xref: /netbsd-src/sys/netinet/tcp_subr.c (revision c9496f6b604074a9451a67df576a5b423068e71e)
1 /*	$NetBSD: tcp_subr.c,v 1.272 2018/01/19 07:53:01 ozaki-r 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 1.272 2018/01/19 07:53:01 ozaki-r Exp $");
95 
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_inet_csum.h"
100 #include "opt_mbuftrace.h"
101 #endif
102 
103 #include <sys/param.h>
104 #include <sys/atomic.h>
105 #include <sys/proc.h>
106 #include <sys/systm.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 int	tcp_rst_ppslim = 100;	/* 100pps */
204 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
205 int	tcp_do_loopback_cksum = 0;
206 int	tcp_do_abc = 1;		/* RFC3465 Appropriate byte counting. */
207 int	tcp_abc_aggressive = 1;	/* 1: L=2*SMSS  0: L=1*SMSS */
208 int	tcp_sack_tp_maxholes = 32;
209 int	tcp_sack_globalmaxholes = 1024;
210 int	tcp_sack_globalholes = 0;
211 int	tcp_ecn_maxretries = 1;
212 int	tcp_msl_enable = 1;		/* enable TIME_WAIT truncation	*/
213 int	tcp_msl_loop   = PR_SLOWHZ;	/* MSL for loopback		*/
214 int	tcp_msl_local  = 5 * PR_SLOWHZ;	/* MSL for 'local'		*/
215 int	tcp_msl_remote = TCPTV_MSL;	/* MSL otherwise		*/
216 int	tcp_msl_remote_threshold = TCPTV_SRTTDFLT;	/* RTT threshold */
217 int	tcp_rttlocal = 0;		/* Use RTT to decide who's 'local' */
218 
219 int	tcp4_vtw_enable = 0;		/* 1 to enable */
220 int	tcp6_vtw_enable = 0;		/* 1 to enable */
221 int	tcp_vtw_was_enabled = 0;
222 int	tcp_vtw_entries = 1 << 4;	/* 16 vestigial TIME_WAIT entries */
223 
224 /* tcb hash */
225 #ifndef TCBHASHSIZE
226 #define	TCBHASHSIZE	128
227 #endif
228 int	tcbhashsize = TCBHASHSIZE;
229 
230 /* syn hash parameters */
231 #define	TCP_SYN_HASH_SIZE	293
232 #define	TCP_SYN_BUCKET_SIZE	35
233 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
234 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
235 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
236 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
237 
238 int	tcp_freeq(struct tcpcb *);
239 static int	tcp_iss_secret_init(void);
240 
241 #ifdef INET
242 static void	tcp_mtudisc_callback(struct in_addr);
243 #endif
244 
245 #ifdef INET6
246 void	tcp6_mtudisc(struct in6pcb *, int);
247 #endif
248 
249 static struct pool tcpcb_pool;
250 
251 static int tcp_drainwanted;
252 
253 #ifdef TCP_CSUM_COUNTERS
254 #include <sys/device.h>
255 
256 #if defined(INET)
257 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
258     NULL, "tcp", "hwcsum bad");
259 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
260     NULL, "tcp", "hwcsum ok");
261 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "hwcsum data");
263 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "swcsum");
265 
266 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
267 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
268 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
269 EVCNT_ATTACH_STATIC(tcp_swcsum);
270 #endif /* defined(INET) */
271 
272 #if defined(INET6)
273 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
274     NULL, "tcp6", "hwcsum bad");
275 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
276     NULL, "tcp6", "hwcsum ok");
277 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278     NULL, "tcp6", "hwcsum data");
279 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280     NULL, "tcp6", "swcsum");
281 
282 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
283 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
284 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
285 EVCNT_ATTACH_STATIC(tcp6_swcsum);
286 #endif /* defined(INET6) */
287 #endif /* TCP_CSUM_COUNTERS */
288 
289 
290 #ifdef TCP_OUTPUT_COUNTERS
291 #include <sys/device.h>
292 
293 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
294     NULL, "tcp", "output big header");
295 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
296     NULL, "tcp", "output predict hit");
297 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
298     NULL, "tcp", "output predict miss");
299 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
300     NULL, "tcp", "output copy small");
301 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
302     NULL, "tcp", "output copy big");
303 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
304     NULL, "tcp", "output reference big");
305 
306 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
307 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
308 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
309 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
310 EVCNT_ATTACH_STATIC(tcp_output_copybig);
311 EVCNT_ATTACH_STATIC(tcp_output_refbig);
312 
313 #endif /* TCP_OUTPUT_COUNTERS */
314 
315 #ifdef TCP_REASS_COUNTERS
316 #include <sys/device.h>
317 
318 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
319     NULL, "tcp_reass", "calls");
320 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
321     &tcp_reass_, "tcp_reass", "insert into empty queue");
322 struct evcnt tcp_reass_iteration[8] = {
323     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
324     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
325     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
326     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
327     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
328     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
329     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
330     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
331 };
332 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
333     &tcp_reass_, "tcp_reass", "prepend to first");
334 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
335     &tcp_reass_, "tcp_reass", "prepend");
336 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
337     &tcp_reass_, "tcp_reass", "insert");
338 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
339     &tcp_reass_, "tcp_reass", "insert at tail");
340 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
341     &tcp_reass_, "tcp_reass", "append");
342 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
343     &tcp_reass_, "tcp_reass", "append to tail fragment");
344 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
345     &tcp_reass_, "tcp_reass", "overlap at end");
346 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
347     &tcp_reass_, "tcp_reass", "overlap at start");
348 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
349     &tcp_reass_, "tcp_reass", "duplicate segment");
350 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
351     &tcp_reass_, "tcp_reass", "duplicate fragment");
352 
353 EVCNT_ATTACH_STATIC(tcp_reass_);
354 EVCNT_ATTACH_STATIC(tcp_reass_empty);
355 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
356 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
357 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
358 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
359 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
360 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
361 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
362 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
363 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
364 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
365 EVCNT_ATTACH_STATIC(tcp_reass_insert);
366 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
367 EVCNT_ATTACH_STATIC(tcp_reass_append);
368 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
369 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
370 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
371 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
372 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
373 
374 #endif /* TCP_REASS_COUNTERS */
375 
376 #ifdef MBUFTRACE
377 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
378 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
379 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
380 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
381 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
382 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
383 #endif
384 
385 static int
386 do_tcpinit(void)
387 {
388 
389 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
390 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
391 	    NULL, IPL_SOFTNET);
392 
393 	tcp_usrreq_init();
394 
395 	/* Initialize timer state. */
396 	tcp_timer_init();
397 
398 	/* Initialize the compressed state engine. */
399 	syn_cache_init();
400 
401 	/* Initialize the congestion control algorithms. */
402 	tcp_congctl_init();
403 
404 	/* Initialize the TCPCB template. */
405 	tcp_tcpcb_template();
406 
407 	/* Initialize reassembly queue */
408 	tcpipqent_init();
409 
410 	/* SACK */
411 	tcp_sack_init();
412 
413 	MOWNER_ATTACH(&tcp_tx_mowner);
414 	MOWNER_ATTACH(&tcp_rx_mowner);
415 	MOWNER_ATTACH(&tcp_reass_mowner);
416 	MOWNER_ATTACH(&tcp_sock_mowner);
417 	MOWNER_ATTACH(&tcp_sock_tx_mowner);
418 	MOWNER_ATTACH(&tcp_sock_rx_mowner);
419 	MOWNER_ATTACH(&tcp_mowner);
420 
421 	tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
422 
423 	vtw_earlyinit();
424 
425 	tcp_slowtimo_init();
426 
427 	return 0;
428 }
429 
430 void
431 tcp_init_common(unsigned basehlen)
432 {
433 	static ONCE_DECL(dotcpinit);
434 	unsigned hlen = basehlen + sizeof(struct tcphdr);
435 	unsigned oldhlen;
436 
437 	if (max_linkhdr + hlen > MHLEN)
438 		panic("tcp_init");
439 	while ((oldhlen = max_protohdr) < hlen)
440 		atomic_cas_uint(&max_protohdr, oldhlen, hlen);
441 
442 	RUN_ONCE(&dotcpinit, do_tcpinit);
443 }
444 
445 /*
446  * Tcp initialization
447  */
448 void
449 tcp_init(void)
450 {
451 
452 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
453 
454 	tcp_init_common(sizeof(struct ip));
455 }
456 
457 /*
458  * Create template to be used to send tcp packets on a connection.
459  * Call after host entry created, allocates an mbuf and fills
460  * in a skeletal tcp/ip header, minimizing the amount of work
461  * necessary when the connection is used.
