xref: /netbsd-src/sys/netinet/tcp_input.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: tcp_input.c,v 1.272 2007/11/09 23:55:58 dyoung 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 1.272 2007/11/09 23:55:58 dyoung Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 #ifdef TCP_SIGNATURE
175 #include <sys/md5.h>
176 #endif
177 
178 #include <net/if.h>
179 #include <net/route.h>
180 #include <net/if_types.h>
181 
182 #include <netinet/in.h>
183 #include <netinet/in_systm.h>
184 #include <netinet/ip.h>
185 #include <netinet/in_pcb.h>
186 #include <netinet/in_var.h>
187 #include <netinet/ip_var.h>
188 #include <netinet/in_offload.h>
189 
190 #ifdef INET6
191 #ifndef INET
192 #include <netinet/in.h>
193 #endif
194 #include <netinet/ip6.h>
195 #include <netinet6/ip6_var.h>
196 #include <netinet6/in6_pcb.h>
197 #include <netinet6/ip6_var.h>
198 #include <netinet6/in6_var.h>
199 #include <netinet/icmp6.h>
200 #include <netinet6/nd6.h>
201 #ifdef TCP_SIGNATURE
202 #include <netinet6/scope6_var.h>
203 #endif
204 #endif
205 
206 #ifndef INET6
207 /* always need ip6.h for IP6_EXTHDR_GET */
208 #include <netinet/ip6.h>
209 #endif
210 
211 #include <netinet/tcp.h>
212 #include <netinet/tcp_fsm.h>
213 #include <netinet/tcp_seq.h>
214 #include <netinet/tcp_timer.h>
215 #include <netinet/tcp_var.h>
216 #include <netinet/tcpip.h>
217 #include <netinet/tcp_congctl.h>
218 #include <netinet/tcp_debug.h>
219 
220 #include <machine/stdarg.h>
221 
222 #ifdef IPSEC
223 #include <netinet6/ipsec.h>
224 #include <netkey/key.h>
225 #endif /*IPSEC*/
226 #ifdef INET6
227 #include "faith.h"
228 #if defined(NFAITH) && NFAITH > 0
229 #include <net/if_faith.h>
230 #endif
231 #endif	/* IPSEC */
232 
233 #ifdef FAST_IPSEC
234 #include <netipsec/ipsec.h>
235 #include <netipsec/ipsec_var.h>			/* XXX ipsecstat namespace */
236 #include <netipsec/key.h>
237 #ifdef INET6
238 #include <netipsec/ipsec6.h>
239 #endif
240 #endif	/* FAST_IPSEC*/
241 
242 int	tcprexmtthresh = 3;
243 int	tcp_log_refused;
244 
245 int	tcp_do_autorcvbuf = 0;
246 int	tcp_autorcvbuf_inc = 16 * 1024;
247 int	tcp_autorcvbuf_max = 256 * 1024;
248 
249 static int tcp_rst_ppslim_count = 0;
250 static struct timeval tcp_rst_ppslim_last;
251 static int tcp_ackdrop_ppslim_count = 0;
252 static struct timeval tcp_ackdrop_ppslim_last;
253 
254 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
255 
256 /* for modulo comparisons of timestamps */
257 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
258 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
259 
260 /*
261  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
262  */
263 #ifdef INET6
264 #define ND6_HINT(tp) \
265 do { \
266 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 && \
267 	    tp->t_in6pcb->in6p_route.ro_rt) { \
268 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
269 	} \
270 } while (/*CONSTCOND*/ 0)
271 #else
272 #define ND6_HINT(tp)
273 #endif
274 
275 /*
276  * Macro to compute ACK transmission behavior.  Delay the ACK unless
277  * we have already delayed an ACK (must send an ACK every two segments).
278  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
279  * option is enabled.
280  */
281 #define	TCP_SETUP_ACK(tp, th) \
282 do { \
283 	if ((tp)->t_flags & TF_DELACK || \
284 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
285 		tp->t_flags |= TF_ACKNOW; \
286 	else \
287 		TCP_SET_DELACK(tp); \
288 } while (/*CONSTCOND*/ 0)
289 
290 #define ICMP_CHECK(tp, th, acked) \
291 do { \
292 	/* \
293 	 * If we had a pending ICMP message that \
294 	 * refers to data that have just been  \
295 	 * acknowledged, disregard the recorded ICMP \
296 	 * message. \
297 	 */ \
298 	if (((tp)->t_flags & TF_PMTUD_PEND) && \
299 	    SEQ_GT((th)->th_ack, (tp)->t_pmtud_th_seq)) \
300 		(tp)->t_flags &= ~TF_PMTUD_PEND; \
301 \
302 	/* \
303 	 * Keep track of the largest chunk of data \
304 	 * acknowledged since last PMTU update \
305 	 */ \
306 	if ((tp)->t_pmtud_mss_acked < (acked)) \
307 		(tp)->t_pmtud_mss_acked = (acked); \
308 } while (/*CONSTCOND*/ 0)
309 
310 /*
311  * Convert TCP protocol fields to host order for easier processing.
312  */
313 #define	TCP_FIELDS_TO_HOST(th)						\
314 do {									\
315 	NTOHL((th)->th_seq);						\
316 	NTOHL((th)->th_ack);						\
317 	NTOHS((th)->th_win);						\
318 	NTOHS((th)->th_urp);						\
319 } while (/*CONSTCOND*/ 0)
320 
321 /*
322  * ... and reverse the above.
323  */
324 #define	TCP_FIELDS_TO_NET(th)						\
325 do {									\
326 	HTONL((th)->th_seq);						\
327 	HTONL((th)->th_ack);						\
328 	HTONS((th)->th_win);						\
329 	HTONS((th)->th_urp);						\
330 } while (/*CONSTCOND*/ 0)
331 
332 #ifdef TCP_CSUM_COUNTERS
333 #include <sys/device.h>
334 
335 #if defined(INET)
336 extern struct evcnt tcp_hwcsum_ok;
337 extern struct evcnt tcp_hwcsum_bad;
338 extern struct evcnt tcp_hwcsum_data;
339 extern struct evcnt tcp_swcsum;
340 #endif /* defined(INET) */
341 #if defined(INET6)
342 extern struct evcnt tcp6_hwcsum_ok;
343 extern struct evcnt tcp6_hwcsum_bad;
344 extern struct evcnt tcp6_hwcsum_data;
345 extern struct evcnt tcp6_swcsum;
346 #endif /* defined(INET6) */
347 
348 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
349 
350 #else
351 
352 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
353 
354 #endif /* TCP_CSUM_COUNTERS */
355 
356 #ifdef TCP_REASS_COUNTERS
357 #include <sys/device.h>
358 
359 extern struct evcnt tcp_reass_;
360 extern struct evcnt tcp_reass_empty;
361 extern struct evcnt tcp_reass_iteration[8];
362 extern struct evcnt tcp_reass_prependfirst;
363 extern struct evcnt tcp_reass_prepend;
364 extern struct evcnt tcp_reass_insert;
365 extern struct evcnt tcp_reass_inserttail;
366 extern struct evcnt tcp_reass_append;
367 extern struct evcnt tcp_reass_appendtail;
368 extern struct evcnt tcp_reass_overlaptail;
369 extern struct evcnt tcp_reass_overlapfront;
370 extern struct evcnt tcp_reass_segdup;
371 extern struct evcnt tcp_reass_fragdup;
372 
373 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
374 
375 #else
376 
377 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
378 
379 #endif /* TCP_REASS_COUNTERS */
380 
381 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *,
382     int *);
383 static int tcp_dooptions(struct tcpcb *, const u_char *, int,
384     struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *);
385 
386 #ifdef INET
387 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
388 #endif
389 #ifdef INET6
390 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
391 #endif
392 
393 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
394 
395 #if defined(MBUFTRACE)
396 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass");
397 #endif /* defined(MBUFTRACE) */
398 
399 static POOL_INIT(tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl",
400     NULL, IPL_VM);
401 
402 struct ipqent *
403 tcpipqent_alloc()
404 {
405 	struct ipqent *ipqe;
406 	int s;
407 
408 	s = splvm();
409 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
410 	splx(s);
411 
412 	return ipqe;
413 }
414 
415 void
416 tcpipqent_free(struct ipqent *ipqe)
417 {
418 	int s;
419 
420 	s = splvm();
421 	pool_put(&tcpipqent_pool, ipqe);
422 	splx(s);
423 }
424 
425 static int
426 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen)
427 {
428 	struct ipqent *p, *q, *nq, *tiqe = NULL;
429 	struct socket *so = NULL;
430 	int pkt_flags;
431 	tcp_seq pkt_seq;
432 	unsigned pkt_len;
433 	u_long rcvpartdupbyte = 0;
434 	u_long rcvoobyte;
435 #ifdef TCP_REASS_COUNTERS
436 	u_int count = 0;
437 #endif
438 
439 	if (tp->t_inpcb)
440 		so = tp->t_inpcb->inp_socket;
441 #ifdef INET6
442 	else if (tp->t_in6pcb)
443 		so = tp->t_in6pcb->in6p_socket;
444 #endif
445 
446 	TCP_REASS_LOCK_CHECK(tp);
447 
448 	/*
449 	 * Call with th==0 after become established to
450 	 * force pre-ESTABLISHED data up to user socket.
451 	 */
452 	if (th == 0)
453 		goto present;
454 
455 	m_claimm(m, &tcp_reass_mowner);
456 
457 	rcvoobyte = *tlen;
458 	/*
459 	 * Copy these to local variables because the tcpiphdr
460 	 * gets munged while we are collapsing mbufs.
461 	 */
462 	pkt_seq = th->th_seq;
463 	pkt_len = *tlen;
464 	pkt_flags = th->th_flags;
465 
466 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
467 
468 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
469 		/*
470 		 * When we miss a packet, the vast majority of time we get
471 		 * packets that follow it in order.  So optimize for that.
472 		 */
473 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
474 			p->ipqe_len += pkt_len;
475 			p->ipqe_flags |= pkt_flags;
476 			m_cat(p->ipre_mlast, m);
477 			TRAVERSE(p->ipre_mlast);
478 			m = NULL;
479 			tiqe = p;
480 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
481 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
482 			goto skip_replacement;
483 		}
484 		/*
485 		 * While we're here, if the pkt is completely beyond
486 		 * anything we have, just insert it at the tail.
487 		 */
488 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
489 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
490 			goto insert_it;
491 		}
492 	}
493 
494 	q = TAILQ_FIRST(&tp->segq);
495 
496 	if (q != NULL) {
497 		/*
498 		 * If this segment immediately precedes the first out-of-order
499 		 * block, simply slap the segment in front of it and (mostly)
500 		 * skip the complicated logic.
501 		 */
502 		if (pkt_seq + pkt_len == q->ipqe_seq) {
503 			q->ipqe_seq = pkt_seq;
504 			q->ipqe_len += pkt_len;
505 			q->ipqe_flags |= pkt_flags;
506 			m_cat(m, q->ipqe_m);
507 			q->ipqe_m = m;
508 			q->ipre_mlast = m; /* last mbuf may have changed */
509 			TRAVERSE(q->ipre_mlast);
510 			tiqe = q;
511 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
512 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
513 			goto skip_replacement;
514 		}
515 	} else {
516 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
517 	}
518 
519 	/*
520 	 * Find a segment which begins after this one does.
521 	 */
522 	for (p = NULL; q != NULL; q = nq) {
523 		nq = TAILQ_NEXT(q, ipqe_q);
524 #ifdef TCP_REASS_COUNTERS
525 		count++;
526 #endif
527 		/*
528 		 * If the received segment is just right after this
529 		 * fragment, merge the two together and then check
530 		 * for further overlaps.
531 		 */
532 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
533 #ifdef TCPREASS_DEBUG
534 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
535 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
536 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
537 #endif
538 			pkt_len += q->ipqe_len;
539 			pkt_flags |= q->ipqe_flags;
540 			pkt_seq = q->ipqe_seq;
541 			m_cat(q->ipre_mlast, m);
542 			TRAVERSE(q->ipre_mlast);
543 			m = q->ipqe_m;
544 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
545 			goto free_ipqe;
546 		}
547 		/*
548 		 * If the received segment is completely past this
549 		 * fragment, we need to go the next fragment.
550 		 */
551 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
552 			p = q;
553 			continue;
554 		}
555 		/*
556 		 * If the fragment is past the received segment,
557 		 * it (or any following) can't be concatenated.
558 		 */
559 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
560 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
561 			break;
562 		}
563 
564 		/*
565 		 * We've received all the data in this segment before.
566 		 * mark it as a duplicate and return.
567 		 */
568 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
569 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
570 			tcpstat.tcps_rcvduppack++;
571 			tcpstat.tcps_rcvdupbyte += pkt_len;
572 			tcp_new_dsack(tp, pkt_seq, pkt_len);
573 			m_freem(m);
574 			if (tiqe != NULL) {
575 				tcpipqent_free(tiqe);
576 			}
577 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
578 			return (0);
579 		}
580 		/*
581 		 * Received segment completely overlaps this fragment
582 		 * so we drop the fragment (this keeps the temporal
583 		 * ordering of segments correct).
584 		 */
585 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
586 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
587 			rcvpartdupbyte += q->ipqe_len;
588 			m_freem(q->ipqe_m);
589 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
590 			goto free_ipqe;
591 		}
592 		/*
593 		 * RX'ed segment extends past the end of the
594 		 * fragment.  Drop the overlapping bytes.  Then
595 		 * merge the fragment and segment then treat as
596 		 * a longer received packet.
597 		 */
598 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
599 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
600 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
601 #ifdef TCPREASS_DEBUG
602 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
603 			       tp, overlap,
604 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
605 #endif
606 			m_adj(m, overlap);
607 			rcvpartdupbyte += overlap;
608 			m_cat(q->ipre_mlast, m);
609 			TRAVERSE(q->ipre_mlast);
610 			m = q->ipqe_m;
611 			pkt_seq = q->ipqe_seq;
612 			pkt_len += q->ipqe_len - overlap;
613 			rcvoobyte -= overlap;
614 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
615 			goto free_ipqe;
616 		}
617 		/*
618 		 * RX'ed segment extends past the front of the
619 		 * fragment.  Drop the overlapping bytes on the
620 		 * received packet.  The packet will then be
621 		 * contatentated with this fragment a bit later.
622 		 */
623 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
624 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
625 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
626 #ifdef TCPREASS_DEBUG
627 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
628 			       tp, overlap,
629 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
630 #endif
631 			m_adj(m, -overlap);
632 			pkt_len -= overlap;
633 			rcvpartdupbyte += overlap;
634 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
635 			rcvoobyte -= overlap;
636 		}
637 		/*
638 		 * If the received segment immediates precedes this
639 		 * fragment then tack the fragment onto this segment
640 		 * and reinsert the data.
641 		 */
642 		if (q->ipqe_seq == pkt_seq + pkt_len) {
643 #ifdef TCPREASS_DEBUG
644 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
645 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
646 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
647 #endif
648 			pkt_len += q->ipqe_len;
649 			pkt_flags |= q->ipqe_flags;
650 			m_cat(m, q->ipqe_m);
651 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
652 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
653 			tp->t_segqlen--;
654 			KASSERT(tp->t_segqlen >= 0);
655 			KASSERT(tp->t_segqlen != 0 ||
656 			    (TAILQ_EMPTY(&tp->segq) &&
657 			    TAILQ_EMPTY(&tp->timeq)));
658 			if (tiqe == NULL) {
659 				tiqe = q;
660 			} else {
661 				tcpipqent_free(q);
662 			}
663 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
664 			break;
665 		}
666 		/*
667 		 * If the fragment is before the segment, remember it.
668 		 * When this loop is terminated, p will contain the
669 		 * pointer to fragment that is right before the received
670 		 * segment.
671 		 */
672 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
673 			p = q;
674 
675 		continue;
676 
677 		/*
678 		 * This is a common operation.  It also will allow
679 		 * to save doing a malloc/free in most instances.
680 		 */
681 	  free_ipqe:
682 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
683 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
684 		tp->t_segqlen--;
685 		KASSERT(tp->t_segqlen >= 0);
686 		KASSERT(tp->t_segqlen != 0 ||
687 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
688 		if (tiqe == NULL) {
689 			tiqe = q;
690 		} else {
691 			tcpipqent_free(q);
692 		}
693 	}
694 
695 #ifdef TCP_REASS_COUNTERS
696 	if (count > 7)
697 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
698 	else if (count > 0)
699 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
700 #endif
701 
702     insert_it:
703 
704 	/*
705 	 * Allocate a new queue entry since the received segment did not
706 	 * collapse onto any other out-of-order block; thus we are allocating
707 	 * a new block.  If it had collapsed, tiqe would not be NULL and
708 	 * we would be reusing it.
709 	 * XXX If we can't, just drop the packet.  XXX
710 	 */
711 	if (tiqe == NULL) {
712 		tiqe = tcpipqent_alloc();
713 		if (tiqe == NULL) {
714 			tcpstat.tcps_rcvmemdrop++;
715 			m_freem(m);
716 			return (0);
717 		}
718 	}
719 
720 	/*
721 	 * Update the counters.
722 	 */
723 	tcpstat.tcps_rcvoopack++;
724 	tcpstat.tcps_rcvoobyte += rcvoobyte;
725 	if (rcvpartdupbyte) {
726 	    tcpstat.tcps_rcvpartduppack++;
727 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
728 	}
729 
730 	/*
731 	 * Insert the new fragment queue entry into both queues.
