xref: /netbsd-src/sys/netinet/tcp_input.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: tcp_input.c,v 1.188 2004/01/02 12:01:39 itojun 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 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  *
80  * Redistribution and use in source and binary forms, with or without
81  * modification, are permitted provided that the following conditions
82  * are met:
83  * 1. Redistributions of source code must retain the above copyright
84  *    notice, this list of conditions and the following disclaimer.
85  * 2. Redistributions in binary form must reproduce the above copyright
86  *    notice, this list of conditions and the following disclaimer in the
87  *    documentation and/or other materials provided with the distribution.
88  * 3. All advertising materials mentioning features or use of this software
89  *    must display the following acknowledgement:
90  *	This product includes software developed by the NetBSD
91  *	Foundation, Inc. and its contributors.
92  * 4. Neither the name of The NetBSD Foundation nor the names of its
93  *    contributors may be used to endorse or promote products derived
94  *    from this software without specific prior written permission.
95  *
96  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
97  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
98  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
99  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
100  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
101  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
102  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
103  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
104  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
105  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
106  * POSSIBILITY OF SUCH DAMAGE.
107  */
108 
109 /*
110  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
111  *	The Regents of the University of California.  All rights reserved.
112  *
113  * Redistribution and use in source and binary forms, with or without
114  * modification, are permitted provided that the following conditions
115  * are met:
116  * 1. Redistributions of source code must retain the above copyright
117  *    notice, this list of conditions and the following disclaimer.
118  * 2. Redistributions in binary form must reproduce the above copyright
119  *    notice, this list of conditions and the following disclaimer in the
120  *    documentation and/or other materials provided with the distribution.
121  * 3. Neither the name of the University nor the names of its contributors
122  *    may be used to endorse or promote products derived from this software
123  *    without specific prior written permission.
124  *
125  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
126  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
127  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
128  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
129  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
130  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
131  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
132  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
133  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
134  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
135  * SUCH DAMAGE.
136  *
137  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
138  */
139 
140 /*
141  *	TODO list for SYN cache stuff:
142  *
143  *	Find room for a "state" field, which is needed to keep a
144  *	compressed state for TIME_WAIT TCBs.  It's been noted already
145  *	that this is fairly important for very high-volume web and
146  *	mail servers, which use a large number of short-lived
147  *	connections.
148  */
149 
150 #include <sys/cdefs.h>
151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.188 2004/01/02 12:01:39 itojun Exp $");
152 
153 #include "opt_inet.h"
154 #include "opt_ipsec.h"
155 #include "opt_inet_csum.h"
156 #include "opt_tcp_debug.h"
157 
158 #include <sys/param.h>
159 #include <sys/systm.h>
160 #include <sys/malloc.h>
161 #include <sys/mbuf.h>
162 #include <sys/protosw.h>
163 #include <sys/socket.h>
164 #include <sys/socketvar.h>
165 #include <sys/errno.h>
166 #include <sys/syslog.h>
167 #include <sys/pool.h>
168 #include <sys/domain.h>
169 #include <sys/kernel.h>
170 
171 #include <net/if.h>
172 #include <net/route.h>
173 #include <net/if_types.h>
174 
175 #include <netinet/in.h>
176 #include <netinet/in_systm.h>
177 #include <netinet/ip.h>
178 #include <netinet/in_pcb.h>
179 #include <netinet/in_var.h>
180 #include <netinet/ip_var.h>
181 
182 #ifdef INET6
183 #ifndef INET
184 #include <netinet/in.h>
185 #endif
186 #include <netinet/ip6.h>
187 #include <netinet6/ip6_var.h>
188 #include <netinet6/in6_pcb.h>
189 #include <netinet6/ip6_var.h>
190 #include <netinet6/in6_var.h>
191 #include <netinet/icmp6.h>
192 #include <netinet6/nd6.h>
193 #endif
194 
195 #ifndef INET6
196 /* always need ip6.h for IP6_EXTHDR_GET */
197 #include <netinet/ip6.h>
198 #endif
199 
200 #include <netinet/tcp.h>
201 #include <netinet/tcp_fsm.h>
202 #include <netinet/tcp_seq.h>
203 #include <netinet/tcp_timer.h>
204 #include <netinet/tcp_var.h>
205 #include <netinet/tcpip.h>
206 #include <netinet/tcp_debug.h>
207 
208 #include <machine/stdarg.h>
209 
210 #ifdef IPSEC
211 #include <netinet6/ipsec.h>
212 #include <netkey/key.h>
213 #endif /*IPSEC*/
214 #ifdef INET6
215 #include "faith.h"
216 #if defined(NFAITH) && NFAITH > 0
217 #include <net/if_faith.h>
218 #endif
219 #endif	/* IPSEC */
220 
221 #ifdef FAST_IPSEC
222 #include <netipsec/ipsec.h>
223 #include <netipsec/key.h>
224 #ifdef INET6
225 #include <netipsec/ipsec6.h>
226 #endif
227 #endif	/* FAST_IPSEC*/
228 
229 
230 int	tcprexmtthresh = 3;
231 int	tcp_log_refused;
232 
233 static int tcp_rst_ppslim_count = 0;
234 static struct timeval tcp_rst_ppslim_last;
235 
236 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
237 
238 /* for modulo comparisons of timestamps */
239 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
240 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
241 
242 /*
243  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
244  */
245 #ifdef INET6
246 #define ND6_HINT(tp) \
247 do { \
248 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \
249 	 && tp->t_in6pcb->in6p_route.ro_rt) { \
250 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
251 	} \
252 } while (/*CONSTCOND*/ 0)
253 #else
254 #define ND6_HINT(tp)
255 #endif
256 
257 /*
258  * Macro to compute ACK transmission behavior.  Delay the ACK unless
259  * we have already delayed an ACK (must send an ACK every two segments).
260  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
261  * option is enabled.
262  */
263 #define	TCP_SETUP_ACK(tp, th) \
264 do { \
265 	if ((tp)->t_flags & TF_DELACK || \
266 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
267 		tp->t_flags |= TF_ACKNOW; \
268 	else \
269 		TCP_SET_DELACK(tp); \
270 } while (/*CONSTCOND*/ 0)
271 
272 /*
273  * Convert TCP protocol fields to host order for easier processing.
274  */
275 #define	TCP_FIELDS_TO_HOST(th)						\
276 do {									\
277 	NTOHL((th)->th_seq);						\
278 	NTOHL((th)->th_ack);						\
279 	NTOHS((th)->th_win);						\
280 	NTOHS((th)->th_urp);						\
281 } while (/*CONSTCOND*/ 0)
282 
283 /*
284  * ... and reverse the above.
285  */
286 #define	TCP_FIELDS_TO_NET(th)						\
287 do {									\
288 	HTONL((th)->th_seq);						\
289 	HTONL((th)->th_ack);						\
290 	HTONS((th)->th_win);						\
291 	HTONS((th)->th_urp);						\
292 } while (/*CONSTCOND*/ 0)
293 
294 #ifdef TCP_CSUM_COUNTERS
295 #include <sys/device.h>
296 
297 extern struct evcnt tcp_hwcsum_ok;
298 extern struct evcnt tcp_hwcsum_bad;
299 extern struct evcnt tcp_hwcsum_data;
300 extern struct evcnt tcp_swcsum;
301 
302 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
303 
304 #else
305 
306 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
307 
308 #endif /* TCP_CSUM_COUNTERS */
309 
310 #ifdef TCP_REASS_COUNTERS
311 #include <sys/device.h>
312 
313 extern struct evcnt tcp_reass_;
314 extern struct evcnt tcp_reass_empty;
315 extern struct evcnt tcp_reass_iteration[8];
316 extern struct evcnt tcp_reass_prependfirst;
317 extern struct evcnt tcp_reass_prepend;
318 extern struct evcnt tcp_reass_insert;
319 extern struct evcnt tcp_reass_inserttail;
320 extern struct evcnt tcp_reass_append;
321 extern struct evcnt tcp_reass_appendtail;
322 extern struct evcnt tcp_reass_overlaptail;
323 extern struct evcnt tcp_reass_overlapfront;
324 extern struct evcnt tcp_reass_segdup;
325 extern struct evcnt tcp_reass_fragdup;
326 
327 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
328 
329 #else
330 
331 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
332 
333 #endif /* TCP_REASS_COUNTERS */
334 
335 #ifdef INET
336 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *));
337 #endif
338 #ifdef INET6
339 static void tcp6_log_refused
340     __P((const struct ip6_hdr *, const struct tcphdr *));
341 #endif
342 
343 int
344 tcp_reass(tp, th, m, tlen)
345 	struct tcpcb *tp;
346 	struct tcphdr *th;
347 	struct mbuf *m;
348 	int *tlen;
349 {
350 	struct ipqent *p, *q, *nq, *tiqe = NULL;
351 	struct socket *so = NULL;
352 	int pkt_flags;
353 	tcp_seq pkt_seq;
354 	unsigned pkt_len;
355 	u_long rcvpartdupbyte = 0;
356 	u_long rcvoobyte;
357 #ifdef TCP_REASS_COUNTERS
358 	u_int count = 0;
359 #endif
360 
361 	if (tp->t_inpcb)
362 		so = tp->t_inpcb->inp_socket;
363 #ifdef INET6
364 	else if (tp->t_in6pcb)
365 		so = tp->t_in6pcb->in6p_socket;
366 #endif
367 
368 	TCP_REASS_LOCK_CHECK(tp);
369 
370 	/*
371 	 * Call with th==0 after become established to
372 	 * force pre-ESTABLISHED data up to user socket.
373 	 */
374 	if (th == 0)
375 		goto present;
376 
377 	rcvoobyte = *tlen;
378 	/*
379 	 * Copy these to local variables because the tcpiphdr
380 	 * gets munged while we are collapsing mbufs.
381 	 */
382 	pkt_seq = th->th_seq;
383 	pkt_len = *tlen;
384 	pkt_flags = th->th_flags;
385 
386 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
387 
388 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
389 		/*
390 		 * When we miss a packet, the vast majority of time we get
391 		 * packets that follow it in order.  So optimize for that.
392 		 */
393 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
394 			p->ipqe_len += pkt_len;
395 			p->ipqe_flags |= pkt_flags;
396 			m_cat(p->ipqe_m, m);
397 			tiqe = p;
398 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
399 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
400 			goto skip_replacement;
401 		}
402 		/*
403 		 * While we're here, if the pkt is completely beyond
404 		 * anything we have, just insert it at the tail.
405 		 */
406 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
407 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
408 			goto insert_it;
409 		}
410 	}
411 
412 	q = TAILQ_FIRST(&tp->segq);
413 
414 	if (q != NULL) {
415 		/*
416 		 * If this segment immediately precedes the first out-of-order
417 		 * block, simply slap the segment in front of it and (mostly)
418 		 * skip the complicated logic.
419 		 */
420 		if (pkt_seq + pkt_len == q->ipqe_seq) {
421 			q->ipqe_seq = pkt_seq;
422 			q->ipqe_len += pkt_len;
423 			q->ipqe_flags |= pkt_flags;
424 			m_cat(m, q->ipqe_m);
425 			q->ipqe_m = m;
426 			tiqe = q;
427 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
428 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
429 			goto skip_replacement;
430 		}
431 	} else {
432 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
433 	}
434 
435 	/*
436 	 * Find a segment which begins after this one does.
437 	 */
438 	for (p = NULL; q != NULL; q = nq) {
439 		nq = TAILQ_NEXT(q, ipqe_q);
440 #ifdef TCP_REASS_COUNTERS
441 		count++;
442 #endif
443 		/*
444 		 * If the received segment is just right after this
445 		 * fragment, merge the two together and then check
446 		 * for further overlaps.
447 		 */
448 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
449 #ifdef TCPREASS_DEBUG
450 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
451 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
452 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
453 #endif
454 			pkt_len += q->ipqe_len;
455 			pkt_flags |= q->ipqe_flags;
456 			pkt_seq = q->ipqe_seq;
457 			m_cat(q->ipqe_m, m);
458 			m = q->ipqe_m;
459 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
460 			goto free_ipqe;
461 		}
462 		/*
463 		 * If the received segment is completely past this
464 		 * fragment, we need to go the next fragment.
465 		 */
466 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
467 			p = q;
468 			continue;
469 		}
470 		/*
471 		 * If the fragment is past the received segment,
472 		 * it (or any following) can't be concatenated.
473 		 */
474 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
475 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
476 			break;
477 		}
478 
479 		/*
480 		 * We've received all the data in this segment before.
481 		 * mark it as a duplicate and return.
482 		 */
483 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
484 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
485 			tcpstat.tcps_rcvduppack++;
486 			tcpstat.tcps_rcvdupbyte += pkt_len;
487 			m_freem(m);
488 			if (tiqe != NULL)
489 				pool_put(&ipqent_pool, tiqe);
490 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
491 			return (0);
492 		}
493 		/*
494 		 * Received segment completely overlaps this fragment
495 		 * so we drop the fragment (this keeps the temporal
496 		 * ordering of segments correct).
497 		 */
498 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
499 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
500 			rcvpartdupbyte += q->ipqe_len;
501 			m_freem(q->ipqe_m);
502 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
503 			goto free_ipqe;
504 		}
505 		/*
506 		 * RX'ed segment extends past the end of the
507 		 * fragment.  Drop the overlapping bytes.  Then
508 		 * merge the fragment and segment then treat as
509 		 * a longer received packet.
510 		 */
511 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
512 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
513 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
514 #ifdef TCPREASS_DEBUG
515 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
516 			       tp, overlap,
517 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
518 #endif
519 			m_adj(m, overlap);
520 			rcvpartdupbyte += overlap;
521 			m_cat(q->ipqe_m, m);
522 			m = q->ipqe_m;
523 			pkt_seq = q->ipqe_seq;
524 			pkt_len += q->ipqe_len - overlap;
525 			rcvoobyte -= overlap;
526 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
527 			goto free_ipqe;
528 		}
529 		/*
530 		 * RX'ed segment extends past the front of the
531 		 * fragment.  Drop the overlapping bytes on the
532 		 * received packet.  The packet will then be
533 		 * contatentated with this fragment a bit later.
534 		 */
535 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
536 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
537 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
538 #ifdef TCPREASS_DEBUG
539 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
540 			       tp, overlap,
541 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
542 #endif
543 			m_adj(m, -overlap);
544 			pkt_len -= overlap;
545 			rcvpartdupbyte += overlap;
546 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
547 			rcvoobyte -= overlap;
548 		}
549 		/*
550 		 * If the received segment immediates precedes this
551 		 * fragment then tack the fragment onto this segment
552 		 * and reinsert the data.
553 		 */
554 		if (q->ipqe_seq == pkt_seq + pkt_len) {
555 #ifdef TCPREASS_DEBUG
556 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
557 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
558 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
559 #endif
560 			pkt_len += q->ipqe_len;
561 			pkt_flags |= q->ipqe_flags;
562 			m_cat(m, q->ipqe_m);
563 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
564 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
565 			if (tiqe == NULL) {
566 			    tiqe = q;
567 			} else {
568 			    pool_put(&ipqent_pool, q);
569 			}
570 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
571 			break;
572 		}
573 		/*
574 		 * If the fragment is before the segment, remember it.
575 		 * When this loop is terminated, p will contain the
576 		 * pointer to fragment that is right before the received
577 		 * segment.
578 		 */
579 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
580 			p = q;
581 
582 		continue;
583 
584 		/*
585 		 * This is a common operation.  It also will allow
586 		 * to save doing a malloc/free in most instances.
