xref: /netbsd-src/sys/netinet/tcp_input.c (revision 37b34d511dea595d3ba03a661cf3b775038ea5f8)
1 /*	$NetBSD: tcp_input.c,v 1.156 2002/10/16 15:15:28 itojun Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  *      This product includes software developed by the University of
47  *      California, Berkeley and its contributors.
48  *      This product includes software developed at the Information
49  *      Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 /*-
73  * Copyright (c) 1997, 1998, 1999, 2001 The NetBSD Foundation, Inc.
74  * All rights reserved.
75  *
76  * This code is derived from software contributed to The NetBSD Foundation
77  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
78  * Facility, NASA Ames Research Center.
79  *
80  * Redistribution and use in source and binary forms, with or without
81  * modification, are permitted provided that the following conditions
82  * are met:
83  * 1. Redistributions of source code must retain the above copyright
84  *    notice, this list of conditions and the following disclaimer.
85  * 2. Redistributions in binary form must reproduce the above copyright
86  *    notice, this list of conditions and the following disclaimer in the
87  *    documentation and/or other materials provided with the distribution.
88  * 3. All advertising materials mentioning features or use of this software
89  *    must display the following acknowledgement:
90  *	This product includes software developed by the NetBSD
91  *	Foundation, Inc. and its contributors.
92  * 4. Neither the name of The NetBSD Foundation nor the names of its
93  *    contributors may be used to endorse or promote products derived
94  *    from this software without specific prior written permission.
95  *
96  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
97  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
98  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
99  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
100  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
101  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
102  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
103  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
104  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
105  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
106  * POSSIBILITY OF SUCH DAMAGE.
107  */
108 
109 /*
110  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
111  *	The Regents of the University of California.  All rights reserved.
112  *
113  * Redistribution and use in source and binary forms, with or without
114  * modification, are permitted provided that the following conditions
115  * are met:
116  * 1. Redistributions of source code must retain the above copyright
117  *    notice, this list of conditions and the following disclaimer.
118  * 2. Redistributions in binary form must reproduce the above copyright
119  *    notice, this list of conditions and the following disclaimer in the
120  *    documentation and/or other materials provided with the distribution.
121  * 3. 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 1.156 2002/10/16 15:15:28 itojun 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/ip_var.h>
184 
185 #ifdef INET6
186 #ifndef INET
187 #include <netinet/in.h>
188 #endif
189 #include <netinet/ip6.h>
190 #include <netinet6/ip6_var.h>
191 #include <netinet6/in6_pcb.h>
192 #include <netinet6/ip6_var.h>
193 #include <netinet6/in6_var.h>
194 #include <netinet/icmp6.h>
195 #include <netinet6/nd6.h>
196 #endif
197 
198 #ifdef PULLDOWN_TEST
199 #ifndef INET6
200 /* always need ip6.h for IP6_EXTHDR_GET */
201 #include <netinet/ip6.h>
202 #endif
203 #endif
204 
205 #include <netinet/tcp.h>
206 #include <netinet/tcp_fsm.h>
207 #include <netinet/tcp_seq.h>
208 #include <netinet/tcp_timer.h>
209 #include <netinet/tcp_var.h>
210 #include <netinet/tcpip.h>
211 #include <netinet/tcp_debug.h>
212 
213 #include <machine/stdarg.h>
214 
215 #ifdef IPSEC
216 #include <netinet6/ipsec.h>
217 #include <netkey/key.h>
218 #endif /*IPSEC*/
219 #ifdef INET6
220 #include "faith.h"
221 #if defined(NFAITH) && NFAITH > 0
222 #include <net/if_faith.h>
223 #endif
224 #endif
225 
226 int	tcprexmtthresh = 3;
227 int	tcp_log_refused;
228 
229 static int tcp_rst_ppslim_count = 0;
230 static struct timeval tcp_rst_ppslim_last;
231 
232 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
233 
234 /* for modulo comparisons of timestamps */
235 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
236 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
237 
238 /*
239  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
240  */
241 #ifdef INET6
242 #define ND6_HINT(tp) \
243 do { \
244 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \
245 	 && tp->t_in6pcb->in6p_route.ro_rt) { \
246 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
247 	} \
248 } while (0)
249 #else
250 #define ND6_HINT(tp)
251 #endif
252 
253 /*
254  * Macro to compute ACK transmission behavior.  Delay the ACK unless
255  * we have already delayed an ACK (must send an ACK every two segments).
256  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
257  * option is enabled.
258  */
259 #define	TCP_SETUP_ACK(tp, th) \
260 do { \
261 	if ((tp)->t_flags & TF_DELACK || \
262 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
263 		tp->t_flags |= TF_ACKNOW; \
264 	else \
265 		TCP_SET_DELACK(tp); \
266 } while (0)
267 
268 /*
269  * Convert TCP protocol fields to host order for easier processing.
270  */
271 #define	TCP_FIELDS_TO_HOST(th)						\
272 do {									\
273 	NTOHL((th)->th_seq);						\
274 	NTOHL((th)->th_ack);						\
275 	NTOHS((th)->th_win);						\
276 	NTOHS((th)->th_urp);						\
277 } while (0)
278 
279 /*
280  * ... and reverse the above.
281  */
282 #define	TCP_FIELDS_TO_NET(th)						\
283 do {									\
284 	HTONL((th)->th_seq);						\
285 	HTONL((th)->th_ack);						\
286 	HTONS((th)->th_win);						\
287 	HTONS((th)->th_urp);						\
288 } while (0)
289 
290 #ifdef TCP_CSUM_COUNTERS
291 #include <sys/device.h>
292 
293 extern struct evcnt tcp_hwcsum_ok;
294 extern struct evcnt tcp_hwcsum_bad;
295 extern struct evcnt tcp_hwcsum_data;
296 extern struct evcnt tcp_swcsum;
297 
298 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
299 
300 #else
301 
302 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
303 
304 #endif /* TCP_CSUM_COUNTERS */
305 
306 #ifdef TCP_REASS_COUNTERS
307 #include <sys/device.h>
308 
309 extern struct evcnt tcp_reass_;
310 extern struct evcnt tcp_reass_empty;
311 extern struct evcnt tcp_reass_iteration[8];
312 extern struct evcnt tcp_reass_prependfirst;
313 extern struct evcnt tcp_reass_prepend;
314 extern struct evcnt tcp_reass_insert;
315 extern struct evcnt tcp_reass_inserttail;
316 extern struct evcnt tcp_reass_append;
317 extern struct evcnt tcp_reass_appendtail;
318 extern struct evcnt tcp_reass_overlaptail;
319 extern struct evcnt tcp_reass_overlapfront;
320 extern struct evcnt tcp_reass_segdup;
321 extern struct evcnt tcp_reass_fragdup;
322 
323 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
324 
325 #else
326 
327 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
328 
329 #endif /* TCP_REASS_COUNTERS */
330 
331 #ifdef INET
332 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *));
333 #endif
334 #ifdef INET6
335 static void tcp6_log_refused
336     __P((const struct ip6_hdr *, const struct tcphdr *));
337 #endif
338 
339 int
340 tcp_reass(tp, th, m, tlen)
341 	struct tcpcb *tp;
342 	struct tcphdr *th;
343 	struct mbuf *m;
344 	int *tlen;
345 {
346 	struct ipqent *p, *q, *nq, *tiqe = NULL;
347 	struct socket *so = NULL;
348 	int pkt_flags;
349 	tcp_seq pkt_seq;
350 	unsigned pkt_len;
351 	u_long rcvpartdupbyte = 0;
352 	u_long rcvoobyte;
353 #ifdef TCP_REASS_COUNTERS
354 	u_int count = 0;
355 #endif
356 
357 	if (tp->t_inpcb)
358 		so = tp->t_inpcb->inp_socket;
359 #ifdef INET6
360 	else if (tp->t_in6pcb)
361 		so = tp->t_in6pcb->in6p_socket;
362 #endif
363 
364 	TCP_REASS_LOCK_CHECK(tp);
365 
366 	/*
367 	 * Call with th==0 after become established to
368 	 * force pre-ESTABLISHED data up to user socket.
369 	 */
370 	if (th == 0)
371 		goto present;
372 
373 	rcvoobyte = *tlen;
374 	/*
375 	 * Copy these to local variables because the tcpiphdr
376 	 * gets munged while we are collapsing mbufs.
377 	 */
378 	pkt_seq = th->th_seq;
379 	pkt_len = *tlen;
380 	pkt_flags = th->th_flags;
381 
382 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
383 
384 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
385 		/*
386 		 * When we miss a packet, the vast majority of time we get
387 		 * packets that follow it in order.  So optimize for that.
388 		 */
389 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
390 			p->ipqe_len += pkt_len;
391 			p->ipqe_flags |= pkt_flags;
392 			m_cat(p->ipqe_m, m);
393 			tiqe = p;
394 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
395 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
396 			goto skip_replacement;
397 		}
398 		/*
399 		 * While we're here, if the pkt is completely beyond
400 		 * anything we have, just insert it at the tail.
401 		 */
402 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
403 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
404 			goto insert_it;
405 		}
406 	}
407 
408 	q = TAILQ_FIRST(&tp->segq);
409 
410 	if (q != NULL) {
411 		/*
412 		 * If this segment immediately precedes the first out-of-order
413 		 * block, simply slap the segment in front of it and (mostly)
414 		 * skip the complicated logic.
415 		 */
416 		if (pkt_seq + pkt_len == q->ipqe_seq) {
417 			q->ipqe_seq = pkt_seq;
418 			q->ipqe_len += pkt_len;
419 			q->ipqe_flags |= pkt_flags;
420 			m_cat(m, q->ipqe_m);
421 			q->ipqe_m = m;
422 			tiqe = q;
423 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
424 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
425 			goto skip_replacement;
426 		}
427 	} else {
428 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
429 	}
430 
431 	/*
432 	 * Find a segment which begins after this one does.
433 	 */
434 	for (p = NULL; q != NULL; q = nq) {
435 		nq = TAILQ_NEXT(q, ipqe_q);
436 #ifdef TCP_REASS_COUNTERS
437 		count++;
438 #endif
439 		/*
440 		 * If the received segment is just right after this
441 		 * fragment, merge the two together and then check
442 		 * for further overlaps.
443 		 */
444 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
445 #ifdef TCPREASS_DEBUG
446 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
447 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
448 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
449 #endif
450 			pkt_len += q->ipqe_len;
451 			pkt_flags |= q->ipqe_flags;
452 			pkt_seq = q->ipqe_seq;
453 			m_cat(q->ipqe_m, m);
454 			m = q->ipqe_m;
455 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
456 			goto free_ipqe;
457 		}
458 		/*
459 		 * If the received segment is completely past this
460 		 * fragment, we need to go the next fragment.
461 		 */
462 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
463 			p = q;
464 			continue;
465 		}
466 		/*
467 		 * If the fragment is past the received segment,
468 		 * it (or any following) can't be concatenated.
469 		 */
470 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
471 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
472 			break;
473 		}
474 
475 		/*
476 		 * We've received all the data in this segment before.
477 		 * mark it as a duplicate and return.
478 		 */
479 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
480 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
481 			tcpstat.tcps_rcvduppack++;
482 			tcpstat.tcps_rcvdupbyte += pkt_len;
483 			m_freem(m);
484 			if (tiqe != NULL)
485 				pool_put(&ipqent_pool, tiqe);
486 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
487 			return (0);
488 		}
489 		/*
490 		 * Received segment completely overlaps this fragment
491 		 * so we drop the fragment (this keeps the temporal
492 		 * ordering of segments correct).
493 		 */
494 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
495 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
496 			rcvpartdupbyte += q->ipqe_len;
497 			m_freem(q->ipqe_m);
498 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
499 			goto free_ipqe;
500 		}
501 		/*
502 		 * RX'ed segment extends past the end of the
503 		 * fragment.  Drop the overlapping bytes.  Then
504 		 * merge the fragment and segment then treat as
505 		 * a longer received packet.
506 		 */
507 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
508 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
509 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
510 #ifdef TCPREASS_DEBUG
511 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
512 			       tp, overlap,
513 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
514 #endif
515 			m_adj(m, overlap);
516 			rcvpartdupbyte += overlap;
517 			m_cat(q->ipqe_m, m);
518 			m = q->ipqe_m;
519 			pkt_seq = q->ipqe_seq;
520 			pkt_len += q->ipqe_len - overlap;
521 			rcvoobyte -= overlap;
522 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
523 			goto free_ipqe;
524 		}
525 		/*
526 		 * RX'ed segment extends past the front of the
527 		 * fragment.  Drop the overlapping bytes on the
528 		 * received packet.  The packet will then be
529 		 * contatentated with this fragment a bit later.
530 		 */
531 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
532 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
533 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
534 #ifdef TCPREASS_DEBUG
535 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
536 			       tp, overlap,
537 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
538 #endif
539 			m_adj(m, -overlap);
540 			pkt_len -= overlap;
541 			rcvpartdupbyte += overlap;
542 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
543 			rcvoobyte -= overlap;
544 		}
545 		/*
546 		 * If the received segment immediates precedes this
547 		 * fragment then tack the fragment onto this segment
548 		 * and reinsert the data.
549 		 */
550 		if (q->ipqe_seq == pkt_seq + pkt_len) {
551 #ifdef TCPREASS_DEBUG
552 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
553 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
554 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
555 #endif
556 			pkt_len += q->ipqe_len;
557 			pkt_flags |= q->ipqe_flags;
558 			m_cat(m, q->ipqe_m);
559 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
560 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
561 			if (tiqe == NULL) {
562 			    tiqe = q;
563 			} else {
564 			    pool_put(&ipqent_pool, q);
565 			}
566 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
567 			break;
568 		}
569 		/*
570 		 * If the fragment is before the segment, remember it.
571 		 * When this loop is terminated, p will contain the
572 		 * pointer to fragment that is right before the received
573 		 * segment.
574 		 */
575 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
576 			p = q;
577 
578 		continue;
579 
580 		/*
581 		 * This is a common operation.  It also will allow
582 		 * to save doing a malloc/free in most instances.
583 		 */
584 	  free_ipqe:
585 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
586 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
587 		if (tiqe == NULL) {
588 		    tiqe = q;
589 		} else {
590 		    pool_put(&ipqent_pool, q);
591 		}
592 	}
593 
594 #ifdef TCP_REASS_COUNTERS
595 	if (count > 7)
596 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
597 	else if (count > 0)
598 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
599 #endif
600 
601     insert_it:
602 
603 	/*
604 	 * Allocate a new queue entry since the received segment did not
605 	 * collapse onto any other out-of-order block; thus we are allocating
606 	 * a new block.  If it had collapsed, tiqe would not be NULL and
607 	 * we would be reusing it.
608 	 * XXX If we can't, just drop the packet.  XXX
609 	 */
610 	if (tiqe == NULL) {
611 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
612 		if (tiqe == NULL) {
613 			tcpstat.tcps_rcvmemdrop++;
614 			m_freem(m);
615 			return (0);
616 		}
617 	}
618 
619 	/*
620 	 * Update the counters.
621 	 */
622 	tcpstat.tcps_rcvoopack++;
623 	tcpstat.tcps_rcvoobyte += rcvoobyte;
624 	if (rcvpartdupbyte) {
625 	    tcpstat.tcps_rcvpartduppack++;
626 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
627 	}
628 
629 	/*
630 	 * Insert the new fragment queue entry into both queues.
