xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 6cf6fe02a981b55727c49c3d37b0d8191a98c0ee)
1 /*	$NetBSD: tcp_subr.c,v 1.256 2014/09/05 06:04:43 matt 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 1.256 2014/09/05 06:04:43 matt 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/malloc.h>
107 #include <sys/mbuf.h>
108 #include <sys/once.h>
109 #include <sys/socket.h>
110 #include <sys/socketvar.h>
111 #include <sys/protosw.h>
112 #include <sys/errno.h>
113 #include <sys/kernel.h>
114 #include <sys/pool.h>
115 #include <sys/md5.h>
116 #include <sys/cprng.h>
117 
118 #include <net/route.h>
119 #include <net/if.h>
120 
121 #include <netinet/in.h>
122 #include <netinet/in_systm.h>
123 #include <netinet/ip.h>
124 #include <netinet/in_pcb.h>
125 #include <netinet/ip_var.h>
126 #include <netinet/ip_icmp.h>
127 
128 #ifdef INET6
129 #ifndef INET
130 #include <netinet/in.h>
131 #endif
132 #include <netinet/ip6.h>
133 #include <netinet6/in6_pcb.h>
134 #include <netinet6/ip6_var.h>
135 #include <netinet6/in6_var.h>
136 #include <netinet6/ip6protosw.h>
137 #include <netinet/icmp6.h>
138 #include <netinet6/nd6.h>
139 #endif
140 
141 #include <netinet/tcp.h>
142 #include <netinet/tcp_fsm.h>
143 #include <netinet/tcp_seq.h>
144 #include <netinet/tcp_timer.h>
145 #include <netinet/tcp_var.h>
146 #include <netinet/tcp_vtw.h>
147 #include <netinet/tcp_private.h>
148 #include <netinet/tcp_congctl.h>
149 #include <netinet/tcpip.h>
150 
151 #ifdef IPSEC
152 #include <netipsec/ipsec.h>
153 #include <netipsec/xform.h>
154 #ifdef INET6
155 #include <netipsec/ipsec6.h>
156 #endif
157  #include <netipsec/key.h>
158 #endif	/* IPSEC*/
159 
160 
161 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
162 u_int32_t tcp_now;		/* slow ticks, for RFC 1323 timestamps */
163 
164 percpu_t *tcpstat_percpu;
165 
166 /* patchable/settable parameters for tcp */
167 int 	tcp_mssdflt = TCP_MSS;
168 int	tcp_minmss = TCP_MINMSS;
169 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
170 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
171 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
172 int	tcp_do_sack = 1;	/* selective acknowledgement */
173 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
174 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
175 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
176 int	tcp_do_ecn = 0;		/* Explicit Congestion Notification */
177 #ifndef TCP_INIT_WIN
178 #define	TCP_INIT_WIN	4	/* initial slow start window */
179 #endif
180 #ifndef TCP_INIT_WIN_LOCAL
181 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
182 #endif
183 /*
184  * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460.
185  * This is to simulate current behavior for iw == 4
186  */
187 int tcp_init_win_max[] = {
188 	 1 * 1460,
189 	 1 * 1460,
190 	 2 * 1460,
191 	 2 * 1460,
192 	 3 * 1460,
193 	 5 * 1460,
194 	 6 * 1460,
195 	 7 * 1460,
196 	 8 * 1460,
197 	 9 * 1460,
198 	10 * 1460
199 };
200 int	tcp_init_win = TCP_INIT_WIN;
201 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
202 int	tcp_mss_ifmtu = 0;
203 #ifdef TCP_COMPAT_42
204 int	tcp_compat_42 = 1;
205 #else
206 int	tcp_compat_42 = 0;
207 #endif
208 int	tcp_rst_ppslim = 100;	/* 100pps */
209 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
210 int	tcp_do_loopback_cksum = 0;
211 int	tcp_do_abc = 1;		/* RFC3465 Appropriate byte counting. */
212 int	tcp_abc_aggressive = 1;	/* 1: L=2*SMSS  0: L=1*SMSS */
213 int	tcp_sack_tp_maxholes = 32;
214 int	tcp_sack_globalmaxholes = 1024;
215 int	tcp_sack_globalholes = 0;
216 int	tcp_ecn_maxretries = 1;
217 int	tcp_msl_enable = 1;		/* enable TIME_WAIT truncation	*/
218 int	tcp_msl_loop   = PR_SLOWHZ;	/* MSL for loopback		*/
219 int	tcp_msl_local  = 5 * PR_SLOWHZ;	/* MSL for 'local'		*/
220 int	tcp_msl_remote = TCPTV_MSL;	/* MSL otherwise		*/
221 int	tcp_msl_remote_threshold = TCPTV_SRTTDFLT;	/* RTT threshold */
222 int	tcp_rttlocal = 0;		/* Use RTT to decide who's 'local' */
223 
224 int	tcp4_vtw_enable = 0;		/* 1 to enable */
225 int	tcp6_vtw_enable = 0;		/* 1 to enable */
226 int	tcp_vtw_was_enabled = 0;
227 int	tcp_vtw_entries = 1 << 4;	/* 16 vestigial TIME_WAIT entries */
228 
229 /* tcb hash */
230 #ifndef TCBHASHSIZE
231 #define	TCBHASHSIZE	128
232 #endif
233 int	tcbhashsize = TCBHASHSIZE;
234 
235 /* syn hash parameters */
236 #define	TCP_SYN_HASH_SIZE	293
237 #define	TCP_SYN_BUCKET_SIZE	35
238 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
239 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
240 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
241 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
242 
243 int	tcp_freeq(struct tcpcb *);
244 
245 #ifdef INET
246 static void	tcp_mtudisc_callback(struct in_addr);
247 #endif
248 
249 #ifdef INET6
250 void	tcp6_mtudisc(struct in6pcb *, int);
251 #endif
252 
253 static struct pool tcpcb_pool;
254 
255 static int tcp_drainwanted;
256 
257 #ifdef TCP_CSUM_COUNTERS
258 #include <sys/device.h>
259 
260 #if defined(INET)
261 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "hwcsum bad");
263 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "hwcsum ok");
265 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     NULL, "tcp", "hwcsum data");
267 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     NULL, "tcp", "swcsum");
269 
270 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
271 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
272 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
273 EVCNT_ATTACH_STATIC(tcp_swcsum);
274 #endif /* defined(INET) */
275 
276 #if defined(INET6)
277 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     NULL, "tcp6", "hwcsum bad");
279 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     NULL, "tcp6", "hwcsum ok");
281 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     NULL, "tcp6", "hwcsum data");
283 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     NULL, "tcp6", "swcsum");
285 
286 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
287 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
288 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
289 EVCNT_ATTACH_STATIC(tcp6_swcsum);
290 #endif /* defined(INET6) */
291 #endif /* TCP_CSUM_COUNTERS */
292 
293 
294 #ifdef TCP_OUTPUT_COUNTERS
295 #include <sys/device.h>
296 
297 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
298     NULL, "tcp", "output big header");
299 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
300     NULL, "tcp", "output predict hit");
301 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
302     NULL, "tcp", "output predict miss");
303 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
304     NULL, "tcp", "output copy small");
305 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
306     NULL, "tcp", "output copy big");
307 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
308     NULL, "tcp", "output reference big");
309 
310 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
311 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
312 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
313 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
314 EVCNT_ATTACH_STATIC(tcp_output_copybig);
315 EVCNT_ATTACH_STATIC(tcp_output_refbig);
316 
317 #endif /* TCP_OUTPUT_COUNTERS */
318 
319 #ifdef TCP_REASS_COUNTERS
320 #include <sys/device.h>
321 
322 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
323     NULL, "tcp_reass", "calls");
324 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
325     &tcp_reass_, "tcp_reass", "insert into empty queue");
326 struct evcnt tcp_reass_iteration[8] = {
327     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
329     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
330     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
331     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
332     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
333     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
334     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
335 };
336 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
337     &tcp_reass_, "tcp_reass", "prepend to first");
338 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
339     &tcp_reass_, "tcp_reass", "prepend");
340 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
341     &tcp_reass_, "tcp_reass", "insert");
342 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
343     &tcp_reass_, "tcp_reass", "insert at tail");
344 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
345     &tcp_reass_, "tcp_reass", "append");
346 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
347     &tcp_reass_, "tcp_reass", "append to tail fragment");
348 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
349     &tcp_reass_, "tcp_reass", "overlap at end");
350 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
351     &tcp_reass_, "tcp_reass", "overlap at start");
352 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
353     &tcp_reass_, "tcp_reass", "duplicate segment");
354 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
355     &tcp_reass_, "tcp_reass", "duplicate fragment");
356 
357 EVCNT_ATTACH_STATIC(tcp_reass_);
358 EVCNT_ATTACH_STATIC(tcp_reass_empty);
359 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
363 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
364 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
365 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
366 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
367 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
368 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
369 EVCNT_ATTACH_STATIC(tcp_reass_insert);
370 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
371 EVCNT_ATTACH_STATIC(tcp_reass_append);
372 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
373 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
374 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
375 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
376 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
377 
378 #endif /* TCP_REASS_COUNTERS */
379 
380 #ifdef MBUFTRACE
381 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
382 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
383 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
384 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
385 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
386 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
387 #endif
388 
389 callout_t tcp_slowtimo_ch;
390 
391 static int
392 do_tcpinit(void)
393 {
394 
395 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
396 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
397 	    NULL, IPL_SOFTNET);
398 
399 	tcp_usrreq_init();
400 
401 	/* Initialize timer state. */
402 	tcp_timer_init();
403 
404 	/* Initialize the compressed state engine. */
405 	syn_cache_init();
406 
407 	/* Initialize the congestion control algorithms. */
408 	tcp_congctl_init();
409 
410 	/* Initialize the TCPCB template. */
411 	tcp_tcpcb_template();
412 
413 	/* Initialize reassembly queue */
414 	tcpipqent_init();
415 
416 	/* SACK */
417 	tcp_sack_init();
418 
419 	MOWNER_ATTACH(&tcp_tx_mowner);
420 	MOWNER_ATTACH(&tcp_rx_mowner);
421 	MOWNER_ATTACH(&tcp_reass_mowner);
422 	MOWNER_ATTACH(&tcp_sock_mowner);
423 	MOWNER_ATTACH(&tcp_sock_tx_mowner);
424 	MOWNER_ATTACH(&tcp_sock_rx_mowner);
425 	MOWNER_ATTACH(&tcp_mowner);
426 
427 	tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
428 
429 	vtw_earlyinit();
430 
431 	callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
432 	callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
433 
434 	return 0;
435 }
436 
437 void
438 tcp_init_common(unsigned basehlen)
439 {
440 	static ONCE_DECL(dotcpinit);
441 	unsigned hlen = basehlen + sizeof(struct tcphdr);
442 	unsigned oldhlen;
443 
444 	if (max_linkhdr + hlen > MHLEN)
445 		panic("tcp_init");
446 	while ((oldhlen = max_protohdr) < hlen)
447 		atomic_cas_uint(&max_protohdr, oldhlen, hlen);
448 
449 	RUN_ONCE(&dotcpinit, do_tcpinit);
450 }
451 
452 /*
453  * Tcp initialization
454  */
455 void
456 tcp_init(void)
457 {
458 
459 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
460 
461 	tcp_init_common(sizeof(struct ip));
462 }
463 
464 /*
465  * Create template to be used to send tcp packets on a connection.