587 		 */
588 	  free_ipqe:
589 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
590 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
591 		if (tiqe == NULL) {
592 		    tiqe = q;
593 		} else {
594 		    pool_put(&ipqent_pool, q);
595 		}
596 	}
597 
598 #ifdef TCP_REASS_COUNTERS
599 	if (count > 7)
600 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
601 	else if (count > 0)
602 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
603 #endif
604 
605     insert_it:
606 
607 	/*
608 	 * Allocate a new queue entry since the received segment did not
609 	 * collapse onto any other out-of-order block; thus we are allocating
610 	 * a new block.  If it had collapsed, tiqe would not be NULL and
611 	 * we would be reusing it.
612 	 * XXX If we can't, just drop the packet.  XXX
613 	 */
614 	if (tiqe == NULL) {
615 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
616 		if (tiqe == NULL) {
617 			tcpstat.tcps_rcvmemdrop++;
618 			m_freem(m);
619 			return (0);
620 		}
621 	}
622 
623 	/*
624 	 * Update the counters.
625 	 */
626 	tcpstat.tcps_rcvoopack++;
627 	tcpstat.tcps_rcvoobyte += rcvoobyte;
628 	if (rcvpartdupbyte) {
629 	    tcpstat.tcps_rcvpartduppack++;
630 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
631 	}
632 
633 	/*
634 	 * Insert the new fragment queue entry into both queues.
635 	 */
636 	tiqe->ipqe_m = m;
637 	tiqe->ipqe_seq = pkt_seq;
638 	tiqe->ipqe_len = pkt_len;
639 	tiqe->ipqe_flags = pkt_flags;
640 	if (p == NULL) {
641 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
642 #ifdef TCPREASS_DEBUG
643 		if (tiqe->ipqe_seq != tp->rcv_nxt)
644 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
645 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
646 #endif
647 	} else {
648 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
649 #ifdef TCPREASS_DEBUG
650 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
651 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
652 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
653 #endif
654 	}
655 
656 skip_replacement:
657 
658 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
659 
660 present:
661 	/*
662 	 * Present data to user, advancing rcv_nxt through
663 	 * completed sequence space.
664 	 */
665 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
666 		return (0);
667 	q = TAILQ_FIRST(&tp->segq);
668 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
669 		return (0);
670 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
671 		return (0);
672 
673 	tp->rcv_nxt += q->ipqe_len;
674 	pkt_flags = q->ipqe_flags & TH_FIN;
675 	ND6_HINT(tp);
676 
677 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
678 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
679 	if (so->so_state & SS_CANTRCVMORE)
680 		m_freem(q->ipqe_m);
681 	else
682 		sbappendstream(&so->so_rcv, q->ipqe_m);
683 	pool_put(&ipqent_pool, q);
684 	sorwakeup(so);
685 	return (pkt_flags);
686 }
687 
688 #ifdef INET6
689 int
690 tcp6_input(mp, offp, proto)
691 	struct mbuf **mp;
692 	int *offp, proto;
693 {
694 	struct mbuf *m = *mp;
695 
696 	/*
697 	 * draft-itojun-ipv6-tcp-to-anycast
698 	 * better place to put this in?
699 	 */
700 	if (m->m_flags & M_ANYCAST6) {
701 		struct ip6_hdr *ip6;
702 		if (m->m_len < sizeof(struct ip6_hdr)) {
703 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
704 				tcpstat.tcps_rcvshort++;
705 				return IPPROTO_DONE;
706 			}
707 		}
708 		ip6 = mtod(m, struct ip6_hdr *);
709 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
710 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
711 		return IPPROTO_DONE;
712 	}
713 
714 	tcp_input(m, *offp, proto);
715 	return IPPROTO_DONE;
716 }
717 #endif
718 
719 #ifdef INET
720 static void
721 tcp4_log_refused(ip, th)
722 	const struct ip *ip;
723 	const struct tcphdr *th;
724 {
725 	char src[4*sizeof "123"];
726 	char dst[4*sizeof "123"];
727 
728 	if (ip) {
729 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
730 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
731 	}
732 	else {
733 		strlcpy(src, "(unknown)", sizeof(src));
734 		strlcpy(dst, "(unknown)", sizeof(dst));
735 	}
736 	log(LOG_INFO,
737 	    "Connection attempt to TCP %s:%d from %s:%d\n",
738 	    dst, ntohs(th->th_dport),
739 	    src, ntohs(th->th_sport));
740 }
741 #endif
742 
743 #ifdef INET6
744 static void
745 tcp6_log_refused(ip6, th)
746 	const struct ip6_hdr *ip6;
747 	const struct tcphdr *th;
748 {
749 	char src[INET6_ADDRSTRLEN];
750 	char dst[INET6_ADDRSTRLEN];
751 
752 	if (ip6) {
753 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
754 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
755 	}
756 	else {
757 		strlcpy(src, "(unknown v6)", sizeof(src));
758 		strlcpy(dst, "(unknown v6)", sizeof(dst));
759 	}
760 	log(LOG_INFO,
761 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
762 	    dst, ntohs(th->th_dport),
763 	    src, ntohs(th->th_sport));
764 }
765 #endif
766 
767 /*
768  * TCP input routine, follows pages 65-76 of the
769  * protocol specification dated September, 1981 very closely.
770  */
771 void
772 #if __STDC__
773 tcp_input(struct mbuf *m, ...)
774 #else
775 tcp_input(m, va_alist)
776 	struct mbuf *m;
777 #endif
778 {
779 	struct tcphdr *th;
780 	struct ip *ip;
781 	struct inpcb *inp;
782 #ifdef INET6
783 	struct ip6_hdr *ip6;
784 	struct in6pcb *in6p;
785 #endif
786 	u_int8_t *optp = NULL;
787 	int optlen = 0;
788 	int len, tlen, toff, hdroptlen = 0;
789 	struct tcpcb *tp = 0;
790 	int tiflags;
791 	struct socket *so = NULL;
792 	int todrop, acked, ourfinisacked, needoutput = 0;
793 #ifdef TCP_DEBUG
794 	short ostate = 0;
795 #endif
796 	int iss = 0;
797 	u_long tiwin;
798 	struct tcp_opt_info opti;
799 	int off, iphlen;
800 	va_list ap;
801 	int af;		/* af on the wire */
802 	struct mbuf *tcp_saveti = NULL;
803 
804 	MCLAIM(m, &tcp_rx_mowner);
805 	va_start(ap, m);
806 	toff = va_arg(ap, int);
807 	(void)va_arg(ap, int);		/* ignore value, advance ap */
808 	va_end(ap);
809 
810 	tcpstat.tcps_rcvtotal++;
811 
812 	bzero(&opti, sizeof(opti));
813 	opti.ts_present = 0;
814 	opti.maxseg = 0;
815 
816 	/*
817 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
818 	 *
819 	 * TCP is, by definition, unicast, so we reject all
820 	 * multicast outright.
821 	 *
822 	 * Note, there are additional src/dst address checks in
823 	 * the AF-specific code below.
824 	 */
825 	if (m->m_flags & (M_BCAST|M_MCAST)) {
826 		/* XXX stat */
827 		goto drop;
828 	}
829 #ifdef INET6
830 	if (m->m_flags & M_ANYCAST6) {
831 		/* XXX stat */
832 		goto drop;
833 	}
834 #endif
835 
836 	/*
837 	 * Get IP and TCP header together in first mbuf.
838 	 * Note: IP leaves IP header in first mbuf.
839 	 */
840 	ip = mtod(m, struct ip *);
841 #ifdef INET6
842 	ip6 = NULL;
843 #endif
844 	switch (ip->ip_v) {
845 #ifdef INET
846 	case 4:
847 		af = AF_INET;
848 		iphlen = sizeof(struct ip);
849 		ip = mtod(m, struct ip *);
850 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
851 			sizeof(struct tcphdr));
852 		if (th == NULL) {
853 			tcpstat.tcps_rcvshort++;
854 			return;
855 		}
856 		/* We do the checksum after PCB lookup... */
857 		len = ntohs(ip->ip_len);
858 		tlen = len - toff;
859 		break;
860 #endif
861 #ifdef INET6
862 	case 6:
863 		ip = NULL;
864 		iphlen = sizeof(struct ip6_hdr);
865 		af = AF_INET6;
866 		ip6 = mtod(m, struct ip6_hdr *);
867 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
868 			sizeof(struct tcphdr));
869 		if (th == NULL) {
870 			tcpstat.tcps_rcvshort++;
871 			return;
872 		}
873 
874 		/* Be proactive about malicious use of IPv4 mapped address */
875 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
876 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
877 			/* XXX stat */
878 			goto drop;
879 		}
880 
881 		/*
882 		 * Be proactive about unspecified IPv6 address in source.
883 		 * As we use all-zero to indicate unbounded/unconnected pcb,
884 		 * unspecified IPv6 address can be used to confuse us.
885 		 *
886 		 * Note that packets with unspecified IPv6 destination is
887 		 * already dropped in ip6_input.
888 		 */
889 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
890 			/* XXX stat */
891 			goto drop;
892 		}
893 
894 		/*
895 		 * Make sure destination address is not multicast.
896 		 * Source address checked in ip6_input().
897 		 */
898 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
899 			/* XXX stat */
900 			goto drop;
901 		}
902 
903 		/* We do the checksum after PCB lookup... */
904 		len = m->m_pkthdr.len;
905 		tlen = len - toff;
906 		break;
907 #endif
908 	default:
909 		m_freem(m);
910 		return;
911 	}
912 
913 	KASSERT(TCP_HDR_ALIGNED_P(th));
914 
915 	/*
916 	 * Check that TCP offset makes sense,
917 	 * pull out TCP options and adjust length.		XXX
918 	 */
919 	off = th->th_off << 2;
920 	if (off < sizeof (struct tcphdr) || off > tlen) {
921 		tcpstat.tcps_rcvbadoff++;
922 		goto drop;
923 	}
924 	tlen -= off;
925 
926 	/*
927 	 * tcp_input() has been modified to use tlen to mean the TCP data
928 	 * length throughout the function.  Other functions can use
929 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
930 	 * rja
931 	 */
932 
933 	if (off > sizeof (struct tcphdr)) {
934 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
935 		if (th == NULL) {
936 			tcpstat.tcps_rcvshort++;
937 			return;
938 		}
939 		/*
940 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
941 		 * (as they're before toff) and we don't need to update those.
942 		 */
943 		KASSERT(TCP_HDR_ALIGNED_P(th));
944 		optlen = off - sizeof (struct tcphdr);
945 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
946 		/*
947 		 * Do quick retrieval of timestamp options ("options
948 		 * prediction?").  If timestamp is the only option and it's
949 		 * formatted as recommended in RFC 1323 appendix A, we
950 		 * quickly get the values now and not bother calling
951 		 * tcp_dooptions(), etc.
952 		 */
953 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
954 		     (optlen > TCPOLEN_TSTAMP_APPA &&
955 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
956 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
957 		     (th->th_flags & TH_SYN) == 0) {
958 			opti.ts_present = 1;
959 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
960 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
961 			optp = NULL;	/* we've parsed the options */
962 		}
963 	}
964 	tiflags = th->th_flags;
965 
966 	/*
967 	 * Locate pcb for segment.
968 	 */
969 findpcb:
970 	inp = NULL;
971 #ifdef INET6
972 	in6p = NULL;
973 #endif
974 	switch (af) {
975 #ifdef INET
976 	case AF_INET:
977 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
978 		    ip->ip_dst, th->th_dport);
979 		if (inp == 0) {
980 			++tcpstat.tcps_pcbhashmiss;
981 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
982 		}
983 #ifdef INET6
984 		if (inp == 0) {
985 			struct in6_addr s, d;
986 
987 			/* mapped addr case */
988 			bzero(&s, sizeof(s));
989 			s.s6_addr16[5] = htons(0xffff);
990 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
991 			bzero(&d, sizeof(d));
992 			d.s6_addr16[5] = htons(0xffff);
993 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
994 			in6p = in6_pcblookup_connect(&tcbtable, &s,
995 			    th->th_sport, &d, th->th_dport, 0);
996 			if (in6p == 0) {
997 				++tcpstat.tcps_pcbhashmiss;
998 				in6p = in6_pcblookup_bind(&tcbtable, &d,
999 				    th->th_dport, 0);
1000 			}
1001 		}
1002 #endif
1003 #ifndef INET6
1004 		if (inp == 0)
1005 #else
1006 		if (inp == 0 && in6p == 0)
1007 #endif
1008 		{
1009 			++tcpstat.tcps_noport;
1010 			if (tcp_log_refused &&
1011 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1012 				tcp4_log_refused(ip, th);
1013 			}
1014 			TCP_FIELDS_TO_HOST(th);
1015 			goto dropwithreset_ratelim;
1016 		}
1017 #if defined(IPSEC) || defined(FAST_IPSEC)
1018 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1019 		    ipsec4_in_reject(m, inp)) {
1020 			ipsecstat.in_polvio++;
1021 			goto drop;
1022 		}
1023 #ifdef INET6
1024 		else if (in6p &&
1025 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1026 		    ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1027 			ipsecstat.in_polvio++;
1028 			goto drop;
1029 		}
1030 #endif
1031 #endif /*IPSEC*/
1032 		break;
1033 #endif /*INET*/
1034 #ifdef INET6
1035 	case AF_INET6:
1036 	    {
1037 		int faith;
1038 
1039 #if defined(NFAITH) && NFAITH > 0
1040 		faith = faithprefix(&ip6->ip6_dst);
1041 #else
1042 		faith = 0;
1043 #endif
1044 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1045 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1046 		if (in6p == NULL) {
1047 			++tcpstat.tcps_pcbhashmiss;
1048 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1049 				th->th_dport, faith);
1050 		}
1051 		if (in6p == NULL) {
1052 			++tcpstat.tcps_noport;
1053 			if (tcp_log_refused &&
1054 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1055 				tcp6_log_refused(ip6, th);
1056 			}
1057 			TCP_FIELDS_TO_HOST(th);
1058 			goto dropwithreset_ratelim;
1059 		}
1060 #if defined(IPSEC) || defined(FAST_IPSEC)
1061 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1062 		    ipsec6_in_reject(m, in6p)) {
1063 			ipsec6stat.in_polvio++;
1064 			goto drop;
1065 		}
1066 #endif /*IPSEC*/
1067 		break;
1068 	    }
1069 #endif
1070 	}
1071 
1072 	/*
1073 	 * If the state is CLOSED (i.e., TCB does not exist) then
1074 	 * all data in the incoming segment is discarded.
1075 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1076 	 * but should either do a listen or a connect soon.
1077 	 */
1078 	tp = NULL;
1079 	so = NULL;
1080 	if (inp) {
1081 		tp = intotcpcb(inp);
1082 		so = inp->inp_socket;
1083 	}
1084 #ifdef INET6
1085 	else if (in6p) {
1086 		tp = in6totcpcb(in6p);
1087 		so = in6p->in6p_socket;
1088 	}
1089 #endif
1090 	if (tp == 0) {
1091 		TCP_FIELDS_TO_HOST(th);
1092 		goto dropwithreset_ratelim;
1093 	}
1094 	if (tp->t_state == TCPS_CLOSED)
1095 		goto drop;
1096 
1097 	/*
1098 	 * Checksum extended TCP header and data.
