xref: /netbsd-src/sys/netinet/tcp_input.c (revision 001c68bd94f75ce9270b69227c4199fbf34ee396)
1 /*	$NetBSD: tcp_input.c,v 1.172 2003/07/02 19:33:20 ragge 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. All advertising materials mentioning features or use of this software
122  *    must display the following acknowledgement:
123  *	This product includes software developed by the University of
124  *	California, Berkeley and its contributors.
125  * 4. Neither the name of the University nor the names of its contributors
126  *    may be used to endorse or promote products derived from this software
127  *    without specific prior written permission.
128  *
129  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
130  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
131  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
132  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
133  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
134  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
135  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
136  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
137  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
138  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
139  * SUCH DAMAGE.
140  *
141  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
142  */
143 
144 /*
145  *	TODO list for SYN cache stuff:
146  *
147  *	Find room for a "state" field, which is needed to keep a
148  *	compressed state for TIME_WAIT TCBs.  It's been noted already
149  *	that this is fairly important for very high-volume web and
150  *	mail servers, which use a large number of short-lived
151  *	connections.
152  */
153 
154 #include <sys/cdefs.h>
155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.172 2003/07/02 19:33:20 ragge Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 
175 #include <net/if.h>
176 #include <net/route.h>
177 #include <net/if_types.h>
178 
179 #include <netinet/in.h>
180 #include <netinet/in_systm.h>
181 #include <netinet/ip.h>
182 #include <netinet/in_pcb.h>
183 #include <netinet/in_var.h>
184 #include <netinet/ip_var.h>
185 
186 #ifdef INET6
187 #ifndef INET
188 #include <netinet/in.h>
189 #endif
190 #include <netinet/ip6.h>
191 #include <netinet6/ip6_var.h>
192 #include <netinet6/in6_pcb.h>
193 #include <netinet6/ip6_var.h>
194 #include <netinet6/in6_var.h>
195 #include <netinet/icmp6.h>
196 #include <netinet6/nd6.h>
197 #endif
198 
199 #ifndef INET6
200 /* always need ip6.h for IP6_EXTHDR_GET */
201 #include <netinet/ip6.h>
202 #endif
203 
204 #include <netinet/tcp.h>
205 #include <netinet/tcp_fsm.h>
206 #include <netinet/tcp_seq.h>
207 #include <netinet/tcp_timer.h>
208 #include <netinet/tcp_var.h>
209 #include <netinet/tcpip.h>
210 #include <netinet/tcp_debug.h>
211 
212 #include <machine/stdarg.h>
213 
214 #ifdef IPSEC
215 #include <netinet6/ipsec.h>
216 #include <netkey/key.h>
217 #endif /*IPSEC*/
218 #ifdef INET6
219 #include "faith.h"
220 #if defined(NFAITH) && NFAITH > 0
221 #include <net/if_faith.h>
222 #endif
223 #endif
224 
225 int	tcprexmtthresh = 3;
226 int	tcp_log_refused;
227 
228 static int tcp_rst_ppslim_count = 0;
229 static struct timeval tcp_rst_ppslim_last;
230 
231 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
232 
233 /* for modulo comparisons of timestamps */
234 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
235 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
236 
237 /*
238  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
239  */
240 #ifdef INET6
241 #define ND6_HINT(tp) \
242 do { \
243 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \
244 	 && tp->t_in6pcb->in6p_route.ro_rt) { \
245 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
246 	} \
247 } while (/*CONSTCOND*/ 0)
248 #else
249 #define ND6_HINT(tp)
250 #endif
251 
252 /*
253  * Macro to compute ACK transmission behavior.  Delay the ACK unless
254  * we have already delayed an ACK (must send an ACK every two segments).
255  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
256  * option is enabled.
257  */
258 #define	TCP_SETUP_ACK(tp, th) \
259 do { \
260 	if ((tp)->t_flags & TF_DELACK || \
261 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
262 		tp->t_flags |= TF_ACKNOW; \
263 	else \
264 		TCP_SET_DELACK(tp); \
265 } while (/*CONSTCOND*/ 0)
266 
267 /*
268  * Convert TCP protocol fields to host order for easier processing.
269  */
270 #define	TCP_FIELDS_TO_HOST(th)						\
271 do {									\
272 	NTOHL((th)->th_seq);						\
273 	NTOHL((th)->th_ack);						\
274 	NTOHS((th)->th_win);						\
275 	NTOHS((th)->th_urp);						\
276 } while (/*CONSTCOND*/ 0)
277 
278 /*
279  * ... and reverse the above.
280  */
281 #define	TCP_FIELDS_TO_NET(th)						\
282 do {									\
283 	HTONL((th)->th_seq);						\
284 	HTONL((th)->th_ack);						\
285 	HTONS((th)->th_win);						\
286 	HTONS((th)->th_urp);						\
287 } while (/*CONSTCOND*/ 0)
288 
289 #ifdef TCP_CSUM_COUNTERS
290 #include <sys/device.h>
291 
292 extern struct evcnt tcp_hwcsum_ok;
293 extern struct evcnt tcp_hwcsum_bad;
294 extern struct evcnt tcp_hwcsum_data;
295 extern struct evcnt tcp_swcsum;
296 
297 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
298 
299 #else
300 
301 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
302 
303 #endif /* TCP_CSUM_COUNTERS */
304 
305 #ifdef TCP_REASS_COUNTERS
306 #include <sys/device.h>
307 
308 extern struct evcnt tcp_reass_;
309 extern struct evcnt tcp_reass_empty;
310 extern struct evcnt tcp_reass_iteration[8];
311 extern struct evcnt tcp_reass_prependfirst;
312 extern struct evcnt tcp_reass_prepend;
313 extern struct evcnt tcp_reass_insert;
314 extern struct evcnt tcp_reass_inserttail;
315 extern struct evcnt tcp_reass_append;
316 extern struct evcnt tcp_reass_appendtail;
317 extern struct evcnt tcp_reass_overlaptail;
318 extern struct evcnt tcp_reass_overlapfront;
319 extern struct evcnt tcp_reass_segdup;
320 extern struct evcnt tcp_reass_fragdup;
321 
322 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
323 
324 #else
325 
326 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
327 
328 #endif /* TCP_REASS_COUNTERS */
329 
330 #ifdef INET
331 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *));
332 #endif
333 #ifdef INET6
334 static void tcp6_log_refused
335     __P((const struct ip6_hdr *, const struct tcphdr *));
336 #endif
337 
338 int
339 tcp_reass(tp, th, m, tlen)
340 	struct tcpcb *tp;
341 	struct tcphdr *th;
342 	struct mbuf *m;
343 	int *tlen;
344 {
345 	struct ipqent *p, *q, *nq, *tiqe = NULL;
346 	struct socket *so = NULL;
347 	int pkt_flags;
348 	tcp_seq pkt_seq;
349 	unsigned pkt_len;
350 	u_long rcvpartdupbyte = 0;
351 	u_long rcvoobyte;
352 #ifdef TCP_REASS_COUNTERS
353 	u_int count = 0;
354 #endif
355 
356 	if (tp->t_inpcb)
357 		so = tp->t_inpcb->inp_socket;
358 #ifdef INET6
359 	else if (tp->t_in6pcb)
360 		so = tp->t_in6pcb->in6p_socket;
361 #endif
362 
363 	TCP_REASS_LOCK_CHECK(tp);
364 
365 	/*
366 	 * Call with th==0 after become established to
367 	 * force pre-ESTABLISHED data up to user socket.
368 	 */
369 	if (th == 0)
370 		goto present;
371 
372 	rcvoobyte = *tlen;
373 	/*
374 	 * Copy these to local variables because the tcpiphdr
375 	 * gets munged while we are collapsing mbufs.
376 	 */
377 	pkt_seq = th->th_seq;
378 	pkt_len = *tlen;
379 	pkt_flags = th->th_flags;
380 
381 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
382 
383 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
384 		/*
385 		 * When we miss a packet, the vast majority of time we get
386 		 * packets that follow it in order.  So optimize for that.
387 		 */
388 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
389 			p->ipqe_len += pkt_len;
390 			p->ipqe_flags |= pkt_flags;
391 			m_cat(p->ipqe_m, m);
392 			tiqe = p;
393 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
394 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
395 			goto skip_replacement;
396 		}
397 		/*
398 		 * While we're here, if the pkt is completely beyond
399 		 * anything we have, just insert it at the tail.
400 		 */
401 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
402 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
403 			goto insert_it;
404 		}
405 	}
406 
407 	q = TAILQ_FIRST(&tp->segq);
408 
409 	if (q != NULL) {
410 		/*
411 		 * If this segment immediately precedes the first out-of-order
412 		 * block, simply slap the segment in front of it and (mostly)
413 		 * skip the complicated logic.
414 		 */
415 		if (pkt_seq + pkt_len == q->ipqe_seq) {
416 			q->ipqe_seq = pkt_seq;
417 			q->ipqe_len += pkt_len;
418 			q->ipqe_flags |= pkt_flags;
419 			m_cat(m, q->ipqe_m);
420 			q->ipqe_m = m;
421 			tiqe = q;
422 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
423 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
424 			goto skip_replacement;
425 		}
426 	} else {
427 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
428 	}
429 
430 	/*
431 	 * Find a segment which begins after this one does.
432 	 */
433 	for (p = NULL; q != NULL; q = nq) {
434 		nq = TAILQ_NEXT(q, ipqe_q);
435 #ifdef TCP_REASS_COUNTERS
436 		count++;
437 #endif
438 		/*
439 		 * If the received segment is just right after this
440 		 * fragment, merge the two together and then check
441 		 * for further overlaps.
442 		 */
443 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
444 #ifdef TCPREASS_DEBUG
445 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
446 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
447 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
448 #endif
449 			pkt_len += q->ipqe_len;
450 			pkt_flags |= q->ipqe_flags;
451 			pkt_seq = q->ipqe_seq;
452 			m_cat(q->ipqe_m, m);
453 			m = q->ipqe_m;
454 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
455 			goto free_ipqe;
456 		}
457 		/*
458 		 * If the received segment is completely past this
459 		 * fragment, we need to go the next fragment.
460 		 */
461 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
462 			p = q;
463 			continue;
464 		}
465 		/*
466 		 * If the fragment is past the received segment,
467 		 * it (or any following) can't be concatenated.
468 		 */
469 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
470 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
471 			break;
472 		}
473 
474 		/*
475 		 * We've received all the data in this segment before.
476 		 * mark it as a duplicate and return.
477 		 */
478 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
479 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
480 			tcpstat.tcps_rcvduppack++;
481 			tcpstat.tcps_rcvdupbyte += pkt_len;
482 			m_freem(m);
483 			if (tiqe != NULL)
484 				pool_put(&ipqent_pool, tiqe);
485 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
486 			return (0);
487 		}
488 		/*
489 		 * Received segment completely overlaps this fragment
490 		 * so we drop the fragment (this keeps the temporal
491 		 * ordering of segments correct).
492 		 */
493 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
494 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
495 			rcvpartdupbyte += q->ipqe_len;
496 			m_freem(q->ipqe_m);
497 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
498 			goto free_ipqe;
499 		}
500 		/*
501 		 * RX'ed segment extends past the end of the
502 		 * fragment.  Drop the overlapping bytes.  Then
503 		 * merge the fragment and segment then treat as
504 		 * a longer received packet.
505 		 */
506 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
507 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
508 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
509 #ifdef TCPREASS_DEBUG
510 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
511 			       tp, overlap,
512 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
513 #endif
514 			m_adj(m, overlap);
515 			rcvpartdupbyte += overlap;
516 			m_cat(q->ipqe_m, m);
517 			m = q->ipqe_m;
518 			pkt_seq = q->ipqe_seq;
519 			pkt_len += q->ipqe_len - overlap;
520 			rcvoobyte -= overlap;
521 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
522 			goto free_ipqe;
523 		}
524 		/*
525 		 * RX'ed segment extends past the front of the
526 		 * fragment.  Drop the overlapping bytes on the
527 		 * received packet.  The packet will then be
528 		 * contatentated with this fragment a bit later.
529 		 */
530 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
531 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
532 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
533 #ifdef TCPREASS_DEBUG
534 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
535 			       tp, overlap,
536 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
537 #endif
538 			m_adj(m, -overlap);
539 			pkt_len -= overlap;
540 			rcvpartdupbyte += overlap;
541 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
542 			rcvoobyte -= overlap;
543 		}
544 		/*
545 		 * If the received segment immediates precedes this
546 		 * fragment then tack the fragment onto this segment
547 		 * and reinsert the data.
548 		 */
549 		if (q->ipqe_seq == pkt_seq + pkt_len) {
550 #ifdef TCPREASS_DEBUG
551 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
552 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
553 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
554 #endif
555 			pkt_len += q->ipqe_len;
556 			pkt_flags |= q->ipqe_flags;
557 			m_cat(m, q->ipqe_m);
558 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
559 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
560 			if (tiqe == NULL) {
561 			    tiqe = q;
562 			} else {
563 			    pool_put(&ipqent_pool, q);
564 			}
565 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
566 			break;
567 		}
568 		/*
569 		 * If the fragment is before the segment, remember it.
570 		 * When this loop is terminated, p will contain the
571 		 * pointer to fragment that is right before the received
572 		 * segment.
573 		 */
574 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
575 			p = q;
576 
577 		continue;
578 
579 		/*
580 		 * This is a common operation.  It also will allow
581 		 * to save doing a malloc/free in most instances.
582 		 */
583 	  free_ipqe:
584 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
585 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
586 		if (tiqe == NULL) {
587 		    tiqe = q;
588 		} else {
589 		    pool_put(&ipqent_pool, q);
590 		}
591 	}
592 
593 #ifdef TCP_REASS_COUNTERS
594 	if (count > 7)
595 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
596 	else if (count > 0)
597 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
598 #endif
599 
600     insert_it:
601 
602 	/*
603 	 * Allocate a new queue entry since the received segment did not
604 	 * collapse onto any other out-of-order block; thus we are allocating
605 	 * a new block.  If it had collapsed, tiqe would not be NULL and
606 	 * we would be reusing it.
607 	 * XXX If we can't, just drop the packet.  XXX
608 	 */
609 	if (tiqe == NULL) {
610 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
611 		if (tiqe == NULL) {
612 			tcpstat.tcps_rcvmemdrop++;
613 			m_freem(m);
614 			return (0);
615 		}
616 	}
617 
618 	/*
619 	 * Update the counters.
620 	 */
621 	tcpstat.tcps_rcvoopack++;
622 	tcpstat.tcps_rcvoobyte += rcvoobyte;
623 	if (rcvpartdupbyte) {
624 	    tcpstat.tcps_rcvpartduppack++;
625 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
626 	}
627 
628 	/*
629 	 * Insert the new fragment queue entry into both queues.
630 	 */
631 	tiqe->ipqe_m = m;
632 	tiqe->ipqe_seq = pkt_seq;
633 	tiqe->ipqe_len = pkt_len;
634 	tiqe->ipqe_flags = pkt_flags;
635 	if (p == NULL) {
636 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
637 #ifdef TCPREASS_DEBUG
638 		if (tiqe->ipqe_seq != tp->rcv_nxt)
639 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
640 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
641 #endif
642 	} else {
643 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
644 #ifdef TCPREASS_DEBUG
645 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
646 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
647 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
648 #endif
649 	}
650 
651 skip_replacement:
652 
653 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
654 
655 present:
656 	/*
657 	 * Present data to user, advancing rcv_nxt through
658 	 * completed sequence space.
