xref: /netbsd-src/sys/netinet/tcp_input.c (revision 7c3f385475147b6e1c4753f2bee961630e2dfc40)
1 /*	$NetBSD: tcp_input.c,v 1.284 2008/04/12 05:58:22 thorpej 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 1.284 2008/04/12 05:58:22 thorpej Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 #ifdef TCP_SIGNATURE
175 #include <sys/md5.h>
176 #endif
177 #include <sys/lwp.h> /* for lwp0 */
178 
179 #include <net/if.h>
180 #include <net/route.h>
181 #include <net/if_types.h>
182 
183 #include <netinet/in.h>
184 #include <netinet/in_systm.h>
185 #include <netinet/ip.h>
186 #include <netinet/in_pcb.h>
187 #include <netinet/in_var.h>
188 #include <netinet/ip_var.h>
189 #include <netinet/in_offload.h>
190 
191 #ifdef INET6
192 #ifndef INET
193 #include <netinet/in.h>
194 #endif
195 #include <netinet/ip6.h>
196 #include <netinet6/ip6_var.h>
197 #include <netinet6/in6_pcb.h>
198 #include <netinet6/ip6_var.h>
199 #include <netinet6/in6_var.h>
200 #include <netinet/icmp6.h>
201 #include <netinet6/nd6.h>
202 #ifdef TCP_SIGNATURE
203 #include <netinet6/scope6_var.h>
204 #endif
205 #endif
206 
207 #ifndef INET6
208 /* always need ip6.h for IP6_EXTHDR_GET */
209 #include <netinet/ip6.h>
210 #endif
211 
212 #include <netinet/tcp.h>
213 #include <netinet/tcp_fsm.h>
214 #include <netinet/tcp_seq.h>
215 #include <netinet/tcp_timer.h>
216 #include <netinet/tcp_var.h>
217 #include <netinet/tcp_private.h>
218 #include <netinet/tcpip.h>
219 #include <netinet/tcp_congctl.h>
220 #include <netinet/tcp_debug.h>
221 
222 #include <machine/stdarg.h>
223 
224 #ifdef IPSEC
225 #include <netinet6/ipsec.h>
226 #include <netkey/key.h>
227 #endif /*IPSEC*/
228 #ifdef INET6
229 #include "faith.h"
230 #if defined(NFAITH) && NFAITH > 0
231 #include <net/if_faith.h>
232 #endif
233 #endif	/* IPSEC */
234 
235 #ifdef FAST_IPSEC
236 #include <netipsec/ipsec.h>
237 #include <netipsec/ipsec_var.h>			/* XXX ipsecstat namespace */
238 #include <netipsec/key.h>
239 #ifdef INET6
240 #include <netipsec/ipsec6.h>
241 #endif
242 #endif	/* FAST_IPSEC*/
243 
244 int	tcprexmtthresh = 3;
245 int	tcp_log_refused;
246 
247 int	tcp_do_autorcvbuf = 0;
248 int	tcp_autorcvbuf_inc = 16 * 1024;
249 int	tcp_autorcvbuf_max = 256 * 1024;
250 
251 static int tcp_rst_ppslim_count = 0;
252 static struct timeval tcp_rst_ppslim_last;
253 static int tcp_ackdrop_ppslim_count = 0;
254 static struct timeval tcp_ackdrop_ppslim_last;
255 
256 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
257 
258 /* for modulo comparisons of timestamps */
259 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
260 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
261 
262 /*
263  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
264  */
265 #ifdef INET6
266 static inline void
267 nd6_hint(struct tcpcb *tp)
268 {
269 	struct rtentry *rt;
270 
271 	if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 &&
272 	    (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL)
273 		nd6_nud_hint(rt, NULL, 0);
274 }
275 #else
276 static inline void
277 nd6_hint(struct tcpcb *tp)
278 {
279 }
280 #endif
281 
282 /*
283  * Compute ACK transmission behavior.  Delay the ACK unless
284  * we have already delayed an ACK (must send an ACK every two segments).
285  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
286  * option is enabled.
287  */
288 static void
289 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th)
290 {
291 
292 	if (tp->t_flags & TF_DELACK ||
293 	    (tcp_ack_on_push && th->th_flags & TH_PUSH))
294 		tp->t_flags |= TF_ACKNOW;
295 	else
296 		TCP_SET_DELACK(tp);
297 }
298 
299 static void
300 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked)
301 {
302 
303 	/*
304 	 * If we had a pending ICMP message that refers to data that have
305 	 * just been acknowledged, disregard the recorded ICMP message.
306 	 */
307 	if ((tp->t_flags & TF_PMTUD_PEND) &&
308 	    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
309 		tp->t_flags &= ~TF_PMTUD_PEND;
310 
311 	/*
312 	 * Keep track of the largest chunk of data
313 	 * acknowledged since last PMTU update
314 	 */
315 	if (tp->t_pmtud_mss_acked < acked)
316 		tp->t_pmtud_mss_acked = acked;
317 }
318 
319 /*
320  * Convert TCP protocol fields to host order for easier processing.
321  */
322 static void
323 tcp_fields_to_host(struct tcphdr *th)
324 {
325 
326 	NTOHL(th->th_seq);
327 	NTOHL(th->th_ack);
328 	NTOHS(th->th_win);
329 	NTOHS(th->th_urp);
330 }
331 
332 /*
333  * ... and reverse the above.
334  */
335 static void
336 tcp_fields_to_net(struct tcphdr *th)
337 {
338 
339 	HTONL(th->th_seq);
340 	HTONL(th->th_ack);
341 	HTONS(th->th_win);
342 	HTONS(th->th_urp);
343 }
344 
345 #ifdef TCP_CSUM_COUNTERS
346 #include <sys/device.h>
347 
348 #if defined(INET)
349 extern struct evcnt tcp_hwcsum_ok;
350 extern struct evcnt tcp_hwcsum_bad;
351 extern struct evcnt tcp_hwcsum_data;
352 extern struct evcnt tcp_swcsum;
353 #endif /* defined(INET) */
354 #if defined(INET6)
355 extern struct evcnt tcp6_hwcsum_ok;
356 extern struct evcnt tcp6_hwcsum_bad;
357 extern struct evcnt tcp6_hwcsum_data;
358 extern struct evcnt tcp6_swcsum;
359 #endif /* defined(INET6) */
360 
361 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
362 
363 #else
364 
365 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
366 
367 #endif /* TCP_CSUM_COUNTERS */
368 
369 #ifdef TCP_REASS_COUNTERS
370 #include <sys/device.h>
371 
372 extern struct evcnt tcp_reass_;
373 extern struct evcnt tcp_reass_empty;
374 extern struct evcnt tcp_reass_iteration[8];
375 extern struct evcnt tcp_reass_prependfirst;
376 extern struct evcnt tcp_reass_prepend;
377 extern struct evcnt tcp_reass_insert;
378 extern struct evcnt tcp_reass_inserttail;
379 extern struct evcnt tcp_reass_append;
380 extern struct evcnt tcp_reass_appendtail;
381 extern struct evcnt tcp_reass_overlaptail;
382 extern struct evcnt tcp_reass_overlapfront;
383 extern struct evcnt tcp_reass_segdup;
384 extern struct evcnt tcp_reass_fragdup;
385 
386 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
387 
388 #else
389 
390 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
391 
392 #endif /* TCP_REASS_COUNTERS */
393 
394 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *,
395     int *);
396 static int tcp_dooptions(struct tcpcb *, const u_char *, int,
397     struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *);
398 
399 #ifdef INET
400 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
401 #endif
402 #ifdef INET6
403 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
404 #endif
405 
406 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
407 
408 #if defined(MBUFTRACE)
409 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass");
410 #endif /* defined(MBUFTRACE) */
411 
412 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl",
413     NULL, IPL_VM);
414 
415 struct ipqent *
416 tcpipqent_alloc(void)
417 {
418 	struct ipqent *ipqe;
419 	int s;
420 
421 	s = splvm();
422 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
423 	splx(s);
424 
425 	return ipqe;
426 }
427 
428 void
429 tcpipqent_free(struct ipqent *ipqe)
430 {
431 	int s;
432 
433 	s = splvm();
434 	pool_put(&tcpipqent_pool, ipqe);
435 	splx(s);
436 }
437 
438 static int
439 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen)
440 {
441 	struct ipqent *p, *q, *nq, *tiqe = NULL;
442 	struct socket *so = NULL;
443 	int pkt_flags;
444 	tcp_seq pkt_seq;
445 	unsigned pkt_len;
446 	u_long rcvpartdupbyte = 0;
447 	u_long rcvoobyte;
448 #ifdef TCP_REASS_COUNTERS
449 	u_int count = 0;
450 #endif
451 	uint64_t *tcps;
452 
453 	if (tp->t_inpcb)
454 		so = tp->t_inpcb->inp_socket;
455 #ifdef INET6
456 	else if (tp->t_in6pcb)
457 		so = tp->t_in6pcb->in6p_socket;
458 #endif
459 
460 	TCP_REASS_LOCK_CHECK(tp);
461 
462 	/*
463 	 * Call with th==0 after become established to
464 	 * force pre-ESTABLISHED data up to user socket.
465 	 */
466 	if (th == 0)
467 		goto present;
468 
469 	m_claimm(m, &tcp_reass_mowner);
470 
471 	rcvoobyte = *tlen;
472 	/*
473 	 * Copy these to local variables because the tcpiphdr
474 	 * gets munged while we are collapsing mbufs.
475 	 */
476 	pkt_seq = th->th_seq;
477 	pkt_len = *tlen;
478 	pkt_flags = th->th_flags;
479 
480 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
481 
482 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
483 		/*
484 		 * When we miss a packet, the vast majority of time we get
485 		 * packets that follow it in order.  So optimize for that.
486 		 */
487 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
488 			p->ipqe_len += pkt_len;
489 			p->ipqe_flags |= pkt_flags;
490 			m_cat(p->ipre_mlast, m);
491 			TRAVERSE(p->ipre_mlast);
492 			m = NULL;
493 			tiqe = p;
494 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
495 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
496 			goto skip_replacement;
497 		}
498 		/*
499 		 * While we're here, if the pkt is completely beyond
500 		 * anything we have, just insert it at the tail.
501 		 */
502 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
503 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
504 			goto insert_it;
505 		}
506 	}
507 
508 	q = TAILQ_FIRST(&tp->segq);
509 
510 	if (q != NULL) {
511 		/*
512 		 * If this segment immediately precedes the first out-of-order
513 		 * block, simply slap the segment in front of it and (mostly)
514 		 * skip the complicated logic.
515 		 */
516 		if (pkt_seq + pkt_len == q->ipqe_seq) {
517 			q->ipqe_seq = pkt_seq;
518 			q->ipqe_len += pkt_len;
519 			q->ipqe_flags |= pkt_flags;
520 			m_cat(m, q->ipqe_m);
521 			q->ipqe_m = m;
522 			q->ipre_mlast = m; /* last mbuf may have changed */
523 			TRAVERSE(q->ipre_mlast);
524 			tiqe = q;
525 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
526 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
527 			goto skip_replacement;
528 		}
529 	} else {
530 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
531 	}
532 
533 	/*
534 	 * Find a segment which begins after this one does.
535 	 */
536 	for (p = NULL; q != NULL; q = nq) {
537 		nq = TAILQ_NEXT(q, ipqe_q);
538 #ifdef TCP_REASS_COUNTERS
539 		count++;
540 #endif
541 		/*
542 		 * If the received segment is just right after this
543 		 * fragment, merge the two together and then check
544 		 * for further overlaps.
545 		 */
546 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
547 #ifdef TCPREASS_DEBUG
548 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
549 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
550 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
551 #endif
552 			pkt_len += q->ipqe_len;
553 			pkt_flags |= q->ipqe_flags;
554 			pkt_seq = q->ipqe_seq;
555 			m_cat(q->ipre_mlast, m);
556 			TRAVERSE(q->ipre_mlast);
557 			m = q->ipqe_m;
558 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
559 			goto free_ipqe;
560 		}
561 		/*
562 		 * If the received segment is completely past this
563 		 * fragment, we need to go the next fragment.
564 		 */
565 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
566 			p = q;
567 			continue;
568 		}
569 		/*
570 		 * If the fragment is past the received segment,
571 		 * it (or any following) can't be concatenated.
572 		 */
573 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
574 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
575 			break;
576 		}
577 
578 		/*
579 		 * We've received all the data in this segment before.
580 		 * mark it as a duplicate and return.
581 		 */
582 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
583 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
584 			tcps = TCP_STAT_GETREF();
585 			tcps[TCP_STAT_RCVDUPPACK]++;
586 			tcps[TCP_STAT_RCVDUPBYTE] += pkt_len;
587 			TCP_STAT_PUTREF();
588 			tcp_new_dsack(tp, pkt_seq, pkt_len);
589 			m_freem(m);
590 			if (tiqe != NULL) {
591 				tcpipqent_free(tiqe);
592 			}
593 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
594 			return (0);
595 		}
596 		/*
597 		 * Received segment completely overlaps this fragment
598 		 * so we drop the fragment (this keeps the temporal
599 		 * ordering of segments correct).
600 		 */
601 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
602 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
603 			rcvpartdupbyte += q->ipqe_len;
604 			m_freem(q->ipqe_m);
605 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
606 			goto free_ipqe;
607 		}
608 		/*
609 		 * RX'ed segment extends past the end of the
610 		 * fragment.  Drop the overlapping bytes.  Then
611 		 * merge the fragment and segment then treat as
612 		 * a longer received packet.
613 		 */
614 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
615 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
616 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
617 #ifdef TCPREASS_DEBUG
618 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
619 			       tp, overlap,
620 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
621 #endif
622 			m_adj(m, overlap);
623 			rcvpartdupbyte += overlap;
624 			m_cat(q->ipre_mlast, m);
625 			TRAVERSE(q->ipre_mlast);
626 			m = q->ipqe_m;
627 			pkt_seq = q->ipqe_seq;
628 			pkt_len += q->ipqe_len - overlap;
629 			rcvoobyte -= overlap;
630 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
631 			goto free_ipqe;
632 		}
633 		/*
634 		 * RX'ed segment extends past the front of the
635 		 * fragment.  Drop the overlapping bytes on the
636 		 * received packet.  The packet will then be
637 		 * contatentated with this fragment a bit later.
638 		 */
639 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
640 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
641 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
642 #ifdef TCPREASS_DEBUG
643 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
644 			       tp, overlap,
645 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
646 #endif
647 			m_adj(m, -overlap);
648 			pkt_len -= overlap;
649 			rcvpartdupbyte += overlap;
650 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
651 			rcvoobyte -= overlap;
652 		}
653 		/*
654 		 * If the received segment immediates precedes this
655 		 * fragment then tack the fragment onto this segment
656 		 * and reinsert the data.
657 		 */
658 		if (q->ipqe_seq == pkt_seq + pkt_len) {
659 #ifdef TCPREASS_DEBUG
660 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
661 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
662 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
663 #endif
664 			pkt_len += q->ipqe_len;
665 			pkt_flags |= q->ipqe_flags;
666 			m_cat(m, q->ipqe_m);
667 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
668 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
669 			tp->t_segqlen--;
670 			KASSERT(tp->t_segqlen >= 0);
671 			KASSERT(tp->t_segqlen != 0 ||
672 			    (TAILQ_EMPTY(&tp->segq) &&
673 			    TAILQ_EMPTY(&tp->timeq)));
674 			if (tiqe == NULL) {
675 				tiqe = q;
676 			} else {
677 				tcpipqent_free(q);
678 			}
679 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
680 			break;
681 		}
682 		/*
683 		 * If the fragment is before the segment, remember it.
684 		 * When this loop is terminated, p will contain the
685 		 * pointer to fragment that is right before the received
686 		 * segment.
687 		 */
688 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
689 			p = q;
690 
691 		continue;
692 
693 		/*
694 		 * This is a common operation.  It also will allow
695 		 * to save doing a malloc/free in most instances.
696 		 */
697 	  free_ipqe:
698 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
699 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
700 		tp->t_segqlen--;
701 		KASSERT(tp->t_segqlen >= 0);
702 		KASSERT(tp->t_segqlen != 0 ||
703 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
704 		if (tiqe == NULL) {
705 			tiqe = q;
706 		} else {
707 			tcpipqent_free(q);
708 		}
709 	}
710 
711 #ifdef TCP_REASS_COUNTERS
712 	if (count > 7)
713 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
714 	else if (count > 0)
715 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
716 #endif
717 
718     insert_it:
719 
720 	/*
721 	 * Allocate a new queue entry since the received segment did not
722 	 * collapse onto any other out-of-order block; thus we are allocating
723 	 * a new block.  If it had collapsed, tiqe would not be NULL and
724 	 * we would be reusing it.
725 	 * XXX If we can't, just drop the packet.  XXX
726 	 */
727 	if (tiqe == NULL) {
728 		tiqe = tcpipqent_alloc();
729 		if (tiqe == NULL) {
730 			TCP_STATINC(TCP_STAT_RCVMEMDROP);
731 			m_freem(m);
732 			return (0);
733 		}
734 	}
735 
736 	/*
737 	 * Update the counters.
738 	 */
739 	tcps = TCP_STAT_GETREF();
740 	tcps[TCP_STAT_RCVOOPACK]++;
741 	tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte;
742 	if (rcvpartdupbyte) {
743 	    tcps[TCP_STAT_RCVPARTDUPPACK]++;
744 	    tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte;
745 	}
746 	TCP_STAT_PUTREF();
747 
748 	/*
749 	 * Insert the new fragment queue entry into both queues.
