xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: tcp_subr.c,v 1.173 2004/09/15 09:21:22 yamt 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 1.173 2004/09/15 09:21:22 yamt 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 
204 /* tcb hash */
205 #ifndef TCBHASHSIZE
206 #define	TCBHASHSIZE	128
207 #endif
208 int	tcbhashsize = TCBHASHSIZE;
209 
210 /* syn hash parameters */
211 #define	TCP_SYN_HASH_SIZE	293
212 #define	TCP_SYN_BUCKET_SIZE	35
213 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
214 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
215 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
216 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
217 
218 int	tcp_freeq __P((struct tcpcb *));
219 
220 #ifdef INET
221 void	tcp_mtudisc_callback __P((struct in_addr));
222 #endif
223 #ifdef INET6
224 void	tcp6_mtudisc_callback __P((struct in6_addr *));
225 #endif
226 
227 void	tcp_mtudisc __P((struct inpcb *, int));
228 #ifdef INET6
229 void	tcp6_mtudisc __P((struct in6pcb *, int));
230 #endif
231 
232 POOL_INIT(tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl", NULL);
233 
234 #ifdef TCP_CSUM_COUNTERS
235 #include <sys/device.h>
236 
237 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
238     NULL, "tcp", "hwcsum bad");
239 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
240     NULL, "tcp", "hwcsum ok");
241 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
242     NULL, "tcp", "hwcsum data");
243 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
244     NULL, "tcp", "swcsum");
245 
246 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
247 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
248 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
249 EVCNT_ATTACH_STATIC(tcp_swcsum);
250 #endif /* TCP_CSUM_COUNTERS */
251 
252 
253 #ifdef TCP_OUTPUT_COUNTERS
254 #include <sys/device.h>
255 
256 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
257     NULL, "tcp", "output big header");
258 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
259     NULL, "tcp", "output predict hit");
260 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
261     NULL, "tcp", "output predict miss");
262 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
263     NULL, "tcp", "output copy small");
264 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
265     NULL, "tcp", "output copy big");
266 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
267     NULL, "tcp", "output reference big");
268 
269 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
270 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
271 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
272 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
273 EVCNT_ATTACH_STATIC(tcp_output_copybig);
274 EVCNT_ATTACH_STATIC(tcp_output_refbig);
275 
276 #endif /* TCP_OUTPUT_COUNTERS */
277 
278 #ifdef TCP_REASS_COUNTERS
279 #include <sys/device.h>
280 
281 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282     NULL, "tcp_reass", "calls");
283 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284     &tcp_reass_, "tcp_reass", "insert into empty queue");
285 struct evcnt tcp_reass_iteration[8] = {
286     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
287     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
288     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
289     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
290     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
291     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
292     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
293     EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
294 };
295 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
296     &tcp_reass_, "tcp_reass", "prepend to first");
297 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
298     &tcp_reass_, "tcp_reass", "prepend");
299 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
300     &tcp_reass_, "tcp_reass", "insert");
301 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
302     &tcp_reass_, "tcp_reass", "insert at tail");
303 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
304     &tcp_reass_, "tcp_reass", "append");
305 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
306     &tcp_reass_, "tcp_reass", "append to tail fragment");
307 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
308     &tcp_reass_, "tcp_reass", "overlap at end");
309 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
310     &tcp_reass_, "tcp_reass", "overlap at start");
311 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
312     &tcp_reass_, "tcp_reass", "duplicate segment");
313 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
314     &tcp_reass_, "tcp_reass", "duplicate fragment");
315 
316 EVCNT_ATTACH_STATIC(tcp_reass_);
317 EVCNT_ATTACH_STATIC(tcp_reass_empty);
318 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
319 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
320 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
321 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
322 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
323 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
324 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
325 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
326 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
327 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
328 EVCNT_ATTACH_STATIC(tcp_reass_insert);
329 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
330 EVCNT_ATTACH_STATIC(tcp_reass_append);
331 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
332 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
333 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
334 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
335 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
336 
337 #endif /* TCP_REASS_COUNTERS */
338 
339 #ifdef MBUFTRACE
340 struct mowner tcp_mowner = { "tcp" };
341 struct mowner tcp_rx_mowner = { "tcp", "rx" };
342 struct mowner tcp_tx_mowner = { "tcp", "tx" };
343 #endif
344 
345 /*
346  * Tcp initialization
347  */
348 void
349 tcp_init()
350 {
351 	int hlen;
352 
353 	/* Initialize the TCPCB template. */
354 	tcp_tcpcb_template();
355 
356 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
357 
358 	hlen = sizeof(struct ip) + sizeof(struct tcphdr);
359 #ifdef INET6
360 	if (sizeof(struct ip) < sizeof(struct ip6_hdr))
361 		hlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
362 #endif
363 	if (max_protohdr < hlen)
364 		max_protohdr = hlen;
365 	if (max_linkhdr + hlen > MHLEN)
366 		panic("tcp_init");
367 
368 #ifdef INET
369 	icmp_mtudisc_callback_register(tcp_mtudisc_callback);
370 #endif
371 #ifdef INET6
372 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
373 #endif
374 
375 	/* Initialize timer state. */
376 	tcp_timer_init();
377 
378 	/* Initialize the compressed state engine. */
379 	syn_cache_init();
380 
381 	MOWNER_ATTACH(&tcp_tx_mowner);
382 	MOWNER_ATTACH(&tcp_rx_mowner);
383 	MOWNER_ATTACH(&tcp_mowner);
384 }
385 
386 /*
387  * Create template to be used to send tcp packets on a connection.
388  * Call after host entry created, allocates an mbuf and fills
389  * in a skeletal tcp/ip header, minimizing the amount of work
390  * necessary when the connection is used.
391  */
392 struct mbuf *
393 tcp_template(tp)
394 	struct tcpcb *tp;
395 {
396 	struct inpcb *inp = tp->t_inpcb;
397 #ifdef INET6
398 	struct in6pcb *in6p = tp->t_in6pcb;
399 #endif
400 	struct tcphdr *n;
401 	struct mbuf *m;
402 	int hlen;
403 
404 	switch (tp->t_family) {
405 	case AF_INET:
406 		hlen = sizeof(struct ip);
407 		if (inp)
408 			break;
409 #ifdef INET6
410 		if (in6p) {
411 			/* mapped addr case */
412 			if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
413 			 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
414 				break;
415 		}
416 #endif
417 		return NULL;	/*EINVAL*/
418 #ifdef INET6
419 	case AF_INET6:
420 		hlen = sizeof(struct ip6_hdr);
421 		if (in6p) {
422 			/* more sainty check? */
423 			break;
424 		}
425 		return NULL;	/*EINVAL*/
426 #endif
427 	default:
428 		hlen = 0;	/*pacify gcc*/
429 		return NULL;	/*EAFNOSUPPORT*/
430 	}
431 #ifdef DIAGNOSTIC
432 	if (hlen + sizeof(struct tcphdr) > MCLBYTES)
433 		panic("mclbytes too small for t_template");
434 #endif
435 	m = tp->t_template;
436 	if (m && m->m_len == hlen + sizeof(struct tcphdr))
437 		;
438 	else {
439 		if (m)
440 			m_freem(m);
441 		m = tp->t_template = NULL;
442 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
443 		if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
444 			MCLGET(m, M_DONTWAIT);
445 			if ((m->m_flags & M_EXT) == 0) {
446 				m_free(m);
447 				m = NULL;
448 			}
449 		}
450 		if (m == NULL)
451 			return NULL;
452 		MCLAIM(m, &tcp_mowner);
453 		m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
454 	}
455 
456 	bzero(mtod(m, caddr_t), m->m_len);
457 
458 	n = (struct tcphdr *)(mtod(m, caddr_t) + hlen);
459 
460 	switch (tp->t_family) {
461 	case AF_INET:
462 	    {
463 		struct ipovly *ipov;
464 		mtod(m, struct ip *)->ip_v = 4;
465 		mtod(m, struct ip *)->ip_hl = hlen >> 2;
466 		ipov = mtod(m, struct ipovly *);
467 		ipov->ih_pr = IPPROTO_TCP;
468 		ipov->ih_len = htons(sizeof(struct tcphdr));
469 		if (inp) {
470 			ipov->ih_src = inp->inp_laddr;
471 			ipov->ih_dst = inp->inp_faddr;
472 		}
473 #ifdef INET6
474 		else if (in6p) {
475 			/* mapped addr case */
476 			bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
477 				sizeof(ipov->ih_src));
478 			bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
479 				sizeof(ipov->ih_dst));
480 		}
481 #endif
482 		/*
483 		 * Compute the pseudo-header portion of the checksum
484 		 * now.  We incrementally add in the TCP option and
485 		 * payload lengths later, and then compute the TCP
486 		 * checksum right before the packet is sent off onto
487 		 * the wire.