466  * Call after host entry created, allocates an mbuf and fills
467  * in a skeletal tcp/ip header, minimizing the amount of work
468  * necessary when the connection is used.
469  */
470 struct mbuf *
471 tcp_template(struct tcpcb *tp)
472 {
473 	struct inpcb *inp = tp->t_inpcb;
474 #ifdef INET6
475 	struct in6pcb *in6p = tp->t_in6pcb;
476 #endif
477 	struct tcphdr *n;
478 	struct mbuf *m;
479 	int hlen;
480 
481 	switch (tp->t_family) {
482 	case AF_INET:
483 		hlen = sizeof(struct ip);
484 		if (inp)
485 			break;
486 #ifdef INET6
487 		if (in6p) {
488 			/* mapped addr case */
489 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
490 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
491 				break;
492 		}
493 #endif
494 		return NULL;	/*EINVAL*/
495 #ifdef INET6
496 	case AF_INET6:
497 		hlen = sizeof(struct ip6_hdr);
498 		if (in6p) {
499 			/* more sainty check? */
500 			break;
501 		}
502 		return NULL;	/*EINVAL*/
503 #endif
504 	default:
505 		hlen = 0;	/*pacify gcc*/
506 		return NULL;	/*EAFNOSUPPORT*/
507 	}
508 #ifdef DIAGNOSTIC
509 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
510 		panic("mclbytes too small for t_template");
511 #endif
512 	m = tp->t_template;
513 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
514 		;
515 	else {
516 		if (m)
517 			m_freem(m);
518 		m = tp->t_template = NULL;
519 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
520 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
521 			MCLGET(m, M_DONTWAIT);
522 			if ((m->m_flags & M_EXT) == 0) {
523 				m_free(m);
524 				m = NULL;
525 			}
526 		}
527 		if (m == NULL)
528 			return NULL;
529 		MCLAIM(m, &tcp_mowner);
530 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
531 	}
532 
533 	memset(mtod(m, void *), 0, m->m_len);
534 
535 	n = (struct tcphdr *)(mtod(m, char *) + hlen);
536 
537 	switch (tp->t_family) {
538 	case AF_INET:
539 	    {
540 		struct ipovly *ipov;
541 		mtod(m, struct ip *)->ip_v = 4;
542 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
543 		ipov = mtod(m, struct ipovly *);
544 		ipov->ih_pr = IPPROTO_TCP;
545 		ipov->ih_len = htons(sizeof(struct tcphdr));
546 		if (inp) {
547 			ipov->ih_src = inp->inp_laddr;
548 			ipov->ih_dst = inp->inp_faddr;
549 		}
550 #ifdef INET6
551 		else if (in6p) {
552 			/* mapped addr case */
553 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
554 				sizeof(ipov->ih_src));
555 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
556 				sizeof(ipov->ih_dst));
557 		}
558 #endif
559 		/*
560 		 * Compute the pseudo-header portion of the checksum
561 		 * now.  We incrementally add in the TCP option and
562 		 * payload lengths later, and then compute the TCP
563 		 * checksum right before the packet is sent off onto
564 		 * the wire.
565 		 */
566 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
567 		    ipov->ih_dst.s_addr,
568 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
569 		break;
570 	    }
571 #ifdef INET6
572 	case AF_INET6:
573 	    {
574 		struct ip6_hdr *ip6;
575 		mtod(m, struct ip *)->ip_v = 6;
576 		ip6 = mtod(m, struct ip6_hdr *);
577 		ip6->ip6_nxt = IPPROTO_TCP;
578 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
579 		ip6->ip6_src = in6p->in6p_laddr;
580 		ip6->ip6_dst = in6p->in6p_faddr;
581 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
582 		if (ip6_auto_flowlabel) {
583 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
584 			ip6->ip6_flow |=
585 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
586 		}
587 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
588 		ip6->ip6_vfc |= IPV6_VERSION;
589 
590 		/*
591 		 * Compute the pseudo-header portion of the checksum
592 		 * now.  We incrementally add in the TCP option and
593 		 * payload lengths later, and then compute the TCP
594 		 * checksum right before the packet is sent off onto
595 		 * the wire.
596 		 */
597 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
598 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
599 		    htonl(IPPROTO_TCP));
600 		break;
601 	    }
602 #endif
603 	}
604 	if (inp) {
605 		n->th_sport = inp->inp_lport;
606 		n->th_dport = inp->inp_fport;
607 	}
608 #ifdef INET6
609 	else if (in6p) {
610 		n->th_sport = in6p->in6p_lport;
611 		n->th_dport = in6p->in6p_fport;
612 	}
613 #endif
614 	n->th_seq = 0;
615 	n->th_ack = 0;
616 	n->th_x2 = 0;
617 	n->th_off = 5;
618 	n->th_flags = 0;
619 	n->th_win = 0;
620 	n->th_urp = 0;
621 	return (m);
622 }
623 
624 /*
625  * Send a single message to the TCP at address specified by
626  * the given TCP/IP header.  If m == 0, then we make a copy
627  * of the tcpiphdr at ti and send directly to the addressed host.
628  * This is used to force keep alive messages out using the TCP
629  * template for a connection tp->t_template.  If flags are given
630  * then we send a message back to the TCP which originated the
631  * segment ti, and discard the mbuf containing it and any other
632  * attached mbufs.
633  *
634  * In any case the ack and sequence number of the transmitted
635  * segment are as specified by the parameters.