750 	 */
751 	tiqe->ipqe_m = m;
752 	tiqe->ipre_mlast = m;
753 	tiqe->ipqe_seq = pkt_seq;
754 	tiqe->ipqe_len = pkt_len;
755 	tiqe->ipqe_flags = pkt_flags;
756 	if (p == NULL) {
757 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
758 #ifdef TCPREASS_DEBUG
759 		if (tiqe->ipqe_seq != tp->rcv_nxt)
760 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
761 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
762 #endif
763 	} else {
764 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
765 #ifdef TCPREASS_DEBUG
766 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
767 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
768 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
769 #endif
770 	}
771 	tp->t_segqlen++;
772 
773 skip_replacement:
774 
775 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
776 
777 present:
778 	/*
779 	 * Present data to user, advancing rcv_nxt through
780 	 * completed sequence space.
781 	 */
782 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
783 		return (0);
784 	q = TAILQ_FIRST(&tp->segq);
785 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
786 		return (0);
787 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
788 		return (0);
789 
790 	tp->rcv_nxt += q->ipqe_len;
791 	pkt_flags = q->ipqe_flags & TH_FIN;
792 	nd6_hint(tp);
793 
794 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
795 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
796 	tp->t_segqlen--;
797 	KASSERT(tp->t_segqlen >= 0);
798 	KASSERT(tp->t_segqlen != 0 ||
799 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
800 	if (so->so_state & SS_CANTRCVMORE)
801 		m_freem(q->ipqe_m);
802 	else
803 		sbappendstream(&so->so_rcv, q->ipqe_m);
804 	tcpipqent_free(q);
805 	sorwakeup(so);
806 	return (pkt_flags);
807 }
808 
809 #ifdef INET6
810 int
811 tcp6_input(struct mbuf **mp, int *offp, int proto)
812 {
813 	struct mbuf *m = *mp;
814 
815 	/*
816 	 * draft-itojun-ipv6-tcp-to-anycast
817 	 * better place to put this in?
818 	 */
819 	if (m->m_flags & M_ANYCAST6) {
820 		struct ip6_hdr *ip6;
821 		if (m->m_len < sizeof(struct ip6_hdr)) {
822 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
823 				TCP_STATINC(TCP_STAT_RCVSHORT);
824 				return IPPROTO_DONE;
825 			}
826 		}
827 		ip6 = mtod(m, struct ip6_hdr *);
828 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
829 		    (char *)&ip6->ip6_dst - (char *)ip6);
830 		return IPPROTO_DONE;
831 	}
832 
833 	tcp_input(m, *offp, proto);
834 	return IPPROTO_DONE;
835 }
836 #endif
837 
838 #ifdef INET
839 static void
840 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
841 {
842 	char src[4*sizeof "123"];
843 	char dst[4*sizeof "123"];
844 
845 	if (ip) {
846 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
847 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
848 	}
849 	else {
850 		strlcpy(src, "(unknown)", sizeof(src));
851 		strlcpy(dst, "(unknown)", sizeof(dst));
852 	}
853 	log(LOG_INFO,
854 	    "Connection attempt to TCP %s:%d from %s:%d\n",
855 	    dst, ntohs(th->th_dport),
856 	    src, ntohs(th->th_sport));
857 }
858 #endif
859 
860 #ifdef INET6
861 static void
862 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
863 {
864 	char src[INET6_ADDRSTRLEN];
865 	char dst[INET6_ADDRSTRLEN];
866 
867 	if (ip6) {
868 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
869 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
870 	}
871 	else {
872 		strlcpy(src, "(unknown v6)", sizeof(src));
873 		strlcpy(dst, "(unknown v6)", sizeof(dst));
874 	}
875 	log(LOG_INFO,
876 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
877 	    dst, ntohs(th->th_dport),
878 	    src, ntohs(th->th_sport));
879 }
880 #endif
881 
882 /*
883  * Checksum extended TCP header and data.
884  */
885 int
886 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th,
887     int toff, int off, int tlen)
888 {
889 
890 	/*
891 	 * XXX it's better to record and check if this mbuf is
892 	 * already checked.
893 	 */
894 
895 	switch (af) {
896 #ifdef INET
897 	case AF_INET:
898 		switch (m->m_pkthdr.csum_flags &
899 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
900 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
901 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
902 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
903 			goto badcsum;
904 
905 		case M_CSUM_TCPv4|M_CSUM_DATA: {
906 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
907 
908 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
909 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
910 				const struct ip *ip =
911 				    mtod(m, const struct ip *);
912 
913 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
914 				    ip->ip_dst.s_addr,
915 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
916 			}
917 			if ((hw_csum ^ 0xffff) != 0)
918 				goto badcsum;
919 			break;
920 		}
921 
922 		case M_CSUM_TCPv4:
923 			/* Checksum was okay. */
924 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
925 			break;
926 
927 		default:
928 			/*
929 			 * Must compute it ourselves.  Maybe skip checksum
930 			 * on loopback interfaces.
931 			 */
932 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
933 					     IFF_LOOPBACK) ||
934 					   tcp_do_loopback_cksum)) {
935 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
936 				if (in4_cksum(m, IPPROTO_TCP, toff,
937 					      tlen + off) != 0)
938 					goto badcsum;
939 			}
940 			break;
941 		}
942 		break;
943 #endif /* INET4 */
944 
945 #ifdef INET6
946 	case AF_INET6:
947 		switch (m->m_pkthdr.csum_flags &
948 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
949 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
950 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
951 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
952 			goto badcsum;
953 
954 #if 0 /* notyet */
955 		case M_CSUM_TCPv6|M_CSUM_DATA:
956 #endif
957 
958 		case M_CSUM_TCPv6:
959 			/* Checksum was okay. */
960 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok);
961 			break;
962 
963 		default:
964 			/*
965 			 * Must compute it ourselves.  Maybe skip checksum
966 			 * on loopback interfaces.
967 			 */
968 			if (__predict_true((m->m_flags & M_LOOP) == 0 ||
969 			    tcp_do_loopback_cksum)) {
970 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
971 				if (in6_cksum(m, IPPROTO_TCP, toff,
972 				    tlen + off) != 0)
973 					goto badcsum;
974 			}
975 		}
976 		break;
977 #endif /* INET6 */
978 	}
979 
980 	return 0;
981 
982 badcsum:
983 	TCP_STATINC(TCP_STAT_RCVBADSUM);
984 	return -1;
985 }
986 
987 /*
988  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
989  */
990 void
991 tcp_input(struct mbuf *m, ...)
992 {
993 	struct tcphdr *th;
994 	struct ip *ip;
995 	struct inpcb *inp;
996 #ifdef INET6
997 	struct ip6_hdr *ip6;
998 	struct in6pcb *in6p;
999 #endif
1000 	u_int8_t *optp = NULL;
1001 	int optlen = 0;
1002 	int len, tlen, toff, hdroptlen = 0;
1003 	struct tcpcb *tp = 0;
1004 	int tiflags;
1005 	struct socket *so = NULL;
1006 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
1007 #ifdef TCP_DEBUG
1008 	short ostate = 0;
1009 #endif
1010 	u_long tiwin;
1011 	struct tcp_opt_info opti;
1012 	int off, iphlen;
1013 	va_list ap;
1014 	int af;		/* af on the wire */
1015 	struct mbuf *tcp_saveti = NULL;
1016 	uint32_t ts_rtt;
1017 	uint8_t iptos;
1018 	uint64_t *tcps;
1019 
1020 	MCLAIM(m, &tcp_rx_mowner);
1021 	va_start(ap, m);
1022 	toff = va_arg(ap, int);
1023 	(void)va_arg(ap, int);		/* ignore value, advance ap */
1024 	va_end(ap);
1025 
1026 	TCP_STATINC(TCP_STAT_RCVTOTAL);
1027 
1028 	bzero(&opti, sizeof(opti));
1029 	opti.ts_present = 0;
1030 	opti.maxseg = 0;
1031 
1032 	/*
1033 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
1034 	 *
1035 	 * TCP is, by definition, unicast, so we reject all
1036 	 * multicast outright.
1037 	 *
1038 	 * Note, there are additional src/dst address checks in
1039 	 * the AF-specific code below.
1040 	 */
1041 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1042 		/* XXX stat */
1043 		goto drop;
1044 	}
1045 #ifdef INET6
1046 	if (m->m_flags & M_ANYCAST6) {
1047 		/* XXX stat */
1048 		goto drop;
1049 	}
1050 #endif
1051 
1052 	/*
1053 	 * Get IP and TCP header.
1054 	 * Note: IP leaves IP header in first mbuf.
1055 	 */
1056 	ip = mtod(m, struct ip *);
1057 #ifdef INET6
1058 	ip6 = NULL;
1059 #endif
1060 	switch (ip->ip_v) {
1061 #ifdef INET
1062 	case 4:
1063 		af = AF_INET;
1064 		iphlen = sizeof(struct ip);
1065 		ip = mtod(m, struct ip *);
1066 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1067 			sizeof(struct tcphdr));
1068 		if (th == NULL) {
1069 			TCP_STATINC(TCP_STAT_RCVSHORT);
1070 			return;
1071 		}
1072 		/* We do the checksum after PCB lookup... */
1073 		len = ntohs(ip->ip_len);
1074 		tlen = len - toff;
1075 		iptos = ip->ip_tos;
1076 		break;
1077 #endif
1078 #ifdef INET6
1079 	case 6:
1080 		ip = NULL;
1081 		iphlen = sizeof(struct ip6_hdr);
1082 		af = AF_INET6;
1083 		ip6 = mtod(m, struct ip6_hdr *);
1084 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1085 			sizeof(struct tcphdr));
1086 		if (th == NULL) {
1087 			TCP_STATINC(TCP_STAT_RCVSHORT);
1088 			return;
1089 		}
1090 
1091 		/* Be proactive about malicious use of IPv4 mapped address */
1092 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1093 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1094 			/* XXX stat */
1095 			goto drop;
1096 		}
1097 
1098 		/*
1099 		 * Be proactive about unspecified IPv6 address in source.
1100 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1101 		 * unspecified IPv6 address can be used to confuse us.
1102 		 *
1103 		 * Note that packets with unspecified IPv6 destination is
1104 		 * already dropped in ip6_input.
1105 		 */
1106 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1107 			/* XXX stat */
1108 			goto drop;
1109 		}
1110 
1111 		/*
1112 		 * Make sure destination address is not multicast.
1113 		 * Source address checked in ip6_input().
1114 		 */
1115 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1116 			/* XXX stat */
1117 			goto drop;
1118 		}
1119 
1120 		/* We do the checksum after PCB lookup... */
1121 		len = m->m_pkthdr.len;
1122 		tlen = len - toff;
1123 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1124 		break;
1125 #endif
1126 	default:
1127 		m_freem(m);
1128 		return;
1129 	}
1130 
1131 	KASSERT(TCP_HDR_ALIGNED_P(th));
1132 
1133 	/*
1134 	 * Check that TCP offset makes sense,
1135 	 * pull out TCP options and adjust length.		XXX
1136 	 */
1137 	off = th->th_off << 2;
1138 	if (off < sizeof (struct tcphdr) || off > tlen) {
1139 		TCP_STATINC(TCP_STAT_RCVBADOFF);
1140 		goto drop;
1141 	}
1142 	tlen -= off;
1143 
1144 	/*
1145 	 * tcp_input() has been modified to use tlen to mean the TCP data
1146 	 * length throughout the function.  Other functions can use
1147 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1148 	 * rja
1149 	 */
1150 
1151 	if (off > sizeof (struct tcphdr)) {
1152 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1153 		if (th == NULL) {
1154 			TCP_STATINC(TCP_STAT_RCVSHORT);
1155 			return;
1156 		}
1157 		/*
1158 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1159 		 * (as they're before toff) and we don't need to update those.
1160 		 */
1161 		KASSERT(TCP_HDR_ALIGNED_P(th));
1162 		optlen = off - sizeof (struct tcphdr);
1163 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1164 		/*
1165 		 * Do quick retrieval of timestamp options ("options
1166 		 * prediction?").  If timestamp is the only option and it's
1167 		 * formatted as recommended in RFC 1323 appendix A, we
1168 		 * quickly get the values now and not bother calling
1169 		 * tcp_dooptions(), etc.
1170 		 */
1171 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1172 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1173 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1174 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1175 		     (th->th_flags & TH_SYN) == 0) {
1176 			opti.ts_present = 1;
1177 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1178 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1179 			optp = NULL;	/* we've parsed the options */
1180 		}
1181 	}
1182 	tiflags = th->th_flags;
1183 
1184 	/*
1185 	 * Locate pcb for segment.
1186 	 */
1187 findpcb:
1188 	inp = NULL;
1189 #ifdef INET6
1190 	in6p = NULL;
1191 #endif
1192 	switch (af) {
1193 #ifdef INET
1194 	case AF_INET:
1195 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1196 		    ip->ip_dst, th->th_dport);
1197 		if (inp == 0) {
1198 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1199 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1200 		}
1201 #ifdef INET6
1202 		if (inp == 0) {
1203 			struct in6_addr s, d;
1204 
1205 			/* mapped addr case */
1206 			bzero(&s, sizeof(s));
1207 			s.s6_addr16[5] = htons(0xffff);
1208 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1209 			bzero(&d, sizeof(d));
1210 			d.s6_addr16[5] = htons(0xffff);
1211 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1212 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1213 			    th->th_sport, &d, th->th_dport, 0);
1214 			if (in6p == 0) {
1215 				TCP_STATINC(TCP_STAT_PCBHASHMISS);
1216 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1217 				    th->th_dport, 0);
1218 			}
1219 		}
1220 #endif
1221 #ifndef INET6
1222 		if (inp == 0)
1223 #else
1224 		if (inp == 0 && in6p == 0)
1225 #endif
1226 		{
1227 			TCP_STATINC(TCP_STAT_NOPORT);
1228 			if (tcp_log_refused &&
1229 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1230 				tcp4_log_refused(ip, th);
1231 			}
1232 			tcp_fields_to_host(th);
1233 			goto dropwithreset_ratelim;
1234 		}
1235 #if defined(IPSEC) || defined(FAST_IPSEC)
1236 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1237 		    ipsec4_in_reject(m, inp)) {
1238 			ipsecstat.in_polvio++;
1239 			goto drop;
1240 		}
1241 #ifdef INET6
1242 		else if (in6p &&
1243 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1244 		    ipsec6_in_reject_so(m, in6p->in6p_socket)) {
1245 			ipsecstat.in_polvio++;
1246 			goto drop;
1247 		}
1248 #endif
1249 #endif /*IPSEC*/
1250 		break;
1251 #endif /*INET*/
1252 #ifdef INET6
1253 	case AF_INET6:
1254 	    {
1255 		int faith;
1256 
1257 #if defined(NFAITH) && NFAITH > 0
1258 		faith = faithprefix(&ip6->ip6_dst);
1259 #else
1260 		faith = 0;
1261 #endif
1262 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1263 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1264 		if (in6p == NULL) {
1265 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1266 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1267 				th->th_dport, faith);
1268 		}
1269 		if (in6p == NULL) {
1270 			TCP_STATINC(TCP_STAT_NOPORT);
1271 			if (tcp_log_refused &&
1272 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1273 				tcp6_log_refused(ip6, th);
1274 			}
1275 			tcp_fields_to_host(th);
1276 			goto dropwithreset_ratelim;
1277 		}
1278 #if defined(IPSEC) || defined(FAST_IPSEC)
1279 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1280 		    ipsec6_in_reject(m, in6p)) {
1281 			ipsec6stat.in_polvio++;
1282 			goto drop;
1283 		}
1284 #endif /*IPSEC*/
1285 		break;
1286 	    }
1287 #endif
1288 	}
1289 
1290 	/*
1291 	 * If the state is CLOSED (i.e., TCB does not exist) then
1292 	 * all data in the incoming segment is discarded.
1293 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1294 	 * but should either do a listen or a connect soon.
1295 	 */
1296 	tp = NULL;
1297 	so = NULL;
1298 	if (inp) {
1299 		tp = intotcpcb(inp);
1300 		so = inp->inp_socket;
1301 	}
1302 #ifdef INET6
1303 	else if (in6p) {
1304 		tp = in6totcpcb(in6p);
1305 		so = in6p->in6p_socket;
1306 	}
1307 #endif
1308 	if (tp == 0) {
1309 		tcp_fields_to_host(th);
1310 		goto dropwithreset_ratelim;
1311 	}
1312 	if (tp->t_state == TCPS_CLOSED)
1313 		goto drop;
1314 
1315 	/*
1316 	 * Checksum extended TCP header and data.
