xref: /netbsd-src/sys/netinet/tcp_input.c (revision e1e74f372d4580af0ad34c25f04bfd9fa29bb82c)
1 /*	$NetBSD: tcp_input.c,v 1.282 2008/03/01 14:16:52 rmind 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   1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  *      This product includes software developed by the University of
47  *      California, Berkeley and its contributors.
48  *      This product includes software developed at the Information
49  *      Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 /*-
73  * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc.
74  * All rights reserved.
75  *
76  * This code is derived from software contributed to The NetBSD Foundation
77  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
78  * Facility, NASA Ames Research Center.
79  * This code is derived from software contributed to The NetBSD Foundation
80  * by Charles M. Hannum.
81  * This code is derived from software contributed to The NetBSD Foundation
82  * by Rui Paulo.
83  *
84  * Redistribution and use in source and binary forms, with or without
85  * modification, are permitted provided that the following conditions
86  * are met:
87  * 1. Redistributions of source code must retain the above copyright
88  *    notice, this list of conditions and the following disclaimer.
89  * 2. Redistributions in binary form must reproduce the above copyright
90  *    notice, this list of conditions and the following disclaimer in the
91  *    documentation and/or other materials provided with the distribution.
92  * 3. All advertising materials mentioning features or use of this software
93  *    must display the following acknowledgement:
94  *	This product includes software developed by the NetBSD
95  *	Foundation, Inc. and its contributors.
96  * 4. Neither the name of The NetBSD Foundation nor the names of its
97  *    contributors may be used to endorse or promote products derived
98  *    from this software without specific prior written permission.
99  *
100  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
101  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
102  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
103  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
104  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
105  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
106  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
107  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
108  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
109  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
110  * POSSIBILITY OF SUCH DAMAGE.
111  */
112 
113 /*
114  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
115  *	The Regents of the University of California.  All rights reserved.
116  *
117  * Redistribution and use in source and binary forms, with or without
118  * modification, are permitted provided that the following conditions
119  * are met:
120  * 1. Redistributions of source code must retain the above copyright
121  *    notice, this list of conditions and the following disclaimer.
122  * 2. Redistributions in binary form must reproduce the above copyright
123  *    notice, this list of conditions and the following disclaimer in the
124  *    documentation and/or other materials provided with the distribution.
125  * 3. Neither the name of the University nor the names of its contributors
126  *    may be used to endorse or promote products derived from this software
127  *    without specific prior written permission.
128  *
129  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
130  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
131  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
132  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
133  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
134  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
135  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
136  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
137  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
138  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
139  * SUCH DAMAGE.
140  *
141  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
142  */
143 
144 /*
145  *	TODO list for SYN cache stuff:
146  *
147  *	Find room for a "state" field, which is needed to keep a
148  *	compressed state for TIME_WAIT TCBs.  It's been noted already
149  *	that this is fairly important for very high-volume web and
150  *	mail servers, which use a large number of short-lived
151  *	connections.
152  */
153 
154 #include <sys/cdefs.h>
155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.282 2008/03/01 14:16:52 rmind Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 #ifdef TCP_SIGNATURE
175 #include <sys/md5.h>
176 #endif
177 #include <sys/lwp.h> /* for lwp0 */
178 
179 #include <net/if.h>
180 #include <net/route.h>
181 #include <net/if_types.h>
182 
183 #include <netinet/in.h>
184 #include <netinet/in_systm.h>
185 #include <netinet/ip.h>
186 #include <netinet/in_pcb.h>
187 #include <netinet/in_var.h>
188 #include <netinet/ip_var.h>
189 #include <netinet/in_offload.h>
190 
191 #ifdef INET6
192 #ifndef INET
193 #include <netinet/in.h>
194 #endif
195 #include <netinet/ip6.h>
196 #include <netinet6/ip6_var.h>
197 #include <netinet6/in6_pcb.h>
198 #include <netinet6/ip6_var.h>
199 #include <netinet6/in6_var.h>
200 #include <netinet/icmp6.h>
201 #include <netinet6/nd6.h>
202 #ifdef TCP_SIGNATURE
203 #include <netinet6/scope6_var.h>
204 #endif
205 #endif
206 
207 #ifndef INET6
208 /* always need ip6.h for IP6_EXTHDR_GET */
209 #include <netinet/ip6.h>
210 #endif
211 
212 #include <netinet/tcp.h>
213 #include <netinet/tcp_fsm.h>
214 #include <netinet/tcp_seq.h>
215 #include <netinet/tcp_timer.h>
216 #include <netinet/tcp_var.h>
217 #include <netinet/tcpip.h>
218 #include <netinet/tcp_congctl.h>
219 #include <netinet/tcp_debug.h>
220 
221 #include <machine/stdarg.h>
222 
223 #ifdef IPSEC
224 #include <netinet6/ipsec.h>
225 #include <netkey/key.h>
226 #endif /*IPSEC*/
227 #ifdef INET6
228 #include "faith.h"
229 #if defined(NFAITH) && NFAITH > 0
230 #include <net/if_faith.h>
231 #endif
232 #endif	/* IPSEC */
233 
234 #ifdef FAST_IPSEC
235 #include <netipsec/ipsec.h>
236 #include <netipsec/ipsec_var.h>			/* XXX ipsecstat namespace */
237 #include <netipsec/key.h>
238 #ifdef INET6
239 #include <netipsec/ipsec6.h>
240 #endif
241 #endif	/* FAST_IPSEC*/
242 
243 int	tcprexmtthresh = 3;
244 int	tcp_log_refused;
245 
246 int	tcp_do_autorcvbuf = 0;
247 int	tcp_autorcvbuf_inc = 16 * 1024;
248 int	tcp_autorcvbuf_max = 256 * 1024;
249 
250 static int tcp_rst_ppslim_count = 0;
251 static struct timeval tcp_rst_ppslim_last;
252 static int tcp_ackdrop_ppslim_count = 0;
253 static struct timeval tcp_ackdrop_ppslim_last;
254 
255 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
256 
257 /* for modulo comparisons of timestamps */
258 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
259 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
260 
261 /*
262  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
263  */
264 #ifdef INET6
265 static inline void
266 nd6_hint(struct tcpcb *tp)
267 {
268 	struct rtentry *rt;
269 
270 	if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 &&
271 	    (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL)
272 		nd6_nud_hint(rt, NULL, 0);
273 }
274 #else
275 static inline void
276 nd6_hint(struct tcpcb *tp)
277 {
278 }
279 #endif
280 
281 /*
282  * Compute ACK transmission behavior.  Delay the ACK unless
283  * we have already delayed an ACK (must send an ACK every two segments).
284  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
285  * option is enabled.
286  */
287 static void
288 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th)
289 {
290 
291 	if (tp->t_flags & TF_DELACK ||
292 	    (tcp_ack_on_push && th->th_flags & TH_PUSH))
293 		tp->t_flags |= TF_ACKNOW;
294 	else
295 		TCP_SET_DELACK(tp);
296 }
297 
298 static void
299 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked)
300 {
301 
302 	/*
303 	 * If we had a pending ICMP message that refers to data that have
304 	 * just been acknowledged, disregard the recorded ICMP message.
305 	 */
306 	if ((tp->t_flags & TF_PMTUD_PEND) &&
307 	    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
308 		tp->t_flags &= ~TF_PMTUD_PEND;
309 
310 	/*
311 	 * Keep track of the largest chunk of data
312 	 * acknowledged since last PMTU update
313 	 */
314 	if (tp->t_pmtud_mss_acked < acked)
315 		tp->t_pmtud_mss_acked = acked;
316 }
317 
318 /*
319  * Convert TCP protocol fields to host order for easier processing.
320  */
321 static void
322 tcp_fields_to_host(struct tcphdr *th)
323 {
324 
325 	NTOHL(th->th_seq);
326 	NTOHL(th->th_ack);
327 	NTOHS(th->th_win);
328 	NTOHS(th->th_urp);
329 }
330 
331 /*
332  * ... and reverse the above.
333  */
334 static void
335 tcp_fields_to_net(struct tcphdr *th)
336 {
337 
338 	HTONL(th->th_seq);
339 	HTONL(th->th_ack);
340 	HTONS(th->th_win);
341 	HTONS(th->th_urp);
342 }
343 
344 #ifdef TCP_CSUM_COUNTERS
345 #include <sys/device.h>
346 
347 #if defined(INET)
348 extern struct evcnt tcp_hwcsum_ok;
349 extern struct evcnt tcp_hwcsum_bad;
350 extern struct evcnt tcp_hwcsum_data;
351 extern struct evcnt tcp_swcsum;
352 #endif /* defined(INET) */
353 #if defined(INET6)
354 extern struct evcnt tcp6_hwcsum_ok;
355 extern struct evcnt tcp6_hwcsum_bad;
356 extern struct evcnt tcp6_hwcsum_data;
357 extern struct evcnt tcp6_swcsum;
358 #endif /* defined(INET6) */
359 
360 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
361 
362 #else
363 
364 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
365 
366 #endif /* TCP_CSUM_COUNTERS */
367 
368 #ifdef TCP_REASS_COUNTERS
369 #include <sys/device.h>
370 
371 extern struct evcnt tcp_reass_;
372 extern struct evcnt tcp_reass_empty;
373 extern struct evcnt tcp_reass_iteration[8];
374 extern struct evcnt tcp_reass_prependfirst;
375 extern struct evcnt tcp_reass_prepend;
376 extern struct evcnt tcp_reass_insert;
377 extern struct evcnt tcp_reass_inserttail;
378 extern struct evcnt tcp_reass_append;
379 extern struct evcnt tcp_reass_appendtail;
380 extern struct evcnt tcp_reass_overlaptail;
381 extern struct evcnt tcp_reass_overlapfront;
382 extern struct evcnt tcp_reass_segdup;
383 extern struct evcnt tcp_reass_fragdup;
384 
385 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
386 
387 #else
388 
389 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
390 
391 #endif /* TCP_REASS_COUNTERS */
392 
393 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *,
394     int *);
395 static int tcp_dooptions(struct tcpcb *, const u_char *, int,
396     struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *);
397 
398 #ifdef INET
399 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
400 #endif
401 #ifdef INET6
402 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
403 #endif
404 
405 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
406 
407 #if defined(MBUFTRACE)
408 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass");
409 #endif /* defined(MBUFTRACE) */
410 
411 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl",
412     NULL, IPL_VM);
413 
414 struct ipqent *
415 tcpipqent_alloc(void)
416 {
417 	struct ipqent *ipqe;
418 	int s;
419 
420 	s = splvm();
421 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
422 	splx(s);
423 
424 	return ipqe;
425 }
426 
427 void
428 tcpipqent_free(struct ipqent *ipqe)
429 {
430 	int s;
431 
432 	s = splvm();
433 	pool_put(&tcpipqent_pool, ipqe);
434 	splx(s);
435 }
436 
437 static int
438 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen)
439 {
440 	struct ipqent *p, *q, *nq, *tiqe = NULL;
441 	struct socket *so = NULL;
442 	int pkt_flags;
443 	tcp_seq pkt_seq;
444 	unsigned pkt_len;
445 	u_long rcvpartdupbyte = 0;
446 	u_long rcvoobyte;
447 #ifdef TCP_REASS_COUNTERS
448 	u_int count = 0;
449 #endif
450 
451 	if (tp->t_inpcb)
452 		so = tp->t_inpcb->inp_socket;
453 #ifdef INET6
454 	else if (tp->t_in6pcb)
455 		so = tp->t_in6pcb->in6p_socket;
456 #endif
457 
458 	TCP_REASS_LOCK_CHECK(tp);
459 
460 	/*
461 	 * Call with th==0 after become established to
462 	 * force pre-ESTABLISHED data up to user socket.
463 	 */
464 	if (th == 0)
465 		goto present;
466 
467 	m_claimm(m, &tcp_reass_mowner);
468 
469 	rcvoobyte = *tlen;
470 	/*
471 	 * Copy these to local variables because the tcpiphdr
472 	 * gets munged while we are collapsing mbufs.
473 	 */
474 	pkt_seq = th->th_seq;
475 	pkt_len = *tlen;
476 	pkt_flags = th->th_flags;
477 
478 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
479 
480 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
481 		/*
482 		 * When we miss a packet, the vast majority of time we get
483 		 * packets that follow it in order.  So optimize for that.
484 		 */
485 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
486 			p->ipqe_len += pkt_len;
487 			p->ipqe_flags |= pkt_flags;
488 			m_cat(p->ipre_mlast, m);
489 			TRAVERSE(p->ipre_mlast);
490 			m = NULL;
491 			tiqe = p;
492 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
493 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
494 			goto skip_replacement;
495 		}
496 		/*
497 		 * While we're here, if the pkt is completely beyond
498 		 * anything we have, just insert it at the tail.
499 		 */
500 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
501 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
502 			goto insert_it;
503 		}
504 	}
505 
506 	q = TAILQ_FIRST(&tp->segq);
507 
508 	if (q != NULL) {
509 		/*
510 		 * If this segment immediately precedes the first out-of-order
511 		 * block, simply slap the segment in front of it and (mostly)
512 		 * skip the complicated logic.
513 		 */
514 		if (pkt_seq + pkt_len == q->ipqe_seq) {
515 			q->ipqe_seq = pkt_seq;
516 			q->ipqe_len += pkt_len;
517 			q->ipqe_flags |= pkt_flags;
518 			m_cat(m, q->ipqe_m);
519 			q->ipqe_m = m;
520 			q->ipre_mlast = m; /* last mbuf may have changed */
521 			TRAVERSE(q->ipre_mlast);
522 			tiqe = q;
523 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
524 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
525 			goto skip_replacement;
526 		}
527 	} else {
528 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
529 	}
530 
531 	/*
532 	 * Find a segment which begins after this one does.
533 	 */
534 	for (p = NULL; q != NULL; q = nq) {
535 		nq = TAILQ_NEXT(q, ipqe_q);
536 #ifdef TCP_REASS_COUNTERS
537 		count++;
538 #endif
539 		/*
540 		 * If the received segment is just right after this
541 		 * fragment, merge the two together and then check
542 		 * for further overlaps.
543 		 */
544 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
545 #ifdef TCPREASS_DEBUG
546 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
547 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
548 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
549 #endif
550 			pkt_len += q->ipqe_len;
551 			pkt_flags |= q->ipqe_flags;
552 			pkt_seq = q->ipqe_seq;
553 			m_cat(q->ipre_mlast, m);
554 			TRAVERSE(q->ipre_mlast);
555 			m = q->ipqe_m;
556 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
557 			goto free_ipqe;
558 		}
559 		/*
560 		 * If the received segment is completely past this
561 		 * fragment, we need to go the next fragment.
562 		 */
563 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
564 			p = q;
565 			continue;
566 		}
567 		/*
568 		 * If the fragment is past the received segment,
569 		 * it (or any following) can't be concatenated.
570 		 */
571 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
572 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
573 			break;
574 		}
575 
576 		/*
577 		 * We've received all the data in this segment before.
578 		 * mark it as a duplicate and return.
579 		 */
580 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
581 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
582 			tcpstat.tcps_rcvduppack++;
583 			tcpstat.tcps_rcvdupbyte += pkt_len;
584 			tcp_new_dsack(tp, pkt_seq, pkt_len);
585 			m_freem(m);
586 			if (tiqe != NULL) {
587 				tcpipqent_free(tiqe);
588 			}
589 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
590 			return (0);
591 		}
592 		/*
593 		 * Received segment completely overlaps this fragment
594 		 * so we drop the fragment (this keeps the temporal
595 		 * ordering of segments correct).
596 		 */
597 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
598 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
599 			rcvpartdupbyte += q->ipqe_len;
600 			m_freem(q->ipqe_m);
601 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
602 			goto free_ipqe;
603 		}
604 		/*
605 		 * RX'ed segment extends past the end of the
606 		 * fragment.  Drop the overlapping bytes.  Then
607 		 * merge the fragment and segment then treat as
608 		 * a longer received packet.
609 		 */
610 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
611 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
612 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
613 #ifdef TCPREASS_DEBUG
614 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
615 			       tp, overlap,
616 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
617 #endif
618 			m_adj(m, overlap);
619 			rcvpartdupbyte += overlap;
620 			m_cat(q->ipre_mlast, m);
621 			TRAVERSE(q->ipre_mlast);
622 			m = q->ipqe_m;
623 			pkt_seq = q->ipqe_seq;
624 			pkt_len += q->ipqe_len - overlap;
625 			rcvoobyte -= overlap;
626 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
627 			goto free_ipqe;
628 		}
629 		/*
630 		 * RX'ed segment extends past the front of the
631 		 * fragment.  Drop the overlapping bytes on the
632 		 * received packet.  The packet will then be
633 		 * contatentated with this fragment a bit later.
634 		 */
635 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
636 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
637 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
638 #ifdef TCPREASS_DEBUG
639 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
640 			       tp, overlap,
641 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
642 #endif
643 			m_adj(m, -overlap);
644 			pkt_len -= overlap;
645 			rcvpartdupbyte += overlap;
646 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
647 			rcvoobyte -= overlap;
648 		}
649 		/*
650 		 * If the received segment immediates precedes this
651 		 * fragment then tack the fragment onto this segment
652 		 * and reinsert the data.
653 		 */
654 		if (q->ipqe_seq == pkt_seq + pkt_len) {
655 #ifdef TCPREASS_DEBUG
656 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
657 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
658 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
659 #endif
660 			pkt_len += q->ipqe_len;
661 			pkt_flags |= q->ipqe_flags;
662 			m_cat(m, q->ipqe_m);
663 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
664 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
665 			tp->t_segqlen--;
666 			KASSERT(tp->t_segqlen >= 0);
667 			KASSERT(tp->t_segqlen != 0 ||
668 			    (TAILQ_EMPTY(&tp->segq) &&
669 			    TAILQ_EMPTY(&tp->timeq)));
670 			if (tiqe == NULL) {
671 				tiqe = q;
672 			} else {
673 				tcpipqent_free(q);
674 			}
675 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
676 			break;
677 		}
678 		/*
679 		 * If the fragment is before the segment, remember it.