636  */
637 int
638 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
639     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
640 {
641 #ifdef INET6
642 	struct rtentry *rt;
643 #endif
644 	struct route *ro;
645 	int error, tlen, win = 0;
646 	int hlen;
647 	struct ip *ip;
648 #ifdef INET6
649 	struct ip6_hdr *ip6;
650 #endif
651 	int family;	/* family on packet, not inpcb/in6pcb! */
652 	struct tcphdr *th;
653 	struct socket *so;
654 
655 	if (tp != NULL && (flags & TH_RST) == 0) {
656 #ifdef DIAGNOSTIC
657 		if (tp->t_inpcb && tp->t_in6pcb)
658 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
659 #endif
660 #ifdef INET
661 		if (tp->t_inpcb)
662 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
663 #endif
664 #ifdef INET6
665 		if (tp->t_in6pcb)
666 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
667 #endif
668 	}
669 
670 	th = NULL;	/* Quell uninitialized warning */
671 	ip = NULL;
672 #ifdef INET6
673 	ip6 = NULL;
674 #endif
675 	if (m == 0) {
676 		if (!mtemplate)
677 			return EINVAL;
678 
679 		/* get family information from template */
680 		switch (mtod(mtemplate, struct ip *)->ip_v) {
681 		case 4:
682 			family = AF_INET;
683 			hlen = sizeof(struct ip);
684 			break;
685 #ifdef INET6
686 		case 6:
687 			family = AF_INET6;
688 			hlen = sizeof(struct ip6_hdr);
689 			break;
690 #endif
691 		default:
692 			return EAFNOSUPPORT;
693 		}
694 
695 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
696 		if (m) {
697 			MCLAIM(m, &tcp_tx_mowner);
698 			MCLGET(m, M_DONTWAIT);
699 			if ((m->m_flags & M_EXT) == 0) {
700 				m_free(m);
701 				m = NULL;
702 			}
703 		}
704 		if (m == NULL)
705 			return (ENOBUFS);
706 
707 		if (tcp_compat_42)
708 			tlen = 1;
709 		else
710 			tlen = 0;
711 
712 		m->m_data += max_linkhdr;
713 		bcopy(mtod(mtemplate, void *), mtod(m, void *),
714 			mtemplate->m_len);
715 		switch (family) {
716 		case AF_INET:
717 			ip = mtod(m, struct ip *);
718 			th = (struct tcphdr *)(ip + 1);
719 			break;
720 #ifdef INET6
721 		case AF_INET6:
722 			ip6 = mtod(m, struct ip6_hdr *);
723 			th = (struct tcphdr *)(ip6 + 1);
724 			break;
725 #endif
726 #if 0
727 		default:
728 			/* noone will visit here */
729 			m_freem(m);
730 			return EAFNOSUPPORT;
731 #endif
732 		}
733 		flags = TH_ACK;
734 	} else {
735 
736 		if ((m->m_flags & M_PKTHDR) == 0) {
737 #if 0
738 			printf("non PKTHDR to tcp_respond\n");
739 #endif
740 			m_freem(m);
741 			return EINVAL;
742 		}
743 #ifdef DIAGNOSTIC
744 		if (!th0)
745 			panic("th0 == NULL in tcp_respond");
746 #endif
747 
748 		/* get family information from m */
749 		switch (mtod(m, struct ip *)->ip_v) {
750 		case 4:
751 			family = AF_INET;
752 			hlen = sizeof(struct ip);
753 			ip = mtod(m, struct ip *);
754 			break;
755 #ifdef INET6
756 		case 6:
757 			family = AF_INET6;
758 			hlen = sizeof(struct ip6_hdr);
759 			ip6 = mtod(m, struct ip6_hdr *);
760 			break;
761 #endif
762 		default:
763 			m_freem(m);
764 			return EAFNOSUPPORT;
765 		}
766 		/* clear h/w csum flags inherited from rx packet */
767 		m->m_pkthdr.csum_flags = 0;
768 
769 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
770 			tlen = sizeof(*th0);
771 		else
772 			tlen = th0->th_off << 2;
773 
774 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
775 		    mtod(m, char *) + hlen == (char *)th0) {
776 			m->m_len = hlen + tlen;
777 			m_freem(m->m_next);
778 			m->m_next = NULL;
779 		} else {
780 			struct mbuf *n;
781 
782 #ifdef DIAGNOSTIC
783 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
784 				m_freem(m);
785 				return EMSGSIZE;
786 			}
787 #endif
788 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
789 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
790 				MCLGET(n, M_DONTWAIT);
791 				if ((n->m_flags & M_EXT) == 0) {
792 					m_freem(n);
793 					n = NULL;
794 				}
795 			}
796 			if (!n) {
797 				m_freem(m);
798 				return ENOBUFS;
799 			}
800 
801 			MCLAIM(n, &tcp_tx_mowner);
802 			n->m_data += max_linkhdr;
803 			n->m_len = hlen + tlen;
804 			m_copyback(n, 0, hlen, mtod(m, void *));
805 			m_copyback(n, hlen, tlen, (void *)th0);
806 
807 			m_freem(m);
808 			m = n;
809 			n = NULL;
810 		}
811 
812 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
813 		switch (family) {
814 		case AF_INET:
815 			ip = mtod(m, struct ip *);
816 			th = (struct tcphdr *)(ip + 1);
817 			ip->ip_p = IPPROTO_TCP;
818 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
819 			ip->ip_p = IPPROTO_TCP;
820 			break;
821 #ifdef INET6
822 		case AF_INET6:
823 			ip6 = mtod(m, struct ip6_hdr *);
824 			th = (struct tcphdr *)(ip6 + 1);
825 			ip6->ip6_nxt = IPPROTO_TCP;
826 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
827 			ip6->ip6_nxt = IPPROTO_TCP;
828 			break;
829 #endif
830 #if 0
831 		default:
832 			/* noone will visit here */
833 			m_freem(m);
834 			return EAFNOSUPPORT;
835 #endif
836 		}
837 		xchg(th->th_dport, th->th_sport, u_int16_t);
838 #undef xchg
839 		tlen = 0;	/*be friendly with the following code*/
840 	}
841 	th->th_seq = htonl(seq);
842 	th->th_ack = htonl(ack);
843 	th->th_x2 = 0;
844 	if ((flags & TH_SYN) == 0) {
845 		if (tp)
846 			win >>= tp->rcv_scale;
847 		if (win > TCP_MAXWIN)
848 			win = TCP_MAXWIN;
849 		th->th_win = htons((u_int16_t)win);
850 		th->th_off = sizeof (struct tcphdr) >> 2;
851 		tlen += sizeof(*th);
852 	} else
853 		tlen += th->th_off << 2;
854 	m->m_len = hlen + tlen;
855 	m->m_pkthdr.len = hlen + tlen;
856 	m->m_pkthdr.rcvif = NULL;
857 	th->th_flags = flags;
858 	th->th_urp = 0;
859 
860 	switch (family) {
861 #ifdef INET
862 	case AF_INET:
863 	    {
864 		struct ipovly *ipov = (struct ipovly *)ip;
865 		memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
866 		ipov->ih_len = htons((u_int16_t)tlen);
867 
868 		th->th_sum = 0;
869 		th->th_sum = in_cksum(m, hlen + tlen);
870 		ip->ip_len = htons(hlen + tlen);
871 		ip->ip_ttl = ip_defttl;
872 		break;
873 	    }
874 #endif
875 #ifdef INET6
876 	case AF_INET6:
877 	    {
878 		th->th_sum = 0;
879 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
880 				tlen);
881 		ip6->ip6_plen = htons(tlen);
882 		if (tp && tp->t_in6pcb) {
883 			struct ifnet *oifp;
884 			ro = &tp->t_in6pcb->in6p_route;
885 			oifp = (rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
886 			                                           : NULL;
887 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
888 		} else
889 			ip6->ip6_hlim = ip6_defhlim;
890 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
891 		if (ip6_auto_flowlabel) {
892 			ip6->ip6_flow |=
893 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
894 		}
895 		break;
896 	    }
897 #endif
898 	}
899 
900 	if (tp && tp->t_inpcb)
901 		so = tp->t_inpcb->inp_socket;
902 #ifdef INET6
903 	else if (tp && tp->t_in6pcb)
904 		so = tp->t_in6pcb->in6p_socket;
905 #endif
906 	else
907 		so = NULL;
908 
909 	if (tp != NULL && tp->t_inpcb != NULL) {
910 		ro = &tp->t_inpcb->inp_route;
911 #ifdef DIAGNOSTIC
912 		if (family != AF_INET)
913 			panic("tcp_respond: address family mismatch");
914 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
915 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
916 			    ntohl(ip->ip_dst.s_addr),
917 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
918 		}
919 #endif
920 	}
921 #ifdef INET6
922 	else if (tp != NULL && tp->t_in6pcb != NULL) {
923 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
924 #ifdef DIAGNOSTIC
925 		if (family == AF_INET) {
926 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
927 				panic("tcp_respond: not mapped addr");
928 			if (memcmp(&ip->ip_dst,
929 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
930 			    sizeof(ip->ip_dst)) != 0) {
931 				panic("tcp_respond: ip_dst != in6p_faddr");
932 			}
933 		} else if (family == AF_INET6) {
934 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
935 			    &tp->t_in6pcb->in6p_faddr))
936 				panic("tcp_respond: ip6_dst != in6p_faddr");
937 		} else
938 			panic("tcp_respond: address family mismatch");
939 #endif
940 	}
941 #endif
942 	else
943 		ro = NULL;
944 
945 	switch (family) {
946 #ifdef INET
947 	case AF_INET:
948 		error = ip_output(m, NULL, ro,
949 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so);
950 		break;
951 #endif
952 #ifdef INET6
953 	case AF_INET6:
954 		error = ip6_output(m, NULL, ro, 0, NULL, so, NULL);
955 		break;
956 #endif
957 	default:
958 		error = EAFNOSUPPORT;
959 		break;
960 	}
961 
962 	return (error);
963 }
964 
965 /*
966  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
967  * a bunch of members individually, we maintain this template for the
968  * static and mostly-static components of the TCPCB, and copy it into
969  * the new TCPCB instead.
970  */
971 static struct tcpcb tcpcb_template = {
972 	.t_srtt = TCPTV_SRTTBASE,
973 	.t_rttmin = TCPTV_MIN,
974 
975 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
976 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
977 	.snd_numholes = 0,
978 	.snd_cubic_wmax = 0,
979 	.snd_cubic_wmax_last = 0,
980 	.snd_cubic_ctime = 0,
981 
982 	.t_partialacks = -1,
983 	.t_bytes_acked = 0,
984 };
985 
986 /*
987  * Updates the TCPCB template whenever a parameter that would affect
988  * the template is changed.
989  */
990 void
991 tcp_tcpcb_template(void)
992 {
993 	struct tcpcb *tp = &tcpcb_template;
994 	int flags;
995 
996 	tp->t_peermss = tcp_mssdflt;
997 	tp->t_ourmss = tcp_mssdflt;
998 	tp->t_segsz = tcp_mssdflt;
999 
1000 	flags = 0;
1001 	if (tcp_do_rfc1323 && tcp_do_win_scale)
1002 		flags |= TF_REQ_SCALE;
1003 	if (tcp_do_rfc1323 && tcp_do_timestamps)
1004 		flags |= TF_REQ_TSTMP;
1005 	tp->t_flags = flags;
1006 
1007 	/*
1008 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1009 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
1010 	 * reasonable initial retransmit time.