1317 	 */
1318 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1319 		goto badcsum;
1320 
1321 	tcp_fields_to_host(th);
1322 
1323 	/* Unscale the window into a 32-bit value. */
1324 	if ((tiflags & TH_SYN) == 0)
1325 		tiwin = th->th_win << tp->snd_scale;
1326 	else
1327 		tiwin = th->th_win;
1328 
1329 #ifdef INET6
1330 	/* save packet options if user wanted */
1331 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1332 		if (in6p->in6p_options) {
1333 			m_freem(in6p->in6p_options);
1334 			in6p->in6p_options = 0;
1335 		}
1336 		KASSERT(ip6 != NULL);
1337 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1338 	}
1339 #endif
1340 
1341 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1342 		union syn_cache_sa src;
1343 		union syn_cache_sa dst;
1344 
1345 		bzero(&src, sizeof(src));
1346 		bzero(&dst, sizeof(dst));
1347 		switch (af) {
1348 #ifdef INET
1349 		case AF_INET:
1350 			src.sin.sin_len = sizeof(struct sockaddr_in);
1351 			src.sin.sin_family = AF_INET;
1352 			src.sin.sin_addr = ip->ip_src;
1353 			src.sin.sin_port = th->th_sport;
1354 
1355 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1356 			dst.sin.sin_family = AF_INET;
1357 			dst.sin.sin_addr = ip->ip_dst;
1358 			dst.sin.sin_port = th->th_dport;
1359 			break;
1360 #endif
1361 #ifdef INET6
1362 		case AF_INET6:
1363 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1364 			src.sin6.sin6_family = AF_INET6;
1365 			src.sin6.sin6_addr = ip6->ip6_src;
1366 			src.sin6.sin6_port = th->th_sport;
1367 
1368 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1369 			dst.sin6.sin6_family = AF_INET6;
1370 			dst.sin6.sin6_addr = ip6->ip6_dst;
1371 			dst.sin6.sin6_port = th->th_dport;
1372 			break;
1373 #endif /* INET6 */
1374 		default:
1375 			goto badsyn;	/*sanity*/
1376 		}
1377 
1378 		if (so->so_options & SO_DEBUG) {
1379 #ifdef TCP_DEBUG
1380 			ostate = tp->t_state;
1381 #endif
1382 
1383 			tcp_saveti = NULL;
1384 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1385 				goto nosave;
1386 
1387 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1388 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1389 				if (!tcp_saveti)
1390 					goto nosave;
1391 			} else {
1392 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1393 				if (!tcp_saveti)
1394 					goto nosave;
1395 				MCLAIM(m, &tcp_mowner);
1396 				tcp_saveti->m_len = iphlen;
1397 				m_copydata(m, 0, iphlen,
1398 				    mtod(tcp_saveti, void *));
1399 			}
1400 
1401 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1402 				m_freem(tcp_saveti);
1403 				tcp_saveti = NULL;
1404 			} else {
1405 				tcp_saveti->m_len += sizeof(struct tcphdr);
1406 				memcpy(mtod(tcp_saveti, char *) + iphlen, th,
1407 				    sizeof(struct tcphdr));
1408 			}
1409 	nosave:;
1410 		}
1411 		if (so->so_options & SO_ACCEPTCONN) {
1412 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1413 				if (tiflags & TH_RST) {
1414 					syn_cache_reset(&src.sa, &dst.sa, th);
1415 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1416 				    (TH_ACK|TH_SYN)) {
1417 					/*
1418 					 * Received a SYN,ACK.  This should
1419 					 * never happen while we are in
1420 					 * LISTEN.  Send an RST.
1421 					 */
1422 					goto badsyn;
1423 				} else if (tiflags & TH_ACK) {
1424 					so = syn_cache_get(&src.sa, &dst.sa,
1425 						th, toff, tlen, so, m);
1426 					if (so == NULL) {
1427 						/*
1428 						 * We don't have a SYN for
1429 						 * this ACK; send an RST.
1430 						 */
1431 						goto badsyn;
1432 					} else if (so ==
1433 					    (struct socket *)(-1)) {
1434 						/*
1435 						 * We were unable to create
1436 						 * the connection.  If the
1437 						 * 3-way handshake was
1438 						 * completed, and RST has
1439 						 * been sent to the peer.
1440 						 * Since the mbuf might be
1441 						 * in use for the reply,
1442 						 * do not free it.
1443 						 */
1444 						m = NULL;
1445 					} else {
1446 						/*
1447 						 * We have created a
1448 						 * full-blown connection.
1449 						 */
1450 						tp = NULL;
1451 						inp = NULL;
1452 #ifdef INET6
1453 						in6p = NULL;
1454 #endif
1455 						switch (so->so_proto->pr_domain->dom_family) {
1456 #ifdef INET
1457 						case AF_INET:
1458 							inp = sotoinpcb(so);
1459 							tp = intotcpcb(inp);
1460 							break;
1461 #endif
1462 #ifdef INET6
1463 						case AF_INET6:
1464 							in6p = sotoin6pcb(so);
1465 							tp = in6totcpcb(in6p);
1466 							break;
1467 #endif
1468 						}
1469 						if (tp == NULL)
1470 							goto badsyn;	/*XXX*/
1471 						tiwin <<= tp->snd_scale;
1472 						goto after_listen;
1473 					}
1474 				} else {
1475 					/*
1476 					 * None of RST, SYN or ACK was set.
1477 					 * This is an invalid packet for a
1478 					 * TCB in LISTEN state.  Send a RST.
1479 					 */
1480 					goto badsyn;
1481 				}
1482 			} else {
1483 				/*
1484 				 * Received a SYN.
1485 				 *
1486 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1487 				 */
1488 				if (m->m_flags & (M_BCAST|M_MCAST))
1489 					goto drop;
1490 
1491 				switch (af) {
1492 #ifdef INET6
1493 				case AF_INET6:
1494 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1495 						goto drop;
1496 					break;
1497 #endif /* INET6 */
1498 				case AF_INET:
1499 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1500 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1501 						goto drop;
1502 				break;
1503 				}
1504 
1505 #ifdef INET6
1506 				/*
1507 				 * If deprecated address is forbidden, we do
1508 				 * not accept SYN to deprecated interface
1509 				 * address to prevent any new inbound
1510 				 * connection from getting established.
1511 				 * When we do not accept SYN, we send a TCP
1512 				 * RST, with deprecated source address (instead
1513 				 * of dropping it).  We compromise it as it is
1514 				 * much better for peer to send a RST, and
1515 				 * RST will be the final packet for the
1516 				 * exchange.
1517 				 *
1518 				 * If we do not forbid deprecated addresses, we
1519 				 * accept the SYN packet.  RFC2462 does not
1520 				 * suggest dropping SYN in this case.
1521 				 * If we decipher RFC2462 5.5.4, it says like
1522 				 * this:
1523 				 * 1. use of deprecated addr with existing
1524 				 *    communication is okay - "SHOULD continue
1525 				 *    to be used"
1526 				 * 2. use of it with new communication:
1527 				 *   (2a) "SHOULD NOT be used if alternate
1528 				 *        address with sufficient scope is
1529 				 *        available"
1530 				 *   (2b) nothing mentioned otherwise.
1531 				 * Here we fall into (2b) case as we have no
1532 				 * choice in our source address selection - we
1533 				 * must obey the peer.
1534 				 *
1535 				 * The wording in RFC2462 is confusing, and
1536 				 * there are multiple description text for
1537 				 * deprecated address handling - worse, they
1538 				 * are not exactly the same.  I believe 5.5.4
1539 				 * is the best one, so we follow 5.5.4.
1540 				 */
1541 				if (af == AF_INET6 && !ip6_use_deprecated) {
1542 					struct in6_ifaddr *ia6;
1543 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1544 					    &ip6->ip6_dst)) &&
1545 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1546 						tp = NULL;
1547 						goto dropwithreset;
1548 					}
1549 				}
1550 #endif
1551 
1552 #if defined(IPSEC) || defined(FAST_IPSEC)
1553 				switch (af) {
1554 #ifdef INET
1555 				case AF_INET:
1556 					if (ipsec4_in_reject_so(m, so)) {
1557 						ipsecstat.in_polvio++;
1558 						tp = NULL;
1559 						goto dropwithreset;
1560 					}
1561 					break;
1562 #endif
1563 #ifdef INET6
1564 				case AF_INET6:
1565 					if (ipsec6_in_reject_so(m, so)) {
1566 						ipsec6stat.in_polvio++;
1567 						tp = NULL;
1568 						goto dropwithreset;
1569 					}
1570 					break;
1571 #endif /*INET6*/
1572 				}
1573 #endif /*IPSEC*/
1574 
1575 				/*
1576 				 * LISTEN socket received a SYN
1577 				 * from itself?  This can't possibly
1578 				 * be valid; drop the packet.
1579 				 */
1580 				if (th->th_sport == th->th_dport) {
1581 					int i;
1582 
1583 					switch (af) {
1584 #ifdef INET
1585 					case AF_INET:
1586 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1587 						break;
1588 #endif
1589 #ifdef INET6
1590 					case AF_INET6:
1591 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1592 						break;
1593 #endif
1594 					default:
1595 						i = 1;
1596 					}
1597 					if (i) {
1598 						TCP_STATINC(TCP_STAT_BADSYN);
1599 						goto drop;
1600 					}
1601 				}
1602 
1603 				/*
1604 				 * SYN looks ok; create compressed TCP
1605 				 * state for it.
1606 				 */
1607 				if (so->so_qlen <= so->so_qlimit &&
1608 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1609 						so, m, optp, optlen, &opti))
1610 					m = NULL;
1611 			}
1612 			goto drop;
1613 		}
1614 	}
1615 
1616 after_listen:
1617 #ifdef DIAGNOSTIC
1618 	/*
1619 	 * Should not happen now that all embryonic connections
1620 	 * are handled with compressed state.
1621 	 */
1622 	if (tp->t_state == TCPS_LISTEN)
1623 		panic("tcp_input: TCPS_LISTEN");
1624 #endif
1625 
1626 	/*
1627 	 * Segment received on connection.
1628 	 * Reset idle time and keep-alive timer.
1629 	 */
1630 	tp->t_rcvtime = tcp_now;
1631 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1632 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1633 
1634 	/*
1635 	 * Process options.
1636 	 */
1637 #ifdef TCP_SIGNATURE
1638 	if (optp || (tp->t_flags & TF_SIGNATURE))
1639 #else
1640 	if (optp)
1641 #endif
1642 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1643 			goto drop;
1644 
1645 	if (TCP_SACK_ENABLED(tp)) {
1646 		tcp_del_sackholes(tp, th);
1647 	}
1648 
1649 	if (TCP_ECN_ALLOWED(tp)) {
1650 		switch (iptos & IPTOS_ECN_MASK) {
1651 		case IPTOS_ECN_CE:
1652 			tp->t_flags |= TF_ECN_SND_ECE;
1653 			TCP_STATINC(TCP_STAT_ECN_CE);
1654 			break;
1655 		case IPTOS_ECN_ECT0:
1656 			TCP_STATINC(TCP_STAT_ECN_ECT);
1657 			break;
1658 		case IPTOS_ECN_ECT1:
1659 			/* XXX: ignore for now -- rpaulo */
1660 			break;
1661 		}
1662 
1663 		if (tiflags & TH_CWR)
1664 			tp->t_flags &= ~TF_ECN_SND_ECE;
1665 
1666 		/*
1667 		 * Congestion experienced.
1668 		 * Ignore if we are already trying to recover.
1669 		 */
1670 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1671 			tp->t_congctl->cong_exp(tp);
1672 	}
1673 
1674 	if (opti.ts_present && opti.ts_ecr) {
1675 		/*
1676 		 * Calculate the RTT from the returned time stamp and the
1677 		 * connection's time base.  If the time stamp is later than
1678 		 * the current time, or is extremely old, fall back to non-1323
1679 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1680 		 * at the same time.
1681 		 */
1682 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1683 		if (ts_rtt > TCP_PAWS_IDLE)
1684 			ts_rtt = 0;
1685 	} else {
1686 		ts_rtt = 0;
1687 	}
1688 
1689 	/*
1690 	 * Header prediction: check for the two common cases
1691 	 * of a uni-directional data xfer.  If the packet has
1692 	 * no control flags, is in-sequence, the window didn't
1693 	 * change and we're not retransmitting, it's a
1694 	 * candidate.  If the length is zero and the ack moved
1695 	 * forward, we're the sender side of the xfer.  Just
1696 	 * free the data acked & wake any higher level process
1697 	 * that was blocked waiting for space.  If the length
1698 	 * is non-zero and the ack didn't move, we're the
1699 	 * receiver side.  If we're getting packets in-order
1700 	 * (the reassembly queue is empty), add the data to
1701 	 * the socket buffer and note that we need a delayed ack.
1702 	 */
1703 	if (tp->t_state == TCPS_ESTABLISHED &&
1704 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1705 	        == TH_ACK &&
1706 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1707 	    th->th_seq == tp->rcv_nxt &&
1708 	    tiwin && tiwin == tp->snd_wnd &&
1709 	    tp->snd_nxt == tp->snd_max) {
1710 
1711 		/*
1712 		 * If last ACK falls within this segment's sequence numbers,
1713 		 * record the timestamp.
1714 		 * NOTE that the test is modified according to the latest
1715 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1716 		 *
1717 		 * note that we already know
1718 		 *	TSTMP_GEQ(opti.ts_val, tp->ts_recent)
1719 		 */
1720 		if (opti.ts_present &&
1721 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1722 			tp->ts_recent_age = tcp_now;
1723 			tp->ts_recent = opti.ts_val;
1724 		}
1725 
1726 		if (tlen == 0) {
1727 			/* Ack prediction. */
1728 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1729 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1730 			    tp->snd_cwnd >= tp->snd_wnd &&
1731 			    tp->t_partialacks < 0) {
1732 				/*
1733 				 * this is a pure ack for outstanding data.
1734 				 */
1735 				if (ts_rtt)
1736 					tcp_xmit_timer(tp, ts_rtt);
1737 				else if (tp->t_rtttime &&
1738 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1739 					tcp_xmit_timer(tp,
1740 					  tcp_now - tp->t_rtttime);
1741 				acked = th->th_ack - tp->snd_una;
1742 				tcps = TCP_STAT_GETREF();
1743 				tcps[TCP_STAT_PREDACK]++;
1744 				tcps[TCP_STAT_RCVACKPACK]++;
1745 				tcps[TCP_STAT_RCVACKBYTE] += acked;
1746 				TCP_STAT_PUTREF();
1747 				nd6_hint(tp);
1748 
1749 				if (acked > (tp->t_lastoff - tp->t_inoff))
1750 					tp->t_lastm = NULL;
1751 				sbdrop(&so->so_snd, acked);
1752 				tp->t_lastoff -= acked;
1753 
1754 				icmp_check(tp, th, acked);
1755 
1756 				tp->snd_una = th->th_ack;
1757 				tp->snd_fack = tp->snd_una;
1758 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1759 					tp->snd_high = tp->snd_una;
1760 				m_freem(m);
1761 
1762 				/*
1763 				 * If all outstanding data are acked, stop
1764 				 * retransmit timer, otherwise restart timer
1765 				 * using current (possibly backed-off) value.
1766 				 * If process is waiting for space,
1767 				 * wakeup/selnotify/signal.  If data
1768 				 * are ready to send, let tcp_output
1769 				 * decide between more output or persist.
1770 				 */
1771 				if (tp->snd_una == tp->snd_max)
1772 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1773 				else if (TCP_TIMER_ISARMED(tp,
1774 				    TCPT_PERSIST) == 0)
1775 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1776 					    tp->t_rxtcur);
1777 
1778 				sowwakeup(so);
1779 				if (so->so_snd.sb_cc)
1780 					(void) tcp_output(tp);
1781 				if (tcp_saveti)
1782 					m_freem(tcp_saveti);
1783 				return;
1784 			}
1785 		} else if (th->th_ack == tp->snd_una &&
1786 		    TAILQ_FIRST(&tp->segq) == NULL &&
1787 		    tlen <= sbspace(&so->so_rcv)) {
1788 			int newsize = 0;	/* automatic sockbuf scaling */
1789 
1790 			/*
1791 			 * this is a pure, in-sequence data packet
1792 			 * with nothing on the reassembly queue and
1793 			 * we have enough buffer space to take it.
1794 			 */
1795 			tp->rcv_nxt += tlen;
1796 			tcps = TCP_STAT_GETREF();
1797 			tcps[TCP_STAT_PREDDAT]++;
1798 			tcps[TCP_STAT_RCVPACK]++;
1799 			tcps[TCP_STAT_RCVBYTE] += tlen;
1800 			TCP_STAT_PUTREF();
1801 			nd6_hint(tp);
1802 
1803 		/*
1804 		 * Automatic sizing enables the performance of large buffers
1805 		 * and most of the efficiency of small ones by only allocating
1806 		 * space when it is needed.
1807 		 *
1808 		 * On the receive side the socket buffer memory is only rarely
1809 		 * used to any significant extent.  This allows us to be much
1810 		 * more aggressive in scaling the receive socket buffer.  For
1811 		 * the case that the buffer space is actually used to a large
1812 		 * extent and we run out of kernel memory we can simply drop
1813 		 * the new segments; TCP on the sender will just retransmit it
1814 		 * later.  Setting the buffer size too big may only consume too
1815 		 * much kernel memory if the application doesn't read() from
1816 		 * the socket or packet loss or reordering makes use of the
1817 		 * reassembly queue.
1818 		 *
1819 		 * The criteria to step up the receive buffer one notch are:
1820 		 *  1. the number of bytes received during the time it takes
1821 		 *     one timestamp to be reflected back to us (the RTT);
1822 		 *  2. received bytes per RTT is within seven eighth of the
1823 		 *     current socket buffer size;
1824 		 *  3. receive buffer size has not hit maximal automatic size;
1825 		 *
1826 		 * This algorithm does one step per RTT at most and only if
1827 		 * we receive a bulk stream w/o packet losses or reorderings.
1828 		 * Shrinking the buffer during idle times is not necessary as
1829 		 * it doesn't consume any memory when idle.
1830 		 *
1831 		 * TODO: Only step up if the application is actually serving
1832 		 * the buffer to better manage the socket buffer resources.
1833 		 */
1834 			if (tcp_do_autorcvbuf &&
1835 			    opti.ts_ecr &&
1836 			    (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1837 				if (opti.ts_ecr > tp->rfbuf_ts &&
1838 				    opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) {
1839 					if (tp->rfbuf_cnt >
1840 					    (so->so_rcv.sb_hiwat / 8 * 7) &&
1841 					    so->so_rcv.sb_hiwat <
1842 					    tcp_autorcvbuf_max) {
1843 						newsize =
1844 						    min(so->so_rcv.sb_hiwat +
1845 						    tcp_autorcvbuf_inc,
1846 						    tcp_autorcvbuf_max);
1847 					}
1848 					/* Start over with next RTT. */
1849 					tp->rfbuf_ts = 0;
1850 					tp->rfbuf_cnt = 0;
1851 				} else
1852 					tp->rfbuf_cnt += tlen;	/* add up */
1853 			}
1854 
1855 			/*
1856 			 * Drop TCP, IP headers and TCP options then add data
1857 			 * to socket buffer.