659 	 */
660 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
661 		return (0);
662 	q = TAILQ_FIRST(&tp->segq);
663 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
664 		return (0);
665 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
666 		return (0);
667 
668 	tp->rcv_nxt += q->ipqe_len;
669 	pkt_flags = q->ipqe_flags & TH_FIN;
670 	ND6_HINT(tp);
671 
672 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
673 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
674 	if (so->so_state & SS_CANTRCVMORE)
675 		m_freem(q->ipqe_m);
676 	else
677 		sbappendstream(&so->so_rcv, q->ipqe_m);
678 	pool_put(&ipqent_pool, q);
679 	sorwakeup(so);
680 	return (pkt_flags);
681 }
682 
683 #ifdef INET6
684 int
685 tcp6_input(mp, offp, proto)
686 	struct mbuf **mp;
687 	int *offp, proto;
688 {
689 	struct mbuf *m = *mp;
690 
691 	/*
692 	 * draft-itojun-ipv6-tcp-to-anycast
693 	 * better place to put this in?
694 	 */
695 	if (m->m_flags & M_ANYCAST6) {
696 		struct ip6_hdr *ip6;
697 		if (m->m_len < sizeof(struct ip6_hdr)) {
698 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
699 				tcpstat.tcps_rcvshort++;
700 				return IPPROTO_DONE;
701 			}
702 		}
703 		ip6 = mtod(m, struct ip6_hdr *);
704 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
705 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
706 		return IPPROTO_DONE;
707 	}
708 
709 	tcp_input(m, *offp, proto);
710 	return IPPROTO_DONE;
711 }
712 #endif
713 
714 #ifdef INET
715 static void
716 tcp4_log_refused(ip, th)
717 	const struct ip *ip;
718 	const struct tcphdr *th;
719 {
720 	char src[4*sizeof "123"];
721 	char dst[4*sizeof "123"];
722 
723 	if (ip) {
724 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
725 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
726 	}
727 	else {
728 		strlcpy(src, "(unknown)", sizeof(src));
729 		strlcpy(dst, "(unknown)", sizeof(dst));
730 	}
731 	log(LOG_INFO,
732 	    "Connection attempt to TCP %s:%d from %s:%d\n",
733 	    dst, ntohs(th->th_dport),
734 	    src, ntohs(th->th_sport));
735 }
736 #endif
737 
738 #ifdef INET6
739 static void
740 tcp6_log_refused(ip6, th)
741 	const struct ip6_hdr *ip6;
742 	const struct tcphdr *th;
743 {
744 	char src[INET6_ADDRSTRLEN];
745 	char dst[INET6_ADDRSTRLEN];
746 
747 	if (ip6) {
748 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
749 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
750 	}
751 	else {
752 		strlcpy(src, "(unknown v6)", sizeof(src));
753 		strlcpy(dst, "(unknown v6)", sizeof(dst));
754 	}
755 	log(LOG_INFO,
756 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
757 	    dst, ntohs(th->th_dport),
758 	    src, ntohs(th->th_sport));
759 }
760 #endif
761 
762 /*
763  * TCP input routine, follows pages 65-76 of the
764  * protocol specification dated September, 1981 very closely.
765  */
766 void
767 #if __STDC__
768 tcp_input(struct mbuf *m, ...)
769 #else
770 tcp_input(m, va_alist)
771 	struct mbuf *m;
772 #endif
773 {
774 	struct tcphdr *th;
775 	struct ip *ip;
776 	struct inpcb *inp;
777 #ifdef INET6
778 	struct ip6_hdr *ip6;
779 	struct in6pcb *in6p;
780 #endif
781 	u_int8_t *optp = NULL;
782 	int optlen = 0;
783 	int len, tlen, toff, hdroptlen = 0;
784 	struct tcpcb *tp = 0;
785 	int tiflags;
786 	struct socket *so = NULL;
787 	int todrop, acked, ourfinisacked, needoutput = 0;
788 #ifdef TCP_DEBUG
789 	short ostate = 0;
790 #endif
791 	int iss = 0;
792 	u_long tiwin;
793 	struct tcp_opt_info opti;
794 	int off, iphlen;
795 	va_list ap;
796 	int af;		/* af on the wire */
797 	struct mbuf *tcp_saveti = NULL;
798 
799 	MCLAIM(m, &tcp_rx_mowner);
800 	va_start(ap, m);
801 	toff = va_arg(ap, int);
802 	(void)va_arg(ap, int);		/* ignore value, advance ap */
803 	va_end(ap);
804 
805 	tcpstat.tcps_rcvtotal++;
806 
807 	bzero(&opti, sizeof(opti));
808 	opti.ts_present = 0;
809 	opti.maxseg = 0;
810 
811 	/*
812 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
813 	 *
814 	 * TCP is, by definition, unicast, so we reject all
815 	 * multicast outright.
816 	 *
817 	 * Note, there are additional src/dst address checks in
818 	 * the AF-specific code below.
819 	 */
820 	if (m->m_flags & (M_BCAST|M_MCAST)) {
821 		/* XXX stat */
822 		goto drop;
823 	}
824 #ifdef INET6
825 	if (m->m_flags & M_ANYCAST6) {
826 		/* XXX stat */
827 		goto drop;
828 	}
829 #endif
830 
831 	/*
832 	 * Get IP and TCP header together in first mbuf.
833 	 * Note: IP leaves IP header in first mbuf.
834 	 */
835 	ip = mtod(m, struct ip *);
836 #ifdef INET6
837 	ip6 = NULL;
838 #endif
839 	switch (ip->ip_v) {
840 #ifdef INET
841 	case 4:
842 		af = AF_INET;
843 		iphlen = sizeof(struct ip);
844 		ip = mtod(m, struct ip *);
845 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
846 			sizeof(struct tcphdr));
847 		if (th == NULL) {
848 			tcpstat.tcps_rcvshort++;
849 			return;
850 		}
851 		/* We do the checksum after PCB lookup... */
852 		len = ntohs(ip->ip_len);
853 		tlen = len - toff;
854 		break;
855 #endif
856 #ifdef INET6
857 	case 6:
858 		ip = NULL;
859 		iphlen = sizeof(struct ip6_hdr);
860 		af = AF_INET6;
861 		ip6 = mtod(m, struct ip6_hdr *);
862 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
863 			sizeof(struct tcphdr));
864 		if (th == NULL) {
865 			tcpstat.tcps_rcvshort++;
866 			return;
867 		}
868 
869 		/* Be proactive about malicious use of IPv4 mapped address */
870 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
871 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
872 			/* XXX stat */
873 			goto drop;
874 		}
875 
876 		/*
877 		 * Be proactive about unspecified IPv6 address in source.
878 		 * As we use all-zero to indicate unbounded/unconnected pcb,
879 		 * unspecified IPv6 address can be used to confuse us.
880 		 *
881 		 * Note that packets with unspecified IPv6 destination is
882 		 * already dropped in ip6_input.
883 		 */
884 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
885 			/* XXX stat */
886 			goto drop;
887 		}
888 
889 		/*
890 		 * Make sure destination address is not multicast.
891 		 * Source address checked in ip6_input().
892 		 */
893 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
894 			/* XXX stat */
895 			goto drop;
896 		}
897 
898 		/* We do the checksum after PCB lookup... */
899 		len = m->m_pkthdr.len;
900 		tlen = len - toff;
901 		break;
902 #endif
903 	default:
904 		m_freem(m);
905 		return;
906 	}
907 
908 	KASSERT(TCP_HDR_ALIGNED_P(th));
909 
910 	/*
911 	 * Check that TCP offset makes sense,
912 	 * pull out TCP options and adjust length.		XXX
913 	 */
914 	off = th->th_off << 2;
915 	if (off < sizeof (struct tcphdr) || off > tlen) {
916 		tcpstat.tcps_rcvbadoff++;
917 		goto drop;
918 	}
919 	tlen -= off;
920 
921 	/*
922 	 * tcp_input() has been modified to use tlen to mean the TCP data
923 	 * length throughout the function.  Other functions can use
924 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
925 	 * rja
926 	 */
927 
928 	if (off > sizeof (struct tcphdr)) {
929 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
930 		if (th == NULL) {
931 			tcpstat.tcps_rcvshort++;
932 			return;
933 		}
934 		/*
935 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
936 		 * (as they're before toff) and we don't need to update those.
937 		 */
938 		KASSERT(TCP_HDR_ALIGNED_P(th));
939 		optlen = off - sizeof (struct tcphdr);
940 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
941 		/*
942 		 * Do quick retrieval of timestamp options ("options
943 		 * prediction?").  If timestamp is the only option and it's
944 		 * formatted as recommended in RFC 1323 appendix A, we
945 		 * quickly get the values now and not bother calling
946 		 * tcp_dooptions(), etc.
947 		 */
948 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
949 		     (optlen > TCPOLEN_TSTAMP_APPA &&
950 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
951 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
952 		     (th->th_flags & TH_SYN) == 0) {
953 			opti.ts_present = 1;
954 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
955 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
956 			optp = NULL;	/* we've parsed the options */
957 		}
958 	}
959 	tiflags = th->th_flags;
960 
961 	/*
962 	 * Locate pcb for segment.
963 	 */
964 findpcb:
965 	inp = NULL;
966 #ifdef INET6
967 	in6p = NULL;
968 #endif
969 	switch (af) {
970 #ifdef INET
971 	case AF_INET:
972 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
973 		    ip->ip_dst, th->th_dport);
974 		if (inp == 0) {
975 			++tcpstat.tcps_pcbhashmiss;
976 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
977 		}
978 #ifdef INET6
979 		if (inp == 0) {
980 			struct in6_addr s, d;
981 
982 			/* mapped addr case */
983 			bzero(&s, sizeof(s));
984 			s.s6_addr16[5] = htons(0xffff);
985 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
986 			bzero(&d, sizeof(d));
987 			d.s6_addr16[5] = htons(0xffff);
988 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
989 			in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport,
990 				&d, th->th_dport, 0);
991 			if (in6p == 0) {
992 				++tcpstat.tcps_pcbhashmiss;
993 				in6p = in6_pcblookup_bind(&tcb6, &d,
994 					th->th_dport, 0);
995 			}
996 		}
997 #endif
998 #ifndef INET6
999 		if (inp == 0)
1000 #else
1001 		if (inp == 0 && in6p == 0)
1002 #endif
1003 		{
1004 			++tcpstat.tcps_noport;
1005 			if (tcp_log_refused &&
1006 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1007 				tcp4_log_refused(ip, th);
1008 			}
1009 			TCP_FIELDS_TO_HOST(th);
1010 			goto dropwithreset_ratelim;
1011 		}
1012 #ifdef IPSEC
1013 		if (inp && ipsec4_in_reject(m, inp)) {
1014 			ipsecstat.in_polvio++;
1015 			goto drop;
1016 		}
1017 #ifdef INET6
1018 		else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1019 			ipsecstat.in_polvio++;
1020 			goto drop;
1021 		}
1022 #endif
1023 #endif /*IPSEC*/
1024 		break;
1025 #endif /*INET*/
1026 #ifdef INET6
1027 	case AF_INET6:
1028 	    {
1029 		int faith;
1030 
1031 #if defined(NFAITH) && NFAITH > 0
1032 		faith = faithprefix(&ip6->ip6_dst);
1033 #else
1034 		faith = 0;
1035 #endif
1036 		in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport,
1037 			&ip6->ip6_dst, th->th_dport, faith);
1038 		if (in6p == NULL) {
1039 			++tcpstat.tcps_pcbhashmiss;
1040 			in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst,
1041 				th->th_dport, faith);
1042 		}
1043 		if (in6p == NULL) {
1044 			++tcpstat.tcps_noport;
1045 			if (tcp_log_refused &&
1046 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1047 				tcp6_log_refused(ip6, th);
1048 			}
1049 			TCP_FIELDS_TO_HOST(th);
1050 			goto dropwithreset_ratelim;
1051 		}
1052 #ifdef IPSEC
1053 		if (ipsec6_in_reject(m, in6p)) {
1054 			ipsec6stat.in_polvio++;
1055 			goto drop;
1056 		}
1057 #endif /*IPSEC*/
1058 		break;
1059 	    }
1060 #endif
1061 	}
1062 
1063 	/*
1064 	 * If the state is CLOSED (i.e., TCB does not exist) then
1065 	 * all data in the incoming segment is discarded.
1066 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1067 	 * but should either do a listen or a connect soon.
1068 	 */
1069 	tp = NULL;
1070 	so = NULL;
1071 	if (inp) {
1072 		tp = intotcpcb(inp);
1073 		so = inp->inp_socket;
1074 	}
1075 #ifdef INET6
1076 	else if (in6p) {
1077 		tp = in6totcpcb(in6p);
1078 		so = in6p->in6p_socket;
1079 	}
1080 #endif
1081 	if (tp == 0) {
1082 		TCP_FIELDS_TO_HOST(th);
1083 		goto dropwithreset_ratelim;
1084 	}
1085 	if (tp->t_state == TCPS_CLOSED)
1086 		goto drop;
1087 
1088 	/*
1089 	 * Checksum extended TCP header and data.