631 	 */
632 	tiqe->ipqe_m = m;
633 	tiqe->ipqe_seq = pkt_seq;
634 	tiqe->ipqe_len = pkt_len;
635 	tiqe->ipqe_flags = pkt_flags;
636 	if (p == NULL) {
637 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
638 #ifdef TCPREASS_DEBUG
639 		if (tiqe->ipqe_seq != tp->rcv_nxt)
640 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
641 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
642 #endif
643 	} else {
644 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
645 #ifdef TCPREASS_DEBUG
646 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
647 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
648 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
649 #endif
650 	}
651 
652 skip_replacement:
653 
654 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
655 
656 present:
657 	/*
658 	 * Present data to user, advancing rcv_nxt through
659 	 * completed sequence space.
660 	 */
661 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
662 		return (0);
663 	q = TAILQ_FIRST(&tp->segq);
664 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
665 		return (0);
666 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
667 		return (0);
668 
669 	tp->rcv_nxt += q->ipqe_len;
670 	pkt_flags = q->ipqe_flags & TH_FIN;
671 	ND6_HINT(tp);
672 
673 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
674 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
675 	if (so->so_state & SS_CANTRCVMORE)
676 		m_freem(q->ipqe_m);
677 	else
678 		sbappendstream(&so->so_rcv, q->ipqe_m);
679 	pool_put(&ipqent_pool, q);
680 	sorwakeup(so);
681 	return (pkt_flags);
682 }
683 
684 #ifdef INET6
685 int
686 tcp6_input(mp, offp, proto)
687 	struct mbuf **mp;
688 	int *offp, proto;
689 {
690 	struct mbuf *m = *mp;
691 
692 	/*
693 	 * draft-itojun-ipv6-tcp-to-anycast
694 	 * better place to put this in?
695 	 */
696 	if (m->m_flags & M_ANYCAST6) {
697 		struct ip6_hdr *ip6;
698 		if (m->m_len < sizeof(struct ip6_hdr)) {
699 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
700 				tcpstat.tcps_rcvshort++;
701 				return IPPROTO_DONE;
702 			}
703 		}
704 		ip6 = mtod(m, struct ip6_hdr *);
705 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
706 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
707 		return IPPROTO_DONE;
708 	}
709 
710 	tcp_input(m, *offp, proto);
711 	return IPPROTO_DONE;
712 }
713 #endif
714 
715 #ifdef INET
716 static void
717 tcp4_log_refused(ip, th)
718 	const struct ip *ip;
719 	const struct tcphdr *th;
720 {
721 	char src[4*sizeof "123"];
722 	char dst[4*sizeof "123"];
723 
724 	if (ip) {
725 		strcpy(src, inet_ntoa(ip->ip_src));
726 		strcpy(dst, inet_ntoa(ip->ip_dst));
727 	}
728 	else {
729 		strcpy(src, "(unknown)");
730 		strcpy(dst, "(unknown)");
731 	}
732 	log(LOG_INFO,
733 	    "Connection attempt to TCP %s:%d from %s:%d\n",
734 	    dst, ntohs(th->th_dport),
735 	    src, ntohs(th->th_sport));
736 }
737 #endif
738 
739 #ifdef INET6
740 static void
741 tcp6_log_refused(ip6, th)
742 	const struct ip6_hdr *ip6;
743 	const struct tcphdr *th;
744 {
745 	char src[INET6_ADDRSTRLEN];
746 	char dst[INET6_ADDRSTRLEN];
747 
748 	if (ip6) {
749 		strcpy(src, ip6_sprintf(&ip6->ip6_src));
750 		strcpy(dst, ip6_sprintf(&ip6->ip6_dst));
751 	}
752 	else {
753 		strcpy(src, "(unknown v6)");
754 		strcpy(dst, "(unknown v6)");
755 	}
756 	log(LOG_INFO,
757 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
758 	    dst, ntohs(th->th_dport),
759 	    src, ntohs(th->th_sport));
760 }
761 #endif
762 
763 /*
764  * TCP input routine, follows pages 65-76 of the
765  * protocol specification dated September, 1981 very closely.
766  */
767 void
768 #if __STDC__
769 tcp_input(struct mbuf *m, ...)
770 #else
771 tcp_input(m, va_alist)
772 	struct mbuf *m;
773 #endif
774 {
775 	int proto;
776 	struct tcphdr *th;
777 	struct ip *ip;
778 	struct inpcb *inp;
779 #ifdef INET6
780 	struct ip6_hdr *ip6;
781 	struct in6pcb *in6p;
782 #endif
783 	u_int8_t *optp = NULL;
784 	int optlen = 0;
785 	int len, tlen, toff, hdroptlen = 0;
786 	struct tcpcb *tp = 0;
787 	int tiflags;
788 	struct socket *so = NULL;
789 	int todrop, acked, ourfinisacked, needoutput = 0;
790 	short ostate = 0;
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 	va_start(ap, m);
800 	toff = va_arg(ap, int);
801 	proto = va_arg(ap, int);
802 	va_end(ap);
803 
804 	tcpstat.tcps_rcvtotal++;
805 
806 	bzero(&opti, sizeof(opti));
807 	opti.ts_present = 0;
808 	opti.maxseg = 0;
809 
810 	/*
811 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
812 	 *
813 	 * TCP is, by definition, unicast, so we reject all
814 	 * multicast outright.
815 	 *
816 	 * Note, there are additional src/dst address checks in
817 	 * the AF-specific code below.
818 	 */
819 	if (m->m_flags & (M_BCAST|M_MCAST)) {
820 		/* XXX stat */
821 		goto drop;
822 	}
823 #ifdef INET6
824 	if (m->m_flags & M_ANYCAST6) {
825 		/* XXX stat */
826 		goto drop;
827 	}
828 #endif
829 
830 	/*
831 	 * Get IP and TCP header together in first mbuf.
832 	 * Note: IP leaves IP header in first mbuf.
833 	 */
834 	ip = mtod(m, struct ip *);
835 #ifdef INET6
836 	ip6 = NULL;
837 #endif
838 	switch (ip->ip_v) {
839 #ifdef INET
840 	case 4:
841 		af = AF_INET;
842 		iphlen = sizeof(struct ip);
843 #ifndef PULLDOWN_TEST
844 		/* would like to get rid of this... */
845 		if (toff > sizeof (struct ip)) {
846 			ip_stripoptions(m, (struct mbuf *)0);
847 			toff = sizeof(struct ip);
848 		}
849 		if (m->m_len < toff + sizeof (struct tcphdr)) {
850 			if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) {
851 				tcpstat.tcps_rcvshort++;
852 				return;
853 			}
854 		}
855 		ip = mtod(m, struct ip *);
856 		th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
857 #else
858 		ip = mtod(m, struct ip *);
859 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
860 			sizeof(struct tcphdr));
861 		if (th == NULL) {
862 			tcpstat.tcps_rcvshort++;
863 			return;
864 		}
865 #endif
866 		/* We do the checksum after PCB lookup... */
867 		len = ntohs(ip->ip_len);
868 		tlen = len - toff;
869 		break;
870 #endif
871 #ifdef INET6
872 	case 6:
873 		ip = NULL;
874 		iphlen = sizeof(struct ip6_hdr);
875 		af = AF_INET6;
876 #ifndef PULLDOWN_TEST
877 		if (m->m_len < toff + sizeof(struct tcphdr)) {
878 			m = m_pullup(m, toff + sizeof(struct tcphdr));	/*XXX*/
879 			if (m == NULL) {
880 				tcpstat.tcps_rcvshort++;
881 				return;
882 			}
883 		}
884 		ip6 = mtod(m, struct ip6_hdr *);
885 		th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
886 #else
887 		ip6 = mtod(m, struct ip6_hdr *);
888 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
889 			sizeof(struct tcphdr));
890 		if (th == NULL) {
891 			tcpstat.tcps_rcvshort++;
892 			return;
893 		}
894 #endif
895 
896 		/* Be proactive about malicious use of IPv4 mapped address */
897 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
898 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
899 			/* XXX stat */
900 			goto drop;
901 		}
902 
903 		/*
904 		 * Be proactive about unspecified IPv6 address in source.
905 		 * As we use all-zero to indicate unbounded/unconnected pcb,
906 		 * unspecified IPv6 address can be used to confuse us.
907 		 *
908 		 * Note that packets with unspecified IPv6 destination is
909 		 * already dropped in ip6_input.
910 		 */
911 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
912 			/* XXX stat */
913 			goto drop;
914 		}
915 
916 		/*
917 		 * Make sure destination address is not multicast.
918 		 * Source address checked in ip6_input().
919 		 */
920 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
921 			/* XXX stat */
922 			goto drop;
923 		}
924 
925 		/* We do the checksum after PCB lookup... */
926 		len = m->m_pkthdr.len;
927 		tlen = len - toff;
928 		break;
929 #endif
930 	default:
931 		m_freem(m);
932 		return;
933 	}
934 
935 	KASSERT(TCP_HDR_ALIGNED_P(th));
936 
937 	/*
938 	 * Check that TCP offset makes sense,
939 	 * pull out TCP options and adjust length.		XXX
940 	 */
941 	off = th->th_off << 2;
942 	if (off < sizeof (struct tcphdr) || off > tlen) {
943 		tcpstat.tcps_rcvbadoff++;
944 		goto drop;
945 	}
946 	tlen -= off;
947 
948 	/*
949 	 * tcp_input() has been modified to use tlen to mean the TCP data
950 	 * length throughout the function.  Other functions can use
951 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
952 	 * rja
953 	 */
954 
955 	if (off > sizeof (struct tcphdr)) {
956 #ifndef PULLDOWN_TEST
957 		if (m->m_len < toff + off) {
958 			if ((m = m_pullup(m, toff + off)) == 0) {
959 				tcpstat.tcps_rcvshort++;
960 				return;
961 			}
962 			switch (af) {
963 #ifdef INET
964 			case AF_INET:
965 				ip = mtod(m, struct ip *);
966 				break;
967 #endif
968 #ifdef INET6
969 			case AF_INET6:
970 				ip6 = mtod(m, struct ip6_hdr *);
971 				break;
972 #endif
973 			}
974 			th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
975 		}
976 #else
977 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
978 		if (th == NULL) {
979 			tcpstat.tcps_rcvshort++;
980 			return;
981 		}
982 		/*
983 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
984 		 * (as they're before toff) and we don't need to update those.
985 		 */
986 #endif
987 		KASSERT(TCP_HDR_ALIGNED_P(th));
988 		optlen = off - sizeof (struct tcphdr);
989 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
990 		/*
991 		 * Do quick retrieval of timestamp options ("options
992 		 * prediction?").  If timestamp is the only option and it's
993 		 * formatted as recommended in RFC 1323 appendix A, we
994 		 * quickly get the values now and not bother calling
995 		 * tcp_dooptions(), etc.
996 		 */
997 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
998 		     (optlen > TCPOLEN_TSTAMP_APPA &&
999 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1000 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1001 		     (th->th_flags & TH_SYN) == 0) {
1002 			opti.ts_present = 1;
1003 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1004 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1005 			optp = NULL;	/* we've parsed the options */
1006 		}
1007 	}
1008 	tiflags = th->th_flags;
1009 
1010 	/*
1011 	 * Locate pcb for segment.
1012 	 */
1013 findpcb:
1014 	inp = NULL;
1015 #ifdef INET6
1016 	in6p = NULL;
1017 #endif
1018 	switch (af) {
1019 #ifdef INET
1020 	case AF_INET:
1021 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1022 		    ip->ip_dst, th->th_dport);
1023 		if (inp == 0) {
1024 			++tcpstat.tcps_pcbhashmiss;
1025 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1026 		}
1027 #ifdef INET6
1028 		if (inp == 0) {
1029 			struct in6_addr s, d;
1030 
1031 			/* mapped addr case */
1032 			bzero(&s, sizeof(s));
1033 			s.s6_addr16[5] = htons(0xffff);
1034 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1035 			bzero(&d, sizeof(d));
1036 			d.s6_addr16[5] = htons(0xffff);
1037 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1038 			in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport,
1039 				&d, th->th_dport, 0);
1040 			if (in6p == 0) {
1041 				++tcpstat.tcps_pcbhashmiss;
1042 				in6p = in6_pcblookup_bind(&tcb6, &d,
1043 					th->th_dport, 0);
1044 			}
1045 		}
1046 #endif
1047 #ifndef INET6
1048 		if (inp == 0)
1049 #else
1050 		if (inp == 0 && in6p == 0)
1051 #endif
1052 		{
1053 			++tcpstat.tcps_noport;
1054 			if (tcp_log_refused &&
1055 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1056 				tcp4_log_refused(ip, th);
1057 			}
1058 			TCP_FIELDS_TO_HOST(th);
1059 			goto dropwithreset_ratelim;
1060 		}
1061 #ifdef IPSEC
1062 		if (inp && ipsec4_in_reject(m, inp)) {
1063 			ipsecstat.in_polvio++;
1064 			goto drop;
1065 		}
1066 #ifdef INET6
1067 		else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1068 			ipsecstat.in_polvio++;
1069 			goto drop;
1070 		}
1071 #endif
1072 #endif /*IPSEC*/
1073 		break;
1074 #endif /*INET*/
1075 #ifdef INET6
1076 	case AF_INET6:
1077 	    {
1078 		int faith;
1079 
1080 #if defined(NFAITH) && NFAITH > 0
1081 		faith = faithprefix(&ip6->ip6_dst);
1082 #else
1083 		faith = 0;
1084 #endif
1085 		in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport,
1086 			&ip6->ip6_dst, th->th_dport, faith);
1087 		if (in6p == NULL) {
1088 			++tcpstat.tcps_pcbhashmiss;
1089 			in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst,
1090 				th->th_dport, faith);
1091 		}
1092 		if (in6p == NULL) {
1093 			++tcpstat.tcps_noport;
1094 			if (tcp_log_refused &&
1095 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1096 				tcp6_log_refused(ip6, th);
1097 			}
1098 			TCP_FIELDS_TO_HOST(th);
1099 			goto dropwithreset_ratelim;
1100 		}
1101 #ifdef IPSEC
1102 		if (ipsec6_in_reject(m, in6p)) {
1103 			ipsec6stat.in_polvio++;
1104 			goto drop;
1105 		}
1106 #endif /*IPSEC*/
1107 		break;
1108 	    }
1109 #endif
1110 	}
1111 
1112 	/*
1113 	 * If the state is CLOSED (i.e., TCB does not exist) then
1114 	 * all data in the incoming segment is discarded.
1115 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1116 	 * but should either do a listen or a connect soon.
1117 	 */
1118 	tp = NULL;
1119 	so = NULL;
1120 	if (inp) {
1121 		tp = intotcpcb(inp);
1122 		so = inp->inp_socket;
1123 	}
1124 #ifdef INET6
1125 	else if (in6p) {
1126 		tp = in6totcpcb(in6p);
1127 		so = in6p->in6p_socket;
1128 	}
1129 #endif
1130 	if (tp == 0) {
1131 		TCP_FIELDS_TO_HOST(th);
1132 		goto dropwithreset_ratelim;
1133 	}
1134 	if (tp->t_state == TCPS_CLOSED)
1135 		goto drop;
1136 
1137 	/*
1138 	 * Checksum extended TCP header and data.
