xref: /netbsd-src/sys/netinet/tcp_input.c (revision 10ad5ffa714ce1a679dcc9dd8159648df2d67b5a)
1 /*	$NetBSD: tcp_input.c,v 1.298 2009/07/18 23:09:53 minskim 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  *
93  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
94  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
95  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
96  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
97  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
98  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
99  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
100  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
101  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
102  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
103  * POSSIBILITY OF SUCH DAMAGE.
104  */
105 
106 /*
107  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
108  *	The Regents of the University of California.  All rights reserved.
109  *
110  * Redistribution and use in source and binary forms, with or without
111  * modification, are permitted provided that the following conditions
112  * are met:
113  * 1. Redistributions of source code must retain the above copyright
114  *    notice, this list of conditions and the following disclaimer.
115  * 2. Redistributions in binary form must reproduce the above copyright
116  *    notice, this list of conditions and the following disclaimer in the
117  *    documentation and/or other materials provided with the distribution.
118  * 3. Neither the name of the University nor the names of its contributors
119  *    may be used to endorse or promote products derived from this software
120  *    without specific prior written permission.
121  *
122  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
123  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
124  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
125  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
126  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
127  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
128  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
129  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
130  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
131  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
132  * SUCH DAMAGE.
133  *
134  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
135  */
136 
137 /*
138  *	TODO list for SYN cache stuff:
139  *
140  *	Find room for a "state" field, which is needed to keep a
141  *	compressed state for TIME_WAIT TCBs.  It's been noted already
142  *	that this is fairly important for very high-volume web and
143  *	mail servers, which use a large number of short-lived
144  *	connections.
145  */
146 
147 #include <sys/cdefs.h>
148 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.298 2009/07/18 23:09:53 minskim Exp $");
149 
150 #include "opt_inet.h"
151 #include "opt_ipsec.h"
152 #include "opt_inet_csum.h"
153 #include "opt_tcp_debug.h"
154 
155 #include <sys/param.h>
156 #include <sys/systm.h>
157 #include <sys/malloc.h>
158 #include <sys/mbuf.h>
159 #include <sys/protosw.h>
160 #include <sys/socket.h>
161 #include <sys/socketvar.h>
162 #include <sys/errno.h>
163 #include <sys/syslog.h>
164 #include <sys/pool.h>
165 #include <sys/domain.h>
166 #include <sys/kernel.h>
167 #ifdef TCP_SIGNATURE
168 #include <sys/md5.h>
169 #endif
170 #include <sys/lwp.h> /* for lwp0 */
171 
172 #include <net/if.h>
173 #include <net/route.h>
174 #include <net/if_types.h>
175 
176 #include <netinet/in.h>
177 #include <netinet/in_systm.h>
178 #include <netinet/ip.h>
179 #include <netinet/in_pcb.h>
180 #include <netinet/in_var.h>
181 #include <netinet/ip_var.h>
182 #include <netinet/in_offload.h>
183 
184 #ifdef INET6
185 #ifndef INET
186 #include <netinet/in.h>
187 #endif
188 #include <netinet/ip6.h>
189 #include <netinet6/ip6_var.h>
190 #include <netinet6/in6_pcb.h>
191 #include <netinet6/ip6_var.h>
192 #include <netinet6/in6_var.h>
193 #include <netinet/icmp6.h>
194 #include <netinet6/nd6.h>
195 #ifdef TCP_SIGNATURE
196 #include <netinet6/scope6_var.h>
197 #endif
198 #endif
199 
200 #ifndef INET6
201 /* always need ip6.h for IP6_EXTHDR_GET */
202 #include <netinet/ip6.h>
203 #endif
204 
205 #include <netinet/tcp.h>
206 #include <netinet/tcp_fsm.h>
207 #include <netinet/tcp_seq.h>
208 #include <netinet/tcp_timer.h>
209 #include <netinet/tcp_var.h>
210 #include <netinet/tcp_private.h>
211 #include <netinet/tcpip.h>
212 #include <netinet/tcp_congctl.h>
213 #include <netinet/tcp_debug.h>
214 
215 #include <machine/stdarg.h>
216 
217 #ifdef IPSEC
218 #include <netinet6/ipsec.h>
219 #include <netinet6/ipsec_private.h>
220 #include <netkey/key.h>
221 #endif /*IPSEC*/
222 #ifdef INET6
223 #include "faith.h"
224 #if defined(NFAITH) && NFAITH > 0
225 #include <net/if_faith.h>
226 #endif
227 #endif	/* IPSEC */
228 
229 #ifdef FAST_IPSEC
230 #include <netipsec/ipsec.h>
231 #include <netipsec/ipsec_var.h>
232 #include <netipsec/ipsec_private.h>
233 #include <netipsec/key.h>
234 #ifdef INET6
235 #include <netipsec/ipsec6.h>
236 #endif
237 #endif	/* FAST_IPSEC*/
238 
239 int	tcprexmtthresh = 3;
240 int	tcp_log_refused;
241 
242 int	tcp_do_autorcvbuf = 0;
243 int	tcp_autorcvbuf_inc = 16 * 1024;
244 int	tcp_autorcvbuf_max = 256 * 1024;
245 
246 static int tcp_rst_ppslim_count = 0;
247 static struct timeval tcp_rst_ppslim_last;
248 static int tcp_ackdrop_ppslim_count = 0;
249 static struct timeval tcp_ackdrop_ppslim_last;
250 
251 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
252 
253 /* for modulo comparisons of timestamps */
254 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
255 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
256 
257 /*
258  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
259  */
260 #ifdef INET6
261 static inline void
262 nd6_hint(struct tcpcb *tp)
263 {
264 	struct rtentry *rt;
265 
266 	if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 &&
267 	    (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL)
268 		nd6_nud_hint(rt, NULL, 0);
269 }
270 #else
271 static inline void
272 nd6_hint(struct tcpcb *tp)
273 {
274 }
275 #endif
276 
277 /*
278  * Compute ACK transmission behavior.  Delay the ACK unless
279  * we have already delayed an ACK (must send an ACK every two segments).
280  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
281  * option is enabled.
282  */
283 static void
284 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th)
285 {
286 
287 	if (tp->t_flags & TF_DELACK ||
288 	    (tcp_ack_on_push && th->th_flags & TH_PUSH))
289 		tp->t_flags |= TF_ACKNOW;
290 	else
291 		TCP_SET_DELACK(tp);
292 }
293 
294 static void
295 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked)
296 {
297 
298 	/*
299 	 * If we had a pending ICMP message that refers to data that have
300 	 * just been acknowledged, disregard the recorded ICMP message.
301 	 */
302 	if ((tp->t_flags & TF_PMTUD_PEND) &&
303 	    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
304 		tp->t_flags &= ~TF_PMTUD_PEND;
305 
306 	/*
307 	 * Keep track of the largest chunk of data
308 	 * acknowledged since last PMTU update
309 	 */
310 	if (tp->t_pmtud_mss_acked < acked)
311 		tp->t_pmtud_mss_acked = acked;
312 }
313 
314 /*
315  * Convert TCP protocol fields to host order for easier processing.
316  */
317 static void
318 tcp_fields_to_host(struct tcphdr *th)
319 {
320 
321 	NTOHL(th->th_seq);
322 	NTOHL(th->th_ack);
323 	NTOHS(th->th_win);
324 	NTOHS(th->th_urp);
325 }
326 
327 /*
328  * ... and reverse the above.
329  */
330 static void
331 tcp_fields_to_net(struct tcphdr *th)
332 {
333 
334 	HTONL(th->th_seq);
335 	HTONL(th->th_ack);
336 	HTONS(th->th_win);
337 	HTONS(th->th_urp);
338 }
339 
340 #ifdef TCP_CSUM_COUNTERS
341 #include <sys/device.h>
342 
343 #if defined(INET)
344 extern struct evcnt tcp_hwcsum_ok;
345 extern struct evcnt tcp_hwcsum_bad;
346 extern struct evcnt tcp_hwcsum_data;
347 extern struct evcnt tcp_swcsum;
348 #endif /* defined(INET) */
349 #if defined(INET6)
350 extern struct evcnt tcp6_hwcsum_ok;
351 extern struct evcnt tcp6_hwcsum_bad;
352 extern struct evcnt tcp6_hwcsum_data;
353 extern struct evcnt tcp6_swcsum;
354 #endif /* defined(INET6) */
355 
356 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
357 
358 #else
359 
360 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
361 
362 #endif /* TCP_CSUM_COUNTERS */
363 
364 #ifdef TCP_REASS_COUNTERS
365 #include <sys/device.h>
366 
367 extern struct evcnt tcp_reass_;
368 extern struct evcnt tcp_reass_empty;
369 extern struct evcnt tcp_reass_iteration[8];
370 extern struct evcnt tcp_reass_prependfirst;
371 extern struct evcnt tcp_reass_prepend;
372 extern struct evcnt tcp_reass_insert;
373 extern struct evcnt tcp_reass_inserttail;
374 extern struct evcnt tcp_reass_append;
375 extern struct evcnt tcp_reass_appendtail;
376 extern struct evcnt tcp_reass_overlaptail;
377 extern struct evcnt tcp_reass_overlapfront;
378 extern struct evcnt tcp_reass_segdup;
379 extern struct evcnt tcp_reass_fragdup;
380 
381 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
382 
383 #else
384 
385 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
386 
387 #endif /* TCP_REASS_COUNTERS */
388 
389 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *,
390     int *);
391 static int tcp_dooptions(struct tcpcb *, const u_char *, int,
392     struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *);
393 
394 #ifdef INET
395 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
396 #endif
397 #ifdef INET6
398 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
399 #endif
400 
401 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
402 
403 #if defined(MBUFTRACE)
404 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass");
405 #endif /* defined(MBUFTRACE) */
406 
407 static struct pool tcpipqent_pool;
408 
409 void
410 tcpipqent_init(void)
411 {
412 
413 	pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl",
414 	    NULL, IPL_VM);
415 }
416 
417 struct ipqent *
418 tcpipqent_alloc(void)
419 {
420 	struct ipqent *ipqe;
421 	int s;
422 
423 	s = splvm();
424 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
425 	splx(s);
426 
427 	return ipqe;
428 }
429 
430 void
431 tcpipqent_free(struct ipqent *ipqe)
432 {
433 	int s;
434 
435 	s = splvm();
436 	pool_put(&tcpipqent_pool, ipqe);
437 	splx(s);
438 }
439 
440 static int
441 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen)
442 {
443 	struct ipqent *p, *q, *nq, *tiqe = NULL;
444 	struct socket *so = NULL;
445 	int pkt_flags;
446 	tcp_seq pkt_seq;
447 	unsigned pkt_len;
448 	u_long rcvpartdupbyte = 0;
449 	u_long rcvoobyte;
450 #ifdef TCP_REASS_COUNTERS
451 	u_int count = 0;
452 #endif
453 	uint64_t *tcps;
454 
455 	if (tp->t_inpcb)
456 		so = tp->t_inpcb->inp_socket;
457 #ifdef INET6
458 	else if (tp->t_in6pcb)
459 		so = tp->t_in6pcb->in6p_socket;
460 #endif
461 
462 	TCP_REASS_LOCK_CHECK(tp);
463 
464 	/*
465 	 * Call with th==0 after become established to
466 	 * force pre-ESTABLISHED data up to user socket.
467 	 */
468 	if (th == 0)
469 		goto present;
470 
471 	m_claimm(m, &tcp_reass_mowner);
472 
473 	rcvoobyte = *tlen;
474 	/*
475 	 * Copy these to local variables because the tcpiphdr
476 	 * gets munged while we are collapsing mbufs.
477 	 */
478 	pkt_seq = th->th_seq;
479 	pkt_len = *tlen;
480 	pkt_flags = th->th_flags;
481 
482 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
483 
484 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
485 		/*
486 		 * When we miss a packet, the vast majority of time we get
487 		 * packets that follow it in order.  So optimize for that.
488 		 */
489 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
490 			p->ipqe_len += pkt_len;
491 			p->ipqe_flags |= pkt_flags;
492 			m_cat(p->ipre_mlast, m);
493 			TRAVERSE(p->ipre_mlast);
494 			m = NULL;
495 			tiqe = p;
496 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
497 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
498 			goto skip_replacement;
499 		}
500 		/*
501 		 * While we're here, if the pkt is completely beyond
502 		 * anything we have, just insert it at the tail.
503 		 */
504 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
505 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
506 			goto insert_it;
507 		}
508 	}
509 
510 	q = TAILQ_FIRST(&tp->segq);
511 
512 	if (q != NULL) {
513 		/*
514 		 * If this segment immediately precedes the first out-of-order
515 		 * block, simply slap the segment in front of it and (mostly)
516 		 * skip the complicated logic.
517 		 */
518 		if (pkt_seq + pkt_len == q->ipqe_seq) {
519 			q->ipqe_seq = pkt_seq;
520 			q->ipqe_len += pkt_len;
521 			q->ipqe_flags |= pkt_flags;
522 			m_cat(m, q->ipqe_m);
523 			q->ipqe_m = m;
524 			q->ipre_mlast = m; /* last mbuf may have changed */
525 			TRAVERSE(q->ipre_mlast);
526 			tiqe = q;
527 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
528 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
529 			goto skip_replacement;
530 		}
531 	} else {
532 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
533 	}
534 
535 	/*
536 	 * Find a segment which begins after this one does.
537 	 */
538 	for (p = NULL; q != NULL; q = nq) {
539 		nq = TAILQ_NEXT(q, ipqe_q);
540 #ifdef TCP_REASS_COUNTERS
541 		count++;
542 #endif
543 		/*
544 		 * If the received segment is just right after this
545 		 * fragment, merge the two together and then check
546 		 * for further overlaps.
547 		 */
548 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
549 #ifdef TCPREASS_DEBUG
550 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
551 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
552 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
553 #endif
554 			pkt_len += q->ipqe_len;
555 			pkt_flags |= q->ipqe_flags;
556 			pkt_seq = q->ipqe_seq;
557 			m_cat(q->ipre_mlast, m);
558 			TRAVERSE(q->ipre_mlast);
559 			m = q->ipqe_m;
560 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
561 			goto free_ipqe;
562 		}
563 		/*
564 		 * If the received segment is completely past this
565 		 * fragment, we need to go the next fragment.
566 		 */
567 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
568 			p = q;
569 			continue;
570 		}
571 		/*
572 		 * If the fragment is past the received segment,
573 		 * it (or any following) can't be concatenated.
574 		 */
575 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
576 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
577 			break;
578 		}
579 
580 		/*
581 		 * We've received all the data in this segment before.
582 		 * mark it as a duplicate and return.
583 		 */
584 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
585 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
586 			tcps = TCP_STAT_GETREF();
587 			tcps[TCP_STAT_RCVDUPPACK]++;
588 			tcps[TCP_STAT_RCVDUPBYTE] += pkt_len;
589 			TCP_STAT_PUTREF();
590 			tcp_new_dsack(tp, pkt_seq, pkt_len);
591 			m_freem(m);
592 			if (tiqe != NULL) {
593 				tcpipqent_free(tiqe);
594 			}
595 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
596 			return (0);
597 		}
598 		/*
599 		 * Received segment completely overlaps this fragment
600 		 * so we drop the fragment (this keeps the temporal
601 		 * ordering of segments correct).
602 		 */
603 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
604 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
605 			rcvpartdupbyte += q->ipqe_len;
606 			m_freem(q->ipqe_m);
607 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
608 			goto free_ipqe;
609 		}
610 		/*
611 		 * RX'ed segment extends past the end of the
612 		 * fragment.  Drop the overlapping bytes.  Then
613 		 * merge the fragment and segment then treat as
614 		 * a longer received packet.
615 		 */
616 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
617 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
618 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
619 #ifdef TCPREASS_DEBUG
620 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
621 			       tp, overlap,
622 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
623 #endif
624 			m_adj(m, overlap);
625 			rcvpartdupbyte += overlap;
626 			m_cat(q->ipre_mlast, m);
627 			TRAVERSE(q->ipre_mlast);
628 			m = q->ipqe_m;
629 			pkt_seq = q->ipqe_seq;
630 			pkt_len += q->ipqe_len - overlap;
631 			rcvoobyte -= overlap;
632 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
633 			goto free_ipqe;
634 		}
635 		/*
636 		 * RX'ed segment extends past the front of the
637 		 * fragment.  Drop the overlapping bytes on the
638 		 * received packet.  The packet will then be
639 		 * contatentated with this fragment a bit later.
640 		 */
641 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
642 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
643 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
644 #ifdef TCPREASS_DEBUG
645 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
646 			       tp, overlap,
647 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
648 #endif
649 			m_adj(m, -overlap);
650 			pkt_len -= overlap;
651 			rcvpartdupbyte += overlap;
652 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
653 			rcvoobyte -= overlap;
654 		}
655 		/*
656 		 * If the received segment immediates precedes this
657 		 * fragment then tack the fragment onto this segment
658 		 * and reinsert the data.
659 		 */
660 		if (q->ipqe_seq == pkt_seq + pkt_len) {
661 #ifdef TCPREASS_DEBUG
662 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
663 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
664 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
665 #endif
666 			pkt_len += q->ipqe_len;
667 			pkt_flags |= q->ipqe_flags;
668 			m_cat(m, q->ipqe_m);
669 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
670 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
671 			tp->t_segqlen--;
672 			KASSERT(tp->t_segqlen >= 0);
673 			KASSERT(tp->t_segqlen != 0 ||
674 			    (TAILQ_EMPTY(&tp->segq) &&
675 			    TAILQ_EMPTY(&tp->timeq)));
676 			if (tiqe == NULL) {
677 				tiqe = q;
678 			} else {
679 				tcpipqent_free(q);
680 			}
681 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
682 			break;
683 		}
684 		/*
685 		 * If the fragment is before the segment, remember it.
686 		 * When this loop is terminated, p will contain the
687 		 * pointer to fragment that is right before the received
688 		 * segment.
689 		 */
690 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
691 			p = q;
692 
693 		continue;
694 
695 		/*
696 		 * This is a common operation.  It also will allow
697 		 * to save doing a malloc/free in most instances.