680 		 * When this loop is terminated, p will contain the
681 		 * pointer to fragment that is right before the received
682 		 * segment.
683 		 */
684 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
685 			p = q;
686 
687 		continue;
688 
689 		/*
690 		 * This is a common operation.  It also will allow
691 		 * to save doing a malloc/free in most instances.
692 		 */
693 	  free_ipqe:
694 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
695 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
696 		tp->t_segqlen--;
697 		KASSERT(tp->t_segqlen >= 0);
698 		KASSERT(tp->t_segqlen != 0 ||
699 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
700 		if (tiqe == NULL) {
701 			tiqe = q;
702 		} else {
703 			tcpipqent_free(q);
704 		}
705 	}
706 
707 #ifdef TCP_REASS_COUNTERS
708 	if (count > 7)
709 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
710 	else if (count > 0)
711 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
712 #endif
713 
714     insert_it:
715 
716 	/*
717 	 * Allocate a new queue entry since the received segment did not
718 	 * collapse onto any other out-of-order block; thus we are allocating
719 	 * a new block.  If it had collapsed, tiqe would not be NULL and
720 	 * we would be reusing it.
721 	 * XXX If we can't, just drop the packet.  XXX
722 	 */
723 	if (tiqe == NULL) {
724 		tiqe = tcpipqent_alloc();
725 		if (tiqe == NULL) {
726 			tcpstat.tcps_rcvmemdrop++;
727 			m_freem(m);
728 			return (0);
729 		}
730 	}
731 
732 	/*
733 	 * Update the counters.
734 	 */
735 	tcpstat.tcps_rcvoopack++;
736 	tcpstat.tcps_rcvoobyte += rcvoobyte;
737 	if (rcvpartdupbyte) {
738 	    tcpstat.tcps_rcvpartduppack++;
739 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
740 	}
741 
742 	/*
743 	 * Insert the new fragment queue entry into both queues.
744 	 */
745 	tiqe->ipqe_m = m;
746 	tiqe->ipre_mlast = m;
747 	tiqe->ipqe_seq = pkt_seq;
748 	tiqe->ipqe_len = pkt_len;
749 	tiqe->ipqe_flags = pkt_flags;
750 	if (p == NULL) {
751 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
752 #ifdef TCPREASS_DEBUG
753 		if (tiqe->ipqe_seq != tp->rcv_nxt)
754 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
755 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
756 #endif
757 	} else {
758 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
759 #ifdef TCPREASS_DEBUG
760 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
761 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
762 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
763 #endif
764 	}
765 	tp->t_segqlen++;
766 
767 skip_replacement:
768 
769 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
770 
771 present:
772 	/*
773 	 * Present data to user, advancing rcv_nxt through
774 	 * completed sequence space.
775 	 */
776 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
777 		return (0);
778 	q = TAILQ_FIRST(&tp->segq);
779 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
780 		return (0);
781 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
782 		return (0);
783 
784 	tp->rcv_nxt += q->ipqe_len;
785 	pkt_flags = q->ipqe_flags & TH_FIN;
786 	nd6_hint(tp);
787 
788 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
789 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
790 	tp->t_segqlen--;
791 	KASSERT(tp->t_segqlen >= 0);
792 	KASSERT(tp->t_segqlen != 0 ||
793 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
794 	if (so->so_state & SS_CANTRCVMORE)
795 		m_freem(q->ipqe_m);
796 	else
797 		sbappendstream(&so->so_rcv, q->ipqe_m);
798 	tcpipqent_free(q);
799 	sorwakeup(so);
800 	return (pkt_flags);
801 }
802 
803 #ifdef INET6
804 int
805 tcp6_input(struct mbuf **mp, int *offp, int proto)
806 {
807 	struct mbuf *m = *mp;
808 
809 	/*
810 	 * draft-itojun-ipv6-tcp-to-anycast
811 	 * better place to put this in?
812 	 */
813 	if (m->m_flags & M_ANYCAST6) {
814 		struct ip6_hdr *ip6;
815 		if (m->m_len < sizeof(struct ip6_hdr)) {
816 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
817 				tcpstat.tcps_rcvshort++;
818 				return IPPROTO_DONE;
819 			}
820 		}
821 		ip6 = mtod(m, struct ip6_hdr *);
822 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
823 		    (char *)&ip6->ip6_dst - (char *)ip6);
824 		return IPPROTO_DONE;
825 	}
826 
827 	tcp_input(m, *offp, proto);
828 	return IPPROTO_DONE;
829 }
830 #endif
831 
832 #ifdef INET
833 static void
834 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
835 {
836 	char src[4*sizeof "123"];
837 	char dst[4*sizeof "123"];
838 
839 	if (ip) {
840 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
841 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
842 	}
843 	else {
844 		strlcpy(src, "(unknown)", sizeof(src));
845 		strlcpy(dst, "(unknown)", sizeof(dst));
846 	}
847 	log(LOG_INFO,
848 	    "Connection attempt to TCP %s:%d from %s:%d\n",
849 	    dst, ntohs(th->th_dport),
850 	    src, ntohs(th->th_sport));
851 }
852 #endif
853 
854 #ifdef INET6
855 static void
856 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
857 {
858 	char src[INET6_ADDRSTRLEN];
859 	char dst[INET6_ADDRSTRLEN];
860 
861 	if (ip6) {
862 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
863 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
864 	}
865 	else {
866 		strlcpy(src, "(unknown v6)", sizeof(src));
867 		strlcpy(dst, "(unknown v6)", sizeof(dst));
868 	}
869 	log(LOG_INFO,
870 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
871 	    dst, ntohs(th->th_dport),
872 	    src, ntohs(th->th_sport));
873 }
874 #endif
875 
876 /*
877  * Checksum extended TCP header and data.
878  */
879 int
880 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th,
881     int toff, int off, int tlen)
882 {
883 
884 	/*
885 	 * XXX it's better to record and check if this mbuf is
886 	 * already checked.
887 	 */
888 
889 	switch (af) {
890 #ifdef INET
891 	case AF_INET:
892 		switch (m->m_pkthdr.csum_flags &
893 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
894 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
895 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
896 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
897 			goto badcsum;
898 
899 		case M_CSUM_TCPv4|M_CSUM_DATA: {
900 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
901 
902 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
903 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
904 				const struct ip *ip =
905 				    mtod(m, const struct ip *);
906 
907 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
908 				    ip->ip_dst.s_addr,
909 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
910 			}
911 			if ((hw_csum ^ 0xffff) != 0)
912 				goto badcsum;
913 			break;
914 		}
915 
916 		case M_CSUM_TCPv4:
917 			/* Checksum was okay. */
918 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
919 			break;
920 
921 		default:
922 			/*
923 			 * Must compute it ourselves.  Maybe skip checksum
924 			 * on loopback interfaces.
925 			 */
926 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
927 					     IFF_LOOPBACK) ||
928 					   tcp_do_loopback_cksum)) {
929 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
930 				if (in4_cksum(m, IPPROTO_TCP, toff,
931 					      tlen + off) != 0)
932 					goto badcsum;
933 			}
934 			break;
935 		}
936 		break;
937 #endif /* INET4 */
938 
939 #ifdef INET6
940 	case AF_INET6:
941 		switch (m->m_pkthdr.csum_flags &
942 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
943 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
944 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
945 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
946 			goto badcsum;
947 
948 #if 0 /* notyet */
949 		case M_CSUM_TCPv6|M_CSUM_DATA:
950 #endif
951 
952 		case M_CSUM_TCPv6:
953 			/* Checksum was okay. */
954 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok);
955 			break;
956 
957 		default:
958 			/*
959 			 * Must compute it ourselves.  Maybe skip checksum
960 			 * on loopback interfaces.
961 			 */
962 			if (__predict_true((m->m_flags & M_LOOP) == 0 ||
963 			    tcp_do_loopback_cksum)) {
964 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
965 				if (in6_cksum(m, IPPROTO_TCP, toff,
966 				    tlen + off) != 0)
967 					goto badcsum;
968 			}
969 		}
970 		break;
971 #endif /* INET6 */
972 	}
973 
974 	return 0;
975 
976 badcsum:
977 	tcpstat.tcps_rcvbadsum++;
978 	return -1;
979 }
980 
981 /*
982  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
983  */
984 void
985 tcp_input(struct mbuf *m, ...)
986 {
987 	struct tcphdr *th;
988 	struct ip *ip;
989 	struct inpcb *inp;
990 #ifdef INET6
991 	struct ip6_hdr *ip6;
992 	struct in6pcb *in6p;
993 #endif
994 	u_int8_t *optp = NULL;
995 	int optlen = 0;
996 	int len, tlen, toff, hdroptlen = 0;
997 	struct tcpcb *tp = 0;
998 	int tiflags;
999 	struct socket *so = NULL;
1000 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
1001 #ifdef TCP_DEBUG
1002 	short ostate = 0;
1003 #endif
1004 	u_long tiwin;
1005 	struct tcp_opt_info opti;
1006 	int off, iphlen;
1007 	va_list ap;
1008 	int af;		/* af on the wire */
1009 	struct mbuf *tcp_saveti = NULL;
1010 	uint32_t ts_rtt;
1011 	uint8_t iptos;
1012 
1013 	MCLAIM(m, &tcp_rx_mowner);
1014 	va_start(ap, m);
1015 	toff = va_arg(ap, int);
1016 	(void)va_arg(ap, int);		/* ignore value, advance ap */
1017 	va_end(ap);
1018 
1019 	tcpstat.tcps_rcvtotal++;
1020 
1021 	bzero(&opti, sizeof(opti));
1022 	opti.ts_present = 0;
1023 	opti.maxseg = 0;
1024 
1025 	/*
1026 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
1027 	 *
1028 	 * TCP is, by definition, unicast, so we reject all
1029 	 * multicast outright.
1030 	 *
1031 	 * Note, there are additional src/dst address checks in
1032 	 * the AF-specific code below.
1033 	 */
1034 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1035 		/* XXX stat */
1036 		goto drop;
1037 	}
1038 #ifdef INET6
1039 	if (m->m_flags & M_ANYCAST6) {
1040 		/* XXX stat */
1041 		goto drop;
1042 	}
1043 #endif
1044 
1045 	/*
1046 	 * Get IP and TCP header.
1047 	 * Note: IP leaves IP header in first mbuf.
1048 	 */
1049 	ip = mtod(m, struct ip *);
1050 #ifdef INET6
1051 	ip6 = NULL;
1052 #endif
1053 	switch (ip->ip_v) {
1054 #ifdef INET
1055 	case 4:
1056 		af = AF_INET;
1057 		iphlen = sizeof(struct ip);
1058 		ip = mtod(m, struct ip *);
1059 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1060 			sizeof(struct tcphdr));
1061 		if (th == NULL) {
1062 			tcpstat.tcps_rcvshort++;
1063 			return;
1064 		}
1065 		/* We do the checksum after PCB lookup... */
1066 		len = ntohs(ip->ip_len);
1067 		tlen = len - toff;
1068 		iptos = ip->ip_tos;
1069 		break;
1070 #endif
1071 #ifdef INET6
1072 	case 6:
1073 		ip = NULL;
1074 		iphlen = sizeof(struct ip6_hdr);
1075 		af = AF_INET6;
1076 		ip6 = mtod(m, struct ip6_hdr *);
1077 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1078 			sizeof(struct tcphdr));
1079 		if (th == NULL) {
1080 			tcpstat.tcps_rcvshort++;
1081 			return;
1082 		}
1083 
1084 		/* Be proactive about malicious use of IPv4 mapped address */
1085 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1086 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1087 			/* XXX stat */
1088 			goto drop;
1089 		}
1090 
1091 		/*
1092 		 * Be proactive about unspecified IPv6 address in source.
1093 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1094 		 * unspecified IPv6 address can be used to confuse us.
1095 		 *
1096 		 * Note that packets with unspecified IPv6 destination is
1097 		 * already dropped in ip6_input.
1098 		 */
1099 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1100 			/* XXX stat */
1101 			goto drop;
1102 		}
1103 
1104 		/*
1105 		 * Make sure destination address is not multicast.
1106 		 * Source address checked in ip6_input().
1107 		 */
1108 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1109 			/* XXX stat */
1110 			goto drop;
1111 		}
1112 
1113 		/* We do the checksum after PCB lookup... */
1114 		len = m->m_pkthdr.len;
1115 		tlen = len - toff;
1116 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1117 		break;
1118 #endif
1119 	default:
1120 		m_freem(m);
1121 		return;
1122 	}
1123 
1124 	KASSERT(TCP_HDR_ALIGNED_P(th));
1125 
1126 	/*
1127 	 * Check that TCP offset makes sense,
1128 	 * pull out TCP options and adjust length.		XXX
1129 	 */
1130 	off = th->th_off << 2;
1131 	if (off < sizeof (struct tcphdr) || off > tlen) {
1132 		tcpstat.tcps_rcvbadoff++;
1133 		goto drop;
1134 	}
1135 	tlen -= off;
1136 
1137 	/*
1138 	 * tcp_input() has been modified to use tlen to mean the TCP data
1139 	 * length throughout the function.  Other functions can use
1140 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1141 	 * rja
1142 	 */
1143 
1144 	if (off > sizeof (struct tcphdr)) {
1145 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1146 		if (th == NULL) {
1147 			tcpstat.tcps_rcvshort++;
1148 			return;
1149 		}
1150 		/*
1151 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1152 		 * (as they're before toff) and we don't need to update those.
1153 		 */
1154 		KASSERT(TCP_HDR_ALIGNED_P(th));
1155 		optlen = off - sizeof (struct tcphdr);
1156 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1157 		/*
1158 		 * Do quick retrieval of timestamp options ("options
1159 		 * prediction?").  If timestamp is the only option and it's
1160 		 * formatted as recommended in RFC 1323 appendix A, we
1161 		 * quickly get the values now and not bother calling
1162 		 * tcp_dooptions(), etc.
1163 		 */
1164 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1165 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1166 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1167 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1168 		     (th->th_flags & TH_SYN) == 0) {
1169 			opti.ts_present = 1;
1170 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1171 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1172 			optp = NULL;	/* we've parsed the options */
1173 		}
1174 	}
1175 	tiflags = th->th_flags;
1176 
1177 	/*
1178 	 * Locate pcb for segment.
1179 	 */
1180 findpcb:
1181 	inp = NULL;
1182 #ifdef INET6
1183 	in6p = NULL;
1184 #endif
1185 	switch (af) {
1186 #ifdef INET
1187 	case AF_INET:
1188 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1189 		    ip->ip_dst, th->th_dport);
1190 		if (inp == 0) {
1191 			++tcpstat.tcps_pcbhashmiss;
1192 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1193 		}
1194 #ifdef INET6
1195 		if (inp == 0) {
1196 			struct in6_addr s, d;
1197 
1198 			/* mapped addr case */
1199 			bzero(&s, sizeof(s));
1200 			s.s6_addr16[5] = htons(0xffff);
1201 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1202 			bzero(&d, sizeof(d));
1203 			d.s6_addr16[5] = htons(0xffff);
1204 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1205 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1206 			    th->th_sport, &d, th->th_dport, 0);
1207 			if (in6p == 0) {
1208 				++tcpstat.tcps_pcbhashmiss;
1209 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1210 				    th->th_dport, 0);
1211 			}
1212 		}
1213 #endif
1214 #ifndef INET6
1215 		if (inp == 0)
1216 #else
1217 		if (inp == 0 && in6p == 0)
1218 #endif
1219 		{
1220 			++tcpstat.tcps_noport;
1221 			if (tcp_log_refused &&
1222 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1223 				tcp4_log_refused(ip, th);
1224 			}
1225 			tcp_fields_to_host(th);
1226 			goto dropwithreset_ratelim;
1227 		}
1228 #if defined(IPSEC) || defined(FAST_IPSEC)
1229 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1230 		    ipsec4_in_reject(m, inp)) {
1231 			ipsecstat.in_polvio++;
1232 			goto drop;
1233 		}
1234 #ifdef INET6
1235 		else if (in6p &&
1236 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1237 		    ipsec6_in_reject_so(m, in6p->in6p_socket)) {
1238 			ipsecstat.in_polvio++;
1239 			goto drop;
1240 		}
1241 #endif
1242 #endif /*IPSEC*/
1243 		break;
1244 #endif /*INET*/
1245 #ifdef INET6
1246 	case AF_INET6:
1247 	    {
1248 		int faith;
1249 
1250 #if defined(NFAITH) && NFAITH > 0
1251 		faith = faithprefix(&ip6->ip6_dst);
1252 #else
1253 		faith = 0;
1254 #endif
1255 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1256 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1257 		if (in6p == NULL) {
1258 			++tcpstat.tcps_pcbhashmiss;
1259 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1260 				th->th_dport, faith);
1261 		}
1262 		if (in6p == NULL) {
1263 			++tcpstat.tcps_noport;
1264 			if (tcp_log_refused &&
1265 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1266 				tcp6_log_refused(ip6, th);
1267 			}
1268 			tcp_fields_to_host(th);
1269 			goto dropwithreset_ratelim;
1270 		}
1271 #if defined(IPSEC) || defined(FAST_IPSEC)
1272 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1273 		    ipsec6_in_reject(m, in6p)) {
1274 			ipsec6stat.in_polvio++;
1275 			goto drop;
1276 		}
1277 #endif /*IPSEC*/
1278 		break;
1279 	    }
1280 #endif
1281 	}
1282 
1283 	/*
1284 	 * If the state is CLOSED (i.e., TCB does not exist) then
1285 	 * all data in the incoming segment is discarded.