1139 	 */
1140 	switch (af) {
1141 #ifdef INET
1142 	case AF_INET:
1143 		switch (m->m_pkthdr.csum_flags &
1144 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
1145 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
1146 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
1147 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
1148 			goto badcsum;
1149 
1150 		case M_CSUM_TCPv4|M_CSUM_DATA:
1151 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
1152 			if ((m->m_pkthdr.csum_data ^ 0xffff) != 0)
1153 				goto badcsum;
1154 			break;
1155 
1156 		case M_CSUM_TCPv4:
1157 			/* Checksum was okay. */
1158 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
1159 			break;
1160 
1161 		default:
1162 			/* Must compute it ourselves. */
1163 			TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
1164 #ifndef PULLDOWN_TEST
1165 		    {
1166 			struct ipovly *ipov;
1167 			ipov = (struct ipovly *)ip;
1168 			bzero(ipov->ih_x1, sizeof ipov->ih_x1);
1169 			ipov->ih_len = htons(tlen + off);
1170 
1171 			if (in_cksum(m, len) != 0)
1172 				goto badcsum;
1173 		    }
1174 #else
1175 			if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1176 				goto badcsum;
1177 #endif /* ! PULLDOWN_TEST */
1178 			break;
1179 		}
1180 		break;
1181 #endif /* INET4 */
1182 
1183 #ifdef INET6
1184 	case AF_INET6:
1185 		if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1186 			goto badcsum;
1187 		break;
1188 #endif /* INET6 */
1189 	}
1190 
1191 	TCP_FIELDS_TO_HOST(th);
1192 
1193 	/* Unscale the window into a 32-bit value. */
1194 	if ((tiflags & TH_SYN) == 0)
1195 		tiwin = th->th_win << tp->snd_scale;
1196 	else
1197 		tiwin = th->th_win;
1198 
1199 #ifdef INET6
1200 	/* save packet options if user wanted */
1201 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1202 		if (in6p->in6p_options) {
1203 			m_freem(in6p->in6p_options);
1204 			in6p->in6p_options = 0;
1205 		}
1206 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1207 	}
1208 #endif
1209 
1210 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1211 		union syn_cache_sa src;
1212 		union syn_cache_sa dst;
1213 
1214 		bzero(&src, sizeof(src));
1215 		bzero(&dst, sizeof(dst));
1216 		switch (af) {
1217 #ifdef INET
1218 		case AF_INET:
1219 			src.sin.sin_len = sizeof(struct sockaddr_in);
1220 			src.sin.sin_family = AF_INET;
1221 			src.sin.sin_addr = ip->ip_src;
1222 			src.sin.sin_port = th->th_sport;
1223 
1224 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1225 			dst.sin.sin_family = AF_INET;
1226 			dst.sin.sin_addr = ip->ip_dst;
1227 			dst.sin.sin_port = th->th_dport;
1228 			break;
1229 #endif
1230 #ifdef INET6
1231 		case AF_INET6:
1232 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1233 			src.sin6.sin6_family = AF_INET6;
1234 			src.sin6.sin6_addr = ip6->ip6_src;
1235 			src.sin6.sin6_port = th->th_sport;
1236 
1237 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1238 			dst.sin6.sin6_family = AF_INET6;
1239 			dst.sin6.sin6_addr = ip6->ip6_dst;
1240 			dst.sin6.sin6_port = th->th_dport;
1241 			break;
1242 #endif /* INET6 */
1243 		default:
1244 			goto badsyn;	/*sanity*/
1245 		}
1246 
1247 		if (so->so_options & SO_DEBUG) {
1248 			ostate = tp->t_state;
1249 
1250 			tcp_saveti = NULL;
1251 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1252 				goto nosave;
1253 
1254 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1255 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1256 				if (!tcp_saveti)
1257 					goto nosave;
1258 			} else {
1259 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1260 				if (!tcp_saveti)
1261 					goto nosave;
1262 				tcp_saveti->m_len = iphlen;
1263 				m_copydata(m, 0, iphlen,
1264 				    mtod(tcp_saveti, caddr_t));
1265 			}
1266 
1267 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1268 				m_freem(tcp_saveti);
1269 				tcp_saveti = NULL;
1270 			} else {
1271 				tcp_saveti->m_len += sizeof(struct tcphdr);
1272 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1273 				    sizeof(struct tcphdr));
1274 			}
1275 			if (tcp_saveti) {
1276 				/*
1277 				 * need to recover version # field, which was
1278 				 * overwritten on ip_cksum computation.
1279 				 */
1280 				struct ip *sip;
1281 				sip = mtod(tcp_saveti, struct ip *);
1282 				switch (af) {
1283 #ifdef INET
1284 				case AF_INET:
1285 					sip->ip_v = 4;
1286 					break;
1287 #endif
1288 #ifdef INET6
1289 				case AF_INET6:
1290 					sip->ip_v = 6;
1291 					break;
1292 #endif
1293 				}
1294 			}
1295 	nosave:;
1296 		}
1297 		if (so->so_options & SO_ACCEPTCONN) {
1298 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1299 				if (tiflags & TH_RST) {
1300 					syn_cache_reset(&src.sa, &dst.sa, th);
1301 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1302 				    (TH_ACK|TH_SYN)) {
1303 					/*
1304 					 * Received a SYN,ACK.  This should
1305 					 * never happen while we are in
1306 					 * LISTEN.  Send an RST.
1307 					 */
1308 					goto badsyn;
1309 				} else if (tiflags & TH_ACK) {
1310 					so = syn_cache_get(&src.sa, &dst.sa,
1311 						th, toff, tlen, so, m);
1312 					if (so == NULL) {
1313 						/*
1314 						 * We don't have a SYN for
1315 						 * this ACK; send an RST.
1316 						 */
1317 						goto badsyn;
1318 					} else if (so ==
1319 					    (struct socket *)(-1)) {
1320 						/*
1321 						 * We were unable to create
1322 						 * the connection.  If the
1323 						 * 3-way handshake was
1324 						 * completed, and RST has
1325 						 * been sent to the peer.
1326 						 * Since the mbuf might be
1327 						 * in use for the reply,
1328 						 * do not free it.
1329 						 */
1330 						m = NULL;
1331 					} else {
1332 						/*
1333 						 * We have created a
1334 						 * full-blown connection.
1335 						 */
1336 						tp = NULL;
1337 						inp = NULL;
1338 #ifdef INET6
1339 						in6p = NULL;
1340 #endif
1341 						switch (so->so_proto->pr_domain->dom_family) {
1342 #ifdef INET
1343 						case AF_INET:
1344 							inp = sotoinpcb(so);
1345 							tp = intotcpcb(inp);
1346 							break;
1347 #endif
1348 #ifdef INET6
1349 						case AF_INET6:
1350 							in6p = sotoin6pcb(so);
1351 							tp = in6totcpcb(in6p);
1352 							break;
1353 #endif
1354 						}
1355 						if (tp == NULL)
1356 							goto badsyn;	/*XXX*/
1357 						tiwin <<= tp->snd_scale;
1358 						goto after_listen;
1359 					}
1360 				} else {
1361 					/*
1362 					 * None of RST, SYN or ACK was set.
1363 					 * This is an invalid packet for a
1364 					 * TCB in LISTEN state.  Send a RST.
1365 					 */
1366 					goto badsyn;
1367 				}
1368 			} else {
1369 				/*
1370 				 * Received a SYN.
1371 				 */
1372 
1373 #ifdef INET6
1374 				/*
1375 				 * If deprecated address is forbidden, we do
1376 				 * not accept SYN to deprecated interface
1377 				 * address to prevent any new inbound
1378 				 * connection from getting established.
1379 				 * When we do not accept SYN, we send a TCP
1380 				 * RST, with deprecated source address (instead
1381 				 * of dropping it).  We compromise it as it is
1382 				 * much better for peer to send a RST, and
1383 				 * RST will be the final packet for the
1384 				 * exchange.
1385 				 *
1386 				 * If we do not forbid deprecated addresses, we
1387 				 * accept the SYN packet.  RFC2462 does not
1388 				 * suggest dropping SYN in this case.
1389 				 * If we decipher RFC2462 5.5.4, it says like
1390 				 * this:
1391 				 * 1. use of deprecated addr with existing
1392 				 *    communication is okay - "SHOULD continue
1393 				 *    to be used"
1394 				 * 2. use of it with new communication:
1395 				 *   (2a) "SHOULD NOT be used if alternate
1396 				 *        address with sufficient scope is
1397 				 *        available"
1398 				 *   (2b) nothing mentioned otherwise.
1399 				 * Here we fall into (2b) case as we have no
1400 				 * choice in our source address selection - we
1401 				 * must obey the peer.
1402 				 *
1403 				 * The wording in RFC2462 is confusing, and
1404 				 * there are multiple description text for
1405 				 * deprecated address handling - worse, they
1406 				 * are not exactly the same.  I believe 5.5.4
1407 				 * is the best one, so we follow 5.5.4.
1408 				 */
1409 				if (af == AF_INET6 && !ip6_use_deprecated) {
1410 					struct in6_ifaddr *ia6;
1411 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1412 					    &ip6->ip6_dst)) &&
1413 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1414 						tp = NULL;
1415 						goto dropwithreset;
1416 					}
1417 				}
1418 #endif
1419 
1420 				/*
1421 				 * LISTEN socket received a SYN
1422 				 * from itself?  This can't possibly
1423 				 * be valid; drop the packet.
1424 				 */
1425 				if (th->th_sport == th->th_dport) {
1426 					int i;
1427 
1428 					switch (af) {
1429 #ifdef INET
1430 					case AF_INET:
1431 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1432 						break;
1433 #endif
1434 #ifdef INET6
1435 					case AF_INET6:
1436 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1437 						break;
1438 #endif
1439 					default:
1440 						i = 1;
1441 					}
1442 					if (i) {
1443 						tcpstat.tcps_badsyn++;
1444 						goto drop;
1445 					}
1446 				}
1447 
1448 				/*
1449 				 * SYN looks ok; create compressed TCP
1450 				 * state for it.
1451 				 */
1452 				if (so->so_qlen <= so->so_qlimit &&
1453 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1454 						so, m, optp, optlen, &opti))
1455 					m = NULL;
1456 			}
1457 			goto drop;
1458 		}
1459 	}
1460 
1461 after_listen:
1462 #ifdef DIAGNOSTIC
1463 	/*
1464 	 * Should not happen now that all embryonic connections
1465 	 * are handled with compressed state.
1466 	 */
1467 	if (tp->t_state == TCPS_LISTEN)
1468 		panic("tcp_input: TCPS_LISTEN");
1469 #endif
1470 
1471 	/*
1472 	 * Segment received on connection.
1473 	 * Reset idle time and keep-alive timer.
1474 	 */
1475 	tp->t_rcvtime = tcp_now;
1476 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1477 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1478 
1479 	/*
1480 	 * Process options.
1481 	 */
1482 	if (optp)
1483 		tcp_dooptions(tp, optp, optlen, th, &opti);
1484 
1485 	/*
1486 	 * Header prediction: check for the two common cases
1487 	 * of a uni-directional data xfer.  If the packet has
1488 	 * no control flags, is in-sequence, the window didn't
1489 	 * change and we're not retransmitting, it's a
1490 	 * candidate.  If the length is zero and the ack moved
1491 	 * forward, we're the sender side of the xfer.  Just
1492 	 * free the data acked & wake any higher level process
1493 	 * that was blocked waiting for space.  If the length
1494 	 * is non-zero and the ack didn't move, we're the
1495 	 * receiver side.  If we're getting packets in-order
1496 	 * (the reassembly queue is empty), add the data to
1497 	 * the socket buffer and note that we need a delayed ack.
1498 	 */
1499 	if (tp->t_state == TCPS_ESTABLISHED &&
1500 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1501 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1502 	    th->th_seq == tp->rcv_nxt &&
1503 	    tiwin && tiwin == tp->snd_wnd &&
1504 	    tp->snd_nxt == tp->snd_max) {
1505 
1506 		/*
1507 		 * If last ACK falls within this segment's sequence numbers,
1508 		 *  record the timestamp.
1509 		 */
1510 		if (opti.ts_present &&
1511 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1512 		    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) {
1513 			tp->ts_recent_age = TCP_TIMESTAMP(tp);
1514 			tp->ts_recent = opti.ts_val;
1515 		}
1516 
1517 		if (tlen == 0) {
1518 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1519 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1520 			    tp->snd_cwnd >= tp->snd_wnd &&
1521 			    tp->t_dupacks < tcprexmtthresh) {
1522 				/*
1523 				 * this is a pure ack for outstanding data.
1524 				 */
1525 				++tcpstat.tcps_predack;
1526 				if (opti.ts_present && opti.ts_ecr)
1527 					tcp_xmit_timer(tp,
1528 					  TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
1529 				else if (tp->t_rtttime &&
1530 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1531 					tcp_xmit_timer(tp,
1532 					tcp_now - tp->t_rtttime);
1533 				acked = th->th_ack - tp->snd_una;
1534 				tcpstat.tcps_rcvackpack++;
1535 				tcpstat.tcps_rcvackbyte += acked;
1536 				ND6_HINT(tp);
1537 				sbdrop(&so->so_snd, acked);
1538 				/*
1539 				 * We want snd_recover to track snd_una to
1540 				 * avoid sequence wraparound problems for
1541 				 * very large transfers.
1542 				 */
1543 				tp->snd_una = tp->snd_recover = th->th_ack;
1544 				m_freem(m);
1545 
1546 				/*
1547 				 * If all outstanding data are acked, stop
1548 				 * retransmit timer, otherwise restart timer
1549 				 * using current (possibly backed-off) value.
1550 				 * If process is waiting for space,
1551 				 * wakeup/selwakeup/signal.  If data
1552 				 * are ready to send, let tcp_output
1553 				 * decide between more output or persist.
1554 				 */
1555 				if (tp->snd_una == tp->snd_max)
1556 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1557 				else if (TCP_TIMER_ISARMED(tp,
1558 				    TCPT_PERSIST) == 0)
1559 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1560 					    tp->t_rxtcur);
1561 
1562 				sowwakeup(so);
1563 				if (so->so_snd.sb_cc)
1564 					(void) tcp_output(tp);
1565 				if (tcp_saveti)
1566 					m_freem(tcp_saveti);
1567 				return;
1568 			}
1569 		} else if (th->th_ack == tp->snd_una &&
1570 		    TAILQ_FIRST(&tp->segq) == NULL &&
1571 		    tlen <= sbspace(&so->so_rcv)) {
1572 			/*
1573 			 * this is a pure, in-sequence data packet
1574 			 * with nothing on the reassembly queue and
1575 			 * we have enough buffer space to take it.
1576 			 */
1577 			++tcpstat.tcps_preddat;
1578 			tp->rcv_nxt += tlen;
1579 			tcpstat.tcps_rcvpack++;
1580 			tcpstat.tcps_rcvbyte += tlen;
1581 			ND6_HINT(tp);
1582 			/*
1583 			 * Drop TCP, IP headers and TCP options then add data
1584 			 * to socket buffer.
1585 			 */
1586 			if (so->so_state & SS_CANTRCVMORE)
1587 				m_freem(m);
1588 			else {
1589 				m_adj(m, toff + off);
1590 				sbappendstream(&so->so_rcv, m);
1591 			}
1592 			sorwakeup(so);
1593 			TCP_SETUP_ACK(tp, th);
1594 			if (tp->t_flags & TF_ACKNOW)
1595 				(void) tcp_output(tp);
1596 			if (tcp_saveti)
1597 				m_freem(tcp_saveti);
1598 			return;
1599 		}
1600 	}
1601 
1602 	/*
1603 	 * Compute mbuf offset to TCP data segment.
1604 	 */
1605 	hdroptlen = toff + off;
1606 
1607 	/*
1608 	 * Calculate amount of space in receive window,
1609 	 * and then do TCP input processing.
1610 	 * Receive window is amount of space in rcv queue,
1611 	 * but not less than advertised window.