698 		 */
699 	  free_ipqe:
700 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
701 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
702 		tp->t_segqlen--;
703 		KASSERT(tp->t_segqlen >= 0);
704 		KASSERT(tp->t_segqlen != 0 ||
705 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
706 		if (tiqe == NULL) {
707 			tiqe = q;
708 		} else {
709 			tcpipqent_free(q);
710 		}
711 	}
712 
713 #ifdef TCP_REASS_COUNTERS
714 	if (count > 7)
715 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
716 	else if (count > 0)
717 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
718 #endif
719 
720     insert_it:
721 
722 	/*
723 	 * Allocate a new queue entry since the received segment did not
724 	 * collapse onto any other out-of-order block; thus we are allocating
725 	 * a new block.  If it had collapsed, tiqe would not be NULL and
726 	 * we would be reusing it.
727 	 * XXX If we can't, just drop the packet.  XXX
728 	 */
729 	if (tiqe == NULL) {
730 		tiqe = tcpipqent_alloc();
731 		if (tiqe == NULL) {
732 			TCP_STATINC(TCP_STAT_RCVMEMDROP);
733 			m_freem(m);
734 			return (0);
735 		}
736 	}
737 
738 	/*
739 	 * Update the counters.
740 	 */
741 	tcps = TCP_STAT_GETREF();
742 	tcps[TCP_STAT_RCVOOPACK]++;
743 	tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte;
744 	if (rcvpartdupbyte) {
745 	    tcps[TCP_STAT_RCVPARTDUPPACK]++;
746 	    tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte;
747 	}
748 	TCP_STAT_PUTREF();
749 
750 	/*
751 	 * Insert the new fragment queue entry into both queues.
752 	 */
753 	tiqe->ipqe_m = m;
754 	tiqe->ipre_mlast = m;
755 	tiqe->ipqe_seq = pkt_seq;
756 	tiqe->ipqe_len = pkt_len;
757 	tiqe->ipqe_flags = pkt_flags;
758 	if (p == NULL) {
759 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
760 #ifdef TCPREASS_DEBUG
761 		if (tiqe->ipqe_seq != tp->rcv_nxt)
762 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
763 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
764 #endif
765 	} else {
766 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
767 #ifdef TCPREASS_DEBUG
768 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
769 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
770 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
771 #endif
772 	}
773 	tp->t_segqlen++;
774 
775 skip_replacement:
776 
777 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
778 
779 present:
780 	/*
781 	 * Present data to user, advancing rcv_nxt through
782 	 * completed sequence space.
783 	 */
784 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
785 		return (0);
786 	q = TAILQ_FIRST(&tp->segq);
787 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
788 		return (0);
789 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
790 		return (0);
791 
792 	tp->rcv_nxt += q->ipqe_len;
793 	pkt_flags = q->ipqe_flags & TH_FIN;
794 	nd6_hint(tp);
795 
796 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
797 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
798 	tp->t_segqlen--;
799 	KASSERT(tp->t_segqlen >= 0);
800 	KASSERT(tp->t_segqlen != 0 ||
801 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
802 	if (so->so_state & SS_CANTRCVMORE)
803 		m_freem(q->ipqe_m);
804 	else
805 		sbappendstream(&so->so_rcv, q->ipqe_m);
806 	tcpipqent_free(q);
807 	sorwakeup(so);
808 	return (pkt_flags);
809 }
810 
811 #ifdef INET6
812 int
813 tcp6_input(struct mbuf **mp, int *offp, int proto)
814 {
815 	struct mbuf *m = *mp;
816 
817 	/*
818 	 * draft-itojun-ipv6-tcp-to-anycast
819 	 * better place to put this in?
820 	 */
821 	if (m->m_flags & M_ANYCAST6) {
822 		struct ip6_hdr *ip6;
823 		if (m->m_len < sizeof(struct ip6_hdr)) {
824 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
825 				TCP_STATINC(TCP_STAT_RCVSHORT);
826 				return IPPROTO_DONE;
827 			}
828 		}
829 		ip6 = mtod(m, struct ip6_hdr *);
830 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
831 		    (char *)&ip6->ip6_dst - (char *)ip6);
832 		return IPPROTO_DONE;
833 	}
834 
835 	tcp_input(m, *offp, proto);
836 	return IPPROTO_DONE;
837 }
838 #endif
839 
840 #ifdef INET
841 static void
842 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
843 {
844 	char src[4*sizeof "123"];
845 	char dst[4*sizeof "123"];
846 
847 	if (ip) {
848 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
849 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
850 	}
851 	else {
852 		strlcpy(src, "(unknown)", sizeof(src));
853 		strlcpy(dst, "(unknown)", sizeof(dst));
854 	}
855 	log(LOG_INFO,
856 	    "Connection attempt to TCP %s:%d from %s:%d\n",
857 	    dst, ntohs(th->th_dport),
858 	    src, ntohs(th->th_sport));
859 }
860 #endif
861 
862 #ifdef INET6
863 static void
864 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
865 {
866 	char src[INET6_ADDRSTRLEN];
867 	char dst[INET6_ADDRSTRLEN];
868 
869 	if (ip6) {
870 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
871 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
872 	}
873 	else {
874 		strlcpy(src, "(unknown v6)", sizeof(src));
875 		strlcpy(dst, "(unknown v6)", sizeof(dst));
876 	}
877 	log(LOG_INFO,
878 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
879 	    dst, ntohs(th->th_dport),
880 	    src, ntohs(th->th_sport));
881 }
882 #endif
883 
884 /*
885  * Checksum extended TCP header and data.
886  */
887 int
888 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th,
889     int toff, int off, int tlen)
890 {
891 
892 	/*
893 	 * XXX it's better to record and check if this mbuf is
894 	 * already checked.
895 	 */
896 
897 	switch (af) {
898 #ifdef INET
899 	case AF_INET:
900 		switch (m->m_pkthdr.csum_flags &
901 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
902 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
903 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
904 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
905 			goto badcsum;
906 
907 		case M_CSUM_TCPv4|M_CSUM_DATA: {
908 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
909 
910 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
911 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
912 				const struct ip *ip =
913 				    mtod(m, const struct ip *);
914 
915 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
916 				    ip->ip_dst.s_addr,
917 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
918 			}
919 			if ((hw_csum ^ 0xffff) != 0)
920 				goto badcsum;
921 			break;
922 		}
923 
924 		case M_CSUM_TCPv4:
925 			/* Checksum was okay. */
926 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
927 			break;
928 
929 		default:
930 			/*
931 			 * Must compute it ourselves.  Maybe skip checksum
932 			 * on loopback interfaces.
933 			 */
934 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
935 					     IFF_LOOPBACK) ||
936 					   tcp_do_loopback_cksum)) {
937 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
938 				if (in4_cksum(m, IPPROTO_TCP, toff,
939 					      tlen + off) != 0)
940 					goto badcsum;
941 			}
942 			break;
943 		}
944 		break;
945 #endif /* INET4 */
946 
947 #ifdef INET6
948 	case AF_INET6:
949 		switch (m->m_pkthdr.csum_flags &
950 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
951 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
952 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
953 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
954 			goto badcsum;
955 
956 #if 0 /* notyet */
957 		case M_CSUM_TCPv6|M_CSUM_DATA:
958 #endif
959 
960 		case M_CSUM_TCPv6:
961 			/* Checksum was okay. */
962 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok);
963 			break;
964 
965 		default:
966 			/*
967 			 * Must compute it ourselves.  Maybe skip checksum
968 			 * on loopback interfaces.
969 			 */
970 			if (__predict_true((m->m_flags & M_LOOP) == 0 ||
971 			    tcp_do_loopback_cksum)) {
972 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
973 				if (in6_cksum(m, IPPROTO_TCP, toff,
974 				    tlen + off) != 0)
975 					goto badcsum;
976 			}
977 		}
978 		break;
979 #endif /* INET6 */
980 	}
981 
982 	return 0;
983 
984 badcsum:
985 	TCP_STATINC(TCP_STAT_RCVBADSUM);
986 	return -1;
987 }
988 
989 /*
990  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
991  */
992 void
993 tcp_input(struct mbuf *m, ...)
994 {
995 	struct tcphdr *th;
996 	struct ip *ip;
997 	struct inpcb *inp;
998 #ifdef INET6
999 	struct ip6_hdr *ip6;
1000 	struct in6pcb *in6p;
1001 #endif
1002 	u_int8_t *optp = NULL;
1003 	int optlen = 0;
1004 	int len, tlen, toff, hdroptlen = 0;
1005 	struct tcpcb *tp = 0;
1006 	int tiflags;
1007 	struct socket *so = NULL;
1008 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
1009 #ifdef TCP_DEBUG
1010 	short ostate = 0;
1011 #endif
1012 	u_long tiwin;
1013 	struct tcp_opt_info opti;
1014 	int off, iphlen;
1015 	va_list ap;
1016 	int af;		/* af on the wire */
1017 	struct mbuf *tcp_saveti = NULL;
1018 	uint32_t ts_rtt;
1019 	uint8_t iptos;
1020 	uint64_t *tcps;
1021 
1022 	MCLAIM(m, &tcp_rx_mowner);
1023 	va_start(ap, m);
1024 	toff = va_arg(ap, int);
1025 	(void)va_arg(ap, int);		/* ignore value, advance ap */
1026 	va_end(ap);
1027 
1028 	TCP_STATINC(TCP_STAT_RCVTOTAL);
1029 
1030 	memset(&opti, 0, sizeof(opti));
1031 	opti.ts_present = 0;
1032 	opti.maxseg = 0;
1033 
1034 	/*
1035 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
1036 	 *
1037 	 * TCP is, by definition, unicast, so we reject all
1038 	 * multicast outright.
1039 	 *
1040 	 * Note, there are additional src/dst address checks in
1041 	 * the AF-specific code below.
1042 	 */
1043 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1044 		/* XXX stat */
1045 		goto drop;
1046 	}
1047 #ifdef INET6
1048 	if (m->m_flags & M_ANYCAST6) {
1049 		/* XXX stat */
1050 		goto drop;
1051 	}
1052 #endif
1053 
1054 	/*
1055 	 * Get IP and TCP header.
1056 	 * Note: IP leaves IP header in first mbuf.
1057 	 */
1058 	ip = mtod(m, struct ip *);
1059 #ifdef INET6
1060 	ip6 = NULL;
1061 #endif
1062 	switch (ip->ip_v) {
1063 #ifdef INET
1064 	case 4:
1065 		af = AF_INET;
1066 		iphlen = sizeof(struct ip);
1067 		ip = mtod(m, struct ip *);
1068 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1069 			sizeof(struct tcphdr));
1070 		if (th == NULL) {
1071 			TCP_STATINC(TCP_STAT_RCVSHORT);
1072 			return;
1073 		}
1074 		/* We do the checksum after PCB lookup... */
1075 		len = ntohs(ip->ip_len);
1076 		tlen = len - toff;
1077 		iptos = ip->ip_tos;
1078 		break;
1079 #endif
1080 #ifdef INET6
1081 	case 6:
1082 		ip = NULL;
1083 		iphlen = sizeof(struct ip6_hdr);
1084 		af = AF_INET6;
1085 		ip6 = mtod(m, struct ip6_hdr *);
1086 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1087 			sizeof(struct tcphdr));
1088 		if (th == NULL) {
1089 			TCP_STATINC(TCP_STAT_RCVSHORT);
1090 			return;
1091 		}
1092 
1093 		/* Be proactive about malicious use of IPv4 mapped address */
1094 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1095 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1096 			/* XXX stat */
1097 			goto drop;
1098 		}
1099 
1100 		/*
1101 		 * Be proactive about unspecified IPv6 address in source.
1102 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1103 		 * unspecified IPv6 address can be used to confuse us.
1104 		 *
1105 		 * Note that packets with unspecified IPv6 destination is
1106 		 * already dropped in ip6_input.
1107 		 */
1108 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1109 			/* XXX stat */
1110 			goto drop;
1111 		}
1112 
1113 		/*
1114 		 * Make sure destination address is not multicast.
1115 		 * Source address checked in ip6_input().
1116 		 */
1117 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1118 			/* XXX stat */
1119 			goto drop;
1120 		}
1121 
1122 		/* We do the checksum after PCB lookup... */
1123 		len = m->m_pkthdr.len;
1124 		tlen = len - toff;
1125 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1126 		break;
1127 #endif
1128 	default:
1129 		m_freem(m);
1130 		return;
1131 	}
1132 
1133 	KASSERT(TCP_HDR_ALIGNED_P(th));
1134 
1135 	/*
1136 	 * Check that TCP offset makes sense,
1137 	 * pull out TCP options and adjust length.		XXX
1138 	 */
1139 	off = th->th_off << 2;
1140 	if (off < sizeof (struct tcphdr) || off > tlen) {
1141 		TCP_STATINC(TCP_STAT_RCVBADOFF);
1142 		goto drop;
1143 	}
1144 	tlen -= off;
1145 
1146 	/*
1147 	 * tcp_input() has been modified to use tlen to mean the TCP data
1148 	 * length throughout the function.  Other functions can use
1149 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1150 	 * rja
1151 	 */
1152 
1153 	if (off > sizeof (struct tcphdr)) {
1154 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1155 		if (th == NULL) {
1156 			TCP_STATINC(TCP_STAT_RCVSHORT);
1157 			return;
1158 		}
1159 		/*
1160 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1161 		 * (as they're before toff) and we don't need to update those.
1162 		 */
1163 		KASSERT(TCP_HDR_ALIGNED_P(th));
1164 		optlen = off - sizeof (struct tcphdr);
1165 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1166 		/*
1167 		 * Do quick retrieval of timestamp options ("options
1168 		 * prediction?").  If timestamp is the only option and it's
1169 		 * formatted as recommended in RFC 1323 appendix A, we
1170 		 * quickly get the values now and not bother calling
1171 		 * tcp_dooptions(), etc.
1172 		 */
1173 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1174 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1175 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1176 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1177 		     (th->th_flags & TH_SYN) == 0) {
1178 			opti.ts_present = 1;
1179 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1180 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1181 			optp = NULL;	/* we've parsed the options */
1182 		}
1183 	}
1184 	tiflags = th->th_flags;
1185 
1186 	/*
1187 	 * Locate pcb for segment.
1188 	 */
1189 findpcb:
1190 	inp = NULL;
1191 #ifdef INET6
1192 	in6p = NULL;
1193 #endif
1194 	switch (af) {
1195 #ifdef INET
1196 	case AF_INET:
1197 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1198 		    ip->ip_dst, th->th_dport);
1199 		if (inp == 0) {
1200 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1201 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1202 		}
1203 #ifdef INET6
1204 		if (inp == 0) {
1205 			struct in6_addr s, d;
1206 
1207 			/* mapped addr case */
1208 			memset(&s, 0, sizeof(s));
1209 			s.s6_addr16[5] = htons(0xffff);
1210 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1211 			memset(&d, 0, sizeof(d));
1212 			d.s6_addr16[5] = htons(0xffff);
1213 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1214 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1215 			    th->th_sport, &d, th->th_dport, 0);
1216 			if (in6p == 0) {
1217 				TCP_STATINC(TCP_STAT_PCBHASHMISS);
1218 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1219 				    th->th_dport, 0);
1220 			}
1221 		}
1222 #endif
1223 #ifndef INET6
1224 		if (inp == 0)
1225 #else
1226 		if (inp == 0 && in6p == 0)
1227 #endif
1228 		{
1229 			TCP_STATINC(TCP_STAT_NOPORT);
1230 			if (tcp_log_refused &&
1231 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1232 				tcp4_log_refused(ip, th);
1233 			}
1234 			tcp_fields_to_host(th);
1235 			goto dropwithreset_ratelim;
1236 		}
1237 #if defined(IPSEC) || defined(FAST_IPSEC)
1238 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1239 		    ipsec4_in_reject(m, inp)) {
1240 			IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1241 			goto drop;
1242 		}
1243 #ifdef INET6
1244 		else if (in6p &&
1245 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1246 		    ipsec6_in_reject_so(m, in6p->in6p_socket)) {
1247 			IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1248 			goto drop;
1249 		}
1250 #endif
1251 #endif /*IPSEC*/
1252 		break;
1253 #endif /*INET*/
1254 #ifdef INET6
1255 	case AF_INET6:
1256 	    {
1257 		int faith;
1258 
1259 #if defined(NFAITH) && NFAITH > 0
1260 		faith = faithprefix(&ip6->ip6_dst);
1261 #else
1262 		faith = 0;
1263 #endif
1264 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1265 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1266 		if (in6p == NULL) {
1267 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1268 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1269 				th->th_dport, faith);
1270 		}
1271 		if (in6p == NULL) {
1272 			TCP_STATINC(TCP_STAT_NOPORT);
1273 			if (tcp_log_refused &&
1274 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1275 				tcp6_log_refused(ip6, th);
1276 			}
1277 			tcp_fields_to_host(th);
1278 			goto dropwithreset_ratelim;
1279 		}
1280 #if defined(IPSEC) || defined(FAST_IPSEC)
1281 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1282 		    ipsec6_in_reject(m, in6p)) {
1283 			IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO);
1284 			goto drop;
1285 		}
1286 #endif /*IPSEC*/
1287 		break;
1288 	    }
1289 #endif
1290 	}
1291 
1292 	/*
1293 	 * If the state is CLOSED (i.e., TCB does not exist) then
1294 	 * all data in the incoming segment is discarded.
1295 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1296 	 * but should either do a listen or a connect soon.
1297 	 */
1298 	tp = NULL;
1299 	so = NULL;
1300 	if (inp) {
1301 		/* Check the minimum TTL for socket. */
1302 		if (ip->ip_ttl < inp->inp_ip_minttl)
1303 			goto drop;
1304 
1305 		tp = intotcpcb(inp);
1306 		so = inp->inp_socket;
1307 	}
1308 #ifdef INET6
1309 	else if (in6p) {
1310 		tp = in6totcpcb(in6p);
1311 		so = in6p->in6p_socket;
1312 	}
1313 #endif
1314 	if (tp == 0) {
1315 		tcp_fields_to_host(th);
1316 		goto dropwithreset_ratelim;
1317 	}
1318 	if (tp->t_state == TCPS_CLOSED)
1319 		goto drop;
1320 
1321 	KASSERT(so->so_lock == softnet_lock);
1322 	KASSERT(solocked(so));
1323 
1324 	/*
1325 	 * Checksum extended TCP header and data.
