_notify(struct inpcb *inp, int error)
1407 {
1408 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1409 	struct socket *so = inp->inp_socket;
1410 
1411 	/*
1412 	 * Ignore some errors if we are hooked up.
1413 	 * If connection hasn't completed, has retransmitted several times,
1414 	 * and receives a second error, give up now.  This is better
1415 	 * than waiting a long time to establish a connection that
1416 	 * can never complete.
1417 	 */
1418 	if (tp->t_state == TCPS_ESTABLISHED &&
1419 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1420 	      error == EHOSTDOWN)) {
1421 		return;
1422 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1423 	    tp->t_rxtshift > 3 && tp->t_softerror)
1424 		so->so_error = error;
1425 	else
1426 		tp->t_softerror = error;
1427 	cv_broadcast(&so->so_cv);
1428 	sorwakeup(so);
1429 	sowwakeup(so);
1430 }
1431 
1432 #ifdef INET6
1433 void
1434 tcp6_notify(struct in6pcb *in6p, int error)
1435 {
1436 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1437 	struct socket *so = in6p->in6p_socket;
1438 
1439 	/*
1440 	 * Ignore some errors if we are hooked up.
1441 	 * If connection hasn't completed, has retransmitted several times,
1442 	 * and receives a second error, give up now.  This is better
1443 	 * than waiting a long time to establish a connection that
1444 	 * can never complete.
1445 	 */
1446 	if (tp->t_state == TCPS_ESTABLISHED &&
1447 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1448 	      error == EHOSTDOWN)) {
1449 		return;
1450 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1451 	    tp->t_rxtshift > 3 && tp->t_softerror)
1452 		so->so_error = error;
1453 	else
1454 		tp->t_softerror = error;
1455 	cv_broadcast(&so->so_cv);
1456 	sorwakeup(so);
1457 	sowwakeup(so);
1458 }
1459 #endif
1460 
1461 #ifdef INET6
1462 void *
1463 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1464 {
1465 	struct tcphdr th;
1466 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1467 	int nmatch;
1468 	struct ip6_hdr *ip6;
1469 	const struct sockaddr_in6 *sa6_src = NULL;
1470 	const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1471 	struct mbuf *m;
1472 	int off;
1473 
1474 	if (sa->sa_family != AF_INET6 ||
1475 	    sa->sa_len != sizeof(struct sockaddr_in6))
1476 		return NULL;
1477 	if ((unsigned)cmd >= PRC_NCMDS)
1478 		return NULL;
1479 	else if (cmd == PRC_QUENCH) {
1480 		/*
1481 		 * Don't honor ICMP Source Quench messages meant for
1482 		 * TCP connections.
1483 		 */
1484 		return NULL;
1485 	} else if (PRC_IS_REDIRECT(cmd))
1486 		notify = in6_rtchange, d = NULL;
1487 	else if (cmd == PRC_MSGSIZE)
1488 		; /* special code is present, see below */
1489 	else if (cmd == PRC_HOSTDEAD)
1490 		d = NULL;
1491 	else if (inet6ctlerrmap[cmd] == 0)
1492 		return NULL;
1493 
1494 	/* if the parameter is from icmp6, decode it. */
1495 	if (d != NULL) {
1496 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1497 		m = ip6cp->ip6c_m;
1498 		ip6 = ip6cp->ip6c_ip6;
1499 		off = ip6cp->ip6c_off;
1500 		sa6_src = ip6cp->ip6c_src;
1501 	} else {
1502 		m = NULL;
1503 		ip6 = NULL;
1504 		sa6_src = &sa6_any;
1505 		off = 0;
1506 	}
1507 
1508 	if (ip6) {
1509 		/*
1510 		 * XXX: We assume that when ip6 is non NULL,
1511 		 * M and OFF are valid.
1512 		 */
1513 
1514 		/* check if we can safely examine src and dst ports */
1515 		if (m->m_pkthdr.len < off + sizeof(th)) {
1516 			if (cmd == PRC_MSGSIZE)
1517 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1518 			return NULL;
1519 		}
1520 
1521 		memset(&th, 0, sizeof(th));
1522 		m_copydata(m, off, sizeof(th), (void *)&th);
1523 
1524 		if (cmd == PRC_MSGSIZE) {
1525 			int valid = 0;
1526 
1527 			/*
1528 			 * Check to see if we have a valid TCP connection
1529 			 * corresponding to the address in the ICMPv6 message
1530 			 * payload.