1612 	 */
1613 	{ int win;
1614 
1615 	win = sbspace(&so->so_rcv);
1616 	if (win < 0)
1617 		win = 0;
1618 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1619 	}
1620 
1621 	switch (tp->t_state) {
1622 	case TCPS_LISTEN:
1623 		/*
1624 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1625 		 */
1626 		if (m->m_flags & (M_BCAST|M_MCAST))
1627 			goto drop;
1628 		switch (af) {
1629 #ifdef INET6
1630 		case AF_INET6:
1631 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1632 				goto drop;
1633 			break;
1634 #endif /* INET6 */
1635 		case AF_INET:
1636 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1637 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1638 				goto drop;
1639 			break;
1640 		}
1641 		break;
1642 
1643 	/*
1644 	 * If the state is SYN_SENT:
1645 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1646 	 *	if seg contains a RST, then drop the connection.
1647 	 *	if seg does not contain SYN, then drop it.
1648 	 * Otherwise this is an acceptable SYN segment
1649 	 *	initialize tp->rcv_nxt and tp->irs
1650 	 *	if seg contains ack then advance tp->snd_una
1651 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1652 	 *	arrange for segment to be acked (eventually)
1653 	 *	continue processing rest of data/controls, beginning with URG
1654 	 */
1655 	case TCPS_SYN_SENT:
1656 		if ((tiflags & TH_ACK) &&
1657 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1658 		     SEQ_GT(th->th_ack, tp->snd_max)))
1659 			goto dropwithreset;
1660 		if (tiflags & TH_RST) {
1661 			if (tiflags & TH_ACK)
1662 				tp = tcp_drop(tp, ECONNREFUSED);
1663 			goto drop;
1664 		}
1665 		if ((tiflags & TH_SYN) == 0)
1666 			goto drop;
1667 		if (tiflags & TH_ACK) {
1668 			tp->snd_una = tp->snd_recover = th->th_ack;
1669 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1670 				tp->snd_nxt = tp->snd_una;
1671 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1672 		}
1673 		tp->irs = th->th_seq;
1674 		tcp_rcvseqinit(tp);
1675 		tp->t_flags |= TF_ACKNOW;
1676 		tcp_mss_from_peer(tp, opti.maxseg);
1677 
1678 		/*
1679 		 * Initialize the initial congestion window.  If we
1680 		 * had to retransmit the SYN, we must initialize cwnd
1681 		 * to 1 segment (i.e. the Loss Window).
1682 		 */
1683 		if (tp->t_flags & TF_SYN_REXMT)
1684 			tp->snd_cwnd = tp->t_peermss;
1685 		else
1686 			tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
1687 			    tp->t_peermss);
1688 
1689 		tcp_rmx_rtt(tp);
1690 		if (tiflags & TH_ACK) {
1691 			tcpstat.tcps_connects++;
1692 			soisconnected(so);
1693 			tcp_established(tp);
1694 			/* Do window scaling on this connection? */
1695 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1696 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1697 				tp->snd_scale = tp->requested_s_scale;
1698 				tp->rcv_scale = tp->request_r_scale;
1699 			}
1700 			TCP_REASS_LOCK(tp);
1701 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1702 			TCP_REASS_UNLOCK(tp);
1703 			/*
1704 			 * if we didn't have to retransmit the SYN,
1705 			 * use its rtt as our initial srtt & rtt var.
1706 			 */
1707 			if (tp->t_rtttime)
1708 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1709 		} else
1710 			tp->t_state = TCPS_SYN_RECEIVED;
1711 
1712 		/*
1713 		 * Advance th->th_seq to correspond to first data byte.
1714 		 * If data, trim to stay within window,
1715 		 * dropping FIN if necessary.
1716 		 */
1717 		th->th_seq++;
1718 		if (tlen > tp->rcv_wnd) {
1719 			todrop = tlen - tp->rcv_wnd;
1720 			m_adj(m, -todrop);
1721 			tlen = tp->rcv_wnd;
1722 			tiflags &= ~TH_FIN;
1723 			tcpstat.tcps_rcvpackafterwin++;
1724 			tcpstat.tcps_rcvbyteafterwin += todrop;
1725 		}
1726 		tp->snd_wl1 = th->th_seq - 1;
1727 		tp->rcv_up = th->th_seq;
1728 		goto step6;
1729 
1730 	/*
1731 	 * If the state is SYN_RECEIVED:
1732 	 *	If seg contains an ACK, but not for our SYN, drop the input
1733 	 *	and generate an RST.  See page 36, rfc793
1734 	 */
1735 	case TCPS_SYN_RECEIVED:
1736 		if ((tiflags & TH_ACK) &&
1737 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1738 		     SEQ_GT(th->th_ack, tp->snd_max)))
1739 			goto dropwithreset;
1740 		break;
1741 	}
1742 
1743 	/*
1744 	 * States other than LISTEN or SYN_SENT.
1745 	 * First check timestamp, if present.
1746 	 * Then check that at least some bytes of segment are within
1747 	 * receive window.  If segment begins before rcv_nxt,
1748 	 * drop leading data (and SYN); if nothing left, just ack.
1749 	 *
1750 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1751 	 * and it's less than ts_recent, drop it.
1752 	 */
1753 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1754 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1755 
1756 		/* Check to see if ts_recent is over 24 days old.  */
1757 		if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) >
1758 		    TCP_PAWS_IDLE) {
1759 			/*
1760 			 * Invalidate ts_recent.  If this segment updates
1761 			 * ts_recent, the age will be reset later and ts_recent
1762 			 * will get a valid value.  If it does not, setting
1763 			 * ts_recent to zero will at least satisfy the
1764 			 * requirement that zero be placed in the timestamp
1765 			 * echo reply when ts_recent isn't valid.  The
1766 			 * age isn't reset until we get a valid ts_recent
1767 			 * because we don't want out-of-order segments to be
1768 			 * dropped when ts_recent is old.
1769 			 */
1770 			tp->ts_recent = 0;
1771 		} else {
1772 			tcpstat.tcps_rcvduppack++;
1773 			tcpstat.tcps_rcvdupbyte += tlen;
1774 			tcpstat.tcps_pawsdrop++;
1775 			goto dropafterack;
1776 		}
1777 	}
1778 
1779 	todrop = tp->rcv_nxt - th->th_seq;
1780 	if (todrop > 0) {
1781 		if (tiflags & TH_SYN) {
1782 			tiflags &= ~TH_SYN;
1783 			th->th_seq++;
1784 			if (th->th_urp > 1)
1785 				th->th_urp--;
1786 			else {
1787 				tiflags &= ~TH_URG;
1788 				th->th_urp = 0;
1789 			}
1790 			todrop--;
1791 		}
1792 		if (todrop > tlen ||
1793 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1794 			/*
1795 			 * Any valid FIN must be to the left of the window.
1796 			 * At this point the FIN must be a duplicate or
1797 			 * out of sequence; drop it.
1798 			 */
1799 			tiflags &= ~TH_FIN;
1800 			/*
1801 			 * Send an ACK to resynchronize and drop any data.
1802 			 * But keep on processing for RST or ACK.
1803 			 */
1804 			tp->t_flags |= TF_ACKNOW;
1805 			todrop = tlen;
1806 			tcpstat.tcps_rcvdupbyte += todrop;
1807 			tcpstat.tcps_rcvduppack++;
1808 		} else {
1809 			tcpstat.tcps_rcvpartduppack++;
1810 			tcpstat.tcps_rcvpartdupbyte += todrop;
1811 		}
1812 		hdroptlen += todrop;	/*drop from head afterwards*/
1813 		th->th_seq += todrop;
1814 		tlen -= todrop;
1815 		if (th->th_urp > todrop)
1816 			th->th_urp -= todrop;
1817 		else {
1818 			tiflags &= ~TH_URG;
1819 			th->th_urp = 0;
1820 		}
1821 	}
1822 
1823 	/*
1824 	 * If new data are received on a connection after the
1825 	 * user processes are gone, then RST the other end.
1826 	 */
1827 	if ((so->so_state & SS_NOFDREF) &&
1828 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1829 		tp = tcp_close(tp);
1830 		tcpstat.tcps_rcvafterclose++;
1831 		goto dropwithreset;
1832 	}
1833 
1834 	/*
1835 	 * If segment ends after window, drop trailing data
1836 	 * (and PUSH and FIN); if nothing left, just ACK.
1837 	 */
1838 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1839 	if (todrop > 0) {
1840 		tcpstat.tcps_rcvpackafterwin++;
1841 		if (todrop >= tlen) {
1842 			tcpstat.tcps_rcvbyteafterwin += tlen;
1843 			/*
1844 			 * If a new connection request is received
1845 			 * while in TIME_WAIT, drop the old connection
1846 			 * and start over if the sequence numbers
1847 			 * are above the previous ones.
1848 			 *
1849 			 * NOTE: We will checksum the packet again, and
1850 			 * so we need to put the header fields back into
1851 			 * network order!
1852 			 * XXX This kind of sucks, but we don't expect
1853 			 * XXX this to happen very often, so maybe it
1854 			 * XXX doesn't matter so much.
1855 			 */
1856 			if (tiflags & TH_SYN &&
1857 			    tp->t_state == TCPS_TIME_WAIT &&
1858 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1859 				iss = tcp_new_iss(tp, tp->snd_nxt);
1860 				tp = tcp_close(tp);
1861 				TCP_FIELDS_TO_NET(th);
1862 				goto findpcb;
1863 			}
1864 			/*
1865 			 * If window is closed can only take segments at
1866 			 * window edge, and have to drop data and PUSH from
1867 			 * incoming segments.  Continue processing, but
1868 			 * remember to ack.  Otherwise, drop segment
1869 			 * and ack.
1870 			 */
1871 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1872 				tp->t_flags |= TF_ACKNOW;
1873 				tcpstat.tcps_rcvwinprobe++;
1874 			} else
1875 				goto dropafterack;
1876 		} else
1877 			tcpstat.tcps_rcvbyteafterwin += todrop;
1878 		m_adj(m, -todrop);
1879 		tlen -= todrop;
1880 		tiflags &= ~(TH_PUSH|TH_FIN);
1881 	}
1882 
1883 	/*
1884 	 * If last ACK falls within this segment's sequence numbers,
1885 	 * and the timestamp is newer, record it.
1886 	 */
1887 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1888 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1889 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
1890 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1891 		tp->ts_recent_age = TCP_TIMESTAMP(tp);
1892 		tp->ts_recent = opti.ts_val;
1893 	}
1894 
1895 	/*
1896 	 * If the RST bit is set examine the state:
1897 	 *    SYN_RECEIVED STATE:
1898 	 *	If passive open, return to LISTEN state.
1899 	 *	If active open, inform user that connection was refused.
1900 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1901 	 *	Inform user that connection was reset, and close tcb.
1902 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1903 	 *	Close the tcb.
1904 	 */
1905 	if (tiflags&TH_RST) switch (tp->t_state) {
1906 
1907 	case TCPS_SYN_RECEIVED:
1908 		so->so_error = ECONNREFUSED;
1909 		goto close;
1910 
1911 	case TCPS_ESTABLISHED:
1912 	case TCPS_FIN_WAIT_1:
1913 	case TCPS_FIN_WAIT_2:
1914 	case TCPS_CLOSE_WAIT:
1915 		so->so_error = ECONNRESET;
1916 	close:
1917 		tp->t_state = TCPS_CLOSED;
1918 		tcpstat.tcps_drops++;
1919 		tp = tcp_close(tp);
1920 		goto drop;
1921 
1922 	case TCPS_CLOSING:
1923 	case TCPS_LAST_ACK:
1924 	case TCPS_TIME_WAIT:
1925 		tp = tcp_close(tp);
1926 		goto drop;
1927 	}
1928 
1929 	/*
1930 	 * If a SYN is in the window, then this is an
1931 	 * error and we send an RST and drop the connection.
1932 	 */
1933 	if (tiflags & TH_SYN) {
1934 		tp = tcp_drop(tp, ECONNRESET);
1935 		goto dropwithreset;
1936 	}
1937 
1938 	/*
1939 	 * If the ACK bit is off we drop the segment and return.
1940 	 */
1941 	if ((tiflags & TH_ACK) == 0) {
1942 		if (tp->t_flags & TF_ACKNOW)
1943 			goto dropafterack;
1944 		else
1945 			goto drop;
1946 	}
1947 
1948 	/*
1949 	 * Ack processing.
1950 	 */
1951 	switch (tp->t_state) {
1952 
1953 	/*
1954 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1955 	 * ESTABLISHED state and continue processing, otherwise
1956 	 * send an RST.
1957 	 */
1958 	case TCPS_SYN_RECEIVED:
1959 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
1960 		    SEQ_GT(th->th_ack, tp->snd_max))
1961 			goto dropwithreset;
1962 		tcpstat.tcps_connects++;
1963 		soisconnected(so);
1964 		tcp_established(tp);
1965 		/* Do window scaling? */
1966 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1967 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1968 			tp->snd_scale = tp->requested_s_scale;
1969 			tp->rcv_scale = tp->request_r_scale;
1970 		}
1971 		TCP_REASS_LOCK(tp);
1972 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1973 		TCP_REASS_UNLOCK(tp);
1974 		tp->snd_wl1 = th->th_seq - 1;
1975 		/* fall into ... */
1976 
1977 	/*
1978 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1979 	 * ACKs.  If the ack is in the range
1980 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1981 	 * then advance tp->snd_una to th->th_ack and drop
1982 	 * data from the retransmission queue.  If this ACK reflects
1983 	 * more up to date window information we update our window information.
1984 	 */
1985 	case TCPS_ESTABLISHED:
1986 	case TCPS_FIN_WAIT_1:
1987 	case TCPS_FIN_WAIT_2:
1988 	case TCPS_CLOSE_WAIT:
1989 	case TCPS_CLOSING:
1990 	case TCPS_LAST_ACK:
1991 	case TCPS_TIME_WAIT:
1992 
1993 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1994 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1995 				tcpstat.tcps_rcvdupack++;
1996 				/*
1997 				 * If we have outstanding data (other than
1998 				 * a window probe), this is a completely
1999 				 * duplicate ack (ie, window info didn't
2000 				 * change), the ack is the biggest we've
2001 				 * seen and we've seen exactly our rexmt
2002 				 * threshhold of them, assume a packet
2003 				 * has been dropped and retransmit it.
2004 				 * Kludge snd_nxt & the congestion
2005 				 * window so we send only this one
2006 				 * packet.
2007 				 *
2008 				 * We know we're losing at the current
2009 				 * window size so do congestion avoidance
2010 				 * (set ssthresh to half the current window
2011 				 * and pull our congestion window back to
2012 				 * the new ssthresh).
2013 				 *
2014 				 * Dup acks mean that packets have left the
2015 				 * network (they're now cached at the receiver)
2016 				 * so bump cwnd by the amount in the receiver
2017 				 * to keep a constant cwnd packets in the
2018 				 * network.
2019 				 */
2020 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2021 				    th->th_ack != tp->snd_una)
2022 					tp->t_dupacks = 0;
2023 				else if (++tp->t_dupacks == tcprexmtthresh) {
2024 					tcp_seq onxt = tp->snd_nxt;
2025 					u_int win =
2026 					    min(tp->snd_wnd, tp->snd_cwnd) /
2027 					    2 /	tp->t_segsz;
2028 					if (tcp_do_newreno && SEQ_LT(th->th_ack,
2029 					    tp->snd_recover)) {
2030 						/*
2031 						 * False fast retransmit after
2032 						 * timeout.  Do not cut window.
