xref: /netbsd-src/sys/netinet/tcp_input.c (revision ce2c90c7c172d95d2402a5b3d96d8f8e6d138a21)
1 /*	$NetBSD: tcp_input.c,v 1.250 2006/10/12 11:46:30 rpaulo Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  *      This product includes software developed by the University of
47  *      California, Berkeley and its contributors.
48  *      This product includes software developed at the Information
49  *      Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 /*-
73  * Copyright (c) 1997, 1998, 1999, 2001, 2005, 2006 The NetBSD Foundation, Inc.
74  * All rights reserved.
75  *
76  * This code is derived from software contributed to The NetBSD Foundation
77  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
78  * Facility, NASA Ames Research Center.
79  * This code is derived from software contributed to The NetBSD Foundation
80  * by Charles M. Hannum.
81  * This code is derived from software contributed to The NetBSD Foundation
82  * by Rui Paulo.
83  *
84  * Redistribution and use in source and binary forms, with or without
85  * modification, are permitted provided that the following conditions
86  * are met:
87  * 1. Redistributions of source code must retain the above copyright
88  *    notice, this list of conditions and the following disclaimer.
89  * 2. Redistributions in binary form must reproduce the above copyright
90  *    notice, this list of conditions and the following disclaimer in the
91  *    documentation and/or other materials provided with the distribution.
92  * 3. All advertising materials mentioning features or use of this software
93  *    must display the following acknowledgement:
94  *	This product includes software developed by the NetBSD
95  *	Foundation, Inc. and its contributors.
96  * 4. Neither the name of The NetBSD Foundation nor the names of its
97  *    contributors may be used to endorse or promote products derived
98  *    from this software without specific prior written permission.
99  *
100  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
101  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
102  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
103  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
104  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
105  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
106  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
107  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
108  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
109  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
110  * POSSIBILITY OF SUCH DAMAGE.
111  */
112 
113 /*
114  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
115  *	The Regents of the University of California.  All rights reserved.
116  *
117  * Redistribution and use in source and binary forms, with or without
118  * modification, are permitted provided that the following conditions
119  * are met:
120  * 1. Redistributions of source code must retain the above copyright
121  *    notice, this list of conditions and the following disclaimer.
122  * 2. Redistributions in binary form must reproduce the above copyright
123  *    notice, this list of conditions and the following disclaimer in the
124  *    documentation and/or other materials provided with the distribution.
125  * 3. Neither the name of the University nor the names of its contributors
126  *    may be used to endorse or promote products derived from this software
127  *    without specific prior written permission.
128  *
129  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
130  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
131  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
132  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
133  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
134  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
135  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
136  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
137  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
138  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
139  * SUCH DAMAGE.
140  *
141  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
142  */
143 
144 /*
145  *	TODO list for SYN cache stuff:
146  *
147  *	Find room for a "state" field, which is needed to keep a
148  *	compressed state for TIME_WAIT TCBs.  It's been noted already
149  *	that this is fairly important for very high-volume web and
150  *	mail servers, which use a large number of short-lived
151  *	connections.
152  */
153 
154 #include <sys/cdefs.h>
155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.250 2006/10/12 11:46:30 rpaulo Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 #ifdef TCP_SIGNATURE
175 #include <sys/md5.h>
176 #endif
177 
178 #include <net/if.h>
179 #include <net/route.h>
180 #include <net/if_types.h>
181 
182 #include <netinet/in.h>
183 #include <netinet/in_systm.h>
184 #include <netinet/ip.h>
185 #include <netinet/in_pcb.h>
186 #include <netinet/in_var.h>
187 #include <netinet/ip_var.h>
188 #include <netinet/in_offload.h>
189 
190 #ifdef INET6
191 #ifndef INET
192 #include <netinet/in.h>
193 #endif
194 #include <netinet/ip6.h>
195 #include <netinet6/ip6_var.h>
196 #include <netinet6/in6_pcb.h>
197 #include <netinet6/ip6_var.h>
198 #include <netinet6/in6_var.h>
199 #include <netinet/icmp6.h>
200 #include <netinet6/nd6.h>
201 #ifdef TCP_SIGNATURE
202 #include <netinet6/scope6_var.h>
203 #endif
204 #endif
205 
206 #ifndef INET6
207 /* always need ip6.h for IP6_EXTHDR_GET */
208 #include <netinet/ip6.h>
209 #endif
210 
211 #include <netinet/tcp.h>
212 #include <netinet/tcp_fsm.h>
213 #include <netinet/tcp_seq.h>
214 #include <netinet/tcp_timer.h>
215 #include <netinet/tcp_var.h>
216 #include <netinet/tcpip.h>
217 #include <netinet/tcp_congctl.h>
218 #include <netinet/tcp_debug.h>
219 
220 #include <machine/stdarg.h>
221 
222 #ifdef IPSEC
223 #include <netinet6/ipsec.h>
224 #include <netkey/key.h>
225 #endif /*IPSEC*/
226 #ifdef INET6
227 #include "faith.h"
228 #if defined(NFAITH) && NFAITH > 0
229 #include <net/if_faith.h>
230 #endif
231 #endif	/* IPSEC */
232 
233 #ifdef FAST_IPSEC
234 #include <netipsec/ipsec.h>
235 #include <netipsec/ipsec_var.h>			/* XXX ipsecstat namespace */
236 #include <netipsec/key.h>
237 #ifdef INET6
238 #include <netipsec/ipsec6.h>
239 #endif
240 #endif	/* FAST_IPSEC*/
241 
242 int	tcprexmtthresh = 3;
243 int	tcp_log_refused;
244 
245 static int tcp_rst_ppslim_count = 0;
246 static struct timeval tcp_rst_ppslim_last;
247 static int tcp_ackdrop_ppslim_count = 0;
248 static struct timeval tcp_ackdrop_ppslim_last;
249 
250 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
251 
252 /* for modulo comparisons of timestamps */
253 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
254 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
255 
256 /*
257  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
258  */
259 #ifdef INET6
260 #define ND6_HINT(tp) \
261 do { \
262 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \
263 	    tp->t_in6pcb->in6p_route.ro_rt) { \
264 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
265 	} \
266 } while (/*CONSTCOND*/ 0)
267 #else
268 #define ND6_HINT(tp)
269 #endif
270 
271 /*
272  * Macro to compute ACK transmission behavior.  Delay the ACK unless
273  * we have already delayed an ACK (must send an ACK every two segments).
274  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
275  * option is enabled.
276  */
277 #define	TCP_SETUP_ACK(tp, th) \
278 do { \
279 	if ((tp)->t_flags & TF_DELACK || \
280 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
281 		tp->t_flags |= TF_ACKNOW; \
282 	else \
283 		TCP_SET_DELACK(tp); \
284 } while (/*CONSTCOND*/ 0)
285 
286 #define ICMP_CHECK(tp, th, acked) \
287 do { \
288 	/* \
289 	 * If we had a pending ICMP message that \
290 	 * refers to data that have just been  \
291 	 * acknowledged, disregard the recorded ICMP \
292 	 * message. \
293 	 */ \
294 	if (((tp)->t_flags & TF_PMTUD_PEND) && \
295 	    SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \
296 		(tp)->t_flags &= ~TF_PMTUD_PEND; \
297 \
298 	/* \
299 	 * Keep track of the largest chunk of data \
300 	 * acknowledged since last PMTU update \
301 	 */ \
302 	if ((tp)->t_pmtud_mss_acked < (acked)) \
303 		(tp)->t_pmtud_mss_acked = (acked); \
304 } while (/*CONSTCOND*/ 0)
305 
306 /*
307  * Convert TCP protocol fields to host order for easier processing.
308  */
309 #define	TCP_FIELDS_TO_HOST(th)						\
310 do {									\
311 	NTOHL((th)->th_seq);						\
312 	NTOHL((th)->th_ack);						\
313 	NTOHS((th)->th_win);						\
314 	NTOHS((th)->th_urp);						\
315 } while (/*CONSTCOND*/ 0)
316 
317 /*
318  * ... and reverse the above.
319  */
320 #define	TCP_FIELDS_TO_NET(th)						\
321 do {									\
322 	HTONL((th)->th_seq);						\
323 	HTONL((th)->th_ack);						\
324 	HTONS((th)->th_win);						\
325 	HTONS((th)->th_urp);						\
326 } while (/*CONSTCOND*/ 0)
327 
328 #ifdef TCP_CSUM_COUNTERS
329 #include <sys/device.h>
330 
331 #if defined(INET)
332 extern struct evcnt tcp_hwcsum_ok;
333 extern struct evcnt tcp_hwcsum_bad;
334 extern struct evcnt tcp_hwcsum_data;
335 extern struct evcnt tcp_swcsum;
336 #endif /* defined(INET) */
337 #if defined(INET6)
338 extern struct evcnt tcp6_hwcsum_ok;
339 extern struct evcnt tcp6_hwcsum_bad;
340 extern struct evcnt tcp6_hwcsum_data;
341 extern struct evcnt tcp6_swcsum;
342 #endif /* defined(INET6) */
343 
344 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
345 
346 #else
347 
348 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
349 
350 #endif /* TCP_CSUM_COUNTERS */
351 
352 #ifdef TCP_REASS_COUNTERS
353 #include <sys/device.h>
354 
355 extern struct evcnt tcp_reass_;
356 extern struct evcnt tcp_reass_empty;
357 extern struct evcnt tcp_reass_iteration[8];
358 extern struct evcnt tcp_reass_prependfirst;
359 extern struct evcnt tcp_reass_prepend;
360 extern struct evcnt tcp_reass_insert;
361 extern struct evcnt tcp_reass_inserttail;
362 extern struct evcnt tcp_reass_append;
363 extern struct evcnt tcp_reass_appendtail;
364 extern struct evcnt tcp_reass_overlaptail;
365 extern struct evcnt tcp_reass_overlapfront;
366 extern struct evcnt tcp_reass_segdup;
367 extern struct evcnt tcp_reass_fragdup;
368 
369 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
370 
371 #else
372 
373 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
374 
375 #endif /* TCP_REASS_COUNTERS */
376 
377 #ifdef INET
378 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
379 #endif
380 #ifdef INET6
381 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
382 #endif
383 
384 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
385 
386 POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl", NULL);
387 
388 struct ipqent *
389 tcpipqent_alloc()
390 {
391 	struct ipqent *ipqe;
392 	int s;
393 
394 	s = splvm();
395 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
396 	splx(s);
397 
398 	return ipqe;
399 }
400 
401 void
402 tcpipqent_free(struct ipqent *ipqe)
403 {
404 	int s;
405 
406 	s = splvm();
407 	pool_put(&tcpipqent_pool, ipqe);
408 	splx(s);
409 }
410 
411 int
412 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen)
413 {
414 	struct ipqent *p, *q, *nq, *tiqe = NULL;
415 	struct socket *so = NULL;
416 	int pkt_flags;
417 	tcp_seq pkt_seq;
418 	unsigned pkt_len;
419 	u_long rcvpartdupbyte = 0;
420 	u_long rcvoobyte;
421 #ifdef TCP_REASS_COUNTERS
422 	u_int count = 0;
423 #endif
424 
425 	if (tp->t_inpcb)
426 		so = tp->t_inpcb->inp_socket;
427 #ifdef INET6
428 	else if (tp->t_in6pcb)
429 		so = tp->t_in6pcb->in6p_socket;
430 #endif
431 
432 	TCP_REASS_LOCK_CHECK(tp);
433 
434 	/*
435 	 * Call with th==0 after become established to
436 	 * force pre-ESTABLISHED data up to user socket.
437 	 */
438 	if (th == 0)
439 		goto present;
440 
441 	rcvoobyte = *tlen;
442 	/*
443 	 * Copy these to local variables because the tcpiphdr
444 	 * gets munged while we are collapsing mbufs.
445 	 */
446 	pkt_seq = th->th_seq;
447 	pkt_len = *tlen;
448 	pkt_flags = th->th_flags;
449 
450 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
451 
452 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
453 		/*
454 		 * When we miss a packet, the vast majority of time we get
455 		 * packets that follow it in order.  So optimize for that.
456 		 */
457 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
458 			p->ipqe_len += pkt_len;
459 			p->ipqe_flags |= pkt_flags;
460 			m_cat(p->ipre_mlast, m);
461 			TRAVERSE(p->ipre_mlast);
462 			m = NULL;
463 			tiqe = p;
464 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
465 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
466 			goto skip_replacement;
467 		}
468 		/*
469 		 * While we're here, if the pkt is completely beyond
470 		 * anything we have, just insert it at the tail.
471 		 */
472 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
473 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
474 			goto insert_it;
475 		}
476 	}
477 
478 	q = TAILQ_FIRST(&tp->segq);
479 
480 	if (q != NULL) {
481 		/*
482 		 * If this segment immediately precedes the first out-of-order
483 		 * block, simply slap the segment in front of it and (mostly)
484 		 * skip the complicated logic.
485 		 */
486 		if (pkt_seq + pkt_len == q->ipqe_seq) {
487 			q->ipqe_seq = pkt_seq;
488 			q->ipqe_len += pkt_len;
489 			q->ipqe_flags |= pkt_flags;
490 			m_cat(m, q->ipqe_m);
491 			q->ipqe_m = m;
492 			q->ipre_mlast = m; /* last mbuf may have changed */
493 			TRAVERSE(q->ipre_mlast);
494 			tiqe = q;
495 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
496 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
497 			goto skip_replacement;
498 		}
499 	} else {
500 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
501 	}
502 
503 	/*
504 	 * Find a segment which begins after this one does.
505 	 */
506 	for (p = NULL; q != NULL; q = nq) {
507 		nq = TAILQ_NEXT(q, ipqe_q);
508 #ifdef TCP_REASS_COUNTERS
509 		count++;
510 #endif
511 		/*
512 		 * If the received segment is just right after this
513 		 * fragment, merge the two together and then check
514 		 * for further overlaps.
515 		 */
516 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
517 #ifdef TCPREASS_DEBUG
518 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
519 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
520 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
521 #endif
522 			pkt_len += q->ipqe_len;
523 			pkt_flags |= q->ipqe_flags;
524 			pkt_seq = q->ipqe_seq;
525 			m_cat(q->ipre_mlast, m);
526 			TRAVERSE(q->ipre_mlast);
527 			m = q->ipqe_m;
528 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
529 			goto free_ipqe;
530 		}
531 		/*
532 		 * If the received segment is completely past this
533 		 * fragment, we need to go the next fragment.
534 		 */
535 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
536 			p = q;
537 			continue;
538 		}
539 		/*
540 		 * If the fragment is past the received segment,
541 		 * it (or any following) can't be concatenated.
542 		 */
543 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
544 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
545 			break;
546 		}
547 
548 		/*
549 		 * We've received all the data in this segment before.
550 		 * mark it as a duplicate and return.
551 		 */
552 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
553 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
554 			tcpstat.tcps_rcvduppack++;
555 			tcpstat.tcps_rcvdupbyte += pkt_len;
556 			tcp_new_dsack(tp, pkt_seq, pkt_len);
557 			m_freem(m);
558 			if (tiqe != NULL) {
559 				tcpipqent_free(tiqe);
560 			}
561 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
562 			return (0);
563 		}
564 		/*
565 		 * Received segment completely overlaps this fragment
566 		 * so we drop the fragment (this keeps the temporal
567 		 * ordering of segments correct).
568 		 */
569 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
570 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
571 			rcvpartdupbyte += q->ipqe_len;
572 			m_freem(q->ipqe_m);
573 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
574 			goto free_ipqe;
575 		}
576 		/*
577 		 * RX'ed segment extends past the end of the
578 		 * fragment.  Drop the overlapping bytes.  Then
579 		 * merge the fragment and segment then treat as
580 		 * a longer received packet.
581 		 */
582 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
583 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
584 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
585 #ifdef TCPREASS_DEBUG
586 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
587 			       tp, overlap,
588 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
589 #endif
590 			m_adj(m, overlap);
591 			rcvpartdupbyte += overlap;
592 			m_cat(q->ipre_mlast, m);
593 			TRAVERSE(q->ipre_mlast);
594 			m = q->ipqe_m;
595 			pkt_seq = q->ipqe_seq;
596 			pkt_len += q->ipqe_len - overlap;
597 			rcvoobyte -= overlap;
598 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
599 			goto free_ipqe;
600 		}
601 		/*
602 		 * RX'ed segment extends past the front of the
603 		 * fragment.  Drop the overlapping bytes on the
604 		 * received packet.  The packet will then be
605 		 * contatentated with this fragment a bit later.
606 		 */
607 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
608 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
609 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
610 #ifdef TCPREASS_DEBUG
611 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
612 			       tp, overlap,
613 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
614 #endif
615 			m_adj(m, -overlap);
616 			pkt_len -= overlap;
617 			rcvpartdupbyte += overlap;
618 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
619 			rcvoobyte -= overlap;
620 		}
621 		/*
622 		 * If the received segment immediates precedes this
623 		 * fragment then tack the fragment onto this segment
624 		 * and reinsert the data.