1531 			 */
1532 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1533 			    th.th_dport,
1534 			    (const struct in6_addr *)&sa6_src->sin6_addr,
1535 						  th.th_sport, 0, 0))
1536 				valid++;
1537 
1538 			/*
1539 			 * Depending on the value of "valid" and routing table
1540 			 * size (mtudisc_{hi,lo}wat), we will:
1541 			 * - recalcurate the new MTU and create the
1542 			 *   corresponding routing entry, or
1543 			 * - ignore the MTU change notification.
1544 			 */
1545 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1546 
1547 			/*
1548 			 * no need to call in6_pcbnotify, it should have been
1549 			 * called via callback if necessary
1550 			 */
1551 			return NULL;
1552 		}
1553 
1554 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1555 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1556 		if (nmatch == 0 && syn_cache_count &&
1557 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1558 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1559 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1560 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1561 					  sa, &th);
1562 	} else {
1563 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1564 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1565 	}
1566 
1567 	return NULL;
1568 }
1569 #endif
1570 
1571 #ifdef INET
1572 /* assumes that ip header and tcp header are contiguous on mbuf */
1573 void *
1574 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1575 {
1576 	struct ip *ip = v;
1577 	struct tcphdr *th;
1578 	struct icmp *icp;
1579 	extern const int inetctlerrmap[];
1580 	void (*notify)(struct inpcb *, int) = tcp_notify;
1581 	int errno;
1582 	int nmatch;
1583 	struct tcpcb *tp;
1584 	u_int mtu;
1585 	tcp_seq seq;
1586 	struct inpcb *inp;
1587 #ifdef INET6
1588 	struct in6pcb *in6p;
1589 	struct in6_addr src6, dst6;
1590 #endif
1591 
1592 	if (sa->sa_family != AF_INET ||
1593 	    sa->sa_len != sizeof(struct sockaddr_in))
1594 		return NULL;
1595 	if ((unsigned)cmd >= PRC_NCMDS)
1596 		return NULL;
1597 	errno = inetctlerrmap[cmd];
1598 	if (cmd == PRC_QUENCH)
1599 		/*
1600 		 * Don't honor ICMP Source Quench messages meant for
1601 		 * TCP connections.
1602 		 */
1603 		return NULL;
1604 	else if (PRC_IS_REDIRECT(cmd))
1605 		notify = in_rtchange, ip = 0;
1606 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1607 		/*
1608 		 * Check to see if we have a valid TCP connection
1609 		 * corresponding to the address in the ICMP message
1610 		 * payload.
1611 		 *
1612 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1613 		 */
1614 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1615 #ifdef INET6
1616 		memset(&src6, 0, sizeof(src6));
1617 		memset(&dst6, 0, sizeof(dst6));
1618 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1619 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1620 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1621 #endif
1622 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1623 						th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1624 #ifdef INET6
1625 			in6p = NULL;
1626 #else
1627 			;
1628 #endif
1629 #ifdef INET6
1630 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1631 						       th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1632 			;
1633 #endif
1634 		else
1635 			return NULL;
1636 
1637 		/*
1638 		 * Now that we've validated that we are actually communicating
1639 		 * with the host indicated in the ICMP message, locate the
1640 		 * ICMP header, recalculate the new MTU, and create the
1641 		 * corresponding routing entry.
1642 		 */
1643 		icp = (struct icmp *)((char *)ip -
1644 		    offsetof(struct icmp, icmp_ip));
1645 		if (inp) {
1646 			if ((tp = intotcpcb(inp)) == NULL)
1647 				return NULL;
1648 		}
1649 #ifdef INET6
1650 		else if (in6p) {
1651 			if ((tp = in6totcpcb(in6p)) == NULL)
1652 				return NULL;
1653 		}
1654 #endif
1655 		else
1656 			return NULL;
1657 		seq = ntohl(th->th_seq);
1658 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1659 			return NULL;
1660 		/*
1661 		 * If the ICMP message advertises a Next-Hop MTU
1662 		 * equal or larger than the maximum packet size we have
1663 		 * ever sent, drop the message.