2033 						 */
2034 						tp->snd_cwnd += tp->t_segsz;
2035 						tp->t_dupacks = 0;
2036 						(void) tcp_output(tp);
2037 						goto drop;
2038 					}
2039 
2040 					if (win < 2)
2041 						win = 2;
2042 					tp->snd_ssthresh = win * tp->t_segsz;
2043 					tp->snd_recover = tp->snd_max;
2044 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
2045 					tp->t_rtttime = 0;
2046 					tp->snd_nxt = th->th_ack;
2047 					tp->snd_cwnd = tp->t_segsz;
2048 					(void) tcp_output(tp);
2049 					tp->snd_cwnd = tp->snd_ssthresh +
2050 					       tp->t_segsz * tp->t_dupacks;
2051 					if (SEQ_GT(onxt, tp->snd_nxt))
2052 						tp->snd_nxt = onxt;
2053 					goto drop;
2054 				} else if (tp->t_dupacks > tcprexmtthresh) {
2055 					tp->snd_cwnd += tp->t_segsz;
2056 					(void) tcp_output(tp);
2057 					goto drop;
2058 				}
2059 			} else
2060 				tp->t_dupacks = 0;
2061 			break;
2062 		}
2063 		/*
2064 		 * If the congestion window was inflated to account
2065 		 * for the other side's cached packets, retract it.
2066 		 */
2067 		if (tcp_do_newreno == 0) {
2068 			if (tp->t_dupacks >= tcprexmtthresh &&
2069 			    tp->snd_cwnd > tp->snd_ssthresh)
2070 				tp->snd_cwnd = tp->snd_ssthresh;
2071 			tp->t_dupacks = 0;
2072 		} else if (tp->t_dupacks >= tcprexmtthresh &&
2073 			   tcp_newreno(tp, th) == 0) {
2074 			tp->snd_cwnd = tp->snd_ssthresh;
2075 			/*
2076 			 * Window inflation should have left us with approx.
2077 			 * snd_ssthresh outstanding data.  But in case we
2078 			 * would be inclined to send a burst, better to do
2079 			 * it via the slow start mechanism.
2080 			 */
2081 			if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
2082 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
2083 				    + tp->t_segsz;
2084 			tp->t_dupacks = 0;
2085 		}
2086 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2087 			tcpstat.tcps_rcvacktoomuch++;
2088 			goto dropafterack;
2089 		}
2090 		acked = th->th_ack - tp->snd_una;
2091 		tcpstat.tcps_rcvackpack++;
2092 		tcpstat.tcps_rcvackbyte += acked;
2093 
2094 		/*
2095 		 * If we have a timestamp reply, update smoothed
2096 		 * round trip time.  If no timestamp is present but
2097 		 * transmit timer is running and timed sequence
2098 		 * number was acked, update smoothed round trip time.
2099 		 * Since we now have an rtt measurement, cancel the
2100 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2101 		 * Recompute the initial retransmit timer.
2102 		 */
2103 		if (opti.ts_present && opti.ts_ecr)
2104 			tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
2105 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2106 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2107 
2108 		/*
2109 		 * If all outstanding data is acked, stop retransmit
2110 		 * timer and remember to restart (more output or persist).
2111 		 * If there is more data to be acked, restart retransmit
2112 		 * timer, using current (possibly backed-off) value.
2113 		 */
2114 		if (th->th_ack == tp->snd_max) {
2115 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2116 			needoutput = 1;
2117 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2118 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2119 		/*
2120 		 * When new data is acked, open the congestion window.
2121 		 * If the window gives us less than ssthresh packets
2122 		 * in flight, open exponentially (segsz per packet).
2123 		 * Otherwise open linearly: segsz per window
2124 		 * (segsz^2 / cwnd per packet), plus a constant
2125 		 * fraction of a packet (segsz/8) to help larger windows
2126 		 * open quickly enough.
2127 		 */
2128 		{
2129 		u_int cw = tp->snd_cwnd;
2130 		u_int incr = tp->t_segsz;
2131 
2132 		if (cw > tp->snd_ssthresh)
2133 			incr = incr * incr / cw;
2134 		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
2135 			tp->snd_cwnd = min(cw + incr,
2136 			    TCP_MAXWIN << tp->snd_scale);
2137 		}
2138 		ND6_HINT(tp);
2139 		if (acked > so->so_snd.sb_cc) {
2140 			tp->snd_wnd -= so->so_snd.sb_cc;
2141 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2142 			ourfinisacked = 1;
2143 		} else {
2144 			sbdrop(&so->so_snd, acked);
2145 			tp->snd_wnd -= acked;
2146 			ourfinisacked = 0;
2147 		}
2148 		sowwakeup(so);
2149 		/*
2150 		 * We want snd_recover to track snd_una to
2151 		 * avoid sequence wraparound problems for
2152 		 * very large transfers.
2153 		 */
2154 		tp->snd_una = tp->snd_recover = th->th_ack;
2155 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2156 			tp->snd_nxt = tp->snd_una;
2157 
2158 		switch (tp->t_state) {
2159 
2160 		/*
2161 		 * In FIN_WAIT_1 STATE in addition to the processing
2162 		 * for the ESTABLISHED state if our FIN is now acknowledged
2163 		 * then enter FIN_WAIT_2.
2164 		 */
2165 		case TCPS_FIN_WAIT_1:
2166 			if (ourfinisacked) {
2167 				/*
2168 				 * If we can't receive any more
2169 				 * data, then closing user can proceed.
2170 				 * Starting the timer is contrary to the
2171 				 * specification, but if we don't get a FIN
2172 				 * we'll hang forever.
2173 				 */
2174 				if (so->so_state & SS_CANTRCVMORE) {
2175 					soisdisconnected(so);
2176 					if (tcp_maxidle > 0)
2177 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2178 						    tcp_maxidle);
2179 				}
2180 				tp->t_state = TCPS_FIN_WAIT_2;
2181 			}
2182 			break;
2183 
2184 	 	/*
2185 		 * In CLOSING STATE in addition to the processing for
2186 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2187 		 * then enter the TIME-WAIT state, otherwise ignore
2188 		 * the segment.
2189 		 */
2190 		case TCPS_CLOSING:
2191 			if (ourfinisacked) {
2192 				tp->t_state = TCPS_TIME_WAIT;
2193 				tcp_canceltimers(tp);
2194 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2195 				soisdisconnected(so);
2196 			}
2197 			break;
2198 
2199 		/*
2200 		 * In LAST_ACK, we may still be waiting for data to drain
2201 		 * and/or to be acked, as well as for the ack of our FIN.
2202 		 * If our FIN is now acknowledged, delete the TCB,
2203 		 * enter the closed state and return.
2204 		 */
2205 		case TCPS_LAST_ACK:
2206 			if (ourfinisacked) {
2207 				tp = tcp_close(tp);
2208 				goto drop;
2209 			}
2210 			break;
2211 
2212 		/*
2213 		 * In TIME_WAIT state the only thing that should arrive
2214 		 * is a retransmission of the remote FIN.  Acknowledge
2215 		 * it and restart the finack timer.
2216 		 */
2217 		case TCPS_TIME_WAIT:
2218 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2219 			goto dropafterack;
2220 		}
2221 	}
2222 
2223 step6:
2224 	/*
2225 	 * Update window information.
2226 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2227 	 */
2228 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2229 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2230 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2231 		/* keep track of pure window updates */
2232 		if (tlen == 0 &&
2233 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2234 			tcpstat.tcps_rcvwinupd++;
2235 		tp->snd_wnd = tiwin;
2236 		tp->snd_wl1 = th->th_seq;
2237 		tp->snd_wl2 = th->th_ack;
2238 		if (tp->snd_wnd > tp->max_sndwnd)
2239 			tp->max_sndwnd = tp->snd_wnd;
2240 		needoutput = 1;
2241 	}
2242 
2243 	/*
2244 	 * Process segments with URG.
2245 	 */
2246 	if ((tiflags & TH_URG) && th->th_urp &&
2247 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2248 		/*
2249 		 * This is a kludge, but if we receive and accept
2250 		 * random urgent pointers, we'll crash in
2251 		 * soreceive.  It's hard to imagine someone
2252 		 * actually wanting to send this much urgent data.
2253 		 */
2254 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2255 			th->th_urp = 0;			/* XXX */
2256 			tiflags &= ~TH_URG;		/* XXX */
2257 			goto dodata;			/* XXX */
2258 		}
2259 		/*
2260 		 * If this segment advances the known urgent pointer,
2261 		 * then mark the data stream.  This should not happen
2262 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2263 		 * a FIN has been received from the remote side.
2264 		 * In these states we ignore the URG.
2265 		 *
2266 		 * According to RFC961 (Assigned Protocols),
2267 		 * the urgent pointer points to the last octet
2268 		 * of urgent data.  We continue, however,
2269 		 * to consider it to indicate the first octet
2270 		 * of data past the urgent section as the original
2271 		 * spec states (in one of two places).
2272 		 */
2273 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2274 			tp->rcv_up = th->th_seq + th->th_urp;
2275 			so->so_oobmark = so->so_rcv.sb_cc +
2276 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2277 			if (so->so_oobmark == 0)
2278 				so->so_state |= SS_RCVATMARK;
2279 			sohasoutofband(so);
2280 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2281 		}
2282 		/*
2283 		 * Remove out of band data so doesn't get presented to user.
2284 		 * This can happen independent of advancing the URG pointer,
2285 		 * but if two URG's are pending at once, some out-of-band
2286 		 * data may creep in... ick.
2287 		 */
2288 		if (th->th_urp <= (u_int16_t) tlen
2289 #ifdef SO_OOBINLINE
2290 		     && (so->so_options & SO_OOBINLINE) == 0
2291 #endif
2292 		     )
2293 			tcp_pulloutofband(so, th, m, hdroptlen);
2294 	} else
2295 		/*
2296 		 * If no out of band data is expected,
2297 		 * pull receive urgent pointer along
2298 		 * with the receive window.
2299 		 */
2300 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2301 			tp->rcv_up = tp->rcv_nxt;
2302 dodata:							/* XXX */
2303 
2304 	/*
2305 	 * Process the segment text, merging it into the TCP sequencing queue,
2306 	 * and arranging for acknowledgement of receipt if necessary.
2307 	 * This process logically involves adjusting tp->rcv_wnd as data
2308 	 * is presented to the user (this happens in tcp_usrreq.c,
2309 	 * case PRU_RCVD).  If a FIN has already been received on this
2310 	 * connection then we just ignore the text.
2311 	 */
2312 	if ((tlen || (tiflags & TH_FIN)) &&
2313 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2314 		/*
2315 		 * Insert segment ti into reassembly queue of tcp with
2316 		 * control block tp.  Return TH_FIN if reassembly now includes
2317 		 * a segment with FIN.  The macro form does the common case
2318 		 * inline (segment is the next to be received on an
2319 		 * established connection, and the queue is empty),
2320 		 * avoiding linkage into and removal from the queue and
2321 		 * repetition of various conversions.
2322 		 * Set DELACK for segments received in order, but ack
2323 		 * immediately when segments are out of order
2324 		 * (so fast retransmit can work).
2325 		 */
2326 		/* NOTE: this was TCP_REASS() macro, but used only once */
2327 		TCP_REASS_LOCK(tp);
2328 		if (th->th_seq == tp->rcv_nxt &&
2329 		    TAILQ_FIRST(&tp->segq) == NULL &&
2330 		    tp->t_state == TCPS_ESTABLISHED) {
2331 			TCP_SETUP_ACK(tp, th);
2332 			tp->rcv_nxt += tlen;
2333 			tiflags = th->th_flags & TH_FIN;
2334 			tcpstat.tcps_rcvpack++;
2335 			tcpstat.tcps_rcvbyte += tlen;
2336 			ND6_HINT(tp);
2337 			if (so->so_state & SS_CANTRCVMORE)
2338 				m_freem(m);
2339 			else {
2340 				m_adj(m, hdroptlen);
2341 				sbappendstream(&(so)->so_rcv, m);
2342 			}
2343 			sorwakeup(so);
2344 		} else {
2345 			m_adj(m, hdroptlen);
2346 			tiflags = tcp_reass(tp, th, m, &tlen);
2347 			tp->t_flags |= TF_ACKNOW;
2348 		}
2349 		TCP_REASS_UNLOCK(tp);
2350 
2351 		/*
2352 		 * Note the amount of data that peer has sent into
2353 		 * our window, in order to estimate the sender's
2354 		 * buffer size.
2355 		 */
2356 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2357 	} else {
2358 		m_freem(m);
2359 		m = NULL;
2360 		tiflags &= ~TH_FIN;
2361 	}
2362 
2363 	/*
2364 	 * If FIN is received ACK the FIN and let the user know
2365 	 * that the connection is closing.  Ignore a FIN received before
2366 	 * the connection is fully established.
2367 	 */
2368 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2369 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2370 			socantrcvmore(so);
2371 			tp->t_flags |= TF_ACKNOW;
2372 			tp->rcv_nxt++;
2373 		}
2374 		switch (tp->t_state) {
2375 
2376 	 	/*
2377 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2378 		 */
2379 		case TCPS_ESTABLISHED:
2380 			tp->t_state = TCPS_CLOSE_WAIT;
2381 			break;
2382 
2383 	 	/*
2384 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2385 		 * enter the CLOSING state.
2386 		 */
2387 		case TCPS_FIN_WAIT_1:
2388 			tp->t_state = TCPS_CLOSING;
2389 			break;
2390 
2391 	 	/*
2392 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2393 		 * starting the time-wait timer, turning off the other
2394 		 * standard timers.
2395 		 */
2396 		case TCPS_FIN_WAIT_2:
2397 			tp->t_state = TCPS_TIME_WAIT;
2398 			tcp_canceltimers(tp);
2399 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2400 			soisdisconnected(so);
2401 			break;
2402 
2403 		/*
2404 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2405 		 */
2406 		case TCPS_TIME_WAIT:
2407 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2408 			break;
2409 		}
2410 	}
2411 #ifdef TCP_DEBUG
2412 	if (so->so_options & SO_DEBUG)
2413 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2414 #endif
2415 
2416 	/*
2417 	 * Return any desired output.
2418 	 */
2419 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2420 		(void) tcp_output(tp);
2421 	if (tcp_saveti)
2422 		m_freem(tcp_saveti);
2423 	return;
2424 
2425 badsyn:
2426 	/*
2427 	 * Received a bad SYN.  Increment counters and dropwithreset.
2428 	 */
2429 	tcpstat.tcps_badsyn++;
2430 	tp = NULL;
2431 	goto dropwithreset;
2432 
2433 dropafterack:
2434 	/*
2435 	 * Generate an ACK dropping incoming segment if it occupies
2436 	 * sequence space, where the ACK reflects our state.
2437 	 */
2438 	if (tiflags & TH_RST)
2439 		goto drop;
2440 	m_freem(m);
2441 	tp->t_flags |= TF_ACKNOW;
2442 	(void) tcp_output(tp);
2443 	if (tcp_saveti)
2444 		m_freem(tcp_saveti);
2445 	return;
2446 
2447 dropwithreset_ratelim:
2448 	/*
2449 	 * We may want to rate-limit RSTs in certain situations,
2450 	 * particularly if we are sending an RST in response to
2451 	 * an attempt to connect to or otherwise communicate with
2452 	 * a port for which we have no socket.