462  */
463 struct mbuf *
464 tcp_template(struct tcpcb *tp)
465 {
466 	struct inpcb *inp = tp->t_inpcb;
467 #ifdef INET6
468 	struct in6pcb *in6p = tp->t_in6pcb;
469 #endif
470 	struct tcphdr *n;
471 	struct mbuf *m;
472 	int hlen;
473 
474 	switch (tp->t_family) {
475 	case AF_INET:
476 		hlen = sizeof(struct ip);
477 		if (inp)
478 			break;
479 #ifdef INET6
480 		if (in6p) {
481 			/* mapped addr case */
482 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
483 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
484 				break;
485 		}
486 #endif
487 		return NULL;	/*EINVAL*/
488 #ifdef INET6
489 	case AF_INET6:
490 		hlen = sizeof(struct ip6_hdr);
491 		if (in6p) {
492 			/* more sainty check? */
493 			break;
494 		}
495 		return NULL;	/*EINVAL*/
496 #endif
497 	default:
498 		hlen = 0;	/*pacify gcc*/
499 		return NULL;	/*EAFNOSUPPORT*/
500 	}
501 #ifdef DIAGNOSTIC
502 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
503 		panic("mclbytes too small for t_template");
504 #endif
505 	m = tp->t_template;
506 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
507 		;
508 	else {
509 		if (m)
510 			m_freem(m);
511 		m = tp->t_template = NULL;
512 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
513 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
514 			MCLGET(m, M_DONTWAIT);
515 			if ((m->m_flags & M_EXT) == 0) {
516 				m_free(m);
517 				m = NULL;
518 			}
519 		}
520 		if (m == NULL)
521 			return NULL;
522 		MCLAIM(m, &tcp_mowner);
523 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
524 	}
525 
526 	memset(mtod(m, void *), 0, m->m_len);
527 
528 	n = (struct tcphdr *)(mtod(m, char *) + hlen);
529 
530 	switch (tp->t_family) {
531 	case AF_INET:
532 	    {
533 		struct ipovly *ipov;
534 		mtod(m, struct ip *)->ip_v = 4;
535 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
536 		ipov = mtod(m, struct ipovly *);
537 		ipov->ih_pr = IPPROTO_TCP;
538 		ipov->ih_len = htons(sizeof(struct tcphdr));
539 		if (inp) {
540 			ipov->ih_src = inp->inp_laddr;
541 			ipov->ih_dst = inp->inp_faddr;
542 		}
543 #ifdef INET6
544 		else if (in6p) {
545 			/* mapped addr case */
546 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
547 				sizeof(ipov->ih_src));
548 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
549 				sizeof(ipov->ih_dst));
550 		}
551 #endif
552 		/*
553 		 * Compute the pseudo-header portion of the checksum
554 		 * now.  We incrementally add in the TCP option and
555 		 * payload lengths later, and then compute the TCP
556 		 * checksum right before the packet is sent off onto
557 		 * the wire.
558 		 */
559 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
560 		    ipov->ih_dst.s_addr,
561 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
562 		break;
563 	    }
564 #ifdef INET6
565 	case AF_INET6:
566 	    {
567 		struct ip6_hdr *ip6;
568 		mtod(m, struct ip *)->ip_v = 6;
569 		ip6 = mtod(m, struct ip6_hdr *);
570 		ip6->ip6_nxt = IPPROTO_TCP;
571 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
572 		ip6->ip6_src = in6p->in6p_laddr;
573 		ip6->ip6_dst = in6p->in6p_faddr;
574 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
575 		if (ip6_auto_flowlabel) {
576 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
577 			ip6->ip6_flow |=
578 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
579 		}
580 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
581 		ip6->ip6_vfc |= IPV6_VERSION;
582 
583 		/*
584 		 * Compute the pseudo-header portion of the checksum
585 		 * now.  We incrementally add in the TCP option and
586 		 * payload lengths later, and then compute the TCP
587 		 * checksum right before the packet is sent off onto
588 		 * the wire.
589 		 */
590 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
591 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
592 		    htonl(IPPROTO_TCP));
593 		break;
594 	    }
595 #endif
596 	}
597 	if (inp) {
598 		n->th_sport = inp->inp_lport;
599 		n->th_dport = inp->inp_fport;
600 	}
601 #ifdef INET6
602 	else if (in6p) {
603 		n->th_sport = in6p->in6p_lport;
604 		n->th_dport = in6p->in6p_fport;
605 	}
606 #endif
607 	n->th_seq = 0;
608 	n->th_ack = 0;
609 	n->th_x2 = 0;
610 	n->th_off = 5;
611 	n->th_flags = 0;
612 	n->th_win = 0;
613 	n->th_urp = 0;
614 	return (m);
615 }
616 
617 /*
618  * Send a single message to the TCP at address specified by
619  * the given TCP/IP header.  If m == 0, then we make a copy
620  * of the tcpiphdr at ti and send directly to the addressed host.
621  * This is used to force keep alive messages out using the TCP
622  * template for a connection tp->t_template.  If flags are given
623  * then we send a message back to the TCP which originated the
624  * segment ti, and discard the mbuf containing it and any other
625  * attached mbufs.
626  *
627  * In any case the ack and sequence number of the transmitted
628  * segment are as specified by the parameters.
629  */
630 int
631 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
632     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
633 {
634 	struct route *ro;
635 	int error, tlen, win = 0;
636 	int hlen;
637 	struct ip *ip;
638 #ifdef INET6
639 	struct ip6_hdr *ip6;
640 #endif
641 	int family;	/* family on packet, not inpcb/in6pcb! */
642 	struct tcphdr *th;
643 
644 	if (tp != NULL && (flags & TH_RST) == 0) {
645 #ifdef DIAGNOSTIC
646 		if (tp->t_inpcb && tp->t_in6pcb)
647 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
648 #endif
649 #ifdef INET
650 		if (tp->t_inpcb)
651 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
652 #endif
653 #ifdef INET6
654 		if (tp->t_in6pcb)
655 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
656 #endif
657 	}
658 
659 	th = NULL;	/* Quell uninitialized warning */
660 	ip = NULL;
661 #ifdef INET6
662 	ip6 = NULL;
663 #endif
664 	if (m == 0) {
665 		if (!mtemplate)
666 			return EINVAL;
667 
668 		/* get family information from template */
669 		switch (mtod(mtemplate, struct ip *)->ip_v) {
670 		case 4:
671 			family = AF_INET;
672 			hlen = sizeof(struct ip);
673 			break;
674 #ifdef INET6
675 		case 6:
676 			family = AF_INET6;
677 			hlen = sizeof(struct ip6_hdr);
678 			break;
679 #endif
680 		default:
681 			return EAFNOSUPPORT;
682 		}
683 
684 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
685 		if (m) {
686 			MCLAIM(m, &tcp_tx_mowner);
687 			MCLGET(m, M_DONTWAIT);
688 			if ((m->m_flags & M_EXT) == 0) {
689 				m_free(m);
690 				m = NULL;
691 			}
692 		}
693 		if (m == NULL)
694 			return (ENOBUFS);
695 
696 		tlen = 0;
697 
698 		m->m_data += max_linkhdr;
699 		bcopy(mtod(mtemplate, void *), mtod(m, void *),
700 			mtemplate->m_len);
701 		switch (family) {
702 		case AF_INET:
703 			ip = mtod(m, struct ip *);
704 			th = (struct tcphdr *)(ip + 1);
705 			break;
706 #ifdef INET6
707 		case AF_INET6:
708 			ip6 = mtod(m, struct ip6_hdr *);
709 			th = (struct tcphdr *)(ip6 + 1);
710 			break;
711 #endif
712 #if 0
713 		default:
714 			/* noone will visit here */
715 			m_freem(m);
716 			return EAFNOSUPPORT;
717 #endif
718 		}
719 		flags = TH_ACK;
720 	} else {
721 
722 		if ((m->m_flags & M_PKTHDR) == 0) {
723 #if 0
724 			printf("non PKTHDR to tcp_respond\n");
725 #endif
726 			m_freem(m);
727 			return EINVAL;
728 		}
729 #ifdef DIAGNOSTIC
730 		if (!th0)
731 			panic("th0 == NULL in tcp_respond");
732 #endif
733 
734 		/* get family information from m */
735 		switch (mtod(m, struct ip *)->ip_v) {
736 		case 4:
737 			family = AF_INET;
738 			hlen = sizeof(struct ip);
739 			ip = mtod(m, struct ip *);
740 			break;
741 #ifdef INET6
742 		case 6:
743 			family = AF_INET6;
744 			hlen = sizeof(struct ip6_hdr);
745 			ip6 = mtod(m, struct ip6_hdr *);
746 			break;
747 #endif
748 		default:
749 			m_freem(m);
750 			return EAFNOSUPPORT;
751 		}
752 		/* clear h/w csum flags inherited from rx packet */
753 		m->m_pkthdr.csum_flags = 0;
754 
755 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
756 			tlen = sizeof(*th0);
757 		else
758 			tlen = th0->th_off << 2;
759 
760 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
761 		    mtod(m, char *) + hlen == (char *)th0) {
762 			m->m_len = hlen + tlen;
763 			m_freem(m->m_next);
764 			m->m_next = NULL;
765 		} else {
766 			struct mbuf *n;
767 
768 #ifdef DIAGNOSTIC
769 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
770 				m_freem(m);
771 				return EMSGSIZE;
772 			}
773 #endif
774 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
775 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
776 				MCLGET(n, M_DONTWAIT);
