xref: /netbsd-src/sys/netinet/tcp_input.c (revision 73704c4ce4ee2a60eb617e693ce7e9f03902613e)
1 /*	$NetBSD: tcp_input.c,v 1.184 2003/09/10 01:46:27 itojun Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  *      This product includes software developed by the University of
47  *      California, Berkeley and its contributors.
48  *      This product includes software developed at the Information
49  *      Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 /*-
73  * Copyright (c) 1997, 1998, 1999, 2001 The NetBSD Foundation, Inc.
74  * All rights reserved.
75  *
76  * This code is derived from software contributed to The NetBSD Foundation
77  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
78  * Facility, NASA Ames Research Center.
79  *
80  * Redistribution and use in source and binary forms, with or without
81  * modification, are permitted provided that the following conditions
82  * are met:
83  * 1. Redistributions of source code must retain the above copyright
84  *    notice, this list of conditions and the following disclaimer.
85  * 2. Redistributions in binary form must reproduce the above copyright
86  *    notice, this list of conditions and the following disclaimer in the
87  *    documentation and/or other materials provided with the distribution.
88  * 3. All advertising materials mentioning features or use of this software
89  *    must display the following acknowledgement:
90  *	This product includes software developed by the NetBSD
91  *	Foundation, Inc. and its contributors.
92  * 4. Neither the name of The NetBSD Foundation nor the names of its
93  *    contributors may be used to endorse or promote products derived
94  *    from this software without specific prior written permission.
95  *
96  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
97  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
98  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
99  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
100  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
101  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
102  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
103  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
104  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
105  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
106  * POSSIBILITY OF SUCH DAMAGE.
107  */
108 
109 /*
110  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
111  *	The Regents of the University of California.  All rights reserved.
112  *
113  * Redistribution and use in source and binary forms, with or without
114  * modification, are permitted provided that the following conditions
115  * are met:
116  * 1. Redistributions of source code must retain the above copyright
117  *    notice, this list of conditions and the following disclaimer.
118  * 2. Redistributions in binary form must reproduce the above copyright
119  *    notice, this list of conditions and the following disclaimer in the
120  *    documentation and/or other materials provided with the distribution.
121  * 3. Neither the name of the University nor the names of its contributors
122  *    may be used to endorse or promote products derived from this software
123  *    without specific prior written permission.
124  *
125  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
126  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
127  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
128  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
129  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
130  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
131  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
132  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
133  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
134  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
135  * SUCH DAMAGE.
136  *
137  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
138  */
139 
140 /*
141  *	TODO list for SYN cache stuff:
142  *
143  *	Find room for a "state" field, which is needed to keep a
144  *	compressed state for TIME_WAIT TCBs.  It's been noted already
145  *	that this is fairly important for very high-volume web and
146  *	mail servers, which use a large number of short-lived
147  *	connections.
148  */
149 
150 #include <sys/cdefs.h>
151 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.184 2003/09/10 01:46:27 itojun Exp $");
152 
153 #include "opt_inet.h"
154 #include "opt_ipsec.h"
155 #include "opt_inet_csum.h"
156 #include "opt_tcp_debug.h"
157 
158 #include <sys/param.h>
159 #include <sys/systm.h>
160 #include <sys/malloc.h>
161 #include <sys/mbuf.h>
162 #include <sys/protosw.h>
163 #include <sys/socket.h>
164 #include <sys/socketvar.h>
165 #include <sys/errno.h>
166 #include <sys/syslog.h>
167 #include <sys/pool.h>
168 #include <sys/domain.h>
169 #include <sys/kernel.h>
170 
171 #include <net/if.h>
172 #include <net/route.h>
173 #include <net/if_types.h>
174 
175 #include <netinet/in.h>
176 #include <netinet/in_systm.h>
177 #include <netinet/ip.h>
178 #include <netinet/in_pcb.h>
179 #include <netinet/in_var.h>
180 #include <netinet/ip_var.h>
181 
182 #ifdef INET6
183 #ifndef INET
184 #include <netinet/in.h>
185 #endif
186 #include <netinet/ip6.h>
187 #include <netinet6/ip6_var.h>
188 #include <netinet6/in6_pcb.h>
189 #include <netinet6/ip6_var.h>
190 #include <netinet6/in6_var.h>
191 #include <netinet/icmp6.h>
192 #include <netinet6/nd6.h>
193 #endif
194 
195 #ifndef INET6
196 /* always need ip6.h for IP6_EXTHDR_GET */
197 #include <netinet/ip6.h>
198 #endif
199 
200 #include <netinet/tcp.h>
201 #include <netinet/tcp_fsm.h>
202 #include <netinet/tcp_seq.h>
203 #include <netinet/tcp_timer.h>
204 #include <netinet/tcp_var.h>
205 #include <netinet/tcpip.h>
206 #include <netinet/tcp_debug.h>
207 
208 #include <machine/stdarg.h>
209 
210 #ifdef IPSEC
211 #include <netinet6/ipsec.h>
212 #include <netkey/key.h>
213 #endif /*IPSEC*/
214 #ifdef INET6
215 #include "faith.h"
216 #if defined(NFAITH) && NFAITH > 0
217 #include <net/if_faith.h>
218 #endif
219 #endif	/* IPSEC */
220 
221 #ifdef FAST_IPSEC
222 #include <netipsec/ipsec.h>
223 #include <netipsec/key.h>
224 #endif	/* FAST_IPSEC*/
225 
226 
227 int	tcprexmtthresh = 3;
228 int	tcp_log_refused;
229 
230 static int tcp_rst_ppslim_count = 0;
231 static struct timeval tcp_rst_ppslim_last;
232 
233 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
234 
235 /* for modulo comparisons of timestamps */
236 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
237 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
238 
239 /*
240  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
241  */
242 #ifdef INET6
243 #define ND6_HINT(tp) \
244 do { \
245 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \
246 	 && tp->t_in6pcb->in6p_route.ro_rt) { \
247 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
248 	} \
249 } while (/*CONSTCOND*/ 0)
250 #else
251 #define ND6_HINT(tp)
252 #endif
253 
254 /*
255  * Macro to compute ACK transmission behavior.  Delay the ACK unless
256  * we have already delayed an ACK (must send an ACK every two segments).
257  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
258  * option is enabled.
259  */
260 #define	TCP_SETUP_ACK(tp, th) \
261 do { \
262 	if ((tp)->t_flags & TF_DELACK || \
263 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
264 		tp->t_flags |= TF_ACKNOW; \
265 	else \
266 		TCP_SET_DELACK(tp); \
267 } while (/*CONSTCOND*/ 0)
268 
269 /*
270  * Convert TCP protocol fields to host order for easier processing.
271  */
272 #define	TCP_FIELDS_TO_HOST(th)						\
273 do {									\
274 	NTOHL((th)->th_seq);						\
275 	NTOHL((th)->th_ack);						\
276 	NTOHS((th)->th_win);						\
277 	NTOHS((th)->th_urp);						\
278 } while (/*CONSTCOND*/ 0)
279 
280 /*
281  * ... and reverse the above.
282  */
283 #define	TCP_FIELDS_TO_NET(th)						\
284 do {									\
285 	HTONL((th)->th_seq);						\
286 	HTONL((th)->th_ack);						\
287 	HTONS((th)->th_win);						\
288 	HTONS((th)->th_urp);						\
289 } while (/*CONSTCOND*/ 0)
290 
291 #ifdef TCP_CSUM_COUNTERS
292 #include <sys/device.h>
293 
294 extern struct evcnt tcp_hwcsum_ok;
295 extern struct evcnt tcp_hwcsum_bad;
296 extern struct evcnt tcp_hwcsum_data;
297 extern struct evcnt tcp_swcsum;
298 
299 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
300 
301 #else
302 
303 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
304 
305 #endif /* TCP_CSUM_COUNTERS */
306 
307 #ifdef TCP_REASS_COUNTERS
308 #include <sys/device.h>
309 
310 extern struct evcnt tcp_reass_;
311 extern struct evcnt tcp_reass_empty;
312 extern struct evcnt tcp_reass_iteration[8];
313 extern struct evcnt tcp_reass_prependfirst;
314 extern struct evcnt tcp_reass_prepend;
315 extern struct evcnt tcp_reass_insert;
316 extern struct evcnt tcp_reass_inserttail;
317 extern struct evcnt tcp_reass_append;
318 extern struct evcnt tcp_reass_appendtail;
319 extern struct evcnt tcp_reass_overlaptail;
320 extern struct evcnt tcp_reass_overlapfront;
321 extern struct evcnt tcp_reass_segdup;
322 extern struct evcnt tcp_reass_fragdup;
323 
324 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
325 
326 #else
327 
328 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
329 
330 #endif /* TCP_REASS_COUNTERS */
331 
332 #ifdef INET
333 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *));
334 #endif
335 #ifdef INET6
336 static void tcp6_log_refused
337     __P((const struct ip6_hdr *, const struct tcphdr *));
338 #endif
339 
340 int
341 tcp_reass(tp, th, m, tlen)
342 	struct tcpcb *tp;
343 	struct tcphdr *th;
344 	struct mbuf *m;
345 	int *tlen;
346 {
347 	struct ipqent *p, *q, *nq, *tiqe = NULL;
348 	struct socket *so = NULL;
349 	int pkt_flags;
350 	tcp_seq pkt_seq;
351 	unsigned pkt_len;
352 	u_long rcvpartdupbyte = 0;
353 	u_long rcvoobyte;
354 #ifdef TCP_REASS_COUNTERS
355 	u_int count = 0;
356 #endif
357 
358 	if (tp->t_inpcb)
359 		so = tp->t_inpcb->inp_socket;
360 #ifdef INET6
361 	else if (tp->t_in6pcb)
362 		so = tp->t_in6pcb->in6p_socket;
363 #endif
364 
365 	TCP_REASS_LOCK_CHECK(tp);
366 
367 	/*
368 	 * Call with th==0 after become established to
369 	 * force pre-ESTABLISHED data up to user socket.
370 	 */
371 	if (th == 0)
372 		goto present;
373 
374 	rcvoobyte = *tlen;
375 	/*
376 	 * Copy these to local variables because the tcpiphdr
377 	 * gets munged while we are collapsing mbufs.
378 	 */
379 	pkt_seq = th->th_seq;
380 	pkt_len = *tlen;
381 	pkt_flags = th->th_flags;
382 
383 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
384 
385 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
386 		/*
387 		 * When we miss a packet, the vast majority of time we get
388 		 * packets that follow it in order.  So optimize for that.
389 		 */
390 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
391 			p->ipqe_len += pkt_len;
392 			p->ipqe_flags |= pkt_flags;
393 			m_cat(p->ipqe_m, m);
394 			tiqe = p;
395 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
396 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
397 			goto skip_replacement;
398 		}
399 		/*
400 		 * While we're here, if the pkt is completely beyond
401 		 * anything we have, just insert it at the tail.
402 		 */
403 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
404 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
405 			goto insert_it;
406 		}
407 	}
408 
409 	q = TAILQ_FIRST(&tp->segq);
410 
411 	if (q != NULL) {
412 		/*
413 		 * If this segment immediately precedes the first out-of-order
414 		 * block, simply slap the segment in front of it and (mostly)
415 		 * skip the complicated logic.
416 		 */
417 		if (pkt_seq + pkt_len == q->ipqe_seq) {
418 			q->ipqe_seq = pkt_seq;
419 			q->ipqe_len += pkt_len;
420 			q->ipqe_flags |= pkt_flags;
421 			m_cat(m, q->ipqe_m);
422 			q->ipqe_m = m;
423 			tiqe = q;
424 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
425 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
426 			goto skip_replacement;
427 		}
428 	} else {
429 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
430 	}
431 
432 	/*
433 	 * Find a segment which begins after this one does.
434 	 */
435 	for (p = NULL; q != NULL; q = nq) {
436 		nq = TAILQ_NEXT(q, ipqe_q);
437 #ifdef TCP_REASS_COUNTERS
438 		count++;
439 #endif
440 		/*
441 		 * If the received segment is just right after this
442 		 * fragment, merge the two together and then check
443 		 * for further overlaps.
444 		 */
445 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
446 #ifdef TCPREASS_DEBUG
447 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
448 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
449 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
450 #endif
451 			pkt_len += q->ipqe_len;
452 			pkt_flags |= q->ipqe_flags;
453 			pkt_seq = q->ipqe_seq;
454 			m_cat(q->ipqe_m, m);
455 			m = q->ipqe_m;
456 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
457 			goto free_ipqe;
458 		}
459 		/*
460 		 * If the received segment is completely past this
461 		 * fragment, we need to go the next fragment.
462 		 */
463 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
464 			p = q;
465 			continue;
466 		}
467 		/*
468 		 * If the fragment is past the received segment,
469 		 * it (or any following) can't be concatenated.
470 		 */
471 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
472 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
473 			break;
474 		}
475 
476 		/*
477 		 * We've received all the data in this segment before.
478 		 * mark it as a duplicate and return.
479 		 */
480 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
481 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
482 			tcpstat.tcps_rcvduppack++;
483 			tcpstat.tcps_rcvdupbyte += pkt_len;
484 			m_freem(m);
485 			if (tiqe != NULL)
486 				pool_put(&ipqent_pool, tiqe);
487 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
488 			return (0);
489 		}
490 		/*
491 		 * Received segment completely overlaps this fragment
492 		 * so we drop the fragment (this keeps the temporal
493 		 * ordering of segments correct).
494 		 */
495 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
496 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
497 			rcvpartdupbyte += q->ipqe_len;
498 			m_freem(q->ipqe_m);
499 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
500 			goto free_ipqe;
501 		}
502 		/*
503 		 * RX'ed segment extends past the end of the
504 		 * fragment.  Drop the overlapping bytes.  Then
505 		 * merge the fragment and segment then treat as
506 		 * a longer received packet.
507 		 */
508 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
509 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
510 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
511 #ifdef TCPREASS_DEBUG
512 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
513 			       tp, overlap,
514 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
515 #endif
516 			m_adj(m, overlap);
517 			rcvpartdupbyte += overlap;
518 			m_cat(q->ipqe_m, m);
519 			m = q->ipqe_m;
520 			pkt_seq = q->ipqe_seq;
521 			pkt_len += q->ipqe_len - overlap;
522 			rcvoobyte -= overlap;
523 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
524 			goto free_ipqe;
525 		}
526 		/*
527 		 * RX'ed segment extends past the front of the
528 		 * fragment.  Drop the overlapping bytes on the
529 		 * received packet.  The packet will then be
530 		 * contatentated with this fragment a bit later.
531 		 */
532 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
533 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
534 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
535 #ifdef TCPREASS_DEBUG
536 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
537 			       tp, overlap,
538 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
539 #endif
540 			m_adj(m, -overlap);
541 			pkt_len -= overlap;
542 			rcvpartdupbyte += overlap;
543 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
544 			rcvoobyte -= overlap;
545 		}
546 		/*
547 		 * If the received segment immediates precedes this
548 		 * fragment then tack the fragment onto this segment
549 		 * and reinsert the data.
550 		 */
551 		if (q->ipqe_seq == pkt_seq + pkt_len) {
552 #ifdef TCPREASS_DEBUG
553 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
554 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
555 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
556 #endif
557 			pkt_len += q->ipqe_len;
558 			pkt_flags |= q->ipqe_flags;
559 			m_cat(m, q->ipqe_m);
560 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
561 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
562 			if (tiqe == NULL) {
563 			    tiqe = q;
564 			} else {
565 			    pool_put(&ipqent_pool, q);
566 			}
567 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
568 			break;
569 		}
570 		/*
571 		 * If the fragment is before the segment, remember it.
572 		 * When this loop is terminated, p will contain the
573 		 * pointer to fragment that is right before the received
574 		 * segment.