732 	 */
733 	tiqe->ipqe_m = m;
734 	tiqe->ipre_mlast = m;
735 	tiqe->ipqe_seq = pkt_seq;
736 	tiqe->ipqe_len = pkt_len;
737 	tiqe->ipqe_flags = pkt_flags;
738 	if (p == NULL) {
739 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
740 #ifdef TCPREASS_DEBUG
741 		if (tiqe->ipqe_seq != tp->rcv_nxt)
742 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
743 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
744 #endif
745 	} else {
746 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
747 #ifdef TCPREASS_DEBUG
748 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
749 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
750 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
751 #endif
752 	}
753 	tp->t_segqlen++;
754 
755 skip_replacement:
756 
757 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
758 
759 present:
760 	/*
761 	 * Present data to user, advancing rcv_nxt through
762 	 * completed sequence space.
763 	 */
764 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
765 		return (0);
766 	q = TAILQ_FIRST(&tp->segq);
767 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
768 		return (0);
769 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
770 		return (0);
771 
772 	tp->rcv_nxt += q->ipqe_len;
773 	pkt_flags = q->ipqe_flags & TH_FIN;
774 	ND6_HINT(tp);
775 
776 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
777 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
778 	tp->t_segqlen--;
779 	KASSERT(tp->t_segqlen >= 0);
780 	KASSERT(tp->t_segqlen != 0 ||
781 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
782 	if (so->so_state & SS_CANTRCVMORE)
783 		m_freem(q->ipqe_m);
784 	else
785 		sbappendstream(&so->so_rcv, q->ipqe_m);
786 	tcpipqent_free(q);
787 	sorwakeup(so);
788 	return (pkt_flags);
789 }
790 
791 #ifdef INET6
792 int
793 tcp6_input(struct mbuf **mp, int *offp, int proto)
794 {
795 	struct mbuf *m = *mp;
796 
797 	/*
798 	 * draft-itojun-ipv6-tcp-to-anycast
799 	 * better place to put this in?
800 	 */
801 	if (m->m_flags & M_ANYCAST6) {
802 		struct ip6_hdr *ip6;
803 		if (m->m_len < sizeof(struct ip6_hdr)) {
804 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
805 				tcpstat.tcps_rcvshort++;
806 				return IPPROTO_DONE;
807 			}
808 		}
809 		ip6 = mtod(m, struct ip6_hdr *);
810 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
811 		    (char *)&ip6->ip6_dst - (char *)ip6);
812 		return IPPROTO_DONE;
813 	}
814 
815 	tcp_input(m, *offp, proto);
816 	return IPPROTO_DONE;
817 }
818 #endif
819 
820 #ifdef INET
821 static void
822 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
823 {
824 	char src[4*sizeof "123"];
825 	char dst[4*sizeof "123"];
826 
827 	if (ip) {
828 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
829 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
830 	}
831 	else {
832 		strlcpy(src, "(unknown)", sizeof(src));
833 		strlcpy(dst, "(unknown)", sizeof(dst));
834 	}
835 	log(LOG_INFO,
836 	    "Connection attempt to TCP %s:%d from %s:%d\n",
837 	    dst, ntohs(th->th_dport),
838 	    src, ntohs(th->th_sport));
839 }
840 #endif
841 
842 #ifdef INET6
843 static void
844 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
845 {
846 	char src[INET6_ADDRSTRLEN];
847 	char dst[INET6_ADDRSTRLEN];
848 
849 	if (ip6) {
850 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
851 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
852 	}
853 	else {
854 		strlcpy(src, "(unknown v6)", sizeof(src));
855 		strlcpy(dst, "(unknown v6)", sizeof(dst));
856 	}
857 	log(LOG_INFO,
858 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
859 	    dst, ntohs(th->th_dport),
860 	    src, ntohs(th->th_sport));
861 }
862 #endif
863 
864 /*
865  * Checksum extended TCP header and data.
866  */
867 int
868 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th,
869     int toff, int off, int tlen)
870 {
871 
872 	/*
873 	 * XXX it's better to record and check if this mbuf is
874 	 * already checked.
875 	 */
876 
877 	switch (af) {
878 #ifdef INET
879 	case AF_INET:
880 		switch (m->m_pkthdr.csum_flags &
881 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
882 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
883 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
884 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
885 			goto badcsum;
886 
887 		case M_CSUM_TCPv4|M_CSUM_DATA: {
888 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
889 
890 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
891 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
892 				const struct ip *ip =
893 				    mtod(m, const struct ip *);
894 
895 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
896 				    ip->ip_dst.s_addr,
897 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
898 			}
899 			if ((hw_csum ^ 0xffff) != 0)
900 				goto badcsum;
901 			break;
902 		}
903 
904 		case M_CSUM_TCPv4:
905 			/* Checksum was okay. */
906 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
907 			break;
908 
909 		default:
910 			/*
911 			 * Must compute it ourselves.  Maybe skip checksum
912 			 * on loopback interfaces.
913 			 */
914 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
915 					     IFF_LOOPBACK) ||
916 					   tcp_do_loopback_cksum)) {
917 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
918 				if (in4_cksum(m, IPPROTO_TCP, toff,
919 					      tlen + off) != 0)
920 					goto badcsum;
921 			}
922 			break;
923 		}
924 		break;
925 #endif /* INET4 */
926 
927 #ifdef INET6
928 	case AF_INET6:
929 		switch (m->m_pkthdr.csum_flags &
930 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
931 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
932 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
933 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
934 			goto badcsum;
935 
936 #if 0 /* notyet */
937 		case M_CSUM_TCPv6|M_CSUM_DATA:
938 #endif
939 
940 		case M_CSUM_TCPv6:
941 			/* Checksum was okay. */
942 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok);
943 			break;
944 
945 		default:
946 			/*
947 			 * Must compute it ourselves.  Maybe skip checksum
948 			 * on loopback interfaces.
949 			 */
950 			if (__predict_true((m->m_flags & M_LOOP) == 0 ||
951 			    tcp_do_loopback_cksum)) {
952 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
953 				if (in6_cksum(m, IPPROTO_TCP, toff,
954 				    tlen + off) != 0)
955 					goto badcsum;
956 			}
957 		}
958 		break;
959 #endif /* INET6 */
960 	}
961 
962 	return 0;
963 
964 badcsum:
965 	tcpstat.tcps_rcvbadsum++;
966 	return -1;
967 }
968 
969 /*
970  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
971  */
972 void
973 tcp_input(struct mbuf *m, ...)
974 {
975 	struct tcphdr *th;
976 	struct ip *ip;
977 	struct inpcb *inp;
978 #ifdef INET6
979 	struct ip6_hdr *ip6;
980 	struct in6pcb *in6p;
981 #endif
982 	u_int8_t *optp = NULL;
983 	int optlen = 0;
984 	int len, tlen, toff, hdroptlen = 0;
985 	struct tcpcb *tp = 0;
986 	int tiflags;
987 	struct socket *so = NULL;
988 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
989 #ifdef TCP_DEBUG
990 	short ostate = 0;
991 #endif
992 	u_long tiwin;
993 	struct tcp_opt_info opti;
994 	int off, iphlen;
995 	va_list ap;
996 	int af;		/* af on the wire */
997 	struct mbuf *tcp_saveti = NULL;
998 	uint32_t ts_rtt;
999 	uint8_t iptos;
1000 
1001 	MCLAIM(m, &tcp_rx_mowner);
1002 	va_start(ap, m);
1003 	toff = va_arg(ap, int);
1004 	(void)va_arg(ap, int);		/* ignore value, advance ap */
1005 	va_end(ap);
1006 
1007 	tcpstat.tcps_rcvtotal++;
1008 
1009 	bzero(&opti, sizeof(opti));
1010 	opti.ts_present = 0;
1011 	opti.maxseg = 0;
1012 
1013 	/*
1014 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
1015 	 *
1016 	 * TCP is, by definition, unicast, so we reject all
1017 	 * multicast outright.
1018 	 *
1019 	 * Note, there are additional src/dst address checks in
1020 	 * the AF-specific code below.
1021 	 */
1022 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1023 		/* XXX stat */
1024 		goto drop;
1025 	}
1026 #ifdef INET6
1027 	if (m->m_flags & M_ANYCAST6) {
1028 		/* XXX stat */
1029 		goto drop;
1030 	}
1031 #endif
1032 
1033 	/*
1034 	 * Get IP and TCP header.
1035 	 * Note: IP leaves IP header in first mbuf.
1036 	 */
1037 	ip = mtod(m, struct ip *);
1038 #ifdef INET6
1039 	ip6 = NULL;
1040 #endif
1041 	switch (ip->ip_v) {
1042 #ifdef INET
1043 	case 4:
1044 		af = AF_INET;
1045 		iphlen = sizeof(struct ip);
1046 		ip = mtod(m, struct ip *);
1047 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1048 			sizeof(struct tcphdr));
1049 		if (th == NULL) {
1050 			tcpstat.tcps_rcvshort++;
1051 			return;
1052 		}
1053 		/* We do the checksum after PCB lookup... */
1054 		len = ntohs(ip->ip_len);
1055 		tlen = len - toff;
1056 		iptos = ip->ip_tos;
1057 		break;
1058 #endif
1059 #ifdef INET6
1060 	case 6:
1061 		ip = NULL;
1062 		iphlen = sizeof(struct ip6_hdr);
1063 		af = AF_INET6;
1064 		ip6 = mtod(m, struct ip6_hdr *);
1065 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1066 			sizeof(struct tcphdr));
1067 		if (th == NULL) {
1068 			tcpstat.tcps_rcvshort++;
1069 			return;
1070 		}
1071 
1072 		/* Be proactive about malicious use of IPv4 mapped address */
1073 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1074 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1075 			/* XXX stat */
1076 			goto drop;
1077 		}
1078 
1079 		/*
1080 		 * Be proactive about unspecified IPv6 address in source.
1081 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1082 		 * unspecified IPv6 address can be used to confuse us.
1083 		 *
1084 		 * Note that packets with unspecified IPv6 destination is
1085 		 * already dropped in ip6_input.
1086 		 */
1087 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1088 			/* XXX stat */
1089 			goto drop;
1090 		}
1091 
1092 		/*
1093 		 * Make sure destination address is not multicast.
1094 		 * Source address checked in ip6_input().
1095 		 */
1096 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1097 			/* XXX stat */
1098 			goto drop;
1099 		}
1100 
1101 		/* We do the checksum after PCB lookup... */
1102 		len = m->m_pkthdr.len;
1103 		tlen = len - toff;
1104 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1105 		break;
1106 #endif
1107 	default:
1108 		m_freem(m);
1109 		return;
1110 	}
1111 
1112 	KASSERT(TCP_HDR_ALIGNED_P(th));
1113 
1114 	/*
1115 	 * Check that TCP offset makes sense,
1116 	 * pull out TCP options and adjust length.		XXX
1117 	 */
1118 	off = th->th_off << 2;
1119 	if (off < sizeof (struct tcphdr) || off > tlen) {
1120 		tcpstat.tcps_rcvbadoff++;
1121 		goto drop;
1122 	}
1123 	tlen -= off;
1124 
1125 	/*
1126 	 * tcp_input() has been modified to use tlen to mean the TCP data
1127 	 * length throughout the function.  Other functions can use
1128 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1129 	 * rja
1130 	 */
1131 
1132 	if (off > sizeof (struct tcphdr)) {
1133 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1134 		if (th == NULL) {
1135 			tcpstat.tcps_rcvshort++;
1136 			return;
1137 		}
1138 		/*
1139 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1140 		 * (as they're before toff) and we don't need to update those.
1141 		 */
1142 		KASSERT(TCP_HDR_ALIGNED_P(th));
1143 		optlen = off - sizeof (struct tcphdr);
1144 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1145 		/*
1146 		 * Do quick retrieval of timestamp options ("options
1147 		 * prediction?").  If timestamp is the only option and it's
1148 		 * formatted as recommended in RFC 1323 appendix A, we
1149 		 * quickly get the values now and not bother calling
1150 		 * tcp_dooptions(), etc.
1151 		 */
1152 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1153 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1154 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1155 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1156 		     (th->th_flags & TH_SYN) == 0) {
1157 			opti.ts_present = 1;
1158 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1159 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1160 			optp = NULL;	/* we've parsed the options */
1161 		}
1162 	}
1163 	tiflags = th->th_flags;
1164 
1165 	/*
1166 	 * Locate pcb for segment.
1167 	 */
1168 findpcb:
1169 	inp = NULL;
1170 #ifdef INET6
1171 	in6p = NULL;
1172 #endif
1173 	switch (af) {
1174 #ifdef INET
1175 	case AF_INET:
1176 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1177 		    ip->ip_dst, th->th_dport);
1178 		if (inp == 0) {
1179 			++tcpstat.tcps_pcbhashmiss;
1180 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1181 		}
1182 #ifdef INET6
1183 		if (inp == 0) {
1184 			struct in6_addr s, d;
1185 
1186 			/* mapped addr case */
1187 			bzero(&s, sizeof(s));
1188 			s.s6_addr16[5] = htons(0xffff);
1189 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1190 			bzero(&d, sizeof(d));
1191 			d.s6_addr16[5] = htons(0xffff);
1192 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1193 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1194 			    th->th_sport, &d, th->th_dport, 0);
1195 			if (in6p == 0) {
1196 				++tcpstat.tcps_pcbhashmiss;
1197 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1198 				    th->th_dport, 0);
1199 			}
1200 		}
1201 #endif
1202 #ifndef INET6
1203 		if (inp == 0)
1204 #else
1205 		if (inp == 0 && in6p == 0)
1206 #endif
1207 		{
1208 			++tcpstat.tcps_noport;
1209 			if (tcp_log_refused &&
1210 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1211 				tcp4_log_refused(ip, th);
1212 			}
1213 			TCP_FIELDS_TO_HOST(th);
1214 			goto dropwithreset_ratelim;
1215 		}
1216 #if defined(IPSEC) || defined(FAST_IPSEC)
1217 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1218 		    ipsec4_in_reject(m, inp)) {
1219 			ipsecstat.in_polvio++;
1220 			goto drop;
1221 		}
1222 #ifdef INET6
1223 		else if (in6p &&
1224 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1225 		    ipsec6_in_reject_so(m, in6p->in6p_socket)) {
1226 			ipsecstat.in_polvio++;
1227 			goto drop;
1228 		}
1229 #endif
1230 #endif /*IPSEC*/
1231 		break;
1232 #endif /*INET*/
1233 #ifdef INET6
1234 	case AF_INET6:
1235 	    {
1236 		int faith;
1237 
1238 #if defined(NFAITH) && NFAITH > 0
1239 		faith = faithprefix(&ip6->ip6_dst);
1240 #else
1241 		faith = 0;
1242 #endif
1243 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1244 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1245 		if (in6p == NULL) {
1246 			++tcpstat.tcps_pcbhashmiss;
1247 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1248 				th->th_dport, faith);
1249 		}
1250 		if (in6p == NULL) {
1251 			++tcpstat.tcps_noport;
1252 			if (tcp_log_refused &&
1253 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1254 				tcp6_log_refused(ip6, th);
1255 			}
1256 			TCP_FIELDS_TO_HOST(th);
1257 			goto dropwithreset_ratelim;
1258 		}
1259 #if defined(IPSEC) || defined(FAST_IPSEC)
1260 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1261 		    ipsec6_in_reject(m, in6p)) {
1262 			ipsec6stat.in_polvio++;
1263 			goto drop;
1264 		}
1265 #endif /*IPSEC*/
1266 		break;
1267 	    }
1268 #endif
1269 	}
1270 
1271 	/*
1272 	 * If the state is CLOSED (i.e., TCB does not exist) then
1273 	 * all data in the incoming segment is discarded.
1274 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1275 	 * but should either do a listen or a connect soon.
1276 	 */
1277 	tp = NULL;
1278 	so = NULL;
1279 	if (inp) {
1280 		tp = intotcpcb(inp);
1281 		so = inp->inp_socket;
1282 	}
1283 #ifdef INET6
1284 	else if (in6p) {
1285 		tp = in6totcpcb(in6p);
1286 		so = in6p->in6p_socket;
1287 	}
1288 #endif
1289 	if (tp == 0) {
1290 		TCP_FIELDS_TO_HOST(th);
1291 		goto dropwithreset_ratelim;
1292 	}
1293 	if (tp->t_state == TCPS_CLOSED)
1294 		goto drop;
1295 
1296 	/*
1297 	 * Checksum extended TCP header and data.