625 		 */
626 		if (q->ipqe_seq == pkt_seq + pkt_len) {
627 #ifdef TCPREASS_DEBUG
628 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
629 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
630 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
631 #endif
632 			pkt_len += q->ipqe_len;
633 			pkt_flags |= q->ipqe_flags;
634 			m_cat(m, q->ipqe_m);
635 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
636 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
637 			tp->t_segqlen--;
638 			KASSERT(tp->t_segqlen >= 0);
639 			KASSERT(tp->t_segqlen != 0 ||
640 			    (TAILQ_EMPTY(&tp->segq) &&
641 			    TAILQ_EMPTY(&tp->timeq)));
642 			if (tiqe == NULL) {
643 				tiqe = q;
644 			} else {
645 				tcpipqent_free(q);
646 			}
647 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
648 			break;
649 		}
650 		/*
651 		 * If the fragment is before the segment, remember it.
652 		 * When this loop is terminated, p will contain the
653 		 * pointer to fragment that is right before the received
654 		 * segment.
655 		 */
656 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
657 			p = q;
658 
659 		continue;
660 
661 		/*
662 		 * This is a common operation.  It also will allow
663 		 * to save doing a malloc/free in most instances.
664 		 */
665 	  free_ipqe:
666 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
667 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
668 		tp->t_segqlen--;
669 		KASSERT(tp->t_segqlen >= 0);
670 		KASSERT(tp->t_segqlen != 0 ||
671 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
672 		if (tiqe == NULL) {
673 			tiqe = q;
674 		} else {
675 			tcpipqent_free(q);
676 		}
677 	}
678 
679 #ifdef TCP_REASS_COUNTERS
680 	if (count > 7)
681 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
682 	else if (count > 0)
683 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
684 #endif
685 
686     insert_it:
687 
688 	/*
689 	 * Allocate a new queue entry since the received segment did not
690 	 * collapse onto any other out-of-order block; thus we are allocating
691 	 * a new block.  If it had collapsed, tiqe would not be NULL and
692 	 * we would be reusing it.
693 	 * XXX If we can't, just drop the packet.  XXX
694 	 */
695 	if (tiqe == NULL) {
696 		tiqe = tcpipqent_alloc();
697 		if (tiqe == NULL) {
698 			tcpstat.tcps_rcvmemdrop++;
699 			m_freem(m);
700 			return (0);
701 		}
702 	}
703 
704 	/*
705 	 * Update the counters.
706 	 */
707 	tcpstat.tcps_rcvoopack++;
708 	tcpstat.tcps_rcvoobyte += rcvoobyte;
709 	if (rcvpartdupbyte) {
710 	    tcpstat.tcps_rcvpartduppack++;
711 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
712 	}
713 
714 	/*
715 	 * Insert the new fragment queue entry into both queues.
716 	 */
717 	tiqe->ipqe_m = m;
718 	tiqe->ipre_mlast = m;
719 	tiqe->ipqe_seq = pkt_seq;
720 	tiqe->ipqe_len = pkt_len;
721 	tiqe->ipqe_flags = pkt_flags;
722 	if (p == NULL) {
723 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
724 #ifdef TCPREASS_DEBUG
725 		if (tiqe->ipqe_seq != tp->rcv_nxt)
726 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
727 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
728 #endif
729 	} else {
730 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
731 #ifdef TCPREASS_DEBUG
732 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
733 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
734 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
735 #endif
736 	}
737 	tp->t_segqlen++;
738 
739 skip_replacement:
740 
741 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
742 
743 present:
744 	/*
745 	 * Present data to user, advancing rcv_nxt through
746 	 * completed sequence space.
747 	 */
748 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
749 		return (0);
750 	q = TAILQ_FIRST(&tp->segq);
751 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
752 		return (0);
753 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
754 		return (0);
755 
756 	tp->rcv_nxt += q->ipqe_len;
757 	pkt_flags = q->ipqe_flags & TH_FIN;
758 	ND6_HINT(tp);
759 
760 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
761 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
762 	tp->t_segqlen--;
763 	KASSERT(tp->t_segqlen >= 0);
764 	KASSERT(tp->t_segqlen != 0 ||
765 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
766 	if (so->so_state & SS_CANTRCVMORE)
767 		m_freem(q->ipqe_m);
768 	else
769 		sbappendstream(&so->so_rcv, q->ipqe_m);
770 	tcpipqent_free(q);
771 	sorwakeup(so);
772 	return (pkt_flags);
773 }
774 
775 #ifdef INET6
776 int
777 tcp6_input(struct mbuf **mp, int *offp, int proto)
778 {
779 	struct mbuf *m = *mp;
780 
781 	/*
782 	 * draft-itojun-ipv6-tcp-to-anycast
783 	 * better place to put this in?
784 	 */
785 	if (m->m_flags & M_ANYCAST6) {
786 		struct ip6_hdr *ip6;
787 		if (m->m_len < sizeof(struct ip6_hdr)) {
788 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
789 				tcpstat.tcps_rcvshort++;
790 				return IPPROTO_DONE;
791 			}
792 		}
793 		ip6 = mtod(m, struct ip6_hdr *);
794 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
795 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
796 		return IPPROTO_DONE;
797 	}
798 
799 	tcp_input(m, *offp, proto);
800 	return IPPROTO_DONE;
801 }
802 #endif
803 
804 #ifdef INET
805 static void
806 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
807 {
808 	char src[4*sizeof "123"];
809 	char dst[4*sizeof "123"];
810 
811 	if (ip) {
812 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
813 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
814 	}
815 	else {
816 		strlcpy(src, "(unknown)", sizeof(src));
817 		strlcpy(dst, "(unknown)", sizeof(dst));
818 	}
819 	log(LOG_INFO,
820 	    "Connection attempt to TCP %s:%d from %s:%d\n",
821 	    dst, ntohs(th->th_dport),
822 	    src, ntohs(th->th_sport));
823 }
824 #endif
825 
826 #ifdef INET6
827 static void
828 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
829 {
830 	char src[INET6_ADDRSTRLEN];
831 	char dst[INET6_ADDRSTRLEN];
832 
833 	if (ip6) {
834 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
835 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
836 	}
837 	else {
838 		strlcpy(src, "(unknown v6)", sizeof(src));
839 		strlcpy(dst, "(unknown v6)", sizeof(dst));
840 	}
841 	log(LOG_INFO,
842 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
843 	    dst, ntohs(th->th_dport),
844 	    src, ntohs(th->th_sport));
845 }
846 #endif
847 
848 /*
849  * Checksum extended TCP header and data.
850  */
851 int
852 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th __unused,
853     int toff, int off, int tlen)
854 {
855 
856 	/*
857 	 * XXX it's better to record and check if this mbuf is
858 	 * already checked.
859 	 */
860 
861 	switch (af) {
862 #ifdef INET
863 	case AF_INET:
864 		switch (m->m_pkthdr.csum_flags &
865 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
866 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
867 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
868 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
869 			goto badcsum;
870 
871 		case M_CSUM_TCPv4|M_CSUM_DATA: {
872 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
873 
874 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
875 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
876 				const struct ip *ip =
877 				    mtod(m, const struct ip *);
878 
879 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
880 				    ip->ip_dst.s_addr,
881 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
882 			}
883 			if ((hw_csum ^ 0xffff) != 0)
884 				goto badcsum;
885 			break;
886 		}
887 
888 		case M_CSUM_TCPv4:
889 			/* Checksum was okay. */
890 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
891 			break;
892 
893 		default:
894 			/*
895 			 * Must compute it ourselves.  Maybe skip checksum
896 			 * on loopback interfaces.
897 			 */
898 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
899 					     IFF_LOOPBACK) ||
900 					   tcp_do_loopback_cksum)) {
901 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
902 				if (in4_cksum(m, IPPROTO_TCP, toff,
903 					      tlen + off) != 0)
904 					goto badcsum;
905 			}
906 			break;
907 		}
908 		break;
909 #endif /* INET4 */
910 
911 #ifdef INET6
912 	case AF_INET6:
913 		switch (m->m_pkthdr.csum_flags &
914 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
915 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
916 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
917 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
918 			goto badcsum;
919 
920 #if 0 /* notyet */
921 		case M_CSUM_TCPv6|M_CSUM_DATA:
922 #endif
923 
924 		case M_CSUM_TCPv6:
925 			/* Checksum was okay. */
926 			TCP_CSUM_COUNTER_INCR(&tcp6_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_flags & M_LOOP) == 0 ||
935 			    tcp_do_loopback_cksum)) {
936 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
937 				if (in6_cksum(m, IPPROTO_TCP, toff,
938 				    tlen + off) != 0)
939 					goto badcsum;
940 			}
941 		}
942 		break;
943 #endif /* INET6 */
944 	}
945 
946 	return 0;
947 
948 badcsum:
949 	tcpstat.tcps_rcvbadsum++;
950 	return -1;
951 }
952 
953 /*
954  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
955  */
956 void
957 tcp_input(struct mbuf *m, ...)
958 {
959 	struct tcphdr *th;
960 	struct ip *ip;
961 	struct inpcb *inp;
962 #ifdef INET6
963 	struct ip6_hdr *ip6;
964 	struct in6pcb *in6p;
965 #endif
966 	u_int8_t *optp = NULL;
967 	int optlen = 0;
968 	int len, tlen, toff, hdroptlen = 0;
969 	struct tcpcb *tp = 0;
970 	int tiflags;
971 	struct socket *so = NULL;
972 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
973 #ifdef TCP_DEBUG
974 	short ostate = 0;
975 #endif
976 	u_long tiwin;
977 	struct tcp_opt_info opti;
978 	int off, iphlen;
979 	va_list ap;
980 	int af;		/* af on the wire */
981 	struct mbuf *tcp_saveti = NULL;
982 	uint32_t ts_rtt;
983 	uint8_t iptos;
984 
985 	MCLAIM(m, &tcp_rx_mowner);
986 	va_start(ap, m);
987 	toff = va_arg(ap, int);
988 	(void)va_arg(ap, int);		/* ignore value, advance ap */
989 	va_end(ap);
990 
991 	tcpstat.tcps_rcvtotal++;
992 
993 	bzero(&opti, sizeof(opti));
994 	opti.ts_present = 0;
995 	opti.maxseg = 0;
996 
997 	/*
998 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
999 	 *
1000 	 * TCP is, by definition, unicast, so we reject all
1001 	 * multicast outright.
1002 	 *
1003 	 * Note, there are additional src/dst address checks in
1004 	 * the AF-specific code below.
1005 	 */
1006 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1007 		/* XXX stat */
1008 		goto drop;
1009 	}
1010 #ifdef INET6
1011 	if (m->m_flags & M_ANYCAST6) {
1012 		/* XXX stat */
1013 		goto drop;
1014 	}
1015 #endif
1016 
1017 	/*
1018 	 * Get IP and TCP header.
1019 	 * Note: IP leaves IP header in first mbuf.
1020 	 */
1021 	ip = mtod(m, struct ip *);
1022 #ifdef INET6
1023 	ip6 = NULL;
1024 #endif
1025 	switch (ip->ip_v) {
1026 #ifdef INET
1027 	case 4:
1028 		af = AF_INET;
1029 		iphlen = sizeof(struct ip);
1030 		ip = mtod(m, struct ip *);
1031 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1032 			sizeof(struct tcphdr));
1033 		if (th == NULL) {
1034 			tcpstat.tcps_rcvshort++;
1035 			return;
1036 		}
1037 		/* We do the checksum after PCB lookup... */
1038 		len = ntohs(ip->ip_len);
1039 		tlen = len - toff;
1040 		iptos = ip->ip_tos;
1041 		break;
1042 #endif
1043 #ifdef INET6
1044 	case 6:
1045 		ip = NULL;
1046 		iphlen = sizeof(struct ip6_hdr);
1047 		af = AF_INET6;
1048 		ip6 = mtod(m, struct ip6_hdr *);
1049 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1050 			sizeof(struct tcphdr));
1051 		if (th == NULL) {
1052 			tcpstat.tcps_rcvshort++;
1053 			return;
1054 		}
1055 
1056 		/* Be proactive about malicious use of IPv4 mapped address */
1057 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1058 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1059 			/* XXX stat */
1060 			goto drop;
1061 		}
1062 
1063 		/*
1064 		 * Be proactive about unspecified IPv6 address in source.
1065 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1066 		 * unspecified IPv6 address can be used to confuse us.
1067 		 *
1068 		 * Note that packets with unspecified IPv6 destination is
1069 		 * already dropped in ip6_input.
1070 		 */
1071 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1072 			/* XXX stat */
1073 			goto drop;
1074 		}
1075 
1076 		/*
1077 		 * Make sure destination address is not multicast.
1078 		 * Source address checked in ip6_input().
1079 		 */
1080 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1081 			/* XXX stat */
1082 			goto drop;
1083 		}
1084 
1085 		/* We do the checksum after PCB lookup... */
1086 		len = m->m_pkthdr.len;
1087 		tlen = len - toff;
1088 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1089 		break;
1090 #endif
1091 	default:
1092 		m_freem(m);
1093 		return;
1094 	}
1095 
1096 	KASSERT(TCP_HDR_ALIGNED_P(th));
1097 
1098 	/*
1099 	 * Check that TCP offset makes sense,
1100 	 * pull out TCP options and adjust length.		XXX
1101 	 */
1102 	off = th->th_off << 2;
1103 	if (off < sizeof (struct tcphdr) || off > tlen) {
1104 		tcpstat.tcps_rcvbadoff++;
1105 		goto drop;
1106 	}
1107 	tlen -= off;
1108 
1109 	/*
1110 	 * tcp_input() has been modified to use tlen to mean the TCP data
1111 	 * length throughout the function.  Other functions can use
1112 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1113 	 * rja
1114 	 */
1115 
1116 	if (off > sizeof (struct tcphdr)) {
1117 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1118 		if (th == NULL) {
1119 			tcpstat.tcps_rcvshort++;
1120 			return;
1121 		}
1122 		/*
1123 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1124 		 * (as they're before toff) and we don't need to update those.
1125 		 */
1126 		KASSERT(TCP_HDR_ALIGNED_P(th));
1127 		optlen = off - sizeof (struct tcphdr);
1128 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1129 		/*
1130 		 * Do quick retrieval of timestamp options ("options
1131 		 * prediction?").  If timestamp is the only option and it's
1132 		 * formatted as recommended in RFC 1323 appendix A, we
1133 		 * quickly get the values now and not bother calling
1134 		 * tcp_dooptions(), etc.
1135 		 */
1136 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1137 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1138 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1139 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1140 		     (th->th_flags & TH_SYN) == 0) {
1141 			opti.ts_present = 1;
1142 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1143 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1144 			optp = NULL;	/* we've parsed the options */
1145 		}
1146 	}
1147 	tiflags = th->th_flags;
1148 
1149 	/*
1150 	 * Locate pcb for segment.
1151 	 */
1152 findpcb:
1153 	inp = NULL;
1154 #ifdef INET6
1155 	in6p = NULL;
1156 #endif
1157 	switch (af) {
1158 #ifdef INET
1159 	case AF_INET:
1160 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1161 		    ip->ip_dst, th->th_dport);
1162 		if (inp == 0) {
1163 			++tcpstat.tcps_pcbhashmiss;
1164 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1165 		}
1166 #ifdef INET6
1167 		if (inp == 0) {
1168 			struct in6_addr s, d;
1169 
1170 			/* mapped addr case */
1171 			bzero(&s, sizeof(s));
1172 			s.s6_addr16[5] = htons(0xffff);
1173 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1174 			bzero(&d, sizeof(d));
1175 			d.s6_addr16[5] = htons(0xffff);
1176 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1177 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1178 			    th->th_sport, &d, th->th_dport, 0);
1179 			if (in6p == 0) {
1180 				++tcpstat.tcps_pcbhashmiss;
1181 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1182 				    th->th_dport, 0);
1183 			}
1184 		}
1185 #endif
1186 #ifndef INET6
1187 		if (inp == 0)
1188 #else
1189 		if (inp == 0 && in6p == 0)
1190 #endif
1191 		{
1192 			++tcpstat.tcps_noport;
1193 			if (tcp_log_refused &&
1194 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1195 				tcp4_log_refused(ip, th);
1196 			}
1197 			TCP_FIELDS_TO_HOST(th);
1198 			goto dropwithreset_ratelim;
1199 		}
1200 #if defined(IPSEC) || defined(FAST_IPSEC)
1201 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1202 		    ipsec4_in_reject(m, inp)) {
1203 			ipsecstat.in_polvio++;
1204 			goto drop;
1205 		}
1206 #ifdef INET6
1207 		else if (in6p &&
1208 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1209 		    ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1210 			ipsecstat.in_polvio++;
1211 			goto drop;
1212 		}
1213 #endif
1214 #endif /*IPSEC*/
1215 		break;
1216 #endif /*INET*/
1217 #ifdef INET6
1218 	case AF_INET6:
1219 	    {
1220 		int faith;
1221 
1222 #if defined(NFAITH) && NFAITH > 0
1223 		faith = faithprefix(&ip6->ip6_dst);
1224 #else
1225 		faith = 0;
1226 #endif
1227 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1228 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1229 		if (in6p == NULL) {
1230 			++tcpstat.tcps_pcbhashmiss;
1231 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1232 				th->th_dport, faith);
1233 		}
1234 		if (in6p == NULL) {
1235 			++tcpstat.tcps_noport;
1236 			if (tcp_log_refused &&
1237 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1238 				tcp6_log_refused(ip6, th);
1239 			}
1240 			TCP_FIELDS_TO_HOST(th);
1241 			goto dropwithreset_ratelim;
1242 		}
1243 #if defined(IPSEC) || defined(FAST_IPSEC)
1244 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1245 		    ipsec6_in_reject(m, in6p)) {
1246 			ipsec6stat.in_polvio++;
1247 			goto drop;
1248 		}
1249 #endif /*IPSEC*/
1250 		break;
1251 	    }
1252 #endif
1253 	}
1254 
1255 	/*
1256 	 * If the state is CLOSED (i.e., TCB does not exist) then
1257 	 * all data in the incoming segment is discarded.
