xref: /netbsd-src/sys/netinet/tcp_subr.c (revision f876c1012ec8280866d3287842a50de849d25967)
1 /*	$NetBSD: tcp_subr.c,v 1.189 2005/04/05 01:07:17 kurahone 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 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  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. Neither the name of the University nor the names of its contributors
82  *    may be used to endorse or promote products derived from this software
83  *    without specific prior written permission.
84  *
85  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
86  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
87  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
88  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
89  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
90  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
91  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
92  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
93  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
94  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
95  * SUCH DAMAGE.
96  *
97  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
98  */
99 
100 #include <sys/cdefs.h>
101 __KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.189 2005/04/05 01:07:17 kurahone Exp $");
102 
103 #include "opt_inet.h"
104 #include "opt_ipsec.h"
105 #include "opt_tcp_compat_42.h"
106 #include "opt_inet_csum.h"
107 #include "opt_mbuftrace.h"
108 #include "rnd.h"
109 
110 #include <sys/param.h>
111 #include <sys/proc.h>
112 #include <sys/systm.h>
113 #include <sys/malloc.h>
114 #include <sys/mbuf.h>
115 #include <sys/socket.h>
116 #include <sys/socketvar.h>
117 #include <sys/protosw.h>
118 #include <sys/errno.h>
119 #include <sys/kernel.h>
120 #include <sys/pool.h>
121 #if NRND > 0
122 #include <sys/md5.h>
123 #include <sys/rnd.h>
124 #endif
125 
126 #include <net/route.h>
127 #include <net/if.h>
128 
129 #include <netinet/in.h>
130 #include <netinet/in_systm.h>
131 #include <netinet/ip.h>
132 #include <netinet/in_pcb.h>
133 #include <netinet/ip_var.h>
134 #include <netinet/ip_icmp.h>
135 
136 #ifdef INET6
137 #ifndef INET
138 #include <netinet/in.h>
139 #endif
140 #include <netinet/ip6.h>
141 #include <netinet6/in6_pcb.h>
142 #include <netinet6/ip6_var.h>
143 #include <netinet6/in6_var.h>
144 #include <netinet6/ip6protosw.h>
145 #include <netinet/icmp6.h>
146 #include <netinet6/nd6.h>
147 #endif
148 
149 #include <netinet/tcp.h>
150 #include <netinet/tcp_fsm.h>
151 #include <netinet/tcp_seq.h>
152 #include <netinet/tcp_timer.h>
153 #include <netinet/tcp_var.h>
154 #include <netinet/tcpip.h>
155 
156 #ifdef IPSEC
157 #include <netinet6/ipsec.h>
158 #include <netkey/key.h>
159 #endif /*IPSEC*/
160 
161 #ifdef FAST_IPSEC
162 #include <netipsec/ipsec.h>
163 #include <netipsec/xform.h>
164 #ifdef INET6
165 #include <netipsec/ipsec6.h>
166 #endif
167  #include <netipsec/key.h>
168 #endif	/* FAST_IPSEC*/
169 
170 
171 struct	inpcbtable tcbtable;	/* head of queue of active tcpcb's */
172 struct	tcpstat tcpstat;	/* tcp statistics */
173 u_int32_t tcp_now;		/* for RFC 1323 timestamps */
174 
175 /* patchable/settable parameters for tcp */
176 int 	tcp_mssdflt = TCP_MSS;
177 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
178 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
179 #if NRND > 0
180 int	tcp_do_rfc1948 = 0;	/* ISS by cryptographic hash */
181 #endif
182 int	tcp_do_sack = 1;	/* selective acknowledgement */
183 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
184 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
185 int	tcp_do_newreno = 1;	/* Use the New Reno algorithms */
186 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
187 #ifndef TCP_INIT_WIN
188 #define	TCP_INIT_WIN	0	/* initial slow start window */
189 #endif
190 #ifndef TCP_INIT_WIN_LOCAL
191 #define	TCP_INIT_WIN_LOCAL 4	/* initial slow start window for local nets */
192 #endif
193 int	tcp_init_win = TCP_INIT_WIN;
194 int	tcp_init_win_local = TCP_INIT_WIN_LOCAL;
195 int	tcp_mss_ifmtu = 0;
196 #ifdef TCP_COMPAT_42
197 int	tcp_compat_42 = 1;
198 #else
199 int	tcp_compat_42 = 0;
200 #endif
201 int	tcp_rst_ppslim = 100;	/* 100pps */
202 int	tcp_ackdrop_ppslim = 100;	/* 100pps */
203 int	tcp_do_loopback_cksum = 0;
204 int	tcp_sack_tp_maxholes = 32;
205 int	tcp_sack_globalmaxholes = 1024;
206 int	tcp_sack_globalholes = 0;
207 
208 
209 /* tcb hash */
210 #ifndef TCBHASHSIZE
211 #define	TCBHASHSIZE	128
212 #endif
213 int	tcbhashsize = TCBHASHSIZE;
214 
215 /* syn hash parameters */
216 #define	TCP_SYN_HASH_SIZE	293
217 #define	TCP_SYN_BUCKET_SIZE	35
218 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
219 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
220 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
221 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
222 
223 int	tcp_freeq(struct tcpcb *);
224 
225 #ifdef INET
226 void	tcp_mtudisc_callback(struct in_addr);
227 #endif
228 #ifdef INET6
229 void	tcp6_mtudisc_callback(struct in6_addr *);
230 #endif
231 
232 void	tcp_mtudisc(struct inpcb *, int);
233 #ifdef INET6
234 void	tcp6_mtudisc(struct in6pcb *, int);
235 #endif
236 
237 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL);
238 
239 #ifdef TCP_CSUM_COUNTERS
240 #include <sys/device.h>
241 
242 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243     NULL, "tcp", "hwcsum bad");
244 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245     NULL, "tcp", "hwcsum ok");
246 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247     NULL, "tcp", "hwcsum data");
248 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
249     NULL, "tcp", "swcsum");
250 
251 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
252 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
253 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
254 EVCNT_ATTACH_STATIC(tcp_swcsum);
255 #endif /* TCP_CSUM_COUNTERS */
256 
257 
258 #ifdef TCP_OUTPUT_COUNTERS
259 #include <sys/device.h>
260 
261 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262     NULL, "tcp", "output big header");
263 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264     NULL, "tcp", "output predict hit");
265 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266     NULL, "tcp", "output predict miss");
267 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268     NULL, "tcp", "output copy small");
269 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
270     NULL, "tcp", "output copy big");
271 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
272     NULL, "tcp", "output reference big");
273 
274 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
275 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
276 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
277 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
278 EVCNT_ATTACH_STATIC(tcp_output_copybig);
279 EVCNT_ATTACH_STATIC(tcp_output_refbig);
280 
281 #endif /* TCP_OUTPUT_COUNTERS */
282 
283 #ifdef TCP_REASS_COUNTERS
284 #include <sys/device.h>
285 
286 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
287     NULL, "tcp_reass", "calls");
288 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
289     &tcp_reass_, "tcp_reass", "insert into empty queue");
290 struct evcnt tcp_reass_iteration[8] = {
291     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
292     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
293     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
294     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
295     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
296     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
297     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
298     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
299 };
300 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
301     &tcp_reass_, "tcp_reass", "prepend to first");
302 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303     &tcp_reass_, "tcp_reass", "prepend");
304 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305     &tcp_reass_, "tcp_reass", "insert");
306 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
307     &tcp_reass_, "tcp_reass", "insert at tail");
308 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
309     &tcp_reass_, "tcp_reass", "append");
310 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
311     &tcp_reass_, "tcp_reass", "append to tail fragment");
312 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
313     &tcp_reass_, "tcp_reass", "overlap at end");
314 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
315     &tcp_reass_, "tcp_reass", "overlap at start");
316 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
317     &tcp_reass_, "tcp_reass", "duplicate segment");
318 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
319     &tcp_reass_, "tcp_reass", "duplicate fragment");
320 
321 EVCNT_ATTACH_STATIC(tcp_reass_);
322 EVCNT_ATTACH_STATIC(tcp_reass_empty);
323 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
324 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
325 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
326 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
327 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
328 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
329 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
330 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
331 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
332 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
333 EVCNT_ATTACH_STATIC(tcp_reass_insert);
334 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
335 EVCNT_ATTACH_STATIC(tcp_reass_append);
336 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
337 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
338 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
339 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
340 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
341 
342 #endif /* TCP_REASS_COUNTERS */
343 
344 #ifdef MBUFTRACE
345 struct mowner tcp_mowner = { "tcp" };
346 struct mowner tcp_rx_mowner = { "tcp", "rx" };
347 struct mowner tcp_tx_mowner = { "tcp", "tx" };
348 #endif
349 
350 /*
351  * Tcp initialization
352  */
353 void
354 tcp_init(void)
355 {
356 	int hlen;
357 
358 	/* Initialize the TCPCB template. */
359 	tcp_tcpcb_template();
360 
361 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
362 
363 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
364 #ifdef INET6
365 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
366 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
367 #endif
368 	if (max_protohdr < hlen)
369 		max_protohdr = hlen;
370 	if (max_linkhdr + hlen > MHLEN)
371 		panic("tcp_init");
372 
373 #ifdef INET
374 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
375 #endif
376 #ifdef INET6
377 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
378 #endif
379 
380 	/* Initialize timer state. */
381 	tcp_timer_init();
382 
383 	/* Initialize the compressed state engine. */
384 	syn_cache_init();
385 
386 	MOWNER_ATTACH(&tcp_tx_mowner);
387 	MOWNER_ATTACH(&tcp_rx_mowner);
388 	MOWNER_ATTACH(&tcp_mowner);
389 }
390 
391 /*
392  * Create template to be used to send tcp packets on a connection.