2453 	 */
2454 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2455 	    tcp_rst_ppslim) == 0) {
2456 		/* XXX stat */
2457 		goto drop;
2458 	}
2459 	/* ...fall into dropwithreset... */
2460 
2461 dropwithreset:
2462 	/*
2463 	 * Generate a RST, dropping incoming segment.
2464 	 * Make ACK acceptable to originator of segment.
2465 	 */
2466 	if (tiflags & TH_RST)
2467 		goto drop;
2468 
2469 	switch (af) {
2470 #ifdef INET6
2471 	case AF_INET6:
2472 		/* For following calls to tcp_respond */
2473 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2474 			goto drop;
2475 		break;
2476 #endif /* INET6 */
2477 	case AF_INET:
2478 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2479 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2480 			goto drop;
2481 	}
2482 
2483     {
2484 	/*
2485 	 * need to recover version # field, which was overwritten on
2486 	 * ip_cksum computation.
2487 	 */
2488 	struct ip *sip;
2489 	sip = mtod(m, struct ip *);
2490 	switch (af) {
2491 #ifdef INET
2492 	case AF_INET:
2493 		sip->ip_v = 4;
2494 		break;
2495 #endif
2496 #ifdef INET6
2497 	case AF_INET6:
2498 		sip->ip_v = 6;
2499 		break;
2500 #endif
2501 	}
2502     }
2503 	if (tiflags & TH_ACK)
2504 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2505 	else {
2506 		if (tiflags & TH_SYN)
2507 			tlen++;
2508 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2509 		    TH_RST|TH_ACK);
2510 	}
2511 	if (tcp_saveti)
2512 		m_freem(tcp_saveti);
2513 	return;
2514 
2515 badcsum:
2516 	tcpstat.tcps_rcvbadsum++;
2517 drop:
2518 	/*
2519 	 * Drop space held by incoming segment and return.
2520 	 */
2521 	if (tp) {
2522 		if (tp->t_inpcb)
2523 			so = tp->t_inpcb->inp_socket;
2524 #ifdef INET6
2525 		else if (tp->t_in6pcb)
2526 			so = tp->t_in6pcb->in6p_socket;
2527 #endif
2528 		else
2529 			so = NULL;
2530 #ifdef TCP_DEBUG
2531 		if (so && (so->so_options & SO_DEBUG) != 0)
2532 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2533 #endif
2534 	}
2535 	if (tcp_saveti)
2536 		m_freem(tcp_saveti);
2537 	m_freem(m);
2538 	return;
2539 }
2540 
2541 void
2542 tcp_dooptions(tp, cp, cnt, th, oi)
2543 	struct tcpcb *tp;
2544 	u_char *cp;
2545 	int cnt;
2546 	struct tcphdr *th;
2547 	struct tcp_opt_info *oi;
2548 {
2549 	u_int16_t mss;
2550 	int opt, optlen;
2551 
2552 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2553 		opt = cp[0];
2554 		if (opt == TCPOPT_EOL)
2555 			break;
2556 		if (opt == TCPOPT_NOP)
2557 			optlen = 1;
2558 		else {
2559 			if (cnt < 2)
2560 				break;
2561 			optlen = cp[1];
2562 			if (optlen < 2 || optlen > cnt)
2563 				break;
2564 		}
2565 		switch (opt) {
2566 
2567 		default:
2568 			continue;
2569 
2570 		case TCPOPT_MAXSEG:
2571 			if (optlen != TCPOLEN_MAXSEG)
2572 				continue;
2573 			if (!(th->th_flags & TH_SYN))
2574 				continue;
2575 			bcopy(cp + 2, &mss, sizeof(mss));
2576 			oi->maxseg = ntohs(mss);
2577 			break;
2578 
2579 		case TCPOPT_WINDOW:
2580 			if (optlen != TCPOLEN_WINDOW)
2581 				continue;
2582 			if (!(th->th_flags & TH_SYN))
2583 				continue;
2584 			tp->t_flags |= TF_RCVD_SCALE;
2585 			tp->requested_s_scale = cp[2];
2586 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2587 #if 0	/*XXX*/
2588 				char *p;
2589 
2590 				if (ip)
2591 					p = ntohl(ip->ip_src);
2592 #ifdef INET6
2593 				else if (ip6)
2594 					p = ip6_sprintf(&ip6->ip6_src);
2595 #endif
2596 				else
2597 					p = "(unknown)";
2598 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2599 				    "assuming %d\n",
2600 				    tp->requested_s_scale, p,
2601 				    TCP_MAX_WINSHIFT);
2602 #else
2603 				log(LOG_ERR, "TCP: invalid wscale %d, "
2604 				    "assuming %d\n",
2605 				    tp->requested_s_scale,
2606 				    TCP_MAX_WINSHIFT);
2607 #endif
2608 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2609 			}
2610 			break;
2611 
2612 		case TCPOPT_TIMESTAMP:
2613 			if (optlen != TCPOLEN_TIMESTAMP)
2614 				continue;
2615 			oi->ts_present = 1;
2616 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2617 			NTOHL(oi->ts_val);
2618 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2619 			NTOHL(oi->ts_ecr);
2620 
2621 			/*
2622 			 * A timestamp received in a SYN makes
2623 			 * it ok to send timestamp requests and replies.
2624 			 */
2625 			if (th->th_flags & TH_SYN) {
2626 				tp->t_flags |= TF_RCVD_TSTMP;
2627 				tp->ts_recent = oi->ts_val;
2628 				tp->ts_recent_age = TCP_TIMESTAMP(tp);
2629 			}
2630 			break;
2631 		case TCPOPT_SACK_PERMITTED:
2632 			if (optlen != TCPOLEN_SACK_PERMITTED)
2633 				continue;
2634 			if (!(th->th_flags & TH_SYN))
2635 				continue;
2636 			tp->t_flags &= ~TF_CANT_TXSACK;
2637 			break;
2638 
2639 		case TCPOPT_SACK:
2640 			if (tp->t_flags & TF_IGNR_RXSACK)
2641 				continue;
2642 			if (optlen % 8 != 2 || optlen < 10)
2643 				continue;
2644 			cp += 2;
2645 			optlen -= 2;
2646 			for (; optlen > 0; cp -= 8, optlen -= 8) {
2647 				tcp_seq lwe, rwe;
2648 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
2649 				NTOHL(lwe);
2650 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
2651 				NTOHL(rwe);
2652 				/* tcp_mark_sacked(tp, lwe, rwe); */
2653 			}
2654 			break;
2655 		}
2656 	}
2657 }
2658 
2659 /*
2660  * Pull out of band byte out of a segment so
2661  * it doesn't appear in the user's data queue.
2662  * It is still reflected in the segment length for
2663  * sequencing purposes.
2664  */
2665 void
2666 tcp_pulloutofband(so, th, m, off)
2667 	struct socket *so;
2668 	struct tcphdr *th;
2669 	struct mbuf *m;
2670 	int off;
2671 {
2672 	int cnt = off + th->th_urp - 1;
2673 
2674 	while (cnt >= 0) {
2675 		if (m->m_len > cnt) {
2676 			char *cp = mtod(m, caddr_t) + cnt;
2677 			struct tcpcb *tp = sototcpcb(so);
2678 
2679 			tp->t_iobc = *cp;
2680 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2681 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2682 			m->m_len--;
2683 			return;
2684 		}
2685 		cnt -= m->m_len;
2686 		m = m->m_next;
2687 		if (m == 0)
2688 			break;
2689 	}
2690 	panic("tcp_pulloutofband");
2691 }
2692 
2693 /*
2694  * Collect new round-trip time estimate
2695  * and update averages and current timeout.
2696  */
2697 void
2698 tcp_xmit_timer(tp, rtt)
2699 	struct tcpcb *tp;
2700 	uint32_t rtt;
2701 {
2702 	int32_t delta;
2703 
2704 	tcpstat.tcps_rttupdated++;
2705 	if (tp->t_srtt != 0) {
2706 		/*
2707 		 * srtt is stored as fixed point with 3 bits after the
2708 		 * binary point (i.e., scaled by 8).  The following magic
2709 		 * is equivalent to the smoothing algorithm in rfc793 with
2710 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2711 		 * point).  Adjust rtt to origin 0.
2712 		 */
2713 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2714 		if ((tp->t_srtt += delta) <= 0)
2715 			tp->t_srtt = 1 << 2;
2716 		/*
2717 		 * We accumulate a smoothed rtt variance (actually, a
2718 		 * smoothed mean difference), then set the retransmit
2719 		 * timer to smoothed rtt + 4 times the smoothed variance.
2720 		 * rttvar is stored as fixed point with 2 bits after the
2721 		 * binary point (scaled by 4).  The following is
2722 		 * equivalent to rfc793 smoothing with an alpha of .75
2723 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2724 		 * rfc793's wired-in beta.
2725 		 */
2726 		if (delta < 0)
2727 			delta = -delta;
2728 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2729 		if ((tp->t_rttvar += delta) <= 0)
2730 			tp->t_rttvar = 1 << 2;
2731 	} else {
2732 		/*
2733 		 * No rtt measurement yet - use the unsmoothed rtt.
2734 		 * Set the variance to half the rtt (so our first
2735 		 * retransmit happens at 3*rtt).
2736 		 */
2737 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2738 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2739 	}
2740 	tp->t_rtttime = 0;
2741 	tp->t_rxtshift = 0;
2742 
2743 	/*
2744 	 * the retransmit should happen at rtt + 4 * rttvar.
2745 	 * Because of the way we do the smoothing, srtt and rttvar
2746 	 * will each average +1/2 tick of bias.  When we compute
2747 	 * the retransmit timer, we want 1/2 tick of rounding and
2748 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2749 	 * firing of the timer.  The bias will give us exactly the
2750 	 * 1.5 tick we need.  But, because the bias is
2751 	 * statistical, we have to test that we don't drop below
2752 	 * the minimum feasible timer (which is 2 ticks).
2753 	 */
2754 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2755 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2756 
2757 	/*
2758 	 * We received an ack for a packet that wasn't retransmitted;
2759 	 * it is probably safe to discard any error indications we've
2760 	 * received recently.  This isn't quite right, but close enough
2761 	 * for now (a route might have failed after we sent a segment,
2762 	 * and the return path might not be symmetrical).
2763 	 */
2764 	tp->t_softerror = 0;
2765 }
2766 
2767 /*
2768  * Checks for partial ack.  If partial ack arrives, force the retransmission
2769  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
2770  * 1.  By setting snd_nxt to th_ack, this forces retransmission timer to
2771  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
2772  */
2773 int
2774 tcp_newreno(tp, th)
2775 	struct tcpcb *tp;
2776 	struct tcphdr *th;
2777 {
2778 	tcp_seq onxt = tp->snd_nxt;
2779 	u_long ocwnd = tp->snd_cwnd;
2780 
2781 	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2782 		/*
2783 		 * snd_una has not yet been updated and the socket's send
2784 		 * buffer has not yet drained off the ACK'd data, so we
2785 		 * have to leave snd_una as it was to get the correct data
2786 		 * offset in tcp_output().
2787 		 */
2788 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2789 	        tp->t_rtttime = 0;
2790 	        tp->snd_nxt = th->th_ack;
2791 		/*
2792 		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
2793 		 * is not yet updated when we're called.
2794 		 */
2795 		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
2796 	        (void) tcp_output(tp);
2797 	        tp->snd_cwnd = ocwnd;
2798 	        if (SEQ_GT(onxt, tp->snd_nxt))
2799 	                tp->snd_nxt = onxt;
2800 	        /*
2801 	         * Partial window deflation.  Relies on fact that tp->snd_una
2802 	         * not updated yet.
2803 	         */
2804 	        tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
2805 	        return 1;
2806 	}
2807 	return 0;
2808 }
2809 
2810 
2811 /*
2812  * TCP compressed state engine.  Currently used to hold compressed
2813  * state for SYN_RECEIVED.
2814  */
2815 
2816 u_long	syn_cache_count;
2817 u_int32_t syn_hash1, syn_hash2;
2818 
2819 #define SYN_HASH(sa, sp, dp) \
2820 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
2821 				     ((u_int32_t)(sp)))^syn_hash2)))
2822 #ifndef INET6
2823 #define	SYN_HASHALL(hash, src, dst) \
2824 do {									\
2825 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
2826 		((struct sockaddr_in *)(src))->sin_port,		\
2827 		((struct sockaddr_in *)(dst))->sin_port);		\
2828 } while (0)
2829 #else
2830 #define SYN_HASH6(sa, sp, dp) \
2831 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
2832 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
2833 	 & 0x7fffffff)
2834 
2835 #define SYN_HASHALL(hash, src, dst) \
2836 do {									\
2837 	switch ((src)->sa_family) {					\
2838 	case AF_INET:							\
2839 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
2840 			((struct sockaddr_in *)(src))->sin_port,	\
2841 			((struct sockaddr_in *)(dst))->sin_port);	\
2842 		break;							\
2843 	case AF_INET6:							\
2844 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
2845 			((struct sockaddr_in6 *)(src))->sin6_port,	\
2846 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
2847 		break;							\
2848 	default:							\
2849 		hash = 0;						\
2850 	}								\
2851 } while (/*CONSTCOND*/0)
2852 #endif /* INET6 */
2853 
2854 #define	SYN_CACHE_RM(sc)						\
2855 do {									\
2856 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
2857 	    (sc), sc_bucketq);						\
2858 	(sc)->sc_tp = NULL;						\
2859 	LIST_REMOVE((sc), sc_tpq);					\
2860 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
2861 	callout_stop(&(sc)->sc_timer);					\
2862 	syn_cache_count--;						\
2863 } while (/*CONSTCOND*/0)
2864 
2865 #define	SYN_CACHE_PUT(sc)						\
2866 do {									\
2867 	if ((sc)->sc_ipopts)						\
2868 		(void) m_free((sc)->sc_ipopts);				\
2869 	if ((sc)->sc_route4.ro_rt != NULL)				\
2870 		RTFREE((sc)->sc_route4.ro_rt);				\
2871 	pool_put(&syn_cache_pool, (sc));				\
2872 } while (/*CONSTCOND*/0)
2873 
2874 struct pool syn_cache_pool;
2875 
2876 /*
2877  * We don't estimate RTT with SYNs, so each packet starts with the default
2878  * RTT and each timer step has a fixed timeout value.
2879  */
2880 #define	SYN_CACHE_TIMER_ARM(sc)						\
2881 do {									\
2882 	TCPT_RANGESET((sc)->sc_rxtcur,					\
2883 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
2884 	    TCPTV_REXMTMAX);						\
2885 	callout_reset(&(sc)->sc_timer,					\
2886 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
2887 } while (/*CONSTCOND*/0)
2888 
2889 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
2890 
2891 void
2892 syn_cache_init()
2893 {
2894 	int i;
2895 
2896 	/* Initialize the hash buckets. */
2897 	for (i = 0; i < tcp_syn_cache_size; i++)
2898 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
2899 
2900 	/* Initialize the syn cache pool. */
2901 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
2902 	    "synpl", NULL);
2903 }
2904 
2905 void
2906 syn_cache_insert(sc, tp)
2907 	struct syn_cache *sc;
2908 	struct tcpcb *tp;
2909 {
2910 	struct syn_cache_head *scp;
2911 	struct syn_cache *sc2;
2912 	int s;
2913 
2914 	/*
2915 	 * If there are no entries in the hash table, reinitialize
2916 	 * the hash secrets.