777 				if ((n->m_flags & M_EXT) == 0) {
778 					m_freem(n);
779 					n = NULL;
780 				}
781 			}
782 			if (!n) {
783 				m_freem(m);
784 				return ENOBUFS;
785 			}
786 
787 			MCLAIM(n, &tcp_tx_mowner);
788 			n->m_data += max_linkhdr;
789 			n->m_len = hlen + tlen;
790 			m_copyback(n, 0, hlen, mtod(m, void *));
791 			m_copyback(n, hlen, tlen, (void *)th0);
792 
793 			m_freem(m);
794 			m = n;
795 			n = NULL;
796 		}
797 
798 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
799 		switch (family) {
800 		case AF_INET:
801 			ip = mtod(m, struct ip *);
802 			th = (struct tcphdr *)(ip + 1);
803 			ip->ip_p = IPPROTO_TCP;
804 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
805 			ip->ip_p = IPPROTO_TCP;
806 			break;
807 #ifdef INET6
808 		case AF_INET6:
809 			ip6 = mtod(m, struct ip6_hdr *);
810 			th = (struct tcphdr *)(ip6 + 1);
811 			ip6->ip6_nxt = IPPROTO_TCP;
812 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
813 			ip6->ip6_nxt = IPPROTO_TCP;
814 			break;
815 #endif
816 #if 0
817 		default:
818 			/* noone will visit here */
819 			m_freem(m);
820 			return EAFNOSUPPORT;
821 #endif
822 		}
823 		xchg(th->th_dport, th->th_sport, u_int16_t);
824 #undef xchg
825 		tlen = 0;	/*be friendly with the following code*/
826 	}
827 	th->th_seq = htonl(seq);
828 	th->th_ack = htonl(ack);
829 	th->th_x2 = 0;
830 	if ((flags & TH_SYN) == 0) {
831 		if (tp)
832 			win >>= tp->rcv_scale;
833 		if (win > TCP_MAXWIN)
834 			win = TCP_MAXWIN;
835 		th->th_win = htons((u_int16_t)win);
836 		th->th_off = sizeof (struct tcphdr) >> 2;
837 		tlen += sizeof(*th);
838 	} else
839 		tlen += th->th_off << 2;
840 	m->m_len = hlen + tlen;
841 	m->m_pkthdr.len = hlen + tlen;
842 	m_reset_rcvif(m);
843 	th->th_flags = flags;
844 	th->th_urp = 0;
845 
846 	switch (family) {
847 #ifdef INET
848 	case AF_INET:
849 	    {
850 		struct ipovly *ipov = (struct ipovly *)ip;
851 		memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
852 		ipov->ih_len = htons((u_int16_t)tlen);
853 
854 		th->th_sum = 0;
855 		th->th_sum = in_cksum(m, hlen + tlen);
856 		ip->ip_len = htons(hlen + tlen);
857 		ip->ip_ttl = ip_defttl;
858 		break;
859 	    }
860 #endif
861 #ifdef INET6
862 	case AF_INET6:
863 	    {
864 		th->th_sum = 0;
865 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
866 				tlen);
867 		ip6->ip6_plen = htons(tlen);
868 		if (tp && tp->t_in6pcb)
869 			ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb);
870 		else
871 			ip6->ip6_hlim = ip6_defhlim;
872 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
873 		if (ip6_auto_flowlabel) {
874 			ip6->ip6_flow |=
875 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
876 		}
877 		break;
878 	    }
879 #endif
880 	}
881 
882 	if (tp != NULL && tp->t_inpcb != NULL) {
883 		ro = &tp->t_inpcb->inp_route;
884 #ifdef DIAGNOSTIC
885 		if (family != AF_INET)
886 			panic("tcp_respond: address family mismatch");
887 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
888 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
889 			    ntohl(ip->ip_dst.s_addr),
890 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
891 		}
892 #endif
893 	}
894 #ifdef INET6
895 	else if (tp != NULL && tp->t_in6pcb != NULL) {
896 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
897 #ifdef DIAGNOSTIC
898 		if (family == AF_INET) {
899 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
900 				panic("tcp_respond: not mapped addr");
901 			if (memcmp(&ip->ip_dst,
902 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
903 			    sizeof(ip->ip_dst)) != 0) {
904 				panic("tcp_respond: ip_dst != in6p_faddr");
905 			}
906 		} else if (family == AF_INET6) {
907 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
908 			    &tp->t_in6pcb->in6p_faddr))
909 				panic("tcp_respond: ip6_dst != in6p_faddr");
910 		} else
911 			panic("tcp_respond: address family mismatch");
912 #endif
913 	}
914 #endif
915 	else
916 		ro = NULL;
917 
918 	switch (family) {
919 #ifdef INET
920 	case AF_INET:
921 		error = ip_output(m, NULL, ro,
922 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL,
923 		    tp ? tp->t_inpcb : NULL);
924 		break;
925 #endif
926 #ifdef INET6
927 	case AF_INET6:
928 		error = ip6_output(m, NULL, ro, 0, NULL,
929 		    tp ? tp->t_in6pcb : NULL, NULL);
930 		break;
931 #endif
932 	default:
933 		error = EAFNOSUPPORT;
934 		break;
935 	}
936 
937 	return (error);
938 }
939 
940 /*
941  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
942  * a bunch of members individually, we maintain this template for the
943  * static and mostly-static components of the TCPCB, and copy it into
944  * the new TCPCB instead.
945  */
946 static struct tcpcb tcpcb_template = {
947 	.t_srtt = TCPTV_SRTTBASE,
948 	.t_rttmin = TCPTV_MIN,
949 
950 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
951 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
952 	.snd_numholes = 0,
953 	.snd_cubic_wmax = 0,
954 	.snd_cubic_wmax_last = 0,
955 	.snd_cubic_ctime = 0,
956 
957 	.t_partialacks = -1,
958 	.t_bytes_acked = 0,
959 	.t_sndrexmitpack = 0,
960 	.t_rcvoopack = 0,
961 	.t_sndzerowin = 0,
962 };
963 
964 /*
965  * Updates the TCPCB template whenever a parameter that would affect
966  * the template is changed.
967  */
968 void
969 tcp_tcpcb_template(void)
970 {
971 	struct tcpcb *tp = &tcpcb_template;
972 	int flags;
973 
974 	tp->t_peermss = tcp_mssdflt;
975 	tp->t_ourmss = tcp_mssdflt;
976 	tp->t_segsz = tcp_mssdflt;
977 
978 	flags = 0;
979 	if (tcp_do_rfc1323 && tcp_do_win_scale)
980 		flags |= TF_REQ_SCALE;
981 	if (tcp_do_rfc1323 && tcp_do_timestamps)
982 		flags |= TF_REQ_TSTMP;
983 	tp->t_flags = flags;
984 
985 	/*
986 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
987 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
988 	 * reasonable initial retransmit time.
989 	 */
990 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
991 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
992 	    TCPTV_MIN, TCPTV_REXMTMAX);
993 
994 	/* Keep Alive */
995 	tp->t_keepinit = tcp_keepinit;
996 	tp->t_keepidle = tcp_keepidle;
997 	tp->t_keepintvl = tcp_keepintvl;
998 	tp->t_keepcnt = tcp_keepcnt;
999 	tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1000 
1001 	/* MSL */
1002 	tp->t_msl = TCPTV_MSL;
1003 }
1004 
1005 /*
1006  * Create a new TCP control block, making an
1007  * empty reassembly queue and hooking it to the argument
1008  * protocol control block.
1009  */
1010 /* family selects inpcb, or in6pcb */
1011 struct tcpcb *
1012 tcp_newtcpcb(int family, void *aux)
1013 {
1014 	struct tcpcb *tp;
1015 	int i;
1016 
1017 	/* XXX Consider using a pool_cache for speed. */
1018 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);	/* splsoftnet via tcp_usrreq */
1019 	if (tp == NULL)
1020 		return (NULL);
1021 	memcpy(tp, &tcpcb_template, sizeof(*tp));
1022 	TAILQ_INIT(&tp->segq);
1023 	TAILQ_INIT(&tp->timeq);
1024 	tp->t_family = family;		/* may be overridden later on */
1025 	TAILQ_INIT(&tp->snd_holes);
1026 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
1027 
1028 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
1029 	for (i = 0; i < TCPT_NTIMERS; i++) {
1030 		callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1031 		TCP_TIMER_INIT(tp, i);
1032 	}
1033 	callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1034 
1035 	switch (family) {
1036 	case AF_INET:
1037 	    {
1038 		struct inpcb *inp = (struct inpcb *)aux;
1039 
1040 		inp->inp_ip.ip_ttl = ip_defttl;
1041 		inp->inp_ppcb = (void *)tp;
1042 
1043 		tp->t_inpcb = inp;
1044 		tp->t_mtudisc = ip_mtudisc;
1045 		break;
1046 	    }
1047 #ifdef INET6
1048 	case AF_INET6:
1049 	    {
1050 		struct in6pcb *in6p = (struct in6pcb *)aux;
1051 
1052 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p);
1053 		in6p->in6p_ppcb = (void *)tp;
1054 
1055 		tp->t_in6pcb = in6p;
1056 		/* for IPv6, always try to run path MTU discovery */
1057 		tp->t_mtudisc = 1;
1058 		break;
1059 	    }
1060 #endif /* INET6 */
1061 	default:
1062 		for (i = 0; i < TCPT_NTIMERS; i++)
1063 			callout_destroy(&tp->t_timer[i]);
1064 		callout_destroy(&tp->t_delack_ch);
1065 		pool_put(&tcpcb_pool, tp);	/* splsoftnet via tcp_usrreq */
1066 		return (NULL);
1067 	}
1068 
1069 	/*
1070 	 * Initialize our timebase.  When we send timestamps, we take
1071 	 * the delta from tcp_now -- this means each connection always
1072 	 * gets a timebase of 1, which makes it, among other things,
1073 	 * more difficult to determine how long a system has been up,
1074 	 * and thus how many TCP sequence increments have occurred.