1090 	 */
1091 	switch (af) {
1092 #ifdef INET
1093 	case AF_INET:
1094 		switch (m->m_pkthdr.csum_flags &
1095 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
1096 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
1097 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
1098 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
1099 			goto badcsum;
1100 
1101 		case M_CSUM_TCPv4|M_CSUM_DATA:
1102 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
1103 			if ((m->m_pkthdr.csum_data ^ 0xffff) != 0)
1104 				goto badcsum;
1105 			break;
1106 
1107 		case M_CSUM_TCPv4:
1108 			/* Checksum was okay. */
1109 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
1110 			break;
1111 
1112 		default:
1113 			/* Must compute it ourselves. */
1114 			TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
1115 			if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1116 				goto badcsum;
1117 			break;
1118 		}
1119 		break;
1120 #endif /* INET4 */
1121 
1122 #ifdef INET6
1123 	case AF_INET6:
1124 		if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1125 			goto badcsum;
1126 		break;
1127 #endif /* INET6 */
1128 	}
1129 
1130 	TCP_FIELDS_TO_HOST(th);
1131 
1132 	/* Unscale the window into a 32-bit value. */
1133 	if ((tiflags & TH_SYN) == 0)
1134 		tiwin = th->th_win << tp->snd_scale;
1135 	else
1136 		tiwin = th->th_win;
1137 
1138 #ifdef INET6
1139 	/* save packet options if user wanted */
1140 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1141 		if (in6p->in6p_options) {
1142 			m_freem(in6p->in6p_options);
1143 			in6p->in6p_options = 0;
1144 		}
1145 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1146 	}
1147 #endif
1148 
1149 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1150 		union syn_cache_sa src;
1151 		union syn_cache_sa dst;
1152 
1153 		bzero(&src, sizeof(src));
1154 		bzero(&dst, sizeof(dst));
1155 		switch (af) {
1156 #ifdef INET
1157 		case AF_INET:
1158 			src.sin.sin_len = sizeof(struct sockaddr_in);
1159 			src.sin.sin_family = AF_INET;
1160 			src.sin.sin_addr = ip->ip_src;
1161 			src.sin.sin_port = th->th_sport;
1162 
1163 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1164 			dst.sin.sin_family = AF_INET;
1165 			dst.sin.sin_addr = ip->ip_dst;
1166 			dst.sin.sin_port = th->th_dport;
1167 			break;
1168 #endif
1169 #ifdef INET6
1170 		case AF_INET6:
1171 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1172 			src.sin6.sin6_family = AF_INET6;
1173 			src.sin6.sin6_addr = ip6->ip6_src;
1174 			src.sin6.sin6_port = th->th_sport;
1175 
1176 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1177 			dst.sin6.sin6_family = AF_INET6;
1178 			dst.sin6.sin6_addr = ip6->ip6_dst;
1179 			dst.sin6.sin6_port = th->th_dport;
1180 			break;
1181 #endif /* INET6 */
1182 		default:
1183 			goto badsyn;	/*sanity*/
1184 		}
1185 
1186 		if (so->so_options & SO_DEBUG) {
1187 #ifdef TCP_DEBUG
1188 			ostate = tp->t_state;
1189 #endif
1190 
1191 			tcp_saveti = NULL;
1192 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1193 				goto nosave;
1194 
1195 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1196 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1197 				if (!tcp_saveti)
1198 					goto nosave;
1199 			} else {
1200 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1201 				if (!tcp_saveti)
1202 					goto nosave;
1203 				MCLAIM(m, &tcp_mowner);
1204 				tcp_saveti->m_len = iphlen;
1205 				m_copydata(m, 0, iphlen,
1206 				    mtod(tcp_saveti, caddr_t));
1207 			}
1208 
1209 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1210 				m_freem(tcp_saveti);
1211 				tcp_saveti = NULL;
1212 			} else {
1213 				tcp_saveti->m_len += sizeof(struct tcphdr);
1214 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1215 				    sizeof(struct tcphdr));
1216 			}
1217 	nosave:;
1218 		}
1219 		if (so->so_options & SO_ACCEPTCONN) {
1220 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1221 				if (tiflags & TH_RST) {
1222 					syn_cache_reset(&src.sa, &dst.sa, th);
1223 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1224 				    (TH_ACK|TH_SYN)) {
1225 					/*
1226 					 * Received a SYN,ACK.  This should
1227 					 * never happen while we are in
1228 					 * LISTEN.  Send an RST.
1229 					 */
1230 					goto badsyn;
1231 				} else if (tiflags & TH_ACK) {
1232 					so = syn_cache_get(&src.sa, &dst.sa,
1233 						th, toff, tlen, so, m);
1234 					if (so == NULL) {
1235 						/*
1236 						 * We don't have a SYN for
1237 						 * this ACK; send an RST.
1238 						 */
1239 						goto badsyn;
1240 					} else if (so ==
1241 					    (struct socket *)(-1)) {
1242 						/*
1243 						 * We were unable to create
1244 						 * the connection.  If the
1245 						 * 3-way handshake was
1246 						 * completed, and RST has
1247 						 * been sent to the peer.
1248 						 * Since the mbuf might be
1249 						 * in use for the reply,
1250 						 * do not free it.
1251 						 */
1252 						m = NULL;
1253 					} else {
1254 						/*
1255 						 * We have created a
1256 						 * full-blown connection.
1257 						 */
1258 						tp = NULL;
1259 						inp = NULL;
1260 #ifdef INET6
1261 						in6p = NULL;
1262 #endif
1263 						switch (so->so_proto->pr_domain->dom_family) {
1264 #ifdef INET
1265 						case AF_INET:
1266 							inp = sotoinpcb(so);
1267 							tp = intotcpcb(inp);
1268 							break;
1269 #endif
1270 #ifdef INET6
1271 						case AF_INET6:
1272 							in6p = sotoin6pcb(so);
1273 							tp = in6totcpcb(in6p);
1274 							break;
1275 #endif
1276 						}
1277 						if (tp == NULL)
1278 							goto badsyn;	/*XXX*/
1279 						tiwin <<= tp->snd_scale;
1280 						goto after_listen;
1281 					}
1282 				} else {
1283 					/*
1284 					 * None of RST, SYN or ACK was set.
1285 					 * This is an invalid packet for a
1286 					 * TCB in LISTEN state.  Send a RST.
1287 					 */
1288 					goto badsyn;
1289 				}
1290 			} else {
1291 				/*
1292 				 * Received a SYN.
1293 				 */
1294 
1295 #ifdef INET6
1296 				/*
1297 				 * If deprecated address is forbidden, we do
1298 				 * not accept SYN to deprecated interface
1299 				 * address to prevent any new inbound
1300 				 * connection from getting established.
1301 				 * When we do not accept SYN, we send a TCP
1302 				 * RST, with deprecated source address (instead
1303 				 * of dropping it).  We compromise it as it is
1304 				 * much better for peer to send a RST, and
1305 				 * RST will be the final packet for the
1306 				 * exchange.
1307 				 *
1308 				 * If we do not forbid deprecated addresses, we
1309 				 * accept the SYN packet.  RFC2462 does not
1310 				 * suggest dropping SYN in this case.
1311 				 * If we decipher RFC2462 5.5.4, it says like
1312 				 * this:
1313 				 * 1. use of deprecated addr with existing
1314 				 *    communication is okay - "SHOULD continue
1315 				 *    to be used"
1316 				 * 2. use of it with new communication:
1317 				 *   (2a) "SHOULD NOT be used if alternate
1318 				 *        address with sufficient scope is
1319 				 *        available"
1320 				 *   (2b) nothing mentioned otherwise.
1321 				 * Here we fall into (2b) case as we have no
1322 				 * choice in our source address selection - we
1323 				 * must obey the peer.
1324 				 *
1325 				 * The wording in RFC2462 is confusing, and
1326 				 * there are multiple description text for
1327 				 * deprecated address handling - worse, they
1328 				 * are not exactly the same.  I believe 5.5.4
1329 				 * is the best one, so we follow 5.5.4.
1330 				 */
1331 				if (af == AF_INET6 && !ip6_use_deprecated) {
1332 					struct in6_ifaddr *ia6;
1333 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1334 					    &ip6->ip6_dst)) &&
1335 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1336 						tp = NULL;
1337 						goto dropwithreset;
1338 					}
1339 				}
1340 #endif
1341 
1342 				/*
1343 				 * LISTEN socket received a SYN
1344 				 * from itself?  This can't possibly
1345 				 * be valid; drop the packet.
1346 				 */
1347 				if (th->th_sport == th->th_dport) {
1348 					int i;
1349 
1350 					switch (af) {
1351 #ifdef INET
1352 					case AF_INET:
1353 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1354 						break;
1355 #endif
1356 #ifdef INET6
1357 					case AF_INET6:
1358 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1359 						break;
1360 #endif
1361 					default:
1362 						i = 1;
1363 					}
1364 					if (i) {
1365 						tcpstat.tcps_badsyn++;
1366 						goto drop;
1367 					}
1368 				}
1369 
1370 				/*
1371 				 * SYN looks ok; create compressed TCP
1372 				 * state for it.
1373 				 */
1374 				if (so->so_qlen <= so->so_qlimit &&
1375 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1376 						so, m, optp, optlen, &opti))
1377 					m = NULL;
1378 			}
1379 			goto drop;
1380 		}
1381 	}
1382 
1383 after_listen:
1384 #ifdef DIAGNOSTIC
1385 	/*
1386 	 * Should not happen now that all embryonic connections
1387 	 * are handled with compressed state.
1388 	 */
1389 	if (tp->t_state == TCPS_LISTEN)
1390 		panic("tcp_input: TCPS_LISTEN");
1391 #endif
1392 
1393 	/*
1394 	 * Segment received on connection.
1395 	 * Reset idle time and keep-alive timer.
1396 	 */
1397 	tp->t_rcvtime = tcp_now;
1398 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1399 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1400 
1401 	/*
1402 	 * Process options.
1403 	 */
1404 	if (optp)
1405 		tcp_dooptions(tp, optp, optlen, th, &opti);
1406 
1407 	/*
1408 	 * Header prediction: check for the two common cases
1409 	 * of a uni-directional data xfer.  If the packet has
1410 	 * no control flags, is in-sequence, the window didn't
1411 	 * change and we're not retransmitting, it's a
1412 	 * candidate.  If the length is zero and the ack moved
1413 	 * forward, we're the sender side of the xfer.  Just
1414 	 * free the data acked & wake any higher level process
1415 	 * that was blocked waiting for space.  If the length
1416 	 * is non-zero and the ack didn't move, we're the
1417 	 * receiver side.  If we're getting packets in-order
1418 	 * (the reassembly queue is empty), add the data to
1419 	 * the socket buffer and note that we need a delayed ack.
1420 	 */
1421 	if (tp->t_state == TCPS_ESTABLISHED &&
1422 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1423 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1424 	    th->th_seq == tp->rcv_nxt &&
1425 	    tiwin && tiwin == tp->snd_wnd &&
1426 	    tp->snd_nxt == tp->snd_max) {
1427 
1428 		/*
1429 		 * If last ACK falls within this segment's sequence numbers,
1430 		 *  record the timestamp.
1431 		 */
1432 		if (opti.ts_present &&
1433 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1434 		    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) {
1435 			tp->ts_recent_age = TCP_TIMESTAMP(tp);
1436 			tp->ts_recent = opti.ts_val;
1437 		}
1438 
1439 		if (tlen == 0) {
1440 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1441 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1442 			    tp->snd_cwnd >= tp->snd_wnd &&
1443 			    tp->t_dupacks < tcprexmtthresh) {
1444 				int slen;
1445 
1446 				/*
1447 				 * this is a pure ack for outstanding data.
1448 				 */
1449 				++tcpstat.tcps_predack;
1450 				if (opti.ts_present && opti.ts_ecr)
1451 					tcp_xmit_timer(tp,
1452 					  TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
1453 				else if (tp->t_rtttime &&
1454 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1455 					tcp_xmit_timer(tp,
1456 					tcp_now - tp->t_rtttime);
1457 				acked = th->th_ack - tp->snd_una;
1458 				tcpstat.tcps_rcvackpack++;
1459 				tcpstat.tcps_rcvackbyte += acked;
1460 				ND6_HINT(tp);
1461 
1462 				slen = tp->t_lastm->m_len;
1463 				sbdrop(&so->so_snd, acked);
1464 				if (so->so_snd.sb_cc != 0) {
1465 					tp->t_lastoff -= acked;
1466 					if (tp->t_lastm->m_len < slen)
1467 						tp->t_inoff -=
1468 						    (slen - tp->t_lastm->m_len);
1469 				}
1470 
1471 				/*
1472 				 * We want snd_recover to track snd_una to
1473 				 * avoid sequence wraparound problems for
1474 				 * very large transfers.
1475 				 */
1476 				tp->snd_una = tp->snd_recover = th->th_ack;
1477 				m_freem(m);
1478 
1479 				/*
1480 				 * If all outstanding data are acked, stop
1481 				 * retransmit timer, otherwise restart timer
1482 				 * using current (possibly backed-off) value.
1483 				 * If process is waiting for space,
1484 				 * wakeup/selwakeup/signal.  If data
1485 				 * are ready to send, let tcp_output
1486 				 * decide between more output or persist.
1487 				 */
1488 				if (tp->snd_una == tp->snd_max)
1489 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1490 				else if (TCP_TIMER_ISARMED(tp,
1491 				    TCPT_PERSIST) == 0)
1492 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1493 					    tp->t_rxtcur);
1494 
1495 				sowwakeup(so);
1496 				if (so->so_snd.sb_cc)
1497 					(void) tcp_output(tp);
1498 				if (tcp_saveti)
1499 					m_freem(tcp_saveti);
1500 				return;
1501 			}
1502 		} else if (th->th_ack == tp->snd_una &&
1503 		    TAILQ_FIRST(&tp->segq) == NULL &&
1504 		    tlen <= sbspace(&so->so_rcv)) {
1505 			/*
1506 			 * this is a pure, in-sequence data packet
1507 			 * with nothing on the reassembly queue and
1508 			 * we have enough buffer space to take it.
1509 			 */
1510 			++tcpstat.tcps_preddat;
1511 			tp->rcv_nxt += tlen;
1512 			tcpstat.tcps_rcvpack++;
1513 			tcpstat.tcps_rcvbyte += tlen;
1514 			ND6_HINT(tp);
1515 			/*
1516 			 * Drop TCP, IP headers and TCP options then add data
1517 			 * to socket buffer.
1518 			 */
1519 			if (so->so_state & SS_CANTRCVMORE)
1520 				m_freem(m);
1521 			else {
1522 				m_adj(m, toff + off);
1523 				sbappendstream(&so->so_rcv, m);
1524 			}
1525 			sorwakeup(so);
1526 			TCP_SETUP_ACK(tp, th);
1527 			if (tp->t_flags & TF_ACKNOW)
1528 				(void) tcp_output(tp);
1529 			if (tcp_saveti)
1530 				m_freem(tcp_saveti);
1531 			return;
1532 		}
1533 	}
1534 
1535 	/*
1536 	 * Compute mbuf offset to TCP data segment.
1537 	 */
1538 	hdroptlen = toff + off;
1539 
1540 	/*
1541 	 * Calculate amount of space in receive window,
1542 	 * and then do TCP input processing.
1543 	 * Receive window is amount of space in rcv queue,
1544 	 * but not less than advertised window.
1545 	 */
1546 	{ int win;
1547 
1548 	win = sbspace(&so->so_rcv);
1549 	if (win < 0)
1550 		win = 0;
1551 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1552 	}
1553 
1554 	switch (tp->t_state) {
1555 	case TCPS_LISTEN:
1556 		/*
1557 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1558 		 */
1559 		if (m->m_flags & (M_BCAST|M_MCAST))
1560 			goto drop;
1561 		switch (af) {
1562 #ifdef INET6
1563 		case AF_INET6:
1564 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1565 				goto drop;
1566 			break;
1567 #endif /* INET6 */
1568 		case AF_INET:
1569 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1570 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1571 				goto drop;
1572 			break;
1573 		}
1574 		break;
1575 
1576 	/*
1577 	 * If the state is SYN_SENT:
1578 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1579 	 *	if seg contains a RST, then drop the connection.
1580 	 *	if seg does not contain SYN, then drop it.
1581 	 * Otherwise this is an acceptable SYN segment
1582 	 *	initialize tp->rcv_nxt and tp->irs
1583 	 *	if seg contains ack then advance tp->snd_una
1584 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1585 	 *	arrange for segment to be acked (eventually)
1586 	 *	continue processing rest of data/controls, beginning with URG
1587 	 */
1588 	case TCPS_SYN_SENT:
1589 		if ((tiflags & TH_ACK) &&
1590 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1591 		     SEQ_GT(th->th_ack, tp->snd_max)))
1592 			goto dropwithreset;
1593 		if (tiflags & TH_RST) {
1594 			if (tiflags & TH_ACK)
1595 				tp = tcp_drop(tp, ECONNREFUSED);
1596 			goto drop;
1597 		}
1598 		if ((tiflags & TH_SYN) == 0)
1599 			goto drop;
1600 		if (tiflags & TH_ACK) {
1601 			tp->snd_una = tp->snd_recover = th->th_ack;
1602 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1603 				tp->snd_nxt = tp->snd_una;
1604 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1605 		}
1606 		tp->irs = th->th_seq;
1607 		tcp_rcvseqinit(tp);
1608 		tp->t_flags |= TF_ACKNOW;
1609 		tcp_mss_from_peer(tp, opti.maxseg);
1610 
1611 		/*
1612 		 * Initialize the initial congestion window.  If we
1613 		 * had to retransmit the SYN, we must initialize cwnd
1614 		 * to 1 segment (i.e. the Loss Window).
