xref: /netbsd-src/sys/netinet/tcp_input.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: tcp_input.c,v 1.163 2003/03/01 04:40:27 thorpej Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  *      @(#)COPYRIGHT   1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  *      This product includes software developed by the University of
47  *      California, Berkeley and its contributors.
48  *      This product includes software developed at the Information
49  *      Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 /*-
73  * Copyright (c) 1997, 1998, 1999, 2001 The NetBSD Foundation, Inc.
74  * All rights reserved.
75  *
76  * This code is derived from software contributed to The NetBSD Foundation
77  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
78  * Facility, NASA Ames Research Center.
79  *
80  * Redistribution and use in source and binary forms, with or without
81  * modification, are permitted provided that the following conditions
82  * are met:
83  * 1. Redistributions of source code must retain the above copyright
84  *    notice, this list of conditions and the following disclaimer.
85  * 2. Redistributions in binary form must reproduce the above copyright
86  *    notice, this list of conditions and the following disclaimer in the
87  *    documentation and/or other materials provided with the distribution.
88  * 3. All advertising materials mentioning features or use of this software
89  *    must display the following acknowledgement:
90  *	This product includes software developed by the NetBSD
91  *	Foundation, Inc. and its contributors.
92  * 4. Neither the name of The NetBSD Foundation nor the names of its
93  *    contributors may be used to endorse or promote products derived
94  *    from this software without specific prior written permission.
95  *
96  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
97  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
98  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
99  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
100  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
101  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
102  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
103  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
104  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
105  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
106  * POSSIBILITY OF SUCH DAMAGE.
107  */
108 
109 /*
110  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
111  *	The Regents of the University of California.  All rights reserved.
112  *
113  * Redistribution and use in source and binary forms, with or without
114  * modification, are permitted provided that the following conditions
115  * are met:
116  * 1. Redistributions of source code must retain the above copyright
117  *    notice, this list of conditions and the following disclaimer.
118  * 2. Redistributions in binary form must reproduce the above copyright
119  *    notice, this list of conditions and the following disclaimer in the
120  *    documentation and/or other materials provided with the distribution.
121  * 3. All advertising materials mentioning features or use of this software
122  *    must display the following acknowledgement:
123  *	This product includes software developed by the University of
124  *	California, Berkeley and its contributors.
125  * 4. Neither the name of the University nor the names of its contributors
126  *    may be used to endorse or promote products derived from this software
127  *    without specific prior written permission.
128  *
129  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
130  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
131  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
132  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
133  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
134  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
135  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
136  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
137  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
138  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
139  * SUCH DAMAGE.
140  *
141  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
142  */
143 
144 /*
145  *	TODO list for SYN cache stuff:
146  *
147  *	Find room for a "state" field, which is needed to keep a
148  *	compressed state for TIME_WAIT TCBs.  It's been noted already
149  *	that this is fairly important for very high-volume web and
150  *	mail servers, which use a large number of short-lived
151  *	connections.
152  */
153 
154 #include <sys/cdefs.h>
155 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.163 2003/03/01 04:40:27 thorpej Exp $");
156 
157 #include "opt_inet.h"
158 #include "opt_ipsec.h"
159 #include "opt_inet_csum.h"
160 #include "opt_tcp_debug.h"
161 
162 #include <sys/param.h>
163 #include <sys/systm.h>
164 #include <sys/malloc.h>
165 #include <sys/mbuf.h>
166 #include <sys/protosw.h>
167 #include <sys/socket.h>
168 #include <sys/socketvar.h>
169 #include <sys/errno.h>
170 #include <sys/syslog.h>
171 #include <sys/pool.h>
172 #include <sys/domain.h>
173 #include <sys/kernel.h>
174 
175 #include <net/if.h>
176 #include <net/route.h>
177 #include <net/if_types.h>
178 
179 #include <netinet/in.h>
180 #include <netinet/in_systm.h>
181 #include <netinet/ip.h>
182 #include <netinet/in_pcb.h>
183 #include <netinet/ip_var.h>
184 
185 #ifdef INET6
186 #ifndef INET
187 #include <netinet/in.h>
188 #endif
189 #include <netinet/ip6.h>
190 #include <netinet6/ip6_var.h>
191 #include <netinet6/in6_pcb.h>
192 #include <netinet6/ip6_var.h>
193 #include <netinet6/in6_var.h>
194 #include <netinet/icmp6.h>
195 #include <netinet6/nd6.h>
196 #endif
197 
198 #ifdef PULLDOWN_TEST
199 #ifndef INET6
200 /* always need ip6.h for IP6_EXTHDR_GET */
201 #include <netinet/ip6.h>
202 #endif
203 #endif
204 
205 #include <netinet/tcp.h>
206 #include <netinet/tcp_fsm.h>
207 #include <netinet/tcp_seq.h>
208 #include <netinet/tcp_timer.h>
209 #include <netinet/tcp_var.h>
210 #include <netinet/tcpip.h>
211 #include <netinet/tcp_debug.h>
212 
213 #include <machine/stdarg.h>
214 
215 #ifdef IPSEC
216 #include <netinet6/ipsec.h>
217 #include <netkey/key.h>
218 #endif /*IPSEC*/
219 #ifdef INET6
220 #include "faith.h"
221 #if defined(NFAITH) && NFAITH > 0
222 #include <net/if_faith.h>
223 #endif
224 #endif
225 
226 int	tcprexmtthresh = 3;
227 int	tcp_log_refused;
228 
229 static int tcp_rst_ppslim_count = 0;
230 static struct timeval tcp_rst_ppslim_last;
231 
232 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
233 
234 /* for modulo comparisons of timestamps */
235 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
236 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
237 
238 /*
239  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
240  */
241 #ifdef INET6
242 #define ND6_HINT(tp) \
243 do { \
244 	if (tp && tp->t_in6pcb && tp->t_family == AF_INET6 \
245 	 && tp->t_in6pcb->in6p_route.ro_rt) { \
246 		nd6_nud_hint(tp->t_in6pcb->in6p_route.ro_rt, NULL, 0); \
247 	} \
248 } while (/*CONSTCOND*/ 0)
249 #else
250 #define ND6_HINT(tp)
251 #endif
252 
253 /*
254  * Macro to compute ACK transmission behavior.  Delay the ACK unless
255  * we have already delayed an ACK (must send an ACK every two segments).
256  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
257  * option is enabled.
258  */
259 #define	TCP_SETUP_ACK(tp, th) \
260 do { \
261 	if ((tp)->t_flags & TF_DELACK || \
262 	    (tcp_ack_on_push && (th)->th_flags & TH_PUSH)) \
263 		tp->t_flags |= TF_ACKNOW; \
264 	else \
265 		TCP_SET_DELACK(tp); \
266 } while (/*CONSTCOND*/ 0)
267 
268 /*
269  * Convert TCP protocol fields to host order for easier processing.
270  */
271 #define	TCP_FIELDS_TO_HOST(th)						\
272 do {									\
273 	NTOHL((th)->th_seq);						\
274 	NTOHL((th)->th_ack);						\
275 	NTOHS((th)->th_win);						\
276 	NTOHS((th)->th_urp);						\
277 } while (/*CONSTCOND*/ 0)
278 
279 /*
280  * ... and reverse the above.
281  */
282 #define	TCP_FIELDS_TO_NET(th)						\
283 do {									\
284 	HTONL((th)->th_seq);						\
285 	HTONL((th)->th_ack);						\
286 	HTONS((th)->th_win);						\
287 	HTONS((th)->th_urp);						\
288 } while (/*CONSTCOND*/ 0)
289 
290 #ifdef TCP_CSUM_COUNTERS
291 #include <sys/device.h>
292 
293 extern struct evcnt tcp_hwcsum_ok;
294 extern struct evcnt tcp_hwcsum_bad;
295 extern struct evcnt tcp_hwcsum_data;
296 extern struct evcnt tcp_swcsum;
297 
298 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
299 
300 #else
301 
302 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
303 
304 #endif /* TCP_CSUM_COUNTERS */
305 
306 #ifdef TCP_REASS_COUNTERS
307 #include <sys/device.h>
308 
309 extern struct evcnt tcp_reass_;
310 extern struct evcnt tcp_reass_empty;
311 extern struct evcnt tcp_reass_iteration[8];
312 extern struct evcnt tcp_reass_prependfirst;
313 extern struct evcnt tcp_reass_prepend;
314 extern struct evcnt tcp_reass_insert;
315 extern struct evcnt tcp_reass_inserttail;
316 extern struct evcnt tcp_reass_append;
317 extern struct evcnt tcp_reass_appendtail;
318 extern struct evcnt tcp_reass_overlaptail;
319 extern struct evcnt tcp_reass_overlapfront;
320 extern struct evcnt tcp_reass_segdup;
321 extern struct evcnt tcp_reass_fragdup;
322 
323 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
324 
325 #else
326 
327 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
328 
329 #endif /* TCP_REASS_COUNTERS */
330 
331 #ifdef INET
332 static void tcp4_log_refused __P((const struct ip *, const struct tcphdr *));
333 #endif
334 #ifdef INET6
335 static void tcp6_log_refused
336     __P((const struct ip6_hdr *, const struct tcphdr *));
337 #endif
338 
339 int
340 tcp_reass(tp, th, m, tlen)
341 	struct tcpcb *tp;
342 	struct tcphdr *th;
343 	struct mbuf *m;
344 	int *tlen;
345 {
346 	struct ipqent *p, *q, *nq, *tiqe = NULL;
347 	struct socket *so = NULL;
348 	int pkt_flags;
349 	tcp_seq pkt_seq;
350 	unsigned pkt_len;
351 	u_long rcvpartdupbyte = 0;
352 	u_long rcvoobyte;
353 #ifdef TCP_REASS_COUNTERS
354 	u_int count = 0;
355 #endif
356 
357 	if (tp->t_inpcb)
358 		so = tp->t_inpcb->inp_socket;
359 #ifdef INET6
360 	else if (tp->t_in6pcb)
361 		so = tp->t_in6pcb->in6p_socket;
362 #endif
363 
364 	TCP_REASS_LOCK_CHECK(tp);
365 
366 	/*
367 	 * Call with th==0 after become established to
368 	 * force pre-ESTABLISHED data up to user socket.
369 	 */
370 	if (th == 0)
371 		goto present;
372 
373 	rcvoobyte = *tlen;
374 	/*
375 	 * Copy these to local variables because the tcpiphdr
376 	 * gets munged while we are collapsing mbufs.
377 	 */
378 	pkt_seq = th->th_seq;
379 	pkt_len = *tlen;
380 	pkt_flags = th->th_flags;
381 
382 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
383 
384 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
385 		/*
386 		 * When we miss a packet, the vast majority of time we get
387 		 * packets that follow it in order.  So optimize for that.
388 		 */
389 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
390 			p->ipqe_len += pkt_len;
391 			p->ipqe_flags |= pkt_flags;
392 			m_cat(p->ipqe_m, m);
393 			tiqe = p;
394 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
395 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
396 			goto skip_replacement;
397 		}
398 		/*
399 		 * While we're here, if the pkt is completely beyond
400 		 * anything we have, just insert it at the tail.
401 		 */
402 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
403 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
404 			goto insert_it;
405 		}
406 	}
407 
408 	q = TAILQ_FIRST(&tp->segq);
409 
410 	if (q != NULL) {
411 		/*
412 		 * If this segment immediately precedes the first out-of-order
413 		 * block, simply slap the segment in front of it and (mostly)
414 		 * skip the complicated logic.
415 		 */
416 		if (pkt_seq + pkt_len == q->ipqe_seq) {
417 			q->ipqe_seq = pkt_seq;
418 			q->ipqe_len += pkt_len;
419 			q->ipqe_flags |= pkt_flags;
420 			m_cat(m, q->ipqe_m);
421 			q->ipqe_m = m;
422 			tiqe = q;
423 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
424 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
425 			goto skip_replacement;
426 		}
427 	} else {
428 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
429 	}
430 
431 	/*
432 	 * Find a segment which begins after this one does.
433 	 */
434 	for (p = NULL; q != NULL; q = nq) {
435 		nq = TAILQ_NEXT(q, ipqe_q);
436 #ifdef TCP_REASS_COUNTERS
437 		count++;
438 #endif
439 		/*
440 		 * If the received segment is just right after this
441 		 * fragment, merge the two together and then check
442 		 * for further overlaps.
443 		 */
444 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
445 #ifdef TCPREASS_DEBUG
446 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
447 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
448 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
449 #endif
450 			pkt_len += q->ipqe_len;
451 			pkt_flags |= q->ipqe_flags;
452 			pkt_seq = q->ipqe_seq;
453 			m_cat(q->ipqe_m, m);
454 			m = q->ipqe_m;
455 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
456 			goto free_ipqe;
457 		}
458 		/*
459 		 * If the received segment is completely past this
460 		 * fragment, we need to go the next fragment.
461 		 */
462 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
463 			p = q;
464 			continue;
465 		}
466 		/*
467 		 * If the fragment is past the received segment,
468 		 * it (or any following) can't be concatenated.
469 		 */
470 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
471 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
472 			break;
473 		}
474 
475 		/*
476 		 * We've received all the data in this segment before.
477 		 * mark it as a duplicate and return.
478 		 */
479 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
480 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
481 			tcpstat.tcps_rcvduppack++;
482 			tcpstat.tcps_rcvdupbyte += pkt_len;
483 			m_freem(m);
484 			if (tiqe != NULL)
485 				pool_put(&ipqent_pool, tiqe);
486 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
487 			return (0);
488 		}
489 		/*
490 		 * Received segment completely overlaps this fragment
491 		 * so we drop the fragment (this keeps the temporal
492 		 * ordering of segments correct).
493 		 */
494 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
495 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
496 			rcvpartdupbyte += q->ipqe_len;
497 			m_freem(q->ipqe_m);
498 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
499 			goto free_ipqe;
500 		}
501 		/*
502 		 * RX'ed segment extends past the end of the
503 		 * fragment.  Drop the overlapping bytes.  Then
504 		 * merge the fragment and segment then treat as
505 		 * a longer received packet.
506 		 */
507 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
508 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
509 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
510 #ifdef TCPREASS_DEBUG
511 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
512 			       tp, overlap,
513 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
514 #endif
515 			m_adj(m, overlap);
516 			rcvpartdupbyte += overlap;
517 			m_cat(q->ipqe_m, m);
518 			m = q->ipqe_m;
519 			pkt_seq = q->ipqe_seq;
520 			pkt_len += q->ipqe_len - overlap;
521 			rcvoobyte -= overlap;
522 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
523 			goto free_ipqe;
524 		}
525 		/*
526 		 * RX'ed segment extends past the front of the
527 		 * fragment.  Drop the overlapping bytes on the
528 		 * received packet.  The packet will then be
529 		 * contatentated with this fragment a bit later.
530 		 */
531 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
532 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
533 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
534 #ifdef TCPREASS_DEBUG
535 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
536 			       tp, overlap,
537 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
538 #endif
539 			m_adj(m, -overlap);
540 			pkt_len -= overlap;
541 			rcvpartdupbyte += overlap;
542 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
543 			rcvoobyte -= overlap;
544 		}
545 		/*
546 		 * If the received segment immediates precedes this
547 		 * fragment then tack the fragment onto this segment
548 		 * and reinsert the data.
549 		 */
550 		if (q->ipqe_seq == pkt_seq + pkt_len) {
551 #ifdef TCPREASS_DEBUG
552 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
553 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
554 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
555 #endif
556 			pkt_len += q->ipqe_len;
557 			pkt_flags |= q->ipqe_flags;
558 			m_cat(m, q->ipqe_m);
559 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
560 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
561 			if (tiqe == NULL) {
562 			    tiqe = q;
563 			} else {
564 			    pool_put(&ipqent_pool, q);
565 			}
566 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
567 			break;
568 		}
569 		/*
570 		 * If the fragment is before the segment, remember it.
571 		 * When this loop is terminated, p will contain the
572 		 * pointer to fragment that is right before the received
573 		 * segment.
574 		 */
575 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
576 			p = q;
577 
578 		continue;
579 
580 		/*
581 		 * This is a common operation.  It also will allow
582 		 * to save doing a malloc/free in most instances.