1326 	 */
1327 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1328 		goto badcsum;
1329 
1330 	tcp_fields_to_host(th);
1331 
1332 	/* Unscale the window into a 32-bit value. */
1333 	if ((tiflags & TH_SYN) == 0)
1334 		tiwin = th->th_win << tp->snd_scale;
1335 	else
1336 		tiwin = th->th_win;
1337 
1338 #ifdef INET6
1339 	/* save packet options if user wanted */
1340 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1341 		if (in6p->in6p_options) {
1342 			m_freem(in6p->in6p_options);
1343 			in6p->in6p_options = 0;
1344 		}
1345 		KASSERT(ip6 != NULL);
1346 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1347 	}
1348 #endif
1349 
1350 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1351 		union syn_cache_sa src;
1352 		union syn_cache_sa dst;
1353 
1354 		memset(&src, 0, sizeof(src));
1355 		memset(&dst, 0, sizeof(dst));
1356 		switch (af) {
1357 #ifdef INET
1358 		case AF_INET:
1359 			src.sin.sin_len = sizeof(struct sockaddr_in);
1360 			src.sin.sin_family = AF_INET;
1361 			src.sin.sin_addr = ip->ip_src;
1362 			src.sin.sin_port = th->th_sport;
1363 
1364 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1365 			dst.sin.sin_family = AF_INET;
1366 			dst.sin.sin_addr = ip->ip_dst;
1367 			dst.sin.sin_port = th->th_dport;
1368 			break;
1369 #endif
1370 #ifdef INET6
1371 		case AF_INET6:
1372 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1373 			src.sin6.sin6_family = AF_INET6;
1374 			src.sin6.sin6_addr = ip6->ip6_src;
1375 			src.sin6.sin6_port = th->th_sport;
1376 
1377 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1378 			dst.sin6.sin6_family = AF_INET6;
1379 			dst.sin6.sin6_addr = ip6->ip6_dst;
1380 			dst.sin6.sin6_port = th->th_dport;
1381 			break;
1382 #endif /* INET6 */
1383 		default:
1384 			goto badsyn;	/*sanity*/
1385 		}
1386 
1387 		if (so->so_options & SO_DEBUG) {
1388 #ifdef TCP_DEBUG
1389 			ostate = tp->t_state;
1390 #endif
1391 
1392 			tcp_saveti = NULL;
1393 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1394 				goto nosave;
1395 
1396 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1397 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1398 				if (!tcp_saveti)
1399 					goto nosave;
1400 			} else {
1401 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1402 				if (!tcp_saveti)
1403 					goto nosave;
1404 				MCLAIM(m, &tcp_mowner);
1405 				tcp_saveti->m_len = iphlen;
1406 				m_copydata(m, 0, iphlen,
1407 				    mtod(tcp_saveti, void *));
1408 			}
1409 
1410 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1411 				m_freem(tcp_saveti);
1412 				tcp_saveti = NULL;
1413 			} else {
1414 				tcp_saveti->m_len += sizeof(struct tcphdr);
1415 				memcpy(mtod(tcp_saveti, char *) + iphlen, th,
1416 				    sizeof(struct tcphdr));
1417 			}
1418 	nosave:;
1419 		}
1420 		if (so->so_options & SO_ACCEPTCONN) {
1421 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1422 				if (tiflags & TH_RST) {
1423 					syn_cache_reset(&src.sa, &dst.sa, th);
1424 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1425 				    (TH_ACK|TH_SYN)) {
1426 					/*
1427 					 * Received a SYN,ACK.  This should
1428 					 * never happen while we are in
1429 					 * LISTEN.  Send an RST.
1430 					 */
1431 					goto badsyn;
1432 				} else if (tiflags & TH_ACK) {
1433 					so = syn_cache_get(&src.sa, &dst.sa,
1434 						th, toff, tlen, so, m);
1435 					if (so == NULL) {
1436 						/*
1437 						 * We don't have a SYN for
1438 						 * this ACK; send an RST.
1439 						 */
1440 						goto badsyn;
1441 					} else if (so ==
1442 					    (struct socket *)(-1)) {
1443 						/*
1444 						 * We were unable to create
1445 						 * the connection.  If the
1446 						 * 3-way handshake was
1447 						 * completed, and RST has
1448 						 * been sent to the peer.
1449 						 * Since the mbuf might be
1450 						 * in use for the reply,
1451 						 * do not free it.
1452 						 */
1453 						m = NULL;
1454 					} else {
1455 						/*
1456 						 * We have created a
1457 						 * full-blown connection.
1458 						 */
1459 						tp = NULL;
1460 						inp = NULL;
1461 #ifdef INET6
1462 						in6p = NULL;
1463 #endif
1464 						switch (so->so_proto->pr_domain->dom_family) {
1465 #ifdef INET
1466 						case AF_INET:
1467 							inp = sotoinpcb(so);
1468 							tp = intotcpcb(inp);
1469 							break;
1470 #endif
1471 #ifdef INET6
1472 						case AF_INET6:
1473 							in6p = sotoin6pcb(so);
1474 							tp = in6totcpcb(in6p);
1475 							break;
1476 #endif
1477 						}
1478 						if (tp == NULL)
1479 							goto badsyn;	/*XXX*/
1480 						tiwin <<= tp->snd_scale;
1481 						goto after_listen;
1482 					}
1483 				} else {
1484 					/*
1485 					 * None of RST, SYN or ACK was set.
1486 					 * This is an invalid packet for a
1487 					 * TCB in LISTEN state.  Send a RST.
1488 					 */
1489 					goto badsyn;
1490 				}
1491 			} else {
1492 				/*
1493 				 * Received a SYN.
1494 				 *
1495 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1496 				 */
1497 				if (m->m_flags & (M_BCAST|M_MCAST))
1498 					goto drop;
1499 
1500 				switch (af) {
1501 #ifdef INET6
1502 				case AF_INET6:
1503 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1504 						goto drop;
1505 					break;
1506 #endif /* INET6 */
1507 				case AF_INET:
1508 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1509 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1510 						goto drop;
1511 				break;
1512 				}
1513 
1514 #ifdef INET6
1515 				/*
1516 				 * If deprecated address is forbidden, we do
1517 				 * not accept SYN to deprecated interface
1518 				 * address to prevent any new inbound
1519 				 * connection from getting established.
1520 				 * When we do not accept SYN, we send a TCP
1521 				 * RST, with deprecated source address (instead
1522 				 * of dropping it).  We compromise it as it is
1523 				 * much better for peer to send a RST, and
1524 				 * RST will be the final packet for the
1525 				 * exchange.
1526 				 *
1527 				 * If we do not forbid deprecated addresses, we
1528 				 * accept the SYN packet.  RFC2462 does not
1529 				 * suggest dropping SYN in this case.
1530 				 * If we decipher RFC2462 5.5.4, it says like
1531 				 * this:
1532 				 * 1. use of deprecated addr with existing
1533 				 *    communication is okay - "SHOULD continue
1534 				 *    to be used"
1535 				 * 2. use of it with new communication:
1536 				 *   (2a) "SHOULD NOT be used if alternate
1537 				 *        address with sufficient scope is
1538 				 *        available"
1539 				 *   (2b) nothing mentioned otherwise.
1540 				 * Here we fall into (2b) case as we have no
1541 				 * choice in our source address selection - we
1542 				 * must obey the peer.
1543 				 *
1544 				 * The wording in RFC2462 is confusing, and
1545 				 * there are multiple description text for
1546 				 * deprecated address handling - worse, they
1547 				 * are not exactly the same.  I believe 5.5.4
1548 				 * is the best one, so we follow 5.5.4.
1549 				 */
1550 				if (af == AF_INET6 && !ip6_use_deprecated) {
1551 					struct in6_ifaddr *ia6;
1552 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1553 					    &ip6->ip6_dst)) &&
1554 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1555 						tp = NULL;
1556 						goto dropwithreset;
1557 					}
1558 				}
1559 #endif
1560 
1561 #if defined(IPSEC) || defined(FAST_IPSEC)
1562 				switch (af) {
1563 #ifdef INET
1564 				case AF_INET:
1565 					if (ipsec4_in_reject_so(m, so)) {
1566 						IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1567 						tp = NULL;
1568 						goto dropwithreset;
1569 					}
1570 					break;
1571 #endif
1572 #ifdef INET6
1573 				case AF_INET6:
1574 					if (ipsec6_in_reject_so(m, so)) {
1575 						IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO);
1576 						tp = NULL;
1577 						goto dropwithreset;
1578 					}
1579 					break;
1580 #endif /*INET6*/
1581 				}
1582 #endif /*IPSEC*/
1583 
1584 				/*
1585 				 * LISTEN socket received a SYN
1586 				 * from itself?  This can't possibly
1587 				 * be valid; drop the packet.
1588 				 */
1589 				if (th->th_sport == th->th_dport) {
1590 					int i;
1591 
1592 					switch (af) {
1593 #ifdef INET
1594 					case AF_INET:
1595 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1596 						break;
1597 #endif
1598 #ifdef INET6
1599 					case AF_INET6:
1600 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1601 						break;
1602 #endif
1603 					default:
1604 						i = 1;
1605 					}
1606 					if (i) {
1607 						TCP_STATINC(TCP_STAT_BADSYN);
1608 						goto drop;
1609 					}
1610 				}
1611 
1612 				/*
1613 				 * SYN looks ok; create compressed TCP
1614 				 * state for it.
1615 				 */
1616 				if (so->so_qlen <= so->so_qlimit &&
1617 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1618 						so, m, optp, optlen, &opti))
1619 					m = NULL;
1620 			}
1621 			goto drop;
1622 		}
1623 	}
1624 
1625 after_listen:
1626 #ifdef DIAGNOSTIC
1627 	/*
1628 	 * Should not happen now that all embryonic connections
1629 	 * are handled with compressed state.
1630 	 */
1631 	if (tp->t_state == TCPS_LISTEN)
1632 		panic("tcp_input: TCPS_LISTEN");
1633 #endif
1634 
1635 	/*
1636 	 * Segment received on connection.
1637 	 * Reset idle time and keep-alive timer.
1638 	 */
1639 	tp->t_rcvtime = tcp_now;
1640 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1641 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1642 
1643 	/*
1644 	 * Process options.
1645 	 */
1646 #ifdef TCP_SIGNATURE
1647 	if (optp || (tp->t_flags & TF_SIGNATURE))
1648 #else
1649 	if (optp)
1650 #endif
1651 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1652 			goto drop;
1653 
1654 	if (TCP_SACK_ENABLED(tp)) {
1655 		tcp_del_sackholes(tp, th);
1656 	}
1657 
1658 	if (TCP_ECN_ALLOWED(tp)) {
1659 		switch (iptos & IPTOS_ECN_MASK) {
1660 		case IPTOS_ECN_CE:
1661 			tp->t_flags |= TF_ECN_SND_ECE;
1662 			TCP_STATINC(TCP_STAT_ECN_CE);
1663 			break;
1664 		case IPTOS_ECN_ECT0:
1665 			TCP_STATINC(TCP_STAT_ECN_ECT);
1666 			break;
1667 		case IPTOS_ECN_ECT1:
1668 			/* XXX: ignore for now -- rpaulo */
1669 			break;
1670 		}
1671 
1672 		if (tiflags & TH_CWR)
1673 			tp->t_flags &= ~TF_ECN_SND_ECE;
1674 
1675 		/*
1676 		 * Congestion experienced.
1677 		 * Ignore if we are already trying to recover.
1678 		 */
1679 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1680 			tp->t_congctl->cong_exp(tp);
1681 	}
1682 
1683 	if (opti.ts_present && opti.ts_ecr) {
1684 		/*
1685 		 * Calculate the RTT from the returned time stamp and the
1686 		 * connection's time base.  If the time stamp is later than
1687 		 * the current time, or is extremely old, fall back to non-1323
1688 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1689 		 * at the same time.
1690 		 */
1691 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1692 		if (ts_rtt > TCP_PAWS_IDLE)
1693 			ts_rtt = 0;
1694 	} else {
1695 		ts_rtt = 0;
1696 	}
1697 
1698 	/*
1699 	 * Header prediction: check for the two common cases
1700 	 * of a uni-directional data xfer.  If the packet has
1701 	 * no control flags, is in-sequence, the window didn't
1702 	 * change and we're not retransmitting, it's a
1703 	 * candidate.  If the length is zero and the ack moved
1704 	 * forward, we're the sender side of the xfer.  Just
1705 	 * free the data acked & wake any higher level process
1706 	 * that was blocked waiting for space.  If the length
1707 	 * is non-zero and the ack didn't move, we're the
1708 	 * receiver side.  If we're getting packets in-order
1709 	 * (the reassembly queue is empty), add the data to
1710 	 * the socket buffer and note that we need a delayed ack.
1711 	 */
1712 	if (tp->t_state == TCPS_ESTABLISHED &&
1713 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1714 	        == TH_ACK &&
1715 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1716 	    th->th_seq == tp->rcv_nxt &&
1717 	    tiwin && tiwin == tp->snd_wnd &&
1718 	    tp->snd_nxt == tp->snd_max) {
1719 
1720 		/*
1721 		 * If last ACK falls within this segment's sequence numbers,
1722 		 * record the timestamp.
1723 		 * NOTE that the test is modified according to the latest
1724 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1725 		 *
1726 		 * note that we already know
1727 		 *	TSTMP_GEQ(opti.ts_val, tp->ts_recent)
1728 		 */
1729 		if (opti.ts_present &&
1730 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1731 			tp->ts_recent_age = tcp_now;
1732 			tp->ts_recent = opti.ts_val;
1733 		}
1734 
1735 		if (tlen == 0) {
1736 			/* Ack prediction. */
1737 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1738 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1739 			    tp->snd_cwnd >= tp->snd_wnd &&
1740 			    tp->t_partialacks < 0) {
1741 				/*
1742 				 * this is a pure ack for outstanding data.
1743 				 */
1744 				if (ts_rtt)
1745 					tcp_xmit_timer(tp, ts_rtt);
1746 				else if (tp->t_rtttime &&
1747 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1748 					tcp_xmit_timer(tp,
1749 					  tcp_now - tp->t_rtttime);
1750 				acked = th->th_ack - tp->snd_una;
1751 				tcps = TCP_STAT_GETREF();
1752 				tcps[TCP_STAT_PREDACK]++;
1753 				tcps[TCP_STAT_RCVACKPACK]++;
1754 				tcps[TCP_STAT_RCVACKBYTE] += acked;
1755 				TCP_STAT_PUTREF();
1756 				nd6_hint(tp);
1757 
1758 				if (acked > (tp->t_lastoff - tp->t_inoff))
1759 					tp->t_lastm = NULL;
1760 				sbdrop(&so->so_snd, acked);
1761 				tp->t_lastoff -= acked;
1762 
1763 				icmp_check(tp, th, acked);
1764 
1765 				tp->snd_una = th->th_ack;
1766 				tp->snd_fack = tp->snd_una;
1767 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1768 					tp->snd_high = tp->snd_una;
1769 				m_freem(m);
1770 
1771 				/*
1772 				 * If all outstanding data are acked, stop
1773 				 * retransmit timer, otherwise restart timer
1774 				 * using current (possibly backed-off) value.
1775 				 * If process is waiting for space,
1776 				 * wakeup/selnotify/signal.  If data
1777 				 * are ready to send, let tcp_output
1778 				 * decide between more output or persist.
1779 				 */
1780 				if (tp->snd_una == tp->snd_max)
1781 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1782 				else if (TCP_TIMER_ISARMED(tp,
1783 				    TCPT_PERSIST) == 0)
1784 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1785 					    tp->t_rxtcur);
1786 
1787 				sowwakeup(so);
1788 				if (so->so_snd.sb_cc)
1789 					(void) tcp_output(tp);
1790 				if (tcp_saveti)
1791 					m_freem(tcp_saveti);
1792 				return;
1793 			}
1794 		} else if (th->th_ack == tp->snd_una &&
1795 		    TAILQ_FIRST(&tp->segq) == NULL &&
1796 		    tlen <= sbspace(&so->so_rcv)) {
1797 			int newsize = 0;	/* automatic sockbuf scaling */
1798 
1799 			/*
1800 			 * this is a pure, in-sequence data packet
1801 			 * with nothing on the reassembly queue and
1802 			 * we have enough buffer space to take it.
1803 			 */
1804 			tp->rcv_nxt += tlen;
1805 			tcps = TCP_STAT_GETREF();
1806 			tcps[TCP_STAT_PREDDAT]++;
1807 			tcps[TCP_STAT_RCVPACK]++;
1808 			tcps[TCP_STAT_RCVBYTE] += tlen;
1809 			TCP_STAT_PUTREF();
1810 			nd6_hint(tp);
1811 
1812 		/*
1813 		 * Automatic sizing enables the performance of large buffers
1814 		 * and most of the efficiency of small ones by only allocating
1815 		 * space when it is needed.
1816 		 *
1817 		 * On the receive side the socket buffer memory is only rarely
1818 		 * used to any significant extent.  This allows us to be much
1819 		 * more aggressive in scaling the receive socket buffer.  For
1820 		 * the case that the buffer space is actually used to a large
1821 		 * extent and we run out of kernel memory we can simply drop
1822 		 * the new segments; TCP on the sender will just retransmit it
1823 		 * later.  Setting the buffer size too big may only consume too
1824 		 * much kernel memory if the application doesn't read() from
1825 		 * the socket or packet loss or reordering makes use of the
1826 		 * reassembly queue.
1827 		 *
1828 		 * The criteria to step up the receive buffer one notch are:
1829 		 *  1. the number of bytes received during the time it takes
1830 		 *     one timestamp to be reflected back to us (the RTT);
1831 		 *  2. received bytes per RTT is within seven eighth of the
1832 		 *     current socket buffer size;
1833 		 *  3. receive buffer size has not hit maximal automatic size;
1834 		 *
1835 		 * This algorithm does one step per RTT at most and only if
1836 		 * we receive a bulk stream w/o packet losses or reorderings.