1858 			 */
1859 			if (so->so_state & SS_CANTRCVMORE)
1860 				m_freem(m);
1861 			else {
1862 				/*
1863 				 * Set new socket buffer size.
1864 				 * Give up when limit is reached.
1865 				 */
1866 				if (newsize)
1867 					if (!sbreserve(&so->so_rcv,
1868 					    newsize, so))
1869 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1870 				m_adj(m, toff + off);
1871 				sbappendstream(&so->so_rcv, m);
1872 			}
1873 			sorwakeup(so);
1874 			tcp_setup_ack(tp, th);
1875 			if (tp->t_flags & TF_ACKNOW)
1876 				(void) tcp_output(tp);
1877 			if (tcp_saveti)
1878 				m_freem(tcp_saveti);
1879 			return;
1880 		}
1881 	}
1882 
1883 	/*
1884 	 * Compute mbuf offset to TCP data segment.
1885 	 */
1886 	hdroptlen = toff + off;
1887 
1888 	/*
1889 	 * Calculate amount of space in receive window,
1890 	 * and then do TCP input processing.
1891 	 * Receive window is amount of space in rcv queue,
1892 	 * but not less than advertised window.
1893 	 */
1894 	{ int win;
1895 
1896 	win = sbspace(&so->so_rcv);
1897 	if (win < 0)
1898 		win = 0;
1899 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1900 	}
1901 
1902 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1903 	tp->rfbuf_ts = 0;
1904 	tp->rfbuf_cnt = 0;
1905 
1906 	switch (tp->t_state) {
1907 	/*
1908 	 * If the state is SYN_SENT:
1909 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1910 	 *	if seg contains a RST, then drop the connection.
1911 	 *	if seg does not contain SYN, then drop it.
1912 	 * Otherwise this is an acceptable SYN segment
1913 	 *	initialize tp->rcv_nxt and tp->irs
1914 	 *	if seg contains ack then advance tp->snd_una
1915 	 *	if seg contains a ECE and ECN support is enabled, the stream
1916 	 *	    is ECN capable.
1917 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1918 	 *	arrange for segment to be acked (eventually)
1919 	 *	continue processing rest of data/controls, beginning with URG
1920 	 */
1921 	case TCPS_SYN_SENT:
1922 		if ((tiflags & TH_ACK) &&
1923 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1924 		     SEQ_GT(th->th_ack, tp->snd_max)))
1925 			goto dropwithreset;
1926 		if (tiflags & TH_RST) {
1927 			if (tiflags & TH_ACK)
1928 				tp = tcp_drop(tp, ECONNREFUSED);
1929 			goto drop;
1930 		}
1931 		if ((tiflags & TH_SYN) == 0)
1932 			goto drop;
1933 		if (tiflags & TH_ACK) {
1934 			tp->snd_una = th->th_ack;
1935 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1936 				tp->snd_nxt = tp->snd_una;
1937 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1938 				tp->snd_high = tp->snd_una;
1939 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1940 
1941 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1942 				tp->t_flags |= TF_ECN_PERMIT;
1943 				TCP_STATINC(TCP_STAT_ECN_SHS);
1944 			}
1945 
1946 		}
1947 		tp->irs = th->th_seq;
1948 		tcp_rcvseqinit(tp);
1949 		tp->t_flags |= TF_ACKNOW;
1950 		tcp_mss_from_peer(tp, opti.maxseg);
1951 
1952 		/*
1953 		 * Initialize the initial congestion window.  If we
1954 		 * had to retransmit the SYN, we must initialize cwnd
1955 		 * to 1 segment (i.e. the Loss Window).
1956 		 */
1957 		if (tp->t_flags & TF_SYN_REXMT)
1958 			tp->snd_cwnd = tp->t_peermss;
1959 		else {
1960 			int ss = tcp_init_win;
1961 #ifdef INET
1962 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1963 				ss = tcp_init_win_local;
1964 #endif
1965 #ifdef INET6
1966 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1967 				ss = tcp_init_win_local;
1968 #endif
1969 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1970 		}
1971 
1972 		tcp_rmx_rtt(tp);
1973 		if (tiflags & TH_ACK) {
1974 			TCP_STATINC(TCP_STAT_CONNECTS);
1975 			soisconnected(so);
1976 			tcp_established(tp);
1977 			/* Do window scaling on this connection? */
1978 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1979 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1980 				tp->snd_scale = tp->requested_s_scale;
1981 				tp->rcv_scale = tp->request_r_scale;
1982 			}
1983 			TCP_REASS_LOCK(tp);
1984 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1985 			TCP_REASS_UNLOCK(tp);
1986 			/*
1987 			 * if we didn't have to retransmit the SYN,
1988 			 * use its rtt as our initial srtt & rtt var.
1989 			 */
1990 			if (tp->t_rtttime)
1991 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1992 		} else
1993 			tp->t_state = TCPS_SYN_RECEIVED;
1994 
1995 		/*
1996 		 * Advance th->th_seq to correspond to first data byte.
1997 		 * If data, trim to stay within window,
1998 		 * dropping FIN if necessary.
1999 		 */
2000 		th->th_seq++;
2001 		if (tlen > tp->rcv_wnd) {
2002 			todrop = tlen - tp->rcv_wnd;
2003 			m_adj(m, -todrop);
2004 			tlen = tp->rcv_wnd;
2005 			tiflags &= ~TH_FIN;
2006 			tcps = TCP_STAT_GETREF();
2007 			tcps[TCP_STAT_RCVPACKAFTERWIN]++;
2008 			tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop;
2009 			TCP_STAT_PUTREF();
2010 		}
2011 		tp->snd_wl1 = th->th_seq - 1;
2012 		tp->rcv_up = th->th_seq;
2013 		goto step6;
2014 
2015 	/*
2016 	 * If the state is SYN_RECEIVED:
2017 	 *	If seg contains an ACK, but not for our SYN, drop the input
2018 	 *	and generate an RST.  See page 36, rfc793
2019 	 */
2020 	case TCPS_SYN_RECEIVED:
2021 		if ((tiflags & TH_ACK) &&
2022 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
2023 		     SEQ_GT(th->th_ack, tp->snd_max)))
2024 			goto dropwithreset;
2025 		break;
2026 	}
2027 
2028 	/*
2029 	 * States other than LISTEN or SYN_SENT.
2030 	 * First check timestamp, if present.
2031 	 * Then check that at least some bytes of segment are within
2032 	 * receive window.  If segment begins before rcv_nxt,
2033 	 * drop leading data (and SYN); if nothing left, just ack.
2034 	 *
2035 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2036 	 * and it's less than ts_recent, drop it.
2037 	 */
2038 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
2039 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
2040 
2041 		/* Check to see if ts_recent is over 24 days old.  */
2042 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
2043 			/*
2044 			 * Invalidate ts_recent.  If this segment updates
2045 			 * ts_recent, the age will be reset later and ts_recent
2046 			 * will get a valid value.  If it does not, setting
2047 			 * ts_recent to zero will at least satisfy the
2048 			 * requirement that zero be placed in the timestamp
2049 			 * echo reply when ts_recent isn't valid.  The
2050 			 * age isn't reset until we get a valid ts_recent
2051 			 * because we don't want out-of-order segments to be
2052 			 * dropped when ts_recent is old.
2053 			 */
2054 			tp->ts_recent = 0;
2055 		} else {
2056 			tcps = TCP_STAT_GETREF();
2057 			tcps[TCP_STAT_RCVDUPPACK]++;
2058 			tcps[TCP_STAT_RCVDUPBYTE] += tlen;
2059 			tcps[TCP_STAT_PAWSDROP]++;
2060 			TCP_STAT_PUTREF();
2061 			tcp_new_dsack(tp, th->th_seq, tlen);
2062 			goto dropafterack;
2063 		}
2064 	}
2065 
2066 	todrop = tp->rcv_nxt - th->th_seq;
2067 	dupseg = false;
2068 	if (todrop > 0) {
2069 		if (tiflags & TH_SYN) {
2070 			tiflags &= ~TH_SYN;
2071 			th->th_seq++;
2072 			if (th->th_urp > 1)
2073 				th->th_urp--;
2074 			else {
2075 				tiflags &= ~TH_URG;
2076 				th->th_urp = 0;
2077 			}
2078 			todrop--;
2079 		}
2080 		if (todrop > tlen ||
2081 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
2082 			/*
2083 			 * Any valid FIN or RST must be to the left of the
2084 			 * window.  At this point the FIN or RST must be a
2085 			 * duplicate or out of sequence; drop it.
2086 			 */
2087 			if (tiflags & TH_RST)
2088 				goto drop;
2089 			tiflags &= ~(TH_FIN|TH_RST);
2090 			/*
2091 			 * Send an ACK to resynchronize and drop any data.
2092 			 * But keep on processing for RST or ACK.
2093 			 */
2094 			tp->t_flags |= TF_ACKNOW;
2095 			todrop = tlen;
2096 			dupseg = true;
2097 			tcps = TCP_STAT_GETREF();
2098 			tcps[TCP_STAT_RCVDUPPACK]++;
2099 			tcps[TCP_STAT_RCVDUPBYTE] += todrop;
2100 			TCP_STAT_PUTREF();
2101 		} else if ((tiflags & TH_RST) &&
2102 			   th->th_seq != tp->last_ack_sent) {
2103 			/*
2104 			 * Test for reset before adjusting the sequence
2105 			 * number for overlapping data.
2106 			 */
2107 			goto dropafterack_ratelim;
2108 		} else {
2109 			tcps = TCP_STAT_GETREF();
2110 			tcps[TCP_STAT_RCVPARTDUPPACK]++;
2111 			tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop;
2112 			TCP_STAT_PUTREF();
2113 		}
2114 		tcp_new_dsack(tp, th->th_seq, todrop);
2115 		hdroptlen += todrop;	/*drop from head afterwards*/
2116 		th->th_seq += todrop;
2117 		tlen -= todrop;
2118 		if (th->th_urp > todrop)
2119 			th->th_urp -= todrop;
2120 		else {
2121 			tiflags &= ~TH_URG;
2122 			th->th_urp = 0;
2123 		}
2124 	}
2125 
2126 	/*
2127 	 * If new data are received on a connection after the
2128 	 * user processes are gone, then RST the other end.
2129 	 */
2130 	if ((so->so_state & SS_NOFDREF) &&
2131 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2132 		tp = tcp_close(tp);
2133 		TCP_STATINC(TCP_STAT_RCVAFTERCLOSE);
2134 		goto dropwithreset;
2135 	}
2136 
2137 	/*
2138 	 * If segment ends after window, drop trailing data
2139 	 * (and PUSH and FIN); if nothing left, just ACK.
2140 	 */
2141 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2142 	if (todrop > 0) {
2143 		TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN);
2144 		if (todrop >= tlen) {
2145 			/*
2146 			 * The segment actually starts after the window.
2147 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2148 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2149 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2150 			 */
2151 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen);
2152 			/*
2153 			 * If a new connection request is received
2154 			 * while in TIME_WAIT, drop the old connection
2155 			 * and start over if the sequence numbers
2156 			 * are above the previous ones.
2157 			 *
2158 			 * NOTE: We will checksum the packet again, and
2159 			 * so we need to put the header fields back into
2160 			 * network order!
2161 			 * XXX This kind of sucks, but we don't expect
2162 			 * XXX this to happen very often, so maybe it
2163 			 * XXX doesn't matter so much.
2164 			 */
2165 			if (tiflags & TH_SYN &&
2166 			    tp->t_state == TCPS_TIME_WAIT &&
2167 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2168 				tp = tcp_close(tp);
2169 				tcp_fields_to_net(th);
2170 				goto findpcb;
2171 			}
2172 			/*
2173 			 * If window is closed can only take segments at
2174 			 * window edge, and have to drop data and PUSH from
2175 			 * incoming segments.  Continue processing, but
2176 			 * remember to ack.  Otherwise, drop segment
2177 			 * and (if not RST) ack.
2178 			 */
2179 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2180 				tp->t_flags |= TF_ACKNOW;
2181 				TCP_STATINC(TCP_STAT_RCVWINPROBE);
2182 			} else
2183 				goto dropafterack;
2184 		} else
2185 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop);
2186 		m_adj(m, -todrop);
2187 		tlen -= todrop;
2188 		tiflags &= ~(TH_PUSH|TH_FIN);
2189 	}
2190 
2191 	/*
2192 	 * If last ACK falls within this segment's sequence numbers,
2193 	 *  record the timestamp.
2194 	 * NOTE:
2195 	 * 1) That the test incorporates suggestions from the latest
2196 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2197 	 * 2) That updating only on newer timestamps interferes with
2198 	 *    our earlier PAWS tests, so this check should be solely
2199 	 *    predicated on the sequence space of this segment.
2200 	 * 3) That we modify the segment boundary check to be
2201 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2202 	 *    instead of RFC1323's
2203 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2204 	 *    This modified check allows us to overcome RFC1323's
2205 	 *    limitations as described in Stevens TCP/IP Illustrated
2206 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2207 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2208 	 */
2209 	if (opti.ts_present &&
2210 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2211 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2212 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2213 		tp->ts_recent_age = tcp_now;
2214 		tp->ts_recent = opti.ts_val;
2215 	}
2216 
2217 	/*
2218 	 * If the RST bit is set examine the state:
2219 	 *    SYN_RECEIVED STATE:
2220 	 *	If passive open, return to LISTEN state.
2221 	 *	If active open, inform user that connection was refused.
2222 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2223 	 *	Inform user that connection was reset, and close tcb.
2224 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2225 	 *	Close the tcb.
2226 	 */
2227 	if (tiflags & TH_RST) {
2228 		if (th->th_seq != tp->last_ack_sent)
2229 			goto dropafterack_ratelim;
2230 
2231 		switch (tp->t_state) {
2232 		case TCPS_SYN_RECEIVED:
2233 			so->so_error = ECONNREFUSED;
2234 			goto close;
2235 
2236 		case TCPS_ESTABLISHED:
2237 		case TCPS_FIN_WAIT_1:
2238 		case TCPS_FIN_WAIT_2:
2239 		case TCPS_CLOSE_WAIT:
2240 			so->so_error = ECONNRESET;
2241 		close:
2242 			tp->t_state = TCPS_CLOSED;
2243 			TCP_STATINC(TCP_STAT_DROPS);
2244 			tp = tcp_close(tp);
2245 			goto drop;
2246 
2247 		case TCPS_CLOSING:
2248 		case TCPS_LAST_ACK:
2249 		case TCPS_TIME_WAIT:
2250 			tp = tcp_close(tp);
2251 			goto drop;
2252 		}
2253 	}
2254 
2255 	/*
2256 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2257 	 * we must be in a synchronized state.  RFC791 states (under RST
2258 	 * generation) that any unacceptable segment (an out-of-order SYN
2259 	 * qualifies) received in a synchronized state must elicit only an
2260 	 * empty acknowledgment segment ... and the connection remains in
2261 	 * the same state.
2262 	 */
2263 	if (tiflags & TH_SYN) {
2264 		if (tp->rcv_nxt == th->th_seq) {
2265 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2266 			    TH_ACK);
2267 			if (tcp_saveti)
2268 				m_freem(tcp_saveti);
2269 			return;
2270 		}
2271 
2272 		goto dropafterack_ratelim;
2273 	}
2274 
2275 	/*
2276 	 * If the ACK bit is off we drop the segment and return.
2277 	 */
2278 	if ((tiflags & TH_ACK) == 0) {
2279 		if (tp->t_flags & TF_ACKNOW)
2280 			goto dropafterack;
2281 		else
2282 			goto drop;
2283 	}
2284 
2285 	/*
2286 	 * Ack processing.
2287 	 */
2288 	switch (tp->t_state) {
2289 
2290 	/*
2291 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2292 	 * ESTABLISHED state and continue processing, otherwise
2293 	 * send an RST.
2294 	 */
2295 	case TCPS_SYN_RECEIVED:
2296 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2297 		    SEQ_GT(th->th_ack, tp->snd_max))
2298 			goto dropwithreset;
2299 		TCP_STATINC(TCP_STAT_CONNECTS);
2300 		soisconnected(so);
2301 		tcp_established(tp);
2302 		/* Do window scaling? */
2303 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2304 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2305 			tp->snd_scale = tp->requested_s_scale;
2306 			tp->rcv_scale = tp->request_r_scale;
2307 		}
2308 		TCP_REASS_LOCK(tp);
2309 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2310 		TCP_REASS_UNLOCK(tp);
2311 		tp->snd_wl1 = th->th_seq - 1;
2312 		/* fall into ... */
2313 
2314 	/*
2315 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2316 	 * ACKs.  If the ack is in the range
2317 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2318 	 * then advance tp->snd_una to th->th_ack and drop
2319 	 * data from the retransmission queue.  If this ACK reflects
2320 	 * more up to date window information we update our window information.
2321 	 */
2322 	case TCPS_ESTABLISHED:
2323 	case TCPS_FIN_WAIT_1:
2324 	case TCPS_FIN_WAIT_2:
2325 	case TCPS_CLOSE_WAIT:
2326 	case TCPS_CLOSING:
2327 	case TCPS_LAST_ACK:
2328 	case TCPS_TIME_WAIT:
2329 
2330 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2331 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2332 				TCP_STATINC(TCP_STAT_RCVDUPPACK);
2333 				/*
2334 				 * If we have outstanding data (other than
2335 				 * a window probe), this is a completely
2336 				 * duplicate ack (ie, window info didn't
2337 				 * change), the ack is the biggest we've
2338 				 * seen and we've seen exactly our rexmt
2339 				 * threshhold of them, assume a packet
2340 				 * has been dropped and retransmit it.