1099 	 */
1100 	switch (af) {
1101 #ifdef INET
1102 	case AF_INET:
1103 		switch (m->m_pkthdr.csum_flags &
1104 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
1105 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
1106 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
1107 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
1108 			goto badcsum;
1109 
1110 		case M_CSUM_TCPv4|M_CSUM_DATA: {
1111 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
1112 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
1113 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
1114 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
1115 				    ip->ip_dst.s_addr,
1116 				    htonl(hw_csum + ntohs(ip->ip_len) +
1117 					  IPPROTO_UDP));
1118 			}
1119 			if ((hw_csum ^ 0xffff) != 0)
1120 				goto badcsum;
1121 			break;
1122 		}
1123 
1124 		case M_CSUM_TCPv4:
1125 			/* Checksum was okay. */
1126 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
1127 			break;
1128 
1129 		default:
1130 			/* Must compute it ourselves. */
1131 			TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
1132 			if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1133 				goto badcsum;
1134 			break;
1135 		}
1136 		break;
1137 #endif /* INET4 */
1138 
1139 #ifdef INET6
1140 	case AF_INET6:
1141 		if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1142 			goto badcsum;
1143 		break;
1144 #endif /* INET6 */
1145 	}
1146 
1147 	TCP_FIELDS_TO_HOST(th);
1148 
1149 	/* Unscale the window into a 32-bit value. */
1150 	if ((tiflags & TH_SYN) == 0)
1151 		tiwin = th->th_win << tp->snd_scale;
1152 	else
1153 		tiwin = th->th_win;
1154 
1155 #ifdef INET6
1156 	/* save packet options if user wanted */
1157 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1158 		if (in6p->in6p_options) {
1159 			m_freem(in6p->in6p_options);
1160 			in6p->in6p_options = 0;
1161 		}
1162 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1163 	}
1164 #endif
1165 
1166 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1167 		union syn_cache_sa src;
1168 		union syn_cache_sa dst;
1169 
1170 		bzero(&src, sizeof(src));
1171 		bzero(&dst, sizeof(dst));
1172 		switch (af) {
1173 #ifdef INET
1174 		case AF_INET:
1175 			src.sin.sin_len = sizeof(struct sockaddr_in);
1176 			src.sin.sin_family = AF_INET;
1177 			src.sin.sin_addr = ip->ip_src;
1178 			src.sin.sin_port = th->th_sport;
1179 
1180 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1181 			dst.sin.sin_family = AF_INET;
1182 			dst.sin.sin_addr = ip->ip_dst;
1183 			dst.sin.sin_port = th->th_dport;
1184 			break;
1185 #endif
1186 #ifdef INET6
1187 		case AF_INET6:
1188 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1189 			src.sin6.sin6_family = AF_INET6;
1190 			src.sin6.sin6_addr = ip6->ip6_src;
1191 			src.sin6.sin6_port = th->th_sport;
1192 
1193 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1194 			dst.sin6.sin6_family = AF_INET6;
1195 			dst.sin6.sin6_addr = ip6->ip6_dst;
1196 			dst.sin6.sin6_port = th->th_dport;
1197 			break;
1198 #endif /* INET6 */
1199 		default:
1200 			goto badsyn;	/*sanity*/
1201 		}
1202 
1203 		if (so->so_options & SO_DEBUG) {
1204 #ifdef TCP_DEBUG
1205 			ostate = tp->t_state;
1206 #endif
1207 
1208 			tcp_saveti = NULL;
1209 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1210 				goto nosave;
1211 
1212 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1213 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1214 				if (!tcp_saveti)
1215 					goto nosave;
1216 			} else {
1217 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1218 				if (!tcp_saveti)
1219 					goto nosave;
1220 				MCLAIM(m, &tcp_mowner);
1221 				tcp_saveti->m_len = iphlen;
1222 				m_copydata(m, 0, iphlen,
1223 				    mtod(tcp_saveti, caddr_t));
1224 			}
1225 
1226 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1227 				m_freem(tcp_saveti);
1228 				tcp_saveti = NULL;
1229 			} else {
1230 				tcp_saveti->m_len += sizeof(struct tcphdr);
1231 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1232 				    sizeof(struct tcphdr));
1233 			}
1234 	nosave:;
1235 		}
1236 		if (so->so_options & SO_ACCEPTCONN) {
1237 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1238 				if (tiflags & TH_RST) {
1239 					syn_cache_reset(&src.sa, &dst.sa, th);
1240 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1241 				    (TH_ACK|TH_SYN)) {
1242 					/*
1243 					 * Received a SYN,ACK.  This should
1244 					 * never happen while we are in
1245 					 * LISTEN.  Send an RST.
1246 					 */
1247 					goto badsyn;
1248 				} else if (tiflags & TH_ACK) {
1249 					so = syn_cache_get(&src.sa, &dst.sa,
1250 						th, toff, tlen, so, m);
1251 					if (so == NULL) {
1252 						/*
1253 						 * We don't have a SYN for
1254 						 * this ACK; send an RST.
1255 						 */
1256 						goto badsyn;
1257 					} else if (so ==
1258 					    (struct socket *)(-1)) {
1259 						/*
1260 						 * We were unable to create
1261 						 * the connection.  If the
1262 						 * 3-way handshake was
1263 						 * completed, and RST has
1264 						 * been sent to the peer.
1265 						 * Since the mbuf might be
1266 						 * in use for the reply,
1267 						 * do not free it.
1268 						 */
1269 						m = NULL;
1270 					} else {
1271 						/*
1272 						 * We have created a
1273 						 * full-blown connection.
1274 						 */
1275 						tp = NULL;
1276 						inp = NULL;
1277 #ifdef INET6
1278 						in6p = NULL;
1279 #endif
1280 						switch (so->so_proto->pr_domain->dom_family) {
1281 #ifdef INET
1282 						case AF_INET:
1283 							inp = sotoinpcb(so);
1284 							tp = intotcpcb(inp);
1285 							break;
1286 #endif
1287 #ifdef INET6
1288 						case AF_INET6:
1289 							in6p = sotoin6pcb(so);
1290 							tp = in6totcpcb(in6p);
1291 							break;
1292 #endif
1293 						}
1294 						if (tp == NULL)
1295 							goto badsyn;	/*XXX*/
1296 						tiwin <<= tp->snd_scale;
1297 						goto after_listen;
1298 					}
1299 				} else {
1300 					/*
1301 					 * None of RST, SYN or ACK was set.
1302 					 * This is an invalid packet for a
1303 					 * TCB in LISTEN state.  Send a RST.
1304 					 */
1305 					goto badsyn;
1306 				}
1307 			} else {
1308 				/*
1309 				 * Received a SYN.
1310 				 */
1311 
1312 #ifdef INET6
1313 				/*
1314 				 * If deprecated address is forbidden, we do
1315 				 * not accept SYN to deprecated interface
1316 				 * address to prevent any new inbound
1317 				 * connection from getting established.
1318 				 * When we do not accept SYN, we send a TCP
1319 				 * RST, with deprecated source address (instead
1320 				 * of dropping it).  We compromise it as it is
1321 				 * much better for peer to send a RST, and
1322 				 * RST will be the final packet for the
1323 				 * exchange.
1324 				 *
1325 				 * If we do not forbid deprecated addresses, we
1326 				 * accept the SYN packet.  RFC2462 does not
1327 				 * suggest dropping SYN in this case.
1328 				 * If we decipher RFC2462 5.5.4, it says like
1329 				 * this:
1330 				 * 1. use of deprecated addr with existing
1331 				 *    communication is okay - "SHOULD continue
1332 				 *    to be used"
1333 				 * 2. use of it with new communication:
1334 				 *   (2a) "SHOULD NOT be used if alternate
1335 				 *        address with sufficient scope is
1336 				 *        available"
1337 				 *   (2b) nothing mentioned otherwise.
1338 				 * Here we fall into (2b) case as we have no
1339 				 * choice in our source address selection - we
1340 				 * must obey the peer.
1341 				 *
1342 				 * The wording in RFC2462 is confusing, and
1343 				 * there are multiple description text for
1344 				 * deprecated address handling - worse, they
1345 				 * are not exactly the same.  I believe 5.5.4
1346 				 * is the best one, so we follow 5.5.4.
1347 				 */
1348 				if (af == AF_INET6 && !ip6_use_deprecated) {
1349 					struct in6_ifaddr *ia6;
1350 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1351 					    &ip6->ip6_dst)) &&
1352 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1353 						tp = NULL;
1354 						goto dropwithreset;
1355 					}
1356 				}
1357 #endif
1358 
1359 #ifdef IPSEC
1360 				switch (af) {
1361 				case AF_INET:
1362 					if (ipsec4_in_reject_so(m, so)) {
1363 						ipsecstat.in_polvio++;
1364 						tp = NULL;
1365 						goto dropwithreset;
1366 					}
1367 					break;
1368 #ifdef INET6
1369 				case AF_INET6:
1370 					if (ipsec6_in_reject_so(m, so)) {
1371 						ipsec6stat.in_polvio++;
1372 						tp = NULL;
1373 						goto dropwithreset;
1374 					}
1375 					break;
1376 #endif
1377 				}
1378 #endif
1379 
1380 				/*
1381 				 * LISTEN socket received a SYN
1382 				 * from itself?  This can't possibly
1383 				 * be valid; drop the packet.
1384 				 */
1385 				if (th->th_sport == th->th_dport) {
1386 					int i;
1387 
1388 					switch (af) {
1389 #ifdef INET
1390 					case AF_INET:
1391 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1392 						break;
1393 #endif
1394 #ifdef INET6
1395 					case AF_INET6:
1396 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1397 						break;
1398 #endif
1399 					default:
1400 						i = 1;
1401 					}
1402 					if (i) {
1403 						tcpstat.tcps_badsyn++;
1404 						goto drop;
1405 					}
1406 				}
1407 
1408 				/*
1409 				 * SYN looks ok; create compressed TCP
1410 				 * state for it.
1411 				 */
1412 				if (so->so_qlen <= so->so_qlimit &&
1413 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1414 						so, m, optp, optlen, &opti))
1415 					m = NULL;
1416 			}
1417 			goto drop;
1418 		}
1419 	}
1420 
1421 after_listen:
1422 #ifdef DIAGNOSTIC
1423 	/*
1424 	 * Should not happen now that all embryonic connections
1425 	 * are handled with compressed state.
1426 	 */
1427 	if (tp->t_state == TCPS_LISTEN)
1428 		panic("tcp_input: TCPS_LISTEN");
1429 #endif
1430 
1431 	/*
1432 	 * Segment received on connection.
1433 	 * Reset idle time and keep-alive timer.
1434 	 */
1435 	tp->t_rcvtime = tcp_now;
1436 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1437 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1438 
1439 	/*
1440 	 * Process options.
1441 	 */
1442 	if (optp)
1443 		tcp_dooptions(tp, optp, optlen, th, &opti);
1444 
1445 	/*
1446 	 * Header prediction: check for the two common cases
1447 	 * of a uni-directional data xfer.  If the packet has
1448 	 * no control flags, is in-sequence, the window didn't
1449 	 * change and we're not retransmitting, it's a
1450 	 * candidate.  If the length is zero and the ack moved
1451 	 * forward, we're the sender side of the xfer.  Just
1452 	 * free the data acked & wake any higher level process
1453 	 * that was blocked waiting for space.  If the length
1454 	 * is non-zero and the ack didn't move, we're the
1455 	 * receiver side.  If we're getting packets in-order
1456 	 * (the reassembly queue is empty), add the data to
1457 	 * the socket buffer and note that we need a delayed ack.
1458 	 */
1459 	if (tp->t_state == TCPS_ESTABLISHED &&
1460 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1461 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1462 	    th->th_seq == tp->rcv_nxt &&
1463 	    tiwin && tiwin == tp->snd_wnd &&
1464 	    tp->snd_nxt == tp->snd_max) {
1465 
1466 		/*
1467 		 * If last ACK falls within this segment's sequence numbers,
1468 		 *  record the timestamp.
1469 		 */
1470 		if (opti.ts_present &&
1471 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1472 		    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) {
1473 			tp->ts_recent_age = TCP_TIMESTAMP(tp);
1474 			tp->ts_recent = opti.ts_val;
1475 		}
1476 
1477 		if (tlen == 0) {
1478 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1479 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1480 			    tp->snd_cwnd >= tp->snd_wnd &&
1481 			    tp->t_dupacks < tcprexmtthresh) {
1482 				/*
1483 				 * this is a pure ack for outstanding data.
1484 				 */
1485 				++tcpstat.tcps_predack;
1486 				if (opti.ts_present && opti.ts_ecr)
1487 					tcp_xmit_timer(tp,
1488 					  TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
1489 				else if (tp->t_rtttime &&
1490 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1491 					tcp_xmit_timer(tp,
1492 					tcp_now - tp->t_rtttime);
1493 				acked = th->th_ack - tp->snd_una;
1494 				tcpstat.tcps_rcvackpack++;
1495 				tcpstat.tcps_rcvackbyte += acked;
1496 				ND6_HINT(tp);
1497 
1498 				if (acked > (tp->t_lastoff - tp->t_inoff))
1499 					tp->t_lastm = NULL;
1500 				sbdrop(&so->so_snd, acked);
1501 				tp->t_lastoff -= acked;
1502 
1503 				/*
1504 				 * We want snd_recover to track snd_una to
1505 				 * avoid sequence wraparound problems for
1506 				 * very large transfers.
1507 				 */
1508 				tp->snd_una = tp->snd_recover = th->th_ack;
1509 				m_freem(m);
1510 
1511 				/*
1512 				 * If all outstanding data are acked, stop
1513 				 * retransmit timer, otherwise restart timer
1514 				 * using current (possibly backed-off) value.
1515 				 * If process is waiting for space,
1516 				 * wakeup/selwakeup/signal.  If data
1517 				 * are ready to send, let tcp_output
1518 				 * decide between more output or persist.
1519 				 */
1520 				if (tp->snd_una == tp->snd_max)
1521 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1522 				else if (TCP_TIMER_ISARMED(tp,
1523 				    TCPT_PERSIST) == 0)
1524 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1525 					    tp->t_rxtcur);
1526 
1527 				sowwakeup(so);
1528 				if (so->so_snd.sb_cc)
1529 					(void) tcp_output(tp);
1530 				if (tcp_saveti)
1531 					m_freem(tcp_saveti);
1532 				return;
1533 			}
1534 		} else if (th->th_ack == tp->snd_una &&
1535 		    TAILQ_FIRST(&tp->segq) == NULL &&
1536 		    tlen <= sbspace(&so->so_rcv)) {
1537 			/*
1538 			 * this is a pure, in-sequence data packet
1539 			 * with nothing on the reassembly queue and
1540 			 * we have enough buffer space to take it.
1541 			 */
1542 			++tcpstat.tcps_preddat;
1543 			tp->rcv_nxt += tlen;
1544 			tcpstat.tcps_rcvpack++;
1545 			tcpstat.tcps_rcvbyte += tlen;
1546 			ND6_HINT(tp);
1547 			/*
1548 			 * Drop TCP, IP headers and TCP options then add data
1549 			 * to socket buffer.
1550 			 */
1551 			if (so->so_state & SS_CANTRCVMORE)
1552 				m_freem(m);
1553 			else {
1554 				m_adj(m, toff + off);
1555 				sbappendstream(&so->so_rcv, m);
1556 			}
1557 			sorwakeup(so);
1558 			TCP_SETUP_ACK(tp, th);
1559 			if (tp->t_flags & TF_ACKNOW)
1560 				(void) tcp_output(tp);
1561 			if (tcp_saveti)
1562 				m_freem(tcp_saveti);
1563 			return;
1564 		}
1565 	}
1566 
1567 	/*
1568 	 * Compute mbuf offset to TCP data segment.
1569 	 */
1570 	hdroptlen = toff + off;
1571 
1572 	/*
1573 	 * Calculate amount of space in receive window,
1574 	 * and then do TCP input processing.