1837 		 * Shrinking the buffer during idle times is not necessary as
1838 		 * it doesn't consume any memory when idle.
1839 		 *
1840 		 * TODO: Only step up if the application is actually serving
1841 		 * the buffer to better manage the socket buffer resources.
1842 		 */
1843 			if (tcp_do_autorcvbuf &&
1844 			    opti.ts_ecr &&
1845 			    (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1846 				if (opti.ts_ecr > tp->rfbuf_ts &&
1847 				    opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) {
1848 					if (tp->rfbuf_cnt >
1849 					    (so->so_rcv.sb_hiwat / 8 * 7) &&
1850 					    so->so_rcv.sb_hiwat <
1851 					    tcp_autorcvbuf_max) {
1852 						newsize =
1853 						    min(so->so_rcv.sb_hiwat +
1854 						    tcp_autorcvbuf_inc,
1855 						    tcp_autorcvbuf_max);
1856 					}
1857 					/* Start over with next RTT. */
1858 					tp->rfbuf_ts = 0;
1859 					tp->rfbuf_cnt = 0;
1860 				} else
1861 					tp->rfbuf_cnt += tlen;	/* add up */
1862 			}
1863 
1864 			/*
1865 			 * Drop TCP, IP headers and TCP options then add data
1866 			 * to socket buffer.
1867 			 */
1868 			if (so->so_state & SS_CANTRCVMORE)
1869 				m_freem(m);
1870 			else {
1871 				/*
1872 				 * Set new socket buffer size.
1873 				 * Give up when limit is reached.
1874 				 */
1875 				if (newsize)
1876 					if (!sbreserve(&so->so_rcv,
1877 					    newsize, so))
1878 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1879 				m_adj(m, toff + off);
1880 				sbappendstream(&so->so_rcv, m);
1881 			}
1882 			sorwakeup(so);
1883 			tcp_setup_ack(tp, th);
1884 			if (tp->t_flags & TF_ACKNOW)
1885 				(void) tcp_output(tp);
1886 			if (tcp_saveti)
1887 				m_freem(tcp_saveti);
1888 			return;
1889 		}
1890 	}
1891 
1892 	/*
1893 	 * Compute mbuf offset to TCP data segment.
1894 	 */
1895 	hdroptlen = toff + off;
1896 
1897 	/*
1898 	 * Calculate amount of space in receive window,
1899 	 * and then do TCP input processing.
1900 	 * Receive window is amount of space in rcv queue,
1901 	 * but not less than advertised window.
1902 	 */
1903 	{ int win;
1904 
1905 	win = sbspace(&so->so_rcv);
1906 	if (win < 0)
1907 		win = 0;
1908 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1909 	}
1910 
1911 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1912 	tp->rfbuf_ts = 0;
1913 	tp->rfbuf_cnt = 0;
1914 
1915 	switch (tp->t_state) {
1916 	/*
1917 	 * If the state is SYN_SENT:
1918 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1919 	 *	if seg contains a RST, then drop the connection.
1920 	 *	if seg does not contain SYN, then drop it.
1921 	 * Otherwise this is an acceptable SYN segment
1922 	 *	initialize tp->rcv_nxt and tp->irs
1923 	 *	if seg contains ack then advance tp->snd_una
1924 	 *	if seg contains a ECE and ECN support is enabled, the stream
1925 	 *	    is ECN capable.
1926 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1927 	 *	arrange for segment to be acked (eventually)
1928 	 *	continue processing rest of data/controls, beginning with URG
1929 	 */
1930 	case TCPS_SYN_SENT:
1931 		if ((tiflags & TH_ACK) &&
1932 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1933 		     SEQ_GT(th->th_ack, tp->snd_max)))
1934 			goto dropwithreset;
1935 		if (tiflags & TH_RST) {
1936 			if (tiflags & TH_ACK)
1937 				tp = tcp_drop(tp, ECONNREFUSED);
1938 			goto drop;
1939 		}
1940 		if ((tiflags & TH_SYN) == 0)
1941 			goto drop;
1942 		if (tiflags & TH_ACK) {
1943 			tp->snd_una = th->th_ack;
1944 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1945 				tp->snd_nxt = tp->snd_una;
1946 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1947 				tp->snd_high = tp->snd_una;
1948 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1949 
1950 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1951 				tp->t_flags |= TF_ECN_PERMIT;
1952 				TCP_STATINC(TCP_STAT_ECN_SHS);
1953 			}
1954 
1955 		}
1956 		tp->irs = th->th_seq;
1957 		tcp_rcvseqinit(tp);
1958 		tp->t_flags |= TF_ACKNOW;
1959 		tcp_mss_from_peer(tp, opti.maxseg);
1960 
1961 		/*
1962 		 * Initialize the initial congestion window.  If we
1963 		 * had to retransmit the SYN, we must initialize cwnd
1964 		 * to 1 segment (i.e. the Loss Window).
1965 		 */
1966 		if (tp->t_flags & TF_SYN_REXMT)
1967 			tp->snd_cwnd = tp->t_peermss;
1968 		else {
1969 			int ss = tcp_init_win;
1970 #ifdef INET
1971 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1972 				ss = tcp_init_win_local;
1973 #endif
1974 #ifdef INET6
1975 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1976 				ss = tcp_init_win_local;
1977 #endif
1978 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1979 		}
1980 
1981 		tcp_rmx_rtt(tp);
1982 		if (tiflags & TH_ACK) {
1983 			TCP_STATINC(TCP_STAT_CONNECTS);
1984 			soisconnected(so);
1985 			tcp_established(tp);
1986 			/* Do window scaling on this connection? */
1987 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1988 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1989 				tp->snd_scale = tp->requested_s_scale;
1990 				tp->rcv_scale = tp->request_r_scale;
1991 			}
1992 			TCP_REASS_LOCK(tp);
1993 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1994 			TCP_REASS_UNLOCK(tp);
1995 			/*
1996 			 * if we didn't have to retransmit the SYN,
1997 			 * use its rtt as our initial srtt & rtt var.
1998 			 */
1999 			if (tp->t_rtttime)
2000 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2001 		} else
2002 			tp->t_state = TCPS_SYN_RECEIVED;
2003 
2004 		/*
2005 		 * Advance th->th_seq to correspond to first data byte.
2006 		 * If data, trim to stay within window,
2007 		 * dropping FIN if necessary.
2008 		 */
2009 		th->th_seq++;
2010 		if (tlen > tp->rcv_wnd) {
2011 			todrop = tlen - tp->rcv_wnd;
2012 			m_adj(m, -todrop);
2013 			tlen = tp->rcv_wnd;
2014 			tiflags &= ~TH_FIN;
2015 			tcps = TCP_STAT_GETREF();
2016 			tcps[TCP_STAT_RCVPACKAFTERWIN]++;
2017 			tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop;
2018 			TCP_STAT_PUTREF();
2019 		}
2020 		tp->snd_wl1 = th->th_seq - 1;
2021 		tp->rcv_up = th->th_seq;
2022 		goto step6;
2023 
2024 	/*
2025 	 * If the state is SYN_RECEIVED:
2026 	 *	If seg contains an ACK, but not for our SYN, drop the input
2027 	 *	and generate an RST.  See page 36, rfc793
2028 	 */
2029 	case TCPS_SYN_RECEIVED:
2030 		if ((tiflags & TH_ACK) &&
2031 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
2032 		     SEQ_GT(th->th_ack, tp->snd_max)))
2033 			goto dropwithreset;
2034 		break;
2035 	}
2036 
2037 	/*
2038 	 * States other than LISTEN or SYN_SENT.
2039 	 * First check timestamp, if present.
2040 	 * Then check that at least some bytes of segment are within
2041 	 * receive window.  If segment begins before rcv_nxt,
2042 	 * drop leading data (and SYN); if nothing left, just ack.
2043 	 *
2044 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2045 	 * and it's less than ts_recent, drop it.
2046 	 */
2047 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
2048 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
2049 
2050 		/* Check to see if ts_recent is over 24 days old.  */
2051 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
2052 			/*
2053 			 * Invalidate ts_recent.  If this segment updates
2054 			 * ts_recent, the age will be reset later and ts_recent
2055 			 * will get a valid value.  If it does not, setting
2056 			 * ts_recent to zero will at least satisfy the
2057 			 * requirement that zero be placed in the timestamp
2058 			 * echo reply when ts_recent isn't valid.  The
2059 			 * age isn't reset until we get a valid ts_recent
2060 			 * because we don't want out-of-order segments to be
2061 			 * dropped when ts_recent is old.
2062 			 */
2063 			tp->ts_recent = 0;
2064 		} else {
2065 			tcps = TCP_STAT_GETREF();
2066 			tcps[TCP_STAT_RCVDUPPACK]++;
2067 			tcps[TCP_STAT_RCVDUPBYTE] += tlen;
2068 			tcps[TCP_STAT_PAWSDROP]++;
2069 			TCP_STAT_PUTREF();
2070 			tcp_new_dsack(tp, th->th_seq, tlen);
2071 			goto dropafterack;
2072 		}
2073 	}
2074 
2075 	todrop = tp->rcv_nxt - th->th_seq;
2076 	dupseg = false;
2077 	if (todrop > 0) {
2078 		if (tiflags & TH_SYN) {
2079 			tiflags &= ~TH_SYN;
2080 			th->th_seq++;
2081 			if (th->th_urp > 1)
2082 				th->th_urp--;
2083 			else {
2084 				tiflags &= ~TH_URG;
2085 				th->th_urp = 0;
2086 			}
2087 			todrop--;
2088 		}
2089 		if (todrop > tlen ||
2090 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
2091 			/*
2092 			 * Any valid FIN or RST must be to the left of the
2093 			 * window.  At this point the FIN or RST must be a
2094 			 * duplicate or out of sequence; drop it.
2095 			 */
2096 			if (tiflags & TH_RST)
2097 				goto drop;
2098 			tiflags &= ~(TH_FIN|TH_RST);
2099 			/*
2100 			 * Send an ACK to resynchronize and drop any data.
2101 			 * But keep on processing for RST or ACK.
2102 			 */
2103 			tp->t_flags |= TF_ACKNOW;
2104 			todrop = tlen;
2105 			dupseg = true;
2106 			tcps = TCP_STAT_GETREF();
2107 			tcps[TCP_STAT_RCVDUPPACK]++;
2108 			tcps[TCP_STAT_RCVDUPBYTE] += todrop;
2109 			TCP_STAT_PUTREF();
2110 		} else if ((tiflags & TH_RST) &&
2111 			   th->th_seq != tp->rcv_nxt) {
2112 			/*
2113 			 * Test for reset before adjusting the sequence
2114 			 * number for overlapping data.
2115 			 */
2116 			goto dropafterack_ratelim;
2117 		} else {
2118 			tcps = TCP_STAT_GETREF();
2119 			tcps[TCP_STAT_RCVPARTDUPPACK]++;
2120 			tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop;
2121 			TCP_STAT_PUTREF();
2122 		}
2123 		tcp_new_dsack(tp, th->th_seq, todrop);
2124 		hdroptlen += todrop;	/*drop from head afterwards*/
2125 		th->th_seq += todrop;
2126 		tlen -= todrop;
2127 		if (th->th_urp > todrop)
2128 			th->th_urp -= todrop;
2129 		else {
2130 			tiflags &= ~TH_URG;
2131 			th->th_urp = 0;
2132 		}
2133 	}
2134 
2135 	/*
2136 	 * If new data are received on a connection after the
2137 	 * user processes are gone, then RST the other end.
2138 	 */
2139 	if ((so->so_state & SS_NOFDREF) &&
2140 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2141 		tp = tcp_close(tp);
2142 		TCP_STATINC(TCP_STAT_RCVAFTERCLOSE);
2143 		goto dropwithreset;
2144 	}
2145 
2146 	/*
2147 	 * If segment ends after window, drop trailing data
2148 	 * (and PUSH and FIN); if nothing left, just ACK.
2149 	 */
2150 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2151 	if (todrop > 0) {
2152 		TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN);
2153 		if (todrop >= tlen) {
2154 			/*
2155 			 * The segment actually starts after the window.
2156 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2157 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2158 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2159 			 */
2160 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen);
2161 			/*
2162 			 * If a new connection request is received
2163 			 * while in TIME_WAIT, drop the old connection
2164 			 * and start over if the sequence numbers
2165 			 * are above the previous ones.
2166 			 *
2167 			 * NOTE: We will checksum the packet again, and
2168 			 * so we need to put the header fields back into
2169 			 * network order!
2170 			 * XXX This kind of sucks, but we don't expect
2171 			 * XXX this to happen very often, so maybe it
2172 			 * XXX doesn't matter so much.
2173 			 */
2174 			if (tiflags & TH_SYN &&
2175 			    tp->t_state == TCPS_TIME_WAIT &&
2176 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2177 				tp = tcp_close(tp);
2178 				tcp_fields_to_net(th);
2179 				goto findpcb;
2180 			}
2181 			/*
2182 			 * If window is closed can only take segments at
2183 			 * window edge, and have to drop data and PUSH from
2184 			 * incoming segments.  Continue processing, but
2185 			 * remember to ack.  Otherwise, drop segment
2186 			 * and (if not RST) ack.
2187 			 */
2188 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2189 				tp->t_flags |= TF_ACKNOW;
2190 				TCP_STATINC(TCP_STAT_RCVWINPROBE);
2191 			} else
2192 				goto dropafterack;
2193 		} else
2194 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop);
2195 		m_adj(m, -todrop);
2196 		tlen -= todrop;
2197 		tiflags &= ~(TH_PUSH|TH_FIN);
2198 	}
2199 
2200 	/*
2201 	 * If last ACK falls within this segment's sequence numbers,
2202 	 *  record the timestamp.
2203 	 * NOTE:
2204 	 * 1) That the test incorporates suggestions from the latest
2205 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2206 	 * 2) That updating only on newer timestamps interferes with
2207 	 *    our earlier PAWS tests, so this check should be solely
2208 	 *    predicated on the sequence space of this segment.
2209 	 * 3) That we modify the segment boundary check to be
2210 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2211 	 *    instead of RFC1323's
2212 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2213 	 *    This modified check allows us to overcome RFC1323's
2214 	 *    limitations as described in Stevens TCP/IP Illustrated
2215 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2216 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2217 	 */
2218 	if (opti.ts_present &&
2219 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2220 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2221 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2222 		tp->ts_recent_age = tcp_now;
2223 		tp->ts_recent = opti.ts_val;
2224 	}
2225 
2226 	/*
2227 	 * If the RST bit is set examine the state:
2228 	 *    SYN_RECEIVED STATE:
2229 	 *	If passive open, return to LISTEN state.
2230 	 *	If active open, inform user that connection was refused.
2231 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2232 	 *	Inform user that connection was reset, and close tcb.
2233 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2234 	 *	Close the tcb.
2235 	 */
2236 	if (tiflags & TH_RST) {
2237 		if (th->th_seq != tp->rcv_nxt)
2238 			goto dropafterack_ratelim;
2239 
2240 		switch (tp->t_state) {
2241 		case TCPS_SYN_RECEIVED:
2242 			so->so_error = ECONNREFUSED;
2243 			goto close;
2244 
2245 		case TCPS_ESTABLISHED:
2246 		case TCPS_FIN_WAIT_1:
2247 		case TCPS_FIN_WAIT_2:
2248 		case TCPS_CLOSE_WAIT:
2249 			so->so_error = ECONNRESET;
2250 		close:
2251 			tp->t_state = TCPS_CLOSED;
2252 			TCP_STATINC(TCP_STAT_DROPS);
2253 			tp = tcp_close(tp);
2254 			goto drop;
2255 
2256 		case TCPS_CLOSING:
2257 		case TCPS_LAST_ACK:
2258 		case TCPS_TIME_WAIT:
2259 			tp = tcp_close(tp);
2260 			goto drop;
2261 		}
2262 	}
2263 
2264 	/*
2265 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2266 	 * we must be in a synchronized state.  RFC791 states (under RST
2267 	 * generation) that any unacceptable segment (an out-of-order SYN
2268 	 * qualifies) received in a synchronized state must elicit only an
2269 	 * empty acknowledgment segment ... and the connection remains in
2270 	 * the same state.
2271 	 */
2272 	if (tiflags & TH_SYN) {
2273 		if (tp->rcv_nxt == th->th_seq) {
2274 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2275 			    TH_ACK);
2276 			if (tcp_saveti)
2277 				m_freem(tcp_saveti);
2278 			return;
2279 		}
2280 
2281 		goto dropafterack_ratelim;
2282 	}
2283 
2284 	/*
2285 	 * If the ACK bit is off we drop the segment and return.
2286 	 */
2287 	if ((tiflags & TH_ACK) == 0) {
2288 		if (tp->t_flags & TF_ACKNOW)
2289 			goto dropafterack;
2290 		else
2291 			goto drop;
2292 	}
2293 
2294 	/*
2295 	 * Ack processing.
2296 	 */
2297 	switch (tp->t_state) {
2298 
2299 	/*
2300 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2301 	 * ESTABLISHED state and continue processing, otherwise
2302 	 * send an RST.
2303 	 */
2304 	case TCPS_SYN_RECEIVED:
2305 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2306 		    SEQ_GT(th->th_ack, tp->snd_max))
2307 			goto dropwithreset;
2308 		TCP_STATINC(TCP_STAT_CONNECTS);
2309 		soisconnected(so);
2310 		tcp_established(tp);
2311 		/* Do window scaling? */
2312 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2313 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2314 			tp->snd_scale = tp->requested_s_scale;
2315 			tp->rcv_scale = tp->request_r_scale;
2316 		}
2317 		TCP_REASS_LOCK(tp);
2318 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2319 		TCP_REASS_UNLOCK(tp);
2320 		tp->snd_wl1 = th->th_seq - 1;
2321 		/* fall into ... */
2322 
2323 	/*
2324 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2325 	 * ACKs.  If the ack is in the range
2326 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2327 	 * then advance tp->snd_una to th->th_ack and drop
2328 	 * data from the retransmission queue.  If this ACK reflects
2329 	 * more up to date window information we update our window information.