488 		 */
489 		n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
490 		    ipov->ih_dst.s_addr,
491 		    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
492 		break;
493 	    }
494 #ifdef INET6
495 	case AF_INET6:
496 	    {
497 		struct ip6_hdr *ip6;
498 		mtod(m, struct ip *)->ip_v = 6;
499 		ip6 = mtod(m, struct ip6_hdr *);
500 		ip6->ip6_nxt = IPPROTO_TCP;
501 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
502 		ip6->ip6_src = in6p->in6p_laddr;
503 		ip6->ip6_dst = in6p->in6p_faddr;
504 		ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
505 		if (ip6_auto_flowlabel) {
506 			ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
507 			ip6->ip6_flow |=
508 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
509 		}
510 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
511 		ip6->ip6_vfc |= IPV6_VERSION;
512 
513 		/*
514 		 * Compute the pseudo-header portion of the checksum
515 		 * now.  We incrementally add in the TCP option and
516 		 * payload lengths later, and then compute the TCP
517 		 * checksum right before the packet is sent off onto
518 		 * the wire.
519 		 */
520 		n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
521 		    &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
522 		    htonl(IPPROTO_TCP));
523 		break;
524 	    }
525 #endif
526 	}
527 	if (inp) {
528 		n->th_sport = inp->inp_lport;
529 		n->th_dport = inp->inp_fport;
530 	}
531 #ifdef INET6
532 	else if (in6p) {
533 		n->th_sport = in6p->in6p_lport;
534 		n->th_dport = in6p->in6p_fport;
535 	}
536 #endif
537 	n->th_seq = 0;
538 	n->th_ack = 0;
539 	n->th_x2 = 0;
540 	n->th_off = 5;
541 	n->th_flags = 0;
542 	n->th_win = 0;
543 	n->th_urp = 0;
544 	return (m);
545 }
546 
547 /*
548  * Send a single message to the TCP at address specified by
549  * the given TCP/IP header.  If m == 0, then we make a copy
550  * of the tcpiphdr at ti and send directly to the addressed host.
551  * This is used to force keep alive messages out using the TCP
552  * template for a connection tp->t_template.  If flags are given
553  * then we send a message back to the TCP which originated the
554  * segment ti, and discard the mbuf containing it and any other
555  * attached mbufs.
556  *
557  * In any case the ack and sequence number of the transmitted
558  * segment are as specified by the parameters.
559  */
560 int
561 tcp_respond(tp, template, m, th0, ack, seq, flags)
562 	struct tcpcb *tp;
563 	struct mbuf *template;
564 	struct mbuf *m;
565 	struct tcphdr *th0;
566 	tcp_seq ack, seq;
567 	int flags;
568 {
569 	struct route *ro;
570 	int error, tlen, win = 0;
571 	int hlen;
572 	struct ip *ip;
573 #ifdef INET6
574 	struct ip6_hdr *ip6;
575 #endif
576 	int family;	/* family on packet, not inpcb/in6pcb! */
577 	struct tcphdr *th;
578 	struct socket *so;
579 
580 	if (tp != NULL && (flags & TH_RST) == 0) {
581 #ifdef DIAGNOSTIC
582 		if (tp->t_inpcb && tp->t_in6pcb)
583 			panic("tcp_respond: both t_inpcb and t_in6pcb are set");
584 #endif
585 #ifdef INET
586 		if (tp->t_inpcb)
587 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
588 #endif
589 #ifdef INET6
590 		if (tp->t_in6pcb)
591 			win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
592 #endif
593 	}
594 
595 	th = NULL;	/* Quell uninitialized warning */
596 	ip = NULL;
597 #ifdef INET6
598 	ip6 = NULL;
599 #endif
600 	if (m == 0) {
601 		if (!template)
602 			return EINVAL;
603 
604 		/* get family information from template */
605 		switch (mtod(template, struct ip *)->ip_v) {
606 		case 4:
607 			family = AF_INET;
608 			hlen = sizeof(struct ip);
609 			break;
610 #ifdef INET6
611 		case 6:
612 			family = AF_INET6;
613 			hlen = sizeof(struct ip6_hdr);
614 			break;
615 #endif
616 		default:
617 			return EAFNOSUPPORT;
618 		}
619 
620 		MGETHDR(m, M_DONTWAIT, MT_HEADER);
621 		if (m) {
622 			MCLAIM(m, &tcp_tx_mowner);
623 			MCLGET(m, M_DONTWAIT);
624 			if ((m->m_flags & M_EXT) == 0) {
625 				m_free(m);
626 				m = NULL;
627 			}
628 		}
629 		if (m == NULL)
630 			return (ENOBUFS);
631 
632 		if (tcp_compat_42)
633 			tlen = 1;
634 		else
635 			tlen = 0;
636 
637 		m->m_data += max_linkhdr;
638 		bcopy(mtod(template, caddr_t), mtod(m, caddr_t),
639 			template->m_len);
640 		switch (family) {
641 		case AF_INET:
642 			ip = mtod(m, struct ip *);
643 			th = (struct tcphdr *)(ip + 1);
644 			break;
645 #ifdef INET6
646 		case AF_INET6:
647 			ip6 = mtod(m, struct ip6_hdr *);
648 			th = (struct tcphdr *)(ip6 + 1);
649 			break;
650 #endif
651 #if 0
652 		default:
653 			/* noone will visit here */
654 			m_freem(m);
655 			return EAFNOSUPPORT;
656 #endif
657 		}
658 		flags = TH_ACK;
659 	} else {
660 
661 		if ((m->m_flags & M_PKTHDR) == 0) {
662 #if 0
663 			printf("non PKTHDR to tcp_respond\n");
664 #endif
665 			m_freem(m);
666 			return EINVAL;
667 		}
668 #ifdef DIAGNOSTIC
669 		if (!th0)
670 			panic("th0 == NULL in tcp_respond");
671 #endif
672 
673 		/* get family information from m */
674 		switch (mtod(m, struct ip *)->ip_v) {
675 		case 4:
676 			family = AF_INET;
677 			hlen = sizeof(struct ip);
678 			ip = mtod(m, struct ip *);
679 			break;
680 #ifdef INET6
681 		case 6:
682 			family = AF_INET6;
683 			hlen = sizeof(struct ip6_hdr);
684 			ip6 = mtod(m, struct ip6_hdr *);
685 			break;
686 #endif
687 		default:
688 			m_freem(m);
689 			return EAFNOSUPPORT;
690 		}
691 		if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
692 			tlen = sizeof(*th0);
693 		else
694 			tlen = th0->th_off << 2;
695 
696 		if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
697 		    mtod(m, caddr_t) + hlen == (caddr_t)th0) {
698 			m->m_len = hlen + tlen;
699 			m_freem(m->m_next);
700 			m->m_next = NULL;
701 		} else {
702 			struct mbuf *n;
703 
704 #ifdef DIAGNOSTIC
705 			if (max_linkhdr + hlen + tlen > MCLBYTES) {
706 				m_freem(m);
707 				return EMSGSIZE;