1011 	 */
1012 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
1013 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1014 	    TCPTV_MIN, TCPTV_REXMTMAX);
1015 
1016 	/* Keep Alive */
1017 	tp->t_keepinit = tcp_keepinit;
1018 	tp->t_keepidle = tcp_keepidle;
1019 	tp->t_keepintvl = tcp_keepintvl;
1020 	tp->t_keepcnt = tcp_keepcnt;
1021 	tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1022 
1023 	/* MSL */
1024 	tp->t_msl = TCPTV_MSL;
1025 }
1026 
1027 /*
1028  * Create a new TCP control block, making an
1029  * empty reassembly queue and hooking it to the argument
1030  * protocol control block.
1031  */
1032 /* family selects inpcb, or in6pcb */
1033 struct tcpcb *
1034 tcp_newtcpcb(int family, void *aux)
1035 {
1036 #ifdef INET6
1037 	struct rtentry *rt;
1038 #endif
1039 	struct tcpcb *tp;
1040 	int i;
1041 
1042 	/* XXX Consider using a pool_cache for speed. */
1043 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);	/* splsoftnet via tcp_usrreq */
1044 	if (tp == NULL)
1045 		return (NULL);
1046 	memcpy(tp, &tcpcb_template, sizeof(*tp));
1047 	TAILQ_INIT(&tp->segq);
1048 	TAILQ_INIT(&tp->timeq);
1049 	tp->t_family = family;		/* may be overridden later on */
1050 	TAILQ_INIT(&tp->snd_holes);
1051 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
1052 
1053 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
1054 	for (i = 0; i < TCPT_NTIMERS; i++) {
1055 		callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1056 		TCP_TIMER_INIT(tp, i);
1057 	}
1058 	callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1059 
1060 	switch (family) {
1061 	case AF_INET:
1062 	    {
1063 		struct inpcb *inp = (struct inpcb *)aux;
1064 
1065 		inp->inp_ip.ip_ttl = ip_defttl;
1066 		inp->inp_ppcb = (void *)tp;
1067 
1068 		tp->t_inpcb = inp;
1069 		tp->t_mtudisc = ip_mtudisc;
1070 		break;
1071 	    }
1072 #ifdef INET6
1073 	case AF_INET6:
1074 	    {
1075 		struct in6pcb *in6p = (struct in6pcb *)aux;
1076 
1077 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
1078 			(rt = rtcache_validate(&in6p->in6p_route)) != NULL
1079 			    ? rt->rt_ifp
1080 			    : NULL);
1081 		in6p->in6p_ppcb = (void *)tp;
1082 
1083 		tp->t_in6pcb = in6p;
1084 		/* for IPv6, always try to run path MTU discovery */
1085 		tp->t_mtudisc = 1;
1086 		break;
1087 	    }
1088 #endif /* INET6 */
1089 	default:
1090 		for (i = 0; i < TCPT_NTIMERS; i++)
1091 			callout_destroy(&tp->t_timer[i]);
1092 		callout_destroy(&tp->t_delack_ch);
1093 		pool_put(&tcpcb_pool, tp);	/* splsoftnet via tcp_usrreq */
1094 		return (NULL);
1095 	}
1096 
1097 	/*
1098 	 * Initialize our timebase.  When we send timestamps, we take
1099 	 * the delta from tcp_now -- this means each connection always
1100 	 * gets a timebase of 1, which makes it, among other things,
1101 	 * more difficult to determine how long a system has been up,
1102 	 * and thus how many TCP sequence increments have occurred.
1103 	 *
1104 	 * We start with 1, because 0 doesn't work with linux, which
1105 	 * considers timestamp 0 in a SYN packet as a bug and disables
1106 	 * timestamps.
1107 	 */
1108 	tp->ts_timebase = tcp_now - 1;
1109 
1110 	tcp_congctl_select(tp, tcp_congctl_global_name);
1111 
1112 	return (tp);
1113 }
1114 
1115 /*
1116  * Drop a TCP connection, reporting
1117  * the specified error.  If connection is synchronized,
1118  * then send a RST to peer.
1119  */
1120 struct tcpcb *
1121 tcp_drop(struct tcpcb *tp, int errno)
1122 {
1123 	struct socket *so = NULL;
1124 
1125 #ifdef DIAGNOSTIC
1126 	if (tp->t_inpcb && tp->t_in6pcb)
1127 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1128 #endif
1129 #ifdef INET
1130 	if (tp->t_inpcb)
1131 		so = tp->t_inpcb->inp_socket;
1132 #endif
1133 #ifdef INET6
1134 	if (tp->t_in6pcb)
1135 		so = tp->t_in6pcb->in6p_socket;
1136 #endif
1137 	if (!so)
1138 		return NULL;
1139 
1140 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1141 		tp->t_state = TCPS_CLOSED;
1142 		(void) tcp_output(tp);
1143 		TCP_STATINC(TCP_STAT_DROPS);
1144 	} else
1145 		TCP_STATINC(TCP_STAT_CONNDROPS);
1146 	if (errno == ETIMEDOUT && tp->t_softerror)
1147 		errno = tp->t_softerror;
1148 	so->so_error = errno;
1149 	return (tcp_close(tp));
1150 }
1151 
1152 /*
1153  * Close a TCP control block:
1154  *	discard all space held by the tcp
1155  *	discard internet protocol block
1156  *	wake up any sleepers
1157  */
1158 struct tcpcb *
1159 tcp_close(struct tcpcb *tp)
1160 {
1161 	struct inpcb *inp;
1162 #ifdef INET6
1163 	struct in6pcb *in6p;
1164 #endif
1165 	struct socket *so;
1166 #ifdef RTV_RTT
1167 	struct rtentry *rt;
1168 #endif
1169 	struct route *ro;
1170 	int j;
1171 
1172 	inp = tp->t_inpcb;
1173 #ifdef INET6
1174 	in6p = tp->t_in6pcb;
1175 #endif
1176 	so = NULL;
1177 	ro = NULL;
1178 	if (inp) {
1179 		so = inp->inp_socket;
1180 		ro = &inp->inp_route;
1181 	}
1182 #ifdef INET6
1183 	else if (in6p) {
1184 		so = in6p->in6p_socket;
1185 		ro = (struct route *)&in6p->in6p_route;
1186 	}
1187 #endif
1188 
1189 #ifdef RTV_RTT
1190 	/*
1191 	 * If we sent enough data to get some meaningful characteristics,
1192 	 * save them in the routing entry.  'Enough' is arbitrarily
1193 	 * defined as the sendpipesize (default 4K) * 16.  This would
1194 	 * give us 16 rtt samples assuming we only get one sample per
1195 	 * window (the usual case on a long haul net).  16 samples is
1196 	 * enough for the srtt filter to converge to within 5% of the correct
1197 	 * value; fewer samples and we could save a very bogus rtt.
1198 	 *
1199 	 * Don't update the default route's characteristics and don't
1200 	 * update anything that the user "locked".
1201 	 */
1202 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1203 	    ro && (rt = rtcache_validate(ro)) != NULL &&
1204 	    !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1205 		u_long i = 0;
1206 
1207 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1208 			i = tp->t_srtt *
1209 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1210 			if (rt->rt_rmx.rmx_rtt && i)
1211 				/*
1212 				 * filter this update to half the old & half
1213 				 * the new values, converting scale.
1214 				 * See route.h and tcp_var.h for a
1215 				 * description of the scaling constants.
1216 				 */
1217 				rt->rt_rmx.rmx_rtt =
1218 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1219 			else
1220 				rt->rt_rmx.rmx_rtt = i;
1221 		}
1222 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1223 			i = tp->t_rttvar *
1224 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1225 			if (rt->rt_rmx.rmx_rttvar && i)
1226 				rt->rt_rmx.rmx_rttvar =
1227 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1228 			else
1229 				rt->rt_rmx.rmx_rttvar = i;
1230 		}
1231 		/*
1232 		 * update the pipelimit (ssthresh) if it has been updated
1233 		 * already or if a pipesize was specified & the threshhold
1234 		 * got below half the pipesize.  I.e., wait for bad news
1235 		 * before we start updating, then update on both good
1236 		 * and bad news.
1237 		 */
1238 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1239 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1240 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1241 			/*
1242 			 * convert the limit from user data bytes to
1243 			 * packets then to packet data bytes.
1244 			 */
1245 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1246 			if (i < 2)
1247 				i = 2;
1248 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1249 			if (rt->rt_rmx.rmx_ssthresh)
1250 				rt->rt_rmx.rmx_ssthresh =
1251 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1252 			else
1253 				rt->rt_rmx.rmx_ssthresh = i;
1254 		}
1255 	}
1256 #endif /* RTV_RTT */
1257 	/* free the reassembly queue, if any */
1258 	TCP_REASS_LOCK(tp);
1259 	(void) tcp_freeq(tp);
1260 	TCP_REASS_UNLOCK(tp);
1261 
1262 	/* free the SACK holes list. */
1263 	tcp_free_sackholes(tp);
1264 	tcp_congctl_release(tp);
1265 	syn_cache_cleanup(tp);
1266 
1267 	if (tp->t_template) {
1268 		m_free(tp->t_template);
1269 		tp->t_template = NULL;
1270 	}
1271 
1272 	/*
1273 	 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1274 	 * the timers can also drop the lock.  We need to prevent access
1275 	 * to the tcpcb as it's half torn down.  Flag the pcb as dead
1276 	 * (prevents access by timers) and only then detach it.