1075 	 *
1076 	 * We start with 1, because 0 doesn't work with linux, which
1077 	 * considers timestamp 0 in a SYN packet as a bug and disables
1078 	 * timestamps.
1079 	 */
1080 	tp->ts_timebase = tcp_now - 1;
1081 
1082 	tcp_congctl_select(tp, tcp_congctl_global_name);
1083 
1084 	return (tp);
1085 }
1086 
1087 /*
1088  * Drop a TCP connection, reporting
1089  * the specified error.  If connection is synchronized,
1090  * then send a RST to peer.
1091  */
1092 struct tcpcb *
1093 tcp_drop(struct tcpcb *tp, int errno)
1094 {
1095 	struct socket *so = NULL;
1096 
1097 #ifdef DIAGNOSTIC
1098 	if (tp->t_inpcb && tp->t_in6pcb)
1099 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1100 #endif
1101 #ifdef INET
1102 	if (tp->t_inpcb)
1103 		so = tp->t_inpcb->inp_socket;
1104 #endif
1105 #ifdef INET6
1106 	if (tp->t_in6pcb)
1107 		so = tp->t_in6pcb->in6p_socket;
1108 #endif
1109 	if (!so)
1110 		return NULL;
1111 
1112 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1113 		tp->t_state = TCPS_CLOSED;
1114 		(void) tcp_output(tp);
1115 		TCP_STATINC(TCP_STAT_DROPS);
1116 	} else
1117 		TCP_STATINC(TCP_STAT_CONNDROPS);
1118 	if (errno == ETIMEDOUT && tp->t_softerror)
1119 		errno = tp->t_softerror;
1120 	so->so_error = errno;
1121 	return (tcp_close(tp));
1122 }
1123 
1124 /*
1125  * Close a TCP control block:
1126  *	discard all space held by the tcp
1127  *	discard internet protocol block
1128  *	wake up any sleepers
1129  */
1130 struct tcpcb *
1131 tcp_close(struct tcpcb *tp)
1132 {
1133 	struct inpcb *inp;
1134 #ifdef INET6
1135 	struct in6pcb *in6p;
1136 #endif
1137 	struct socket *so;
1138 #ifdef RTV_RTT
1139 	struct rtentry *rt = NULL;
1140 #endif
1141 	struct route *ro;
1142 	int j;
1143 
1144 	inp = tp->t_inpcb;
1145 #ifdef INET6
1146 	in6p = tp->t_in6pcb;
1147 #endif
1148 	so = NULL;
1149 	ro = NULL;
1150 	if (inp) {
1151 		so = inp->inp_socket;
1152 		ro = &inp->inp_route;
1153 	}
1154 #ifdef INET6
1155 	else if (in6p) {
1156 		so = in6p->in6p_socket;
1157 		ro = (struct route *)&in6p->in6p_route;
1158 	}
1159 #endif
1160 
1161 #ifdef RTV_RTT
1162 	/*
1163 	 * If we sent enough data to get some meaningful characteristics,
1164 	 * save them in the routing entry.  'Enough' is arbitrarily
1165 	 * defined as the sendpipesize (default 4K) * 16.  This would
1166 	 * give us 16 rtt samples assuming we only get one sample per
1167 	 * window (the usual case on a long haul net).  16 samples is
1168 	 * enough for the srtt filter to converge to within 5% of the correct
1169 	 * value; fewer samples and we could save a very bogus rtt.
1170 	 *
1171 	 * Don't update the default route's characteristics and don't
1172 	 * update anything that the user "locked".
1173 	 */
1174 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1175 	    ro && (rt = rtcache_validate(ro)) != NULL &&
1176 	    !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1177 		u_long i = 0;
1178 
1179 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1180 			i = tp->t_srtt *
1181 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1182 			if (rt->rt_rmx.rmx_rtt && i)
1183 				/*
1184 				 * filter this update to half the old & half
1185 				 * the new values, converting scale.
1186 				 * See route.h and tcp_var.h for a
1187 				 * description of the scaling constants.
1188 				 */
1189 				rt->rt_rmx.rmx_rtt =
1190 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1191 			else
1192 				rt->rt_rmx.rmx_rtt = i;
1193 		}
1194 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1195 			i = tp->t_rttvar *
1196 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1197 			if (rt->rt_rmx.rmx_rttvar && i)
1198 				rt->rt_rmx.rmx_rttvar =
1199 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1200 			else
1201 				rt->rt_rmx.rmx_rttvar = i;
1202 		}
1203 		/*
1204 		 * update the pipelimit (ssthresh) if it has been updated
1205 		 * already or if a pipesize was specified & the threshhold
1206 		 * got below half the pipesize.  I.e., wait for bad news
1207 		 * before we start updating, then update on both good
1208 		 * and bad news.
1209 		 */
1210 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1211 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1212 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1213 			/*
1214 			 * convert the limit from user data bytes to
1215 			 * packets then to packet data bytes.
1216 			 */
1217 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1218 			if (i < 2)
1219 				i = 2;
1220 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1221 			if (rt->rt_rmx.rmx_ssthresh)
1222 				rt->rt_rmx.rmx_ssthresh =
1223 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1224 			else
1225 				rt->rt_rmx.rmx_ssthresh = i;
1226 		}
1227 	}
1228 	rtcache_unref(rt, ro);
1229 #endif /* RTV_RTT */
1230 	/* free the reassembly queue, if any */
1231 	TCP_REASS_LOCK(tp);
1232 	(void) tcp_freeq(tp);
1233 	TCP_REASS_UNLOCK(tp);
1234 
1235 	/* free the SACK holes list. */
1236 	tcp_free_sackholes(tp);
1237 	tcp_congctl_release(tp);
1238 	syn_cache_cleanup(tp);
1239 
1240 	if (tp->t_template) {
1241 		m_free(tp->t_template);
1242 		tp->t_template = NULL;
1243 	}
1244 
1245 	/*
1246 	 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1247 	 * the timers can also drop the lock.  We need to prevent access
1248 	 * to the tcpcb as it's half torn down.  Flag the pcb as dead
1249 	 * (prevents access by timers) and only then detach it.
1250 	 */
1251 	tp->t_flags |= TF_DEAD;
1252 	if (inp) {
1253 		inp->inp_ppcb = 0;
1254 		soisdisconnected(so);
1255 		in_pcbdetach(inp);
1256 	}
1257 #ifdef INET6
1258 	else if (in6p) {
1259 		in6p->in6p_ppcb = 0;
1260 		soisdisconnected(so);
1261 		in6_pcbdetach(in6p);
1262 	}
1263 #endif
1264 	/*
1265 	 * pcb is no longer visble elsewhere, so we can safely release
1266 	 * the lock in callout_halt() if needed.
1267 	 */
1268 	TCP_STATINC(TCP_STAT_CLOSED);
1269 	for (j = 0; j < TCPT_NTIMERS; j++) {
1270 		callout_halt(&tp->t_timer[j], softnet_lock);
1271 		callout_destroy(&tp->t_timer[j]);
1272 	}
1273 	callout_halt(&tp->t_delack_ch, softnet_lock);
1274 	callout_destroy(&tp->t_delack_ch);
1275 	pool_put(&tcpcb_pool, tp);
1276 
1277 	return NULL;
1278 }
1279 
1280 int
1281 tcp_freeq(struct tcpcb *tp)
1282 {
1283 	struct ipqent *qe;
1284 	int rv = 0;
1285 #ifdef TCPREASS_DEBUG
1286 	int i = 0;
1287 #endif
1288 
1289 	TCP_REASS_LOCK_CHECK(tp);
1290 
1291 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1292 #ifdef TCPREASS_DEBUG
1293 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1294 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1295 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1296 #endif
1297 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1298 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1299 		m_freem(qe->ipqe_m);
1300 		tcpipqent_free(qe);
1301 		rv = 1;
1302 	}
1303 	tp->t_segqlen = 0;
1304 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1305 	return (rv);
1306 }
1307 
1308 void
1309 tcp_fasttimo(void)
1310 {
1311 	if (tcp_drainwanted) {
1312 		tcp_drain();
1313 		tcp_drainwanted = 0;
1314 	}
1315 }
1316 
1317 void
1318 tcp_drainstub(void)
1319 {
1320 	tcp_drainwanted = 1;
1321 }
1322 
1323 /*
1324  * Protocol drain routine.  Called when memory is in short supply.