583 		 */
584 	  free_ipqe:
585 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
586 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
587 		if (tiqe == NULL) {
588 		    tiqe = q;
589 		} else {
590 		    pool_put(&ipqent_pool, q);
591 		}
592 	}
593 
594 #ifdef TCP_REASS_COUNTERS
595 	if (count > 7)
596 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
597 	else if (count > 0)
598 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
599 #endif
600 
601     insert_it:
602 
603 	/*
604 	 * Allocate a new queue entry since the received segment did not
605 	 * collapse onto any other out-of-order block; thus we are allocating
606 	 * a new block.  If it had collapsed, tiqe would not be NULL and
607 	 * we would be reusing it.
608 	 * XXX If we can't, just drop the packet.  XXX
609 	 */
610 	if (tiqe == NULL) {
611 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
612 		if (tiqe == NULL) {
613 			tcpstat.tcps_rcvmemdrop++;
614 			m_freem(m);
615 			return (0);
616 		}
617 	}
618 
619 	/*
620 	 * Update the counters.
621 	 */
622 	tcpstat.tcps_rcvoopack++;
623 	tcpstat.tcps_rcvoobyte += rcvoobyte;
624 	if (rcvpartdupbyte) {
625 	    tcpstat.tcps_rcvpartduppack++;
626 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
627 	}
628 
629 	/*
630 	 * Insert the new fragment queue entry into both queues.
631 	 */
632 	tiqe->ipqe_m = m;
633 	tiqe->ipqe_seq = pkt_seq;
634 	tiqe->ipqe_len = pkt_len;
635 	tiqe->ipqe_flags = pkt_flags;
636 	if (p == NULL) {
637 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
638 #ifdef TCPREASS_DEBUG
639 		if (tiqe->ipqe_seq != tp->rcv_nxt)
640 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
641 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
642 #endif
643 	} else {
644 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
645 #ifdef TCPREASS_DEBUG
646 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
647 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
648 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
649 #endif
650 	}
651 
652 skip_replacement:
653 
654 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
655 
656 present:
657 	/*
658 	 * Present data to user, advancing rcv_nxt through
659 	 * completed sequence space.
660 	 */
661 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
662 		return (0);
663 	q = TAILQ_FIRST(&tp->segq);
664 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
665 		return (0);
666 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
667 		return (0);
668 
669 	tp->rcv_nxt += q->ipqe_len;
670 	pkt_flags = q->ipqe_flags & TH_FIN;
671 	ND6_HINT(tp);
672 
673 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
674 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
675 	if (so->so_state & SS_CANTRCVMORE)
676 		m_freem(q->ipqe_m);
677 	else
678 		sbappendstream(&so->so_rcv, q->ipqe_m);
679 	pool_put(&ipqent_pool, q);
680 	sorwakeup(so);
681 	return (pkt_flags);
682 }
683 
684 #ifdef INET6
685 int
686 tcp6_input(mp, offp, proto)
687 	struct mbuf **mp;
688 	int *offp, proto;
689 {
690 	struct mbuf *m = *mp;
691 
692 	/*
693 	 * draft-itojun-ipv6-tcp-to-anycast
694 	 * better place to put this in?
695 	 */
696 	if (m->m_flags & M_ANYCAST6) {
697 		struct ip6_hdr *ip6;
698 		if (m->m_len < sizeof(struct ip6_hdr)) {
699 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
700 				tcpstat.tcps_rcvshort++;
701 				return IPPROTO_DONE;
702 			}
703 		}
704 		ip6 = mtod(m, struct ip6_hdr *);
705 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
706 		    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
707 		return IPPROTO_DONE;
708 	}
709 
710 	tcp_input(m, *offp, proto);
711 	return IPPROTO_DONE;
712 }
713 #endif
714 
715 #ifdef INET
716 static void
717 tcp4_log_refused(ip, th)
718 	const struct ip *ip;
719 	const struct tcphdr *th;
720 {
721 	char src[4*sizeof "123"];
722 	char dst[4*sizeof "123"];
723 
724 	if (ip) {
725 		strcpy(src, inet_ntoa(ip->ip_src));
726 		strcpy(dst, inet_ntoa(ip->ip_dst));
727 	}
728 	else {
729 		strcpy(src, "(unknown)");
730 		strcpy(dst, "(unknown)");
731 	}
732 	log(LOG_INFO,
733 	    "Connection attempt to TCP %s:%d from %s:%d\n",
734 	    dst, ntohs(th->th_dport),
735 	    src, ntohs(th->th_sport));
736 }
737 #endif
738 
739 #ifdef INET6
740 static void
741 tcp6_log_refused(ip6, th)
742 	const struct ip6_hdr *ip6;
743 	const struct tcphdr *th;
744 {
745 	char src[INET6_ADDRSTRLEN];
746 	char dst[INET6_ADDRSTRLEN];
747 
748 	if (ip6) {
749 		strcpy(src, ip6_sprintf(&ip6->ip6_src));
750 		strcpy(dst, ip6_sprintf(&ip6->ip6_dst));
751 	}
752 	else {
753 		strcpy(src, "(unknown v6)");
754 		strcpy(dst, "(unknown v6)");
755 	}
756 	log(LOG_INFO,
757 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
758 	    dst, ntohs(th->th_dport),
759 	    src, ntohs(th->th_sport));
760 }
761 #endif
762 
763 /*
764  * TCP input routine, follows pages 65-76 of the
765  * protocol specification dated September, 1981 very closely.
766  */
767 void
768 #if __STDC__
769 tcp_input(struct mbuf *m, ...)
770 #else
771 tcp_input(m, va_alist)
772 	struct mbuf *m;
773 #endif
774 {
775 	struct tcphdr *th;
776 	struct ip *ip;
777 	struct inpcb *inp;
778 #ifdef INET6
779 	struct ip6_hdr *ip6;
780 	struct in6pcb *in6p;
781 #endif
782 	u_int8_t *optp = NULL;
783 	int optlen = 0;
784 	int len, tlen, toff, hdroptlen = 0;
785 	struct tcpcb *tp = 0;
786 	int tiflags;
787 	struct socket *so = NULL;
788 	int todrop, acked, ourfinisacked, needoutput = 0;
789 #ifdef TCP_DEBUG
790 	short ostate = 0;
791 #endif
792 	int iss = 0;
793 	u_long tiwin;
794 	struct tcp_opt_info opti;
795 	int off, iphlen;
796 	va_list ap;
797 	int af;		/* af on the wire */
798 	struct mbuf *tcp_saveti = NULL;
799 
800 	MCLAIM(m, &tcp_rx_mowner);
801 	va_start(ap, m);
802 	toff = va_arg(ap, int);
803 	(void)va_arg(ap, int);		/* ignore value, advance ap */
804 	va_end(ap);
805 
806 	tcpstat.tcps_rcvtotal++;
807 
808 	bzero(&opti, sizeof(opti));
809 	opti.ts_present = 0;
810 	opti.maxseg = 0;
811 
812 	/*
813 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
814 	 *
815 	 * TCP is, by definition, unicast, so we reject all
816 	 * multicast outright.
817 	 *
818 	 * Note, there are additional src/dst address checks in
819 	 * the AF-specific code below.
820 	 */
821 	if (m->m_flags & (M_BCAST|M_MCAST)) {
822 		/* XXX stat */
823 		goto drop;
824 	}
825 #ifdef INET6
826 	if (m->m_flags & M_ANYCAST6) {
827 		/* XXX stat */
828 		goto drop;
829 	}
830 #endif
831 
832 	/*
833 	 * Get IP and TCP header together in first mbuf.
834 	 * Note: IP leaves IP header in first mbuf.
835 	 */
836 	ip = mtod(m, struct ip *);
837 #ifdef INET6
838 	ip6 = NULL;
839 #endif
840 	switch (ip->ip_v) {
841 #ifdef INET
842 	case 4:
843 		af = AF_INET;
844 		iphlen = sizeof(struct ip);
845 #ifndef PULLDOWN_TEST
846 		/* would like to get rid of this... */
847 		if (toff > sizeof (struct ip)) {
848 			ip_stripoptions(m, (struct mbuf *)0);
849 			toff = sizeof(struct ip);
850 		}
851 		if (m->m_len < toff + sizeof (struct tcphdr)) {
852 			if ((m = m_pullup(m, toff + sizeof (struct tcphdr))) == 0) {
853 				tcpstat.tcps_rcvshort++;
854 				return;
855 			}
856 		}
857 		ip = mtod(m, struct ip *);
858 		th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
859 #else
860 		ip = mtod(m, struct ip *);
861 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
862 			sizeof(struct tcphdr));
863 		if (th == NULL) {
864 			tcpstat.tcps_rcvshort++;
865 			return;
866 		}
867 #endif
868 		/* We do the checksum after PCB lookup... */
869 		len = ntohs(ip->ip_len);
870 		tlen = len - toff;
871 		break;
872 #endif
873 #ifdef INET6
874 	case 6:
875 		ip = NULL;
876 		iphlen = sizeof(struct ip6_hdr);
877 		af = AF_INET6;
878 #ifndef PULLDOWN_TEST
879 		if (m->m_len < toff + sizeof(struct tcphdr)) {
880 			m = m_pullup(m, toff + sizeof(struct tcphdr));	/*XXX*/
881 			if (m == NULL) {
882 				tcpstat.tcps_rcvshort++;
883 				return;
884 			}
885 		}
886 		ip6 = mtod(m, struct ip6_hdr *);
887 		th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
888 #else
889 		ip6 = mtod(m, struct ip6_hdr *);
890 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
891 			sizeof(struct tcphdr));
892 		if (th == NULL) {
893 			tcpstat.tcps_rcvshort++;
894 			return;
895 		}
896 #endif
897 
898 		/* Be proactive about malicious use of IPv4 mapped address */
899 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
900 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
901 			/* XXX stat */
902 			goto drop;
903 		}
904 
905 		/*
906 		 * Be proactive about unspecified IPv6 address in source.
907 		 * As we use all-zero to indicate unbounded/unconnected pcb,
908 		 * unspecified IPv6 address can be used to confuse us.
909 		 *
910 		 * Note that packets with unspecified IPv6 destination is
911 		 * already dropped in ip6_input.
912 		 */
913 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
914 			/* XXX stat */
915 			goto drop;
916 		}
917 
918 		/*
919 		 * Make sure destination address is not multicast.
920 		 * Source address checked in ip6_input().
921 		 */
922 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
923 			/* XXX stat */
924 			goto drop;
925 		}
926 
927 		/* We do the checksum after PCB lookup... */
928 		len = m->m_pkthdr.len;
929 		tlen = len - toff;
930 		break;
931 #endif
932 	default:
933 		m_freem(m);
934 		return;
935 	}
936 
937 	KASSERT(TCP_HDR_ALIGNED_P(th));
938 
939 	/*
940 	 * Check that TCP offset makes sense,
941 	 * pull out TCP options and adjust length.		XXX
942 	 */
943 	off = th->th_off << 2;
944 	if (off < sizeof (struct tcphdr) || off > tlen) {
945 		tcpstat.tcps_rcvbadoff++;
946 		goto drop;
947 	}
948 	tlen -= off;
949 
950 	/*
951 	 * tcp_input() has been modified to use tlen to mean the TCP data
952 	 * length throughout the function.  Other functions can use
953 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
954 	 * rja
955 	 */
956 
957 	if (off > sizeof (struct tcphdr)) {
958 #ifndef PULLDOWN_TEST
959 		if (m->m_len < toff + off) {
960 			if ((m = m_pullup(m, toff + off)) == 0) {
961 				tcpstat.tcps_rcvshort++;
962 				return;
963 			}
964 			switch (af) {
965 #ifdef INET
966 			case AF_INET:
967 				ip = mtod(m, struct ip *);
968 				break;
969 #endif
970 #ifdef INET6
971 			case AF_INET6:
972 				ip6 = mtod(m, struct ip6_hdr *);
973 				break;
974 #endif
975 			}
976 			th = (struct tcphdr *)(mtod(m, caddr_t) + toff);
977 		}
978 #else
979 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
980 		if (th == NULL) {
981 			tcpstat.tcps_rcvshort++;
982 			return;
983 		}
984 		/*
985 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
986 		 * (as they're before toff) and we don't need to update those.
987 		 */
988 #endif
989 		KASSERT(TCP_HDR_ALIGNED_P(th));
990 		optlen = off - sizeof (struct tcphdr);
991 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
992 		/*
993 		 * Do quick retrieval of timestamp options ("options
994 		 * prediction?").  If timestamp is the only option and it's
995 		 * formatted as recommended in RFC 1323 appendix A, we
996 		 * quickly get the values now and not bother calling
997 		 * tcp_dooptions(), etc.
998 		 */
999 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1000 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1001 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1002 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1003 		     (th->th_flags & TH_SYN) == 0) {
1004 			opti.ts_present = 1;
1005 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1006 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1007 			optp = NULL;	/* we've parsed the options */
1008 		}
1009 	}
1010 	tiflags = th->th_flags;
1011 
1012 	/*
1013 	 * Locate pcb for segment.
1014 	 */
1015 findpcb:
1016 	inp = NULL;
1017 #ifdef INET6
1018 	in6p = NULL;
1019 #endif
1020 	switch (af) {
1021 #ifdef INET
1022 	case AF_INET:
1023 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1024 		    ip->ip_dst, th->th_dport);
1025 		if (inp == 0) {
1026 			++tcpstat.tcps_pcbhashmiss;
1027 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1028 		}
1029 #ifdef INET6
1030 		if (inp == 0) {
1031 			struct in6_addr s, d;
1032 
1033 			/* mapped addr case */
1034 			bzero(&s, sizeof(s));
1035 			s.s6_addr16[5] = htons(0xffff);
1036 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1037 			bzero(&d, sizeof(d));
1038 			d.s6_addr16[5] = htons(0xffff);
1039 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1040 			in6p = in6_pcblookup_connect(&tcb6, &s, th->th_sport,
1041 				&d, th->th_dport, 0);
1042 			if (in6p == 0) {
1043 				++tcpstat.tcps_pcbhashmiss;
1044 				in6p = in6_pcblookup_bind(&tcb6, &d,
1045 					th->th_dport, 0);
1046 			}
1047 		}
1048 #endif
1049 #ifndef INET6
1050 		if (inp == 0)
1051 #else
1052 		if (inp == 0 && in6p == 0)
1053 #endif
1054 		{
1055 			++tcpstat.tcps_noport;
1056 			if (tcp_log_refused &&
1057 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1058 				tcp4_log_refused(ip, th);
1059 			}
1060 			TCP_FIELDS_TO_HOST(th);
1061 			goto dropwithreset_ratelim;
1062 		}
1063 #ifdef IPSEC
1064 		if (inp && ipsec4_in_reject(m, inp)) {
1065 			ipsecstat.in_polvio++;
1066 			goto drop;
1067 		}
1068 #ifdef INET6
1069 		else if (in6p && ipsec4_in_reject_so(m, in6p->in6p_socket)) {
1070 			ipsecstat.in_polvio++;
1071 			goto drop;
1072 		}
1073 #endif
1074 #endif /*IPSEC*/
1075 		break;
1076 #endif /*INET*/
1077 #ifdef INET6
1078 	case AF_INET6:
1079 	    {
1080 		int faith;
1081 
1082 #if defined(NFAITH) && NFAITH > 0
1083 		faith = faithprefix(&ip6->ip6_dst);
1084 #else
1085 		faith = 0;
1086 #endif
1087 		in6p = in6_pcblookup_connect(&tcb6, &ip6->ip6_src, th->th_sport,
1088 			&ip6->ip6_dst, th->th_dport, faith);
1089 		if (in6p == NULL) {
1090 			++tcpstat.tcps_pcbhashmiss;
1091 			in6p = in6_pcblookup_bind(&tcb6, &ip6->ip6_dst,
1092 				th->th_dport, faith);
1093 		}
1094 		if (in6p == NULL) {
1095 			++tcpstat.tcps_noport;
1096 			if (tcp_log_refused &&
1097 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1098 				tcp6_log_refused(ip6, th);
1099 			}
1100 			TCP_FIELDS_TO_HOST(th);
1101 			goto dropwithreset_ratelim;
1102 		}
1103 #ifdef IPSEC
1104 		if (ipsec6_in_reject(m, in6p)) {
1105 			ipsec6stat.in_polvio++;
1106 			goto drop;
1107 		}
1108 #endif /*IPSEC*/
1109 		break;
1110 	    }
1111 #endif
1112 	}
1113 
1114 	/*
1115 	 * If the state is CLOSED (i.e., TCB does not exist) then
1116 	 * all data in the incoming segment is discarded.