393  * Call after host entry created, allocates an mbuf and fills
394  * in a skeletal tcp/ip header, minimizing the amount of work
395  * necessary when the connection is used.
396  */
397 struct mbuf *
398 tcp_template(struct tcpcb *tp)
399 {
400 	struct inpcb *inp = tp->t_inpcb;
401 #ifdef INET6
402 	struct in6pcb *in6p = tp->t_in6pcb;
403 #endif
404 	struct tcphdr *n;
405 	struct mbuf *m;
406 	int hlen;
407 
408 	switch (tp->t_family) {
409 	case AF_INET:
410 		hlen = sizeof(struct ip);
411 		if (inp)
412 			break;
413 #ifdef INET6
414 		if (in6p) {
415 			/* mapped addr case */
416 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
417 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
418 				break;
419 		}
420 #endif
421 		return NULL;	/*EINVAL*/
422 #ifdef INET6
423 	case AF_INET6:
424 		hlen = sizeof(struct ip6_hdr);
425 		if (in6p) {
426 			/* more sainty check? */
427 			break;
428 		}
429 		return NULL;	/*EINVAL*/
430 #endif
431 	default:
432 		hlen = 0;	/*pacify gcc*/
433 		return NULL;	/*EAFNOSUPPORT*/
434 	}
435 #ifdef DIAGNOSTIC
436 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
437 		panic("mclbytes too small for t_template");
438 #endif
439 	m = tp->t_template;
440 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
441 		;
442 	else {
443 		if (m)
444 			m_freem(m);
445 		m = tp->t_template = NULL;
446 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
447 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
448 			MCLGET(m, M_DONTWAIT);
449 			if ((m->m_flags & M_EXT) == 0) {
450 				m_free(m);
451 				m = NULL;
452 			}
453 		}
454 		if (m == NULL)
455 			return NULL;
456 		MCLAIM(m, &tcp_mowner);
457 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
458 	}
459 
460 	bzero(mtod(m, caddr_t), m->m_len);
461 
462 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
463 
464 	switch (tp->t_family) {
465 	case AF_INET:
466 	    {
467 		struct ipovly *ipov;
468 		mtod(m, struct ip *)->ip_v = 4;
469 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
470 		ipov = mtod(m, struct ipovly *);
471 		ipov->ih_pr = IPPROTO_TCP;
472 		ipov->ih_len = htons(sizeof(struct tcphdr));
473 		if (inp) {
474 			ipov->ih_src = inp->inp_laddr;
475 			ipov->ih_dst = inp->inp_faddr;
476 		}
477 #ifdef INET6
478 		else if (in6p) {
479 			/* mapped addr case */
480 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
481 				sizeof(ipov->ih_src));
482 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
483 				sizeof(ipov->ih_dst));
484 		}
485 #endif
486 		/*
487 		 * Compute the pseudo-header portion of the checksum
488 		 * now.  We incrementally add in the TCP option and
489 		 * payload lengths later, and then compute the TCP
490 		 * checksum right before the packet is sent off onto
491 		 * the wire.
492 		 */
493 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
494 		    ipov->ih_dst.s_addr,
495 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
496 		break;
497 	    }
498 #ifdef INET6
499 	case AF_INET6:
500 	    {
501 		struct ip6_hdr *ip6;
502 		mtod(m, struct ip *)->ip_v = 6;
503 		ip6 = mtod(m, struct ip6_hdr *);
504 		ip6->ip6_nxt = IPPROTO_TCP;
505 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
506 		ip6->ip6_src = in6p->in6p_laddr;
507 		ip6->ip6_dst = in6p->in6p_faddr;
508 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
509 		if (ip6_auto_flowlabel) {
510 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
511 			ip6->ip6_flow |=
512 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
513 		}
514 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
515 		ip6->ip6_vfc |= IPV6_VERSION;
516 
517 		/*
518 		 * Compute the pseudo-header portion of the checksum
519 		 * now.  We incrementally add in the TCP option and
520 		 * payload lengths later, and then compute the TCP
521 		 * checksum right before the packet is sent off onto
522 		 * the wire.
523 		 */
524 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
525 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
526 		    htonl(IPPROTO_TCP));
527 		break;
528 	    }
529 #endif
530 	}
531 	if (inp) {
532 		n->th_sport = inp->inp_lport;
533 		n->th_dport = inp->inp_fport;
534 	}
535 #ifdef INET6
536 	else if (in6p) {
537 		n->th_sport = in6p->in6p_lport;
538 		n->th_dport = in6p->in6p_fport;
539 	}
540 #endif
541 	n->th_seq = 0;
542 	n->th_ack = 0;
543 	n->th_x2 = 0;
544 	n->th_off = 5;
545 	n->th_flags = 0;
546 	n->th_win = 0;
547 	n->th_urp = 0;
548 	return (m);
549 }
550 
551 /*
552  * Send a single message to the TCP at address specified by
553  * the given TCP/IP header.  If m == 0, then we make a copy
554  * of the tcpiphdr at ti and send directly to the addressed host.
555  * This is used to force keep alive messages out using the TCP
556  * template for a connection tp->t_template.  If flags are given
557  * then we send a message back to the TCP which originated the
558  * segment ti, and discard the mbuf containing it and any other
559  * attached mbufs.
560  *
561  * In any case the ack and sequence number of the transmitted
562  * segment are as specified by the parameters.