1575 	 * Receive window is amount of space in rcv queue,
1576 	 * but not less than advertised window.
1577 	 */
1578 	{ int win;
1579 
1580 	win = sbspace(&so->so_rcv);
1581 	if (win < 0)
1582 		win = 0;
1583 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1584 	}
1585 
1586 	switch (tp->t_state) {
1587 	case TCPS_LISTEN:
1588 		/*
1589 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1590 		 */
1591 		if (m->m_flags & (M_BCAST|M_MCAST))
1592 			goto drop;
1593 		switch (af) {
1594 #ifdef INET6
1595 		case AF_INET6:
1596 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1597 				goto drop;
1598 			break;
1599 #endif /* INET6 */
1600 		case AF_INET:
1601 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1602 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1603 				goto drop;
1604 			break;
1605 		}
1606 		break;
1607 
1608 	/*
1609 	 * If the state is SYN_SENT:
1610 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1611 	 *	if seg contains a RST, then drop the connection.
1612 	 *	if seg does not contain SYN, then drop it.
1613 	 * Otherwise this is an acceptable SYN segment
1614 	 *	initialize tp->rcv_nxt and tp->irs
1615 	 *	if seg contains ack then advance tp->snd_una
1616 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1617 	 *	arrange for segment to be acked (eventually)
1618 	 *	continue processing rest of data/controls, beginning with URG
1619 	 */
1620 	case TCPS_SYN_SENT:
1621 		if ((tiflags & TH_ACK) &&
1622 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1623 		     SEQ_GT(th->th_ack, tp->snd_max)))
1624 			goto dropwithreset;
1625 		if (tiflags & TH_RST) {
1626 			if (tiflags & TH_ACK)
1627 				tp = tcp_drop(tp, ECONNREFUSED);
1628 			goto drop;
1629 		}
1630 		if ((tiflags & TH_SYN) == 0)
1631 			goto drop;
1632 		if (tiflags & TH_ACK) {
1633 			tp->snd_una = tp->snd_recover = th->th_ack;
1634 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1635 				tp->snd_nxt = tp->snd_una;
1636 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1637 		}
1638 		tp->irs = th->th_seq;
1639 		tcp_rcvseqinit(tp);
1640 		tp->t_flags |= TF_ACKNOW;
1641 		tcp_mss_from_peer(tp, opti.maxseg);
1642 
1643 		/*
1644 		 * Initialize the initial congestion window.  If we
1645 		 * had to retransmit the SYN, we must initialize cwnd
1646 		 * to 1 segment (i.e. the Loss Window).
1647 		 */
1648 		if (tp->t_flags & TF_SYN_REXMT)
1649 			tp->snd_cwnd = tp->t_peermss;
1650 		else {
1651 			int ss = tcp_init_win;
1652 #ifdef INET
1653 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1654 				ss = tcp_init_win_local;
1655 #endif
1656 #ifdef INET6
1657 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1658 				ss = tcp_init_win_local;
1659 #endif
1660 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1661 		}
1662 
1663 		tcp_rmx_rtt(tp);
1664 		if (tiflags & TH_ACK) {
1665 			tcpstat.tcps_connects++;
1666 			soisconnected(so);
1667 			tcp_established(tp);
1668 			/* Do window scaling on this connection? */
1669 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1670 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1671 				tp->snd_scale = tp->requested_s_scale;
1672 				tp->rcv_scale = tp->request_r_scale;
1673 			}
1674 			TCP_REASS_LOCK(tp);
1675 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1676 			TCP_REASS_UNLOCK(tp);
1677 			/*
1678 			 * if we didn't have to retransmit the SYN,
1679 			 * use its rtt as our initial srtt & rtt var.
1680 			 */
1681 			if (tp->t_rtttime)
1682 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1683 		} else
1684 			tp->t_state = TCPS_SYN_RECEIVED;
1685 
1686 		/*
1687 		 * Advance th->th_seq to correspond to first data byte.
1688 		 * If data, trim to stay within window,
1689 		 * dropping FIN if necessary.
1690 		 */
1691 		th->th_seq++;
1692 		if (tlen > tp->rcv_wnd) {
1693 			todrop = tlen - tp->rcv_wnd;
1694 			m_adj(m, -todrop);
1695 			tlen = tp->rcv_wnd;
1696 			tiflags &= ~TH_FIN;
1697 			tcpstat.tcps_rcvpackafterwin++;
1698 			tcpstat.tcps_rcvbyteafterwin += todrop;
1699 		}
1700 		tp->snd_wl1 = th->th_seq - 1;
1701 		tp->rcv_up = th->th_seq;
1702 		goto step6;
1703 
1704 	/*
1705 	 * If the state is SYN_RECEIVED:
1706 	 *	If seg contains an ACK, but not for our SYN, drop the input
1707 	 *	and generate an RST.  See page 36, rfc793
1708 	 */
1709 	case TCPS_SYN_RECEIVED:
1710 		if ((tiflags & TH_ACK) &&
1711 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1712 		     SEQ_GT(th->th_ack, tp->snd_max)))
1713 			goto dropwithreset;
1714 		break;
1715 	}
1716 
1717 	/*
1718 	 * States other than LISTEN or SYN_SENT.
1719 	 * First check timestamp, if present.
1720 	 * Then check that at least some bytes of segment are within
1721 	 * receive window.  If segment begins before rcv_nxt,
1722 	 * drop leading data (and SYN); if nothing left, just ack.
1723 	 *
1724 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1725 	 * and it's less than ts_recent, drop it.
1726 	 */
1727 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1728 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1729 
1730 		/* Check to see if ts_recent is over 24 days old.  */
1731 		if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) >
1732 		    TCP_PAWS_IDLE) {
1733 			/*
1734 			 * Invalidate ts_recent.  If this segment updates
1735 			 * ts_recent, the age will be reset later and ts_recent
1736 			 * will get a valid value.  If it does not, setting
1737 			 * ts_recent to zero will at least satisfy the
1738 			 * requirement that zero be placed in the timestamp
1739 			 * echo reply when ts_recent isn't valid.  The
1740 			 * age isn't reset until we get a valid ts_recent
1741 			 * because we don't want out-of-order segments to be
1742 			 * dropped when ts_recent is old.
1743 			 */
1744 			tp->ts_recent = 0;
1745 		} else {
1746 			tcpstat.tcps_rcvduppack++;
1747 			tcpstat.tcps_rcvdupbyte += tlen;
1748 			tcpstat.tcps_pawsdrop++;
1749 			goto dropafterack;
1750 		}
1751 	}
1752 
1753 	todrop = tp->rcv_nxt - th->th_seq;
1754 	if (todrop > 0) {
1755 		if (tiflags & TH_SYN) {
1756 			tiflags &= ~TH_SYN;
1757 			th->th_seq++;
1758 			if (th->th_urp > 1)
1759 				th->th_urp--;
1760 			else {
1761 				tiflags &= ~TH_URG;
1762 				th->th_urp = 0;
1763 			}
1764 			todrop--;
1765 		}
1766 		if (todrop > tlen ||
1767 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1768 			/*
1769 			 * Any valid FIN must be to the left of the window.
1770 			 * At this point the FIN must be a duplicate or
1771 			 * out of sequence; drop it.
1772 			 */
1773 			tiflags &= ~TH_FIN;
1774 			/*
1775 			 * Send an ACK to resynchronize and drop any data.
1776 			 * But keep on processing for RST or ACK.
1777 			 */
1778 			tp->t_flags |= TF_ACKNOW;
1779 			todrop = tlen;
1780 			tcpstat.tcps_rcvdupbyte += todrop;
1781 			tcpstat.tcps_rcvduppack++;
1782 		} else {
1783 			tcpstat.tcps_rcvpartduppack++;
1784 			tcpstat.tcps_rcvpartdupbyte += todrop;
1785 		}
1786 		hdroptlen += todrop;	/*drop from head afterwards*/
1787 		th->th_seq += todrop;
1788 		tlen -= todrop;
1789 		if (th->th_urp > todrop)
1790 			th->th_urp -= todrop;
1791 		else {
1792 			tiflags &= ~TH_URG;
1793 			th->th_urp = 0;
1794 		}
1795 	}
1796 
1797 	/*
1798 	 * If new data are received on a connection after the
1799 	 * user processes are gone, then RST the other end.
1800 	 */
1801 	if ((so->so_state & SS_NOFDREF) &&
1802 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1803 		tp = tcp_close(tp);
1804 		tcpstat.tcps_rcvafterclose++;
1805 		goto dropwithreset;
1806 	}
1807 
1808 	/*
1809 	 * If segment ends after window, drop trailing data
1810 	 * (and PUSH and FIN); if nothing left, just ACK.
1811 	 */
1812 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1813 	if (todrop > 0) {
1814 		tcpstat.tcps_rcvpackafterwin++;
1815 		if (todrop >= tlen) {
1816 			tcpstat.tcps_rcvbyteafterwin += tlen;
1817 			/*
1818 			 * If a new connection request is received
1819 			 * while in TIME_WAIT, drop the old connection
1820 			 * and start over if the sequence numbers
1821 			 * are above the previous ones.
1822 			 *
1823 			 * NOTE: We will checksum the packet again, and
1824 			 * so we need to put the header fields back into
1825 			 * network order!
1826 			 * XXX This kind of sucks, but we don't expect
1827 			 * XXX this to happen very often, so maybe it
1828 			 * XXX doesn't matter so much.
1829 			 */
1830 			if (tiflags & TH_SYN &&
1831 			    tp->t_state == TCPS_TIME_WAIT &&
1832 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1833 				iss = tcp_new_iss(tp, tp->snd_nxt);
1834 				tp = tcp_close(tp);
1835 				TCP_FIELDS_TO_NET(th);
1836 				goto findpcb;
1837 			}
1838 			/*
1839 			 * If window is closed can only take segments at
1840 			 * window edge, and have to drop data and PUSH from
1841 			 * incoming segments.  Continue processing, but
1842 			 * remember to ack.  Otherwise, drop segment
1843 			 * and ack.
1844 			 */
1845 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1846 				tp->t_flags |= TF_ACKNOW;
1847 				tcpstat.tcps_rcvwinprobe++;
1848 			} else
1849 				goto dropafterack;
1850 		} else
1851 			tcpstat.tcps_rcvbyteafterwin += todrop;
1852 		m_adj(m, -todrop);
1853 		tlen -= todrop;
1854 		tiflags &= ~(TH_PUSH|TH_FIN);
1855 	}
1856 
1857 	/*
1858 	 * If last ACK falls within this segment's sequence numbers,
1859 	 * and the timestamp is newer, record it.
1860 	 */
1861 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1862 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1863 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
1864 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1865 		tp->ts_recent_age = TCP_TIMESTAMP(tp);
1866 		tp->ts_recent = opti.ts_val;
1867 	}
1868 
1869 	/*
1870 	 * If the RST bit is set examine the state:
1871 	 *    SYN_RECEIVED STATE:
1872 	 *	If passive open, return to LISTEN state.
1873 	 *	If active open, inform user that connection was refused.
1874 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1875 	 *	Inform user that connection was reset, and close tcb.
1876 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1877 	 *	Close the tcb.
1878 	 */
1879 	if (tiflags&TH_RST) switch (tp->t_state) {
1880 
1881 	case TCPS_SYN_RECEIVED:
1882 		so->so_error = ECONNREFUSED;
1883 		goto close;
1884 
1885 	case TCPS_ESTABLISHED:
1886 	case TCPS_FIN_WAIT_1:
1887 	case TCPS_FIN_WAIT_2:
1888 	case TCPS_CLOSE_WAIT:
1889 		so->so_error = ECONNRESET;
1890 	close:
1891 		tp->t_state = TCPS_CLOSED;
1892 		tcpstat.tcps_drops++;
1893 		tp = tcp_close(tp);
1894 		goto drop;
1895 
1896 	case TCPS_CLOSING:
1897 	case TCPS_LAST_ACK:
1898 	case TCPS_TIME_WAIT:
1899 		tp = tcp_close(tp);
1900 		goto drop;
1901 	}
1902 
1903 	/*
1904 	 * If a SYN is in the window, then this is an
1905 	 * error and we send an RST and drop the connection.
1906 	 */
1907 	if (tiflags & TH_SYN) {
1908 		tp = tcp_drop(tp, ECONNRESET);
1909 		goto dropwithreset;
1910 	}
1911 
1912 	/*
1913 	 * If the ACK bit is off we drop the segment and return.
1914 	 */
1915 	if ((tiflags & TH_ACK) == 0) {
1916 		if (tp->t_flags & TF_ACKNOW)
1917 			goto dropafterack;
1918 		else
1919 			goto drop;
1920 	}
1921 
1922 	/*
1923 	 * Ack processing.
1924 	 */
1925 	switch (tp->t_state) {
1926 
1927 	/*
1928 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1929 	 * ESTABLISHED state and continue processing, otherwise
1930 	 * send an RST.
1931 	 */
1932 	case TCPS_SYN_RECEIVED:
1933 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
1934 		    SEQ_GT(th->th_ack, tp->snd_max))
1935 			goto dropwithreset;
1936 		tcpstat.tcps_connects++;
1937 		soisconnected(so);
1938 		tcp_established(tp);
1939 		/* Do window scaling? */
1940 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1941 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1942 			tp->snd_scale = tp->requested_s_scale;
1943 			tp->rcv_scale = tp->request_r_scale;
1944 		}
1945 		TCP_REASS_LOCK(tp);
1946 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1947 		TCP_REASS_UNLOCK(tp);
1948 		tp->snd_wl1 = th->th_seq - 1;
1949 		/* fall into ... */
1950 
1951 	/*
1952 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1953 	 * ACKs.  If the ack is in the range
1954 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1955 	 * then advance tp->snd_una to th->th_ack and drop
1956 	 * data from the retransmission queue.  If this ACK reflects
1957 	 * more up to date window information we update our window information.
1958 	 */
1959 	case TCPS_ESTABLISHED:
1960 	case TCPS_FIN_WAIT_1:
1961 	case TCPS_FIN_WAIT_2:
1962 	case TCPS_CLOSE_WAIT:
1963 	case TCPS_CLOSING:
1964 	case TCPS_LAST_ACK:
1965 	case TCPS_TIME_WAIT:
1966 
1967 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1968 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1969 				tcpstat.tcps_rcvdupack++;
1970 				/*
1971 				 * If we have outstanding data (other than
1972 				 * a window probe), this is a completely
1973 				 * duplicate ack (ie, window info didn't
1974 				 * change), the ack is the biggest we've
1975 				 * seen and we've seen exactly our rexmt
1976 				 * threshhold of them, assume a packet
1977 				 * has been dropped and retransmit it.
1978 				 * Kludge snd_nxt & the congestion
1979 				 * window so we send only this one
1980 				 * packet.
1981 				 *
1982 				 * We know we're losing at the current
1983 				 * window size so do congestion avoidance
1984 				 * (set ssthresh to half the current window
1985 				 * and pull our congestion window back to
1986 				 * the new ssthresh).
1987 				 *
1988 				 * Dup acks mean that packets have left the
1989 				 * network (they're now cached at the receiver)
1990 				 * so bump cwnd by the amount in the receiver
1991 				 * to keep a constant cwnd packets in the
1992 				 * network.
1993 				 */
1994 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
1995 				    th->th_ack != tp->snd_una)
1996 					tp->t_dupacks = 0;
1997 				else if (++tp->t_dupacks == tcprexmtthresh) {
1998 					tcp_seq onxt = tp->snd_nxt;
1999 					u_int win =
2000 					    min(tp->snd_wnd, tp->snd_cwnd) /
2001 					    2 /	tp->t_segsz;
2002 					if (tcp_do_newreno && SEQ_LT(th->th_ack,
2003 					    tp->snd_recover)) {
2004 						/*
2005 						 * False fast retransmit after
2006 						 * timeout.  Do not cut window.