1117 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1118 	 * but should either do a listen or a connect soon.
1119 	 */
1120 	tp = NULL;
1121 	so = NULL;
1122 	if (inp) {
1123 		tp = intotcpcb(inp);
1124 		so = inp->inp_socket;
1125 	}
1126 #ifdef INET6
1127 	else if (in6p) {
1128 		tp = in6totcpcb(in6p);
1129 		so = in6p->in6p_socket;
1130 	}
1131 #endif
1132 	if (tp == 0) {
1133 		TCP_FIELDS_TO_HOST(th);
1134 		goto dropwithreset_ratelim;
1135 	}
1136 	if (tp->t_state == TCPS_CLOSED)
1137 		goto drop;
1138 
1139 	/*
1140 	 * Checksum extended TCP header and data.
1141 	 */
1142 	switch (af) {
1143 #ifdef INET
1144 	case AF_INET:
1145 		switch (m->m_pkthdr.csum_flags &
1146 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
1147 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
1148 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
1149 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
1150 			goto badcsum;
1151 
1152 		case M_CSUM_TCPv4|M_CSUM_DATA:
1153 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
1154 			if ((m->m_pkthdr.csum_data ^ 0xffff) != 0)
1155 				goto badcsum;
1156 			break;
1157 
1158 		case M_CSUM_TCPv4:
1159 			/* Checksum was okay. */
1160 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
1161 			break;
1162 
1163 		default:
1164 			/* Must compute it ourselves. */
1165 			TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
1166 #ifndef PULLDOWN_TEST
1167 		    {
1168 			struct ipovly *ipov;
1169 			ipov = (struct ipovly *)ip;
1170 			bzero(ipov->ih_x1, sizeof ipov->ih_x1);
1171 			ipov->ih_len = htons(tlen + off);
1172 
1173 			if (in_cksum(m, len) != 0)
1174 				goto badcsum;
1175 		    }
1176 #else
1177 			if (in4_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1178 				goto badcsum;
1179 #endif /* ! PULLDOWN_TEST */
1180 			break;
1181 		}
1182 		break;
1183 #endif /* INET4 */
1184 
1185 #ifdef INET6
1186 	case AF_INET6:
1187 		if (in6_cksum(m, IPPROTO_TCP, toff, tlen + off) != 0)
1188 			goto badcsum;
1189 		break;
1190 #endif /* INET6 */
1191 	}
1192 
1193 	TCP_FIELDS_TO_HOST(th);
1194 
1195 	/* Unscale the window into a 32-bit value. */
1196 	if ((tiflags & TH_SYN) == 0)
1197 		tiwin = th->th_win << tp->snd_scale;
1198 	else
1199 		tiwin = th->th_win;
1200 
1201 #ifdef INET6
1202 	/* save packet options if user wanted */
1203 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1204 		if (in6p->in6p_options) {
1205 			m_freem(in6p->in6p_options);
1206 			in6p->in6p_options = 0;
1207 		}
1208 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1209 	}
1210 #endif
1211 
1212 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1213 		union syn_cache_sa src;
1214 		union syn_cache_sa dst;
1215 
1216 		bzero(&src, sizeof(src));
1217 		bzero(&dst, sizeof(dst));
1218 		switch (af) {
1219 #ifdef INET
1220 		case AF_INET:
1221 			src.sin.sin_len = sizeof(struct sockaddr_in);
1222 			src.sin.sin_family = AF_INET;
1223 			src.sin.sin_addr = ip->ip_src;
1224 			src.sin.sin_port = th->th_sport;
1225 
1226 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1227 			dst.sin.sin_family = AF_INET;
1228 			dst.sin.sin_addr = ip->ip_dst;
1229 			dst.sin.sin_port = th->th_dport;
1230 			break;
1231 #endif
1232 #ifdef INET6
1233 		case AF_INET6:
1234 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1235 			src.sin6.sin6_family = AF_INET6;
1236 			src.sin6.sin6_addr = ip6->ip6_src;
1237 			src.sin6.sin6_port = th->th_sport;
1238 
1239 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1240 			dst.sin6.sin6_family = AF_INET6;
1241 			dst.sin6.sin6_addr = ip6->ip6_dst;
1242 			dst.sin6.sin6_port = th->th_dport;
1243 			break;
1244 #endif /* INET6 */
1245 		default:
1246 			goto badsyn;	/*sanity*/
1247 		}
1248 
1249 		if (so->so_options & SO_DEBUG) {
1250 #ifdef TCP_DEBUG
1251 			ostate = tp->t_state;
1252 #endif
1253 
1254 			tcp_saveti = NULL;
1255 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1256 				goto nosave;
1257 
1258 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1259 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1260 				if (!tcp_saveti)
1261 					goto nosave;
1262 			} else {
1263 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1264 				if (!tcp_saveti)
1265 					goto nosave;
1266 				MCLAIM(m, &tcp_mowner);
1267 				tcp_saveti->m_len = iphlen;
1268 				m_copydata(m, 0, iphlen,
1269 				    mtod(tcp_saveti, caddr_t));
1270 			}
1271 
1272 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1273 				m_freem(tcp_saveti);
1274 				tcp_saveti = NULL;
1275 			} else {
1276 				tcp_saveti->m_len += sizeof(struct tcphdr);
1277 				bcopy(th, mtod(tcp_saveti, caddr_t) + iphlen,
1278 				    sizeof(struct tcphdr));
1279 			}
1280 			if (tcp_saveti) {
1281 				/*
1282 				 * need to recover version # field, which was
1283 				 * overwritten on ip_cksum computation.
1284 				 */
1285 				struct ip *sip;
1286 				sip = mtod(tcp_saveti, struct ip *);
1287 				switch (af) {
1288 #ifdef INET
1289 				case AF_INET:
1290 					sip->ip_v = 4;
1291 					break;
1292 #endif
1293 #ifdef INET6
1294 				case AF_INET6:
1295 					sip->ip_v = 6;
1296 					break;
1297 #endif
1298 				}
1299 			}
1300 	nosave:;
1301 		}
1302 		if (so->so_options & SO_ACCEPTCONN) {
1303 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1304 				if (tiflags & TH_RST) {
1305 					syn_cache_reset(&src.sa, &dst.sa, th);
1306 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1307 				    (TH_ACK|TH_SYN)) {
1308 					/*
1309 					 * Received a SYN,ACK.  This should
1310 					 * never happen while we are in
1311 					 * LISTEN.  Send an RST.
1312 					 */
1313 					goto badsyn;
1314 				} else if (tiflags & TH_ACK) {
1315 					so = syn_cache_get(&src.sa, &dst.sa,
1316 						th, toff, tlen, so, m);
1317 					if (so == NULL) {
1318 						/*
1319 						 * We don't have a SYN for
1320 						 * this ACK; send an RST.
1321 						 */
1322 						goto badsyn;
1323 					} else if (so ==
1324 					    (struct socket *)(-1)) {
1325 						/*
1326 						 * We were unable to create
1327 						 * the connection.  If the
1328 						 * 3-way handshake was
1329 						 * completed, and RST has
1330 						 * been sent to the peer.
1331 						 * Since the mbuf might be
1332 						 * in use for the reply,
1333 						 * do not free it.
1334 						 */
1335 						m = NULL;
1336 					} else {
1337 						/*
1338 						 * We have created a
1339 						 * full-blown connection.
1340 						 */
1341 						tp = NULL;
1342 						inp = NULL;
1343 #ifdef INET6
1344 						in6p = NULL;
1345 #endif
1346 						switch (so->so_proto->pr_domain->dom_family) {
1347 #ifdef INET
1348 						case AF_INET:
1349 							inp = sotoinpcb(so);
1350 							tp = intotcpcb(inp);
1351 							break;
1352 #endif
1353 #ifdef INET6
1354 						case AF_INET6:
1355 							in6p = sotoin6pcb(so);
1356 							tp = in6totcpcb(in6p);
1357 							break;
1358 #endif
1359 						}
1360 						if (tp == NULL)
1361 							goto badsyn;	/*XXX*/
1362 						tiwin <<= tp->snd_scale;
1363 						goto after_listen;
1364 					}
1365 				} else {
1366 					/*
1367 					 * None of RST, SYN or ACK was set.
1368 					 * This is an invalid packet for a
1369 					 * TCB in LISTEN state.  Send a RST.
1370 					 */
1371 					goto badsyn;
1372 				}
1373 			} else {
1374 				/*
1375 				 * Received a SYN.
1376 				 */
1377 
1378 #ifdef INET6
1379 				/*
1380 				 * If deprecated address is forbidden, we do
1381 				 * not accept SYN to deprecated interface
1382 				 * address to prevent any new inbound
1383 				 * connection from getting established.
1384 				 * When we do not accept SYN, we send a TCP
1385 				 * RST, with deprecated source address (instead
1386 				 * of dropping it).  We compromise it as it is
1387 				 * much better for peer to send a RST, and
1388 				 * RST will be the final packet for the
1389 				 * exchange.
1390 				 *
1391 				 * If we do not forbid deprecated addresses, we
1392 				 * accept the SYN packet.  RFC2462 does not
1393 				 * suggest dropping SYN in this case.
1394 				 * If we decipher RFC2462 5.5.4, it says like
1395 				 * this:
1396 				 * 1. use of deprecated addr with existing
1397 				 *    communication is okay - "SHOULD continue
1398 				 *    to be used"
1399 				 * 2. use of it with new communication:
1400 				 *   (2a) "SHOULD NOT be used if alternate
1401 				 *        address with sufficient scope is
1402 				 *        available"
1403 				 *   (2b) nothing mentioned otherwise.
1404 				 * Here we fall into (2b) case as we have no
1405 				 * choice in our source address selection - we
1406 				 * must obey the peer.
1407 				 *
1408 				 * The wording in RFC2462 is confusing, and
1409 				 * there are multiple description text for
1410 				 * deprecated address handling - worse, they
1411 				 * are not exactly the same.  I believe 5.5.4
1412 				 * is the best one, so we follow 5.5.4.
1413 				 */
1414 				if (af == AF_INET6 && !ip6_use_deprecated) {
1415 					struct in6_ifaddr *ia6;
1416 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1417 					    &ip6->ip6_dst)) &&
1418 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1419 						tp = NULL;
1420 						goto dropwithreset;
1421 					}
1422 				}
1423 #endif
1424 
1425 				/*
1426 				 * LISTEN socket received a SYN
1427 				 * from itself?  This can't possibly
1428 				 * be valid; drop the packet.
1429 				 */
1430 				if (th->th_sport == th->th_dport) {
1431 					int i;
1432 
1433 					switch (af) {
1434 #ifdef INET
1435 					case AF_INET:
1436 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1437 						break;
1438 #endif
1439 #ifdef INET6
1440 					case AF_INET6:
1441 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1442 						break;
1443 #endif
1444 					default:
1445 						i = 1;
1446 					}
1447 					if (i) {
1448 						tcpstat.tcps_badsyn++;
1449 						goto drop;
1450 					}
1451 				}
1452 
1453 				/*
1454 				 * SYN looks ok; create compressed TCP
1455 				 * state for it.
1456 				 */
1457 				if (so->so_qlen <= so->so_qlimit &&
1458 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1459 						so, m, optp, optlen, &opti))
1460 					m = NULL;
1461 			}
1462 			goto drop;
1463 		}
1464 	}
1465 
1466 after_listen:
1467 #ifdef DIAGNOSTIC
1468 	/*
1469 	 * Should not happen now that all embryonic connections
1470 	 * are handled with compressed state.
1471 	 */
1472 	if (tp->t_state == TCPS_LISTEN)
1473 		panic("tcp_input: TCPS_LISTEN");
1474 #endif
1475 
1476 	/*
1477 	 * Segment received on connection.
1478 	 * Reset idle time and keep-alive timer.
1479 	 */
1480 	tp->t_rcvtime = tcp_now;
1481 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1482 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1483 
1484 	/*
1485 	 * Process options.
1486 	 */
1487 	if (optp)
1488 		tcp_dooptions(tp, optp, optlen, th, &opti);
1489 
1490 	/*
1491 	 * Header prediction: check for the two common cases
1492 	 * of a uni-directional data xfer.  If the packet has
1493 	 * no control flags, is in-sequence, the window didn't
1494 	 * change and we're not retransmitting, it's a
1495 	 * candidate.  If the length is zero and the ack moved
1496 	 * forward, we're the sender side of the xfer.  Just
1497 	 * free the data acked & wake any higher level process
1498 	 * that was blocked waiting for space.  If the length
1499 	 * is non-zero and the ack didn't move, we're the
1500 	 * receiver side.  If we're getting packets in-order
1501 	 * (the reassembly queue is empty), add the data to
1502 	 * the socket buffer and note that we need a delayed ack.
1503 	 */
1504 	if (tp->t_state == TCPS_ESTABLISHED &&
1505 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1506 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1507 	    th->th_seq == tp->rcv_nxt &&
1508 	    tiwin && tiwin == tp->snd_wnd &&
1509 	    tp->snd_nxt == tp->snd_max) {
1510 
1511 		/*
1512 		 * If last ACK falls within this segment's sequence numbers,
1513 		 *  record the timestamp.
1514 		 */
1515 		if (opti.ts_present &&
1516 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1517 		    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen)) {
1518 			tp->ts_recent_age = TCP_TIMESTAMP(tp);
1519 			tp->ts_recent = opti.ts_val;
1520 		}
1521 
1522 		if (tlen == 0) {
1523 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1524 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1525 			    tp->snd_cwnd >= tp->snd_wnd &&
1526 			    tp->t_dupacks < tcprexmtthresh) {
1527 				/*
1528 				 * this is a pure ack for outstanding data.
1529 				 */
1530 				++tcpstat.tcps_predack;
1531 				if (opti.ts_present && opti.ts_ecr)
1532 					tcp_xmit_timer(tp,
1533 					  TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
1534 				else if (tp->t_rtttime &&
1535 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1536 					tcp_xmit_timer(tp,
1537 					tcp_now - tp->t_rtttime);
1538 				acked = th->th_ack - tp->snd_una;
1539 				tcpstat.tcps_rcvackpack++;
1540 				tcpstat.tcps_rcvackbyte += acked;
1541 				ND6_HINT(tp);
1542 				sbdrop(&so->so_snd, acked);
1543 				/*
1544 				 * We want snd_recover to track snd_una to
1545 				 * avoid sequence wraparound problems for
1546 				 * very large transfers.
1547 				 */
1548 				tp->snd_una = tp->snd_recover = th->th_ack;
1549 				m_freem(m);
1550 
1551 				/*
1552 				 * If all outstanding data are acked, stop
1553 				 * retransmit timer, otherwise restart timer
1554 				 * using current (possibly backed-off) value.
1555 				 * If process is waiting for space,
1556 				 * wakeup/selwakeup/signal.  If data
1557 				 * are ready to send, let tcp_output
1558 				 * decide between more output or persist.
1559 				 */
1560 				if (tp->snd_una == tp->snd_max)
1561 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1562 				else if (TCP_TIMER_ISARMED(tp,
1563 				    TCPT_PERSIST) == 0)
1564 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1565 					    tp->t_rxtcur);
1566 
1567 				sowwakeup(so);
1568 				if (so->so_snd.sb_cc)
1569 					(void) tcp_output(tp);
1570 				if (tcp_saveti)
1571 					m_freem(tcp_saveti);
1572 				return;
1573 			}
1574 		} else if (th->th_ack == tp->snd_una &&
1575 		    TAILQ_FIRST(&tp->segq) == NULL &&
1576 		    tlen <= sbspace(&so->so_rcv)) {
1577 			/*
1578 			 * this is a pure, in-sequence data packet
1579 			 * with nothing on the reassembly queue and
1580 			 * we have enough buffer space to take it.