1664 		 */
1665 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1666 		if (mtu >= tp->t_pmtud_mtu_sent)
1667 			return NULL;
1668 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1669 			/*
1670 			 * Calculate new MTU, and create corresponding
1671 			 * route (traditional PMTUD).
1672 			 */
1673 			tp->t_flags &= ~TF_PMTUD_PEND;
1674 			icmp_mtudisc(icp, ip->ip_dst);
1675 		} else {
1676 			/*
1677 			 * Record the information got in the ICMP
1678 			 * message; act on it later.
1679 			 * If we had already recorded an ICMP message,
1680 			 * replace the old one only if the new message
1681 			 * refers to an older TCP segment
1682 			 */
1683 			if (tp->t_flags & TF_PMTUD_PEND) {
1684 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1685 					return NULL;
1686 			} else
1687 				tp->t_flags |= TF_PMTUD_PEND;
1688 			tp->t_pmtud_th_seq = seq;
1689 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1690 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1691 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1692 		}
1693 		return NULL;
1694 	} else if (cmd == PRC_HOSTDEAD)
1695 		ip = 0;
1696 	else if (errno == 0)
1697 		return NULL;
1698 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1699 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1700 		nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1701 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1702 		if (nmatch == 0 && syn_cache_count &&
1703 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1704 		    inetctlerrmap[cmd] == ENETUNREACH ||
1705 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1706 			struct sockaddr_in sin;
1707 			memset(&sin, 0, sizeof(sin));
1708 			sin.sin_len = sizeof(sin);
1709 			sin.sin_family = AF_INET;
1710 			sin.sin_port = th->th_sport;
1711 			sin.sin_addr = ip->ip_src;
1712 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1713 		}
1714 
1715 		/* XXX mapped address case */
1716 	} else
1717 		in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1718 		    notify);
1719 	return NULL;
1720 }
1721 
1722 /*
1723  * When a source quench is received, we are being notified of congestion.
1724  * Close the congestion window down to the Loss Window (one segment).
1725  * We will gradually open it again as we proceed.
1726  */
1727 void
1728 tcp_quench(struct inpcb *inp, int errno)
1729 {
1730 	struct tcpcb *tp = intotcpcb(inp);
1731 
1732 	if (tp) {
1733 		tp->snd_cwnd = tp->t_segsz;
1734 		tp->t_bytes_acked = 0;
1735 	}
1736 }
1737 #endif
1738 
1739 #ifdef INET6
1740 void
1741 tcp6_quench(struct in6pcb *in6p, int errno)
1742 {
1743 	struct tcpcb *tp = in6totcpcb(in6p);
1744 
1745 	if (tp) {
1746 		tp->snd_cwnd = tp->t_segsz;
1747 		tp->t_bytes_acked = 0;
1748 	}
1749 }
1750 #endif
1751 
1752 #ifdef INET
1753 /*
1754  * Path MTU Discovery handlers.
1755  */
1756 void
1757 tcp_mtudisc_callback(struct in_addr faddr)
1758 {
1759 #ifdef INET6
1760 	struct in6_addr in6;
1761 #endif
1762 
1763 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1764 #ifdef INET6
1765 	memset(&in6, 0, sizeof(in6));
1766 	in6.s6_addr16[5] = 0xffff;
1767 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1768 	tcp6_mtudisc_callback(&in6);
1769 #endif
1770 }
1771 
1772 /*
1773  * On receipt of path MTU corrections, flush old route and replace it
1774  * with the new one.  Retransmit all unacknowledged packets, to ensure
1775  * that all packets will be received.
1776  */
1777 void
1778 tcp_mtudisc(struct inpcb *inp, int errno)
1779 {
1780 	struct tcpcb *tp = intotcpcb(inp);
1781 	struct rtentry *rt = in_pcbrtentry(inp);
1782 
1783 	if (tp != 0) {
1784 		if (rt != 0) {
1785 			/*
1786 			 * If this was not a host route, remove and realloc.