563  */
564 int
565 tcp_respond(struct tcpcb *tp, struct mbuf *template, struct mbuf *m,
566     struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
567 {
568 	struct route *ro;
569 	int error, tlen, win = 0;
570 	int hlen;
571 	struct ip *ip;
572 #ifdef INET6
573 	struct ip6_hdr *ip6;
574 #endif
575 	int family;	/* family on packet, not inpcb/in6pcb! */
576 	struct tcphdr *th;
577 	struct socket *so;
578 
579 	if (tp != NULL && (flags & TH_RST) == 0) {
580 #ifdef DIAGNOSTIC
581 		if (tp->t_inpcb && tp->t_in6pcb)
582 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
583 #endif
584 #ifdef INET
585 		if (tp->t_inpcb)
586 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
587 #endif
588 #ifdef INET6
589 		if (tp->t_in6pcb)
590 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
591 #endif
592 	}
593 
594 	th = NULL;	/* Quell uninitialized warning */
595 	ip = NULL;
596 #ifdef INET6
597 	ip6 = NULL;
598 #endif
599 	if (m == 0) {
600 		if (!template)
601 			return EINVAL;
602 
603 		/* get family information from template */
604 		switch (mtod(template, struct ip *)->ip_v) {
605 		case 4:
606 			family = AF_INET;
607 			hlen = sizeof(struct ip);
608 			break;
609 #ifdef INET6
610 		case 6:
611 			family = AF_INET6;
612 			hlen = sizeof(struct ip6_hdr);
613 			break;
614 #endif
615 		default:
616 			return EAFNOSUPPORT;
617 		}
618 
619 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
620 		if (m) {
621 			MCLAIM(m, &tcp_tx_mowner);
622 			MCLGET(m, M_DONTWAIT);
623 			if ((m->m_flags & M_EXT) == 0) {
624 				m_free(m);
625 				m = NULL;
626 			}
627 		}
628 		if (m == NULL)
629 			return (ENOBUFS);
630 
631 		if (tcp_compat_42)
632 			tlen = 1;
633 		else
634 			tlen = 0;
635 
636 		m->m_data += max_linkhdr;
637 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
638 			template->m_len);
639 		switch (family) {
640 		case AF_INET:
641 			ip = mtod(m, struct ip *);
642 			th = (struct tcphdr *)(ip + 1);
643 			break;
644 #ifdef INET6
645 		case AF_INET6:
646 			ip6 = mtod(m, struct ip6_hdr *);
647 			th = (struct tcphdr *)(ip6 + 1);
648 			break;
649 #endif
650 #if 0
651 		default:
652 			/* noone will visit here */
653 			m_freem(m);
654 			return EAFNOSUPPORT;
655 #endif
656 		}
657 		flags = TH_ACK;
658 	} else {
659 
660 		if ((m->m_flags & M_PKTHDR) == 0) {
661 #if 0
662 			printf("non PKTHDR to tcp_respond\n");
663 #endif
664 			m_freem(m);
665 			return EINVAL;
666 		}
667 #ifdef DIAGNOSTIC
668 		if (!th0)
669 			panic("th0 == NULL in tcp_respond");
670 #endif
671 
672 		/* get family information from m */
673 		switch (mtod(m, struct ip *)->ip_v) {
674 		case 4:
675 			family = AF_INET;
676 			hlen = sizeof(struct ip);
677 			ip = mtod(m, struct ip *);
678 			break;
679 #ifdef INET6
680 		case 6:
681 			family = AF_INET6;
682 			hlen = sizeof(struct ip6_hdr);
683 			ip6 = mtod(m, struct ip6_hdr *);
684 			break;
685 #endif
686 		default:
687 			m_freem(m);
688 			return EAFNOSUPPORT;
689 		}
690 		/* clear h/w csum flags inherited from rx packet */
691 		m->m_pkthdr.csum_flags = 0;
692 
693 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
694 			tlen = sizeof(*th0);
695 		else
696 			tlen = th0->th_off << 2;
697 
698 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
699 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
700 			m->m_len = hlen + tlen;
701 			m_freem(m->m_next);
702 			m->m_next = NULL;
703 		} else {
704 			struct mbuf *n;
705 
706 #ifdef DIAGNOSTIC
707 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
708 				m_freem(m);
709 				return EMSGSIZE;
710 			}
711 #endif
712 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
713 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
714 				MCLGET(n, M_DONTWAIT);
715 				if ((n->m_flags & M_EXT) == 0) {
716 					m_freem(n);
717 					n = NULL;
718 				}
719 			}
720 			if (!n) {
721 				m_freem(m);
722 				return ENOBUFS;
723 			}
724 
725 			MCLAIM(n, &tcp_tx_mowner);
726 			n->m_data += max_linkhdr;
727 			n->m_len = hlen + tlen;
728 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
729 			m_copyback(n, hlen, tlen, (caddr_t)th0);
730 
731 			m_freem(m);
732 			m = n;
733 			n = NULL;
734 		}
735 
736 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
737 		switch (family) {
738 		case AF_INET:
739 			ip = mtod(m, struct ip *);
740 			th = (struct tcphdr *)(ip + 1);
741 			ip->ip_p = IPPROTO_TCP;
742 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
743 			ip->ip_p = IPPROTO_TCP;
744 			break;
745 #ifdef INET6
746 		case AF_INET6:
747 			ip6 = mtod(m, struct ip6_hdr *);
748 			th = (struct tcphdr *)(ip6 + 1);
749 			ip6->ip6_nxt = IPPROTO_TCP;
750 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
751 			ip6->ip6_nxt = IPPROTO_TCP;
752 			break;
753 #endif
754 #if 0
755 		default:
756 			/* noone will visit here */
757 			m_freem(m);
758 			return EAFNOSUPPORT;
759 #endif
760 		}
761 		xchg(th->th_dport, th->th_sport, u_int16_t);
762 #undef xchg
763 		tlen = 0;	/*be friendly with the following code*/
764 	}
765 	th->th_seq = htonl(seq);
766 	th->th_ack = htonl(ack);
767 	th->th_x2 = 0;
768 	if ((flags & TH_SYN) == 0) {
769 		if (tp)
770 			win >>= tp->rcv_scale;
771 		if (win > TCP_MAXWIN)
772 			win = TCP_MAXWIN;
773 		th->th_win = htons((u_int16_t)win);
774 		th->th_off = sizeof (struct tcphdr) >> 2;
775 		tlen += sizeof(*th);
776 	} else
777 		tlen += th->th_off << 2;
778 	m->m_len = hlen + tlen;
779 	m->m_pkthdr.len = hlen + tlen;
780 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
781 	th->th_flags = flags;
782 	th->th_urp = 0;
783 
784 	switch (family) {
785 #ifdef INET
786 	case AF_INET:
787 	    {
788 		struct ipovly *ipov = (struct ipovly *)ip;
789 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
790 		ipov->ih_len = htons((u_int16_t)tlen);
791 
792 		th->th_sum = 0;
793 		th->th_sum = in_cksum(m, hlen + tlen);
794 		ip->ip_len = htons(hlen + tlen);
795 		ip->ip_ttl = ip_defttl;
796 		break;
797 	    }
798 #endif
799 #ifdef INET6
800 	case AF_INET6:
801 	    {
802 		th->th_sum = 0;
803 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
804 				tlen);
805 		ip6->ip6_plen = htons(tlen);
806 		if (tp && tp->t_in6pcb) {
807 			struct ifnet *oifp;
808 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
809 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
810 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
811 		} else
812 			ip6->ip6_hlim = ip6_defhlim;
813 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
814 		if (ip6_auto_flowlabel) {
815 			ip6->ip6_flow |=
816 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
817 		}
818 		break;
819 	    }
820 #endif
821 	}
822 
823 	if (tp && tp->t_inpcb)
824 		so = tp->t_inpcb->inp_socket;
825 #ifdef INET6
826 	else if (tp && tp->t_in6pcb)
827 		so = tp->t_in6pcb->in6p_socket;
828 #endif
829 	else
830 		so = NULL;
831 
832 	if (tp != NULL && tp->t_inpcb != NULL) {
833 		ro = &tp->t_inpcb->inp_route;
834 #ifdef DIAGNOSTIC
835 		if (family != AF_INET)
836 			panic("tcp_respond: address family mismatch");
837 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
838 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
839 			    ntohl(ip->ip_dst.s_addr),
840 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
841 		}
842 #endif
843 	}
844 #ifdef INET6
845 	else if (tp != NULL && tp->t_in6pcb != NULL) {
846 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
847 #ifdef DIAGNOSTIC
848 		if (family == AF_INET) {
849 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
850 				panic("tcp_respond: not mapped addr");
851 			if (bcmp(&ip->ip_dst,
852 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
853 			    sizeof(ip->ip_dst)) != 0) {
854 				panic("tcp_respond: ip_dst != in6p_faddr");
855 			}
856 		} else if (family == AF_INET6) {
857 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
858 			    &tp->t_in6pcb->in6p_faddr))
859 				panic("tcp_respond: ip6_dst != in6p_faddr");
860 		} else
861 			panic("tcp_respond: address family mismatch");
862 #endif
863 	}
864 #endif
865 	else
866 		ro = NULL;
867 
868 	switch (family) {
869 #ifdef INET
870 	case AF_INET:
871 		error = ip_output(m, NULL, ro,
872 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
873 		    (struct ip_moptions *)0, so);
874 		break;
875 #endif
876 #ifdef INET6
877 	case AF_INET6:
878 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
879 		    (struct ip6_moptions *)0, so, NULL);
880 		break;
881 #endif
882 	default:
883 		error = EAFNOSUPPORT;
884 		break;
885 	}
886 
887 	return (error);
888 }
889 
890 /*
891  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
892  * a bunch of members individually, we maintain this template for the
893  * static and mostly-static components of the TCPCB, and copy it into
894  * the new TCPCB instead.
895  */
896 static struct tcpcb tcpcb_template = {
897 	/*
898 	 * If TCP_NTIMERS ever changes, we'll need to update this
899 	 * initializer.