1581 			 */
1582 			++tcpstat.tcps_preddat;
1583 			tp->rcv_nxt += tlen;
1584 			tcpstat.tcps_rcvpack++;
1585 			tcpstat.tcps_rcvbyte += tlen;
1586 			ND6_HINT(tp);
1587 			/*
1588 			 * Drop TCP, IP headers and TCP options then add data
1589 			 * to socket buffer.
1590 			 */
1591 			if (so->so_state & SS_CANTRCVMORE)
1592 				m_freem(m);
1593 			else {
1594 				m_adj(m, toff + off);
1595 				sbappendstream(&so->so_rcv, m);
1596 			}
1597 			sorwakeup(so);
1598 			TCP_SETUP_ACK(tp, th);
1599 			if (tp->t_flags & TF_ACKNOW)
1600 				(void) tcp_output(tp);
1601 			if (tcp_saveti)
1602 				m_freem(tcp_saveti);
1603 			return;
1604 		}
1605 	}
1606 
1607 	/*
1608 	 * Compute mbuf offset to TCP data segment.
1609 	 */
1610 	hdroptlen = toff + off;
1611 
1612 	/*
1613 	 * Calculate amount of space in receive window,
1614 	 * and then do TCP input processing.
1615 	 * Receive window is amount of space in rcv queue,
1616 	 * but not less than advertised window.
1617 	 */
1618 	{ int win;
1619 
1620 	win = sbspace(&so->so_rcv);
1621 	if (win < 0)
1622 		win = 0;
1623 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1624 	}
1625 
1626 	switch (tp->t_state) {
1627 	case TCPS_LISTEN:
1628 		/*
1629 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1630 		 */
1631 		if (m->m_flags & (M_BCAST|M_MCAST))
1632 			goto drop;
1633 		switch (af) {
1634 #ifdef INET6
1635 		case AF_INET6:
1636 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1637 				goto drop;
1638 			break;
1639 #endif /* INET6 */
1640 		case AF_INET:
1641 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1642 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1643 				goto drop;
1644 			break;
1645 		}
1646 		break;
1647 
1648 	/*
1649 	 * If the state is SYN_SENT:
1650 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1651 	 *	if seg contains a RST, then drop the connection.
1652 	 *	if seg does not contain SYN, then drop it.
1653 	 * Otherwise this is an acceptable SYN segment
1654 	 *	initialize tp->rcv_nxt and tp->irs
1655 	 *	if seg contains ack then advance tp->snd_una
1656 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1657 	 *	arrange for segment to be acked (eventually)
1658 	 *	continue processing rest of data/controls, beginning with URG
1659 	 */
1660 	case TCPS_SYN_SENT:
1661 		if ((tiflags & TH_ACK) &&
1662 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1663 		     SEQ_GT(th->th_ack, tp->snd_max)))
1664 			goto dropwithreset;
1665 		if (tiflags & TH_RST) {
1666 			if (tiflags & TH_ACK)
1667 				tp = tcp_drop(tp, ECONNREFUSED);
1668 			goto drop;
1669 		}
1670 		if ((tiflags & TH_SYN) == 0)
1671 			goto drop;
1672 		if (tiflags & TH_ACK) {
1673 			tp->snd_una = tp->snd_recover = th->th_ack;
1674 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1675 				tp->snd_nxt = tp->snd_una;
1676 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1677 		}
1678 		tp->irs = th->th_seq;
1679 		tcp_rcvseqinit(tp);
1680 		tp->t_flags |= TF_ACKNOW;
1681 		tcp_mss_from_peer(tp, opti.maxseg);
1682 
1683 		/*
1684 		 * Initialize the initial congestion window.  If we
1685 		 * had to retransmit the SYN, we must initialize cwnd
1686 		 * to 1 segment (i.e. the Loss Window).
1687 		 */
1688 		if (tp->t_flags & TF_SYN_REXMT)
1689 			tp->snd_cwnd = tp->t_peermss;
1690 		else {
1691 			int ss = tcp_init_win;
1692 #ifdef INET
1693 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1694 				ss = tcp_init_win_local;
1695 #endif
1696 #ifdef INET6
1697 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1698 				ss = tcp_init_win_local;
1699 #endif
1700 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1701 		}
1702 
1703 		tcp_rmx_rtt(tp);
1704 		if (tiflags & TH_ACK) {
1705 			tcpstat.tcps_connects++;
1706 			soisconnected(so);
1707 			tcp_established(tp);
1708 			/* Do window scaling on this connection? */
1709 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1710 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1711 				tp->snd_scale = tp->requested_s_scale;
1712 				tp->rcv_scale = tp->request_r_scale;
1713 			}
1714 			TCP_REASS_LOCK(tp);
1715 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1716 			TCP_REASS_UNLOCK(tp);
1717 			/*
1718 			 * if we didn't have to retransmit the SYN,
1719 			 * use its rtt as our initial srtt & rtt var.
1720 			 */
1721 			if (tp->t_rtttime)
1722 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1723 		} else
1724 			tp->t_state = TCPS_SYN_RECEIVED;
1725 
1726 		/*
1727 		 * Advance th->th_seq to correspond to first data byte.
1728 		 * If data, trim to stay within window,
1729 		 * dropping FIN if necessary.
1730 		 */
1731 		th->th_seq++;
1732 		if (tlen > tp->rcv_wnd) {
1733 			todrop = tlen - tp->rcv_wnd;
1734 			m_adj(m, -todrop);
1735 			tlen = tp->rcv_wnd;
1736 			tiflags &= ~TH_FIN;
1737 			tcpstat.tcps_rcvpackafterwin++;
1738 			tcpstat.tcps_rcvbyteafterwin += todrop;
1739 		}
1740 		tp->snd_wl1 = th->th_seq - 1;
1741 		tp->rcv_up = th->th_seq;
1742 		goto step6;
1743 
1744 	/*
1745 	 * If the state is SYN_RECEIVED:
1746 	 *	If seg contains an ACK, but not for our SYN, drop the input
1747 	 *	and generate an RST.  See page 36, rfc793
1748 	 */
1749 	case TCPS_SYN_RECEIVED:
1750 		if ((tiflags & TH_ACK) &&
1751 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1752 		     SEQ_GT(th->th_ack, tp->snd_max)))
1753 			goto dropwithreset;
1754 		break;
1755 	}
1756 
1757 	/*
1758 	 * States other than LISTEN or SYN_SENT.
1759 	 * First check timestamp, if present.
1760 	 * Then check that at least some bytes of segment are within
1761 	 * receive window.  If segment begins before rcv_nxt,
1762 	 * drop leading data (and SYN); if nothing left, just ack.
1763 	 *
1764 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1765 	 * and it's less than ts_recent, drop it.
1766 	 */
1767 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1768 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1769 
1770 		/* Check to see if ts_recent is over 24 days old.  */
1771 		if ((int)(TCP_TIMESTAMP(tp) - tp->ts_recent_age) >
1772 		    TCP_PAWS_IDLE) {
1773 			/*
1774 			 * Invalidate ts_recent.  If this segment updates
1775 			 * ts_recent, the age will be reset later and ts_recent
1776 			 * will get a valid value.  If it does not, setting
1777 			 * ts_recent to zero will at least satisfy the
1778 			 * requirement that zero be placed in the timestamp
1779 			 * echo reply when ts_recent isn't valid.  The
1780 			 * age isn't reset until we get a valid ts_recent
1781 			 * because we don't want out-of-order segments to be
1782 			 * dropped when ts_recent is old.
1783 			 */
1784 			tp->ts_recent = 0;
1785 		} else {
1786 			tcpstat.tcps_rcvduppack++;
1787 			tcpstat.tcps_rcvdupbyte += tlen;
1788 			tcpstat.tcps_pawsdrop++;
1789 			goto dropafterack;
1790 		}
1791 	}
1792 
1793 	todrop = tp->rcv_nxt - th->th_seq;
1794 	if (todrop > 0) {
1795 		if (tiflags & TH_SYN) {
1796 			tiflags &= ~TH_SYN;
1797 			th->th_seq++;
1798 			if (th->th_urp > 1)
1799 				th->th_urp--;
1800 			else {
1801 				tiflags &= ~TH_URG;
1802 				th->th_urp = 0;
1803 			}
1804 			todrop--;
1805 		}
1806 		if (todrop > tlen ||
1807 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1808 			/*
1809 			 * Any valid FIN must be to the left of the window.
1810 			 * At this point the FIN must be a duplicate or
1811 			 * out of sequence; drop it.
1812 			 */
1813 			tiflags &= ~TH_FIN;
1814 			/*
1815 			 * Send an ACK to resynchronize and drop any data.
1816 			 * But keep on processing for RST or ACK.
1817 			 */
1818 			tp->t_flags |= TF_ACKNOW;
1819 			todrop = tlen;
1820 			tcpstat.tcps_rcvdupbyte += todrop;
1821 			tcpstat.tcps_rcvduppack++;
1822 		} else {
1823 			tcpstat.tcps_rcvpartduppack++;
1824 			tcpstat.tcps_rcvpartdupbyte += todrop;
1825 		}
1826 		hdroptlen += todrop;	/*drop from head afterwards*/
1827 		th->th_seq += todrop;
1828 		tlen -= todrop;
1829 		if (th->th_urp > todrop)
1830 			th->th_urp -= todrop;
1831 		else {
1832 			tiflags &= ~TH_URG;
1833 			th->th_urp = 0;
1834 		}
1835 	}
1836 
1837 	/*
1838 	 * If new data are received on a connection after the
1839 	 * user processes are gone, then RST the other end.
1840 	 */
1841 	if ((so->so_state & SS_NOFDREF) &&
1842 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1843 		tp = tcp_close(tp);
1844 		tcpstat.tcps_rcvafterclose++;
1845 		goto dropwithreset;
1846 	}
1847 
1848 	/*
1849 	 * If segment ends after window, drop trailing data
1850 	 * (and PUSH and FIN); if nothing left, just ACK.
1851 	 */
1852 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1853 	if (todrop > 0) {
1854 		tcpstat.tcps_rcvpackafterwin++;
1855 		if (todrop >= tlen) {
1856 			tcpstat.tcps_rcvbyteafterwin += tlen;
1857 			/*
1858 			 * If a new connection request is received
1859 			 * while in TIME_WAIT, drop the old connection
1860 			 * and start over if the sequence numbers
1861 			 * are above the previous ones.
1862 			 *
1863 			 * NOTE: We will checksum the packet again, and
1864 			 * so we need to put the header fields back into
1865 			 * network order!
1866 			 * XXX This kind of sucks, but we don't expect
1867 			 * XXX this to happen very often, so maybe it
1868 			 * XXX doesn't matter so much.
1869 			 */
1870 			if (tiflags & TH_SYN &&
1871 			    tp->t_state == TCPS_TIME_WAIT &&
1872 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1873 				iss = tcp_new_iss(tp, tp->snd_nxt);
1874 				tp = tcp_close(tp);
1875 				TCP_FIELDS_TO_NET(th);
1876 				goto findpcb;
1877 			}
1878 			/*
1879 			 * If window is closed can only take segments at
1880 			 * window edge, and have to drop data and PUSH from
1881 			 * incoming segments.  Continue processing, but
1882 			 * remember to ack.  Otherwise, drop segment
1883 			 * and ack.
1884 			 */
1885 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1886 				tp->t_flags |= TF_ACKNOW;
1887 				tcpstat.tcps_rcvwinprobe++;
1888 			} else
1889 				goto dropafterack;
1890 		} else
1891 			tcpstat.tcps_rcvbyteafterwin += todrop;
1892 		m_adj(m, -todrop);
1893 		tlen -= todrop;
1894 		tiflags &= ~(TH_PUSH|TH_FIN);
1895 	}
1896 
1897 	/*
1898 	 * If last ACK falls within this segment's sequence numbers,
1899 	 * and the timestamp is newer, record it.
1900 	 */
1901 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1902 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1903 	    SEQ_LT(tp->last_ack_sent, th->th_seq + tlen +
1904 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1905 		tp->ts_recent_age = TCP_TIMESTAMP(tp);
1906 		tp->ts_recent = opti.ts_val;
1907 	}
1908 
1909 	/*
1910 	 * If the RST bit is set examine the state:
1911 	 *    SYN_RECEIVED STATE:
1912 	 *	If passive open, return to LISTEN state.
1913 	 *	If active open, inform user that connection was refused.
1914 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1915 	 *	Inform user that connection was reset, and close tcb.
1916 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1917 	 *	Close the tcb.
1918 	 */
1919 	if (tiflags&TH_RST) switch (tp->t_state) {
1920 
1921 	case TCPS_SYN_RECEIVED:
1922 		so->so_error = ECONNREFUSED;
1923 		goto close;
1924 
1925 	case TCPS_ESTABLISHED:
1926 	case TCPS_FIN_WAIT_1:
1927 	case TCPS_FIN_WAIT_2:
1928 	case TCPS_CLOSE_WAIT:
1929 		so->so_error = ECONNRESET;
1930 	close:
1931 		tp->t_state = TCPS_CLOSED;
1932 		tcpstat.tcps_drops++;
1933 		tp = tcp_close(tp);
1934 		goto drop;
1935 
1936 	case TCPS_CLOSING:
1937 	case TCPS_LAST_ACK:
1938 	case TCPS_TIME_WAIT:
1939 		tp = tcp_close(tp);
1940 		goto drop;
1941 	}
1942 
1943 	/*
1944 	 * If a SYN is in the window, then this is an
1945 	 * error and we send an RST and drop the connection.
1946 	 */
1947 	if (tiflags & TH_SYN) {
1948 		tp = tcp_drop(tp, ECONNRESET);
1949 		goto dropwithreset;
1950 	}
1951 
1952 	/*
1953 	 * If the ACK bit is off we drop the segment and return.
1954 	 */
1955 	if ((tiflags & TH_ACK) == 0) {
1956 		if (tp->t_flags & TF_ACKNOW)
1957 			goto dropafterack;
1958 		else
1959 			goto drop;
1960 	}
1961 
1962 	/*
1963 	 * Ack processing.
1964 	 */
1965 	switch (tp->t_state) {
1966 
1967 	/*
1968 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1969 	 * ESTABLISHED state and continue processing, otherwise
1970 	 * send an RST.
1971 	 */
1972 	case TCPS_SYN_RECEIVED:
1973 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
1974 		    SEQ_GT(th->th_ack, tp->snd_max))
1975 			goto dropwithreset;
1976 		tcpstat.tcps_connects++;
1977 		soisconnected(so);
1978 		tcp_established(tp);
1979 		/* Do window scaling? */
1980 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1981 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1982 			tp->snd_scale = tp->requested_s_scale;
1983 			tp->rcv_scale = tp->request_r_scale;
1984 		}
1985 		TCP_REASS_LOCK(tp);
1986 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1987 		TCP_REASS_UNLOCK(tp);
1988 		tp->snd_wl1 = th->th_seq - 1;
1989 		/* fall into ... */
1990 
1991 	/*
1992 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1993 	 * ACKs.  If the ack is in the range
1994 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1995 	 * then advance tp->snd_una to th->th_ack and drop
1996 	 * data from the retransmission queue.  If this ACK reflects
1997 	 * more up to date window information we update our window information.
1998 	 */
1999 	case TCPS_ESTABLISHED:
2000 	case TCPS_FIN_WAIT_1:
2001 	case TCPS_FIN_WAIT_2:
2002 	case TCPS_CLOSE_WAIT:
2003 	case TCPS_CLOSING:
2004 	case TCPS_LAST_ACK:
2005 	case TCPS_TIME_WAIT:
2006 
2007 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2008 			if (tlen == 0 && tiwin == tp->snd_wnd) {
2009 				tcpstat.tcps_rcvdupack++;
2010 				/*
2011 				 * If we have outstanding data (other than
2012 				 * a window probe), this is a completely
2013 				 * duplicate ack (ie, window info didn't
2014 				 * change), the ack is the biggest we've
2015 				 * seen and we've seen exactly our rexmt
2016 				 * threshhold of them, assume a packet
2017 				 * has been dropped and retransmit it.