1787 			 */
1788 			if ((rt->rt_flags & RTF_HOST) == 0) {
1789 				in_rtchange(inp, errno);
1790 				if ((rt = in_pcbrtentry(inp)) == 0)
1791 					return;
1792 			}
1793 
1794 			/*
1795 			 * Slow start out of the error condition.  We
1796 			 * use the MTU because we know it's smaller
1797 			 * than the previously transmitted segment.
1798 			 *
1799 			 * Note: This is more conservative than the
1800 			 * suggestion in draft-floyd-incr-init-win-03.
1801 			 */
1802 			if (rt->rt_rmx.rmx_mtu != 0)
1803 				tp->snd_cwnd =
1804 				    TCP_INITIAL_WINDOW(tcp_init_win,
1805 				    rt->rt_rmx.rmx_mtu);
1806 		}
1807 
1808 		/*
1809 		 * Resend unacknowledged packets.
1810 		 */
1811 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1812 		tcp_output(tp);
1813 	}
1814 }
1815 #endif
1816 
1817 #ifdef INET6
1818 /*
1819  * Path MTU Discovery handlers.
1820  */
1821 void
1822 tcp6_mtudisc_callback(struct in6_addr *faddr)
1823 {
1824 	struct sockaddr_in6 sin6;
1825 
1826 	memset(&sin6, 0, sizeof(sin6));
1827 	sin6.sin6_family = AF_INET6;
1828 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1829 	sin6.sin6_addr = *faddr;
1830 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1831 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1832 }
1833 
1834 void
1835 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1836 {
1837 	struct tcpcb *tp = in6totcpcb(in6p);
1838 	struct rtentry *rt = in6_pcbrtentry(in6p);
1839 
1840 	if (tp != 0) {
1841 		if (rt != 0) {
1842 			/*
1843 			 * If this was not a host route, remove and realloc.
1844 			 */
1845 			if ((rt->rt_flags & RTF_HOST) == 0) {
1846 				in6_rtchange(in6p, errno);
1847 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1848 					return;
1849 			}
1850 
1851 			/*
1852 			 * Slow start out of the error condition.  We
1853 			 * use the MTU because we know it's smaller
1854 			 * than the previously transmitted segment.
1855 			 *
1856 			 * Note: This is more conservative than the
1857 			 * suggestion in draft-floyd-incr-init-win-03.
1858 			 */
1859 			if (rt->rt_rmx.rmx_mtu != 0)
1860 				tp->snd_cwnd =
1861 				    TCP_INITIAL_WINDOW(tcp_init_win,
1862 				    rt->rt_rmx.rmx_mtu);
1863 		}
1864 
1865 		/*
1866 		 * Resend unacknowledged packets.
1867 		 */
1868 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1869 		tcp_output(tp);
1870 	}
1871 }
1872 #endif /* INET6 */
1873 
1874 /*
1875  * Compute the MSS to advertise to the peer.  Called only during
1876  * the 3-way handshake.  If we are the server (peer initiated
1877  * connection), we are called with a pointer to the interface
1878  * on which the SYN packet arrived.  If we are the client (we
1879  * initiated connection), we are called with a pointer to the
1880  * interface out which this connection should go.
1881  *
1882  * NOTE: Do not subtract IP option/extension header size nor IPsec
1883  * header size from MSS advertisement.  MSS option must hold the maximum
1884  * segment size we can accept, so it must always be:
1885  *	 max(if mtu) - ip header - tcp header
1886  */
1887 u_long
1888 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1889 {
1890 	extern u_long in_maxmtu;
1891 	u_long mss = 0;
1892 	u_long hdrsiz;
1893 
1894 	/*
1895 	 * In order to avoid defeating path MTU discovery on the peer,
1896 	 * we advertise the max MTU of all attached networks as our MSS,
1897 	 * per RFC 1191, section 3.1.
1898 	 *
1899 	 * We provide the option to advertise just the MTU of
1900 	 * the interface on which we hope this connection will
1901 	 * be receiving.  If we are responding to a SYN, we
1902 	 * will have a pretty good idea about this, but when
1903 	 * initiating a connection there is a bit more doubt.