2341 				 * Kludge snd_nxt & the congestion
2342 				 * window so we send only this one
2343 				 * packet.
2344 				 */
2345 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2346 				    th->th_ack != tp->snd_una)
2347 					tp->t_dupacks = 0;
2348 				else if (tp->t_partialacks < 0 &&
2349 					 (++tp->t_dupacks == tcprexmtthresh ||
2350 					 TCP_FACK_FASTRECOV(tp))) {
2351 					/*
2352 					 * Do the fast retransmit, and adjust
2353 					 * congestion control paramenters.
2354 					 */
2355 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2356 						/* False fast retransmit */
2357 						break;
2358 					} else
2359 						goto drop;
2360 				} else if (tp->t_dupacks > tcprexmtthresh) {
2361 					tp->snd_cwnd += tp->t_segsz;
2362 					(void) tcp_output(tp);
2363 					goto drop;
2364 				}
2365 			} else {
2366 				/*
2367 				 * If the ack appears to be very old, only
2368 				 * allow data that is in-sequence.  This
2369 				 * makes it somewhat more difficult to insert
2370 				 * forged data by guessing sequence numbers.
2371 				 * Sent an ack to try to update the send
2372 				 * sequence number on the other side.
2373 				 */
2374 				if (tlen && th->th_seq != tp->rcv_nxt &&
2375 				    SEQ_LT(th->th_ack,
2376 				    tp->snd_una - tp->max_sndwnd))
2377 					goto dropafterack;
2378 			}
2379 			break;
2380 		}
2381 		/*
2382 		 * If the congestion window was inflated to account
2383 		 * for the other side's cached packets, retract it.
2384 		 */
2385 		/* XXX: make SACK have his own congestion control
2386 		 * struct -- rpaulo */
2387 		if (TCP_SACK_ENABLED(tp))
2388 			tcp_sack_newack(tp, th);
2389 		else
2390 			tp->t_congctl->fast_retransmit_newack(tp, th);
2391 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2392 			TCP_STATINC(TCP_STAT_RCVACKTOOMUCH);
2393 			goto dropafterack;
2394 		}
2395 		acked = th->th_ack - tp->snd_una;
2396 		tcps = TCP_STAT_GETREF();
2397 		tcps[TCP_STAT_RCVACKPACK]++;
2398 		tcps[TCP_STAT_RCVACKBYTE] += acked;
2399 		TCP_STAT_PUTREF();
2400 
2401 		/*
2402 		 * If we have a timestamp reply, update smoothed
2403 		 * round trip time.  If no timestamp is present but
2404 		 * transmit timer is running and timed sequence
2405 		 * number was acked, update smoothed round trip time.
2406 		 * Since we now have an rtt measurement, cancel the
2407 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2408 		 * Recompute the initial retransmit timer.
2409 		 */
2410 		if (ts_rtt)
2411 			tcp_xmit_timer(tp, ts_rtt);
2412 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2413 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2414 
2415 		/*
2416 		 * If all outstanding data is acked, stop retransmit
2417 		 * timer and remember to restart (more output or persist).
2418 		 * If there is more data to be acked, restart retransmit
2419 		 * timer, using current (possibly backed-off) value.
2420 		 */
2421 		if (th->th_ack == tp->snd_max) {
2422 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2423 			needoutput = 1;
2424 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2425 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2426 
2427 		/*
2428 		 * New data has been acked, adjust the congestion window.
2429 		 */
2430 		tp->t_congctl->newack(tp, th);
2431 
2432 		nd6_hint(tp);
2433 		if (acked > so->so_snd.sb_cc) {
2434 			tp->snd_wnd -= so->so_snd.sb_cc;
2435 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2436 			ourfinisacked = 1;
2437 		} else {
2438 			if (acked > (tp->t_lastoff - tp->t_inoff))
2439 				tp->t_lastm = NULL;
2440 			sbdrop(&so->so_snd, acked);
2441 			tp->t_lastoff -= acked;
2442 			tp->snd_wnd -= acked;
2443 			ourfinisacked = 0;
2444 		}
2445 		sowwakeup(so);
2446 
2447 		icmp_check(tp, th, acked);
2448 
2449 		tp->snd_una = th->th_ack;
2450 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2451 			tp->snd_fack = tp->snd_una;
2452 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2453 			tp->snd_nxt = tp->snd_una;
2454 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2455 			tp->snd_high = tp->snd_una;
2456 
2457 		switch (tp->t_state) {
2458 
2459 		/*
2460 		 * In FIN_WAIT_1 STATE in addition to the processing
2461 		 * for the ESTABLISHED state if our FIN is now acknowledged
2462 		 * then enter FIN_WAIT_2.
2463 		 */
2464 		case TCPS_FIN_WAIT_1:
2465 			if (ourfinisacked) {
2466 				/*
2467 				 * If we can't receive any more
2468 				 * data, then closing user can proceed.
2469 				 * Starting the timer is contrary to the
2470 				 * specification, but if we don't get a FIN
2471 				 * we'll hang forever.
2472 				 */
2473 				if (so->so_state & SS_CANTRCVMORE) {
2474 					soisdisconnected(so);
2475 					if (tp->t_maxidle > 0)
2476 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2477 						    tp->t_maxidle);
2478 				}
2479 				tp->t_state = TCPS_FIN_WAIT_2;
2480 			}
2481 			break;
2482 
2483 	 	/*
2484 		 * In CLOSING STATE in addition to the processing for
2485 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2486 		 * then enter the TIME-WAIT state, otherwise ignore
2487 		 * the segment.
2488 		 */
2489 		case TCPS_CLOSING:
2490 			if (ourfinisacked) {
2491 				tp->t_state = TCPS_TIME_WAIT;
2492 				tcp_canceltimers(tp);
2493 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2494 				soisdisconnected(so);
2495 			}
2496 			break;
2497 
2498 		/*
2499 		 * In LAST_ACK, we may still be waiting for data to drain
2500 		 * and/or to be acked, as well as for the ack of our FIN.
2501 		 * If our FIN is now acknowledged, delete the TCB,
2502 		 * enter the closed state and return.
2503 		 */
2504 		case TCPS_LAST_ACK:
2505 			if (ourfinisacked) {
2506 				tp = tcp_close(tp);
2507 				goto drop;
2508 			}
2509 			break;
2510 
2511 		/*
2512 		 * In TIME_WAIT state the only thing that should arrive
2513 		 * is a retransmission of the remote FIN.  Acknowledge
2514 		 * it and restart the finack timer.
2515 		 */
2516 		case TCPS_TIME_WAIT:
2517 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2518 			goto dropafterack;
2519 		}
2520 	}
2521 
2522 step6:
2523 	/*
2524 	 * Update window information.
2525 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2526 	 */
2527 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2528 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2529 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2530 		/* keep track of pure window updates */
2531 		if (tlen == 0 &&
2532 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2533 			TCP_STATINC(TCP_STAT_RCVWINUPD);
2534 		tp->snd_wnd = tiwin;
2535 		tp->snd_wl1 = th->th_seq;
2536 		tp->snd_wl2 = th->th_ack;
2537 		if (tp->snd_wnd > tp->max_sndwnd)
2538 			tp->max_sndwnd = tp->snd_wnd;
2539 		needoutput = 1;
2540 	}
2541 
2542 	/*
2543 	 * Process segments with URG.
2544 	 */
2545 	if ((tiflags & TH_URG) && th->th_urp &&
2546 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2547 		/*
2548 		 * This is a kludge, but if we receive and accept
2549 		 * random urgent pointers, we'll crash in
2550 		 * soreceive.  It's hard to imagine someone
2551 		 * actually wanting to send this much urgent data.
2552 		 */
2553 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2554 			th->th_urp = 0;			/* XXX */
2555 			tiflags &= ~TH_URG;		/* XXX */
2556 			goto dodata;			/* XXX */
2557 		}
2558 		/*
2559 		 * If this segment advances the known urgent pointer,
2560 		 * then mark the data stream.  This should not happen
2561 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2562 		 * a FIN has been received from the remote side.
2563 		 * In these states we ignore the URG.
2564 		 *
2565 		 * According to RFC961 (Assigned Protocols),
2566 		 * the urgent pointer points to the last octet
2567 		 * of urgent data.  We continue, however,
2568 		 * to consider it to indicate the first octet
2569 		 * of data past the urgent section as the original
2570 		 * spec states (in one of two places).
2571 		 */
2572 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2573 			tp->rcv_up = th->th_seq + th->th_urp;
2574 			so->so_oobmark = so->so_rcv.sb_cc +
2575 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2576 			if (so->so_oobmark == 0)
2577 				so->so_state |= SS_RCVATMARK;
2578 			sohasoutofband(so);
2579 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2580 		}
2581 		/*
2582 		 * Remove out of band data so doesn't get presented to user.
2583 		 * This can happen independent of advancing the URG pointer,
2584 		 * but if two URG's are pending at once, some out-of-band
2585 		 * data may creep in... ick.
2586 		 */
2587 		if (th->th_urp <= (u_int16_t) tlen
2588 #ifdef SO_OOBINLINE
2589 		     && (so->so_options & SO_OOBINLINE) == 0
2590 #endif
2591 		     )
2592 			tcp_pulloutofband(so, th, m, hdroptlen);
2593 	} else
2594 		/*
2595 		 * If no out of band data is expected,
2596 		 * pull receive urgent pointer along
2597 		 * with the receive window.
2598 		 */
2599 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2600 			tp->rcv_up = tp->rcv_nxt;
2601 dodata:							/* XXX */
2602 
2603 	/*
2604 	 * Process the segment text, merging it into the TCP sequencing queue,
2605 	 * and arranging for acknowledgement of receipt if necessary.
2606 	 * This process logically involves adjusting tp->rcv_wnd as data
2607 	 * is presented to the user (this happens in tcp_usrreq.c,
2608 	 * case PRU_RCVD).  If a FIN has already been received on this
2609 	 * connection then we just ignore the text.
2610 	 */
2611 	if ((tlen || (tiflags & TH_FIN)) &&
2612 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2613 		/*
2614 		 * Insert segment ti into reassembly queue of tcp with
2615 		 * control block tp.  Return TH_FIN if reassembly now includes
2616 		 * a segment with FIN.  The macro form does the common case
2617 		 * inline (segment is the next to be received on an
2618 		 * established connection, and the queue is empty),
2619 		 * avoiding linkage into and removal from the queue and
2620 		 * repetition of various conversions.
2621 		 * Set DELACK for segments received in order, but ack
2622 		 * immediately when segments are out of order
2623 		 * (so fast retransmit can work).
2624 		 */
2625 		/* NOTE: this was TCP_REASS() macro, but used only once */
2626 		TCP_REASS_LOCK(tp);
2627 		if (th->th_seq == tp->rcv_nxt &&
2628 		    TAILQ_FIRST(&tp->segq) == NULL &&
2629 		    tp->t_state == TCPS_ESTABLISHED) {
2630 			tcp_setup_ack(tp, th);
2631 			tp->rcv_nxt += tlen;
2632 			tiflags = th->th_flags & TH_FIN;
2633 			tcps = TCP_STAT_GETREF();
2634 			tcps[TCP_STAT_RCVPACK]++;
2635 			tcps[TCP_STAT_RCVBYTE] += tlen;
2636 			TCP_STAT_PUTREF();
2637 			nd6_hint(tp);
2638 			if (so->so_state & SS_CANTRCVMORE)
2639 				m_freem(m);
2640 			else {
2641 				m_adj(m, hdroptlen);
2642 				sbappendstream(&(so)->so_rcv, m);
2643 			}
2644 			sorwakeup(so);
2645 		} else {
2646 			m_adj(m, hdroptlen);
2647 			tiflags = tcp_reass(tp, th, m, &tlen);
2648 			tp->t_flags |= TF_ACKNOW;
2649 		}
2650 		TCP_REASS_UNLOCK(tp);
2651 
2652 		/*
2653 		 * Note the amount of data that peer has sent into
2654 		 * our window, in order to estimate the sender's
2655 		 * buffer size.
2656 		 */
2657 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2658 	} else {
2659 		m_freem(m);
2660 		m = NULL;
2661 		tiflags &= ~TH_FIN;
2662 	}
2663 
2664 	/*
2665 	 * If FIN is received ACK the FIN and let the user know
2666 	 * that the connection is closing.  Ignore a FIN received before
2667 	 * the connection is fully established.
2668 	 */
2669 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2670 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2671 			socantrcvmore(so);
2672 			tp->t_flags |= TF_ACKNOW;
2673 			tp->rcv_nxt++;
2674 		}
2675 		switch (tp->t_state) {
2676 
2677 	 	/*
2678 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2679 		 */
2680 		case TCPS_ESTABLISHED:
2681 			tp->t_state = TCPS_CLOSE_WAIT;
2682 			break;
2683 
2684 	 	/*
2685 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2686 		 * enter the CLOSING state.
2687 		 */
2688 		case TCPS_FIN_WAIT_1:
2689 			tp->t_state = TCPS_CLOSING;
2690 			break;
2691 
2692 	 	/*
2693 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2694 		 * starting the time-wait timer, turning off the other
2695 		 * standard timers.
2696 		 */
2697 		case TCPS_FIN_WAIT_2:
2698 			tp->t_state = TCPS_TIME_WAIT;
2699 			tcp_canceltimers(tp);
2700 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2701 			soisdisconnected(so);
2702 			break;
2703 
2704 		/*
2705 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2706 		 */
2707 		case TCPS_TIME_WAIT:
2708 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2709 			break;
2710 		}
2711 	}
2712 #ifdef TCP_DEBUG
2713 	if (so->so_options & SO_DEBUG)
2714 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2715 #endif
2716 
2717 	/*
2718 	 * Return any desired output.
2719 	 */
2720 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2721 		(void) tcp_output(tp);
2722 	}
2723 	if (tcp_saveti)
2724 		m_freem(tcp_saveti);
2725 	return;
2726 
2727 badsyn:
2728 	/*
2729 	 * Received a bad SYN.  Increment counters and dropwithreset.
2730 	 */
2731 	TCP_STATINC(TCP_STAT_BADSYN);
2732 	tp = NULL;
2733 	goto dropwithreset;
2734 
2735 dropafterack:
2736 	/*
2737 	 * Generate an ACK dropping incoming segment if it occupies
2738 	 * sequence space, where the ACK reflects our state.
2739 	 */
2740 	if (tiflags & TH_RST)
2741 		goto drop;
2742 	goto dropafterack2;
2743 
2744 dropafterack_ratelim:
2745 	/*
2746 	 * We may want to rate-limit ACKs against SYN/RST attack.
2747 	 */
2748 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2749 	    tcp_ackdrop_ppslim) == 0) {
2750 		/* XXX stat */
2751 		goto drop;
2752 	}
2753 	/* ...fall into dropafterack2... */
2754 
2755 dropafterack2:
2756 	m_freem(m);
2757 	tp->t_flags |= TF_ACKNOW;
2758 	(void) tcp_output(tp);
2759 	if (tcp_saveti)
2760 		m_freem(tcp_saveti);
2761 	return;
2762 
2763 dropwithreset_ratelim:
2764 	/*
2765 	 * We may want to rate-limit RSTs in certain situations,
2766 	 * particularly if we are sending an RST in response to
2767 	 * an attempt to connect to or otherwise communicate with
2768 	 * a port for which we have no socket.
2769 	 */
2770 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2771 	    tcp_rst_ppslim) == 0) {
2772 		/* XXX stat */
2773 		goto drop;
2774 	}
2775 	/* ...fall into dropwithreset... */
2776 
2777 dropwithreset:
2778 	/*
2779 	 * Generate a RST, dropping incoming segment.
2780 	 * Make ACK acceptable to originator of segment.
2781 	 */
2782 	if (tiflags & TH_RST)
2783 		goto drop;
2784 
2785 	switch (af) {
2786 #ifdef INET6
2787 	case AF_INET6:
2788 		/* For following calls to tcp_respond */
2789 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2790 			goto drop;
2791 		break;
2792 #endif /* INET6 */
2793 	case AF_INET:
2794 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2795 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2796 			goto drop;
2797 	}
2798 
2799 	if (tiflags & TH_ACK)
2800 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2801 	else {
2802 		if (tiflags & TH_SYN)
2803 			tlen++;
2804 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2805 		    TH_RST|TH_ACK);
2806 	}
2807 	if (tcp_saveti)
2808 		m_freem(tcp_saveti);
2809 	return;
2810 
2811 badcsum:
2812 drop:
2813 	/*
2814 	 * Drop space held by incoming segment and return.
2815 	 */
2816 	if (tp) {
2817 		if (tp->t_inpcb)
2818 			so = tp->t_inpcb->inp_socket;
2819 #ifdef INET6
2820 		else if (tp->t_in6pcb)
2821 			so = tp->t_in6pcb->in6p_socket;
2822 #endif
2823 		else
2824 			so = NULL;
2825 #ifdef TCP_DEBUG
2826 		if (so && (so->so_options & SO_DEBUG) != 0)
2827 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2828 #endif
2829 	}
2830 	if (tcp_saveti)
2831 		m_freem(tcp_saveti);
2832 	m_freem(m);
2833 	return;
2834 }
2835 
2836 #ifdef TCP_SIGNATURE
2837 int
2838 tcp_signature_apply(void *fstate, void *data, u_int len)
2839 {
2840 
2841 	MD5Update(fstate, (u_char *)data, len);
2842 	return (0);
2843 }
2844 
2845 struct secasvar *
2846 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2847 {
2848 	struct secasvar *sav;
2849 #ifdef FAST_IPSEC
2850 	union sockaddr_union dst;
2851 #endif
2852 	struct ip *ip;
2853 	struct ip6_hdr *ip6;
2854 
2855 	ip = mtod(m, struct ip *);
2856 	switch (ip->ip_v) {
2857 	case 4:
2858 		ip = mtod(m, struct ip *);
2859 		ip6 = NULL;
2860 		break;
2861 	case 6:
2862 		ip = NULL;
2863 		ip6 = mtod(m, struct ip6_hdr *);
2864 		break;
2865 	default:
2866 		return (NULL);
2867 	}
2868 
2869 #ifdef FAST_IPSEC
2870 	/* Extract the destination from the IP header in the mbuf. */
2871 	bzero(&dst, sizeof(union sockaddr_union));
2872 	if (ip !=NULL) {
2873 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2874 		dst.sa.sa_family = AF_INET;
2875 		dst.sin.sin_addr = ip->ip_dst;
2876 	} else {
2877 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2878 		dst.sa.sa_family = AF_INET6;
2879 		dst.sin6.sin6_addr = ip6->ip6_dst;
2880 	}
2881 
2882 	/*
2883 	 * Look up an SADB entry which matches the address of the peer.