900 	 */
901 	.t_timer = {
902 		CALLOUT_INITIALIZER,
903 		CALLOUT_INITIALIZER,
904 		CALLOUT_INITIALIZER,
905 		CALLOUT_INITIALIZER,
906 	},
907 	.t_delack_ch = CALLOUT_INITIALIZER,
908 
909 	.t_srtt = TCPTV_SRTTBASE,
910 	.t_rttmin = TCPTV_MIN,
911 
912 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
913 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
914 	.snd_numholes = 0,
915 
916 	.t_partialacks = -1,
917 };
918 
919 /*
920  * Updates the TCPCB template whenever a parameter that would affect
921  * the template is changed.
922  */
923 void
924 tcp_tcpcb_template(void)
925 {
926 	struct tcpcb *tp = &tcpcb_template;
927 	int flags;
928 
929 	tp->t_peermss = tcp_mssdflt;
930 	tp->t_ourmss = tcp_mssdflt;
931 	tp->t_segsz = tcp_mssdflt;
932 
933 	flags = 0;
934 	if (tcp_do_rfc1323 && tcp_do_win_scale)
935 		flags |= TF_REQ_SCALE;
936 	if (tcp_do_rfc1323 && tcp_do_timestamps)
937 		flags |= TF_REQ_TSTMP;
938 	tp->t_flags = flags;
939 
940 	/*
941 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
942 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
943 	 * reasonable initial retransmit time.
944 	 */
945 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
946 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
947 	    TCPTV_MIN, TCPTV_REXMTMAX);
948 }
949 
950 /*
951  * Create a new TCP control block, making an
952  * empty reassembly queue and hooking it to the argument
953  * protocol control block.
954  */
955 /* family selects inpcb, or in6pcb */
956 struct tcpcb *
957 tcp_newtcpcb(int family, void *aux)
958 {
959 	struct tcpcb *tp;
960 	int i;
961 
962 	/* XXX Consider using a pool_cache for speed. */
963 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
964 	if (tp == NULL)
965 		return (NULL);
966 	memcpy(tp, &tcpcb_template, sizeof(*tp));
967 	TAILQ_INIT(&tp->segq);
968 	TAILQ_INIT(&tp->timeq);
969 	tp->t_family = family;		/* may be overridden later on */
970 	TAILQ_INIT(&tp->snd_holes);
971 	LIST_INIT(&tp->t_sc);		/* XXX can template this */
972 
973 	/* Don't sweat this loop; hopefully the compiler will unroll it. */
974 	for (i = 0; i < TCPT_NTIMERS; i++)
975 		TCP_TIMER_INIT(tp, i);
976 
977 	switch (family) {
978 	case AF_INET:
979 	    {
980 		struct inpcb *inp = (struct inpcb *)aux;
981 
982 		inp->inp_ip.ip_ttl = ip_defttl;
983 		inp->inp_ppcb = (caddr_t)tp;
984 
985 		tp->t_inpcb = inp;
986 		tp->t_mtudisc = ip_mtudisc;
987 		break;
988 	    }
989 #ifdef INET6
990 	case AF_INET6:
991 	    {
992 		struct in6pcb *in6p = (struct in6pcb *)aux;
993 
994 		in6p->in6p_ip6.ip6_hlim = in6_selecthlim(in6p,
995 			in6p->in6p_route.ro_rt ? in6p->in6p_route.ro_rt->rt_ifp
996 					       : NULL);
997 		in6p->in6p_ppcb = (caddr_t)tp;
998 
999 		tp->t_in6pcb = in6p;
1000 		/* for IPv6, always try to run path MTU discovery */
1001 		tp->t_mtudisc = 1;
1002 		break;
1003 	    }
1004 #endif /* INET6 */
1005 	default:
1006 		pool_put(&tcpcb_pool, tp);
1007 		return (NULL);
1008 	}
1009 
1010 	/*
1011 	 * Initialize our timebase.  When we send timestamps, we take
1012 	 * the delta from tcp_now -- this means each connection always
1013 	 * gets a timebase of 0, which makes it, among other things,
1014 	 * more difficult to determine how long a system has been up,
1015 	 * and thus how many TCP sequence increments have occurred.
1016 	 */
1017 	tp->ts_timebase = tcp_now;
1018 
1019 	return (tp);
1020 }
1021 
1022 /*
1023  * Drop a TCP connection, reporting
1024  * the specified error.  If connection is synchronized,
1025  * then send a RST to peer.
1026  */
1027 struct tcpcb *
1028 tcp_drop(struct tcpcb *tp, int errno)
1029 {
1030 	struct socket *so = NULL;
1031 
1032 #ifdef DIAGNOSTIC
1033 	if (tp->t_inpcb && tp->t_in6pcb)
1034 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1035 #endif
1036 #ifdef INET
1037 	if (tp->t_inpcb)
1038 		so = tp->t_inpcb->inp_socket;
1039 #endif
1040 #ifdef INET6
1041 	if (tp->t_in6pcb)
1042 		so = tp->t_in6pcb->in6p_socket;
1043 #endif
1044 	if (!so)
1045 		return NULL;
1046 
1047 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1048 		tp->t_state = TCPS_CLOSED;
1049 		(void) tcp_output(tp);
1050 		tcpstat.tcps_drops++;
1051 	} else
1052 		tcpstat.tcps_conndrops++;
1053 	if (errno == ETIMEDOUT && tp->t_softerror)
1054 		errno = tp->t_softerror;
1055 	so->so_error = errno;
1056 	return (tcp_close(tp));
1057 }
1058 
1059 /*
1060  * Return whether this tcpcb is marked as dead, indicating
1061  * to the calling timer function that no further action should
1062  * be taken, as we are about to release this tcpcb.  The release
1063  * of the storage will be done if this is the last timer running.
1064  *
1065  * This should be called from the callout handler function after
1066  * callout_ack() is done, so that the number of invoking timer
1067  * functions is 0.
1068  */
1069 int
1070 tcp_isdead(struct tcpcb *tp)
1071 {
1072 	int dead = (tp->t_flags & TF_DEAD);
1073 
1074 	if (__predict_false(dead)) {
1075 		if (tcp_timers_invoking(tp) > 0)
1076 				/* not quite there yet -- count separately? */
1077 			return dead;
1078 		tcpstat.tcps_delayed_free++;
1079 		pool_put(&tcpcb_pool, tp);
1080 	}
1081 	return dead;
1082 }
1083 
1084 /*
1085  * Close a TCP control block:
1086  *	discard all space held by the tcp
1087  *	discard internet protocol block
1088  *	wake up any sleepers
1089  */
1090 struct tcpcb *
1091 tcp_close(struct tcpcb *tp)
1092 {
1093 	struct inpcb *inp;
1094 #ifdef INET6
1095 	struct in6pcb *in6p;
1096 #endif
1097 	struct socket *so;
1098 #ifdef RTV_RTT
1099 	struct rtentry *rt;
1100 #endif
1101 	struct route *ro;
1102 
1103 	inp = tp->t_inpcb;
1104 #ifdef INET6
1105 	in6p = tp->t_in6pcb;
1106 #endif
1107 	so = NULL;
1108 	ro = NULL;
1109 	if (inp) {
1110 		so = inp->inp_socket;
1111 		ro = &inp->inp_route;
1112 	}
1113 #ifdef INET6
1114 	else if (in6p) {
1115 		so = in6p->in6p_socket;
1116 		ro = (struct route *)&in6p->in6p_route;
1117 	}
1118 #endif
1119 
1120 #ifdef RTV_RTT
1121 	/*
1122 	 * If we sent enough data to get some meaningful characteristics,
1123 	 * save them in the routing entry.  'Enough' is arbitrarily
1124 	 * defined as the sendpipesize (default 4K) * 16.  This would
1125 	 * give us 16 rtt samples assuming we only get one sample per
1126 	 * window (the usual case on a long haul net).  16 samples is
1127 	 * enough for the srtt filter to converge to within 5% of the correct
1128 	 * value; fewer samples and we could save a very bogus rtt.
1129 	 *
1130 	 * Don't update the default route's characteristics and don't
1131 	 * update anything that the user "locked".
1132 	 */
1133 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1134 	    ro && (rt = ro->ro_rt) &&
1135 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1136 		u_long i = 0;
1137 
1138 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1139 			i = tp->t_srtt *
1140 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1141 			if (rt->rt_rmx.rmx_rtt && i)
1142 				/*
1143 				 * filter this update to half the old & half
1144 				 * the new values, converting scale.
1145 				 * See route.h and tcp_var.h for a
1146 				 * description of the scaling constants.