1904 	 *
1905 	 * We also need to ensure that loopback has a large enough
1906 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1907 	 */
1908 
1909 	if (ifp != NULL)
1910 		switch (af) {
1911 		case AF_INET:
1912 			mss = ifp->if_mtu;
1913 			break;
1914 #ifdef INET6
1915 		case AF_INET6:
1916 			mss = IN6_LINKMTU(ifp);
1917 			break;
1918 #endif
1919 		}
1920 
1921 	if (tcp_mss_ifmtu == 0)
1922 		switch (af) {
1923 		case AF_INET:
1924 			mss = max(in_maxmtu, mss);
1925 			break;
1926 #ifdef INET6
1927 		case AF_INET6:
1928 			mss = max(in6_maxmtu, mss);
1929 			break;
1930 #endif
1931 		}
1932 
1933 	switch (af) {
1934 	case AF_INET:
1935 		hdrsiz = sizeof(struct ip);
1936 		break;
1937 #ifdef INET6
1938 	case AF_INET6:
1939 		hdrsiz = sizeof(struct ip6_hdr);
1940 		break;
1941 #endif
1942 	default:
1943 		hdrsiz = 0;
1944 		break;
1945 	}
1946 	hdrsiz += sizeof(struct tcphdr);
1947 	if (mss > hdrsiz)
1948 		mss -= hdrsiz;
1949 
1950 	mss = max(tcp_mssdflt, mss);
1951 	return (mss);
1952 }
1953 
1954 /*
1955  * Set connection variables based on the peer's advertised MSS.
1956  * We are passed the TCPCB for the actual connection.  If we
1957  * are the server, we are called by the compressed state engine
1958  * when the 3-way handshake is complete.  If we are the client,
1959  * we are called when we receive the SYN,ACK from the server.
1960  *
1961  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1962  * before this routine is called!
1963  */
1964 void
1965 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1966 {
1967 	struct socket *so;
1968 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1969 	struct rtentry *rt;
1970 #endif
1971 	u_long bufsize;
1972 	int mss;
1973 
1974 #ifdef DIAGNOSTIC
1975 	if (tp->t_inpcb && tp->t_in6pcb)
1976 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1977 #endif
1978 	so = NULL;
1979 	rt = NULL;
1980 #ifdef INET
1981 	if (tp->t_inpcb) {
1982 		so = tp->t_inpcb->inp_socket;
1983 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1984 		rt = in_pcbrtentry(tp->t_inpcb);
1985 #endif
1986 	}
1987 #endif
1988 #ifdef INET6
1989 	if (tp->t_in6pcb) {
1990 		so = tp->t_in6pcb->in6p_socket;
1991 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1992 		rt = in6_pcbrtentry(tp->t_in6pcb);
1993 #endif
1994 	}
1995 #endif
1996 
1997 	/*
1998 	 * As per RFC1122, use the default MSS value, unless they
1999 	 * sent us an offer.  Do not accept offers less than 256 bytes.
2000 	 */
2001 	mss = tcp_mssdflt;
2002 	if (offer)
2003 		mss = offer;
2004 	mss = max(mss, 256);		/* sanity */
2005 	tp->t_peermss = mss;
2006 	mss -= tcp_optlen(tp);
2007 #ifdef INET
2008 	if (tp->t_inpcb)
2009 		mss -= ip_optlen(tp->t_inpcb);
2010 #endif
2011 #ifdef INET6
2012 	if (tp->t_in6pcb)
2013 		mss -= ip6_optlen(tp->t_in6pcb);
2014 #endif
2015 
2016 	/*
2017 	 * If there's a pipesize, change the socket buffer to that size.
2018 	 * Make the socket buffer an integral number of MSS units.  If
2019 	 * the MSS is larger than the socket buffer, artificially decrease
2020 	 * the MSS.
2021 	 */
2022 #ifdef RTV_SPIPE
2023 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
2024 		bufsize = rt->rt_rmx.rmx_sendpipe;
2025 	else
2026 #endif
2027 	{
2028 		KASSERT(so != NULL);
2029 		bufsize = so->so_snd.sb_hiwat;
2030 	}
2031 	if (bufsize < mss)
2032 		mss = bufsize;
2033 	else {
2034 		bufsize = roundup(bufsize, mss);
2035 		if (bufsize > sb_max)
2036 			bufsize = sb_max;
2037 		(void) sbreserve(&so->so_snd, bufsize, so);
2038 	}
2039 	tp->t_segsz = mss;
2040 
2041 #ifdef RTV_SSTHRESH
2042 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2043 		/*
2044 		 * There's some sort of gateway or interface buffer
2045 		 * limit on the path.  Use this to set the slow
2046 		 * start threshold, but set the threshold to no less
2047 		 * than 2 * MSS.