2330 	 */
2331 	case TCPS_ESTABLISHED:
2332 	case TCPS_FIN_WAIT_1:
2333 	case TCPS_FIN_WAIT_2:
2334 	case TCPS_CLOSE_WAIT:
2335 	case TCPS_CLOSING:
2336 	case TCPS_LAST_ACK:
2337 	case TCPS_TIME_WAIT:
2338 
2339 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2340 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2341 				TCP_STATINC(TCP_STAT_RCVDUPPACK);
2342 				/*
2343 				 * If we have outstanding data (other than
2344 				 * a window probe), this is a completely
2345 				 * duplicate ack (ie, window info didn't
2346 				 * change), the ack is the biggest we've
2347 				 * seen and we've seen exactly our rexmt
2348 				 * threshhold of them, assume a packet
2349 				 * has been dropped and retransmit it.
2350 				 * Kludge snd_nxt & the congestion
2351 				 * window so we send only this one
2352 				 * packet.
2353 				 */
2354 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2355 				    th->th_ack != tp->snd_una)
2356 					tp->t_dupacks = 0;
2357 				else if (tp->t_partialacks < 0 &&
2358 					 (++tp->t_dupacks == tcprexmtthresh ||
2359 					 TCP_FACK_FASTRECOV(tp))) {
2360 					/*
2361 					 * Do the fast retransmit, and adjust
2362 					 * congestion control paramenters.
2363 					 */
2364 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2365 						/* False fast retransmit */
2366 						break;
2367 					} else
2368 						goto drop;
2369 				} else if (tp->t_dupacks > tcprexmtthresh) {
2370 					tp->snd_cwnd += tp->t_segsz;
2371 					(void) tcp_output(tp);
2372 					goto drop;
2373 				}
2374 			} else {
2375 				/*
2376 				 * If the ack appears to be very old, only
2377 				 * allow data that is in-sequence.  This
2378 				 * makes it somewhat more difficult to insert
2379 				 * forged data by guessing sequence numbers.
2380 				 * Sent an ack to try to update the send
2381 				 * sequence number on the other side.
2382 				 */
2383 				if (tlen && th->th_seq != tp->rcv_nxt &&
2384 				    SEQ_LT(th->th_ack,
2385 				    tp->snd_una - tp->max_sndwnd))
2386 					goto dropafterack;
2387 			}
2388 			break;
2389 		}
2390 		/*
2391 		 * If the congestion window was inflated to account
2392 		 * for the other side's cached packets, retract it.
2393 		 */
2394 		/* XXX: make SACK have his own congestion control
2395 		 * struct -- rpaulo */
2396 		if (TCP_SACK_ENABLED(tp))
2397 			tcp_sack_newack(tp, th);
2398 		else
2399 			tp->t_congctl->fast_retransmit_newack(tp, th);
2400 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2401 			TCP_STATINC(TCP_STAT_RCVACKTOOMUCH);
2402 			goto dropafterack;
2403 		}
2404 		acked = th->th_ack - tp->snd_una;
2405 		tcps = TCP_STAT_GETREF();
2406 		tcps[TCP_STAT_RCVACKPACK]++;
2407 		tcps[TCP_STAT_RCVACKBYTE] += acked;
2408 		TCP_STAT_PUTREF();
2409 
2410 		/*
2411 		 * If we have a timestamp reply, update smoothed
2412 		 * round trip time.  If no timestamp is present but
2413 		 * transmit timer is running and timed sequence
2414 		 * number was acked, update smoothed round trip time.
2415 		 * Since we now have an rtt measurement, cancel the
2416 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2417 		 * Recompute the initial retransmit timer.
2418 		 */
2419 		if (ts_rtt)
2420 			tcp_xmit_timer(tp, ts_rtt);
2421 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2422 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2423 
2424 		/*
2425 		 * If all outstanding data is acked, stop retransmit
2426 		 * timer and remember to restart (more output or persist).
2427 		 * If there is more data to be acked, restart retransmit
2428 		 * timer, using current (possibly backed-off) value.
2429 		 */
2430 		if (th->th_ack == tp->snd_max) {
2431 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2432 			needoutput = 1;
2433 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2434 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2435 
2436 		/*
2437 		 * New data has been acked, adjust the congestion window.
2438 		 */
2439 		tp->t_congctl->newack(tp, th);
2440 
2441 		nd6_hint(tp);
2442 		if (acked > so->so_snd.sb_cc) {
2443 			tp->snd_wnd -= so->so_snd.sb_cc;
2444 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2445 			ourfinisacked = 1;
2446 		} else {
2447 			if (acked > (tp->t_lastoff - tp->t_inoff))
2448 				tp->t_lastm = NULL;
2449 			sbdrop(&so->so_snd, acked);
2450 			tp->t_lastoff -= acked;
2451 			tp->snd_wnd -= acked;
2452 			ourfinisacked = 0;
2453 		}
2454 		sowwakeup(so);
2455 
2456 		icmp_check(tp, th, acked);
2457 
2458 		tp->snd_una = th->th_ack;
2459 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2460 			tp->snd_fack = tp->snd_una;
2461 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2462 			tp->snd_nxt = tp->snd_una;
2463 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2464 			tp->snd_high = tp->snd_una;
2465 
2466 		switch (tp->t_state) {
2467 
2468 		/*
2469 		 * In FIN_WAIT_1 STATE in addition to the processing
2470 		 * for the ESTABLISHED state if our FIN is now acknowledged
2471 		 * then enter FIN_WAIT_2.
2472 		 */
2473 		case TCPS_FIN_WAIT_1:
2474 			if (ourfinisacked) {
2475 				/*
2476 				 * If we can't receive any more
2477 				 * data, then closing user can proceed.
2478 				 * Starting the timer is contrary to the
2479 				 * specification, but if we don't get a FIN
2480 				 * we'll hang forever.
2481 				 */
2482 				if (so->so_state & SS_CANTRCVMORE) {
2483 					soisdisconnected(so);
2484 					if (tp->t_maxidle > 0)
2485 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2486 						    tp->t_maxidle);
2487 				}
2488 				tp->t_state = TCPS_FIN_WAIT_2;
2489 			}
2490 			break;
2491 
2492 	 	/*
2493 		 * In CLOSING STATE in addition to the processing for
2494 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2495 		 * then enter the TIME-WAIT state, otherwise ignore
2496 		 * the segment.
2497 		 */
2498 		case TCPS_CLOSING:
2499 			if (ourfinisacked) {
2500 				tp->t_state = TCPS_TIME_WAIT;
2501 				tcp_canceltimers(tp);
2502 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2503 				soisdisconnected(so);
2504 			}
2505 			break;
2506 
2507 		/*
2508 		 * In LAST_ACK, we may still be waiting for data to drain
2509 		 * and/or to be acked, as well as for the ack of our FIN.
2510 		 * If our FIN is now acknowledged, delete the TCB,
2511 		 * enter the closed state and return.
2512 		 */
2513 		case TCPS_LAST_ACK:
2514 			if (ourfinisacked) {
2515 				tp = tcp_close(tp);
2516 				goto drop;
2517 			}
2518 			break;
2519 
2520 		/*
2521 		 * In TIME_WAIT state the only thing that should arrive
2522 		 * is a retransmission of the remote FIN.  Acknowledge
2523 		 * it and restart the finack timer.
2524 		 */
2525 		case TCPS_TIME_WAIT:
2526 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2527 			goto dropafterack;
2528 		}
2529 	}
2530 
2531 step6:
2532 	/*
2533 	 * Update window information.
2534 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2535 	 */
2536 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2537 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2538 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2539 		/* keep track of pure window updates */
2540 		if (tlen == 0 &&
2541 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2542 			TCP_STATINC(TCP_STAT_RCVWINUPD);
2543 		tp->snd_wnd = tiwin;
2544 		tp->snd_wl1 = th->th_seq;
2545 		tp->snd_wl2 = th->th_ack;
2546 		if (tp->snd_wnd > tp->max_sndwnd)
2547 			tp->max_sndwnd = tp->snd_wnd;
2548 		needoutput = 1;
2549 	}
2550 
2551 	/*
2552 	 * Process segments with URG.
2553 	 */
2554 	if ((tiflags & TH_URG) && th->th_urp &&
2555 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2556 		/*
2557 		 * This is a kludge, but if we receive and accept
2558 		 * random urgent pointers, we'll crash in
2559 		 * soreceive.  It's hard to imagine someone
2560 		 * actually wanting to send this much urgent data.
2561 		 */
2562 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2563 			th->th_urp = 0;			/* XXX */
2564 			tiflags &= ~TH_URG;		/* XXX */
2565 			goto dodata;			/* XXX */
2566 		}
2567 		/*
2568 		 * If this segment advances the known urgent pointer,
2569 		 * then mark the data stream.  This should not happen
2570 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2571 		 * a FIN has been received from the remote side.
2572 		 * In these states we ignore the URG.
2573 		 *
2574 		 * According to RFC961 (Assigned Protocols),
2575 		 * the urgent pointer points to the last octet
2576 		 * of urgent data.  We continue, however,
2577 		 * to consider it to indicate the first octet
2578 		 * of data past the urgent section as the original
2579 		 * spec states (in one of two places).
2580 		 */
2581 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2582 			tp->rcv_up = th->th_seq + th->th_urp;
2583 			so->so_oobmark = so->so_rcv.sb_cc +
2584 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2585 			if (so->so_oobmark == 0)
2586 				so->so_state |= SS_RCVATMARK;
2587 			sohasoutofband(so);
2588 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2589 		}
2590 		/*
2591 		 * Remove out of band data so doesn't get presented to user.
2592 		 * This can happen independent of advancing the URG pointer,
2593 		 * but if two URG's are pending at once, some out-of-band
2594 		 * data may creep in... ick.
2595 		 */
2596 		if (th->th_urp <= (u_int16_t) tlen
2597 #ifdef SO_OOBINLINE
2598 		     && (so->so_options & SO_OOBINLINE) == 0
2599 #endif
2600 		     )
2601 			tcp_pulloutofband(so, th, m, hdroptlen);
2602 	} else
2603 		/*
2604 		 * If no out of band data is expected,
2605 		 * pull receive urgent pointer along
2606 		 * with the receive window.
2607 		 */
2608 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2609 			tp->rcv_up = tp->rcv_nxt;
2610 dodata:							/* XXX */
2611 
2612 	/*
2613 	 * Process the segment text, merging it into the TCP sequencing queue,
2614 	 * and arranging for acknowledgement of receipt if necessary.
2615 	 * This process logically involves adjusting tp->rcv_wnd as data
2616 	 * is presented to the user (this happens in tcp_usrreq.c,
2617 	 * case PRU_RCVD).  If a FIN has already been received on this
2618 	 * connection then we just ignore the text.
2619 	 */
2620 	if ((tlen || (tiflags & TH_FIN)) &&
2621 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2622 		/*
2623 		 * Insert segment ti into reassembly queue of tcp with
2624 		 * control block tp.  Return TH_FIN if reassembly now includes
2625 		 * a segment with FIN.  The macro form does the common case
2626 		 * inline (segment is the next to be received on an
2627 		 * established connection, and the queue is empty),
2628 		 * avoiding linkage into and removal from the queue and
2629 		 * repetition of various conversions.
2630 		 * Set DELACK for segments received in order, but ack
2631 		 * immediately when segments are out of order
2632 		 * (so fast retransmit can work).
2633 		 */
2634 		/* NOTE: this was TCP_REASS() macro, but used only once */
2635 		TCP_REASS_LOCK(tp);
2636 		if (th->th_seq == tp->rcv_nxt &&
2637 		    TAILQ_FIRST(&tp->segq) == NULL &&
2638 		    tp->t_state == TCPS_ESTABLISHED) {
2639 			tcp_setup_ack(tp, th);
2640 			tp->rcv_nxt += tlen;
2641 			tiflags = th->th_flags & TH_FIN;
2642 			tcps = TCP_STAT_GETREF();
2643 			tcps[TCP_STAT_RCVPACK]++;
2644 			tcps[TCP_STAT_RCVBYTE] += tlen;
2645 			TCP_STAT_PUTREF();
2646 			nd6_hint(tp);
2647 			if (so->so_state & SS_CANTRCVMORE)
2648 				m_freem(m);
2649 			else {
2650 				m_adj(m, hdroptlen);
2651 				sbappendstream(&(so)->so_rcv, m);
2652 			}
2653 			TCP_REASS_UNLOCK(tp);
2654 			sorwakeup(so);
2655 		} else {
2656 			m_adj(m, hdroptlen);
2657 			tiflags = tcp_reass(tp, th, m, &tlen);
2658 			tp->t_flags |= TF_ACKNOW;
2659 			TCP_REASS_UNLOCK(tp);
2660 		}
2661 
2662 		/*
2663 		 * Note the amount of data that peer has sent into
2664 		 * our window, in order to estimate the sender's
2665 		 * buffer size.
2666 		 */
2667 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2668 	} else {
2669 		m_freem(m);
2670 		m = NULL;
2671 		tiflags &= ~TH_FIN;
2672 	}
2673 
2674 	/*
2675 	 * If FIN is received ACK the FIN and let the user know
2676 	 * that the connection is closing.  Ignore a FIN received before
2677 	 * the connection is fully established.
2678 	 */
2679 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2680 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2681 			socantrcvmore(so);
2682 			tp->t_flags |= TF_ACKNOW;
2683 			tp->rcv_nxt++;
2684 		}
2685 		switch (tp->t_state) {
2686 
2687 	 	/*
2688 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2689 		 */
2690 		case TCPS_ESTABLISHED:
2691 			tp->t_state = TCPS_CLOSE_WAIT;
2692 			break;
2693 
2694 	 	/*
2695 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2696 		 * enter the CLOSING state.
2697 		 */
2698 		case TCPS_FIN_WAIT_1:
2699 			tp->t_state = TCPS_CLOSING;
2700 			break;
2701 
2702 	 	/*
2703 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2704 		 * starting the time-wait timer, turning off the other
2705 		 * standard timers.
2706 		 */
2707 		case TCPS_FIN_WAIT_2:
2708 			tp->t_state = TCPS_TIME_WAIT;
2709 			tcp_canceltimers(tp);
2710 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2711 			soisdisconnected(so);
2712 			break;
2713 
2714 		/*
2715 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2716 		 */
2717 		case TCPS_TIME_WAIT:
2718 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2719 			break;
2720 		}
2721 	}
2722 #ifdef TCP_DEBUG
2723 	if (so->so_options & SO_DEBUG)
2724 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2725 #endif
2726 
2727 	/*
2728 	 * Return any desired output.
2729 	 */
2730 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2731 		(void) tcp_output(tp);
2732 	}
2733 	if (tcp_saveti)
2734 		m_freem(tcp_saveti);
2735 	return;
2736 
2737 badsyn:
2738 	/*
2739 	 * Received a bad SYN.  Increment counters and dropwithreset.
2740 	 */
2741 	TCP_STATINC(TCP_STAT_BADSYN);
2742 	tp = NULL;
2743 	goto dropwithreset;
2744 
2745 dropafterack:
2746 	/*
2747 	 * Generate an ACK dropping incoming segment if it occupies
2748 	 * sequence space, where the ACK reflects our state.
2749 	 */
2750 	if (tiflags & TH_RST)
2751 		goto drop;
2752 	goto dropafterack2;
2753 
2754 dropafterack_ratelim:
2755 	/*
2756 	 * We may want to rate-limit ACKs against SYN/RST attack.
2757 	 */
2758 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2759 	    tcp_ackdrop_ppslim) == 0) {
2760 		/* XXX stat */
2761 		goto drop;
2762 	}
2763 	/* ...fall into dropafterack2... */
2764 
2765 dropafterack2:
2766 	m_freem(m);
2767 	tp->t_flags |= TF_ACKNOW;
2768 	(void) tcp_output(tp);
2769 	if (tcp_saveti)
2770 		m_freem(tcp_saveti);
2771 	return;
2772 
2773 dropwithreset_ratelim:
2774 	/*
2775 	 * We may want to rate-limit RSTs in certain situations,
2776 	 * particularly if we are sending an RST in response to
2777 	 * an attempt to connect to or otherwise communicate with
2778 	 * a port for which we have no socket.
2779 	 */
2780 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2781 	    tcp_rst_ppslim) == 0) {
2782 		/* XXX stat */
2783 		goto drop;
2784 	}
2785 	/* ...fall into dropwithreset... */
2786 
2787 dropwithreset:
2788 	/*
2789 	 * Generate a RST, dropping incoming segment.
2790 	 * Make ACK acceptable to originator of segment.
2791 	 */
2792 	if (tiflags & TH_RST)
2793 		goto drop;
2794 
2795 	switch (af) {
2796 #ifdef INET6
2797 	case AF_INET6:
2798 		/* For following calls to tcp_respond */
2799 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2800 			goto drop;
2801 		break;
2802 #endif /* INET6 */
2803 	case AF_INET:
2804 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2805 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2806 			goto drop;
2807 	}
2808 
2809 	if (tiflags & TH_ACK)
2810 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2811 	else {
2812 		if (tiflags & TH_SYN)
2813 			tlen++;
2814 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2815 		    TH_RST|TH_ACK);
2816 	}
2817 	if (tcp_saveti)
2818 		m_freem(tcp_saveti);
2819 	return;
2820 
2821 badcsum:
2822 drop:
2823 	/*
2824 	 * Drop space held by incoming segment and return.