575 		 */
576 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
577 			p = q;
578 
579 		continue;
580 
581 		/*
582 		 * This is a common operation.  It also will allow
583 		 * to save doing a malloc/free in most instances.
584 		 */
585 	  free_ipqe:
586 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
587 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
588 		if (tiqe == NULL) {
589 		    tiqe = q;
590 		} else {
591 		    pool_put(&ipqent_pool, q);
592 		}
593 	}
594 
595 #ifdef TCP_REASS_COUNTERS
596 	if (count > 7)
597 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
598 	else if (count > 0)
599 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
600 #endif
601 
602     insert_it:
603 
604 	/*
605 	 * Allocate a new queue entry since the received segment did not
606 	 * collapse onto any other out-of-order block; thus we are allocating
607 	 * a new block.  If it had collapsed, tiqe would not be NULL and
608 	 * we would be reusing it.
609 	 * XXX If we can't, just drop the packet.  XXX
610 	 */
611 	if (tiqe == NULL) {
612 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
613 		if (tiqe == NULL) {
614 			tcpstat.tcps_rcvmemdrop++;
615 			m_freem(m);
616 			return (0);
617 		}
618 	}
619 
620 	/*
621 	 * Update the counters.
622 	 */
623 	tcpstat.tcps_rcvoopack++;
624 	tcpstat.tcps_rcvoobyte += rcvoobyte;
625 	if (rcvpartdupbyte) {
626 	    tcpstat.tcps_rcvpartduppack++;
627 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
628 	}
629 
630 	/*
631 	 * Insert the new fragment queue entry into both queues.
632 	 */
633 	tiqe->ipqe_m = m;
634 	tiqe->ipqe_seq = pkt_seq;
635 	tiqe->ipqe_len = pkt_len;
636 	tiqe->ipqe_flags = pkt_flags;
637 	if (p == NULL) {
638 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
639 #ifdef TCPREASS_DEBUG
640 		if (tiqe->ipqe_seq != tp->rcv_nxt)
641 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
642 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
643 #endif
644 	} else {
645 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
646 #ifdef TCPREASS_DEBUG
647 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
648 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
649 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
650 #endif
651 	}
652 
653 skip_replacement:
654 
655 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
656 
657 present:
658 	/*
659 	 * Present data to user, advancing rcv_nxt through
660 	 * completed sequence space.
661 	 */
662 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
663 		return (0);
664 	q = TAILQ_FIRST(&tp->segq);
665 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
666 		return (0);
667 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
668 		return (0);
669 
670 	tp->rcv_nxt += q->ipqe_len;
671 	pkt_flags = q->ipqe_flags & TH_FIN;
672 	ND6_HINT(tp);
673 
674 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
675 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
676 	if (so->so_state & SS_CANTRCVMORE)
677 		m_freem(q->ipqe_m);
678 	else
679 		sbappendstream(&so->so_rcv, q->ipqe_m);
680 	pool_put(&ipqent_pool, q);
681 	sorwakeup(so);
682 	return (pkt_flags);
683 }
684 
685 #ifdef INET6
686 int
687 tcp6_input(mp, offp, proto)
688 	struct mbuf **mp;
689 	int *offp, proto;
690 {
691 	struct mbuf *m = *mp;
692 
693 	/*
694 	 * draft-itojun-ipv6-tcp-to-anycast
695 	 * better place to put this in?
696 	 */
697 	if (m->m_flags & M_ANYCAST6) {
698 		struct ip6_hdr *ip6;
699 		if (m->m_len < sizeof(struct ip6_hdr)) {
700 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
701 				tcpstat.tcps_rcvshort++;
702 				return IPPROTO_DONE;
703 			}
704 		}
705 		ip6 = mtod(m, struct ip6_hdr *);
706 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
707 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
708 		return IPPROTO_DONE;
709 	}
710 
711 	tcp_input(m, *offp, proto);
712 	return IPPROTO_DONE;
713 }
714 #endif
715 
716 #ifdef INET
717 static void
718 tcp4_log_refused(ip, th)
719 	const struct ip *ip;
720 	const struct tcphdr *th;
721 {
722 	char src[4*sizeof "123"];
723 	char dst[4*sizeof "123"];
724 
725 	if (ip) {
726 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
727 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
728 	}
729 	else {
730 		strlcpy(src, "(unknown)", sizeof(src));
731 		strlcpy(dst, "(unknown)", sizeof(dst));
732 	}
733 	log(LOG_INFO,
734 	    "Connection attempt to TCP %s:%d from %s:%d\n",
735 	    dst, ntohs(th->th_dport),
736 	    src, ntohs(th->th_sport));
737 }
738 #endif
739 
740 #ifdef INET6
741 static void
742 tcp6_log_refused(ip6, th)
743 	const struct ip6_hdr *ip6;
744 	const struct tcphdr *th;
745 {
746 	char src[INET6_ADDRSTRLEN];
747 	char dst[INET6_ADDRSTRLEN];
748 
749 	if (ip6) {
750 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
751 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
752 	}
753 	else {
754 		strlcpy(src, "(unknown v6)", sizeof(src));
755 		strlcpy(dst, "(unknown v6)", sizeof(dst));
756 	}
757 	log(LOG_INFO,
758 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
759 	    dst, ntohs(th->th_dport),
760 	    src, ntohs(th->th_sport));
761 }
762 #endif
763 
764 /*
765  * TCP input routine, follows pages 65-76 of the
766  * protocol specification dated September, 1981 very closely.
767  */
768 void
769 #if __STDC__
770 tcp_input(struct mbuf *m, ...)
771 #else
772 tcp_input(m, va_alist)
773 	struct mbuf *m;
774 #endif
775 {
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 #ifdef TCP_DEBUG
791 	short ostate = 0;
792 #endif
793 	int iss = 0;
794 	u_long tiwin;
795 	struct tcp_opt_info opti;
796 	int off, iphlen;
797 	va_list ap;
798 	int af;		/* af on the wire */
799 	struct mbuf *tcp_saveti = NULL;
800 
801 	MCLAIM(m, &tcp_rx_mowner);
802 	va_start(ap, m);
803 	toff = va_arg(ap, int);
804 	(void)va_arg(ap, int);		/* ignore value, advance ap */
805 	va_end(ap);
806 
807 	tcpstat.tcps_rcvtotal++;
808 
809 	bzero(&opti, sizeof(opti));
810 	opti.ts_present = 0;
811 	opti.maxseg = 0;
812 
813 	/*
814 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
815 	 *
816 	 * TCP is, by definition, unicast, so we reject all
817 	 * multicast outright.
818 	 *
819 	 * Note, there are additional src/dst address checks in
820 	 * the AF-specific code below.
821 	 */
822 	if (m->m_flags & (M_BCAST|M_MCAST)) {
823 		/* XXX stat */
824 		goto drop;
825 	}
826 #ifdef INET6
827 	if (m->m_flags & M_ANYCAST6) {
828 		/* XXX stat */
829 		goto drop;
830 	}
831 #endif
832 
833 	/*
834 	 * Get IP and TCP header together in first mbuf.
835 	 * Note: IP leaves IP header in first mbuf.
836 	 */
837 	ip = mtod(m, struct ip *);
838 #ifdef INET6
839 	ip6 = NULL;
840 #endif
841 	switch (ip->ip_v) {
842 #ifdef INET
843 	case 4:
844 		af = AF_INET;
845 		iphlen = sizeof(struct ip);
846 		ip = mtod(m, struct ip *);
847 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
848 			sizeof(struct tcphdr));
849 		if (th == NULL) {
850 			tcpstat.tcps_rcvshort++;
851 			return;
852 		}
853 		/* We do the checksum after PCB lookup... */
854 		len = ntohs(ip->ip_len);
855 		tlen = len - toff;
856 		break;
857 #endif
858 #ifdef INET6
859 	case 6:
860 		ip = NULL;
861 		iphlen = sizeof(struct ip6_hdr);
862 		af = AF_INET6;
863 		ip6 = mtod(m, struct ip6_hdr *);
864 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
865 			sizeof(struct tcphdr));
866 		if (th == NULL) {
867 			tcpstat.tcps_rcvshort++;
868 			return;
869 		}
870 
871 		/* Be proactive about malicious use of IPv4 mapped address */
872 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
873 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
874 			/* XXX stat */
875 			goto drop;
876 		}
877 
878 		/*
879 		 * Be proactive about unspecified IPv6 address in source.
880 		 * As we use all-zero to indicate unbounded/unconnected pcb,
881 		 * unspecified IPv6 address can be used to confuse us.
882 		 *
883 		 * Note that packets with unspecified IPv6 destination is
884 		 * already dropped in ip6_input.
885 		 */
886 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
887 			/* XXX stat */
888 			goto drop;
889 		}
890 
891 		/*
892 		 * Make sure destination address is not multicast.
893 		 * Source address checked in ip6_input().
894 		 */
895 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
896 			/* XXX stat */
897 			goto drop;
898 		}
899 
900 		/* We do the checksum after PCB lookup... */
901 		len = m->m_pkthdr.len;
902 		tlen = len - toff;
903 		break;
904 #endif
905 	default:
906 		m_freem(m);
907 		return;
908 	}
909 
910 	KASSERT(TCP_HDR_ALIGNED_P(th));
911 
912 	/*
913 	 * Check that TCP offset makes sense,
914 	 * pull out TCP options and adjust length.		XXX
915 	 */
916 	off = th->th_off << 2;
917 	if (off < sizeof (struct tcphdr) || off > tlen) {
918 		tcpstat.tcps_rcvbadoff++;
919 		goto drop;
920 	}
921 	tlen -= off;
922 
923 	/*
924 	 * tcp_input() has been modified to use tlen to mean the TCP data
925 	 * length throughout the function.  Other functions can use
926 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
927 	 * rja
928 	 */
929 
930 	if (off > sizeof (struct tcphdr)) {
931 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
932 		if (th == NULL) {
933 			tcpstat.tcps_rcvshort++;
934 			return;
935 		}
936 		/*
937 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
938 		 * (as they're before toff) and we don't need to update those.
939 		 */
940 		KASSERT(TCP_HDR_ALIGNED_P(th));
941 		optlen = off - sizeof (struct tcphdr);
942 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
943 		/*
944 		 * Do quick retrieval of timestamp options ("options
945 		 * prediction?").  If timestamp is the only option and it's
946 		 * formatted as recommended in RFC 1323 appendix A, we
947 		 * quickly get the values now and not bother calling
948 		 * tcp_dooptions(), etc.
949 		 */
950 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
951 		     (optlen > TCPOLEN_TSTAMP_APPA &&
952 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
953 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
954 		     (th->th_flags & TH_SYN) == 0) {
955 			opti.ts_present = 1;
956 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
957 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
958 			optp = NULL;	/* we've parsed the options */
959 		}
960 	}
961 	tiflags = th->th_flags;
962 
963 	/*
964 	 * Locate pcb for segment.
965 	 */
966 findpcb:
967 	inp = NULL;
968 #ifdef INET6
969 	in6p = NULL;
970 #endif
971 	switch (af) {
972 #ifdef INET
973 	case AF_INET:
974 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
975 		    ip->ip_dst, th->th_dport);
976 		if (inp == 0) {
977 			++tcpstat.tcps_pcbhashmiss;
978 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
979 		}
980 #ifdef INET6
981 		if (inp == 0) {
982 			struct in6_addr s, d;
983 
984 			/* mapped addr case */
985 			bzero(&s, sizeof(s));
986 			s.s6_addr16[5] = htons(0xffff);
987 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
988 			bzero(&d, sizeof(d));
989 			d.s6_addr16[5] = htons(0xffff);
990 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
991 			in6p = in6_pcblookup_connect(&tcbtable, &s,
992 			    th->th_sport, &d, th->th_dport, 0);
993 			if (in6p == 0) {
994 				++tcpstat.tcps_pcbhashmiss;
995 				in6p = in6_pcblookup_bind(&tcbtable, &d,
996 				    th->th_dport, 0);
997 			}
998 		}
999 #endif
1000 #ifndef INET6
1001 		if (inp == 0)
1002 #else
1003 		if (inp == 0 && in6p == 0)
1004 #endif
1005 		{
1006 			++tcpstat.tcps_noport;
1007 			if (tcp_log_refused &&
1008 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1009 				tcp4_log_refused(ip, th);
1010 			}
1011 			TCP_FIELDS_TO_HOST(th);
1012 			goto dropwithreset_ratelim;
1013 		}
1014 #if defined(IPSEC) || defined(FAST_IPSEC)
1015 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1016 		    ipsec4_in_reject(m, inp)) {
1017 			ipsecstat.in_polvio++;
1018 			goto drop;
1019 		}
1020 #ifdef INET6
1021 		else if (in6p &&
1022 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1023 		    ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1024 			ipsecstat.in_polvio++;
1025 			goto drop;
1026 		}
1027 #endif
1028 #endif /*IPSEC*/
1029 		break;
1030 #endif /*INET*/
1031 #ifdef INET6
1032 	case AF_INET6:
1033 	    {
1034 		int faith;
1035 
1036 #if defined(NFAITH) && NFAITH > 0
1037 		faith = faithprefix(&ip6->ip6_dst);
1038 #else
1039 		faith = 0;
1040 #endif
1041 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1042 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1043 		if (in6p == NULL) {
1044 			++tcpstat.tcps_pcbhashmiss;
1045 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1046 				th->th_dport, faith);
1047 		}
1048 		if (in6p == NULL) {
1049 			++tcpstat.tcps_noport;
1050 			if (tcp_log_refused &&
1051 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1052 				tcp6_log_refused(ip6, th);
1053 			}
1054 			TCP_FIELDS_TO_HOST(th);
1055 			goto dropwithreset_ratelim;
1056 		}
1057 #if defined(IPSEC) || defined(FAST_IPSEC)
1058 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1059 		    ipsec6_in_reject(m, in6p)) {
1060 			ipsec6stat.in_polvio++;
1061 			goto drop;
1062 		}
1063 #endif /*IPSEC*/
1064 		break;
1065 	    }
1066 #endif
1067 	}
1068 
1069 	/*
1070 	 * If the state is CLOSED (i.e., TCB does not exist) then
1071 	 * all data in the incoming segment is discarded.
1072 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1073 	 * but should either do a listen or a connect soon.
1074 	 */
1075 	tp = NULL;
1076 	so = NULL;
1077 	if (inp) {
1078 		tp = intotcpcb(inp);
1079 		so = inp->inp_socket;
1080 	}
1081 #ifdef INET6
1082 	else if (in6p) {
1083 		tp = in6totcpcb(in6p);
1084 		so = in6p->in6p_socket;
1085 	}
1086 #endif
1087 	if (tp == 0) {
1088 		TCP_FIELDS_TO_HOST(th);
1089 		goto dropwithreset_ratelim;
1090 	}
1091 	if (tp->t_state == TCPS_CLOSED)
1092 		goto drop;
1093 
1094 	/*
1095 	 * Checksum extended TCP header and data.