2018 				 * Kludge snd_nxt & the congestion
2019 				 * window so we send only this one
2020 				 * packet.
2021 				 *
2022 				 * We know we're losing at the current
2023 				 * window size so do congestion avoidance
2024 				 * (set ssthresh to half the current window
2025 				 * and pull our congestion window back to
2026 				 * the new ssthresh).
2027 				 *
2028 				 * Dup acks mean that packets have left the
2029 				 * network (they're now cached at the receiver)
2030 				 * so bump cwnd by the amount in the receiver
2031 				 * to keep a constant cwnd packets in the
2032 				 * network.
2033 				 */
2034 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2035 				    th->th_ack != tp->snd_una)
2036 					tp->t_dupacks = 0;
2037 				else if (++tp->t_dupacks == tcprexmtthresh) {
2038 					tcp_seq onxt = tp->snd_nxt;
2039 					u_int win =
2040 					    min(tp->snd_wnd, tp->snd_cwnd) /
2041 					    2 /	tp->t_segsz;
2042 					if (tcp_do_newreno && SEQ_LT(th->th_ack,
2043 					    tp->snd_recover)) {
2044 						/*
2045 						 * False fast retransmit after
2046 						 * timeout.  Do not cut window.
2047 						 */
2048 						tp->snd_cwnd += tp->t_segsz;
2049 						tp->t_dupacks = 0;
2050 						(void) tcp_output(tp);
2051 						goto drop;
2052 					}
2053 
2054 					if (win < 2)
2055 						win = 2;
2056 					tp->snd_ssthresh = win * tp->t_segsz;
2057 					tp->snd_recover = tp->snd_max;
2058 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
2059 					tp->t_rtttime = 0;
2060 					tp->snd_nxt = th->th_ack;
2061 					tp->snd_cwnd = tp->t_segsz;
2062 					(void) tcp_output(tp);
2063 					tp->snd_cwnd = tp->snd_ssthresh +
2064 					       tp->t_segsz * tp->t_dupacks;
2065 					if (SEQ_GT(onxt, tp->snd_nxt))
2066 						tp->snd_nxt = onxt;
2067 					goto drop;
2068 				} else if (tp->t_dupacks > tcprexmtthresh) {
2069 					tp->snd_cwnd += tp->t_segsz;
2070 					(void) tcp_output(tp);
2071 					goto drop;
2072 				}
2073 			} else
2074 				tp->t_dupacks = 0;
2075 			break;
2076 		}
2077 		/*
2078 		 * If the congestion window was inflated to account
2079 		 * for the other side's cached packets, retract it.
2080 		 */
2081 		if (tcp_do_newreno == 0) {
2082 			if (tp->t_dupacks >= tcprexmtthresh &&
2083 			    tp->snd_cwnd > tp->snd_ssthresh)
2084 				tp->snd_cwnd = tp->snd_ssthresh;
2085 			tp->t_dupacks = 0;
2086 		} else if (tp->t_dupacks >= tcprexmtthresh &&
2087 			   tcp_newreno(tp, th) == 0) {
2088 			tp->snd_cwnd = tp->snd_ssthresh;
2089 			/*
2090 			 * Window inflation should have left us with approx.
2091 			 * snd_ssthresh outstanding data.  But in case we
2092 			 * would be inclined to send a burst, better to do
2093 			 * it via the slow start mechanism.
2094 			 */
2095 			if (SEQ_SUB(tp->snd_max, th->th_ack) < tp->snd_ssthresh)
2096 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, th->th_ack)
2097 				    + tp->t_segsz;
2098 			tp->t_dupacks = 0;
2099 		}
2100 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2101 			tcpstat.tcps_rcvacktoomuch++;
2102 			goto dropafterack;
2103 		}
2104 		acked = th->th_ack - tp->snd_una;
2105 		tcpstat.tcps_rcvackpack++;
2106 		tcpstat.tcps_rcvackbyte += acked;
2107 
2108 		/*
2109 		 * If we have a timestamp reply, update smoothed
2110 		 * round trip time.  If no timestamp is present but
2111 		 * transmit timer is running and timed sequence
2112 		 * number was acked, update smoothed round trip time.
2113 		 * Since we now have an rtt measurement, cancel the
2114 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2115 		 * Recompute the initial retransmit timer.
2116 		 */
2117 		if (opti.ts_present && opti.ts_ecr)
2118 			tcp_xmit_timer(tp, TCP_TIMESTAMP(tp) - opti.ts_ecr + 1);
2119 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2120 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2121 
2122 		/*
2123 		 * If all outstanding data is acked, stop retransmit
2124 		 * timer and remember to restart (more output or persist).
2125 		 * If there is more data to be acked, restart retransmit
2126 		 * timer, using current (possibly backed-off) value.
2127 		 */
2128 		if (th->th_ack == tp->snd_max) {
2129 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2130 			needoutput = 1;
2131 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2132 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2133 		/*
2134 		 * When new data is acked, open the congestion window.
2135 		 * If the window gives us less than ssthresh packets
2136 		 * in flight, open exponentially (segsz per packet).
2137 		 * Otherwise open linearly: segsz per window
2138 		 * (segsz^2 / cwnd per packet), plus a constant
2139 		 * fraction of a packet (segsz/8) to help larger windows
2140 		 * open quickly enough.
2141 		 */
2142 		{
2143 		u_int cw = tp->snd_cwnd;
2144 		u_int incr = tp->t_segsz;
2145 
2146 		if (cw > tp->snd_ssthresh)
2147 			incr = incr * incr / cw;
2148 		if (tcp_do_newreno == 0 || SEQ_GEQ(th->th_ack, tp->snd_recover))
2149 			tp->snd_cwnd = min(cw + incr,
2150 			    TCP_MAXWIN << tp->snd_scale);
2151 		}
2152 		ND6_HINT(tp);
2153 		if (acked > so->so_snd.sb_cc) {
2154 			tp->snd_wnd -= so->so_snd.sb_cc;
2155 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2156 			ourfinisacked = 1;
2157 		} else {
2158 			sbdrop(&so->so_snd, acked);
2159 			tp->snd_wnd -= acked;
2160 			ourfinisacked = 0;
2161 		}
2162 		sowwakeup(so);
2163 		/*
2164 		 * We want snd_recover to track snd_una to
2165 		 * avoid sequence wraparound problems for
2166 		 * very large transfers.
2167 		 */
2168 		tp->snd_una = tp->snd_recover = th->th_ack;
2169 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2170 			tp->snd_nxt = tp->snd_una;
2171 
2172 		switch (tp->t_state) {
2173 
2174 		/*
2175 		 * In FIN_WAIT_1 STATE in addition to the processing
2176 		 * for the ESTABLISHED state if our FIN is now acknowledged
2177 		 * then enter FIN_WAIT_2.
2178 		 */
2179 		case TCPS_FIN_WAIT_1:
2180 			if (ourfinisacked) {
2181 				/*
2182 				 * If we can't receive any more
2183 				 * data, then closing user can proceed.
2184 				 * Starting the timer is contrary to the
2185 				 * specification, but if we don't get a FIN
2186 				 * we'll hang forever.
2187 				 */
2188 				if (so->so_state & SS_CANTRCVMORE) {
2189 					soisdisconnected(so);
2190 					if (tcp_maxidle > 0)
2191 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2192 						    tcp_maxidle);
2193 				}
2194 				tp->t_state = TCPS_FIN_WAIT_2;
2195 			}
2196 			break;
2197 
2198 	 	/*
2199 		 * In CLOSING STATE in addition to the processing for
2200 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2201 		 * then enter the TIME-WAIT state, otherwise ignore
2202 		 * the segment.
2203 		 */
2204 		case TCPS_CLOSING:
2205 			if (ourfinisacked) {
2206 				tp->t_state = TCPS_TIME_WAIT;
2207 				tcp_canceltimers(tp);
2208 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2209 				soisdisconnected(so);
2210 			}
2211 			break;
2212 
2213 		/*
2214 		 * In LAST_ACK, we may still be waiting for data to drain
2215 		 * and/or to be acked, as well as for the ack of our FIN.
2216 		 * If our FIN is now acknowledged, delete the TCB,
2217 		 * enter the closed state and return.
2218 		 */
2219 		case TCPS_LAST_ACK:
2220 			if (ourfinisacked) {
2221 				tp = tcp_close(tp);
2222 				goto drop;
2223 			}
2224 			break;
2225 
2226 		/*
2227 		 * In TIME_WAIT state the only thing that should arrive
2228 		 * is a retransmission of the remote FIN.  Acknowledge
2229 		 * it and restart the finack timer.
2230 		 */
2231 		case TCPS_TIME_WAIT:
2232 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2233 			goto dropafterack;
2234 		}
2235 	}
2236 
2237 step6:
2238 	/*
2239 	 * Update window information.
2240 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2241 	 */
2242 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2243 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2244 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2245 		/* keep track of pure window updates */
2246 		if (tlen == 0 &&
2247 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2248 			tcpstat.tcps_rcvwinupd++;
2249 		tp->snd_wnd = tiwin;
2250 		tp->snd_wl1 = th->th_seq;
2251 		tp->snd_wl2 = th->th_ack;
2252 		if (tp->snd_wnd > tp->max_sndwnd)
2253 			tp->max_sndwnd = tp->snd_wnd;
2254 		needoutput = 1;
2255 	}
2256 
2257 	/*
2258 	 * Process segments with URG.
2259 	 */
2260 	if ((tiflags & TH_URG) && th->th_urp &&
2261 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2262 		/*
2263 		 * This is a kludge, but if we receive and accept
2264 		 * random urgent pointers, we'll crash in
2265 		 * soreceive.  It's hard to imagine someone
2266 		 * actually wanting to send this much urgent data.
2267 		 */
2268 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2269 			th->th_urp = 0;			/* XXX */
2270 			tiflags &= ~TH_URG;		/* XXX */
2271 			goto dodata;			/* XXX */
2272 		}
2273 		/*
2274 		 * If this segment advances the known urgent pointer,
2275 		 * then mark the data stream.  This should not happen
2276 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2277 		 * a FIN has been received from the remote side.
2278 		 * In these states we ignore the URG.
2279 		 *
2280 		 * According to RFC961 (Assigned Protocols),
2281 		 * the urgent pointer points to the last octet
2282 		 * of urgent data.  We continue, however,
2283 		 * to consider it to indicate the first octet
2284 		 * of data past the urgent section as the original
2285 		 * spec states (in one of two places).
2286 		 */
2287 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2288 			tp->rcv_up = th->th_seq + th->th_urp;
2289 			so->so_oobmark = so->so_rcv.sb_cc +
2290 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2291 			if (so->so_oobmark == 0)
2292 				so->so_state |= SS_RCVATMARK;
2293 			sohasoutofband(so);
2294 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2295 		}
2296 		/*
2297 		 * Remove out of band data so doesn't get presented to user.
2298 		 * This can happen independent of advancing the URG pointer,
2299 		 * but if two URG's are pending at once, some out-of-band
2300 		 * data may creep in... ick.
2301 		 */
2302 		if (th->th_urp <= (u_int16_t) tlen
2303 #ifdef SO_OOBINLINE
2304 		     && (so->so_options & SO_OOBINLINE) == 0
2305 #endif
2306 		     )
2307 			tcp_pulloutofband(so, th, m, hdroptlen);
2308 	} else
2309 		/*
2310 		 * If no out of band data is expected,
2311 		 * pull receive urgent pointer along
2312 		 * with the receive window.
2313 		 */
2314 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2315 			tp->rcv_up = tp->rcv_nxt;
2316 dodata:							/* XXX */
2317 
2318 	/*
2319 	 * Process the segment text, merging it into the TCP sequencing queue,
2320 	 * and arranging for acknowledgement of receipt if necessary.
2321 	 * This process logically involves adjusting tp->rcv_wnd as data
2322 	 * is presented to the user (this happens in tcp_usrreq.c,
2323 	 * case PRU_RCVD).  If a FIN has already been received on this
2324 	 * connection then we just ignore the text.
2325 	 */
2326 	if ((tlen || (tiflags & TH_FIN)) &&
2327 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2328 		/*
2329 		 * Insert segment ti into reassembly queue of tcp with
2330 		 * control block tp.  Return TH_FIN if reassembly now includes
2331 		 * a segment with FIN.  The macro form does the common case
2332 		 * inline (segment is the next to be received on an
2333 		 * established connection, and the queue is empty),
2334 		 * avoiding linkage into and removal from the queue and
2335 		 * repetition of various conversions.
2336 		 * Set DELACK for segments received in order, but ack
2337 		 * immediately when segments are out of order
2338 		 * (so fast retransmit can work).
2339 		 */
2340 		/* NOTE: this was TCP_REASS() macro, but used only once */
2341 		TCP_REASS_LOCK(tp);
2342 		if (th->th_seq == tp->rcv_nxt &&
2343 		    TAILQ_FIRST(&tp->segq) == NULL &&
2344 		    tp->t_state == TCPS_ESTABLISHED) {
2345 			TCP_SETUP_ACK(tp, th);
2346 			tp->rcv_nxt += tlen;
2347 			tiflags = th->th_flags & TH_FIN;
2348 			tcpstat.tcps_rcvpack++;
2349 			tcpstat.tcps_rcvbyte += tlen;
2350 			ND6_HINT(tp);
2351 			if (so->so_state & SS_CANTRCVMORE)
2352 				m_freem(m);
2353 			else {
2354 				m_adj(m, hdroptlen);
2355 				sbappendstream(&(so)->so_rcv, m);
2356 			}
2357 			sorwakeup(so);
2358 		} else {
2359 			m_adj(m, hdroptlen);
2360 			tiflags = tcp_reass(tp, th, m, &tlen);
2361 			tp->t_flags |= TF_ACKNOW;
2362 		}
2363 		TCP_REASS_UNLOCK(tp);
2364 
2365 		/*
2366 		 * Note the amount of data that peer has sent into
2367 		 * our window, in order to estimate the sender's
2368 		 * buffer size.
2369 		 */
2370 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2371 	} else {
2372 		m_freem(m);
2373 		m = NULL;
2374 		tiflags &= ~TH_FIN;
2375 	}
2376 
2377 	/*
2378 	 * If FIN is received ACK the FIN and let the user know
2379 	 * that the connection is closing.  Ignore a FIN received before
2380 	 * the connection is fully established.
2381 	 */
2382 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2383 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2384 			socantrcvmore(so);
2385 			tp->t_flags |= TF_ACKNOW;
2386 			tp->rcv_nxt++;
2387 		}
2388 		switch (tp->t_state) {
2389 
2390 	 	/*
2391 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2392 		 */
2393 		case TCPS_ESTABLISHED:
2394 			tp->t_state = TCPS_CLOSE_WAIT;
2395 			break;
2396 
2397 	 	/*
2398 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2399 		 * enter the CLOSING state.
2400 		 */
2401 		case TCPS_FIN_WAIT_1:
2402 			tp->t_state = TCPS_CLOSING;
2403 			break;
2404 
2405 	 	/*
2406 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2407 		 * starting the time-wait timer, turning off the other
2408 		 * standard timers.
2409 		 */
2410 		case TCPS_FIN_WAIT_2:
2411 			tp->t_state = TCPS_TIME_WAIT;
2412 			tcp_canceltimers(tp);
2413 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2414 			soisdisconnected(so);
2415 			break;
2416 
2417 		/*
2418 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2419 		 */
2420 		case TCPS_TIME_WAIT:
2421 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2422 			break;
2423 		}
2424 	}
2425 #ifdef TCP_DEBUG
2426 	if (so->so_options & SO_DEBUG)
2427 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2428 #endif
2429 
2430 	/*
2431 	 * Return any desired output.
2432 	 */
2433 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2434 		(void) tcp_output(tp);
2435 	if (tcp_saveti)
2436 		m_freem(tcp_saveti);
2437 	return;
2438 
2439 badsyn:
2440 	/*
2441 	 * Received a bad SYN.  Increment counters and dropwithreset.