2825 	 */
2826 	if (tp) {
2827 		if (tp->t_inpcb)
2828 			so = tp->t_inpcb->inp_socket;
2829 #ifdef INET6
2830 		else if (tp->t_in6pcb)
2831 			so = tp->t_in6pcb->in6p_socket;
2832 #endif
2833 		else
2834 			so = NULL;
2835 #ifdef TCP_DEBUG
2836 		if (so && (so->so_options & SO_DEBUG) != 0)
2837 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2838 #endif
2839 	}
2840 	if (tcp_saveti)
2841 		m_freem(tcp_saveti);
2842 	m_freem(m);
2843 	return;
2844 }
2845 
2846 #ifdef TCP_SIGNATURE
2847 int
2848 tcp_signature_apply(void *fstate, void *data, u_int len)
2849 {
2850 
2851 	MD5Update(fstate, (u_char *)data, len);
2852 	return (0);
2853 }
2854 
2855 struct secasvar *
2856 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2857 {
2858 	struct secasvar *sav;
2859 #ifdef FAST_IPSEC
2860 	union sockaddr_union dst;
2861 #endif
2862 	struct ip *ip;
2863 	struct ip6_hdr *ip6;
2864 
2865 	ip = mtod(m, struct ip *);
2866 	switch (ip->ip_v) {
2867 	case 4:
2868 		ip = mtod(m, struct ip *);
2869 		ip6 = NULL;
2870 		break;
2871 	case 6:
2872 		ip = NULL;
2873 		ip6 = mtod(m, struct ip6_hdr *);
2874 		break;
2875 	default:
2876 		return (NULL);
2877 	}
2878 
2879 #ifdef FAST_IPSEC
2880 	/* Extract the destination from the IP header in the mbuf. */
2881 	memset(&dst, 0, sizeof(union sockaddr_union));
2882 	if (ip !=NULL) {
2883 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2884 		dst.sa.sa_family = AF_INET;
2885 		dst.sin.sin_addr = ip->ip_dst;
2886 	} else {
2887 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2888 		dst.sa.sa_family = AF_INET6;
2889 		dst.sin6.sin6_addr = ip6->ip6_dst;
2890 	}
2891 
2892 	/*
2893 	 * Look up an SADB entry which matches the address of the peer.
2894 	 */
2895 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2896 #else
2897 	if (ip)
2898 		sav = key_allocsa(AF_INET, (void *)&ip->ip_src,
2899 		    (void *)&ip->ip_dst, IPPROTO_TCP,
2900 		    htonl(TCP_SIG_SPI), 0, 0);
2901 	else
2902 		sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src,
2903 		    (void *)&ip6->ip6_dst, IPPROTO_TCP,
2904 		    htonl(TCP_SIG_SPI), 0, 0);
2905 #endif
2906 
2907 	return (sav);	/* freesav must be performed by caller */
2908 }
2909 
2910 int
2911 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2912     struct secasvar *sav, char *sig)
2913 {
2914 	MD5_CTX ctx;
2915 	struct ip *ip;
2916 	struct ipovly *ipovly;
2917 	struct ip6_hdr *ip6;
2918 	struct ippseudo ippseudo;
2919 	struct ip6_hdr_pseudo ip6pseudo;
2920 	struct tcphdr th0;
2921 	int l, tcphdrlen;
2922 
2923 	if (sav == NULL)
2924 		return (-1);
2925 
2926 	tcphdrlen = th->th_off * 4;
2927 
2928 	switch (mtod(m, struct ip *)->ip_v) {
2929 	case 4:
2930 		ip = mtod(m, struct ip *);
2931 		ip6 = NULL;
2932 		break;
2933 	case 6:
2934 		ip = NULL;
2935 		ip6 = mtod(m, struct ip6_hdr *);
2936 		break;
2937 	default:
2938 		return (-1);
2939 	}
2940 
2941 	MD5Init(&ctx);
2942 
2943 	if (ip) {
2944 		memset(&ippseudo, 0, sizeof(ippseudo));
2945 		ipovly = (struct ipovly *)ip;
2946 		ippseudo.ippseudo_src = ipovly->ih_src;
2947 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2948 		ippseudo.ippseudo_pad = 0;
2949 		ippseudo.ippseudo_p = IPPROTO_TCP;
2950 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2951 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2952 	} else {
2953 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2954 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2955 		in6_clearscope(&ip6pseudo.ip6ph_src);
2956 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2957 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2958 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2959 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2960 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2961 	}
2962 
2963 	th0 = *th;
2964 	th0.th_sum = 0;
2965 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2966 
2967 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2968 	if (l > 0)
2969 		m_apply(m, thoff + tcphdrlen,
2970 		    m->m_pkthdr.len - thoff - tcphdrlen,
2971 		    tcp_signature_apply, &ctx);
2972 
2973 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2974 	MD5Final(sig, &ctx);
2975 
2976 	return (0);
2977 }
2978 #endif
2979 
2980 static int
2981 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt,
2982     struct tcphdr *th,
2983     struct mbuf *m, int toff, struct tcp_opt_info *oi)
2984 {
2985 	u_int16_t mss;
2986 	int opt, optlen = 0;
2987 #ifdef TCP_SIGNATURE
2988 	void *sigp = NULL;
2989 	char sigbuf[TCP_SIGLEN];
2990 	struct secasvar *sav = NULL;
2991 #endif
2992 
2993 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2994 		opt = cp[0];
2995 		if (opt == TCPOPT_EOL)
2996 			break;
2997 		if (opt == TCPOPT_NOP)
2998 			optlen = 1;
2999 		else {
3000 			if (cnt < 2)
3001 				break;
3002 			optlen = cp[1];
3003 			if (optlen < 2 || optlen > cnt)
3004 				break;
3005 		}
3006 		switch (opt) {
3007 
3008 		default:
3009 			continue;
3010 
3011 		case TCPOPT_MAXSEG:
3012 			if (optlen != TCPOLEN_MAXSEG)
3013 				continue;
3014 			if (!(th->th_flags & TH_SYN))
3015 				continue;
3016 			if (TCPS_HAVERCVDSYN(tp->t_state))
3017 				continue;
3018 			bcopy(cp + 2, &mss, sizeof(mss));
3019 			oi->maxseg = ntohs(mss);
3020 			break;
3021 
3022 		case TCPOPT_WINDOW:
3023 			if (optlen != TCPOLEN_WINDOW)
3024 				continue;
3025 			if (!(th->th_flags & TH_SYN))
3026 				continue;
3027 			if (TCPS_HAVERCVDSYN(tp->t_state))
3028 				continue;
3029 			tp->t_flags |= TF_RCVD_SCALE;
3030 			tp->requested_s_scale = cp[2];
3031 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
3032 #if 0	/*XXX*/
3033 				char *p;
3034 
3035 				if (ip)
3036 					p = ntohl(ip->ip_src);
3037 #ifdef INET6
3038 				else if (ip6)
3039 					p = ip6_sprintf(&ip6->ip6_src);
3040 #endif
3041 				else
3042 					p = "(unknown)";
3043 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
3044 				    "assuming %d\n",
3045 				    tp->requested_s_scale, p,
3046 				    TCP_MAX_WINSHIFT);
3047 #else
3048 				log(LOG_ERR, "TCP: invalid wscale %d, "
3049 				    "assuming %d\n",
3050 				    tp->requested_s_scale,
3051 				    TCP_MAX_WINSHIFT);
3052 #endif
3053 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
3054 			}
3055 			break;
3056 
3057 		case TCPOPT_TIMESTAMP:
3058 			if (optlen != TCPOLEN_TIMESTAMP)
3059 				continue;
3060 			oi->ts_present = 1;
3061 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
3062 			NTOHL(oi->ts_val);
3063 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
3064 			NTOHL(oi->ts_ecr);
3065 
3066 			if (!(th->th_flags & TH_SYN))
3067 				continue;
3068 			if (TCPS_HAVERCVDSYN(tp->t_state))
3069 				continue;
3070 			/*
3071 			 * A timestamp received in a SYN makes
3072 			 * it ok to send timestamp requests and replies.
3073 			 */
3074 			tp->t_flags |= TF_RCVD_TSTMP;
3075 			tp->ts_recent = oi->ts_val;
3076 			tp->ts_recent_age = tcp_now;
3077                         break;
3078 
3079 		case TCPOPT_SACK_PERMITTED:
3080 			if (optlen != TCPOLEN_SACK_PERMITTED)
3081 				continue;
3082 			if (!(th->th_flags & TH_SYN))
3083 				continue;
3084 			if (TCPS_HAVERCVDSYN(tp->t_state))
3085 				continue;
3086 			if (tcp_do_sack) {
3087 				tp->t_flags |= TF_SACK_PERMIT;
3088 				tp->t_flags |= TF_WILL_SACK;
3089 			}
3090 			break;
3091 
3092 		case TCPOPT_SACK:
3093 			tcp_sack_option(tp, th, cp, optlen);
3094 			break;
3095 #ifdef TCP_SIGNATURE
3096 		case TCPOPT_SIGNATURE:
3097 			if (optlen != TCPOLEN_SIGNATURE)
3098 				continue;
3099 			if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN))
3100 				return (-1);
3101 
3102 			sigp = sigbuf;
3103 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
3104 			tp->t_flags |= TF_SIGNATURE;
3105 			break;
3106 #endif
3107 		}
3108 	}
3109 
3110 #ifdef TCP_SIGNATURE
3111 	if (tp->t_flags & TF_SIGNATURE) {
3112 
3113 		sav = tcp_signature_getsav(m, th);
3114 
3115 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3116 			return (-1);
3117 	}
3118 
3119 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3120 		if (sav == NULL)
3121 			return (-1);
3122 #ifdef FAST_IPSEC
3123 		KEY_FREESAV(&sav);
3124 #else
3125 		key_freesav(sav);
3126 #endif
3127 		return (-1);
3128 	}
3129 
3130 	if (sigp) {
3131 		char sig[TCP_SIGLEN];
3132 
3133 		tcp_fields_to_net(th);
3134 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3135 			tcp_fields_to_host(th);
3136 			if (sav == NULL)
3137 				return (-1);
3138 #ifdef FAST_IPSEC
3139 			KEY_FREESAV(&sav);
3140 #else
3141 			key_freesav(sav);
3142 #endif
3143 			return (-1);
3144 		}
3145 		tcp_fields_to_host(th);
3146 
3147 		if (memcmp(sig, sigp, TCP_SIGLEN)) {
3148 			TCP_STATINC(TCP_STAT_BADSIG);
3149 			if (sav == NULL)
3150 				return (-1);
3151 #ifdef FAST_IPSEC
3152 			KEY_FREESAV(&sav);
3153 #else
3154 			key_freesav(sav);
3155 #endif
3156 			return (-1);
3157 		} else
3158 			TCP_STATINC(TCP_STAT_GOODSIG);
3159 
3160 		key_sa_recordxfer(sav, m);
3161 #ifdef FAST_IPSEC
3162 		KEY_FREESAV(&sav);
3163 #else
3164 		key_freesav(sav);
3165 #endif
3166 	}
3167 #endif
3168 
3169 	return (0);
3170 }
3171 
3172 /*
3173  * Pull out of band byte out of a segment so
3174  * it doesn't appear in the user's data queue.
3175  * It is still reflected in the segment length for
3176  * sequencing purposes.
3177  */
3178 void
3179 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3180     struct mbuf *m, int off)
3181 {
3182 	int cnt = off + th->th_urp - 1;
3183 
3184 	while (cnt >= 0) {
3185 		if (m->m_len > cnt) {
3186 			char *cp = mtod(m, char *) + cnt;
3187 			struct tcpcb *tp = sototcpcb(so);
3188 
3189 			tp->t_iobc = *cp;
3190 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3191 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3192 			m->m_len--;
3193 			return;
3194 		}
3195 		cnt -= m->m_len;
3196 		m = m->m_next;
3197 		if (m == 0)
3198 			break;
3199 	}
3200 	panic("tcp_pulloutofband");
3201 }
3202 
3203 /*
3204  * Collect new round-trip time estimate
3205  * and update averages and current timeout.
3206  */
3207 void
3208 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3209 {
3210 	int32_t delta;
3211 
3212 	TCP_STATINC(TCP_STAT_RTTUPDATED);
3213 	if (tp->t_srtt != 0) {
3214 		/*
3215 		 * srtt is stored as fixed point with 3 bits after the
3216 		 * binary point (i.e., scaled by 8).  The following magic
3217 		 * is equivalent to the smoothing algorithm in rfc793 with
3218 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3219 		 * point).  Adjust rtt to origin 0.
3220 		 */
3221 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3222 		if ((tp->t_srtt += delta) <= 0)
3223 			tp->t_srtt = 1 << 2;
3224 		/*
3225 		 * We accumulate a smoothed rtt variance (actually, a
3226 		 * smoothed mean difference), then set the retransmit
3227 		 * timer to smoothed rtt + 4 times the smoothed variance.
3228 		 * rttvar is stored as fixed point with 2 bits after the
3229 		 * binary point (scaled by 4).  The following is
3230 		 * equivalent to rfc793 smoothing with an alpha of .75
3231 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3232 		 * rfc793's wired-in beta.
3233 		 */
3234 		if (delta < 0)
3235 			delta = -delta;
3236 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3237 		if ((tp->t_rttvar += delta) <= 0)
3238 			tp->t_rttvar = 1 << 2;
3239 	} else {
3240 		/*
3241 		 * No rtt measurement yet - use the unsmoothed rtt.
3242 		 * Set the variance to half the rtt (so our first
3243 		 * retransmit happens at 3*rtt).
3244 		 */
3245 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3246 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3247 	}
3248 	tp->t_rtttime = 0;
3249 	tp->t_rxtshift = 0;
3250 
3251 	/*
3252 	 * the retransmit should happen at rtt + 4 * rttvar.
3253 	 * Because of the way we do the smoothing, srtt and rttvar
3254 	 * will each average +1/2 tick of bias.  When we compute
3255 	 * the retransmit timer, we want 1/2 tick of rounding and
3256 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3257 	 * firing of the timer.  The bias will give us exactly the
3258 	 * 1.5 tick we need.  But, because the bias is
3259 	 * statistical, we have to test that we don't drop below
3260 	 * the minimum feasible timer (which is 2 ticks).
3261 	 */
3262 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3263 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3264 
3265 	/*
3266 	 * We received an ack for a packet that wasn't retransmitted;
3267 	 * it is probably safe to discard any error indications we've
3268 	 * received recently.  This isn't quite right, but close enough
3269 	 * for now (a route might have failed after we sent a segment,
3270 	 * and the return path might not be symmetrical).
3271 	 */
3272 	tp->t_softerror = 0;
3273 }
3274 
3275 
3276 /*
3277  * TCP compressed state engine.  Currently used to hold compressed
3278  * state for SYN_RECEIVED.
3279  */
3280 
3281 u_long	syn_cache_count;
3282 u_int32_t syn_hash1, syn_hash2;
3283 
3284 #define SYN_HASH(sa, sp, dp) \
3285 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3286 				     ((u_int32_t)(sp)))^syn_hash2)))
3287 #ifndef INET6
3288 #define	SYN_HASHALL(hash, src, dst) \
3289 do {									\
3290 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3291 		((const struct sockaddr_in *)(src))->sin_port,		\
3292 		((const struct sockaddr_in *)(dst))->sin_port);		\
3293 } while (/*CONSTCOND*/ 0)
3294 #else
3295 #define SYN_HASH6(sa, sp, dp) \
3296 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3297 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3298 	 & 0x7fffffff)
3299 
3300 #define SYN_HASHALL(hash, src, dst) \
3301 do {									\
3302 	switch ((src)->sa_family) {					\
3303 	case AF_INET:							\
3304 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3305 			((const struct sockaddr_in *)(src))->sin_port,	\
3306 			((const struct sockaddr_in *)(dst))->sin_port);	\
3307 		break;							\
3308 	case AF_INET6:							\
3309 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3310 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3311 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3312 		break;							\
3313 	default:							\
3314 		hash = 0;						\
3315 	}								\
3316 } while (/*CONSTCOND*/0)
3317 #endif /* INET6 */
3318 
3319 static struct pool syn_cache_pool;
3320 
3321 /*
3322  * We don't estimate RTT with SYNs, so each packet starts with the default
3323  * RTT and each timer step has a fixed timeout value.
3324  */
3325 #define	SYN_CACHE_TIMER_ARM(sc)						\
3326 do {									\
3327 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3328 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3329 	    TCPTV_REXMTMAX);						\
3330 	callout_reset(&(sc)->sc_timer,					\
3331 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3332 } while (/*CONSTCOND*/0)
3333 
3334 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3335 
3336 static inline void
3337 syn_cache_rm(struct syn_cache *sc)
3338 {
3339 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3340 	    sc, sc_bucketq);
3341 	sc->sc_tp = NULL;
3342 	LIST_REMOVE(sc, sc_tpq);
3343 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3344 	callout_stop(&sc->sc_timer);
3345 	syn_cache_count--;
3346 }
3347 
3348 static inline void
3349 syn_cache_put(struct syn_cache *sc)
3350 {
3351 	if (sc->sc_ipopts)
3352 		(void) m_free(sc->sc_ipopts);
3353 	rtcache_free(&sc->sc_route);
3354 	if (callout_invoking(&sc->sc_timer))
3355 		sc->sc_flags |= SCF_DEAD;
3356 	else {
3357 		callout_destroy(&sc->sc_timer);
3358 		pool_put(&syn_cache_pool, sc);
3359 	}
3360 }
3361 
3362 void
3363 syn_cache_init(void)
3364 {
3365 	int i;
3366 
3367 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3368 	    "synpl", NULL, IPL_SOFTNET);
3369 
3370 	/* Initialize the hash buckets. */
3371 	for (i = 0; i < tcp_syn_cache_size; i++)
3372 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3373 }
3374 
3375 void
3376 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3377 {
3378 	struct syn_cache_head *scp;
3379 	struct syn_cache *sc2;
3380 	int s;
3381 
3382 	/*
3383 	 * If there are no entries in the hash table, reinitialize
3384 	 * the hash secrets.
3385 	 */
3386 	if (syn_cache_count == 0) {
3387 		syn_hash1 = arc4random();
3388 		syn_hash2 = arc4random();
3389 	}
3390 
3391 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3392 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3393 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3394 
3395 	/*
3396 	 * Make sure that we don't overflow the per-bucket
3397 	 * limit or the total cache size limit.
3398 	 */
3399 	s = splsoftnet();
3400 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3401 		TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW);
3402 		/*
3403 		 * The bucket is full.  Toss the oldest element in the
3404 		 * bucket.  This will be the first entry in the bucket.
3405 		 */
3406 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3407 #ifdef DIAGNOSTIC
3408 		/*
3409 		 * This should never happen; we should always find an
3410 		 * entry in our bucket.