2007 						 */
2008 						tp->snd_cwnd += tp->t_segsz;
2009 						tp->t_dupacks = 0;
2010 						(void) tcp_output(tp);
2011 						goto drop;
2012 					}
2013 
2014 					if (win < 2)
2015 						win = 2;
2016 					tp->snd_ssthresh = win * tp->t_segsz;
2017 					tp->snd_recover = tp->snd_max;
2018 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
2019 					tp->t_rtttime = 0;
2020 					tp->snd_nxt = th->th_ack;
2021 					tp->snd_cwnd = tp->t_segsz;
2022 					(void) tcp_output(tp);
2023 					tp->snd_cwnd = tp->snd_ssthresh +
2024 					       tp->t_segsz * tp->t_dupacks;
2025 					if (SEQ_GT(onxt, tp->snd_nxt))
2026 						tp->snd_nxt = onxt;
2027 					goto drop;
2028 				} else if (tp->t_dupacks > tcprexmtthresh) {
2029 					tp->snd_cwnd += tp->t_segsz;
2030 					(void) tcp_output(tp);
2031 					goto drop;
2032 				}
2033 			} else
2034 				tp->t_dupacks = 0;
2035 			break;
2036 		}
2037 		/*
2038 		 * If the congestion window was inflated to account
2039 		 * for the other side's cached packets, retract it.
2040 		 */
2041 		if (tcp_do_newreno == 0) {
2042 			if (tp->t_dupacks >= tcprexmtthresh &&
2043 			    tp->snd_cwnd > tp->snd_ssthresh)
2044 				tp->snd_cwnd = tp->snd_ssthresh;
2045 			tp->t_dupacks = 0;
2046 		} else if (tp->t_dupacks >= tcprexmtthresh &&
2047 			   tcp_newreno(tp, th) == 0) {
2048 			tp->snd_cwnd = tp->snd_ssthresh;
2049 			/*
2050 			 * Window inflation should have left us with approx.
2051 			 * snd_ssthresh outstanding data.  But in case we
2052 			 * would be inclined to send a burst, better to do
2053 			 * it via the slow start mechanism.
2054 			 */
2055 			if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
2056 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
2057 				    + tp->t_segsz;
2058 			tp->t_dupacks = 0;
2059 		}
2060 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2061 			tcpstat.tcps_rcvacktoomuch++;
2062 			goto dropafterack;
2063 		}
2064 		acked = th->th_ack - tp->snd_una;
2065 		tcpstat.tcps_rcvackpack++;
2066 		tcpstat.tcps_rcvackbyte += acked;
2067 
2068 		/*
2069 		 * If we have a timestamp reply, update smoothed
2070 		 * round trip time.  If no timestamp is present but
2071 		 * transmit timer is running and timed sequence
2072 		 * number was acked, update smoothed round trip time.
2073 		 * Since we now have an rtt measurement, cancel the
2074 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2075 		 * Recompute the initial retransmit timer.
2076 		 */
2077 		if (opti.ts_present && opti.ts_ecr)
2078 			tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
2079 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2080 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2081 
2082 		/*
2083 		 * If all outstanding data is acked, stop retransmit
2084 		 * timer and remember to restart (more output or persist).
2085 		 * If there is more data to be acked, restart retransmit
2086 		 * timer, using current (possibly backed-off) value.
2087 		 */
2088 		if (th->th_ack == tp->snd_max) {
2089 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2090 			needoutput = 1;
2091 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2092 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2093 		/*
2094 		 * When new data is acked, open the congestion window.
2095 		 * If the window gives us less than ssthresh packets
2096 		 * in flight, open exponentially (segsz per packet).
2097 		 * Otherwise open linearly: segsz per window
2098 		 * (segsz^2 / cwnd per packet), plus a constant
2099 		 * fraction of a packet (segsz/8) to help larger windows
2100 		 * open quickly enough.
2101 		 */
2102 		{
2103 		u_int cw = tp->snd_cwnd;
2104 		u_int incr = tp->t_segsz;
2105 
2106 		if (cw > tp->snd_ssthresh)
2107 			incr = incr * incr / cw;
2108 		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
2109 			tp->snd_cwnd = min(cw + incr,
2110 			    TCP_MAXWIN << tp->snd_scale);
2111 		}
2112 		ND6_HINT(tp);
2113 		if (acked > so->so_snd.sb_cc) {
2114 			tp->snd_wnd -= so->so_snd.sb_cc;
2115 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2116 			ourfinisacked = 1;
2117 		} else {
2118 			if (acked > (tp->t_lastoff - tp->t_inoff))
2119 				tp->t_lastm = NULL;
2120 			sbdrop(&so->so_snd, acked);
2121 			tp->t_lastoff -= acked;
2122 			ourfinisacked = 0;
2123 		}
2124 		sowwakeup(so);
2125 		/*
2126 		 * We want snd_recover to track snd_una to
2127 		 * avoid sequence wraparound problems for
2128 		 * very large transfers.
2129 		 */
2130 		tp->snd_una = tp->snd_recover = th->th_ack;
2131 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2132 			tp->snd_nxt = tp->snd_una;
2133 
2134 		switch (tp->t_state) {
2135 
2136 		/*
2137 		 * In FIN_WAIT_1 STATE in addition to the processing
2138 		 * for the ESTABLISHED state if our FIN is now acknowledged
2139 		 * then enter FIN_WAIT_2.
2140 		 */
2141 		case TCPS_FIN_WAIT_1:
2142 			if (ourfinisacked) {
2143 				/*
2144 				 * If we can't receive any more
2145 				 * data, then closing user can proceed.
2146 				 * Starting the timer is contrary to the
2147 				 * specification, but if we don't get a FIN
2148 				 * we'll hang forever.
2149 				 */
2150 				if (so->so_state & SS_CANTRCVMORE) {
2151 					soisdisconnected(so);
2152 					if (tcp_maxidle > 0)
2153 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2154 						    tcp_maxidle);
2155 				}
2156 				tp->t_state = TCPS_FIN_WAIT_2;
2157 			}
2158 			break;
2159 
2160 	 	/*
2161 		 * In CLOSING STATE in addition to the processing for
2162 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2163 		 * then enter the TIME-WAIT state, otherwise ignore
2164 		 * the segment.
2165 		 */
2166 		case TCPS_CLOSING:
2167 			if (ourfinisacked) {
2168 				tp->t_state = TCPS_TIME_WAIT;
2169 				tcp_canceltimers(tp);
2170 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2171 				soisdisconnected(so);
2172 			}
2173 			break;
2174 
2175 		/*
2176 		 * In LAST_ACK, we may still be waiting for data to drain
2177 		 * and/or to be acked, as well as for the ack of our FIN.
2178 		 * If our FIN is now acknowledged, delete the TCB,
2179 		 * enter the closed state and return.
2180 		 */
2181 		case TCPS_LAST_ACK:
2182 			if (ourfinisacked) {
2183 				tp = tcp_close(tp);
2184 				goto drop;
2185 			}
2186 			break;
2187 
2188 		/*
2189 		 * In TIME_WAIT state the only thing that should arrive
2190 		 * is a retransmission of the remote FIN.  Acknowledge
2191 		 * it and restart the finack timer.
2192 		 */
2193 		case TCPS_TIME_WAIT:
2194 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2195 			goto dropafterack;
2196 		}
2197 	}
2198 
2199 step6:
2200 	/*
2201 	 * Update window information.
2202 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2203 	 */
2204 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2205 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2206 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2207 		/* keep track of pure window updates */
2208 		if (tlen == 0 &&
2209 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2210 			tcpstat.tcps_rcvwinupd++;
2211 		tp->snd_wnd = tiwin;
2212 		tp->snd_wl1 = th->th_seq;
2213 		tp->snd_wl2 = th->th_ack;
2214 		if (tp->snd_wnd > tp->max_sndwnd)
2215 			tp->max_sndwnd = tp->snd_wnd;
2216 		needoutput = 1;
2217 	}
2218 
2219 	/*
2220 	 * Process segments with URG.
2221 	 */
2222 	if ((tiflags & TH_URG) && th->th_urp &&
2223 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2224 		/*
2225 		 * This is a kludge, but if we receive and accept
2226 		 * random urgent pointers, we'll crash in
2227 		 * soreceive.  It's hard to imagine someone
2228 		 * actually wanting to send this much urgent data.
2229 		 */
2230 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2231 			th->th_urp = 0;			/* XXX */
2232 			tiflags &= ~TH_URG;		/* XXX */
2233 			goto dodata;			/* XXX */
2234 		}
2235 		/*
2236 		 * If this segment advances the known urgent pointer,
2237 		 * then mark the data stream.  This should not happen
2238 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2239 		 * a FIN has been received from the remote side.
2240 		 * In these states we ignore the URG.
2241 		 *
2242 		 * According to RFC961 (Assigned Protocols),
2243 		 * the urgent pointer points to the last octet
2244 		 * of urgent data.  We continue, however,
2245 		 * to consider it to indicate the first octet
2246 		 * of data past the urgent section as the original
2247 		 * spec states (in one of two places).
2248 		 */
2249 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2250 			tp->rcv_up = th->th_seq + th->th_urp;
2251 			so->so_oobmark = so->so_rcv.sb_cc +
2252 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2253 			if (so->so_oobmark == 0)
2254 				so->so_state |= SS_RCVATMARK;
2255 			sohasoutofband(so);
2256 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2257 		}
2258 		/*
2259 		 * Remove out of band data so doesn't get presented to user.
2260 		 * This can happen independent of advancing the URG pointer,
2261 		 * but if two URG's are pending at once, some out-of-band
2262 		 * data may creep in... ick.
2263 		 */
2264 		if (th->th_urp <= (u_int16_t) tlen
2265 #ifdef SO_OOBINLINE
2266 		     && (so->so_options & SO_OOBINLINE) == 0
2267 #endif
2268 		     )
2269 			tcp_pulloutofband(so, th, m, hdroptlen);
2270 	} else
2271 		/*
2272 		 * If no out of band data is expected,
2273 		 * pull receive urgent pointer along
2274 		 * with the receive window.
2275 		 */
2276 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2277 			tp->rcv_up = tp->rcv_nxt;
2278 dodata:							/* XXX */
2279 
2280 	/*
2281 	 * Process the segment text, merging it into the TCP sequencing queue,
2282 	 * and arranging for acknowledgement of receipt if necessary.
2283 	 * This process logically involves adjusting tp->rcv_wnd as data
2284 	 * is presented to the user (this happens in tcp_usrreq.c,
2285 	 * case PRU_RCVD).  If a FIN has already been received on this
2286 	 * connection then we just ignore the text.
2287 	 */
2288 	if ((tlen || (tiflags & TH_FIN)) &&
2289 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2290 		/*
2291 		 * Insert segment ti into reassembly queue of tcp with
2292 		 * control block tp.  Return TH_FIN if reassembly now includes
2293 		 * a segment with FIN.  The macro form does the common case
2294 		 * inline (segment is the next to be received on an
2295 		 * established connection, and the queue is empty),
2296 		 * avoiding linkage into and removal from the queue and
2297 		 * repetition of various conversions.
2298 		 * Set DELACK for segments received in order, but ack
2299 		 * immediately when segments are out of order
2300 		 * (so fast retransmit can work).
2301 		 */
2302 		/* NOTE: this was TCP_REASS() macro, but used only once */
2303 		TCP_REASS_LOCK(tp);
2304 		if (th->th_seq == tp->rcv_nxt &&
2305 		    TAILQ_FIRST(&tp->segq) == NULL &&
2306 		    tp->t_state == TCPS_ESTABLISHED) {
2307 			TCP_SETUP_ACK(tp, th);
2308 			tp->rcv_nxt += tlen;
2309 			tiflags = th->th_flags & TH_FIN;
2310 			tcpstat.tcps_rcvpack++;
2311 			tcpstat.tcps_rcvbyte += tlen;
2312 			ND6_HINT(tp);
2313 			if (so->so_state & SS_CANTRCVMORE)
2314 				m_freem(m);
2315 			else {
2316 				m_adj(m, hdroptlen);
2317 				sbappendstream(&(so)->so_rcv, m);
2318 			}
2319 			sorwakeup(so);
2320 		} else {
2321 			m_adj(m, hdroptlen);
2322 			tiflags = tcp_reass(tp, th, m, &tlen);
2323 			tp->t_flags |= TF_ACKNOW;
2324 		}
2325 		TCP_REASS_UNLOCK(tp);
2326 
2327 		/*
2328 		 * Note the amount of data that peer has sent into
2329 		 * our window, in order to estimate the sender's
2330 		 * buffer size.
2331 		 */
2332 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2333 	} else {
2334 		m_freem(m);
2335 		m = NULL;
2336 		tiflags &= ~TH_FIN;
2337 	}
2338 
2339 	/*
2340 	 * If FIN is received ACK the FIN and let the user know
2341 	 * that the connection is closing.  Ignore a FIN received before
2342 	 * the connection is fully established.
2343 	 */
2344 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2345 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2346 			socantrcvmore(so);
2347 			tp->t_flags |= TF_ACKNOW;
2348 			tp->rcv_nxt++;
2349 		}
2350 		switch (tp->t_state) {
2351 
2352 	 	/*
2353 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2354 		 */
2355 		case TCPS_ESTABLISHED:
2356 			tp->t_state = TCPS_CLOSE_WAIT;
2357 			break;
2358 
2359 	 	/*
2360 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2361 		 * enter the CLOSING state.
2362 		 */
2363 		case TCPS_FIN_WAIT_1:
2364 			tp->t_state = TCPS_CLOSING;
2365 			break;
2366 
2367 	 	/*
2368 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2369 		 * starting the time-wait timer, turning off the other
2370 		 * standard timers.
2371 		 */
2372 		case TCPS_FIN_WAIT_2:
2373 			tp->t_state = TCPS_TIME_WAIT;
2374 			tcp_canceltimers(tp);
2375 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2376 			soisdisconnected(so);
2377 			break;
2378 
2379 		/*
2380 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2381 		 */
2382 		case TCPS_TIME_WAIT:
2383 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2384 			break;
2385 		}
2386 	}
2387 #ifdef TCP_DEBUG
2388 	if (so->so_options & SO_DEBUG)
2389 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2390 #endif
2391 
2392 	/*
2393 	 * Return any desired output.
2394 	 */
2395 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2396 		(void) tcp_output(tp);
2397 	if (tcp_saveti)
2398 		m_freem(tcp_saveti);
2399 	return;
2400 
2401 badsyn:
2402 	/*
2403 	 * Received a bad SYN.  Increment counters and dropwithreset.
2404 	 */
2405 	tcpstat.tcps_badsyn++;
2406 	tp = NULL;
2407 	goto dropwithreset;
2408 
2409 dropafterack:
2410 	/*
2411 	 * Generate an ACK dropping incoming segment if it occupies
2412 	 * sequence space, where the ACK reflects our state.
2413 	 */
2414 	if (tiflags & TH_RST)
2415 		goto drop;
2416 	m_freem(m);
2417 	tp->t_flags |= TF_ACKNOW;
2418 	(void) tcp_output(tp);
2419 	if (tcp_saveti)
2420 		m_freem(tcp_saveti);
2421 	return;
2422 
2423 dropwithreset_ratelim:
2424 	/*
2425 	 * We may want to rate-limit RSTs in certain situations,
2426 	 * particularly if we are sending an RST in response to
2427 	 * an attempt to connect to or otherwise communicate with
2428 	 * a port for which we have no socket.