1096 	 */
1097 	switch (af) {
1098 #ifdef INET
1099 	case AF_INET:
1100 		switch (m->m_pkthdr.csum_flags &
1101 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
1102 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
1103 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
1104 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
1105 			goto badcsum;
1106 
1107 		case M_CSUM_TCPv4|M_CSUM_DATA: {
1108 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
1109 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
1110 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
1111 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
1112 				    ip->ip_dst.s_addr,
1113 				    htonl(hw_csum + ntohs(ip->ip_len) +
1114 					  IPPROTO_UDP));
1115 			}
1116 			if ((hw_csum ^ 0xffff) != 0)
1117 				goto badcsum;
1118 			break;
1119 		}
1120 
1121 		case M_CSUM_TCPv4:
1122 			/* Checksum was okay. */
1123 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
1124 			break;
1125 
1126 		default:
1127 			/* Must compute it ourselves. */
1128 			TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
1129 			if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1130 				goto badcsum;
1131 			break;
1132 		}
1133 		break;
1134 #endif /* INET4 */
1135 
1136 #ifdef INET6
1137 	case AF_INET6:
1138 		if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1139 			goto badcsum;
1140 		break;
1141 #endif /* INET6 */
1142 	}
1143 
1144 	TCP_FIELDS_TO_HOST(th);
1145 
1146 	/* Unscale the window into a 32-bit value. */
1147 	if ((tiflags & TH_SYN) == 0)
1148 		tiwin = th->th_win << tp->snd_scale;
1149 	else
1150 		tiwin = th->th_win;
1151 
1152 #ifdef INET6
1153 	/* save packet options if user wanted */
1154 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1155 		if (in6p->in6p_options) {
1156 			m_freem(in6p->in6p_options);
1157 			in6p->in6p_options = 0;
1158 		}
1159 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1160 	}
1161 #endif
1162 
1163 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1164 		union syn_cache_sa src;
1165 		union syn_cache_sa dst;
1166 
1167 		bzero(&src, sizeof(src));
1168 		bzero(&dst, sizeof(dst));
1169 		switch (af) {
1170 #ifdef INET
1171 		case AF_INET:
1172 			src.sin.sin_len = sizeof(struct sockaddr_in);
1173 			src.sin.sin_family = AF_INET;
1174 			src.sin.sin_addr = ip->ip_src;
1175 			src.sin.sin_port = th->th_sport;
1176 
1177 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1178 			dst.sin.sin_family = AF_INET;
1179 			dst.sin.sin_addr = ip->ip_dst;
1180 			dst.sin.sin_port = th->th_dport;
1181 			break;
1182 #endif
1183 #ifdef INET6
1184 		case AF_INET6:
1185 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1186 			src.sin6.sin6_family = AF_INET6;
1187 			src.sin6.sin6_addr = ip6->ip6_src;
1188 			src.sin6.sin6_port = th->th_sport;
1189 
1190 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1191 			dst.sin6.sin6_family = AF_INET6;
1192 			dst.sin6.sin6_addr = ip6->ip6_dst;
1193 			dst.sin6.sin6_port = th->th_dport;
1194 			break;
1195 #endif /* INET6 */
1196 		default:
1197 			goto badsyn;	/*sanity*/
1198 		}
1199 
1200 		if (so->so_options & SO_DEBUG) {
1201 #ifdef TCP_DEBUG
1202 			ostate = tp->t_state;
1203 #endif
1204 
1205 			tcp_saveti = NULL;
1206 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1207 				goto nosave;
1208 
1209 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1210 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1211 				if (!tcp_saveti)
1212 					goto nosave;
1213 			} else {
1214 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1215 				if (!tcp_saveti)
1216 					goto nosave;
1217 				MCLAIM(m, &tcp_mowner);
1218 				tcp_saveti->m_len = iphlen;
1219 				m_copydata(m, 0, iphlen,
1220 				    mtod(tcp_saveti, caddr_t));
1221 			}
1222 
1223 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1224 				m_freem(tcp_saveti);
1225 				tcp_saveti = NULL;
1226 			} else {
1227 				tcp_saveti->m_len += sizeof(struct tcphdr);
1228 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1229 				    sizeof(struct tcphdr));
1230 			}
1231 	nosave:;
1232 		}
1233 		if (so->so_options & SO_ACCEPTCONN) {
1234 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1235 				if (tiflags & TH_RST) {
1236 					syn_cache_reset(&src.sa, &dst.sa, th);
1237 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1238 				    (TH_ACK|TH_SYN)) {
1239 					/*
1240 					 * Received a SYN,ACK.  This should
1241 					 * never happen while we are in
1242 					 * LISTEN.  Send an RST.
1243 					 */
1244 					goto badsyn;
1245 				} else if (tiflags & TH_ACK) {
1246 					so = syn_cache_get(&src.sa, &dst.sa,
1247 						th, toff, tlen, so, m);
1248 					if (so == NULL) {
1249 						/*
1250 						 * We don't have a SYN for
1251 						 * this ACK; send an RST.
1252 						 */
1253 						goto badsyn;
1254 					} else if (so ==
1255 					    (struct socket *)(-1)) {
1256 						/*
1257 						 * We were unable to create
1258 						 * the connection.  If the
1259 						 * 3-way handshake was
1260 						 * completed, and RST has
1261 						 * been sent to the peer.
1262 						 * Since the mbuf might be
1263 						 * in use for the reply,
1264 						 * do not free it.
1265 						 */
1266 						m = NULL;
1267 					} else {
1268 						/*
1269 						 * We have created a
1270 						 * full-blown connection.
1271 						 */
1272 						tp = NULL;
1273 						inp = NULL;
1274 #ifdef INET6
1275 						in6p = NULL;
1276 #endif
1277 						switch (so->so_proto->pr_domain->dom_family) {
1278 #ifdef INET
1279 						case AF_INET:
1280 							inp = sotoinpcb(so);
1281 							tp = intotcpcb(inp);
1282 							break;
1283 #endif
1284 #ifdef INET6
1285 						case AF_INET6:
1286 							in6p = sotoin6pcb(so);
1287 							tp = in6totcpcb(in6p);
1288 							break;
1289 #endif
1290 						}
1291 						if (tp == NULL)
1292 							goto badsyn;	/*XXX*/
1293 						tiwin <<= tp->snd_scale;
1294 						goto after_listen;
1295 					}
1296 				} else {
1297 					/*
1298 					 * None of RST, SYN or ACK was set.
1299 					 * This is an invalid packet for a
1300 					 * TCB in LISTEN state.  Send a RST.
1301 					 */
1302 					goto badsyn;
1303 				}
1304 			} else {
1305 				/*
1306 				 * Received a SYN.
1307 				 */
1308 
1309 #ifdef INET6
1310 				/*
1311 				 * If deprecated address is forbidden, we do
1312 				 * not accept SYN to deprecated interface
1313 				 * address to prevent any new inbound
1314 				 * connection from getting established.
1315 				 * When we do not accept SYN, we send a TCP
1316 				 * RST, with deprecated source address (instead
1317 				 * of dropping it).  We compromise it as it is
1318 				 * much better for peer to send a RST, and
1319 				 * RST will be the final packet for the
1320 				 * exchange.
1321 				 *
1322 				 * If we do not forbid deprecated addresses, we
1323 				 * accept the SYN packet.  RFC2462 does not
1324 				 * suggest dropping SYN in this case.
1325 				 * If we decipher RFC2462 5.5.4, it says like
1326 				 * this:
1327 				 * 1. use of deprecated addr with existing
1328 				 *    communication is okay - "SHOULD continue
1329 				 *    to be used"
1330 				 * 2. use of it with new communication:
1331 				 *   (2a) "SHOULD NOT be used if alternate
1332 				 *        address with sufficient scope is
1333 				 *        available"
1334 				 *   (2b) nothing mentioned otherwise.
1335 				 * Here we fall into (2b) case as we have no
1336 				 * choice in our source address selection - we
1337 				 * must obey the peer.
1338 				 *
1339 				 * The wording in RFC2462 is confusing, and
1340 				 * there are multiple description text for
1341 				 * deprecated address handling - worse, they
1342 				 * are not exactly the same.  I believe 5.5.4
1343 				 * is the best one, so we follow 5.5.4.
1344 				 */
1345 				if (af == AF_INET6 && !ip6_use_deprecated) {
1346 					struct in6_ifaddr *ia6;
1347 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1348 					    &ip6->ip6_dst)) &&
1349 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1350 						tp = NULL;
1351 						goto dropwithreset;
1352 					}
1353 				}
1354 #endif
1355 
1356 #ifdef IPSEC
1357 				switch (af) {
1358 				case AF_INET:
1359 					if (ipsec4_in_reject_so(m, so)) {
1360 						ipsecstat.in_polvio++;
1361 						tp = NULL;
1362 						goto dropwithreset;
1363 					}
1364 					break;
1365 #ifdef INET6
1366 				case AF_INET6:
1367 					if (ipsec6_in_reject_so(m, so)) {
1368 						ipsec6stat.in_polvio++;
1369 						tp = NULL;
1370 						goto dropwithreset;
1371 					}
1372 					break;
1373 #endif
1374 				}
1375 #endif
1376 
1377 				/*
1378 				 * LISTEN socket received a SYN
1379 				 * from itself?  This can't possibly
1380 				 * be valid; drop the packet.
1381 				 */
1382 				if (th->th_sport == th->th_dport) {
1383 					int i;
1384 
1385 					switch (af) {
1386 #ifdef INET
1387 					case AF_INET:
1388 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1389 						break;
1390 #endif
1391 #ifdef INET6
1392 					case AF_INET6:
1393 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1394 						break;
1395 #endif
1396 					default:
1397 						i = 1;
1398 					}
1399 					if (i) {
1400 						tcpstat.tcps_badsyn++;
1401 						goto drop;
1402 					}
1403 				}
1404 
1405 				/*
1406 				 * SYN looks ok; create compressed TCP
1407 				 * state for it.
1408 				 */
1409 				if (so->so_qlen <= so->so_qlimit &&
1410 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1411 						so, m, optp, optlen, &opti))
1412 					m = NULL;
1413 			}
1414 			goto drop;
1415 		}
1416 	}
1417 
1418 after_listen:
1419 #ifdef DIAGNOSTIC
1420 	/*
1421 	 * Should not happen now that all embryonic connections
1422 	 * are handled with compressed state.
1423 	 */
1424 	if (tp->t_state == TCPS_LISTEN)
1425 		panic("tcp_input: TCPS_LISTEN");
1426 #endif
1427 
1428 	/*
1429 	 * Segment received on connection.
1430 	 * Reset idle time and keep-alive timer.
1431 	 */
1432 	tp->t_rcvtime = tcp_now;
1433 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1434 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1435 
1436 	/*
1437 	 * Process options.
1438 	 */
1439 	if (optp)
1440 		tcp_dooptions(tp, optp, optlen, th, &opti);
1441 
1442 	/*
1443 	 * Header prediction: check for the two common cases
1444 	 * of a uni-directional data xfer.  If the packet has
1445 	 * no control flags, is in-sequence, the window didn't
1446 	 * change and we're not retransmitting, it's a
1447 	 * candidate.  If the length is zero and the ack moved
1448 	 * forward, we're the sender side of the xfer.  Just
1449 	 * free the data acked & wake any higher level process
1450 	 * that was blocked waiting for space.  If the length
1451 	 * is non-zero and the ack didn't move, we're the
1452 	 * receiver side.  If we're getting packets in-order
1453 	 * (the reassembly queue is empty), add the data to
1454 	 * the socket buffer and note that we need a delayed ack.
1455 	 */
1456 	if (tp->t_state == TCPS_ESTABLISHED &&
1457 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1458 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1459 	    th->th_seq == tp->rcv_nxt &&
1460 	    tiwin && tiwin == tp->snd_wnd &&
1461 	    tp->snd_nxt == tp->snd_max) {
1462 
1463 		/*
1464 		 * If last ACK falls within this segment's sequence numbers,
1465 		 *  record the timestamp.
1466 		 */
1467 		if (opti.ts_present &&
1468 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1469 		    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) {
1470 			tp->ts_recent_age = TCP_TIMESTAMP(tp);
1471 			tp->ts_recent = opti.ts_val;
1472 		}
1473 
1474 		if (tlen == 0) {
1475 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1476 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1477 			    tp->snd_cwnd >= tp->snd_wnd &&
1478 			    tp->t_dupacks < tcprexmtthresh) {
1479 				int slen;
1480 
1481 				/*
1482 				 * this is a pure ack for outstanding data.
1483 				 */
1484 				++tcpstat.tcps_predack;
1485 				if (opti.ts_present && opti.ts_ecr)
1486 					tcp_xmit_timer(tp,
1487 					  TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
1488 				else if (tp->t_rtttime &&
1489 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1490 					tcp_xmit_timer(tp,
1491 					tcp_now - tp->t_rtttime);
1492 				acked = th->th_ack - tp->snd_una;
1493 				tcpstat.tcps_rcvackpack++;
1494 				tcpstat.tcps_rcvackbyte += acked;
1495 				ND6_HINT(tp);
1496 
1497 				slen = tp->t_lastm->m_len;
1498 				sbdrop(&so->so_snd, acked);
1499 				if (so->so_snd.sb_cc != 0) {
1500 					tp->t_lastoff -= acked;
1501 					if (tp->t_lastm->m_len < slen)
1502 						tp->t_inoff -=
1503 						    (slen - tp->t_lastm->m_len);
1504 				}
1505 
1506 				/*
1507 				 * We want snd_recover to track snd_una to
1508 				 * avoid sequence wraparound problems for
1509 				 * very large transfers.
1510 				 */
1511 				tp->snd_una = tp->snd_recover = th->th_ack;
1512 				m_freem(m);
1513 
1514 				/*
1515 				 * If all outstanding data are acked, stop
1516 				 * retransmit timer, otherwise restart timer
1517 				 * using current (possibly backed-off) value.
1518 				 * If process is waiting for space,
1519 				 * wakeup/selwakeup/signal.  If data
1520 				 * are ready to send, let tcp_output
1521 				 * decide between more output or persist.
1522 				 */
1523 				if (tp->snd_una == tp->snd_max)
1524 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1525 				else if (TCP_TIMER_ISARMED(tp,
1526 				    TCPT_PERSIST) == 0)
1527 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1528 					    tp->t_rxtcur);
1529 
1530 				sowwakeup(so);
1531 				if (so->so_snd.sb_cc)
1532 					(void) tcp_output(tp);
1533 				if (tcp_saveti)
1534 					m_freem(tcp_saveti);
1535 				return;
1536 			}
1537 		} else if (th->th_ack == tp->snd_una &&
1538 		    TAILQ_FIRST(&tp->segq) == NULL &&
1539 		    tlen <= sbspace(&so->so_rcv)) {
1540 			/*
1541 			 * this is a pure, in-sequence data packet
1542 			 * with nothing on the reassembly queue and
1543 			 * we have enough buffer space to take it.
1544 			 */
1545 			++tcpstat.tcps_preddat;
1546 			tp->rcv_nxt += tlen;
1547 			tcpstat.tcps_rcvpack++;
1548 			tcpstat.tcps_rcvbyte += tlen;
1549 			ND6_HINT(tp);
1550 			/*
1551 			 * Drop TCP, IP headers and TCP options then add data
1552 			 * to socket buffer.
1553 			 */
1554 			if (so->so_state & SS_CANTRCVMORE)
1555 				m_freem(m);
1556 			else {
1557 				m_adj(m, toff + off);
1558 				sbappendstream(&so->so_rcv, m);
1559 			}
1560 			sorwakeup(so);
1561 			TCP_SETUP_ACK(tp, th);
1562 			if (tp->t_flags & TF_ACKNOW)
1563 				(void) tcp_output(tp);
1564 			if (tcp_saveti)
1565 				m_freem(tcp_saveti);
1566 			return;
1567 		}
1568 	}
1569 
1570 	/*
1571 	 * Compute mbuf offset to TCP data segment.
1572 	 */
1573 	hdroptlen = toff + off;
1574 
1575 	/*
1576 	 * Calculate amount of space in receive window,
1577 	 * and then do TCP input processing.
1578 	 * Receive window is amount of space in rcv queue,
1579 	 * but not less than advertised window.