1298 	 */
1299 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1300 		goto badcsum;
1301 
1302 	TCP_FIELDS_TO_HOST(th);
1303 
1304 	/* Unscale the window into a 32-bit value. */
1305 	if ((tiflags & TH_SYN) == 0)
1306 		tiwin = th->th_win << tp->snd_scale;
1307 	else
1308 		tiwin = th->th_win;
1309 
1310 #ifdef INET6
1311 	/* save packet options if user wanted */
1312 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1313 		if (in6p->in6p_options) {
1314 			m_freem(in6p->in6p_options);
1315 			in6p->in6p_options = 0;
1316 		}
1317 		KASSERT(ip6 != NULL);
1318 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1319 	}
1320 #endif
1321 
1322 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1323 		union syn_cache_sa src;
1324 		union syn_cache_sa dst;
1325 
1326 		bzero(&src, sizeof(src));
1327 		bzero(&dst, sizeof(dst));
1328 		switch (af) {
1329 #ifdef INET
1330 		case AF_INET:
1331 			src.sin.sin_len = sizeof(struct sockaddr_in);
1332 			src.sin.sin_family = AF_INET;
1333 			src.sin.sin_addr = ip->ip_src;
1334 			src.sin.sin_port = th->th_sport;
1335 
1336 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1337 			dst.sin.sin_family = AF_INET;
1338 			dst.sin.sin_addr = ip->ip_dst;
1339 			dst.sin.sin_port = th->th_dport;
1340 			break;
1341 #endif
1342 #ifdef INET6
1343 		case AF_INET6:
1344 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1345 			src.sin6.sin6_family = AF_INET6;
1346 			src.sin6.sin6_addr = ip6->ip6_src;
1347 			src.sin6.sin6_port = th->th_sport;
1348 
1349 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1350 			dst.sin6.sin6_family = AF_INET6;
1351 			dst.sin6.sin6_addr = ip6->ip6_dst;
1352 			dst.sin6.sin6_port = th->th_dport;
1353 			break;
1354 #endif /* INET6 */
1355 		default:
1356 			goto badsyn;	/*sanity*/
1357 		}
1358 
1359 		if (so->so_options & SO_DEBUG) {
1360 #ifdef TCP_DEBUG
1361 			ostate = tp->t_state;
1362 #endif
1363 
1364 			tcp_saveti = NULL;
1365 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1366 				goto nosave;
1367 
1368 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1369 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1370 				if (!tcp_saveti)
1371 					goto nosave;
1372 			} else {
1373 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1374 				if (!tcp_saveti)
1375 					goto nosave;
1376 				MCLAIM(m, &tcp_mowner);
1377 				tcp_saveti->m_len = iphlen;
1378 				m_copydata(m, 0, iphlen,
1379 				    mtod(tcp_saveti, void *));
1380 			}
1381 
1382 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1383 				m_freem(tcp_saveti);
1384 				tcp_saveti = NULL;
1385 			} else {
1386 				tcp_saveti->m_len += sizeof(struct tcphdr);
1387 				memcpy(mtod(tcp_saveti, char *) + iphlen, th,
1388 				    sizeof(struct tcphdr));
1389 			}
1390 	nosave:;
1391 		}
1392 		if (so->so_options & SO_ACCEPTCONN) {
1393 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1394 				if (tiflags & TH_RST) {
1395 					syn_cache_reset(&src.sa, &dst.sa, th);
1396 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1397 				    (TH_ACK|TH_SYN)) {
1398 					/*
1399 					 * Received a SYN,ACK.  This should
1400 					 * never happen while we are in
1401 					 * LISTEN.  Send an RST.
1402 					 */
1403 					goto badsyn;
1404 				} else if (tiflags & TH_ACK) {
1405 					so = syn_cache_get(&src.sa, &dst.sa,
1406 						th, toff, tlen, so, m);
1407 					if (so == NULL) {
1408 						/*
1409 						 * We don't have a SYN for
1410 						 * this ACK; send an RST.
1411 						 */
1412 						goto badsyn;
1413 					} else if (so ==
1414 					    (struct socket *)(-1)) {
1415 						/*
1416 						 * We were unable to create
1417 						 * the connection.  If the
1418 						 * 3-way handshake was
1419 						 * completed, and RST has
1420 						 * been sent to the peer.
1421 						 * Since the mbuf might be
1422 						 * in use for the reply,
1423 						 * do not free it.
1424 						 */
1425 						m = NULL;
1426 					} else {
1427 						/*
1428 						 * We have created a
1429 						 * full-blown connection.
1430 						 */
1431 						tp = NULL;
1432 						inp = NULL;
1433 #ifdef INET6
1434 						in6p = NULL;
1435 #endif
1436 						switch (so->so_proto->pr_domain->dom_family) {
1437 #ifdef INET
1438 						case AF_INET:
1439 							inp = sotoinpcb(so);
1440 							tp = intotcpcb(inp);
1441 							break;
1442 #endif
1443 #ifdef INET6
1444 						case AF_INET6:
1445 							in6p = sotoin6pcb(so);
1446 							tp = in6totcpcb(in6p);
1447 							break;
1448 #endif
1449 						}
1450 						if (tp == NULL)
1451 							goto badsyn;	/*XXX*/
1452 						tiwin <<= tp->snd_scale;
1453 						goto after_listen;
1454 					}
1455 				} else {
1456 					/*
1457 					 * None of RST, SYN or ACK was set.
1458 					 * This is an invalid packet for a
1459 					 * TCB in LISTEN state.  Send a RST.
1460 					 */
1461 					goto badsyn;
1462 				}
1463 			} else {
1464 				/*
1465 				 * Received a SYN.
1466 				 *
1467 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1468 				 */
1469 				if (m->m_flags & (M_BCAST|M_MCAST))
1470 					goto drop;
1471 
1472 				switch (af) {
1473 #ifdef INET6
1474 				case AF_INET6:
1475 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1476 						goto drop;
1477 					break;
1478 #endif /* INET6 */
1479 				case AF_INET:
1480 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1481 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1482 						goto drop;
1483 				break;
1484 				}
1485 
1486 #ifdef INET6
1487 				/*
1488 				 * If deprecated address is forbidden, we do
1489 				 * not accept SYN to deprecated interface
1490 				 * address to prevent any new inbound
1491 				 * connection from getting established.
1492 				 * When we do not accept SYN, we send a TCP
1493 				 * RST, with deprecated source address (instead
1494 				 * of dropping it).  We compromise it as it is
1495 				 * much better for peer to send a RST, and
1496 				 * RST will be the final packet for the
1497 				 * exchange.
1498 				 *
1499 				 * If we do not forbid deprecated addresses, we
1500 				 * accept the SYN packet.  RFC2462 does not
1501 				 * suggest dropping SYN in this case.
1502 				 * If we decipher RFC2462 5.5.4, it says like
1503 				 * this:
1504 				 * 1. use of deprecated addr with existing
1505 				 *    communication is okay - "SHOULD continue
1506 				 *    to be used"
1507 				 * 2. use of it with new communication:
1508 				 *   (2a) "SHOULD NOT be used if alternate
1509 				 *        address with sufficient scope is
1510 				 *        available"
1511 				 *   (2b) nothing mentioned otherwise.
1512 				 * Here we fall into (2b) case as we have no
1513 				 * choice in our source address selection - we
1514 				 * must obey the peer.
1515 				 *
1516 				 * The wording in RFC2462 is confusing, and
1517 				 * there are multiple description text for
1518 				 * deprecated address handling - worse, they
1519 				 * are not exactly the same.  I believe 5.5.4
1520 				 * is the best one, so we follow 5.5.4.
1521 				 */
1522 				if (af == AF_INET6 && !ip6_use_deprecated) {
1523 					struct in6_ifaddr *ia6;
1524 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1525 					    &ip6->ip6_dst)) &&
1526 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1527 						tp = NULL;
1528 						goto dropwithreset;
1529 					}
1530 				}
1531 #endif
1532 
1533 #if defined(IPSEC) || defined(FAST_IPSEC)
1534 				switch (af) {
1535 #ifdef INET
1536 				case AF_INET:
1537 					if (ipsec4_in_reject_so(m, so)) {
1538 						ipsecstat.in_polvio++;
1539 						tp = NULL;
1540 						goto dropwithreset;
1541 					}
1542 					break;
1543 #endif
1544 #ifdef INET6
1545 				case AF_INET6:
1546 					if (ipsec6_in_reject_so(m, so)) {
1547 						ipsec6stat.in_polvio++;
1548 						tp = NULL;
1549 						goto dropwithreset;
1550 					}
1551 					break;
1552 #endif /*INET6*/
1553 				}
1554 #endif /*IPSEC*/
1555 
1556 				/*
1557 				 * LISTEN socket received a SYN
1558 				 * from itself?  This can't possibly
1559 				 * be valid; drop the packet.
1560 				 */
1561 				if (th->th_sport == th->th_dport) {
1562 					int i;
1563 
1564 					switch (af) {
1565 #ifdef INET
1566 					case AF_INET:
1567 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1568 						break;
1569 #endif
1570 #ifdef INET6
1571 					case AF_INET6:
1572 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1573 						break;
1574 #endif
1575 					default:
1576 						i = 1;
1577 					}
1578 					if (i) {
1579 						tcpstat.tcps_badsyn++;
1580 						goto drop;
1581 					}
1582 				}
1583 
1584 				/*
1585 				 * SYN looks ok; create compressed TCP
1586 				 * state for it.
1587 				 */
1588 				if (so->so_qlen <= so->so_qlimit &&
1589 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1590 						so, m, optp, optlen, &opti))
1591 					m = NULL;
1592 			}
1593 			goto drop;
1594 		}
1595 	}
1596 
1597 after_listen:
1598 #ifdef DIAGNOSTIC
1599 	/*
1600 	 * Should not happen now that all embryonic connections
1601 	 * are handled with compressed state.
1602 	 */
1603 	if (tp->t_state == TCPS_LISTEN)
1604 		panic("tcp_input: TCPS_LISTEN");
1605 #endif
1606 
1607 	/*
1608 	 * Segment received on connection.
1609 	 * Reset idle time and keep-alive timer.
1610 	 */
1611 	tp->t_rcvtime = tcp_now;
1612 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1613 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1614 
1615 	/*
1616 	 * Process options.
1617 	 */
1618 #ifdef TCP_SIGNATURE
1619 	if (optp || (tp->t_flags & TF_SIGNATURE))
1620 #else
1621 	if (optp)
1622 #endif
1623 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1624 			goto drop;
1625 
1626 	if (TCP_SACK_ENABLED(tp)) {
1627 		tcp_del_sackholes(tp, th);
1628 	}
1629 
1630 	if (TCP_ECN_ALLOWED(tp)) {
1631 		switch (iptos & IPTOS_ECN_MASK) {
1632 		case IPTOS_ECN_CE:
1633 			tp->t_flags |= TF_ECN_SND_ECE;
1634 			tcpstat.tcps_ecn_ce++;
1635 			break;
1636 		case IPTOS_ECN_ECT0:
1637 			tcpstat.tcps_ecn_ect++;
1638 			break;
1639 		case IPTOS_ECN_ECT1:
1640 			/* XXX: ignore for now -- rpaulo */
1641 			break;
1642 		}
1643 
1644 		if (tiflags & TH_CWR)
1645 			tp->t_flags &= ~TF_ECN_SND_ECE;
1646 
1647 		/*
1648 		 * Congestion experienced.
1649 		 * Ignore if we are already trying to recover.
1650 		 */
1651 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1652 			tp->t_congctl->cong_exp(tp);
1653 	}
1654 
1655 	if (opti.ts_present && opti.ts_ecr) {
1656 		/*
1657 		 * Calculate the RTT from the returned time stamp and the
1658 		 * connection's time base.  If the time stamp is later than
1659 		 * the current time, or is extremely old, fall back to non-1323
1660 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1661 		 * at the same time.
1662 		 */
1663 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1664 		if (ts_rtt > TCP_PAWS_IDLE)
1665 			ts_rtt = 0;
1666 	} else {
1667 		ts_rtt = 0;
1668 	}
1669 
1670 	/*
1671 	 * Header prediction: check for the two common cases
1672 	 * of a uni-directional data xfer.  If the packet has
1673 	 * no control flags, is in-sequence, the window didn't
1674 	 * change and we're not retransmitting, it's a
1675 	 * candidate.  If the length is zero and the ack moved
1676 	 * forward, we're the sender side of the xfer.  Just
1677 	 * free the data acked & wake any higher level process
1678 	 * that was blocked waiting for space.  If the length
1679 	 * is non-zero and the ack didn't move, we're the
1680 	 * receiver side.  If we're getting packets in-order
1681 	 * (the reassembly queue is empty), add the data to
1682 	 * the socket buffer and note that we need a delayed ack.
1683 	 */
1684 	if (tp->t_state == TCPS_ESTABLISHED &&
1685 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1686 	        == TH_ACK &&
1687 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1688 	    th->th_seq == tp->rcv_nxt &&
1689 	    tiwin && tiwin == tp->snd_wnd &&
1690 	    tp->snd_nxt == tp->snd_max) {
1691 
1692 		/*
1693 		 * If last ACK falls within this segment's sequence numbers,
1694 		 *  record the timestamp.
1695 		 * NOTE:
1696 		 * 1) That the test incorporates suggestions from the latest
1697 		 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1698 		 * 2) That updating only on newer timestamps interferes with
1699 		 *    our earlier PAWS tests, so this check should be solely
1700 		 *    predicated on the sequence space of this segment.
1701 		 * 3) That we modify the segment boundary check to be
1702 		 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
1703 		 *    instead of RFC1323's
1704 		 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
1705 		 *    This modified check allows us to overcome RFC1323's
1706 		 *    limitations as described in Stevens TCP/IP Illustrated
1707 		 *    Vol. 2 p.869. In such cases, we can still calculate the
1708 		 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1709 		 */
1710 		if (opti.ts_present &&
1711 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1712 		    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1713 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1714 			tp->ts_recent_age = tcp_now;
1715 			tp->ts_recent = opti.ts_val;
1716 		}
1717 
1718 		if (tlen == 0) {
1719 			/* Ack prediction. */
1720 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1721 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1722 			    tp->snd_cwnd >= tp->snd_wnd &&
1723 			    tp->t_partialacks < 0) {
1724 				/*
1725 				 * this is a pure ack for outstanding data.
1726 				 */
1727 				++tcpstat.tcps_predack;
1728 				if (ts_rtt)
1729 					tcp_xmit_timer(tp, ts_rtt);
1730 				else if (tp->t_rtttime &&
1731 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1732 					tcp_xmit_timer(tp,
1733 					  tcp_now - tp->t_rtttime);
1734 				acked = th->th_ack - tp->snd_una;
1735 				tcpstat.tcps_rcvackpack++;
1736 				tcpstat.tcps_rcvackbyte += acked;
1737 				ND6_HINT(tp);
1738 
1739 				if (acked > (tp->t_lastoff - tp->t_inoff))
1740 					tp->t_lastm = NULL;
1741 				sbdrop(&so->so_snd, acked);
1742 				tp->t_lastoff -= acked;
1743 
1744 				ICMP_CHECK(tp, th, acked);
1745 
1746 				tp->snd_una = th->th_ack;
1747 				tp->snd_fack = tp->snd_una;
1748 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1749 					tp->snd_high = tp->snd_una;
1750 				m_freem(m);
1751 
1752 				/*
1753 				 * If all outstanding data are acked, stop
1754 				 * retransmit timer, otherwise restart timer
1755 				 * using current (possibly backed-off) value.
1756 				 * If process is waiting for space,
1757 				 * wakeup/selwakeup/signal.  If data
1758 				 * are ready to send, let tcp_output
1759 				 * decide between more output or persist.
1760 				 */
1761 				if (tp->snd_una == tp->snd_max)
1762 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1763 				else if (TCP_TIMER_ISARMED(tp,
1764 				    TCPT_PERSIST) == 0)
1765 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1766 					    tp->t_rxtcur);
1767 
1768 				sowwakeup(so);
1769 				if (so->so_snd.sb_cc)
1770 					(void) tcp_output(tp);
1771 				if (tcp_saveti)
1772 					m_freem(tcp_saveti);
1773 				return;
1774 			}
1775 		} else if (th->th_ack == tp->snd_una &&
1776 		    TAILQ_FIRST(&tp->segq) == NULL &&
1777 		    tlen <= sbspace(&so->so_rcv)) {
1778 			int newsize = 0;	/* automatic sockbuf scaling */
1779 
1780 			/*
1781 			 * this is a pure, in-sequence data packet
1782 			 * with nothing on the reassembly queue and
1783 			 * we have enough buffer space to take it.
1784 			 */
1785 			++tcpstat.tcps_preddat;
1786 			tp->rcv_nxt += tlen;
1787 			tcpstat.tcps_rcvpack++;
1788 			tcpstat.tcps_rcvbyte += tlen;
1789 			ND6_HINT(tp);
1790 
1791 		/*
1792 		 * Automatic sizing enables the performance of large buffers
1793 		 * and most of the efficiency of small ones by only allocating
1794 		 * space when it is needed.
1795 		 *
1796 		 * On the receive side the socket buffer memory is only rarely
1797 		 * used to any significant extent.  This allows us to be much
1798 		 * more aggressive in scaling the receive socket buffer.  For
1799 		 * the case that the buffer space is actually used to a large
1800 		 * extent and we run out of kernel memory we can simply drop
1801 		 * the new segments; TCP on the sender will just retransmit it
1802 		 * later.  Setting the buffer size too big may only consume too
1803 		 * much kernel memory if the application doesn't read() from
1804 		 * the socket or packet loss or reordering makes use of the
1805 		 * reassembly queue.
1806 		 *
1807 		 * The criteria to step up the receive buffer one notch are:
1808 		 *  1. the number of bytes received during the time it takes
1809 		 *     one timestamp to be reflected back to us (the RTT);
1810 		 *  2. received bytes per RTT is within seven eighth of the
1811 		 *     current socket buffer size;
1812 		 *  3. receive buffer size has not hit maximal automatic size;
1813 		 *
1814 		 * This algorithm does one step per RTT at most and only if
1815 		 * we receive a bulk stream w/o packet losses or reorderings.
1816 		 * Shrinking the buffer during idle times is not necessary as
1817 		 * it doesn't consume any memory when idle.
1818 		 *
1819 		 * TODO: Only step up if the application is actually serving
1820 		 * the buffer to better manage the socket buffer resources.