2429 	 */
2430 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2431 	    tcp_rst_ppslim) == 0) {
2432 		/* XXX stat */
2433 		goto drop;
2434 	}
2435 	/* ...fall into dropwithreset... */
2436 
2437 dropwithreset:
2438 	/*
2439 	 * Generate a RST, dropping incoming segment.
2440 	 * Make ACK acceptable to originator of segment.
2441 	 */
2442 	if (tiflags & TH_RST)
2443 		goto drop;
2444 
2445 	switch (af) {
2446 #ifdef INET6
2447 	case AF_INET6:
2448 		/* For following calls to tcp_respond */
2449 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2450 			goto drop;
2451 		break;
2452 #endif /* INET6 */
2453 	case AF_INET:
2454 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2455 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2456 			goto drop;
2457 	}
2458 
2459 	if (tiflags & TH_ACK)
2460 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2461 	else {
2462 		if (tiflags & TH_SYN)
2463 			tlen++;
2464 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2465 		    TH_RST|TH_ACK);
2466 	}
2467 	if (tcp_saveti)
2468 		m_freem(tcp_saveti);
2469 	return;
2470 
2471 badcsum:
2472 	tcpstat.tcps_rcvbadsum++;
2473 drop:
2474 	/*
2475 	 * Drop space held by incoming segment and return.
2476 	 */
2477 	if (tp) {
2478 		if (tp->t_inpcb)
2479 			so = tp->t_inpcb->inp_socket;
2480 #ifdef INET6
2481 		else if (tp->t_in6pcb)
2482 			so = tp->t_in6pcb->in6p_socket;
2483 #endif
2484 		else
2485 			so = NULL;
2486 #ifdef TCP_DEBUG
2487 		if (so && (so->so_options & SO_DEBUG) != 0)
2488 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2489 #endif
2490 	}
2491 	if (tcp_saveti)
2492 		m_freem(tcp_saveti);
2493 	m_freem(m);
2494 	return;
2495 }
2496 
2497 void
2498 tcp_dooptions(tp, cp, cnt, th, oi)
2499 	struct tcpcb *tp;
2500 	u_char *cp;
2501 	int cnt;
2502 	struct tcphdr *th;
2503 	struct tcp_opt_info *oi;
2504 {
2505 	u_int16_t mss;
2506 	int opt, optlen;
2507 
2508 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2509 		opt = cp[0];
2510 		if (opt == TCPOPT_EOL)
2511 			break;
2512 		if (opt == TCPOPT_NOP)
2513 			optlen = 1;
2514 		else {
2515 			if (cnt < 2)
2516 				break;
2517 			optlen = cp[1];
2518 			if (optlen < 2 || optlen > cnt)
2519 				break;
2520 		}
2521 		switch (opt) {
2522 
2523 		default:
2524 			continue;
2525 
2526 		case TCPOPT_MAXSEG:
2527 			if (optlen != TCPOLEN_MAXSEG)
2528 				continue;
2529 			if (!(th->th_flags & TH_SYN))
2530 				continue;
2531 			bcopy(cp + 2, &mss, sizeof(mss));
2532 			oi->maxseg = ntohs(mss);
2533 			break;
2534 
2535 		case TCPOPT_WINDOW:
2536 			if (optlen != TCPOLEN_WINDOW)
2537 				continue;
2538 			if (!(th->th_flags & TH_SYN))
2539 				continue;
2540 			tp->t_flags |= TF_RCVD_SCALE;
2541 			tp->requested_s_scale = cp[2];
2542 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2543 #if 0	/*XXX*/
2544 				char *p;
2545 
2546 				if (ip)
2547 					p = ntohl(ip->ip_src);
2548 #ifdef INET6
2549 				else if (ip6)
2550 					p = ip6_sprintf(&ip6->ip6_src);
2551 #endif
2552 				else
2553 					p = "(unknown)";
2554 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2555 				    "assuming %d\n",
2556 				    tp->requested_s_scale, p,
2557 				    TCP_MAX_WINSHIFT);
2558 #else
2559 				log(LOG_ERR, "TCP: invalid wscale %d, "
2560 				    "assuming %d\n",
2561 				    tp->requested_s_scale,
2562 				    TCP_MAX_WINSHIFT);
2563 #endif
2564 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2565 			}
2566 			break;
2567 
2568 		case TCPOPT_TIMESTAMP:
2569 			if (optlen != TCPOLEN_TIMESTAMP)
2570 				continue;
2571 			oi->ts_present = 1;
2572 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2573 			NTOHL(oi->ts_val);
2574 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2575 			NTOHL(oi->ts_ecr);
2576 
2577 			/*
2578 			 * A timestamp received in a SYN makes
2579 			 * it ok to send timestamp requests and replies.
2580 			 */
2581 			if (th->th_flags & TH_SYN) {
2582 				tp->t_flags |= TF_RCVD_TSTMP;
2583 				tp->ts_recent = oi->ts_val;
2584 				tp->ts_recent_age = TCP_TIMESTAMP(tp);
2585 			}
2586 			break;
2587 		case TCPOPT_SACK_PERMITTED:
2588 			if (optlen != TCPOLEN_SACK_PERMITTED)
2589 				continue;
2590 			if (!(th->th_flags & TH_SYN))
2591 				continue;
2592 			tp->t_flags &= ~TF_CANT_TXSACK;
2593 			break;
2594 
2595 		case TCPOPT_SACK:
2596 			if (tp->t_flags & TF_IGNR_RXSACK)
2597 				continue;
2598 			if (optlen % 8 != 2 || optlen < 10)
2599 				continue;
2600 			cp += 2;
2601 			optlen -= 2;
2602 			for (; optlen > 0; cp -= 8, optlen -= 8) {
2603 				tcp_seq lwe, rwe;
2604 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
2605 				NTOHL(lwe);
2606 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
2607 				NTOHL(rwe);
2608 				/* tcp_mark_sacked(tp, lwe, rwe); */
2609 			}
2610 			break;
2611 		}
2612 	}
2613 }
2614 
2615 /*
2616  * Pull out of band byte out of a segment so
2617  * it doesn't appear in the user's data queue.
2618  * It is still reflected in the segment length for
2619  * sequencing purposes.
2620  */
2621 void
2622 tcp_pulloutofband(so, th, m, off)
2623 	struct socket *so;
2624 	struct tcphdr *th;
2625 	struct mbuf *m;
2626 	int off;
2627 {
2628 	int cnt = off + th->th_urp - 1;
2629 
2630 	while (cnt >= 0) {
2631 		if (m->m_len > cnt) {
2632 			char *cp = mtod(m, caddr_t) + cnt;
2633 			struct tcpcb *tp = sototcpcb(so);
2634 
2635 			tp->t_iobc = *cp;
2636 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2637 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2638 			m->m_len--;
2639 			return;
2640 		}
2641 		cnt -= m->m_len;
2642 		m = m->m_next;
2643 		if (m == 0)
2644 			break;
2645 	}
2646 	panic("tcp_pulloutofband");
2647 }
2648 
2649 /*
2650  * Collect new round-trip time estimate
2651  * and update averages and current timeout.
2652  */
2653 void
2654 tcp_xmit_timer(tp, rtt)
2655 	struct tcpcb *tp;
2656 	uint32_t rtt;
2657 {
2658 	int32_t delta;
2659 
2660 	tcpstat.tcps_rttupdated++;
2661 	if (tp->t_srtt != 0) {
2662 		/*
2663 		 * srtt is stored as fixed point with 3 bits after the
2664 		 * binary point (i.e., scaled by 8).  The following magic
2665 		 * is equivalent to the smoothing algorithm in rfc793 with
2666 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2667 		 * point).  Adjust rtt to origin 0.
2668 		 */
2669 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2670 		if ((tp->t_srtt += delta) <= 0)
2671 			tp->t_srtt = 1 << 2;
2672 		/*
2673 		 * We accumulate a smoothed rtt variance (actually, a
2674 		 * smoothed mean difference), then set the retransmit
2675 		 * timer to smoothed rtt + 4 times the smoothed variance.
2676 		 * rttvar is stored as fixed point with 2 bits after the
2677 		 * binary point (scaled by 4).  The following is
2678 		 * equivalent to rfc793 smoothing with an alpha of .75
2679 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2680 		 * rfc793's wired-in beta.
2681 		 */
2682 		if (delta < 0)
2683 			delta = -delta;
2684 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2685 		if ((tp->t_rttvar += delta) <= 0)
2686 			tp->t_rttvar = 1 << 2;
2687 	} else {
2688 		/*
2689 		 * No rtt measurement yet - use the unsmoothed rtt.
2690 		 * Set the variance to half the rtt (so our first
2691 		 * retransmit happens at 3*rtt).
2692 		 */
2693 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2694 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2695 	}
2696 	tp->t_rtttime = 0;
2697 	tp->t_rxtshift = 0;
2698 
2699 	/*
2700 	 * the retransmit should happen at rtt + 4 * rttvar.
2701 	 * Because of the way we do the smoothing, srtt and rttvar
2702 	 * will each average +1/2 tick of bias.  When we compute
2703 	 * the retransmit timer, we want 1/2 tick of rounding and
2704 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2705 	 * firing of the timer.  The bias will give us exactly the
2706 	 * 1.5 tick we need.  But, because the bias is
2707 	 * statistical, we have to test that we don't drop below
2708 	 * the minimum feasible timer (which is 2 ticks).
2709 	 */
2710 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2711 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2712 
2713 	/*
2714 	 * We received an ack for a packet that wasn't retransmitted;
2715 	 * it is probably safe to discard any error indications we've
2716 	 * received recently.  This isn't quite right, but close enough
2717 	 * for now (a route might have failed after we sent a segment,
2718 	 * and the return path might not be symmetrical).
2719 	 */
2720 	tp->t_softerror = 0;
2721 }
2722 
2723 /*
2724  * Checks for partial ack.  If partial ack arrives, force the retransmission
2725  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
2726  * 1.  By setting snd_nxt to th_ack, this forces retransmission timer to
2727  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
2728  */
2729 int
2730 tcp_newreno(tp, th)
2731 	struct tcpcb *tp;
2732 	struct tcphdr *th;
2733 {
2734 	tcp_seq onxt = tp->snd_nxt;
2735 	u_long ocwnd = tp->snd_cwnd;
2736 
2737 	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2738 		/*
2739 		 * snd_una has not yet been updated and the socket's send
2740 		 * buffer has not yet drained off the ACK'd data, so we
2741 		 * have to leave snd_una as it was to get the correct data
2742 		 * offset in tcp_output().
2743 		 */
2744 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2745 		tp->t_rtttime = 0;
2746 		tp->snd_nxt = th->th_ack;
2747 		/*
2748 		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
2749 		 * is not yet updated when we're called.
2750 		 */
2751 		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
2752 		(void) tcp_output(tp);
2753 		tp->snd_cwnd = ocwnd;
2754 		if (SEQ_GT(onxt, tp->snd_nxt))
2755 			tp->snd_nxt = onxt;
2756 		/*
2757 		 * Partial window deflation.  Relies on fact that tp->snd_una
2758 		 * not updated yet.
2759 		 */
2760 		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
2761 		return 1;
2762 	}
2763 	return 0;
2764 }
2765 
2766 
2767 /*
2768  * TCP compressed state engine.  Currently used to hold compressed
2769  * state for SYN_RECEIVED.
2770  */
2771 
2772 u_long	syn_cache_count;
2773 u_int32_t syn_hash1, syn_hash2;
2774 
2775 #define SYN_HASH(sa, sp, dp) \
2776 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
2777 				     ((u_int32_t)(sp)))^syn_hash2)))
2778 #ifndef INET6
2779 #define	SYN_HASHALL(hash, src, dst) \
2780 do {									\
2781 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
2782 		((struct sockaddr_in *)(src))->sin_port,		\
2783 		((struct sockaddr_in *)(dst))->sin_port);		\
2784 } while (/*CONSTCOND*/ 0)
2785 #else
2786 #define SYN_HASH6(sa, sp, dp) \
2787 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
2788 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
2789 	 & 0x7fffffff)
2790 
2791 #define SYN_HASHALL(hash, src, dst) \
2792 do {									\
2793 	switch ((src)->sa_family) {					\
2794 	case AF_INET:							\
2795 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
2796 			((struct sockaddr_in *)(src))->sin_port,	\
2797 			((struct sockaddr_in *)(dst))->sin_port);	\
2798 		break;							\
2799 	case AF_INET6:							\
2800 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
2801 			((struct sockaddr_in6 *)(src))->sin6_port,	\
2802 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
2803 		break;							\
2804 	default:							\
2805 		hash = 0;						\
2806 	}								\
2807 } while (/*CONSTCOND*/0)
2808 #endif /* INET6 */
2809 
2810 #define	SYN_CACHE_RM(sc)						\
2811 do {									\
2812 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
2813 	    (sc), sc_bucketq);						\
2814 	(sc)->sc_tp = NULL;						\
2815 	LIST_REMOVE((sc), sc_tpq);					\
2816 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
2817 	callout_stop(&(sc)->sc_timer);					\
2818 	syn_cache_count--;						\
2819 } while (/*CONSTCOND*/0)
2820 
2821 #define	SYN_CACHE_PUT(sc)						\
2822 do {									\
2823 	if ((sc)->sc_ipopts)						\
2824 		(void) m_free((sc)->sc_ipopts);				\
2825 	if ((sc)->sc_route4.ro_rt != NULL)				\
2826 		RTFREE((sc)->sc_route4.ro_rt);				\
2827 	if (callout_invoking(&(sc)->sc_timer))				\
2828 		(sc)->sc_flags |= SCF_DEAD;				\
2829 	else								\
2830 		pool_put(&syn_cache_pool, (sc));			\
2831 } while (/*CONSTCOND*/0)
2832 
2833 struct pool syn_cache_pool;
2834 
2835 /*
2836  * We don't estimate RTT with SYNs, so each packet starts with the default
2837  * RTT and each timer step has a fixed timeout value.
2838  */
2839 #define	SYN_CACHE_TIMER_ARM(sc)						\
2840 do {									\
2841 	TCPT_RANGESET((sc)->sc_rxtcur,					\
2842 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
2843 	    TCPTV_REXMTMAX);						\
2844 	callout_reset(&(sc)->sc_timer,					\
2845 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
2846 } while (/*CONSTCOND*/0)
2847 
2848 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
2849 
2850 void
2851 syn_cache_init()
2852 {
2853 	int i;
2854 
2855 	/* Initialize the hash buckets. */
2856 	for (i = 0; i < tcp_syn_cache_size; i++)
2857 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
2858 
2859 	/* Initialize the syn cache pool. */
2860 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
2861 	    "synpl", NULL);
2862 }
2863 
2864 void
2865 syn_cache_insert(sc, tp)
2866 	struct syn_cache *sc;
2867 	struct tcpcb *tp;
2868 {
2869 	struct syn_cache_head *scp;
2870 	struct syn_cache *sc2;
2871 	int s;
2872 
2873 	/*
2874 	 * If there are no entries in the hash table, reinitialize
2875 	 * the hash secrets.
2876 	 */
2877 	if (syn_cache_count == 0) {
2878 		syn_hash1 = arc4random();
2879 		syn_hash2 = arc4random();
2880 	}
2881 
2882 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
2883 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
2884 	scp = &tcp_syn_cache[sc->sc_bucketidx];
2885 
2886 	/*
2887 	 * Make sure that we don't overflow the per-bucket
2888 	 * limit or the total cache size limit.