1615 		 */
1616 		if (tp->t_flags & TF_SYN_REXMT)
1617 			tp->snd_cwnd = tp->t_peermss;
1618 		else {
1619 			int ss = tcp_init_win;
1620 #ifdef INET
1621 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1622 				ss = tcp_init_win_local;
1623 #endif
1624 #ifdef INET6
1625 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1626 				ss = tcp_init_win_local;
1627 #endif
1628 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1629 		}
1630 
1631 		tcp_rmx_rtt(tp);
1632 		if (tiflags & TH_ACK) {
1633 			tcpstat.tcps_connects++;
1634 			soisconnected(so);
1635 			tcp_established(tp);
1636 			/* Do window scaling on this connection? */
1637 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1638 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1639 				tp->snd_scale = tp->requested_s_scale;
1640 				tp->rcv_scale = tp->request_r_scale;
1641 			}
1642 			TCP_REASS_LOCK(tp);
1643 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1644 			TCP_REASS_UNLOCK(tp);
1645 			/*
1646 			 * if we didn't have to retransmit the SYN,
1647 			 * use its rtt as our initial srtt & rtt var.
1648 			 */
1649 			if (tp->t_rtttime)
1650 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1651 		} else
1652 			tp->t_state = TCPS_SYN_RECEIVED;
1653 
1654 		/*
1655 		 * Advance th->th_seq to correspond to first data byte.
1656 		 * If data, trim to stay within window,
1657 		 * dropping FIN if necessary.
1658 		 */
1659 		th->th_seq++;
1660 		if (tlen > tp->rcv_wnd) {
1661 			todrop = tlen - tp->rcv_wnd;
1662 			m_adj(m, -todrop);
1663 			tlen = tp->rcv_wnd;
1664 			tiflags &= ~TH_FIN;
1665 			tcpstat.tcps_rcvpackafterwin++;
1666 			tcpstat.tcps_rcvbyteafterwin += todrop;
1667 		}
1668 		tp->snd_wl1 = th->th_seq - 1;
1669 		tp->rcv_up = th->th_seq;
1670 		goto step6;
1671 
1672 	/*
1673 	 * If the state is SYN_RECEIVED:
1674 	 *	If seg contains an ACK, but not for our SYN, drop the input
1675 	 *	and generate an RST.  See page 36, rfc793
1676 	 */
1677 	case TCPS_SYN_RECEIVED:
1678 		if ((tiflags & TH_ACK) &&
1679 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1680 		     SEQ_GT(th->th_ack, tp->snd_max)))
1681 			goto dropwithreset;
1682 		break;
1683 	}
1684 
1685 	/*
1686 	 * States other than LISTEN or SYN_SENT.
1687 	 * First check timestamp, if present.
1688 	 * Then check that at least some bytes of segment are within
1689 	 * receive window.  If segment begins before rcv_nxt,
1690 	 * drop leading data (and SYN); if nothing left, just ack.
1691 	 *
1692 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1693 	 * and it's less than ts_recent, drop it.
1694 	 */
1695 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1696 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1697 
1698 		/* Check to see if ts_recent is over 24 days old.  */
1699 		if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) >
1700 		    TCP_PAWS_IDLE) {
1701 			/*
1702 			 * Invalidate ts_recent.  If this segment updates
1703 			 * ts_recent, the age will be reset later and ts_recent
1704 			 * will get a valid value.  If it does not, setting
1705 			 * ts_recent to zero will at least satisfy the
1706 			 * requirement that zero be placed in the timestamp
1707 			 * echo reply when ts_recent isn't valid.  The
1708 			 * age isn't reset until we get a valid ts_recent
1709 			 * because we don't want out-of-order segments to be
1710 			 * dropped when ts_recent is old.
1711 			 */
1712 			tp->ts_recent = 0;
1713 		} else {
1714 			tcpstat.tcps_rcvduppack++;
1715 			tcpstat.tcps_rcvdupbyte += tlen;
1716 			tcpstat.tcps_pawsdrop++;
1717 			goto dropafterack;
1718 		}
1719 	}
1720 
1721 	todrop = tp->rcv_nxt - th->th_seq;
1722 	if (todrop > 0) {
1723 		if (tiflags & TH_SYN) {
1724 			tiflags &= ~TH_SYN;
1725 			th->th_seq++;
1726 			if (th->th_urp > 1)
1727 				th->th_urp--;
1728 			else {
1729 				tiflags &= ~TH_URG;
1730 				th->th_urp = 0;
1731 			}
1732 			todrop--;
1733 		}
1734 		if (todrop > tlen ||
1735 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1736 			/*
1737 			 * Any valid FIN must be to the left of the window.
1738 			 * At this point the FIN must be a duplicate or
1739 			 * out of sequence; drop it.
1740 			 */
1741 			tiflags &= ~TH_FIN;
1742 			/*
1743 			 * Send an ACK to resynchronize and drop any data.
1744 			 * But keep on processing for RST or ACK.
1745 			 */
1746 			tp->t_flags |= TF_ACKNOW;
1747 			todrop = tlen;
1748 			tcpstat.tcps_rcvdupbyte += todrop;
1749 			tcpstat.tcps_rcvduppack++;
1750 		} else {
1751 			tcpstat.tcps_rcvpartduppack++;
1752 			tcpstat.tcps_rcvpartdupbyte += todrop;
1753 		}
1754 		hdroptlen += todrop;	/*drop from head afterwards*/
1755 		th->th_seq += todrop;
1756 		tlen -= todrop;
1757 		if (th->th_urp > todrop)
1758 			th->th_urp -= todrop;
1759 		else {
1760 			tiflags &= ~TH_URG;
1761 			th->th_urp = 0;
1762 		}
1763 	}
1764 
1765 	/*
1766 	 * If new data are received on a connection after the
1767 	 * user processes are gone, then RST the other end.
1768 	 */
1769 	if ((so->so_state & SS_NOFDREF) &&
1770 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1771 		tp = tcp_close(tp);
1772 		tcpstat.tcps_rcvafterclose++;
1773 		goto dropwithreset;
1774 	}
1775 
1776 	/*
1777 	 * If segment ends after window, drop trailing data
1778 	 * (and PUSH and FIN); if nothing left, just ACK.
1779 	 */
1780 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1781 	if (todrop > 0) {
1782 		tcpstat.tcps_rcvpackafterwin++;
1783 		if (todrop >= tlen) {
1784 			tcpstat.tcps_rcvbyteafterwin += tlen;
1785 			/*
1786 			 * If a new connection request is received
1787 			 * while in TIME_WAIT, drop the old connection
1788 			 * and start over if the sequence numbers
1789 			 * are above the previous ones.
1790 			 *
1791 			 * NOTE: We will checksum the packet again, and
1792 			 * so we need to put the header fields back into
1793 			 * network order!
1794 			 * XXX This kind of sucks, but we don't expect
1795 			 * XXX this to happen very often, so maybe it
1796 			 * XXX doesn't matter so much.
1797 			 */
1798 			if (tiflags & TH_SYN &&
1799 			    tp->t_state == TCPS_TIME_WAIT &&
1800 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1801 				iss = tcp_new_iss(tp, tp->snd_nxt);
1802 				tp = tcp_close(tp);
1803 				TCP_FIELDS_TO_NET(th);
1804 				goto findpcb;
1805 			}
1806 			/*
1807 			 * If window is closed can only take segments at
1808 			 * window edge, and have to drop data and PUSH from
1809 			 * incoming segments.  Continue processing, but
1810 			 * remember to ack.  Otherwise, drop segment
1811 			 * and ack.
1812 			 */
1813 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1814 				tp->t_flags |= TF_ACKNOW;
1815 				tcpstat.tcps_rcvwinprobe++;
1816 			} else
1817 				goto dropafterack;
1818 		} else
1819 			tcpstat.tcps_rcvbyteafterwin += todrop;
1820 		m_adj(m, -todrop);
1821 		tlen -= todrop;
1822 		tiflags &= ~(TH_PUSH|TH_FIN);
1823 	}
1824 
1825 	/*
1826 	 * If last ACK falls within this segment's sequence numbers,
1827 	 * and the timestamp is newer, record it.
1828 	 */
1829 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1830 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1831 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
1832 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1833 		tp->ts_recent_age = TCP_TIMESTAMP(tp);
1834 		tp->ts_recent = opti.ts_val;
1835 	}
1836 
1837 	/*
1838 	 * If the RST bit is set examine the state:
1839 	 *    SYN_RECEIVED STATE:
1840 	 *	If passive open, return to LISTEN state.
1841 	 *	If active open, inform user that connection was refused.
1842 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1843 	 *	Inform user that connection was reset, and close tcb.
1844 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1845 	 *	Close the tcb.
1846 	 */
1847 	if (tiflags&TH_RST) switch (tp->t_state) {
1848 
1849 	case TCPS_SYN_RECEIVED:
1850 		so->so_error = ECONNREFUSED;
1851 		goto close;
1852 
1853 	case TCPS_ESTABLISHED:
1854 	case TCPS_FIN_WAIT_1:
1855 	case TCPS_FIN_WAIT_2:
1856 	case TCPS_CLOSE_WAIT:
1857 		so->so_error = ECONNRESET;
1858 	close:
1859 		tp->t_state = TCPS_CLOSED;
1860 		tcpstat.tcps_drops++;
1861 		tp = tcp_close(tp);
1862 		goto drop;
1863 
1864 	case TCPS_CLOSING:
1865 	case TCPS_LAST_ACK:
1866 	case TCPS_TIME_WAIT:
1867 		tp = tcp_close(tp);
1868 		goto drop;
1869 	}
1870 
1871 	/*
1872 	 * If a SYN is in the window, then this is an
1873 	 * error and we send an RST and drop the connection.
1874 	 */
1875 	if (tiflags & TH_SYN) {
1876 		tp = tcp_drop(tp, ECONNRESET);
1877 		goto dropwithreset;
1878 	}
1879 
1880 	/*
1881 	 * If the ACK bit is off we drop the segment and return.
1882 	 */
1883 	if ((tiflags & TH_ACK) == 0) {
1884 		if (tp->t_flags & TF_ACKNOW)
1885 			goto dropafterack;
1886 		else
1887 			goto drop;
1888 	}
1889 
1890 	/*
1891 	 * Ack processing.
1892 	 */
1893 	switch (tp->t_state) {
1894 
1895 	/*
1896 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1897 	 * ESTABLISHED state and continue processing, otherwise
1898 	 * send an RST.
1899 	 */
1900 	case TCPS_SYN_RECEIVED:
1901 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
1902 		    SEQ_GT(th->th_ack, tp->snd_max))
1903 			goto dropwithreset;
1904 		tcpstat.tcps_connects++;
1905 		soisconnected(so);
1906 		tcp_established(tp);
1907 		/* Do window scaling? */
1908 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1909 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1910 			tp->snd_scale = tp->requested_s_scale;
1911 			tp->rcv_scale = tp->request_r_scale;
1912 		}
1913 		TCP_REASS_LOCK(tp);
1914 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1915 		TCP_REASS_UNLOCK(tp);
1916 		tp->snd_wl1 = th->th_seq - 1;
1917 		/* fall into ... */
1918 
1919 	/*
1920 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1921 	 * ACKs.  If the ack is in the range
1922 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1923 	 * then advance tp->snd_una to th->th_ack and drop
1924 	 * data from the retransmission queue.  If this ACK reflects
1925 	 * more up to date window information we update our window information.
1926 	 */
1927 	case TCPS_ESTABLISHED:
1928 	case TCPS_FIN_WAIT_1:
1929 	case TCPS_FIN_WAIT_2:
1930 	case TCPS_CLOSE_WAIT:
1931 	case TCPS_CLOSING:
1932 	case TCPS_LAST_ACK:
1933 	case TCPS_TIME_WAIT:
1934 
1935 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1936 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1937 				tcpstat.tcps_rcvdupack++;
1938 				/*
1939 				 * If we have outstanding data (other than
1940 				 * a window probe), this is a completely
1941 				 * duplicate ack (ie, window info didn't
1942 				 * change), the ack is the biggest we've
1943 				 * seen and we've seen exactly our rexmt
1944 				 * threshhold of them, assume a packet
1945 				 * has been dropped and retransmit it.
1946 				 * Kludge snd_nxt & the congestion
1947 				 * window so we send only this one
1948 				 * packet.
1949 				 *
1950 				 * We know we're losing at the current
1951 				 * window size so do congestion avoidance
1952 				 * (set ssthresh to half the current window
1953 				 * and pull our congestion window back to
1954 				 * the new ssthresh).
1955 				 *
1956 				 * Dup acks mean that packets have left the
1957 				 * network (they're now cached at the receiver)
1958 				 * so bump cwnd by the amount in the receiver
1959 				 * to keep a constant cwnd packets in the
1960 				 * network.
1961 				 */
1962 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
1963 				    th->th_ack != tp->snd_una)
1964 					tp->t_dupacks = 0;
1965 				else if (++tp->t_dupacks == tcprexmtthresh) {
1966 					tcp_seq onxt = tp->snd_nxt;
1967 					u_int win =
1968 					    min(tp->snd_wnd, tp->snd_cwnd) /
1969 					    2 /	tp->t_segsz;
1970 					if (tcp_do_newreno && SEQ_LT(th->th_ack,
1971 					    tp->snd_recover)) {
1972 						/*
1973 						 * False fast retransmit after
1974 						 * timeout.  Do not cut window.
1975 						 */
1976 						tp->snd_cwnd += tp->t_segsz;
1977 						tp->t_dupacks = 0;
1978 						(void) tcp_output(tp);
1979 						goto drop;
1980 					}
1981 
1982 					if (win < 2)
1983 						win = 2;
1984 					tp->snd_ssthresh = win * tp->t_segsz;
1985 					tp->snd_recover = tp->snd_max;
1986 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1987 					tp->t_rtttime = 0;
1988 					tp->snd_nxt = th->th_ack;
1989 					tp->snd_cwnd = tp->t_segsz;
1990 					(void) tcp_output(tp);
1991 					tp->snd_cwnd = tp->snd_ssthresh +
1992 					       tp->t_segsz * tp->t_dupacks;
1993 					if (SEQ_GT(onxt, tp->snd_nxt))
1994 						tp->snd_nxt = onxt;
1995 					goto drop;
1996 				} else if (tp->t_dupacks > tcprexmtthresh) {
1997 					tp->snd_cwnd += tp->t_segsz;
1998 					(void) tcp_output(tp);
1999 					goto drop;
2000 				}
2001 			} else
2002 				tp->t_dupacks = 0;
2003 			break;
2004 		}
2005 		/*
2006 		 * If the congestion window was inflated to account
2007 		 * for the other side's cached packets, retract it.