1580 	 */
1581 	{ int win;
1582 
1583 	win = sbspace(&so->so_rcv);
1584 	if (win < 0)
1585 		win = 0;
1586 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1587 	}
1588 
1589 	switch (tp->t_state) {
1590 	case TCPS_LISTEN:
1591 		/*
1592 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1593 		 */
1594 		if (m->m_flags & (M_BCAST|M_MCAST))
1595 			goto drop;
1596 		switch (af) {
1597 #ifdef INET6
1598 		case AF_INET6:
1599 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1600 				goto drop;
1601 			break;
1602 #endif /* INET6 */
1603 		case AF_INET:
1604 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1605 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1606 				goto drop;
1607 			break;
1608 		}
1609 		break;
1610 
1611 	/*
1612 	 * If the state is SYN_SENT:
1613 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1614 	 *	if seg contains a RST, then drop the connection.
1615 	 *	if seg does not contain SYN, then drop it.
1616 	 * Otherwise this is an acceptable SYN segment
1617 	 *	initialize tp->rcv_nxt and tp->irs
1618 	 *	if seg contains ack then advance tp->snd_una
1619 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1620 	 *	arrange for segment to be acked (eventually)
1621 	 *	continue processing rest of data/controls, beginning with URG
1622 	 */
1623 	case TCPS_SYN_SENT:
1624 		if ((tiflags & TH_ACK) &&
1625 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1626 		     SEQ_GT(th->th_ack, tp->snd_max)))
1627 			goto dropwithreset;
1628 		if (tiflags & TH_RST) {
1629 			if (tiflags & TH_ACK)
1630 				tp = tcp_drop(tp, ECONNREFUSED);
1631 			goto drop;
1632 		}
1633 		if ((tiflags & TH_SYN) == 0)
1634 			goto drop;
1635 		if (tiflags & TH_ACK) {
1636 			tp->snd_una = tp->snd_recover = th->th_ack;
1637 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1638 				tp->snd_nxt = tp->snd_una;
1639 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1640 		}
1641 		tp->irs = th->th_seq;
1642 		tcp_rcvseqinit(tp);
1643 		tp->t_flags |= TF_ACKNOW;
1644 		tcp_mss_from_peer(tp, opti.maxseg);
1645 
1646 		/*
1647 		 * Initialize the initial congestion window.  If we
1648 		 * had to retransmit the SYN, we must initialize cwnd
1649 		 * to 1 segment (i.e. the Loss Window).
1650 		 */
1651 		if (tp->t_flags & TF_SYN_REXMT)
1652 			tp->snd_cwnd = tp->t_peermss;
1653 		else {
1654 			int ss = tcp_init_win;
1655 #ifdef INET
1656 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1657 				ss = tcp_init_win_local;
1658 #endif
1659 #ifdef INET6
1660 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1661 				ss = tcp_init_win_local;
1662 #endif
1663 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1664 		}
1665 
1666 		tcp_rmx_rtt(tp);
1667 		if (tiflags & TH_ACK) {
1668 			tcpstat.tcps_connects++;
1669 			soisconnected(so);
1670 			tcp_established(tp);
1671 			/* Do window scaling on this connection? */
1672 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1673 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1674 				tp->snd_scale = tp->requested_s_scale;
1675 				tp->rcv_scale = tp->request_r_scale;
1676 			}
1677 			TCP_REASS_LOCK(tp);
1678 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1679 			TCP_REASS_UNLOCK(tp);
1680 			/*
1681 			 * if we didn't have to retransmit the SYN,
1682 			 * use its rtt as our initial srtt & rtt var.
1683 			 */
1684 			if (tp->t_rtttime)
1685 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1686 		} else
1687 			tp->t_state = TCPS_SYN_RECEIVED;
1688 
1689 		/*
1690 		 * Advance th->th_seq to correspond to first data byte.
1691 		 * If data, trim to stay within window,
1692 		 * dropping FIN if necessary.
1693 		 */
1694 		th->th_seq++;
1695 		if (tlen > tp->rcv_wnd) {
1696 			todrop = tlen - tp->rcv_wnd;
1697 			m_adj(m, -todrop);
1698 			tlen = tp->rcv_wnd;
1699 			tiflags &= ~TH_FIN;
1700 			tcpstat.tcps_rcvpackafterwin++;
1701 			tcpstat.tcps_rcvbyteafterwin += todrop;
1702 		}
1703 		tp->snd_wl1 = th->th_seq - 1;
1704 		tp->rcv_up = th->th_seq;
1705 		goto step6;
1706 
1707 	/*
1708 	 * If the state is SYN_RECEIVED:
1709 	 *	If seg contains an ACK, but not for our SYN, drop the input
1710 	 *	and generate an RST.  See page 36, rfc793
1711 	 */
1712 	case TCPS_SYN_RECEIVED:
1713 		if ((tiflags & TH_ACK) &&
1714 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1715 		     SEQ_GT(th->th_ack, tp->snd_max)))
1716 			goto dropwithreset;
1717 		break;
1718 	}
1719 
1720 	/*
1721 	 * States other than LISTEN or SYN_SENT.
1722 	 * First check timestamp, if present.
1723 	 * Then check that at least some bytes of segment are within
1724 	 * receive window.  If segment begins before rcv_nxt,
1725 	 * drop leading data (and SYN); if nothing left, just ack.
1726 	 *
1727 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1728 	 * and it's less than ts_recent, drop it.
1729 	 */
1730 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1731 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1732 
1733 		/* Check to see if ts_recent is over 24 days old.  */
1734 		if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) >
1735 		    TCP_PAWS_IDLE) {
1736 			/*
1737 			 * Invalidate ts_recent.  If this segment updates
1738 			 * ts_recent, the age will be reset later and ts_recent
1739 			 * will get a valid value.  If it does not, setting
1740 			 * ts_recent to zero will at least satisfy the
1741 			 * requirement that zero be placed in the timestamp
1742 			 * echo reply when ts_recent isn't valid.  The
1743 			 * age isn't reset until we get a valid ts_recent
1744 			 * because we don't want out-of-order segments to be
1745 			 * dropped when ts_recent is old.
1746 			 */
1747 			tp->ts_recent = 0;
1748 		} else {
1749 			tcpstat.tcps_rcvduppack++;
1750 			tcpstat.tcps_rcvdupbyte += tlen;
1751 			tcpstat.tcps_pawsdrop++;
1752 			goto dropafterack;
1753 		}
1754 	}
1755 
1756 	todrop = tp->rcv_nxt - th->th_seq;
1757 	if (todrop > 0) {
1758 		if (tiflags & TH_SYN) {
1759 			tiflags &= ~TH_SYN;
1760 			th->th_seq++;
1761 			if (th->th_urp > 1)
1762 				th->th_urp--;
1763 			else {
1764 				tiflags &= ~TH_URG;
1765 				th->th_urp = 0;
1766 			}
1767 			todrop--;
1768 		}
1769 		if (todrop > tlen ||
1770 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1771 			/*
1772 			 * Any valid FIN must be to the left of the window.
1773 			 * At this point the FIN must be a duplicate or
1774 			 * out of sequence; drop it.
1775 			 */
1776 			tiflags &= ~TH_FIN;
1777 			/*
1778 			 * Send an ACK to resynchronize and drop any data.
1779 			 * But keep on processing for RST or ACK.
1780 			 */
1781 			tp->t_flags |= TF_ACKNOW;
1782 			todrop = tlen;
1783 			tcpstat.tcps_rcvdupbyte += todrop;
1784 			tcpstat.tcps_rcvduppack++;
1785 		} else {
1786 			tcpstat.tcps_rcvpartduppack++;
1787 			tcpstat.tcps_rcvpartdupbyte += todrop;
1788 		}
1789 		hdroptlen += todrop;	/*drop from head afterwards*/
1790 		th->th_seq += todrop;
1791 		tlen -= todrop;
1792 		if (th->th_urp > todrop)
1793 			th->th_urp -= todrop;
1794 		else {
1795 			tiflags &= ~TH_URG;
1796 			th->th_urp = 0;
1797 		}
1798 	}
1799 
1800 	/*
1801 	 * If new data are received on a connection after the
1802 	 * user processes are gone, then RST the other end.
1803 	 */
1804 	if ((so->so_state & SS_NOFDREF) &&
1805 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1806 		tp = tcp_close(tp);
1807 		tcpstat.tcps_rcvafterclose++;
1808 		goto dropwithreset;
1809 	}
1810 
1811 	/*
1812 	 * If segment ends after window, drop trailing data
1813 	 * (and PUSH and FIN); if nothing left, just ACK.
1814 	 */
1815 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1816 	if (todrop > 0) {
1817 		tcpstat.tcps_rcvpackafterwin++;
1818 		if (todrop >= tlen) {
1819 			tcpstat.tcps_rcvbyteafterwin += tlen;
1820 			/*
1821 			 * If a new connection request is received
1822 			 * while in TIME_WAIT, drop the old connection
1823 			 * and start over if the sequence numbers
1824 			 * are above the previous ones.
1825 			 *
1826 			 * NOTE: We will checksum the packet again, and
1827 			 * so we need to put the header fields back into
1828 			 * network order!
1829 			 * XXX This kind of sucks, but we don't expect
1830 			 * XXX this to happen very often, so maybe it
1831 			 * XXX doesn't matter so much.
1832 			 */
1833 			if (tiflags & TH_SYN &&
1834 			    tp->t_state == TCPS_TIME_WAIT &&
1835 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1836 				iss = tcp_new_iss(tp, tp->snd_nxt);
1837 				tp = tcp_close(tp);
1838 				TCP_FIELDS_TO_NET(th);
1839 				goto findpcb;
1840 			}
1841 			/*
1842 			 * If window is closed can only take segments at
1843 			 * window edge, and have to drop data and PUSH from
1844 			 * incoming segments.  Continue processing, but
1845 			 * remember to ack.  Otherwise, drop segment
1846 			 * and ack.
1847 			 */
1848 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1849 				tp->t_flags |= TF_ACKNOW;
1850 				tcpstat.tcps_rcvwinprobe++;
1851 			} else
1852 				goto dropafterack;
1853 		} else
1854 			tcpstat.tcps_rcvbyteafterwin += todrop;
1855 		m_adj(m, -todrop);
1856 		tlen -= todrop;
1857 		tiflags &= ~(TH_PUSH|TH_FIN);
1858 	}
1859 
1860 	/*
1861 	 * If last ACK falls within this segment's sequence numbers,
1862 	 * and the timestamp is newer, record it.
1863 	 */
1864 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1865 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1866 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
1867 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1868 		tp->ts_recent_age = TCP_TIMESTAMP(tp);
1869 		tp->ts_recent = opti.ts_val;
1870 	}
1871 
1872 	/*
1873 	 * If the RST bit is set examine the state:
1874 	 *    SYN_RECEIVED STATE:
1875 	 *	If passive open, return to LISTEN state.
1876 	 *	If active open, inform user that connection was refused.
1877 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1878 	 *	Inform user that connection was reset, and close tcb.
1879 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1880 	 *	Close the tcb.
1881 	 */
1882 	if (tiflags&TH_RST) switch (tp->t_state) {
1883 
1884 	case TCPS_SYN_RECEIVED:
1885 		so->so_error = ECONNREFUSED;
1886 		goto close;
1887 
1888 	case TCPS_ESTABLISHED:
1889 	case TCPS_FIN_WAIT_1:
1890 	case TCPS_FIN_WAIT_2:
1891 	case TCPS_CLOSE_WAIT:
1892 		so->so_error = ECONNRESET;
1893 	close:
1894 		tp->t_state = TCPS_CLOSED;
1895 		tcpstat.tcps_drops++;
1896 		tp = tcp_close(tp);
1897 		goto drop;
1898 
1899 	case TCPS_CLOSING:
1900 	case TCPS_LAST_ACK:
1901 	case TCPS_TIME_WAIT:
1902 		tp = tcp_close(tp);
1903 		goto drop;
1904 	}
1905 
1906 	/*
1907 	 * If a SYN is in the window, then this is an
1908 	 * error and we send an RST and drop the connection.
1909 	 */
1910 	if (tiflags & TH_SYN) {
1911 		tp = tcp_drop(tp, ECONNRESET);
1912 		goto dropwithreset;
1913 	}
1914 
1915 	/*
1916 	 * If the ACK bit is off we drop the segment and return.
1917 	 */
1918 	if ((tiflags & TH_ACK) == 0) {
1919 		if (tp->t_flags & TF_ACKNOW)
1920 			goto dropafterack;
1921 		else
1922 			goto drop;
1923 	}
1924 
1925 	/*
1926 	 * Ack processing.
1927 	 */
1928 	switch (tp->t_state) {
1929 
1930 	/*
1931 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1932 	 * ESTABLISHED state and continue processing, otherwise
1933 	 * send an RST.
1934 	 */
1935 	case TCPS_SYN_RECEIVED:
1936 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
1937 		    SEQ_GT(th->th_ack, tp->snd_max))
1938 			goto dropwithreset;
1939 		tcpstat.tcps_connects++;
1940 		soisconnected(so);
1941 		tcp_established(tp);
1942 		/* Do window scaling? */
1943 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1944 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1945 			tp->snd_scale = tp->requested_s_scale;
1946 			tp->rcv_scale = tp->request_r_scale;
1947 		}
1948 		TCP_REASS_LOCK(tp);
1949 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1950 		TCP_REASS_UNLOCK(tp);
1951 		tp->snd_wl1 = th->th_seq - 1;
1952 		/* fall into ... */
1953 
1954 	/*
1955 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1956 	 * ACKs.  If the ack is in the range
1957 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1958 	 * then advance tp->snd_una to th->th_ack and drop
1959 	 * data from the retransmission queue.  If this ACK reflects
1960 	 * more up to date window information we update our window information.
1961 	 */
1962 	case TCPS_ESTABLISHED:
1963 	case TCPS_FIN_WAIT_1:
1964 	case TCPS_FIN_WAIT_2:
1965 	case TCPS_CLOSE_WAIT:
1966 	case TCPS_CLOSING:
1967 	case TCPS_LAST_ACK:
1968 	case TCPS_TIME_WAIT:
1969 
1970 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1971 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1972 				tcpstat.tcps_rcvdupack++;
1973 				/*
1974 				 * If we have outstanding data (other than
1975 				 * a window probe), this is a completely
1976 				 * duplicate ack (ie, window info didn't
1977 				 * change), the ack is the biggest we've
1978 				 * seen and we've seen exactly our rexmt
1979 				 * threshhold of them, assume a packet
1980 				 * has been dropped and retransmit it.
1981 				 * Kludge snd_nxt & the congestion
1982 				 * window so we send only this one
1983 				 * packet.
1984 				 *
1985 				 * We know we're losing at the current
1986 				 * window size so do congestion avoidance
1987 				 * (set ssthresh to half the current window
1988 				 * and pull our congestion window back to
1989 				 * the new ssthresh).
1990 				 *
1991 				 * Dup acks mean that packets have left the
1992 				 * network (they're now cached at the receiver)
1993 				 * so bump cwnd by the amount in the receiver
1994 				 * to keep a constant cwnd packets in the
1995 				 * network.
1996 				 */
1997 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
1998 				    th->th_ack != tp->snd_una)
1999 					tp->t_dupacks = 0;
2000 				else if (++tp->t_dupacks == tcprexmtthresh) {
2001 					tcp_seq onxt = tp->snd_nxt;
2002 					u_int win =
2003 					    min(tp->snd_wnd, tp->snd_cwnd) /
2004 					    2 /	tp->t_segsz;
2005 					if (tcp_do_newreno && SEQ_LT(th->th_ack,
2006 					    tp->snd_recover)) {
2007 						/*
2008 						 * False fast retransmit after
2009 						 * timeout.  Do not cut window.