2889 	 */
2890 	s = splsoftnet();
2891 	if (scp->sch_length >= tcp_syn_bucket_limit) {
2892 		tcpstat.tcps_sc_bucketoverflow++;
2893 		/*
2894 		 * The bucket is full.  Toss the oldest element in the
2895 		 * bucket.  This will be the first entry in the bucket.
2896 		 */
2897 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
2898 #ifdef DIAGNOSTIC
2899 		/*
2900 		 * This should never happen; we should always find an
2901 		 * entry in our bucket.
2902 		 */
2903 		if (sc2 == NULL)
2904 			panic("syn_cache_insert: bucketoverflow: impossible");
2905 #endif
2906 		SYN_CACHE_RM(sc2);
2907 		SYN_CACHE_PUT(sc2);
2908 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
2909 		struct syn_cache_head *scp2, *sce;
2910 
2911 		tcpstat.tcps_sc_overflowed++;
2912 		/*
2913 		 * The cache is full.  Toss the oldest entry in the
2914 		 * first non-empty bucket we can find.
2915 		 *
2916 		 * XXX We would really like to toss the oldest
2917 		 * entry in the cache, but we hope that this
2918 		 * condition doesn't happen very often.
2919 		 */
2920 		scp2 = scp;
2921 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
2922 			sce = &tcp_syn_cache[tcp_syn_cache_size];
2923 			for (++scp2; scp2 != scp; scp2++) {
2924 				if (scp2 >= sce)
2925 					scp2 = &tcp_syn_cache[0];
2926 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
2927 					break;
2928 			}
2929 #ifdef DIAGNOSTIC
2930 			/*
2931 			 * This should never happen; we should always find a
2932 			 * non-empty bucket.
2933 			 */
2934 			if (scp2 == scp)
2935 				panic("syn_cache_insert: cacheoverflow: "
2936 				    "impossible");
2937 #endif
2938 		}
2939 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
2940 		SYN_CACHE_RM(sc2);
2941 		SYN_CACHE_PUT(sc2);
2942 	}
2943 
2944 	/*
2945 	 * Initialize the entry's timer.
2946 	 */
2947 	sc->sc_rxttot = 0;
2948 	sc->sc_rxtshift = 0;
2949 	SYN_CACHE_TIMER_ARM(sc);
2950 
2951 	/* Link it from tcpcb entry */
2952 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
2953 
2954 	/* Put it into the bucket. */
2955 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
2956 	scp->sch_length++;
2957 	syn_cache_count++;
2958 
2959 	tcpstat.tcps_sc_added++;
2960 	splx(s);
2961 }
2962 
2963 /*
2964  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
2965  * If we have retransmitted an entry the maximum number of times, expire
2966  * that entry.
2967  */
2968 void
2969 syn_cache_timer(void *arg)
2970 {
2971 	struct syn_cache *sc = arg;
2972 	int s;
2973 
2974 	s = splsoftnet();
2975 	callout_ack(&sc->sc_timer);
2976 
2977 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
2978 		tcpstat.tcps_sc_delayed_free++;
2979 		pool_put(&syn_cache_pool, sc);
2980 		splx(s);
2981 		return;
2982 	}
2983 
2984 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
2985 		/* Drop it -- too many retransmissions. */
2986 		goto dropit;
2987 	}
2988 
2989 	/*
2990 	 * Compute the total amount of time this entry has
2991 	 * been on a queue.  If this entry has been on longer
2992 	 * than the keep alive timer would allow, expire it.
2993 	 */
2994 	sc->sc_rxttot += sc->sc_rxtcur;
2995 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
2996 		goto dropit;
2997 
2998 	tcpstat.tcps_sc_retransmitted++;
2999 	(void) syn_cache_respond(sc, NULL);
3000 
3001 	/* Advance the timer back-off. */
3002 	sc->sc_rxtshift++;
3003 	SYN_CACHE_TIMER_ARM(sc);
3004 
3005 	splx(s);
3006 	return;
3007 
3008  dropit:
3009 	tcpstat.tcps_sc_timed_out++;
3010 	SYN_CACHE_RM(sc);
3011 	SYN_CACHE_PUT(sc);
3012 	splx(s);
3013 }
3014 
3015 /*
3016  * Remove syn cache created by the specified tcb entry,
3017  * because this does not make sense to keep them
3018  * (if there's no tcb entry, syn cache entry will never be used)
3019  */
3020 void
3021 syn_cache_cleanup(tp)
3022 	struct tcpcb *tp;
3023 {
3024 	struct syn_cache *sc, *nsc;
3025 	int s;
3026 
3027 	s = splsoftnet();
3028 
3029 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3030 		nsc = LIST_NEXT(sc, sc_tpq);
3031 
3032 #ifdef DIAGNOSTIC
3033 		if (sc->sc_tp != tp)
3034 			panic("invalid sc_tp in syn_cache_cleanup");
3035 #endif
3036 		SYN_CACHE_RM(sc);
3037 		SYN_CACHE_PUT(sc);
3038 	}
3039 	/* just for safety */
3040 	LIST_INIT(&tp->t_sc);
3041 
3042 	splx(s);
3043 }
3044 
3045 /*
3046  * Find an entry in the syn cache.
3047  */
3048 struct syn_cache *
3049 syn_cache_lookup(src, dst, headp)
3050 	struct sockaddr *src;
3051 	struct sockaddr *dst;
3052 	struct syn_cache_head **headp;
3053 {
3054 	struct syn_cache *sc;
3055 	struct syn_cache_head *scp;
3056 	u_int32_t hash;
3057 	int s;
3058 
3059 	SYN_HASHALL(hash, src, dst);
3060 
3061 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3062 	*headp = scp;
3063 	s = splsoftnet();
3064 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3065 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3066 		if (sc->sc_hash != hash)
3067 			continue;
3068 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3069 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3070 			splx(s);
3071 			return (sc);
3072 		}
3073 	}
3074 	splx(s);
3075 	return (NULL);
3076 }
3077 
3078 /*
3079  * This function gets called when we receive an ACK for a
3080  * socket in the LISTEN state.  We look up the connection
3081  * in the syn cache, and if its there, we pull it out of
3082  * the cache and turn it into a full-blown connection in
3083  * the SYN-RECEIVED state.
3084  *
3085  * The return values may not be immediately obvious, and their effects
3086  * can be subtle, so here they are:
3087  *
3088  *	NULL	SYN was not found in cache; caller should drop the
3089  *		packet and send an RST.
3090  *
3091  *	-1	We were unable to create the new connection, and are
3092  *		aborting it.  An ACK,RST is being sent to the peer
3093  *		(unless we got screwey sequence numbners; see below),
3094  *		because the 3-way handshake has been completed.  Caller
3095  *		should not free the mbuf, since we may be using it.  If
3096  *		we are not, we will free it.
3097  *
3098  *	Otherwise, the return value is a pointer to the new socket
3099  *	associated with the connection.
3100  */
3101 struct socket *
3102 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3103 	struct sockaddr *src;
3104 	struct sockaddr *dst;
3105 	struct tcphdr *th;
3106 	unsigned int hlen, tlen;
3107 	struct socket *so;
3108 	struct mbuf *m;
3109 {
3110 	struct syn_cache *sc;
3111 	struct syn_cache_head *scp;
3112 	struct inpcb *inp = NULL;
3113 #ifdef INET6
3114 	struct in6pcb *in6p = NULL;
3115 #endif
3116 	struct tcpcb *tp = 0;
3117 	struct mbuf *am;
3118 	int s;
3119 	struct socket *oso;
3120 
3121 	s = splsoftnet();
3122 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3123 		splx(s);
3124 		return (NULL);
3125 	}
3126 
3127 	/*
3128 	 * Verify the sequence and ack numbers.  Try getting the correct
3129 	 * response again.
3130 	 */
3131 	if ((th->th_ack != sc->sc_iss + 1) ||
3132 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3133 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3134 		(void) syn_cache_respond(sc, m);
3135 		splx(s);
3136 		return ((struct socket *)(-1));
3137 	}
3138 
3139 	/* Remove this cache entry */
3140 	SYN_CACHE_RM(sc);
3141 	splx(s);
3142 
3143 	/*
3144 	 * Ok, create the full blown connection, and set things up
3145 	 * as they would have been set up if we had created the
3146 	 * connection when the SYN arrived.  If we can't create
3147 	 * the connection, abort it.
3148 	 */
3149 	/*
3150 	 * inp still has the OLD in_pcb stuff, set the
3151 	 * v6-related flags on the new guy, too.   This is
3152 	 * done particularly for the case where an AF_INET6
3153 	 * socket is bound only to a port, and a v4 connection
3154 	 * comes in on that port.
3155 	 * we also copy the flowinfo from the original pcb
3156 	 * to the new one.
3157 	 */
3158 	oso = so;
3159 	so = sonewconn(so, SS_ISCONNECTED);
3160 	if (so == NULL)
3161 		goto resetandabort;
3162 
3163 	switch (so->so_proto->pr_domain->dom_family) {
3164 #ifdef INET
3165 	case AF_INET:
3166 		inp = sotoinpcb(so);
3167 		break;
3168 #endif
3169 #ifdef INET6
3170 	case AF_INET6:
3171 		in6p = sotoin6pcb(so);
3172 		break;
3173 #endif
3174 	}
3175 	switch (src->sa_family) {
3176 #ifdef INET
3177 	case AF_INET:
3178 		if (inp) {
3179 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3180 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3181 			inp->inp_options = ip_srcroute();
3182 			in_pcbstate(inp, INP_BOUND);
3183 			if (inp->inp_options == NULL) {
3184 				inp->inp_options = sc->sc_ipopts;
3185 				sc->sc_ipopts = NULL;
3186 			}
3187 		}
3188 #ifdef INET6
3189 		else if (in6p) {
3190 			/* IPv4 packet to AF_INET6 socket */
3191 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3192 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3193 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3194 				&in6p->in6p_laddr.s6_addr32[3],
3195 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3196 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3197 			in6totcpcb(in6p)->t_family = AF_INET;
3198 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3199 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3200 			else
3201 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3202 			in6_pcbstate(in6p, IN6P_BOUND);
3203 		}
3204 #endif
3205 		break;
3206 #endif
3207 #ifdef INET6
3208 	case AF_INET6:
3209 		if (in6p) {
3210 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3211 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3212 			in6_pcbstate(in6p, IN6P_BOUND);
3213 		}
3214 		break;
3215 #endif
3216 	}
3217 #ifdef INET6
3218 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3219 		struct in6pcb *oin6p = sotoin6pcb(oso);
3220 		/* inherit socket options from the listening socket */
3221 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3222 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3223 			m_freem(in6p->in6p_options);
3224 			in6p->in6p_options = 0;
3225 		}
3226 		ip6_savecontrol(in6p, &in6p->in6p_options,
3227 			mtod(m, struct ip6_hdr *), m);
3228 	}
3229 #endif
3230 
3231 #if defined(IPSEC) || defined(FAST_IPSEC)
3232 	/*
3233 	 * we make a copy of policy, instead of sharing the policy,
3234 	 * for better behavior in terms of SA lookup and dead SA removal.
3235 	 */
3236 	if (inp) {
3237 		/* copy old policy into new socket's */
3238 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3239 			printf("tcp_input: could not copy policy\n");
3240 	}
3241 #ifdef INET6
3242 	else if (in6p) {
3243 		/* copy old policy into new socket's */
3244 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3245 		    in6p->in6p_sp))
3246 			printf("tcp_input: could not copy policy\n");
3247 	}
3248 #endif
3249 #endif
3250 
3251 	/*
3252 	 * Give the new socket our cached route reference.
3253 	 */
3254 	if (inp)
3255 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3256 #ifdef INET6
3257 	else
3258 		in6p->in6p_route = sc->sc_route6;
3259 #endif
3260 	sc->sc_route4.ro_rt = NULL;
3261 
3262 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3263 	if (am == NULL)
3264 		goto resetandabort;
3265 	MCLAIM(am, &tcp_mowner);
3266 	am->m_len = src->sa_len;
3267 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3268 	if (inp) {
3269 		if (in_pcbconnect(inp, am)) {
3270 			(void) m_free(am);
3271 			goto resetandabort;
3272 		}
3273 	}
3274 #ifdef INET6
3275 	else if (in6p) {
3276 		if (src->sa_family == AF_INET) {
3277 			/* IPv4 packet to AF_INET6 socket */
3278 			struct sockaddr_in6 *sin6;
3279 			sin6 = mtod(am, struct sockaddr_in6 *);
3280 			am->m_len = sizeof(*sin6);
3281 			bzero(sin6, sizeof(*sin6));
3282 			sin6->sin6_family = AF_INET6;
3283 			sin6->sin6_len = sizeof(*sin6);
3284 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3285 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3286 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3287 				&sin6->sin6_addr.s6_addr32[3],
3288 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3289 		}
3290 		if (in6_pcbconnect(in6p, am)) {
3291 			(void) m_free(am);
3292 			goto resetandabort;
3293 		}
3294 	}
3295 #endif
3296 	else {
3297 		(void) m_free(am);
3298 		goto resetandabort;
3299 	}
3300 	(void) m_free(am);
3301 
3302 	if (inp)
3303 		tp = intotcpcb(inp);
3304 #ifdef INET6
3305 	else if (in6p)
3306 		tp = in6totcpcb(in6p);
3307 #endif
3308 	else
3309 		tp = NULL;
3310 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3311 	if (sc->sc_request_r_scale != 15) {
3312 		tp->requested_s_scale = sc->sc_requested_s_scale;
3313 		tp->request_r_scale = sc->sc_request_r_scale;
3314 		tp->snd_scale = sc->sc_requested_s_scale;
3315 		tp->rcv_scale = sc->sc_request_r_scale;
3316 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3317 	}
3318 	if (sc->sc_flags & SCF_TIMESTAMP)
3319 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3320 	tp->ts_timebase = sc->sc_timebase;
3321 
3322 	tp->t_template = tcp_template(tp);
3323 	if (tp->t_template == 0) {
3324 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3325 		so = NULL;
3326 		m_freem(m);
3327 		goto abort;
3328 	}
3329 
3330 	tp->iss = sc->sc_iss;
3331 	tp->irs = sc->sc_irs;
3332 	tcp_sendseqinit(tp);
3333 	tcp_rcvseqinit(tp);
3334 	tp->t_state = TCPS_SYN_RECEIVED;
3335 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3336 	tcpstat.tcps_accepts++;
3337 
3338 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3339 	tp->t_ourmss = sc->sc_ourmaxseg;
3340 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3341 
3342 	/*
3343 	 * Initialize the initial congestion window.  If we
3344 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3345 	 * to 1 segment (i.e. the Loss Window).
3346 	 */
3347 	if (sc->sc_rxtshift)
3348 		tp->snd_cwnd = tp->t_peermss;
3349 	else {
3350 		int ss = tcp_init_win;
3351 #ifdef INET
3352 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3353 			ss = tcp_init_win_local;
3354 #endif
3355 #ifdef INET6
3356 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3357 			ss = tcp_init_win_local;
3358 #endif
3359 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3360 	}
3361 
3362 	tcp_rmx_rtt(tp);
3363 	tp->snd_wl1 = sc->sc_irs;
3364 	tp->rcv_up = sc->sc_irs + 1;
3365 
3366 	/*
3367 	 * This is what whould have happened in tcp_output() when
3368 	 * the SYN,ACK was sent.