1258 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1259 	 * but should either do a listen or a connect soon.
1260 	 */
1261 	tp = NULL;
1262 	so = NULL;
1263 	if (inp) {
1264 		tp = intotcpcb(inp);
1265 		so = inp->inp_socket;
1266 	}
1267 #ifdef INET6
1268 	else if (in6p) {
1269 		tp = in6totcpcb(in6p);
1270 		so = in6p->in6p_socket;
1271 	}
1272 #endif
1273 	if (tp == 0) {
1274 		TCP_FIELDS_TO_HOST(th);
1275 		goto dropwithreset_ratelim;
1276 	}
1277 	if (tp->t_state == TCPS_CLOSED)
1278 		goto drop;
1279 
1280 	/*
1281 	 * Checksum extended TCP header and data.
1282 	 */
1283 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1284 		goto badcsum;
1285 
1286 	TCP_FIELDS_TO_HOST(th);
1287 
1288 	/* Unscale the window into a 32-bit value. */
1289 	if ((tiflags & TH_SYN) == 0)
1290 		tiwin = th->th_win << tp->snd_scale;
1291 	else
1292 		tiwin = th->th_win;
1293 
1294 #ifdef INET6
1295 	/* save packet options if user wanted */
1296 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1297 		if (in6p->in6p_options) {
1298 			m_freem(in6p->in6p_options);
1299 			in6p->in6p_options = 0;
1300 		}
1301 		KASSERT(ip6 != NULL);
1302 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1303 	}
1304 #endif
1305 
1306 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1307 		union syn_cache_sa src;
1308 		union syn_cache_sa dst;
1309 
1310 		bzero(&src, sizeof(src));
1311 		bzero(&dst, sizeof(dst));
1312 		switch (af) {
1313 #ifdef INET
1314 		case AF_INET:
1315 			src.sin.sin_len = sizeof(struct sockaddr_in);
1316 			src.sin.sin_family = AF_INET;
1317 			src.sin.sin_addr = ip->ip_src;
1318 			src.sin.sin_port = th->th_sport;
1319 
1320 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1321 			dst.sin.sin_family = AF_INET;
1322 			dst.sin.sin_addr = ip->ip_dst;
1323 			dst.sin.sin_port = th->th_dport;
1324 			break;
1325 #endif
1326 #ifdef INET6
1327 		case AF_INET6:
1328 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1329 			src.sin6.sin6_family = AF_INET6;
1330 			src.sin6.sin6_addr = ip6->ip6_src;
1331 			src.sin6.sin6_port = th->th_sport;
1332 
1333 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1334 			dst.sin6.sin6_family = AF_INET6;
1335 			dst.sin6.sin6_addr = ip6->ip6_dst;
1336 			dst.sin6.sin6_port = th->th_dport;
1337 			break;
1338 #endif /* INET6 */
1339 		default:
1340 			goto badsyn;	/*sanity*/
1341 		}
1342 
1343 		if (so->so_options & SO_DEBUG) {
1344 #ifdef TCP_DEBUG
1345 			ostate = tp->t_state;
1346 #endif
1347 
1348 			tcp_saveti = NULL;
1349 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1350 				goto nosave;
1351 
1352 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1353 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1354 				if (!tcp_saveti)
1355 					goto nosave;
1356 			} else {
1357 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1358 				if (!tcp_saveti)
1359 					goto nosave;
1360 				MCLAIM(m, &tcp_mowner);
1361 				tcp_saveti->m_len = iphlen;
1362 				m_copydata(m, 0, iphlen,
1363 				    mtod(tcp_saveti, caddr_t));
1364 			}
1365 
1366 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1367 				m_freem(tcp_saveti);
1368 				tcp_saveti = NULL;
1369 			} else {
1370 				tcp_saveti->m_len += sizeof(struct tcphdr);
1371 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1372 				    sizeof(struct tcphdr));
1373 			}
1374 	nosave:;
1375 		}
1376 		if (so->so_options & SO_ACCEPTCONN) {
1377 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1378 				if (tiflags & TH_RST) {
1379 					syn_cache_reset(&src.sa, &dst.sa, th);
1380 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1381 				    (TH_ACK|TH_SYN)) {
1382 					/*
1383 					 * Received a SYN,ACK.  This should
1384 					 * never happen while we are in
1385 					 * LISTEN.  Send an RST.
1386 					 */
1387 					goto badsyn;
1388 				} else if (tiflags & TH_ACK) {
1389 					so = syn_cache_get(&src.sa, &dst.sa,
1390 						th, toff, tlen, so, m);
1391 					if (so == NULL) {
1392 						/*
1393 						 * We don't have a SYN for
1394 						 * this ACK; send an RST.
1395 						 */
1396 						goto badsyn;
1397 					} else if (so ==
1398 					    (struct socket *)(-1)) {
1399 						/*
1400 						 * We were unable to create
1401 						 * the connection.  If the
1402 						 * 3-way handshake was
1403 						 * completed, and RST has
1404 						 * been sent to the peer.
1405 						 * Since the mbuf might be
1406 						 * in use for the reply,
1407 						 * do not free it.
1408 						 */
1409 						m = NULL;
1410 					} else {
1411 						/*
1412 						 * We have created a
1413 						 * full-blown connection.
1414 						 */
1415 						tp = NULL;
1416 						inp = NULL;
1417 #ifdef INET6
1418 						in6p = NULL;
1419 #endif
1420 						switch (so->so_proto->pr_domain->dom_family) {
1421 #ifdef INET
1422 						case AF_INET:
1423 							inp = sotoinpcb(so);
1424 							tp = intotcpcb(inp);
1425 							break;
1426 #endif
1427 #ifdef INET6
1428 						case AF_INET6:
1429 							in6p = sotoin6pcb(so);
1430 							tp = in6totcpcb(in6p);
1431 							break;
1432 #endif
1433 						}
1434 						if (tp == NULL)
1435 							goto badsyn;	/*XXX*/
1436 						tiwin <<= tp->snd_scale;
1437 						goto after_listen;
1438 					}
1439 				} else {
1440 					/*
1441 					 * None of RST, SYN or ACK was set.
1442 					 * This is an invalid packet for a
1443 					 * TCB in LISTEN state.  Send a RST.
1444 					 */
1445 					goto badsyn;
1446 				}
1447 			} else {
1448 				/*
1449 				 * Received a SYN.
1450 				 *
1451 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1452 				 */
1453 				if (m->m_flags & (M_BCAST|M_MCAST))
1454 					goto drop;
1455 
1456 				switch (af) {
1457 #ifdef INET6
1458 				case AF_INET6:
1459 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1460 						goto drop;
1461 					break;
1462 #endif /* INET6 */
1463 				case AF_INET:
1464 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1465 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1466 						goto drop;
1467 				break;
1468 				}
1469 
1470 #ifdef INET6
1471 				/*
1472 				 * If deprecated address is forbidden, we do
1473 				 * not accept SYN to deprecated interface
1474 				 * address to prevent any new inbound
1475 				 * connection from getting established.
1476 				 * When we do not accept SYN, we send a TCP
1477 				 * RST, with deprecated source address (instead
1478 				 * of dropping it).  We compromise it as it is
1479 				 * much better for peer to send a RST, and
1480 				 * RST will be the final packet for the
1481 				 * exchange.
1482 				 *
1483 				 * If we do not forbid deprecated addresses, we
1484 				 * accept the SYN packet.  RFC2462 does not
1485 				 * suggest dropping SYN in this case.
1486 				 * If we decipher RFC2462 5.5.4, it says like
1487 				 * this:
1488 				 * 1. use of deprecated addr with existing
1489 				 *    communication is okay - "SHOULD continue
1490 				 *    to be used"
1491 				 * 2. use of it with new communication:
1492 				 *   (2a) "SHOULD NOT be used if alternate
1493 				 *        address with sufficient scope is
1494 				 *        available"
1495 				 *   (2b) nothing mentioned otherwise.
1496 				 * Here we fall into (2b) case as we have no
1497 				 * choice in our source address selection - we
1498 				 * must obey the peer.
1499 				 *
1500 				 * The wording in RFC2462 is confusing, and
1501 				 * there are multiple description text for
1502 				 * deprecated address handling - worse, they
1503 				 * are not exactly the same.  I believe 5.5.4
1504 				 * is the best one, so we follow 5.5.4.
1505 				 */
1506 				if (af == AF_INET6 && !ip6_use_deprecated) {
1507 					struct in6_ifaddr *ia6;
1508 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1509 					    &ip6->ip6_dst)) &&
1510 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1511 						tp = NULL;
1512 						goto dropwithreset;
1513 					}
1514 				}
1515 #endif
1516 
1517 #ifdef IPSEC
1518 				switch (af) {
1519 #ifdef INET
1520 				case AF_INET:
1521 					if (ipsec4_in_reject_so(m, so)) {
1522 						ipsecstat.in_polvio++;
1523 						tp = NULL;
1524 						goto dropwithreset;
1525 					}
1526 					break;
1527 #endif
1528 #ifdef INET6
1529 				case AF_INET6:
1530 					if (ipsec6_in_reject_so(m, so)) {
1531 						ipsec6stat.in_polvio++;
1532 						tp = NULL;
1533 						goto dropwithreset;
1534 					}
1535 					break;
1536 #endif
1537 				}
1538 #endif
1539 
1540 				/*
1541 				 * LISTEN socket received a SYN
1542 				 * from itself?  This can't possibly
1543 				 * be valid; drop the packet.
1544 				 */
1545 				if (th->th_sport == th->th_dport) {
1546 					int i;
1547 
1548 					switch (af) {
1549 #ifdef INET
1550 					case AF_INET:
1551 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1552 						break;
1553 #endif
1554 #ifdef INET6
1555 					case AF_INET6:
1556 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1557 						break;
1558 #endif
1559 					default:
1560 						i = 1;
1561 					}
1562 					if (i) {
1563 						tcpstat.tcps_badsyn++;
1564 						goto drop;
1565 					}
1566 				}
1567 
1568 				/*
1569 				 * SYN looks ok; create compressed TCP
1570 				 * state for it.
1571 				 */
1572 				if (so->so_qlen <= so->so_qlimit &&
1573 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1574 						so, m, optp, optlen, &opti))
1575 					m = NULL;
1576 			}
1577 			goto drop;
1578 		}
1579 	}
1580 
1581 after_listen:
1582 #ifdef DIAGNOSTIC
1583 	/*
1584 	 * Should not happen now that all embryonic connections
1585 	 * are handled with compressed state.
1586 	 */
1587 	if (tp->t_state == TCPS_LISTEN)
1588 		panic("tcp_input: TCPS_LISTEN");
1589 #endif
1590 
1591 	/*
1592 	 * Segment received on connection.
1593 	 * Reset idle time and keep-alive timer.
1594 	 */
1595 	tp->t_rcvtime = tcp_now;
1596 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1597 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1598 
1599 	/*
1600 	 * Process options.
1601 	 */
1602 #ifdef TCP_SIGNATURE
1603 	if (optp || (tp->t_flags & TF_SIGNATURE))
1604 #else
1605 	if (optp)
1606 #endif
1607 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1608 			goto drop;
1609 
1610 	if (TCP_SACK_ENABLED(tp)) {
1611 		tcp_del_sackholes(tp, th);
1612 	}
1613 
1614 	if (TCP_ECN_ALLOWED(tp)) {
1615 		switch (iptos & IPTOS_ECN_MASK) {
1616 		case IPTOS_ECN_CE:
1617 			tp->t_flags |= TF_ECN_SND_ECE;
1618 			tcpstat.tcps_ecn_ce++;
1619 			break;
1620 		case IPTOS_ECN_ECT0:
1621 			tcpstat.tcps_ecn_ect++;
1622 			break;
1623 		case IPTOS_ECN_ECT1:
1624 			/* XXX: ignore for now -- rpaulo */
1625 			break;
1626 		}
1627 
1628 		if (tiflags & TH_CWR)
1629 			tp->t_flags &= ~TF_ECN_SND_ECE;
1630 
1631 		/*
1632 		 * Congestion experienced.
1633 		 * Ignore if we are already trying to recover.
1634 		 */
1635 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1636 			tcp_reno_congestion_exp(tp);
1637 	}
1638 
1639 	if (opti.ts_present && opti.ts_ecr) {
1640 		/*
1641 		 * Calculate the RTT from the returned time stamp and the
1642 		 * connection's time base.  If the time stamp is later than
1643 		 * the current time, or is extremely old, fall back to non-1323
1644 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1645 		 * at the same time.
1646 		 */
1647 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1648 		if (ts_rtt > TCP_PAWS_IDLE)
1649 			ts_rtt = 0;
1650 	} else {
1651 		ts_rtt = 0;
1652 	}
1653 
1654 	/*
1655 	 * Header prediction: check for the two common cases
1656 	 * of a uni-directional data xfer.  If the packet has
1657 	 * no control flags, is in-sequence, the window didn't
1658 	 * change and we're not retransmitting, it's a
1659 	 * candidate.  If the length is zero and the ack moved
1660 	 * forward, we're the sender side of the xfer.  Just
1661 	 * free the data acked & wake any higher level process
1662 	 * that was blocked waiting for space.  If the length
1663 	 * is non-zero and the ack didn't move, we're the
1664 	 * receiver side.  If we're getting packets in-order
1665 	 * (the reassembly queue is empty), add the data to
1666 	 * the socket buffer and note that we need a delayed ack.
1667 	 */
1668 	if (tp->t_state == TCPS_ESTABLISHED &&
1669 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1670 	        == TH_ACK &&
1671 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1672 	    th->th_seq == tp->rcv_nxt &&
1673 	    tiwin && tiwin == tp->snd_wnd &&
1674 	    tp->snd_nxt == tp->snd_max) {
1675 
1676 		/*
1677 		 * If last ACK falls within this segment's sequence numbers,
1678 		 *  record the timestamp.
1679 		 * NOTE:
1680 		 * 1) That the test incorporates suggestions from the latest
1681 		 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1682 		 * 2) That updating only on newer timestamps interferes with
1683 		 *    our earlier PAWS tests, so this check should be solely
1684 		 *    predicated on the sequence space of this segment.
1685 		 * 3) That we modify the segment boundary check to be
1686 		 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
1687 		 *    instead of RFC1323's
1688 		 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
1689 		 *    This modified check allows us to overcome RFC1323's
1690 		 *    limitations as described in Stevens TCP/IP Illustrated
1691 		 *    Vol. 2 p.869. In such cases, we can still calculate the
1692 		 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1693 		 */
1694 		if (opti.ts_present &&
1695 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1696 		    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1697 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1698 			tp->ts_recent_age = tcp_now;
1699 			tp->ts_recent = opti.ts_val;
1700 		}
1701 
1702 		if (tlen == 0) {
1703 			/* Ack prediction. */
1704 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1705 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1706 			    tp->snd_cwnd >= tp->snd_wnd &&
1707 			    tp->t_partialacks < 0) {
1708 				/*
1709 				 * this is a pure ack for outstanding data.
1710 				 */
1711 				++tcpstat.tcps_predack;
1712 				if (ts_rtt)
1713 					tcp_xmit_timer(tp, ts_rtt);
1714 				else if (tp->t_rtttime &&
1715 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1716 					tcp_xmit_timer(tp,
1717 					  tcp_now - tp->t_rtttime);
1718 				acked = th->th_ack - tp->snd_una;
1719 				tcpstat.tcps_rcvackpack++;
1720 				tcpstat.tcps_rcvackbyte += acked;
1721 				ND6_HINT(tp);
1722 
1723 				if (acked > (tp->t_lastoff - tp->t_inoff))
1724 					tp->t_lastm = NULL;
1725 				sbdrop(&so->so_snd, acked);
1726 				tp->t_lastoff -= acked;
1727 
1728 				ICMP_CHECK(tp, th, acked);
1729 
1730 				tp->snd_una = th->th_ack;
1731 				tp->snd_fack = tp->snd_una;
1732 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1733 					tp->snd_high = tp->snd_una;
1734 				m_freem(m);
1735 
1736 				/*
1737 				 * If all outstanding data are acked, stop
1738 				 * retransmit timer, otherwise restart timer
1739 				 * using current (possibly backed-off) value.
1740 				 * If process is waiting for space,
1741 				 * wakeup/selwakeup/signal.  If data
1742 				 * are ready to send, let tcp_output
1743 				 * decide between more output or persist.
1744 				 */
1745 				if (tp->snd_una == tp->snd_max)
1746 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1747 				else if (TCP_TIMER_ISARMED(tp,
1748 				    TCPT_PERSIST) == 0)
1749 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1750 					    tp->t_rxtcur);
1751 
1752 				sowwakeup(so);
1753 				if (so->so_snd.sb_cc)
1754 					(void) tcp_output(tp);
1755 				if (tcp_saveti)
1756 					m_freem(tcp_saveti);
1757 				return;
1758 			}
1759 		} else if (th->th_ack == tp->snd_una &&
1760 		    TAILQ_FIRST(&tp->segq) == NULL &&
1761 		    tlen <= sbspace(&so->so_rcv)) {
1762 			/*
1763 			 * this is a pure, in-sequence data packet
1764 			 * with nothing on the reassembly queue and
1765 			 * we have enough buffer space to take it.