1821 		 */
1822 			if (tcp_do_autorcvbuf &&
1823 			    opti.ts_ecr &&
1824 			    (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1825 				if (opti.ts_ecr > tp->rfbuf_ts &&
1826 				    opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) {
1827 					if (tp->rfbuf_cnt >
1828 					    (so->so_rcv.sb_hiwat / 8 * 7) &&
1829 					    so->so_rcv.sb_hiwat <
1830 					    tcp_autorcvbuf_max) {
1831 						newsize =
1832 						    min(so->so_rcv.sb_hiwat +
1833 						    tcp_autorcvbuf_inc,
1834 						    tcp_autorcvbuf_max);
1835 					}
1836 					/* Start over with next RTT. */
1837 					tp->rfbuf_ts = 0;
1838 					tp->rfbuf_cnt = 0;
1839 				} else
1840 					tp->rfbuf_cnt += tlen;	/* add up */
1841 			}
1842 
1843 			/*
1844 			 * Drop TCP, IP headers and TCP options then add data
1845 			 * to socket buffer.
1846 			 */
1847 			if (so->so_state & SS_CANTRCVMORE)
1848 				m_freem(m);
1849 			else {
1850 				/*
1851 				 * Set new socket buffer size.
1852 				 * Give up when limit is reached.
1853 				 */
1854 				if (newsize)
1855 					if (!sbreserve(&so->so_rcv,
1856 					    newsize, so))
1857 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1858 				m_adj(m, toff + off);
1859 				sbappendstream(&so->so_rcv, m);
1860 			}
1861 			sorwakeup(so);
1862 			TCP_SETUP_ACK(tp, th);
1863 			if (tp->t_flags & TF_ACKNOW)
1864 				(void) tcp_output(tp);
1865 			if (tcp_saveti)
1866 				m_freem(tcp_saveti);
1867 			return;
1868 		}
1869 	}
1870 
1871 	/*
1872 	 * Compute mbuf offset to TCP data segment.
1873 	 */
1874 	hdroptlen = toff + off;
1875 
1876 	/*
1877 	 * Calculate amount of space in receive window,
1878 	 * and then do TCP input processing.
1879 	 * Receive window is amount of space in rcv queue,
1880 	 * but not less than advertised window.
1881 	 */
1882 	{ int win;
1883 
1884 	win = sbspace(&so->so_rcv);
1885 	if (win < 0)
1886 		win = 0;
1887 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1888 	}
1889 
1890 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1891 	tp->rfbuf_ts = 0;
1892 	tp->rfbuf_cnt = 0;
1893 
1894 	switch (tp->t_state) {
1895 	/*
1896 	 * If the state is SYN_SENT:
1897 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1898 	 *	if seg contains a RST, then drop the connection.
1899 	 *	if seg does not contain SYN, then drop it.
1900 	 * Otherwise this is an acceptable SYN segment
1901 	 *	initialize tp->rcv_nxt and tp->irs
1902 	 *	if seg contains ack then advance tp->snd_una
1903 	 *	if seg contains a ECE and ECN support is enabled, the stream
1904 	 *	    is ECN capable.
1905 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1906 	 *	arrange for segment to be acked (eventually)
1907 	 *	continue processing rest of data/controls, beginning with URG
1908 	 */
1909 	case TCPS_SYN_SENT:
1910 		if ((tiflags & TH_ACK) &&
1911 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1912 		     SEQ_GT(th->th_ack, tp->snd_max)))
1913 			goto dropwithreset;
1914 		if (tiflags & TH_RST) {
1915 			if (tiflags & TH_ACK)
1916 				tp = tcp_drop(tp, ECONNREFUSED);
1917 			goto drop;
1918 		}
1919 		if ((tiflags & TH_SYN) == 0)
1920 			goto drop;
1921 		if (tiflags & TH_ACK) {
1922 			tp->snd_una = th->th_ack;
1923 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1924 				tp->snd_nxt = tp->snd_una;
1925 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1926 				tp->snd_high = tp->snd_una;
1927 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1928 
1929 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1930 				tp->t_flags |= TF_ECN_PERMIT;
1931 				tcpstat.tcps_ecn_shs++;
1932 			}
1933 
1934 		}
1935 		tp->irs = th->th_seq;
1936 		tcp_rcvseqinit(tp);
1937 		tp->t_flags |= TF_ACKNOW;
1938 		tcp_mss_from_peer(tp, opti.maxseg);
1939 
1940 		/*
1941 		 * Initialize the initial congestion window.  If we
1942 		 * had to retransmit the SYN, we must initialize cwnd
1943 		 * to 1 segment (i.e. the Loss Window).
1944 		 */
1945 		if (tp->t_flags & TF_SYN_REXMT)
1946 			tp->snd_cwnd = tp->t_peermss;
1947 		else {
1948 			int ss = tcp_init_win;
1949 #ifdef INET
1950 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1951 				ss = tcp_init_win_local;
1952 #endif
1953 #ifdef INET6
1954 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1955 				ss = tcp_init_win_local;
1956 #endif
1957 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1958 		}
1959 
1960 		tcp_rmx_rtt(tp);
1961 		if (tiflags & TH_ACK) {
1962 			tcpstat.tcps_connects++;
1963 			soisconnected(so);
1964 			tcp_established(tp);
1965 			/* Do window scaling on this connection? */
1966 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1967 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1968 				tp->snd_scale = tp->requested_s_scale;
1969 				tp->rcv_scale = tp->request_r_scale;
1970 			}
1971 			TCP_REASS_LOCK(tp);
1972 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1973 			TCP_REASS_UNLOCK(tp);
1974 			/*
1975 			 * if we didn't have to retransmit the SYN,
1976 			 * use its rtt as our initial srtt & rtt var.
1977 			 */
1978 			if (tp->t_rtttime)
1979 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1980 		} else
1981 			tp->t_state = TCPS_SYN_RECEIVED;
1982 
1983 		/*
1984 		 * Advance th->th_seq to correspond to first data byte.
1985 		 * If data, trim to stay within window,
1986 		 * dropping FIN if necessary.
1987 		 */
1988 		th->th_seq++;
1989 		if (tlen > tp->rcv_wnd) {
1990 			todrop = tlen - tp->rcv_wnd;
1991 			m_adj(m, -todrop);
1992 			tlen = tp->rcv_wnd;
1993 			tiflags &= ~TH_FIN;
1994 			tcpstat.tcps_rcvpackafterwin++;
1995 			tcpstat.tcps_rcvbyteafterwin += todrop;
1996 		}
1997 		tp->snd_wl1 = th->th_seq - 1;
1998 		tp->rcv_up = th->th_seq;
1999 		goto step6;
2000 
2001 	/*
2002 	 * If the state is SYN_RECEIVED:
2003 	 *	If seg contains an ACK, but not for our SYN, drop the input
2004 	 *	and generate an RST.  See page 36, rfc793
2005 	 */
2006 	case TCPS_SYN_RECEIVED:
2007 		if ((tiflags & TH_ACK) &&
2008 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
2009 		     SEQ_GT(th->th_ack, tp->snd_max)))
2010 			goto dropwithreset;
2011 		break;
2012 	}
2013 
2014 	/*
2015 	 * States other than LISTEN or SYN_SENT.
2016 	 * First check timestamp, if present.
2017 	 * Then check that at least some bytes of segment are within
2018 	 * receive window.  If segment begins before rcv_nxt,
2019 	 * drop leading data (and SYN); if nothing left, just ack.
2020 	 *
2021 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2022 	 * and it's less than ts_recent, drop it.
2023 	 */
2024 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
2025 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
2026 
2027 		/* Check to see if ts_recent is over 24 days old.  */
2028 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
2029 			/*
2030 			 * Invalidate ts_recent.  If this segment updates
2031 			 * ts_recent, the age will be reset later and ts_recent
2032 			 * will get a valid value.  If it does not, setting
2033 			 * ts_recent to zero will at least satisfy the
2034 			 * requirement that zero be placed in the timestamp
2035 			 * echo reply when ts_recent isn't valid.  The
2036 			 * age isn't reset until we get a valid ts_recent
2037 			 * because we don't want out-of-order segments to be
2038 			 * dropped when ts_recent is old.
2039 			 */
2040 			tp->ts_recent = 0;
2041 		} else {
2042 			tcpstat.tcps_rcvduppack++;
2043 			tcpstat.tcps_rcvdupbyte += tlen;
2044 			tcpstat.tcps_pawsdrop++;
2045 			tcp_new_dsack(tp, th->th_seq, tlen);
2046 			goto dropafterack;
2047 		}
2048 	}
2049 
2050 	todrop = tp->rcv_nxt - th->th_seq;
2051 	dupseg = false;
2052 	if (todrop > 0) {
2053 		if (tiflags & TH_SYN) {
2054 			tiflags &= ~TH_SYN;
2055 			th->th_seq++;
2056 			if (th->th_urp > 1)
2057 				th->th_urp--;
2058 			else {
2059 				tiflags &= ~TH_URG;
2060 				th->th_urp = 0;
2061 			}
2062 			todrop--;
2063 		}
2064 		if (todrop > tlen ||
2065 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
2066 			/*
2067 			 * Any valid FIN or RST must be to the left of the
2068 			 * window.  At this point the FIN or RST must be a
2069 			 * duplicate or out of sequence; drop it.
2070 			 */
2071 			if (tiflags & TH_RST)
2072 				goto drop;
2073 			tiflags &= ~(TH_FIN|TH_RST);
2074 			/*
2075 			 * Send an ACK to resynchronize and drop any data.
2076 			 * But keep on processing for RST or ACK.
2077 			 */
2078 			tp->t_flags |= TF_ACKNOW;
2079 			todrop = tlen;
2080 			dupseg = true;
2081 			tcpstat.tcps_rcvdupbyte += todrop;
2082 			tcpstat.tcps_rcvduppack++;
2083 		} else if ((tiflags & TH_RST) &&
2084 			   th->th_seq != tp->last_ack_sent) {
2085 			/*
2086 			 * Test for reset before adjusting the sequence
2087 			 * number for overlapping data.
2088 			 */
2089 			goto dropafterack_ratelim;
2090 		} else {
2091 			tcpstat.tcps_rcvpartduppack++;
2092 			tcpstat.tcps_rcvpartdupbyte += todrop;
2093 		}
2094 		tcp_new_dsack(tp, th->th_seq, todrop);
2095 		hdroptlen += todrop;	/*drop from head afterwards*/
2096 		th->th_seq += todrop;
2097 		tlen -= todrop;
2098 		if (th->th_urp > todrop)
2099 			th->th_urp -= todrop;
2100 		else {
2101 			tiflags &= ~TH_URG;
2102 			th->th_urp = 0;
2103 		}
2104 	}
2105 
2106 	/*
2107 	 * If new data are received on a connection after the
2108 	 * user processes are gone, then RST the other end.
2109 	 */
2110 	if ((so->so_state & SS_NOFDREF) &&
2111 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2112 		tp = tcp_close(tp);
2113 		tcpstat.tcps_rcvafterclose++;
2114 		goto dropwithreset;
2115 	}
2116 
2117 	/*
2118 	 * If segment ends after window, drop trailing data
2119 	 * (and PUSH and FIN); if nothing left, just ACK.
2120 	 */
2121 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2122 	if (todrop > 0) {
2123 		tcpstat.tcps_rcvpackafterwin++;
2124 		if (todrop >= tlen) {
2125 			/*
2126 			 * The segment actually starts after the window.
2127 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2128 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2129 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2130 			 */
2131 			tcpstat.tcps_rcvbyteafterwin += tlen;
2132 			/*
2133 			 * If a new connection request is received
2134 			 * while in TIME_WAIT, drop the old connection
2135 			 * and start over if the sequence numbers
2136 			 * are above the previous ones.
2137 			 *
2138 			 * NOTE: We will checksum the packet again, and
2139 			 * so we need to put the header fields back into
2140 			 * network order!
2141 			 * XXX This kind of sucks, but we don't expect
2142 			 * XXX this to happen very often, so maybe it
2143 			 * XXX doesn't matter so much.
2144 			 */
2145 			if (tiflags & TH_SYN &&
2146 			    tp->t_state == TCPS_TIME_WAIT &&
2147 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2148 				tp = tcp_close(tp);
2149 				TCP_FIELDS_TO_NET(th);
2150 				goto findpcb;
2151 			}
2152 			/*
2153 			 * If window is closed can only take segments at
2154 			 * window edge, and have to drop data and PUSH from
2155 			 * incoming segments.  Continue processing, but
2156 			 * remember to ack.  Otherwise, drop segment
2157 			 * and (if not RST) ack.
2158 			 */
2159 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2160 				tp->t_flags |= TF_ACKNOW;
2161 				tcpstat.tcps_rcvwinprobe++;
2162 			} else
2163 				goto dropafterack;
2164 		} else
2165 			tcpstat.tcps_rcvbyteafterwin += todrop;
2166 		m_adj(m, -todrop);
2167 		tlen -= todrop;
2168 		tiflags &= ~(TH_PUSH|TH_FIN);
2169 	}
2170 
2171 	/*
2172 	 * If last ACK falls within this segment's sequence numbers,
2173 	 * and the timestamp is newer, record it.
2174 	 */
2175 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
2176 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2177 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
2178 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2179 		tp->ts_recent_age = tcp_now;
2180 		tp->ts_recent = opti.ts_val;
2181 	}
2182 
2183 	/*
2184 	 * If the RST bit is set examine the state:
2185 	 *    SYN_RECEIVED STATE:
2186 	 *	If passive open, return to LISTEN state.
2187 	 *	If active open, inform user that connection was refused.
2188 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2189 	 *	Inform user that connection was reset, and close tcb.
2190 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2191 	 *	Close the tcb.
2192 	 */
2193 	if (tiflags & TH_RST) {
2194 		if (th->th_seq != tp->last_ack_sent)
2195 			goto dropafterack_ratelim;
2196 
2197 		switch (tp->t_state) {
2198 		case TCPS_SYN_RECEIVED:
2199 			so->so_error = ECONNREFUSED;
2200 			goto close;
2201 
2202 		case TCPS_ESTABLISHED:
2203 		case TCPS_FIN_WAIT_1:
2204 		case TCPS_FIN_WAIT_2:
2205 		case TCPS_CLOSE_WAIT:
2206 			so->so_error = ECONNRESET;
2207 		close:
2208 			tp->t_state = TCPS_CLOSED;
2209 			tcpstat.tcps_drops++;
2210 			tp = tcp_close(tp);
2211 			goto drop;
2212 
2213 		case TCPS_CLOSING:
2214 		case TCPS_LAST_ACK:
2215 		case TCPS_TIME_WAIT:
2216 			tp = tcp_close(tp);
2217 			goto drop;
2218 		}
2219 	}
2220 
2221 	/*
2222 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2223 	 * we must be in a synchronized state.  RFC791 states (under RST
2224 	 * generation) that any unacceptable segment (an out-of-order SYN
2225 	 * qualifies) received in a synchronized state must elicit only an
2226 	 * empty acknowledgment segment ... and the connection remains in
2227 	 * the same state.
2228 	 */
2229 	if (tiflags & TH_SYN) {
2230 		if (tp->rcv_nxt == th->th_seq) {
2231 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2232 			    TH_ACK);
2233 			if (tcp_saveti)
2234 				m_freem(tcp_saveti);
2235 			return;
2236 		}
2237 
2238 		goto dropafterack_ratelim;
2239 	}
2240 
2241 	/*
2242 	 * If the ACK bit is off we drop the segment and return.
2243 	 */
2244 	if ((tiflags & TH_ACK) == 0) {
2245 		if (tp->t_flags & TF_ACKNOW)
2246 			goto dropafterack;
2247 		else
2248 			goto drop;
2249 	}
2250 
2251 	/*
2252 	 * Ack processing.
2253 	 */
2254 	switch (tp->t_state) {
2255 
2256 	/*
2257 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2258 	 * ESTABLISHED state and continue processing, otherwise
2259 	 * send an RST.
2260 	 */
2261 	case TCPS_SYN_RECEIVED:
2262 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2263 		    SEQ_GT(th->th_ack, tp->snd_max))
2264 			goto dropwithreset;
2265 		tcpstat.tcps_connects++;
2266 		soisconnected(so);
2267 		tcp_established(tp);
2268 		/* Do window scaling? */
2269 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2270 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2271 			tp->snd_scale = tp->requested_s_scale;
2272 			tp->rcv_scale = tp->request_r_scale;
2273 		}
2274 		TCP_REASS_LOCK(tp);
2275 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2276 		TCP_REASS_UNLOCK(tp);
2277 		tp->snd_wl1 = th->th_seq - 1;
2278 		/* fall into ... */
2279 
2280 	/*
2281 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2282 	 * ACKs.  If the ack is in the range
2283 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2284 	 * then advance tp->snd_una to th->th_ack and drop
2285 	 * data from the retransmission queue.  If this ACK reflects
2286 	 * more up to date window information we update our window information.
2287 	 */
2288 	case TCPS_ESTABLISHED:
2289 	case TCPS_FIN_WAIT_1:
2290 	case TCPS_FIN_WAIT_2:
2291 	case TCPS_CLOSE_WAIT:
2292 	case TCPS_CLOSING:
2293 	case TCPS_LAST_ACK:
2294 	case TCPS_TIME_WAIT:
2295 
2296 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2297 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2298 				tcpstat.tcps_rcvdupack++;
2299 				/*
2300 				 * If we have outstanding data (other than
2301 				 * a window probe), this is a completely
2302 				 * duplicate ack (ie, window info didn't
2303 				 * change), the ack is the biggest we've
2304 				 * seen and we've seen exactly our rexmt
2305 				 * threshhold of them, assume a packet
2306 				 * has been dropped and retransmit it.