1325  * Called from pr_fasttimo thus a callout context.
1326  */
1327 void
1328 tcp_drain(void)
1329 {
1330 	struct inpcb_hdr *inph;
1331 	struct tcpcb *tp;
1332 
1333 	mutex_enter(softnet_lock);
1334 	KERNEL_LOCK(1, NULL);
1335 
1336 	/*
1337 	 * Free the sequence queue of all TCP connections.
1338 	 */
1339 	TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1340 		switch (inph->inph_af) {
1341 		case AF_INET:
1342 			tp = intotcpcb((struct inpcb *)inph);
1343 			break;
1344 #ifdef INET6
1345 		case AF_INET6:
1346 			tp = in6totcpcb((struct in6pcb *)inph);
1347 			break;
1348 #endif
1349 		default:
1350 			tp = NULL;
1351 			break;
1352 		}
1353 		if (tp != NULL) {
1354 			/*
1355 			 * We may be called from a device's interrupt
1356 			 * context.  If the tcpcb is already busy,
1357 			 * just bail out now.
1358 			 */
1359 			if (tcp_reass_lock_try(tp) == 0)
1360 				continue;
1361 			if (tcp_freeq(tp))
1362 				TCP_STATINC(TCP_STAT_CONNSDRAINED);
1363 			TCP_REASS_UNLOCK(tp);
1364 		}
1365 	}
1366 
1367 	KERNEL_UNLOCK_ONE(NULL);
1368 	mutex_exit(softnet_lock);
1369 }
1370 
1371 /*
1372  * Notify a tcp user of an asynchronous error;
1373  * store error as soft error, but wake up user
1374  * (for now, won't do anything until can select for soft error).
1375  */
1376 void
1377 tcp_notify(struct inpcb *inp, int error)
1378 {
1379 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1380 	struct socket *so = inp->inp_socket;
1381 
1382 	/*
1383 	 * Ignore some errors if we are hooked up.
1384 	 * If connection hasn't completed, has retransmitted several times,
1385 	 * and receives a second error, give up now.  This is better
1386 	 * than waiting a long time to establish a connection that
1387 	 * can never complete.
1388 	 */
1389 	if (tp->t_state == TCPS_ESTABLISHED &&
1390 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1391 	      error == EHOSTDOWN)) {
1392 		return;
1393 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1394 	    tp->t_rxtshift > 3 && tp->t_softerror)
1395 		so->so_error = error;
1396 	else
1397 		tp->t_softerror = error;
1398 	cv_broadcast(&so->so_cv);
1399 	sorwakeup(so);
1400 	sowwakeup(so);
1401 }
1402 
1403 #ifdef INET6
1404 void
1405 tcp6_notify(struct in6pcb *in6p, int error)
1406 {
1407 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1408 	struct socket *so = in6p->in6p_socket;
1409 
1410 	/*
1411 	 * Ignore some errors if we are hooked up.
1412 	 * If connection hasn't completed, has retransmitted several times,
1413 	 * and receives a second error, give up now.  This is better
1414 	 * than waiting a long time to establish a connection that
1415 	 * can never complete.
1416 	 */
1417 	if (tp->t_state == TCPS_ESTABLISHED &&
1418 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1419 	      error == EHOSTDOWN)) {
1420 		return;
1421 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1422 	    tp->t_rxtshift > 3 && tp->t_softerror)
1423 		so->so_error = error;
1424 	else
1425 		tp->t_softerror = error;
1426 	cv_broadcast(&so->so_cv);
1427 	sorwakeup(so);
1428 	sowwakeup(so);
1429 }
1430 #endif
1431 
1432 #ifdef INET6
1433 void *
1434 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1435 {
1436 	struct tcphdr th;
1437 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1438 	int nmatch;
1439 	struct ip6_hdr *ip6;
1440 	const struct sockaddr_in6 *sa6_src = NULL;
1441 	const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1442 	struct mbuf *m;
1443 	int off;
1444 
1445 	if (sa->sa_family != AF_INET6 ||
1446 	    sa->sa_len != sizeof(struct sockaddr_in6))
1447 		return NULL;
1448 	if ((unsigned)cmd >= PRC_NCMDS)
1449 		return NULL;
1450 	else if (cmd == PRC_QUENCH) {
1451 		/*
1452 		 * Don't honor ICMP Source Quench messages meant for
1453 		 * TCP connections.
1454 		 */
1455 		return NULL;
1456 	} else if (PRC_IS_REDIRECT(cmd))
1457 		notify = in6_rtchange, d = NULL;
1458 	else if (cmd == PRC_MSGSIZE)
1459 		; /* special code is present, see below */
1460 	else if (cmd == PRC_HOSTDEAD)
1461 		d = NULL;
1462 	else if (inet6ctlerrmap[cmd] == 0)
1463 		return NULL;
1464 
1465 	/* if the parameter is from icmp6, decode it. */
1466 	if (d != NULL) {
1467 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1468 		m = ip6cp->ip6c_m;
1469 		ip6 = ip6cp->ip6c_ip6;
1470 		off = ip6cp->ip6c_off;
1471 		sa6_src = ip6cp->ip6c_src;
1472 	} else {
1473 		m = NULL;
1474 		ip6 = NULL;
1475 		sa6_src = &sa6_any;
1476 		off = 0;
1477 	}
1478 
1479 	if (ip6) {
1480 		/*
1481 		 * XXX: We assume that when ip6 is non NULL,
1482 		 * M and OFF are valid.
1483 		 */
1484 
1485 		/* check if we can safely examine src and dst ports */
1486 		if (m->m_pkthdr.len < off + sizeof(th)) {
1487 			if (cmd == PRC_MSGSIZE)
1488 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1489 			return NULL;
1490 		}
1491 
1492 		memset(&th, 0, sizeof(th));
1493 		m_copydata(m, off, sizeof(th), (void *)&th);
1494 
1495 		if (cmd == PRC_MSGSIZE) {
1496 			int valid = 0;
1497 
1498 			/*
1499 			 * Check to see if we have a valid TCP connection
1500 			 * corresponding to the address in the ICMPv6 message
1501 			 * payload.
1502 			 */
1503 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1504 			    th.th_dport,
1505 			    (const struct in6_addr *)&sa6_src->sin6_addr,
1506 						  th.th_sport, 0, 0))
1507 				valid++;
1508 
1509 			/*
1510 			 * Depending on the value of "valid" and routing table
1511 			 * size (mtudisc_{hi,lo}wat), we will:
1512 			 * - recalcurate the new MTU and create the
1513 			 *   corresponding routing entry, or
1514 			 * - ignore the MTU change notification.
1515 			 */
1516 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1517 
1518 			/*
1519 			 * no need to call in6_pcbnotify, it should have been
1520 			 * called via callback if necessary
1521 			 */
1522 			return NULL;
1523 		}
1524 
1525 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1526 		    (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1527 		if (nmatch == 0 && syn_cache_count &&
1528 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1529 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1530 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1531 			syn_cache_unreach((const struct sockaddr *)sa6_src,
1532 					  sa, &th);
1533 	} else {
1534 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1535 		    (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1536 	}
1537 
1538 	return NULL;
1539 }
1540 #endif
1541 
1542 #ifdef INET
1543 /* assumes that ip header and tcp header are contiguous on mbuf */
1544 void *
1545 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1546 {
1547 	struct ip *ip = v;
1548 	struct tcphdr *th;
1549 	struct icmp *icp;
1550 	extern const int inetctlerrmap[];
1551 	void (*notify)(struct inpcb *, int) = tcp_notify;
1552 	int errno;
1553 	int nmatch;
1554 	struct tcpcb *tp;
1555 	u_int mtu;
1556 	tcp_seq seq;
1557 	struct inpcb *inp;
1558 #ifdef INET6
1559 	struct in6pcb *in6p;
1560 	struct in6_addr src6, dst6;
1561 #endif
1562 
1563 	if (sa->sa_family != AF_INET ||
1564 	    sa->sa_len != sizeof(struct sockaddr_in))
1565 		return NULL;
1566 	if ((unsigned)cmd >= PRC_NCMDS)
1567 		return NULL;
1568 	errno = inetctlerrmap[cmd];
1569 	if (cmd == PRC_QUENCH)
1570 		/*
1571 		 * Don't honor ICMP Source Quench messages meant for
1572 		 * TCP connections.
1573 		 */
1574 		return NULL;
1575 	else if (PRC_IS_REDIRECT(cmd))
1576 		notify = in_rtchange, ip = 0;
1577 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1578 		/*
1579 		 * Check to see if we have a valid TCP connection
1580 		 * corresponding to the address in the ICMP message
1581 		 * payload.