2010 						 */
2011 						tp->snd_cwnd += tp->t_segsz;
2012 						tp->t_dupacks = 0;
2013 						(void) tcp_output(tp);
2014 						goto drop;
2015 					}
2016 
2017 					if (win < 2)
2018 						win = 2;
2019 					tp->snd_ssthresh = win * tp->t_segsz;
2020 					tp->snd_recover = tp->snd_max;
2021 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
2022 					tp->t_rtttime = 0;
2023 					tp->snd_nxt = th->th_ack;
2024 					tp->snd_cwnd = tp->t_segsz;
2025 					(void) tcp_output(tp);
2026 					tp->snd_cwnd = tp->snd_ssthresh +
2027 					       tp->t_segsz * tp->t_dupacks;
2028 					if (SEQ_GT(onxt, tp->snd_nxt))
2029 						tp->snd_nxt = onxt;
2030 					goto drop;
2031 				} else if (tp->t_dupacks > tcprexmtthresh) {
2032 					tp->snd_cwnd += tp->t_segsz;
2033 					(void) tcp_output(tp);
2034 					goto drop;
2035 				}
2036 			} else
2037 				tp->t_dupacks = 0;
2038 			break;
2039 		}
2040 		/*
2041 		 * If the congestion window was inflated to account
2042 		 * for the other side's cached packets, retract it.
2043 		 */
2044 		if (tcp_do_newreno == 0) {
2045 			if (tp->t_dupacks >= tcprexmtthresh &&
2046 			    tp->snd_cwnd > tp->snd_ssthresh)
2047 				tp->snd_cwnd = tp->snd_ssthresh;
2048 			tp->t_dupacks = 0;
2049 		} else if (tp->t_dupacks >= tcprexmtthresh &&
2050 			   tcp_newreno(tp, th) == 0) {
2051 			tp->snd_cwnd = tp->snd_ssthresh;
2052 			/*
2053 			 * Window inflation should have left us with approx.
2054 			 * snd_ssthresh outstanding data.  But in case we
2055 			 * would be inclined to send a burst, better to do
2056 			 * it via the slow start mechanism.
2057 			 */
2058 			if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
2059 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
2060 				    + tp->t_segsz;
2061 			tp->t_dupacks = 0;
2062 		}
2063 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2064 			tcpstat.tcps_rcvacktoomuch++;
2065 			goto dropafterack;
2066 		}
2067 		acked = th->th_ack - tp->snd_una;
2068 		tcpstat.tcps_rcvackpack++;
2069 		tcpstat.tcps_rcvackbyte += acked;
2070 
2071 		/*
2072 		 * If we have a timestamp reply, update smoothed
2073 		 * round trip time.  If no timestamp is present but
2074 		 * transmit timer is running and timed sequence
2075 		 * number was acked, update smoothed round trip time.
2076 		 * Since we now have an rtt measurement, cancel the
2077 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2078 		 * Recompute the initial retransmit timer.
2079 		 */
2080 		if (opti.ts_present && opti.ts_ecr)
2081 			tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
2082 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2083 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2084 
2085 		/*
2086 		 * If all outstanding data is acked, stop retransmit
2087 		 * timer and remember to restart (more output or persist).
2088 		 * If there is more data to be acked, restart retransmit
2089 		 * timer, using current (possibly backed-off) value.
2090 		 */
2091 		if (th->th_ack == tp->snd_max) {
2092 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2093 			needoutput = 1;
2094 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2095 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2096 		/*
2097 		 * When new data is acked, open the congestion window.
2098 		 * If the window gives us less than ssthresh packets
2099 		 * in flight, open exponentially (segsz per packet).
2100 		 * Otherwise open linearly: segsz per window
2101 		 * (segsz^2 / cwnd per packet), plus a constant
2102 		 * fraction of a packet (segsz/8) to help larger windows
2103 		 * open quickly enough.
2104 		 */
2105 		{
2106 		u_int cw = tp->snd_cwnd;
2107 		u_int incr = tp->t_segsz;
2108 
2109 		if (cw > tp->snd_ssthresh)
2110 			incr = incr * incr / cw;
2111 		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
2112 			tp->snd_cwnd = min(cw + incr,
2113 			    TCP_MAXWIN << tp->snd_scale);
2114 		}
2115 		ND6_HINT(tp);
2116 		if (acked > so->so_snd.sb_cc) {
2117 			tp->snd_wnd -= so->so_snd.sb_cc;
2118 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2119 			ourfinisacked = 1;
2120 		} else {
2121 			int slen;
2122 
2123 			slen = tp->t_lastm->m_len;
2124 			sbdrop(&so->so_snd, acked);
2125 			tp->snd_wnd -= acked;
2126 			if (so->so_snd.sb_cc != 0) {
2127 				tp->t_lastoff -= acked;
2128 				if (tp->t_lastm->m_len != slen)
2129 					tp->t_inoff -=
2130 					    (slen - tp->t_lastm->m_len);
2131 			}
2132 			ourfinisacked = 0;
2133 		}
2134 		sowwakeup(so);
2135 		/*
2136 		 * We want snd_recover to track snd_una to
2137 		 * avoid sequence wraparound problems for
2138 		 * very large transfers.
2139 		 */
2140 		tp->snd_una = tp->snd_recover = th->th_ack;
2141 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2142 			tp->snd_nxt = tp->snd_una;
2143 
2144 		switch (tp->t_state) {
2145 
2146 		/*
2147 		 * In FIN_WAIT_1 STATE in addition to the processing
2148 		 * for the ESTABLISHED state if our FIN is now acknowledged
2149 		 * then enter FIN_WAIT_2.
2150 		 */
2151 		case TCPS_FIN_WAIT_1:
2152 			if (ourfinisacked) {
2153 				/*
2154 				 * If we can't receive any more
2155 				 * data, then closing user can proceed.
2156 				 * Starting the timer is contrary to the
2157 				 * specification, but if we don't get a FIN
2158 				 * we'll hang forever.
2159 				 */
2160 				if (so->so_state & SS_CANTRCVMORE) {
2161 					soisdisconnected(so);
2162 					if (tcp_maxidle > 0)
2163 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2164 						    tcp_maxidle);
2165 				}
2166 				tp->t_state = TCPS_FIN_WAIT_2;
2167 			}
2168 			break;
2169 
2170 	 	/*
2171 		 * In CLOSING STATE in addition to the processing for
2172 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2173 		 * then enter the TIME-WAIT state, otherwise ignore
2174 		 * the segment.
2175 		 */
2176 		case TCPS_CLOSING:
2177 			if (ourfinisacked) {
2178 				tp->t_state = TCPS_TIME_WAIT;
2179 				tcp_canceltimers(tp);
2180 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2181 				soisdisconnected(so);
2182 			}
2183 			break;
2184 
2185 		/*
2186 		 * In LAST_ACK, we may still be waiting for data to drain
2187 		 * and/or to be acked, as well as for the ack of our FIN.
2188 		 * If our FIN is now acknowledged, delete the TCB,
2189 		 * enter the closed state and return.
2190 		 */
2191 		case TCPS_LAST_ACK:
2192 			if (ourfinisacked) {
2193 				tp = tcp_close(tp);
2194 				goto drop;
2195 			}
2196 			break;
2197 
2198 		/*
2199 		 * In TIME_WAIT state the only thing that should arrive
2200 		 * is a retransmission of the remote FIN.  Acknowledge
2201 		 * it and restart the finack timer.
2202 		 */
2203 		case TCPS_TIME_WAIT:
2204 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2205 			goto dropafterack;
2206 		}
2207 	}
2208 
2209 step6:
2210 	/*
2211 	 * Update window information.
2212 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2213 	 */
2214 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2215 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2216 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2217 		/* keep track of pure window updates */
2218 		if (tlen == 0 &&
2219 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2220 			tcpstat.tcps_rcvwinupd++;
2221 		tp->snd_wnd = tiwin;
2222 		tp->snd_wl1 = th->th_seq;
2223 		tp->snd_wl2 = th->th_ack;
2224 		if (tp->snd_wnd > tp->max_sndwnd)
2225 			tp->max_sndwnd = tp->snd_wnd;
2226 		needoutput = 1;
2227 	}
2228 
2229 	/*
2230 	 * Process segments with URG.
2231 	 */
2232 	if ((tiflags & TH_URG) && th->th_urp &&
2233 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2234 		/*
2235 		 * This is a kludge, but if we receive and accept
2236 		 * random urgent pointers, we'll crash in
2237 		 * soreceive.  It's hard to imagine someone
2238 		 * actually wanting to send this much urgent data.
2239 		 */
2240 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2241 			th->th_urp = 0;			/* XXX */
2242 			tiflags &= ~TH_URG;		/* XXX */
2243 			goto dodata;			/* XXX */
2244 		}
2245 		/*
2246 		 * If this segment advances the known urgent pointer,
2247 		 * then mark the data stream.  This should not happen
2248 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2249 		 * a FIN has been received from the remote side.
2250 		 * In these states we ignore the URG.
2251 		 *
2252 		 * According to RFC961 (Assigned Protocols),
2253 		 * the urgent pointer points to the last octet
2254 		 * of urgent data.  We continue, however,
2255 		 * to consider it to indicate the first octet
2256 		 * of data past the urgent section as the original
2257 		 * spec states (in one of two places).
2258 		 */
2259 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2260 			tp->rcv_up = th->th_seq + th->th_urp;
2261 			so->so_oobmark = so->so_rcv.sb_cc +
2262 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2263 			if (so->so_oobmark == 0)
2264 				so->so_state |= SS_RCVATMARK;
2265 			sohasoutofband(so);
2266 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2267 		}
2268 		/*
2269 		 * Remove out of band data so doesn't get presented to user.
2270 		 * This can happen independent of advancing the URG pointer,
2271 		 * but if two URG's are pending at once, some out-of-band
2272 		 * data may creep in... ick.
2273 		 */
2274 		if (th->th_urp <= (u_int16_t) tlen
2275 #ifdef SO_OOBINLINE
2276 		     && (so->so_options & SO_OOBINLINE) == 0
2277 #endif
2278 		     )
2279 			tcp_pulloutofband(so, th, m, hdroptlen);
2280 	} else
2281 		/*
2282 		 * If no out of band data is expected,
2283 		 * pull receive urgent pointer along
2284 		 * with the receive window.
2285 		 */
2286 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2287 			tp->rcv_up = tp->rcv_nxt;
2288 dodata:							/* XXX */
2289 
2290 	/*
2291 	 * Process the segment text, merging it into the TCP sequencing queue,
2292 	 * and arranging for acknowledgement of receipt if necessary.
2293 	 * This process logically involves adjusting tp->rcv_wnd as data
2294 	 * is presented to the user (this happens in tcp_usrreq.c,
2295 	 * case PRU_RCVD).  If a FIN has already been received on this
2296 	 * connection then we just ignore the text.
2297 	 */
2298 	if ((tlen || (tiflags & TH_FIN)) &&
2299 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2300 		/*
2301 		 * Insert segment ti into reassembly queue of tcp with
2302 		 * control block tp.  Return TH_FIN if reassembly now includes
2303 		 * a segment with FIN.  The macro form does the common case
2304 		 * inline (segment is the next to be received on an
2305 		 * established connection, and the queue is empty),
2306 		 * avoiding linkage into and removal from the queue and
2307 		 * repetition of various conversions.
2308 		 * Set DELACK for segments received in order, but ack
2309 		 * immediately when segments are out of order
2310 		 * (so fast retransmit can work).
2311 		 */
2312 		/* NOTE: this was TCP_REASS() macro, but used only once */
2313 		TCP_REASS_LOCK(tp);
2314 		if (th->th_seq == tp->rcv_nxt &&
2315 		    TAILQ_FIRST(&tp->segq) == NULL &&
2316 		    tp->t_state == TCPS_ESTABLISHED) {
2317 			TCP_SETUP_ACK(tp, th);
2318 			tp->rcv_nxt += tlen;
2319 			tiflags = th->th_flags & TH_FIN;
2320 			tcpstat.tcps_rcvpack++;
2321 			tcpstat.tcps_rcvbyte += tlen;
2322 			ND6_HINT(tp);
2323 			if (so->so_state & SS_CANTRCVMORE)
2324 				m_freem(m);
2325 			else {
2326 				m_adj(m, hdroptlen);
2327 				sbappendstream(&(so)->so_rcv, m);
2328 			}
2329 			sorwakeup(so);
2330 		} else {
2331 			m_adj(m, hdroptlen);
2332 			tiflags = tcp_reass(tp, th, m, &tlen);
2333 			tp->t_flags |= TF_ACKNOW;
2334 		}
2335 		TCP_REASS_UNLOCK(tp);
2336 
2337 		/*
2338 		 * Note the amount of data that peer has sent into
2339 		 * our window, in order to estimate the sender's
2340 		 * buffer size.
2341 		 */
2342 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2343 	} else {
2344 		m_freem(m);
2345 		m = NULL;
2346 		tiflags &= ~TH_FIN;
2347 	}
2348 
2349 	/*
2350 	 * If FIN is received ACK the FIN and let the user know
2351 	 * that the connection is closing.  Ignore a FIN received before
2352 	 * the connection is fully established.
2353 	 */
2354 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2355 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2356 			socantrcvmore(so);
2357 			tp->t_flags |= TF_ACKNOW;
2358 			tp->rcv_nxt++;
2359 		}
2360 		switch (tp->t_state) {
2361 
2362 	 	/*
2363 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2364 		 */
2365 		case TCPS_ESTABLISHED:
2366 			tp->t_state = TCPS_CLOSE_WAIT;
2367 			break;
2368 
2369 	 	/*
2370 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2371 		 * enter the CLOSING state.
2372 		 */
2373 		case TCPS_FIN_WAIT_1:
2374 			tp->t_state = TCPS_CLOSING;
2375 			break;
2376 
2377 	 	/*
2378 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2379 		 * starting the time-wait timer, turning off the other
2380 		 * standard timers.
2381 		 */
2382 		case TCPS_FIN_WAIT_2:
2383 			tp->t_state = TCPS_TIME_WAIT;
2384 			tcp_canceltimers(tp);
2385 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2386 			soisdisconnected(so);
2387 			break;
2388 
2389 		/*
2390 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2391 		 */
2392 		case TCPS_TIME_WAIT:
2393 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2394 			break;
2395 		}
2396 	}
2397 #ifdef TCP_DEBUG
2398 	if (so->so_options & SO_DEBUG)
2399 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2400 #endif
2401 
2402 	/*
2403 	 * Return any desired output.
2404 	 */
2405 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2406 		(void) tcp_output(tp);
2407 	if (tcp_saveti)
2408 		m_freem(tcp_saveti);
2409 	return;
2410 
2411 badsyn:
2412 	/*
2413 	 * Received a bad SYN.  Increment counters and dropwithreset.
2414 	 */
2415 	tcpstat.tcps_badsyn++;
2416 	tp = NULL;
2417 	goto dropwithreset;
2418 
2419 dropafterack:
2420 	/*
2421 	 * Generate an ACK dropping incoming segment if it occupies
2422 	 * sequence space, where the ACK reflects our state.
2423 	 */
2424 	if (tiflags & TH_RST)
2425 		goto drop;
2426 	m_freem(m);
2427 	tp->t_flags |= TF_ACKNOW;
2428 	(void) tcp_output(tp);
2429 	if (tcp_saveti)
2430 		m_freem(tcp_saveti);
2431 	return;
2432 
2433 dropwithreset_ratelim:
2434 	/*
2435 	 * We may want to rate-limit RSTs in certain situations,
2436 	 * particularly if we are sending an RST in response to
2437 	 * an attempt to connect to or otherwise communicate with
2438 	 * a port for which we have no socket.