1286 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1287 	 * but should either do a listen or a connect soon.
1288 	 */
1289 	tp = NULL;
1290 	so = NULL;
1291 	if (inp) {
1292 		tp = intotcpcb(inp);
1293 		so = inp->inp_socket;
1294 	}
1295 #ifdef INET6
1296 	else if (in6p) {
1297 		tp = in6totcpcb(in6p);
1298 		so = in6p->in6p_socket;
1299 	}
1300 #endif
1301 	if (tp == 0) {
1302 		tcp_fields_to_host(th);
1303 		goto dropwithreset_ratelim;
1304 	}
1305 	if (tp->t_state == TCPS_CLOSED)
1306 		goto drop;
1307 
1308 	/*
1309 	 * Checksum extended TCP header and data.
1310 	 */
1311 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1312 		goto badcsum;
1313 
1314 	tcp_fields_to_host(th);
1315 
1316 	/* Unscale the window into a 32-bit value. */
1317 	if ((tiflags & TH_SYN) == 0)
1318 		tiwin = th->th_win << tp->snd_scale;
1319 	else
1320 		tiwin = th->th_win;
1321 
1322 #ifdef INET6
1323 	/* save packet options if user wanted */
1324 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1325 		if (in6p->in6p_options) {
1326 			m_freem(in6p->in6p_options);
1327 			in6p->in6p_options = 0;
1328 		}
1329 		KASSERT(ip6 != NULL);
1330 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1331 	}
1332 #endif
1333 
1334 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1335 		union syn_cache_sa src;
1336 		union syn_cache_sa dst;
1337 
1338 		bzero(&src, sizeof(src));
1339 		bzero(&dst, sizeof(dst));
1340 		switch (af) {
1341 #ifdef INET
1342 		case AF_INET:
1343 			src.sin.sin_len = sizeof(struct sockaddr_in);
1344 			src.sin.sin_family = AF_INET;
1345 			src.sin.sin_addr = ip->ip_src;
1346 			src.sin.sin_port = th->th_sport;
1347 
1348 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1349 			dst.sin.sin_family = AF_INET;
1350 			dst.sin.sin_addr = ip->ip_dst;
1351 			dst.sin.sin_port = th->th_dport;
1352 			break;
1353 #endif
1354 #ifdef INET6
1355 		case AF_INET6:
1356 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1357 			src.sin6.sin6_family = AF_INET6;
1358 			src.sin6.sin6_addr = ip6->ip6_src;
1359 			src.sin6.sin6_port = th->th_sport;
1360 
1361 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1362 			dst.sin6.sin6_family = AF_INET6;
1363 			dst.sin6.sin6_addr = ip6->ip6_dst;
1364 			dst.sin6.sin6_port = th->th_dport;
1365 			break;
1366 #endif /* INET6 */
1367 		default:
1368 			goto badsyn;	/*sanity*/
1369 		}
1370 
1371 		if (so->so_options & SO_DEBUG) {
1372 #ifdef TCP_DEBUG
1373 			ostate = tp->t_state;
1374 #endif
1375 
1376 			tcp_saveti = NULL;
1377 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1378 				goto nosave;
1379 
1380 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1381 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1382 				if (!tcp_saveti)
1383 					goto nosave;
1384 			} else {
1385 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1386 				if (!tcp_saveti)
1387 					goto nosave;
1388 				MCLAIM(m, &tcp_mowner);
1389 				tcp_saveti->m_len = iphlen;
1390 				m_copydata(m, 0, iphlen,
1391 				    mtod(tcp_saveti, void *));
1392 			}
1393 
1394 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1395 				m_freem(tcp_saveti);
1396 				tcp_saveti = NULL;
1397 			} else {
1398 				tcp_saveti->m_len += sizeof(struct tcphdr);
1399 				memcpy(mtod(tcp_saveti, char *) + iphlen, th,
1400 				    sizeof(struct tcphdr));
1401 			}
1402 	nosave:;
1403 		}
1404 		if (so->so_options & SO_ACCEPTCONN) {
1405 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1406 				if (tiflags & TH_RST) {
1407 					syn_cache_reset(&src.sa, &dst.sa, th);
1408 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1409 				    (TH_ACK|TH_SYN)) {
1410 					/*
1411 					 * Received a SYN,ACK.  This should
1412 					 * never happen while we are in
1413 					 * LISTEN.  Send an RST.
1414 					 */
1415 					goto badsyn;
1416 				} else if (tiflags & TH_ACK) {
1417 					so = syn_cache_get(&src.sa, &dst.sa,
1418 						th, toff, tlen, so, m);
1419 					if (so == NULL) {
1420 						/*
1421 						 * We don't have a SYN for
1422 						 * this ACK; send an RST.
1423 						 */
1424 						goto badsyn;
1425 					} else if (so ==
1426 					    (struct socket *)(-1)) {
1427 						/*
1428 						 * We were unable to create
1429 						 * the connection.  If the
1430 						 * 3-way handshake was
1431 						 * completed, and RST has
1432 						 * been sent to the peer.
1433 						 * Since the mbuf might be
1434 						 * in use for the reply,
1435 						 * do not free it.
1436 						 */
1437 						m = NULL;
1438 					} else {
1439 						/*
1440 						 * We have created a
1441 						 * full-blown connection.
1442 						 */
1443 						tp = NULL;
1444 						inp = NULL;
1445 #ifdef INET6
1446 						in6p = NULL;
1447 #endif
1448 						switch (so->so_proto->pr_domain->dom_family) {
1449 #ifdef INET
1450 						case AF_INET:
1451 							inp = sotoinpcb(so);
1452 							tp = intotcpcb(inp);
1453 							break;
1454 #endif
1455 #ifdef INET6
1456 						case AF_INET6:
1457 							in6p = sotoin6pcb(so);
1458 							tp = in6totcpcb(in6p);
1459 							break;
1460 #endif
1461 						}
1462 						if (tp == NULL)
1463 							goto badsyn;	/*XXX*/
1464 						tiwin <<= tp->snd_scale;
1465 						goto after_listen;
1466 					}
1467 				} else {
1468 					/*
1469 					 * None of RST, SYN or ACK was set.
1470 					 * This is an invalid packet for a
1471 					 * TCB in LISTEN state.  Send a RST.
1472 					 */
1473 					goto badsyn;
1474 				}
1475 			} else {
1476 				/*
1477 				 * Received a SYN.
1478 				 *
1479 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1480 				 */
1481 				if (m->m_flags & (M_BCAST|M_MCAST))
1482 					goto drop;
1483 
1484 				switch (af) {
1485 #ifdef INET6
1486 				case AF_INET6:
1487 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1488 						goto drop;
1489 					break;
1490 #endif /* INET6 */
1491 				case AF_INET:
1492 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1493 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1494 						goto drop;
1495 				break;
1496 				}
1497 
1498 #ifdef INET6
1499 				/*
1500 				 * If deprecated address is forbidden, we do
1501 				 * not accept SYN to deprecated interface
1502 				 * address to prevent any new inbound
1503 				 * connection from getting established.
1504 				 * When we do not accept SYN, we send a TCP
1505 				 * RST, with deprecated source address (instead
1506 				 * of dropping it).  We compromise it as it is
1507 				 * much better for peer to send a RST, and
1508 				 * RST will be the final packet for the
1509 				 * exchange.
1510 				 *
1511 				 * If we do not forbid deprecated addresses, we
1512 				 * accept the SYN packet.  RFC2462 does not
1513 				 * suggest dropping SYN in this case.
1514 				 * If we decipher RFC2462 5.5.4, it says like
1515 				 * this:
1516 				 * 1. use of deprecated addr with existing
1517 				 *    communication is okay - "SHOULD continue
1518 				 *    to be used"
1519 				 * 2. use of it with new communication:
1520 				 *   (2a) "SHOULD NOT be used if alternate
1521 				 *        address with sufficient scope is
1522 				 *        available"
1523 				 *   (2b) nothing mentioned otherwise.
1524 				 * Here we fall into (2b) case as we have no
1525 				 * choice in our source address selection - we
1526 				 * must obey the peer.
1527 				 *
1528 				 * The wording in RFC2462 is confusing, and
1529 				 * there are multiple description text for
1530 				 * deprecated address handling - worse, they
1531 				 * are not exactly the same.  I believe 5.5.4
1532 				 * is the best one, so we follow 5.5.4.
1533 				 */
1534 				if (af == AF_INET6 && !ip6_use_deprecated) {
1535 					struct in6_ifaddr *ia6;
1536 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1537 					    &ip6->ip6_dst)) &&
1538 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1539 						tp = NULL;
1540 						goto dropwithreset;
1541 					}
1542 				}
1543 #endif
1544 
1545 #if defined(IPSEC) || defined(FAST_IPSEC)
1546 				switch (af) {
1547 #ifdef INET
1548 				case AF_INET:
1549 					if (ipsec4_in_reject_so(m, so)) {
1550 						ipsecstat.in_polvio++;
1551 						tp = NULL;
1552 						goto dropwithreset;
1553 					}
1554 					break;
1555 #endif
1556 #ifdef INET6
1557 				case AF_INET6:
1558 					if (ipsec6_in_reject_so(m, so)) {
1559 						ipsec6stat.in_polvio++;
1560 						tp = NULL;
1561 						goto dropwithreset;
1562 					}
1563 					break;
1564 #endif /*INET6*/
1565 				}
1566 #endif /*IPSEC*/
1567 
1568 				/*
1569 				 * LISTEN socket received a SYN
1570 				 * from itself?  This can't possibly
1571 				 * be valid; drop the packet.
1572 				 */
1573 				if (th->th_sport == th->th_dport) {
1574 					int i;
1575 
1576 					switch (af) {
1577 #ifdef INET
1578 					case AF_INET:
1579 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1580 						break;
1581 #endif
1582 #ifdef INET6
1583 					case AF_INET6:
1584 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1585 						break;
1586 #endif
1587 					default:
1588 						i = 1;
1589 					}
1590 					if (i) {
1591 						tcpstat.tcps_badsyn++;
1592 						goto drop;
1593 					}
1594 				}
1595 
1596 				/*
1597 				 * SYN looks ok; create compressed TCP
1598 				 * state for it.
1599 				 */
1600 				if (so->so_qlen <= so->so_qlimit &&
1601 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1602 						so, m, optp, optlen, &opti))
1603 					m = NULL;
1604 			}
1605 			goto drop;
1606 		}
1607 	}
1608 
1609 after_listen:
1610 #ifdef DIAGNOSTIC
1611 	/*
1612 	 * Should not happen now that all embryonic connections
1613 	 * are handled with compressed state.
1614 	 */
1615 	if (tp->t_state == TCPS_LISTEN)
1616 		panic("tcp_input: TCPS_LISTEN");
1617 #endif
1618 
1619 	/*
1620 	 * Segment received on connection.
1621 	 * Reset idle time and keep-alive timer.
1622 	 */
1623 	tp->t_rcvtime = tcp_now;
1624 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1625 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1626 
1627 	/*
1628 	 * Process options.
1629 	 */
1630 #ifdef TCP_SIGNATURE
1631 	if (optp || (tp->t_flags & TF_SIGNATURE))
1632 #else
1633 	if (optp)
1634 #endif
1635 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1636 			goto drop;
1637 
1638 	if (TCP_SACK_ENABLED(tp)) {
1639 		tcp_del_sackholes(tp, th);
1640 	}
1641 
1642 	if (TCP_ECN_ALLOWED(tp)) {
1643 		switch (iptos & IPTOS_ECN_MASK) {
1644 		case IPTOS_ECN_CE:
1645 			tp->t_flags |= TF_ECN_SND_ECE;
1646 			tcpstat.tcps_ecn_ce++;
1647 			break;
1648 		case IPTOS_ECN_ECT0:
1649 			tcpstat.tcps_ecn_ect++;
1650 			break;
1651 		case IPTOS_ECN_ECT1:
1652 			/* XXX: ignore for now -- rpaulo */
1653 			break;
1654 		}
1655 
1656 		if (tiflags & TH_CWR)
1657 			tp->t_flags &= ~TF_ECN_SND_ECE;
1658 
1659 		/*
1660 		 * Congestion experienced.
1661 		 * Ignore if we are already trying to recover.
1662 		 */
1663 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1664 			tp->t_congctl->cong_exp(tp);
1665 	}
1666 
1667 	if (opti.ts_present && opti.ts_ecr) {
1668 		/*
1669 		 * Calculate the RTT from the returned time stamp and the
1670 		 * connection's time base.  If the time stamp is later than
1671 		 * the current time, or is extremely old, fall back to non-1323
1672 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1673 		 * at the same time.
1674 		 */
1675 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1676 		if (ts_rtt > TCP_PAWS_IDLE)
1677 			ts_rtt = 0;
1678 	} else {
1679 		ts_rtt = 0;
1680 	}
1681 
1682 	/*
1683 	 * Header prediction: check for the two common cases
1684 	 * of a uni-directional data xfer.  If the packet has
1685 	 * no control flags, is in-sequence, the window didn't
1686 	 * change and we're not retransmitting, it's a
1687 	 * candidate.  If the length is zero and the ack moved
1688 	 * forward, we're the sender side of the xfer.  Just
1689 	 * free the data acked & wake any higher level process
1690 	 * that was blocked waiting for space.  If the length
1691 	 * is non-zero and the ack didn't move, we're the
1692 	 * receiver side.  If we're getting packets in-order
1693 	 * (the reassembly queue is empty), add the data to
1694 	 * the socket buffer and note that we need a delayed ack.
1695 	 */
1696 	if (tp->t_state == TCPS_ESTABLISHED &&
1697 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1698 	        == TH_ACK &&
1699 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1700 	    th->th_seq == tp->rcv_nxt &&
1701 	    tiwin && tiwin == tp->snd_wnd &&
1702 	    tp->snd_nxt == tp->snd_max) {
1703 
1704 		/*
1705 		 * If last ACK falls within this segment's sequence numbers,
1706 		 * record the timestamp.
1707 		 * NOTE that the test is modified according to the latest
1708 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1709 		 *
1710 		 * note that we already know
1711 		 *	TSTMP_GEQ(opti.ts_val, tp->ts_recent)
1712 		 */
1713 		if (opti.ts_present &&
1714 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1715 			tp->ts_recent_age = tcp_now;
1716 			tp->ts_recent = opti.ts_val;
1717 		}
1718 
1719 		if (tlen == 0) {
1720 			/* Ack prediction. */
1721 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1722 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1723 			    tp->snd_cwnd >= tp->snd_wnd &&
1724 			    tp->t_partialacks < 0) {
1725 				/*
1726 				 * this is a pure ack for outstanding data.
1727 				 */
1728 				++tcpstat.tcps_predack;
1729 				if (ts_rtt)
1730 					tcp_xmit_timer(tp, ts_rtt);
1731 				else if (tp->t_rtttime &&
1732 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1733 					tcp_xmit_timer(tp,
1734 					  tcp_now - tp->t_rtttime);
1735 				acked = th->th_ack - tp->snd_una;
1736 				tcpstat.tcps_rcvackpack++;
1737 				tcpstat.tcps_rcvackbyte += acked;
1738 				nd6_hint(tp);
1739 
1740 				if (acked > (tp->t_lastoff - tp->t_inoff))
1741 					tp->t_lastm = NULL;
1742 				sbdrop(&so->so_snd, acked);
1743 				tp->t_lastoff -= acked;
1744 
1745 				icmp_check(tp, th, acked);
1746 
1747 				tp->snd_una = th->th_ack;
1748 				tp->snd_fack = tp->snd_una;
1749 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1750 					tp->snd_high = tp->snd_una;
1751 				m_freem(m);
1752 
1753 				/*
1754 				 * If all outstanding data are acked, stop
1755 				 * retransmit timer, otherwise restart timer
1756 				 * using current (possibly backed-off) value.
1757 				 * If process is waiting for space,
1758 				 * wakeup/selnotify/signal.  If data
1759 				 * are ready to send, let tcp_output
1760 				 * decide between more output or persist.
1761 				 */
1762 				if (tp->snd_una == tp->snd_max)
1763 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1764 				else if (TCP_TIMER_ISARMED(tp,
1765 				    TCPT_PERSIST) == 0)
1766 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1767 					    tp->t_rxtcur);
1768 
1769 				sowwakeup(so);
1770 				if (so->so_snd.sb_cc)
1771 					(void) tcp_output(tp);
1772 				if (tcp_saveti)
1773 					m_freem(tcp_saveti);
1774 				return;
1775 			}
1776 		} else if (th->th_ack == tp->snd_una &&
1777 		    TAILQ_FIRST(&tp->segq) == NULL &&
1778 		    tlen <= sbspace(&so->so_rcv)) {
1779 			int newsize = 0;	/* automatic sockbuf scaling */
1780 
1781 			/*
1782 			 * this is a pure, in-sequence data packet
1783 			 * with nothing on the reassembly queue and
1784 			 * we have enough buffer space to take it.
1785 			 */
1786 			++tcpstat.tcps_preddat;
1787 			tp->rcv_nxt += tlen;
1788 			tcpstat.tcps_rcvpack++;
1789 			tcpstat.tcps_rcvbyte += tlen;
1790 			nd6_hint(tp);
1791 
1792 		/*
1793 		 * Automatic sizing enables the performance of large buffers
1794 		 * and most of the efficiency of small ones by only allocating
1795 		 * space when it is needed.
1796 		 *
1797 		 * On the receive side the socket buffer memory is only rarely
1798 		 * used to any significant extent.  This allows us to be much
1799 		 * more aggressive in scaling the receive socket buffer.  For
1800 		 * the case that the buffer space is actually used to a large
1801 		 * extent and we run out of kernel memory we can simply drop
1802 		 * the new segments; TCP on the sender will just retransmit it
1803 		 * later.  Setting the buffer size too big may only consume too
1804 		 * much kernel memory if the application doesn't read() from
1805 		 * the socket or packet loss or reordering makes use of the
1806 		 * reassembly queue.