2008 		 */
2009 		if (tcp_do_newreno == 0) {
2010 			if (tp->t_dupacks >= tcprexmtthresh &&
2011 			    tp->snd_cwnd > tp->snd_ssthresh)
2012 				tp->snd_cwnd = tp->snd_ssthresh;
2013 			tp->t_dupacks = 0;
2014 		} else if (tp->t_dupacks >= tcprexmtthresh &&
2015 			   tcp_newreno(tp, th) == 0) {
2016 			tp->snd_cwnd = tp->snd_ssthresh;
2017 			/*
2018 			 * Window inflation should have left us with approx.
2019 			 * snd_ssthresh outstanding data.  But in case we
2020 			 * would be inclined to send a burst, better to do
2021 			 * it via the slow start mechanism.
2022 			 */
2023 			if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
2024 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
2025 				    + tp->t_segsz;
2026 			tp->t_dupacks = 0;
2027 		}
2028 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2029 			tcpstat.tcps_rcvacktoomuch++;
2030 			goto dropafterack;
2031 		}
2032 		acked = th->th_ack - tp->snd_una;
2033 		tcpstat.tcps_rcvackpack++;
2034 		tcpstat.tcps_rcvackbyte += acked;
2035 
2036 		/*
2037 		 * If we have a timestamp reply, update smoothed
2038 		 * round trip time.  If no timestamp is present but
2039 		 * transmit timer is running and timed sequence
2040 		 * number was acked, update smoothed round trip time.
2041 		 * Since we now have an rtt measurement, cancel the
2042 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2043 		 * Recompute the initial retransmit timer.
2044 		 */
2045 		if (opti.ts_present && opti.ts_ecr)
2046 			tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
2047 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2048 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2049 
2050 		/*
2051 		 * If all outstanding data is acked, stop retransmit
2052 		 * timer and remember to restart (more output or persist).
2053 		 * If there is more data to be acked, restart retransmit
2054 		 * timer, using current (possibly backed-off) value.
2055 		 */
2056 		if (th->th_ack == tp->snd_max) {
2057 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2058 			needoutput = 1;
2059 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2060 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2061 		/*
2062 		 * When new data is acked, open the congestion window.
2063 		 * If the window gives us less than ssthresh packets
2064 		 * in flight, open exponentially (segsz per packet).
2065 		 * Otherwise open linearly: segsz per window
2066 		 * (segsz^2 / cwnd per packet), plus a constant
2067 		 * fraction of a packet (segsz/8) to help larger windows
2068 		 * open quickly enough.
2069 		 */
2070 		{
2071 		u_int cw = tp->snd_cwnd;
2072 		u_int incr = tp->t_segsz;
2073 
2074 		if (cw > tp->snd_ssthresh)
2075 			incr = incr * incr / cw;
2076 		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
2077 			tp->snd_cwnd = min(cw + incr,
2078 			    TCP_MAXWIN << tp->snd_scale);
2079 		}
2080 		ND6_HINT(tp);
2081 		if (acked > so->so_snd.sb_cc) {
2082 			tp->snd_wnd -= so->so_snd.sb_cc;
2083 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2084 			ourfinisacked = 1;
2085 		} else {
2086 			int slen;
2087 
2088 			slen = tp->t_lastm->m_len;
2089 			sbdrop(&so->so_snd, acked);
2090 			tp->snd_wnd -= acked;
2091 			if (so->so_snd.sb_cc != 0) {
2092 				tp->t_lastoff -= acked;
2093 				if (tp->t_lastm->m_len != slen)
2094 					tp->t_inoff -=
2095 					    (slen - tp->t_lastm->m_len);
2096 			}
2097 			ourfinisacked = 0;
2098 		}
2099 		sowwakeup(so);
2100 		/*
2101 		 * We want snd_recover to track snd_una to
2102 		 * avoid sequence wraparound problems for
2103 		 * very large transfers.
2104 		 */
2105 		tp->snd_una = tp->snd_recover = th->th_ack;
2106 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2107 			tp->snd_nxt = tp->snd_una;
2108 
2109 		switch (tp->t_state) {
2110 
2111 		/*
2112 		 * In FIN_WAIT_1 STATE in addition to the processing
2113 		 * for the ESTABLISHED state if our FIN is now acknowledged
2114 		 * then enter FIN_WAIT_2.
2115 		 */
2116 		case TCPS_FIN_WAIT_1:
2117 			if (ourfinisacked) {
2118 				/*
2119 				 * If we can't receive any more
2120 				 * data, then closing user can proceed.
2121 				 * Starting the timer is contrary to the
2122 				 * specification, but if we don't get a FIN
2123 				 * we'll hang forever.
2124 				 */
2125 				if (so->so_state & SS_CANTRCVMORE) {
2126 					soisdisconnected(so);
2127 					if (tcp_maxidle > 0)
2128 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2129 						    tcp_maxidle);
2130 				}
2131 				tp->t_state = TCPS_FIN_WAIT_2;
2132 			}
2133 			break;
2134 
2135 	 	/*
2136 		 * In CLOSING STATE in addition to the processing for
2137 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2138 		 * then enter the TIME-WAIT state, otherwise ignore
2139 		 * the segment.
2140 		 */
2141 		case TCPS_CLOSING:
2142 			if (ourfinisacked) {
2143 				tp->t_state = TCPS_TIME_WAIT;
2144 				tcp_canceltimers(tp);
2145 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2146 				soisdisconnected(so);
2147 			}
2148 			break;
2149 
2150 		/*
2151 		 * In LAST_ACK, we may still be waiting for data to drain
2152 		 * and/or to be acked, as well as for the ack of our FIN.
2153 		 * If our FIN is now acknowledged, delete the TCB,
2154 		 * enter the closed state and return.
2155 		 */
2156 		case TCPS_LAST_ACK:
2157 			if (ourfinisacked) {
2158 				tp = tcp_close(tp);
2159 				goto drop;
2160 			}
2161 			break;
2162 
2163 		/*
2164 		 * In TIME_WAIT state the only thing that should arrive
2165 		 * is a retransmission of the remote FIN.  Acknowledge
2166 		 * it and restart the finack timer.
2167 		 */
2168 		case TCPS_TIME_WAIT:
2169 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2170 			goto dropafterack;
2171 		}
2172 	}
2173 
2174 step6:
2175 	/*
2176 	 * Update window information.
2177 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2178 	 */
2179 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2180 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2181 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2182 		/* keep track of pure window updates */
2183 		if (tlen == 0 &&
2184 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2185 			tcpstat.tcps_rcvwinupd++;
2186 		tp->snd_wnd = tiwin;
2187 		tp->snd_wl1 = th->th_seq;
2188 		tp->snd_wl2 = th->th_ack;
2189 		if (tp->snd_wnd > tp->max_sndwnd)
2190 			tp->max_sndwnd = tp->snd_wnd;
2191 		needoutput = 1;
2192 	}
2193 
2194 	/*
2195 	 * Process segments with URG.
2196 	 */
2197 	if ((tiflags & TH_URG) && th->th_urp &&
2198 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2199 		/*
2200 		 * This is a kludge, but if we receive and accept
2201 		 * random urgent pointers, we'll crash in
2202 		 * soreceive.  It's hard to imagine someone
2203 		 * actually wanting to send this much urgent data.
2204 		 */
2205 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2206 			th->th_urp = 0;			/* XXX */
2207 			tiflags &= ~TH_URG;		/* XXX */
2208 			goto dodata;			/* XXX */
2209 		}
2210 		/*
2211 		 * If this segment advances the known urgent pointer,
2212 		 * then mark the data stream.  This should not happen
2213 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2214 		 * a FIN has been received from the remote side.
2215 		 * In these states we ignore the URG.
2216 		 *
2217 		 * According to RFC961 (Assigned Protocols),
2218 		 * the urgent pointer points to the last octet
2219 		 * of urgent data.  We continue, however,
2220 		 * to consider it to indicate the first octet
2221 		 * of data past the urgent section as the original
2222 		 * spec states (in one of two places).
2223 		 */
2224 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2225 			tp->rcv_up = th->th_seq + th->th_urp;
2226 			so->so_oobmark = so->so_rcv.sb_cc +
2227 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2228 			if (so->so_oobmark == 0)
2229 				so->so_state |= SS_RCVATMARK;
2230 			sohasoutofband(so);
2231 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2232 		}
2233 		/*
2234 		 * Remove out of band data so doesn't get presented to user.
2235 		 * This can happen independent of advancing the URG pointer,
2236 		 * but if two URG's are pending at once, some out-of-band
2237 		 * data may creep in... ick.
2238 		 */
2239 		if (th->th_urp <= (u_int16_t) tlen
2240 #ifdef SO_OOBINLINE
2241 		     && (so->so_options & SO_OOBINLINE) == 0
2242 #endif
2243 		     )
2244 			tcp_pulloutofband(so, th, m, hdroptlen);
2245 	} else
2246 		/*
2247 		 * If no out of band data is expected,
2248 		 * pull receive urgent pointer along
2249 		 * with the receive window.
2250 		 */
2251 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2252 			tp->rcv_up = tp->rcv_nxt;
2253 dodata:							/* XXX */
2254 
2255 	/*
2256 	 * Process the segment text, merging it into the TCP sequencing queue,
2257 	 * and arranging for acknowledgement of receipt if necessary.
2258 	 * This process logically involves adjusting tp->rcv_wnd as data
2259 	 * is presented to the user (this happens in tcp_usrreq.c,
2260 	 * case PRU_RCVD).  If a FIN has already been received on this
2261 	 * connection then we just ignore the text.
2262 	 */
2263 	if ((tlen || (tiflags & TH_FIN)) &&
2264 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2265 		/*
2266 		 * Insert segment ti into reassembly queue of tcp with
2267 		 * control block tp.  Return TH_FIN if reassembly now includes
2268 		 * a segment with FIN.  The macro form does the common case
2269 		 * inline (segment is the next to be received on an
2270 		 * established connection, and the queue is empty),
2271 		 * avoiding linkage into and removal from the queue and
2272 		 * repetition of various conversions.
2273 		 * Set DELACK for segments received in order, but ack
2274 		 * immediately when segments are out of order
2275 		 * (so fast retransmit can work).
2276 		 */
2277 		/* NOTE: this was TCP_REASS() macro, but used only once */
2278 		TCP_REASS_LOCK(tp);
2279 		if (th->th_seq == tp->rcv_nxt &&
2280 		    TAILQ_FIRST(&tp->segq) == NULL &&
2281 		    tp->t_state == TCPS_ESTABLISHED) {
2282 			TCP_SETUP_ACK(tp, th);
2283 			tp->rcv_nxt += tlen;
2284 			tiflags = th->th_flags & TH_FIN;
2285 			tcpstat.tcps_rcvpack++;
2286 			tcpstat.tcps_rcvbyte += tlen;
2287 			ND6_HINT(tp);
2288 			if (so->so_state & SS_CANTRCVMORE)
2289 				m_freem(m);
2290 			else {
2291 				m_adj(m, hdroptlen);
2292 				sbappendstream(&(so)->so_rcv, m);
2293 			}
2294 			sorwakeup(so);
2295 		} else {
2296 			m_adj(m, hdroptlen);
2297 			tiflags = tcp_reass(tp, th, m, &tlen);
2298 			tp->t_flags |= TF_ACKNOW;
2299 		}
2300 		TCP_REASS_UNLOCK(tp);
2301 
2302 		/*
2303 		 * Note the amount of data that peer has sent into
2304 		 * our window, in order to estimate the sender's
2305 		 * buffer size.
2306 		 */
2307 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2308 	} else {
2309 		m_freem(m);
2310 		m = NULL;
2311 		tiflags &= ~TH_FIN;
2312 	}
2313 
2314 	/*
2315 	 * If FIN is received ACK the FIN and let the user know
2316 	 * that the connection is closing.  Ignore a FIN received before
2317 	 * the connection is fully established.
2318 	 */
2319 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2320 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2321 			socantrcvmore(so);
2322 			tp->t_flags |= TF_ACKNOW;
2323 			tp->rcv_nxt++;
2324 		}
2325 		switch (tp->t_state) {
2326 
2327 	 	/*
2328 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2329 		 */
2330 		case TCPS_ESTABLISHED:
2331 			tp->t_state = TCPS_CLOSE_WAIT;
2332 			break;
2333 
2334 	 	/*
2335 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2336 		 * enter the CLOSING state.
2337 		 */
2338 		case TCPS_FIN_WAIT_1:
2339 			tp->t_state = TCPS_CLOSING;
2340 			break;
2341 
2342 	 	/*
2343 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2344 		 * starting the time-wait timer, turning off the other
2345 		 * standard timers.
2346 		 */
2347 		case TCPS_FIN_WAIT_2:
2348 			tp->t_state = TCPS_TIME_WAIT;
2349 			tcp_canceltimers(tp);
2350 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2351 			soisdisconnected(so);
2352 			break;
2353 
2354 		/*
2355 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2356 		 */
2357 		case TCPS_TIME_WAIT:
2358 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2359 			break;
2360 		}
2361 	}
2362 #ifdef TCP_DEBUG
2363 	if (so->so_options & SO_DEBUG)
2364 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2365 #endif
2366 
2367 	/*
2368 	 * Return any desired output.
2369 	 */
2370 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2371 		(void) tcp_output(tp);
2372 	if (tcp_saveti)
2373 		m_freem(tcp_saveti);
2374 	return;
2375 
2376 badsyn:
2377 	/*
2378 	 * Received a bad SYN.  Increment counters and dropwithreset.
2379 	 */
2380 	tcpstat.tcps_badsyn++;
2381 	tp = NULL;
2382 	goto dropwithreset;
2383 
2384 dropafterack:
2385 	/*
2386 	 * Generate an ACK dropping incoming segment if it occupies
2387 	 * sequence space, where the ACK reflects our state.
2388 	 */
2389 	if (tiflags & TH_RST)
2390 		goto drop;
2391 	m_freem(m);
2392 	tp->t_flags |= TF_ACKNOW;
2393 	(void) tcp_output(tp);
2394 	if (tcp_saveti)
2395 		m_freem(tcp_saveti);
2396 	return;
2397 
2398 dropwithreset_ratelim:
2399 	/*
2400 	 * We may want to rate-limit RSTs in certain situations,
2401 	 * particularly if we are sending an RST in response to
2402 	 * an attempt to connect to or otherwise communicate with
2403 	 * a port for which we have no socket.
2404 	 */
2405 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2406 	    tcp_rst_ppslim) == 0) {
2407 		/* XXX stat */
2408 		goto drop;
2409 	}
2410 	/* ...fall into dropwithreset... */
2411 
2412 dropwithreset:
2413 	/*
2414 	 * Generate a RST, dropping incoming segment.
2415 	 * Make ACK acceptable to originator of segment.