1766 			 */
1767 			++tcpstat.tcps_preddat;
1768 			tp->rcv_nxt += tlen;
1769 			tcpstat.tcps_rcvpack++;
1770 			tcpstat.tcps_rcvbyte += tlen;
1771 			ND6_HINT(tp);
1772 			/*
1773 			 * Drop TCP, IP headers and TCP options then add data
1774 			 * to socket buffer.
1775 			 */
1776 			if (so->so_state & SS_CANTRCVMORE)
1777 				m_freem(m);
1778 			else {
1779 				m_adj(m, toff + off);
1780 				sbappendstream(&so->so_rcv, m);
1781 			}
1782 			sorwakeup(so);
1783 			TCP_SETUP_ACK(tp, th);
1784 			if (tp->t_flags & TF_ACKNOW)
1785 				(void) tcp_output(tp);
1786 			if (tcp_saveti)
1787 				m_freem(tcp_saveti);
1788 			return;
1789 		}
1790 	}
1791 
1792 	/*
1793 	 * Compute mbuf offset to TCP data segment.
1794 	 */
1795 	hdroptlen = toff + off;
1796 
1797 	/*
1798 	 * Calculate amount of space in receive window,
1799 	 * and then do TCP input processing.
1800 	 * Receive window is amount of space in rcv queue,
1801 	 * but not less than advertised window.
1802 	 */
1803 	{ int win;
1804 
1805 	win = sbspace(&so->so_rcv);
1806 	if (win < 0)
1807 		win = 0;
1808 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1809 	}
1810 
1811 	switch (tp->t_state) {
1812 	/*
1813 	 * If the state is SYN_SENT:
1814 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1815 	 *	if seg contains a RST, then drop the connection.
1816 	 *	if seg does not contain SYN, then drop it.
1817 	 * Otherwise this is an acceptable SYN segment
1818 	 *	initialize tp->rcv_nxt and tp->irs
1819 	 *	if seg contains ack then advance tp->snd_una
1820 	 *	if seg contains a ECE and ECN support is enabled, the stream
1821 	 *	    is ECN capable.
1822 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1823 	 *	arrange for segment to be acked (eventually)
1824 	 *	continue processing rest of data/controls, beginning with URG
1825 	 */
1826 	case TCPS_SYN_SENT:
1827 		if ((tiflags & TH_ACK) &&
1828 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1829 		     SEQ_GT(th->th_ack, tp->snd_max)))
1830 			goto dropwithreset;
1831 		if (tiflags & TH_RST) {
1832 			if (tiflags & TH_ACK)
1833 				tp = tcp_drop(tp, ECONNREFUSED);
1834 			goto drop;
1835 		}
1836 		if ((tiflags & TH_SYN) == 0)
1837 			goto drop;
1838 		if (tiflags & TH_ACK) {
1839 			tp->snd_una = th->th_ack;
1840 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1841 				tp->snd_nxt = tp->snd_una;
1842 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1843 				tp->snd_high = tp->snd_una;
1844 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1845 
1846 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1847 				tp->t_flags |= TF_ECN_PERMIT;
1848 				tcpstat.tcps_ecn_shs++;
1849 			}
1850 
1851 		}
1852 		tp->irs = th->th_seq;
1853 		tcp_rcvseqinit(tp);
1854 		tp->t_flags |= TF_ACKNOW;
1855 		tcp_mss_from_peer(tp, opti.maxseg);
1856 
1857 		/*
1858 		 * Initialize the initial congestion window.  If we
1859 		 * had to retransmit the SYN, we must initialize cwnd
1860 		 * to 1 segment (i.e. the Loss Window).
1861 		 */
1862 		if (tp->t_flags & TF_SYN_REXMT)
1863 			tp->snd_cwnd = tp->t_peermss;
1864 		else {
1865 			int ss = tcp_init_win;
1866 #ifdef INET
1867 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1868 				ss = tcp_init_win_local;
1869 #endif
1870 #ifdef INET6
1871 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1872 				ss = tcp_init_win_local;
1873 #endif
1874 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1875 		}
1876 
1877 		tcp_rmx_rtt(tp);
1878 		if (tiflags & TH_ACK) {
1879 			tcpstat.tcps_connects++;
1880 			soisconnected(so);
1881 			tcp_established(tp);
1882 			/* Do window scaling on this connection? */
1883 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1884 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1885 				tp->snd_scale = tp->requested_s_scale;
1886 				tp->rcv_scale = tp->request_r_scale;
1887 			}
1888 			TCP_REASS_LOCK(tp);
1889 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1890 			TCP_REASS_UNLOCK(tp);
1891 			/*
1892 			 * if we didn't have to retransmit the SYN,
1893 			 * use its rtt as our initial srtt & rtt var.
1894 			 */
1895 			if (tp->t_rtttime)
1896 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1897 		} else
1898 			tp->t_state = TCPS_SYN_RECEIVED;
1899 
1900 		/*
1901 		 * Advance th->th_seq to correspond to first data byte.
1902 		 * If data, trim to stay within window,
1903 		 * dropping FIN if necessary.
1904 		 */
1905 		th->th_seq++;
1906 		if (tlen > tp->rcv_wnd) {
1907 			todrop = tlen - tp->rcv_wnd;
1908 			m_adj(m, -todrop);
1909 			tlen = tp->rcv_wnd;
1910 			tiflags &= ~TH_FIN;
1911 			tcpstat.tcps_rcvpackafterwin++;
1912 			tcpstat.tcps_rcvbyteafterwin += todrop;
1913 		}
1914 		tp->snd_wl1 = th->th_seq - 1;
1915 		tp->rcv_up = th->th_seq;
1916 		goto step6;
1917 
1918 	/*
1919 	 * If the state is SYN_RECEIVED:
1920 	 *	If seg contains an ACK, but not for our SYN, drop the input
1921 	 *	and generate an RST.  See page 36, rfc793
1922 	 */
1923 	case TCPS_SYN_RECEIVED:
1924 		if ((tiflags & TH_ACK) &&
1925 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1926 		     SEQ_GT(th->th_ack, tp->snd_max)))
1927 			goto dropwithreset;
1928 		break;
1929 	}
1930 
1931 	/*
1932 	 * States other than LISTEN or SYN_SENT.
1933 	 * First check timestamp, if present.
1934 	 * Then check that at least some bytes of segment are within
1935 	 * receive window.  If segment begins before rcv_nxt,
1936 	 * drop leading data (and SYN); if nothing left, just ack.
1937 	 *
1938 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1939 	 * and it's less than ts_recent, drop it.
1940 	 */
1941 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1942 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1943 
1944 		/* Check to see if ts_recent is over 24 days old.  */
1945 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
1946 			/*
1947 			 * Invalidate ts_recent.  If this segment updates
1948 			 * ts_recent, the age will be reset later and ts_recent
1949 			 * will get a valid value.  If it does not, setting
1950 			 * ts_recent to zero will at least satisfy the
1951 			 * requirement that zero be placed in the timestamp
1952 			 * echo reply when ts_recent isn't valid.  The
1953 			 * age isn't reset until we get a valid ts_recent
1954 			 * because we don't want out-of-order segments to be
1955 			 * dropped when ts_recent is old.
1956 			 */
1957 			tp->ts_recent = 0;
1958 		} else {
1959 			tcpstat.tcps_rcvduppack++;
1960 			tcpstat.tcps_rcvdupbyte += tlen;
1961 			tcpstat.tcps_pawsdrop++;
1962 			tcp_new_dsack(tp, th->th_seq, tlen);
1963 			goto dropafterack;
1964 		}
1965 	}
1966 
1967 	todrop = tp->rcv_nxt - th->th_seq;
1968 	dupseg = FALSE;
1969 	if (todrop > 0) {
1970 		if (tiflags & TH_SYN) {
1971 			tiflags &= ~TH_SYN;
1972 			th->th_seq++;
1973 			if (th->th_urp > 1)
1974 				th->th_urp--;
1975 			else {
1976 				tiflags &= ~TH_URG;
1977 				th->th_urp = 0;
1978 			}
1979 			todrop--;
1980 		}
1981 		if (todrop > tlen ||
1982 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1983 			/*
1984 			 * Any valid FIN or RST must be to the left of the
1985 			 * window.  At this point the FIN or RST must be a
1986 			 * duplicate or out of sequence; drop it.
1987 			 */
1988 			if (tiflags & TH_RST)
1989 				goto drop;
1990 			tiflags &= ~(TH_FIN|TH_RST);
1991 			/*
1992 			 * Send an ACK to resynchronize and drop any data.
1993 			 * But keep on processing for RST or ACK.
1994 			 */
1995 			tp->t_flags |= TF_ACKNOW;
1996 			todrop = tlen;
1997 			dupseg = TRUE;
1998 			tcpstat.tcps_rcvdupbyte += todrop;
1999 			tcpstat.tcps_rcvduppack++;
2000 		} else if ((tiflags & TH_RST) &&
2001 			   th->th_seq != tp->last_ack_sent) {
2002 			/*
2003 			 * Test for reset before adjusting the sequence
2004 			 * number for overlapping data.
2005 			 */
2006 			goto dropafterack_ratelim;
2007 		} else {
2008 			tcpstat.tcps_rcvpartduppack++;
2009 			tcpstat.tcps_rcvpartdupbyte += todrop;
2010 		}
2011 		tcp_new_dsack(tp, th->th_seq, todrop);
2012 		hdroptlen += todrop;	/*drop from head afterwards*/
2013 		th->th_seq += todrop;
2014 		tlen -= todrop;
2015 		if (th->th_urp > todrop)
2016 			th->th_urp -= todrop;
2017 		else {
2018 			tiflags &= ~TH_URG;
2019 			th->th_urp = 0;
2020 		}
2021 	}
2022 
2023 	/*
2024 	 * If new data are received on a connection after the
2025 	 * user processes are gone, then RST the other end.
2026 	 */
2027 	if ((so->so_state & SS_NOFDREF) &&
2028 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2029 		tp = tcp_close(tp);
2030 		tcpstat.tcps_rcvafterclose++;
2031 		goto dropwithreset;
2032 	}
2033 
2034 	/*
2035 	 * If segment ends after window, drop trailing data
2036 	 * (and PUSH and FIN); if nothing left, just ACK.
2037 	 */
2038 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2039 	if (todrop > 0) {
2040 		tcpstat.tcps_rcvpackafterwin++;
2041 		if (todrop >= tlen) {
2042 			/*
2043 			 * The segment actually starts after the window.
2044 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2045 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2046 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2047 			 */
2048 			tcpstat.tcps_rcvbyteafterwin += tlen;
2049 			/*
2050 			 * If a new connection request is received
2051 			 * while in TIME_WAIT, drop the old connection
2052 			 * and start over if the sequence numbers
2053 			 * are above the previous ones.
2054 			 *
2055 			 * NOTE: We will checksum the packet again, and
2056 			 * so we need to put the header fields back into
2057 			 * network order!
2058 			 * XXX This kind of sucks, but we don't expect
2059 			 * XXX this to happen very often, so maybe it
2060 			 * XXX doesn't matter so much.
2061 			 */
2062 			if (tiflags & TH_SYN &&
2063 			    tp->t_state == TCPS_TIME_WAIT &&
2064 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2065 				tp = tcp_close(tp);
2066 				TCP_FIELDS_TO_NET(th);
2067 				goto findpcb;
2068 			}
2069 			/*
2070 			 * If window is closed can only take segments at
2071 			 * window edge, and have to drop data and PUSH from
2072 			 * incoming segments.  Continue processing, but
2073 			 * remember to ack.  Otherwise, drop segment
2074 			 * and (if not RST) ack.
2075 			 */
2076 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2077 				tp->t_flags |= TF_ACKNOW;
2078 				tcpstat.tcps_rcvwinprobe++;
2079 			} else
2080 				goto dropafterack;
2081 		} else
2082 			tcpstat.tcps_rcvbyteafterwin += todrop;
2083 		m_adj(m, -todrop);
2084 		tlen -= todrop;
2085 		tiflags &= ~(TH_PUSH|TH_FIN);
2086 	}
2087 
2088 	/*
2089 	 * If last ACK falls within this segment's sequence numbers,
2090 	 * and the timestamp is newer, record it.
2091 	 */
2092 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
2093 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2094 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
2095 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2096 		tp->ts_recent_age = tcp_now;
2097 		tp->ts_recent = opti.ts_val;
2098 	}
2099 
2100 	/*
2101 	 * If the RST bit is set examine the state:
2102 	 *    SYN_RECEIVED STATE:
2103 	 *	If passive open, return to LISTEN state.
2104 	 *	If active open, inform user that connection was refused.
2105 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2106 	 *	Inform user that connection was reset, and close tcb.
2107 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2108 	 *	Close the tcb.
2109 	 */
2110 	if (tiflags & TH_RST) {
2111 		if (th->th_seq != tp->last_ack_sent)
2112 			goto dropafterack_ratelim;
2113 
2114 		switch (tp->t_state) {
2115 		case TCPS_SYN_RECEIVED:
2116 			so->so_error = ECONNREFUSED;
2117 			goto close;
2118 
2119 		case TCPS_ESTABLISHED:
2120 		case TCPS_FIN_WAIT_1:
2121 		case TCPS_FIN_WAIT_2:
2122 		case TCPS_CLOSE_WAIT:
2123 			so->so_error = ECONNRESET;
2124 		close:
2125 			tp->t_state = TCPS_CLOSED;
2126 			tcpstat.tcps_drops++;
2127 			tp = tcp_close(tp);
2128 			goto drop;
2129 
2130 		case TCPS_CLOSING:
2131 		case TCPS_LAST_ACK:
2132 		case TCPS_TIME_WAIT:
2133 			tp = tcp_close(tp);
2134 			goto drop;
2135 		}
2136 	}
2137 
2138 	/*
2139 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2140 	 * we must be in a synchronized state.  RFC791 states (under RST
2141 	 * generation) that any unacceptable segment (an out-of-order SYN
2142 	 * qualifies) received in a synchronized state must elicit only an
2143 	 * empty acknowledgment segment ... and the connection remains in
2144 	 * the same state.
2145 	 */
2146 	if (tiflags & TH_SYN) {
2147 		if (tp->rcv_nxt == th->th_seq) {
2148 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2149 			    TH_ACK);
2150 			if (tcp_saveti)
2151 				m_freem(tcp_saveti);
2152 			return;
2153 		}
2154 
2155 		goto dropafterack_ratelim;
2156 	}
2157 
2158 	/*
2159 	 * If the ACK bit is off we drop the segment and return.
2160 	 */
2161 	if ((tiflags & TH_ACK) == 0) {
2162 		if (tp->t_flags & TF_ACKNOW)
2163 			goto dropafterack;
2164 		else
2165 			goto drop;
2166 	}
2167 
2168 	/*
2169 	 * Ack processing.
2170 	 */
2171 	switch (tp->t_state) {
2172 
2173 	/*
2174 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2175 	 * ESTABLISHED state and continue processing, otherwise
2176 	 * send an RST.
2177 	 */
2178 	case TCPS_SYN_RECEIVED:
2179 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2180 		    SEQ_GT(th->th_ack, tp->snd_max))
2181 			goto dropwithreset;
2182 		tcpstat.tcps_connects++;
2183 		soisconnected(so);
2184 		tcp_established(tp);
2185 		/* Do window scaling? */
2186 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2187 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2188 			tp->snd_scale = tp->requested_s_scale;
2189 			tp->rcv_scale = tp->request_r_scale;
2190 		}
2191 		TCP_REASS_LOCK(tp);
2192 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2193 		TCP_REASS_UNLOCK(tp);
2194 		tp->snd_wl1 = th->th_seq - 1;
2195 		/* fall into ... */
2196 
2197 	/*
2198 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2199 	 * ACKs.  If the ack is in the range
2200 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2201 	 * then advance tp->snd_una to th->th_ack and drop
2202 	 * data from the retransmission queue.  If this ACK reflects
2203 	 * more up to date window information we update our window information.
2204 	 */
2205 	case TCPS_ESTABLISHED:
2206 	case TCPS_FIN_WAIT_1:
2207 	case TCPS_FIN_WAIT_2:
2208 	case TCPS_CLOSE_WAIT:
2209 	case TCPS_CLOSING:
2210 	case TCPS_LAST_ACK:
2211 	case TCPS_TIME_WAIT:
2212 
2213 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2214 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2215 				tcpstat.tcps_rcvdupack++;
2216 				/*
2217 				 * If we have outstanding data (other than
2218 				 * a window probe), this is a completely
2219 				 * duplicate ack (ie, window info didn't
2220 				 * change), the ack is the biggest we've
2221 				 * seen and we've seen exactly our rexmt
2222 				 * threshhold of them, assume a packet
2223 				 * has been dropped and retransmit it.
2224 				 * Kludge snd_nxt & the congestion
2225 				 * window so we send only this one
2226 				 * packet.
2227 				 *
2228 				 * We know we're losing at the current
2229 				 * window size so do congestion avoidance
2230 				 * (set ssthresh to half the current window
2231 				 * and pull our congestion window back to
2232 				 * the new ssthresh).
2233 				 *
2234 				 * Dup acks mean that packets have left the
2235 				 * network (they're now cached at the receiver)
2236 				 * so bump cwnd by the amount in the receiver
2237 				 * to keep a constant cwnd packets in the
2238 				 * network.