1807 		 *
1808 		 * The criteria to step up the receive buffer one notch are:
1809 		 *  1. the number of bytes received during the time it takes
1810 		 *     one timestamp to be reflected back to us (the RTT);
1811 		 *  2. received bytes per RTT is within seven eighth of the
1812 		 *     current socket buffer size;
1813 		 *  3. receive buffer size has not hit maximal automatic size;
1814 		 *
1815 		 * This algorithm does one step per RTT at most and only if
1816 		 * we receive a bulk stream w/o packet losses or reorderings.
1817 		 * Shrinking the buffer during idle times is not necessary as
1818 		 * it doesn't consume any memory when idle.
1819 		 *
1820 		 * TODO: Only step up if the application is actually serving
1821 		 * the buffer to better manage the socket buffer resources.
1822 		 */
1823 			if (tcp_do_autorcvbuf &&
1824 			    opti.ts_ecr &&
1825 			    (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1826 				if (opti.ts_ecr > tp->rfbuf_ts &&
1827 				    opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) {
1828 					if (tp->rfbuf_cnt >
1829 					    (so->so_rcv.sb_hiwat / 8 * 7) &&
1830 					    so->so_rcv.sb_hiwat <
1831 					    tcp_autorcvbuf_max) {
1832 						newsize =
1833 						    min(so->so_rcv.sb_hiwat +
1834 						    tcp_autorcvbuf_inc,
1835 						    tcp_autorcvbuf_max);
1836 					}
1837 					/* Start over with next RTT. */
1838 					tp->rfbuf_ts = 0;
1839 					tp->rfbuf_cnt = 0;
1840 				} else
1841 					tp->rfbuf_cnt += tlen;	/* add up */
1842 			}
1843 
1844 			/*
1845 			 * Drop TCP, IP headers and TCP options then add data
1846 			 * to socket buffer.
1847 			 */
1848 			if (so->so_state & SS_CANTRCVMORE)
1849 				m_freem(m);
1850 			else {
1851 				/*
1852 				 * Set new socket buffer size.
1853 				 * Give up when limit is reached.
1854 				 */
1855 				if (newsize)
1856 					if (!sbreserve(&so->so_rcv,
1857 					    newsize, so))
1858 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1859 				m_adj(m, toff + off);
1860 				sbappendstream(&so->so_rcv, m);
1861 			}
1862 			sorwakeup(so);
1863 			tcp_setup_ack(tp, th);
1864 			if (tp->t_flags & TF_ACKNOW)
1865 				(void) tcp_output(tp);
1866 			if (tcp_saveti)
1867 				m_freem(tcp_saveti);
1868 			return;
1869 		}
1870 	}
1871 
1872 	/*
1873 	 * Compute mbuf offset to TCP data segment.
1874 	 */
1875 	hdroptlen = toff + off;
1876 
1877 	/*
1878 	 * Calculate amount of space in receive window,
1879 	 * and then do TCP input processing.
1880 	 * Receive window is amount of space in rcv queue,
1881 	 * but not less than advertised window.
1882 	 */
1883 	{ int win;
1884 
1885 	win = sbspace(&so->so_rcv);
1886 	if (win < 0)
1887 		win = 0;
1888 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1889 	}
1890 
1891 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1892 	tp->rfbuf_ts = 0;
1893 	tp->rfbuf_cnt = 0;
1894 
1895 	switch (tp->t_state) {
1896 	/*
1897 	 * If the state is SYN_SENT:
1898 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1899 	 *	if seg contains a RST, then drop the connection.
1900 	 *	if seg does not contain SYN, then drop it.
1901 	 * Otherwise this is an acceptable SYN segment
1902 	 *	initialize tp->rcv_nxt and tp->irs
1903 	 *	if seg contains ack then advance tp->snd_una
1904 	 *	if seg contains a ECE and ECN support is enabled, the stream
1905 	 *	    is ECN capable.
1906 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1907 	 *	arrange for segment to be acked (eventually)
1908 	 *	continue processing rest of data/controls, beginning with URG
1909 	 */
1910 	case TCPS_SYN_SENT:
1911 		if ((tiflags & TH_ACK) &&
1912 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1913 		     SEQ_GT(th->th_ack, tp->snd_max)))
1914 			goto dropwithreset;
1915 		if (tiflags & TH_RST) {
1916 			if (tiflags & TH_ACK)
1917 				tp = tcp_drop(tp, ECONNREFUSED);
1918 			goto drop;
1919 		}
1920 		if ((tiflags & TH_SYN) == 0)
1921 			goto drop;
1922 		if (tiflags & TH_ACK) {
1923 			tp->snd_una = th->th_ack;
1924 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1925 				tp->snd_nxt = tp->snd_una;
1926 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1927 				tp->snd_high = tp->snd_una;
1928 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1929 
1930 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1931 				tp->t_flags |= TF_ECN_PERMIT;
1932 				tcpstat.tcps_ecn_shs++;
1933 			}
1934 
1935 		}
1936 		tp->irs = th->th_seq;
1937 		tcp_rcvseqinit(tp);
1938 		tp->t_flags |= TF_ACKNOW;
1939 		tcp_mss_from_peer(tp, opti.maxseg);
1940 
1941 		/*
1942 		 * Initialize the initial congestion window.  If we
1943 		 * had to retransmit the SYN, we must initialize cwnd
1944 		 * to 1 segment (i.e. the Loss Window).
1945 		 */
1946 		if (tp->t_flags & TF_SYN_REXMT)
1947 			tp->snd_cwnd = tp->t_peermss;
1948 		else {
1949 			int ss = tcp_init_win;
1950 #ifdef INET
1951 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1952 				ss = tcp_init_win_local;
1953 #endif
1954 #ifdef INET6
1955 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1956 				ss = tcp_init_win_local;
1957 #endif
1958 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1959 		}
1960 
1961 		tcp_rmx_rtt(tp);
1962 		if (tiflags & TH_ACK) {
1963 			tcpstat.tcps_connects++;
1964 			soisconnected(so);
1965 			tcp_established(tp);
1966 			/* Do window scaling on this connection? */
1967 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1968 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1969 				tp->snd_scale = tp->requested_s_scale;
1970 				tp->rcv_scale = tp->request_r_scale;
1971 			}
1972 			TCP_REASS_LOCK(tp);
1973 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1974 			TCP_REASS_UNLOCK(tp);
1975 			/*
1976 			 * if we didn't have to retransmit the SYN,
1977 			 * use its rtt as our initial srtt & rtt var.
1978 			 */
1979 			if (tp->t_rtttime)
1980 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1981 		} else
1982 			tp->t_state = TCPS_SYN_RECEIVED;
1983 
1984 		/*
1985 		 * Advance th->th_seq to correspond to first data byte.
1986 		 * If data, trim to stay within window,
1987 		 * dropping FIN if necessary.
1988 		 */
1989 		th->th_seq++;
1990 		if (tlen > tp->rcv_wnd) {
1991 			todrop = tlen - tp->rcv_wnd;
1992 			m_adj(m, -todrop);
1993 			tlen = tp->rcv_wnd;
1994 			tiflags &= ~TH_FIN;
1995 			tcpstat.tcps_rcvpackafterwin++;
1996 			tcpstat.tcps_rcvbyteafterwin += todrop;
1997 		}
1998 		tp->snd_wl1 = th->th_seq - 1;
1999 		tp->rcv_up = th->th_seq;
2000 		goto step6;
2001 
2002 	/*
2003 	 * If the state is SYN_RECEIVED:
2004 	 *	If seg contains an ACK, but not for our SYN, drop the input
2005 	 *	and generate an RST.  See page 36, rfc793
2006 	 */
2007 	case TCPS_SYN_RECEIVED:
2008 		if ((tiflags & TH_ACK) &&
2009 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
2010 		     SEQ_GT(th->th_ack, tp->snd_max)))
2011 			goto dropwithreset;
2012 		break;
2013 	}
2014 
2015 	/*
2016 	 * States other than LISTEN or SYN_SENT.
2017 	 * First check timestamp, if present.
2018 	 * Then check that at least some bytes of segment are within
2019 	 * receive window.  If segment begins before rcv_nxt,
2020 	 * drop leading data (and SYN); if nothing left, just ack.
2021 	 *
2022 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2023 	 * and it's less than ts_recent, drop it.
2024 	 */
2025 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
2026 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
2027 
2028 		/* Check to see if ts_recent is over 24 days old.  */
2029 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
2030 			/*
2031 			 * Invalidate ts_recent.  If this segment updates
2032 			 * ts_recent, the age will be reset later and ts_recent
2033 			 * will get a valid value.  If it does not, setting
2034 			 * ts_recent to zero will at least satisfy the
2035 			 * requirement that zero be placed in the timestamp
2036 			 * echo reply when ts_recent isn't valid.  The
2037 			 * age isn't reset until we get a valid ts_recent
2038 			 * because we don't want out-of-order segments to be
2039 			 * dropped when ts_recent is old.
2040 			 */
2041 			tp->ts_recent = 0;
2042 		} else {
2043 			tcpstat.tcps_rcvduppack++;
2044 			tcpstat.tcps_rcvdupbyte += tlen;
2045 			tcpstat.tcps_pawsdrop++;
2046 			tcp_new_dsack(tp, th->th_seq, tlen);
2047 			goto dropafterack;
2048 		}
2049 	}
2050 
2051 	todrop = tp->rcv_nxt - th->th_seq;
2052 	dupseg = false;
2053 	if (todrop > 0) {
2054 		if (tiflags & TH_SYN) {
2055 			tiflags &= ~TH_SYN;
2056 			th->th_seq++;
2057 			if (th->th_urp > 1)
2058 				th->th_urp--;
2059 			else {
2060 				tiflags &= ~TH_URG;
2061 				th->th_urp = 0;
2062 			}
2063 			todrop--;
2064 		}
2065 		if (todrop > tlen ||
2066 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
2067 			/*
2068 			 * Any valid FIN or RST must be to the left of the
2069 			 * window.  At this point the FIN or RST must be a
2070 			 * duplicate or out of sequence; drop it.
2071 			 */
2072 			if (tiflags & TH_RST)
2073 				goto drop;
2074 			tiflags &= ~(TH_FIN|TH_RST);
2075 			/*
2076 			 * Send an ACK to resynchronize and drop any data.
2077 			 * But keep on processing for RST or ACK.
2078 			 */
2079 			tp->t_flags |= TF_ACKNOW;
2080 			todrop = tlen;
2081 			dupseg = true;
2082 			tcpstat.tcps_rcvdupbyte += todrop;
2083 			tcpstat.tcps_rcvduppack++;
2084 		} else if ((tiflags & TH_RST) &&
2085 			   th->th_seq != tp->last_ack_sent) {
2086 			/*
2087 			 * Test for reset before adjusting the sequence
2088 			 * number for overlapping data.
2089 			 */
2090 			goto dropafterack_ratelim;
2091 		} else {
2092 			tcpstat.tcps_rcvpartduppack++;
2093 			tcpstat.tcps_rcvpartdupbyte += todrop;
2094 		}
2095 		tcp_new_dsack(tp, th->th_seq, todrop);
2096 		hdroptlen += todrop;	/*drop from head afterwards*/
2097 		th->th_seq += todrop;
2098 		tlen -= todrop;
2099 		if (th->th_urp > todrop)
2100 			th->th_urp -= todrop;
2101 		else {
2102 			tiflags &= ~TH_URG;
2103 			th->th_urp = 0;
2104 		}
2105 	}
2106 
2107 	/*
2108 	 * If new data are received on a connection after the
2109 	 * user processes are gone, then RST the other end.
2110 	 */
2111 	if ((so->so_state & SS_NOFDREF) &&
2112 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2113 		tp = tcp_close(tp);
2114 		tcpstat.tcps_rcvafterclose++;
2115 		goto dropwithreset;
2116 	}
2117 
2118 	/*
2119 	 * If segment ends after window, drop trailing data
2120 	 * (and PUSH and FIN); if nothing left, just ACK.
2121 	 */
2122 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2123 	if (todrop > 0) {
2124 		tcpstat.tcps_rcvpackafterwin++;
2125 		if (todrop >= tlen) {
2126 			/*
2127 			 * The segment actually starts after the window.
2128 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2129 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2130 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2131 			 */
2132 			tcpstat.tcps_rcvbyteafterwin += tlen;
2133 			/*
2134 			 * If a new connection request is received
2135 			 * while in TIME_WAIT, drop the old connection
2136 			 * and start over if the sequence numbers
2137 			 * are above the previous ones.
2138 			 *
2139 			 * NOTE: We will checksum the packet again, and
2140 			 * so we need to put the header fields back into
2141 			 * network order!
2142 			 * XXX This kind of sucks, but we don't expect
2143 			 * XXX this to happen very often, so maybe it
2144 			 * XXX doesn't matter so much.
2145 			 */
2146 			if (tiflags & TH_SYN &&
2147 			    tp->t_state == TCPS_TIME_WAIT &&
2148 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2149 				tp = tcp_close(tp);
2150 				tcp_fields_to_net(th);
2151 				goto findpcb;
2152 			}
2153 			/*
2154 			 * If window is closed can only take segments at
2155 			 * window edge, and have to drop data and PUSH from
2156 			 * incoming segments.  Continue processing, but
2157 			 * remember to ack.  Otherwise, drop segment
2158 			 * and (if not RST) ack.
2159 			 */
2160 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2161 				tp->t_flags |= TF_ACKNOW;
2162 				tcpstat.tcps_rcvwinprobe++;
2163 			} else
2164 				goto dropafterack;
2165 		} else
2166 			tcpstat.tcps_rcvbyteafterwin += todrop;
2167 		m_adj(m, -todrop);
2168 		tlen -= todrop;
2169 		tiflags &= ~(TH_PUSH|TH_FIN);
2170 	}
2171 
2172 	/*
2173 	 * If last ACK falls within this segment's sequence numbers,
2174 	 *  record the timestamp.
2175 	 * NOTE:
2176 	 * 1) That the test incorporates suggestions from the latest
2177 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2178 	 * 2) That updating only on newer timestamps interferes with
2179 	 *    our earlier PAWS tests, so this check should be solely
2180 	 *    predicated on the sequence space of this segment.
2181 	 * 3) That we modify the segment boundary check to be
2182 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2183 	 *    instead of RFC1323's
2184 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2185 	 *    This modified check allows us to overcome RFC1323's
2186 	 *    limitations as described in Stevens TCP/IP Illustrated
2187 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2188 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2189 	 */
2190 	if (opti.ts_present &&
2191 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2192 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2193 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2194 		tp->ts_recent_age = tcp_now;
2195 		tp->ts_recent = opti.ts_val;
2196 	}
2197 
2198 	/*
2199 	 * If the RST bit is set examine the state:
2200 	 *    SYN_RECEIVED STATE:
2201 	 *	If passive open, return to LISTEN state.
2202 	 *	If active open, inform user that connection was refused.
2203 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2204 	 *	Inform user that connection was reset, and close tcb.
2205 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2206 	 *	Close the tcb.
2207 	 */
2208 	if (tiflags & TH_RST) {
2209 		if (th->th_seq != tp->last_ack_sent)
2210 			goto dropafterack_ratelim;
2211 
2212 		switch (tp->t_state) {
2213 		case TCPS_SYN_RECEIVED:
2214 			so->so_error = ECONNREFUSED;
2215 			goto close;
2216 
2217 		case TCPS_ESTABLISHED:
2218 		case TCPS_FIN_WAIT_1:
2219 		case TCPS_FIN_WAIT_2:
2220 		case TCPS_CLOSE_WAIT:
2221 			so->so_error = ECONNRESET;
2222 		close:
2223 			tp->t_state = TCPS_CLOSED;
2224 			tcpstat.tcps_drops++;
2225 			tp = tcp_close(tp);
2226 			goto drop;
2227 
2228 		case TCPS_CLOSING:
2229 		case TCPS_LAST_ACK:
2230 		case TCPS_TIME_WAIT:
2231 			tp = tcp_close(tp);
2232 			goto drop;
2233 		}
2234 	}
2235 
2236 	/*
2237 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2238 	 * we must be in a synchronized state.  RFC791 states (under RST
2239 	 * generation) that any unacceptable segment (an out-of-order SYN
2240 	 * qualifies) received in a synchronized state must elicit only an
2241 	 * empty acknowledgment segment ... and the connection remains in
2242 	 * the same state.
2243 	 */
2244 	if (tiflags & TH_SYN) {
2245 		if (tp->rcv_nxt == th->th_seq) {
2246 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2247 			    TH_ACK);
2248 			if (tcp_saveti)
2249 				m_freem(tcp_saveti);
2250 			return;
2251 		}
2252 
2253 		goto dropafterack_ratelim;
2254 	}
2255 
2256 	/*
2257 	 * If the ACK bit is off we drop the segment and return.
2258 	 */
2259 	if ((tiflags & TH_ACK) == 0) {
2260 		if (tp->t_flags & TF_ACKNOW)
2261 			goto dropafterack;
2262 		else
2263 			goto drop;
2264 	}
2265 
2266 	/*
2267 	 * Ack processing.
2268 	 */
2269 	switch (tp->t_state) {
2270 
2271 	/*
2272 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2273 	 * ESTABLISHED state and continue processing, otherwise
2274 	 * send an RST.