708 			}
709 #endif
710 			MGETHDR(n, M_DONTWAIT, MT_HEADER);
711 			if (n && max_linkhdr + hlen + tlen > MHLEN) {
712 				MCLGET(n, M_DONTWAIT);
713 				if ((n->m_flags & M_EXT) == 0) {
714 					m_freem(n);
715 					n = NULL;
716 				}
717 			}
718 			if (!n) {
719 				m_freem(m);
720 				return ENOBUFS;
721 			}
722 
723 			MCLAIM(n, &tcp_tx_mowner);
724 			n->m_data += max_linkhdr;
725 			n->m_len = hlen + tlen;
726 			m_copyback(n, 0, hlen, mtod(m, caddr_t));
727 			m_copyback(n, hlen, tlen, (caddr_t)th0);
728 
729 			m_freem(m);
730 			m = n;
731 			n = NULL;
732 		}
733 
734 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
735 		switch (family) {
736 		case AF_INET:
737 			ip = mtod(m, struct ip *);
738 			th = (struct tcphdr *)(ip + 1);
739 			ip->ip_p = IPPROTO_TCP;
740 			xchg(ip->ip_dst, ip->ip_src, struct in_addr);
741 			ip->ip_p = IPPROTO_TCP;
742 			break;
743 #ifdef INET6
744 		case AF_INET6:
745 			ip6 = mtod(m, struct ip6_hdr *);
746 			th = (struct tcphdr *)(ip6 + 1);
747 			ip6->ip6_nxt = IPPROTO_TCP;
748 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
749 			ip6->ip6_nxt = IPPROTO_TCP;
750 			break;
751 #endif
752 #if 0
753 		default:
754 			/* noone will visit here */
755 			m_freem(m);
756 			return EAFNOSUPPORT;
757 #endif
758 		}
759 		xchg(th->th_dport, th->th_sport, u_int16_t);
760 #undef xchg
761 		tlen = 0;	/*be friendly with the following code*/
762 	}
763 	th->th_seq = htonl(seq);
764 	th->th_ack = htonl(ack);
765 	th->th_x2 = 0;
766 	if ((flags & TH_SYN) == 0) {
767 		if (tp)
768 			win >>= tp->rcv_scale;
769 		if (win > TCP_MAXWIN)
770 			win = TCP_MAXWIN;
771 		th->th_win = htons((u_int16_t)win);
772 		th->th_off = sizeof (struct tcphdr) >> 2;
773 		tlen += sizeof(*th);
774 	} else
775 		tlen += th->th_off << 2;
776 	m->m_len = hlen + tlen;
777 	m->m_pkthdr.len = hlen + tlen;
778 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
779 	th->th_flags = flags;
780 	th->th_urp = 0;
781 
782 	switch (family) {
783 #ifdef INET
784 	case AF_INET:
785 	    {
786 		struct ipovly *ipov = (struct ipovly *)ip;
787 		bzero(ipov->ih_x1, sizeof ipov->ih_x1);
788 		ipov->ih_len = htons((u_int16_t)tlen);
789 
790 		th->th_sum = 0;
791 		th->th_sum = in_cksum(m, hlen + tlen);
792 		ip->ip_len = htons(hlen + tlen);
793 		ip->ip_ttl = ip_defttl;
794 		break;
795 	    }
796 #endif
797 #ifdef INET6
798 	case AF_INET6:
799 	    {
800 		th->th_sum = 0;
801 		th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
802 				tlen);
803 		ip6->ip6_plen = ntohs(tlen);
804 		if (tp && tp->t_in6pcb) {
805 			struct ifnet *oifp;
806 			ro = (struct route *)&tp->t_in6pcb->in6p_route;
807 			oifp = ro->ro_rt ? ro->ro_rt->rt_ifp : NULL;
808 			ip6->ip6_hlim = in6_selecthlim(tp->t_in6pcb, oifp);
809 		} else
810 			ip6->ip6_hlim = ip6_defhlim;
811 		ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
812 		if (ip6_auto_flowlabel) {
813 			ip6->ip6_flow |=
814 			    (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
815 		}
816 		break;
817 	    }
818 #endif
819 	}
820 
821 	if (tp && tp->t_inpcb)
822 		so = tp->t_inpcb->inp_socket;
823 #ifdef INET6
824 	else if (tp && tp->t_in6pcb)
825 		so = tp->t_in6pcb->in6p_socket;
826 #endif
827 	else
828 		so = NULL;
829 
830 	if (tp != NULL && tp->t_inpcb != NULL) {
831 		ro = &tp->t_inpcb->inp_route;
832 #ifdef DIAGNOSTIC
833 		if (family != AF_INET)
834 			panic("tcp_respond: address family mismatch");
835 		if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
836 			panic("tcp_respond: ip_dst %x != inp_faddr %x",
837 			    ntohl(ip->ip_dst.s_addr),
838 			    ntohl(tp->t_inpcb->inp_faddr.s_addr));
839 		}
840 #endif
841 	}
842 #ifdef INET6
843 	else if (tp != NULL && tp->t_in6pcb != NULL) {
844 		ro = (struct route *)&tp->t_in6pcb->in6p_route;
845 #ifdef DIAGNOSTIC
846 		if (family == AF_INET) {
847 			if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
848 				panic("tcp_respond: not mapped addr");
849 			if (bcmp(&ip->ip_dst,
850 			    &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
851 			    sizeof(ip->ip_dst)) != 0) {
852 				panic("tcp_respond: ip_dst != in6p_faddr");
853 			}
854 		} else if (family == AF_INET6) {
855 			if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
856 			    &tp->t_in6pcb->in6p_faddr))
857 				panic("tcp_respond: ip6_dst != in6p_faddr");
858 		} else
859 			panic("tcp_respond: address family mismatch");
860 #endif
861 	}
862 #endif
863 	else
864 		ro = NULL;
865 
866 	switch (family) {
867 #ifdef INET
868 	case AF_INET:
869 		error = ip_output(m, NULL, ro,
870 		    (tp && tp->t_mtudisc ? IP_MTUDISC : 0),
871 		    (struct ip_moptions *)0, so);
872 		break;
873 #endif
874 #ifdef INET6
875 	case AF_INET6:
876 		error = ip6_output(m, NULL, (struct route_in6 *)ro, 0,
877 		    (struct ip6_moptions *)0, so, NULL);
878 		break;
879 #endif
880 	default:
881 		error = EAFNOSUPPORT;
882 		break;
883 	}
884 
885 	return (error);
886 }
887 
888 /*
889  * Template TCPCB.  Rather than zeroing a new TCPCB and initializing
890  * a bunch of members individually, we maintain this template for the
891  * static and mostly-static components of the TCPCB, and copy it into
892  * the new TCPCB instead.
893  */
894 static struct tcpcb tcpcb_template = {
895 	/*
896 	 * If TCP_NTIMERS ever changes, we'll need to update this
897 	 * initializer.
898 	 */
899 	.t_timer = {
900 		CALLOUT_INITIALIZER,
901 		CALLOUT_INITIALIZER,
902 		CALLOUT_INITIALIZER,
903 		CALLOUT_INITIALIZER,
904 	},
905 	.t_delack_ch = CALLOUT_INITIALIZER,
906 
907 	.t_srtt = TCPTV_SRTTBASE,
908 	.t_rttmin = TCPTV_MIN,
909 
910 	.snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
911 	.snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
912 };
913 
914 /*
915  * Updates the TCPCB template whenever a parameter that would affect
916  * the template is changed.