1277 	 */
1278 	tp->t_flags |= TF_DEAD;
1279 	if (inp) {
1280 		inp->inp_ppcb = 0;
1281 		soisdisconnected(so);
1282 		in_pcbdetach(inp);
1283 	}
1284 #ifdef INET6
1285 	else if (in6p) {
1286 		in6p->in6p_ppcb = 0;
1287 		soisdisconnected(so);
1288 		in6_pcbdetach(in6p);
1289 	}
1290 #endif
1291 	/*
1292 	 * pcb is no longer visble elsewhere, so we can safely release
1293 	 * the lock in callout_halt() if needed.
1294 	 */
1295 	TCP_STATINC(TCP_STAT_CLOSED);
1296 	for (j = 0; j < TCPT_NTIMERS; j++) {
1297 		callout_halt(&tp->t_timer[j], softnet_lock);
1298 		callout_destroy(&tp->t_timer[j]);
1299 	}
1300 	callout_halt(&tp->t_delack_ch, softnet_lock);
1301 	callout_destroy(&tp->t_delack_ch);
1302 	pool_put(&tcpcb_pool, tp);
1303 
1304 	return NULL;
1305 }
1306 
1307 int
1308 tcp_freeq(struct tcpcb *tp)
1309 {
1310 	struct ipqent *qe;
1311 	int rv = 0;
1312 #ifdef TCPREASS_DEBUG
1313 	int i = 0;
1314 #endif
1315 
1316 	TCP_REASS_LOCK_CHECK(tp);
1317 
1318 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1319 #ifdef TCPREASS_DEBUG
1320 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1321 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1322 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1323 #endif
1324 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1325 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1326 		m_freem(qe->ipqe_m);
1327 		tcpipqent_free(qe);
1328 		rv = 1;
1329 	}
1330 	tp->t_segqlen = 0;
1331 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1332 	return (rv);
1333 }
1334 
1335 void
1336 tcp_fasttimo(void)
1337 {
1338 	if (tcp_drainwanted) {
1339 		tcp_drain();
1340 		tcp_drainwanted = 0;
1341 	}
1342 }
1343 
1344 void
1345 tcp_drainstub(void)
1346 {
1347 	tcp_drainwanted = 1;
1348 }
1349 
1350 /*
1351  * Protocol drain routine.  Called when memory is in short supply.
1352  * Called from pr_fasttimo thus a callout context.
1353  */
1354 void
1355 tcp_drain(void)
1356 {
1357 	struct inpcb_hdr *inph;
1358 	struct tcpcb *tp;
1359 
1360 	mutex_enter(softnet_lock);
1361 	KERNEL_LOCK(1, NULL);
1362 
1363 	/*
1364 	 * Free the sequence queue of all TCP connections.
1365 	 */
1366 	TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1367 		switch (inph->inph_af) {
1368 		case AF_INET:
1369 			tp = intotcpcb((struct inpcb *)inph);
1370 			break;
1371 #ifdef INET6
1372 		case AF_INET6:
1373 			tp = in6totcpcb((struct in6pcb *)inph);
1374 			break;
1375 #endif
1376 		default:
1377 			tp = NULL;
1378 			break;
1379 		}
1380 		if (tp != NULL) {
1381 			/*
1382 			 * We may be called from a device's interrupt
1383 			 * context.  If the tcpcb is already busy,
1384 			 * just bail out now.
1385 			 */
1386 			if (tcp_reass_lock_try(tp) == 0)
1387 				continue;
1388 			if (tcp_freeq(tp))
1389 				TCP_STATINC(TCP_STAT_CONNSDRAINED);
1390 			TCP_REASS_UNLOCK(tp);
1391 		}
1392 	}
1393 
1394 	KERNEL_UNLOCK_ONE(NULL);
1395 	mutex_exit(softnet_lock);
1396 }
1397 
1398 /*
1399  * Notify a tcp user of an asynchronous error;
1400  * store error as soft error, but wake up user
1401  * (for now, won't do anything until can select for soft error).
1402  */
1403 void
1404 tcp_notify(struct inpcb *inp, int error)
1405 {
1406 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1407 	struct socket *so = inp->inp_socket;
1408 
1409 	/*
1410 	 * Ignore some errors if we are hooked up.
1411 	 * If connection hasn't completed, has retransmitted several times,
1412 	 * and receives a second error, give up now.  This is better
1413 	 * than waiting a long time to establish a connection that
1414 	 * can never complete.
1415 	 */
1416 	if (tp->t_state == TCPS_ESTABLISHED &&
1417 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1418 	      error == EHOSTDOWN)) {
1419 		return;
1420 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1421 	    tp->t_rxtshift > 3 && tp->t_softerror)
1422 		so->so_error = error;
1423 	else
1424 		tp->t_softerror = error;
1425 	cv_broadcast(&so->so_cv);
1426 	sorwakeup(so);
1427 	sowwakeup(so);
1428 }
1429 
1430 #ifdef INET6
1431 void
1432 tcp6_notify(struct in6pcb *in6p, int error)
1433 {
1434 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1435 	struct socket *so = in6p->in6p_socket;
1436 
1437 	/*
1438 	 * Ignore some errors if we are hooked up.
1439 	 * If connection hasn't completed, has retransmitted several times,
1440 	 * and receives a second error, give up now.  This is better
1441 	 * than waiting a long time to establish a connection that
1442 	 * can never complete.
1443 	 */
1444 	if (tp->t_state == TCPS_ESTABLISHED &&
1445 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1446 	      error == EHOSTDOWN)) {
1447 		return;
1448 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1449 	    tp->t_rxtshift > 3 && tp->t_softerror)
1450 		so->so_error = error;
1451 	else
1452 		tp->t_softerror = error;
1453 	cv_broadcast(&so->so_cv);
1454 	sorwakeup(so);
1455 	sowwakeup(so);
1456 }
1457 #endif
1458 
1459 #ifdef INET6
1460 void *
1461 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1462 {
1463 	struct tcphdr th;
1464 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1465 	int nmatch;
1466 	struct ip6_hdr *ip6;
1467 	const struct sockaddr_in6 *sa6_src = NULL;
1468 	const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1469 	struct mbuf *m;
1470 	int off;
1471 
1472 	if (sa->sa_family != AF_INET6 ||
1473 	    sa->sa_len != sizeof(struct sockaddr_in6))
1474 		return NULL;
1475 	if ((unsigned)cmd >= PRC_NCMDS)
1476 		return NULL;
1477 	else if (cmd == PRC_QUENCH) {
1478 		/*
1479 		 * Don't honor ICMP Source Quench messages meant for
1480 		 * TCP connections.
1481 		 */
1482 		return NULL;
1483 	} else if (PRC_IS_REDIRECT(cmd))
1484 		notify = in6_rtchange, d = NULL;
1485 	else if (cmd == PRC_MSGSIZE)
1486 		; /* special code is present, see below */
1487 	else if (cmd == PRC_HOSTDEAD)
1488 		d = NULL;
1489 	else if (inet6ctlerrmap[cmd] == 0)
1490 		return NULL;
1491 
1492 	/* if the parameter is from icmp6, decode it. */
1493 	if (d != NULL) {
1494 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1495 		m = ip6cp->ip6c_m;
1496 		ip6 = ip6cp->ip6c_ip6;
1497 		off = ip6cp->ip6c_off;
1498 		sa6_src = ip6cp->ip6c_src;
1499 	} else {
1500 		m = NULL;
1501 		ip6 = NULL;
1502 		sa6_src = &sa6_any;
1503 		off = 0;
1504 	}
1505 
1506 	if (ip6) {
1507 		/*
1508 		 * XXX: We assume that when ip6 is non NULL,
1509 		 * M and OFF are valid.
1510 		 */
1511 
1512 		/* check if we can safely examine src and dst ports */
1513 		if (m->m_pkthdr.len < off + sizeof(th)) {
1514 			if (cmd == PRC_MSGSIZE)
1515 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1516 			return NULL;
1517 		}
1518 
1519 		memset(&th, 0, sizeof(th));
1520 		m_copydata(m, off, sizeof(th), (void *)&th);
1521 
1522 		if (cmd == PRC_MSGSIZE) {
1523 			int valid = 0;
1524 
1525 			/*
1526 			 * Check to see if we have a valid TCP connection
1527 			 * corresponding to the address in the ICMPv6 message
1528 			 * payload.
1529 			 */
1530 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1531 			    th.th_dport,
1532 			    (const struct in6_addr *)&sa6_src->sin6_addr,
1533 						  th.th_sport, 0, 0))
1534 				valid++;
1535 
1536 			/*
1537 			 * Depending on the value of "valid" and routing table
1538 			 * size (mtudisc_{hi,lo}wat), we will:
1539 			 * - recalcurate the new MTU and create the
1540 			 *   corresponding routing entry, or
1541 			 * - ignore the MTU change notification.