1147 				 */
1148 				rt->rt_rmx.rmx_rtt =
1149 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1150 			else
1151 				rt->rt_rmx.rmx_rtt = i;
1152 		}
1153 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1154 			i = tp->t_rttvar *
1155 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1156 			if (rt->rt_rmx.rmx_rttvar && i)
1157 				rt->rt_rmx.rmx_rttvar =
1158 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1159 			else
1160 				rt->rt_rmx.rmx_rttvar = i;
1161 		}
1162 		/*
1163 		 * update the pipelimit (ssthresh) if it has been updated
1164 		 * already or if a pipesize was specified & the threshhold
1165 		 * got below half the pipesize.  I.e., wait for bad news
1166 		 * before we start updating, then update on both good
1167 		 * and bad news.
1168 		 */
1169 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1170 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1171 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1172 			/*
1173 			 * convert the limit from user data bytes to
1174 			 * packets then to packet data bytes.
1175 			 */
1176 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1177 			if (i < 2)
1178 				i = 2;
1179 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1180 			if (rt->rt_rmx.rmx_ssthresh)
1181 				rt->rt_rmx.rmx_ssthresh =
1182 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1183 			else
1184 				rt->rt_rmx.rmx_ssthresh = i;
1185 		}
1186 	}
1187 #endif /* RTV_RTT */
1188 	/* free the reassembly queue, if any */
1189 	TCP_REASS_LOCK(tp);
1190 	(void) tcp_freeq(tp);
1191 	TCP_REASS_UNLOCK(tp);
1192 
1193 	/* free the SACK holes list. */
1194 	tcp_free_sackholes(tp);
1195 
1196 	tcp_canceltimers(tp);
1197 	TCP_CLEAR_DELACK(tp);
1198 	syn_cache_cleanup(tp);
1199 
1200 	if (tp->t_template) {
1201 		m_free(tp->t_template);
1202 		tp->t_template = NULL;
1203 	}
1204 	if (tcp_timers_invoking(tp))
1205 		tp->t_flags |= TF_DEAD;
1206 	else
1207 		pool_put(&tcpcb_pool, tp);
1208 
1209 	if (inp) {
1210 		inp->inp_ppcb = 0;
1211 		soisdisconnected(so);
1212 		in_pcbdetach(inp);
1213 	}
1214 #ifdef INET6
1215 	else if (in6p) {
1216 		in6p->in6p_ppcb = 0;
1217 		soisdisconnected(so);
1218 		in6_pcbdetach(in6p);
1219 	}
1220 #endif
1221 	tcpstat.tcps_closed++;
1222 	return ((struct tcpcb *)0);
1223 }
1224 
1225 int
1226 tcp_freeq(tp)
1227 	struct tcpcb *tp;
1228 {
1229 	struct ipqent *qe;
1230 	int rv = 0;
1231 #ifdef TCPREASS_DEBUG
1232 	int i = 0;
1233 #endif
1234 
1235 	TCP_REASS_LOCK_CHECK(tp);
1236 
1237 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1238 #ifdef TCPREASS_DEBUG
1239 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1240 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1241 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1242 #endif
1243 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1244 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1245 		m_freem(qe->ipqe_m);
1246 		tcpipqent_free(qe);
1247 		rv = 1;
1248 	}
1249 	tp->t_segqlen = 0;
1250 	KASSERT(TAILQ_EMPTY(&tp->timeq));
1251 	return (rv);
1252 }
1253 
1254 /*
1255  * Protocol drain routine.  Called when memory is in short supply.
1256  */
1257 void
1258 tcp_drain(void)
1259 {
1260 	struct inpcb_hdr *inph;
1261 	struct tcpcb *tp;
1262 
1263 	/*
1264 	 * Free the sequence queue of all TCP connections.
1265 	 */
1266 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1267 		switch (inph->inph_af) {
1268 		case AF_INET:
1269 			tp = intotcpcb((struct inpcb *)inph);
1270 			break;
1271 #ifdef INET6
1272 		case AF_INET6:
1273 			tp = in6totcpcb((struct in6pcb *)inph);
1274 			break;
1275 #endif
1276 		default:
1277 			tp = NULL;
1278 			break;
1279 		}
1280 		if (tp != NULL) {
1281 			/*
1282 			 * We may be called from a device's interrupt
1283 			 * context.  If the tcpcb is already busy,
1284 			 * just bail out now.
1285 			 */
1286 			if (tcp_reass_lock_try(tp) == 0)
1287 				continue;
1288 			if (tcp_freeq(tp))
1289 				tcpstat.tcps_connsdrained++;
1290 			TCP_REASS_UNLOCK(tp);
1291 		}
1292 	}
1293 }
1294 
1295 /*
1296  * Notify a tcp user of an asynchronous error;
1297  * store error as soft error, but wake up user
1298  * (for now, won't do anything until can select for soft error).
1299  */
1300 void
1301 tcp_notify(struct inpcb *inp, int error)
1302 {
1303 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1304 	struct socket *so = inp->inp_socket;
1305 
1306 	/*
1307 	 * Ignore some errors if we are hooked up.
1308 	 * If connection hasn't completed, has retransmitted several times,
1309 	 * and receives a second error, give up now.  This is better
1310 	 * than waiting a long time to establish a connection that
1311 	 * can never complete.
1312 	 */
1313 	if (tp->t_state == TCPS_ESTABLISHED &&
1314 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1315 	      error == EHOSTDOWN)) {
1316 		return;
1317 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1318 	    tp->t_rxtshift > 3 && tp->t_softerror)
1319 		so->so_error = error;
1320 	else
1321 		tp->t_softerror = error;
1322 	wakeup((caddr_t) &so->so_timeo);
1323 	sorwakeup(so);
1324 	sowwakeup(so);
1325 }
1326 
1327 #ifdef INET6
1328 void
1329 tcp6_notify(struct in6pcb *in6p, int error)
1330 {
1331 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1332 	struct socket *so = in6p->in6p_socket;
1333 
1334 	/*
1335 	 * Ignore some errors if we are hooked up.
1336 	 * If connection hasn't completed, has retransmitted several times,
1337 	 * and receives a second error, give up now.  This is better
1338 	 * than waiting a long time to establish a connection that
1339 	 * can never complete.
1340 	 */
1341 	if (tp->t_state == TCPS_ESTABLISHED &&
1342 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1343 	      error == EHOSTDOWN)) {
1344 		return;
1345 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1346 	    tp->t_rxtshift > 3 && tp->t_softerror)
1347 		so->so_error = error;
1348 	else
1349 		tp->t_softerror = error;
1350 	wakeup((caddr_t) &so->so_timeo);
1351 	sorwakeup(so);
1352 	sowwakeup(so);
1353 }
1354 #endif
1355 
1356 #ifdef INET6
1357 void
1358 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1359 {
1360 	struct tcphdr th;
1361 	void (*notify)(struct in6pcb *, int) = tcp6_notify;
1362 	int nmatch;
1363 	struct ip6_hdr *ip6;
1364 	const struct sockaddr_in6 *sa6_src = NULL;
1365 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1366 	struct mbuf *m;
1367 	int off;
1368 
1369 	if (sa->sa_family != AF_INET6 ||
1370 	    sa->sa_len != sizeof(struct sockaddr_in6))
1371 		return;
1372 	if ((unsigned)cmd >= PRC_NCMDS)
1373 		return;
1374 	else if (cmd == PRC_QUENCH) {
1375 		/* XXX there's no PRC_QUENCH in IPv6 */
1376 		notify = tcp6_quench;
1377 	} else if (PRC_IS_REDIRECT(cmd))
1378 		notify = in6_rtchange, d = NULL;
1379 	else if (cmd == PRC_MSGSIZE)
1380 		; /* special code is present, see below */
1381 	else if (cmd == PRC_HOSTDEAD)
1382 		d = NULL;
1383 	else if (inet6ctlerrmap[cmd] == 0)
1384 		return;
1385 
1386 	/* if the parameter is from icmp6, decode it. */
1387 	if (d != NULL) {
1388 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1389 		m = ip6cp->ip6c_m;
1390 		ip6 = ip6cp->ip6c_ip6;
1391 		off = ip6cp->ip6c_off;
1392 		sa6_src = ip6cp->ip6c_src;
1393 	} else {
1394 		m = NULL;
1395 		ip6 = NULL;
1396 		sa6_src = &sa6_any;
1397 		off = 0;
1398 	}
1399 
1400 	if (ip6) {
1401 		/*
1402 		 * XXX: We assume that when ip6 is non NULL,
1403 		 * M and OFF are valid.