2416 	 */
2417 	if (tiflags & TH_RST)
2418 		goto drop;
2419 
2420 	switch (af) {
2421 #ifdef INET6
2422 	case AF_INET6:
2423 		/* For following calls to tcp_respond */
2424 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2425 			goto drop;
2426 		break;
2427 #endif /* INET6 */
2428 	case AF_INET:
2429 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2430 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2431 			goto drop;
2432 	}
2433 
2434 	if (tiflags & TH_ACK)
2435 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2436 	else {
2437 		if (tiflags & TH_SYN)
2438 			tlen++;
2439 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2440 		    TH_RST|TH_ACK);
2441 	}
2442 	if (tcp_saveti)
2443 		m_freem(tcp_saveti);
2444 	return;
2445 
2446 badcsum:
2447 	tcpstat.tcps_rcvbadsum++;
2448 drop:
2449 	/*
2450 	 * Drop space held by incoming segment and return.
2451 	 */
2452 	if (tp) {
2453 		if (tp->t_inpcb)
2454 			so = tp->t_inpcb->inp_socket;
2455 #ifdef INET6
2456 		else if (tp->t_in6pcb)
2457 			so = tp->t_in6pcb->in6p_socket;
2458 #endif
2459 		else
2460 			so = NULL;
2461 #ifdef TCP_DEBUG
2462 		if (so && (so->so_options & SO_DEBUG) != 0)
2463 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2464 #endif
2465 	}
2466 	if (tcp_saveti)
2467 		m_freem(tcp_saveti);
2468 	m_freem(m);
2469 	return;
2470 }
2471 
2472 void
2473 tcp_dooptions(tp, cp, cnt, th, oi)
2474 	struct tcpcb *tp;
2475 	u_char *cp;
2476 	int cnt;
2477 	struct tcphdr *th;
2478 	struct tcp_opt_info *oi;
2479 {
2480 	u_int16_t mss;
2481 	int opt, optlen;
2482 
2483 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2484 		opt = cp[0];
2485 		if (opt == TCPOPT_EOL)
2486 			break;
2487 		if (opt == TCPOPT_NOP)
2488 			optlen = 1;
2489 		else {
2490 			if (cnt < 2)
2491 				break;
2492 			optlen = cp[1];
2493 			if (optlen < 2 || optlen > cnt)
2494 				break;
2495 		}
2496 		switch (opt) {
2497 
2498 		default:
2499 			continue;
2500 
2501 		case TCPOPT_MAXSEG:
2502 			if (optlen != TCPOLEN_MAXSEG)
2503 				continue;
2504 			if (!(th->th_flags & TH_SYN))
2505 				continue;
2506 			bcopy(cp + 2, &mss, sizeof(mss));
2507 			oi->maxseg = ntohs(mss);
2508 			break;
2509 
2510 		case TCPOPT_WINDOW:
2511 			if (optlen != TCPOLEN_WINDOW)
2512 				continue;
2513 			if (!(th->th_flags & TH_SYN))
2514 				continue;
2515 			tp->t_flags |= TF_RCVD_SCALE;
2516 			tp->requested_s_scale = cp[2];
2517 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2518 #if 0	/*XXX*/
2519 				char *p;
2520 
2521 				if (ip)
2522 					p = ntohl(ip->ip_src);
2523 #ifdef INET6
2524 				else if (ip6)
2525 					p = ip6_sprintf(&ip6->ip6_src);
2526 #endif
2527 				else
2528 					p = "(unknown)";
2529 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2530 				    "assuming %d\n",
2531 				    tp->requested_s_scale, p,
2532 				    TCP_MAX_WINSHIFT);
2533 #else
2534 				log(LOG_ERR, "TCP: invalid wscale %d, "
2535 				    "assuming %d\n",
2536 				    tp->requested_s_scale,
2537 				    TCP_MAX_WINSHIFT);
2538 #endif
2539 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2540 			}
2541 			break;
2542 
2543 		case TCPOPT_TIMESTAMP:
2544 			if (optlen != TCPOLEN_TIMESTAMP)
2545 				continue;
2546 			oi->ts_present = 1;
2547 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2548 			NTOHL(oi->ts_val);
2549 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2550 			NTOHL(oi->ts_ecr);
2551 
2552 			/*
2553 			 * A timestamp received in a SYN makes
2554 			 * it ok to send timestamp requests and replies.
2555 			 */
2556 			if (th->th_flags & TH_SYN) {
2557 				tp->t_flags |= TF_RCVD_TSTMP;
2558 				tp->ts_recent = oi->ts_val;
2559 				tp->ts_recent_age = TCP_TIMESTAMP(tp);
2560 			}
2561 			break;
2562 		case TCPOPT_SACK_PERMITTED:
2563 			if (optlen != TCPOLEN_SACK_PERMITTED)
2564 				continue;
2565 			if (!(th->th_flags & TH_SYN))
2566 				continue;
2567 			tp->t_flags &= ~TF_CANT_TXSACK;
2568 			break;
2569 
2570 		case TCPOPT_SACK:
2571 			if (tp->t_flags & TF_IGNR_RXSACK)
2572 				continue;
2573 			if (optlen % 8 != 2 || optlen < 10)
2574 				continue;
2575 			cp += 2;
2576 			optlen -= 2;
2577 			for (; optlen > 0; cp -= 8, optlen -= 8) {
2578 				tcp_seq lwe, rwe;
2579 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
2580 				NTOHL(lwe);
2581 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
2582 				NTOHL(rwe);
2583 				/* tcp_mark_sacked(tp, lwe, rwe); */
2584 			}
2585 			break;
2586 		}
2587 	}
2588 }
2589 
2590 /*
2591  * Pull out of band byte out of a segment so
2592  * it doesn't appear in the user's data queue.
2593  * It is still reflected in the segment length for
2594  * sequencing purposes.
2595  */
2596 void
2597 tcp_pulloutofband(so, th, m, off)
2598 	struct socket *so;
2599 	struct tcphdr *th;
2600 	struct mbuf *m;
2601 	int off;
2602 {
2603 	int cnt = off + th->th_urp - 1;
2604 
2605 	while (cnt >= 0) {
2606 		if (m->m_len > cnt) {
2607 			char *cp = mtod(m, caddr_t) + cnt;
2608 			struct tcpcb *tp = sototcpcb(so);
2609 
2610 			tp->t_iobc = *cp;
2611 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2612 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2613 			m->m_len--;
2614 			return;
2615 		}
2616 		cnt -= m->m_len;
2617 		m = m->m_next;
2618 		if (m == 0)
2619 			break;
2620 	}
2621 	panic("tcp_pulloutofband");
2622 }
2623 
2624 /*
2625  * Collect new round-trip time estimate
2626  * and update averages and current timeout.
2627  */
2628 void
2629 tcp_xmit_timer(tp, rtt)
2630 	struct tcpcb *tp;
2631 	uint32_t rtt;
2632 {
2633 	int32_t delta;
2634 
2635 	tcpstat.tcps_rttupdated++;
2636 	if (tp->t_srtt != 0) {
2637 		/*
2638 		 * srtt is stored as fixed point with 3 bits after the
2639 		 * binary point (i.e., scaled by 8).  The following magic
2640 		 * is equivalent to the smoothing algorithm in rfc793 with
2641 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2642 		 * point).  Adjust rtt to origin 0.
2643 		 */
2644 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2645 		if ((tp->t_srtt += delta) <= 0)
2646 			tp->t_srtt = 1 << 2;
2647 		/*
2648 		 * We accumulate a smoothed rtt variance (actually, a
2649 		 * smoothed mean difference), then set the retransmit
2650 		 * timer to smoothed rtt + 4 times the smoothed variance.
2651 		 * rttvar is stored as fixed point with 2 bits after the
2652 		 * binary point (scaled by 4).  The following is
2653 		 * equivalent to rfc793 smoothing with an alpha of .75
2654 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2655 		 * rfc793's wired-in beta.
2656 		 */
2657 		if (delta < 0)
2658 			delta = -delta;
2659 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2660 		if ((tp->t_rttvar += delta) <= 0)
2661 			tp->t_rttvar = 1 << 2;
2662 	} else {
2663 		/*
2664 		 * No rtt measurement yet - use the unsmoothed rtt.
2665 		 * Set the variance to half the rtt (so our first
2666 		 * retransmit happens at 3*rtt).
2667 		 */
2668 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2669 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2670 	}
2671 	tp->t_rtttime = 0;
2672 	tp->t_rxtshift = 0;
2673 
2674 	/*
2675 	 * the retransmit should happen at rtt + 4 * rttvar.
2676 	 * Because of the way we do the smoothing, srtt and rttvar
2677 	 * will each average +1/2 tick of bias.  When we compute
2678 	 * the retransmit timer, we want 1/2 tick of rounding and
2679 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2680 	 * firing of the timer.  The bias will give us exactly the
2681 	 * 1.5 tick we need.  But, because the bias is
2682 	 * statistical, we have to test that we don't drop below
2683 	 * the minimum feasible timer (which is 2 ticks).
2684 	 */
2685 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2686 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2687 
2688 	/*
2689 	 * We received an ack for a packet that wasn't retransmitted;
2690 	 * it is probably safe to discard any error indications we've
2691 	 * received recently.  This isn't quite right, but close enough
2692 	 * for now (a route might have failed after we sent a segment,
2693 	 * and the return path might not be symmetrical).
2694 	 */
2695 	tp->t_softerror = 0;
2696 }
2697 
2698 /*
2699  * Checks for partial ack.  If partial ack arrives, force the retransmission
2700  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
2701  * 1.  By setting snd_nxt to th_ack, this forces retransmission timer to
2702  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
2703  */
2704 int
2705 tcp_newreno(tp, th)
2706 	struct tcpcb *tp;
2707 	struct tcphdr *th;
2708 {
2709 	tcp_seq onxt = tp->snd_nxt;
2710 	u_long ocwnd = tp->snd_cwnd;
2711 
2712 	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2713 		/*
2714 		 * snd_una has not yet been updated and the socket's send
2715 		 * buffer has not yet drained off the ACK'd data, so we
2716 		 * have to leave snd_una as it was to get the correct data
2717 		 * offset in tcp_output().
2718 		 */
2719 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2720 		tp->t_rtttime = 0;
2721 		tp->snd_nxt = th->th_ack;
2722 		/*
2723 		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
2724 		 * is not yet updated when we're called.
2725 		 */
2726 		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
2727 		(void) tcp_output(tp);
2728 		tp->snd_cwnd = ocwnd;
2729 		if (SEQ_GT(onxt, tp->snd_nxt))
2730 			tp->snd_nxt = onxt;
2731 		/*
2732 		 * Partial window deflation.  Relies on fact that tp->snd_una
2733 		 * not updated yet.
2734 		 */
2735 		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
2736 		return 1;
2737 	}
2738 	return 0;
2739 }
2740 
2741 
2742 /*
2743  * TCP compressed state engine.  Currently used to hold compressed
2744  * state for SYN_RECEIVED.
2745  */
2746 
2747 u_long	syn_cache_count;
2748 u_int32_t syn_hash1, syn_hash2;
2749 
2750 #define SYN_HASH(sa, sp, dp) \
2751 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
2752 				     ((u_int32_t)(sp)))^syn_hash2)))
2753 #ifndef INET6
2754 #define	SYN_HASHALL(hash, src, dst) \
2755 do {									\
2756 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
2757 		((struct sockaddr_in *)(src))->sin_port,		\
2758 		((struct sockaddr_in *)(dst))->sin_port);		\
2759 } while (/*CONSTCOND*/ 0)
2760 #else
2761 #define SYN_HASH6(sa, sp, dp) \
2762 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
2763 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
2764 	 & 0x7fffffff)
2765 
2766 #define SYN_HASHALL(hash, src, dst) \
2767 do {									\
2768 	switch ((src)->sa_family) {					\
2769 	case AF_INET:							\
2770 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
2771 			((struct sockaddr_in *)(src))->sin_port,	\
2772 			((struct sockaddr_in *)(dst))->sin_port);	\
2773 		break;							\
2774 	case AF_INET6:							\
2775 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
2776 			((struct sockaddr_in6 *)(src))->sin6_port,	\
2777 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
2778 		break;							\
2779 	default:							\
2780 		hash = 0;						\
2781 	}								\
2782 } while (/*CONSTCOND*/0)
2783 #endif /* INET6 */
2784 
2785 #define	SYN_CACHE_RM(sc)						\
2786 do {									\
2787 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
2788 	    (sc), sc_bucketq);						\
2789 	(sc)->sc_tp = NULL;						\
2790 	LIST_REMOVE((sc), sc_tpq);					\
2791 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
2792 	callout_stop(&(sc)->sc_timer);					\
2793 	syn_cache_count--;						\
2794 } while (/*CONSTCOND*/0)
2795 
2796 #define	SYN_CACHE_PUT(sc)						\
2797 do {									\
2798 	if ((sc)->sc_ipopts)						\
2799 		(void) m_free((sc)->sc_ipopts);				\
2800 	if ((sc)->sc_route4.ro_rt != NULL)				\
2801 		RTFREE((sc)->sc_route4.ro_rt);				\
2802 	pool_put(&syn_cache_pool, (sc));				\
2803 } while (/*CONSTCOND*/0)
2804 
2805 struct pool syn_cache_pool;
2806 
2807 /*
2808  * We don't estimate RTT with SYNs, so each packet starts with the default
2809  * RTT and each timer step has a fixed timeout value.
2810  */
2811 #define	SYN_CACHE_TIMER_ARM(sc)						\
2812 do {									\
2813 	TCPT_RANGESET((sc)->sc_rxtcur,					\
2814 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
2815 	    TCPTV_REXMTMAX);						\
2816 	callout_reset(&(sc)->sc_timer,					\
2817 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
2818 } while (/*CONSTCOND*/0)
2819 
2820 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
2821 
2822 void
2823 syn_cache_init()
2824 {
2825 	int i;
2826 
2827 	/* Initialize the hash buckets. */
2828 	for (i = 0; i < tcp_syn_cache_size; i++)
2829 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
2830 
2831 	/* Initialize the syn cache pool. */
2832 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
2833 	    "synpl", NULL);
2834 }
2835 
2836 void
2837 syn_cache_insert(sc, tp)
2838 	struct syn_cache *sc;
2839 	struct tcpcb *tp;
2840 {
2841 	struct syn_cache_head *scp;
2842 	struct syn_cache *sc2;
2843 	int s;
2844 
2845 	/*
2846 	 * If there are no entries in the hash table, reinitialize
2847 	 * the hash secrets.
2848 	 */
2849 	if (syn_cache_count == 0) {
2850 		struct timeval tv;
2851 		microtime(&tv);
2852 		syn_hash1 = arc4random() ^ (u_long)&sc;
2853 		syn_hash2 = arc4random() ^ tv.tv_usec;
2854 	}
2855 
2856 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
2857 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
2858 	scp = &tcp_syn_cache[sc->sc_bucketidx];
2859 
2860 	/*
2861 	 * Make sure that we don't overflow the per-bucket
2862 	 * limit or the total cache size limit.
2863 	 */
2864 	s = splsoftnet();
2865 	if (scp->sch_length >= tcp_syn_bucket_limit) {
2866 		tcpstat.tcps_sc_bucketoverflow++;
2867 		/*
2868 		 * The bucket is full.  Toss the oldest element in the
2869 		 * bucket.  This will be the first entry in the bucket.
2870 		 */
2871 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
2872 #ifdef DIAGNOSTIC
2873 		/*
2874 		 * This should never happen; we should always find an
2875 		 * entry in our bucket.