917  */
918 void
919 tcp_tcpcb_template(void)
920 {
921 	struct tcpcb *tp = &tcpcb_template;
922 	int flags;
923 
924 	tp->t_peermss = tcp_mssdflt;
925 	tp->t_ourmss = tcp_mssdflt;
926 	tp->t_segsz = tcp_mssdflt;
927 
928 	flags = 0;
929 	if (tcp_do_rfc1323 && tcp_do_win_scale)
930 		flags |= TF_REQ_SCALE;
931 	if (tcp_do_rfc1323 && tcp_do_timestamps)
932 		flags |= TF_REQ_TSTMP;
933 	if (tcp_do_sack == 2)
934 		flags |= TF_WILL_SACK;
935 	else if (tcp_do_sack == 1)
936 		flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
937 	flags |= TF_CANT_TXSACK;
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 struct tcpcb *
956 tcp_newtcpcb(family, aux)
957 	int family;	/* selects inpcb, or in6pcb */
958 	void *aux;
959 {
960 	struct tcpcb *tp;
961 	int i;
962 
963 	/* XXX Consider using a pool_cache for speed. */
964 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
965 	if (tp == NULL)
966 		return (NULL);
967 	memcpy(tp, &tcpcb_template, sizeof(*tp));
968 	TAILQ_INIT(&tp->segq);
969 	TAILQ_INIT(&tp->timeq);
970 	tp->t_family = family;		/* may be overridden later on */
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(tp, errno)
1029 	struct tcpcb *tp;
1030 	int errno;
1031 {
1032 	struct socket *so = NULL;
1033 
1034 #ifdef DIAGNOSTIC
1035 	if (tp->t_inpcb && tp->t_in6pcb)
1036 		panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1037 #endif
1038 #ifdef INET
1039 	if (tp->t_inpcb)
1040 		so = tp->t_inpcb->inp_socket;
1041 #endif
1042 #ifdef INET6
1043 	if (tp->t_in6pcb)
1044 		so = tp->t_in6pcb->in6p_socket;
1045 #endif
1046 	if (!so)
1047 		return NULL;
1048 
1049 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1050 		tp->t_state = TCPS_CLOSED;
1051 		(void) tcp_output(tp);
1052 		tcpstat.tcps_drops++;
1053 	} else
1054 		tcpstat.tcps_conndrops++;
1055 	if (errno == ETIMEDOUT && tp->t_softerror)
1056 		errno = tp->t_softerror;
1057 	so->so_error = errno;
1058 	return (tcp_close(tp));
1059 }
1060 
1061 /*
1062  * Return whether this tcpcb is marked as dead, indicating
1063  * to the calling timer function that no further action should
1064  * be taken, as we are about to release this tcpcb.  The release
1065  * of the storage will be done if this is the last timer running.
1066  *
1067  * This should be called from the callout handler function after
1068  * callout_ack() is done, so that the number of invoking timer
1069  * functions is 0.
1070  */
1071 int
1072 tcp_isdead(tp)
1073 	struct tcpcb *tp;
1074 {
1075 	int dead = (tp->t_flags & TF_DEAD);
1076 
1077 	if (__predict_false(dead)) {
1078 		if (tcp_timers_invoking(tp) > 0)
1079 				/* not quite there yet -- count separately? */
1080 			return dead;
1081 		tcpstat.tcps_delayed_free++;
1082 		pool_put(&tcpcb_pool, tp);
1083 	}
1084 	return dead;
1085 }
1086 
1087 /*
1088  * Close a TCP control block:
1089  *	discard all space held by the tcp
1090  *	discard internet protocol block
1091  *	wake up any sleepers
1092  */
1093 struct tcpcb *
1094 tcp_close(tp)
1095 	struct tcpcb *tp;
1096 {
1097 	struct inpcb *inp;
1098 #ifdef INET6
1099 	struct in6pcb *in6p;
1100 #endif
1101 	struct socket *so;
1102 #ifdef RTV_RTT
1103 	struct rtentry *rt;
1104 #endif
1105 	struct route *ro;
1106 
1107 	inp = tp->t_inpcb;
1108 #ifdef INET6
1109 	in6p = tp->t_in6pcb;
1110 #endif
1111 	so = NULL;
1112 	ro = NULL;
1113 	if (inp) {
1114 		so = inp->inp_socket;
1115 		ro = &inp->inp_route;
1116 	}
1117 #ifdef INET6
1118 	else if (in6p) {
1119 		so = in6p->in6p_socket;
1120 		ro = (struct route *)&in6p->in6p_route;
1121 	}
1122 #endif
1123 
1124 #ifdef RTV_RTT
1125 	/*
1126 	 * If we sent enough data to get some meaningful characteristics,
1127 	 * save them in the routing entry.  'Enough' is arbitrarily
1128 	 * defined as the sendpipesize (default 4K) * 16.  This would
1129 	 * give us 16 rtt samples assuming we only get one sample per
1130 	 * window (the usual case on a long haul net).  16 samples is
1131 	 * enough for the srtt filter to converge to within 5% of the correct
1132 	 * value; fewer samples and we could save a very bogus rtt.
1133 	 *
1134 	 * Don't update the default route's characteristics and don't
1135 	 * update anything that the user "locked".
1136 	 */
1137 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1138 	    ro && (rt = ro->ro_rt) &&
1139 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
1140 		u_long i = 0;
1141 
1142 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1143 			i = tp->t_srtt *
1144 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1145 			if (rt->rt_rmx.rmx_rtt && i)
1146 				/*
1147 				 * filter this update to half the old & half
1148 				 * the new values, converting scale.
1149 				 * See route.h and tcp_var.h for a
1150 				 * description of the scaling constants.
1151 				 */
1152 				rt->rt_rmx.rmx_rtt =
1153 				    (rt->rt_rmx.rmx_rtt + i) / 2;
1154 			else
1155 				rt->rt_rmx.rmx_rtt = i;
1156 		}
1157 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1158 			i = tp->t_rttvar *
1159 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1160 			if (rt->rt_rmx.rmx_rttvar && i)
1161 				rt->rt_rmx.rmx_rttvar =
1162 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
1163 			else
1164 				rt->rt_rmx.rmx_rttvar = i;
1165 		}
1166 		/*
1167 		 * update the pipelimit (ssthresh) if it has been updated
1168 		 * already or if a pipesize was specified & the threshhold
1169 		 * got below half the pipesize.  I.e., wait for bad news
1170 		 * before we start updating, then update on both good
1171 		 * and bad news.
1172 		 */
1173 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1174 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1175 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1176 			/*
1177 			 * convert the limit from user data bytes to
1178 			 * packets then to packet data bytes.
1179 			 */
1180 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
1181 			if (i < 2)
1182 				i = 2;
1183 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1184 			if (rt->rt_rmx.rmx_ssthresh)
1185 				rt->rt_rmx.rmx_ssthresh =
1186 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
1187 			else
1188 				rt->rt_rmx.rmx_ssthresh = i;
1189 		}
1190 	}
1191 #endif /* RTV_RTT */
1192 	/* free the reassembly queue, if any */
1193 	TCP_REASS_LOCK(tp);
1194 	(void) tcp_freeq(tp);
1195 	TCP_REASS_UNLOCK(tp);
1196 
1197 	tcp_canceltimers(tp);
1198 	TCP_CLEAR_DELACK(tp);
1199 	syn_cache_cleanup(tp);
1200 
1201 	if (tp->t_template) {
1202 		m_free(tp->t_template);
1203 		tp->t_template = NULL;
1204 	}
1205 	if (tcp_timers_invoking(tp))
1206 		tp->t_flags |= TF_DEAD;
1207 	else
1208 		pool_put(&tcpcb_pool, tp);
1209 
1210 	if (inp) {
1211 		inp->inp_ppcb = 0;
1212 		soisdisconnected(so);
1213 		in_pcbdetach(inp);
1214 	}
1215 #ifdef INET6
1216 	else if (in6p) {
1217 		in6p->in6p_ppcb = 0;
1218 		soisdisconnected(so);
1219 		in6_pcbdetach(in6p);
1220 	}
1221 #endif
1222 	tcpstat.tcps_closed++;
1223 	return ((struct tcpcb *)0);
1224 }
1225 
1226 int
1227 tcp_freeq(tp)
1228 	struct tcpcb *tp;
1229 {
1230 	struct ipqent *qe;
1231 	int rv = 0;
1232 #ifdef TCPREASS_DEBUG
1233 	int i = 0;
1234 #endif
1235 
1236 	TCP_REASS_LOCK_CHECK(tp);
1237 
1238 	while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1239 #ifdef TCPREASS_DEBUG
1240 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1241 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1242 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1243 #endif
1244 		TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1245 		TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1246 		m_freem(qe->ipqe_m);
1247 		pool_put(&tcpipqent_pool, qe);
1248 		rv = 1;
1249 	}
1250 	return (rv);
1251 }
1252 
1253 /*
1254  * Protocol drain routine.  Called when memory is in short supply.
1255  */
1256 void
1257 tcp_drain()
1258 {
1259 	struct inpcb_hdr *inph;
1260 	struct tcpcb *tp;
1261 
1262 	/*
1263 	 * Free the sequence queue of all TCP connections.