2442 	 */
2443 	tcpstat.tcps_badsyn++;
2444 	tp = NULL;
2445 	goto dropwithreset;
2446 
2447 dropafterack:
2448 	/*
2449 	 * Generate an ACK dropping incoming segment if it occupies
2450 	 * sequence space, where the ACK reflects our state.
2451 	 */
2452 	if (tiflags & TH_RST)
2453 		goto drop;
2454 	m_freem(m);
2455 	tp->t_flags |= TF_ACKNOW;
2456 	(void) tcp_output(tp);
2457 	if (tcp_saveti)
2458 		m_freem(tcp_saveti);
2459 	return;
2460 
2461 dropwithreset_ratelim:
2462 	/*
2463 	 * We may want to rate-limit RSTs in certain situations,
2464 	 * particularly if we are sending an RST in response to
2465 	 * an attempt to connect to or otherwise communicate with
2466 	 * a port for which we have no socket.
2467 	 */
2468 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2469 	    tcp_rst_ppslim) == 0) {
2470 		/* XXX stat */
2471 		goto drop;
2472 	}
2473 	/* ...fall into dropwithreset... */
2474 
2475 dropwithreset:
2476 	/*
2477 	 * Generate a RST, dropping incoming segment.
2478 	 * Make ACK acceptable to originator of segment.
2479 	 */
2480 	if (tiflags & TH_RST)
2481 		goto drop;
2482 
2483 	switch (af) {
2484 #ifdef INET6
2485 	case AF_INET6:
2486 		/* For following calls to tcp_respond */
2487 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2488 			goto drop;
2489 		break;
2490 #endif /* INET6 */
2491 	case AF_INET:
2492 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2493 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2494 			goto drop;
2495 	}
2496 
2497     {
2498 	/*
2499 	 * need to recover version # field, which was overwritten on
2500 	 * ip_cksum computation.
2501 	 */
2502 	struct ip *sip;
2503 	sip = mtod(m, struct ip *);
2504 	switch (af) {
2505 #ifdef INET
2506 	case AF_INET:
2507 		sip->ip_v = 4;
2508 		break;
2509 #endif
2510 #ifdef INET6
2511 	case AF_INET6:
2512 		sip->ip_v = 6;
2513 		break;
2514 #endif
2515 	}
2516     }
2517 	if (tiflags & TH_ACK)
2518 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2519 	else {
2520 		if (tiflags & TH_SYN)
2521 			tlen++;
2522 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2523 		    TH_RST|TH_ACK);
2524 	}
2525 	if (tcp_saveti)
2526 		m_freem(tcp_saveti);
2527 	return;
2528 
2529 badcsum:
2530 	tcpstat.tcps_rcvbadsum++;
2531 drop:
2532 	/*
2533 	 * Drop space held by incoming segment and return.
2534 	 */
2535 	if (tp) {
2536 		if (tp->t_inpcb)
2537 			so = tp->t_inpcb->inp_socket;
2538 #ifdef INET6
2539 		else if (tp->t_in6pcb)
2540 			so = tp->t_in6pcb->in6p_socket;
2541 #endif
2542 		else
2543 			so = NULL;
2544 #ifdef TCP_DEBUG
2545 		if (so && (so->so_options & SO_DEBUG) != 0)
2546 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2547 #endif
2548 	}
2549 	if (tcp_saveti)
2550 		m_freem(tcp_saveti);
2551 	m_freem(m);
2552 	return;
2553 }
2554 
2555 void
2556 tcp_dooptions(tp, cp, cnt, th, oi)
2557 	struct tcpcb *tp;
2558 	u_char *cp;
2559 	int cnt;
2560 	struct tcphdr *th;
2561 	struct tcp_opt_info *oi;
2562 {
2563 	u_int16_t mss;
2564 	int opt, optlen;
2565 
2566 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2567 		opt = cp[0];
2568 		if (opt == TCPOPT_EOL)
2569 			break;
2570 		if (opt == TCPOPT_NOP)
2571 			optlen = 1;
2572 		else {
2573 			if (cnt < 2)
2574 				break;
2575 			optlen = cp[1];
2576 			if (optlen < 2 || optlen > cnt)
2577 				break;
2578 		}
2579 		switch (opt) {
2580 
2581 		default:
2582 			continue;
2583 
2584 		case TCPOPT_MAXSEG:
2585 			if (optlen != TCPOLEN_MAXSEG)
2586 				continue;
2587 			if (!(th->th_flags & TH_SYN))
2588 				continue;
2589 			bcopy(cp + 2, &mss, sizeof(mss));
2590 			oi->maxseg = ntohs(mss);
2591 			break;
2592 
2593 		case TCPOPT_WINDOW:
2594 			if (optlen != TCPOLEN_WINDOW)
2595 				continue;
2596 			if (!(th->th_flags & TH_SYN))
2597 				continue;
2598 			tp->t_flags |= TF_RCVD_SCALE;
2599 			tp->requested_s_scale = cp[2];
2600 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
2601 #if 0	/*XXX*/
2602 				char *p;
2603 
2604 				if (ip)
2605 					p = ntohl(ip->ip_src);
2606 #ifdef INET6
2607 				else if (ip6)
2608 					p = ip6_sprintf(&ip6->ip6_src);
2609 #endif
2610 				else
2611 					p = "(unknown)";
2612 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
2613 				    "assuming %d\n",
2614 				    tp->requested_s_scale, p,
2615 				    TCP_MAX_WINSHIFT);
2616 #else
2617 				log(LOG_ERR, "TCP: invalid wscale %d, "
2618 				    "assuming %d\n",
2619 				    tp->requested_s_scale,
2620 				    TCP_MAX_WINSHIFT);
2621 #endif
2622 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
2623 			}
2624 			break;
2625 
2626 		case TCPOPT_TIMESTAMP:
2627 			if (optlen != TCPOLEN_TIMESTAMP)
2628 				continue;
2629 			oi->ts_present = 1;
2630 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2631 			NTOHL(oi->ts_val);
2632 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2633 			NTOHL(oi->ts_ecr);
2634 
2635 			/*
2636 			 * A timestamp received in a SYN makes
2637 			 * it ok to send timestamp requests and replies.
2638 			 */
2639 			if (th->th_flags & TH_SYN) {
2640 				tp->t_flags |= TF_RCVD_TSTMP;
2641 				tp->ts_recent = oi->ts_val;
2642 				tp->ts_recent_age = TCP_TIMESTAMP(tp);
2643 			}
2644 			break;
2645 		case TCPOPT_SACK_PERMITTED:
2646 			if (optlen != TCPOLEN_SACK_PERMITTED)
2647 				continue;
2648 			if (!(th->th_flags & TH_SYN))
2649 				continue;
2650 			tp->t_flags &= ~TF_CANT_TXSACK;
2651 			break;
2652 
2653 		case TCPOPT_SACK:
2654 			if (tp->t_flags & TF_IGNR_RXSACK)
2655 				continue;
2656 			if (optlen % 8 != 2 || optlen < 10)
2657 				continue;
2658 			cp += 2;
2659 			optlen -= 2;
2660 			for (; optlen > 0; cp -= 8, optlen -= 8) {
2661 				tcp_seq lwe, rwe;
2662 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
2663 				NTOHL(lwe);
2664 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
2665 				NTOHL(rwe);
2666 				/* tcp_mark_sacked(tp, lwe, rwe); */
2667 			}
2668 			break;
2669 		}
2670 	}
2671 }
2672 
2673 /*
2674  * Pull out of band byte out of a segment so
2675  * it doesn't appear in the user's data queue.
2676  * It is still reflected in the segment length for
2677  * sequencing purposes.
2678  */
2679 void
2680 tcp_pulloutofband(so, th, m, off)
2681 	struct socket *so;
2682 	struct tcphdr *th;
2683 	struct mbuf *m;
2684 	int off;
2685 {
2686 	int cnt = off + th->th_urp - 1;
2687 
2688 	while (cnt >= 0) {
2689 		if (m->m_len > cnt) {
2690 			char *cp = mtod(m, caddr_t) + cnt;
2691 			struct tcpcb *tp = sototcpcb(so);
2692 
2693 			tp->t_iobc = *cp;
2694 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2695 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2696 			m->m_len--;
2697 			return;
2698 		}
2699 		cnt -= m->m_len;
2700 		m = m->m_next;
2701 		if (m == 0)
2702 			break;
2703 	}
2704 	panic("tcp_pulloutofband");
2705 }
2706 
2707 /*
2708  * Collect new round-trip time estimate
2709  * and update averages and current timeout.
2710  */
2711 void
2712 tcp_xmit_timer(tp, rtt)
2713 	struct tcpcb *tp;
2714 	uint32_t rtt;
2715 {
2716 	int32_t delta;
2717 
2718 	tcpstat.tcps_rttupdated++;
2719 	if (tp->t_srtt != 0) {
2720 		/*
2721 		 * srtt is stored as fixed point with 3 bits after the
2722 		 * binary point (i.e., scaled by 8).  The following magic
2723 		 * is equivalent to the smoothing algorithm in rfc793 with
2724 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2725 		 * point).  Adjust rtt to origin 0.
2726 		 */
2727 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2728 		if ((tp->t_srtt += delta) <= 0)
2729 			tp->t_srtt = 1 << 2;
2730 		/*
2731 		 * We accumulate a smoothed rtt variance (actually, a
2732 		 * smoothed mean difference), then set the retransmit
2733 		 * timer to smoothed rtt + 4 times the smoothed variance.
2734 		 * rttvar is stored as fixed point with 2 bits after the
2735 		 * binary point (scaled by 4).  The following is
2736 		 * equivalent to rfc793 smoothing with an alpha of .75
2737 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2738 		 * rfc793's wired-in beta.
2739 		 */
2740 		if (delta < 0)
2741 			delta = -delta;
2742 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2743 		if ((tp->t_rttvar += delta) <= 0)
2744 			tp->t_rttvar = 1 << 2;
2745 	} else {
2746 		/*
2747 		 * No rtt measurement yet - use the unsmoothed rtt.
2748 		 * Set the variance to half the rtt (so our first
2749 		 * retransmit happens at 3*rtt).
2750 		 */
2751 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2752 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2753 	}
2754 	tp->t_rtttime = 0;
2755 	tp->t_rxtshift = 0;
2756 
2757 	/*
2758 	 * the retransmit should happen at rtt + 4 * rttvar.
2759 	 * Because of the way we do the smoothing, srtt and rttvar
2760 	 * will each average +1/2 tick of bias.  When we compute
2761 	 * the retransmit timer, we want 1/2 tick of rounding and
2762 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2763 	 * firing of the timer.  The bias will give us exactly the
2764 	 * 1.5 tick we need.  But, because the bias is
2765 	 * statistical, we have to test that we don't drop below
2766 	 * the minimum feasible timer (which is 2 ticks).
2767 	 */
2768 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2769 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2770 
2771 	/*
2772 	 * We received an ack for a packet that wasn't retransmitted;
2773 	 * it is probably safe to discard any error indications we've
2774 	 * received recently.  This isn't quite right, but close enough
2775 	 * for now (a route might have failed after we sent a segment,
2776 	 * and the return path might not be symmetrical).
2777 	 */
2778 	tp->t_softerror = 0;
2779 }
2780 
2781 /*
2782  * Checks for partial ack.  If partial ack arrives, force the retransmission
2783  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
2784  * 1.  By setting snd_nxt to th_ack, this forces retransmission timer to
2785  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
2786  */
2787 int
2788 tcp_newreno(tp, th)
2789 	struct tcpcb *tp;
2790 	struct tcphdr *th;
2791 {
2792 	tcp_seq onxt = tp->snd_nxt;
2793 	u_long ocwnd = tp->snd_cwnd;
2794 
2795 	if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2796 		/*
2797 		 * snd_una has not yet been updated and the socket's send
2798 		 * buffer has not yet drained off the ACK'd data, so we
2799 		 * have to leave snd_una as it was to get the correct data
2800 		 * offset in tcp_output().
2801 		 */
2802 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2803 	        tp->t_rtttime = 0;
2804 	        tp->snd_nxt = th->th_ack;
2805 		/*
2806 		 * Set snd_cwnd to one segment beyond ACK'd offset.  snd_una
2807 		 * is not yet updated when we're called.
2808 		 */
2809 		tp->snd_cwnd = tp->t_segsz + (th->th_ack - tp->snd_una);
2810 	        (void) tcp_output(tp);
2811 	        tp->snd_cwnd = ocwnd;
2812 	        if (SEQ_GT(onxt, tp->snd_nxt))
2813 	                tp->snd_nxt = onxt;
2814 	        /*
2815 	         * Partial window deflation.  Relies on fact that tp->snd_una
2816 	         * not updated yet.
2817 	         */
2818 	        tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_segsz);
2819 	        return 1;
2820 	}
2821 	return 0;
2822 }
2823 
2824 
2825 /*
2826  * TCP compressed state engine.  Currently used to hold compressed
2827  * state for SYN_RECEIVED.
2828  */
2829 
2830 u_long	syn_cache_count;
2831 u_int32_t syn_hash1, syn_hash2;
2832 
2833 #define SYN_HASH(sa, sp, dp) \
2834 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
2835 				     ((u_int32_t)(sp)))^syn_hash2)))
2836 #ifndef INET6
2837 #define	SYN_HASHALL(hash, src, dst) \
2838 do {									\
2839 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
2840 		((struct sockaddr_in *)(src))->sin_port,		\
2841 		((struct sockaddr_in *)(dst))->sin_port);		\
2842 } while (/*CONSTCOND*/ 0)
2843 #else
2844 #define SYN_HASH6(sa, sp, dp) \
2845 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
2846 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
2847 	 & 0x7fffffff)
2848 
2849 #define SYN_HASHALL(hash, src, dst) \
2850 do {									\
2851 	switch ((src)->sa_family) {					\
2852 	case AF_INET:							\
2853 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
2854 			((struct sockaddr_in *)(src))->sin_port,	\
2855 			((struct sockaddr_in *)(dst))->sin_port);	\
2856 		break;							\
2857 	case AF_INET6:							\
2858 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
2859 			((struct sockaddr_in6 *)(src))->sin6_port,	\
2860 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
2861 		break;							\
2862 	default:							\
2863 		hash = 0;						\
2864 	}								\
2865 } while (/*CONSTCOND*/0)
2866 #endif /* INET6 */
2867 
2868 #define	SYN_CACHE_RM(sc)						\
2869 do {									\
2870 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
2871 	    (sc), sc_bucketq);						\
2872 	(sc)->sc_tp = NULL;						\
2873 	LIST_REMOVE((sc), sc_tpq);					\
2874 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
2875 	callout_stop(&(sc)->sc_timer);					\
2876 	syn_cache_count--;						\
2877 } while (/*CONSTCOND*/0)
2878 
2879 #define	SYN_CACHE_PUT(sc)						\
2880 do {									\
2881 	if ((sc)->sc_ipopts)						\
2882 		(void) m_free((sc)->sc_ipopts);				\
2883 	if ((sc)->sc_route4.ro_rt != NULL)				\
2884 		RTFREE((sc)->sc_route4.ro_rt);				\
2885 	pool_put(&syn_cache_pool, (sc));				\
2886 } while (/*CONSTCOND*/0)
2887 
2888 struct pool syn_cache_pool;
2889 
2890 /*
2891  * We don't estimate RTT with SYNs, so each packet starts with the default
2892  * RTT and each timer step has a fixed timeout value.
2893  */
2894 #define	SYN_CACHE_TIMER_ARM(sc)						\
2895 do {									\
2896 	TCPT_RANGESET((sc)->sc_rxtcur,					\
2897 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
2898 	    TCPTV_REXMTMAX);						\
2899 	callout_reset(&(sc)->sc_timer,					\
2900 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
2901 } while (/*CONSTCOND*/0)
2902 
2903 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
2904 
2905 void
2906 syn_cache_init()
2907 {
2908 	int i;
2909 
2910 	/* Initialize the hash buckets. */
2911 	for (i = 0; i < tcp_syn_cache_size; i++)
2912 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
2913 
2914 	/* Initialize the syn cache pool. */
2915 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
2916 	    "synpl", NULL);
2917 }
2918 
2919 void
2920 syn_cache_insert(sc, tp)
2921 	struct syn_cache *sc;
2922 	struct tcpcb *tp;
2923 {
2924 	struct syn_cache_head *scp;
2925 	struct syn_cache *sc2;
2926 	int s;
2927 
2928 	/*
2929 	 * If there are no entries in the hash table, reinitialize
2930 	 * the hash secrets.