1404 		 */
1405 
1406 		/* check if we can safely examine src and dst ports */
1407 		if (m->m_pkthdr.len < off + sizeof(th)) {
1408 			if (cmd == PRC_MSGSIZE)
1409 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1410 			return;
1411 		}
1412 
1413 		bzero(&th, sizeof(th));
1414 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1415 
1416 		if (cmd == PRC_MSGSIZE) {
1417 			int valid = 0;
1418 
1419 			/*
1420 			 * Check to see if we have a valid TCP connection
1421 			 * corresponding to the address in the ICMPv6 message
1422 			 * payload.
1423 			 */
1424 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1425 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1426 			    th.th_sport, 0))
1427 				valid++;
1428 
1429 			/*
1430 			 * Depending on the value of "valid" and routing table
1431 			 * size (mtudisc_{hi,lo}wat), we will:
1432 			 * - recalcurate the new MTU and create the
1433 			 *   corresponding routing entry, or
1434 			 * - ignore the MTU change notification.
1435 			 */
1436 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1437 
1438 			/*
1439 			 * no need to call in6_pcbnotify, it should have been
1440 			 * called via callback if necessary
1441 			 */
1442 			return;
1443 		}
1444 
1445 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1446 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1447 		if (nmatch == 0 && syn_cache_count &&
1448 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1449 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1450 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1451 			syn_cache_unreach((struct sockaddr *)sa6_src,
1452 					  sa, &th);
1453 	} else {
1454 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1455 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1456 	}
1457 }
1458 #endif
1459 
1460 #ifdef INET
1461 /* assumes that ip header and tcp header are contiguous on mbuf */
1462 void *
1463 tcp_ctlinput(int cmd, struct sockaddr *sa, void *v)
1464 {
1465 	struct ip *ip = v;
1466 	struct tcphdr *th;
1467 	struct icmp *icp;
1468 	extern const int inetctlerrmap[];
1469 	void (*notify)(struct inpcb *, int) = tcp_notify;
1470 	int errno;
1471 	int nmatch;
1472 #ifdef INET6
1473 	struct in6_addr src6, dst6;
1474 #endif
1475 
1476 	if (sa->sa_family != AF_INET ||
1477 	    sa->sa_len != sizeof(struct sockaddr_in))
1478 		return NULL;
1479 	if ((unsigned)cmd >= PRC_NCMDS)
1480 		return NULL;
1481 	errno = inetctlerrmap[cmd];
1482 	if (cmd == PRC_QUENCH)
1483 		notify = tcp_quench;
1484 	else if (PRC_IS_REDIRECT(cmd))
1485 		notify = in_rtchange, ip = 0;
1486 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1487 		/*
1488 		 * Check to see if we have a valid TCP connection
1489 		 * corresponding to the address in the ICMP message
1490 		 * payload.
1491 		 *
1492 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1493 		 */
1494 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1495 #ifdef INET6
1496 		memset(&src6, 0, sizeof(src6));
1497 		memset(&dst6, 0, sizeof(dst6));
1498 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1499 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1500 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1501 #endif
1502 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1503 		    ip->ip_src, th->th_sport) != NULL)
1504 			;
1505 #ifdef INET6
1506 		else if (in6_pcblookup_connect(&tcbtable, &dst6,
1507 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1508 			;
1509 #endif
1510 		else
1511 			return NULL;
1512 
1513 		/*
1514 		 * Now that we've validated that we are actually communicating
1515 		 * with the host indicated in the ICMP message, locate the
1516 		 * ICMP header, recalculate the new MTU, and create the
1517 		 * corresponding routing entry.
1518 		 */
1519 		icp = (struct icmp *)((caddr_t)ip -
1520 		    offsetof(struct icmp, icmp_ip));
1521 		icmp_mtudisc(icp, ip->ip_dst);
1522 
1523 		return NULL;
1524 	} else if (cmd == PRC_HOSTDEAD)
1525 		ip = 0;
1526 	else if (errno == 0)
1527 		return NULL;
1528 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1529 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1530 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1531 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1532 		if (nmatch == 0 && syn_cache_count &&
1533 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1534 		    inetctlerrmap[cmd] == ENETUNREACH ||
1535 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1536 			struct sockaddr_in sin;
1537 			bzero(&sin, sizeof(sin));
1538 			sin.sin_len = sizeof(sin);
1539 			sin.sin_family = AF_INET;
1540 			sin.sin_port = th->th_sport;
1541 			sin.sin_addr = ip->ip_src;
1542 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1543 		}
1544 
1545 		/* XXX mapped address case */
1546 	} else
1547 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1548 		    notify);
1549 	return NULL;
1550 }
1551 
1552 /*
1553  * When a source quench is received, we are being notified of congestion.
1554  * Close the congestion window down to the Loss Window (one segment).
1555  * We will gradually open it again as we proceed.
1556  */
1557 void
1558 tcp_quench(struct inpcb *inp, int errno)
1559 {
1560 	struct tcpcb *tp = intotcpcb(inp);
1561 
1562 	if (tp)
1563 		tp->snd_cwnd = tp->t_segsz;
1564 }
1565 #endif
1566 
1567 #ifdef INET6
1568 void
1569 tcp6_quench(struct in6pcb *in6p, int errno)
1570 {
1571 	struct tcpcb *tp = in6totcpcb(in6p);
1572 
1573 	if (tp)
1574 		tp->snd_cwnd = tp->t_segsz;
1575 }
1576 #endif
1577 
1578 #ifdef INET
1579 /*
1580  * Path MTU Discovery handlers.
1581  */
1582 void
1583 tcp_mtudisc_callback(struct in_addr faddr)
1584 {
1585 #ifdef INET6
1586 	struct in6_addr in6;
1587 #endif
1588 
1589 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1590 #ifdef INET6
1591 	memset(&in6, 0, sizeof(in6));
1592 	in6.s6_addr16[5] = 0xffff;
1593 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1594 	tcp6_mtudisc_callback(&in6);
1595 #endif
1596 }
1597 
1598 /*
1599  * On receipt of path MTU corrections, flush old route and replace it
1600  * with the new one.  Retransmit all unacknowledged packets, to ensure
1601  * that all packets will be received.
1602  */
1603 void
1604 tcp_mtudisc(struct inpcb *inp, int errno)
1605 {
1606 	struct tcpcb *tp = intotcpcb(inp);
1607 	struct rtentry *rt = in_pcbrtentry(inp);
1608 
1609 	if (tp != 0) {
1610 		if (rt != 0) {
1611 			/*
1612 			 * If this was not a host route, remove and realloc.
1613 			 */
1614 			if ((rt->rt_flags & RTF_HOST) == 0) {
1615 				in_rtchange(inp, errno);
1616 				if ((rt = in_pcbrtentry(inp)) == 0)
1617 					return;
1618 			}
1619 
1620 			/*
1621 			 * Slow start out of the error condition.  We
1622 			 * use the MTU because we know it's smaller
1623 			 * than the previously transmitted segment.
1624 			 *
1625 			 * Note: This is more conservative than the
1626 			 * suggestion in draft-floyd-incr-init-win-03.
1627 			 */
1628 			if (rt->rt_rmx.rmx_mtu != 0)
1629 				tp->snd_cwnd =
1630 				    TCP_INITIAL_WINDOW(tcp_init_win,
1631 				    rt->rt_rmx.rmx_mtu);
1632 		}
1633 
1634 		/*
1635 		 * Resend unacknowledged packets.
1636 		 */
1637 		tp->snd_nxt = tp->snd_una;
1638 		tcp_output(tp);
1639 	}
1640 }
1641 #endif
1642 
1643 #ifdef INET6
1644 /*
1645  * Path MTU Discovery handlers.
1646  */
1647 void
1648 tcp6_mtudisc_callback(struct in6_addr *faddr)
1649 {
1650 	struct sockaddr_in6 sin6;
1651 
1652 	bzero(&sin6, sizeof(sin6));
1653 	sin6.sin6_family = AF_INET6;
1654 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1655 	sin6.sin6_addr = *faddr;
1656 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1657 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1658 }
1659 
1660 void
1661 tcp6_mtudisc(struct in6pcb *in6p, int errno)
1662 {
1663 	struct tcpcb *tp = in6totcpcb(in6p);
1664 	struct rtentry *rt = in6_pcbrtentry(in6p);
1665 
1666 	if (tp != 0) {
1667 		if (rt != 0) {
1668 			/*
1669 			 * If this was not a host route, remove and realloc.