1264 	 */
1265 	CIRCLEQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1266 		switch (inph->inph_af) {
1267 		case AF_INET:
1268 			tp = intotcpcb((struct inpcb *)inph);
1269 			break;
1270 #ifdef INET6
1271 		case AF_INET6:
1272 			tp = in6totcpcb((struct in6pcb *)inph);
1273 			break;
1274 #endif
1275 		default:
1276 			tp = NULL;
1277 			break;
1278 		}
1279 		if (tp != NULL) {
1280 			/*
1281 			 * We may be called from a device's interrupt
1282 			 * context.  If the tcpcb is already busy,
1283 			 * just bail out now.
1284 			 */
1285 			if (tcp_reass_lock_try(tp) == 0)
1286 				continue;
1287 			if (tcp_freeq(tp))
1288 				tcpstat.tcps_connsdrained++;
1289 			TCP_REASS_UNLOCK(tp);
1290 		}
1291 	}
1292 }
1293 
1294 /*
1295  * Notify a tcp user of an asynchronous error;
1296  * store error as soft error, but wake up user
1297  * (for now, won't do anything until can select for soft error).
1298  */
1299 void
1300 tcp_notify(inp, error)
1301 	struct inpcb *inp;
1302 	int error;
1303 {
1304 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1305 	struct socket *so = inp->inp_socket;
1306 
1307 	/*
1308 	 * Ignore some errors if we are hooked up.
1309 	 * If connection hasn't completed, has retransmitted several times,
1310 	 * and receives a second error, give up now.  This is better
1311 	 * than waiting a long time to establish a connection that
1312 	 * can never complete.
1313 	 */
1314 	if (tp->t_state == TCPS_ESTABLISHED &&
1315 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1316 	      error == EHOSTDOWN)) {
1317 		return;
1318 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1319 	    tp->t_rxtshift > 3 && tp->t_softerror)
1320 		so->so_error = error;
1321 	else
1322 		tp->t_softerror = error;
1323 	wakeup((caddr_t) &so->so_timeo);
1324 	sorwakeup(so);
1325 	sowwakeup(so);
1326 }
1327 
1328 #ifdef INET6
1329 void
1330 tcp6_notify(in6p, error)
1331 	struct in6pcb *in6p;
1332 	int error;
1333 {
1334 	struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1335 	struct socket *so = in6p->in6p_socket;
1336 
1337 	/*
1338 	 * Ignore some errors if we are hooked up.
1339 	 * If connection hasn't completed, has retransmitted several times,
1340 	 * and receives a second error, give up now.  This is better
1341 	 * than waiting a long time to establish a connection that
1342 	 * can never complete.
1343 	 */
1344 	if (tp->t_state == TCPS_ESTABLISHED &&
1345 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
1346 	      error == EHOSTDOWN)) {
1347 		return;
1348 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1349 	    tp->t_rxtshift > 3 && tp->t_softerror)
1350 		so->so_error = error;
1351 	else
1352 		tp->t_softerror = error;
1353 	wakeup((caddr_t) &so->so_timeo);
1354 	sorwakeup(so);
1355 	sowwakeup(so);
1356 }
1357 #endif
1358 
1359 #ifdef INET6
1360 void
1361 tcp6_ctlinput(cmd, sa, d)
1362 	int cmd;
1363 	struct sockaddr *sa;
1364 	void *d;
1365 {
1366 	struct tcphdr th;
1367 	void (*notify) __P((struct in6pcb *, int)) = tcp6_notify;
1368 	int nmatch;
1369 	struct ip6_hdr *ip6;
1370 	const struct sockaddr_in6 *sa6_src = NULL;
1371 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
1372 	struct mbuf *m;
1373 	int off;
1374 
1375 	if (sa->sa_family != AF_INET6 ||
1376 	    sa->sa_len != sizeof(struct sockaddr_in6))
1377 		return;
1378 	if ((unsigned)cmd >= PRC_NCMDS)
1379 		return;
1380 	else if (cmd == PRC_QUENCH) {
1381 		/* XXX there's no PRC_QUENCH in IPv6 */
1382 		notify = tcp6_quench;
1383 	} else if (PRC_IS_REDIRECT(cmd))
1384 		notify = in6_rtchange, d = NULL;
1385 	else if (cmd == PRC_MSGSIZE)
1386 		; /* special code is present, see below */
1387 	else if (cmd == PRC_HOSTDEAD)
1388 		d = NULL;
1389 	else if (inet6ctlerrmap[cmd] == 0)
1390 		return;
1391 
1392 	/* if the parameter is from icmp6, decode it. */
1393 	if (d != NULL) {
1394 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1395 		m = ip6cp->ip6c_m;
1396 		ip6 = ip6cp->ip6c_ip6;
1397 		off = ip6cp->ip6c_off;
1398 		sa6_src = ip6cp->ip6c_src;
1399 	} else {
1400 		m = NULL;
1401 		ip6 = NULL;
1402 		sa6_src = &sa6_any;
1403 		off = 0;
1404 	}
1405 
1406 	if (ip6) {
1407 		/*
1408 		 * XXX: We assume that when ip6 is non NULL,
1409 		 * M and OFF are valid.
1410 		 */
1411 
1412 		/* check if we can safely examine src and dst ports */
1413 		if (m->m_pkthdr.len < off + sizeof(th)) {
1414 			if (cmd == PRC_MSGSIZE)
1415 				icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1416 			return;
1417 		}
1418 
1419 		bzero(&th, sizeof(th));
1420 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
1421 
1422 		if (cmd == PRC_MSGSIZE) {
1423 			int valid = 0;
1424 
1425 			/*
1426 			 * Check to see if we have a valid TCP connection
1427 			 * corresponding to the address in the ICMPv6 message
1428 			 * payload.
1429 			 */
1430 			if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1431 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
1432 			    th.th_sport, 0))
1433 				valid++;
1434 
1435 			/*
1436 			 * Depending on the value of "valid" and routing table
1437 			 * size (mtudisc_{hi,lo}wat), we will:
1438 			 * - recalcurate the new MTU and create the
1439 			 *   corresponding routing entry, or
1440 			 * - ignore the MTU change notification.
1441 			 */
1442 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1443 
1444 			/*
1445 			 * no need to call in6_pcbnotify, it should have been
1446 			 * called via callback if necessary
1447 			 */
1448 			return;
1449 		}
1450 
1451 		nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1452 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1453 		if (nmatch == 0 && syn_cache_count &&
1454 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1455 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
1456 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
1457 			syn_cache_unreach((struct sockaddr *)sa6_src,
1458 					  sa, &th);
1459 	} else {
1460 		(void) in6_pcbnotify(&tcbtable, sa, 0,
1461 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1462 	}
1463 }
1464 #endif
1465 
1466 #ifdef INET
1467 /* assumes that ip header and tcp header are contiguous on mbuf */
1468 void *
1469 tcp_ctlinput(cmd, sa, v)
1470 	int cmd;
1471 	struct sockaddr *sa;
1472 	void *v;
1473 {
1474 	struct ip *ip = v;
1475 	struct tcphdr *th;
1476 	struct icmp *icp;
1477 	extern const int inetctlerrmap[];
1478 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1479 	int errno;
1480 	int nmatch;
1481 #ifdef INET6
1482 	struct in6_addr src6, dst6;
1483 #endif
1484 
1485 	if (sa->sa_family != AF_INET ||
1486 	    sa->sa_len != sizeof(struct sockaddr_in))
1487 		return NULL;
1488 	if ((unsigned)cmd >= PRC_NCMDS)
1489 		return NULL;
1490 	errno = inetctlerrmap[cmd];
1491 	if (cmd == PRC_QUENCH)
1492 		notify = tcp_quench;
1493 	else if (PRC_IS_REDIRECT(cmd))
1494 		notify = in_rtchange, ip = 0;
1495 	else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1496 		/*
1497 		 * Check to see if we have a valid TCP connection
1498 		 * corresponding to the address in the ICMP message
1499 		 * payload.