3411 		 */
3412 		if (sc2 == NULL)
3413 			panic("syn_cache_insert: bucketoverflow: impossible");
3414 #endif
3415 		syn_cache_rm(sc2);
3416 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3417 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3418 		struct syn_cache_head *scp2, *sce;
3419 
3420 		TCP_STATINC(TCP_STAT_SC_OVERFLOWED);
3421 		/*
3422 		 * The cache is full.  Toss the oldest entry in the
3423 		 * first non-empty bucket we can find.
3424 		 *
3425 		 * XXX We would really like to toss the oldest
3426 		 * entry in the cache, but we hope that this
3427 		 * condition doesn't happen very often.
3428 		 */
3429 		scp2 = scp;
3430 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3431 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3432 			for (++scp2; scp2 != scp; scp2++) {
3433 				if (scp2 >= sce)
3434 					scp2 = &tcp_syn_cache[0];
3435 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3436 					break;
3437 			}
3438 #ifdef DIAGNOSTIC
3439 			/*
3440 			 * This should never happen; we should always find a
3441 			 * non-empty bucket.
3442 			 */
3443 			if (scp2 == scp)
3444 				panic("syn_cache_insert: cacheoverflow: "
3445 				    "impossible");
3446 #endif
3447 		}
3448 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3449 		syn_cache_rm(sc2);
3450 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3451 	}
3452 
3453 	/*
3454 	 * Initialize the entry's timer.
3455 	 */
3456 	sc->sc_rxttot = 0;
3457 	sc->sc_rxtshift = 0;
3458 	SYN_CACHE_TIMER_ARM(sc);
3459 
3460 	/* Link it from tcpcb entry */
3461 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3462 
3463 	/* Put it into the bucket. */
3464 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3465 	scp->sch_length++;
3466 	syn_cache_count++;
3467 
3468 	TCP_STATINC(TCP_STAT_SC_ADDED);
3469 	splx(s);
3470 }
3471 
3472 /*
3473  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3474  * If we have retransmitted an entry the maximum number of times, expire
3475  * that entry.
3476  */
3477 void
3478 syn_cache_timer(void *arg)
3479 {
3480 	struct syn_cache *sc = arg;
3481 
3482 	mutex_enter(softnet_lock);
3483 	KERNEL_LOCK(1, NULL);
3484 	callout_ack(&sc->sc_timer);
3485 
3486 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3487 		TCP_STATINC(TCP_STAT_SC_DELAYED_FREE);
3488 		callout_destroy(&sc->sc_timer);
3489 		pool_put(&syn_cache_pool, sc);
3490 		KERNEL_UNLOCK_ONE(NULL);
3491 		mutex_exit(softnet_lock);
3492 		return;
3493 	}
3494 
3495 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3496 		/* Drop it -- too many retransmissions. */
3497 		goto dropit;
3498 	}
3499 
3500 	/*
3501 	 * Compute the total amount of time this entry has
3502 	 * been on a queue.  If this entry has been on longer
3503 	 * than the keep alive timer would allow, expire it.
3504 	 */
3505 	sc->sc_rxttot += sc->sc_rxtcur;
3506 	if (sc->sc_rxttot >= tcp_keepinit)
3507 		goto dropit;
3508 
3509 	TCP_STATINC(TCP_STAT_SC_RETRANSMITTED);
3510 	(void) syn_cache_respond(sc, NULL);
3511 
3512 	/* Advance the timer back-off. */
3513 	sc->sc_rxtshift++;
3514 	SYN_CACHE_TIMER_ARM(sc);
3515 
3516 	KERNEL_UNLOCK_ONE(NULL);
3517 	mutex_exit(softnet_lock);
3518 	return;
3519 
3520  dropit:
3521 	TCP_STATINC(TCP_STAT_SC_TIMED_OUT);
3522 	syn_cache_rm(sc);
3523 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3524 	KERNEL_UNLOCK_ONE(NULL);
3525 	mutex_exit(softnet_lock);
3526 }
3527 
3528 /*
3529  * Remove syn cache created by the specified tcb entry,
3530  * because this does not make sense to keep them
3531  * (if there's no tcb entry, syn cache entry will never be used)
3532  */
3533 void
3534 syn_cache_cleanup(struct tcpcb *tp)
3535 {
3536 	struct syn_cache *sc, *nsc;
3537 	int s;
3538 
3539 	s = splsoftnet();
3540 
3541 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3542 		nsc = LIST_NEXT(sc, sc_tpq);
3543 
3544 #ifdef DIAGNOSTIC
3545 		if (sc->sc_tp != tp)
3546 			panic("invalid sc_tp in syn_cache_cleanup");
3547 #endif
3548 		syn_cache_rm(sc);
3549 		syn_cache_put(sc);	/* calls pool_put but see spl above */
3550 	}
3551 	/* just for safety */
3552 	LIST_INIT(&tp->t_sc);
3553 
3554 	splx(s);
3555 }
3556 
3557 /*
3558  * Find an entry in the syn cache.
3559  */
3560 struct syn_cache *
3561 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3562     struct syn_cache_head **headp)
3563 {
3564 	struct syn_cache *sc;
3565 	struct syn_cache_head *scp;
3566 	u_int32_t hash;
3567 	int s;
3568 
3569 	SYN_HASHALL(hash, src, dst);
3570 
3571 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3572 	*headp = scp;
3573 	s = splsoftnet();
3574 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3575 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3576 		if (sc->sc_hash != hash)
3577 			continue;
3578 		if (!memcmp(&sc->sc_src, src, src->sa_len) &&
3579 		    !memcmp(&sc->sc_dst, dst, dst->sa_len)) {
3580 			splx(s);
3581 			return (sc);
3582 		}
3583 	}
3584 	splx(s);
3585 	return (NULL);
3586 }
3587 
3588 /*
3589  * This function gets called when we receive an ACK for a
3590  * socket in the LISTEN state.  We look up the connection
3591  * in the syn cache, and if its there, we pull it out of
3592  * the cache and turn it into a full-blown connection in
3593  * the SYN-RECEIVED state.
3594  *
3595  * The return values may not be immediately obvious, and their effects
3596  * can be subtle, so here they are:
3597  *
3598  *	NULL	SYN was not found in cache; caller should drop the
3599  *		packet and send an RST.
3600  *
3601  *	-1	We were unable to create the new connection, and are
3602  *		aborting it.  An ACK,RST is being sent to the peer
3603  *		(unless we got screwey sequence numbners; see below),
3604  *		because the 3-way handshake has been completed.  Caller
3605  *		should not free the mbuf, since we may be using it.  If
3606  *		we are not, we will free it.
3607  *
3608  *	Otherwise, the return value is a pointer to the new socket
3609  *	associated with the connection.
3610  */
3611 struct socket *
3612 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3613     struct tcphdr *th, unsigned int hlen, unsigned int tlen,
3614     struct socket *so, struct mbuf *m)
3615 {
3616 	struct syn_cache *sc;
3617 	struct syn_cache_head *scp;
3618 	struct inpcb *inp = NULL;
3619 #ifdef INET6
3620 	struct in6pcb *in6p = NULL;
3621 #endif
3622 	struct tcpcb *tp = 0;
3623 	struct mbuf *am;
3624 	int s;
3625 	struct socket *oso;
3626 
3627 	s = splsoftnet();
3628 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3629 		splx(s);
3630 		return (NULL);
3631 	}
3632 
3633 	/*
3634 	 * Verify the sequence and ack numbers.  Try getting the correct
3635 	 * response again.
3636 	 */
3637 	if ((th->th_ack != sc->sc_iss + 1) ||
3638 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3639 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3640 		(void) syn_cache_respond(sc, m);
3641 		splx(s);
3642 		return ((struct socket *)(-1));
3643 	}
3644 
3645 	/* Remove this cache entry */
3646 	syn_cache_rm(sc);
3647 	splx(s);
3648 
3649 	/*
3650 	 * Ok, create the full blown connection, and set things up
3651 	 * as they would have been set up if we had created the
3652 	 * connection when the SYN arrived.  If we can't create
3653 	 * the connection, abort it.
3654 	 */
3655 	/*
3656 	 * inp still has the OLD in_pcb stuff, set the
3657 	 * v6-related flags on the new guy, too.   This is
3658 	 * done particularly for the case where an AF_INET6
3659 	 * socket is bound only to a port, and a v4 connection
3660 	 * comes in on that port.
3661 	 * we also copy the flowinfo from the original pcb
3662 	 * to the new one.
3663 	 */
3664 	oso = so;
3665 	so = sonewconn(so, SS_ISCONNECTED);
3666 	if (so == NULL)
3667 		goto resetandabort;
3668 
3669 	switch (so->so_proto->pr_domain->dom_family) {
3670 #ifdef INET
3671 	case AF_INET:
3672 		inp = sotoinpcb(so);
3673 		break;
3674 #endif
3675 #ifdef INET6
3676 	case AF_INET6:
3677 		in6p = sotoin6pcb(so);
3678 		break;
3679 #endif
3680 	}
3681 	switch (src->sa_family) {
3682 #ifdef INET
3683 	case AF_INET:
3684 		if (inp) {
3685 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3686 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3687 			inp->inp_options = ip_srcroute();
3688 			in_pcbstate(inp, INP_BOUND);
3689 			if (inp->inp_options == NULL) {
3690 				inp->inp_options = sc->sc_ipopts;
3691 				sc->sc_ipopts = NULL;
3692 			}
3693 		}
3694 #ifdef INET6
3695 		else if (in6p) {
3696 			/* IPv4 packet to AF_INET6 socket */
3697 			memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr));
3698 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3699 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3700 				&in6p->in6p_laddr.s6_addr32[3],
3701 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3702 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3703 			in6totcpcb(in6p)->t_family = AF_INET;
3704 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3705 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3706 			else
3707 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3708 			in6_pcbstate(in6p, IN6P_BOUND);
3709 		}
3710 #endif
3711 		break;
3712 #endif
3713 #ifdef INET6
3714 	case AF_INET6:
3715 		if (in6p) {
3716 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3717 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3718 			in6_pcbstate(in6p, IN6P_BOUND);
3719 		}
3720 		break;
3721 #endif
3722 	}
3723 #ifdef INET6
3724 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3725 		struct in6pcb *oin6p = sotoin6pcb(oso);
3726 		/* inherit socket options from the listening socket */
3727 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3728 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3729 			m_freem(in6p->in6p_options);
3730 			in6p->in6p_options = 0;
3731 		}
3732 		ip6_savecontrol(in6p, &in6p->in6p_options,
3733 			mtod(m, struct ip6_hdr *), m);
3734 	}
3735 #endif
3736 
3737 #if defined(IPSEC) || defined(FAST_IPSEC)
3738 	/*
3739 	 * we make a copy of policy, instead of sharing the policy,
3740 	 * for better behavior in terms of SA lookup and dead SA removal.
3741 	 */
3742 	if (inp) {
3743 		/* copy old policy into new socket's */
3744 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3745 			printf("tcp_input: could not copy policy\n");
3746 	}
3747 #ifdef INET6
3748 	else if (in6p) {
3749 		/* copy old policy into new socket's */
3750 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3751 		    in6p->in6p_sp))
3752 			printf("tcp_input: could not copy policy\n");
3753 	}
3754 #endif
3755 #endif
3756 
3757 	/*
3758 	 * Give the new socket our cached route reference.
3759 	 */
3760 	if (inp) {
3761 		rtcache_copy(&inp->inp_route, &sc->sc_route);
3762 		rtcache_free(&sc->sc_route);
3763 	}
3764 #ifdef INET6
3765 	else {
3766 		rtcache_copy(&in6p->in6p_route, &sc->sc_route);
3767 		rtcache_free(&sc->sc_route);
3768 	}
3769 #endif
3770 
3771 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3772 	if (am == NULL)
3773 		goto resetandabort;
3774 	MCLAIM(am, &tcp_mowner);
3775 	am->m_len = src->sa_len;
3776 	bcopy(src, mtod(am, void *), src->sa_len);
3777 	if (inp) {
3778 		if (in_pcbconnect(inp, am, &lwp0)) {
3779 			(void) m_free(am);
3780 			goto resetandabort;
3781 		}
3782 	}
3783 #ifdef INET6
3784 	else if (in6p) {
3785 		if (src->sa_family == AF_INET) {
3786 			/* IPv4 packet to AF_INET6 socket */
3787 			struct sockaddr_in6 *sin6;
3788 			sin6 = mtod(am, struct sockaddr_in6 *);
3789 			am->m_len = sizeof(*sin6);
3790 			memset(sin6, 0, sizeof(*sin6));
3791 			sin6->sin6_family = AF_INET6;
3792 			sin6->sin6_len = sizeof(*sin6);
3793 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3794 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3795 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3796 				&sin6->sin6_addr.s6_addr32[3],
3797 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3798 		}
3799 		if (in6_pcbconnect(in6p, am, NULL)) {
3800 			(void) m_free(am);
3801 			goto resetandabort;
3802 		}
3803 	}
3804 #endif
3805 	else {
3806 		(void) m_free(am);
3807 		goto resetandabort;
3808 	}
3809 	(void) m_free(am);
3810 
3811 	if (inp)
3812 		tp = intotcpcb(inp);
3813 #ifdef INET6
3814 	else if (in6p)
3815 		tp = in6totcpcb(in6p);
3816 #endif
3817 	else
3818 		tp = NULL;
3819 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3820 	if (sc->sc_request_r_scale != 15) {
3821 		tp->requested_s_scale = sc->sc_requested_s_scale;
3822 		tp->request_r_scale = sc->sc_request_r_scale;
3823 		tp->snd_scale = sc->sc_requested_s_scale;
3824 		tp->rcv_scale = sc->sc_request_r_scale;
3825 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3826 	}
3827 	if (sc->sc_flags & SCF_TIMESTAMP)
3828 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3829 	tp->ts_timebase = sc->sc_timebase;
3830 
3831 	tp->t_template = tcp_template(tp);
3832 	if (tp->t_template == 0) {
3833 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3834 		so = NULL;
3835 		m_freem(m);
3836 		goto abort;
3837 	}
3838 
3839 	tp->iss = sc->sc_iss;
3840 	tp->irs = sc->sc_irs;
3841 	tcp_sendseqinit(tp);
3842 	tcp_rcvseqinit(tp);
3843 	tp->t_state = TCPS_SYN_RECEIVED;
3844 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit);
3845 	TCP_STATINC(TCP_STAT_ACCEPTS);
3846 
3847 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3848 		tp->t_flags |= TF_WILL_SACK;
3849 
3850 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3851 		tp->t_flags |= TF_ECN_PERMIT;
3852 
3853 #ifdef TCP_SIGNATURE
3854 	if (sc->sc_flags & SCF_SIGNATURE)
3855 		tp->t_flags |= TF_SIGNATURE;
3856 #endif
3857 
3858 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3859 	tp->t_ourmss = sc->sc_ourmaxseg;
3860 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3861 
3862 	/*
3863 	 * Initialize the initial congestion window.  If we
3864 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3865 	 * to 1 segment (i.e. the Loss Window).
3866 	 */
3867 	if (sc->sc_rxtshift)
3868 		tp->snd_cwnd = tp->t_peermss;
3869 	else {
3870 		int ss = tcp_init_win;
3871 #ifdef INET
3872 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3873 			ss = tcp_init_win_local;
3874 #endif
3875 #ifdef INET6
3876 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3877 			ss = tcp_init_win_local;
3878 #endif
3879 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3880 	}
3881 
3882 	tcp_rmx_rtt(tp);
3883 	tp->snd_wl1 = sc->sc_irs;
3884 	tp->rcv_up = sc->sc_irs + 1;
3885 
3886 	/*
3887 	 * This is what whould have happened in tcp_output() when
3888 	 * the SYN,ACK was sent.
3889 	 */
3890 	tp->snd_up = tp->snd_una;
3891 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3892 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3893 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3894 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3895 	tp->last_ack_sent = tp->rcv_nxt;
3896 	tp->t_partialacks = -1;
3897 	tp->t_dupacks = 0;
3898 
3899 	TCP_STATINC(TCP_STAT_SC_COMPLETED);
3900 	s = splsoftnet();
3901 	syn_cache_put(sc);
3902 	splx(s);
3903 	return (so);
3904 
3905 resetandabort:
3906 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3907 abort:
3908 	if (so != NULL) {
3909 		(void) soqremque(so, 1);
3910 		(void) soabort(so);
3911 		mutex_enter(softnet_lock);
3912 	}
3913 	s = splsoftnet();
3914 	syn_cache_put(sc);
3915 	splx(s);
3916 	TCP_STATINC(TCP_STAT_SC_ABORTED);
3917 	return ((struct socket *)(-1));
3918 }
3919 
3920 /*
3921  * This function is called when we get a RST for a
3922  * non-existent connection, so that we can see if the
3923  * connection is in the syn cache.  If it is, zap it.
3924  */
3925 
3926 void
3927 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3928 {
3929 	struct syn_cache *sc;
3930 	struct syn_cache_head *scp;
3931 	int s = splsoftnet();
3932 
3933 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3934 		splx(s);
3935 		return;
3936 	}
3937 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3938 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3939 		splx(s);
3940 		return;
3941 	}
3942 	syn_cache_rm(sc);
3943 	TCP_STATINC(TCP_STAT_SC_RESET);
3944 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3945 	splx(s);
3946 }
3947 
3948 void
3949 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3950     struct tcphdr *th)
3951 {
3952 	struct syn_cache *sc;
3953 	struct syn_cache_head *scp;
3954 	int s;
3955 
3956 	s = splsoftnet();
3957 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3958 		splx(s);
3959 		return;
3960 	}
3961 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3962 	if (ntohl (th->th_seq) != sc->sc_iss) {
3963 		splx(s);
3964 		return;
3965 	}
3966 
3967 	/*
3968 	 * If we've retransmitted 3 times and this is our second error,
3969 	 * we remove the entry.  Otherwise, we allow it to continue on.
3970 	 * This prevents us from incorrectly nuking an entry during a
3971 	 * spurious network outage.
3972 	 *
3973 	 * See tcp_notify().
