xref: /netbsd-src/sys/netinet/tcp_input.c (revision 313c6c94c424eed90c7b7e494aa83308a0a5d0ce)
1 /*	$NetBSD: tcp_input.c,v 1.295 2009/03/18 16:00:22 cegger 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, 2005, 2006 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  * This code is derived from software contributed to The NetBSD Foundation
80  * by Charles M. Hannum.
81  * This code is derived from software contributed to The NetBSD Foundation
82  * by Rui Paulo.
83  *
84  * Redistribution and use in source and binary forms, with or without
85  * modification, are permitted provided that the following conditions
86  * are met:
87  * 1. Redistributions of source code must retain the above copyright
88  *    notice, this list of conditions and the following disclaimer.
89  * 2. Redistributions in binary form must reproduce the above copyright
90  *    notice, this list of conditions and the following disclaimer in the
91  *    documentation and/or other materials provided with the distribution.
92  *
93  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
94  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
95  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
96  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
97  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
98  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
99  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
100  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
101  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
102  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
103  * POSSIBILITY OF SUCH DAMAGE.
104  */
105 
106 /*
107  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
108  *	The Regents of the University of California.  All rights reserved.
109  *
110  * Redistribution and use in source and binary forms, with or without
111  * modification, are permitted provided that the following conditions
112  * are met:
113  * 1. Redistributions of source code must retain the above copyright
114  *    notice, this list of conditions and the following disclaimer.
115  * 2. Redistributions in binary form must reproduce the above copyright
116  *    notice, this list of conditions and the following disclaimer in the
117  *    documentation and/or other materials provided with the distribution.
118  * 3. Neither the name of the University nor the names of its contributors
119  *    may be used to endorse or promote products derived from this software
120  *    without specific prior written permission.
121  *
122  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
123  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
124  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
125  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
126  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
127  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
128  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
129  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
130  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
131  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
132  * SUCH DAMAGE.
133  *
134  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
135  */
136 
137 /*
138  *	TODO list for SYN cache stuff:
139  *
140  *	Find room for a "state" field, which is needed to keep a
141  *	compressed state for TIME_WAIT TCBs.  It's been noted already
142  *	that this is fairly important for very high-volume web and
143  *	mail servers, which use a large number of short-lived
144  *	connections.
145  */
146 
147 #include <sys/cdefs.h>
148 __KERNEL_RCSID(0, "$NetBSD: tcp_input.c,v 1.295 2009/03/18 16:00:22 cegger Exp $");
149 
150 #include "opt_inet.h"
151 #include "opt_ipsec.h"
152 #include "opt_inet_csum.h"
153 #include "opt_tcp_debug.h"
154 
155 #include <sys/param.h>
156 #include <sys/systm.h>
157 #include <sys/malloc.h>
158 #include <sys/mbuf.h>
159 #include <sys/protosw.h>
160 #include <sys/socket.h>
161 #include <sys/socketvar.h>
162 #include <sys/errno.h>
163 #include <sys/syslog.h>
164 #include <sys/pool.h>
165 #include <sys/domain.h>