1582 		 *
1583 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1584 		 */
1585 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1586 #ifdef INET6
1587 		in6_in_2_v4mapin6(&ip->ip_src, &src6);
1588 		in6_in_2_v4mapin6(&ip->ip_dst, &dst6);
1589 #endif
1590 		if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1591 						th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1592 #ifdef INET6
1593 			in6p = NULL;
1594 #else
1595 			;
1596 #endif
1597 #ifdef INET6
1598 		else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1599 						       th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1600 			;
1601 #endif
1602 		else
1603 			return NULL;
1604 
1605 		/*
1606 		 * Now that we've validated that we are actually communicating
1607 		 * with the host indicated in the ICMP message, locate the
1608 		 * ICMP header, recalculate the new MTU, and create the
1609 		 * corresponding routing entry.
1610 		 */
1611 		icp = (struct icmp *)((char *)ip -
1612 		    offsetof(struct icmp, icmp_ip));
1613 		if (inp) {
1614 			if ((tp = intotcpcb(inp)) == NULL)
1615 				return NULL;
1616 		}
1617 #ifdef INET6
1618 		else if (in6p) {
1619 			if ((tp = in6totcpcb(in6p)) == NULL)
1620 				return NULL;
1621 		}
1622 #endif
1623 		else
1624 			return NULL;
1625 		seq = ntohl(th->th_seq);
1626 		if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1627 			return NULL;
1628 		/*
1629 		 * If the ICMP message advertises a Next-Hop MTU
1630 		 * equal or larger than the maximum packet size we have
1631 		 * ever sent, drop the message.
1632 		 */
1633 		mtu = (u_int)ntohs(icp->icmp_nextmtu);
1634 		if (mtu >= tp->t_pmtud_mtu_sent)
1635 			return NULL;
1636 		if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1637 			/*
1638 			 * Calculate new MTU, and create corresponding
1639 			 * route (traditional PMTUD).
1640 			 */
1641 			tp->t_flags &= ~TF_PMTUD_PEND;
1642 			icmp_mtudisc(icp, ip->ip_dst);
1643 		} else {
1644 			/*
1645 			 * Record the information got in the ICMP
1646 			 * message; act on it later.
1647 			 * If we had already recorded an ICMP message,
1648 			 * replace the old one only if the new message
1649 			 * refers to an older TCP segment
1650 			 */
1651 			if (tp->t_flags & TF_PMTUD_PEND) {
1652 				if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1653 					return NULL;
1654 			} else
1655 				tp->t_flags |= TF_PMTUD_PEND;
1656 			tp->t_pmtud_th_seq = seq;
1657 			tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1658 			tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1659 			tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1660 		}
1661 		return NULL;
1662 	} else if (cmd == PRC_HOSTDEAD)
1663 		ip = 0;
1664 	else if (errno == 0)
1665 		return NULL;
1666 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1667 		th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1668 		nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1669 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1670 		if (nmatch == 0 && syn_cache_count &&
1671 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1672 		    inetctlerrmap[cmd] == ENETUNREACH ||
1673 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1674 			struct sockaddr_in sin;
1675 			memset(&sin, 0, sizeof(sin));
1676 			sin.sin_len = sizeof(sin);
1677 			sin.sin_family = AF_INET;
1678 			sin.sin_port = th->th_sport;
1679 			sin.sin_addr = ip->ip_src;
1680 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1681 		}
1682 
1683 		/* XXX mapped address case */
1684 	} else
1685 		in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1686 		    notify);
1687 	return NULL;
1688 }
1689 
1690 /*
1691  * When a source quench is received, we are being notified of congestion.
1692  * Close the congestion window down to the Loss Window (one segment).
1693  * We will gradually open it again as we proceed.
1694  */
1695 void
1696 tcp_quench(struct inpcb *inp, int errno)
1697 {
1698 	struct tcpcb *tp = intotcpcb(inp);
1699 
1700 	if (tp) {
1701 		tp->snd_cwnd = tp->t_segsz;
1702 		tp->t_bytes_acked = 0;
1703 	}
1704 }
1705 #endif
1706 
1707 #ifdef INET6
1708 void
1709 tcp6_quench(struct in6pcb *in6p, int errno)
1710 {
1711 	struct tcpcb *tp = in6totcpcb(in6p);
1712 
1713 	if (tp) {
1714 		tp->snd_cwnd = tp->t_segsz;
1715 		tp->t_bytes_acked = 0;
1716 	}
1717 }
1718 #endif
1719 
1720 #ifdef INET
1721 /*
1722  * Path MTU Discovery handlers.
1723  */
1724 void
1725 tcp_mtudisc_callback(struct in_addr faddr)
1726 {
1727 #ifdef INET6
1728 	struct in6_addr in6;
1729 #endif
1730 
1731 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1732 #ifdef INET6
1733 	in6_in_2_v4mapin6(&faddr, &in6);
1734 	tcp6_mtudisc_callback(&in6);
1735 #endif
1736 }
1737 
1738 /*
1739  * On receipt of path MTU corrections, flush old route and replace it
1740  * with the new one.  Retransmit all unacknowledged packets, to ensure
1741  * that all packets will be received.
1742  */
1743 void
1744 tcp_mtudisc(struct inpcb *inp, int errno)
1745 {
1746 	struct tcpcb *tp = intotcpcb(inp);
1747 	struct rtentry *rt;
1748 
1749 	if (tp == NULL)
1750 		return;
1751 
1752 	rt = in_pcbrtentry(inp);
1753 	if (rt != NULL) {
1754 		/*
1755 		 * If this was not a host route, remove and realloc.
1756 		 */
1757 		if ((rt->rt_flags & RTF_HOST) == 0) {
1758 			in_pcbrtentry_unref(rt, inp);
1759 			in_rtchange(inp, errno);
1760 			if ((rt = in_pcbrtentry(inp)) == NULL)
1761 				return;
1762 		}
1763 
1764 		/*
1765 		 * Slow start out of the error condition.  We
1766 		 * use the MTU because we know it's smaller
1767 		 * than the previously transmitted segment.
1768 		 *
1769 		 * Note: This is more conservative than the
1770 		 * suggestion in draft-floyd-incr-init-win-03.
1771 		 */
1772 		if (rt->rt_rmx.rmx_mtu != 0)
1773 			tp->snd_cwnd =
1774 			    TCP_INITIAL_WINDOW(tcp_init_win,
1775 			    rt->rt_rmx.rmx_mtu);
1776 		in_pcbrtentry_unref(rt, inp);
1777 	}
1778 
1779 	/*
1780 	 * Resend unacknowledged packets.
1781 	 */
1782 	tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1783 	tcp_output(tp);
1784 }
1785 #endif /* INET */
1786 
1787 #ifdef INET6
1788 /*
1789  * Path MTU Discovery handlers.
1790  */
1791 void
1792 tcp6_mtudisc_callback(struct in6_addr *faddr)
1793 {
1794 	struct sockaddr_in6 sin6;
1795 
1796 	memset(&sin6, 0, sizeof(sin6));
1797 	sin6.sin6_family = AF_INET6;
1798 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1799 	sin6.sin6_addr = *faddr;
1800 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1801 	    (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1802 }
1803 
1804 void
1805 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1806 {
1807 	struct tcpcb *tp = in6totcpcb(in6p);
1808 	struct rtentry *rt;
1809 
1810 	if (tp == NULL)
1811 		return;
1812 
1813 	rt = in6_pcbrtentry(in6p);
1814 	if (rt != NULL) {
1815 		/*
1816 		 * If this was not a host route, remove and realloc.
1817 		 */
1818 		if ((rt->rt_flags & RTF_HOST) == 0) {
1819 			in6_pcbrtentry_unref(rt, in6p);
1820 			in6_rtchange(in6p, errno);
1821 			rt = in6_pcbrtentry(in6p);
1822 			if (rt == NULL)
1823 				return;
1824 		}
1825 
1826 		/*
1827 		 * Slow start out of the error condition.  We
1828 		 * use the MTU because we know it's smaller
1829 		 * than the previously transmitted segment.
1830 		 *
1831 		 * Note: This is more conservative than the
1832 		 * suggestion in draft-floyd-incr-init-win-03.
1833 		 */
1834 		if (rt->rt_rmx.rmx_mtu != 0) {
1835 			tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
1836 			    rt->rt_rmx.rmx_mtu);
1837 		}
1838 		in6_pcbrtentry_unref(rt, in6p);
1839 	}
1840 
1841 	/*
1842 	 * Resend unacknowledged packets.
1843 	 */
1844 	tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1845 	tcp_output(tp);
1846 }
1847 #endif /* INET6 */
1848 
1849 /*
1850  * Compute the MSS to advertise to the peer.  Called only during
1851  * the 3-way handshake.  If we are the server (peer initiated
1852  * connection), we are called with a pointer to the interface
1853  * on which the SYN packet arrived.  If we are the client (we
1854  * initiated connection), we are called with a pointer to the
1855  * interface out which this connection should go.
1856  *
1857  * NOTE: Do not subtract IP option/extension header size nor IPsec
1858  * header size from MSS advertisement.  MSS option must hold the maximum
1859  * segment size we can accept, so it must always be:
1860  *	 max(if mtu) - ip header - tcp header
1861  */
1862 u_long
1863 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1864 {
1865 	extern u_long in_maxmtu;
1866 	u_long mss = 0;
1867 	u_long hdrsiz;
1868 
1869 	/*
1870 	 * In order to avoid defeating path MTU discovery on the peer,
1871 	 * we advertise the max MTU of all attached networks as our MSS,
1872 	 * per RFC 1191, section 3.1.