1542 			 */
1543 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1544 
1545 			/*
1546 			 * no need to call in6_pcbnotify, it should have been
1547 			 * called via callback if necessary
1548 			 */
1549 			return NULL;
1550 		}
1551 
1552 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1553 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1554 		if (nmatch == 0 && syn_cache_count &&
1555 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1556 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1557 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1558 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1559 					  sa, &th);
1560 	} else {
1561 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1562 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1563 	}
1564 
1565 	return NULL;
1566 }
1567 #endif
1568 
1569 #ifdef INET
1570 /* assumes that ip header and tcp header are contiguous on mbuf */
1571 void *
1572 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1573 {
1574 	struct ip *ip = v;
1575 	struct tcphdr *th;
1576 	struct icmp *icp;
1577 	extern const int inetctlerrmap[];
1578 	void (*notify)(struct inpcb *, int) = tcp_notify;
1579 	int errno;
1580 	int nmatch;
1581 	struct tcpcb *tp;
1582 	u_int mtu;
1583 	tcp_seq seq;
1584 	struct inpcb *inp;
1585 #ifdef INET6
1586 	struct in6pcb *in6p;
1587 	struct in6_addr src6, dst6;
1588 #endif
1589 
1590 	if (sa->sa_family != AF_INET ||
1591 	    sa->sa_len != sizeof(struct sockaddr_in))
1592 		return NULL;
1593 	if ((unsigned)cmd >= PRC_NCMDS)
1594 		return NULL;
1595 	errno = inetctlerrmap[cmd];
1596 	if (cmd == PRC_QUENCH)
1597 		/*
1598 		 * Don't honor ICMP Source Quench messages meant for
1599 		 * TCP connections.
1600 		 */
1601 		return NULL;
1602 	else if (PRC_IS_REDIRECT(cmd))
1603 		notify = in_rtchange, ip = 0;
1604 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1605 		/*
1606 		 * Check to see if we have a valid TCP connection
1607 		 * corresponding to the address in the ICMP message
1608 		 * payload.
1609 		 *
1610 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1611 		 */
1612 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1613 #ifdef INET6
1614 		memset(&src6, 0, sizeof(src6));
1615 		memset(&dst6, 0, sizeof(dst6));
1616 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1617 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1618 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1619 #endif
1620 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1621 						th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1622 #ifdef INET6
1623 			in6p = NULL;
1624 #else
1625 			;
1626 #endif
1627 #ifdef INET6
1628 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1629 						       th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1630 			;
1631 #endif
1632 		else
1633 			return NULL;
1634 
1635 		/*
1636 		 * Now that we've validated that we are actually communicating
1637 		 * with the host indicated in the ICMP message, locate the
1638 		 * ICMP header, recalculate the new MTU, and create the
1639 		 * corresponding routing entry.
1640 		 */
1641 		icp = (struct icmp *)((char *)ip -
1642 		    offsetof(struct icmp, icmp_ip));
1643 		if (inp) {
1644 			if ((tp = intotcpcb(inp)) == NULL)
1645 				return NULL;
1646 		}
1647 #ifdef INET6
1648 		else if (in6p) {
1649 			if ((tp = in6totcpcb(in6p)) == NULL)
1650 				return NULL;
1651 		}
1652 #endif
1653 		else
1654 			return NULL;
1655 		seq = ntohl(th->th_seq);
1656 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1657 			return NULL;
1658 		/*
1659 		 * If the ICMP message advertises a Next-Hop MTU
1660 		 * equal or larger than the maximum packet size we have
1661 		 * ever sent, drop the message.
1662 		 */
1663 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1664 		if (mtu >= tp->t_pmtud_mtu_sent)
1665 			return NULL;
1666 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1667 			/*
1668 			 * Calculate new MTU, and create corresponding
1669 			 * route (traditional PMTUD).
1670 			 */
1671 			tp->t_flags &= ~TF_PMTUD_PEND;
1672 			icmp_mtudisc(icp, ip->ip_dst);
1673 		} else {
1674 			/*
1675 			 * Record the information got in the ICMP
1676 			 * message; act on it later.
1677 			 * If we had already recorded an ICMP message,
1678 			 * replace the old one only if the new message
1679 			 * refers to an older TCP segment
1680 			 */
1681 			if (tp->t_flags & TF_PMTUD_PEND) {
1682 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1683 					return NULL;
1684 			} else
1685 				tp->t_flags |= TF_PMTUD_PEND;
1686 			tp->t_pmtud_th_seq = seq;
1687 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1688 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1689 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1690 		}
1691 		return NULL;
1692 	} else if (cmd == PRC_HOSTDEAD)
1693 		ip = 0;
1694 	else if (errno == 0)
1695 		return NULL;
1696 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1697 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1698 		nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1699 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1700 		if (nmatch == 0 && syn_cache_count &&
1701 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1702 		    inetctlerrmap[cmd] == ENETUNREACH ||
1703 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1704 			struct sockaddr_in sin;
1705 			memset(&sin, 0, sizeof(sin));
1706 			sin.sin_len = sizeof(sin);
1707 			sin.sin_family = AF_INET;
1708 			sin.sin_port = th->th_sport;
1709 			sin.sin_addr = ip->ip_src;
1710 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1711 		}
1712 
1713 		/* XXX mapped address case */
1714 	} else
1715 		in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1716 		    notify);
1717 	return NULL;
1718 }
1719 
1720 /*
1721  * When a source quench is received, we are being notified of congestion.
1722  * Close the congestion window down to the Loss Window (one segment).
1723  * We will gradually open it again as we proceed.
1724  */
1725 void
1726 tcp_quench(struct inpcb *inp, int errno)
1727 {
1728 	struct tcpcb *tp = intotcpcb(inp);
1729 
1730 	if (tp) {
1731 		tp->snd_cwnd = tp->t_segsz;
1732 		tp->t_bytes_acked = 0;
1733 	}
1734 }
1735 #endif
1736 
1737 #ifdef INET6
1738 void
1739 tcp6_quench(struct in6pcb *in6p, int errno)
1740 {
1741 	struct tcpcb *tp = in6totcpcb(in6p);
1742 
1743 	if (tp) {
1744 		tp->snd_cwnd = tp->t_segsz;
1745 		tp->t_bytes_acked = 0;
1746 	}
1747 }
1748 #endif
1749 
1750 #ifdef INET
1751 /*
1752  * Path MTU Discovery handlers.
1753  */
1754 void
1755 tcp_mtudisc_callback(struct in_addr faddr)
1756 {
1757 #ifdef INET6
1758 	struct in6_addr in6;
1759 #endif
1760 
1761 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1762 #ifdef INET6
1763 	memset(&in6, 0, sizeof(in6));
1764 	in6.s6_addr16[5] = 0xffff;
1765 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1766 	tcp6_mtudisc_callback(&in6);
1767 #endif
1768 }
1769 
1770 /*
1771  * On receipt of path MTU corrections, flush old route and replace it
1772  * with the new one.  Retransmit all unacknowledged packets, to ensure
1773  * that all packets will be received.
1774  */
1775 void
1776 tcp_mtudisc(struct inpcb *inp, int errno)
1777 {
1778 	struct tcpcb *tp = intotcpcb(inp);
1779 	struct rtentry *rt = in_pcbrtentry(inp);
1780 
1781 	if (tp != 0) {
1782 		if (rt != 0) {
1783 			/*
1784 			 * If this was not a host route, remove and realloc.
1785 			 */
1786 			if ((rt->rt_flags & RTF_HOST) == 0) {
1787 				in_rtchange(inp, errno);
1788 				if ((rt = in_pcbrtentry(inp)) == 0)
1789 					return;
1790 			}
1791 
1792 			/*
1793 			 * Slow start out of the error condition.  We
1794 			 * use the MTU because we know it's smaller
1795 			 * than the previously transmitted segment.
1796 			 *
1797 			 * Note: This is more conservative than the
1798 			 * suggestion in draft-floyd-incr-init-win-03.
1799 			 */
1800 			if (rt->rt_rmx.rmx_mtu != 0)
1801 				tp->snd_cwnd =
1802 				    TCP_INITIAL_WINDOW(tcp_init_win,
1803 				    rt->rt_rmx.rmx_mtu);
1804 		}
1805 
1806 		/*
1807 		 * Resend unacknowledged packets.
1808 		 */
1809 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1810 		tcp_output(tp);
1811 	}
1812 }
1813 #endif
1814 
1815 #ifdef INET6
1816 /*
1817  * Path MTU Discovery handlers.
1818  */
1819 void
1820 tcp6_mtudisc_callback(struct in6_addr *faddr)
1821 {
1822 	struct sockaddr_in6 sin6;
1823 
1824 	memset(&sin6, 0, sizeof(sin6));
1825 	sin6.sin6_family = AF_INET6;
1826 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1827 	sin6.sin6_addr = *faddr;
1828 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1829 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1830 }
1831 
1832 void
1833 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1834 {
1835 	struct tcpcb *tp = in6totcpcb(in6p);
1836 	struct rtentry *rt = in6_pcbrtentry(in6p);
1837 
1838 	if (tp != 0) {
1839 		if (rt != 0) {
1840 			/*
1841 			 * If this was not a host route, remove and realloc.
1842 			 */
1843 			if ((rt->rt_flags & RTF_HOST) == 0) {
1844 				in6_rtchange(in6p, errno);
1845 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1846 					return;
1847 			}
1848 
1849 			/*
1850 			 * Slow start out of the error condition.  We
1851 			 * use the MTU because we know it's smaller
1852 			 * than the previously transmitted segment.