2307 				 * Kludge snd_nxt & the congestion
2308 				 * window so we send only this one
2309 				 * packet.
2310 				 */
2311 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2312 				    th->th_ack != tp->snd_una)
2313 					tp->t_dupacks = 0;
2314 				else if (tp->t_partialacks < 0 &&
2315 					 ((!TCP_SACK_ENABLED(tp) &&
2316 					 ++tp->t_dupacks == tcprexmtthresh) ||
2317 					 TCP_FACK_FASTRECOV(tp))) {
2318 					/*
2319 					 * Do the fast retransmit, and adjust
2320 					 * congestion control paramenters.
2321 					 */
2322 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2323 						/* False fast retransmit */
2324 						break;
2325 					} else
2326 						goto drop;
2327 				} else if (tp->t_dupacks > tcprexmtthresh) {
2328 					tp->snd_cwnd += tp->t_segsz;
2329 					(void) tcp_output(tp);
2330 					goto drop;
2331 				}
2332 			} else {
2333 				/*
2334 				 * If the ack appears to be very old, only
2335 				 * allow data that is in-sequence.  This
2336 				 * makes it somewhat more difficult to insert
2337 				 * forged data by guessing sequence numbers.
2338 				 * Sent an ack to try to update the send
2339 				 * sequence number on the other side.
2340 				 */
2341 				if (tlen && th->th_seq != tp->rcv_nxt &&
2342 				    SEQ_LT(th->th_ack,
2343 				    tp->snd_una - tp->max_sndwnd))
2344 					goto dropafterack;
2345 			}
2346 			break;
2347 		}
2348 		/*
2349 		 * If the congestion window was inflated to account
2350 		 * for the other side's cached packets, retract it.
2351 		 */
2352 		/* XXX: make SACK have his own congestion control
2353 		 * struct -- rpaulo */
2354 		if (TCP_SACK_ENABLED(tp))
2355 			tcp_sack_newack(tp, th);
2356 		else
2357 			tp->t_congctl->fast_retransmit_newack(tp, th);
2358 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2359 			tcpstat.tcps_rcvacktoomuch++;
2360 			goto dropafterack;
2361 		}
2362 		acked = th->th_ack - tp->snd_una;
2363 		tcpstat.tcps_rcvackpack++;
2364 		tcpstat.tcps_rcvackbyte += acked;
2365 
2366 		/*
2367 		 * If we have a timestamp reply, update smoothed
2368 		 * round trip time.  If no timestamp is present but
2369 		 * transmit timer is running and timed sequence
2370 		 * number was acked, update smoothed round trip time.
2371 		 * Since we now have an rtt measurement, cancel the
2372 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2373 		 * Recompute the initial retransmit timer.
2374 		 */
2375 		if (ts_rtt)
2376 			tcp_xmit_timer(tp, ts_rtt);
2377 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2378 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2379 
2380 		/*
2381 		 * If all outstanding data is acked, stop retransmit
2382 		 * timer and remember to restart (more output or persist).
2383 		 * If there is more data to be acked, restart retransmit
2384 		 * timer, using current (possibly backed-off) value.
2385 		 */
2386 		if (th->th_ack == tp->snd_max) {
2387 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2388 			needoutput = 1;
2389 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2390 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2391 
2392 		/*
2393 		 * New data has been acked, adjust the congestion window.
2394 		 */
2395 		tp->t_congctl->newack(tp, th);
2396 
2397 		ND6_HINT(tp);
2398 		if (acked > so->so_snd.sb_cc) {
2399 			tp->snd_wnd -= so->so_snd.sb_cc;
2400 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2401 			ourfinisacked = 1;
2402 		} else {
2403 			if (acked > (tp->t_lastoff - tp->t_inoff))
2404 				tp->t_lastm = NULL;
2405 			sbdrop(&so->so_snd, acked);
2406 			tp->t_lastoff -= acked;
2407 			tp->snd_wnd -= acked;
2408 			ourfinisacked = 0;
2409 		}
2410 		sowwakeup(so);
2411 
2412 		ICMP_CHECK(tp, th, acked);
2413 
2414 		tp->snd_una = th->th_ack;
2415 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2416 			tp->snd_fack = tp->snd_una;
2417 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2418 			tp->snd_nxt = tp->snd_una;
2419 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2420 			tp->snd_high = tp->snd_una;
2421 
2422 		switch (tp->t_state) {
2423 
2424 		/*
2425 		 * In FIN_WAIT_1 STATE in addition to the processing
2426 		 * for the ESTABLISHED state if our FIN is now acknowledged
2427 		 * then enter FIN_WAIT_2.
2428 		 */
2429 		case TCPS_FIN_WAIT_1:
2430 			if (ourfinisacked) {
2431 				/*
2432 				 * If we can't receive any more
2433 				 * data, then closing user can proceed.
2434 				 * Starting the timer is contrary to the
2435 				 * specification, but if we don't get a FIN
2436 				 * we'll hang forever.
2437 				 */
2438 				if (so->so_state & SS_CANTRCVMORE) {
2439 					soisdisconnected(so);
2440 					if (tp->t_maxidle > 0)
2441 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2442 						    tp->t_maxidle);
2443 				}
2444 				tp->t_state = TCPS_FIN_WAIT_2;
2445 			}
2446 			break;
2447 
2448 	 	/*
2449 		 * In CLOSING STATE in addition to the processing for
2450 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2451 		 * then enter the TIME-WAIT state, otherwise ignore
2452 		 * the segment.
2453 		 */
2454 		case TCPS_CLOSING:
2455 			if (ourfinisacked) {
2456 				tp->t_state = TCPS_TIME_WAIT;
2457 				tcp_canceltimers(tp);
2458 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2459 				soisdisconnected(so);
2460 			}
2461 			break;
2462 
2463 		/*
2464 		 * In LAST_ACK, we may still be waiting for data to drain
2465 		 * and/or to be acked, as well as for the ack of our FIN.
2466 		 * If our FIN is now acknowledged, delete the TCB,
2467 		 * enter the closed state and return.
2468 		 */
2469 		case TCPS_LAST_ACK:
2470 			if (ourfinisacked) {
2471 				tp = tcp_close(tp);
2472 				goto drop;
2473 			}
2474 			break;
2475 
2476 		/*
2477 		 * In TIME_WAIT state the only thing that should arrive
2478 		 * is a retransmission of the remote FIN.  Acknowledge
2479 		 * it and restart the finack timer.
2480 		 */
2481 		case TCPS_TIME_WAIT:
2482 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2483 			goto dropafterack;
2484 		}
2485 	}
2486 
2487 step6:
2488 	/*
2489 	 * Update window information.
2490 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2491 	 */
2492 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2493 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2494 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2495 		/* keep track of pure window updates */
2496 		if (tlen == 0 &&
2497 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2498 			tcpstat.tcps_rcvwinupd++;
2499 		tp->snd_wnd = tiwin;
2500 		tp->snd_wl1 = th->th_seq;
2501 		tp->snd_wl2 = th->th_ack;
2502 		if (tp->snd_wnd > tp->max_sndwnd)
2503 			tp->max_sndwnd = tp->snd_wnd;
2504 		needoutput = 1;
2505 	}
2506 
2507 	/*
2508 	 * Process segments with URG.
2509 	 */
2510 	if ((tiflags & TH_URG) && th->th_urp &&
2511 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2512 		/*
2513 		 * This is a kludge, but if we receive and accept
2514 		 * random urgent pointers, we'll crash in
2515 		 * soreceive.  It's hard to imagine someone
2516 		 * actually wanting to send this much urgent data.
2517 		 */
2518 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2519 			th->th_urp = 0;			/* XXX */
2520 			tiflags &= ~TH_URG;		/* XXX */
2521 			goto dodata;			/* XXX */
2522 		}
2523 		/*
2524 		 * If this segment advances the known urgent pointer,
2525 		 * then mark the data stream.  This should not happen
2526 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2527 		 * a FIN has been received from the remote side.
2528 		 * In these states we ignore the URG.
2529 		 *
2530 		 * According to RFC961 (Assigned Protocols),
2531 		 * the urgent pointer points to the last octet
2532 		 * of urgent data.  We continue, however,
2533 		 * to consider it to indicate the first octet
2534 		 * of data past the urgent section as the original
2535 		 * spec states (in one of two places).
2536 		 */
2537 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2538 			tp->rcv_up = th->th_seq + th->th_urp;
2539 			so->so_oobmark = so->so_rcv.sb_cc +
2540 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2541 			if (so->so_oobmark == 0)
2542 				so->so_state |= SS_RCVATMARK;
2543 			sohasoutofband(so);
2544 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2545 		}
2546 		/*
2547 		 * Remove out of band data so doesn't get presented to user.
2548 		 * This can happen independent of advancing the URG pointer,
2549 		 * but if two URG's are pending at once, some out-of-band
2550 		 * data may creep in... ick.
2551 		 */
2552 		if (th->th_urp <= (u_int16_t) tlen
2553 #ifdef SO_OOBINLINE
2554 		     && (so->so_options & SO_OOBINLINE) == 0
2555 #endif
2556 		     )
2557 			tcp_pulloutofband(so, th, m, hdroptlen);
2558 	} else
2559 		/*
2560 		 * If no out of band data is expected,
2561 		 * pull receive urgent pointer along
2562 		 * with the receive window.
2563 		 */
2564 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2565 			tp->rcv_up = tp->rcv_nxt;
2566 dodata:							/* XXX */
2567 
2568 	/*
2569 	 * Process the segment text, merging it into the TCP sequencing queue,
2570 	 * and arranging for acknowledgement of receipt if necessary.
2571 	 * This process logically involves adjusting tp->rcv_wnd as data
2572 	 * is presented to the user (this happens in tcp_usrreq.c,
2573 	 * case PRU_RCVD).  If a FIN has already been received on this
2574 	 * connection then we just ignore the text.
2575 	 */
2576 	if ((tlen || (tiflags & TH_FIN)) &&
2577 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2578 		/*
2579 		 * Insert segment ti into reassembly queue of tcp with
2580 		 * control block tp.  Return TH_FIN if reassembly now includes
2581 		 * a segment with FIN.  The macro form does the common case
2582 		 * inline (segment is the next to be received on an
2583 		 * established connection, and the queue is empty),
2584 		 * avoiding linkage into and removal from the queue and
2585 		 * repetition of various conversions.
2586 		 * Set DELACK for segments received in order, but ack
2587 		 * immediately when segments are out of order
2588 		 * (so fast retransmit can work).
2589 		 */
2590 		/* NOTE: this was TCP_REASS() macro, but used only once */
2591 		TCP_REASS_LOCK(tp);
2592 		if (th->th_seq == tp->rcv_nxt &&
2593 		    TAILQ_FIRST(&tp->segq) == NULL &&
2594 		    tp->t_state == TCPS_ESTABLISHED) {
2595 			TCP_SETUP_ACK(tp, th);
2596 			tp->rcv_nxt += tlen;
2597 			tiflags = th->th_flags & TH_FIN;
2598 			tcpstat.tcps_rcvpack++;
2599 			tcpstat.tcps_rcvbyte += tlen;
2600 			ND6_HINT(tp);
2601 			if (so->so_state & SS_CANTRCVMORE)
2602 				m_freem(m);
2603 			else {
2604 				m_adj(m, hdroptlen);
2605 				sbappendstream(&(so)->so_rcv, m);
2606 			}
2607 			sorwakeup(so);
2608 		} else {
2609 			m_adj(m, hdroptlen);
2610 			tiflags = tcp_reass(tp, th, m, &tlen);
2611 			tp->t_flags |= TF_ACKNOW;
2612 		}
2613 		TCP_REASS_UNLOCK(tp);
2614 
2615 		/*
2616 		 * Note the amount of data that peer has sent into
2617 		 * our window, in order to estimate the sender's
2618 		 * buffer size.
2619 		 */
2620 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2621 	} else {
2622 		m_freem(m);
2623 		m = NULL;
2624 		tiflags &= ~TH_FIN;
2625 	}
2626 
2627 	/*
2628 	 * If FIN is received ACK the FIN and let the user know
2629 	 * that the connection is closing.  Ignore a FIN received before
2630 	 * the connection is fully established.
2631 	 */
2632 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2633 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2634 			socantrcvmore(so);
2635 			tp->t_flags |= TF_ACKNOW;
2636 			tp->rcv_nxt++;
2637 		}
2638 		switch (tp->t_state) {
2639 
2640 	 	/*
2641 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2642 		 */
2643 		case TCPS_ESTABLISHED:
2644 			tp->t_state = TCPS_CLOSE_WAIT;
2645 			break;
2646 
2647 	 	/*
2648 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2649 		 * enter the CLOSING state.
2650 		 */
2651 		case TCPS_FIN_WAIT_1:
2652 			tp->t_state = TCPS_CLOSING;
2653 			break;
2654 
2655 	 	/*
2656 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2657 		 * starting the time-wait timer, turning off the other
2658 		 * standard timers.
2659 		 */
2660 		case TCPS_FIN_WAIT_2:
2661 			tp->t_state = TCPS_TIME_WAIT;
2662 			tcp_canceltimers(tp);
2663 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2664 			soisdisconnected(so);
2665 			break;
2666 
2667 		/*
2668 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2669 		 */
2670 		case TCPS_TIME_WAIT:
2671 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2672 			break;
2673 		}
2674 	}
2675 #ifdef TCP_DEBUG
2676 	if (so->so_options & SO_DEBUG)
2677 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2678 #endif
2679 
2680 	/*
2681 	 * Return any desired output.
2682 	 */
2683 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2684 		(void) tcp_output(tp);
2685 	}
2686 	if (tcp_saveti)
2687 		m_freem(tcp_saveti);
2688 	return;
2689 
2690 badsyn:
2691 	/*
2692 	 * Received a bad SYN.  Increment counters and dropwithreset.
2693 	 */
2694 	tcpstat.tcps_badsyn++;
2695 	tp = NULL;
2696 	goto dropwithreset;
2697 
2698 dropafterack:
2699 	/*
2700 	 * Generate an ACK dropping incoming segment if it occupies
2701 	 * sequence space, where the ACK reflects our state.
2702 	 */
2703 	if (tiflags & TH_RST)
2704 		goto drop;
2705 	goto dropafterack2;
2706 
2707 dropafterack_ratelim:
2708 	/*
2709 	 * We may want to rate-limit ACKs against SYN/RST attack.
2710 	 */
2711 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2712 	    tcp_ackdrop_ppslim) == 0) {
2713 		/* XXX stat */
2714 		goto drop;
2715 	}
2716 	/* ...fall into dropafterack2... */
2717 
2718 dropafterack2:
2719 	m_freem(m);
2720 	tp->t_flags |= TF_ACKNOW;
2721 	(void) tcp_output(tp);
2722 	if (tcp_saveti)
2723 		m_freem(tcp_saveti);
2724 	return;
2725 
2726 dropwithreset_ratelim:
2727 	/*
2728 	 * We may want to rate-limit RSTs in certain situations,
2729 	 * particularly if we are sending an RST in response to
2730 	 * an attempt to connect to or otherwise communicate with
2731 	 * a port for which we have no socket.
2732 	 */
2733 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2734 	    tcp_rst_ppslim) == 0) {
2735 		/* XXX stat */
2736 		goto drop;
2737 	}
2738 	/* ...fall into dropwithreset... */
2739 
2740 dropwithreset:
2741 	/*
2742 	 * Generate a RST, dropping incoming segment.
2743 	 * Make ACK acceptable to originator of segment.
2744 	 */
2745 	if (tiflags & TH_RST)
2746 		goto drop;
2747 
2748 	switch (af) {
2749 #ifdef INET6
2750 	case AF_INET6:
2751 		/* For following calls to tcp_respond */
2752 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2753 			goto drop;
2754 		break;
2755 #endif /* INET6 */
2756 	case AF_INET:
2757 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2758 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2759 			goto drop;
2760 	}
2761 
2762 	if (tiflags & TH_ACK)
2763 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2764 	else {
2765 		if (tiflags & TH_SYN)
2766 			tlen++;
2767 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2768 		    TH_RST|TH_ACK);
2769 	}
2770 	if (tcp_saveti)
2771 		m_freem(tcp_saveti);
2772 	return;
2773 
2774 badcsum:
2775 drop:
2776 	/*
2777 	 * Drop space held by incoming segment and return.
2778 	 */
2779 	if (tp) {
2780 		if (tp->t_inpcb)
2781 			so = tp->t_inpcb->inp_socket;
2782 #ifdef INET6
2783 		else if (tp->t_in6pcb)
2784 			so = tp->t_in6pcb->in6p_socket;
2785 #endif
2786 		else
2787 			so = NULL;
2788 #ifdef TCP_DEBUG
2789 		if (so && (so->so_options & SO_DEBUG) != 0)
2790 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2791 #endif
2792 	}
2793 	if (tcp_saveti)
2794 		m_freem(tcp_saveti);
2795 	m_freem(m);
2796 	return;
2797 }
2798 
2799 #ifdef TCP_SIGNATURE
2800 int
2801 tcp_signature_apply(void *fstate, void *data, u_int len)
2802 {
2803 
2804 	MD5Update(fstate, (u_char *)data, len);
2805 	return (0);
2806 }
2807 
2808 struct secasvar *
2809 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2810 {
2811 	struct secasvar *sav;
2812 #ifdef FAST_IPSEC
2813 	union sockaddr_union dst;
2814 #endif
2815 	struct ip *ip;
2816 	struct ip6_hdr *ip6;
2817 
2818 	ip = mtod(m, struct ip *);
2819 	switch (ip->ip_v) {
2820 	case 4:
2821 		ip = mtod(m, struct ip *);
2822 		ip6 = NULL;
2823 		break;
2824 	case 6:
2825 		ip = NULL;
2826 		ip6 = mtod(m, struct ip6_hdr *);
2827 		break;
2828 	default:
2829 		return (NULL);
2830 	}
2831 
2832 #ifdef FAST_IPSEC
2833 	/* Extract the destination from the IP header in the mbuf. */
2834 	bzero(&dst, sizeof(union sockaddr_union));
2835 	if (ip !=NULL) {
2836 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2837 		dst.sa.sa_family = AF_INET;
2838 		dst.sin.sin_addr = ip->ip_dst;
2839 	} else {
2840 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2841 		dst.sa.sa_family = AF_INET6;
2842 		dst.sin6.sin6_addr = ip6->ip6_dst;
2843 	}
2844 
2845 	/*
2846 	 * Look up an SADB entry which matches the address of the peer.