2275 	 */
2276 	case TCPS_SYN_RECEIVED:
2277 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2278 		    SEQ_GT(th->th_ack, tp->snd_max))
2279 			goto dropwithreset;
2280 		tcpstat.tcps_connects++;
2281 		soisconnected(so);
2282 		tcp_established(tp);
2283 		/* Do window scaling? */
2284 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2285 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2286 			tp->snd_scale = tp->requested_s_scale;
2287 			tp->rcv_scale = tp->request_r_scale;
2288 		}
2289 		TCP_REASS_LOCK(tp);
2290 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2291 		TCP_REASS_UNLOCK(tp);
2292 		tp->snd_wl1 = th->th_seq - 1;
2293 		/* fall into ... */
2294 
2295 	/*
2296 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2297 	 * ACKs.  If the ack is in the range
2298 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2299 	 * then advance tp->snd_una to th->th_ack and drop
2300 	 * data from the retransmission queue.  If this ACK reflects
2301 	 * more up to date window information we update our window information.
2302 	 */
2303 	case TCPS_ESTABLISHED:
2304 	case TCPS_FIN_WAIT_1:
2305 	case TCPS_FIN_WAIT_2:
2306 	case TCPS_CLOSE_WAIT:
2307 	case TCPS_CLOSING:
2308 	case TCPS_LAST_ACK:
2309 	case TCPS_TIME_WAIT:
2310 
2311 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2312 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2313 				tcpstat.tcps_rcvdupack++;
2314 				/*
2315 				 * If we have outstanding data (other than
2316 				 * a window probe), this is a completely
2317 				 * duplicate ack (ie, window info didn't
2318 				 * change), the ack is the biggest we've
2319 				 * seen and we've seen exactly our rexmt
2320 				 * threshhold of them, assume a packet
2321 				 * has been dropped and retransmit it.
2322 				 * Kludge snd_nxt & the congestion
2323 				 * window so we send only this one
2324 				 * packet.
2325 				 */
2326 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2327 				    th->th_ack != tp->snd_una)
2328 					tp->t_dupacks = 0;
2329 				else if (tp->t_partialacks < 0 &&
2330 					 (++tp->t_dupacks == tcprexmtthresh ||
2331 					 TCP_FACK_FASTRECOV(tp))) {
2332 					/*
2333 					 * Do the fast retransmit, and adjust
2334 					 * congestion control paramenters.
2335 					 */
2336 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2337 						/* False fast retransmit */
2338 						break;
2339 					} else
2340 						goto drop;
2341 				} else if (tp->t_dupacks > tcprexmtthresh) {
2342 					tp->snd_cwnd += tp->t_segsz;
2343 					(void) tcp_output(tp);
2344 					goto drop;
2345 				}
2346 			} else {
2347 				/*
2348 				 * If the ack appears to be very old, only
2349 				 * allow data that is in-sequence.  This
2350 				 * makes it somewhat more difficult to insert
2351 				 * forged data by guessing sequence numbers.
2352 				 * Sent an ack to try to update the send
2353 				 * sequence number on the other side.
2354 				 */
2355 				if (tlen && th->th_seq != tp->rcv_nxt &&
2356 				    SEQ_LT(th->th_ack,
2357 				    tp->snd_una - tp->max_sndwnd))
2358 					goto dropafterack;
2359 			}
2360 			break;
2361 		}
2362 		/*
2363 		 * If the congestion window was inflated to account
2364 		 * for the other side's cached packets, retract it.
2365 		 */
2366 		/* XXX: make SACK have his own congestion control
2367 		 * struct -- rpaulo */
2368 		if (TCP_SACK_ENABLED(tp))
2369 			tcp_sack_newack(tp, th);
2370 		else
2371 			tp->t_congctl->fast_retransmit_newack(tp, th);
2372 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2373 			tcpstat.tcps_rcvacktoomuch++;
2374 			goto dropafterack;
2375 		}
2376 		acked = th->th_ack - tp->snd_una;
2377 		tcpstat.tcps_rcvackpack++;
2378 		tcpstat.tcps_rcvackbyte += acked;
2379 
2380 		/*
2381 		 * If we have a timestamp reply, update smoothed
2382 		 * round trip time.  If no timestamp is present but
2383 		 * transmit timer is running and timed sequence
2384 		 * number was acked, update smoothed round trip time.
2385 		 * Since we now have an rtt measurement, cancel the
2386 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2387 		 * Recompute the initial retransmit timer.
2388 		 */
2389 		if (ts_rtt)
2390 			tcp_xmit_timer(tp, ts_rtt);
2391 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2392 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2393 
2394 		/*
2395 		 * If all outstanding data is acked, stop retransmit
2396 		 * timer and remember to restart (more output or persist).
2397 		 * If there is more data to be acked, restart retransmit
2398 		 * timer, using current (possibly backed-off) value.
2399 		 */
2400 		if (th->th_ack == tp->snd_max) {
2401 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2402 			needoutput = 1;
2403 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2404 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2405 
2406 		/*
2407 		 * New data has been acked, adjust the congestion window.
2408 		 */
2409 		tp->t_congctl->newack(tp, th);
2410 
2411 		nd6_hint(tp);
2412 		if (acked > so->so_snd.sb_cc) {
2413 			tp->snd_wnd -= so->so_snd.sb_cc;
2414 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2415 			ourfinisacked = 1;
2416 		} else {
2417 			if (acked > (tp->t_lastoff - tp->t_inoff))
2418 				tp->t_lastm = NULL;
2419 			sbdrop(&so->so_snd, acked);
2420 			tp->t_lastoff -= acked;
2421 			tp->snd_wnd -= acked;
2422 			ourfinisacked = 0;
2423 		}
2424 		sowwakeup(so);
2425 
2426 		icmp_check(tp, th, acked);
2427 
2428 		tp->snd_una = th->th_ack;
2429 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2430 			tp->snd_fack = tp->snd_una;
2431 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2432 			tp->snd_nxt = tp->snd_una;
2433 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2434 			tp->snd_high = tp->snd_una;
2435 
2436 		switch (tp->t_state) {
2437 
2438 		/*
2439 		 * In FIN_WAIT_1 STATE in addition to the processing
2440 		 * for the ESTABLISHED state if our FIN is now acknowledged
2441 		 * then enter FIN_WAIT_2.
2442 		 */
2443 		case TCPS_FIN_WAIT_1:
2444 			if (ourfinisacked) {
2445 				/*
2446 				 * If we can't receive any more
2447 				 * data, then closing user can proceed.
2448 				 * Starting the timer is contrary to the
2449 				 * specification, but if we don't get a FIN
2450 				 * we'll hang forever.
2451 				 */
2452 				if (so->so_state & SS_CANTRCVMORE) {
2453 					soisdisconnected(so);
2454 					if (tp->t_maxidle > 0)
2455 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2456 						    tp->t_maxidle);
2457 				}
2458 				tp->t_state = TCPS_FIN_WAIT_2;
2459 			}
2460 			break;
2461 
2462 	 	/*
2463 		 * In CLOSING STATE in addition to the processing for
2464 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2465 		 * then enter the TIME-WAIT state, otherwise ignore
2466 		 * the segment.
2467 		 */
2468 		case TCPS_CLOSING:
2469 			if (ourfinisacked) {
2470 				tp->t_state = TCPS_TIME_WAIT;
2471 				tcp_canceltimers(tp);
2472 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2473 				soisdisconnected(so);
2474 			}
2475 			break;
2476 
2477 		/*
2478 		 * In LAST_ACK, we may still be waiting for data to drain
2479 		 * and/or to be acked, as well as for the ack of our FIN.
2480 		 * If our FIN is now acknowledged, delete the TCB,
2481 		 * enter the closed state and return.
2482 		 */
2483 		case TCPS_LAST_ACK:
2484 			if (ourfinisacked) {
2485 				tp = tcp_close(tp);
2486 				goto drop;
2487 			}
2488 			break;
2489 
2490 		/*
2491 		 * In TIME_WAIT state the only thing that should arrive
2492 		 * is a retransmission of the remote FIN.  Acknowledge
2493 		 * it and restart the finack timer.
2494 		 */
2495 		case TCPS_TIME_WAIT:
2496 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2497 			goto dropafterack;
2498 		}
2499 	}
2500 
2501 step6:
2502 	/*
2503 	 * Update window information.
2504 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2505 	 */
2506 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2507 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2508 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2509 		/* keep track of pure window updates */
2510 		if (tlen == 0 &&
2511 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2512 			tcpstat.tcps_rcvwinupd++;
2513 		tp->snd_wnd = tiwin;
2514 		tp->snd_wl1 = th->th_seq;
2515 		tp->snd_wl2 = th->th_ack;
2516 		if (tp->snd_wnd > tp->max_sndwnd)
2517 			tp->max_sndwnd = tp->snd_wnd;
2518 		needoutput = 1;
2519 	}
2520 
2521 	/*
2522 	 * Process segments with URG.
2523 	 */
2524 	if ((tiflags & TH_URG) && th->th_urp &&
2525 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2526 		/*
2527 		 * This is a kludge, but if we receive and accept
2528 		 * random urgent pointers, we'll crash in
2529 		 * soreceive.  It's hard to imagine someone
2530 		 * actually wanting to send this much urgent data.
2531 		 */
2532 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2533 			th->th_urp = 0;			/* XXX */
2534 			tiflags &= ~TH_URG;		/* XXX */
2535 			goto dodata;			/* XXX */
2536 		}
2537 		/*
2538 		 * If this segment advances the known urgent pointer,
2539 		 * then mark the data stream.  This should not happen
2540 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2541 		 * a FIN has been received from the remote side.
2542 		 * In these states we ignore the URG.
2543 		 *
2544 		 * According to RFC961 (Assigned Protocols),
2545 		 * the urgent pointer points to the last octet
2546 		 * of urgent data.  We continue, however,
2547 		 * to consider it to indicate the first octet
2548 		 * of data past the urgent section as the original
2549 		 * spec states (in one of two places).
2550 		 */
2551 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2552 			tp->rcv_up = th->th_seq + th->th_urp;
2553 			so->so_oobmark = so->so_rcv.sb_cc +
2554 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2555 			if (so->so_oobmark == 0)
2556 				so->so_state |= SS_RCVATMARK;
2557 			sohasoutofband(so);
2558 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2559 		}
2560 		/*
2561 		 * Remove out of band data so doesn't get presented to user.
2562 		 * This can happen independent of advancing the URG pointer,
2563 		 * but if two URG's are pending at once, some out-of-band
2564 		 * data may creep in... ick.
2565 		 */
2566 		if (th->th_urp <= (u_int16_t) tlen
2567 #ifdef SO_OOBINLINE
2568 		     && (so->so_options & SO_OOBINLINE) == 0
2569 #endif
2570 		     )
2571 			tcp_pulloutofband(so, th, m, hdroptlen);
2572 	} else
2573 		/*
2574 		 * If no out of band data is expected,
2575 		 * pull receive urgent pointer along
2576 		 * with the receive window.
2577 		 */
2578 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2579 			tp->rcv_up = tp->rcv_nxt;
2580 dodata:							/* XXX */
2581 
2582 	/*
2583 	 * Process the segment text, merging it into the TCP sequencing queue,
2584 	 * and arranging for acknowledgement of receipt if necessary.
2585 	 * This process logically involves adjusting tp->rcv_wnd as data
2586 	 * is presented to the user (this happens in tcp_usrreq.c,
2587 	 * case PRU_RCVD).  If a FIN has already been received on this
2588 	 * connection then we just ignore the text.
2589 	 */
2590 	if ((tlen || (tiflags & TH_FIN)) &&
2591 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2592 		/*
2593 		 * Insert segment ti into reassembly queue of tcp with
2594 		 * control block tp.  Return TH_FIN if reassembly now includes
2595 		 * a segment with FIN.  The macro form does the common case
2596 		 * inline (segment is the next to be received on an
2597 		 * established connection, and the queue is empty),
2598 		 * avoiding linkage into and removal from the queue and
2599 		 * repetition of various conversions.
2600 		 * Set DELACK for segments received in order, but ack
2601 		 * immediately when segments are out of order
2602 		 * (so fast retransmit can work).
2603 		 */
2604 		/* NOTE: this was TCP_REASS() macro, but used only once */
2605 		TCP_REASS_LOCK(tp);
2606 		if (th->th_seq == tp->rcv_nxt &&
2607 		    TAILQ_FIRST(&tp->segq) == NULL &&
2608 		    tp->t_state == TCPS_ESTABLISHED) {
2609 			tcp_setup_ack(tp, th);
2610 			tp->rcv_nxt += tlen;
2611 			tiflags = th->th_flags & TH_FIN;
2612 			tcpstat.tcps_rcvpack++;
2613 			tcpstat.tcps_rcvbyte += tlen;
2614 			nd6_hint(tp);
2615 			if (so->so_state & SS_CANTRCVMORE)
2616 				m_freem(m);
2617 			else {
2618 				m_adj(m, hdroptlen);
2619 				sbappendstream(&(so)->so_rcv, m);
2620 			}
2621 			sorwakeup(so);
2622 		} else {
2623 			m_adj(m, hdroptlen);
2624 			tiflags = tcp_reass(tp, th, m, &tlen);
2625 			tp->t_flags |= TF_ACKNOW;
2626 		}
2627 		TCP_REASS_UNLOCK(tp);
2628 
2629 		/*
2630 		 * Note the amount of data that peer has sent into
2631 		 * our window, in order to estimate the sender's
2632 		 * buffer size.
2633 		 */
2634 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2635 	} else {
2636 		m_freem(m);
2637 		m = NULL;
2638 		tiflags &= ~TH_FIN;
2639 	}
2640 
2641 	/*
2642 	 * If FIN is received ACK the FIN and let the user know
2643 	 * that the connection is closing.  Ignore a FIN received before
2644 	 * the connection is fully established.
2645 	 */
2646 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2647 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2648 			socantrcvmore(so);
2649 			tp->t_flags |= TF_ACKNOW;
2650 			tp->rcv_nxt++;
2651 		}
2652 		switch (tp->t_state) {
2653 
2654 	 	/*
2655 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2656 		 */
2657 		case TCPS_ESTABLISHED:
2658 			tp->t_state = TCPS_CLOSE_WAIT;
2659 			break;
2660 
2661 	 	/*
2662 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2663 		 * enter the CLOSING state.
2664 		 */
2665 		case TCPS_FIN_WAIT_1:
2666 			tp->t_state = TCPS_CLOSING;
2667 			break;
2668 
2669 	 	/*
2670 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2671 		 * starting the time-wait timer, turning off the other
2672 		 * standard timers.
2673 		 */
2674 		case TCPS_FIN_WAIT_2:
2675 			tp->t_state = TCPS_TIME_WAIT;
2676 			tcp_canceltimers(tp);
2677 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2678 			soisdisconnected(so);
2679 			break;
2680 
2681 		/*
2682 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2683 		 */
2684 		case TCPS_TIME_WAIT:
2685 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2686 			break;
2687 		}
2688 	}
2689 #ifdef TCP_DEBUG
2690 	if (so->so_options & SO_DEBUG)
2691 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2692 #endif
2693 
2694 	/*
2695 	 * Return any desired output.
2696 	 */
2697 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2698 		(void) tcp_output(tp);
2699 	}
2700 	if (tcp_saveti)
2701 		m_freem(tcp_saveti);
2702 	return;
2703 
2704 badsyn:
2705 	/*
2706 	 * Received a bad SYN.  Increment counters and dropwithreset.
2707 	 */
2708 	tcpstat.tcps_badsyn++;
2709 	tp = NULL;
2710 	goto dropwithreset;
2711 
2712 dropafterack:
2713 	/*
2714 	 * Generate an ACK dropping incoming segment if it occupies
2715 	 * sequence space, where the ACK reflects our state.
2716 	 */
2717 	if (tiflags & TH_RST)
2718 		goto drop;
2719 	goto dropafterack2;
2720 
2721 dropafterack_ratelim:
2722 	/*
2723 	 * We may want to rate-limit ACKs against SYN/RST attack.
2724 	 */
2725 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2726 	    tcp_ackdrop_ppslim) == 0) {
2727 		/* XXX stat */
2728 		goto drop;
2729 	}
2730 	/* ...fall into dropafterack2... */
2731 
2732 dropafterack2:
2733 	m_freem(m);
2734 	tp->t_flags |= TF_ACKNOW;
2735 	(void) tcp_output(tp);
2736 	if (tcp_saveti)
2737 		m_freem(tcp_saveti);
2738 	return;
2739 
2740 dropwithreset_ratelim:
2741 	/*
2742 	 * We may want to rate-limit RSTs in certain situations,
2743 	 * particularly if we are sending an RST in response to
2744 	 * an attempt to connect to or otherwise communicate with
2745 	 * a port for which we have no socket.
2746 	 */
2747 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2748 	    tcp_rst_ppslim) == 0) {
2749 		/* XXX stat */
2750 		goto drop;
2751 	}
2752 	/* ...fall into dropwithreset... */
2753 
2754 dropwithreset:
2755 	/*
2756 	 * Generate a RST, dropping incoming segment.
2757 	 * Make ACK acceptable to originator of segment.
2758 	 */
2759 	if (tiflags & TH_RST)
2760 		goto drop;
2761 
2762 	switch (af) {
2763 #ifdef INET6
2764 	case AF_INET6:
2765 		/* For following calls to tcp_respond */
2766 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2767 			goto drop;
2768 		break;
2769 #endif /* INET6 */
2770 	case AF_INET:
2771 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2772 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2773 			goto drop;
2774 	}
2775 
2776 	if (tiflags & TH_ACK)
2777 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2778 	else {
2779 		if (tiflags & TH_SYN)
2780 			tlen++;
2781 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2782 		    TH_RST|TH_ACK);
2783 	}
2784 	if (tcp_saveti)
2785 		m_freem(tcp_saveti);
2786 	return;
2787 
2788 badcsum:
2789 drop:
2790 	/*
2791 	 * Drop space held by incoming segment and return.