2439 	 */
2440 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2441 	    tcp_rst_ppslim) == 0) {
2442 		/* XXX stat */
2443 		goto drop;
2444 	}
2445 	/* ...fall into dropwithreset... */
2446 
2447 dropwithreset:
2448 	/*
2449 	 * Generate a RST, dropping incoming segment.
2450 	 * Make ACK acceptable to originator of segment.
2451 	 */
2452 	if (tiflags & TH_RST)
2453 		goto drop;
2454 
2455 	switch (af) {
2456 #ifdef INET6
2457 	case AF_INET6:
2458 		/* For following calls to tcp_respond */
2459 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2460 			goto drop;
2461 		break;
2462 #endif /* INET6 */
2463 	case AF_INET:
2464 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2465 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2466 			goto drop;
2467 	}
2468 
2469 	if (tiflags & TH_ACK)
2470 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2471 	else {
2472 		if (tiflags & TH_SYN)
2473 			tlen++;
2474 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2475 		    TH_RST|TH_ACK);
2476 	}
2477 	if (tcp_saveti)
2478 		m_freem(tcp_saveti);
2479 	return;
2480 
2481 badcsum:
2482 	tcpstat.tcps_rcvbadsum++;
2483 drop:
2484 	/*
2485 	 * Drop space held by incoming segment and return.
2486 	 */
2487 	if (tp) {
2488 		if (tp->t_inpcb)
2489 			so = tp->t_inpcb->inp_socket;
2490 #ifdef INET6
2491 		else if (tp->t_in6pcb)
2492 			so = tp->t_in6pcb->in6p_socket;
2493 #endif
2494 		else
2495 			so = NULL;
2496 #ifdef TCP_DEBUG
2497 		if (so && (so->so_options & SO_DEBUG) != 0)
2498 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2499 #endif
2500 	}
2501 	if (tcp_saveti)
2502 		m_freem(tcp_saveti);
2503 	m_freem(m);
2504 	return;
2505 }
2506 
2507 void
2508 tcp_dooptions(tp, cp, cnt, th, oi)
2509 	struct tcpcb *tp;
2510 	u_char *cp;
2511 	int cnt;
2512 	struct tcphdr *th;
2513 	struct tcp_opt_info *oi;
2514 {
2515 	u_int16_t mss;
2516 	int opt, optlen;
2517 
2518 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2519 		opt = cp[0];
2520 		if (opt == TCPOPT_EOL)
2521 			break;
2522 		if (opt == TCPOPT_NOP)
2523 			optlen = 1;
2524 		else {
2525 			if (cnt < 2)
2526 				break;
2527 			optlen = cp[1];
2528 			if (optlen < 2 || optlen > cnt)
2529 				break;
2530 		}
2531 		switch (opt) {
2532 
2533 		default:
2534 			continue;
2535 
2536 		case TCPOPT_MAXSEG:
2537 			if (optlen != TCPOLEN_MAXSEG)
2538 				continue;
2539 			if (!(th->th_flags & TH_SYN))
2540 				continue;
2541 			bcopy(cp + 2, &mss, sizeof(mss));
2542 			oi->maxseg = ntohs(mss);
2543 			break;
2544 
2545 		case TCPOPT_WINDOW:
2546 			if (optlen != TCPOLEN_WINDOW)
2547 				continue;
2548 			if (!(th->th_flags & TH_SYN))
2549 				continue;
2550 			tp->t_flags |= TF_RCVD_SCALE;
2551 			tp->requested_s_scale = cp[2];
2552 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2553 #if 0	/*XXX*/
2554 				char *p;
2555 
2556 				if (ip)
2557 					p = ntohl(ip->ip_src);
2558 #ifdef INET6
2559 				else if (ip6)
2560 					p = ip6_sprintf(&ip6->ip6_src);
2561 #endif
2562 				else
2563 					p = "(unknown)";
2564 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2565 				    "assuming %d\n",
2566 				    tp->requested_s_scale, p,
2567 				    TCP_MAX_WINSHIFT);
2568 #else
2569 				log(LOG_ERR, "TCP: invalid wscale %d, "
2570 				    "assuming %d\n",
2571 				    tp->requested_s_scale,
2572 				    TCP_MAX_WINSHIFT);
2573 #endif
2574 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2575 			}
2576 			break;
2577 
2578 		case TCPOPT_TIMESTAMP:
2579 			if (optlen != TCPOLEN_TIMESTAMP)
2580 				continue;
2581 			oi->ts_present = 1;
2582 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2583 			NTOHL(oi->ts_val);
2584 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2585 			NTOHL(oi->ts_ecr);
2586 
2587 			/*
2588 			 * A timestamp received in a SYN makes
2589 			 * it ok to send timestamp requests and replies.
2590 			 */
2591 			if (th->th_flags & TH_SYN) {
2592 				tp->t_flags |= TF_RCVD_TSTMP;
2593 				tp->ts_recent = oi->ts_val;
2594 				tp->ts_recent_age = TCP_TIMESTAMP(tp);
2595 			}
2596 			break;
2597 		case TCPOPT_SACK_PERMITTED:
2598 			if (optlen != TCPOLEN_SACK_PERMITTED)
2599 				continue;
2600 			if (!(th->th_flags & TH_SYN))
2601 				continue;
2602 			tp->t_flags &= ~TF_CANT_TXSACK;
2603 			break;
2604 
2605 		case TCPOPT_SACK:
2606 			if (tp->t_flags & TF_IGNR_RXSACK)
2607 				continue;
2608 			if (optlen % 8 != 2 || optlen < 10)
2609 				continue;
2610 			cp += 2;
2611 			optlen -= 2;
2612 			for (; optlen > 0; cp -= 8, optlen -= 8) {
2613 				tcp_seq lwe, rwe;
2614 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
2615 				NTOHL(lwe);
2616 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
2617 				NTOHL(rwe);
2618 				/* tcp_mark_sacked(tp, lwe, rwe); */
2619 			}
2620 			break;
2621 		}
2622 	}
2623 }
2624 
2625 /*
2626  * Pull out of band byte out of a segment so
2627  * it doesn't appear in the user's data queue.
2628  * It is still reflected in the segment length for
2629  * sequencing purposes.
2630  */
2631 void
2632 tcp_pulloutofband(so, th, m, off)
2633 	struct socket *so;
2634 	struct tcphdr *th;
2635 	struct mbuf *m;
2636 	int off;
2637 {
2638 	int cnt = off + th->th_urp - 1;
2639 
2640 	while (cnt >= 0) {
2641 		if (m->m_len > cnt) {
2642 			char *cp = mtod(m, caddr_t) + cnt;
2643 			struct tcpcb *tp = sototcpcb(so);
2644 
2645 			tp->t_iobc = *cp;
2646 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2647 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2648 			m->m_len--;
2649 			return;
2650 		}
2651 		cnt -= m->m_len;
2652 		m = m->m_next;
2653 		if (m == 0)
2654 			break;
2655 	}
2656 	panic("tcp_pulloutofband");
2657 }
2658 
2659 /*
2660  * Collect new round-trip time estimate
2661  * and update averages and current timeout.
2662  */
2663 void
2664 tcp_xmit_timer(tp, rtt)
2665 	struct tcpcb *tp;
2666 	uint32_t rtt;
2667 {
2668 	int32_t delta;
2669 
2670 	tcpstat.tcps_rttupdated++;
2671 	if (tp->t_srtt != 0) {
2672 		/*
2673 		 * srtt is stored as fixed point with 3 bits after the
2674 		 * binary point (i.e., scaled by 8).  The following magic
2675 		 * is equivalent to the smoothing algorithm in rfc793 with
2676 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2677 		 * point).  Adjust rtt to origin 0.
2678 		 */
2679 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2680 		if ((tp->t_srtt += delta) <= 0)
2681 			tp->t_srtt = 1 << 2;
2682 		/*
2683 		 * We accumulate a smoothed rtt variance (actually, a
2684 		 * smoothed mean difference), then set the retransmit
2685 		 * timer to smoothed rtt + 4 times the smoothed variance.
2686 		 * rttvar is stored as fixed point with 2 bits after the
2687 		 * binary point (scaled by 4).  The following is
2688 		 * equivalent to rfc793 smoothing with an alpha of .75
2689 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2690 		 * rfc793's wired-in beta.
2691 		 */
2692 		if (delta < 0)
2693 			delta = -delta;
2694 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2695 		if ((tp->t_rttvar += delta) <= 0)
2696 			tp->t_rttvar = 1 << 2;
2697 	} else {
2698 		/*
2699 		 * No rtt measurement yet - use the unsmoothed rtt.
2700 		 * Set the variance to half the rtt (so our first
2701 		 * retransmit happens at 3*rtt).
2702 		 */
2703 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2704 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2705 	}
2706 	tp->t_rtttime = 0;
2707 	tp->t_rxtshift = 0;
2708 
2709 	/*
2710 	 * the retransmit should happen at rtt + 4 * rttvar.
2711 	 * Because of the way we do the smoothing, srtt and rttvar
2712 	 * will each average +1/2 tick of bias.  When we compute
2713 	 * the retransmit timer, we want 1/2 tick of rounding and
2714 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2715 	 * firing of the timer.  The bias will give us exactly the
2716 	 * 1.5 tick we need.  But, because the bias is
2717 	 * statistical, we have to test that we don't drop below
2718 	 * the minimum feasible timer (which is 2 ticks).
2719 	 */
2720 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2721 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2722 
2723 	/*
2724 	 * We received an ack for a packet that wasn't retransmitted;
2725 	 * it is probably safe to discard any error indications we've
2726 	 * received recently.  This isn't quite right, but close enough
2727 	 * for now (a route might have failed after we sent a segment,
2728 	 * and the return path might not be symmetrical).
2729 	 */
2730 	tp->t_softerror = 0;
2731 }
2732 
2733 /*
2734  * Checks for partial ack.  If partial ack arrives, force the retransmission
2735  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
2736  * 1.  By setting snd_nxt to th_ack, this forces retransmission timer to
2737  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
2738  */
2739 int
2740 tcp_newreno(tp, th)
2741 	struct tcpcb *tp;
2742 	struct tcphdr *th;
2743 {
2744 	tcp_seq onxt = tp->snd_nxt;
2745 	u_long ocwnd = tp->snd_cwnd;
2746 
2747 	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2748 		/*
2749 		 * snd_una has not yet been updated and the socket's send
2750 		 * buffer has not yet drained off the ACK'd data, so we
2751 		 * have to leave snd_una as it was to get the correct data
2752 		 * offset in tcp_output().
2753 		 */
2754 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2755 		tp->t_rtttime = 0;
2756 		tp->snd_nxt = th->th_ack;
2757 		/*
2758 		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
2759 		 * is not yet updated when we're called.
2760 		 */
2761 		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
2762 		(void) tcp_output(tp);
2763 		tp->snd_cwnd = ocwnd;
2764 		if (SEQ_GT(onxt, tp->snd_nxt))
2765 			tp->snd_nxt = onxt;
2766 		/*
2767 		 * Partial window deflation.  Relies on fact that tp->snd_una
2768 		 * not updated yet.
2769 		 */
2770 		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
2771 		return 1;
2772 	}
2773 	return 0;
2774 }
2775 
2776 
2777 /*
2778  * TCP compressed state engine.  Currently used to hold compressed
2779  * state for SYN_RECEIVED.
2780  */
2781 
2782 u_long	syn_cache_count;
2783 u_int32_t syn_hash1, syn_hash2;
2784 
2785 #define SYN_HASH(sa, sp, dp) \
2786 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
2787 				     ((u_int32_t)(sp)))^syn_hash2)))
2788 #ifndef INET6
2789 #define	SYN_HASHALL(hash, src, dst) \
2790 do {									\
2791 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
2792 		((struct sockaddr_in *)(src))->sin_port,		\
2793 		((struct sockaddr_in *)(dst))->sin_port);		\
2794 } while (/*CONSTCOND*/ 0)
2795 #else
2796 #define SYN_HASH6(sa, sp, dp) \
2797 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
2798 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
2799 	 & 0x7fffffff)
2800 
2801 #define SYN_HASHALL(hash, src, dst) \
2802 do {									\
2803 	switch ((src)->sa_family) {					\
2804 	case AF_INET:							\
2805 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
2806 			((struct sockaddr_in *)(src))->sin_port,	\
2807 			((struct sockaddr_in *)(dst))->sin_port);	\
2808 		break;							\
2809 	case AF_INET6:							\
2810 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
2811 			((struct sockaddr_in6 *)(src))->sin6_port,	\
2812 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
2813 		break;							\
2814 	default:							\
2815 		hash = 0;						\
2816 	}								\
2817 } while (/*CONSTCOND*/0)
2818 #endif /* INET6 */
2819 
2820 #define	SYN_CACHE_RM(sc)						\
2821 do {									\
2822 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
2823 	    (sc), sc_bucketq);						\
2824 	(sc)->sc_tp = NULL;						\
2825 	LIST_REMOVE((sc), sc_tpq);					\
2826 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
2827 	callout_stop(&(sc)->sc_timer);					\
2828 	syn_cache_count--;						\
2829 } while (/*CONSTCOND*/0)
2830 
2831 #define	SYN_CACHE_PUT(sc)						\
2832 do {									\
2833 	if ((sc)->sc_ipopts)						\
2834 		(void) m_free((sc)->sc_ipopts);				\
2835 	if ((sc)->sc_route4.ro_rt != NULL)				\
2836 		RTFREE((sc)->sc_route4.ro_rt);				\
2837 	if (callout_invoking(&(sc)->sc_timer))				\
2838 		(sc)->sc_flags |= SCF_DEAD;				\
2839 	else								\
2840 		pool_put(&syn_cache_pool, (sc));			\
2841 } while (/*CONSTCOND*/0)
2842 
2843 struct pool syn_cache_pool;
2844 
2845 /*
2846  * We don't estimate RTT with SYNs, so each packet starts with the default
2847  * RTT and each timer step has a fixed timeout value.
2848  */
2849 #define	SYN_CACHE_TIMER_ARM(sc)						\
2850 do {									\
2851 	TCPT_RANGESET((sc)->sc_rxtcur,					\
2852 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
2853 	    TCPTV_REXMTMAX);						\
2854 	callout_reset(&(sc)->sc_timer,					\
2855 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
2856 } while (/*CONSTCOND*/0)
2857 
2858 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
2859 
2860 void
2861 syn_cache_init()
2862 {
2863 	int i;
2864 
2865 	/* Initialize the hash buckets. */
2866 	for (i = 0; i < tcp_syn_cache_size; i++)
2867 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
2868 
2869 	/* Initialize the syn cache pool. */
2870 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
2871 	    "synpl", NULL);
2872 }
2873 
2874 void
2875 syn_cache_insert(sc, tp)
2876 	struct syn_cache *sc;
2877 	struct tcpcb *tp;
2878 {
2879 	struct syn_cache_head *scp;
2880 	struct syn_cache *sc2;
2881 	int s;
2882 
2883 	/*
2884 	 * If there are no entries in the hash table, reinitialize
2885 	 * the hash secrets.
2886 	 */
2887 	if (syn_cache_count == 0) {
2888 		struct timeval tv;
2889 		microtime(&tv);
2890 		syn_hash1 = arc4random() ^ (u_long)&sc;
2891 		syn_hash2 = arc4random() ^ tv.tv_usec;
2892 	}
2893 
2894 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
2895 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
2896 	scp = &tcp_syn_cache[sc->sc_bucketidx];
2897 
2898 	/*
2899 	 * Make sure that we don't overflow the per-bucket
2900 	 * limit or the total cache size limit.