2847 	 */
2848 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2849 #else
2850 	if (ip)
2851 		sav = key_allocsa(AF_INET, (void *)&ip->ip_src,
2852 		    (void *)&ip->ip_dst, IPPROTO_TCP,
2853 		    htonl(TCP_SIG_SPI), 0, 0);
2854 	else
2855 		sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src,
2856 		    (void *)&ip6->ip6_dst, IPPROTO_TCP,
2857 		    htonl(TCP_SIG_SPI), 0, 0);
2858 #endif
2859 
2860 	return (sav);	/* freesav must be performed by caller */
2861 }
2862 
2863 int
2864 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2865     struct secasvar *sav, char *sig)
2866 {
2867 	MD5_CTX ctx;
2868 	struct ip *ip;
2869 	struct ipovly *ipovly;
2870 	struct ip6_hdr *ip6;
2871 	struct ippseudo ippseudo;
2872 	struct ip6_hdr_pseudo ip6pseudo;
2873 	struct tcphdr th0;
2874 	int l, tcphdrlen;
2875 
2876 	if (sav == NULL)
2877 		return (-1);
2878 
2879 	tcphdrlen = th->th_off * 4;
2880 
2881 	switch (mtod(m, struct ip *)->ip_v) {
2882 	case 4:
2883 		ip = mtod(m, struct ip *);
2884 		ip6 = NULL;
2885 		break;
2886 	case 6:
2887 		ip = NULL;
2888 		ip6 = mtod(m, struct ip6_hdr *);
2889 		break;
2890 	default:
2891 		return (-1);
2892 	}
2893 
2894 	MD5Init(&ctx);
2895 
2896 	if (ip) {
2897 		memset(&ippseudo, 0, sizeof(ippseudo));
2898 		ipovly = (struct ipovly *)ip;
2899 		ippseudo.ippseudo_src = ipovly->ih_src;
2900 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2901 		ippseudo.ippseudo_pad = 0;
2902 		ippseudo.ippseudo_p = IPPROTO_TCP;
2903 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2904 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2905 	} else {
2906 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2907 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2908 		in6_clearscope(&ip6pseudo.ip6ph_src);
2909 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2910 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2911 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2912 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2913 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2914 	}
2915 
2916 	th0 = *th;
2917 	th0.th_sum = 0;
2918 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2919 
2920 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2921 	if (l > 0)
2922 		m_apply(m, thoff + tcphdrlen,
2923 		    m->m_pkthdr.len - thoff - tcphdrlen,
2924 		    tcp_signature_apply, &ctx);
2925 
2926 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2927 	MD5Final(sig, &ctx);
2928 
2929 	return (0);
2930 }
2931 #endif
2932 
2933 static int
2934 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt,
2935     struct tcphdr *th,
2936     struct mbuf *m, int toff, struct tcp_opt_info *oi)
2937 {
2938 	u_int16_t mss;
2939 	int opt, optlen = 0;
2940 #ifdef TCP_SIGNATURE
2941 	void *sigp = NULL;
2942 	char sigbuf[TCP_SIGLEN];
2943 	struct secasvar *sav = NULL;
2944 #endif
2945 
2946 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2947 		opt = cp[0];
2948 		if (opt == TCPOPT_EOL)
2949 			break;
2950 		if (opt == TCPOPT_NOP)
2951 			optlen = 1;
2952 		else {
2953 			if (cnt < 2)
2954 				break;
2955 			optlen = cp[1];
2956 			if (optlen < 2 || optlen > cnt)
2957 				break;
2958 		}
2959 		switch (opt) {
2960 
2961 		default:
2962 			continue;
2963 
2964 		case TCPOPT_MAXSEG:
2965 			if (optlen != TCPOLEN_MAXSEG)
2966 				continue;
2967 			if (!(th->th_flags & TH_SYN))
2968 				continue;
2969 			if (TCPS_HAVERCVDSYN(tp->t_state))
2970 				continue;
2971 			bcopy(cp + 2, &mss, sizeof(mss));
2972 			oi->maxseg = ntohs(mss);
2973 			break;
2974 
2975 		case TCPOPT_WINDOW:
2976 			if (optlen != TCPOLEN_WINDOW)
2977 				continue;
2978 			if (!(th->th_flags & TH_SYN))
2979 				continue;
2980 			if (TCPS_HAVERCVDSYN(tp->t_state))
2981 				continue;
2982 			tp->t_flags |= TF_RCVD_SCALE;
2983 			tp->requested_s_scale = cp[2];
2984 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2985 #if 0	/*XXX*/
2986 				char *p;
2987 
2988 				if (ip)
2989 					p = ntohl(ip->ip_src);
2990 #ifdef INET6
2991 				else if (ip6)
2992 					p = ip6_sprintf(&ip6->ip6_src);
2993 #endif
2994 				else
2995 					p = "(unknown)";
2996 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2997 				    "assuming %d\n",
2998 				    tp->requested_s_scale, p,
2999 				    TCP_MAX_WINSHIFT);
3000 #else
3001 				log(LOG_ERR, "TCP: invalid wscale %d, "
3002 				    "assuming %d\n",
3003 				    tp->requested_s_scale,
3004 				    TCP_MAX_WINSHIFT);
3005 #endif
3006 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
3007 			}
3008 			break;
3009 
3010 		case TCPOPT_TIMESTAMP:
3011 			if (optlen != TCPOLEN_TIMESTAMP)
3012 				continue;
3013 			oi->ts_present = 1;
3014 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
3015 			NTOHL(oi->ts_val);
3016 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
3017 			NTOHL(oi->ts_ecr);
3018 
3019 			if (!(th->th_flags & TH_SYN))
3020 				continue;
3021 			if (TCPS_HAVERCVDSYN(tp->t_state))
3022 				continue;
3023 			/*
3024 			 * A timestamp received in a SYN makes
3025 			 * it ok to send timestamp requests and replies.
3026 			 */
3027 			tp->t_flags |= TF_RCVD_TSTMP;
3028 			tp->ts_recent = oi->ts_val;
3029 			tp->ts_recent_age = tcp_now;
3030                         break;
3031 
3032 		case TCPOPT_SACK_PERMITTED:
3033 			if (optlen != TCPOLEN_SACK_PERMITTED)
3034 				continue;
3035 			if (!(th->th_flags & TH_SYN))
3036 				continue;
3037 			if (TCPS_HAVERCVDSYN(tp->t_state))
3038 				continue;
3039 			if (tcp_do_sack) {
3040 				tp->t_flags |= TF_SACK_PERMIT;
3041 				tp->t_flags |= TF_WILL_SACK;
3042 			}
3043 			break;
3044 
3045 		case TCPOPT_SACK:
3046 			tcp_sack_option(tp, th, cp, optlen);
3047 			break;
3048 #ifdef TCP_SIGNATURE
3049 		case TCPOPT_SIGNATURE:
3050 			if (optlen != TCPOLEN_SIGNATURE)
3051 				continue;
3052 			if (sigp && bcmp(sigp, cp + 2, TCP_SIGLEN))
3053 				return (-1);
3054 
3055 			sigp = sigbuf;
3056 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
3057 			tp->t_flags |= TF_SIGNATURE;
3058 			break;
3059 #endif
3060 		}
3061 	}
3062 
3063 #ifdef TCP_SIGNATURE
3064 	if (tp->t_flags & TF_SIGNATURE) {
3065 
3066 		sav = tcp_signature_getsav(m, th);
3067 
3068 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3069 			return (-1);
3070 	}
3071 
3072 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3073 		if (sav == NULL)
3074 			return (-1);
3075 #ifdef FAST_IPSEC
3076 		KEY_FREESAV(&sav);
3077 #else
3078 		key_freesav(sav);
3079 #endif
3080 		return (-1);
3081 	}
3082 
3083 	if (sigp) {
3084 		char sig[TCP_SIGLEN];
3085 
3086 		TCP_FIELDS_TO_NET(th);
3087 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3088 			TCP_FIELDS_TO_HOST(th);
3089 			if (sav == NULL)
3090 				return (-1);
3091 #ifdef FAST_IPSEC
3092 			KEY_FREESAV(&sav);
3093 #else
3094 			key_freesav(sav);
3095 #endif
3096 			return (-1);
3097 		}
3098 		TCP_FIELDS_TO_HOST(th);
3099 
3100 		if (bcmp(sig, sigp, TCP_SIGLEN)) {
3101 			tcpstat.tcps_badsig++;
3102 			if (sav == NULL)
3103 				return (-1);
3104 #ifdef FAST_IPSEC
3105 			KEY_FREESAV(&sav);
3106 #else
3107 			key_freesav(sav);
3108 #endif
3109 			return (-1);
3110 		} else
3111 			tcpstat.tcps_goodsig++;
3112 
3113 		key_sa_recordxfer(sav, m);
3114 #ifdef FAST_IPSEC
3115 		KEY_FREESAV(&sav);
3116 #else
3117 		key_freesav(sav);
3118 #endif
3119 	}
3120 #endif
3121 
3122 	return (0);
3123 }
3124 
3125 /*
3126  * Pull out of band byte out of a segment so
3127  * it doesn't appear in the user's data queue.
3128  * It is still reflected in the segment length for
3129  * sequencing purposes.
3130  */
3131 void
3132 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3133     struct mbuf *m, int off)
3134 {
3135 	int cnt = off + th->th_urp - 1;
3136 
3137 	while (cnt >= 0) {
3138 		if (m->m_len > cnt) {
3139 			char *cp = mtod(m, char *) + cnt;
3140 			struct tcpcb *tp = sototcpcb(so);
3141 
3142 			tp->t_iobc = *cp;
3143 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3144 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3145 			m->m_len--;
3146 			return;
3147 		}
3148 		cnt -= m->m_len;
3149 		m = m->m_next;
3150 		if (m == 0)
3151 			break;
3152 	}
3153 	panic("tcp_pulloutofband");
3154 }
3155 
3156 /*
3157  * Collect new round-trip time estimate
3158  * and update averages and current timeout.
3159  */
3160 void
3161 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3162 {
3163 	int32_t delta;
3164 
3165 	tcpstat.tcps_rttupdated++;
3166 	if (tp->t_srtt != 0) {
3167 		/*
3168 		 * srtt is stored as fixed point with 3 bits after the
3169 		 * binary point (i.e., scaled by 8).  The following magic
3170 		 * is equivalent to the smoothing algorithm in rfc793 with
3171 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3172 		 * point).  Adjust rtt to origin 0.
3173 		 */
3174 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3175 		if ((tp->t_srtt += delta) <= 0)
3176 			tp->t_srtt = 1 << 2;
3177 		/*
3178 		 * We accumulate a smoothed rtt variance (actually, a
3179 		 * smoothed mean difference), then set the retransmit
3180 		 * timer to smoothed rtt + 4 times the smoothed variance.
3181 		 * rttvar is stored as fixed point with 2 bits after the
3182 		 * binary point (scaled by 4).  The following is
3183 		 * equivalent to rfc793 smoothing with an alpha of .75
3184 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3185 		 * rfc793's wired-in beta.
3186 		 */
3187 		if (delta < 0)
3188 			delta = -delta;
3189 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3190 		if ((tp->t_rttvar += delta) <= 0)
3191 			tp->t_rttvar = 1 << 2;
3192 	} else {
3193 		/*
3194 		 * No rtt measurement yet - use the unsmoothed rtt.
3195 		 * Set the variance to half the rtt (so our first
3196 		 * retransmit happens at 3*rtt).
3197 		 */
3198 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3199 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3200 	}
3201 	tp->t_rtttime = 0;
3202 	tp->t_rxtshift = 0;
3203 
3204 	/*
3205 	 * the retransmit should happen at rtt + 4 * rttvar.
3206 	 * Because of the way we do the smoothing, srtt and rttvar
3207 	 * will each average +1/2 tick of bias.  When we compute
3208 	 * the retransmit timer, we want 1/2 tick of rounding and
3209 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3210 	 * firing of the timer.  The bias will give us exactly the
3211 	 * 1.5 tick we need.  But, because the bias is
3212 	 * statistical, we have to test that we don't drop below
3213 	 * the minimum feasible timer (which is 2 ticks).
3214 	 */
3215 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3216 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3217 
3218 	/*
3219 	 * We received an ack for a packet that wasn't retransmitted;
3220 	 * it is probably safe to discard any error indications we've
3221 	 * received recently.  This isn't quite right, but close enough
3222 	 * for now (a route might have failed after we sent a segment,
3223 	 * and the return path might not be symmetrical).
3224 	 */
3225 	tp->t_softerror = 0;
3226 }
3227 
3228 
3229 /*
3230  * TCP compressed state engine.  Currently used to hold compressed
3231  * state for SYN_RECEIVED.
3232  */
3233 
3234 u_long	syn_cache_count;
3235 u_int32_t syn_hash1, syn_hash2;
3236 
3237 #define SYN_HASH(sa, sp, dp) \
3238 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3239 				     ((u_int32_t)(sp)))^syn_hash2)))
3240 #ifndef INET6
3241 #define	SYN_HASHALL(hash, src, dst) \
3242 do {									\
3243 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3244 		((const struct sockaddr_in *)(src))->sin_port,		\
3245 		((const struct sockaddr_in *)(dst))->sin_port);		\
3246 } while (/*CONSTCOND*/ 0)
3247 #else
3248 #define SYN_HASH6(sa, sp, dp) \
3249 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3250 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3251 	 & 0x7fffffff)
3252 
3253 #define SYN_HASHALL(hash, src, dst) \
3254 do {									\
3255 	switch ((src)->sa_family) {					\
3256 	case AF_INET:							\
3257 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3258 			((const struct sockaddr_in *)(src))->sin_port,	\
3259 			((const struct sockaddr_in *)(dst))->sin_port);	\
3260 		break;							\
3261 	case AF_INET6:							\
3262 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3263 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3264 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3265 		break;							\
3266 	default:							\
3267 		hash = 0;						\
3268 	}								\
3269 } while (/*CONSTCOND*/0)
3270 #endif /* INET6 */
3271 
3272 POOL_INIT(syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, "synpl", NULL,
3273     IPL_SOFTNET);
3274 
3275 /*
3276  * We don't estimate RTT with SYNs, so each packet starts with the default
3277  * RTT and each timer step has a fixed timeout value.
3278  */
3279 #define	SYN_CACHE_TIMER_ARM(sc)						\
3280 do {									\
3281 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3282 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3283 	    TCPTV_REXMTMAX);						\
3284 	callout_reset(&(sc)->sc_timer,					\
3285 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3286 } while (/*CONSTCOND*/0)
3287 
3288 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3289 
3290 static inline void
3291 syn_cache_rm(struct syn_cache *sc)
3292 {
3293 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3294 	    sc, sc_bucketq);
3295 	sc->sc_tp = NULL;
3296 	LIST_REMOVE(sc, sc_tpq);
3297 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3298 	callout_stop(&sc->sc_timer);
3299 	syn_cache_count--;
3300 }
3301 
3302 static inline void
3303 syn_cache_put(struct syn_cache *sc)
3304 {
3305 	if (sc->sc_ipopts)
3306 		(void) m_free(sc->sc_ipopts);
3307 	rtcache_free(&sc->sc_route);
3308 	if (callout_invoking(&sc->sc_timer))
3309 		sc->sc_flags |= SCF_DEAD;
3310 	else {
3311 		callout_destroy(&sc->sc_timer);
3312 		pool_put(&syn_cache_pool, sc);
3313 	}
3314 }
3315 
3316 void
3317 syn_cache_init(void)
3318 {
3319 	int i;
3320 
3321 	/* Initialize the hash buckets. */
3322 	for (i = 0; i < tcp_syn_cache_size; i++)
3323 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3324 }
3325 
3326 void
3327 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3328 {
3329 	struct syn_cache_head *scp;
3330 	struct syn_cache *sc2;
3331 	int s;
3332 
3333 	/*
3334 	 * If there are no entries in the hash table, reinitialize
3335 	 * the hash secrets.
3336 	 */
3337 	if (syn_cache_count == 0) {
3338 		syn_hash1 = arc4random();
3339 		syn_hash2 = arc4random();
3340 	}
3341 
3342 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3343 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3344 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3345 
3346 	/*
3347 	 * Make sure that we don't overflow the per-bucket
3348 	 * limit or the total cache size limit.
3349 	 */
3350 	s = splsoftnet();
3351 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3352 		tcpstat.tcps_sc_bucketoverflow++;
3353 		/*
3354 		 * The bucket is full.  Toss the oldest element in the
3355 		 * bucket.  This will be the first entry in the bucket.