1500 		 *
1501 		 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1502 		 */
1503 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1504 #ifdef INET6
1505 		memset(&src6, 0, sizeof(src6));
1506 		memset(&dst6, 0, sizeof(dst6));
1507 		src6.s6_addr16[5] = dst6.s6_addr16[5] = 0xffff;
1508 		memcpy(&src6.s6_addr32[3], &ip->ip_src, sizeof(struct in_addr));
1509 		memcpy(&dst6.s6_addr32[3], &ip->ip_dst, sizeof(struct in_addr));
1510 #endif
1511 		if (in_pcblookup_connect(&tcbtable, ip->ip_dst, th->th_dport,
1512 		    ip->ip_src, th->th_sport) != NULL)
1513 			;
1514 #ifdef INET6
1515 		else if (in6_pcblookup_connect(&tcbtable, &dst6,
1516 		    th->th_dport, &src6, th->th_sport, 0) != NULL)
1517 			;
1518 #endif
1519 		else
1520 			return NULL;
1521 
1522 		/*
1523 		 * Now that we've validated that we are actually communicating
1524 		 * with the host indicated in the ICMP message, locate the
1525 		 * ICMP header, recalculate the new MTU, and create the
1526 		 * corresponding routing entry.
1527 		 */
1528 		icp = (struct icmp *)((caddr_t)ip -
1529 		    offsetof(struct icmp, icmp_ip));
1530 		icmp_mtudisc(icp, ip->ip_dst);
1531 
1532 		return NULL;
1533 	} else if (cmd == PRC_HOSTDEAD)
1534 		ip = 0;
1535 	else if (errno == 0)
1536 		return NULL;
1537 	if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1538 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1539 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
1540 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1541 		if (nmatch == 0 && syn_cache_count &&
1542 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
1543 		    inetctlerrmap[cmd] == ENETUNREACH ||
1544 		    inetctlerrmap[cmd] == EHOSTDOWN)) {
1545 			struct sockaddr_in sin;
1546 			bzero(&sin, sizeof(sin));
1547 			sin.sin_len = sizeof(sin);
1548 			sin.sin_family = AF_INET;
1549 			sin.sin_port = th->th_sport;
1550 			sin.sin_addr = ip->ip_src;
1551 			syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1552 		}
1553 
1554 		/* XXX mapped address case */
1555 	} else
1556 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
1557 		    notify);
1558 	return NULL;
1559 }
1560 
1561 /*
1562  * When a source quence is received, we are being notifed of congestion.
1563  * Close the congestion window down to the Loss Window (one segment).
1564  * We will gradually open it again as we proceed.
1565  */
1566 void
1567 tcp_quench(inp, errno)
1568 	struct inpcb *inp;
1569 	int errno;
1570 {
1571 	struct tcpcb *tp = intotcpcb(inp);
1572 
1573 	if (tp)
1574 		tp->snd_cwnd = tp->t_segsz;
1575 }
1576 #endif
1577 
1578 #ifdef INET6
1579 void
1580 tcp6_quench(in6p, errno)
1581 	struct in6pcb *in6p;
1582 	int errno;
1583 {
1584 	struct tcpcb *tp = in6totcpcb(in6p);
1585 
1586 	if (tp)
1587 		tp->snd_cwnd = tp->t_segsz;
1588 }
1589 #endif
1590 
1591 #ifdef INET
1592 /*
1593  * Path MTU Discovery handlers.
1594  */
1595 void
1596 tcp_mtudisc_callback(faddr)
1597 	struct in_addr faddr;
1598 {
1599 #ifdef INET6
1600 	struct in6_addr in6;
1601 #endif
1602 
1603 	in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1604 #ifdef INET6
1605 	memset(&in6, 0, sizeof(in6));
1606 	in6.s6_addr16[5] = 0xffff;
1607 	memcpy(&in6.s6_addr32[3], &faddr, sizeof(struct in_addr));
1608 	tcp6_mtudisc_callback(&in6);
1609 #endif
1610 }
1611 
1612 /*
1613  * On receipt of path MTU corrections, flush old route and replace it
1614  * with the new one.  Retransmit all unacknowledged packets, to ensure
1615  * that all packets will be received.
1616  */
1617 void
1618 tcp_mtudisc(inp, errno)
1619 	struct inpcb *inp;
1620 	int errno;
1621 {
1622 	struct tcpcb *tp = intotcpcb(inp);
1623 	struct rtentry *rt = in_pcbrtentry(inp);
1624 
1625 	if (tp != 0) {
1626 		if (rt != 0) {
1627 			/*
1628 			 * If this was not a host route, remove and realloc.
1629 			 */
1630 			if ((rt->rt_flags & RTF_HOST) == 0) {
1631 				in_rtchange(inp, errno);
1632 				if ((rt = in_pcbrtentry(inp)) == 0)
1633 					return;
1634 			}
1635 
1636 			/*
1637 			 * Slow start out of the error condition.  We
1638 			 * use the MTU because we know it's smaller
1639 			 * than the previously transmitted segment.
1640 			 *
1641 			 * Note: This is more conservative than the
1642 			 * suggestion in draft-floyd-incr-init-win-03.
1643 			 */
1644 			if (rt->rt_rmx.rmx_mtu != 0)
1645 				tp->snd_cwnd =
1646 				    TCP_INITIAL_WINDOW(tcp_init_win,
1647 				    rt->rt_rmx.rmx_mtu);
1648 		}
1649 
1650 		/*
1651 		 * Resend unacknowledged packets.
1652 		 */
1653 		tp->snd_nxt = tp->snd_una;
1654 		tcp_output(tp);
1655 	}
1656 }
1657 #endif
1658 
1659 #ifdef INET6
1660 /*
1661  * Path MTU Discovery handlers.
1662  */
1663 void
1664 tcp6_mtudisc_callback(faddr)
1665 	struct in6_addr *faddr;
1666 {
1667 	struct sockaddr_in6 sin6;
1668 
1669 	bzero(&sin6, sizeof(sin6));
1670 	sin6.sin6_family = AF_INET6;
1671 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1672 	sin6.sin6_addr = *faddr;
1673 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1674 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1675 }
1676 
1677 void
1678 tcp6_mtudisc(in6p, errno)
1679 	struct in6pcb *in6p;
1680 	int errno;
1681 {
1682 	struct tcpcb *tp = in6totcpcb(in6p);
1683 	struct rtentry *rt = in6_pcbrtentry(in6p);
1684 
1685 	if (tp != 0) {
1686 		if (rt != 0) {
1687 			/*
1688 			 * If this was not a host route, remove and realloc.
1689 			 */
1690 			if ((rt->rt_flags & RTF_HOST) == 0) {
1691 				in6_rtchange(in6p, errno);
1692 				if ((rt = in6_pcbrtentry(in6p)) == 0)
1693 					return;
1694 			}
1695 
1696 			/*
1697 			 * Slow start out of the error condition.  We
1698 			 * use the MTU because we know it's smaller
1699 			 * than the previously transmitted segment.
1700 			 *
1701 			 * Note: This is more conservative than the
1702 			 * suggestion in draft-floyd-incr-init-win-03.
1703 			 */
1704 			if (rt->rt_rmx.rmx_mtu != 0)
1705 				tp->snd_cwnd =
1706 				    TCP_INITIAL_WINDOW(tcp_init_win,
1707 				    rt->rt_rmx.rmx_mtu);
1708 		}
1709 
1710 		/*
1711 		 * Resend unacknowledged packets.
1712 		 */
1713 		tp->snd_nxt = tp->snd_una;
1714 		tcp_output(tp);
1715 	}
1716 }
1717 #endif /* INET6 */
1718 
1719 /*
1720  * Compute the MSS to advertise to the peer.  Called only during
1721  * the 3-way handshake.  If we are the server (peer initiated
1722  * connection), we are called with a pointer to the interface
1723  * on which the SYN packet arrived.  If we are the client (we
1724  * initiated connection), we are called with a pointer to the
1725  * interface out which this connection should go.
1726  *
1727  * NOTE: Do not subtract IP option/extension header size nor IPsec
1728  * header size from MSS advertisement.  MSS option must hold the maximum
1729  * segment size we can accept, so it must always be:
1730  *	 max(if mtu) - ip header - tcp header
1731  */
1732 u_long
1733 tcp_mss_to_advertise(ifp, af)
1734 	const struct ifnet *ifp;
1735 	int af;
1736 {
1737 	extern u_long in_maxmtu;
1738 	u_long mss = 0;
1739 	u_long hdrsiz;
1740 
1741 	/*
1742 	 * In order to avoid defeating path MTU discovery on the peer,
1743 	 * we advertise the max MTU of all attached networks as our MSS,
1744 	 * per RFC 1191, section 3.1.