2901 	 */
2902 	s = splsoftnet();
2903 	if (scp->sch_length >= tcp_syn_bucket_limit) {
2904 		tcpstat.tcps_sc_bucketoverflow++;
2905 		/*
2906 		 * The bucket is full.  Toss the oldest element in the
2907 		 * bucket.  This will be the first entry in the bucket.
2908 		 */
2909 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
2910 #ifdef DIAGNOSTIC
2911 		/*
2912 		 * This should never happen; we should always find an
2913 		 * entry in our bucket.
2914 		 */
2915 		if (sc2 == NULL)
2916 			panic("syn_cache_insert: bucketoverflow: impossible");
2917 #endif
2918 		SYN_CACHE_RM(sc2);
2919 		SYN_CACHE_PUT(sc2);
2920 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
2921 		struct syn_cache_head *scp2, *sce;
2922 
2923 		tcpstat.tcps_sc_overflowed++;
2924 		/*
2925 		 * The cache is full.  Toss the oldest entry in the
2926 		 * first non-empty bucket we can find.
2927 		 *
2928 		 * XXX We would really like to toss the oldest
2929 		 * entry in the cache, but we hope that this
2930 		 * condition doesn't happen very often.
2931 		 */
2932 		scp2 = scp;
2933 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
2934 			sce = &tcp_syn_cache[tcp_syn_cache_size];
2935 			for (++scp2; scp2 != scp; scp2++) {
2936 				if (scp2 >= sce)
2937 					scp2 = &tcp_syn_cache[0];
2938 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
2939 					break;
2940 			}
2941 #ifdef DIAGNOSTIC
2942 			/*
2943 			 * This should never happen; we should always find a
2944 			 * non-empty bucket.
2945 			 */
2946 			if (scp2 == scp)
2947 				panic("syn_cache_insert: cacheoverflow: "
2948 				    "impossible");
2949 #endif
2950 		}
2951 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
2952 		SYN_CACHE_RM(sc2);
2953 		SYN_CACHE_PUT(sc2);
2954 	}
2955 
2956 	/*
2957 	 * Initialize the entry's timer.
2958 	 */
2959 	sc->sc_rxttot = 0;
2960 	sc->sc_rxtshift = 0;
2961 	SYN_CACHE_TIMER_ARM(sc);
2962 
2963 	/* Link it from tcpcb entry */
2964 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
2965 
2966 	/* Put it into the bucket. */
2967 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
2968 	scp->sch_length++;
2969 	syn_cache_count++;
2970 
2971 	tcpstat.tcps_sc_added++;
2972 	splx(s);
2973 }
2974 
2975 /*
2976  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
2977  * If we have retransmitted an entry the maximum number of times, expire
2978  * that entry.
2979  */
2980 void
2981 syn_cache_timer(void *arg)
2982 {
2983 	struct syn_cache *sc = arg;
2984 	int s;
2985 
2986 	s = splsoftnet();
2987 	callout_ack(&sc->sc_timer);
2988 
2989 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
2990 		tcpstat.tcps_sc_delayed_free++;
2991 		pool_put(&syn_cache_pool, sc);
2992 		splx(s);
2993 		return;
2994 	}
2995 
2996 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
2997 		/* Drop it -- too many retransmissions. */
2998 		goto dropit;
2999 	}
3000 
3001 	/*
3002 	 * Compute the total amount of time this entry has
3003 	 * been on a queue.  If this entry has been on longer
3004 	 * than the keep alive timer would allow, expire it.
3005 	 */
3006 	sc->sc_rxttot += sc->sc_rxtcur;
3007 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
3008 		goto dropit;
3009 
3010 	tcpstat.tcps_sc_retransmitted++;
3011 	(void) syn_cache_respond(sc, NULL);
3012 
3013 	/* Advance the timer back-off. */
3014 	sc->sc_rxtshift++;
3015 	SYN_CACHE_TIMER_ARM(sc);
3016 
3017 	splx(s);
3018 	return;
3019 
3020  dropit:
3021 	tcpstat.tcps_sc_timed_out++;
3022 	SYN_CACHE_RM(sc);
3023 	SYN_CACHE_PUT(sc);
3024 	splx(s);
3025 }
3026 
3027 /*
3028  * Remove syn cache created by the specified tcb entry,
3029  * because this does not make sense to keep them
3030  * (if there's no tcb entry, syn cache entry will never be used)
3031  */
3032 void
3033 syn_cache_cleanup(tp)
3034 	struct tcpcb *tp;
3035 {
3036 	struct syn_cache *sc, *nsc;
3037 	int s;
3038 
3039 	s = splsoftnet();
3040 
3041 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3042 		nsc = LIST_NEXT(sc, sc_tpq);
3043 
3044 #ifdef DIAGNOSTIC
3045 		if (sc->sc_tp != tp)
3046 			panic("invalid sc_tp in syn_cache_cleanup");
3047 #endif
3048 		SYN_CACHE_RM(sc);
3049 		SYN_CACHE_PUT(sc);
3050 	}
3051 	/* just for safety */
3052 	LIST_INIT(&tp->t_sc);
3053 
3054 	splx(s);
3055 }
3056 
3057 /*
3058  * Find an entry in the syn cache.
3059  */
3060 struct syn_cache *
3061 syn_cache_lookup(src, dst, headp)
3062 	struct sockaddr *src;
3063 	struct sockaddr *dst;
3064 	struct syn_cache_head **headp;
3065 {
3066 	struct syn_cache *sc;
3067 	struct syn_cache_head *scp;
3068 	u_int32_t hash;
3069 	int s;
3070 
3071 	SYN_HASHALL(hash, src, dst);
3072 
3073 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3074 	*headp = scp;
3075 	s = splsoftnet();
3076 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3077 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3078 		if (sc->sc_hash != hash)
3079 			continue;
3080 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3081 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3082 			splx(s);
3083 			return (sc);
3084 		}
3085 	}
3086 	splx(s);
3087 	return (NULL);
3088 }
3089 
3090 /*
3091  * This function gets called when we receive an ACK for a
3092  * socket in the LISTEN state.  We look up the connection
3093  * in the syn cache, and if its there, we pull it out of
3094  * the cache and turn it into a full-blown connection in
3095  * the SYN-RECEIVED state.
3096  *
3097  * The return values may not be immediately obvious, and their effects
3098  * can be subtle, so here they are:
3099  *
3100  *	NULL	SYN was not found in cache; caller should drop the
3101  *		packet and send an RST.
3102  *
3103  *	-1	We were unable to create the new connection, and are
3104  *		aborting it.  An ACK,RST is being sent to the peer
3105  *		(unless we got screwey sequence numbners; see below),
3106  *		because the 3-way handshake has been completed.  Caller
3107  *		should not free the mbuf, since we may be using it.  If
3108  *		we are not, we will free it.
3109  *
3110  *	Otherwise, the return value is a pointer to the new socket
3111  *	associated with the connection.
3112  */
3113 struct socket *
3114 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3115 	struct sockaddr *src;
3116 	struct sockaddr *dst;
3117 	struct tcphdr *th;
3118 	unsigned int hlen, tlen;
3119 	struct socket *so;
3120 	struct mbuf *m;
3121 {
3122 	struct syn_cache *sc;
3123 	struct syn_cache_head *scp;
3124 	struct inpcb *inp = NULL;
3125 #ifdef INET6
3126 	struct in6pcb *in6p = NULL;
3127 #endif
3128 	struct tcpcb *tp = 0;
3129 	struct mbuf *am;
3130 	int s;
3131 	struct socket *oso;
3132 
3133 	s = splsoftnet();
3134 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3135 		splx(s);
3136 		return (NULL);
3137 	}
3138 
3139 	/*
3140 	 * Verify the sequence and ack numbers.  Try getting the correct
3141 	 * response again.
3142 	 */
3143 	if ((th->th_ack != sc->sc_iss + 1) ||
3144 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3145 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3146 		(void) syn_cache_respond(sc, m);
3147 		splx(s);
3148 		return ((struct socket *)(-1));
3149 	}
3150 
3151 	/* Remove this cache entry */
3152 	SYN_CACHE_RM(sc);
3153 	splx(s);
3154 
3155 	/*
3156 	 * Ok, create the full blown connection, and set things up
3157 	 * as they would have been set up if we had created the
3158 	 * connection when the SYN arrived.  If we can't create
3159 	 * the connection, abort it.
3160 	 */
3161 	/*
3162 	 * inp still has the OLD in_pcb stuff, set the
3163 	 * v6-related flags on the new guy, too.   This is
3164 	 * done particularly for the case where an AF_INET6
3165 	 * socket is bound only to a port, and a v4 connection
3166 	 * comes in on that port.
3167 	 * we also copy the flowinfo from the original pcb
3168 	 * to the new one.
3169 	 */
3170 	oso = so;
3171 	so = sonewconn(so, SS_ISCONNECTED);
3172 	if (so == NULL)
3173 		goto resetandabort;
3174 
3175 	switch (so->so_proto->pr_domain->dom_family) {
3176 #ifdef INET
3177 	case AF_INET:
3178 		inp = sotoinpcb(so);
3179 		break;
3180 #endif
3181 #ifdef INET6
3182 	case AF_INET6:
3183 		in6p = sotoin6pcb(so);
3184 		break;
3185 #endif
3186 	}
3187 	switch (src->sa_family) {
3188 #ifdef INET
3189 	case AF_INET:
3190 		if (inp) {
3191 			struct in_ifaddr *ia;
3192 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3193 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3194 			inp->inp_options = ip_srcroute();
3195 			INADDR_TO_IA(inp->inp_laddr, ia);
3196 			KASSERT(ia != NULL);
3197 			KASSERT(inp->inp_ia == NULL);
3198 			inp->inp_ia = ia;
3199 			LIST_INSERT_HEAD(&ia->ia_inpcbs, inp, inp_ialink);
3200 			IFAREF(&ia->ia_ifa);
3201 			in_pcbstate(inp, INP_BOUND);
3202 			if (inp->inp_options == NULL) {
3203 				inp->inp_options = sc->sc_ipopts;
3204 				sc->sc_ipopts = NULL;
3205 			}
3206 		}
3207 #ifdef INET6
3208 		else if (in6p) {
3209 			/* IPv4 packet to AF_INET6 socket */
3210 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3211 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3212 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3213 				&in6p->in6p_laddr.s6_addr32[3],
3214 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3215 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3216 			in6totcpcb(in6p)->t_family = AF_INET;
3217 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3218 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3219 			else
3220 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3221 			in6_pcbstate(in6p, IN6P_BOUND);
3222 		}
3223 #endif
3224 		break;
3225 #endif
3226 #ifdef INET6
3227 	case AF_INET6:
3228 		if (in6p) {
3229 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3230 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3231 			in6_pcbstate(in6p, IN6P_BOUND);
3232 		}
3233 		break;
3234 #endif
3235 	}
3236 #ifdef INET6
3237 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3238 		struct in6pcb *oin6p = sotoin6pcb(oso);
3239 		/* inherit socket options from the listening socket */
3240 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3241 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3242 			m_freem(in6p->in6p_options);
3243 			in6p->in6p_options = 0;
3244 		}
3245 		ip6_savecontrol(in6p, &in6p->in6p_options,
3246 			mtod(m, struct ip6_hdr *), m);
3247 	}
3248 #endif
3249 
3250 #if defined(IPSEC) || defined(FAST_IPSEC)
3251 	/*
3252 	 * we make a copy of policy, instead of sharing the policy,
3253 	 * for better behavior in terms of SA lookup and dead SA removal.
3254 	 */
3255 	if (inp) {
3256 		/* copy old policy into new socket's */
3257 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3258 			printf("tcp_input: could not copy policy\n");
3259 	}
3260 #ifdef INET6
3261 	else if (in6p) {
3262 		/* copy old policy into new socket's */
3263 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3264 		    in6p->in6p_sp))
3265 			printf("tcp_input: could not copy policy\n");
3266 	}
3267 #endif
3268 #endif
3269 
3270 	/*
3271 	 * Give the new socket our cached route reference.
3272 	 */
3273 	if (inp)
3274 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3275 #ifdef INET6
3276 	else
3277 		in6p->in6p_route = sc->sc_route6;
3278 #endif
3279 	sc->sc_route4.ro_rt = NULL;
3280 
3281 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3282 	if (am == NULL)
3283 		goto resetandabort;
3284 	MCLAIM(am, &tcp_mowner);
3285 	am->m_len = src->sa_len;
3286 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3287 	if (inp) {
3288 		if (in_pcbconnect(inp, am)) {
3289 			(void) m_free(am);
3290 			goto resetandabort;
3291 		}
3292 	}
3293 #ifdef INET6
3294 	else if (in6p) {
3295 		if (src->sa_family == AF_INET) {
3296 			/* IPv4 packet to AF_INET6 socket */
3297 			struct sockaddr_in6 *sin6;
3298 			sin6 = mtod(am, struct sockaddr_in6 *);
3299 			am->m_len = sizeof(*sin6);
3300 			bzero(sin6, sizeof(*sin6));
3301 			sin6->sin6_family = AF_INET6;
3302 			sin6->sin6_len = sizeof(*sin6);
3303 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3304 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3305 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3306 				&sin6->sin6_addr.s6_addr32[3],
3307 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3308 		}
3309 		if (in6_pcbconnect(in6p, am)) {
3310 			(void) m_free(am);
3311 			goto resetandabort;
3312 		}
3313 	}
3314 #endif
3315 	else {
3316 		(void) m_free(am);
3317 		goto resetandabort;
3318 	}
3319 	(void) m_free(am);
3320 
3321 	if (inp)
3322 		tp = intotcpcb(inp);
3323 #ifdef INET6
3324 	else if (in6p)
3325 		tp = in6totcpcb(in6p);
3326 #endif
3327 	else
3328 		tp = NULL;
3329 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3330 	if (sc->sc_request_r_scale != 15) {
3331 		tp->requested_s_scale = sc->sc_requested_s_scale;
3332 		tp->request_r_scale = sc->sc_request_r_scale;
3333 		tp->snd_scale = sc->sc_requested_s_scale;
3334 		tp->rcv_scale = sc->sc_request_r_scale;
3335 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3336 	}
3337 	if (sc->sc_flags & SCF_TIMESTAMP)
3338 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3339 	tp->ts_timebase = sc->sc_timebase;
3340 
3341 	tp->t_template = tcp_template(tp);
3342 	if (tp->t_template == 0) {
3343 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3344 		so = NULL;
3345 		m_freem(m);
3346 		goto abort;
3347 	}
3348 
3349 	tp->iss = sc->sc_iss;
3350 	tp->irs = sc->sc_irs;
3351 	tcp_sendseqinit(tp);
3352 	tcp_rcvseqinit(tp);
3353 	tp->t_state = TCPS_SYN_RECEIVED;
3354 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3355 	tcpstat.tcps_accepts++;
3356 
3357 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3358 	tp->t_ourmss = sc->sc_ourmaxseg;
3359 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3360 
3361 	/*
3362 	 * Initialize the initial congestion window.  If we
3363 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3364 	 * to 1 segment (i.e. the Loss Window).
3365 	 */
3366 	if (sc->sc_rxtshift)
3367 		tp->snd_cwnd = tp->t_peermss;
3368 	else {
3369 		int ss = tcp_init_win;
3370 #ifdef INET
3371 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3372 			ss = tcp_init_win_local;
3373 #endif
3374 #ifdef INET6
3375 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3376 			ss = tcp_init_win_local;
3377 #endif
3378 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3379 	}
3380 
3381 	tcp_rmx_rtt(tp);
3382 	tp->snd_wl1 = sc->sc_irs;
3383 	tp->rcv_up = sc->sc_irs + 1;
3384 
3385 	/*
3386 	 * This is what whould have happened in tcp_output() when
3387 	 * the SYN,ACK was sent.