166 #include <sys/kernel.h>
167 #ifdef TCP_SIGNATURE
168 #include <sys/md5.h>
169 #endif
170 #include <sys/lwp.h> /* for lwp0 */
171 
172 #include <net/if.h>
173 #include <net/route.h>
174 #include <net/if_types.h>
175 
176 #include <netinet/in.h>
177 #include <netinet/in_systm.h>
178 #include <netinet/ip.h>
179 #include <netinet/in_pcb.h>
180 #include <netinet/in_var.h>
181 #include <netinet/ip_var.h>
182 #include <netinet/in_offload.h>
183 
184 #ifdef INET6
185 #ifndef INET
186 #include <netinet/in.h>
187 #endif
188 #include <netinet/ip6.h>
189 #include <netinet6/ip6_var.h>
190 #include <netinet6/in6_pcb.h>
191 #include <netinet6/ip6_var.h>
192 #include <netinet6/in6_var.h>
193 #include <netinet/icmp6.h>
194 #include <netinet6/nd6.h>
195 #ifdef TCP_SIGNATURE
196 #include <netinet6/scope6_var.h>
197 #endif
198 #endif
199 
200 #ifndef INET6
201 /* always need ip6.h for IP6_EXTHDR_GET */
202 #include <netinet/ip6.h>
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/tcp_private.h>
211 #include <netinet/tcpip.h>
212 #include <netinet/tcp_congctl.h>
213 #include <netinet/tcp_debug.h>
214 
215 #include <machine/stdarg.h>
216 
217 #ifdef IPSEC
218 #include <netinet6/ipsec.h>
219 #include <netinet6/ipsec_private.h>
220 #include <netkey/key.h>
221 #endif /*IPSEC*/
222 #ifdef INET6
223 #include "faith.h"
224 #if defined(NFAITH) && NFAITH > 0
225 #include <net/if_faith.h>
226 #endif
227 #endif	/* IPSEC */
228 
229 #ifdef FAST_IPSEC
230 #include <netipsec/ipsec.h>
231 #include <netipsec/ipsec_var.h>
232 #include <netipsec/ipsec_private.h>
233 #include <netipsec/key.h>
234 #ifdef INET6
235 #include <netipsec/ipsec6.h>
236 #endif
237 #endif	/* FAST_IPSEC*/
238 
239 int	tcprexmtthresh = 3;
240 int	tcp_log_refused;
241 
242 int	tcp_do_autorcvbuf = 0;
243 int	tcp_autorcvbuf_inc = 16 * 1024;
244 int	tcp_autorcvbuf_max = 256 * 1024;
245 
246 static int tcp_rst_ppslim_count = 0;
247 static struct timeval tcp_rst_ppslim_last;
248 static int tcp_ackdrop_ppslim_count = 0;
249 static struct timeval tcp_ackdrop_ppslim_last;
250 
251 #define TCP_PAWS_IDLE	(24U * 24 * 60 * 60 * PR_SLOWHZ)
252 
253 /* for modulo comparisons of timestamps */
254 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
255 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
256 
257 /*
258  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
259  */
260 #ifdef INET6
261 static inline void
262 nd6_hint(struct tcpcb *tp)
263 {
264 	struct rtentry *rt;
265 
266 	if (tp != NULL && tp->t_in6pcb != NULL && tp->t_family == AF_INET6 &&
267 	    (rt = rtcache_validate(&tp->t_in6pcb->in6p_route)) != NULL)
268 		nd6_nud_hint(rt, NULL, 0);
269 }
270 #else
271 static inline void
272 nd6_hint(struct tcpcb *tp)
273 {
274 }
275 #endif
276 
277 /*
278  * Compute ACK transmission behavior.  Delay the ACK unless
279  * we have already delayed an ACK (must send an ACK every two segments).
280  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
281  * option is enabled.
282  */
283 static void
284 tcp_setup_ack(struct tcpcb *tp, const struct tcphdr *th)
285 {
286 
287 	if (tp->t_flags & TF_DELACK ||
288 	    (tcp_ack_on_push && th->th_flags & TH_PUSH))
289 		tp->t_flags |= TF_ACKNOW;
290 	else
291 		TCP_SET_DELACK(tp);
292 }
293 
294 static void
295 icmp_check(struct tcpcb *tp, const struct tcphdr *th, int acked)
296 {
297 
298 	/*
299 	 * If we had a pending ICMP message that refers to data that have
300 	 * just been acknowledged, disregard the recorded ICMP message.
301 	 */
302 	if ((tp->t_flags & TF_PMTUD_PEND) &&
303 	    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
304 		tp->t_flags &= ~TF_PMTUD_PEND;
305 
306 	/*
307 	 * Keep track of the largest chunk of data
308 	 * acknowledged since last PMTU update
309 	 */
310 	if (tp->t_pmtud_mss_acked < acked)
311 		tp->t_pmtud_mss_acked = acked;
312 }
313 
314 /*
315  * Convert TCP protocol fields to host order for easier processing.
316  */
317 static void
318 tcp_fields_to_host(struct tcphdr *th)
319 {
320 
321 	NTOHL(th->th_seq);
322 	NTOHL(th->th_ack);
323 	NTOHS(th->th_win);
324 	NTOHS(th->th_urp);
325 }
326 
327 /*
328  * ... and reverse the above.