2876 		 */
2877 		if (sc2 == NULL)
2878 			panic("syn_cache_insert: bucketoverflow: impossible");
2879 #endif
2880 		SYN_CACHE_RM(sc2);
2881 		SYN_CACHE_PUT(sc2);
2882 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
2883 		struct syn_cache_head *scp2, *sce;
2884 
2885 		tcpstat.tcps_sc_overflowed++;
2886 		/*
2887 		 * The cache is full.  Toss the oldest entry in the
2888 		 * first non-empty bucket we can find.
2889 		 *
2890 		 * XXX We would really like to toss the oldest
2891 		 * entry in the cache, but we hope that this
2892 		 * condition doesn't happen very often.
2893 		 */
2894 		scp2 = scp;
2895 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
2896 			sce = &tcp_syn_cache[tcp_syn_cache_size];
2897 			for (++scp2; scp2 != scp; scp2++) {
2898 				if (scp2 >= sce)
2899 					scp2 = &tcp_syn_cache[0];
2900 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
2901 					break;
2902 			}
2903 #ifdef DIAGNOSTIC
2904 			/*
2905 			 * This should never happen; we should always find a
2906 			 * non-empty bucket.
2907 			 */
2908 			if (scp2 == scp)
2909 				panic("syn_cache_insert: cacheoverflow: "
2910 				    "impossible");
2911 #endif
2912 		}
2913 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
2914 		SYN_CACHE_RM(sc2);
2915 		SYN_CACHE_PUT(sc2);
2916 	}
2917 
2918 	/*
2919 	 * Initialize the entry's timer.
2920 	 */
2921 	sc->sc_rxttot = 0;
2922 	sc->sc_rxtshift = 0;
2923 	SYN_CACHE_TIMER_ARM(sc);
2924 
2925 	/* Link it from tcpcb entry */
2926 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
2927 
2928 	/* Put it into the bucket. */
2929 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
2930 	scp->sch_length++;
2931 	syn_cache_count++;
2932 
2933 	tcpstat.tcps_sc_added++;
2934 	splx(s);
2935 }
2936 
2937 /*
2938  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
2939  * If we have retransmitted an entry the maximum number of times, expire
2940  * that entry.
2941  */
2942 void
2943 syn_cache_timer(void *arg)
2944 {
2945 	struct syn_cache *sc = arg;
2946 	int s;
2947 
2948 	s = splsoftnet();
2949 
2950 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
2951 		/* Drop it -- too many retransmissions. */
2952 		goto dropit;
2953 	}
2954 
2955 	/*
2956 	 * Compute the total amount of time this entry has
2957 	 * been on a queue.  If this entry has been on longer
2958 	 * than the keep alive timer would allow, expire it.
2959 	 */
2960 	sc->sc_rxttot += sc->sc_rxtcur;
2961 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
2962 		goto dropit;
2963 
2964 	tcpstat.tcps_sc_retransmitted++;
2965 	(void) syn_cache_respond(sc, NULL);
2966 
2967 	/* Advance the timer back-off. */
2968 	sc->sc_rxtshift++;
2969 	SYN_CACHE_TIMER_ARM(sc);
2970 
2971 	splx(s);
2972 	return;
2973 
2974  dropit:
2975 	tcpstat.tcps_sc_timed_out++;
2976 	SYN_CACHE_RM(sc);
2977 	SYN_CACHE_PUT(sc);
2978 	splx(s);
2979 }
2980 
2981 /*
2982  * Remove syn cache created by the specified tcb entry,
2983  * because this does not make sense to keep them
2984  * (if there's no tcb entry, syn cache entry will never be used)
2985  */
2986 void
2987 syn_cache_cleanup(tp)
2988 	struct tcpcb *tp;
2989 {
2990 	struct syn_cache *sc, *nsc;
2991 	int s;
2992 
2993 	s = splsoftnet();
2994 
2995 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
2996 		nsc = LIST_NEXT(sc, sc_tpq);
2997 
2998 #ifdef DIAGNOSTIC
2999 		if (sc->sc_tp != tp)
3000 			panic("invalid sc_tp in syn_cache_cleanup");
3001 #endif
3002 		SYN_CACHE_RM(sc);
3003 		SYN_CACHE_PUT(sc);
3004 	}
3005 	/* just for safety */
3006 	LIST_INIT(&tp->t_sc);
3007 
3008 	splx(s);
3009 }
3010 
3011 /*
3012  * Find an entry in the syn cache.
3013  */
3014 struct syn_cache *
3015 syn_cache_lookup(src, dst, headp)
3016 	struct sockaddr *src;
3017 	struct sockaddr *dst;
3018 	struct syn_cache_head **headp;
3019 {
3020 	struct syn_cache *sc;
3021 	struct syn_cache_head *scp;
3022 	u_int32_t hash;
3023 	int s;
3024 
3025 	SYN_HASHALL(hash, src, dst);
3026 
3027 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3028 	*headp = scp;
3029 	s = splsoftnet();
3030 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3031 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3032 		if (sc->sc_hash != hash)
3033 			continue;
3034 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3035 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3036 			splx(s);
3037 			return (sc);
3038 		}
3039 	}
3040 	splx(s);
3041 	return (NULL);
3042 }
3043 
3044 /*
3045  * This function gets called when we receive an ACK for a
3046  * socket in the LISTEN state.  We look up the connection
3047  * in the syn cache, and if its there, we pull it out of
3048  * the cache and turn it into a full-blown connection in
3049  * the SYN-RECEIVED state.
3050  *
3051  * The return values may not be immediately obvious, and their effects
3052  * can be subtle, so here they are:
3053  *
3054  *	NULL	SYN was not found in cache; caller should drop the
3055  *		packet and send an RST.
3056  *
3057  *	-1	We were unable to create the new connection, and are
3058  *		aborting it.  An ACK,RST is being sent to the peer
3059  *		(unless we got screwey sequence numbners; see below),
3060  *		because the 3-way handshake has been completed.  Caller
3061  *		should not free the mbuf, since we may be using it.  If
3062  *		we are not, we will free it.
3063  *
3064  *	Otherwise, the return value is a pointer to the new socket
3065  *	associated with the connection.
3066  */
3067 struct socket *
3068 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3069 	struct sockaddr *src;
3070 	struct sockaddr *dst;
3071 	struct tcphdr *th;
3072 	unsigned int hlen, tlen;
3073 	struct socket *so;
3074 	struct mbuf *m;
3075 {
3076 	struct syn_cache *sc;
3077 	struct syn_cache_head *scp;
3078 	struct inpcb *inp = NULL;
3079 #ifdef INET6
3080 	struct in6pcb *in6p = NULL;
3081 #endif
3082 	struct tcpcb *tp = 0;
3083 	struct mbuf *am;
3084 	int s;
3085 	struct socket *oso;
3086 
3087 	s = splsoftnet();
3088 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3089 		splx(s);
3090 		return (NULL);
3091 	}
3092 
3093 	/*
3094 	 * Verify the sequence and ack numbers.  Try getting the correct
3095 	 * response again.
3096 	 */
3097 	if ((th->th_ack != sc->sc_iss + 1) ||
3098 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3099 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3100 		(void) syn_cache_respond(sc, m);
3101 		splx(s);
3102 		return ((struct socket *)(-1));
3103 	}
3104 
3105 	/* Remove this cache entry */
3106 	SYN_CACHE_RM(sc);
3107 	splx(s);
3108 
3109 	/*
3110 	 * Ok, create the full blown connection, and set things up
3111 	 * as they would have been set up if we had created the
3112 	 * connection when the SYN arrived.  If we can't create
3113 	 * the connection, abort it.
3114 	 */
3115 	/*
3116 	 * inp still has the OLD in_pcb stuff, set the
3117 	 * v6-related flags on the new guy, too.   This is
3118 	 * done particularly for the case where an AF_INET6
3119 	 * socket is bound only to a port, and a v4 connection
3120 	 * comes in on that port.
3121 	 * we also copy the flowinfo from the original pcb
3122 	 * to the new one.
3123 	 */
3124 	oso = so;
3125 	so = sonewconn(so, SS_ISCONNECTED);
3126 	if (so == NULL)
3127 		goto resetandabort;
3128 
3129 	switch (so->so_proto->pr_domain->dom_family) {
3130 #ifdef INET
3131 	case AF_INET:
3132 		inp = sotoinpcb(so);
3133 		break;
3134 #endif
3135 #ifdef INET6
3136 	case AF_INET6:
3137 		in6p = sotoin6pcb(so);
3138 		break;
3139 #endif
3140 	}
3141 	switch (src->sa_family) {
3142 #ifdef INET
3143 	case AF_INET:
3144 		if (inp) {
3145 			struct in_ifaddr *ia;
3146 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3147 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3148 			inp->inp_options = ip_srcroute();
3149 			INADDR_TO_IA(inp->inp_laddr, ia);
3150 			KASSERT(ia != NULL);
3151 			KASSERT(inp->inp_ia == NULL);
3152 			inp->inp_ia = ia;
3153 			LIST_INSERT_HEAD(&ia->ia_inpcbs, inp, inp_ialink);
3154 			IFAREF(&ia->ia_ifa);
3155 			in_pcbstate(inp, INP_BOUND);
3156 			if (inp->inp_options == NULL) {
3157 				inp->inp_options = sc->sc_ipopts;
3158 				sc->sc_ipopts = NULL;
3159 			}
3160 		}
3161 #ifdef INET6
3162 		else if (in6p) {
3163 			/* IPv4 packet to AF_INET6 socket */
3164 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3165 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3166 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3167 				&in6p->in6p_laddr.s6_addr32[3],
3168 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3169 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3170 			in6totcpcb(in6p)->t_family = AF_INET;
3171 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3172 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3173 			else
3174 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3175 		}
3176 #endif
3177 		break;
3178 #endif
3179 #ifdef INET6
3180 	case AF_INET6:
3181 		if (in6p) {
3182 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3183 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3184 #if 0
3185 			in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK;
3186 			/*inp->inp_options = ip6_srcroute();*/ /* soon. */
3187 #endif
3188 		}
3189 		break;
3190 #endif
3191 	}
3192 #ifdef INET6
3193 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3194 		struct in6pcb *oin6p = sotoin6pcb(oso);
3195 		/* inherit socket options from the listening socket */
3196 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3197 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3198 			m_freem(in6p->in6p_options);
3199 			in6p->in6p_options = 0;
3200 		}
3201 		ip6_savecontrol(in6p, &in6p->in6p_options,
3202 			mtod(m, struct ip6_hdr *), m);
3203 	}
3204 #endif
3205 
3206 #ifdef IPSEC
3207 	/*
3208 	 * we make a copy of policy, instead of sharing the policy,
3209 	 * for better behavior in terms of SA lookup and dead SA removal.
3210 	 */
3211 	if (inp) {
3212 		/* copy old policy into new socket's */
3213 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3214 			printf("tcp_input: could not copy policy\n");
3215 	}
3216 #ifdef INET6
3217 	else if (in6p) {
3218 		/* copy old policy into new socket's */
3219 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3220 		    in6p->in6p_sp))
3221 			printf("tcp_input: could not copy policy\n");
3222 	}
3223 #endif
3224 #endif
3225 
3226 	/*
3227 	 * Give the new socket our cached route reference.
3228 	 */
3229 	if (inp)
3230 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3231 #ifdef INET6
3232 	else
3233 		in6p->in6p_route = sc->sc_route6;
3234 #endif
3235 	sc->sc_route4.ro_rt = NULL;
3236 
3237 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3238 	if (am == NULL)
3239 		goto resetandabort;
3240 	MCLAIM(am, &tcp_mowner);
3241 	am->m_len = src->sa_len;
3242 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3243 	if (inp) {
3244 		if (in_pcbconnect(inp, am)) {
3245 			(void) m_free(am);
3246 			goto resetandabort;
3247 		}
3248 	}
3249 #ifdef INET6
3250 	else if (in6p) {
3251 		if (src->sa_family == AF_INET) {
3252 			/* IPv4 packet to AF_INET6 socket */
3253 			struct sockaddr_in6 *sin6;
3254 			sin6 = mtod(am, struct sockaddr_in6 *);
3255 			am->m_len = sizeof(*sin6);
3256 			bzero(sin6, sizeof(*sin6));
3257 			sin6->sin6_family = AF_INET6;
3258 			sin6->sin6_len = sizeof(*sin6);
3259 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3260 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3261 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3262 				&sin6->sin6_addr.s6_addr32[3],
3263 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3264 		}
3265 		if (in6_pcbconnect(in6p, am)) {
3266 			(void) m_free(am);
3267 			goto resetandabort;
3268 		}
3269 	}
3270 #endif
3271 	else {
3272 		(void) m_free(am);
3273 		goto resetandabort;
3274 	}
3275 	(void) m_free(am);
3276 
3277 	if (inp)
3278 		tp = intotcpcb(inp);
3279 #ifdef INET6
3280 	else if (in6p)
3281 		tp = in6totcpcb(in6p);
3282 #endif
3283 	else
3284 		tp = NULL;
3285 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3286 	if (sc->sc_request_r_scale != 15) {
3287 		tp->requested_s_scale = sc->sc_requested_s_scale;
3288 		tp->request_r_scale = sc->sc_request_r_scale;
3289 		tp->snd_scale = sc->sc_requested_s_scale;
3290 		tp->rcv_scale = sc->sc_request_r_scale;
3291 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3292 	}
3293 	if (sc->sc_flags & SCF_TIMESTAMP)
3294 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3295 	tp->ts_timebase = sc->sc_timebase;
3296 
3297 	tp->t_template = tcp_template(tp);
3298 	if (tp->t_template == 0) {
3299 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3300 		so = NULL;
3301 		m_freem(m);
3302 		goto abort;
3303 	}
3304 
3305 	tp->iss = sc->sc_iss;
3306 	tp->irs = sc->sc_irs;
3307 	tcp_sendseqinit(tp);
3308 	tcp_rcvseqinit(tp);
3309 	tp->t_state = TCPS_SYN_RECEIVED;
3310 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3311 	tcpstat.tcps_accepts++;
3312 
3313 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3314 	tp->t_ourmss = sc->sc_ourmaxseg;
3315 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3316 
3317 	/*
3318 	 * Initialize the initial congestion window.  If we
3319 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3320 	 * to 1 segment (i.e. the Loss Window).
3321 	 */
3322 	if (sc->sc_rxtshift)
3323 		tp->snd_cwnd = tp->t_peermss;
3324 	else {
3325 		int ss = tcp_init_win;
3326 #ifdef INET
3327 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3328 			ss = tcp_init_win_local;
3329 #endif
3330 #ifdef INET6
3331 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3332 			ss = tcp_init_win_local;
3333 #endif
3334 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3335 	}
3336 
3337 	tcp_rmx_rtt(tp);
3338 	tp->snd_wl1 = sc->sc_irs;
3339 	tp->rcv_up = sc->sc_irs + 1;
3340 
3341 	/*
3342 	 * This is what whould have happened in tcp_output() when
3343 	 * the SYN,ACK was sent.