2884 	 */
2885 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2886 #else
2887 	if (ip)
2888 		sav = key_allocsa(AF_INET, (void *)&ip->ip_src,
2889 		    (void *)&ip->ip_dst, IPPROTO_TCP,
2890 		    htonl(TCP_SIG_SPI), 0, 0);
2891 	else
2892 		sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src,
2893 		    (void *)&ip6->ip6_dst, IPPROTO_TCP,
2894 		    htonl(TCP_SIG_SPI), 0, 0);
2895 #endif
2896 
2897 	return (sav);	/* freesav must be performed by caller */
2898 }
2899 
2900 int
2901 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2902     struct secasvar *sav, char *sig)
2903 {
2904 	MD5_CTX ctx;
2905 	struct ip *ip;
2906 	struct ipovly *ipovly;
2907 	struct ip6_hdr *ip6;
2908 	struct ippseudo ippseudo;
2909 	struct ip6_hdr_pseudo ip6pseudo;
2910 	struct tcphdr th0;
2911 	int l, tcphdrlen;
2912 
2913 	if (sav == NULL)
2914 		return (-1);
2915 
2916 	tcphdrlen = th->th_off * 4;
2917 
2918 	switch (mtod(m, struct ip *)->ip_v) {
2919 	case 4:
2920 		ip = mtod(m, struct ip *);
2921 		ip6 = NULL;
2922 		break;
2923 	case 6:
2924 		ip = NULL;
2925 		ip6 = mtod(m, struct ip6_hdr *);
2926 		break;
2927 	default:
2928 		return (-1);
2929 	}
2930 
2931 	MD5Init(&ctx);
2932 
2933 	if (ip) {
2934 		memset(&ippseudo, 0, sizeof(ippseudo));
2935 		ipovly = (struct ipovly *)ip;
2936 		ippseudo.ippseudo_src = ipovly->ih_src;
2937 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2938 		ippseudo.ippseudo_pad = 0;
2939 		ippseudo.ippseudo_p = IPPROTO_TCP;
2940 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2941 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2942 	} else {
2943 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2944 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2945 		in6_clearscope(&ip6pseudo.ip6ph_src);
2946 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2947 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2948 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2949 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2950 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2951 	}
2952 
2953 	th0 = *th;
2954 	th0.th_sum = 0;
2955 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2956 
2957 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2958 	if (l > 0)
2959 		m_apply(m, thoff + tcphdrlen,
2960 		    m->m_pkthdr.len - thoff - tcphdrlen,
2961 		    tcp_signature_apply, &ctx);
2962 
2963 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2964 	MD5Final(sig, &ctx);
2965 
2966 	return (0);
2967 }
2968 #endif
2969 
2970 static int
2971 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt,
2972     struct tcphdr *th,
2973     struct mbuf *m, int toff, struct tcp_opt_info *oi)
2974 {
2975 	u_int16_t mss;
2976 	int opt, optlen = 0;
2977 #ifdef TCP_SIGNATURE
2978 	void *sigp = NULL;
2979 	char sigbuf[TCP_SIGLEN];
2980 	struct secasvar *sav = NULL;
2981 #endif
2982 
2983 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2984 		opt = cp[0];
2985 		if (opt == TCPOPT_EOL)
2986 			break;
2987 		if (opt == TCPOPT_NOP)
2988 			optlen = 1;
2989 		else {
2990 			if (cnt < 2)
2991 				break;
2992 			optlen = cp[1];
2993 			if (optlen < 2 || optlen > cnt)
2994 				break;
2995 		}
2996 		switch (opt) {
2997 
2998 		default:
2999 			continue;
3000 
3001 		case TCPOPT_MAXSEG:
3002 			if (optlen != TCPOLEN_MAXSEG)
3003 				continue;
3004 			if (!(th->th_flags & TH_SYN))
3005 				continue;
3006 			if (TCPS_HAVERCVDSYN(tp->t_state))
3007 				continue;
3008 			bcopy(cp + 2, &mss, sizeof(mss));
3009 			oi->maxseg = ntohs(mss);
3010 			break;
3011 
3012 		case TCPOPT_WINDOW:
3013 			if (optlen != TCPOLEN_WINDOW)
3014 				continue;
3015 			if (!(th->th_flags & TH_SYN))
3016 				continue;
3017 			if (TCPS_HAVERCVDSYN(tp->t_state))
3018 				continue;
3019 			tp->t_flags |= TF_RCVD_SCALE;
3020 			tp->requested_s_scale = cp[2];
3021 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
3022 #if 0	/*XXX*/
3023 				char *p;
3024 
3025 				if (ip)
3026 					p = ntohl(ip->ip_src);
3027 #ifdef INET6
3028 				else if (ip6)
3029 					p = ip6_sprintf(&ip6->ip6_src);
3030 #endif
3031 				else
3032 					p = "(unknown)";
3033 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
3034 				    "assuming %d\n",
3035 				    tp->requested_s_scale, p,
3036 				    TCP_MAX_WINSHIFT);
3037 #else
3038 				log(LOG_ERR, "TCP: invalid wscale %d, "
3039 				    "assuming %d\n",
3040 				    tp->requested_s_scale,
3041 				    TCP_MAX_WINSHIFT);
3042 #endif
3043 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
3044 			}
3045 			break;
3046 
3047 		case TCPOPT_TIMESTAMP:
3048 			if (optlen != TCPOLEN_TIMESTAMP)
3049 				continue;
3050 			oi->ts_present = 1;
3051 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
3052 			NTOHL(oi->ts_val);
3053 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
3054 			NTOHL(oi->ts_ecr);
3055 
3056 			if (!(th->th_flags & TH_SYN))
3057 				continue;
3058 			if (TCPS_HAVERCVDSYN(tp->t_state))
3059 				continue;
3060 			/*
3061 			 * A timestamp received in a SYN makes
3062 			 * it ok to send timestamp requests and replies.
3063 			 */
3064 			tp->t_flags |= TF_RCVD_TSTMP;
3065 			tp->ts_recent = oi->ts_val;
3066 			tp->ts_recent_age = tcp_now;
3067                         break;
3068 
3069 		case TCPOPT_SACK_PERMITTED:
3070 			if (optlen != TCPOLEN_SACK_PERMITTED)
3071 				continue;
3072 			if (!(th->th_flags & TH_SYN))
3073 				continue;
3074 			if (TCPS_HAVERCVDSYN(tp->t_state))
3075 				continue;
3076 			if (tcp_do_sack) {
3077 				tp->t_flags |= TF_SACK_PERMIT;
3078 				tp->t_flags |= TF_WILL_SACK;
3079 			}
3080 			break;
3081 
3082 		case TCPOPT_SACK:
3083 			tcp_sack_option(tp, th, cp, optlen);
3084 			break;
3085 #ifdef TCP_SIGNATURE
3086 		case TCPOPT_SIGNATURE:
3087 			if (optlen != TCPOLEN_SIGNATURE)
3088 				continue;
3089 			if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN))
3090 				return (-1);
3091 
3092 			sigp = sigbuf;
3093 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
3094 			tp->t_flags |= TF_SIGNATURE;
3095 			break;
3096 #endif
3097 		}
3098 	}
3099 
3100 #ifdef TCP_SIGNATURE
3101 	if (tp->t_flags & TF_SIGNATURE) {
3102 
3103 		sav = tcp_signature_getsav(m, th);
3104 
3105 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3106 			return (-1);
3107 	}
3108 
3109 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3110 		if (sav == NULL)
3111 			return (-1);
3112 #ifdef FAST_IPSEC
3113 		KEY_FREESAV(&sav);
3114 #else
3115 		key_freesav(sav);
3116 #endif
3117 		return (-1);
3118 	}
3119 
3120 	if (sigp) {
3121 		char sig[TCP_SIGLEN];
3122 
3123 		tcp_fields_to_net(th);
3124 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3125 			tcp_fields_to_host(th);
3126 			if (sav == NULL)
3127 				return (-1);
3128 #ifdef FAST_IPSEC
3129 			KEY_FREESAV(&sav);
3130 #else
3131 			key_freesav(sav);
3132 #endif
3133 			return (-1);
3134 		}
3135 		tcp_fields_to_host(th);
3136 
3137 		if (bcmp(sig, sigp, TCP_SIGLEN)) {
3138 			TCP_STATINC(TCP_STAT_BADSIG);
3139 			if (sav == NULL)
3140 				return (-1);
3141 #ifdef FAST_IPSEC
3142 			KEY_FREESAV(&sav);
3143 #else
3144 			key_freesav(sav);
3145 #endif
3146 			return (-1);
3147 		} else
3148 			TCP_STATINC(TCP_STAT_GOODSIG);
3149 
3150 		key_sa_recordxfer(sav, m);
3151 #ifdef FAST_IPSEC
3152 		KEY_FREESAV(&sav);
3153 #else
3154 		key_freesav(sav);
3155 #endif
3156 	}
3157 #endif
3158 
3159 	return (0);
3160 }
3161 
3162 /*
3163  * Pull out of band byte out of a segment so
3164  * it doesn't appear in the user's data queue.
3165  * It is still reflected in the segment length for
3166  * sequencing purposes.
3167  */
3168 void
3169 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3170     struct mbuf *m, int off)
3171 {
3172 	int cnt = off + th->th_urp - 1;
3173 
3174 	while (cnt >= 0) {
3175 		if (m->m_len > cnt) {
3176 			char *cp = mtod(m, char *) + cnt;
3177 			struct tcpcb *tp = sototcpcb(so);
3178 
3179 			tp->t_iobc = *cp;
3180 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3181 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3182 			m->m_len--;
3183 			return;
3184 		}
3185 		cnt -= m->m_len;
3186 		m = m->m_next;
3187 		if (m == 0)
3188 			break;
3189 	}
3190 	panic("tcp_pulloutofband");
3191 }
3192 
3193 /*
3194  * Collect new round-trip time estimate
3195  * and update averages and current timeout.
3196  */
3197 void
3198 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3199 {
3200 	int32_t delta;
3201 
3202 	TCP_STATINC(TCP_STAT_RTTUPDATED);
3203 	if (tp->t_srtt != 0) {
3204 		/*
3205 		 * srtt is stored as fixed point with 3 bits after the
3206 		 * binary point (i.e., scaled by 8).  The following magic
3207 		 * is equivalent to the smoothing algorithm in rfc793 with
3208 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3209 		 * point).  Adjust rtt to origin 0.
3210 		 */
3211 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3212 		if ((tp->t_srtt += delta) <= 0)
3213 			tp->t_srtt = 1 << 2;
3214 		/*
3215 		 * We accumulate a smoothed rtt variance (actually, a
3216 		 * smoothed mean difference), then set the retransmit
3217 		 * timer to smoothed rtt + 4 times the smoothed variance.
3218 		 * rttvar is stored as fixed point with 2 bits after the
3219 		 * binary point (scaled by 4).  The following is
3220 		 * equivalent to rfc793 smoothing with an alpha of .75
3221 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3222 		 * rfc793's wired-in beta.
3223 		 */
3224 		if (delta < 0)
3225 			delta = -delta;
3226 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3227 		if ((tp->t_rttvar += delta) <= 0)
3228 			tp->t_rttvar = 1 << 2;
3229 	} else {
3230 		/*
3231 		 * No rtt measurement yet - use the unsmoothed rtt.
3232 		 * Set the variance to half the rtt (so our first
3233 		 * retransmit happens at 3*rtt).
3234 		 */
3235 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3236 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3237 	}
3238 	tp->t_rtttime = 0;
3239 	tp->t_rxtshift = 0;
3240 
3241 	/*
3242 	 * the retransmit should happen at rtt + 4 * rttvar.
3243 	 * Because of the way we do the smoothing, srtt and rttvar
3244 	 * will each average +1/2 tick of bias.  When we compute
3245 	 * the retransmit timer, we want 1/2 tick of rounding and
3246 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3247 	 * firing of the timer.  The bias will give us exactly the
3248 	 * 1.5 tick we need.  But, because the bias is
3249 	 * statistical, we have to test that we don't drop below
3250 	 * the minimum feasible timer (which is 2 ticks).
3251 	 */
3252 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3253 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3254 
3255 	/*
3256 	 * We received an ack for a packet that wasn't retransmitted;
3257 	 * it is probably safe to discard any error indications we've
3258 	 * received recently.  This isn't quite right, but close enough
3259 	 * for now (a route might have failed after we sent a segment,
3260 	 * and the return path might not be symmetrical).
3261 	 */
3262 	tp->t_softerror = 0;
3263 }
3264 
3265 
3266 /*
3267  * TCP compressed state engine.  Currently used to hold compressed
3268  * state for SYN_RECEIVED.
3269  */
3270 
3271 u_long	syn_cache_count;
3272 u_int32_t syn_hash1, syn_hash2;
3273 
3274 #define SYN_HASH(sa, sp, dp) \
3275 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3276 				     ((u_int32_t)(sp)))^syn_hash2)))
3277 #ifndef INET6
3278 #define	SYN_HASHALL(hash, src, dst) \
3279 do {									\
3280 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3281 		((const struct sockaddr_in *)(src))->sin_port,		\
3282 		((const struct sockaddr_in *)(dst))->sin_port);		\
3283 } while (/*CONSTCOND*/ 0)
3284 #else
3285 #define SYN_HASH6(sa, sp, dp) \
3286 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3287 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3288 	 & 0x7fffffff)
3289 
3290 #define SYN_HASHALL(hash, src, dst) \
3291 do {									\
3292 	switch ((src)->sa_family) {					\
3293 	case AF_INET:							\
3294 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3295 			((const struct sockaddr_in *)(src))->sin_port,	\
3296 			((const struct sockaddr_in *)(dst))->sin_port);	\
3297 		break;							\
3298 	case AF_INET6:							\
3299 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3300 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3301 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3302 		break;							\
3303 	default:							\
3304 		hash = 0;						\
3305 	}								\
3306 } while (/*CONSTCOND*/0)
3307 #endif /* INET6 */
3308 
3309 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL,
3310     IPL_SOFTNET);
3311 
3312 /*
3313  * We don't estimate RTT with SYNs, so each packet starts with the default
3314  * RTT and each timer step has a fixed timeout value.
3315  */
3316 #define	SYN_CACHE_TIMER_ARM(sc)						\
3317 do {									\
3318 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3319 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3320 	    TCPTV_REXMTMAX);						\
3321 	callout_reset(&(sc)->sc_timer,					\
3322 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3323 } while (/*CONSTCOND*/0)
3324 
3325 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3326 
3327 static inline void
3328 syn_cache_rm(struct syn_cache *sc)
3329 {
3330 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3331 	    sc, sc_bucketq);
3332 	sc->sc_tp = NULL;
3333 	LIST_REMOVE(sc, sc_tpq);
3334 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3335 	callout_stop(&sc->sc_timer);
3336 	syn_cache_count--;
3337 }
3338 
3339 static inline void
3340 syn_cache_put(struct syn_cache *sc)
3341 {
3342 	if (sc->sc_ipopts)
3343 		(void) m_free(sc->sc_ipopts);
3344 	rtcache_free(&sc->sc_route);
3345 	if (callout_invoking(&sc->sc_timer))
3346 		sc->sc_flags |= SCF_DEAD;
3347 	else {
3348 		callout_destroy(&sc->sc_timer);
3349 		pool_put(&syn_cache_pool, sc);
3350 	}
3351 }
3352 
3353 void
3354 syn_cache_init(void)
3355 {
3356 	int i;
3357 
3358 	/* Initialize the hash buckets. */
3359 	for (i = 0; i < tcp_syn_cache_size; i++)
3360 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3361 }
3362 
3363 void
3364 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3365 {
3366 	struct syn_cache_head *scp;
3367 	struct syn_cache *sc2;
3368 	int s;
3369 
3370 	/*
3371 	 * If there are no entries in the hash table, reinitialize
3372 	 * the hash secrets.
3373 	 */
3374 	if (syn_cache_count == 0) {
3375 		syn_hash1 = arc4random();
3376 		syn_hash2 = arc4random();
3377 	}
3378 
3379 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3380 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3381 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3382 
3383 	/*
3384 	 * Make sure that we don't overflow the per-bucket
3385 	 * limit or the total cache size limit.
3386 	 */
3387 	s = splsoftnet();
3388 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3389 		TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW);
3390 		/*
3391 		 * The bucket is full.  Toss the oldest element in the
3392 		 * bucket.  This will be the first entry in the bucket.
3393 		 */
3394 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3395 #ifdef DIAGNOSTIC
3396 		/*
3397 		 * This should never happen; we should always find an
3398 		 * entry in our bucket.