3388 	 */
3389 	tp->snd_up = tp->snd_una;
3390 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3391 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3392 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3393 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3394 	tp->last_ack_sent = tp->rcv_nxt;
3395 
3396 	tcpstat.tcps_sc_completed++;
3397 	SYN_CACHE_PUT(sc);
3398 	return (so);
3399 
3400 resetandabort:
3401 	(void) tcp_respond(NULL, m, m, th,
3402 			   th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
3403 abort:
3404 	if (so != NULL)
3405 		(void) soabort(so);
3406 	SYN_CACHE_PUT(sc);
3407 	tcpstat.tcps_sc_aborted++;
3408 	return ((struct socket *)(-1));
3409 }
3410 
3411 /*
3412  * This function is called when we get a RST for a
3413  * non-existent connection, so that we can see if the
3414  * connection is in the syn cache.  If it is, zap it.
3415  */
3416 
3417 void
3418 syn_cache_reset(src, dst, th)
3419 	struct sockaddr *src;
3420 	struct sockaddr *dst;
3421 	struct tcphdr *th;
3422 {
3423 	struct syn_cache *sc;
3424 	struct syn_cache_head *scp;
3425 	int s = splsoftnet();
3426 
3427 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3428 		splx(s);
3429 		return;
3430 	}
3431 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3432 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3433 		splx(s);
3434 		return;
3435 	}
3436 	SYN_CACHE_RM(sc);
3437 	splx(s);
3438 	tcpstat.tcps_sc_reset++;
3439 	SYN_CACHE_PUT(sc);
3440 }
3441 
3442 void
3443 syn_cache_unreach(src, dst, th)
3444 	struct sockaddr *src;
3445 	struct sockaddr *dst;
3446 	struct tcphdr *th;
3447 {
3448 	struct syn_cache *sc;
3449 	struct syn_cache_head *scp;
3450 	int s;
3451 
3452 	s = splsoftnet();
3453 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3454 		splx(s);
3455 		return;
3456 	}
3457 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3458 	if (ntohl (th->th_seq) != sc->sc_iss) {
3459 		splx(s);
3460 		return;
3461 	}
3462 
3463 	/*
3464 	 * If we've rertransmitted 3 times and this is our second error,
3465 	 * we remove the entry.  Otherwise, we allow it to continue on.
3466 	 * This prevents us from incorrectly nuking an entry during a
3467 	 * spurious network outage.
3468 	 *
3469 	 * See tcp_notify().
3470 	 */
3471 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3472 		sc->sc_flags |= SCF_UNREACH;
3473 		splx(s);
3474 		return;
3475 	}
3476 
3477 	SYN_CACHE_RM(sc);
3478 	splx(s);
3479 	tcpstat.tcps_sc_unreach++;
3480 	SYN_CACHE_PUT(sc);
3481 }
3482 
3483 /*
3484  * Given a LISTEN socket and an inbound SYN request, add
3485  * this to the syn cache, and send back a segment:
3486  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3487  * to the source.
3488  *
3489  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3490  * Doing so would require that we hold onto the data and deliver it
3491  * to the application.  However, if we are the target of a SYN-flood
3492  * DoS attack, an attacker could send data which would eventually
3493  * consume all available buffer space if it were ACKed.  By not ACKing
3494  * the data, we avoid this DoS scenario.
3495  */
3496 
3497 int
3498 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi)
3499 	struct sockaddr *src;
3500 	struct sockaddr *dst;
3501 	struct tcphdr *th;
3502 	unsigned int hlen;
3503 	struct socket *so;
3504 	struct mbuf *m;
3505 	u_char *optp;
3506 	int optlen;
3507 	struct tcp_opt_info *oi;
3508 {
3509 	struct tcpcb tb, *tp;
3510 	long win;
3511 	struct syn_cache *sc;
3512 	struct syn_cache_head *scp;
3513 	struct mbuf *ipopts;
3514 
3515 	tp = sototcpcb(so);
3516 
3517 	/*
3518 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3519 	 *
3520 	 * Note this check is performed in tcp_input() very early on.
3521 	 */
3522 
3523 	/*
3524 	 * Initialize some local state.
3525 	 */
3526 	win = sbspace(&so->so_rcv);
3527 	if (win > TCP_MAXWIN)
3528 		win = TCP_MAXWIN;
3529 
3530 	switch (src->sa_family) {
3531 #ifdef INET
3532 	case AF_INET:
3533 		/*
3534 		 * Remember the IP options, if any.
3535 		 */
3536 		ipopts = ip_srcroute();
3537 		break;
3538 #endif
3539 	default:
3540 		ipopts = NULL;
3541 	}
3542 
3543 	if (optp) {
3544 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3545 		tcp_dooptions(&tb, optp, optlen, th, oi);
3546 	} else
3547 		tb.t_flags = 0;
3548 
3549 	/*
3550 	 * See if we already have an entry for this connection.
3551 	 * If we do, resend the SYN,ACK.  We do not count this
3552 	 * as a retransmission (XXX though maybe we should).
3553 	 */
3554 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3555 		tcpstat.tcps_sc_dupesyn++;
3556 		if (ipopts) {
3557 			/*
3558 			 * If we were remembering a previous source route,
3559 			 * forget it and use the new one we've been given.
3560 			 */
3561 			if (sc->sc_ipopts)
3562 				(void) m_free(sc->sc_ipopts);
3563 			sc->sc_ipopts = ipopts;
3564 		}
3565 		sc->sc_timestamp = tb.ts_recent;
3566 		if (syn_cache_respond(sc, m) == 0) {
3567 			tcpstat.tcps_sndacks++;
3568 			tcpstat.tcps_sndtotal++;
3569 		}
3570 		return (1);
3571 	}
3572 
3573 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3574 	if (sc == NULL) {
3575 		if (ipopts)
3576 			(void) m_free(ipopts);
3577 		return (0);
3578 	}
3579 
3580 	/*
3581 	 * Fill in the cache, and put the necessary IP and TCP
3582 	 * options into the reply.
3583 	 */
3584 	bzero(sc, sizeof(struct syn_cache));
3585 	callout_init(&sc->sc_timer);
3586 	bcopy(src, &sc->sc_src, src->sa_len);
3587 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3588 	sc->sc_flags = 0;
3589 	sc->sc_ipopts = ipopts;
3590 	sc->sc_irs = th->th_seq;
3591 	switch (src->sa_family) {
3592 #ifdef INET
3593 	case AF_INET:
3594 	    {
3595 		struct sockaddr_in *srcin = (void *) src;
3596 		struct sockaddr_in *dstin = (void *) dst;
3597 
3598 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3599 		    &srcin->sin_addr, dstin->sin_port,
3600 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3601 		break;
3602 	    }
3603 #endif /* INET */
3604 #ifdef INET6
3605 	case AF_INET6:
3606 	    {
3607 		struct sockaddr_in6 *srcin6 = (void *) src;
3608 		struct sockaddr_in6 *dstin6 = (void *) dst;
3609 
3610 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3611 		    &srcin6->sin6_addr, dstin6->sin6_port,
3612 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3613 		break;
3614 	    }
3615 #endif /* INET6 */
3616 	}
3617 	sc->sc_peermaxseg = oi->maxseg;
3618 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
3619 						m->m_pkthdr.rcvif : NULL,
3620 						sc->sc_src.sa.sa_family);
3621 	sc->sc_win = win;
3622 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
3623 	sc->sc_timestamp = tb.ts_recent;
3624 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
3625 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
3626 		sc->sc_flags |= SCF_TIMESTAMP;
3627 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
3628 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
3629 		sc->sc_requested_s_scale = tb.requested_s_scale;
3630 		sc->sc_request_r_scale = 0;
3631 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
3632 		    TCP_MAXWIN << sc->sc_request_r_scale <
3633 		    so->so_rcv.sb_hiwat)
3634 			sc->sc_request_r_scale++;
3635 	} else {
3636 		sc->sc_requested_s_scale = 15;
3637 		sc->sc_request_r_scale = 15;
3638 	}
3639 	sc->sc_tp = tp;
3640 	if (syn_cache_respond(sc, m) == 0) {
3641 		syn_cache_insert(sc, tp);
3642 		tcpstat.tcps_sndacks++;
3643 		tcpstat.tcps_sndtotal++;
3644 	} else {
3645 		SYN_CACHE_PUT(sc);
3646 		tcpstat.tcps_sc_dropped++;
3647 	}
3648 	return (1);
3649 }
3650 
3651 int
3652 syn_cache_respond(sc, m)
3653 	struct syn_cache *sc;
3654 	struct mbuf *m;
3655 {
3656 	struct route *ro;
3657 	u_int8_t *optp;
3658 	int optlen, error;
3659 	u_int16_t tlen;
3660 	struct ip *ip = NULL;
3661 #ifdef INET6
3662 	struct ip6_hdr *ip6 = NULL;
3663 #endif
3664 	struct tcpcb *tp;
3665 	struct tcphdr *th;
3666 	u_int hlen;
3667 	struct socket *so;
3668 
3669 	switch (sc->sc_src.sa.sa_family) {
3670 	case AF_INET:
3671 		hlen = sizeof(struct ip);
3672 		ro = &sc->sc_route4;
3673 		break;
3674 #ifdef INET6
3675 	case AF_INET6:
3676 		hlen = sizeof(struct ip6_hdr);
3677 		ro = (struct route *)&sc->sc_route6;
3678 		break;
3679 #endif
3680 	default:
3681 		if (m)
3682 			m_freem(m);
3683 		return EAFNOSUPPORT;
3684 	}
3685 
3686 	/* Compute the size of the TCP options. */
3687 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
3688 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
3689 
3690 	tlen = hlen + sizeof(struct tcphdr) + optlen;
3691 
3692 	/*
3693 	 * Create the IP+TCP header from scratch.
3694 	 */
3695 	if (m)
3696 		m_freem(m);
3697 #ifdef DIAGNOSTIC
3698 	if (max_linkhdr + tlen > MCLBYTES)
3699 		return (ENOBUFS);
3700 #endif
3701 	MGETHDR(m, M_DONTWAIT, MT_DATA);
3702 	if (m && tlen > MHLEN) {
3703 		MCLGET(m, M_DONTWAIT);
3704 		if ((m->m_flags & M_EXT) == 0) {
3705 			m_freem(m);
3706 			m = NULL;
3707 		}
3708 	}
3709 	if (m == NULL)
3710 		return (ENOBUFS);
3711 	MCLAIM(m, &tcp_tx_mowner);
3712 
3713 	/* Fixup the mbuf. */
3714 	m->m_data += max_linkhdr;
3715 	m->m_len = m->m_pkthdr.len = tlen;
3716 	if (sc->sc_tp) {
3717 		tp = sc->sc_tp;
3718 		if (tp->t_inpcb)
3719 			so = tp->t_inpcb->inp_socket;
3720 #ifdef INET6
3721 		else if (tp->t_in6pcb)
3722 			so = tp->t_in6pcb->in6p_socket;
3723 #endif
3724 		else
3725 			so = NULL;
3726 	} else
3727 		so = NULL;
3728 	m->m_pkthdr.rcvif = NULL;
3729 	memset(mtod(m, u_char *), 0, tlen);
3730 
3731 	switch (sc->sc_src.sa.sa_family) {
3732 	case AF_INET:
3733 		ip = mtod(m, struct ip *);
3734 		ip->ip_dst = sc->sc_src.sin.sin_addr;
3735 		ip->ip_src = sc->sc_dst.sin.sin_addr;
3736 		ip->ip_p = IPPROTO_TCP;
3737 		th = (struct tcphdr *)(ip + 1);
3738 		th->th_dport = sc->sc_src.sin.sin_port;
3739 		th->th_sport = sc->sc_dst.sin.sin_port;
3740 		break;
3741 #ifdef INET6
3742 	case AF_INET6:
3743 		ip6 = mtod(m, struct ip6_hdr *);
3744 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
3745 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
3746 		ip6->ip6_nxt = IPPROTO_TCP;
3747 		/* ip6_plen will be updated in ip6_output() */
3748 		th = (struct tcphdr *)(ip6 + 1);
3749 		th->th_dport = sc->sc_src.sin6.sin6_port;
3750 		th->th_sport = sc->sc_dst.sin6.sin6_port;
3751 		break;
3752 #endif
3753 	default:
3754 		th = NULL;
3755 	}
3756 
3757 	th->th_seq = htonl(sc->sc_iss);
3758 	th->th_ack = htonl(sc->sc_irs + 1);
3759 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
3760 	th->th_flags = TH_SYN|TH_ACK;
3761 	th->th_win = htons(sc->sc_win);
3762 	/* th_sum already 0 */
3763 	/* th_urp already 0 */
3764 
3765 	/* Tack on the TCP options. */
3766 	optp = (u_int8_t *)(th + 1);
3767 	*optp++ = TCPOPT_MAXSEG;
3768 	*optp++ = 4;
3769 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
3770 	*optp++ = sc->sc_ourmaxseg & 0xff;
3771 
3772 	if (sc->sc_request_r_scale != 15) {
3773 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
3774 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
3775 		    sc->sc_request_r_scale);
3776 		optp += 4;
3777 	}
3778 
3779 	if (sc->sc_flags & SCF_TIMESTAMP) {
3780 		u_int32_t *lp = (u_int32_t *)(optp);
3781 		/* Form timestamp option as shown in appendix A of RFC 1323. */
3782 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
3783 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
3784 		*lp   = htonl(sc->sc_timestamp);
3785 		optp += TCPOLEN_TSTAMP_APPA;
3786 	}
3787 
3788 	/* Compute the packet's checksum. */
3789 	switch (sc->sc_src.sa.sa_family) {
3790 	case AF_INET:
3791 		ip->ip_len = htons(tlen - hlen);
3792 		th->th_sum = 0;
3793 		th->th_sum = in_cksum(m, tlen);
3794 		break;
3795 #ifdef INET6
3796 	case AF_INET6:
3797 		ip6->ip6_plen = htons(tlen - hlen);
3798 		th->th_sum = 0;
3799 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
3800 		break;
3801 #endif
3802 	}
3803 
3804 	/*
3805 	 * Fill in some straggling IP bits.  Note the stack expects
3806 	 * ip_len to be in host order, for convenience.
3807 	 */
3808 	switch (sc->sc_src.sa.sa_family) {
3809 #ifdef INET
3810 	case AF_INET:
3811 		ip->ip_len = htons(tlen);
3812 		ip->ip_ttl = ip_defttl;
3813 		/* XXX tos? */
3814 		break;
3815 #endif
3816 #ifdef INET6
3817 	case AF_INET6:
3818 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3819 		ip6->ip6_vfc |= IPV6_VERSION;
3820 		ip6->ip6_plen = htons(tlen - hlen);
3821 		/* ip6_hlim will be initialized afterwards */
3822 		/* XXX flowlabel? */
3823 		break;
3824 #endif
3825 	}
3826 
3827 	/* XXX use IPsec policy on listening socket, on SYN ACK */
3828 	tp = sc->sc_tp;
3829 
3830 	switch (sc->sc_src.sa.sa_family) {
3831 #ifdef INET
3832 	case AF_INET:
3833 		error = ip_output(m, sc->sc_ipopts, ro,
3834 		    (ip_mtudisc ? IP_MTUDISC : 0),
3835 		    (struct ip_moptions *)NULL, so);
3836 		break;
3837 #endif
3838 #ifdef INET6
3839 	case AF_INET6:
3840 		ip6->ip6_hlim = in6_selecthlim(NULL,
3841 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
3842 
3843 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
3844 			(struct ip6_moptions *)0, so, NULL);
3845 		break;
3846 #endif
3847 	default:
3848 		error = EAFNOSUPPORT;
3849 		break;
3850 	}
3851 	return (error);
3852 }
3853