1853 			 *
1854 			 * Note: This is more conservative than the
1855 			 * suggestion in draft-floyd-incr-init-win-03.
1856 			 */
1857 			if (rt->rt_rmx.rmx_mtu != 0)
1858 				tp->snd_cwnd =
1859 				    TCP_INITIAL_WINDOW(tcp_init_win,
1860 				    rt->rt_rmx.rmx_mtu);
1861 		}
1862 
1863 		/*
1864 		 * Resend unacknowledged packets.
1865 		 */
1866 		tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1867 		tcp_output(tp);
1868 	}
1869 }
1870 #endif /* INET6 */
1871 
1872 /*
1873  * Compute the MSS to advertise to the peer.  Called only during
1874  * the 3-way handshake.  If we are the server (peer initiated
1875  * connection), we are called with a pointer to the interface
1876  * on which the SYN packet arrived.  If we are the client (we
1877  * initiated connection), we are called with a pointer to the
1878  * interface out which this connection should go.
1879  *
1880  * NOTE: Do not subtract IP option/extension header size nor IPsec
1881  * header size from MSS advertisement.  MSS option must hold the maximum
1882  * segment size we can accept, so it must always be:
1883  *	 max(if mtu) - ip header - tcp header
1884  */
1885 u_long
1886 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1887 {
1888 	extern u_long in_maxmtu;
1889 	u_long mss = 0;
1890 	u_long hdrsiz;
1891 
1892 	/*
1893 	 * In order to avoid defeating path MTU discovery on the peer,
1894 	 * we advertise the max MTU of all attached networks as our MSS,
1895 	 * per RFC 1191, section 3.1.
1896 	 *
1897 	 * We provide the option to advertise just the MTU of
1898 	 * the interface on which we hope this connection will
1899 	 * be receiving.  If we are responding to a SYN, we
1900 	 * will have a pretty good idea about this, but when
1901 	 * initiating a connection there is a bit more doubt.
1902 	 *
1903 	 * We also need to ensure that loopback has a large enough
1904 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1905 	 */
1906 
1907 	if (ifp != NULL)
1908 		switch (af) {
1909 		case AF_INET:
1910 			mss = ifp->if_mtu;
1911 			break;
1912 #ifdef INET6
1913 		case AF_INET6:
1914 			mss = IN6_LINKMTU(ifp);
1915 			break;
1916 #endif
1917 		}
1918 
1919 	if (tcp_mss_ifmtu == 0)
1920 		switch (af) {
1921 		case AF_INET:
1922 			mss = max(in_maxmtu, mss);
1923 			break;
1924 #ifdef INET6
1925 		case AF_INET6:
1926 			mss = max(in6_maxmtu, mss);
1927 			break;
1928 #endif
1929 		}
1930 
1931 	switch (af) {
1932 	case AF_INET:
1933 		hdrsiz = sizeof(struct ip);
1934 		break;
1935 #ifdef INET6
1936 	case AF_INET6:
1937 		hdrsiz = sizeof(struct ip6_hdr);
1938 		break;
1939 #endif
1940 	default:
1941 		hdrsiz = 0;
1942 		break;
1943 	}
1944 	hdrsiz += sizeof(struct tcphdr);
1945 	if (mss > hdrsiz)
1946 		mss -= hdrsiz;
1947 
1948 	mss = max(tcp_mssdflt, mss);
1949 	return (mss);
1950 }
1951 
1952 /*
1953  * Set connection variables based on the peer's advertised MSS.
1954  * We are passed the TCPCB for the actual connection.  If we
1955  * are the server, we are called by the compressed state engine
1956  * when the 3-way handshake is complete.  If we are the client,
1957  * we are called when we receive the SYN,ACK from the server.
1958  *
1959  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1960  * before this routine is called!
1961  */
1962 void
1963 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1964 {
1965 	struct socket *so;
1966 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1967 	struct rtentry *rt;
1968 #endif
1969 	u_long bufsize;
1970 	int mss;
1971 
1972 #ifdef DIAGNOSTIC
1973 	if (tp->t_inpcb && tp->t_in6pcb)
1974 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1975 #endif
1976 	so = NULL;
1977 	rt = NULL;
1978 #ifdef INET
1979 	if (tp->t_inpcb) {
1980 		so = tp->t_inpcb->inp_socket;
1981 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1982 		rt = in_pcbrtentry(tp->t_inpcb);
1983 #endif
1984 	}
1985 #endif
1986 #ifdef INET6
1987 	if (tp->t_in6pcb) {
1988 		so = tp->t_in6pcb->in6p_socket;
1989 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1990 		rt = in6_pcbrtentry(tp->t_in6pcb);
1991 #endif
1992 	}
1993 #endif
1994 
1995 	/*
1996 	 * As per RFC1122, use the default MSS value, unless they
1997 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1998 	 */
1999 	mss = tcp_mssdflt;
2000 	if (offer)
2001 		mss = offer;
2002 	mss = max(mss, 256);		/* sanity */
2003 	tp->t_peermss = mss;
2004 	mss -= tcp_optlen(tp);
2005 #ifdef INET
2006 	if (tp->t_inpcb)
2007 		mss -= ip_optlen(tp->t_inpcb);
2008 #endif
2009 #ifdef INET6
2010 	if (tp->t_in6pcb)
2011 		mss -= ip6_optlen(tp->t_in6pcb);
2012 #endif
2013 
2014 	/*
2015 	 * If there's a pipesize, change the socket buffer to that size.
2016 	 * Make the socket buffer an integral number of MSS units.  If
2017 	 * the MSS is larger than the socket buffer, artificially decrease
2018 	 * the MSS.
2019 	 */
2020 #ifdef RTV_SPIPE
2021 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
2022 		bufsize = rt->rt_rmx.rmx_sendpipe;
2023 	else
2024 #endif
2025 	{
2026 		KASSERT(so != NULL);
2027 		bufsize = so->so_snd.sb_hiwat;
2028 	}
2029 	if (bufsize < mss)
2030 		mss = bufsize;
2031 	else {
2032 		bufsize = roundup(bufsize, mss);
2033 		if (bufsize > sb_max)
2034 			bufsize = sb_max;
2035 		(void) sbreserve(&so->so_snd, bufsize, so);
2036 	}
2037 	tp->t_segsz = mss;
2038 
2039 #ifdef RTV_SSTHRESH
2040 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2041 		/*
2042 		 * There's some sort of gateway or interface buffer
2043 		 * limit on the path.  Use this to set the slow
2044 		 * start threshold, but set the threshold to no less
2045 		 * than 2 * MSS.
2046 		 */
2047 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2048 	}
2049 #endif
2050 }
2051 
2052 /*
2053  * Processing necessary when a TCP connection is established.
2054  */
2055 void
2056 tcp_established(struct tcpcb *tp)
2057 {
2058 	struct socket *so;
2059 #ifdef RTV_RPIPE
2060 	struct rtentry *rt;
2061 #endif
2062 	u_long bufsize;
2063 
2064 #ifdef DIAGNOSTIC
2065 	if (tp->t_inpcb && tp->t_in6pcb)
2066 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
2067 #endif
2068 	so = NULL;
2069 	rt = NULL;
2070 #ifdef INET
2071 	/* This is a while() to reduce the dreadful stairstepping below */
2072 	while (tp->t_inpcb) {
2073 		so = tp->t_inpcb->inp_socket;
2074 #if defined(RTV_RPIPE)
2075 		rt = in_pcbrtentry(tp->t_inpcb);
2076 #endif
2077 		if (__predict_true(tcp_msl_enable)) {
2078 			if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2079 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2080 				break;
2081 			}
2082 
2083 			if (__predict_false(tcp_rttlocal)) {
2084 				/* This may be adjusted by tcp_input */
2085 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2086 				break;
2087 			}
2088 			if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2089 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2090 				break;
2091 			}
2092 		}
2093 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2094 		break;
2095 	}
2096 #endif
2097 #ifdef INET6
2098 	/* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2099 	while (!tp->t_inpcb && tp->t_in6pcb) {
2100 		so = tp->t_in6pcb->in6p_socket;
2101 #if defined(RTV_RPIPE)
2102 		rt = in6_pcbrtentry(tp->t_in6pcb);
2103 #endif
2104 		if (__predict_true(tcp_msl_enable)) {
2105 			extern const struct in6_addr in6addr_loopback;
2106 
2107 			if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2108 					       &in6addr_loopback)) {
2109 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2110 				break;
2111 			}
2112 
2113 			if (__predict_false(tcp_rttlocal)) {
2114 				/* This may be adjusted by tcp_input */
2115 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2116 				break;
2117 			}
2118 			if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2119 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2120 				break;
2121 			}
2122 		}
2123 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2124 		break;
2125 	}
2126 #endif
2127 
2128 	tp->t_state = TCPS_ESTABLISHED;
2129 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2130 
2131 #ifdef RTV_RPIPE
2132 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2133 		bufsize = rt->rt_rmx.rmx_recvpipe;
2134 	else
2135 #endif
2136 	{
2137 		KASSERT(so != NULL);
2138 		bufsize = so->so_rcv.sb_hiwat;
2139 	}
2140 	if (bufsize > tp->t_ourmss) {
2141 		bufsize = roundup(bufsize, tp->t_ourmss);
2142 		if (bufsize > sb_max)
2143 			bufsize = sb_max;
2144 		(void) sbreserve(&so->so_rcv, bufsize, so);
2145 	}
2146 }
2147 
2148 /*
2149  * Check if there's an initial rtt or rttvar.  Convert from the
2150  * route-table units to scaled multiples of the slow timeout timer.