1745 	 *
1746 	 * We provide the option to advertise just the MTU of
1747 	 * the interface on which we hope this connection will
1748 	 * be receiving.  If we are responding to a SYN, we
1749 	 * will have a pretty good idea about this, but when
1750 	 * initiating a connection there is a bit more doubt.
1751 	 *
1752 	 * We also need to ensure that loopback has a large enough
1753 	 * MSS, as the loopback MTU is never included in in_maxmtu.
1754 	 */
1755 
1756 	if (ifp != NULL)
1757 		switch (af) {
1758 		case AF_INET:
1759 			mss = ifp->if_mtu;
1760 			break;
1761 #ifdef INET6
1762 		case AF_INET6:
1763 			mss = IN6_LINKMTU(ifp);
1764 			break;
1765 #endif
1766 		}
1767 
1768 	if (tcp_mss_ifmtu == 0)
1769 		switch (af) {
1770 		case AF_INET:
1771 			mss = max(in_maxmtu, mss);
1772 			break;
1773 #ifdef INET6
1774 		case AF_INET6:
1775 			mss = max(in6_maxmtu, mss);
1776 			break;
1777 #endif
1778 		}
1779 
1780 	switch (af) {
1781 	case AF_INET:
1782 		hdrsiz = sizeof(struct ip);
1783 		break;
1784 #ifdef INET6
1785 	case AF_INET6:
1786 		hdrsiz = sizeof(struct ip6_hdr);
1787 		break;
1788 #endif
1789 	default:
1790 		hdrsiz = 0;
1791 		break;
1792 	}
1793 	hdrsiz += sizeof(struct tcphdr);
1794 	if (mss > hdrsiz)
1795 		mss -= hdrsiz;
1796 
1797 	mss = max(tcp_mssdflt, mss);
1798 	return (mss);
1799 }
1800 
1801 /*
1802  * Set connection variables based on the peer's advertised MSS.
1803  * We are passed the TCPCB for the actual connection.  If we
1804  * are the server, we are called by the compressed state engine
1805  * when the 3-way handshake is complete.  If we are the client,
1806  * we are called when we receive the SYN,ACK from the server.
1807  *
1808  * NOTE: Our advertised MSS value must be initialized in the TCPCB
1809  * before this routine is called!
1810  */
1811 void
1812 tcp_mss_from_peer(tp, offer)
1813 	struct tcpcb *tp;
1814 	int offer;
1815 {
1816 	struct socket *so;
1817 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1818 	struct rtentry *rt;
1819 #endif
1820 	u_long bufsize;
1821 	int mss;
1822 
1823 #ifdef DIAGNOSTIC
1824 	if (tp->t_inpcb && tp->t_in6pcb)
1825 		panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1826 #endif
1827 	so = NULL;
1828 	rt = NULL;
1829 #ifdef INET
1830 	if (tp->t_inpcb) {
1831 		so = tp->t_inpcb->inp_socket;
1832 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1833 		rt = in_pcbrtentry(tp->t_inpcb);
1834 #endif
1835 	}
1836 #endif
1837 #ifdef INET6
1838 	if (tp->t_in6pcb) {
1839 		so = tp->t_in6pcb->in6p_socket;
1840 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1841 		rt = in6_pcbrtentry(tp->t_in6pcb);
1842 #endif
1843 	}
1844 #endif
1845 
1846 	/*
1847 	 * As per RFC1122, use the default MSS value, unless they
1848 	 * sent us an offer.  Do not accept offers less than 256 bytes.
1849 	 */
1850 	mss = tcp_mssdflt;
1851 	if (offer)
1852 		mss = offer;
1853 	mss = max(mss, 256);		/* sanity */
1854 	tp->t_peermss = mss;
1855 	mss -= tcp_optlen(tp);
1856 #ifdef INET
1857 	if (tp->t_inpcb)
1858 		mss -= ip_optlen(tp->t_inpcb);
1859 #endif
1860 #ifdef INET6
1861 	if (tp->t_in6pcb)
1862 		mss -= ip6_optlen(tp->t_in6pcb);
1863 #endif
1864 
1865 	/*
1866 	 * If there's a pipesize, change the socket buffer to that size.
1867 	 * Make the socket buffer an integral number of MSS units.  If
1868 	 * the MSS is larger than the socket buffer, artificially decrease
1869 	 * the MSS.
1870 	 */
1871 #ifdef RTV_SPIPE
1872 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1873 		bufsize = rt->rt_rmx.rmx_sendpipe;
1874 	else
1875 #endif
1876 		bufsize = so->so_snd.sb_hiwat;
1877 	if (bufsize < mss)
1878 		mss = bufsize;
1879 	else {
1880 		bufsize = roundup(bufsize, mss);
1881 		if (bufsize > sb_max)
1882 			bufsize = sb_max;
1883 		(void) sbreserve(&so->so_snd, bufsize, so);
1884 	}
1885 	tp->t_segsz = mss;
1886 
1887 #ifdef RTV_SSTHRESH
1888 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1889 		/*
1890 		 * There's some sort of gateway or interface buffer
1891 		 * limit on the path.  Use this to set the slow
1892 		 * start threshold, but set the threshold to no less
1893 		 * than 2 * MSS.
1894 		 */
1895 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1896 	}
1897 #endif
1898 }
1899 
1900 /*
1901  * Processing necessary when a TCP connection is established.
1902  */
1903 void
1904 tcp_established(tp)
1905 	struct tcpcb *tp;
1906 {
1907 	struct socket *so;
1908 #ifdef RTV_RPIPE
1909 	struct rtentry *rt;
1910 #endif
1911 	u_long bufsize;
1912 
1913 #ifdef DIAGNOSTIC
1914 	if (tp->t_inpcb && tp->t_in6pcb)
1915 		panic("tcp_established: both t_inpcb and t_in6pcb are set");
1916 #endif
1917 	so = NULL;
1918 	rt = NULL;
1919 #ifdef INET
1920 	if (tp->t_inpcb) {
1921 		so = tp->t_inpcb->inp_socket;
1922 #if defined(RTV_RPIPE)
1923 		rt = in_pcbrtentry(tp->t_inpcb);
1924 #endif
1925 	}
1926 #endif
1927 #ifdef INET6
1928 	if (tp->t_in6pcb) {
1929 		so = tp->t_in6pcb->in6p_socket;
1930 #if defined(RTV_RPIPE)
1931 		rt = in6_pcbrtentry(tp->t_in6pcb);
1932 #endif
1933 	}
1934 #endif
1935 
1936 	tp->t_state = TCPS_ESTABLISHED;
1937 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1938 
1939 #ifdef RTV_RPIPE
1940 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1941 		bufsize = rt->rt_rmx.rmx_recvpipe;
1942 	else
1943 #endif
1944 		bufsize = so->so_rcv.sb_hiwat;
1945 	if (bufsize > tp->t_ourmss) {
1946 		bufsize = roundup(bufsize, tp->t_ourmss);
1947 		if (bufsize > sb_max)
1948 			bufsize = sb_max;
1949 		(void) sbreserve(&so->so_rcv, bufsize, so);
1950 	}
1951 }
1952 
1953 /*
1954  * Check if there's an initial rtt or rttvar.  Convert from the
1955  * route-table units to scaled multiples of the slow timeout timer.
1956  * Called only during the 3-way handshake.
1957  */
1958 void
1959 tcp_rmx_rtt(tp)
1960 	struct tcpcb *tp;
1961 {
1962 #ifdef RTV_RTT
1963 	struct rtentry *rt = NULL;
1964 	int rtt;
1965 
1966 #ifdef DIAGNOSTIC
1967 	if (tp->t_inpcb && tp->t_in6pcb)
1968 		panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
1969 #endif
1970 #ifdef INET
1971 	if (tp->t_inpcb)
1972 		rt = in_pcbrtentry(tp->t_inpcb);
1973 #endif
1974 #ifdef INET6
1975 	if (tp->t_in6pcb)
1976 		rt = in6_pcbrtentry(tp->t_in6pcb);
1977 #endif
1978 	if (rt == NULL)
1979 		return;
1980 
1981 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1982 		/*
1983 		 * XXX The lock bit for MTU indicates that the value
1984 		 * is also a minimum value; this is subject to time.