2048 		 */
2049 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2050 	}
2051 #endif
2052 }
2053 
2054 /*
2055  * Processing necessary when a TCP connection is established.
2056  */
2057 void
2058 tcp_established(struct tcpcb *tp)
2059 {
2060 	struct socket *so;
2061 #ifdef RTV_RPIPE
2062 	struct rtentry *rt;
2063 #endif
2064 	u_long bufsize;
2065 
2066 #ifdef DIAGNOSTIC
2067 	if (tp->t_inpcb && tp->t_in6pcb)
2068 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
2069 #endif
2070 	so = NULL;
2071 	rt = NULL;
2072 #ifdef INET
2073 	/* This is a while() to reduce the dreadful stairstepping below */
2074 	while (tp->t_inpcb) {
2075 		so = tp->t_inpcb->inp_socket;
2076 #if defined(RTV_RPIPE)
2077 		rt = in_pcbrtentry(tp->t_inpcb);
2078 #endif
2079 		if (__predict_true(tcp_msl_enable)) {
2080 			if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2081 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2082 				break;
2083 			}
2084 
2085 			if (__predict_false(tcp_rttlocal)) {
2086 				/* This may be adjusted by tcp_input */
2087 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2088 				break;
2089 			}
2090 			if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2091 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2092 				break;
2093 			}
2094 		}
2095 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2096 		break;
2097 	}
2098 #endif
2099 #ifdef INET6
2100 	/* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2101 	while (!tp->t_inpcb && tp->t_in6pcb) {
2102 		so = tp->t_in6pcb->in6p_socket;
2103 #if defined(RTV_RPIPE)
2104 		rt = in6_pcbrtentry(tp->t_in6pcb);
2105 #endif
2106 		if (__predict_true(tcp_msl_enable)) {
2107 			extern const struct in6_addr in6addr_loopback;
2108 
2109 			if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2110 					       &in6addr_loopback)) {
2111 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2112 				break;
2113 			}
2114 
2115 			if (__predict_false(tcp_rttlocal)) {
2116 				/* This may be adjusted by tcp_input */
2117 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2118 				break;
2119 			}
2120 			if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2121 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2122 				break;
2123 			}
2124 		}
2125 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2126 		break;
2127 	}
2128 #endif
2129 
2130 	tp->t_state = TCPS_ESTABLISHED;
2131 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2132 
2133 #ifdef RTV_RPIPE
2134 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2135 		bufsize = rt->rt_rmx.rmx_recvpipe;
2136 	else
2137 #endif
2138 	{
2139 		KASSERT(so != NULL);
2140 		bufsize = so->so_rcv.sb_hiwat;
2141 	}
2142 	if (bufsize > tp->t_ourmss) {
2143 		bufsize = roundup(bufsize, tp->t_ourmss);
2144 		if (bufsize > sb_max)
2145 			bufsize = sb_max;
2146 		(void) sbreserve(&so->so_rcv, bufsize, so);
2147 	}
2148 }
2149 
2150 /*
2151  * Check if there's an initial rtt or rttvar.  Convert from the
2152  * route-table units to scaled multiples of the slow timeout timer.
2153  * Called only during the 3-way handshake.
2154  */
2155 void
2156 tcp_rmx_rtt(struct tcpcb *tp)
2157 {
2158 #ifdef RTV_RTT
2159 	struct rtentry *rt = NULL;
2160 	int rtt;
2161 
2162 #ifdef DIAGNOSTIC
2163 	if (tp->t_inpcb && tp->t_in6pcb)
2164 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2165 #endif
2166 #ifdef INET
2167 	if (tp->t_inpcb)
2168 		rt = in_pcbrtentry(tp->t_inpcb);
2169 #endif
2170 #ifdef INET6
2171 	if (tp->t_in6pcb)
2172 		rt = in6_pcbrtentry(tp->t_in6pcb);
2173 #endif
2174 	if (rt == NULL)
2175 		return;
2176 
2177 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2178 		/*
2179 		 * XXX The lock bit for MTU indicates that the value
2180 		 * is also a minimum value; this is subject to time.