2239 				 *
2240 				 * When using TCP ECN, notify the peer that
2241 				 * we reduced the cwnd.
2242 				 *
2243 				 * If we are using TCP/SACK, then enter
2244 				 * Fast Recovery if the receiver SACKs
2245 				 * data that is tcprexmtthresh * MSS
2246 				 * bytes past the last ACKed segment,
2247 				 * irrespective of the number of DupAcks.
2248 				 */
2249 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2250 				    th->th_ack != tp->snd_una)
2251 					tp->t_dupacks = 0;
2252 				else if (tp->t_partialacks < 0 &&
2253 					 (++tp->t_dupacks == tcprexmtthresh ||
2254 					 TCP_FACK_FASTRECOV(tp))) {
2255 					/*
2256 					 * Do the fast retransmit, and adjust
2257 					 * congestion control paramenters.
2258 					 */
2259 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2260 						/* False fast retransmit */
2261 						break;
2262 					} else
2263 						goto drop;
2264 				} else if (tp->t_dupacks > tcprexmtthresh) {
2265 					tp->snd_cwnd += tp->t_segsz;
2266 					(void) tcp_output(tp);
2267 					goto drop;
2268 				}
2269 			} else {
2270 				/*
2271 				 * If the ack appears to be very old, only
2272 				 * allow data that is in-sequence.  This
2273 				 * makes it somewhat more difficult to insert
2274 				 * forged data by guessing sequence numbers.
2275 				 * Sent an ack to try to update the send
2276 				 * sequence number on the other side.
2277 				 */
2278 				if (tlen && th->th_seq != tp->rcv_nxt &&
2279 				    SEQ_LT(th->th_ack,
2280 				    tp->snd_una - tp->max_sndwnd))
2281 					goto dropafterack;
2282 			}
2283 			break;
2284 		}
2285 		/*
2286 		 * If the congestion window was inflated to account
2287 		 * for the other side's cached packets, retract it.
2288 		 */
2289 		/* XXX: make SACK have his own congestion control
2290 		 * struct -- rpaulo */
2291 		if (TCP_SACK_ENABLED(tp))
2292 			tcp_sack_newack(tp, th);
2293 		else
2294 			tp->t_congctl->fast_retransmit_newack(tp, th);
2295 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2296 			tcpstat.tcps_rcvacktoomuch++;
2297 			goto dropafterack;
2298 		}
2299 		acked = th->th_ack - tp->snd_una;
2300 		tcpstat.tcps_rcvackpack++;
2301 		tcpstat.tcps_rcvackbyte += acked;
2302 
2303 		/*
2304 		 * If we have a timestamp reply, update smoothed
2305 		 * round trip time.  If no timestamp is present but
2306 		 * transmit timer is running and timed sequence
2307 		 * number was acked, update smoothed round trip time.
2308 		 * Since we now have an rtt measurement, cancel the
2309 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2310 		 * Recompute the initial retransmit timer.
2311 		 */
2312 		if (ts_rtt)
2313 			tcp_xmit_timer(tp, ts_rtt);
2314 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2315 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2316 
2317 		/*
2318 		 * If all outstanding data is acked, stop retransmit
2319 		 * timer and remember to restart (more output or persist).
2320 		 * If there is more data to be acked, restart retransmit
2321 		 * timer, using current (possibly backed-off) value.
2322 		 */
2323 		if (th->th_ack == tp->snd_max) {
2324 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2325 			needoutput = 1;
2326 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2327 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2328 
2329 		/*
2330 		 * New data has been acked, adjust the congestion window.
2331 		 */
2332 		tp->t_congctl->newack(tp, th);
2333 
2334 		ND6_HINT(tp);
2335 		if (acked > so->so_snd.sb_cc) {
2336 			tp->snd_wnd -= so->so_snd.sb_cc;
2337 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2338 			ourfinisacked = 1;
2339 		} else {
2340 			if (acked > (tp->t_lastoff - tp->t_inoff))
2341 				tp->t_lastm = NULL;
2342 			sbdrop(&so->so_snd, acked);
2343 			tp->t_lastoff -= acked;
2344 			tp->snd_wnd -= acked;
2345 			ourfinisacked = 0;
2346 		}
2347 		sowwakeup(so);
2348 
2349 		ICMP_CHECK(tp, th, acked);
2350 
2351 		tp->snd_una = th->th_ack;
2352 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2353 			tp->snd_fack = tp->snd_una;
2354 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2355 			tp->snd_nxt = tp->snd_una;
2356 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2357 			tp->snd_high = tp->snd_una;
2358 
2359 		switch (tp->t_state) {
2360 
2361 		/*
2362 		 * In FIN_WAIT_1 STATE in addition to the processing
2363 		 * for the ESTABLISHED state if our FIN is now acknowledged
2364 		 * then enter FIN_WAIT_2.
2365 		 */
2366 		case TCPS_FIN_WAIT_1:
2367 			if (ourfinisacked) {
2368 				/*
2369 				 * If we can't receive any more
2370 				 * data, then closing user can proceed.
2371 				 * Starting the timer is contrary to the
2372 				 * specification, but if we don't get a FIN
2373 				 * we'll hang forever.
2374 				 */
2375 				if (so->so_state & SS_CANTRCVMORE) {
2376 					soisdisconnected(so);
2377 					if (tcp_maxidle > 0)
2378 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2379 						    tcp_maxidle);
2380 				}
2381 				tp->t_state = TCPS_FIN_WAIT_2;
2382 			}
2383 			break;
2384 
2385 	 	/*
2386 		 * In CLOSING STATE in addition to the processing for
2387 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2388 		 * then enter the TIME-WAIT state, otherwise ignore
2389 		 * the segment.
2390 		 */
2391 		case TCPS_CLOSING:
2392 			if (ourfinisacked) {
2393 				tp->t_state = TCPS_TIME_WAIT;
2394 				tcp_canceltimers(tp);
2395 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2396 				soisdisconnected(so);
2397 			}
2398 			break;
2399 
2400 		/*
2401 		 * In LAST_ACK, we may still be waiting for data to drain
2402 		 * and/or to be acked, as well as for the ack of our FIN.
2403 		 * If our FIN is now acknowledged, delete the TCB,
2404 		 * enter the closed state and return.
2405 		 */
2406 		case TCPS_LAST_ACK:
2407 			if (ourfinisacked) {
2408 				tp = tcp_close(tp);
2409 				goto drop;
2410 			}
2411 			break;
2412 
2413 		/*
2414 		 * In TIME_WAIT state the only thing that should arrive
2415 		 * is a retransmission of the remote FIN.  Acknowledge
2416 		 * it and restart the finack timer.
2417 		 */
2418 		case TCPS_TIME_WAIT:
2419 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2420 			goto dropafterack;
2421 		}
2422 	}
2423 
2424 step6:
2425 	/*
2426 	 * Update window information.
2427 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2428 	 */
2429 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2430 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2431 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2432 		/* keep track of pure window updates */
2433 		if (tlen == 0 &&
2434 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2435 			tcpstat.tcps_rcvwinupd++;
2436 		tp->snd_wnd = tiwin;
2437 		tp->snd_wl1 = th->th_seq;
2438 		tp->snd_wl2 = th->th_ack;
2439 		if (tp->snd_wnd > tp->max_sndwnd)
2440 			tp->max_sndwnd = tp->snd_wnd;
2441 		needoutput = 1;
2442 	}
2443 
2444 	/*
2445 	 * Process segments with URG.
2446 	 */
2447 	if ((tiflags & TH_URG) && th->th_urp &&
2448 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2449 		/*
2450 		 * This is a kludge, but if we receive and accept
2451 		 * random urgent pointers, we'll crash in
2452 		 * soreceive.  It's hard to imagine someone
2453 		 * actually wanting to send this much urgent data.
2454 		 */
2455 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2456 			th->th_urp = 0;			/* XXX */
2457 			tiflags &= ~TH_URG;		/* XXX */
2458 			goto dodata;			/* XXX */
2459 		}
2460 		/*
2461 		 * If this segment advances the known urgent pointer,
2462 		 * then mark the data stream.  This should not happen
2463 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2464 		 * a FIN has been received from the remote side.
2465 		 * In these states we ignore the URG.
2466 		 *
2467 		 * According to RFC961 (Assigned Protocols),
2468 		 * the urgent pointer points to the last octet
2469 		 * of urgent data.  We continue, however,
2470 		 * to consider it to indicate the first octet
2471 		 * of data past the urgent section as the original
2472 		 * spec states (in one of two places).
2473 		 */
2474 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2475 			tp->rcv_up = th->th_seq + th->th_urp;
2476 			so->so_oobmark = so->so_rcv.sb_cc +
2477 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2478 			if (so->so_oobmark == 0)
2479 				so->so_state |= SS_RCVATMARK;
2480 			sohasoutofband(so);
2481 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2482 		}
2483 		/*
2484 		 * Remove out of band data so doesn't get presented to user.
2485 		 * This can happen independent of advancing the URG pointer,
2486 		 * but if two URG's are pending at once, some out-of-band
2487 		 * data may creep in... ick.
2488 		 */
2489 		if (th->th_urp <= (u_int16_t) tlen
2490 #ifdef SO_OOBINLINE
2491 		     && (so->so_options & SO_OOBINLINE) == 0
2492 #endif
2493 		     )
2494 			tcp_pulloutofband(so, th, m, hdroptlen);
2495 	} else
2496 		/*
2497 		 * If no out of band data is expected,
2498 		 * pull receive urgent pointer along
2499 		 * with the receive window.
2500 		 */
2501 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2502 			tp->rcv_up = tp->rcv_nxt;
2503 dodata:							/* XXX */
2504 
2505 	/*
2506 	 * Process the segment text, merging it into the TCP sequencing queue,
2507 	 * and arranging for acknowledgement of receipt if necessary.
2508 	 * This process logically involves adjusting tp->rcv_wnd as data
2509 	 * is presented to the user (this happens in tcp_usrreq.c,
2510 	 * case PRU_RCVD).  If a FIN has already been received on this
2511 	 * connection then we just ignore the text.
2512 	 */
2513 	if ((tlen || (tiflags & TH_FIN)) &&
2514 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2515 		/*
2516 		 * Insert segment ti into reassembly queue of tcp with
2517 		 * control block tp.  Return TH_FIN if reassembly now includes
2518 		 * a segment with FIN.  The macro form does the common case
2519 		 * inline (segment is the next to be received on an
2520 		 * established connection, and the queue is empty),
2521 		 * avoiding linkage into and removal from the queue and
2522 		 * repetition of various conversions.
2523 		 * Set DELACK for segments received in order, but ack
2524 		 * immediately when segments are out of order
2525 		 * (so fast retransmit can work).
2526 		 */
2527 		/* NOTE: this was TCP_REASS() macro, but used only once */
2528 		TCP_REASS_LOCK(tp);
2529 		if (th->th_seq == tp->rcv_nxt &&
2530 		    TAILQ_FIRST(&tp->segq) == NULL &&
2531 		    tp->t_state == TCPS_ESTABLISHED) {
2532 			TCP_SETUP_ACK(tp, th);
2533 			tp->rcv_nxt += tlen;
2534 			tiflags = th->th_flags & TH_FIN;
2535 			tcpstat.tcps_rcvpack++;
2536 			tcpstat.tcps_rcvbyte += tlen;
2537 			ND6_HINT(tp);
2538 			if (so->so_state & SS_CANTRCVMORE)
2539 				m_freem(m);
2540 			else {
2541 				m_adj(m, hdroptlen);
2542 				sbappendstream(&(so)->so_rcv, m);
2543 			}
2544 			sorwakeup(so);
2545 		} else {
2546 			m_adj(m, hdroptlen);
2547 			tiflags = tcp_reass(tp, th, m, &tlen);
2548 			tp->t_flags |= TF_ACKNOW;
2549 		}
2550 		TCP_REASS_UNLOCK(tp);
2551 
2552 		/*
2553 		 * Note the amount of data that peer has sent into
2554 		 * our window, in order to estimate the sender's
2555 		 * buffer size.
2556 		 */
2557 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2558 	} else {
2559 		m_freem(m);
2560 		m = NULL;
2561 		tiflags &= ~TH_FIN;
2562 	}
2563 
2564 	/*
2565 	 * If FIN is received ACK the FIN and let the user know
2566 	 * that the connection is closing.  Ignore a FIN received before
2567 	 * the connection is fully established.
2568 	 */
2569 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2570 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2571 			socantrcvmore(so);
2572 			tp->t_flags |= TF_ACKNOW;
2573 			tp->rcv_nxt++;
2574 		}
2575 		switch (tp->t_state) {
2576 
2577 	 	/*
2578 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2579 		 */
2580 		case TCPS_ESTABLISHED:
2581 			tp->t_state = TCPS_CLOSE_WAIT;
2582 			break;
2583 
2584 	 	/*
2585 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2586 		 * enter the CLOSING state.
2587 		 */
2588 		case TCPS_FIN_WAIT_1:
2589 			tp->t_state = TCPS_CLOSING;
2590 			break;
2591 
2592 	 	/*
2593 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2594 		 * starting the time-wait timer, turning off the other
2595 		 * standard timers.
2596 		 */
2597 		case TCPS_FIN_WAIT_2:
2598 			tp->t_state = TCPS_TIME_WAIT;
2599 			tcp_canceltimers(tp);
2600 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2601 			soisdisconnected(so);
2602 			break;
2603 
2604 		/*
2605 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2606 		 */
2607 		case TCPS_TIME_WAIT:
2608 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2609 			break;
2610 		}
2611 	}
2612 #ifdef TCP_DEBUG
2613 	if (so->so_options & SO_DEBUG)
2614 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2615 #endif
2616 
2617 	/*
2618 	 * Return any desired output.
2619 	 */
2620 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2621 		(void) tcp_output(tp);
2622 	}
2623 	if (tcp_saveti)
2624 		m_freem(tcp_saveti);
2625 	return;
2626 
2627 badsyn:
2628 	/*
2629 	 * Received a bad SYN.  Increment counters and dropwithreset.
2630 	 */
2631 	tcpstat.tcps_badsyn++;
2632 	tp = NULL;
2633 	goto dropwithreset;
2634 
2635 dropafterack:
2636 	/*
2637 	 * Generate an ACK dropping incoming segment if it occupies
2638 	 * sequence space, where the ACK reflects our state.
2639 	 */
2640 	if (tiflags & TH_RST)
2641 		goto drop;
2642 	goto dropafterack2;
2643 
2644 dropafterack_ratelim:
2645 	/*
2646 	 * We may want to rate-limit ACKs against SYN/RST attack.
2647 	 */
2648 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2649 	    tcp_ackdrop_ppslim) == 0) {
2650 		/* XXX stat */
2651 		goto drop;
2652 	}
2653 	/* ...fall into dropafterack2... */
2654 
2655 dropafterack2:
2656 	m_freem(m);
2657 	tp->t_flags |= TF_ACKNOW;
2658 	(void) tcp_output(tp);
2659 	if (tcp_saveti)
2660 		m_freem(tcp_saveti);
2661 	return;
2662 
2663 dropwithreset_ratelim:
2664 	/*
2665 	 * We may want to rate-limit RSTs in certain situations,
2666 	 * particularly if we are sending an RST in response to
2667 	 * an attempt to connect to or otherwise communicate with
2668 	 * a port for which we have no socket.
2669 	 */
2670 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2671 	    tcp_rst_ppslim) == 0) {
2672 		/* XXX stat */
2673 		goto drop;
2674 	}
2675 	/* ...fall into dropwithreset... */
2676 
2677 dropwithreset:
2678 	/*
2679 	 * Generate a RST, dropping incoming segment.
2680 	 * Make ACK acceptable to originator of segment.
2681 	 */
2682 	if (tiflags & TH_RST)
2683 		goto drop;
2684 
2685 	switch (af) {
2686 #ifdef INET6
2687 	case AF_INET6:
2688 		/* For following calls to tcp_respond */
2689 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2690 			goto drop;
2691 		break;
2692 #endif /* INET6 */
2693 	case AF_INET:
2694 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2695 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2696 			goto drop;
2697 	}
2698 
2699 	if (tiflags & TH_ACK)
2700 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2701 	else {
2702 		if (tiflags & TH_SYN)
2703 			tlen++;
2704 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2705 		    TH_RST|TH_ACK);
2706 	}
2707 	if (tcp_saveti)
2708 		m_freem(tcp_saveti);
2709 	return;
2710 
2711 badcsum:
2712 drop:
2713 	/*
2714 	 * Drop space held by incoming segment and return.