1985 		 */
1986 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1987 			TCPT_RANGESET(tp->t_rttmin,
1988 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1989 			    TCPTV_MIN, TCPTV_REXMTMAX);
1990 		tp->t_srtt = rtt /
1991 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1992 		if (rt->rt_rmx.rmx_rttvar) {
1993 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1994 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
1995 				(TCP_RTTVAR_SHIFT + 2));
1996 		} else {
1997 			/* Default variation is +- 1 rtt */
1998 			tp->t_rttvar =
1999 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2000 		}
2001 		TCPT_RANGESET(tp->t_rxtcur,
2002 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2003 		    tp->t_rttmin, TCPTV_REXMTMAX);
2004 	}
2005 #endif
2006 }
2007 
2008 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
2009 #if NRND > 0
2010 u_int8_t tcp_iss_secret[16];	/* 128 bits; should be plenty */
2011 #endif
2012 
2013 /*
2014  * Get a new sequence value given a tcp control block
2015  */
2016 tcp_seq
2017 tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2018 {
2019 
2020 #ifdef INET
2021 	if (tp->t_inpcb != NULL) {
2022 		return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2023 		    &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2024 		    tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2025 		    addin));
2026 	}
2027 #endif
2028 #ifdef INET6
2029 	if (tp->t_in6pcb != NULL) {
2030 		return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2031 		    &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2032 		    tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2033 		    addin));
2034 	}
2035 #endif
2036 	/* Not possible. */
2037 	panic("tcp_new_iss");
2038 }
2039 
2040 /*
2041  * This routine actually generates a new TCP initial sequence number.
2042  */
2043 tcp_seq
2044 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2045     size_t addrsz, tcp_seq addin)
2046 {
2047 	tcp_seq tcp_iss;
2048 
2049 #if NRND > 0
2050 	static int beenhere;
2051 
2052 	/*
2053 	 * If we haven't been here before, initialize our cryptographic
2054 	 * hash secret.
2055 	 */
2056 	if (beenhere == 0) {
2057 		rnd_extract_data(tcp_iss_secret, sizeof(tcp_iss_secret),
2058 		    RND_EXTRACT_ANY);
2059 		beenhere = 1;
2060 	}
2061 
2062 	if (tcp_do_rfc1948) {
2063 		MD5_CTX ctx;
2064 		u_int8_t hash[16];	/* XXX MD5 knowledge */
2065 
2066 		/*
2067 		 * Compute the base value of the ISS.  It is a hash
2068 		 * of (saddr, sport, daddr, dport, secret).
2069 		 */
2070 		MD5Init(&ctx);
2071 
2072 		MD5Update(&ctx, (u_char *) laddr, addrsz);
2073 		MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2074 
2075 		MD5Update(&ctx, (u_char *) faddr, addrsz);
2076 		MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2077 
2078 		MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2079 
2080 		MD5Final(hash, &ctx);
2081 
2082 		memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2083 
2084 		/*
2085 		 * Now increment our "timer", and add it in to
2086 		 * the computed value.
2087 		 *
2088 		 * XXX Use `addin'?
2089 		 * XXX TCP_ISSINCR too large to use?
2090 		 */
2091 		tcp_iss_seq += TCP_ISSINCR;
2092 #ifdef TCPISS_DEBUG
2093 		printf("ISS hash 0x%08x, ", tcp_iss);
2094 #endif
2095 		tcp_iss += tcp_iss_seq + addin;
2096 #ifdef TCPISS_DEBUG
2097 		printf("new ISS 0x%08x\n", tcp_iss);
2098 #endif
2099 	} else
2100 #endif /* NRND > 0 */
2101 	{
2102 		/*
2103 		 * Randomize.
2104 		 */
2105 #if NRND > 0
2106 		rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
2107 #else
2108 		tcp_iss = arc4random();
2109 #endif
2110 
2111 		/*
2112 		 * If we were asked to add some amount to a known value,
2113 		 * we will take a random value obtained above, mask off
2114 		 * the upper bits, and add in the known value.  We also
2115 		 * add in a constant to ensure that we are at least a
2116 		 * certain distance from the original value.
2117 		 *
2118 		 * This is used when an old connection is in timed wait
2119 		 * and we have a new one coming in, for instance.
2120 		 */
2121 		if (addin != 0) {
2122 #ifdef TCPISS_DEBUG
2123 			printf("Random %08x, ", tcp_iss);
2124 #endif
2125 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2126 			tcp_iss += addin + TCP_ISSINCR;
2127 #ifdef TCPISS_DEBUG
2128 			printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2129 #endif
2130 		} else {
2131 			tcp_iss &= TCP_ISS_RANDOM_MASK;
2132 			tcp_iss += tcp_iss_seq;
2133 			tcp_iss_seq += TCP_ISSINCR;
2134 #ifdef TCPISS_DEBUG
2135 			printf("ISS %08x\n", tcp_iss);
2136 #endif
2137 		}
2138 	}
2139 
2140 	if (tcp_compat_42) {
2141 		/*
2142 		 * Limit it to the positive range for really old TCP
2143 		 * implementations.
2144 		 * Just AND off the top bit instead of checking if
2145 		 * is set first - saves a branch 50% of the time.
2146 		 */
2147 		tcp_iss &= 0x7fffffff;		/* XXX */
2148 	}
2149 
2150 	return (tcp_iss);
2151 }
2152 
2153 #if defined(IPSEC) || defined(FAST_IPSEC)
2154 /* compute ESP/AH header size for TCP, including outer IP header. */
2155 size_t
2156 ipsec4_hdrsiz_tcp(tp)
2157 	struct tcpcb *tp;
2158 {
2159 	struct inpcb *inp;
2160 	size_t hdrsiz;
2161 
2162 	/* XXX mapped addr case (tp->t_in6pcb) */
2163 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2164 		return 0;
2165 	switch (tp->t_family) {
2166 	case AF_INET:
2167 		/* XXX: should use currect direction. */
2168 		hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2169 		break;
2170 	default:
2171 		hdrsiz = 0;
2172 		break;
2173 	}
2174 
2175 	return hdrsiz;
2176 }
2177 
2178 #ifdef INET6
2179 size_t
2180 ipsec6_hdrsiz_tcp(tp)
2181 	struct tcpcb *tp;
2182 {
2183 	struct in6pcb *in6p;
2184 	size_t hdrsiz;
2185 
2186 	if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2187 		return 0;
2188 	switch (tp->t_family) {
2189 	case AF_INET6:
2190 		/* XXX: should use currect direction. */
2191 		hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2192 		break;
2193 	case AF_INET:
2194 		/* mapped address case - tricky */
2195 	default:
2196 		hdrsiz = 0;
2197 		break;
2198 	}
2199 
2200 	return hdrsiz;
2201 }
2202 #endif
2203 #endif /*IPSEC*/
2204 
2205 /*
2206  * Determine the length of the TCP options for this connection.
2207  *
2208  * XXX:  What do we do for SACK, when we add that?  Just reserve
2209  *       all of the space?  Otherwise we can't exactly be incrementing
2210  *       cwnd by an amount that varies depending on the amount we last
2211  *       had to SACK!
2212  */
2213 
2214 u_int
2215 tcp_optlen(tp)
2216 	struct tcpcb *tp;
2217 {
2218 	u_int optlen;
2219 
2220 	optlen = 0;
2221 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2222 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2223 		optlen += TCPOLEN_TSTAMP_APPA;
2224 
2225 #ifdef TCP_SIGNATURE
2226 #if defined(INET6) && defined(FAST_IPSEC)
2227 	if (tp->t_family == AF_INET)
2228 #endif
2229 	if (tp->t_flags & TF_SIGNATURE)
2230 		optlen += TCPOLEN_SIGNATURE + 2;
2231 #endif /* TCP_SIGNATURE */
2232 
2233 	return optlen;
2234 }
2235