2917 	 */
2918 	if (syn_cache_count == 0) {
2919 		struct timeval tv;
2920 		microtime(&tv);
2921 		syn_hash1 = arc4random() ^ (u_long)&sc;
2922 		syn_hash2 = arc4random() ^ tv.tv_usec;
2923 	}
2924 
2925 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
2926 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
2927 	scp = &tcp_syn_cache[sc->sc_bucketidx];
2928 
2929 	/*
2930 	 * Make sure that we don't overflow the per-bucket
2931 	 * limit or the total cache size limit.
2932 	 */
2933 	s = splsoftnet();
2934 	if (scp->sch_length >= tcp_syn_bucket_limit) {
2935 		tcpstat.tcps_sc_bucketoverflow++;
2936 		/*
2937 		 * The bucket is full.  Toss the oldest element in the
2938 		 * bucket.  This will be the first entry in the bucket.
2939 		 */
2940 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
2941 #ifdef DIAGNOSTIC
2942 		/*
2943 		 * This should never happen; we should always find an
2944 		 * entry in our bucket.
2945 		 */
2946 		if (sc2 == NULL)
2947 			panic("syn_cache_insert: bucketoverflow: impossible");
2948 #endif
2949 		SYN_CACHE_RM(sc2);
2950 		SYN_CACHE_PUT(sc2);
2951 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
2952 		struct syn_cache_head *scp2, *sce;
2953 
2954 		tcpstat.tcps_sc_overflowed++;
2955 		/*
2956 		 * The cache is full.  Toss the oldest entry in the
2957 		 * first non-empty bucket we can find.
2958 		 *
2959 		 * XXX We would really like to toss the oldest
2960 		 * entry in the cache, but we hope that this
2961 		 * condition doesn't happen very often.
2962 		 */
2963 		scp2 = scp;
2964 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
2965 			sce = &tcp_syn_cache[tcp_syn_cache_size];
2966 			for (++scp2; scp2 != scp; scp2++) {
2967 				if (scp2 >= sce)
2968 					scp2 = &tcp_syn_cache[0];
2969 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
2970 					break;
2971 			}
2972 #ifdef DIAGNOSTIC
2973 			/*
2974 			 * This should never happen; we should always find a
2975 			 * non-empty bucket.
2976 			 */
2977 			if (scp2 == scp)
2978 				panic("syn_cache_insert: cacheoverflow: "
2979 				    "impossible");
2980 #endif
2981 		}
2982 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
2983 		SYN_CACHE_RM(sc2);
2984 		SYN_CACHE_PUT(sc2);
2985 	}
2986 
2987 	/*
2988 	 * Initialize the entry's timer.
2989 	 */
2990 	sc->sc_rxttot = 0;
2991 	sc->sc_rxtshift = 0;
2992 	SYN_CACHE_TIMER_ARM(sc);
2993 
2994 	/* Link it from tcpcb entry */
2995 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
2996 
2997 	/* Put it into the bucket. */
2998 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
2999 	scp->sch_length++;
3000 	syn_cache_count++;
3001 
3002 	tcpstat.tcps_sc_added++;
3003 	splx(s);
3004 }
3005 
3006 /*
3007  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3008  * If we have retransmitted an entry the maximum number of times, expire
3009  * that entry.
3010  */
3011 void
3012 syn_cache_timer(void *arg)
3013 {
3014 	struct syn_cache *sc = arg;
3015 	int s;
3016 
3017 	s = splsoftnet();
3018 
3019 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3020 		/* Drop it -- too many retransmissions. */
3021 		goto dropit;
3022 	}
3023 
3024 	/*
3025 	 * Compute the total amount of time this entry has
3026 	 * been on a queue.  If this entry has been on longer
3027 	 * than the keep alive timer would allow, expire it.
3028 	 */
3029 	sc->sc_rxttot += sc->sc_rxtcur;
3030 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
3031 		goto dropit;
3032 
3033 	tcpstat.tcps_sc_retransmitted++;
3034 	(void) syn_cache_respond(sc, NULL);
3035 
3036 	/* Advance the timer back-off. */
3037 	sc->sc_rxtshift++;
3038 	SYN_CACHE_TIMER_ARM(sc);
3039 
3040 	splx(s);
3041 	return;
3042 
3043  dropit:
3044 	tcpstat.tcps_sc_timed_out++;
3045 	SYN_CACHE_RM(sc);
3046 	SYN_CACHE_PUT(sc);
3047 	splx(s);
3048 }
3049 
3050 /*
3051  * Remove syn cache created by the specified tcb entry,
3052  * because this does not make sense to keep them
3053  * (if there's no tcb entry, syn cache entry will never be used)
3054  */
3055 void
3056 syn_cache_cleanup(tp)
3057 	struct tcpcb *tp;
3058 {
3059 	struct syn_cache *sc, *nsc;
3060 	int s;
3061 
3062 	s = splsoftnet();
3063 
3064 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3065 		nsc = LIST_NEXT(sc, sc_tpq);
3066 
3067 #ifdef DIAGNOSTIC
3068 		if (sc->sc_tp != tp)
3069 			panic("invalid sc_tp in syn_cache_cleanup");
3070 #endif
3071 		SYN_CACHE_RM(sc);
3072 		SYN_CACHE_PUT(sc);
3073 	}
3074 	/* just for safety */
3075 	LIST_INIT(&tp->t_sc);
3076 
3077 	splx(s);
3078 }
3079 
3080 /*
3081  * Find an entry in the syn cache.
3082  */
3083 struct syn_cache *
3084 syn_cache_lookup(src, dst, headp)
3085 	struct sockaddr *src;
3086 	struct sockaddr *dst;
3087 	struct syn_cache_head **headp;
3088 {
3089 	struct syn_cache *sc;
3090 	struct syn_cache_head *scp;
3091 	u_int32_t hash;
3092 	int s;
3093 
3094 	SYN_HASHALL(hash, src, dst);
3095 
3096 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3097 	*headp = scp;
3098 	s = splsoftnet();
3099 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3100 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3101 		if (sc->sc_hash != hash)
3102 			continue;
3103 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3104 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3105 			splx(s);
3106 			return (sc);
3107 		}
3108 	}
3109 	splx(s);
3110 	return (NULL);
3111 }
3112 
3113 /*
3114  * This function gets called when we receive an ACK for a
3115  * socket in the LISTEN state.  We look up the connection
3116  * in the syn cache, and if its there, we pull it out of
3117  * the cache and turn it into a full-blown connection in
3118  * the SYN-RECEIVED state.
3119  *
3120  * The return values may not be immediately obvious, and their effects
3121  * can be subtle, so here they are:
3122  *
3123  *	NULL	SYN was not found in cache; caller should drop the
3124  *		packet and send an RST.
3125  *
3126  *	-1	We were unable to create the new connection, and are
3127  *		aborting it.  An ACK,RST is being sent to the peer
3128  *		(unless we got screwey sequence numbners; see below),
3129  *		because the 3-way handshake has been completed.  Caller
3130  *		should not free the mbuf, since we may be using it.  If
3131  *		we are not, we will free it.
3132  *
3133  *	Otherwise, the return value is a pointer to the new socket
3134  *	associated with the connection.
3135  */
3136 struct socket *
3137 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3138 	struct sockaddr *src;
3139 	struct sockaddr *dst;
3140 	struct tcphdr *th;
3141 	unsigned int hlen, tlen;
3142 	struct socket *so;
3143 	struct mbuf *m;
3144 {
3145 	struct syn_cache *sc;
3146 	struct syn_cache_head *scp;
3147 	struct inpcb *inp = NULL;
3148 #ifdef INET6
3149 	struct in6pcb *in6p = NULL;
3150 #endif
3151 	struct tcpcb *tp = 0;
3152 	struct mbuf *am;
3153 	int s;
3154 	struct socket *oso;
3155 
3156 	s = splsoftnet();
3157 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3158 		splx(s);
3159 		return (NULL);
3160 	}
3161 
3162 	/*
3163 	 * Verify the sequence and ack numbers.  Try getting the correct
3164 	 * response again.
3165 	 */
3166 	if ((th->th_ack != sc->sc_iss + 1) ||
3167 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3168 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3169 		(void) syn_cache_respond(sc, m);
3170 		splx(s);
3171 		return ((struct socket *)(-1));
3172 	}
3173 
3174 	/* Remove this cache entry */
3175 	SYN_CACHE_RM(sc);
3176 	splx(s);
3177 
3178 	/*
3179 	 * Ok, create the full blown connection, and set things up
3180 	 * as they would have been set up if we had created the
3181 	 * connection when the SYN arrived.  If we can't create
3182 	 * the connection, abort it.
3183 	 */
3184 	/*
3185 	 * inp still has the OLD in_pcb stuff, set the
3186 	 * v6-related flags on the new guy, too.   This is
3187 	 * done particularly for the case where an AF_INET6
3188 	 * socket is bound only to a port, and a v4 connection
3189 	 * comes in on that port.
3190 	 * we also copy the flowinfo from the original pcb
3191 	 * to the new one.
3192 	 */
3193     {
3194 	struct inpcb *parentinpcb;
3195 
3196 	parentinpcb = (struct inpcb *)so->so_pcb;
3197 
3198 	oso = so;
3199 	so = sonewconn(so, SS_ISCONNECTED);
3200 	if (so == NULL)
3201 		goto resetandabort;
3202 
3203 	switch (so->so_proto->pr_domain->dom_family) {
3204 #ifdef INET
3205 	case AF_INET:
3206 		inp = sotoinpcb(so);
3207 		break;
3208 #endif
3209 #ifdef INET6
3210 	case AF_INET6:
3211 		in6p = sotoin6pcb(so);
3212 		break;
3213 #endif
3214 	}
3215     }
3216 	switch (src->sa_family) {
3217 #ifdef INET
3218 	case AF_INET:
3219 		if (inp) {
3220 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3221 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3222 			inp->inp_options = ip_srcroute();
3223 			in_pcbstate(inp, INP_BOUND);
3224 			if (inp->inp_options == NULL) {
3225 				inp->inp_options = sc->sc_ipopts;
3226 				sc->sc_ipopts = NULL;
3227 			}
3228 		}
3229 #ifdef INET6
3230 		else if (in6p) {
3231 			/* IPv4 packet to AF_INET6 socket */
3232 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3233 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3234 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3235 				&in6p->in6p_laddr.s6_addr32[3],
3236 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3237 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3238 			in6totcpcb(in6p)->t_family = AF_INET;
3239 		}
3240 #endif
3241 		break;
3242 #endif
3243 #ifdef INET6
3244 	case AF_INET6:
3245 		if (in6p) {
3246 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3247 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3248 #if 0
3249 			in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK;
3250 			/*inp->inp_options = ip6_srcroute();*/ /* soon. */
3251 #endif
3252 		}
3253 		break;
3254 #endif
3255 	}
3256 #ifdef INET6
3257 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3258 		struct in6pcb *oin6p = sotoin6pcb(oso);
3259 		/* inherit socket options from the listening socket */
3260 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3261 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3262 			m_freem(in6p->in6p_options);
3263 			in6p->in6p_options = 0;
3264 		}
3265 		ip6_savecontrol(in6p, &in6p->in6p_options,
3266 			mtod(m, struct ip6_hdr *), m);
3267 	}
3268 #endif
3269 
3270 #ifdef IPSEC
3271 	/*
3272 	 * we make a copy of policy, instead of sharing the policy,
3273 	 * for better behavior in terms of SA lookup and dead SA removal.
3274 	 */
3275 	if (inp) {
3276 		/* copy old policy into new socket's */
3277 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3278 			printf("tcp_input: could not copy policy\n");
3279 	}
3280 #ifdef INET6
3281 	else if (in6p) {
3282 		/* copy old policy into new socket's */
3283 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3284 		    in6p->in6p_sp))
3285 			printf("tcp_input: could not copy policy\n");
3286 	}
3287 #endif
3288 #endif
3289 
3290 	/*
3291 	 * Give the new socket our cached route reference.
3292 	 */
3293 	if (inp)
3294 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3295 #ifdef INET6
3296 	else
3297 		in6p->in6p_route = sc->sc_route6;
3298 #endif
3299 	sc->sc_route4.ro_rt = NULL;
3300 
3301 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3302 	if (am == NULL)
3303 		goto resetandabort;
3304 	am->m_len = src->sa_len;
3305 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3306 	if (inp) {
3307 		if (in_pcbconnect(inp, am)) {
3308 			(void) m_free(am);
3309 			goto resetandabort;
3310 		}
3311 	}
3312 #ifdef INET6
3313 	else if (in6p) {
3314 		if (src->sa_family == AF_INET) {
3315 			/* IPv4 packet to AF_INET6 socket */
3316 			struct sockaddr_in6 *sin6;
3317 			sin6 = mtod(am, struct sockaddr_in6 *);
3318 			am->m_len = sizeof(*sin6);
3319 			bzero(sin6, sizeof(*sin6));
3320 			sin6->sin6_family = AF_INET6;
3321 			sin6->sin6_len = sizeof(*sin6);
3322 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3323 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3324 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3325 				&sin6->sin6_addr.s6_addr32[3],
3326 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3327 		}
3328 		if (in6_pcbconnect(in6p, am)) {
3329 			(void) m_free(am);
3330 			goto resetandabort;
3331 		}
3332 	}
3333 #endif
3334 	else {
3335 		(void) m_free(am);
3336 		goto resetandabort;
3337 	}
3338 	(void) m_free(am);
3339 
3340 	if (inp)
3341 		tp = intotcpcb(inp);
3342 #ifdef INET6
3343 	else if (in6p)
3344 		tp = in6totcpcb(in6p);
3345 #endif
3346 	else
3347 		tp = NULL;
3348 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3349 	if (sc->sc_request_r_scale != 15) {
3350 		tp->requested_s_scale = sc->sc_requested_s_scale;
3351 		tp->request_r_scale = sc->sc_request_r_scale;
3352 		tp->snd_scale = sc->sc_requested_s_scale;
3353 		tp->rcv_scale = sc->sc_request_r_scale;
3354 		tp->t_flags |= TF_RCVD_SCALE;
3355 	}
3356 	if (sc->sc_flags & SCF_TIMESTAMP)
3357 		tp->t_flags |= TF_RCVD_TSTMP;
3358 	tp->ts_timebase = sc->sc_timebase;
3359 
3360 	tp->t_template = tcp_template(tp);
3361 	if (tp->t_template == 0) {
3362 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3363 		so = NULL;
3364 		m_freem(m);
3365 		goto abort;
3366 	}
3367 
3368 	tp->iss = sc->sc_iss;
3369 	tp->irs = sc->sc_irs;
3370 	tcp_sendseqinit(tp);
3371 	tcp_rcvseqinit(tp);
3372 	tp->t_state = TCPS_SYN_RECEIVED;
3373 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3374 	tcpstat.tcps_accepts++;
3375 
3376 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3377 	tp->t_ourmss = sc->sc_ourmaxseg;
3378 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3379 
3380 	/*
3381 	 * Initialize the initial congestion window.  If we
3382 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3383 	 * to 1 segment (i.e. the Loss Window).
3384 	 */
3385 	if (sc->sc_rxtshift)
3386 		tp->snd_cwnd = tp->t_peermss;
3387 	else
3388 		tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss);
3389 
3390 	tcp_rmx_rtt(tp);
3391 	tp->snd_wl1 = sc->sc_irs;
3392 	tp->rcv_up = sc->sc_irs + 1;
3393 
3394 	/*
3395 	 * This is what whould have happened in tcp_ouput() when
3396 	 * the SYN,ACK was sent.