2715 	 */
2716 	if (tp) {
2717 		if (tp->t_inpcb)
2718 			so = tp->t_inpcb->inp_socket;
2719 #ifdef INET6
2720 		else if (tp->t_in6pcb)
2721 			so = tp->t_in6pcb->in6p_socket;
2722 #endif
2723 		else
2724 			so = NULL;
2725 #ifdef TCP_DEBUG
2726 		if (so && (so->so_options & SO_DEBUG) != 0)
2727 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2728 #endif
2729 	}
2730 	if (tcp_saveti)
2731 		m_freem(tcp_saveti);
2732 	m_freem(m);
2733 	return;
2734 }
2735 
2736 #ifdef TCP_SIGNATURE
2737 int
2738 tcp_signature_apply(void *fstate, caddr_t data, u_int len)
2739 {
2740 
2741 	MD5Update(fstate, (u_char *)data, len);
2742 	return (0);
2743 }
2744 
2745 struct secasvar *
2746 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2747 {
2748 	struct secasvar *sav;
2749 #ifdef FAST_IPSEC
2750 	union sockaddr_union dst;
2751 #endif
2752 	struct ip *ip;
2753 	struct ip6_hdr *ip6;
2754 
2755 	ip = mtod(m, struct ip *);
2756 	switch (ip->ip_v) {
2757 	case 4:
2758 		ip = mtod(m, struct ip *);
2759 		ip6 = NULL;
2760 		break;
2761 	case 6:
2762 		ip = NULL;
2763 		ip6 = mtod(m, struct ip6_hdr *);
2764 		break;
2765 	default:
2766 		return (NULL);
2767 	}
2768 
2769 #ifdef FAST_IPSEC
2770 	/* Extract the destination from the IP header in the mbuf. */
2771 	bzero(&dst, sizeof(union sockaddr_union));
2772 	dst.sa.sa_len = sizeof(struct sockaddr_in);
2773 	dst.sa.sa_family = AF_INET;
2774 	dst.sin.sin_addr = ip->ip_dst;
2775 
2776 	/*
2777 	 * Look up an SADB entry which matches the address of the peer.
2778 	 */
2779 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2780 #else
2781 	if (ip)
2782 		sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src,
2783 		    (caddr_t)&ip->ip_dst, IPPROTO_TCP,
2784 		    htonl(TCP_SIG_SPI), 0, 0);
2785 	else
2786 		sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src,
2787 		    (caddr_t)&ip6->ip6_dst, IPPROTO_TCP,
2788 		    htonl(TCP_SIG_SPI), 0, 0);
2789 #endif
2790 
2791 	return (sav);	/* freesav must be performed by caller */
2792 }
2793 
2794 int
2795 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2796     struct secasvar *sav, char *sig)
2797 {
2798 	MD5_CTX ctx;
2799 	struct ip *ip;
2800 	struct ipovly *ipovly;
2801 	struct ip6_hdr *ip6;
2802 	struct ippseudo ippseudo;
2803 	struct ip6_hdr_pseudo ip6pseudo;
2804 	struct tcphdr th0;
2805 	int l, tcphdrlen;
2806 
2807 	if (sav == NULL)
2808 		return (-1);
2809 
2810 	tcphdrlen = th->th_off * 4;
2811 
2812 	switch (mtod(m, struct ip *)->ip_v) {
2813 	case 4:
2814 		ip = mtod(m, struct ip *);
2815 		ip6 = NULL;
2816 		break;
2817 	case 6:
2818 		ip = NULL;
2819 		ip6 = mtod(m, struct ip6_hdr *);
2820 		break;
2821 	default:
2822 		return (-1);
2823 	}
2824 
2825 	MD5Init(&ctx);
2826 
2827 	if (ip) {
2828 		memset(&ippseudo, 0, sizeof(ippseudo));
2829 		ipovly = (struct ipovly *)ip;
2830 		ippseudo.ippseudo_src = ipovly->ih_src;
2831 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2832 		ippseudo.ippseudo_pad = 0;
2833 		ippseudo.ippseudo_p = IPPROTO_TCP;
2834 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2835 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2836 	} else {
2837 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2838 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2839 		in6_clearscope(&ip6pseudo.ip6ph_src);
2840 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2841 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2842 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2843 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2844 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2845 	}
2846 
2847 	th0 = *th;
2848 	th0.th_sum = 0;
2849 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2850 
2851 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2852 	if (l > 0)
2853 		m_apply(m, thoff + tcphdrlen,
2854 		    m->m_pkthdr.len - thoff - tcphdrlen,
2855 		    tcp_signature_apply, &ctx);
2856 
2857 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2858 	MD5Final(sig, &ctx);
2859 
2860 	return (0);
2861 }
2862 #endif
2863 
2864 int
2865 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th,
2866     struct mbuf *m __unused, int toff __unused, struct tcp_opt_info *oi)
2867 {
2868 	u_int16_t mss;
2869 	int opt, optlen = 0;
2870 #ifdef TCP_SIGNATURE
2871 	caddr_t sigp = NULL;
2872 	char sigbuf[TCP_SIGLEN];
2873 	struct secasvar *sav = NULL;
2874 #endif
2875 
2876 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2877 		opt = cp[0];
2878 		if (opt == TCPOPT_EOL)
2879 			break;
2880 		if (opt == TCPOPT_NOP)
2881 			optlen = 1;
2882 		else {
2883 			if (cnt < 2)
2884 				break;
2885 			optlen = cp[1];
2886 			if (optlen < 2 || optlen > cnt)
2887 				break;
2888 		}
2889 		switch (opt) {
2890 
2891 		default:
2892 			continue;
2893 
2894 		case TCPOPT_MAXSEG:
2895 			if (optlen != TCPOLEN_MAXSEG)
2896 				continue;
2897 			if (!(th->th_flags & TH_SYN))
2898 				continue;
2899 			if (TCPS_HAVERCVDSYN(tp->t_state))
2900 				continue;
2901 			bcopy(cp + 2, &mss, sizeof(mss));
2902 			oi->maxseg = ntohs(mss);
2903 			break;
2904 
2905 		case TCPOPT_WINDOW:
2906 			if (optlen != TCPOLEN_WINDOW)
2907 				continue;
2908 			if (!(th->th_flags & TH_SYN))
2909 				continue;
2910 			if (TCPS_HAVERCVDSYN(tp->t_state))
2911 				continue;
2912 			tp->t_flags |= TF_RCVD_SCALE;
2913 			tp->requested_s_scale = cp[2];
2914 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2915 #if 0	/*XXX*/
2916 				char *p;
2917 
2918 				if (ip)
2919 					p = ntohl(ip->ip_src);
2920 #ifdef INET6
2921 				else if (ip6)
2922 					p = ip6_sprintf(&ip6->ip6_src);
2923 #endif
2924 				else
2925 					p = "(unknown)";
2926 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2927 				    "assuming %d\n",
2928 				    tp->requested_s_scale, p,
2929 				    TCP_MAX_WINSHIFT);
2930 #else
2931 				log(LOG_ERR, "TCP: invalid wscale %d, "
2932 				    "assuming %d\n",
2933 				    tp->requested_s_scale,
2934 				    TCP_MAX_WINSHIFT);
2935 #endif
2936 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2937 			}
2938 			break;
2939 
2940 		case TCPOPT_TIMESTAMP:
2941 			if (optlen != TCPOLEN_TIMESTAMP)
2942 				continue;
2943 			oi->ts_present = 1;
2944 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2945 			NTOHL(oi->ts_val);
2946 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2947 			NTOHL(oi->ts_ecr);
2948 
2949 			if (!(th->th_flags & TH_SYN))
2950 				continue;
2951 			if (TCPS_HAVERCVDSYN(tp->t_state))
2952 				continue;
2953 			/*
2954 			 * A timestamp received in a SYN makes
2955 			 * it ok to send timestamp requests and replies.
2956 			 */
2957 			tp->t_flags |= TF_RCVD_TSTMP;
2958 			tp->ts_recent = oi->ts_val;
2959 			tp->ts_recent_age = tcp_now;
2960                         break;
2961 
2962 		case TCPOPT_SACK_PERMITTED:
2963 			if (optlen != TCPOLEN_SACK_PERMITTED)
2964 				continue;
2965 			if (!(th->th_flags & TH_SYN))
2966 				continue;
2967 			if (TCPS_HAVERCVDSYN(tp->t_state))
2968 				continue;
2969 			if (tcp_do_sack) {
2970 				tp->t_flags |= TF_SACK_PERMIT;
2971 				tp->t_flags |= TF_WILL_SACK;
2972 			}
2973 			break;
2974 
2975 		case TCPOPT_SACK:
2976 			tcp_sack_option(tp, th, cp, optlen);
2977 			break;
2978 #ifdef TCP_SIGNATURE
2979 		case TCPOPT_SIGNATURE:
2980 			if (optlen != TCPOLEN_SIGNATURE)
2981 				continue;
2982 			if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN))
2983 				return (-1);
2984 
2985 			sigp = sigbuf;
2986 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
2987 			memset(cp + 2, 0, TCP_SIGLEN);
2988 			tp->t_flags |= TF_SIGNATURE;
2989 			break;
2990 #endif
2991 		}
2992 	}
2993 
2994 #ifdef TCP_SIGNATURE
2995 	if (tp->t_flags & TF_SIGNATURE) {
2996 
2997 		sav = tcp_signature_getsav(m, th);
2998 
2999 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3000 			return (-1);
3001 	}
3002 
3003 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3004 		if (sav == NULL)
3005 			return (-1);
3006 #ifdef FAST_IPSEC
3007 		KEY_FREESAV(&sav);
3008 #else
3009 		key_freesav(sav);
3010 #endif
3011 		return (-1);
3012 	}
3013 
3014 	if (sigp) {
3015 		char sig[TCP_SIGLEN];
3016 
3017 		TCP_FIELDS_TO_NET(th);
3018 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3019 			TCP_FIELDS_TO_HOST(th);
3020 			if (sav == NULL)
3021 				return (-1);
3022 #ifdef FAST_IPSEC
3023 			KEY_FREESAV(&sav);
3024 #else
3025 			key_freesav(sav);
3026 #endif
3027 			return (-1);
3028 		}
3029 		TCP_FIELDS_TO_HOST(th);
3030 
3031 		if (bcmp(sig, sigp, TCP_SIGLEN)) {
3032 			tcpstat.tcps_badsig++;
3033 			if (sav == NULL)
3034 				return (-1);
3035 #ifdef FAST_IPSEC
3036 			KEY_FREESAV(&sav);
3037 #else
3038 			key_freesav(sav);
3039 #endif
3040 			return (-1);
3041 		} else
3042 			tcpstat.tcps_goodsig++;
3043 
3044 		key_sa_recordxfer(sav, m);
3045 #ifdef FAST_IPSEC
3046 		KEY_FREESAV(&sav);
3047 #else
3048 		key_freesav(sav);
3049 #endif
3050 	}
3051 #endif
3052 
3053 	return (0);
3054 }
3055 
3056 /*
3057  * Pull out of band byte out of a segment so
3058  * it doesn't appear in the user's data queue.
3059  * It is still reflected in the segment length for
3060  * sequencing purposes.
3061  */
3062 void
3063 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3064     struct mbuf *m, int off)
3065 {
3066 	int cnt = off + th->th_urp - 1;
3067 
3068 	while (cnt >= 0) {
3069 		if (m->m_len > cnt) {
3070 			char *cp = mtod(m, caddr_t) + cnt;
3071 			struct tcpcb *tp = sototcpcb(so);
3072 
3073 			tp->t_iobc = *cp;
3074 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3075 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3076 			m->m_len--;
3077 			return;
3078 		}
3079 		cnt -= m->m_len;
3080 		m = m->m_next;
3081 		if (m == 0)
3082 			break;
3083 	}
3084 	panic("tcp_pulloutofband");
3085 }
3086 
3087 /*
3088  * Collect new round-trip time estimate
3089  * and update averages and current timeout.
3090  */
3091 void
3092 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3093 {
3094 	int32_t delta;
3095 
3096 	tcpstat.tcps_rttupdated++;
3097 	if (tp->t_srtt != 0) {
3098 		/*
3099 		 * srtt is stored as fixed point with 3 bits after the
3100 		 * binary point (i.e., scaled by 8).  The following magic
3101 		 * is equivalent to the smoothing algorithm in rfc793 with
3102 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3103 		 * point).  Adjust rtt to origin 0.
3104 		 */
3105 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3106 		if ((tp->t_srtt += delta) <= 0)
3107 			tp->t_srtt = 1 << 2;
3108 		/*
3109 		 * We accumulate a smoothed rtt variance (actually, a
3110 		 * smoothed mean difference), then set the retransmit
3111 		 * timer to smoothed rtt + 4 times the smoothed variance.
3112 		 * rttvar is stored as fixed point with 2 bits after the
3113 		 * binary point (scaled by 4).  The following is
3114 		 * equivalent to rfc793 smoothing with an alpha of .75
3115 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3116 		 * rfc793's wired-in beta.
3117 		 */
3118 		if (delta < 0)
3119 			delta = -delta;
3120 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3121 		if ((tp->t_rttvar += delta) <= 0)
3122 			tp->t_rttvar = 1 << 2;
3123 	} else {
3124 		/*
3125 		 * No rtt measurement yet - use the unsmoothed rtt.
3126 		 * Set the variance to half the rtt (so our first
3127 		 * retransmit happens at 3*rtt).
3128 		 */
3129 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3130 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3131 	}
3132 	tp->t_rtttime = 0;
3133 	tp->t_rxtshift = 0;
3134 
3135 	/*
3136 	 * the retransmit should happen at rtt + 4 * rttvar.
3137 	 * Because of the way we do the smoothing, srtt and rttvar
3138 	 * will each average +1/2 tick of bias.  When we compute
3139 	 * the retransmit timer, we want 1/2 tick of rounding and
3140 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3141 	 * firing of the timer.  The bias will give us exactly the
3142 	 * 1.5 tick we need.  But, because the bias is
3143 	 * statistical, we have to test that we don't drop below
3144 	 * the minimum feasible timer (which is 2 ticks).
3145 	 */
3146 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3147 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3148 
3149 	/*
3150 	 * We received an ack for a packet that wasn't retransmitted;
3151 	 * it is probably safe to discard any error indications we've
3152 	 * received recently.  This isn't quite right, but close enough
3153 	 * for now (a route might have failed after we sent a segment,
3154 	 * and the return path might not be symmetrical).
3155 	 */
3156 	tp->t_softerror = 0;
3157 }
3158 
3159 
3160 /*
3161  * TCP compressed state engine.  Currently used to hold compressed
3162  * state for SYN_RECEIVED.
3163  */
3164 
3165 u_long	syn_cache_count;
3166 u_int32_t syn_hash1, syn_hash2;
3167 
3168 #define SYN_HASH(sa, sp, dp) \
3169 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3170 				     ((u_int32_t)(sp)))^syn_hash2)))
3171 #ifndef INET6
3172 #define	SYN_HASHALL(hash, src, dst) \
3173 do {									\
3174 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3175 		((const struct sockaddr_in *)(src))->sin_port,		\
3176 		((const struct sockaddr_in *)(dst))->sin_port);		\
3177 } while (/*CONSTCOND*/ 0)
3178 #else
3179 #define SYN_HASH6(sa, sp, dp) \
3180 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3181 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3182 	 & 0x7fffffff)
3183 
3184 #define SYN_HASHALL(hash, src, dst) \
3185 do {									\
3186 	switch ((src)->sa_family) {					\
3187 	case AF_INET:							\
3188 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3189 			((const struct sockaddr_in *)(src))->sin_port,	\
3190 			((const struct sockaddr_in *)(dst))->sin_port);	\
3191 		break;							\
3192 	case AF_INET6:							\
3193 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3194 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3195 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3196 		break;							\
3197 	default:							\
3198 		hash = 0;						\
3199 	}								\
3200 } while (/*CONSTCOND*/0)
3201 #endif /* INET6 */
3202 
3203 #define	SYN_CACHE_RM(sc)						\
3204 do {									\
3205 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
3206 	    (sc), sc_bucketq);						\
3207 	(sc)->sc_tp = NULL;						\
3208 	LIST_REMOVE((sc), sc_tpq);					\
3209 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
3210 	callout_stop(&(sc)->sc_timer);					\
3211 	syn_cache_count--;						\
3212 } while (/*CONSTCOND*/0)
3213 
3214 #define	SYN_CACHE_PUT(sc)						\
3215 do {									\
3216 	if ((sc)->sc_ipopts)						\
3217 		(void) m_free((sc)->sc_ipopts);				\
3218 	if ((sc)->sc_route4.ro_rt != NULL)				\
3219 		RTFREE((sc)->sc_route4.ro_rt);				\
3220 	if (callout_invoking(&(sc)->sc_timer))				\
3221 		(sc)->sc_flags |= SCF_DEAD;				\
3222 	else								\
3223 		pool_put(&syn_cache_pool, (sc));			\
3224 } while (/*CONSTCOND*/0)
3225 
3226 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL);
3227 
3228 /*
3229  * We don't estimate RTT with SYNs, so each packet starts with the default
3230  * RTT and each timer step has a fixed timeout value.
3231  */
3232 #define	SYN_CACHE_TIMER_ARM(sc)						\
3233 do {									\
3234 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3235 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3236 	    TCPTV_REXMTMAX);						\
3237 	callout_reset(&(sc)->sc_timer,					\
3238 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3239 } while (/*CONSTCOND*/0)
3240 
3241 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3242 
3243 void
3244 syn_cache_init(void)
3245 {
3246 	int i;
3247 
3248 	/* Initialize the hash buckets. */
3249 	for (i = 0; i < tcp_syn_cache_size; i++)
3250 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3251 }
3252 
3253 void
3254 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3255 {
3256 	struct syn_cache_head *scp;
3257 	struct syn_cache *sc2;
3258 	int s;
3259 
3260 	/*
3261 	 * If there are no entries in the hash table, reinitialize
3262 	 * the hash secrets.
3263 	 */
3264 	if (syn_cache_count == 0) {
3265 		syn_hash1 = arc4random();
3266 		syn_hash2 = arc4random();
3267 	}
3268 
3269 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3270 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3271 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3272 
3273 	/*
3274 	 * Make sure that we don't overflow the per-bucket
3275 	 * limit or the total cache size limit.