329  */
330 static void
331 tcp_fields_to_net(struct tcphdr *th)
332 {
333 
334 	HTONL(th->th_seq);
335 	HTONL(th->th_ack);
336 	HTONS(th->th_win);
337 	HTONS(th->th_urp);
338 }
339 
340 #ifdef TCP_CSUM_COUNTERS
341 #include <sys/device.h>
342 
343 #if defined(INET)
344 extern struct evcnt tcp_hwcsum_ok;
345 extern struct evcnt tcp_hwcsum_bad;
346 extern struct evcnt tcp_hwcsum_data;
347 extern struct evcnt tcp_swcsum;
348 #endif /* defined(INET) */
349 #if defined(INET6)
350 extern struct evcnt tcp6_hwcsum_ok;
351 extern struct evcnt tcp6_hwcsum_bad;
352 extern struct evcnt tcp6_hwcsum_data;
353 extern struct evcnt tcp6_swcsum;
354 #endif /* defined(INET6) */
355 
356 #define	TCP_CSUM_COUNTER_INCR(ev)	(ev)->ev_count++
357 
358 #else
359 
360 #define	TCP_CSUM_COUNTER_INCR(ev)	/* nothing */
361 
362 #endif /* TCP_CSUM_COUNTERS */
363 
364 #ifdef TCP_REASS_COUNTERS
365 #include <sys/device.h>
366 
367 extern struct evcnt tcp_reass_;
368 extern struct evcnt tcp_reass_empty;
369 extern struct evcnt tcp_reass_iteration[8];
370 extern struct evcnt tcp_reass_prependfirst;
371 extern struct evcnt tcp_reass_prepend;
372 extern struct evcnt tcp_reass_insert;
373 extern struct evcnt tcp_reass_inserttail;
374 extern struct evcnt tcp_reass_append;
375 extern struct evcnt tcp_reass_appendtail;
376 extern struct evcnt tcp_reass_overlaptail;
377 extern struct evcnt tcp_reass_overlapfront;
378 extern struct evcnt tcp_reass_segdup;
379 extern struct evcnt tcp_reass_fragdup;
380 
381 #define	TCP_REASS_COUNTER_INCR(ev)	(ev)->ev_count++
382 
383 #else
384 
385 #define	TCP_REASS_COUNTER_INCR(ev)	/* nothing */
386 
387 #endif /* TCP_REASS_COUNTERS */
388 
389 static int tcp_reass(struct tcpcb *, const struct tcphdr *, struct mbuf *,
390     int *);
391 static int tcp_dooptions(struct tcpcb *, const u_char *, int,
392     struct tcphdr *, struct mbuf *, int, struct tcp_opt_info *);
393 
394 #ifdef INET
395 static void tcp4_log_refused(const struct ip *, const struct tcphdr *);
396 #endif
397 #ifdef INET6
398 static void tcp6_log_refused(const struct ip6_hdr *, const struct tcphdr *);
399 #endif
400 
401 #define	TRAVERSE(x) while ((x)->m_next) (x) = (x)->m_next
402 
403 #if defined(MBUFTRACE)
404 struct mowner tcp_reass_mowner = MOWNER_INIT("tcp", "reass");
405 #endif /* defined(MBUFTRACE) */
406 
407 static struct pool tcpipqent_pool;
408 
409 void
410 tcpipqent_init(void)
411 {
412 
413 	pool_init(&tcpipqent_pool, sizeof(struct ipqent), 0, 0, 0, "tcpipqepl",
414 	    NULL, IPL_VM);
415 }
416 
417 struct ipqent *
418 tcpipqent_alloc(void)
419 {
420 	struct ipqent *ipqe;
421 	int s;
422 
423 	s = splvm();
424 	ipqe = pool_get(&tcpipqent_pool, PR_NOWAIT);
425 	splx(s);
426 
427 	return ipqe;
428 }
429 
430 void
431 tcpipqent_free(struct ipqent *ipqe)
432 {
433 	int s;
434 
435 	s = splvm();
436 	pool_put(&tcpipqent_pool, ipqe);
437 	splx(s);
438 }
439 
440 static int
441 tcp_reass(struct tcpcb *tp, const struct tcphdr *th, struct mbuf *m, int *tlen)
442 {
443 	struct ipqent *p, *q, *nq, *tiqe = NULL;
444 	struct socket *so = NULL;
445 	int pkt_flags;
446 	tcp_seq pkt_seq;
447 	unsigned pkt_len;
448 	u_long rcvpartdupbyte = 0;
449 	u_long rcvoobyte;
450 #ifdef TCP_REASS_COUNTERS
451 	u_int count = 0;
452 #endif
453 	uint64_t *tcps;
454 
455 	if (tp->t_inpcb)
456 		so = tp->t_inpcb->inp_socket;
457 #ifdef INET6
458 	else if (tp->t_in6pcb)
459 		so = tp->t_in6pcb->in6p_socket;
460 #endif
461 
462 	TCP_REASS_LOCK_CHECK(tp);
463 
464 	/*
465 	 * Call with th==0 after become established to
466 	 * force pre-ESTABLISHED data up to user socket.