3356 		 */
3357 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3358 #ifdef DIAGNOSTIC
3359 		/*
3360 		 * This should never happen; we should always find an
3361 		 * entry in our bucket.
3362 		 */
3363 		if (sc2 == NULL)
3364 			panic("syn_cache_insert: bucketoverflow: impossible");
3365 #endif
3366 		syn_cache_rm(sc2);
3367 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3368 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3369 		struct syn_cache_head *scp2, *sce;
3370 
3371 		tcpstat.tcps_sc_overflowed++;
3372 		/*
3373 		 * The cache is full.  Toss the oldest entry in the
3374 		 * first non-empty bucket we can find.
3375 		 *
3376 		 * XXX We would really like to toss the oldest
3377 		 * entry in the cache, but we hope that this
3378 		 * condition doesn't happen very often.
3379 		 */
3380 		scp2 = scp;
3381 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3382 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3383 			for (++scp2; scp2 != scp; scp2++) {
3384 				if (scp2 >= sce)
3385 					scp2 = &tcp_syn_cache[0];
3386 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3387 					break;
3388 			}
3389 #ifdef DIAGNOSTIC
3390 			/*
3391 			 * This should never happen; we should always find a
3392 			 * non-empty bucket.
3393 			 */
3394 			if (scp2 == scp)
3395 				panic("syn_cache_insert: cacheoverflow: "
3396 				    "impossible");
3397 #endif
3398 		}
3399 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3400 		syn_cache_rm(sc2);
3401 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3402 	}
3403 
3404 	/*
3405 	 * Initialize the entry's timer.
3406 	 */
3407 	sc->sc_rxttot = 0;
3408 	sc->sc_rxtshift = 0;
3409 	SYN_CACHE_TIMER_ARM(sc);
3410 
3411 	/* Link it from tcpcb entry */
3412 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3413 
3414 	/* Put it into the bucket. */
3415 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3416 	scp->sch_length++;
3417 	syn_cache_count++;
3418 
3419 	tcpstat.tcps_sc_added++;
3420 	splx(s);
3421 }
3422 
3423 /*
3424  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3425  * If we have retransmitted an entry the maximum number of times, expire
3426  * that entry.
3427  */
3428 void
3429 syn_cache_timer(void *arg)
3430 {
3431 	struct syn_cache *sc = arg;
3432 	int s;
3433 
3434 	s = splsoftnet();
3435 	callout_ack(&sc->sc_timer);
3436 
3437 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3438 		tcpstat.tcps_sc_delayed_free++;
3439 		callout_destroy(&sc->sc_timer);
3440 		pool_put(&syn_cache_pool, sc);
3441 		splx(s);
3442 		return;
3443 	}
3444 
3445 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3446 		/* Drop it -- too many retransmissions. */
3447 		goto dropit;
3448 	}
3449 
3450 	/*
3451 	 * Compute the total amount of time this entry has
3452 	 * been on a queue.  If this entry has been on longer
3453 	 * than the keep alive timer would allow, expire it.
3454 	 */
3455 	sc->sc_rxttot += sc->sc_rxtcur;
3456 	if (sc->sc_rxttot >= tcp_keepinit)
3457 		goto dropit;
3458 
3459 	tcpstat.tcps_sc_retransmitted++;
3460 	(void) syn_cache_respond(sc, NULL);
3461 
3462 	/* Advance the timer back-off. */
3463 	sc->sc_rxtshift++;
3464 	SYN_CACHE_TIMER_ARM(sc);
3465 
3466 	splx(s);
3467 	return;
3468 
3469  dropit:
3470 	tcpstat.tcps_sc_timed_out++;
3471 	syn_cache_rm(sc);
3472 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3473 	splx(s);
3474 }
3475 
3476 /*
3477  * Remove syn cache created by the specified tcb entry,
3478  * because this does not make sense to keep them
3479  * (if there's no tcb entry, syn cache entry will never be used)
3480  */
3481 void
3482 syn_cache_cleanup(struct tcpcb *tp)
3483 {
3484 	struct syn_cache *sc, *nsc;
3485 	int s;
3486 
3487 	s = splsoftnet();
3488 
3489 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3490 		nsc = LIST_NEXT(sc, sc_tpq);
3491 
3492 #ifdef DIAGNOSTIC
3493 		if (sc->sc_tp != tp)
3494 			panic("invalid sc_tp in syn_cache_cleanup");
3495 #endif
3496 		syn_cache_rm(sc);
3497 		syn_cache_put(sc);	/* calls pool_put but see spl above */
3498 	}
3499 	/* just for safety */
3500 	LIST_INIT(&tp->t_sc);
3501 
3502 	splx(s);
3503 }
3504 
3505 /*
3506  * Find an entry in the syn cache.
3507  */
3508 struct syn_cache *
3509 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3510     struct syn_cache_head **headp)
3511 {
3512 	struct syn_cache *sc;
3513 	struct syn_cache_head *scp;
3514 	u_int32_t hash;
3515 	int s;
3516 
3517 	SYN_HASHALL(hash, src, dst);
3518 
3519 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3520 	*headp = scp;
3521 	s = splsoftnet();
3522 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3523 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3524 		if (sc->sc_hash != hash)
3525 			continue;
3526 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3527 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3528 			splx(s);
3529 			return (sc);
3530 		}
3531 	}
3532 	splx(s);
3533 	return (NULL);
3534 }
3535 
3536 /*
3537  * This function gets called when we receive an ACK for a
3538  * socket in the LISTEN state.  We look up the connection
3539  * in the syn cache, and if its there, we pull it out of
3540  * the cache and turn it into a full-blown connection in
3541  * the SYN-RECEIVED state.
3542  *
3543  * The return values may not be immediately obvious, and their effects
3544  * can be subtle, so here they are:
3545  *
3546  *	NULL	SYN was not found in cache; caller should drop the
3547  *		packet and send an RST.
3548  *
3549  *	-1	We were unable to create the new connection, and are
3550  *		aborting it.  An ACK,RST is being sent to the peer
3551  *		(unless we got screwey sequence numbners; see below),
3552  *		because the 3-way handshake has been completed.  Caller
3553  *		should not free the mbuf, since we may be using it.  If
3554  *		we are not, we will free it.
3555  *
3556  *	Otherwise, the return value is a pointer to the new socket
3557  *	associated with the connection.
3558  */
3559 struct socket *
3560 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3561     struct tcphdr *th, unsigned int hlen, unsigned int tlen,
3562     struct socket *so, struct mbuf *m)
3563 {
3564 	struct syn_cache *sc;
3565 	struct syn_cache_head *scp;
3566 	struct inpcb *inp = NULL;
3567 #ifdef INET6
3568 	struct in6pcb *in6p = NULL;
3569 #endif
3570 	struct tcpcb *tp = 0;
3571 	struct mbuf *am;
3572 	int s;
3573 	struct socket *oso;
3574 
3575 	s = splsoftnet();
3576 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3577 		splx(s);
3578 		return (NULL);
3579 	}
3580 
3581 	/*
3582 	 * Verify the sequence and ack numbers.  Try getting the correct
3583 	 * response again.
3584 	 */
3585 	if ((th->th_ack != sc->sc_iss + 1) ||
3586 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3587 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3588 		(void) syn_cache_respond(sc, m);
3589 		splx(s);
3590 		return ((struct socket *)(-1));
3591 	}
3592 
3593 	/* Remove this cache entry */
3594 	syn_cache_rm(sc);
3595 	splx(s);
3596 
3597 	/*
3598 	 * Ok, create the full blown connection, and set things up
3599 	 * as they would have been set up if we had created the
3600 	 * connection when the SYN arrived.  If we can't create
3601 	 * the connection, abort it.
3602 	 */
3603 	/*
3604 	 * inp still has the OLD in_pcb stuff, set the
3605 	 * v6-related flags on the new guy, too.   This is
3606 	 * done particularly for the case where an AF_INET6
3607 	 * socket is bound only to a port, and a v4 connection
3608 	 * comes in on that port.
3609 	 * we also copy the flowinfo from the original pcb
3610 	 * to the new one.
3611 	 */
3612 	oso = so;
3613 	so = sonewconn(so, SS_ISCONNECTED);
3614 	if (so == NULL)
3615 		goto resetandabort;
3616 
3617 	switch (so->so_proto->pr_domain->dom_family) {
3618 #ifdef INET
3619 	case AF_INET:
3620 		inp = sotoinpcb(so);
3621 		break;
3622 #endif
3623 #ifdef INET6
3624 	case AF_INET6:
3625 		in6p = sotoin6pcb(so);
3626 		break;
3627 #endif
3628 	}
3629 	switch (src->sa_family) {
3630 #ifdef INET
3631 	case AF_INET:
3632 		if (inp) {
3633 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3634 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3635 			inp->inp_options = ip_srcroute();
3636 			in_pcbstate(inp, INP_BOUND);
3637 			if (inp->inp_options == NULL) {
3638 				inp->inp_options = sc->sc_ipopts;
3639 				sc->sc_ipopts = NULL;
3640 			}
3641 		}
3642 #ifdef INET6
3643 		else if (in6p) {
3644 			/* IPv4 packet to AF_INET6 socket */
3645 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3646 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3647 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3648 				&in6p->in6p_laddr.s6_addr32[3],
3649 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3650 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3651 			in6totcpcb(in6p)->t_family = AF_INET;
3652 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3653 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3654 			else
3655 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3656 			in6_pcbstate(in6p, IN6P_BOUND);
3657 		}
3658 #endif
3659 		break;
3660 #endif
3661 #ifdef INET6
3662 	case AF_INET6:
3663 		if (in6p) {
3664 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3665 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3666 			in6_pcbstate(in6p, IN6P_BOUND);
3667 		}
3668 		break;
3669 #endif
3670 	}
3671 #ifdef INET6
3672 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3673 		struct in6pcb *oin6p = sotoin6pcb(oso);
3674 		/* inherit socket options from the listening socket */
3675 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3676 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3677 			m_freem(in6p->in6p_options);
3678 			in6p->in6p_options = 0;
3679 		}
3680 		ip6_savecontrol(in6p, &in6p->in6p_options,
3681 			mtod(m, struct ip6_hdr *), m);
3682 	}
3683 #endif
3684 
3685 #if defined(IPSEC) || defined(FAST_IPSEC)
3686 	/*
3687 	 * we make a copy of policy, instead of sharing the policy,
3688 	 * for better behavior in terms of SA lookup and dead SA removal.
3689 	 */
3690 	if (inp) {
3691 		/* copy old policy into new socket's */
3692 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3693 			printf("tcp_input: could not copy policy\n");
3694 	}
3695 #ifdef INET6
3696 	else if (in6p) {
3697 		/* copy old policy into new socket's */
3698 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3699 		    in6p->in6p_sp))
3700 			printf("tcp_input: could not copy policy\n");
3701 	}
3702 #endif
3703 #endif
3704 
3705 	/*
3706 	 * Give the new socket our cached route reference.
3707 	 */
3708 	if (inp) {
3709 		rtcache_copy(&inp->inp_route, &sc->sc_route);
3710 		rtcache_free(&sc->sc_route);
3711 	}
3712 #ifdef INET6
3713 	else {
3714 		rtcache_copy(&in6p->in6p_route, &sc->sc_route);
3715 		rtcache_free(&sc->sc_route);
3716 	}
3717 #endif
3718 
3719 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3720 	if (am == NULL)
3721 		goto resetandabort;
3722 	MCLAIM(am, &tcp_mowner);
3723 	am->m_len = src->sa_len;
3724 	bcopy(src, mtod(am, void *), src->sa_len);
3725 	if (inp) {
3726 		if (in_pcbconnect(inp, am, NULL)) {
3727 			(void) m_free(am);
3728 			goto resetandabort;
3729 		}
3730 	}
3731 #ifdef INET6
3732 	else if (in6p) {
3733 		if (src->sa_family == AF_INET) {
3734 			/* IPv4 packet to AF_INET6 socket */
3735 			struct sockaddr_in6 *sin6;
3736 			sin6 = mtod(am, struct sockaddr_in6 *);
3737 			am->m_len = sizeof(*sin6);
3738 			bzero(sin6, sizeof(*sin6));
3739 			sin6->sin6_family = AF_INET6;
3740 			sin6->sin6_len = sizeof(*sin6);
3741 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3742 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3743 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3744 				&sin6->sin6_addr.s6_addr32[3],
3745 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3746 		}
3747 		if (in6_pcbconnect(in6p, am, NULL)) {
3748 			(void) m_free(am);
3749 			goto resetandabort;
3750 		}
3751 	}
3752 #endif
3753 	else {
3754 		(void) m_free(am);
3755 		goto resetandabort;
3756 	}
3757 	(void) m_free(am);
3758 
3759 	if (inp)
3760 		tp = intotcpcb(inp);
3761 #ifdef INET6
3762 	else if (in6p)
3763 		tp = in6totcpcb(in6p);
3764 #endif
3765 	else
3766 		tp = NULL;
3767 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3768 	if (sc->sc_request_r_scale != 15) {
3769 		tp->requested_s_scale = sc->sc_requested_s_scale;
3770 		tp->request_r_scale = sc->sc_request_r_scale;
3771 		tp->snd_scale = sc->sc_requested_s_scale;
3772 		tp->rcv_scale = sc->sc_request_r_scale;
3773 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3774 	}
3775 	if (sc->sc_flags & SCF_TIMESTAMP)
3776 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3777 	tp->ts_timebase = sc->sc_timebase;
3778 
3779 	tp->t_template = tcp_template(tp);
3780 	if (tp->t_template == 0) {
3781 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3782 		so = NULL;
3783 		m_freem(m);
3784 		goto abort;
3785 	}
3786 
3787 	tp->iss = sc->sc_iss;
3788 	tp->irs = sc->sc_irs;
3789 	tcp_sendseqinit(tp);
3790 	tcp_rcvseqinit(tp);
3791 	tp->t_state = TCPS_SYN_RECEIVED;
3792 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit);
3793 	tcpstat.tcps_accepts++;
3794 
3795 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3796 		tp->t_flags |= TF_WILL_SACK;
3797 
3798 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3799 		tp->t_flags |= TF_ECN_PERMIT;
3800 
3801 #ifdef TCP_SIGNATURE
3802 	if (sc->sc_flags & SCF_SIGNATURE)
3803 		tp->t_flags |= TF_SIGNATURE;
3804 #endif
3805 
3806 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3807 	tp->t_ourmss = sc->sc_ourmaxseg;
3808 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3809 
3810 	/*
3811 	 * Initialize the initial congestion window.  If we
3812 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3813 	 * to 1 segment (i.e. the Loss Window).
3814 	 */
3815 	if (sc->sc_rxtshift)
3816 		tp->snd_cwnd = tp->t_peermss;
3817 	else {
3818 		int ss = tcp_init_win;
3819 #ifdef INET
3820 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3821 			ss = tcp_init_win_local;
3822 #endif
3823 #ifdef INET6
3824 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3825 			ss = tcp_init_win_local;
3826 #endif
3827 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3828 	}
3829 
3830 	tcp_rmx_rtt(tp);
3831 	tp->snd_wl1 = sc->sc_irs;
3832 	tp->rcv_up = sc->sc_irs + 1;
3833 
3834 	/*
3835 	 * This is what whould have happened in tcp_output() when
3836 	 * the SYN,ACK was sent.
3837 	 */
3838 	tp->snd_up = tp->snd_una;
3839 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3840 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3841 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3842 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3843 	tp->last_ack_sent = tp->rcv_nxt;
3844 	tp->t_partialacks = -1;
3845 	tp->t_dupacks = 0;
3846 
3847 	tcpstat.tcps_sc_completed++;
3848 	s = splsoftnet();
3849 	syn_cache_put(sc);
3850 	splx(s);
3851 	return (so);
3852 
3853 resetandabort:
3854 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3855 abort:
3856 	if (so != NULL)
3857 		(void) soabort(so);
3858 	s = splsoftnet();
3859 	syn_cache_put(sc);
3860 	splx(s);
3861 	tcpstat.tcps_sc_aborted++;
3862 	return ((struct socket *)(-1));
3863 }
3864 
3865 /*
3866  * This function is called when we get a RST for a
3867  * non-existent connection, so that we can see if the
3868  * connection is in the syn cache.  If it is, zap it.
3869  */
3870 
3871 void
3872 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3873 {
3874 	struct syn_cache *sc;
3875 	struct syn_cache_head *scp;
3876 	int s = splsoftnet();
3877 
3878 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3879 		splx(s);
3880 		return;
3881 	}
3882 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3883 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3884 		splx(s);
3885 		return;
3886 	}
3887 	syn_cache_rm(sc);
3888 	tcpstat.tcps_sc_reset++;
3889 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3890 	splx(s);
3891 }
3892 
3893 void
3894 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3895     struct tcphdr *th)
3896 {
3897 	struct syn_cache *sc;
3898 	struct syn_cache_head *scp;
3899 	int s;
3900 
3901 	s = splsoftnet();
3902 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3903 		splx(s);
3904 		return;
3905 	}
3906 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3907 	if (ntohl (th->th_seq) != sc->sc_iss) {
3908 		splx(s);
3909 		return;
3910 	}
3911 
3912 	/*
3913 	 * If we've retransmitted 3 times and this is our second error,
3914 	 * we remove the entry.  Otherwise, we allow it to continue on.
3915 	 * This prevents us from incorrectly nuking an entry during a
3916 	 * spurious network outage.
3917 	 *
3918 	 * See tcp_notify().