3276 	 */
3277 	s = splsoftnet();
3278 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3279 		tcpstat.tcps_sc_bucketoverflow++;
3280 		/*
3281 		 * The bucket is full.  Toss the oldest element in the
3282 		 * bucket.  This will be the first entry in the bucket.
3283 		 */
3284 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3285 #ifdef DIAGNOSTIC
3286 		/*
3287 		 * This should never happen; we should always find an
3288 		 * entry in our bucket.
3289 		 */
3290 		if (sc2 == NULL)
3291 			panic("syn_cache_insert: bucketoverflow: impossible");
3292 #endif
3293 		SYN_CACHE_RM(sc2);
3294 		SYN_CACHE_PUT(sc2);	/* calls pool_put but see spl above */
3295 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3296 		struct syn_cache_head *scp2, *sce;
3297 
3298 		tcpstat.tcps_sc_overflowed++;
3299 		/*
3300 		 * The cache is full.  Toss the oldest entry in the
3301 		 * first non-empty bucket we can find.
3302 		 *
3303 		 * XXX We would really like to toss the oldest
3304 		 * entry in the cache, but we hope that this
3305 		 * condition doesn't happen very often.
3306 		 */
3307 		scp2 = scp;
3308 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3309 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3310 			for (++scp2; scp2 != scp; scp2++) {
3311 				if (scp2 >= sce)
3312 					scp2 = &tcp_syn_cache[0];
3313 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3314 					break;
3315 			}
3316 #ifdef DIAGNOSTIC
3317 			/*
3318 			 * This should never happen; we should always find a
3319 			 * non-empty bucket.
3320 			 */
3321 			if (scp2 == scp)
3322 				panic("syn_cache_insert: cacheoverflow: "
3323 				    "impossible");
3324 #endif
3325 		}
3326 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3327 		SYN_CACHE_RM(sc2);
3328 		SYN_CACHE_PUT(sc2);	/* calls pool_put but see spl above */
3329 	}
3330 
3331 	/*
3332 	 * Initialize the entry's timer.
3333 	 */
3334 	sc->sc_rxttot = 0;
3335 	sc->sc_rxtshift = 0;
3336 	SYN_CACHE_TIMER_ARM(sc);
3337 
3338 	/* Link it from tcpcb entry */
3339 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3340 
3341 	/* Put it into the bucket. */
3342 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3343 	scp->sch_length++;
3344 	syn_cache_count++;
3345 
3346 	tcpstat.tcps_sc_added++;
3347 	splx(s);
3348 }
3349 
3350 /*
3351  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3352  * If we have retransmitted an entry the maximum number of times, expire
3353  * that entry.
3354  */
3355 void
3356 syn_cache_timer(void *arg)
3357 {
3358 	struct syn_cache *sc = arg;
3359 	int s;
3360 
3361 	s = splsoftnet();
3362 	callout_ack(&sc->sc_timer);
3363 
3364 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3365 		tcpstat.tcps_sc_delayed_free++;
3366 		pool_put(&syn_cache_pool, sc);
3367 		splx(s);
3368 		return;
3369 	}
3370 
3371 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3372 		/* Drop it -- too many retransmissions. */
3373 		goto dropit;
3374 	}
3375 
3376 	/*
3377 	 * Compute the total amount of time this entry has
3378 	 * been on a queue.  If this entry has been on longer
3379 	 * than the keep alive timer would allow, expire it.
3380 	 */
3381 	sc->sc_rxttot += sc->sc_rxtcur;
3382 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
3383 		goto dropit;
3384 
3385 	tcpstat.tcps_sc_retransmitted++;
3386 	(void) syn_cache_respond(sc, NULL);
3387 
3388 	/* Advance the timer back-off. */
3389 	sc->sc_rxtshift++;
3390 	SYN_CACHE_TIMER_ARM(sc);
3391 
3392 	splx(s);
3393 	return;
3394 
3395  dropit:
3396 	tcpstat.tcps_sc_timed_out++;
3397 	SYN_CACHE_RM(sc);
3398 	SYN_CACHE_PUT(sc);	/* calls pool_put but see spl above */
3399 	splx(s);
3400 }
3401 
3402 /*
3403  * Remove syn cache created by the specified tcb entry,
3404  * because this does not make sense to keep them
3405  * (if there's no tcb entry, syn cache entry will never be used)
3406  */
3407 void
3408 syn_cache_cleanup(struct tcpcb *tp)
3409 {
3410 	struct syn_cache *sc, *nsc;
3411 	int s;
3412 
3413 	s = splsoftnet();
3414 
3415 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3416 		nsc = LIST_NEXT(sc, sc_tpq);
3417 
3418 #ifdef DIAGNOSTIC
3419 		if (sc->sc_tp != tp)
3420 			panic("invalid sc_tp in syn_cache_cleanup");
3421 #endif
3422 		SYN_CACHE_RM(sc);
3423 		SYN_CACHE_PUT(sc);	/* calls pool_put but see spl above */
3424 	}
3425 	/* just for safety */
3426 	LIST_INIT(&tp->t_sc);
3427 
3428 	splx(s);
3429 }
3430 
3431 /*
3432  * Find an entry in the syn cache.
3433  */
3434 struct syn_cache *
3435 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3436     struct syn_cache_head **headp)
3437 {
3438 	struct syn_cache *sc;
3439 	struct syn_cache_head *scp;
3440 	u_int32_t hash;
3441 	int s;
3442 
3443 	SYN_HASHALL(hash, src, dst);
3444 
3445 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3446 	*headp = scp;
3447 	s = splsoftnet();
3448 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3449 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3450 		if (sc->sc_hash != hash)
3451 			continue;
3452 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3453 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3454 			splx(s);
3455 			return (sc);
3456 		}
3457 	}
3458 	splx(s);
3459 	return (NULL);
3460 }
3461 
3462 /*
3463  * This function gets called when we receive an ACK for a
3464  * socket in the LISTEN state.  We look up the connection
3465  * in the syn cache, and if its there, we pull it out of
3466  * the cache and turn it into a full-blown connection in
3467  * the SYN-RECEIVED state.
3468  *
3469  * The return values may not be immediately obvious, and their effects
3470  * can be subtle, so here they are:
3471  *
3472  *	NULL	SYN was not found in cache; caller should drop the
3473  *		packet and send an RST.
3474  *
3475  *	-1	We were unable to create the new connection, and are
3476  *		aborting it.  An ACK,RST is being sent to the peer
3477  *		(unless we got screwey sequence numbners; see below),
3478  *		because the 3-way handshake has been completed.  Caller
3479  *		should not free the mbuf, since we may be using it.  If
3480  *		we are not, we will free it.
3481  *
3482  *	Otherwise, the return value is a pointer to the new socket
3483  *	associated with the connection.
3484  */
3485 struct socket *
3486 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3487     struct tcphdr *th, unsigned int hlen __unused, unsigned int tlen __unused,
3488     struct socket *so, struct mbuf *m)
3489 {
3490 	struct syn_cache *sc;
3491 	struct syn_cache_head *scp;
3492 	struct inpcb *inp = NULL;
3493 #ifdef INET6
3494 	struct in6pcb *in6p = NULL;
3495 #endif
3496 	struct tcpcb *tp = 0;
3497 	struct mbuf *am;
3498 	int s;
3499 	struct socket *oso;
3500 
3501 	s = splsoftnet();
3502 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3503 		splx(s);
3504 		return (NULL);
3505 	}
3506 
3507 	/*
3508 	 * Verify the sequence and ack numbers.  Try getting the correct
3509 	 * response again.
3510 	 */
3511 	if ((th->th_ack != sc->sc_iss + 1) ||
3512 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3513 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3514 		(void) syn_cache_respond(sc, m);
3515 		splx(s);
3516 		return ((struct socket *)(-1));
3517 	}
3518 
3519 	/* Remove this cache entry */
3520 	SYN_CACHE_RM(sc);
3521 	splx(s);
3522 
3523 	/*
3524 	 * Ok, create the full blown connection, and set things up
3525 	 * as they would have been set up if we had created the
3526 	 * connection when the SYN arrived.  If we can't create
3527 	 * the connection, abort it.
3528 	 */
3529 	/*
3530 	 * inp still has the OLD in_pcb stuff, set the
3531 	 * v6-related flags on the new guy, too.   This is
3532 	 * done particularly for the case where an AF_INET6
3533 	 * socket is bound only to a port, and a v4 connection
3534 	 * comes in on that port.
3535 	 * we also copy the flowinfo from the original pcb
3536 	 * to the new one.
3537 	 */
3538 	oso = so;
3539 	so = sonewconn(so, SS_ISCONNECTED);
3540 	if (so == NULL)
3541 		goto resetandabort;
3542 
3543 	switch (so->so_proto->pr_domain->dom_family) {
3544 #ifdef INET
3545 	case AF_INET:
3546 		inp = sotoinpcb(so);
3547 		break;
3548 #endif
3549 #ifdef INET6
3550 	case AF_INET6:
3551 		in6p = sotoin6pcb(so);
3552 		break;
3553 #endif
3554 	}
3555 	switch (src->sa_family) {
3556 #ifdef INET
3557 	case AF_INET:
3558 		if (inp) {
3559 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3560 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3561 			inp->inp_options = ip_srcroute();
3562 			in_pcbstate(inp, INP_BOUND);
3563 			if (inp->inp_options == NULL) {
3564 				inp->inp_options = sc->sc_ipopts;
3565 				sc->sc_ipopts = NULL;
3566 			}
3567 		}
3568 #ifdef INET6
3569 		else if (in6p) {
3570 			/* IPv4 packet to AF_INET6 socket */
3571 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3572 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3573 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3574 				&in6p->in6p_laddr.s6_addr32[3],
3575 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3576 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3577 			in6totcpcb(in6p)->t_family = AF_INET;
3578 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3579 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3580 			else
3581 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3582 			in6_pcbstate(in6p, IN6P_BOUND);
3583 		}
3584 #endif
3585 		break;
3586 #endif
3587 #ifdef INET6
3588 	case AF_INET6:
3589 		if (in6p) {
3590 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3591 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3592 			in6_pcbstate(in6p, IN6P_BOUND);
3593 		}
3594 		break;
3595 #endif
3596 	}
3597 #ifdef INET6
3598 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3599 		struct in6pcb *oin6p = sotoin6pcb(oso);
3600 		/* inherit socket options from the listening socket */
3601 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3602 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3603 			m_freem(in6p->in6p_options);
3604 			in6p->in6p_options = 0;
3605 		}
3606 		ip6_savecontrol(in6p, &in6p->in6p_options,
3607 			mtod(m, struct ip6_hdr *), m);
3608 	}
3609 #endif
3610 
3611 #if defined(IPSEC) || defined(FAST_IPSEC)
3612 	/*
3613 	 * we make a copy of policy, instead of sharing the policy,
3614 	 * for better behavior in terms of SA lookup and dead SA removal.
3615 	 */
3616 	if (inp) {
3617 		/* copy old policy into new socket's */
3618 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3619 			printf("tcp_input: could not copy policy\n");
3620 	}
3621 #ifdef INET6
3622 	else if (in6p) {
3623 		/* copy old policy into new socket's */
3624 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3625 		    in6p->in6p_sp))
3626 			printf("tcp_input: could not copy policy\n");
3627 	}
3628 #endif
3629 #endif
3630 
3631 	/*
3632 	 * Give the new socket our cached route reference.
3633 	 */
3634 	if (inp)
3635 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3636 #ifdef INET6
3637 	else
3638 		in6p->in6p_route = sc->sc_route6;
3639 #endif
3640 	sc->sc_route4.ro_rt = NULL;
3641 
3642 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3643 	if (am == NULL)
3644 		goto resetandabort;
3645 	MCLAIM(am, &tcp_mowner);
3646 	am->m_len = src->sa_len;
3647 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3648 	if (inp) {
3649 		if (in_pcbconnect(inp, am, NULL)) {
3650 			(void) m_free(am);
3651 			goto resetandabort;
3652 		}
3653 	}
3654 #ifdef INET6
3655 	else if (in6p) {
3656 		if (src->sa_family == AF_INET) {
3657 			/* IPv4 packet to AF_INET6 socket */
3658 			struct sockaddr_in6 *sin6;
3659 			sin6 = mtod(am, struct sockaddr_in6 *);
3660 			am->m_len = sizeof(*sin6);
3661 			bzero(sin6, sizeof(*sin6));
3662 			sin6->sin6_family = AF_INET6;
3663 			sin6->sin6_len = sizeof(*sin6);
3664 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3665 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3666 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3667 				&sin6->sin6_addr.s6_addr32[3],
3668 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3669 		}
3670 		if (in6_pcbconnect(in6p, am, NULL)) {
3671 			(void) m_free(am);
3672 			goto resetandabort;
3673 		}
3674 	}
3675 #endif
3676 	else {
3677 		(void) m_free(am);
3678 		goto resetandabort;
3679 	}
3680 	(void) m_free(am);
3681 
3682 	if (inp)
3683 		tp = intotcpcb(inp);
3684 #ifdef INET6
3685 	else if (in6p)
3686 		tp = in6totcpcb(in6p);
3687 #endif
3688 	else
3689 		tp = NULL;
3690 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3691 	if (sc->sc_request_r_scale != 15) {
3692 		tp->requested_s_scale = sc->sc_requested_s_scale;
3693 		tp->request_r_scale = sc->sc_request_r_scale;
3694 		tp->snd_scale = sc->sc_requested_s_scale;
3695 		tp->rcv_scale = sc->sc_request_r_scale;
3696 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3697 	}
3698 	if (sc->sc_flags & SCF_TIMESTAMP)
3699 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3700 	tp->ts_timebase = sc->sc_timebase;
3701 
3702 	tp->t_template = tcp_template(tp);
3703 	if (tp->t_template == 0) {
3704 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3705 		so = NULL;
3706 		m_freem(m);
3707 		goto abort;
3708 	}
3709 
3710 	tp->iss = sc->sc_iss;
3711 	tp->irs = sc->sc_irs;
3712 	tcp_sendseqinit(tp);
3713 	tcp_rcvseqinit(tp);
3714 	tp->t_state = TCPS_SYN_RECEIVED;
3715 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3716 	tcpstat.tcps_accepts++;
3717 
3718 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3719 		tp->t_flags |= TF_WILL_SACK;
3720 
3721 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3722 		tp->t_flags |= TF_ECN_PERMIT;
3723 
3724 #ifdef TCP_SIGNATURE
3725 	if (sc->sc_flags & SCF_SIGNATURE)
3726 		tp->t_flags |= TF_SIGNATURE;
3727 #endif
3728 
3729 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3730 	tp->t_ourmss = sc->sc_ourmaxseg;
3731 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3732 
3733 	/*
3734 	 * Initialize the initial congestion window.  If we
3735 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3736 	 * to 1 segment (i.e. the Loss Window).
3737 	 */
3738 	if (sc->sc_rxtshift)
3739 		tp->snd_cwnd = tp->t_peermss;
3740 	else {
3741 		int ss = tcp_init_win;
3742 #ifdef INET
3743 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3744 			ss = tcp_init_win_local;
3745 #endif
3746 #ifdef INET6
3747 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3748 			ss = tcp_init_win_local;
3749 #endif
3750 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3751 	}
3752 
3753 	tcp_rmx_rtt(tp);
3754 	tp->snd_wl1 = sc->sc_irs;
3755 	tp->rcv_up = sc->sc_irs + 1;
3756 
3757 	/*
3758 	 * This is what whould have happened in tcp_output() when
3759 	 * the SYN,ACK was sent.
3760 	 */
3761 	tp->snd_up = tp->snd_una;
3762 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3763 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3764 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3765 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3766 	tp->last_ack_sent = tp->rcv_nxt;
3767 	tp->t_partialacks = -1;
3768 	tp->t_dupacks = 0;
3769 
3770 	tcpstat.tcps_sc_completed++;
3771 	s = splsoftnet();
3772 	SYN_CACHE_PUT(sc);
3773 	splx(s);
3774 	return (so);
3775 
3776 resetandabort:
3777 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3778 abort:
3779 	if (so != NULL)
3780 		(void) soabort(so);
3781 	s = splsoftnet();
3782 	SYN_CACHE_PUT(sc);
3783 	splx(s);
3784 	tcpstat.tcps_sc_aborted++;
3785 	return ((struct socket *)(-1));
3786 }
3787 
3788 /*
3789  * This function is called when we get a RST for a
3790  * non-existent connection, so that we can see if the
3791  * connection is in the syn cache.  If it is, zap it.
3792  */
3793 
3794 void
3795 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3796 {
3797 	struct syn_cache *sc;
3798 	struct syn_cache_head *scp;
3799 	int s = splsoftnet();
3800 
3801 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3802 		splx(s);
3803 		return;
3804 	}
3805 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3806 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3807 		splx(s);
3808 		return;
3809 	}
3810 	SYN_CACHE_RM(sc);
3811 	tcpstat.tcps_sc_reset++;
3812 	SYN_CACHE_PUT(sc);	/* calls pool_put but see spl above */
3813 	splx(s);
3814 }
3815 
3816 void
3817 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3818     struct tcphdr *th)
3819 {
3820 	struct syn_cache *sc;
3821 	struct syn_cache_head *scp;
3822 	int s;
3823 
3824 	s = splsoftnet();
3825 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3826 		splx(s);
3827 		return;
3828 	}
3829 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3830 	if (ntohl (th->th_seq) != sc->sc_iss) {
3831 		splx(s);
3832 		return;
3833 	}
3834 
3835 	/*
3836 	 * If we've retransmitted 3 times and this is our second error,
3837 	 * we remove the entry.  Otherwise, we allow it to continue on.