2792 	 */
2793 	if (tp) {
2794 		if (tp->t_inpcb)
2795 			so = tp->t_inpcb->inp_socket;
2796 #ifdef INET6
2797 		else if (tp->t_in6pcb)
2798 			so = tp->t_in6pcb->in6p_socket;
2799 #endif
2800 		else
2801 			so = NULL;
2802 #ifdef TCP_DEBUG
2803 		if (so && (so->so_options & SO_DEBUG) != 0)
2804 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2805 #endif
2806 	}
2807 	if (tcp_saveti)
2808 		m_freem(tcp_saveti);
2809 	m_freem(m);
2810 	return;
2811 }
2812 
2813 #ifdef TCP_SIGNATURE
2814 int
2815 tcp_signature_apply(void *fstate, void *data, u_int len)
2816 {
2817 
2818 	MD5Update(fstate, (u_char *)data, len);
2819 	return (0);
2820 }
2821 
2822 struct secasvar *
2823 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2824 {
2825 	struct secasvar *sav;
2826 #ifdef FAST_IPSEC
2827 	union sockaddr_union dst;
2828 #endif
2829 	struct ip *ip;
2830 	struct ip6_hdr *ip6;
2831 
2832 	ip = mtod(m, struct ip *);
2833 	switch (ip->ip_v) {
2834 	case 4:
2835 		ip = mtod(m, struct ip *);
2836 		ip6 = NULL;
2837 		break;
2838 	case 6:
2839 		ip = NULL;
2840 		ip6 = mtod(m, struct ip6_hdr *);
2841 		break;
2842 	default:
2843 		return (NULL);
2844 	}
2845 
2846 #ifdef FAST_IPSEC
2847 	/* Extract the destination from the IP header in the mbuf. */
2848 	bzero(&dst, sizeof(union sockaddr_union));
2849 	if (ip !=NULL) {
2850 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2851 		dst.sa.sa_family = AF_INET;
2852 		dst.sin.sin_addr = ip->ip_dst;
2853 	} else {
2854 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2855 		dst.sa.sa_family = AF_INET6;
2856 		dst.sin6.sin6_addr = ip6->ip6_dst;
2857 	}
2858 
2859 	/*
2860 	 * Look up an SADB entry which matches the address of the peer.
2861 	 */
2862 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2863 #else
2864 	if (ip)
2865 		sav = key_allocsa(AF_INET, (void *)&ip->ip_src,
2866 		    (void *)&ip->ip_dst, IPPROTO_TCP,
2867 		    htonl(TCP_SIG_SPI), 0, 0);
2868 	else
2869 		sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src,
2870 		    (void *)&ip6->ip6_dst, IPPROTO_TCP,
2871 		    htonl(TCP_SIG_SPI), 0, 0);
2872 #endif
2873 
2874 	return (sav);	/* freesav must be performed by caller */
2875 }
2876 
2877 int
2878 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2879     struct secasvar *sav, char *sig)
2880 {
2881 	MD5_CTX ctx;
2882 	struct ip *ip;
2883 	struct ipovly *ipovly;
2884 	struct ip6_hdr *ip6;
2885 	struct ippseudo ippseudo;
2886 	struct ip6_hdr_pseudo ip6pseudo;
2887 	struct tcphdr th0;
2888 	int l, tcphdrlen;
2889 
2890 	if (sav == NULL)
2891 		return (-1);
2892 
2893 	tcphdrlen = th->th_off * 4;
2894 
2895 	switch (mtod(m, struct ip *)->ip_v) {
2896 	case 4:
2897 		ip = mtod(m, struct ip *);
2898 		ip6 = NULL;
2899 		break;
2900 	case 6:
2901 		ip = NULL;
2902 		ip6 = mtod(m, struct ip6_hdr *);
2903 		break;
2904 	default:
2905 		return (-1);
2906 	}
2907 
2908 	MD5Init(&ctx);
2909 
2910 	if (ip) {
2911 		memset(&ippseudo, 0, sizeof(ippseudo));
2912 		ipovly = (struct ipovly *)ip;
2913 		ippseudo.ippseudo_src = ipovly->ih_src;
2914 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2915 		ippseudo.ippseudo_pad = 0;
2916 		ippseudo.ippseudo_p = IPPROTO_TCP;
2917 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2918 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2919 	} else {
2920 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2921 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2922 		in6_clearscope(&ip6pseudo.ip6ph_src);
2923 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2924 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2925 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2926 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2927 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2928 	}
2929 
2930 	th0 = *th;
2931 	th0.th_sum = 0;
2932 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2933 
2934 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2935 	if (l > 0)
2936 		m_apply(m, thoff + tcphdrlen,
2937 		    m->m_pkthdr.len - thoff - tcphdrlen,
2938 		    tcp_signature_apply, &ctx);
2939 
2940 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2941 	MD5Final(sig, &ctx);
2942 
2943 	return (0);
2944 }
2945 #endif
2946 
2947 static int
2948 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt,
2949     struct tcphdr *th,
2950     struct mbuf *m, int toff, struct tcp_opt_info *oi)
2951 {
2952 	u_int16_t mss;
2953 	int opt, optlen = 0;
2954 #ifdef TCP_SIGNATURE
2955 	void *sigp = NULL;
2956 	char sigbuf[TCP_SIGLEN];
2957 	struct secasvar *sav = NULL;
2958 #endif
2959 
2960 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2961 		opt = cp[0];
2962 		if (opt == TCPOPT_EOL)
2963 			break;
2964 		if (opt == TCPOPT_NOP)
2965 			optlen = 1;
2966 		else {
2967 			if (cnt < 2)
2968 				break;
2969 			optlen = cp[1];
2970 			if (optlen < 2 || optlen > cnt)
2971 				break;
2972 		}
2973 		switch (opt) {
2974 
2975 		default:
2976 			continue;
2977 
2978 		case TCPOPT_MAXSEG:
2979 			if (optlen != TCPOLEN_MAXSEG)
2980 				continue;
2981 			if (!(th->th_flags & TH_SYN))
2982 				continue;
2983 			if (TCPS_HAVERCVDSYN(tp->t_state))
2984 				continue;
2985 			bcopy(cp + 2, &mss, sizeof(mss));
2986 			oi->maxseg = ntohs(mss);
2987 			break;
2988 
2989 		case TCPOPT_WINDOW:
2990 			if (optlen != TCPOLEN_WINDOW)
2991 				continue;
2992 			if (!(th->th_flags & TH_SYN))
2993 				continue;
2994 			if (TCPS_HAVERCVDSYN(tp->t_state))
2995 				continue;
2996 			tp->t_flags |= TF_RCVD_SCALE;
2997 			tp->requested_s_scale = cp[2];
2998 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2999 #if 0	/*XXX*/
3000 				char *p;
3001 
3002 				if (ip)
3003 					p = ntohl(ip->ip_src);
3004 #ifdef INET6
3005 				else if (ip6)
3006 					p = ip6_sprintf(&ip6->ip6_src);
3007 #endif
3008 				else
3009 					p = "(unknown)";
3010 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
3011 				    "assuming %d\n",
3012 				    tp->requested_s_scale, p,
3013 				    TCP_MAX_WINSHIFT);
3014 #else
3015 				log(LOG_ERR, "TCP: invalid wscale %d, "
3016 				    "assuming %d\n",
3017 				    tp->requested_s_scale,
3018 				    TCP_MAX_WINSHIFT);
3019 #endif
3020 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
3021 			}
3022 			break;
3023 
3024 		case TCPOPT_TIMESTAMP:
3025 			if (optlen != TCPOLEN_TIMESTAMP)
3026 				continue;
3027 			oi->ts_present = 1;
3028 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
3029 			NTOHL(oi->ts_val);
3030 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
3031 			NTOHL(oi->ts_ecr);
3032 
3033 			if (!(th->th_flags & TH_SYN))
3034 				continue;
3035 			if (TCPS_HAVERCVDSYN(tp->t_state))
3036 				continue;
3037 			/*
3038 			 * A timestamp received in a SYN makes
3039 			 * it ok to send timestamp requests and replies.
3040 			 */
3041 			tp->t_flags |= TF_RCVD_TSTMP;
3042 			tp->ts_recent = oi->ts_val;
3043 			tp->ts_recent_age = tcp_now;
3044                         break;
3045 
3046 		case TCPOPT_SACK_PERMITTED:
3047 			if (optlen != TCPOLEN_SACK_PERMITTED)
3048 				continue;
3049 			if (!(th->th_flags & TH_SYN))
3050 				continue;
3051 			if (TCPS_HAVERCVDSYN(tp->t_state))
3052 				continue;
3053 			if (tcp_do_sack) {
3054 				tp->t_flags |= TF_SACK_PERMIT;
3055 				tp->t_flags |= TF_WILL_SACK;
3056 			}
3057 			break;
3058 
3059 		case TCPOPT_SACK:
3060 			tcp_sack_option(tp, th, cp, optlen);
3061 			break;
3062 #ifdef TCP_SIGNATURE
3063 		case TCPOPT_SIGNATURE:
3064 			if (optlen != TCPOLEN_SIGNATURE)
3065 				continue;
3066 			if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN))
3067 				return (-1);
3068 
3069 			sigp = sigbuf;
3070 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
3071 			tp->t_flags |= TF_SIGNATURE;
3072 			break;
3073 #endif
3074 		}
3075 	}
3076 
3077 #ifdef TCP_SIGNATURE
3078 	if (tp->t_flags & TF_SIGNATURE) {
3079 
3080 		sav = tcp_signature_getsav(m, th);
3081 
3082 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3083 			return (-1);
3084 	}
3085 
3086 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3087 		if (sav == NULL)
3088 			return (-1);
3089 #ifdef FAST_IPSEC
3090 		KEY_FREESAV(&sav);
3091 #else
3092 		key_freesav(sav);
3093 #endif
3094 		return (-1);
3095 	}
3096 
3097 	if (sigp) {
3098 		char sig[TCP_SIGLEN];
3099 
3100 		tcp_fields_to_net(th);
3101 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3102 			tcp_fields_to_host(th);
3103 			if (sav == NULL)
3104 				return (-1);
3105 #ifdef FAST_IPSEC
3106 			KEY_FREESAV(&sav);
3107 #else
3108 			key_freesav(sav);
3109 #endif
3110 			return (-1);
3111 		}
3112 		tcp_fields_to_host(th);
3113 
3114 		if (bcmp(sig, sigp, TCP_SIGLEN)) {
3115 			tcpstat.tcps_badsig++;
3116 			if (sav == NULL)
3117 				return (-1);
3118 #ifdef FAST_IPSEC
3119 			KEY_FREESAV(&sav);
3120 #else
3121 			key_freesav(sav);
3122 #endif
3123 			return (-1);
3124 		} else
3125 			tcpstat.tcps_goodsig++;
3126 
3127 		key_sa_recordxfer(sav, m);
3128 #ifdef FAST_IPSEC
3129 		KEY_FREESAV(&sav);
3130 #else
3131 		key_freesav(sav);
3132 #endif
3133 	}
3134 #endif
3135 
3136 	return (0);
3137 }
3138 
3139 /*
3140  * Pull out of band byte out of a segment so
3141  * it doesn't appear in the user's data queue.
3142  * It is still reflected in the segment length for
3143  * sequencing purposes.
3144  */
3145 void
3146 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3147     struct mbuf *m, int off)
3148 {
3149 	int cnt = off + th->th_urp - 1;
3150 
3151 	while (cnt >= 0) {
3152 		if (m->m_len > cnt) {
3153 			char *cp = mtod(m, char *) + cnt;
3154 			struct tcpcb *tp = sototcpcb(so);
3155 
3156 			tp->t_iobc = *cp;
3157 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3158 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3159 			m->m_len--;
3160 			return;
3161 		}
3162 		cnt -= m->m_len;
3163 		m = m->m_next;
3164 		if (m == 0)
3165 			break;
3166 	}
3167 	panic("tcp_pulloutofband");
3168 }
3169 
3170 /*
3171  * Collect new round-trip time estimate
3172  * and update averages and current timeout.
3173  */
3174 void
3175 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3176 {
3177 	int32_t delta;
3178 
3179 	tcpstat.tcps_rttupdated++;
3180 	if (tp->t_srtt != 0) {
3181 		/*
3182 		 * srtt is stored as fixed point with 3 bits after the
3183 		 * binary point (i.e., scaled by 8).  The following magic
3184 		 * is equivalent to the smoothing algorithm in rfc793 with
3185 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3186 		 * point).  Adjust rtt to origin 0.
3187 		 */
3188 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3189 		if ((tp->t_srtt += delta) <= 0)
3190 			tp->t_srtt = 1 << 2;
3191 		/*
3192 		 * We accumulate a smoothed rtt variance (actually, a
3193 		 * smoothed mean difference), then set the retransmit
3194 		 * timer to smoothed rtt + 4 times the smoothed variance.
3195 		 * rttvar is stored as fixed point with 2 bits after the
3196 		 * binary point (scaled by 4).  The following is
3197 		 * equivalent to rfc793 smoothing with an alpha of .75
3198 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3199 		 * rfc793's wired-in beta.
3200 		 */
3201 		if (delta < 0)
3202 			delta = -delta;
3203 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3204 		if ((tp->t_rttvar += delta) <= 0)
3205 			tp->t_rttvar = 1 << 2;
3206 	} else {
3207 		/*
3208 		 * No rtt measurement yet - use the unsmoothed rtt.
3209 		 * Set the variance to half the rtt (so our first
3210 		 * retransmit happens at 3*rtt).
3211 		 */
3212 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3213 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3214 	}
3215 	tp->t_rtttime = 0;
3216 	tp->t_rxtshift = 0;
3217 
3218 	/*
3219 	 * the retransmit should happen at rtt + 4 * rttvar.
3220 	 * Because of the way we do the smoothing, srtt and rttvar
3221 	 * will each average +1/2 tick of bias.  When we compute
3222 	 * the retransmit timer, we want 1/2 tick of rounding and
3223 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3224 	 * firing of the timer.  The bias will give us exactly the
3225 	 * 1.5 tick we need.  But, because the bias is
3226 	 * statistical, we have to test that we don't drop below
3227 	 * the minimum feasible timer (which is 2 ticks).
3228 	 */
3229 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3230 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3231 
3232 	/*
3233 	 * We received an ack for a packet that wasn't retransmitted;
3234 	 * it is probably safe to discard any error indications we've
3235 	 * received recently.  This isn't quite right, but close enough
3236 	 * for now (a route might have failed after we sent a segment,
3237 	 * and the return path might not be symmetrical).
3238 	 */
3239 	tp->t_softerror = 0;
3240 }
3241 
3242 
3243 /*
3244  * TCP compressed state engine.  Currently used to hold compressed
3245  * state for SYN_RECEIVED.
3246  */
3247 
3248 u_long	syn_cache_count;
3249 u_int32_t syn_hash1, syn_hash2;
3250 
3251 #define SYN_HASH(sa, sp, dp) \
3252 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3253 				     ((u_int32_t)(sp)))^syn_hash2)))
3254 #ifndef INET6
3255 #define	SYN_HASHALL(hash, src, dst) \
3256 do {									\
3257 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3258 		((const struct sockaddr_in *)(src))->sin_port,		\
3259 		((const struct sockaddr_in *)(dst))->sin_port);		\
3260 } while (/*CONSTCOND*/ 0)
3261 #else
3262 #define SYN_HASH6(sa, sp, dp) \
3263 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3264 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3265 	 & 0x7fffffff)
3266 
3267 #define SYN_HASHALL(hash, src, dst) \
3268 do {									\
3269 	switch ((src)->sa_family) {					\
3270 	case AF_INET:							\
3271 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3272 			((const struct sockaddr_in *)(src))->sin_port,	\
3273 			((const struct sockaddr_in *)(dst))->sin_port);	\
3274 		break;							\
3275 	case AF_INET6:							\
3276 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3277 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3278 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3279 		break;							\
3280 	default:							\
3281 		hash = 0;						\
3282 	}								\
3283 } while (/*CONSTCOND*/0)
3284 #endif /* INET6 */
3285 
3286 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL,
3287     IPL_SOFTNET);
3288 
3289 /*
3290  * We don't estimate RTT with SYNs, so each packet starts with the default
3291  * RTT and each timer step has a fixed timeout value.
3292  */
3293 #define	SYN_CACHE_TIMER_ARM(sc)						\
3294 do {									\
3295 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3296 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3297 	    TCPTV_REXMTMAX);						\
3298 	callout_reset(&(sc)->sc_timer,					\
3299 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3300 } while (/*CONSTCOND*/0)
3301 
3302 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3303 
3304 static inline void
3305 syn_cache_rm(struct syn_cache *sc)
3306 {
3307 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3308 	    sc, sc_bucketq);
3309 	sc->sc_tp = NULL;
3310 	LIST_REMOVE(sc, sc_tpq);
3311 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3312 	callout_stop(&sc->sc_timer);
3313 	syn_cache_count--;
3314 }
3315 
3316 static inline void
3317 syn_cache_put(struct syn_cache *sc)
3318 {
3319 	if (sc->sc_ipopts)
3320 		(void) m_free(sc->sc_ipopts);
3321 	rtcache_free(&sc->sc_route);
3322 	if (callout_invoking(&sc->sc_timer))
3323 		sc->sc_flags |= SCF_DEAD;
3324 	else {
3325 		callout_destroy(&sc->sc_timer);
3326 		pool_put(&syn_cache_pool, sc);
3327 	}
3328 }
3329 
3330 void
3331 syn_cache_init(void)
3332 {
3333 	int i;
3334 
3335 	/* Initialize the hash buckets. */
3336 	for (i = 0; i < tcp_syn_cache_size; i++)
3337 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3338 }
3339 
3340 void
3341 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3342 {
3343 	struct syn_cache_head *scp;
3344 	struct syn_cache *sc2;
3345 	int s;
3346 
3347 	/*
3348 	 * If there are no entries in the hash table, reinitialize
3349 	 * the hash secrets.