1670 			 */
1671 			if ((rt->rt_flags & RTF_HOST) == 0) {
1672 				in6_rtchange(in6p, errno);
1673 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1674 					return;
1675 			}
1676 
1677 			/*
1678 			 * Slow start out of the error condition.  We
1679 			 * use the MTU because we know it's smaller
1680 			 * than the previously transmitted segment.
1681 			 *
1682 			 * Note: This is more conservative than the
1683 			 * suggestion in draft-floyd-incr-init-win-03.
1684 			 */
1685 			if (rt->rt_rmx.rmx_mtu != 0)
1686 				tp->snd_cwnd =
1687 				    TCP_INITIAL_WINDOW(tcp_init_win,
1688 				    rt->rt_rmx.rmx_mtu);
1689 		}
1690 
1691 		/*
1692 		 * Resend unacknowledged packets.
1693 		 */
1694 		tp->snd_nxt = tp->snd_una;
1695 		tcp_output(tp);
1696 	}
1697 }
1698 #endif /* INET6 */
1699 
1700 /*
1701  * Compute the MSS to advertise to the peer.  Called only during
1702  * the 3-way handshake.  If we are the server (peer initiated
1703  * connection), we are called with a pointer to the interface
1704  * on which the SYN packet arrived.  If we are the client (we
1705  * initiated connection), we are called with a pointer to the
1706  * interface out which this connection should go.
1707  *
1708  * NOTE: Do not subtract IP option/extension header size nor IPsec
1709  * header size from MSS advertisement.  MSS option must hold the maximum
1710  * segment size we can accept, so it must always be:
1711  *	 max(if mtu) - ip header - tcp header
1712  */
1713 u_long
1714 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1715 {
1716 	extern u_long in_maxmtu;
1717 	u_long mss = 0;
1718 	u_long hdrsiz;
1719 
1720 	/*
1721 	 * In order to avoid defeating path MTU discovery on the peer,
1722 	 * we advertise the max MTU of all attached networks as our MSS,
1723 	 * per RFC 1191, section 3.1.
1724 	 *
1725 	 * We provide the option to advertise just the MTU of
1726 	 * the interface on which we hope this connection will
1727 	 * be receiving.  If we are responding to a SYN, we
1728 	 * will have a pretty good idea about this, but when
1729 	 * initiating a connection there is a bit more doubt.
1730 	 *
1731 	 * We also need to ensure that loopback has a large enough
1732 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1733 	 */
1734 
1735 	if (ifp != NULL)
1736 		switch (af) {
1737 		case AF_INET:
1738 			mss = ifp->if_mtu;
1739 			break;
1740 #ifdef INET6
1741 		case AF_INET6:
1742 			mss = IN6_LINKMTU(ifp);
1743 			break;
1744 #endif
1745 		}
1746 
1747 	if (tcp_mss_ifmtu == 0)
1748 		switch (af) {
1749 		case AF_INET:
1750 			mss = max(in_maxmtu, mss);
1751 			break;
1752 #ifdef INET6
1753 		case AF_INET6:
1754 			mss = max(in6_maxmtu, mss);
1755 			break;
1756 #endif
1757 		}
1758 
1759 	switch (af) {
1760 	case AF_INET:
1761 		hdrsiz = sizeof(struct ip);
1762 		break;
1763 #ifdef INET6
1764 	case AF_INET6:
1765 		hdrsiz = sizeof(struct ip6_hdr);
1766 		break;
1767 #endif
1768 	default:
1769 		hdrsiz = 0;
1770 		break;
1771 	}
1772 	hdrsiz += sizeof(struct tcphdr);
1773 	if (mss > hdrsiz)
1774 		mss -= hdrsiz;
1775 
1776 	mss = max(tcp_mssdflt, mss);
1777 	return (mss);
1778 }
1779 
1780 /*
1781  * Set connection variables based on the peer's advertised MSS.
1782  * We are passed the TCPCB for the actual connection.  If we
1783  * are the server, we are called by the compressed state engine
1784  * when the 3-way handshake is complete.  If we are the client,
1785  * we are called when we receive the SYN,ACK from the server.
1786  *
1787  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1788  * before this routine is called!
1789  */
1790 void
1791 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1792 {
1793 	struct socket *so;
1794 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1795 	struct rtentry *rt;
1796 #endif
1797 	u_long bufsize;
1798 	int mss;
1799 
1800 #ifdef DIAGNOSTIC
1801 	if (tp->t_inpcb && tp->t_in6pcb)
1802 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1803 #endif
1804 	so = NULL;
1805 	rt = NULL;
1806 #ifdef INET
1807 	if (tp->t_inpcb) {
1808 		so = tp->t_inpcb->inp_socket;
1809 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1810 		rt = in_pcbrtentry(tp->t_inpcb);
1811 #endif
1812 	}
1813 #endif
1814 #ifdef INET6
1815 	if (tp->t_in6pcb) {
1816 		so = tp->t_in6pcb->in6p_socket;
1817 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1818 		rt = in6_pcbrtentry(tp->t_in6pcb);
1819 #endif
1820 	}
1821 #endif
1822 
1823 	/*
1824 	 * As per RFC1122, use the default MSS value, unless they
1825 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1826 	 */
1827 	mss = tcp_mssdflt;
1828 	if (offer)
1829 		mss = offer;
1830 	mss = max(mss, 256);		/* sanity */
1831 	tp->t_peermss = mss;
1832 	mss -= tcp_optlen(tp);
1833 #ifdef INET
1834 	if (tp->t_inpcb)
1835 		mss -= ip_optlen(tp->t_inpcb);
1836 #endif
1837 #ifdef INET6
1838 	if (tp->t_in6pcb)
1839 		mss -= ip6_optlen(tp->t_in6pcb);
1840 #endif
1841 
1842 	/*
1843 	 * If there's a pipesize, change the socket buffer to that size.
1844 	 * Make the socket buffer an integral number of MSS units.  If
1845 	 * the MSS is larger than the socket buffer, artificially decrease
1846 	 * the MSS.
1847 	 */
1848 #ifdef RTV_SPIPE
1849 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1850 		bufsize = rt->rt_rmx.rmx_sendpipe;
1851 	else
1852 #endif
1853 		bufsize = so->so_snd.sb_hiwat;
1854 	if (bufsize < mss)
1855 		mss = bufsize;
1856 	else {
1857 		bufsize = roundup(bufsize, mss);
1858 		if (bufsize > sb_max)
1859 			bufsize = sb_max;
1860 		(void) sbreserve(&so->so_snd, bufsize, so);
1861 	}
1862 	tp->t_segsz = mss;
1863 
1864 #ifdef RTV_SSTHRESH
1865 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1866 		/*
1867 		 * There's some sort of gateway or interface buffer
1868 		 * limit on the path.  Use this to set the slow
1869 		 * start threshold, but set the threshold to no less
1870 		 * than 2 * MSS.
1871 		 */
1872 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1873 	}
1874 #endif
1875 }
1876 
1877 /*
1878  * Processing necessary when a TCP connection is established.
1879  */
1880 void
1881 tcp_established(struct tcpcb *tp)
1882 {
1883 	struct socket *so;
1884 #ifdef RTV_RPIPE
1885 	struct rtentry *rt;
1886 #endif
1887 	u_long bufsize;
1888 
1889 #ifdef DIAGNOSTIC
1890 	if (tp->t_inpcb && tp->t_in6pcb)
1891 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1892 #endif
1893 	so = NULL;
1894 	rt = NULL;
1895 #ifdef INET
1896 	if (tp->t_inpcb) {
1897 		so = tp->t_inpcb->inp_socket;
1898 #if defined(RTV_RPIPE)
1899 		rt = in_pcbrtentry(tp->t_inpcb);
1900 #endif
1901 	}
1902 #endif
1903 #ifdef INET6
1904 	if (tp->t_in6pcb) {
1905 		so = tp->t_in6pcb->in6p_socket;
1906 #if defined(RTV_RPIPE)
1907 		rt = in6_pcbrtentry(tp->t_in6pcb);
1908 #endif
1909 	}
1910 #endif
1911 
1912 	tp->t_state = TCPS_ESTABLISHED;
1913 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1914 
1915 #ifdef RTV_RPIPE
1916 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1917 		bufsize = rt->rt_rmx.rmx_recvpipe;
1918 	else
1919 #endif
1920 		bufsize = so->so_rcv.sb_hiwat;
1921 	if (bufsize > tp->t_ourmss) {
1922 		bufsize = roundup(bufsize, tp->t_ourmss);
1923 		if (bufsize > sb_max)
1924 			bufsize = sb_max;
1925 		(void) sbreserve(&so->so_rcv, bufsize, so);
1926 	}
1927 }
1928 
1929 /*
1930  * Check if there's an initial rtt or rttvar.  Convert from the
1931  * route-table units to scaled multiples of the slow timeout timer.