3838 	 * This prevents us from incorrectly nuking an entry during a
3839 	 * spurious network outage.
3840 	 *
3841 	 * See tcp_notify().
3842 	 */
3843 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3844 		sc->sc_flags |= SCF_UNREACH;
3845 		splx(s);
3846 		return;
3847 	}
3848 
3849 	SYN_CACHE_RM(sc);
3850 	tcpstat.tcps_sc_unreach++;
3851 	SYN_CACHE_PUT(sc);	/* calls pool_put but see spl above */
3852 	splx(s);
3853 }
3854 
3855 /*
3856  * Given a LISTEN socket and an inbound SYN request, add
3857  * this to the syn cache, and send back a segment:
3858  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3859  * to the source.
3860  *
3861  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3862  * Doing so would require that we hold onto the data and deliver it
3863  * to the application.  However, if we are the target of a SYN-flood
3864  * DoS attack, an attacker could send data which would eventually
3865  * consume all available buffer space if it were ACKed.  By not ACKing
3866  * the data, we avoid this DoS scenario.
3867  */
3868 
3869 int
3870 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3871     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
3872     int optlen, struct tcp_opt_info *oi)
3873 {
3874 	struct tcpcb tb, *tp;
3875 	long win;
3876 	struct syn_cache *sc;
3877 	struct syn_cache_head *scp;
3878 	struct mbuf *ipopts;
3879 	struct tcp_opt_info opti;
3880 	int s;
3881 
3882 	tp = sototcpcb(so);
3883 
3884 	bzero(&opti, sizeof(opti));
3885 
3886 	/*
3887 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3888 	 *
3889 	 * Note this check is performed in tcp_input() very early on.
3890 	 */
3891 
3892 	/*
3893 	 * Initialize some local state.
3894 	 */
3895 	win = sbspace(&so->so_rcv);
3896 	if (win > TCP_MAXWIN)
3897 		win = TCP_MAXWIN;
3898 
3899 	switch (src->sa_family) {
3900 #ifdef INET
3901 	case AF_INET:
3902 		/*
3903 		 * Remember the IP options, if any.
3904 		 */
3905 		ipopts = ip_srcroute();
3906 		break;
3907 #endif
3908 	default:
3909 		ipopts = NULL;
3910 	}
3911 
3912 #ifdef TCP_SIGNATURE
3913 	if (optp || (tp->t_flags & TF_SIGNATURE))
3914 #else
3915 	if (optp)
3916 #endif
3917 	{
3918 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3919 #ifdef TCP_SIGNATURE
3920 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
3921 #endif
3922 		tb.t_state = TCPS_LISTEN;
3923 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
3924 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
3925 			return (0);
3926 	} else
3927 		tb.t_flags = 0;
3928 
3929 	/*
3930 	 * See if we already have an entry for this connection.
3931 	 * If we do, resend the SYN,ACK.  We do not count this
3932 	 * as a retransmission (XXX though maybe we should).
3933 	 */
3934 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3935 		tcpstat.tcps_sc_dupesyn++;
3936 		if (ipopts) {
3937 			/*
3938 			 * If we were remembering a previous source route,
3939 			 * forget it and use the new one we've been given.
3940 			 */
3941 			if (sc->sc_ipopts)
3942 				(void) m_free(sc->sc_ipopts);
3943 			sc->sc_ipopts = ipopts;
3944 		}
3945 		sc->sc_timestamp = tb.ts_recent;
3946 		if (syn_cache_respond(sc, m) == 0) {
3947 			tcpstat.tcps_sndacks++;
3948 			tcpstat.tcps_sndtotal++;
3949 		}
3950 		return (1);
3951 	}
3952 
3953 	s = splsoftnet();
3954 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3955 	splx(s);
3956 	if (sc == NULL) {
3957 		if (ipopts)
3958 			(void) m_free(ipopts);
3959 		return (0);
3960 	}
3961 
3962 	/*
3963 	 * Fill in the cache, and put the necessary IP and TCP
3964 	 * options into the reply.
3965 	 */
3966 	bzero(sc, sizeof(struct syn_cache));
3967 	callout_init(&sc->sc_timer);
3968 	bcopy(src, &sc->sc_src, src->sa_len);
3969 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3970 	sc->sc_flags = 0;
3971 	sc->sc_ipopts = ipopts;
3972 	sc->sc_irs = th->th_seq;
3973 	switch (src->sa_family) {
3974 #ifdef INET
3975 	case AF_INET:
3976 	    {
3977 		struct sockaddr_in *srcin = (void *) src;
3978 		struct sockaddr_in *dstin = (void *) dst;
3979 
3980 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3981 		    &srcin->sin_addr, dstin->sin_port,
3982 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3983 		break;
3984 	    }
3985 #endif /* INET */
3986 #ifdef INET6
3987 	case AF_INET6:
3988 	    {
3989 		struct sockaddr_in6 *srcin6 = (void *) src;
3990 		struct sockaddr_in6 *dstin6 = (void *) dst;
3991 
3992 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3993 		    &srcin6->sin6_addr, dstin6->sin6_port,
3994 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3995 		break;
3996 	    }
3997 #endif /* INET6 */
3998 	}
3999 	sc->sc_peermaxseg = oi->maxseg;
4000 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4001 						m->m_pkthdr.rcvif : NULL,
4002 						sc->sc_src.sa.sa_family);
4003 	sc->sc_win = win;
4004 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
4005 	sc->sc_timestamp = tb.ts_recent;
4006 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4007 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4008 		sc->sc_flags |= SCF_TIMESTAMP;
4009 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4010 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4011 		sc->sc_requested_s_scale = tb.requested_s_scale;
4012 		sc->sc_request_r_scale = 0;
4013 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4014 		    TCP_MAXWIN << sc->sc_request_r_scale <
4015 		    so->so_rcv.sb_hiwat)
4016 			sc->sc_request_r_scale++;
4017 	} else {
4018 		sc->sc_requested_s_scale = 15;
4019 		sc->sc_request_r_scale = 15;
4020 	}
4021 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4022 		sc->sc_flags |= SCF_SACK_PERMIT;
4023 
4024 	/*
4025 	 * ECN setup packet recieved.
4026 	 */
4027 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4028 		sc->sc_flags |= SCF_ECN_PERMIT;
4029 
4030 #ifdef TCP_SIGNATURE
4031 	if (tb.t_flags & TF_SIGNATURE)
4032 		sc->sc_flags |= SCF_SIGNATURE;
4033 #endif
4034 	sc->sc_tp = tp;
4035 	if (syn_cache_respond(sc, m) == 0) {
4036 		syn_cache_insert(sc, tp);
4037 		tcpstat.tcps_sndacks++;
4038 		tcpstat.tcps_sndtotal++;
4039 	} else {
4040 		s = splsoftnet();
4041 		SYN_CACHE_PUT(sc);
4042 		splx(s);
4043 		tcpstat.tcps_sc_dropped++;
4044 	}
4045 	return (1);
4046 }
4047 
4048 int
4049 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4050 {
4051 	struct route *ro;
4052 	u_int8_t *optp;
4053 	int optlen, error;
4054 	u_int16_t tlen;
4055 	struct ip *ip = NULL;
4056 #ifdef INET6
4057 	struct ip6_hdr *ip6 = NULL;
4058 #endif
4059 	struct tcpcb *tp = NULL;
4060 	struct tcphdr *th;
4061 	u_int hlen;
4062 	struct socket *so;
4063 
4064 	switch (sc->sc_src.sa.sa_family) {
4065 	case AF_INET:
4066 		hlen = sizeof(struct ip);
4067 		ro = &sc->sc_route4;
4068 		break;
4069 #ifdef INET6
4070 	case AF_INET6:
4071 		hlen = sizeof(struct ip6_hdr);
4072 		ro = (struct route *)&sc->sc_route6;
4073 		break;
4074 #endif
4075 	default:
4076 		if (m)
4077 			m_freem(m);
4078 		return (EAFNOSUPPORT);
4079 	}
4080 
4081 	/* Compute the size of the TCP options. */
4082 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4083 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4084 #ifdef TCP_SIGNATURE
4085 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4086 #endif
4087 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4088 
4089 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4090 
4091 	/*
4092 	 * Create the IP+TCP header from scratch.
4093 	 */
4094 	if (m)
4095 		m_freem(m);
4096 #ifdef DIAGNOSTIC
4097 	if (max_linkhdr + tlen > MCLBYTES)
4098 		return (ENOBUFS);
4099 #endif
4100 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4101 	if (m && tlen > MHLEN) {
4102 		MCLGET(m, M_DONTWAIT);
4103 		if ((m->m_flags & M_EXT) == 0) {
4104 			m_freem(m);
4105 			m = NULL;
4106 		}
4107 	}
4108 	if (m == NULL)
4109 		return (ENOBUFS);
4110 	MCLAIM(m, &tcp_tx_mowner);
4111 
4112 	/* Fixup the mbuf. */
4113 	m->m_data += max_linkhdr;
4114 	m->m_len = m->m_pkthdr.len = tlen;
4115 	if (sc->sc_tp) {
4116 		tp = sc->sc_tp;
4117 		if (tp->t_inpcb)
4118 			so = tp->t_inpcb->inp_socket;
4119 #ifdef INET6
4120 		else if (tp->t_in6pcb)
4121 			so = tp->t_in6pcb->in6p_socket;
4122 #endif
4123 		else
4124 			so = NULL;
4125 	} else
4126 		so = NULL;
4127 	m->m_pkthdr.rcvif = NULL;
4128 	memset(mtod(m, u_char *), 0, tlen);
4129 
4130 	switch (sc->sc_src.sa.sa_family) {
4131 	case AF_INET:
4132 		ip = mtod(m, struct ip *);
4133 		ip->ip_v = 4;
4134 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4135 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4136 		ip->ip_p = IPPROTO_TCP;
4137 		th = (struct tcphdr *)(ip + 1);
4138 		th->th_dport = sc->sc_src.sin.sin_port;
4139 		th->th_sport = sc->sc_dst.sin.sin_port;
4140 		break;
4141 #ifdef INET6
4142 	case AF_INET6:
4143 		ip6 = mtod(m, struct ip6_hdr *);
4144 		ip6->ip6_vfc = IPV6_VERSION;
4145 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4146 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4147 		ip6->ip6_nxt = IPPROTO_TCP;
4148 		/* ip6_plen will be updated in ip6_output() */
4149 		th = (struct tcphdr *)(ip6 + 1);
4150 		th->th_dport = sc->sc_src.sin6.sin6_port;
4151 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4152 		break;
4153 #endif
4154 	default:
4155 		th = NULL;
4156 	}
4157 
4158 	th->th_seq = htonl(sc->sc_iss);
4159 	th->th_ack = htonl(sc->sc_irs + 1);
4160 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4161 	th->th_flags = TH_SYN|TH_ACK;
4162 	th->th_win = htons(sc->sc_win);
4163 	/* th_sum already 0 */
4164 	/* th_urp already 0 */
4165 
4166 	/* Tack on the TCP options. */
4167 	optp = (u_int8_t *)(th + 1);
4168 	*optp++ = TCPOPT_MAXSEG;
4169 	*optp++ = 4;
4170 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4171 	*optp++ = sc->sc_ourmaxseg & 0xff;
4172 
4173 	if (sc->sc_request_r_scale != 15) {
4174 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4175 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4176 		    sc->sc_request_r_scale);
4177 		optp += 4;
4178 	}
4179 
4180 	if (sc->sc_flags & SCF_TIMESTAMP) {
4181 		u_int32_t *lp = (u_int32_t *)(optp);
4182 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4183 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4184 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4185 		*lp   = htonl(sc->sc_timestamp);
4186 		optp += TCPOLEN_TSTAMP_APPA;
4187 	}
4188 
4189 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4190 		u_int8_t *p = optp;
4191 
4192 		/* Let the peer know that we will SACK. */
4193 		p[0] = TCPOPT_SACK_PERMITTED;
4194 		p[1] = 2;
4195 		p[2] = TCPOPT_NOP;
4196 		p[3] = TCPOPT_NOP;
4197 		optp += 4;
4198 	}
4199 
4200 	/*
4201 	 * Send ECN SYN-ACK setup packet.
4202 	 * Routes can be asymetric, so, even if we receive a packet
4203 	 * with ECE and CWR set, we must not assume no one will block
4204 	 * the ECE packet we are about to send.
4205 	 */
4206 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4207 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4208 		th->th_flags |= TH_ECE;
4209 		tcpstat.tcps_ecn_shs++;
4210 
4211 		/*
4212 		 * draft-ietf-tcpm-ecnsyn-00.txt
4213 		 *
4214 		 * "[...] a TCP node MAY respond to an ECN-setup
4215 		 * SYN packet by setting ECT in the responding
4216 		 * ECN-setup SYN/ACK packet, indicating to routers
4217 		 * that the SYN/ACK packet is ECN-Capable.
4218 		 * This allows a congested router along the path
4219 		 * to mark the packet instead of dropping the
4220 		 * packet as an indication of congestion."
4221 		 *
4222 		 * "[...] There can be a great benefit in setting
4223 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4224 		 * Congestion is  most likely to occur in
4225 		 * the server-to-client direction.  As a result,
4226 		 * setting an ECN-capable codepoint in SYN/ACK
4227 		 * packets can reduce the occurence of three-second
4228 		 * retransmit timeouts resulting from the drop
4229 		 * of SYN/ACK packets."
4230 		 *
4231 		 * Page 4 and 6, January 2006.
4232 		 */
4233 
4234 		switch (sc->sc_src.sa.sa_family) {
4235 #ifdef INET
4236 		case AF_INET:
4237 			ip->ip_tos |= IPTOS_ECN_ECT0;
4238 			break;
4239 #endif
4240 #ifdef INET6
4241 		case AF_INET6:
4242 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4243 			break;
4244 #endif
4245 		}
4246 		tcpstat.tcps_ecn_ect++;
4247 	}
4248 
4249 #ifdef TCP_SIGNATURE
4250 	if (sc->sc_flags & SCF_SIGNATURE) {
4251 		struct secasvar *sav;
4252 		u_int8_t *sigp;
4253 
4254 		sav = tcp_signature_getsav(m, th);
4255 
4256 		if (sav == NULL) {
4257 			if (m)
4258 				m_freem(m);
4259 			return (EPERM);
4260 		}
4261 
4262 		*optp++ = TCPOPT_SIGNATURE;
4263 		*optp++ = TCPOLEN_SIGNATURE;
4264 		sigp = optp;
4265 		bzero(optp, TCP_SIGLEN);
4266 		optp += TCP_SIGLEN;
4267 		*optp++ = TCPOPT_NOP;
4268 		*optp++ = TCPOPT_EOL;
4269 
4270 		(void)tcp_signature(m, th, hlen, sav, sigp);
4271 
4272 		key_sa_recordxfer(sav, m);
4273 #ifdef FAST_IPSEC
4274 		KEY_FREESAV(&sav);
4275 #else
4276 		key_freesav(sav);
4277 #endif
4278 	}
4279 #endif
4280 
4281 	/* Compute the packet's checksum. */
4282 	switch (sc->sc_src.sa.sa_family) {
4283 	case AF_INET:
4284 		ip->ip_len = htons(tlen - hlen);
4285 		th->th_sum = 0;
4286 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4287 		break;
4288 #ifdef INET6
4289 	case AF_INET6:
4290 		ip6->ip6_plen = htons(tlen - hlen);
4291 		th->th_sum = 0;
4292 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4293 		break;
4294 #endif
4295 	}
4296 
4297 	/*
4298 	 * Fill in some straggling IP bits.  Note the stack expects
4299 	 * ip_len to be in host order, for convenience.
4300 	 */
4301 	switch (sc->sc_src.sa.sa_family) {
4302 #ifdef INET
4303 	case AF_INET:
4304 		ip->ip_len = htons(tlen);
4305 		ip->ip_ttl = ip_defttl;
4306 		/* XXX tos? */
4307 		break;
4308 #endif
4309 #ifdef INET6
4310 	case AF_INET6:
4311 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4312 		ip6->ip6_vfc |= IPV6_VERSION;
4313 		ip6->ip6_plen = htons(tlen - hlen);
4314 		/* ip6_hlim will be initialized afterwards */
4315 		/* XXX flowlabel? */
4316 		break;
4317 #endif
4318 	}
4319 
4320 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4321 	tp = sc->sc_tp;
4322 
4323 	switch (sc->sc_src.sa.sa_family) {
4324 #ifdef INET
4325 	case AF_INET:
4326 		error = ip_output(m, sc->sc_ipopts, ro,
4327 		    (ip_mtudisc ? IP_MTUDISC : 0),
4328 		    (struct ip_moptions *)NULL, so);
4329 		break;
4330 #endif
4331 #ifdef INET6
4332 	case AF_INET6:
4333 		ip6->ip6_hlim = in6_selecthlim(NULL,
4334 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4335 
4336 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4337 			(struct ip6_moptions *)0, so, NULL);
4338 		break;
4339 #endif
4340 	default:
4341 		error = EAFNOSUPPORT;
4342 		break;
4343 	}
4344 	return (error);
4345 }
4346