3397 	 */
3398 	tp->snd_up = tp->snd_una;
3399 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3400 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3401 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3402 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3403 	tp->last_ack_sent = tp->rcv_nxt;
3404 
3405 	tcpstat.tcps_sc_completed++;
3406 	SYN_CACHE_PUT(sc);
3407 	return (so);
3408 
3409 resetandabort:
3410 	(void) tcp_respond(NULL, m, m, th,
3411 			   th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
3412 abort:
3413 	if (so != NULL)
3414 		(void) soabort(so);
3415 	SYN_CACHE_PUT(sc);
3416 	tcpstat.tcps_sc_aborted++;
3417 	return ((struct socket *)(-1));
3418 }
3419 
3420 /*
3421  * This function is called when we get a RST for a
3422  * non-existent connection, so that we can see if the
3423  * connection is in the syn cache.  If it is, zap it.
3424  */
3425 
3426 void
3427 syn_cache_reset(src, dst, th)
3428 	struct sockaddr *src;
3429 	struct sockaddr *dst;
3430 	struct tcphdr *th;
3431 {
3432 	struct syn_cache *sc;
3433 	struct syn_cache_head *scp;
3434 	int s = splsoftnet();
3435 
3436 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3437 		splx(s);
3438 		return;
3439 	}
3440 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3441 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3442 		splx(s);
3443 		return;
3444 	}
3445 	SYN_CACHE_RM(sc);
3446 	splx(s);
3447 	tcpstat.tcps_sc_reset++;
3448 	SYN_CACHE_PUT(sc);
3449 }
3450 
3451 void
3452 syn_cache_unreach(src, dst, th)
3453 	struct sockaddr *src;
3454 	struct sockaddr *dst;
3455 	struct tcphdr *th;
3456 {
3457 	struct syn_cache *sc;
3458 	struct syn_cache_head *scp;
3459 	int s;
3460 
3461 	s = splsoftnet();
3462 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3463 		splx(s);
3464 		return;
3465 	}
3466 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3467 	if (ntohl (th->th_seq) != sc->sc_iss) {
3468 		splx(s);
3469 		return;
3470 	}
3471 
3472 	/*
3473 	 * If we've rertransmitted 3 times and this is our second error,
3474 	 * we remove the entry.  Otherwise, we allow it to continue on.
3475 	 * This prevents us from incorrectly nuking an entry during a
3476 	 * spurious network outage.
3477 	 *
3478 	 * See tcp_notify().
3479 	 */
3480 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3481 		sc->sc_flags |= SCF_UNREACH;
3482 		splx(s);
3483 		return;
3484 	}
3485 
3486 	SYN_CACHE_RM(sc);
3487 	splx(s);
3488 	tcpstat.tcps_sc_unreach++;
3489 	SYN_CACHE_PUT(sc);
3490 }
3491 
3492 /*
3493  * Given a LISTEN socket and an inbound SYN request, add
3494  * this to the syn cache, and send back a segment:
3495  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3496  * to the source.
3497  *
3498  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3499  * Doing so would require that we hold onto the data and deliver it
3500  * to the application.  However, if we are the target of a SYN-flood
3501  * DoS attack, an attacker could send data which would eventually
3502  * consume all available buffer space if it were ACKed.  By not ACKing
3503  * the data, we avoid this DoS scenario.
3504  */
3505 
3506 int
3507 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi)
3508 	struct sockaddr *src;
3509 	struct sockaddr *dst;
3510 	struct tcphdr *th;
3511 	unsigned int hlen;
3512 	struct socket *so;
3513 	struct mbuf *m;
3514 	u_char *optp;
3515 	int optlen;
3516 	struct tcp_opt_info *oi;
3517 {
3518 	struct tcpcb tb, *tp;
3519 	long win;
3520 	struct syn_cache *sc;
3521 	struct syn_cache_head *scp;
3522 	struct mbuf *ipopts;
3523 
3524 	tp = sototcpcb(so);
3525 
3526 	/*
3527 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3528 	 *
3529 	 * Note this check is performed in tcp_input() very early on.
3530 	 */
3531 
3532 	/*
3533 	 * Initialize some local state.
3534 	 */
3535 	win = sbspace(&so->so_rcv);
3536 	if (win > TCP_MAXWIN)
3537 		win = TCP_MAXWIN;
3538 
3539 	switch (src->sa_family) {
3540 #ifdef INET
3541 	case AF_INET:
3542 		/*
3543 		 * Remember the IP options, if any.
3544 		 */
3545 		ipopts = ip_srcroute();
3546 		break;
3547 #endif
3548 	default:
3549 		ipopts = NULL;
3550 	}
3551 
3552 	if (optp) {
3553 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3554 		tcp_dooptions(&tb, optp, optlen, th, oi);
3555 	} else
3556 		tb.t_flags = 0;
3557 
3558 	/*
3559 	 * See if we already have an entry for this connection.
3560 	 * If we do, resend the SYN,ACK.  We do not count this
3561 	 * as a retransmission (XXX though maybe we should).
3562 	 */
3563 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3564 		tcpstat.tcps_sc_dupesyn++;
3565 		if (ipopts) {
3566 			/*
3567 			 * If we were remembering a previous source route,
3568 			 * forget it and use the new one we've been given.
3569 			 */
3570 			if (sc->sc_ipopts)
3571 				(void) m_free(sc->sc_ipopts);
3572 			sc->sc_ipopts = ipopts;
3573 		}
3574 		sc->sc_timestamp = tb.ts_recent;
3575 		if (syn_cache_respond(sc, m) == 0) {
3576 			tcpstat.tcps_sndacks++;
3577 			tcpstat.tcps_sndtotal++;
3578 		}
3579 		return (1);
3580 	}
3581 
3582 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3583 	if (sc == NULL) {
3584 		if (ipopts)
3585 			(void) m_free(ipopts);
3586 		return (0);
3587 	}
3588 
3589 	/*
3590 	 * Fill in the cache, and put the necessary IP and TCP
3591 	 * options into the reply.
3592 	 */
3593 	callout_init(&sc->sc_timer);
3594 	bzero(sc, sizeof(struct syn_cache));
3595 	bcopy(src, &sc->sc_src, src->sa_len);
3596 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3597 	sc->sc_flags = 0;
3598 	sc->sc_ipopts = ipopts;
3599 	sc->sc_irs = th->th_seq;
3600 	switch (src->sa_family) {
3601 #ifdef INET
3602 	case AF_INET:
3603 	    {
3604 		struct sockaddr_in *srcin = (void *) src;
3605 		struct sockaddr_in *dstin = (void *) dst;
3606 
3607 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3608 		    &srcin->sin_addr, dstin->sin_port,
3609 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3610 		break;
3611 	    }
3612 #endif /* INET */
3613 #ifdef INET6
3614 	case AF_INET6:
3615 	    {
3616 		struct sockaddr_in6 *srcin6 = (void *) src;
3617 		struct sockaddr_in6 *dstin6 = (void *) dst;
3618 
3619 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3620 		    &srcin6->sin6_addr, dstin6->sin6_port,
3621 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3622 		break;
3623 	    }
3624 #endif /* INET6 */
3625 	}
3626 	sc->sc_peermaxseg = oi->maxseg;
3627 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
3628 						m->m_pkthdr.rcvif : NULL,
3629 						sc->sc_src.sa.sa_family);
3630 	sc->sc_win = win;
3631 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
3632 	sc->sc_timestamp = tb.ts_recent;
3633 	if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP))
3634 		sc->sc_flags |= SCF_TIMESTAMP;
3635 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
3636 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
3637 		sc->sc_requested_s_scale = tb.requested_s_scale;
3638 		sc->sc_request_r_scale = 0;
3639 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
3640 		    TCP_MAXWIN << sc->sc_request_r_scale <
3641 		    so->so_rcv.sb_hiwat)
3642 			sc->sc_request_r_scale++;
3643 	} else {
3644 		sc->sc_requested_s_scale = 15;
3645 		sc->sc_request_r_scale = 15;
3646 	}
3647 	sc->sc_tp = tp;
3648 	if (syn_cache_respond(sc, m) == 0) {
3649 		syn_cache_insert(sc, tp);
3650 		tcpstat.tcps_sndacks++;
3651 		tcpstat.tcps_sndtotal++;
3652 	} else {
3653 		SYN_CACHE_PUT(sc);
3654 		tcpstat.tcps_sc_dropped++;
3655 	}
3656 	return (1);
3657 }
3658 
3659 int
3660 syn_cache_respond(sc, m)
3661 	struct syn_cache *sc;
3662 	struct mbuf *m;
3663 {
3664 	struct route *ro;
3665 	u_int8_t *optp;
3666 	int optlen, error;
3667 	u_int16_t tlen;
3668 	struct ip *ip = NULL;
3669 #ifdef INET6
3670 	struct ip6_hdr *ip6 = NULL;
3671 #endif
3672 	struct tcphdr *th;
3673 	u_int hlen;
3674 
3675 	switch (sc->sc_src.sa.sa_family) {
3676 	case AF_INET:
3677 		hlen = sizeof(struct ip);
3678 		ro = &sc->sc_route4;
3679 		break;
3680 #ifdef INET6
3681 	case AF_INET6:
3682 		hlen = sizeof(struct ip6_hdr);
3683 		ro = (struct route *)&sc->sc_route6;
3684 		break;
3685 #endif
3686 	default:
3687 		if (m)
3688 			m_freem(m);
3689 		return EAFNOSUPPORT;
3690 	}
3691 
3692 	/* Compute the size of the TCP options. */
3693 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
3694 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
3695 
3696 	tlen = hlen + sizeof(struct tcphdr) + optlen;
3697 
3698 	/*
3699 	 * Create the IP+TCP header from scratch.
3700 	 */
3701 	if (m)
3702 		m_freem(m);
3703 #ifdef DIAGNOSTIC
3704 	if (max_linkhdr + tlen > MCLBYTES)
3705 		return (ENOBUFS);
3706 #endif
3707 	MGETHDR(m, M_DONTWAIT, MT_DATA);
3708 	if (m && tlen > MHLEN) {
3709 		MCLGET(m, M_DONTWAIT);
3710 		if ((m->m_flags & M_EXT) == 0) {
3711 			m_freem(m);
3712 			m = NULL;
3713 		}
3714 	}
3715 	if (m == NULL)
3716 		return (ENOBUFS);
3717 
3718 	/* Fixup the mbuf. */
3719 	m->m_data += max_linkhdr;
3720 	m->m_len = m->m_pkthdr.len = tlen;
3721 #ifdef IPSEC
3722 	if (sc->sc_tp) {
3723 		struct tcpcb *tp;
3724 		struct socket *so;
3725 
3726 		tp = sc->sc_tp;
3727 		if (tp->t_inpcb)
3728 			so = tp->t_inpcb->inp_socket;
3729 #ifdef INET6
3730 		else if (tp->t_in6pcb)
3731 			so = tp->t_in6pcb->in6p_socket;
3732 #endif
3733 		else
3734 			so = NULL;
3735 		/* use IPsec policy on listening socket, on SYN ACK */
3736 		if (ipsec_setsocket(m, so) != 0) {
3737 			m_freem(m);
3738 			return ENOBUFS;
3739 		}
3740 	}
3741 #endif
3742 	m->m_pkthdr.rcvif = NULL;
3743 	memset(mtod(m, u_char *), 0, tlen);
3744 
3745 	switch (sc->sc_src.sa.sa_family) {
3746 	case AF_INET:
3747 		ip = mtod(m, struct ip *);
3748 		ip->ip_dst = sc->sc_src.sin.sin_addr;
3749 		ip->ip_src = sc->sc_dst.sin.sin_addr;
3750 		ip->ip_p = IPPROTO_TCP;
3751 		th = (struct tcphdr *)(ip + 1);
3752 		th->th_dport = sc->sc_src.sin.sin_port;
3753 		th->th_sport = sc->sc_dst.sin.sin_port;
3754 		break;
3755 #ifdef INET6
3756 	case AF_INET6:
3757 		ip6 = mtod(m, struct ip6_hdr *);
3758 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
3759 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
3760 		ip6->ip6_nxt = IPPROTO_TCP;
3761 		/* ip6_plen will be updated in ip6_output() */
3762 		th = (struct tcphdr *)(ip6 + 1);
3763 		th->th_dport = sc->sc_src.sin6.sin6_port;
3764 		th->th_sport = sc->sc_dst.sin6.sin6_port;
3765 		break;
3766 #endif
3767 	default:
3768 		th = NULL;
3769 	}
3770 
3771 	th->th_seq = htonl(sc->sc_iss);
3772 	th->th_ack = htonl(sc->sc_irs + 1);
3773 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
3774 	th->th_flags = TH_SYN|TH_ACK;
3775 	th->th_win = htons(sc->sc_win);
3776 	/* th_sum already 0 */
3777 	/* th_urp already 0 */
3778 
3779 	/* Tack on the TCP options. */
3780 	optp = (u_int8_t *)(th + 1);
3781 	*optp++ = TCPOPT_MAXSEG;
3782 	*optp++ = 4;
3783 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
3784 	*optp++ = sc->sc_ourmaxseg & 0xff;
3785 
3786 	if (sc->sc_request_r_scale != 15) {
3787 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
3788 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
3789 		    sc->sc_request_r_scale);
3790 		optp += 4;
3791 	}
3792 
3793 	if (sc->sc_flags & SCF_TIMESTAMP) {
3794 		u_int32_t *lp = (u_int32_t *)(optp);
3795 		/* Form timestamp option as shown in appendix A of RFC 1323. */
3796 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
3797 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
3798 		*lp   = htonl(sc->sc_timestamp);
3799 		optp += TCPOLEN_TSTAMP_APPA;
3800 	}
3801 
3802 	/* Compute the packet's checksum. */
3803 	switch (sc->sc_src.sa.sa_family) {
3804 	case AF_INET:
3805 		ip->ip_len = htons(tlen - hlen);
3806 		th->th_sum = 0;
3807 		th->th_sum = in_cksum(m, tlen);
3808 		break;
3809 #ifdef INET6
3810 	case AF_INET6:
3811 		ip6->ip6_plen = htons(tlen - hlen);
3812 		th->th_sum = 0;
3813 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
3814 		break;
3815 #endif
3816 	}
3817 
3818 	/*
3819 	 * Fill in some straggling IP bits.  Note the stack expects
3820 	 * ip_len to be in host order, for convenience.
3821 	 */
3822 	switch (sc->sc_src.sa.sa_family) {
3823 #ifdef INET
3824 	case AF_INET:
3825 		ip->ip_len = htons(tlen);
3826 		ip->ip_ttl = ip_defttl;
3827 		/* XXX tos? */
3828 		break;
3829 #endif
3830 #ifdef INET6
3831 	case AF_INET6:
3832 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3833 		ip6->ip6_vfc |= IPV6_VERSION;
3834 		ip6->ip6_plen = htons(tlen - hlen);
3835 		/* ip6_hlim will be initialized afterwards */
3836 		/* XXX flowlabel? */
3837 		break;
3838 #endif
3839 	}
3840 
3841 	switch (sc->sc_src.sa.sa_family) {
3842 #ifdef INET
3843 	case AF_INET:
3844 		error = ip_output(m, sc->sc_ipopts, ro,
3845 		    (ip_mtudisc ? IP_MTUDISC : 0),
3846 		    NULL);
3847 		break;
3848 #endif
3849 #ifdef INET6
3850 	case AF_INET6:
3851 		ip6->ip6_hlim = in6_selecthlim(NULL,
3852 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
3853 
3854 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro,
3855 			0, NULL, NULL);
3856 		break;
3857 #endif
3858 	default:
3859 		error = EAFNOSUPPORT;
3860 		break;
3861 	}
3862 	return (error);
3863 }
3864