2931 	 */
2932 	if (syn_cache_count == 0) {
2933 		struct timeval tv;
2934 		microtime(&tv);
2935 		syn_hash1 = arc4random() ^ (u_long)&sc;
2936 		syn_hash2 = arc4random() ^ tv.tv_usec;
2937 	}
2938 
2939 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
2940 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
2941 	scp = &tcp_syn_cache[sc->sc_bucketidx];
2942 
2943 	/*
2944 	 * Make sure that we don't overflow the per-bucket
2945 	 * limit or the total cache size limit.
2946 	 */
2947 	s = splsoftnet();
2948 	if (scp->sch_length >= tcp_syn_bucket_limit) {
2949 		tcpstat.tcps_sc_bucketoverflow++;
2950 		/*
2951 		 * The bucket is full.  Toss the oldest element in the
2952 		 * bucket.  This will be the first entry in the bucket.
2953 		 */
2954 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
2955 #ifdef DIAGNOSTIC
2956 		/*
2957 		 * This should never happen; we should always find an
2958 		 * entry in our bucket.
2959 		 */
2960 		if (sc2 == NULL)
2961 			panic("syn_cache_insert: bucketoverflow: impossible");
2962 #endif
2963 		SYN_CACHE_RM(sc2);
2964 		SYN_CACHE_PUT(sc2);
2965 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
2966 		struct syn_cache_head *scp2, *sce;
2967 
2968 		tcpstat.tcps_sc_overflowed++;
2969 		/*
2970 		 * The cache is full.  Toss the oldest entry in the
2971 		 * first non-empty bucket we can find.
2972 		 *
2973 		 * XXX We would really like to toss the oldest
2974 		 * entry in the cache, but we hope that this
2975 		 * condition doesn't happen very often.
2976 		 */
2977 		scp2 = scp;
2978 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
2979 			sce = &tcp_syn_cache[tcp_syn_cache_size];
2980 			for (++scp2; scp2 != scp; scp2++) {
2981 				if (scp2 >= sce)
2982 					scp2 = &tcp_syn_cache[0];
2983 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
2984 					break;
2985 			}
2986 #ifdef DIAGNOSTIC
2987 			/*
2988 			 * This should never happen; we should always find a
2989 			 * non-empty bucket.
2990 			 */
2991 			if (scp2 == scp)
2992 				panic("syn_cache_insert: cacheoverflow: "
2993 				    "impossible");
2994 #endif
2995 		}
2996 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
2997 		SYN_CACHE_RM(sc2);
2998 		SYN_CACHE_PUT(sc2);
2999 	}
3000 
3001 	/*
3002 	 * Initialize the entry's timer.
3003 	 */
3004 	sc->sc_rxttot = 0;
3005 	sc->sc_rxtshift = 0;
3006 	SYN_CACHE_TIMER_ARM(sc);
3007 
3008 	/* Link it from tcpcb entry */
3009 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3010 
3011 	/* Put it into the bucket. */
3012 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3013 	scp->sch_length++;
3014 	syn_cache_count++;
3015 
3016 	tcpstat.tcps_sc_added++;
3017 	splx(s);
3018 }
3019 
3020 /*
3021  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3022  * If we have retransmitted an entry the maximum number of times, expire
3023  * that entry.
3024  */
3025 void
3026 syn_cache_timer(void *arg)
3027 {
3028 	struct syn_cache *sc = arg;
3029 	int s;
3030 
3031 	s = splsoftnet();
3032 
3033 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3034 		/* Drop it -- too many retransmissions. */
3035 		goto dropit;
3036 	}
3037 
3038 	/*
3039 	 * Compute the total amount of time this entry has
3040 	 * been on a queue.  If this entry has been on longer
3041 	 * than the keep alive timer would allow, expire it.
3042 	 */
3043 	sc->sc_rxttot += sc->sc_rxtcur;
3044 	if (sc->sc_rxttot >= TCPTV_KEEP_INIT)
3045 		goto dropit;
3046 
3047 	tcpstat.tcps_sc_retransmitted++;
3048 	(void) syn_cache_respond(sc, NULL);
3049 
3050 	/* Advance the timer back-off. */
3051 	sc->sc_rxtshift++;
3052 	SYN_CACHE_TIMER_ARM(sc);
3053 
3054 	splx(s);
3055 	return;
3056 
3057  dropit:
3058 	tcpstat.tcps_sc_timed_out++;
3059 	SYN_CACHE_RM(sc);
3060 	SYN_CACHE_PUT(sc);
3061 	splx(s);
3062 }
3063 
3064 /*
3065  * Remove syn cache created by the specified tcb entry,
3066  * because this does not make sense to keep them
3067  * (if there's no tcb entry, syn cache entry will never be used)
3068  */
3069 void
3070 syn_cache_cleanup(tp)
3071 	struct tcpcb *tp;
3072 {
3073 	struct syn_cache *sc, *nsc;
3074 	int s;
3075 
3076 	s = splsoftnet();
3077 
3078 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3079 		nsc = LIST_NEXT(sc, sc_tpq);
3080 
3081 #ifdef DIAGNOSTIC
3082 		if (sc->sc_tp != tp)
3083 			panic("invalid sc_tp in syn_cache_cleanup");
3084 #endif
3085 		SYN_CACHE_RM(sc);
3086 		SYN_CACHE_PUT(sc);
3087 	}
3088 	/* just for safety */
3089 	LIST_INIT(&tp->t_sc);
3090 
3091 	splx(s);
3092 }
3093 
3094 /*
3095  * Find an entry in the syn cache.
3096  */
3097 struct syn_cache *
3098 syn_cache_lookup(src, dst, headp)
3099 	struct sockaddr *src;
3100 	struct sockaddr *dst;
3101 	struct syn_cache_head **headp;
3102 {
3103 	struct syn_cache *sc;
3104 	struct syn_cache_head *scp;
3105 	u_int32_t hash;
3106 	int s;
3107 
3108 	SYN_HASHALL(hash, src, dst);
3109 
3110 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3111 	*headp = scp;
3112 	s = splsoftnet();
3113 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3114 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3115 		if (sc->sc_hash != hash)
3116 			continue;
3117 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3118 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3119 			splx(s);
3120 			return (sc);
3121 		}
3122 	}
3123 	splx(s);
3124 	return (NULL);
3125 }
3126 
3127 /*
3128  * This function gets called when we receive an ACK for a
3129  * socket in the LISTEN state.  We look up the connection
3130  * in the syn cache, and if its there, we pull it out of
3131  * the cache and turn it into a full-blown connection in
3132  * the SYN-RECEIVED state.
3133  *
3134  * The return values may not be immediately obvious, and their effects
3135  * can be subtle, so here they are:
3136  *
3137  *	NULL	SYN was not found in cache; caller should drop the
3138  *		packet and send an RST.
3139  *
3140  *	-1	We were unable to create the new connection, and are
3141  *		aborting it.  An ACK,RST is being sent to the peer
3142  *		(unless we got screwey sequence numbners; see below),
3143  *		because the 3-way handshake has been completed.  Caller
3144  *		should not free the mbuf, since we may be using it.  If
3145  *		we are not, we will free it.
3146  *
3147  *	Otherwise, the return value is a pointer to the new socket
3148  *	associated with the connection.
3149  */
3150 struct socket *
3151 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3152 	struct sockaddr *src;
3153 	struct sockaddr *dst;
3154 	struct tcphdr *th;
3155 	unsigned int hlen, tlen;
3156 	struct socket *so;
3157 	struct mbuf *m;
3158 {
3159 	struct syn_cache *sc;
3160 	struct syn_cache_head *scp;
3161 	struct inpcb *inp = NULL;
3162 #ifdef INET6
3163 	struct in6pcb *in6p = NULL;
3164 #endif
3165 	struct tcpcb *tp = 0;
3166 	struct mbuf *am;
3167 	int s;
3168 	struct socket *oso;
3169 
3170 	s = splsoftnet();
3171 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3172 		splx(s);
3173 		return (NULL);
3174 	}
3175 
3176 	/*
3177 	 * Verify the sequence and ack numbers.  Try getting the correct
3178 	 * response again.
3179 	 */
3180 	if ((th->th_ack != sc->sc_iss + 1) ||
3181 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3182 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3183 		(void) syn_cache_respond(sc, m);
3184 		splx(s);
3185 		return ((struct socket *)(-1));
3186 	}
3187 
3188 	/* Remove this cache entry */
3189 	SYN_CACHE_RM(sc);
3190 	splx(s);
3191 
3192 	/*
3193 	 * Ok, create the full blown connection, and set things up
3194 	 * as they would have been set up if we had created the
3195 	 * connection when the SYN arrived.  If we can't create
3196 	 * the connection, abort it.
3197 	 */
3198 	/*
3199 	 * inp still has the OLD in_pcb stuff, set the
3200 	 * v6-related flags on the new guy, too.   This is
3201 	 * done particularly for the case where an AF_INET6
3202 	 * socket is bound only to a port, and a v4 connection
3203 	 * comes in on that port.
3204 	 * we also copy the flowinfo from the original pcb
3205 	 * to the new one.
3206 	 */
3207 	oso = so;
3208 	so = sonewconn(so, SS_ISCONNECTED);
3209 	if (so == NULL)
3210 		goto resetandabort;
3211 
3212 	switch (so->so_proto->pr_domain->dom_family) {
3213 #ifdef INET
3214 	case AF_INET:
3215 		inp = sotoinpcb(so);
3216 		break;
3217 #endif
3218 #ifdef INET6
3219 	case AF_INET6:
3220 		in6p = sotoin6pcb(so);
3221 		break;
3222 #endif
3223 	}
3224 	switch (src->sa_family) {
3225 #ifdef INET
3226 	case AF_INET:
3227 		if (inp) {
3228 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3229 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3230 			inp->inp_options = ip_srcroute();
3231 			in_pcbstate(inp, INP_BOUND);
3232 			if (inp->inp_options == NULL) {
3233 				inp->inp_options = sc->sc_ipopts;
3234 				sc->sc_ipopts = NULL;
3235 			}
3236 		}
3237 #ifdef INET6
3238 		else if (in6p) {
3239 			/* IPv4 packet to AF_INET6 socket */
3240 			bzero(&in6p->in6p_laddr, sizeof(in6p->in6p_laddr));
3241 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3242 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3243 				&in6p->in6p_laddr.s6_addr32[3],
3244 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3245 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3246 			in6totcpcb(in6p)->t_family = AF_INET;
3247 		}
3248 #endif
3249 		break;
3250 #endif
3251 #ifdef INET6
3252 	case AF_INET6:
3253 		if (in6p) {
3254 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3255 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3256 #if 0
3257 			in6p->in6p_flowinfo = ip6->ip6_flow & IPV6_FLOWINFO_MASK;
3258 			/*inp->inp_options = ip6_srcroute();*/ /* soon. */
3259 #endif
3260 		}
3261 		break;
3262 #endif
3263 	}
3264 #ifdef INET6
3265 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3266 		struct in6pcb *oin6p = sotoin6pcb(oso);
3267 		/* inherit socket options from the listening socket */
3268 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3269 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3270 			m_freem(in6p->in6p_options);
3271 			in6p->in6p_options = 0;
3272 		}
3273 		ip6_savecontrol(in6p, &in6p->in6p_options,
3274 			mtod(m, struct ip6_hdr *), m);
3275 	}
3276 #endif
3277 
3278 #ifdef IPSEC
3279 	/*
3280 	 * we make a copy of policy, instead of sharing the policy,
3281 	 * for better behavior in terms of SA lookup and dead SA removal.
3282 	 */
3283 	if (inp) {
3284 		/* copy old policy into new socket's */
3285 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3286 			printf("tcp_input: could not copy policy\n");
3287 	}
3288 #ifdef INET6
3289 	else if (in6p) {
3290 		/* copy old policy into new socket's */
3291 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3292 		    in6p->in6p_sp))
3293 			printf("tcp_input: could not copy policy\n");
3294 	}
3295 #endif
3296 #endif
3297 
3298 	/*
3299 	 * Give the new socket our cached route reference.
3300 	 */
3301 	if (inp)
3302 		inp->inp_route = sc->sc_route4;		/* struct assignment */
3303 #ifdef INET6
3304 	else
3305 		in6p->in6p_route = sc->sc_route6;
3306 #endif
3307 	sc->sc_route4.ro_rt = NULL;
3308 
3309 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3310 	if (am == NULL)
3311 		goto resetandabort;
3312 	MCLAIM(am, &tcp_mowner);
3313 	am->m_len = src->sa_len;
3314 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3315 	if (inp) {
3316 		if (in_pcbconnect(inp, am)) {
3317 			(void) m_free(am);
3318 			goto resetandabort;
3319 		}
3320 	}
3321 #ifdef INET6
3322 	else if (in6p) {
3323 		if (src->sa_family == AF_INET) {
3324 			/* IPv4 packet to AF_INET6 socket */
3325 			struct sockaddr_in6 *sin6;
3326 			sin6 = mtod(am, struct sockaddr_in6 *);
3327 			am->m_len = sizeof(*sin6);
3328 			bzero(sin6, sizeof(*sin6));
3329 			sin6->sin6_family = AF_INET6;
3330 			sin6->sin6_len = sizeof(*sin6);
3331 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3332 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3333 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3334 				&sin6->sin6_addr.s6_addr32[3],
3335 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3336 		}
3337 		if (in6_pcbconnect(in6p, am)) {
3338 			(void) m_free(am);
3339 			goto resetandabort;
3340 		}
3341 	}
3342 #endif
3343 	else {
3344 		(void) m_free(am);
3345 		goto resetandabort;
3346 	}
3347 	(void) m_free(am);
3348 
3349 	if (inp)
3350 		tp = intotcpcb(inp);
3351 #ifdef INET6
3352 	else if (in6p)
3353 		tp = in6totcpcb(in6p);
3354 #endif
3355 	else
3356 		tp = NULL;
3357 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3358 	if (sc->sc_request_r_scale != 15) {
3359 		tp->requested_s_scale = sc->sc_requested_s_scale;
3360 		tp->request_r_scale = sc->sc_request_r_scale;
3361 		tp->snd_scale = sc->sc_requested_s_scale;
3362 		tp->rcv_scale = sc->sc_request_r_scale;
3363 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3364 	}
3365 	if (sc->sc_flags & SCF_TIMESTAMP)
3366 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3367 	tp->ts_timebase = sc->sc_timebase;
3368 
3369 	tp->t_template = tcp_template(tp);
3370 	if (tp->t_template == 0) {
3371 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3372 		so = NULL;
3373 		m_freem(m);
3374 		goto abort;
3375 	}
3376 
3377 	tp->iss = sc->sc_iss;
3378 	tp->irs = sc->sc_irs;
3379 	tcp_sendseqinit(tp);
3380 	tcp_rcvseqinit(tp);
3381 	tp->t_state = TCPS_SYN_RECEIVED;
3382 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
3383 	tcpstat.tcps_accepts++;
3384 
3385 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3386 	tp->t_ourmss = sc->sc_ourmaxseg;
3387 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3388 
3389 	/*
3390 	 * Initialize the initial congestion window.  If we
3391 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3392 	 * to 1 segment (i.e. the Loss Window).
3393 	 */
3394 	if (sc->sc_rxtshift)
3395 		tp->snd_cwnd = tp->t_peermss;
3396 	else {
3397 		int ss = tcp_init_win;
3398 #ifdef INET
3399 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3400 			ss = tcp_init_win_local;
3401 #endif
3402 #ifdef INET6
3403 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3404 			ss = tcp_init_win_local;
3405 #endif
3406 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3407 	}
3408 
3409 	tcp_rmx_rtt(tp);
3410 	tp->snd_wl1 = sc->sc_irs;
3411 	tp->rcv_up = sc->sc_irs + 1;
3412 
3413 	/*
3414 	 * This is what whould have happened in tcp_output() when
3415 	 * the SYN,ACK was sent.