2151  * Called only during the 3-way handshake.
2152  */
2153 void
2154 tcp_rmx_rtt(struct tcpcb *tp)
2155 {
2156 #ifdef RTV_RTT
2157 	struct rtentry *rt = NULL;
2158 	int rtt;
2159 
2160 #ifdef DIAGNOSTIC
2161 	if (tp->t_inpcb && tp->t_in6pcb)
2162 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2163 #endif
2164 #ifdef INET
2165 	if (tp->t_inpcb)
2166 		rt = in_pcbrtentry(tp->t_inpcb);
2167 #endif
2168 #ifdef INET6
2169 	if (tp->t_in6pcb)
2170 		rt = in6_pcbrtentry(tp->t_in6pcb);
2171 #endif
2172 	if (rt == NULL)
2173 		return;
2174 
2175 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2176 		/*
2177 		 * XXX The lock bit for MTU indicates that the value
2178 		 * is also a minimum value; this is subject to time.
2179 		 */
2180 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2181 			TCPT_RANGESET(tp->t_rttmin,
2182 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2183 			    TCPTV_MIN, TCPTV_REXMTMAX);
2184 		tp->t_srtt = rtt /
2185 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2186 		if (rt->rt_rmx.rmx_rttvar) {
2187 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2188 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2189 				(TCP_RTTVAR_SHIFT + 2));
2190 		} else {
2191 			/* Default variation is +- 1 rtt */
2192 			tp->t_rttvar =
2193 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2194 		}
2195 		TCPT_RANGESET(tp->t_rxtcur,
2196 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2197 		    tp->t_rttmin, TCPTV_REXMTMAX);
2198 	}
2199 #endif
2200 }
2201 
2202 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2203 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2204 
2205 /*
2206  * Get a new sequence value given a tcp control block
2207  */
2208 tcp_seq
2209 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2210 {
2211 
2212 #ifdef INET
2213 	if (tp->t_inpcb != NULL) {
2214 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2215 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2216 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2217 		    addin));
2218 	}
2219 #endif
2220 #ifdef INET6
2221 	if (tp->t_in6pcb != NULL) {
2222 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2223 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2224 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2225 		    addin));
2226 	}
2227 #endif
2228 	/* Not possible. */
2229 	panic("tcp_new_iss");
2230 }
2231 
2232 /*
2233  * This routine actually generates a new TCP initial sequence number.
2234  */
2235 tcp_seq
2236 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2237     size_t addrsz, tcp_seq addin)
2238 {
2239 	tcp_seq tcp_iss;
2240 
2241 	static bool tcp_iss_gotten_secret;
2242 
2243 	/*
2244 	 * If we haven't been here before, initialize our cryptographic
2245 	 * hash secret.
2246 	 */
2247 	if (tcp_iss_gotten_secret == false) {
2248 		cprng_strong(kern_cprng,
2249 			     tcp_iss_secret, sizeof(tcp_iss_secret), FASYNC);
2250 		tcp_iss_gotten_secret = true;
2251 	}
2252 
2253 	if (tcp_do_rfc1948) {
2254 		MD5_CTX ctx;
2255 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2256 
2257 		/*
2258 		 * Compute the base value of the ISS.  It is a hash
2259 		 * of (saddr, sport, daddr, dport, secret).
2260 		 */
2261 		MD5Init(&ctx);
2262 
2263 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2264 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2265 
2266 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2267 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2268 
2269 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2270 
2271 		MD5Final(hash, &ctx);
2272 
2273 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2274 
2275 		/*
2276 		 * Now increment our "timer", and add it in to
2277 		 * the computed value.
2278 		 *
2279 		 * XXX Use `addin'?
2280 		 * XXX TCP_ISSINCR too large to use?
2281 		 */
2282 		tcp_iss_seq += TCP_ISSINCR;
2283 #ifdef TCPISS_DEBUG
2284 		printf("ISS hash 0x%08x, ", tcp_iss);
2285 #endif
2286 		tcp_iss += tcp_iss_seq + addin;
2287 #ifdef TCPISS_DEBUG
2288 		printf("new ISS 0x%08x\n", tcp_iss);
2289 #endif
2290 	} else {
2291 		/*
2292 		 * Randomize.
2293 		 */
2294 		tcp_iss = cprng_fast32();
2295 
2296 		/*
2297 		 * If we were asked to add some amount to a known value,
2298 		 * we will take a random value obtained above, mask off
2299 		 * the upper bits, and add in the known value.  We also
2300 		 * add in a constant to ensure that we are at least a
2301 		 * certain distance from the original value.
2302 		 *
2303 		 * This is used when an old connection is in timed wait
2304 		 * and we have a new one coming in, for instance.
2305 		 */
2306 		if (addin != 0) {
2307 #ifdef TCPISS_DEBUG
2308 			printf("Random %08x, ", tcp_iss);
2309 #endif
2310 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2311 			tcp_iss += addin + TCP_ISSINCR;
2312 #ifdef TCPISS_DEBUG
2313 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2314 #endif
2315 		} else {
2316 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2317 			tcp_iss += tcp_iss_seq;
2318 			tcp_iss_seq += TCP_ISSINCR;
2319 #ifdef TCPISS_DEBUG
2320 			printf("ISS %08x\n", tcp_iss);
2321 #endif
2322 		}
2323 	}
2324 
2325 	if (tcp_compat_42) {
2326 		/*
2327 		 * Limit it to the positive range for really old TCP
2328 		 * implementations.
2329 		 * Just AND off the top bit instead of checking if
2330 		 * is set first - saves a branch 50% of the time.
2331 		 */
2332 		tcp_iss &= 0x7fffffff;		/* XXX */
2333 	}
2334 
2335 	return (tcp_iss);
2336 }
2337 
2338 #if defined(IPSEC)
2339 /* compute ESP/AH header size for TCP, including outer IP header. */
2340 size_t
2341 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2342 {
2343 	struct inpcb *inp;
2344 	size_t hdrsiz;
2345 
2346 	/* XXX mapped addr case (tp->t_in6pcb) */
2347 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2348 		return 0;
2349 	switch (tp->t_family) {
2350 	case AF_INET:
2351 		/* XXX: should use currect direction. */
2352 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2353 		break;
2354 	default:
2355 		hdrsiz = 0;
2356 		break;
2357 	}
2358 
2359 	return hdrsiz;
2360 }
2361 
2362 #ifdef INET6
2363 size_t
2364 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2365 {
2366 	struct in6pcb *in6p;
2367 	size_t hdrsiz;
2368 
2369 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2370 		return 0;
2371 	switch (tp->t_family) {
2372 	case AF_INET6:
2373 		/* XXX: should use currect direction. */
2374 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2375 		break;
2376 	case AF_INET:
2377 		/* mapped address case - tricky */
2378 	default:
2379 		hdrsiz = 0;
2380 		break;
2381 	}
2382 
2383 	return hdrsiz;
2384 }
2385 #endif
2386 #endif /*IPSEC*/
2387 
2388 /*
2389  * Determine the length of the TCP options for this connection.
2390  *
2391  * XXX:  What do we do for SACK, when we add that?  Just reserve
2392  *       all of the space?  Otherwise we can't exactly be incrementing
2393  *       cwnd by an amount that varies depending on the amount we last
2394  *       had to SACK!
2395  */
2396 
2397 u_int
2398 tcp_optlen(struct tcpcb *tp)
2399 {
2400 	u_int optlen;
2401 
2402 	optlen = 0;
2403 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2404 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2405 		optlen += TCPOLEN_TSTAMP_APPA;
2406 
2407 #ifdef TCP_SIGNATURE
2408 	if (tp->t_flags & TF_SIGNATURE)
2409 		optlen += TCPOLEN_SIGNATURE + 2;
2410 #endif /* TCP_SIGNATURE */
2411 
2412 	return optlen;
2413 }
2414 
2415 u_int
2416 tcp_hdrsz(struct tcpcb *tp)
2417 {
2418 	u_int hlen;
2419 
2420 	switch (tp->t_family) {
2421 #ifdef INET6
2422 	case AF_INET6:
2423 		hlen = sizeof(struct ip6_hdr);
2424 		break;
2425 #endif
2426 	case AF_INET:
2427 		hlen = sizeof(struct ip);
2428 		break;
2429 	default:
2430 		hlen = 0;
2431 		break;
2432 	}
2433 	hlen += sizeof(struct tcphdr);
2434 
2435 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2436 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2437 		hlen += TCPOLEN_TSTAMP_APPA;
2438 #ifdef TCP_SIGNATURE
2439 	if (tp->t_flags & TF_SIGNATURE)
2440 		hlen += TCPOLEN_SIGLEN;
2441 #endif
2442 	return hlen;
2443 }
2444 
2445 void
2446 tcp_statinc(u_int stat)
2447 {
2448 
2449 	KASSERT(stat < TCP_NSTATS);
2450 	TCP_STATINC(stat);
2451 }
2452 
2453 void
2454 tcp_statadd(u_int stat, uint64_t val)
2455 {
2456 
2457 	KASSERT(stat < TCP_NSTATS);
2458 	TCP_STATADD(stat, val);
2459 }
2460