3344 	 */
3345 	tp->snd_up = tp->snd_una;
3346 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3347 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3348 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3349 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3350 	tp->last_ack_sent = tp->rcv_nxt;
3351 
3352 	tcpstat.tcps_sc_completed++;
3353 	SYN_CACHE_PUT(sc);
3354 	return (so);
3355 
3356 resetandabort:
3357 	(void) tcp_respond(NULL, m, m, th,
3358 			   th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
3359 abort:
3360 	if (so != NULL)
3361 		(void) soabort(so);
3362 	SYN_CACHE_PUT(sc);
3363 	tcpstat.tcps_sc_aborted++;
3364 	return ((struct socket *)(-1));
3365 }
3366 
3367 /*
3368  * This function is called when we get a RST for a
3369  * non-existent connection, so that we can see if the
3370  * connection is in the syn cache.  If it is, zap it.
3371  */
3372 
3373 void
3374 syn_cache_reset(src, dst, th)
3375 	struct sockaddr *src;
3376 	struct sockaddr *dst;
3377 	struct tcphdr *th;
3378 {
3379 	struct syn_cache *sc;
3380 	struct syn_cache_head *scp;
3381 	int s = splsoftnet();
3382 
3383 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3384 		splx(s);
3385 		return;
3386 	}
3387 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3388 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3389 		splx(s);
3390 		return;
3391 	}
3392 	SYN_CACHE_RM(sc);
3393 	splx(s);
3394 	tcpstat.tcps_sc_reset++;
3395 	SYN_CACHE_PUT(sc);
3396 }
3397 
3398 void
3399 syn_cache_unreach(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;
3407 
3408 	s = splsoftnet();
3409 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3410 		splx(s);
3411 		return;
3412 	}
3413 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3414 	if (ntohl (th->th_seq) != sc->sc_iss) {
3415 		splx(s);
3416 		return;
3417 	}
3418 
3419 	/*
3420 	 * If we've rertransmitted 3 times and this is our second error,
3421 	 * we remove the entry.  Otherwise, we allow it to continue on.
3422 	 * This prevents us from incorrectly nuking an entry during a
3423 	 * spurious network outage.
3424 	 *
3425 	 * See tcp_notify().
3426 	 */
3427 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3428 		sc->sc_flags |= SCF_UNREACH;
3429 		splx(s);
3430 		return;
3431 	}
3432 
3433 	SYN_CACHE_RM(sc);
3434 	splx(s);
3435 	tcpstat.tcps_sc_unreach++;
3436 	SYN_CACHE_PUT(sc);
3437 }
3438 
3439 /*
3440  * Given a LISTEN socket and an inbound SYN request, add
3441  * this to the syn cache, and send back a segment:
3442  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3443  * to the source.
3444  *
3445  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3446  * Doing so would require that we hold onto the data and deliver it
3447  * to the application.  However, if we are the target of a SYN-flood
3448  * DoS attack, an attacker could send data which would eventually
3449  * consume all available buffer space if it were ACKed.  By not ACKing
3450  * the data, we avoid this DoS scenario.
3451  */
3452 
3453 int
3454 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi)
3455 	struct sockaddr *src;
3456 	struct sockaddr *dst;
3457 	struct tcphdr *th;
3458 	unsigned int hlen;
3459 	struct socket *so;
3460 	struct mbuf *m;
3461 	u_char *optp;
3462 	int optlen;
3463 	struct tcp_opt_info *oi;
3464 {
3465 	struct tcpcb tb, *tp;
3466 	long win;
3467 	struct syn_cache *sc;
3468 	struct syn_cache_head *scp;
3469 	struct mbuf *ipopts;
3470 
3471 	tp = sototcpcb(so);
3472 
3473 	/*
3474 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3475 	 *
3476 	 * Note this check is performed in tcp_input() very early on.
3477 	 */
3478 
3479 	/*
3480 	 * Initialize some local state.
3481 	 */
3482 	win = sbspace(&so->so_rcv);
3483 	if (win > TCP_MAXWIN)
3484 		win = TCP_MAXWIN;
3485 
3486 	switch (src->sa_family) {
3487 #ifdef INET
3488 	case AF_INET:
3489 		/*
3490 		 * Remember the IP options, if any.
3491 		 */
3492 		ipopts = ip_srcroute();
3493 		break;
3494 #endif
3495 	default:
3496 		ipopts = NULL;
3497 	}
3498 
3499 	if (optp) {
3500 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3501 		tcp_dooptions(&tb, optp, optlen, th, oi);
3502 	} else
3503 		tb.t_flags = 0;
3504 
3505 	/*
3506 	 * See if we already have an entry for this connection.
3507 	 * If we do, resend the SYN,ACK.  We do not count this
3508 	 * as a retransmission (XXX though maybe we should).
3509 	 */
3510 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3511 		tcpstat.tcps_sc_dupesyn++;
3512 		if (ipopts) {
3513 			/*
3514 			 * If we were remembering a previous source route,
3515 			 * forget it and use the new one we've been given.
3516 			 */
3517 			if (sc->sc_ipopts)
3518 				(void) m_free(sc->sc_ipopts);
3519 			sc->sc_ipopts = ipopts;
3520 		}
3521 		sc->sc_timestamp = tb.ts_recent;
3522 		if (syn_cache_respond(sc, m) == 0) {
3523 			tcpstat.tcps_sndacks++;
3524 			tcpstat.tcps_sndtotal++;
3525 		}
3526 		return (1);
3527 	}
3528 
3529 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3530 	if (sc == NULL) {
3531 		if (ipopts)
3532 			(void) m_free(ipopts);
3533 		return (0);
3534 	}
3535 
3536 	/*
3537 	 * Fill in the cache, and put the necessary IP and TCP
3538 	 * options into the reply.
3539 	 */
3540 	bzero(sc, sizeof(struct syn_cache));
3541 	callout_init(&sc->sc_timer);
3542 	bcopy(src, &sc->sc_src, src->sa_len);
3543 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3544 	sc->sc_flags = 0;
3545 	sc->sc_ipopts = ipopts;
3546 	sc->sc_irs = th->th_seq;
3547 	switch (src->sa_family) {
3548 #ifdef INET
3549 	case AF_INET:
3550 	    {
3551 		struct sockaddr_in *srcin = (void *) src;
3552 		struct sockaddr_in *dstin = (void *) dst;
3553 
3554 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3555 		    &srcin->sin_addr, dstin->sin_port,
3556 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3557 		break;
3558 	    }
3559 #endif /* INET */
3560 #ifdef INET6
3561 	case AF_INET6:
3562 	    {
3563 		struct sockaddr_in6 *srcin6 = (void *) src;
3564 		struct sockaddr_in6 *dstin6 = (void *) dst;
3565 
3566 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3567 		    &srcin6->sin6_addr, dstin6->sin6_port,
3568 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3569 		break;
3570 	    }
3571 #endif /* INET6 */
3572 	}
3573 	sc->sc_peermaxseg = oi->maxseg;
3574 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
3575 						m->m_pkthdr.rcvif : NULL,
3576 						sc->sc_src.sa.sa_family);
3577 	sc->sc_win = win;
3578 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
3579 	sc->sc_timestamp = tb.ts_recent;
3580 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
3581 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
3582 		sc->sc_flags |= SCF_TIMESTAMP;
3583 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
3584 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
3585 		sc->sc_requested_s_scale = tb.requested_s_scale;
3586 		sc->sc_request_r_scale = 0;
3587 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
3588 		    TCP_MAXWIN << sc->sc_request_r_scale <
3589 		    so->so_rcv.sb_hiwat)
3590 			sc->sc_request_r_scale++;
3591 	} else {
3592 		sc->sc_requested_s_scale = 15;
3593 		sc->sc_request_r_scale = 15;
3594 	}
3595 	sc->sc_tp = tp;
3596 	if (syn_cache_respond(sc, m) == 0) {
3597 		syn_cache_insert(sc, tp);
3598 		tcpstat.tcps_sndacks++;
3599 		tcpstat.tcps_sndtotal++;
3600 	} else {
3601 		SYN_CACHE_PUT(sc);
3602 		tcpstat.tcps_sc_dropped++;
3603 	}
3604 	return (1);
3605 }
3606 
3607 int
3608 syn_cache_respond(sc, m)
3609 	struct syn_cache *sc;
3610 	struct mbuf *m;
3611 {
3612 	struct route *ro;
3613 	u_int8_t *optp;
3614 	int optlen, error;
3615 	u_int16_t tlen;
3616 	struct ip *ip = NULL;
3617 #ifdef INET6
3618 	struct ip6_hdr *ip6 = NULL;
3619 #endif
3620 	struct tcphdr *th;
3621 	u_int hlen;
3622 
3623 	switch (sc->sc_src.sa.sa_family) {
3624 	case AF_INET:
3625 		hlen = sizeof(struct ip);
3626 		ro = &sc->sc_route4;
3627 		break;
3628 #ifdef INET6
3629 	case AF_INET6:
3630 		hlen = sizeof(struct ip6_hdr);
3631 		ro = (struct route *)&sc->sc_route6;
3632 		break;
3633 #endif
3634 	default:
3635 		if (m)
3636 			m_freem(m);
3637 		return EAFNOSUPPORT;
3638 	}
3639 
3640 	/* Compute the size of the TCP options. */
3641 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
3642 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
3643 
3644 	tlen = hlen + sizeof(struct tcphdr) + optlen;
3645 
3646 	/*
3647 	 * Create the IP+TCP header from scratch.
3648 	 */
3649 	if (m)
3650 		m_freem(m);
3651 #ifdef DIAGNOSTIC
3652 	if (max_linkhdr + tlen > MCLBYTES)
3653 		return (ENOBUFS);
3654 #endif
3655 	MGETHDR(m, M_DONTWAIT, MT_DATA);
3656 	if (m && tlen > MHLEN) {
3657 		MCLGET(m, M_DONTWAIT);
3658 		if ((m->m_flags & M_EXT) == 0) {
3659 			m_freem(m);
3660 			m = NULL;
3661 		}
3662 	}
3663 	if (m == NULL)
3664 		return (ENOBUFS);
3665 	MCLAIM(m, &tcp_tx_mowner);
3666 
3667 	/* Fixup the mbuf. */
3668 	m->m_data += max_linkhdr;
3669 	m->m_len = m->m_pkthdr.len = tlen;
3670 #ifdef IPSEC
3671 	if (sc->sc_tp) {
3672 		struct tcpcb *tp;
3673 		struct socket *so;
3674 
3675 		tp = sc->sc_tp;
3676 		if (tp->t_inpcb)
3677 			so = tp->t_inpcb->inp_socket;
3678 #ifdef INET6
3679 		else if (tp->t_in6pcb)
3680 			so = tp->t_in6pcb->in6p_socket;
3681 #endif
3682 		else
3683 			so = NULL;
3684 		/* use IPsec policy on listening socket, on SYN ACK */
3685 		if (ipsec_setsocket(m, so) != 0) {
3686 			m_freem(m);
3687 			return ENOBUFS;
3688 		}
3689 	}
3690 #endif
3691 	m->m_pkthdr.rcvif = NULL;
3692 	memset(mtod(m, u_char *), 0, tlen);
3693 
3694 	switch (sc->sc_src.sa.sa_family) {
3695 	case AF_INET:
3696 		ip = mtod(m, struct ip *);
3697 		ip->ip_dst = sc->sc_src.sin.sin_addr;
3698 		ip->ip_src = sc->sc_dst.sin.sin_addr;
3699 		ip->ip_p = IPPROTO_TCP;
3700 		th = (struct tcphdr *)(ip + 1);
3701 		th->th_dport = sc->sc_src.sin.sin_port;
3702 		th->th_sport = sc->sc_dst.sin.sin_port;
3703 		break;
3704 #ifdef INET6
3705 	case AF_INET6:
3706 		ip6 = mtod(m, struct ip6_hdr *);
3707 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
3708 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
3709 		ip6->ip6_nxt = IPPROTO_TCP;
3710 		/* ip6_plen will be updated in ip6_output() */
3711 		th = (struct tcphdr *)(ip6 + 1);
3712 		th->th_dport = sc->sc_src.sin6.sin6_port;
3713 		th->th_sport = sc->sc_dst.sin6.sin6_port;
3714 		break;
3715 #endif
3716 	default:
3717 		th = NULL;
3718 	}
3719 
3720 	th->th_seq = htonl(sc->sc_iss);
3721 	th->th_ack = htonl(sc->sc_irs + 1);
3722 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
3723 	th->th_flags = TH_SYN|TH_ACK;
3724 	th->th_win = htons(sc->sc_win);
3725 	/* th_sum already 0 */
3726 	/* th_urp already 0 */
3727 
3728 	/* Tack on the TCP options. */
3729 	optp = (u_int8_t *)(th + 1);
3730 	*optp++ = TCPOPT_MAXSEG;
3731 	*optp++ = 4;
3732 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
3733 	*optp++ = sc->sc_ourmaxseg & 0xff;
3734 
3735 	if (sc->sc_request_r_scale != 15) {
3736 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
3737 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
3738 		    sc->sc_request_r_scale);
3739 		optp += 4;
3740 	}
3741 
3742 	if (sc->sc_flags & SCF_TIMESTAMP) {
3743 		u_int32_t *lp = (u_int32_t *)(optp);
3744 		/* Form timestamp option as shown in appendix A of RFC 1323. */
3745 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
3746 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
3747 		*lp   = htonl(sc->sc_timestamp);
3748 		optp += TCPOLEN_TSTAMP_APPA;
3749 	}
3750 
3751 	/* Compute the packet's checksum. */
3752 	switch (sc->sc_src.sa.sa_family) {
3753 	case AF_INET:
3754 		ip->ip_len = htons(tlen - hlen);
3755 		th->th_sum = 0;
3756 		th->th_sum = in_cksum(m, tlen);
3757 		break;
3758 #ifdef INET6
3759 	case AF_INET6:
3760 		ip6->ip6_plen = htons(tlen - hlen);
3761 		th->th_sum = 0;
3762 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
3763 		break;
3764 #endif
3765 	}
3766 
3767 	/*
3768 	 * Fill in some straggling IP bits.  Note the stack expects
3769 	 * ip_len to be in host order, for convenience.
3770 	 */
3771 	switch (sc->sc_src.sa.sa_family) {
3772 #ifdef INET
3773 	case AF_INET:
3774 		ip->ip_len = htons(tlen);
3775 		ip->ip_ttl = ip_defttl;
3776 		/* XXX tos? */
3777 		break;
3778 #endif
3779 #ifdef INET6
3780 	case AF_INET6:
3781 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3782 		ip6->ip6_vfc |= IPV6_VERSION;
3783 		ip6->ip6_plen = htons(tlen - hlen);
3784 		/* ip6_hlim will be initialized afterwards */
3785 		/* XXX flowlabel? */
3786 		break;
3787 #endif
3788 	}
3789 
3790 	switch (sc->sc_src.sa.sa_family) {
3791 #ifdef INET
3792 	case AF_INET:
3793 		error = ip_output(m, sc->sc_ipopts, ro,
3794 		    (ip_mtudisc ? IP_MTUDISC : 0),
3795 		    NULL);
3796 		break;
3797 #endif
3798 #ifdef INET6
3799 	case AF_INET6:
3800 		ip6->ip6_hlim = in6_selecthlim(NULL,
3801 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
3802 
3803 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro,
3804 			0, NULL, NULL);
3805 		break;
3806 #endif
3807 	default:
3808 		error = EAFNOSUPPORT;
3809 		break;
3810 	}
3811 	return (error);
3812 }
3813