3399 		 */
3400 		if (sc2 == NULL)
3401 			panic("syn_cache_insert: bucketoverflow: impossible");
3402 #endif
3403 		syn_cache_rm(sc2);
3404 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3405 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3406 		struct syn_cache_head *scp2, *sce;
3407 
3408 		TCP_STATINC(TCP_STAT_SC_OVERFLOWED);
3409 		/*
3410 		 * The cache is full.  Toss the oldest entry in the
3411 		 * first non-empty bucket we can find.
3412 		 *
3413 		 * XXX We would really like to toss the oldest
3414 		 * entry in the cache, but we hope that this
3415 		 * condition doesn't happen very often.
3416 		 */
3417 		scp2 = scp;
3418 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3419 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3420 			for (++scp2; scp2 != scp; scp2++) {
3421 				if (scp2 >= sce)
3422 					scp2 = &tcp_syn_cache[0];
3423 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3424 					break;
3425 			}
3426 #ifdef DIAGNOSTIC
3427 			/*
3428 			 * This should never happen; we should always find a
3429 			 * non-empty bucket.
3430 			 */
3431 			if (scp2 == scp)
3432 				panic("syn_cache_insert: cacheoverflow: "
3433 				    "impossible");
3434 #endif
3435 		}
3436 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3437 		syn_cache_rm(sc2);
3438 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3439 	}
3440 
3441 	/*
3442 	 * Initialize the entry's timer.
3443 	 */
3444 	sc->sc_rxttot = 0;
3445 	sc->sc_rxtshift = 0;
3446 	SYN_CACHE_TIMER_ARM(sc);
3447 
3448 	/* Link it from tcpcb entry */
3449 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3450 
3451 	/* Put it into the bucket. */
3452 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3453 	scp->sch_length++;
3454 	syn_cache_count++;
3455 
3456 	TCP_STATINC(TCP_STAT_SC_ADDED);
3457 	splx(s);
3458 }
3459 
3460 /*
3461  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3462  * If we have retransmitted an entry the maximum number of times, expire
3463  * that entry.
3464  */
3465 void
3466 syn_cache_timer(void *arg)
3467 {
3468 	struct syn_cache *sc = arg;
3469 	int s;
3470 
3471 	s = splsoftnet();
3472 	callout_ack(&sc->sc_timer);
3473 
3474 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3475 		TCP_STATINC(TCP_STAT_SC_DELAYED_FREE);
3476 		callout_destroy(&sc->sc_timer);
3477 		pool_put(&syn_cache_pool, sc);
3478 		splx(s);
3479 		return;
3480 	}
3481 
3482 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3483 		/* Drop it -- too many retransmissions. */
3484 		goto dropit;
3485 	}
3486 
3487 	/*
3488 	 * Compute the total amount of time this entry has
3489 	 * been on a queue.  If this entry has been on longer
3490 	 * than the keep alive timer would allow, expire it.
3491 	 */
3492 	sc->sc_rxttot += sc->sc_rxtcur;
3493 	if (sc->sc_rxttot >= tcp_keepinit)
3494 		goto dropit;
3495 
3496 	TCP_STATINC(TCP_STAT_SC_RETRANSMITTED);
3497 	(void) syn_cache_respond(sc, NULL);
3498 
3499 	/* Advance the timer back-off. */
3500 	sc->sc_rxtshift++;
3501 	SYN_CACHE_TIMER_ARM(sc);
3502 
3503 	splx(s);
3504 	return;
3505 
3506  dropit:
3507 	TCP_STATINC(TCP_STAT_SC_TIMED_OUT);
3508 	syn_cache_rm(sc);
3509 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3510 	splx(s);
3511 }
3512 
3513 /*
3514  * Remove syn cache created by the specified tcb entry,
3515  * because this does not make sense to keep them
3516  * (if there's no tcb entry, syn cache entry will never be used)
3517  */
3518 void
3519 syn_cache_cleanup(struct tcpcb *tp)
3520 {
3521 	struct syn_cache *sc, *nsc;
3522 	int s;
3523 
3524 	s = splsoftnet();
3525 
3526 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3527 		nsc = LIST_NEXT(sc, sc_tpq);
3528 
3529 #ifdef DIAGNOSTIC
3530 		if (sc->sc_tp != tp)
3531 			panic("invalid sc_tp in syn_cache_cleanup");
3532 #endif
3533 		syn_cache_rm(sc);
3534 		syn_cache_put(sc);	/* calls pool_put but see spl above */
3535 	}
3536 	/* just for safety */
3537 	LIST_INIT(&tp->t_sc);
3538 
3539 	splx(s);
3540 }
3541 
3542 /*
3543  * Find an entry in the syn cache.
3544  */
3545 struct syn_cache *
3546 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3547     struct syn_cache_head **headp)
3548 {
3549 	struct syn_cache *sc;
3550 	struct syn_cache_head *scp;
3551 	u_int32_t hash;
3552 	int s;
3553 
3554 	SYN_HASHALL(hash, src, dst);
3555 
3556 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3557 	*headp = scp;
3558 	s = splsoftnet();
3559 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3560 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3561 		if (sc->sc_hash != hash)
3562 			continue;
3563 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3564 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3565 			splx(s);
3566 			return (sc);
3567 		}
3568 	}
3569 	splx(s);
3570 	return (NULL);
3571 }
3572 
3573 /*
3574  * This function gets called when we receive an ACK for a
3575  * socket in the LISTEN state.  We look up the connection
3576  * in the syn cache, and if its there, we pull it out of
3577  * the cache and turn it into a full-blown connection in
3578  * the SYN-RECEIVED state.
3579  *
3580  * The return values may not be immediately obvious, and their effects
3581  * can be subtle, so here they are:
3582  *
3583  *	NULL	SYN was not found in cache; caller should drop the
3584  *		packet and send an RST.
3585  *
3586  *	-1	We were unable to create the new connection, and are
3587  *		aborting it.  An ACK,RST is being sent to the peer
3588  *		(unless we got screwey sequence numbners; see below),
3589  *		because the 3-way handshake has been completed.  Caller
3590  *		should not free the mbuf, since we may be using it.  If
3591  *		we are not, we will free it.
3592  *
3593  *	Otherwise, the return value is a pointer to the new socket
3594  *	associated with the connection.
3595  */
3596 struct socket *
3597 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3598     struct tcphdr *th, unsigned int hlen, unsigned int tlen,
3599     struct socket *so, struct mbuf *m)
3600 {
3601 	struct syn_cache *sc;
3602 	struct syn_cache_head *scp;
3603 	struct inpcb *inp = NULL;
3604 #ifdef INET6
3605 	struct in6pcb *in6p = NULL;
3606 #endif
3607 	struct tcpcb *tp = 0;
3608 	struct mbuf *am;
3609 	int s;
3610 	struct socket *oso;
3611 
3612 	s = splsoftnet();
3613 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3614 		splx(s);
3615 		return (NULL);
3616 	}
3617 
3618 	/*
3619 	 * Verify the sequence and ack numbers.  Try getting the correct
3620 	 * response again.
3621 	 */
3622 	if ((th->th_ack != sc->sc_iss + 1) ||
3623 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3624 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3625 		(void) syn_cache_respond(sc, m);
3626 		splx(s);
3627 		return ((struct socket *)(-1));
3628 	}
3629 
3630 	/* Remove this cache entry */
3631 	syn_cache_rm(sc);
3632 	splx(s);
3633 
3634 	/*
3635 	 * Ok, create the full blown connection, and set things up
3636 	 * as they would have been set up if we had created the
3637 	 * connection when the SYN arrived.  If we can't create
3638 	 * the connection, abort it.
3639 	 */
3640 	/*
3641 	 * inp still has the OLD in_pcb stuff, set the
3642 	 * v6-related flags on the new guy, too.   This is
3643 	 * done particularly for the case where an AF_INET6
3644 	 * socket is bound only to a port, and a v4 connection
3645 	 * comes in on that port.
3646 	 * we also copy the flowinfo from the original pcb
3647 	 * to the new one.
3648 	 */
3649 	oso = so;
3650 	so = sonewconn(so, SS_ISCONNECTED);
3651 	if (so == NULL)
3652 		goto resetandabort;
3653 
3654 	switch (so->so_proto->pr_domain->dom_family) {
3655 #ifdef INET
3656 	case AF_INET:
3657 		inp = sotoinpcb(so);
3658 		break;
3659 #endif
3660 #ifdef INET6
3661 	case AF_INET6:
3662 		in6p = sotoin6pcb(so);
3663 		break;
3664 #endif
3665 	}
3666 	switch (src->sa_family) {
3667 #ifdef INET
3668 	case AF_INET:
3669 		if (inp) {
3670 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3671 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3672 			inp->inp_options = ip_srcroute();
3673 			in_pcbstate(inp, INP_BOUND);
3674 			if (inp->inp_options == NULL) {
3675 				inp->inp_options = sc->sc_ipopts;
3676 				sc->sc_ipopts = NULL;
3677 			}
3678 		}
3679 #ifdef INET6
3680 		else if (in6p) {
3681 			/* IPv4 packet to AF_INET6 socket */
3682 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3683 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3684 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3685 				&in6p->in6p_laddr.s6_addr32[3],
3686 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3687 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3688 			in6totcpcb(in6p)->t_family = AF_INET;
3689 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3690 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3691 			else
3692 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3693 			in6_pcbstate(in6p, IN6P_BOUND);
3694 		}
3695 #endif
3696 		break;
3697 #endif
3698 #ifdef INET6
3699 	case AF_INET6:
3700 		if (in6p) {
3701 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3702 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3703 			in6_pcbstate(in6p, IN6P_BOUND);
3704 		}
3705 		break;
3706 #endif
3707 	}
3708 #ifdef INET6
3709 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3710 		struct in6pcb *oin6p = sotoin6pcb(oso);
3711 		/* inherit socket options from the listening socket */
3712 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3713 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3714 			m_freem(in6p->in6p_options);
3715 			in6p->in6p_options = 0;
3716 		}
3717 		ip6_savecontrol(in6p, &in6p->in6p_options,
3718 			mtod(m, struct ip6_hdr *), m);
3719 	}
3720 #endif
3721 
3722 #if defined(IPSEC) || defined(FAST_IPSEC)
3723 	/*
3724 	 * we make a copy of policy, instead of sharing the policy,
3725 	 * for better behavior in terms of SA lookup and dead SA removal.
3726 	 */
3727 	if (inp) {
3728 		/* copy old policy into new socket's */
3729 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3730 			printf("tcp_input: could not copy policy\n");
3731 	}
3732 #ifdef INET6
3733 	else if (in6p) {
3734 		/* copy old policy into new socket's */
3735 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3736 		    in6p->in6p_sp))
3737 			printf("tcp_input: could not copy policy\n");
3738 	}
3739 #endif
3740 #endif
3741 
3742 	/*
3743 	 * Give the new socket our cached route reference.
3744 	 */
3745 	if (inp) {
3746 		rtcache_copy(&inp->inp_route, &sc->sc_route);
3747 		rtcache_free(&sc->sc_route);
3748 	}
3749 #ifdef INET6
3750 	else {
3751 		rtcache_copy(&in6p->in6p_route, &sc->sc_route);
3752 		rtcache_free(&sc->sc_route);
3753 	}
3754 #endif
3755 
3756 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3757 	if (am == NULL)
3758 		goto resetandabort;
3759 	MCLAIM(am, &tcp_mowner);
3760 	am->m_len = src->sa_len;
3761 	bcopy(src, mtod(am, void *), src->sa_len);
3762 	if (inp) {
3763 		if (in_pcbconnect(inp, am, &lwp0)) {
3764 			(void) m_free(am);
3765 			goto resetandabort;
3766 		}
3767 	}
3768 #ifdef INET6
3769 	else if (in6p) {
3770 		if (src->sa_family == AF_INET) {
3771 			/* IPv4 packet to AF_INET6 socket */
3772 			struct sockaddr_in6 *sin6;
3773 			sin6 = mtod(am, struct sockaddr_in6 *);
3774 			am->m_len = sizeof(*sin6);
3775 			bzero(sin6, sizeof(*sin6));
3776 			sin6->sin6_family = AF_INET6;
3777 			sin6->sin6_len = sizeof(*sin6);
3778 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3779 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3780 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3781 				&sin6->sin6_addr.s6_addr32[3],
3782 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3783 		}
3784 		if (in6_pcbconnect(in6p, am, NULL)) {
3785 			(void) m_free(am);
3786 			goto resetandabort;
3787 		}
3788 	}
3789 #endif
3790 	else {
3791 		(void) m_free(am);
3792 		goto resetandabort;
3793 	}
3794 	(void) m_free(am);
3795 
3796 	if (inp)
3797 		tp = intotcpcb(inp);
3798 #ifdef INET6
3799 	else if (in6p)
3800 		tp = in6totcpcb(in6p);
3801 #endif
3802 	else
3803 		tp = NULL;
3804 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3805 	if (sc->sc_request_r_scale != 15) {
3806 		tp->requested_s_scale = sc->sc_requested_s_scale;
3807 		tp->request_r_scale = sc->sc_request_r_scale;
3808 		tp->snd_scale = sc->sc_requested_s_scale;
3809 		tp->rcv_scale = sc->sc_request_r_scale;
3810 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3811 	}
3812 	if (sc->sc_flags & SCF_TIMESTAMP)
3813 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3814 	tp->ts_timebase = sc->sc_timebase;
3815 
3816 	tp->t_template = tcp_template(tp);
3817 	if (tp->t_template == 0) {
3818 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3819 		so = NULL;
3820 		m_freem(m);
3821 		goto abort;
3822 	}
3823 
3824 	tp->iss = sc->sc_iss;
3825 	tp->irs = sc->sc_irs;
3826 	tcp_sendseqinit(tp);
3827 	tcp_rcvseqinit(tp);
3828 	tp->t_state = TCPS_SYN_RECEIVED;
3829 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit);
3830 	TCP_STATINC(TCP_STAT_ACCEPTS);
3831 
3832 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3833 		tp->t_flags |= TF_WILL_SACK;
3834 
3835 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3836 		tp->t_flags |= TF_ECN_PERMIT;
3837 
3838 #ifdef TCP_SIGNATURE
3839 	if (sc->sc_flags & SCF_SIGNATURE)
3840 		tp->t_flags |= TF_SIGNATURE;
3841 #endif
3842 
3843 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3844 	tp->t_ourmss = sc->sc_ourmaxseg;
3845 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3846 
3847 	/*
3848 	 * Initialize the initial congestion window.  If we
3849 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3850 	 * to 1 segment (i.e. the Loss Window).
3851 	 */
3852 	if (sc->sc_rxtshift)
3853 		tp->snd_cwnd = tp->t_peermss;
3854 	else {
3855 		int ss = tcp_init_win;
3856 #ifdef INET
3857 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3858 			ss = tcp_init_win_local;
3859 #endif
3860 #ifdef INET6
3861 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3862 			ss = tcp_init_win_local;
3863 #endif
3864 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3865 	}
3866 
3867 	tcp_rmx_rtt(tp);
3868 	tp->snd_wl1 = sc->sc_irs;
3869 	tp->rcv_up = sc->sc_irs + 1;
3870 
3871 	/*
3872 	 * This is what whould have happened in tcp_output() when
3873 	 * the SYN,ACK was sent.
3874 	 */
3875 	tp->snd_up = tp->snd_una;
3876 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3877 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3878 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3879 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3880 	tp->last_ack_sent = tp->rcv_nxt;
3881 	tp->t_partialacks = -1;
3882 	tp->t_dupacks = 0;
3883 
3884 	TCP_STATINC(TCP_STAT_SC_COMPLETED);
3885 	s = splsoftnet();
3886 	syn_cache_put(sc);
3887 	splx(s);
3888 	return (so);
3889 
3890 resetandabort:
3891 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3892 abort:
3893 	if (so != NULL)
3894 		(void) soabort(so);
3895 	s = splsoftnet();
3896 	syn_cache_put(sc);
3897 	splx(s);
3898 	TCP_STATINC(TCP_STAT_SC_ABORTED);
3899 	return ((struct socket *)(-1));
3900 }
3901 
3902 /*
3903  * This function is called when we get a RST for a
3904  * non-existent connection, so that we can see if the
3905  * connection is in the syn cache.  If it is, zap it.
3906  */
3907 
3908 void
3909 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3910 {
3911 	struct syn_cache *sc;
3912 	struct syn_cache_head *scp;
3913 	int s = splsoftnet();
3914 
3915 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3916 		splx(s);
3917 		return;
3918 	}
3919 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3920 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3921 		splx(s);
3922 		return;
3923 	}
3924 	syn_cache_rm(sc);
3925 	TCP_STATINC(TCP_STAT_SC_RESET);
3926 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3927 	splx(s);
3928 }
3929 
3930 void
3931 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3932     struct tcphdr *th)
3933 {
3934 	struct syn_cache *sc;
3935 	struct syn_cache_head *scp;
3936 	int s;
3937 
3938 	s = splsoftnet();
3939 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3940 		splx(s);
3941 		return;
3942 	}
3943 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3944 	if (ntohl (th->th_seq) != sc->sc_iss) {
3945 		splx(s);
3946 		return;
3947 	}
3948 
3949 	/*
3950 	 * If we've retransmitted 3 times and this is our second error,
3951 	 * we remove the entry.  Otherwise, we allow it to continue on.
3952 	 * This prevents us from incorrectly nuking an entry during a
3953 	 * spurious network outage.
3954 	 *
3955 	 * See tcp_notify().
3956 	 */
3957 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3958 		sc->sc_flags |= SCF_UNREACH;
3959 		splx(s);
3960 		return;
3961 	}
3962 
3963 	syn_cache_rm(sc);
3964 	TCP_STATINC(TCP_STAT_SC_UNREACH);
3965 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3966 	splx(s);
3967 }
3968 
3969 /*
3970  * Given a LISTEN socket and an inbound SYN request, add
3971  * this to the syn cache, and send back a segment:
3972  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3973  * to the source.