1873 	 *
1874 	 * We provide the option to advertise just the MTU of
1875 	 * the interface on which we hope this connection will
1876 	 * be receiving.  If we are responding to a SYN, we
1877 	 * will have a pretty good idea about this, but when
1878 	 * initiating a connection there is a bit more doubt.
1879 	 *
1880 	 * We also need to ensure that loopback has a large enough
1881 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1882 	 */
1883 
1884 	if (ifp != NULL)
1885 		switch (af) {
1886 		case AF_INET:
1887 			mss = ifp->if_mtu;
1888 			break;
1889 #ifdef INET6
1890 		case AF_INET6:
1891 			mss = IN6_LINKMTU(ifp);
1892 			break;
1893 #endif
1894 		}
1895 
1896 	if (tcp_mss_ifmtu == 0)
1897 		switch (af) {
1898 		case AF_INET:
1899 			mss = max(in_maxmtu, mss);
1900 			break;
1901 #ifdef INET6
1902 		case AF_INET6:
1903 			mss = max(in6_maxmtu, mss);
1904 			break;
1905 #endif
1906 		}
1907 
1908 	switch (af) {
1909 	case AF_INET:
1910 		hdrsiz = sizeof(struct ip);
1911 		break;
1912 #ifdef INET6
1913 	case AF_INET6:
1914 		hdrsiz = sizeof(struct ip6_hdr);
1915 		break;
1916 #endif
1917 	default:
1918 		hdrsiz = 0;
1919 		break;
1920 	}
1921 	hdrsiz += sizeof(struct tcphdr);
1922 	if (mss > hdrsiz)
1923 		mss -= hdrsiz;
1924 
1925 	mss = max(tcp_mssdflt, mss);
1926 	return (mss);
1927 }
1928 
1929 /*
1930  * Set connection variables based on the peer's advertised MSS.
1931  * We are passed the TCPCB for the actual connection.  If we
1932  * are the server, we are called by the compressed state engine
1933  * when the 3-way handshake is complete.  If we are the client,
1934  * we are called when we receive the SYN,ACK from the server.
1935  *
1936  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1937  * before this routine is called!
1938  */
1939 void
1940 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1941 {
1942 	struct socket *so;
1943 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1944 	struct rtentry *rt;
1945 #endif
1946 	u_long bufsize;
1947 	int mss;
1948 
1949 #ifdef DIAGNOSTIC
1950 	if (tp->t_inpcb && tp->t_in6pcb)
1951 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1952 #endif
1953 	so = NULL;
1954 	rt = NULL;
1955 #ifdef INET
1956 	if (tp->t_inpcb) {
1957 		so = tp->t_inpcb->inp_socket;
1958 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1959 		rt = in_pcbrtentry(tp->t_inpcb);
1960 #endif
1961 	}
1962 #endif
1963 #ifdef INET6
1964 	if (tp->t_in6pcb) {
1965 		so = tp->t_in6pcb->in6p_socket;
1966 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1967 		rt = in6_pcbrtentry(tp->t_in6pcb);
1968 #endif
1969 	}
1970 #endif
1971 
1972 	/*
1973 	 * As per RFC1122, use the default MSS value, unless they
1974 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1975 	 */
1976 	mss = tcp_mssdflt;
1977 	if (offer)
1978 		mss = offer;
1979 	mss = max(mss, 256);		/* sanity */
1980 	tp->t_peermss = mss;
1981 	mss -= tcp_optlen(tp);
1982 #ifdef INET
1983 	if (tp->t_inpcb)
1984 		mss -= ip_optlen(tp->t_inpcb);
1985 #endif
1986 #ifdef INET6
1987 	if (tp->t_in6pcb)
1988 		mss -= ip6_optlen(tp->t_in6pcb);
1989 #endif
1990 
1991 	/*
1992 	 * If there's a pipesize, change the socket buffer to that size.
1993 	 * Make the socket buffer an integral number of MSS units.  If
1994 	 * the MSS is larger than the socket buffer, artificially decrease
1995 	 * the MSS.
1996 	 */
1997 #ifdef RTV_SPIPE
1998 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1999 		bufsize = rt->rt_rmx.rmx_sendpipe;
2000 	else
2001 #endif
2002 	{
2003 		KASSERT(so != NULL);
2004 		bufsize = so->so_snd.sb_hiwat;
2005 	}
2006 	if (bufsize < mss)
2007 		mss = bufsize;
2008 	else {
2009 		bufsize = roundup(bufsize, mss);
2010 		if (bufsize > sb_max)
2011 			bufsize = sb_max;
2012 		(void) sbreserve(&so->so_snd, bufsize, so);
2013 	}
2014 	tp->t_segsz = mss;
2015 
2016 #ifdef RTV_SSTHRESH
2017 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2018 		/*
2019 		 * There's some sort of gateway or interface buffer
2020 		 * limit on the path.  Use this to set the slow
2021 		 * start threshold, but set the threshold to no less
2022 		 * than 2 * MSS.
2023 		 */
2024 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2025 	}
2026 #endif
2027 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
2028 #ifdef INET
2029 	if (tp->t_inpcb)
2030 		in_pcbrtentry_unref(rt, tp->t_inpcb);
2031 #endif
2032 #ifdef INET6
2033 	if (tp->t_in6pcb)
2034 		in6_pcbrtentry_unref(rt, tp->t_in6pcb);
2035 #endif
2036 #endif
2037 }
2038 
2039 /*
2040  * Processing necessary when a TCP connection is established.
2041  */
2042 void
2043 tcp_established(struct tcpcb *tp)
2044 {
2045 	struct socket *so;
2046 #ifdef RTV_RPIPE
2047 	struct rtentry *rt;
2048 #endif
2049 	u_long bufsize;
2050 
2051 #ifdef DIAGNOSTIC
2052 	if (tp->t_inpcb && tp->t_in6pcb)
2053 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
2054 #endif
2055 	so = NULL;
2056 	rt = NULL;
2057 #ifdef INET
2058 	/* This is a while() to reduce the dreadful stairstepping below */
2059 	while (tp->t_inpcb) {
2060 		so = tp->t_inpcb->inp_socket;
2061 #if defined(RTV_RPIPE)
2062 		rt = in_pcbrtentry(tp->t_inpcb);
2063 #endif
2064 		if (__predict_true(tcp_msl_enable)) {
2065 			if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2066 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2067 				break;
2068 			}
2069 
2070 			if (__predict_false(tcp_rttlocal)) {
2071 				/* This may be adjusted by tcp_input */
2072 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2073 				break;
2074 			}
2075 			if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2076 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2077 				break;
2078 			}
2079 		}
2080 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2081 		break;
2082 	}
2083 #endif
2084 #ifdef INET6
2085 	/* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2086 	while (!tp->t_inpcb && tp->t_in6pcb) {
2087 		so = tp->t_in6pcb->in6p_socket;
2088 #if defined(RTV_RPIPE)
2089 		rt = in6_pcbrtentry(tp->t_in6pcb);
2090 #endif
2091 		if (__predict_true(tcp_msl_enable)) {
2092 			extern const struct in6_addr in6addr_loopback;
2093 
2094 			if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2095 					       &in6addr_loopback)) {
2096 				tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2097 				break;
2098 			}
2099 
2100 			if (__predict_false(tcp_rttlocal)) {
2101 				/* This may be adjusted by tcp_input */
2102 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2103 				break;
2104 			}
2105 			if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2106 				tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2107 				break;
2108 			}
2109 		}
2110 		tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2111 		break;
2112 	}
2113 #endif
2114 
2115 	tp->t_state = TCPS_ESTABLISHED;
2116 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2117 
2118 #ifdef RTV_RPIPE
2119 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2120 		bufsize = rt->rt_rmx.rmx_recvpipe;
2121 	else
2122 #endif
2123 	{
2124 		KASSERT(so != NULL);
2125 		bufsize = so->so_rcv.sb_hiwat;
2126 	}
2127 	if (bufsize > tp->t_ourmss) {
2128 		bufsize = roundup(bufsize, tp->t_ourmss);
2129 		if (bufsize > sb_max)
2130 			bufsize = sb_max;
2131 		(void) sbreserve(&so->so_rcv, bufsize, so);
2132 	}
2133 #ifdef RTV_RPIPE
2134 #ifdef INET
2135 	if (tp->t_inpcb)
2136 		in_pcbrtentry_unref(rt, tp->t_inpcb);
2137 #endif
2138 #ifdef INET6
2139 	if (tp->t_in6pcb)
2140 		in6_pcbrtentry_unref(rt, tp->t_in6pcb);
2141 #endif
2142 #endif
2143 }
2144 
2145 /*
2146  * Check if there's an initial rtt or rttvar.  Convert from the
2147  * route-table units to scaled multiples of the slow timeout timer.
2148  * Called only during the 3-way handshake.