3369 	 */
3370 	tp->snd_up = tp->snd_una;
3371 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3372 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3373 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3374 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3375 	tp->last_ack_sent = tp->rcv_nxt;
3376 
3377 	tcpstat.tcps_sc_completed++;
3378 	SYN_CACHE_PUT(sc);
3379 	return (so);
3380 
3381 resetandabort:
3382 	(void) tcp_respond(NULL, m, m, th,
3383 			   th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
3384 abort:
3385 	if (so != NULL)
3386 		(void) soabort(so);
3387 	SYN_CACHE_PUT(sc);
3388 	tcpstat.tcps_sc_aborted++;
3389 	return ((struct socket *)(-1));
3390 }
3391 
3392 /*
3393  * This function is called when we get a RST for a
3394  * non-existent connection, so that we can see if the
3395  * connection is in the syn cache.  If it is, zap it.
3396  */
3397 
3398 void
3399 syn_cache_reset(src, dst, th)
3400 	struct sockaddr *src;
3401 	struct sockaddr *dst;
3402 	struct tcphdr *th;
3403 {
3404 	struct syn_cache *sc;
3405 	struct syn_cache_head *scp;
3406 	int s = splsoftnet();
3407 
3408 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3409 		splx(s);
3410 		return;
3411 	}
3412 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3413 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3414 		splx(s);
3415 		return;
3416 	}
3417 	SYN_CACHE_RM(sc);
3418 	splx(s);
3419 	tcpstat.tcps_sc_reset++;
3420 	SYN_CACHE_PUT(sc);
3421 }
3422 
3423 void
3424 syn_cache_unreach(src, dst, th)
3425 	struct sockaddr *src;
3426 	struct sockaddr *dst;
3427 	struct tcphdr *th;
3428 {
3429 	struct syn_cache *sc;
3430 	struct syn_cache_head *scp;
3431 	int s;
3432 
3433 	s = splsoftnet();
3434 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3435 		splx(s);
3436 		return;
3437 	}
3438 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3439 	if (ntohl (th->th_seq) != sc->sc_iss) {
3440 		splx(s);
3441 		return;
3442 	}
3443 
3444 	/*
3445 	 * If we've retransmitted 3 times and this is our second error,
3446 	 * we remove the entry.  Otherwise, we allow it to continue on.
3447 	 * This prevents us from incorrectly nuking an entry during a
3448 	 * spurious network outage.
3449 	 *
3450 	 * See tcp_notify().
3451 	 */
3452 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3453 		sc->sc_flags |= SCF_UNREACH;
3454 		splx(s);
3455 		return;
3456 	}
3457 
3458 	SYN_CACHE_RM(sc);
3459 	splx(s);
3460 	tcpstat.tcps_sc_unreach++;
3461 	SYN_CACHE_PUT(sc);
3462 }
3463 
3464 /*
3465  * Given a LISTEN socket and an inbound SYN request, add
3466  * this to the syn cache, and send back a segment:
3467  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3468  * to the source.
3469  *
3470  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3471  * Doing so would require that we hold onto the data and deliver it
3472  * to the application.  However, if we are the target of a SYN-flood
3473  * DoS attack, an attacker could send data which would eventually
3474  * consume all available buffer space if it were ACKed.  By not ACKing
3475  * the data, we avoid this DoS scenario.
3476  */
3477 
3478 int
3479 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi)
3480 	struct sockaddr *src;
3481 	struct sockaddr *dst;
3482 	struct tcphdr *th;
3483 	unsigned int hlen;
3484 	struct socket *so;
3485 	struct mbuf *m;
3486 	u_char *optp;
3487 	int optlen;
3488 	struct tcp_opt_info *oi;
3489 {
3490 	struct tcpcb tb, *tp;
3491 	long win;
3492 	struct syn_cache *sc;
3493 	struct syn_cache_head *scp;
3494 	struct mbuf *ipopts;
3495 
3496 	tp = sototcpcb(so);
3497 
3498 	/*
3499 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3500 	 *
3501 	 * Note this check is performed in tcp_input() very early on.
3502 	 */
3503 
3504 	/*
3505 	 * Initialize some local state.
3506 	 */
3507 	win = sbspace(&so->so_rcv);
3508 	if (win > TCP_MAXWIN)
3509 		win = TCP_MAXWIN;
3510 
3511 	switch (src->sa_family) {
3512 #ifdef INET
3513 	case AF_INET:
3514 		/*
3515 		 * Remember the IP options, if any.
3516 		 */
3517 		ipopts = ip_srcroute();
3518 		break;
3519 #endif
3520 	default:
3521 		ipopts = NULL;
3522 	}
3523 
3524 	if (optp) {
3525 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3526 		tcp_dooptions(&tb, optp, optlen, th, oi);
3527 	} else
3528 		tb.t_flags = 0;
3529 
3530 	/*
3531 	 * See if we already have an entry for this connection.
3532 	 * If we do, resend the SYN,ACK.  We do not count this
3533 	 * as a retransmission (XXX though maybe we should).
3534 	 */
3535 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3536 		tcpstat.tcps_sc_dupesyn++;
3537 		if (ipopts) {
3538 			/*
3539 			 * If we were remembering a previous source route,
3540 			 * forget it and use the new one we've been given.
3541 			 */
3542 			if (sc->sc_ipopts)
3543 				(void) m_free(sc->sc_ipopts);
3544 			sc->sc_ipopts = ipopts;
3545 		}
3546 		sc->sc_timestamp = tb.ts_recent;
3547 		if (syn_cache_respond(sc, m) == 0) {
3548 			tcpstat.tcps_sndacks++;
3549 			tcpstat.tcps_sndtotal++;
3550 		}
3551 		return (1);
3552 	}
3553 
3554 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3555 	if (sc == NULL) {
3556 		if (ipopts)
3557 			(void) m_free(ipopts);
3558 		return (0);
3559 	}
3560 
3561 	/*
3562 	 * Fill in the cache, and put the necessary IP and TCP
3563 	 * options into the reply.
3564 	 */
3565 	bzero(sc, sizeof(struct syn_cache));
3566 	callout_init(&sc->sc_timer);
3567 	bcopy(src, &sc->sc_src, src->sa_len);
3568 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3569 	sc->sc_flags = 0;
3570 	sc->sc_ipopts = ipopts;
3571 	sc->sc_irs = th->th_seq;
3572 	switch (src->sa_family) {
3573 #ifdef INET
3574 	case AF_INET:
3575 	    {
3576 		struct sockaddr_in *srcin = (void *) src;
3577 		struct sockaddr_in *dstin = (void *) dst;
3578 
3579 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3580 		    &srcin->sin_addr, dstin->sin_port,
3581 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3582 		break;
3583 	    }
3584 #endif /* INET */
3585 #ifdef INET6
3586 	case AF_INET6:
3587 	    {
3588 		struct sockaddr_in6 *srcin6 = (void *) src;
3589 		struct sockaddr_in6 *dstin6 = (void *) dst;
3590 
3591 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3592 		    &srcin6->sin6_addr, dstin6->sin6_port,
3593 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3594 		break;
3595 	    }
3596 #endif /* INET6 */
3597 	}
3598 	sc->sc_peermaxseg = oi->maxseg;
3599 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
3600 						m->m_pkthdr.rcvif : NULL,
3601 						sc->sc_src.sa.sa_family);
3602 	sc->sc_win = win;
3603 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
3604 	sc->sc_timestamp = tb.ts_recent;
3605 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
3606 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
3607 		sc->sc_flags |= SCF_TIMESTAMP;
3608 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
3609 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
3610 		sc->sc_requested_s_scale = tb.requested_s_scale;
3611 		sc->sc_request_r_scale = 0;
3612 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
3613 		    TCP_MAXWIN << sc->sc_request_r_scale <
3614 		    so->so_rcv.sb_hiwat)
3615 			sc->sc_request_r_scale++;
3616 	} else {
3617 		sc->sc_requested_s_scale = 15;
3618 		sc->sc_request_r_scale = 15;
3619 	}
3620 	sc->sc_tp = tp;
3621 	if (syn_cache_respond(sc, m) == 0) {
3622 		syn_cache_insert(sc, tp);
3623 		tcpstat.tcps_sndacks++;
3624 		tcpstat.tcps_sndtotal++;
3625 	} else {
3626 		SYN_CACHE_PUT(sc);
3627 		tcpstat.tcps_sc_dropped++;
3628 	}
3629 	return (1);
3630 }
3631 
3632 int
3633 syn_cache_respond(sc, m)
3634 	struct syn_cache *sc;
3635 	struct mbuf *m;
3636 {
3637 	struct route *ro;
3638 	u_int8_t *optp;
3639 	int optlen, error;
3640 	u_int16_t tlen;
3641 	struct ip *ip = NULL;
3642 #ifdef INET6
3643 	struct ip6_hdr *ip6 = NULL;
3644 #endif
3645 	struct tcpcb *tp;
3646 	struct tcphdr *th;
3647 	u_int hlen;
3648 	struct socket *so;
3649 
3650 	switch (sc->sc_src.sa.sa_family) {
3651 	case AF_INET:
3652 		hlen = sizeof(struct ip);
3653 		ro = &sc->sc_route4;
3654 		break;
3655 #ifdef INET6
3656 	case AF_INET6:
3657 		hlen = sizeof(struct ip6_hdr);
3658 		ro = (struct route *)&sc->sc_route6;
3659 		break;
3660 #endif
3661 	default:
3662 		if (m)
3663 			m_freem(m);
3664 		return (EAFNOSUPPORT);
3665 	}
3666 
3667 	/* Compute the size of the TCP options. */
3668 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
3669 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
3670 
3671 	tlen = hlen + sizeof(struct tcphdr) + optlen;
3672 
3673 	/*
3674 	 * Create the IP+TCP header from scratch.
3675 	 */
3676 	if (m)
3677 		m_freem(m);
3678 #ifdef DIAGNOSTIC
3679 	if (max_linkhdr + tlen > MCLBYTES)
3680 		return (ENOBUFS);
3681 #endif
3682 	MGETHDR(m, M_DONTWAIT, MT_DATA);
3683 	if (m && tlen > MHLEN) {
3684 		MCLGET(m, M_DONTWAIT);
3685 		if ((m->m_flags & M_EXT) == 0) {
3686 			m_freem(m);
3687 			m = NULL;
3688 		}
3689 	}
3690 	if (m == NULL)
3691 		return (ENOBUFS);
3692 	MCLAIM(m, &tcp_tx_mowner);
3693 
3694 	/* Fixup the mbuf. */
3695 	m->m_data += max_linkhdr;
3696 	m->m_len = m->m_pkthdr.len = tlen;
3697 	if (sc->sc_tp) {
3698 		tp = sc->sc_tp;
3699 		if (tp->t_inpcb)
3700 			so = tp->t_inpcb->inp_socket;
3701 #ifdef INET6
3702 		else if (tp->t_in6pcb)
3703 			so = tp->t_in6pcb->in6p_socket;
3704 #endif
3705 		else
3706 			so = NULL;
3707 	} else
3708 		so = NULL;
3709 	m->m_pkthdr.rcvif = NULL;
3710 	memset(mtod(m, u_char *), 0, tlen);
3711 
3712 	switch (sc->sc_src.sa.sa_family) {
3713 	case AF_INET:
3714 		ip = mtod(m, struct ip *);
3715 		ip->ip_dst = sc->sc_src.sin.sin_addr;
3716 		ip->ip_src = sc->sc_dst.sin.sin_addr;
3717 		ip->ip_p = IPPROTO_TCP;
3718 		th = (struct tcphdr *)(ip + 1);
3719 		th->th_dport = sc->sc_src.sin.sin_port;
3720 		th->th_sport = sc->sc_dst.sin.sin_port;
3721 		break;
3722 #ifdef INET6
3723 	case AF_INET6:
3724 		ip6 = mtod(m, struct ip6_hdr *);
3725 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
3726 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
3727 		ip6->ip6_nxt = IPPROTO_TCP;
3728 		/* ip6_plen will be updated in ip6_output() */
3729 		th = (struct tcphdr *)(ip6 + 1);
3730 		th->th_dport = sc->sc_src.sin6.sin6_port;
3731 		th->th_sport = sc->sc_dst.sin6.sin6_port;
3732 		break;
3733 #endif
3734 	default:
3735 		th = NULL;
3736 	}
3737 
3738 	th->th_seq = htonl(sc->sc_iss);
3739 	th->th_ack = htonl(sc->sc_irs + 1);
3740 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
3741 	th->th_flags = TH_SYN|TH_ACK;
3742 	th->th_win = htons(sc->sc_win);
3743 	/* th_sum already 0 */
3744 	/* th_urp already 0 */
3745 
3746 	/* Tack on the TCP options. */
3747 	optp = (u_int8_t *)(th + 1);
3748 	*optp++ = TCPOPT_MAXSEG;
3749 	*optp++ = 4;
3750 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
3751 	*optp++ = sc->sc_ourmaxseg & 0xff;
3752 
3753 	if (sc->sc_request_r_scale != 15) {
3754 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
3755 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
3756 		    sc->sc_request_r_scale);
3757 		optp += 4;
3758 	}
3759 
3760 	if (sc->sc_flags & SCF_TIMESTAMP) {
3761 		u_int32_t *lp = (u_int32_t *)(optp);
3762 		/* Form timestamp option as shown in appendix A of RFC 1323. */
3763 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
3764 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
3765 		*lp   = htonl(sc->sc_timestamp);
3766 		optp += TCPOLEN_TSTAMP_APPA;
3767 	}
3768 
3769 	/* Compute the packet's checksum. */
3770 	switch (sc->sc_src.sa.sa_family) {
3771 	case AF_INET:
3772 		ip->ip_len = htons(tlen - hlen);
3773 		th->th_sum = 0;
3774 		th->th_sum = in_cksum(m, tlen);
3775 		break;
3776 #ifdef INET6
3777 	case AF_INET6:
3778 		ip6->ip6_plen = htons(tlen - hlen);
3779 		th->th_sum = 0;
3780 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
3781 		break;
3782 #endif
3783 	}
3784 
3785 	/*
3786 	 * Fill in some straggling IP bits.  Note the stack expects
3787 	 * ip_len to be in host order, for convenience.
3788 	 */
3789 	switch (sc->sc_src.sa.sa_family) {
3790 #ifdef INET
3791 	case AF_INET:
3792 		ip->ip_len = htons(tlen);
3793 		ip->ip_ttl = ip_defttl;
3794 		/* XXX tos? */
3795 		break;
3796 #endif
3797 #ifdef INET6
3798 	case AF_INET6:
3799 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3800 		ip6->ip6_vfc |= IPV6_VERSION;
3801 		ip6->ip6_plen = htons(tlen - hlen);
3802 		/* ip6_hlim will be initialized afterwards */
3803 		/* XXX flowlabel? */
3804 		break;
3805 #endif
3806 	}
3807 
3808 	/* XXX use IPsec policy on listening socket, on SYN ACK */
3809 	tp = sc->sc_tp;
3810 
3811 	switch (sc->sc_src.sa.sa_family) {
3812 #ifdef INET
3813 	case AF_INET:
3814 		error = ip_output(m, sc->sc_ipopts, ro,
3815 		    (ip_mtudisc ? IP_MTUDISC : 0),
3816 		    (struct ip_moptions *)NULL, so);
3817 		break;
3818 #endif
3819 #ifdef INET6
3820 	case AF_INET6:
3821 		ip6->ip6_hlim = in6_selecthlim(NULL,
3822 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
3823 
3824 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
3825 			(struct ip6_moptions *)0, so, NULL);
3826 		break;
3827 #endif
3828 	default:
3829 		error = EAFNOSUPPORT;
3830 		break;
3831 	}
3832 	return (error);
3833 }
3834