3974  *
3975  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3976  * Doing so would require that we hold onto the data and deliver it
3977  * to the application.  However, if we are the target of a SYN-flood
3978  * DoS attack, an attacker could send data which would eventually
3979  * consume all available buffer space if it were ACKed.  By not ACKing
3980  * the data, we avoid this DoS scenario.
3981  */
3982 
3983 int
3984 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3985     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
3986     int optlen, struct tcp_opt_info *oi)
3987 {
3988 	struct tcpcb tb, *tp;
3989 	long win;
3990 	struct syn_cache *sc;
3991 	struct syn_cache_head *scp;
3992 	struct mbuf *ipopts;
3993 	struct tcp_opt_info opti;
3994 	int s;
3995 
3996 	tp = sototcpcb(so);
3997 
3998 	bzero(&opti, sizeof(opti));
3999 
4000 	/*
4001 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
4002 	 *
4003 	 * Note this check is performed in tcp_input() very early on.
4004 	 */
4005 
4006 	/*
4007 	 * Initialize some local state.
4008 	 */
4009 	win = sbspace(&so->so_rcv);
4010 	if (win > TCP_MAXWIN)
4011 		win = TCP_MAXWIN;
4012 
4013 	switch (src->sa_family) {
4014 #ifdef INET
4015 	case AF_INET:
4016 		/*
4017 		 * Remember the IP options, if any.
4018 		 */
4019 		ipopts = ip_srcroute();
4020 		break;
4021 #endif
4022 	default:
4023 		ipopts = NULL;
4024 	}
4025 
4026 #ifdef TCP_SIGNATURE
4027 	if (optp || (tp->t_flags & TF_SIGNATURE))
4028 #else
4029 	if (optp)
4030 #endif
4031 	{
4032 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4033 #ifdef TCP_SIGNATURE
4034 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
4035 #endif
4036 		tb.t_state = TCPS_LISTEN;
4037 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
4038 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
4039 			return (0);
4040 	} else
4041 		tb.t_flags = 0;
4042 
4043 	/*
4044 	 * See if we already have an entry for this connection.
4045 	 * If we do, resend the SYN,ACK.  We do not count this
4046 	 * as a retransmission (XXX though maybe we should).
4047 	 */
4048 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4049 		TCP_STATINC(TCP_STAT_SC_DUPESYN);
4050 		if (ipopts) {
4051 			/*
4052 			 * If we were remembering a previous source route,
4053 			 * forget it and use the new one we've been given.
4054 			 */
4055 			if (sc->sc_ipopts)
4056 				(void) m_free(sc->sc_ipopts);
4057 			sc->sc_ipopts = ipopts;
4058 		}
4059 		sc->sc_timestamp = tb.ts_recent;
4060 		if (syn_cache_respond(sc, m) == 0) {
4061 			uint64_t *tcps = TCP_STAT_GETREF();
4062 			tcps[TCP_STAT_SNDACKS]++;
4063 			tcps[TCP_STAT_SNDTOTAL]++;
4064 			TCP_STAT_PUTREF();
4065 		}
4066 		return (1);
4067 	}
4068 
4069 	s = splsoftnet();
4070 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4071 	splx(s);
4072 	if (sc == NULL) {
4073 		if (ipopts)
4074 			(void) m_free(ipopts);
4075 		return (0);
4076 	}
4077 
4078 	/*
4079 	 * Fill in the cache, and put the necessary IP and TCP
4080 	 * options into the reply.
4081 	 */
4082 	bzero(sc, sizeof(struct syn_cache));
4083 	callout_init(&sc->sc_timer, 0);
4084 	bcopy(src, &sc->sc_src, src->sa_len);
4085 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4086 	sc->sc_flags = 0;
4087 	sc->sc_ipopts = ipopts;
4088 	sc->sc_irs = th->th_seq;
4089 	switch (src->sa_family) {
4090 #ifdef INET
4091 	case AF_INET:
4092 	    {
4093 		struct sockaddr_in *srcin = (void *) src;
4094 		struct sockaddr_in *dstin = (void *) dst;
4095 
4096 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
4097 		    &srcin->sin_addr, dstin->sin_port,
4098 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
4099 		break;
4100 	    }
4101 #endif /* INET */
4102 #ifdef INET6
4103 	case AF_INET6:
4104 	    {
4105 		struct sockaddr_in6 *srcin6 = (void *) src;
4106 		struct sockaddr_in6 *dstin6 = (void *) dst;
4107 
4108 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
4109 		    &srcin6->sin6_addr, dstin6->sin6_port,
4110 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
4111 		break;
4112 	    }
4113 #endif /* INET6 */
4114 	}
4115 	sc->sc_peermaxseg = oi->maxseg;
4116 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4117 						m->m_pkthdr.rcvif : NULL,
4118 						sc->sc_src.sa.sa_family);
4119 	sc->sc_win = win;
4120 	sc->sc_timebase = tcp_now - 1;	/* see tcp_newtcpcb() */
4121 	sc->sc_timestamp = tb.ts_recent;
4122 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4123 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4124 		sc->sc_flags |= SCF_TIMESTAMP;
4125 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4126 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4127 		sc->sc_requested_s_scale = tb.requested_s_scale;
4128 		sc->sc_request_r_scale = 0;
4129 		/*
4130 		 * Pick the smallest possible scaling factor that
4131 		 * will still allow us to scale up to sb_max.
4132 		 *
4133 		 * We do this because there are broken firewalls that
4134 		 * will corrupt the window scale option, leading to
4135 		 * the other endpoint believing that our advertised
4136 		 * window is unscaled.  At scale factors larger than
4137 		 * 5 the unscaled window will drop below 1500 bytes,
4138 		 * leading to serious problems when traversing these
4139 		 * broken firewalls.
4140 		 *
4141 		 * With the default sbmax of 256K, a scale factor
4142 		 * of 3 will be chosen by this algorithm.  Those who
4143 		 * choose a larger sbmax should watch out
4144 		 * for the compatiblity problems mentioned above.
4145 		 *
4146 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4147 		 * or <SYN,ACK>) segment itself is never scaled.
4148 		 */
4149 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4150 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4151 			sc->sc_request_r_scale++;
4152 	} else {
4153 		sc->sc_requested_s_scale = 15;
4154 		sc->sc_request_r_scale = 15;
4155 	}
4156 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4157 		sc->sc_flags |= SCF_SACK_PERMIT;
4158 
4159 	/*
4160 	 * ECN setup packet recieved.
4161 	 */
4162 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4163 		sc->sc_flags |= SCF_ECN_PERMIT;
4164 
4165 #ifdef TCP_SIGNATURE
4166 	if (tb.t_flags & TF_SIGNATURE)
4167 		sc->sc_flags |= SCF_SIGNATURE;
4168 #endif
4169 	sc->sc_tp = tp;
4170 	if (syn_cache_respond(sc, m) == 0) {
4171 		uint64_t *tcps = TCP_STAT_GETREF();
4172 		tcps[TCP_STAT_SNDACKS]++;
4173 		tcps[TCP_STAT_SNDTOTAL]++;
4174 		TCP_STAT_PUTREF();
4175 		syn_cache_insert(sc, tp);
4176 	} else {
4177 		s = splsoftnet();
4178 		syn_cache_put(sc);
4179 		splx(s);
4180 		TCP_STATINC(TCP_STAT_SC_DROPPED);
4181 	}
4182 	return (1);
4183 }
4184 
4185 int
4186 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4187 {
4188 #ifdef INET6
4189 	struct rtentry *rt;
4190 #endif
4191 	struct route *ro;
4192 	u_int8_t *optp;
4193 	int optlen, error;
4194 	u_int16_t tlen;
4195 	struct ip *ip = NULL;
4196 #ifdef INET6
4197 	struct ip6_hdr *ip6 = NULL;
4198 #endif
4199 	struct tcpcb *tp = NULL;
4200 	struct tcphdr *th;
4201 	u_int hlen;
4202 	struct socket *so;
4203 
4204 	ro = &sc->sc_route;
4205 	switch (sc->sc_src.sa.sa_family) {
4206 	case AF_INET:
4207 		hlen = sizeof(struct ip);
4208 		break;
4209 #ifdef INET6
4210 	case AF_INET6:
4211 		hlen = sizeof(struct ip6_hdr);
4212 		break;
4213 #endif
4214 	default:
4215 		if (m)
4216 			m_freem(m);
4217 		return (EAFNOSUPPORT);
4218 	}
4219 
4220 	/* Compute the size of the TCP options. */
4221 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4222 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4223 #ifdef TCP_SIGNATURE
4224 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4225 #endif
4226 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4227 
4228 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4229 
4230 	/*
4231 	 * Create the IP+TCP header from scratch.
4232 	 */
4233 	if (m)
4234 		m_freem(m);
4235 #ifdef DIAGNOSTIC
4236 	if (max_linkhdr + tlen > MCLBYTES)
4237 		return (ENOBUFS);
4238 #endif
4239 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4240 	if (m && tlen > MHLEN) {
4241 		MCLGET(m, M_DONTWAIT);
4242 		if ((m->m_flags & M_EXT) == 0) {
4243 			m_freem(m);
4244 			m = NULL;
4245 		}
4246 	}
4247 	if (m == NULL)
4248 		return (ENOBUFS);
4249 	MCLAIM(m, &tcp_tx_mowner);
4250 
4251 	/* Fixup the mbuf. */
4252 	m->m_data += max_linkhdr;
4253 	m->m_len = m->m_pkthdr.len = tlen;
4254 	if (sc->sc_tp) {
4255 		tp = sc->sc_tp;
4256 		if (tp->t_inpcb)
4257 			so = tp->t_inpcb->inp_socket;
4258 #ifdef INET6
4259 		else if (tp->t_in6pcb)
4260 			so = tp->t_in6pcb->in6p_socket;
4261 #endif
4262 		else
4263 			so = NULL;
4264 	} else
4265 		so = NULL;
4266 	m->m_pkthdr.rcvif = NULL;
4267 	memset(mtod(m, u_char *), 0, tlen);
4268 
4269 	switch (sc->sc_src.sa.sa_family) {
4270 	case AF_INET:
4271 		ip = mtod(m, struct ip *);
4272 		ip->ip_v = 4;
4273 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4274 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4275 		ip->ip_p = IPPROTO_TCP;
4276 		th = (struct tcphdr *)(ip + 1);
4277 		th->th_dport = sc->sc_src.sin.sin_port;
4278 		th->th_sport = sc->sc_dst.sin.sin_port;
4279 		break;
4280 #ifdef INET6
4281 	case AF_INET6:
4282 		ip6 = mtod(m, struct ip6_hdr *);
4283 		ip6->ip6_vfc = IPV6_VERSION;
4284 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4285 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4286 		ip6->ip6_nxt = IPPROTO_TCP;
4287 		/* ip6_plen will be updated in ip6_output() */
4288 		th = (struct tcphdr *)(ip6 + 1);
4289 		th->th_dport = sc->sc_src.sin6.sin6_port;
4290 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4291 		break;
4292 #endif
4293 	default:
4294 		th = NULL;
4295 	}
4296 
4297 	th->th_seq = htonl(sc->sc_iss);
4298 	th->th_ack = htonl(sc->sc_irs + 1);
4299 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4300 	th->th_flags = TH_SYN|TH_ACK;
4301 	th->th_win = htons(sc->sc_win);
4302 	/* th_sum already 0 */
4303 	/* th_urp already 0 */
4304 
4305 	/* Tack on the TCP options. */
4306 	optp = (u_int8_t *)(th + 1);
4307 	*optp++ = TCPOPT_MAXSEG;
4308 	*optp++ = 4;
4309 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4310 	*optp++ = sc->sc_ourmaxseg & 0xff;
4311 
4312 	if (sc->sc_request_r_scale != 15) {
4313 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4314 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4315 		    sc->sc_request_r_scale);
4316 		optp += 4;
4317 	}
4318 
4319 	if (sc->sc_flags & SCF_TIMESTAMP) {
4320 		u_int32_t *lp = (u_int32_t *)(optp);
4321 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4322 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4323 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4324 		*lp   = htonl(sc->sc_timestamp);
4325 		optp += TCPOLEN_TSTAMP_APPA;
4326 	}
4327 
4328 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4329 		u_int8_t *p = optp;
4330 
4331 		/* Let the peer know that we will SACK. */
4332 		p[0] = TCPOPT_SACK_PERMITTED;
4333 		p[1] = 2;
4334 		p[2] = TCPOPT_NOP;
4335 		p[3] = TCPOPT_NOP;
4336 		optp += 4;
4337 	}
4338 
4339 	/*
4340 	 * Send ECN SYN-ACK setup packet.
4341 	 * Routes can be asymetric, so, even if we receive a packet
4342 	 * with ECE and CWR set, we must not assume no one will block
4343 	 * the ECE packet we are about to send.
4344 	 */
4345 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4346 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4347 		th->th_flags |= TH_ECE;
4348 		TCP_STATINC(TCP_STAT_ECN_SHS);
4349 
4350 		/*
4351 		 * draft-ietf-tcpm-ecnsyn-00.txt
4352 		 *
4353 		 * "[...] a TCP node MAY respond to an ECN-setup
4354 		 * SYN packet by setting ECT in the responding
4355 		 * ECN-setup SYN/ACK packet, indicating to routers
4356 		 * that the SYN/ACK packet is ECN-Capable.
4357 		 * This allows a congested router along the path
4358 		 * to mark the packet instead of dropping the
4359 		 * packet as an indication of congestion."
4360 		 *
4361 		 * "[...] There can be a great benefit in setting
4362 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4363 		 * Congestion is  most likely to occur in
4364 		 * the server-to-client direction.  As a result,
4365 		 * setting an ECN-capable codepoint in SYN/ACK
4366 		 * packets can reduce the occurence of three-second
4367 		 * retransmit timeouts resulting from the drop
4368 		 * of SYN/ACK packets."
4369 		 *
4370 		 * Page 4 and 6, January 2006.
4371 		 */
4372 
4373 		switch (sc->sc_src.sa.sa_family) {
4374 #ifdef INET
4375 		case AF_INET:
4376 			ip->ip_tos |= IPTOS_ECN_ECT0;
4377 			break;
4378 #endif
4379 #ifdef INET6
4380 		case AF_INET6:
4381 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4382 			break;
4383 #endif
4384 		}
4385 		TCP_STATINC(TCP_STAT_ECN_ECT);
4386 	}
4387 
4388 #ifdef TCP_SIGNATURE
4389 	if (sc->sc_flags & SCF_SIGNATURE) {
4390 		struct secasvar *sav;
4391 		u_int8_t *sigp;
4392 
4393 		sav = tcp_signature_getsav(m, th);
4394 
4395 		if (sav == NULL) {
4396 			if (m)
4397 				m_freem(m);
4398 			return (EPERM);
4399 		}
4400 
4401 		*optp++ = TCPOPT_SIGNATURE;
4402 		*optp++ = TCPOLEN_SIGNATURE;
4403 		sigp = optp;
4404 		bzero(optp, TCP_SIGLEN);
4405 		optp += TCP_SIGLEN;
4406 		*optp++ = TCPOPT_NOP;
4407 		*optp++ = TCPOPT_EOL;
4408 
4409 		(void)tcp_signature(m, th, hlen, sav, sigp);
4410 
4411 		key_sa_recordxfer(sav, m);
4412 #ifdef FAST_IPSEC
4413 		KEY_FREESAV(&sav);
4414 #else
4415 		key_freesav(sav);
4416 #endif
4417 	}
4418 #endif
4419 
4420 	/* Compute the packet's checksum. */
4421 	switch (sc->sc_src.sa.sa_family) {
4422 	case AF_INET:
4423 		ip->ip_len = htons(tlen - hlen);
4424 		th->th_sum = 0;
4425 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4426 		break;
4427 #ifdef INET6
4428 	case AF_INET6:
4429 		ip6->ip6_plen = htons(tlen - hlen);
4430 		th->th_sum = 0;
4431 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4432 		break;
4433 #endif
4434 	}
4435 
4436 	/*
4437 	 * Fill in some straggling IP bits.  Note the stack expects
4438 	 * ip_len to be in host order, for convenience.
4439 	 */
4440 	switch (sc->sc_src.sa.sa_family) {
4441 #ifdef INET
4442 	case AF_INET:
4443 		ip->ip_len = htons(tlen);
4444 		ip->ip_ttl = ip_defttl;
4445 		/* XXX tos? */
4446 		break;
4447 #endif
4448 #ifdef INET6
4449 	case AF_INET6:
4450 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4451 		ip6->ip6_vfc |= IPV6_VERSION;
4452 		ip6->ip6_plen = htons(tlen - hlen);
4453 		/* ip6_hlim will be initialized afterwards */
4454 		/* XXX flowlabel? */
4455 		break;
4456 #endif
4457 	}
4458 
4459 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4460 	tp = sc->sc_tp;
4461 
4462 	switch (sc->sc_src.sa.sa_family) {
4463 #ifdef INET
4464 	case AF_INET:
4465 		error = ip_output(m, sc->sc_ipopts, ro,
4466 		    (ip_mtudisc ? IP_MTUDISC : 0),
4467 		    (struct ip_moptions *)NULL, so);
4468 		break;
4469 #endif
4470 #ifdef INET6
4471 	case AF_INET6:
4472 		ip6->ip6_hlim = in6_selecthlim(NULL,
4473 				(rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
4474 				                                    : NULL);
4475 
4476 		error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL);
4477 		break;
4478 #endif
4479 	default:
4480 		error = EAFNOSUPPORT;
4481 		break;
4482 	}
4483 	return (error);
4484 }
4485