3350 	 */
3351 	if (syn_cache_count == 0) {
3352 		syn_hash1 = arc4random();
3353 		syn_hash2 = arc4random();
3354 	}
3355 
3356 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3357 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3358 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3359 
3360 	/*
3361 	 * Make sure that we don't overflow the per-bucket
3362 	 * limit or the total cache size limit.
3363 	 */
3364 	s = splsoftnet();
3365 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3366 		tcpstat.tcps_sc_bucketoverflow++;
3367 		/*
3368 		 * The bucket is full.  Toss the oldest element in the
3369 		 * bucket.  This will be the first entry in the bucket.
3370 		 */
3371 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3372 #ifdef DIAGNOSTIC
3373 		/*
3374 		 * This should never happen; we should always find an
3375 		 * entry in our bucket.
3376 		 */
3377 		if (sc2 == NULL)
3378 			panic("syn_cache_insert: bucketoverflow: impossible");
3379 #endif
3380 		syn_cache_rm(sc2);
3381 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3382 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3383 		struct syn_cache_head *scp2, *sce;
3384 
3385 		tcpstat.tcps_sc_overflowed++;
3386 		/*
3387 		 * The cache is full.  Toss the oldest entry in the
3388 		 * first non-empty bucket we can find.
3389 		 *
3390 		 * XXX We would really like to toss the oldest
3391 		 * entry in the cache, but we hope that this
3392 		 * condition doesn't happen very often.
3393 		 */
3394 		scp2 = scp;
3395 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3396 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3397 			for (++scp2; scp2 != scp; scp2++) {
3398 				if (scp2 >= sce)
3399 					scp2 = &tcp_syn_cache[0];
3400 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3401 					break;
3402 			}
3403 #ifdef DIAGNOSTIC
3404 			/*
3405 			 * This should never happen; we should always find a
3406 			 * non-empty bucket.
3407 			 */
3408 			if (scp2 == scp)
3409 				panic("syn_cache_insert: cacheoverflow: "
3410 				    "impossible");
3411 #endif
3412 		}
3413 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3414 		syn_cache_rm(sc2);
3415 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3416 	}
3417 
3418 	/*
3419 	 * Initialize the entry's timer.
3420 	 */
3421 	sc->sc_rxttot = 0;
3422 	sc->sc_rxtshift = 0;
3423 	SYN_CACHE_TIMER_ARM(sc);
3424 
3425 	/* Link it from tcpcb entry */
3426 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3427 
3428 	/* Put it into the bucket. */
3429 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3430 	scp->sch_length++;
3431 	syn_cache_count++;
3432 
3433 	tcpstat.tcps_sc_added++;
3434 	splx(s);
3435 }
3436 
3437 /*
3438  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3439  * If we have retransmitted an entry the maximum number of times, expire
3440  * that entry.
3441  */
3442 void
3443 syn_cache_timer(void *arg)
3444 {
3445 	struct syn_cache *sc = arg;
3446 	int s;
3447 
3448 	s = splsoftnet();
3449 	callout_ack(&sc->sc_timer);
3450 
3451 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3452 		tcpstat.tcps_sc_delayed_free++;
3453 		callout_destroy(&sc->sc_timer);
3454 		pool_put(&syn_cache_pool, sc);
3455 		splx(s);
3456 		return;
3457 	}
3458 
3459 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3460 		/* Drop it -- too many retransmissions. */
3461 		goto dropit;
3462 	}
3463 
3464 	/*
3465 	 * Compute the total amount of time this entry has
3466 	 * been on a queue.  If this entry has been on longer
3467 	 * than the keep alive timer would allow, expire it.
3468 	 */
3469 	sc->sc_rxttot += sc->sc_rxtcur;
3470 	if (sc->sc_rxttot >= tcp_keepinit)
3471 		goto dropit;
3472 
3473 	tcpstat.tcps_sc_retransmitted++;
3474 	(void) syn_cache_respond(sc, NULL);
3475 
3476 	/* Advance the timer back-off. */
3477 	sc->sc_rxtshift++;
3478 	SYN_CACHE_TIMER_ARM(sc);
3479 
3480 	splx(s);
3481 	return;
3482 
3483  dropit:
3484 	tcpstat.tcps_sc_timed_out++;
3485 	syn_cache_rm(sc);
3486 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3487 	splx(s);
3488 }
3489 
3490 /*
3491  * Remove syn cache created by the specified tcb entry,
3492  * because this does not make sense to keep them
3493  * (if there's no tcb entry, syn cache entry will never be used)
3494  */
3495 void
3496 syn_cache_cleanup(struct tcpcb *tp)
3497 {
3498 	struct syn_cache *sc, *nsc;
3499 	int s;
3500 
3501 	s = splsoftnet();
3502 
3503 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3504 		nsc = LIST_NEXT(sc, sc_tpq);
3505 
3506 #ifdef DIAGNOSTIC
3507 		if (sc->sc_tp != tp)
3508 			panic("invalid sc_tp in syn_cache_cleanup");
3509 #endif
3510 		syn_cache_rm(sc);
3511 		syn_cache_put(sc);	/* calls pool_put but see spl above */
3512 	}
3513 	/* just for safety */
3514 	LIST_INIT(&tp->t_sc);
3515 
3516 	splx(s);
3517 }
3518 
3519 /*
3520  * Find an entry in the syn cache.
3521  */
3522 struct syn_cache *
3523 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3524     struct syn_cache_head **headp)
3525 {
3526 	struct syn_cache *sc;
3527 	struct syn_cache_head *scp;
3528 	u_int32_t hash;
3529 	int s;
3530 
3531 	SYN_HASHALL(hash, src, dst);
3532 
3533 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3534 	*headp = scp;
3535 	s = splsoftnet();
3536 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3537 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3538 		if (sc->sc_hash != hash)
3539 			continue;
3540 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3541 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3542 			splx(s);
3543 			return (sc);
3544 		}
3545 	}
3546 	splx(s);
3547 	return (NULL);
3548 }
3549 
3550 /*
3551  * This function gets called when we receive an ACK for a
3552  * socket in the LISTEN state.  We look up the connection
3553  * in the syn cache, and if its there, we pull it out of
3554  * the cache and turn it into a full-blown connection in
3555  * the SYN-RECEIVED state.
3556  *
3557  * The return values may not be immediately obvious, and their effects
3558  * can be subtle, so here they are:
3559  *
3560  *	NULL	SYN was not found in cache; caller should drop the
3561  *		packet and send an RST.
3562  *
3563  *	-1	We were unable to create the new connection, and are
3564  *		aborting it.  An ACK,RST is being sent to the peer
3565  *		(unless we got screwey sequence numbners; see below),
3566  *		because the 3-way handshake has been completed.  Caller
3567  *		should not free the mbuf, since we may be using it.  If
3568  *		we are not, we will free it.
3569  *
3570  *	Otherwise, the return value is a pointer to the new socket
3571  *	associated with the connection.
3572  */
3573 struct socket *
3574 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3575     struct tcphdr *th, unsigned int hlen, unsigned int tlen,
3576     struct socket *so, struct mbuf *m)
3577 {
3578 	struct syn_cache *sc;
3579 	struct syn_cache_head *scp;
3580 	struct inpcb *inp = NULL;
3581 #ifdef INET6
3582 	struct in6pcb *in6p = NULL;
3583 #endif
3584 	struct tcpcb *tp = 0;
3585 	struct mbuf *am;
3586 	int s;
3587 	struct socket *oso;
3588 
3589 	s = splsoftnet();
3590 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3591 		splx(s);
3592 		return (NULL);
3593 	}
3594 
3595 	/*
3596 	 * Verify the sequence and ack numbers.  Try getting the correct
3597 	 * response again.
3598 	 */
3599 	if ((th->th_ack != sc->sc_iss + 1) ||
3600 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3601 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3602 		(void) syn_cache_respond(sc, m);
3603 		splx(s);
3604 		return ((struct socket *)(-1));
3605 	}
3606 
3607 	/* Remove this cache entry */
3608 	syn_cache_rm(sc);
3609 	splx(s);
3610 
3611 	/*
3612 	 * Ok, create the full blown connection, and set things up
3613 	 * as they would have been set up if we had created the
3614 	 * connection when the SYN arrived.  If we can't create
3615 	 * the connection, abort it.
3616 	 */
3617 	/*
3618 	 * inp still has the OLD in_pcb stuff, set the
3619 	 * v6-related flags on the new guy, too.   This is
3620 	 * done particularly for the case where an AF_INET6
3621 	 * socket is bound only to a port, and a v4 connection
3622 	 * comes in on that port.
3623 	 * we also copy the flowinfo from the original pcb
3624 	 * to the new one.
3625 	 */
3626 	oso = so;
3627 	so = sonewconn(so, SS_ISCONNECTED);
3628 	if (so == NULL)
3629 		goto resetandabort;
3630 
3631 	switch (so->so_proto->pr_domain->dom_family) {
3632 #ifdef INET
3633 	case AF_INET:
3634 		inp = sotoinpcb(so);
3635 		break;
3636 #endif
3637 #ifdef INET6
3638 	case AF_INET6:
3639 		in6p = sotoin6pcb(so);
3640 		break;
3641 #endif
3642 	}
3643 	switch (src->sa_family) {
3644 #ifdef INET
3645 	case AF_INET:
3646 		if (inp) {
3647 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3648 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3649 			inp->inp_options = ip_srcroute();
3650 			in_pcbstate(inp, INP_BOUND);
3651 			if (inp->inp_options == NULL) {
3652 				inp->inp_options = sc->sc_ipopts;
3653 				sc->sc_ipopts = NULL;
3654 			}
3655 		}
3656 #ifdef INET6
3657 		else if (in6p) {
3658 			/* IPv4 packet to AF_INET6 socket */
3659 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3660 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3661 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3662 				&in6p->in6p_laddr.s6_addr32[3],
3663 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3664 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3665 			in6totcpcb(in6p)->t_family = AF_INET;
3666 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3667 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3668 			else
3669 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3670 			in6_pcbstate(in6p, IN6P_BOUND);
3671 		}
3672 #endif
3673 		break;
3674 #endif
3675 #ifdef INET6
3676 	case AF_INET6:
3677 		if (in6p) {
3678 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3679 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3680 			in6_pcbstate(in6p, IN6P_BOUND);
3681 		}
3682 		break;
3683 #endif
3684 	}
3685 #ifdef INET6
3686 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3687 		struct in6pcb *oin6p = sotoin6pcb(oso);
3688 		/* inherit socket options from the listening socket */
3689 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3690 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3691 			m_freem(in6p->in6p_options);
3692 			in6p->in6p_options = 0;
3693 		}
3694 		ip6_savecontrol(in6p, &in6p->in6p_options,
3695 			mtod(m, struct ip6_hdr *), m);
3696 	}
3697 #endif
3698 
3699 #if defined(IPSEC) || defined(FAST_IPSEC)
3700 	/*
3701 	 * we make a copy of policy, instead of sharing the policy,
3702 	 * for better behavior in terms of SA lookup and dead SA removal.
3703 	 */
3704 	if (inp) {
3705 		/* copy old policy into new socket's */
3706 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3707 			printf("tcp_input: could not copy policy\n");
3708 	}
3709 #ifdef INET6
3710 	else if (in6p) {
3711 		/* copy old policy into new socket's */
3712 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3713 		    in6p->in6p_sp))
3714 			printf("tcp_input: could not copy policy\n");
3715 	}
3716 #endif
3717 #endif
3718 
3719 	/*
3720 	 * Give the new socket our cached route reference.
3721 	 */
3722 	if (inp) {
3723 		rtcache_copy(&inp->inp_route, &sc->sc_route);
3724 		rtcache_free(&sc->sc_route);
3725 	}
3726 #ifdef INET6
3727 	else {
3728 		rtcache_copy(&in6p->in6p_route, &sc->sc_route);
3729 		rtcache_free(&sc->sc_route);
3730 	}
3731 #endif
3732 
3733 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3734 	if (am == NULL)
3735 		goto resetandabort;
3736 	MCLAIM(am, &tcp_mowner);
3737 	am->m_len = src->sa_len;
3738 	bcopy(src, mtod(am, void *), src->sa_len);
3739 	if (inp) {
3740 		if (in_pcbconnect(inp, am, &lwp0)) {
3741 			(void) m_free(am);
3742 			goto resetandabort;
3743 		}
3744 	}
3745 #ifdef INET6
3746 	else if (in6p) {
3747 		if (src->sa_family == AF_INET) {
3748 			/* IPv4 packet to AF_INET6 socket */
3749 			struct sockaddr_in6 *sin6;
3750 			sin6 = mtod(am, struct sockaddr_in6 *);
3751 			am->m_len = sizeof(*sin6);
3752 			bzero(sin6, sizeof(*sin6));
3753 			sin6->sin6_family = AF_INET6;
3754 			sin6->sin6_len = sizeof(*sin6);
3755 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3756 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3757 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3758 				&sin6->sin6_addr.s6_addr32[3],
3759 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3760 		}
3761 		if (in6_pcbconnect(in6p, am, NULL)) {
3762 			(void) m_free(am);
3763 			goto resetandabort;
3764 		}
3765 	}
3766 #endif
3767 	else {
3768 		(void) m_free(am);
3769 		goto resetandabort;
3770 	}
3771 	(void) m_free(am);
3772 
3773 	if (inp)
3774 		tp = intotcpcb(inp);
3775 #ifdef INET6
3776 	else if (in6p)
3777 		tp = in6totcpcb(in6p);
3778 #endif
3779 	else
3780 		tp = NULL;
3781 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3782 	if (sc->sc_request_r_scale != 15) {
3783 		tp->requested_s_scale = sc->sc_requested_s_scale;
3784 		tp->request_r_scale = sc->sc_request_r_scale;
3785 		tp->snd_scale = sc->sc_requested_s_scale;
3786 		tp->rcv_scale = sc->sc_request_r_scale;
3787 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3788 	}
3789 	if (sc->sc_flags & SCF_TIMESTAMP)
3790 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3791 	tp->ts_timebase = sc->sc_timebase;
3792 
3793 	tp->t_template = tcp_template(tp);
3794 	if (tp->t_template == 0) {
3795 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3796 		so = NULL;
3797 		m_freem(m);
3798 		goto abort;
3799 	}
3800 
3801 	tp->iss = sc->sc_iss;
3802 	tp->irs = sc->sc_irs;
3803 	tcp_sendseqinit(tp);
3804 	tcp_rcvseqinit(tp);
3805 	tp->t_state = TCPS_SYN_RECEIVED;
3806 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit);
3807 	tcpstat.tcps_accepts++;
3808 
3809 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3810 		tp->t_flags |= TF_WILL_SACK;
3811 
3812 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3813 		tp->t_flags |= TF_ECN_PERMIT;
3814 
3815 #ifdef TCP_SIGNATURE
3816 	if (sc->sc_flags & SCF_SIGNATURE)
3817 		tp->t_flags |= TF_SIGNATURE;
3818 #endif
3819 
3820 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3821 	tp->t_ourmss = sc->sc_ourmaxseg;
3822 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3823 
3824 	/*
3825 	 * Initialize the initial congestion window.  If we
3826 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3827 	 * to 1 segment (i.e. the Loss Window).
3828 	 */
3829 	if (sc->sc_rxtshift)
3830 		tp->snd_cwnd = tp->t_peermss;
3831 	else {
3832 		int ss = tcp_init_win;
3833 #ifdef INET
3834 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3835 			ss = tcp_init_win_local;
3836 #endif
3837 #ifdef INET6
3838 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3839 			ss = tcp_init_win_local;
3840 #endif
3841 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3842 	}
3843 
3844 	tcp_rmx_rtt(tp);
3845 	tp->snd_wl1 = sc->sc_irs;
3846 	tp->rcv_up = sc->sc_irs + 1;
3847 
3848 	/*
3849 	 * This is what whould have happened in tcp_output() when
3850 	 * the SYN,ACK was sent.
3851 	 */
3852 	tp->snd_up = tp->snd_una;
3853 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3854 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3855 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3856 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3857 	tp->last_ack_sent = tp->rcv_nxt;
3858 	tp->t_partialacks = -1;
3859 	tp->t_dupacks = 0;
3860 
3861 	tcpstat.tcps_sc_completed++;
3862 	s = splsoftnet();
3863 	syn_cache_put(sc);
3864 	splx(s);
3865 	return (so);
3866 
3867 resetandabort:
3868 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3869 abort:
3870 	if (so != NULL)
3871 		(void) soabort(so);
3872 	s = splsoftnet();
3873 	syn_cache_put(sc);
3874 	splx(s);
3875 	tcpstat.tcps_sc_aborted++;
3876 	return ((struct socket *)(-1));
3877 }
3878 
3879 /*
3880  * This function is called when we get a RST for a
3881  * non-existent connection, so that we can see if the
3882  * connection is in the syn cache.  If it is, zap it.
3883  */
3884 
3885 void
3886 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3887 {
3888 	struct syn_cache *sc;
3889 	struct syn_cache_head *scp;
3890 	int s = splsoftnet();
3891 
3892 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3893 		splx(s);
3894 		return;
3895 	}
3896 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3897 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3898 		splx(s);
3899 		return;
3900 	}
3901 	syn_cache_rm(sc);
3902 	tcpstat.tcps_sc_reset++;
3903 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3904 	splx(s);
3905 }
3906 
3907 void
3908 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3909     struct tcphdr *th)
3910 {
3911 	struct syn_cache *sc;
3912 	struct syn_cache_head *scp;
3913 	int s;
3914 
3915 	s = splsoftnet();
3916 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3917 		splx(s);
3918 		return;
3919 	}
3920 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3921 	if (ntohl (th->th_seq) != sc->sc_iss) {
3922 		splx(s);
3923 		return;
3924 	}
3925 
3926 	/*
3927 	 * If we've retransmitted 3 times and this is our second error,
3928 	 * we remove the entry.  Otherwise, we allow it to continue on.