3974 	 */
3975 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3976 		sc->sc_flags |= SCF_UNREACH;
3977 		splx(s);
3978 		return;
3979 	}
3980 
3981 	syn_cache_rm(sc);
3982 	TCP_STATINC(TCP_STAT_SC_UNREACH);
3983 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3984 	splx(s);
3985 }
3986 
3987 /*
3988  * Given a LISTEN socket and an inbound SYN request, add
3989  * this to the syn cache, and send back a segment:
3990  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3991  * to the source.
3992  *
3993  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3994  * Doing so would require that we hold onto the data and deliver it
3995  * to the application.  However, if we are the target of a SYN-flood
3996  * DoS attack, an attacker could send data which would eventually
3997  * consume all available buffer space if it were ACKed.  By not ACKing
3998  * the data, we avoid this DoS scenario.
3999  */
4000 
4001 int
4002 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
4003     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
4004     int optlen, struct tcp_opt_info *oi)
4005 {
4006 	struct tcpcb tb, *tp;
4007 	long win;
4008 	struct syn_cache *sc;
4009 	struct syn_cache_head *scp;
4010 	struct mbuf *ipopts;
4011 	struct tcp_opt_info opti;
4012 	int s;
4013 
4014 	tp = sototcpcb(so);
4015 
4016 	memset(&opti, 0, sizeof(opti));
4017 
4018 	/*
4019 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
4020 	 *
4021 	 * Note this check is performed in tcp_input() very early on.
4022 	 */
4023 
4024 	/*
4025 	 * Initialize some local state.
4026 	 */
4027 	win = sbspace(&so->so_rcv);
4028 	if (win > TCP_MAXWIN)
4029 		win = TCP_MAXWIN;
4030 
4031 	switch (src->sa_family) {
4032 #ifdef INET
4033 	case AF_INET:
4034 		/*
4035 		 * Remember the IP options, if any.
4036 		 */
4037 		ipopts = ip_srcroute();
4038 		break;
4039 #endif
4040 	default:
4041 		ipopts = NULL;
4042 	}
4043 
4044 #ifdef TCP_SIGNATURE
4045 	if (optp || (tp->t_flags & TF_SIGNATURE))
4046 #else
4047 	if (optp)
4048 #endif
4049 	{
4050 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4051 #ifdef TCP_SIGNATURE
4052 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
4053 #endif
4054 		tb.t_state = TCPS_LISTEN;
4055 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
4056 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
4057 			return (0);
4058 	} else
4059 		tb.t_flags = 0;
4060 
4061 	/*
4062 	 * See if we already have an entry for this connection.
4063 	 * If we do, resend the SYN,ACK.  We do not count this
4064 	 * as a retransmission (XXX though maybe we should).
4065 	 */
4066 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4067 		TCP_STATINC(TCP_STAT_SC_DUPESYN);
4068 		if (ipopts) {
4069 			/*
4070 			 * If we were remembering a previous source route,
4071 			 * forget it and use the new one we've been given.
4072 			 */
4073 			if (sc->sc_ipopts)
4074 				(void) m_free(sc->sc_ipopts);
4075 			sc->sc_ipopts = ipopts;
4076 		}
4077 		sc->sc_timestamp = tb.ts_recent;
4078 		if (syn_cache_respond(sc, m) == 0) {
4079 			uint64_t *tcps = TCP_STAT_GETREF();
4080 			tcps[TCP_STAT_SNDACKS]++;
4081 			tcps[TCP_STAT_SNDTOTAL]++;
4082 			TCP_STAT_PUTREF();
4083 		}
4084 		return (1);
4085 	}
4086 
4087 	s = splsoftnet();
4088 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4089 	splx(s);
4090 	if (sc == NULL) {
4091 		if (ipopts)
4092 			(void) m_free(ipopts);
4093 		return (0);
4094 	}
4095 
4096 	/*
4097 	 * Fill in the cache, and put the necessary IP and TCP
4098 	 * options into the reply.
4099 	 */
4100 	memset(sc, 0, sizeof(struct syn_cache));
4101 	callout_init(&sc->sc_timer, CALLOUT_MPSAFE);
4102 	bcopy(src, &sc->sc_src, src->sa_len);
4103 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4104 	sc->sc_flags = 0;
4105 	sc->sc_ipopts = ipopts;
4106 	sc->sc_irs = th->th_seq;
4107 	switch (src->sa_family) {
4108 #ifdef INET
4109 	case AF_INET:
4110 	    {
4111 		struct sockaddr_in *srcin = (void *) src;
4112 		struct sockaddr_in *dstin = (void *) dst;
4113 
4114 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
4115 		    &srcin->sin_addr, dstin->sin_port,
4116 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
4117 		break;
4118 	    }
4119 #endif /* INET */
4120 #ifdef INET6
4121 	case AF_INET6:
4122 	    {
4123 		struct sockaddr_in6 *srcin6 = (void *) src;
4124 		struct sockaddr_in6 *dstin6 = (void *) dst;
4125 
4126 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
4127 		    &srcin6->sin6_addr, dstin6->sin6_port,
4128 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
4129 		break;
4130 	    }
4131 #endif /* INET6 */
4132 	}
4133 	sc->sc_peermaxseg = oi->maxseg;
4134 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4135 						m->m_pkthdr.rcvif : NULL,
4136 						sc->sc_src.sa.sa_family);
4137 	sc->sc_win = win;
4138 	sc->sc_timebase = tcp_now - 1;	/* see tcp_newtcpcb() */
4139 	sc->sc_timestamp = tb.ts_recent;
4140 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4141 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4142 		sc->sc_flags |= SCF_TIMESTAMP;
4143 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4144 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4145 		sc->sc_requested_s_scale = tb.requested_s_scale;
4146 		sc->sc_request_r_scale = 0;
4147 		/*
4148 		 * Pick the smallest possible scaling factor that
4149 		 * will still allow us to scale up to sb_max.
4150 		 *
4151 		 * We do this because there are broken firewalls that
4152 		 * will corrupt the window scale option, leading to
4153 		 * the other endpoint believing that our advertised
4154 		 * window is unscaled.  At scale factors larger than
4155 		 * 5 the unscaled window will drop below 1500 bytes,
4156 		 * leading to serious problems when traversing these
4157 		 * broken firewalls.
4158 		 *
4159 		 * With the default sbmax of 256K, a scale factor
4160 		 * of 3 will be chosen by this algorithm.  Those who
4161 		 * choose a larger sbmax should watch out
4162 		 * for the compatiblity problems mentioned above.
4163 		 *
4164 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4165 		 * or <SYN,ACK>) segment itself is never scaled.
4166 		 */
4167 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4168 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4169 			sc->sc_request_r_scale++;
4170 	} else {
4171 		sc->sc_requested_s_scale = 15;
4172 		sc->sc_request_r_scale = 15;
4173 	}
4174 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4175 		sc->sc_flags |= SCF_SACK_PERMIT;
4176 
4177 	/*
4178 	 * ECN setup packet recieved.
4179 	 */
4180 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4181 		sc->sc_flags |= SCF_ECN_PERMIT;
4182 
4183 #ifdef TCP_SIGNATURE
4184 	if (tb.t_flags & TF_SIGNATURE)
4185 		sc->sc_flags |= SCF_SIGNATURE;
4186 #endif
4187 	sc->sc_tp = tp;
4188 	if (syn_cache_respond(sc, m) == 0) {
4189 		uint64_t *tcps = TCP_STAT_GETREF();
4190 		tcps[TCP_STAT_SNDACKS]++;
4191 		tcps[TCP_STAT_SNDTOTAL]++;
4192 		TCP_STAT_PUTREF();
4193 		syn_cache_insert(sc, tp);
4194 	} else {
4195 		s = splsoftnet();
4196 		syn_cache_put(sc);
4197 		splx(s);
4198 		TCP_STATINC(TCP_STAT_SC_DROPPED);
4199 	}
4200 	return (1);
4201 }
4202 
4203 int
4204 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4205 {
4206 #ifdef INET6
4207 	struct rtentry *rt;
4208 #endif
4209 	struct route *ro;
4210 	u_int8_t *optp;
4211 	int optlen, error;
4212 	u_int16_t tlen;
4213 	struct ip *ip = NULL;
4214 #ifdef INET6
4215 	struct ip6_hdr *ip6 = NULL;
4216 #endif
4217 	struct tcpcb *tp = NULL;
4218 	struct tcphdr *th;
4219 	u_int hlen;
4220 	struct socket *so;
4221 
4222 	ro = &sc->sc_route;
4223 	switch (sc->sc_src.sa.sa_family) {
4224 	case AF_INET:
4225 		hlen = sizeof(struct ip);
4226 		break;
4227 #ifdef INET6
4228 	case AF_INET6:
4229 		hlen = sizeof(struct ip6_hdr);
4230 		break;
4231 #endif
4232 	default:
4233 		if (m)
4234 			m_freem(m);
4235 		return (EAFNOSUPPORT);
4236 	}
4237 
4238 	/* Compute the size of the TCP options. */
4239 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4240 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4241 #ifdef TCP_SIGNATURE
4242 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4243 #endif
4244 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4245 
4246 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4247 
4248 	/*
4249 	 * Create the IP+TCP header from scratch.
4250 	 */
4251 	if (m)
4252 		m_freem(m);
4253 #ifdef DIAGNOSTIC
4254 	if (max_linkhdr + tlen > MCLBYTES)
4255 		return (ENOBUFS);
4256 #endif
4257 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4258 	if (m && tlen > MHLEN) {
4259 		MCLGET(m, M_DONTWAIT);
4260 		if ((m->m_flags & M_EXT) == 0) {
4261 			m_freem(m);
4262 			m = NULL;
4263 		}
4264 	}
4265 	if (m == NULL)
4266 		return (ENOBUFS);
4267 	MCLAIM(m, &tcp_tx_mowner);
4268 
4269 	/* Fixup the mbuf. */
4270 	m->m_data += max_linkhdr;
4271 	m->m_len = m->m_pkthdr.len = tlen;
4272 	if (sc->sc_tp) {
4273 		tp = sc->sc_tp;
4274 		if (tp->t_inpcb)
4275 			so = tp->t_inpcb->inp_socket;
4276 #ifdef INET6
4277 		else if (tp->t_in6pcb)
4278 			so = tp->t_in6pcb->in6p_socket;
4279 #endif
4280 		else
4281 			so = NULL;
4282 	} else
4283 		so = NULL;
4284 	m->m_pkthdr.rcvif = NULL;
4285 	memset(mtod(m, u_char *), 0, tlen);
4286 
4287 	switch (sc->sc_src.sa.sa_family) {
4288 	case AF_INET:
4289 		ip = mtod(m, struct ip *);
4290 		ip->ip_v = 4;
4291 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4292 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4293 		ip->ip_p = IPPROTO_TCP;
4294 		th = (struct tcphdr *)(ip + 1);
4295 		th->th_dport = sc->sc_src.sin.sin_port;
4296 		th->th_sport = sc->sc_dst.sin.sin_port;
4297 		break;
4298 #ifdef INET6
4299 	case AF_INET6:
4300 		ip6 = mtod(m, struct ip6_hdr *);
4301 		ip6->ip6_vfc = IPV6_VERSION;
4302 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4303 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4304 		ip6->ip6_nxt = IPPROTO_TCP;
4305 		/* ip6_plen will be updated in ip6_output() */
4306 		th = (struct tcphdr *)(ip6 + 1);
4307 		th->th_dport = sc->sc_src.sin6.sin6_port;
4308 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4309 		break;
4310 #endif
4311 	default:
4312 		th = NULL;
4313 	}
4314 
4315 	th->th_seq = htonl(sc->sc_iss);
4316 	th->th_ack = htonl(sc->sc_irs + 1);
4317 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4318 	th->th_flags = TH_SYN|TH_ACK;
4319 	th->th_win = htons(sc->sc_win);
4320 	/* th_sum already 0 */
4321 	/* th_urp already 0 */
4322 
4323 	/* Tack on the TCP options. */
4324 	optp = (u_int8_t *)(th + 1);
4325 	*optp++ = TCPOPT_MAXSEG;
4326 	*optp++ = 4;
4327 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4328 	*optp++ = sc->sc_ourmaxseg & 0xff;
4329 
4330 	if (sc->sc_request_r_scale != 15) {
4331 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4332 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4333 		    sc->sc_request_r_scale);
4334 		optp += 4;
4335 	}
4336 
4337 	if (sc->sc_flags & SCF_TIMESTAMP) {
4338 		u_int32_t *lp = (u_int32_t *)(optp);
4339 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4340 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4341 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4342 		*lp   = htonl(sc->sc_timestamp);
4343 		optp += TCPOLEN_TSTAMP_APPA;
4344 	}
4345 
4346 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4347 		u_int8_t *p = optp;
4348 
4349 		/* Let the peer know that we will SACK. */
4350 		p[0] = TCPOPT_SACK_PERMITTED;
4351 		p[1] = 2;
4352 		p[2] = TCPOPT_NOP;
4353 		p[3] = TCPOPT_NOP;
4354 		optp += 4;
4355 	}
4356 
4357 	/*
4358 	 * Send ECN SYN-ACK setup packet.
4359 	 * Routes can be asymetric, so, even if we receive a packet
4360 	 * with ECE and CWR set, we must not assume no one will block
4361 	 * the ECE packet we are about to send.
4362 	 */
4363 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4364 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4365 		th->th_flags |= TH_ECE;
4366 		TCP_STATINC(TCP_STAT_ECN_SHS);
4367 
4368 		/*
4369 		 * draft-ietf-tcpm-ecnsyn-00.txt
4370 		 *
4371 		 * "[...] a TCP node MAY respond to an ECN-setup
4372 		 * SYN packet by setting ECT in the responding
4373 		 * ECN-setup SYN/ACK packet, indicating to routers
4374 		 * that the SYN/ACK packet is ECN-Capable.
4375 		 * This allows a congested router along the path
4376 		 * to mark the packet instead of dropping the
4377 		 * packet as an indication of congestion."
4378 		 *
4379 		 * "[...] There can be a great benefit in setting
4380 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4381 		 * Congestion is  most likely to occur in
4382 		 * the server-to-client direction.  As a result,
4383 		 * setting an ECN-capable codepoint in SYN/ACK
4384 		 * packets can reduce the occurence of three-second
4385 		 * retransmit timeouts resulting from the drop
4386 		 * of SYN/ACK packets."
4387 		 *
4388 		 * Page 4 and 6, January 2006.
4389 		 */
4390 
4391 		switch (sc->sc_src.sa.sa_family) {
4392 #ifdef INET
4393 		case AF_INET:
4394 			ip->ip_tos |= IPTOS_ECN_ECT0;
4395 			break;
4396 #endif
4397 #ifdef INET6
4398 		case AF_INET6:
4399 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4400 			break;
4401 #endif
4402 		}
4403 		TCP_STATINC(TCP_STAT_ECN_ECT);
4404 	}
4405 
4406 #ifdef TCP_SIGNATURE
4407 	if (sc->sc_flags & SCF_SIGNATURE) {
4408 		struct secasvar *sav;
4409 		u_int8_t *sigp;
4410 
4411 		sav = tcp_signature_getsav(m, th);
4412 
4413 		if (sav == NULL) {
4414 			if (m)
4415 				m_freem(m);
4416 			return (EPERM);
4417 		}
4418 
4419 		*optp++ = TCPOPT_SIGNATURE;
4420 		*optp++ = TCPOLEN_SIGNATURE;
4421 		sigp = optp;
4422 		memset(optp, 0, TCP_SIGLEN);
4423 		optp += TCP_SIGLEN;
4424 		*optp++ = TCPOPT_NOP;
4425 		*optp++ = TCPOPT_EOL;
4426 
4427 		(void)tcp_signature(m, th, hlen, sav, sigp);
4428 
4429 		key_sa_recordxfer(sav, m);
4430 #ifdef FAST_IPSEC
4431 		KEY_FREESAV(&sav);
4432 #else
4433 		key_freesav(sav);
4434 #endif
4435 	}
4436 #endif
4437 
4438 	/* Compute the packet's checksum. */
4439 	switch (sc->sc_src.sa.sa_family) {
4440 	case AF_INET:
4441 		ip->ip_len = htons(tlen - hlen);
4442 		th->th_sum = 0;
4443 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4444 		break;
4445 #ifdef INET6
4446 	case AF_INET6:
4447 		ip6->ip6_plen = htons(tlen - hlen);
4448 		th->th_sum = 0;
4449 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4450 		break;
4451 #endif
4452 	}
4453 
4454 	/*
4455 	 * Fill in some straggling IP bits.  Note the stack expects
4456 	 * ip_len to be in host order, for convenience.
4457 	 */
4458 	switch (sc->sc_src.sa.sa_family) {
4459 #ifdef INET
4460 	case AF_INET:
4461 		ip->ip_len = htons(tlen);
4462 		ip->ip_ttl = ip_defttl;
4463 		/* XXX tos? */
4464 		break;
4465 #endif
4466 #ifdef INET6
4467 	case AF_INET6:
4468 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4469 		ip6->ip6_vfc |= IPV6_VERSION;
4470 		ip6->ip6_plen = htons(tlen - hlen);
4471 		/* ip6_hlim will be initialized afterwards */
4472 		/* XXX flowlabel? */
4473 		break;
4474 #endif
4475 	}
4476 
4477 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4478 	tp = sc->sc_tp;
4479 
4480 	switch (sc->sc_src.sa.sa_family) {
4481 #ifdef INET
4482 	case AF_INET:
4483 		error = ip_output(m, sc->sc_ipopts, ro,
4484 		    (ip_mtudisc ? IP_MTUDISC : 0),
4485 		    (struct ip_moptions *)NULL, so);
4486 		break;
4487 #endif
4488 #ifdef INET6
4489 	case AF_INET6:
4490 		ip6->ip6_hlim = in6_selecthlim(NULL,
4491 				(rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
4492 				                                    : NULL);
4493 
4494 		error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL);
4495 		break;
4496 #endif
4497 	default:
4498 		error = EAFNOSUPPORT;
4499 		break;
4500 	}
4501 	return (error);
4502 }
4503