3416 	 */
3417 	tp->snd_up = tp->snd_una;
3418 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3419 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3420 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3421 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3422 	tp->last_ack_sent = tp->rcv_nxt;
3423 
3424 	tcpstat.tcps_sc_completed++;
3425 	SYN_CACHE_PUT(sc);
3426 	return (so);
3427 
3428 resetandabort:
3429 	(void) tcp_respond(NULL, m, m, th,
3430 			   th->th_seq + tlen, (tcp_seq)0, TH_RST|TH_ACK);
3431 abort:
3432 	if (so != NULL)
3433 		(void) soabort(so);
3434 	SYN_CACHE_PUT(sc);
3435 	tcpstat.tcps_sc_aborted++;
3436 	return ((struct socket *)(-1));
3437 }
3438 
3439 /*
3440  * This function is called when we get a RST for a
3441  * non-existent connection, so that we can see if the
3442  * connection is in the syn cache.  If it is, zap it.
3443  */
3444 
3445 void
3446 syn_cache_reset(src, dst, th)
3447 	struct sockaddr *src;
3448 	struct sockaddr *dst;
3449 	struct tcphdr *th;
3450 {
3451 	struct syn_cache *sc;
3452 	struct syn_cache_head *scp;
3453 	int s = splsoftnet();
3454 
3455 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3456 		splx(s);
3457 		return;
3458 	}
3459 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3460 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3461 		splx(s);
3462 		return;
3463 	}
3464 	SYN_CACHE_RM(sc);
3465 	splx(s);
3466 	tcpstat.tcps_sc_reset++;
3467 	SYN_CACHE_PUT(sc);
3468 }
3469 
3470 void
3471 syn_cache_unreach(src, dst, th)
3472 	struct sockaddr *src;
3473 	struct sockaddr *dst;
3474 	struct tcphdr *th;
3475 {
3476 	struct syn_cache *sc;
3477 	struct syn_cache_head *scp;
3478 	int s;
3479 
3480 	s = splsoftnet();
3481 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3482 		splx(s);
3483 		return;
3484 	}
3485 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3486 	if (ntohl (th->th_seq) != sc->sc_iss) {
3487 		splx(s);
3488 		return;
3489 	}
3490 
3491 	/*
3492 	 * If we've rertransmitted 3 times and this is our second error,
3493 	 * we remove the entry.  Otherwise, we allow it to continue on.
3494 	 * This prevents us from incorrectly nuking an entry during a
3495 	 * spurious network outage.
3496 	 *
3497 	 * See tcp_notify().
3498 	 */
3499 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3500 		sc->sc_flags |= SCF_UNREACH;
3501 		splx(s);
3502 		return;
3503 	}
3504 
3505 	SYN_CACHE_RM(sc);
3506 	splx(s);
3507 	tcpstat.tcps_sc_unreach++;
3508 	SYN_CACHE_PUT(sc);
3509 }
3510 
3511 /*
3512  * Given a LISTEN socket and an inbound SYN request, add
3513  * this to the syn cache, and send back a segment:
3514  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3515  * to the source.
3516  *
3517  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3518  * Doing so would require that we hold onto the data and deliver it
3519  * to the application.  However, if we are the target of a SYN-flood
3520  * DoS attack, an attacker could send data which would eventually
3521  * consume all available buffer space if it were ACKed.  By not ACKing
3522  * the data, we avoid this DoS scenario.
3523  */
3524 
3525 int
3526 syn_cache_add(src, dst, th, hlen, so, m, optp, optlen, oi)
3527 	struct sockaddr *src;
3528 	struct sockaddr *dst;
3529 	struct tcphdr *th;
3530 	unsigned int hlen;
3531 	struct socket *so;
3532 	struct mbuf *m;
3533 	u_char *optp;
3534 	int optlen;
3535 	struct tcp_opt_info *oi;
3536 {
3537 	struct tcpcb tb, *tp;
3538 	long win;
3539 	struct syn_cache *sc;
3540 	struct syn_cache_head *scp;
3541 	struct mbuf *ipopts;
3542 
3543 	tp = sototcpcb(so);
3544 
3545 	/*
3546 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3547 	 *
3548 	 * Note this check is performed in tcp_input() very early on.
3549 	 */
3550 
3551 	/*
3552 	 * Initialize some local state.
3553 	 */
3554 	win = sbspace(&so->so_rcv);
3555 	if (win > TCP_MAXWIN)
3556 		win = TCP_MAXWIN;
3557 
3558 	switch (src->sa_family) {
3559 #ifdef INET
3560 	case AF_INET:
3561 		/*
3562 		 * Remember the IP options, if any.
3563 		 */
3564 		ipopts = ip_srcroute();
3565 		break;
3566 #endif
3567 	default:
3568 		ipopts = NULL;
3569 	}
3570 
3571 	if (optp) {
3572 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3573 		tcp_dooptions(&tb, optp, optlen, th, oi);
3574 	} else
3575 		tb.t_flags = 0;
3576 
3577 	/*
3578 	 * See if we already have an entry for this connection.
3579 	 * If we do, resend the SYN,ACK.  We do not count this
3580 	 * as a retransmission (XXX though maybe we should).
3581 	 */
3582 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3583 		tcpstat.tcps_sc_dupesyn++;
3584 		if (ipopts) {
3585 			/*
3586 			 * If we were remembering a previous source route,
3587 			 * forget it and use the new one we've been given.
3588 			 */
3589 			if (sc->sc_ipopts)
3590 				(void) m_free(sc->sc_ipopts);
3591 			sc->sc_ipopts = ipopts;
3592 		}
3593 		sc->sc_timestamp = tb.ts_recent;
3594 		if (syn_cache_respond(sc, m) == 0) {
3595 			tcpstat.tcps_sndacks++;
3596 			tcpstat.tcps_sndtotal++;
3597 		}
3598 		return (1);
3599 	}
3600 
3601 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
3602 	if (sc == NULL) {
3603 		if (ipopts)
3604 			(void) m_free(ipopts);
3605 		return (0);
3606 	}
3607 
3608 	/*
3609 	 * Fill in the cache, and put the necessary IP and TCP
3610 	 * options into the reply.
3611 	 */
3612 	bzero(sc, sizeof(struct syn_cache));
3613 	callout_init(&sc->sc_timer);
3614 	bcopy(src, &sc->sc_src, src->sa_len);
3615 	bcopy(dst, &sc->sc_dst, dst->sa_len);
3616 	sc->sc_flags = 0;
3617 	sc->sc_ipopts = ipopts;
3618 	sc->sc_irs = th->th_seq;
3619 	switch (src->sa_family) {
3620 #ifdef INET
3621 	case AF_INET:
3622 	    {
3623 		struct sockaddr_in *srcin = (void *) src;
3624 		struct sockaddr_in *dstin = (void *) dst;
3625 
3626 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
3627 		    &srcin->sin_addr, dstin->sin_port,
3628 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
3629 		break;
3630 	    }
3631 #endif /* INET */
3632 #ifdef INET6
3633 	case AF_INET6:
3634 	    {
3635 		struct sockaddr_in6 *srcin6 = (void *) src;
3636 		struct sockaddr_in6 *dstin6 = (void *) dst;
3637 
3638 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
3639 		    &srcin6->sin6_addr, dstin6->sin6_port,
3640 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
3641 		break;
3642 	    }
3643 #endif /* INET6 */
3644 	}
3645 	sc->sc_peermaxseg = oi->maxseg;
3646 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
3647 						m->m_pkthdr.rcvif : NULL,
3648 						sc->sc_src.sa.sa_family);
3649 	sc->sc_win = win;
3650 	sc->sc_timebase = tcp_now;	/* see tcp_newtcpcb() */
3651 	sc->sc_timestamp = tb.ts_recent;
3652 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
3653 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
3654 		sc->sc_flags |= SCF_TIMESTAMP;
3655 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
3656 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
3657 		sc->sc_requested_s_scale = tb.requested_s_scale;
3658 		sc->sc_request_r_scale = 0;
3659 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
3660 		    TCP_MAXWIN << sc->sc_request_r_scale <
3661 		    so->so_rcv.sb_hiwat)
3662 			sc->sc_request_r_scale++;
3663 	} else {
3664 		sc->sc_requested_s_scale = 15;
3665 		sc->sc_request_r_scale = 15;
3666 	}
3667 	sc->sc_tp = tp;
3668 	if (syn_cache_respond(sc, m) == 0) {
3669 		syn_cache_insert(sc, tp);
3670 		tcpstat.tcps_sndacks++;
3671 		tcpstat.tcps_sndtotal++;
3672 	} else {
3673 		SYN_CACHE_PUT(sc);
3674 		tcpstat.tcps_sc_dropped++;
3675 	}
3676 	return (1);
3677 }
3678 
3679 int
3680 syn_cache_respond(sc, m)
3681 	struct syn_cache *sc;
3682 	struct mbuf *m;
3683 {
3684 	struct route *ro;
3685 	u_int8_t *optp;
3686 	int optlen, error;
3687 	u_int16_t tlen;
3688 	struct ip *ip = NULL;
3689 #ifdef INET6
3690 	struct ip6_hdr *ip6 = NULL;
3691 #endif
3692 	struct tcphdr *th;
3693 	u_int hlen;
3694 
3695 	switch (sc->sc_src.sa.sa_family) {
3696 	case AF_INET:
3697 		hlen = sizeof(struct ip);
3698 		ro = &sc->sc_route4;
3699 		break;
3700 #ifdef INET6
3701 	case AF_INET6:
3702 		hlen = sizeof(struct ip6_hdr);
3703 		ro = (struct route *)&sc->sc_route6;
3704 		break;
3705 #endif
3706 	default:
3707 		if (m)
3708 			m_freem(m);
3709 		return EAFNOSUPPORT;
3710 	}
3711 
3712 	/* Compute the size of the TCP options. */
3713 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
3714 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
3715 
3716 	tlen = hlen + sizeof(struct tcphdr) + optlen;
3717 
3718 	/*
3719 	 * Create the IP+TCP header from scratch.
3720 	 */
3721 	if (m)
3722 		m_freem(m);
3723 #ifdef DIAGNOSTIC
3724 	if (max_linkhdr + tlen > MCLBYTES)
3725 		return (ENOBUFS);
3726 #endif
3727 	MGETHDR(m, M_DONTWAIT, MT_DATA);
3728 	if (m && tlen > MHLEN) {
3729 		MCLGET(m, M_DONTWAIT);
3730 		if ((m->m_flags & M_EXT) == 0) {
3731 			m_freem(m);
3732 			m = NULL;
3733 		}
3734 	}
3735 	if (m == NULL)
3736 		return (ENOBUFS);
3737 	MCLAIM(m, &tcp_tx_mowner);
3738 
3739 	/* Fixup the mbuf. */
3740 	m->m_data += max_linkhdr;
3741 	m->m_len = m->m_pkthdr.len = tlen;
3742 #ifdef IPSEC
3743 	if (sc->sc_tp) {
3744 		struct tcpcb *tp;
3745 		struct socket *so;
3746 
3747 		tp = sc->sc_tp;
3748 		if (tp->t_inpcb)
3749 			so = tp->t_inpcb->inp_socket;
3750 #ifdef INET6
3751 		else if (tp->t_in6pcb)
3752 			so = tp->t_in6pcb->in6p_socket;
3753 #endif
3754 		else
3755 			so = NULL;
3756 		/* use IPsec policy on listening socket, on SYN ACK */
3757 		if (ipsec_setsocket(m, so) != 0) {
3758 			m_freem(m);
3759 			return ENOBUFS;
3760 		}
3761 	}
3762 #endif
3763 	m->m_pkthdr.rcvif = NULL;
3764 	memset(mtod(m, u_char *), 0, tlen);
3765 
3766 	switch (sc->sc_src.sa.sa_family) {
3767 	case AF_INET:
3768 		ip = mtod(m, struct ip *);
3769 		ip->ip_dst = sc->sc_src.sin.sin_addr;
3770 		ip->ip_src = sc->sc_dst.sin.sin_addr;
3771 		ip->ip_p = IPPROTO_TCP;
3772 		th = (struct tcphdr *)(ip + 1);
3773 		th->th_dport = sc->sc_src.sin.sin_port;
3774 		th->th_sport = sc->sc_dst.sin.sin_port;
3775 		break;
3776 #ifdef INET6
3777 	case AF_INET6:
3778 		ip6 = mtod(m, struct ip6_hdr *);
3779 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
3780 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
3781 		ip6->ip6_nxt = IPPROTO_TCP;
3782 		/* ip6_plen will be updated in ip6_output() */
3783 		th = (struct tcphdr *)(ip6 + 1);
3784 		th->th_dport = sc->sc_src.sin6.sin6_port;
3785 		th->th_sport = sc->sc_dst.sin6.sin6_port;
3786 		break;
3787 #endif
3788 	default:
3789 		th = NULL;
3790 	}
3791 
3792 	th->th_seq = htonl(sc->sc_iss);
3793 	th->th_ack = htonl(sc->sc_irs + 1);
3794 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
3795 	th->th_flags = TH_SYN|TH_ACK;
3796 	th->th_win = htons(sc->sc_win);
3797 	/* th_sum already 0 */
3798 	/* th_urp already 0 */
3799 
3800 	/* Tack on the TCP options. */
3801 	optp = (u_int8_t *)(th + 1);
3802 	*optp++ = TCPOPT_MAXSEG;
3803 	*optp++ = 4;
3804 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
3805 	*optp++ = sc->sc_ourmaxseg & 0xff;
3806 
3807 	if (sc->sc_request_r_scale != 15) {
3808 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
3809 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
3810 		    sc->sc_request_r_scale);
3811 		optp += 4;
3812 	}
3813 
3814 	if (sc->sc_flags & SCF_TIMESTAMP) {
3815 		u_int32_t *lp = (u_int32_t *)(optp);
3816 		/* Form timestamp option as shown in appendix A of RFC 1323. */
3817 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
3818 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
3819 		*lp   = htonl(sc->sc_timestamp);
3820 		optp += TCPOLEN_TSTAMP_APPA;
3821 	}
3822 
3823 	/* Compute the packet's checksum. */
3824 	switch (sc->sc_src.sa.sa_family) {
3825 	case AF_INET:
3826 		ip->ip_len = htons(tlen - hlen);
3827 		th->th_sum = 0;
3828 		th->th_sum = in_cksum(m, tlen);
3829 		break;
3830 #ifdef INET6
3831 	case AF_INET6:
3832 		ip6->ip6_plen = htons(tlen - hlen);
3833 		th->th_sum = 0;
3834 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
3835 		break;
3836 #endif
3837 	}
3838 
3839 	/*
3840 	 * Fill in some straggling IP bits.  Note the stack expects
3841 	 * ip_len to be in host order, for convenience.
3842 	 */
3843 	switch (sc->sc_src.sa.sa_family) {
3844 #ifdef INET
3845 	case AF_INET:
3846 		ip->ip_len = htons(tlen);
3847 		ip->ip_ttl = ip_defttl;
3848 		/* XXX tos? */
3849 		break;
3850 #endif
3851 #ifdef INET6
3852 	case AF_INET6:
3853 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3854 		ip6->ip6_vfc |= IPV6_VERSION;
3855 		ip6->ip6_plen = htons(tlen - hlen);
3856 		/* ip6_hlim will be initialized afterwards */
3857 		/* XXX flowlabel? */
3858 		break;
3859 #endif
3860 	}
3861 
3862 	switch (sc->sc_src.sa.sa_family) {
3863 #ifdef INET
3864 	case AF_INET:
3865 		error = ip_output(m, sc->sc_ipopts, ro,
3866 		    (ip_mtudisc ? IP_MTUDISC : 0),
3867 		    NULL);
3868 		break;
3869 #endif
3870 #ifdef INET6
3871 	case AF_INET6:
3872 		ip6->ip6_hlim = in6_selecthlim(NULL,
3873 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
3874 
3875 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro,
3876 			0, NULL, NULL);
3877 		break;
3878 #endif
3879 	default:
3880 		error = EAFNOSUPPORT;
3881 		break;
3882 	}
3883 	return (error);
3884 }
3885