3929 	 * This prevents us from incorrectly nuking an entry during a
3930 	 * spurious network outage.
3931 	 *
3932 	 * See tcp_notify().
3933 	 */
3934 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3935 		sc->sc_flags |= SCF_UNREACH;
3936 		splx(s);
3937 		return;
3938 	}
3939 
3940 	syn_cache_rm(sc);
3941 	tcpstat.tcps_sc_unreach++;
3942 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3943 	splx(s);
3944 }
3945 
3946 /*
3947  * Given a LISTEN socket and an inbound SYN request, add
3948  * this to the syn cache, and send back a segment:
3949  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3950  * to the source.
3951  *
3952  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3953  * Doing so would require that we hold onto the data and deliver it
3954  * to the application.  However, if we are the target of a SYN-flood
3955  * DoS attack, an attacker could send data which would eventually
3956  * consume all available buffer space if it were ACKed.  By not ACKing
3957  * the data, we avoid this DoS scenario.
3958  */
3959 
3960 int
3961 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3962     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
3963     int optlen, struct tcp_opt_info *oi)
3964 {
3965 	struct tcpcb tb, *tp;
3966 	long win;
3967 	struct syn_cache *sc;
3968 	struct syn_cache_head *scp;
3969 	struct mbuf *ipopts;
3970 	struct tcp_opt_info opti;
3971 	int s;
3972 
3973 	tp = sototcpcb(so);
3974 
3975 	bzero(&opti, sizeof(opti));
3976 
3977 	/*
3978 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3979 	 *
3980 	 * Note this check is performed in tcp_input() very early on.
3981 	 */
3982 
3983 	/*
3984 	 * Initialize some local state.
3985 	 */
3986 	win = sbspace(&so->so_rcv);
3987 	if (win > TCP_MAXWIN)
3988 		win = TCP_MAXWIN;
3989 
3990 	switch (src->sa_family) {
3991 #ifdef INET
3992 	case AF_INET:
3993 		/*
3994 		 * Remember the IP options, if any.
3995 		 */
3996 		ipopts = ip_srcroute();
3997 		break;
3998 #endif
3999 	default:
4000 		ipopts = NULL;
4001 	}
4002 
4003 #ifdef TCP_SIGNATURE
4004 	if (optp || (tp->t_flags & TF_SIGNATURE))
4005 #else
4006 	if (optp)
4007 #endif
4008 	{
4009 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4010 #ifdef TCP_SIGNATURE
4011 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
4012 #endif
4013 		tb.t_state = TCPS_LISTEN;
4014 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
4015 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
4016 			return (0);
4017 	} else
4018 		tb.t_flags = 0;
4019 
4020 	/*
4021 	 * See if we already have an entry for this connection.
4022 	 * If we do, resend the SYN,ACK.  We do not count this
4023 	 * as a retransmission (XXX though maybe we should).
4024 	 */
4025 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4026 		tcpstat.tcps_sc_dupesyn++;
4027 		if (ipopts) {
4028 			/*
4029 			 * If we were remembering a previous source route,
4030 			 * forget it and use the new one we've been given.
4031 			 */
4032 			if (sc->sc_ipopts)
4033 				(void) m_free(sc->sc_ipopts);
4034 			sc->sc_ipopts = ipopts;
4035 		}
4036 		sc->sc_timestamp = tb.ts_recent;
4037 		if (syn_cache_respond(sc, m) == 0) {
4038 			tcpstat.tcps_sndacks++;
4039 			tcpstat.tcps_sndtotal++;
4040 		}
4041 		return (1);
4042 	}
4043 
4044 	s = splsoftnet();
4045 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4046 	splx(s);
4047 	if (sc == NULL) {
4048 		if (ipopts)
4049 			(void) m_free(ipopts);
4050 		return (0);
4051 	}
4052 
4053 	/*
4054 	 * Fill in the cache, and put the necessary IP and TCP
4055 	 * options into the reply.
4056 	 */
4057 	bzero(sc, sizeof(struct syn_cache));
4058 	callout_init(&sc->sc_timer, 0);
4059 	bcopy(src, &sc->sc_src, src->sa_len);
4060 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4061 	sc->sc_flags = 0;
4062 	sc->sc_ipopts = ipopts;
4063 	sc->sc_irs = th->th_seq;
4064 	switch (src->sa_family) {
4065 #ifdef INET
4066 	case AF_INET:
4067 	    {
4068 		struct sockaddr_in *srcin = (void *) src;
4069 		struct sockaddr_in *dstin = (void *) dst;
4070 
4071 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
4072 		    &srcin->sin_addr, dstin->sin_port,
4073 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
4074 		break;
4075 	    }
4076 #endif /* INET */
4077 #ifdef INET6
4078 	case AF_INET6:
4079 	    {
4080 		struct sockaddr_in6 *srcin6 = (void *) src;
4081 		struct sockaddr_in6 *dstin6 = (void *) dst;
4082 
4083 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
4084 		    &srcin6->sin6_addr, dstin6->sin6_port,
4085 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
4086 		break;
4087 	    }
4088 #endif /* INET6 */
4089 	}
4090 	sc->sc_peermaxseg = oi->maxseg;
4091 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4092 						m->m_pkthdr.rcvif : NULL,
4093 						sc->sc_src.sa.sa_family);
4094 	sc->sc_win = win;
4095 	sc->sc_timebase = tcp_now - 1;	/* see tcp_newtcpcb() */
4096 	sc->sc_timestamp = tb.ts_recent;
4097 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4098 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4099 		sc->sc_flags |= SCF_TIMESTAMP;
4100 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4101 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4102 		sc->sc_requested_s_scale = tb.requested_s_scale;
4103 		sc->sc_request_r_scale = 0;
4104 		/*
4105 		 * Pick the smallest possible scaling factor that
4106 		 * will still allow us to scale up to sb_max.
4107 		 *
4108 		 * We do this because there are broken firewalls that
4109 		 * will corrupt the window scale option, leading to
4110 		 * the other endpoint believing that our advertised
4111 		 * window is unscaled.  At scale factors larger than
4112 		 * 5 the unscaled window will drop below 1500 bytes,
4113 		 * leading to serious problems when traversing these
4114 		 * broken firewalls.
4115 		 *
4116 		 * With the default sbmax of 256K, a scale factor
4117 		 * of 3 will be chosen by this algorithm.  Those who
4118 		 * choose a larger sbmax should watch out
4119 		 * for the compatiblity problems mentioned above.
4120 		 *
4121 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4122 		 * or <SYN,ACK>) segment itself is never scaled.
4123 		 */
4124 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4125 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4126 			sc->sc_request_r_scale++;
4127 	} else {
4128 		sc->sc_requested_s_scale = 15;
4129 		sc->sc_request_r_scale = 15;
4130 	}
4131 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4132 		sc->sc_flags |= SCF_SACK_PERMIT;
4133 
4134 	/*
4135 	 * ECN setup packet recieved.
4136 	 */
4137 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4138 		sc->sc_flags |= SCF_ECN_PERMIT;
4139 
4140 #ifdef TCP_SIGNATURE
4141 	if (tb.t_flags & TF_SIGNATURE)
4142 		sc->sc_flags |= SCF_SIGNATURE;
4143 #endif
4144 	sc->sc_tp = tp;
4145 	if (syn_cache_respond(sc, m) == 0) {
4146 		syn_cache_insert(sc, tp);
4147 		tcpstat.tcps_sndacks++;
4148 		tcpstat.tcps_sndtotal++;
4149 	} else {
4150 		s = splsoftnet();
4151 		syn_cache_put(sc);
4152 		splx(s);
4153 		tcpstat.tcps_sc_dropped++;
4154 	}
4155 	return (1);
4156 }
4157 
4158 int
4159 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4160 {
4161 #ifdef INET6
4162 	struct rtentry *rt;
4163 #endif
4164 	struct route *ro;
4165 	u_int8_t *optp;
4166 	int optlen, error;
4167 	u_int16_t tlen;
4168 	struct ip *ip = NULL;
4169 #ifdef INET6
4170 	struct ip6_hdr *ip6 = NULL;
4171 #endif
4172 	struct tcpcb *tp = NULL;
4173 	struct tcphdr *th;
4174 	u_int hlen;
4175 	struct socket *so;
4176 
4177 	ro = &sc->sc_route;
4178 	switch (sc->sc_src.sa.sa_family) {
4179 	case AF_INET:
4180 		hlen = sizeof(struct ip);
4181 		break;
4182 #ifdef INET6
4183 	case AF_INET6:
4184 		hlen = sizeof(struct ip6_hdr);
4185 		break;
4186 #endif
4187 	default:
4188 		if (m)
4189 			m_freem(m);
4190 		return (EAFNOSUPPORT);
4191 	}
4192 
4193 	/* Compute the size of the TCP options. */
4194 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4195 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4196 #ifdef TCP_SIGNATURE
4197 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4198 #endif
4199 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4200 
4201 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4202 
4203 	/*
4204 	 * Create the IP+TCP header from scratch.
4205 	 */
4206 	if (m)
4207 		m_freem(m);
4208 #ifdef DIAGNOSTIC
4209 	if (max_linkhdr + tlen > MCLBYTES)
4210 		return (ENOBUFS);
4211 #endif
4212 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4213 	if (m && tlen > MHLEN) {
4214 		MCLGET(m, M_DONTWAIT);
4215 		if ((m->m_flags & M_EXT) == 0) {
4216 			m_freem(m);
4217 			m = NULL;
4218 		}
4219 	}
4220 	if (m == NULL)
4221 		return (ENOBUFS);
4222 	MCLAIM(m, &tcp_tx_mowner);
4223 
4224 	/* Fixup the mbuf. */
4225 	m->m_data += max_linkhdr;
4226 	m->m_len = m->m_pkthdr.len = tlen;
4227 	if (sc->sc_tp) {
4228 		tp = sc->sc_tp;
4229 		if (tp->t_inpcb)
4230 			so = tp->t_inpcb->inp_socket;
4231 #ifdef INET6
4232 		else if (tp->t_in6pcb)
4233 			so = tp->t_in6pcb->in6p_socket;
4234 #endif
4235 		else
4236 			so = NULL;
4237 	} else
4238 		so = NULL;
4239 	m->m_pkthdr.rcvif = NULL;
4240 	memset(mtod(m, u_char *), 0, tlen);
4241 
4242 	switch (sc->sc_src.sa.sa_family) {
4243 	case AF_INET:
4244 		ip = mtod(m, struct ip *);
4245 		ip->ip_v = 4;
4246 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4247 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4248 		ip->ip_p = IPPROTO_TCP;
4249 		th = (struct tcphdr *)(ip + 1);
4250 		th->th_dport = sc->sc_src.sin.sin_port;
4251 		th->th_sport = sc->sc_dst.sin.sin_port;
4252 		break;
4253 #ifdef INET6
4254 	case AF_INET6:
4255 		ip6 = mtod(m, struct ip6_hdr *);
4256 		ip6->ip6_vfc = IPV6_VERSION;
4257 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4258 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4259 		ip6->ip6_nxt = IPPROTO_TCP;
4260 		/* ip6_plen will be updated in ip6_output() */
4261 		th = (struct tcphdr *)(ip6 + 1);
4262 		th->th_dport = sc->sc_src.sin6.sin6_port;
4263 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4264 		break;
4265 #endif
4266 	default:
4267 		th = NULL;
4268 	}
4269 
4270 	th->th_seq = htonl(sc->sc_iss);
4271 	th->th_ack = htonl(sc->sc_irs + 1);
4272 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4273 	th->th_flags = TH_SYN|TH_ACK;
4274 	th->th_win = htons(sc->sc_win);
4275 	/* th_sum already 0 */
4276 	/* th_urp already 0 */
4277 
4278 	/* Tack on the TCP options. */
4279 	optp = (u_int8_t *)(th + 1);
4280 	*optp++ = TCPOPT_MAXSEG;
4281 	*optp++ = 4;
4282 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4283 	*optp++ = sc->sc_ourmaxseg & 0xff;
4284 
4285 	if (sc->sc_request_r_scale != 15) {
4286 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4287 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4288 		    sc->sc_request_r_scale);
4289 		optp += 4;
4290 	}
4291 
4292 	if (sc->sc_flags & SCF_TIMESTAMP) {
4293 		u_int32_t *lp = (u_int32_t *)(optp);
4294 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4295 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4296 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4297 		*lp   = htonl(sc->sc_timestamp);
4298 		optp += TCPOLEN_TSTAMP_APPA;
4299 	}
4300 
4301 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4302 		u_int8_t *p = optp;
4303 
4304 		/* Let the peer know that we will SACK. */
4305 		p[0] = TCPOPT_SACK_PERMITTED;
4306 		p[1] = 2;
4307 		p[2] = TCPOPT_NOP;
4308 		p[3] = TCPOPT_NOP;
4309 		optp += 4;
4310 	}
4311 
4312 	/*
4313 	 * Send ECN SYN-ACK setup packet.
4314 	 * Routes can be asymetric, so, even if we receive a packet
4315 	 * with ECE and CWR set, we must not assume no one will block
4316 	 * the ECE packet we are about to send.
4317 	 */
4318 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4319 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4320 		th->th_flags |= TH_ECE;
4321 		tcpstat.tcps_ecn_shs++;
4322 
4323 		/*
4324 		 * draft-ietf-tcpm-ecnsyn-00.txt
4325 		 *
4326 		 * "[...] a TCP node MAY respond to an ECN-setup
4327 		 * SYN packet by setting ECT in the responding
4328 		 * ECN-setup SYN/ACK packet, indicating to routers
4329 		 * that the SYN/ACK packet is ECN-Capable.
4330 		 * This allows a congested router along the path
4331 		 * to mark the packet instead of dropping the
4332 		 * packet as an indication of congestion."
4333 		 *
4334 		 * "[...] There can be a great benefit in setting
4335 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4336 		 * Congestion is  most likely to occur in
4337 		 * the server-to-client direction.  As a result,
4338 		 * setting an ECN-capable codepoint in SYN/ACK
4339 		 * packets can reduce the occurence of three-second
4340 		 * retransmit timeouts resulting from the drop
4341 		 * of SYN/ACK packets."
4342 		 *
4343 		 * Page 4 and 6, January 2006.
4344 		 */
4345 
4346 		switch (sc->sc_src.sa.sa_family) {
4347 #ifdef INET
4348 		case AF_INET:
4349 			ip->ip_tos |= IPTOS_ECN_ECT0;
4350 			break;
4351 #endif
4352 #ifdef INET6
4353 		case AF_INET6:
4354 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4355 			break;
4356 #endif
4357 		}
4358 		tcpstat.tcps_ecn_ect++;
4359 	}
4360 
4361 #ifdef TCP_SIGNATURE
4362 	if (sc->sc_flags & SCF_SIGNATURE) {
4363 		struct secasvar *sav;
4364 		u_int8_t *sigp;
4365 
4366 		sav = tcp_signature_getsav(m, th);
4367 
4368 		if (sav == NULL) {
4369 			if (m)
4370 				m_freem(m);
4371 			return (EPERM);
4372 		}
4373 
4374 		*optp++ = TCPOPT_SIGNATURE;
4375 		*optp++ = TCPOLEN_SIGNATURE;
4376 		sigp = optp;
4377 		bzero(optp, TCP_SIGLEN);
4378 		optp += TCP_SIGLEN;
4379 		*optp++ = TCPOPT_NOP;
4380 		*optp++ = TCPOPT_EOL;
4381 
4382 		(void)tcp_signature(m, th, hlen, sav, sigp);
4383 
4384 		key_sa_recordxfer(sav, m);
4385 #ifdef FAST_IPSEC
4386 		KEY_FREESAV(&sav);
4387 #else
4388 		key_freesav(sav);
4389 #endif
4390 	}
4391 #endif
4392 
4393 	/* Compute the packet's checksum. */
4394 	switch (sc->sc_src.sa.sa_family) {
4395 	case AF_INET:
4396 		ip->ip_len = htons(tlen - hlen);
4397 		th->th_sum = 0;
4398 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4399 		break;
4400 #ifdef INET6
4401 	case AF_INET6:
4402 		ip6->ip6_plen = htons(tlen - hlen);
4403 		th->th_sum = 0;
4404 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4405 		break;
4406 #endif
4407 	}
4408 
4409 	/*
4410 	 * Fill in some straggling IP bits.  Note the stack expects
4411 	 * ip_len to be in host order, for convenience.
4412 	 */
4413 	switch (sc->sc_src.sa.sa_family) {
4414 #ifdef INET
4415 	case AF_INET:
4416 		ip->ip_len = htons(tlen);
4417 		ip->ip_ttl = ip_defttl;
4418 		/* XXX tos? */
4419 		break;
4420 #endif
4421 #ifdef INET6
4422 	case AF_INET6:
4423 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4424 		ip6->ip6_vfc |= IPV6_VERSION;
4425 		ip6->ip6_plen = htons(tlen - hlen);
4426 		/* ip6_hlim will be initialized afterwards */
4427 		/* XXX flowlabel? */
4428 		break;
4429 #endif
4430 	}
4431 
4432 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4433 	tp = sc->sc_tp;
4434 
4435 	switch (sc->sc_src.sa.sa_family) {
4436 #ifdef INET
4437 	case AF_INET:
4438 		error = ip_output(m, sc->sc_ipopts, ro,
4439 		    (ip_mtudisc ? IP_MTUDISC : 0),
4440 		    (struct ip_moptions *)NULL, so);
4441 		break;
4442 #endif
4443 #ifdef INET6
4444 	case AF_INET6:
4445 		ip6->ip6_hlim = in6_selecthlim(NULL,
4446 				(rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
4447 				                                    : NULL);
4448 
4449 		error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL);
4450 		break;
4451 #endif
4452 	default:
4453 		error = EAFNOSUPPORT;
4454 		break;
4455 	}
4456 	return (error);
4457 }
4458