1932  * Called only during the 3-way handshake.
1933  */
1934 void
1935 tcp_rmx_rtt(struct tcpcb *tp)
1936 {
1937 #ifdef RTV_RTT
1938 	struct rtentry *rt = NULL;
1939 	int rtt;
1940 
1941 #ifdef DIAGNOSTIC
1942 	if (tp->t_inpcb && tp->t_in6pcb)
1943 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1944 #endif
1945 #ifdef INET
1946 	if (tp->t_inpcb)
1947 		rt = in_pcbrtentry(tp->t_inpcb);
1948 #endif
1949 #ifdef INET6
1950 	if (tp->t_in6pcb)
1951 		rt = in6_pcbrtentry(tp->t_in6pcb);
1952 #endif
1953 	if (rt == NULL)
1954 		return;
1955 
1956 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1957 		/*
1958 		 * XXX The lock bit for MTU indicates that the value
1959 		 * is also a minimum value; this is subject to time.
1960 		 */
1961 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1962 			TCPT_RANGESET(tp->t_rttmin,
1963 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1964 			    TCPTV_MIN, TCPTV_REXMTMAX);
1965 		tp->t_srtt = rtt /
1966 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1967 		if (rt->rt_rmx.rmx_rttvar) {
1968 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1969 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1970 				(TCP_RTTVAR_SHIFT + 2));
1971 		} else {
1972 			/* Default variation is +- 1 rtt */
1973 			tp->t_rttvar =
1974 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1975 		}
1976 		TCPT_RANGESET(tp->t_rxtcur,
1977 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1978 		    tp->t_rttmin, TCPTV_REXMTMAX);
1979 	}
1980 #endif
1981 }
1982 
1983 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
1984 #if NRND > 0
1985 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
1986 #endif
1987 
1988 /*
1989  * Get a new sequence value given a tcp control block
1990  */
1991 tcp_seq
1992 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
1993 {
1994 
1995 #ifdef INET
1996 	if (tp->t_inpcb != NULL) {
1997 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
1998 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
1999 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2000 		    addin));
2001 	}
2002 #endif
2003 #ifdef INET6
2004 	if (tp->t_in6pcb != NULL) {
2005 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2006 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2007 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2008 		    addin));
2009 	}
2010 #endif
2011 	/* Not possible. */
2012 	panic("tcp_new_iss");
2013 }
2014 
2015 /*
2016  * This routine actually generates a new TCP initial sequence number.
2017  */
2018 tcp_seq
2019 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2020     size_t addrsz, tcp_seq addin)
2021 {
2022 	tcp_seq tcp_iss;
2023 
2024 #if NRND > 0
2025 	static int beenhere;
2026 
2027 	/*
2028 	 * If we haven't been here before, initialize our cryptographic
2029 	 * hash secret.
2030 	 */
2031 	if (beenhere == 0) {
2032 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2033 		    RND_EXTRACT_ANY);
2034 		beenhere = 1;
2035 	}
2036 
2037 	if (tcp_do_rfc1948) {
2038 		MD5_CTX ctx;
2039 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2040 
2041 		/*
2042 		 * Compute the base value of the ISS.  It is a hash
2043 		 * of (saddr, sport, daddr, dport, secret).
2044 		 */
2045 		MD5Init(&ctx);
2046 
2047 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2048 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2049 
2050 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2051 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2052 
2053 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2054 
2055 		MD5Final(hash, &ctx);
2056 
2057 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2058 
2059 		/*
2060 		 * Now increment our "timer", and add it in to
2061 		 * the computed value.
2062 		 *
2063 		 * XXX Use `addin'?
2064 		 * XXX TCP_ISSINCR too large to use?
2065 		 */
2066 		tcp_iss_seq += TCP_ISSINCR;
2067 #ifdef TCPISS_DEBUG
2068 		printf("ISS hash 0x%08x, ", tcp_iss);
2069 #endif
2070 		tcp_iss += tcp_iss_seq + addin;
2071 #ifdef TCPISS_DEBUG
2072 		printf("new ISS 0x%08x\n", tcp_iss);
2073 #endif
2074 	} else
2075 #endif /* NRND > 0 */
2076 	{
2077 		/*
2078 		 * Randomize.
2079 		 */
2080 #if NRND > 0
2081 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2082 #else
2083 		tcp_iss = arc4random();
2084 #endif
2085 
2086 		/*
2087 		 * If we were asked to add some amount to a known value,
2088 		 * we will take a random value obtained above, mask off
2089 		 * the upper bits, and add in the known value.  We also
2090 		 * add in a constant to ensure that we are at least a
2091 		 * certain distance from the original value.
2092 		 *
2093 		 * This is used when an old connection is in timed wait
2094 		 * and we have a new one coming in, for instance.
2095 		 */
2096 		if (addin != 0) {
2097 #ifdef TCPISS_DEBUG
2098 			printf("Random %08x, ", tcp_iss);
2099 #endif
2100 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2101 			tcp_iss += addin + TCP_ISSINCR;
2102 #ifdef TCPISS_DEBUG
2103 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2104 #endif
2105 		} else {
2106 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2107 			tcp_iss += tcp_iss_seq;
2108 			tcp_iss_seq += TCP_ISSINCR;
2109 #ifdef TCPISS_DEBUG
2110 			printf("ISS %08x\n", tcp_iss);
2111 #endif
2112 		}
2113 	}
2114 
2115 	if (tcp_compat_42) {
2116 		/*
2117 		 * Limit it to the positive range for really old TCP
2118 		 * implementations.
2119 		 * Just AND off the top bit instead of checking if
2120 		 * is set first - saves a branch 50% of the time.
2121 		 */
2122 		tcp_iss &= 0x7fffffff;		/* XXX */
2123 	}
2124 
2125 	return (tcp_iss);
2126 }
2127 
2128 #if defined(IPSEC) || defined(FAST_IPSEC)
2129 /* compute ESP/AH header size for TCP, including outer IP header. */
2130 size_t
2131 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2132 {
2133 	struct inpcb *inp;
2134 	size_t hdrsiz;
2135 
2136 	/* XXX mapped addr case (tp->t_in6pcb) */
2137 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2138 		return 0;
2139 	switch (tp->t_family) {
2140 	case AF_INET:
2141 		/* XXX: should use currect direction. */
2142 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2143 		break;
2144 	default:
2145 		hdrsiz = 0;
2146 		break;
2147 	}
2148 
2149 	return hdrsiz;
2150 }
2151 
2152 #ifdef INET6
2153 size_t
2154 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2155 {
2156 	struct in6pcb *in6p;
2157 	size_t hdrsiz;
2158 
2159 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2160 		return 0;
2161 	switch (tp->t_family) {
2162 	case AF_INET6:
2163 		/* XXX: should use currect direction. */
2164 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2165 		break;
2166 	case AF_INET:
2167 		/* mapped address case - tricky */
2168 	default:
2169 		hdrsiz = 0;
2170 		break;
2171 	}
2172 
2173 	return hdrsiz;
2174 }
2175 #endif
2176 #endif /*IPSEC*/
2177 
2178 /*
2179  * Determine the length of the TCP options for this connection.
2180  *
2181  * XXX:  What do we do for SACK, when we add that?  Just reserve
2182  *       all of the space?  Otherwise we can't exactly be incrementing
2183  *       cwnd by an amount that varies depending on the amount we last
2184  *       had to SACK!
2185  */
2186 
2187 u_int
2188 tcp_optlen(struct tcpcb *tp)
2189 {
2190 	u_int optlen;
2191 
2192 	optlen = 0;
2193 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2194 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2195 		optlen += TCPOLEN_TSTAMP_APPA;
2196 
2197 #ifdef TCP_SIGNATURE
2198 #if defined(INET6) && defined(FAST_IPSEC)
2199 	if (tp->t_family == AF_INET)
2200 #endif
2201 	if (tp->t_flags & TF_SIGNATURE)
2202 		optlen += TCPOLEN_SIGNATURE + 2;
2203 #endif /* TCP_SIGNATURE */
2204 
2205 	return optlen;
2206 }
2207