3919 	 */
3920 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3921 		sc->sc_flags |= SCF_UNREACH;
3922 		splx(s);
3923 		return;
3924 	}
3925 
3926 	syn_cache_rm(sc);
3927 	tcpstat.tcps_sc_unreach++;
3928 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3929 	splx(s);
3930 }
3931 
3932 /*
3933  * Given a LISTEN socket and an inbound SYN request, add
3934  * this to the syn cache, and send back a segment:
3935  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3936  * to the source.
3937  *
3938  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3939  * Doing so would require that we hold onto the data and deliver it
3940  * to the application.  However, if we are the target of a SYN-flood
3941  * DoS attack, an attacker could send data which would eventually
3942  * consume all available buffer space if it were ACKed.  By not ACKing
3943  * the data, we avoid this DoS scenario.
3944  */
3945 
3946 int
3947 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3948     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
3949     int optlen, struct tcp_opt_info *oi)
3950 {
3951 	struct tcpcb tb, *tp;
3952 	long win;
3953 	struct syn_cache *sc;
3954 	struct syn_cache_head *scp;
3955 	struct mbuf *ipopts;
3956 	struct tcp_opt_info opti;
3957 	int s;
3958 
3959 	tp = sototcpcb(so);
3960 
3961 	bzero(&opti, sizeof(opti));
3962 
3963 	/*
3964 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3965 	 *
3966 	 * Note this check is performed in tcp_input() very early on.
3967 	 */
3968 
3969 	/*
3970 	 * Initialize some local state.
3971 	 */
3972 	win = sbspace(&so->so_rcv);
3973 	if (win > TCP_MAXWIN)
3974 		win = TCP_MAXWIN;
3975 
3976 	switch (src->sa_family) {
3977 #ifdef INET
3978 	case AF_INET:
3979 		/*
3980 		 * Remember the IP options, if any.
3981 		 */
3982 		ipopts = ip_srcroute();
3983 		break;
3984 #endif
3985 	default:
3986 		ipopts = NULL;
3987 	}
3988 
3989 #ifdef TCP_SIGNATURE
3990 	if (optp || (tp->t_flags & TF_SIGNATURE))
3991 #else
3992 	if (optp)
3993 #endif
3994 	{
3995 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3996 #ifdef TCP_SIGNATURE
3997 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
3998 #endif
3999 		tb.t_state = TCPS_LISTEN;
4000 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
4001 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
4002 			return (0);
4003 	} else
4004 		tb.t_flags = 0;
4005 
4006 	/*
4007 	 * See if we already have an entry for this connection.
4008 	 * If we do, resend the SYN,ACK.  We do not count this
4009 	 * as a retransmission (XXX though maybe we should).
4010 	 */
4011 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4012 		tcpstat.tcps_sc_dupesyn++;
4013 		if (ipopts) {
4014 			/*
4015 			 * If we were remembering a previous source route,
4016 			 * forget it and use the new one we've been given.
4017 			 */
4018 			if (sc->sc_ipopts)
4019 				(void) m_free(sc->sc_ipopts);
4020 			sc->sc_ipopts = ipopts;
4021 		}
4022 		sc->sc_timestamp = tb.ts_recent;
4023 		if (syn_cache_respond(sc, m) == 0) {
4024 			tcpstat.tcps_sndacks++;
4025 			tcpstat.tcps_sndtotal++;
4026 		}
4027 		return (1);
4028 	}
4029 
4030 	s = splsoftnet();
4031 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4032 	splx(s);
4033 	if (sc == NULL) {
4034 		if (ipopts)
4035 			(void) m_free(ipopts);
4036 		return (0);
4037 	}
4038 
4039 	/*
4040 	 * Fill in the cache, and put the necessary IP and TCP
4041 	 * options into the reply.
4042 	 */
4043 	bzero(sc, sizeof(struct syn_cache));
4044 	callout_init(&sc->sc_timer, 0);
4045 	bcopy(src, &sc->sc_src, src->sa_len);
4046 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4047 	sc->sc_flags = 0;
4048 	sc->sc_ipopts = ipopts;
4049 	sc->sc_irs = th->th_seq;
4050 	switch (src->sa_family) {
4051 #ifdef INET
4052 	case AF_INET:
4053 	    {
4054 		struct sockaddr_in *srcin = (void *) src;
4055 		struct sockaddr_in *dstin = (void *) dst;
4056 
4057 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
4058 		    &srcin->sin_addr, dstin->sin_port,
4059 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
4060 		break;
4061 	    }
4062 #endif /* INET */
4063 #ifdef INET6
4064 	case AF_INET6:
4065 	    {
4066 		struct sockaddr_in6 *srcin6 = (void *) src;
4067 		struct sockaddr_in6 *dstin6 = (void *) dst;
4068 
4069 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
4070 		    &srcin6->sin6_addr, dstin6->sin6_port,
4071 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
4072 		break;
4073 	    }
4074 #endif /* INET6 */
4075 	}
4076 	sc->sc_peermaxseg = oi->maxseg;
4077 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4078 						m->m_pkthdr.rcvif : NULL,
4079 						sc->sc_src.sa.sa_family);
4080 	sc->sc_win = win;
4081 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
4082 	sc->sc_timestamp = tb.ts_recent;
4083 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4084 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4085 		sc->sc_flags |= SCF_TIMESTAMP;
4086 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4087 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4088 		sc->sc_requested_s_scale = tb.requested_s_scale;
4089 		sc->sc_request_r_scale = 0;
4090 		/*
4091 		 * Pick the smallest possible scaling factor that
4092 		 * will still allow us to scale up to sb_max.
4093 		 *
4094 		 * We do this because there are broken firewalls that
4095 		 * will corrupt the window scale option, leading to
4096 		 * the other endpoint believing that our advertised
4097 		 * window is unscaled.  At scale factors larger than
4098 		 * 5 the unscaled window will drop below 1500 bytes,
4099 		 * leading to serious problems when traversing these
4100 		 * broken firewalls.
4101 		 *
4102 		 * With the default sbmax of 256K, a scale factor
4103 		 * of 3 will be chosen by this algorithm.  Those who
4104 		 * choose a larger sbmax should watch out
4105 		 * for the compatiblity problems mentioned above.
4106 		 *
4107 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4108 		 * or <SYN,ACK>) segment itself is never scaled.
4109 		 */
4110 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4111 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4112 			sc->sc_request_r_scale++;
4113 	} else {
4114 		sc->sc_requested_s_scale = 15;
4115 		sc->sc_request_r_scale = 15;
4116 	}
4117 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4118 		sc->sc_flags |= SCF_SACK_PERMIT;
4119 
4120 	/*
4121 	 * ECN setup packet recieved.
4122 	 */
4123 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4124 		sc->sc_flags |= SCF_ECN_PERMIT;
4125 
4126 #ifdef TCP_SIGNATURE
4127 	if (tb.t_flags & TF_SIGNATURE)
4128 		sc->sc_flags |= SCF_SIGNATURE;
4129 #endif
4130 	sc->sc_tp = tp;
4131 	if (syn_cache_respond(sc, m) == 0) {
4132 		syn_cache_insert(sc, tp);
4133 		tcpstat.tcps_sndacks++;
4134 		tcpstat.tcps_sndtotal++;
4135 	} else {
4136 		s = splsoftnet();
4137 		syn_cache_put(sc);
4138 		splx(s);
4139 		tcpstat.tcps_sc_dropped++;
4140 	}
4141 	return (1);
4142 }
4143 
4144 int
4145 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4146 {
4147 	struct route *ro;
4148 	u_int8_t *optp;
4149 	int optlen, error;
4150 	u_int16_t tlen;
4151 	struct ip *ip = NULL;
4152 #ifdef INET6
4153 	struct ip6_hdr *ip6 = NULL;
4154 #endif
4155 	struct tcpcb *tp = NULL;
4156 	struct tcphdr *th;
4157 	u_int hlen;
4158 	struct socket *so;
4159 
4160 	ro = &sc->sc_route;
4161 	switch (sc->sc_src.sa.sa_family) {
4162 	case AF_INET:
4163 		hlen = sizeof(struct ip);
4164 		break;
4165 #ifdef INET6
4166 	case AF_INET6:
4167 		hlen = sizeof(struct ip6_hdr);
4168 		break;
4169 #endif
4170 	default:
4171 		if (m)
4172 			m_freem(m);
4173 		return (EAFNOSUPPORT);
4174 	}
4175 
4176 	/* Compute the size of the TCP options. */
4177 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4178 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4179 #ifdef TCP_SIGNATURE
4180 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4181 #endif
4182 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4183 
4184 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4185 
4186 	/*
4187 	 * Create the IP+TCP header from scratch.
4188 	 */
4189 	if (m)
4190 		m_freem(m);
4191 #ifdef DIAGNOSTIC
4192 	if (max_linkhdr + tlen > MCLBYTES)
4193 		return (ENOBUFS);
4194 #endif
4195 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4196 	if (m && tlen > MHLEN) {
4197 		MCLGET(m, M_DONTWAIT);
4198 		if ((m->m_flags & M_EXT) == 0) {
4199 			m_freem(m);
4200 			m = NULL;
4201 		}
4202 	}
4203 	if (m == NULL)
4204 		return (ENOBUFS);
4205 	MCLAIM(m, &tcp_tx_mowner);
4206 
4207 	/* Fixup the mbuf. */
4208 	m->m_data += max_linkhdr;
4209 	m->m_len = m->m_pkthdr.len = tlen;
4210 	if (sc->sc_tp) {
4211 		tp = sc->sc_tp;
4212 		if (tp->t_inpcb)
4213 			so = tp->t_inpcb->inp_socket;
4214 #ifdef INET6
4215 		else if (tp->t_in6pcb)
4216 			so = tp->t_in6pcb->in6p_socket;
4217 #endif
4218 		else
4219 			so = NULL;
4220 	} else
4221 		so = NULL;
4222 	m->m_pkthdr.rcvif = NULL;
4223 	memset(mtod(m, u_char *), 0, tlen);
4224 
4225 	switch (sc->sc_src.sa.sa_family) {
4226 	case AF_INET:
4227 		ip = mtod(m, struct ip *);
4228 		ip->ip_v = 4;
4229 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4230 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4231 		ip->ip_p = IPPROTO_TCP;
4232 		th = (struct tcphdr *)(ip + 1);
4233 		th->th_dport = sc->sc_src.sin.sin_port;
4234 		th->th_sport = sc->sc_dst.sin.sin_port;
4235 		break;
4236 #ifdef INET6
4237 	case AF_INET6:
4238 		ip6 = mtod(m, struct ip6_hdr *);
4239 		ip6->ip6_vfc = IPV6_VERSION;
4240 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4241 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4242 		ip6->ip6_nxt = IPPROTO_TCP;
4243 		/* ip6_plen will be updated in ip6_output() */
4244 		th = (struct tcphdr *)(ip6 + 1);
4245 		th->th_dport = sc->sc_src.sin6.sin6_port;
4246 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4247 		break;
4248 #endif
4249 	default:
4250 		th = NULL;
4251 	}
4252 
4253 	th->th_seq = htonl(sc->sc_iss);
4254 	th->th_ack = htonl(sc->sc_irs + 1);
4255 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4256 	th->th_flags = TH_SYN|TH_ACK;
4257 	th->th_win = htons(sc->sc_win);
4258 	/* th_sum already 0 */
4259 	/* th_urp already 0 */
4260 
4261 	/* Tack on the TCP options. */
4262 	optp = (u_int8_t *)(th + 1);
4263 	*optp++ = TCPOPT_MAXSEG;
4264 	*optp++ = 4;
4265 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4266 	*optp++ = sc->sc_ourmaxseg & 0xff;
4267 
4268 	if (sc->sc_request_r_scale != 15) {
4269 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4270 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4271 		    sc->sc_request_r_scale);
4272 		optp += 4;
4273 	}
4274 
4275 	if (sc->sc_flags & SCF_TIMESTAMP) {
4276 		u_int32_t *lp = (u_int32_t *)(optp);
4277 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4278 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4279 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4280 		*lp   = htonl(sc->sc_timestamp);
4281 		optp += TCPOLEN_TSTAMP_APPA;
4282 	}
4283 
4284 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4285 		u_int8_t *p = optp;
4286 
4287 		/* Let the peer know that we will SACK. */
4288 		p[0] = TCPOPT_SACK_PERMITTED;
4289 		p[1] = 2;
4290 		p[2] = TCPOPT_NOP;
4291 		p[3] = TCPOPT_NOP;
4292 		optp += 4;
4293 	}
4294 
4295 	/*
4296 	 * Send ECN SYN-ACK setup packet.
4297 	 * Routes can be asymetric, so, even if we receive a packet
4298 	 * with ECE and CWR set, we must not assume no one will block
4299 	 * the ECE packet we are about to send.
4300 	 */
4301 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4302 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4303 		th->th_flags |= TH_ECE;
4304 		tcpstat.tcps_ecn_shs++;
4305 
4306 		/*
4307 		 * draft-ietf-tcpm-ecnsyn-00.txt
4308 		 *
4309 		 * "[...] a TCP node MAY respond to an ECN-setup
4310 		 * SYN packet by setting ECT in the responding
4311 		 * ECN-setup SYN/ACK packet, indicating to routers
4312 		 * that the SYN/ACK packet is ECN-Capable.
4313 		 * This allows a congested router along the path
4314 		 * to mark the packet instead of dropping the
4315 		 * packet as an indication of congestion."
4316 		 *
4317 		 * "[...] There can be a great benefit in setting
4318 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4319 		 * Congestion is  most likely to occur in
4320 		 * the server-to-client direction.  As a result,
4321 		 * setting an ECN-capable codepoint in SYN/ACK
4322 		 * packets can reduce the occurence of three-second
4323 		 * retransmit timeouts resulting from the drop
4324 		 * of SYN/ACK packets."
4325 		 *
4326 		 * Page 4 and 6, January 2006.
4327 		 */
4328 
4329 		switch (sc->sc_src.sa.sa_family) {
4330 #ifdef INET
4331 		case AF_INET:
4332 			ip->ip_tos |= IPTOS_ECN_ECT0;
4333 			break;
4334 #endif
4335 #ifdef INET6
4336 		case AF_INET6:
4337 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4338 			break;
4339 #endif
4340 		}
4341 		tcpstat.tcps_ecn_ect++;
4342 	}
4343 
4344 #ifdef TCP_SIGNATURE
4345 	if (sc->sc_flags & SCF_SIGNATURE) {
4346 		struct secasvar *sav;
4347 		u_int8_t *sigp;
4348 
4349 		sav = tcp_signature_getsav(m, th);
4350 
4351 		if (sav == NULL) {
4352 			if (m)
4353 				m_freem(m);
4354 			return (EPERM);
4355 		}
4356 
4357 		*optp++ = TCPOPT_SIGNATURE;
4358 		*optp++ = TCPOLEN_SIGNATURE;
4359 		sigp = optp;
4360 		bzero(optp, TCP_SIGLEN);
4361 		optp += TCP_SIGLEN;
4362 		*optp++ = TCPOPT_NOP;
4363 		*optp++ = TCPOPT_EOL;
4364 
4365 		(void)tcp_signature(m, th, hlen, sav, sigp);
4366 
4367 		key_sa_recordxfer(sav, m);
4368 #ifdef FAST_IPSEC
4369 		KEY_FREESAV(&sav);
4370 #else
4371 		key_freesav(sav);
4372 #endif
4373 	}
4374 #endif
4375 
4376 	/* Compute the packet's checksum. */
4377 	switch (sc->sc_src.sa.sa_family) {
4378 	case AF_INET:
4379 		ip->ip_len = htons(tlen - hlen);
4380 		th->th_sum = 0;
4381 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4382 		break;
4383 #ifdef INET6
4384 	case AF_INET6:
4385 		ip6->ip6_plen = htons(tlen - hlen);
4386 		th->th_sum = 0;
4387 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4388 		break;
4389 #endif
4390 	}
4391 
4392 	/*
4393 	 * Fill in some straggling IP bits.  Note the stack expects
4394 	 * ip_len to be in host order, for convenience.
4395 	 */
4396 	switch (sc->sc_src.sa.sa_family) {
4397 #ifdef INET
4398 	case AF_INET:
4399 		ip->ip_len = htons(tlen);
4400 		ip->ip_ttl = ip_defttl;
4401 		/* XXX tos? */
4402 		break;
4403 #endif
4404 #ifdef INET6
4405 	case AF_INET6:
4406 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4407 		ip6->ip6_vfc |= IPV6_VERSION;
4408 		ip6->ip6_plen = htons(tlen - hlen);
4409 		/* ip6_hlim will be initialized afterwards */
4410 		/* XXX flowlabel? */
4411 		break;
4412 #endif
4413 	}
4414 
4415 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4416 	tp = sc->sc_tp;
4417 
4418 	switch (sc->sc_src.sa.sa_family) {
4419 #ifdef INET
4420 	case AF_INET:
4421 		error = ip_output(m, sc->sc_ipopts, ro,
4422 		    (ip_mtudisc ? IP_MTUDISC : 0),
4423 		    (struct ip_moptions *)NULL, so);
4424 		break;
4425 #endif
4426 #ifdef INET6
4427 	case AF_INET6:
4428 		ip6->ip6_hlim = in6_selecthlim(NULL,
4429 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4430 
4431 		error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL);
4432 		break;
4433 #endif
4434 	default:
4435 		error = EAFNOSUPPORT;
4436 		break;
4437 	}
4438 	return (error);
4439 }
4440