2181 		 */
2182 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2183 			TCPT_RANGESET(tp->t_rttmin,
2184 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2185 			    TCPTV_MIN, TCPTV_REXMTMAX);
2186 		tp->t_srtt = rtt /
2187 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2188 		if (rt->rt_rmx.rmx_rttvar) {
2189 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2190 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2191 				(TCP_RTTVAR_SHIFT + 2));
2192 		} else {
2193 			/* Default variation is +- 1 rtt */
2194 			tp->t_rttvar =
2195 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2196 		}
2197 		TCPT_RANGESET(tp->t_rxtcur,
2198 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2199 		    tp->t_rttmin, TCPTV_REXMTMAX);
2200 	}
2201 #endif
2202 }
2203 
2204 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2205 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2206 
2207 /*
2208  * Get a new sequence value given a tcp control block
2209  */
2210 tcp_seq
2211 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2212 {
2213 
2214 #ifdef INET
2215 	if (tp->t_inpcb != NULL) {
2216 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2217 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2218 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2219 		    addin));
2220 	}
2221 #endif
2222 #ifdef INET6
2223 	if (tp->t_in6pcb != NULL) {
2224 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2225 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2226 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2227 		    addin));
2228 	}
2229 #endif
2230 	/* Not possible. */
2231 	panic("tcp_new_iss");
2232 }
2233 
2234 /*
2235  * This routine actually generates a new TCP initial sequence number.
2236  */
2237 tcp_seq
2238 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2239     size_t addrsz, tcp_seq addin)
2240 {
2241 	tcp_seq tcp_iss;
2242 
2243 	static bool tcp_iss_gotten_secret;
2244 
2245 	/*
2246 	 * If we haven't been here before, initialize our cryptographic
2247 	 * hash secret.
2248 	 */
2249 	if (tcp_iss_gotten_secret == false) {
2250 		cprng_strong(kern_cprng,
2251 			     tcp_iss_secret, sizeof(tcp_iss_secret), 0);
2252 		tcp_iss_gotten_secret = true;
2253 	}
2254 
2255 	if (tcp_do_rfc1948) {
2256 		MD5_CTX ctx;
2257 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2258 
2259 		/*
2260 		 * Compute the base value of the ISS.  It is a hash
2261 		 * of (saddr, sport, daddr, dport, secret).
2262 		 */
2263 		MD5Init(&ctx);
2264 
2265 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2266 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2267 
2268 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2269 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2270 
2271 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2272 
2273 		MD5Final(hash, &ctx);
2274 
2275 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2276 
2277 		/*
2278 		 * Now increment our "timer", and add it in to
2279 		 * the computed value.
2280 		 *
2281 		 * XXX Use `addin'?
2282 		 * XXX TCP_ISSINCR too large to use?
2283 		 */
2284 		tcp_iss_seq += TCP_ISSINCR;
2285 #ifdef TCPISS_DEBUG
2286 		printf("ISS hash 0x%08x, ", tcp_iss);
2287 #endif
2288 		tcp_iss += tcp_iss_seq + addin;
2289 #ifdef TCPISS_DEBUG
2290 		printf("new ISS 0x%08x\n", tcp_iss);
2291 #endif
2292 	} else {
2293 		/*
2294 		 * Randomize.
2295 		 */
2296 		tcp_iss = cprng_fast32();
2297 
2298 		/*
2299 		 * If we were asked to add some amount to a known value,
2300 		 * we will take a random value obtained above, mask off
2301 		 * the upper bits, and add in the known value.  We also
2302 		 * add in a constant to ensure that we are at least a
2303 		 * certain distance from the original value.
2304 		 *
2305 		 * This is used when an old connection is in timed wait
2306 		 * and we have a new one coming in, for instance.