2149  */
2150 void
2151 tcp_rmx_rtt(struct tcpcb *tp)
2152 {
2153 #ifdef RTV_RTT
2154 	struct rtentry *rt = NULL;
2155 	int rtt;
2156 
2157 #ifdef DIAGNOSTIC
2158 	if (tp->t_inpcb && tp->t_in6pcb)
2159 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2160 #endif
2161 #ifdef INET
2162 	if (tp->t_inpcb)
2163 		rt = in_pcbrtentry(tp->t_inpcb);
2164 #endif
2165 #ifdef INET6
2166 	if (tp->t_in6pcb)
2167 		rt = in6_pcbrtentry(tp->t_in6pcb);
2168 #endif
2169 	if (rt == NULL)
2170 		return;
2171 
2172 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2173 		/*
2174 		 * XXX The lock bit for MTU indicates that the value
2175 		 * is also a minimum value; this is subject to time.
2176 		 */
2177 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2178 			TCPT_RANGESET(tp->t_rttmin,
2179 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2180 			    TCPTV_MIN, TCPTV_REXMTMAX);
2181 		tp->t_srtt = rtt /
2182 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2183 		if (rt->rt_rmx.rmx_rttvar) {
2184 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2185 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
2186 				(TCP_RTTVAR_SHIFT + 2));
2187 		} else {
2188 			/* Default variation is +- 1 rtt */
2189 			tp->t_rttvar =
2190 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2191 		}
2192 		TCPT_RANGESET(tp->t_rxtcur,
2193 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2194 		    tp->t_rttmin, TCPTV_REXMTMAX);
2195 	}
2196 #ifdef INET
2197 	if (tp->t_inpcb)
2198 		in_pcbrtentry_unref(rt, tp->t_inpcb);
2199 #endif
2200 #ifdef INET6
2201 	if (tp->t_in6pcb)
2202 		in6_pcbrtentry_unref(rt, tp->t_in6pcb);
2203 #endif
2204 #endif
2205 }
2206 
2207 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2208 
2209 /*
2210  * Get a new sequence value given a tcp control block
2211  */
2212 tcp_seq
2213 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2214 {
2215 
2216 #ifdef INET
2217 	if (tp->t_inpcb != NULL) {
2218 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2219 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2220 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2221 		    addin));
2222 	}
2223 #endif
2224 #ifdef INET6
2225 	if (tp->t_in6pcb != NULL) {
2226 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2227 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2228 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2229 		    addin));
2230 	}
2231 #endif
2232 	/* Not possible. */
2233 	panic("tcp_new_iss");
2234 }
2235 
2236 static u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2237 
2238 /*
2239  * Initialize RFC 1948 ISS Secret
2240  */
2241 static int
2242 tcp_iss_secret_init(void)
2243 {
2244 	cprng_strong(kern_cprng,
2245 	    tcp_iss_secret, sizeof(tcp_iss_secret), 0);
2246 
2247 	return 0;
2248 }
2249 
2250 /*
2251  * This routine actually generates a new TCP initial sequence number.
2252  */
2253 tcp_seq
2254 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2255     size_t addrsz, tcp_seq addin)
2256 {
2257 	tcp_seq tcp_iss;
2258 
2259 	if (tcp_do_rfc1948) {
2260 		MD5_CTX ctx;
2261 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2262 		static ONCE_DECL(tcp_iss_secret_control);
2263 
2264 		/*
2265 		 * If we haven't been here before, initialize our cryptographic
2266 		 * hash secret.
2267 		 */
2268 		RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init);
2269 
2270 		/*
2271 		 * Compute the base value of the ISS.  It is a hash
2272 		 * of (saddr, sport, daddr, dport, secret).
2273 		 */
2274 		MD5Init(&ctx);
2275 
2276 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2277 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2278 
2279 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2280 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2281 
2282 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2283 
2284 		MD5Final(hash, &ctx);
2285 
2286 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2287 
2288 		/*
2289 		 * Now increment our "timer", and add it in to
2290 		 * the computed value.
2291 		 *
2292 		 * XXX Use `addin'?
2293 		 * XXX TCP_ISSINCR too large to use?
2294 		 */
2295 		tcp_iss_seq += TCP_ISSINCR;
2296 #ifdef TCPISS_DEBUG
2297 		printf("ISS hash 0x%08x, ", tcp_iss);
2298 #endif
2299 		tcp_iss += tcp_iss_seq + addin;
2300 #ifdef TCPISS_DEBUG
2301 		printf("new ISS 0x%08x\n", tcp_iss);
2302 #endif
2303 	} else {
2304 		/*
2305 		 * Randomize.
2306 		 */
2307 		tcp_iss = cprng_fast32();
2308 
2309 		/*
2310 		 * If we were asked to add some amount to a known value,
2311 		 * we will take a random value obtained above, mask off
2312 		 * the upper bits, and add in the known value.  We also
2313 		 * add in a constant to ensure that we are at least a
2314 		 * certain distance from the original value.
2315 		 *
2316 		 * This is used when an old connection is in timed wait
2317 		 * and we have a new one coming in, for instance.
2318 		 */
2319 		if (addin != 0) {
2320 #ifdef TCPISS_DEBUG
2321 			printf("Random %08x, ", tcp_iss);
2322 #endif
2323 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2324 			tcp_iss += addin + TCP_ISSINCR;
2325 #ifdef TCPISS_DEBUG
2326 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2327 #endif
2328 		} else {
2329 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2330 			tcp_iss += tcp_iss_seq;
2331 			tcp_iss_seq += TCP_ISSINCR;
2332 #ifdef TCPISS_DEBUG
2333 			printf("ISS %08x\n", tcp_iss);
2334 #endif
2335 		}
2336 	}
2337 
2338 	return (tcp_iss);
2339 }
2340 
2341 #if defined(IPSEC)
2342 /* compute ESP/AH header size for TCP, including outer IP header. */
2343 size_t
2344 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2345 {
2346 	struct inpcb *inp;
2347 	size_t hdrsiz;
2348 
2349 	/* XXX mapped addr case (tp->t_in6pcb) */
2350 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2351 		return 0;
2352 	switch (tp->t_family) {
2353 	case AF_INET:
2354 		/* XXX: should use currect direction. */
2355 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2356 		break;
2357 	default:
2358 		hdrsiz = 0;
2359 		break;
2360 	}
2361 
2362 	return hdrsiz;
2363 }
2364 
2365 #ifdef INET6
2366 size_t
2367 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2368 {
2369 	struct in6pcb *in6p;
2370 	size_t hdrsiz;
2371 
2372 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2373 		return 0;
2374 	switch (tp->t_family) {
2375 	case AF_INET6:
2376 		/* XXX: should use currect direction. */
2377 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2378 		break;
2379 	case AF_INET:
2380 		/* mapped address case - tricky */
2381 	default:
2382 		hdrsiz = 0;
2383 		break;
2384 	}
2385 
2386 	return hdrsiz;
2387 }
2388 #endif
2389 #endif /*IPSEC*/
2390 
2391 /*
2392  * Determine the length of the TCP options for this connection.
2393  *
2394  * XXX:  What do we do for SACK, when we add that?  Just reserve
2395  *       all of the space?  Otherwise we can't exactly be incrementing
2396  *       cwnd by an amount that varies depending on the amount we last
2397  *       had to SACK!
2398  */
2399 
2400 u_int
2401 tcp_optlen(struct tcpcb *tp)
2402 {
2403 	u_int optlen;
2404 
2405 	optlen = 0;
2406 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2407 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2408 		optlen += TCPOLEN_TSTAMP_APPA;
2409 
2410 #ifdef TCP_SIGNATURE
2411 	if (tp->t_flags & TF_SIGNATURE)
2412 		optlen += TCPOLEN_SIGLEN;
2413 #endif /* TCP_SIGNATURE */
2414 
2415 	return optlen;
2416 }
2417 
2418 u_int
2419 tcp_hdrsz(struct tcpcb *tp)
2420 {
2421 	u_int hlen;
2422 
2423 	switch (tp->t_family) {
2424 #ifdef INET6
2425 	case AF_INET6:
2426 		hlen = sizeof(struct ip6_hdr);
2427 		break;
2428 #endif
2429 	case AF_INET:
2430 		hlen = sizeof(struct ip);
2431 		break;
2432 	default:
2433 		hlen = 0;
2434 		break;
2435 	}
2436 	hlen += sizeof(struct tcphdr);
2437 
2438 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2439 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2440 		hlen += TCPOLEN_TSTAMP_APPA;
2441 #ifdef TCP_SIGNATURE
2442 	if (tp->t_flags & TF_SIGNATURE)
2443 		hlen += TCPOLEN_SIGLEN;
2444 #endif
2445 	return hlen;
2446 }
2447 
2448 void
2449 tcp_statinc(u_int stat)
2450 {
2451 
2452 	KASSERT(stat < TCP_NSTATS);
2453 	TCP_STATINC(stat);
2454 }
2455 
2456 void
2457 tcp_statadd(u_int stat, uint64_t val)
2458 {
2459 
2460 	KASSERT(stat < TCP_NSTATS);
2461 	TCP_STATADD(stat, val);
2462 }
2463