2307 		 */
2308 		if (addin != 0) {
2309 #ifdef TCPISS_DEBUG
2310 			printf("Random %08x, ", tcp_iss);
2311 #endif
2312 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2313 			tcp_iss += addin + TCP_ISSINCR;
2314 #ifdef TCPISS_DEBUG
2315 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2316 #endif
2317 		} else {
2318 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2319 			tcp_iss += tcp_iss_seq;
2320 			tcp_iss_seq += TCP_ISSINCR;
2321 #ifdef TCPISS_DEBUG
2322 			printf("ISS %08x\n", tcp_iss);
2323 #endif
2324 		}
2325 	}
2326 
2327 	if (tcp_compat_42) {
2328 		/*
2329 		 * Limit it to the positive range for really old TCP
2330 		 * implementations.
2331 		 * Just AND off the top bit instead of checking if
2332 		 * is set first - saves a branch 50% of the time.
2333 		 */
2334 		tcp_iss &= 0x7fffffff;		/* XXX */
2335 	}
2336 
2337 	return (tcp_iss);
2338 }
2339 
2340 #if defined(IPSEC)
2341 /* compute ESP/AH header size for TCP, including outer IP header. */
2342 size_t
2343 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2344 {
2345 	struct inpcb *inp;
2346 	size_t hdrsiz;
2347 
2348 	/* XXX mapped addr case (tp->t_in6pcb) */
2349 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2350 		return 0;
2351 	switch (tp->t_family) {
2352 	case AF_INET:
2353 		/* XXX: should use currect direction. */
2354 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2355 		break;
2356 	default:
2357 		hdrsiz = 0;
2358 		break;
2359 	}
2360 
2361 	return hdrsiz;
2362 }
2363 
2364 #ifdef INET6
2365 size_t
2366 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2367 {
2368 	struct in6pcb *in6p;
2369 	size_t hdrsiz;
2370 
2371 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2372 		return 0;
2373 	switch (tp->t_family) {
2374 	case AF_INET6:
2375 		/* XXX: should use currect direction. */
2376 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2377 		break;
2378 	case AF_INET:
2379 		/* mapped address case - tricky */
2380 	default:
2381 		hdrsiz = 0;
2382 		break;
2383 	}
2384 
2385 	return hdrsiz;
2386 }
2387 #endif
2388 #endif /*IPSEC*/
2389 
2390 /*
2391  * Determine the length of the TCP options for this connection.
2392  *
2393  * XXX:  What do we do for SACK, when we add that?  Just reserve
2394  *       all of the space?  Otherwise we can't exactly be incrementing
2395  *       cwnd by an amount that varies depending on the amount we last
2396  *       had to SACK!
2397  */
2398 
2399 u_int
2400 tcp_optlen(struct tcpcb *tp)
2401 {
2402 	u_int optlen;
2403 
2404 	optlen = 0;
2405 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2406 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2407 		optlen += TCPOLEN_TSTAMP_APPA;
2408 
2409 #ifdef TCP_SIGNATURE
2410 	if (tp->t_flags & TF_SIGNATURE)
2411 		optlen += TCPOLEN_SIGNATURE + 2;
2412 #endif /* TCP_SIGNATURE */
2413 
2414 	return optlen;
2415 }
2416 
2417 u_int
2418 tcp_hdrsz(struct tcpcb *tp)
2419 {
2420 	u_int hlen;
2421 
2422 	switch (tp->t_family) {
2423 #ifdef INET6
2424 	case AF_INET6:
2425 		hlen = sizeof(struct ip6_hdr);
2426 		break;
2427 #endif
2428 	case AF_INET:
2429 		hlen = sizeof(struct ip);
2430 		break;
2431 	default:
2432 		hlen = 0;
2433 		break;
2434 	}
2435 	hlen += sizeof(struct tcphdr);
2436 
2437 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2438 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2439 		hlen += TCPOLEN_TSTAMP_APPA;
2440 #ifdef TCP_SIGNATURE
2441 	if (tp->t_flags & TF_SIGNATURE)
2442 		hlen += TCPOLEN_SIGLEN;
2443 #endif
2444 	return hlen;
2445 }
2446 
2447 void
2448 tcp_statinc(u_int stat)
2449 {
2450 
2451 	KASSERT(stat < TCP_NSTATS);
2452 	TCP_STATINC(stat);
2453 }
2454 
2455 void
2456 tcp_statadd(u_int stat, uint64_t val)
2457 {
2458 
2459 	KASSERT(stat < TCP_NSTATS);
2460 	TCP_STATADD(stat, val);
2461 }
2462