467 	 */
468 	if (th == 0)
469 		goto present;
470 
471 	m_claimm(m, &tcp_reass_mowner);
472 
473 	rcvoobyte = *tlen;
474 	/*
475 	 * Copy these to local variables because the tcpiphdr
476 	 * gets munged while we are collapsing mbufs.
477 	 */
478 	pkt_seq = th->th_seq;
479 	pkt_len = *tlen;
480 	pkt_flags = th->th_flags;
481 
482 	TCP_REASS_COUNTER_INCR(&tcp_reass_);
483 
484 	if ((p = TAILQ_LAST(&tp->segq, ipqehead)) != NULL) {
485 		/*
486 		 * When we miss a packet, the vast majority of time we get
487 		 * packets that follow it in order.  So optimize for that.
488 		 */
489 		if (pkt_seq == p->ipqe_seq + p->ipqe_len) {
490 			p->ipqe_len += pkt_len;
491 			p->ipqe_flags |= pkt_flags;
492 			m_cat(p->ipre_mlast, m);
493 			TRAVERSE(p->ipre_mlast);
494 			m = NULL;
495 			tiqe = p;
496 			TAILQ_REMOVE(&tp->timeq, p, ipqe_timeq);
497 			TCP_REASS_COUNTER_INCR(&tcp_reass_appendtail);
498 			goto skip_replacement;
499 		}
500 		/*
501 		 * While we're here, if the pkt is completely beyond
502 		 * anything we have, just insert it at the tail.
503 		 */
504 		if (SEQ_GT(pkt_seq, p->ipqe_seq + p->ipqe_len)) {
505 			TCP_REASS_COUNTER_INCR(&tcp_reass_inserttail);
506 			goto insert_it;
507 		}
508 	}
509 
510 	q = TAILQ_FIRST(&tp->segq);
511 
512 	if (q != NULL) {
513 		/*
514 		 * If this segment immediately precedes the first out-of-order
515 		 * block, simply slap the segment in front of it and (mostly)
516 		 * skip the complicated logic.
517 		 */
518 		if (pkt_seq + pkt_len == q->ipqe_seq) {
519 			q->ipqe_seq = pkt_seq;
520 			q->ipqe_len += pkt_len;
521 			q->ipqe_flags |= pkt_flags;
522 			m_cat(m, q->ipqe_m);
523 			q->ipqe_m = m;
524 			q->ipre_mlast = m; /* last mbuf may have changed */
525 			TRAVERSE(q->ipre_mlast);
526 			tiqe = q;
527 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
528 			TCP_REASS_COUNTER_INCR(&tcp_reass_prependfirst);
529 			goto skip_replacement;
530 		}
531 	} else {
532 		TCP_REASS_COUNTER_INCR(&tcp_reass_empty);
533 	}
534 
535 	/*
536 	 * Find a segment which begins after this one does.
537 	 */
538 	for (p = NULL; q != NULL; q = nq) {
539 		nq = TAILQ_NEXT(q, ipqe_q);
540 #ifdef TCP_REASS_COUNTERS
541 		count++;
542 #endif
543 		/*
544 		 * If the received segment is just right after this
545 		 * fragment, merge the two together and then check
546 		 * for further overlaps.
547 		 */
548 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
549 #ifdef TCPREASS_DEBUG
550 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
551 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
552 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
553 #endif
554 			pkt_len += q->ipqe_len;
555 			pkt_flags |= q->ipqe_flags;
556 			pkt_seq = q->ipqe_seq;
557 			m_cat(q->ipre_mlast, m);
558 			TRAVERSE(q->ipre_mlast);
559 			m = q->ipqe_m;
560 			TCP_REASS_COUNTER_INCR(&tcp_reass_append);
561 			goto free_ipqe;
562 		}
563 		/*
564 		 * If the received segment is completely past this
565 		 * fragment, we need to go the next fragment.
566 		 */
567 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
568 			p = q;
569 			continue;
570 		}
571 		/*
572 		 * If the fragment is past the received segment,
573 		 * it (or any following) can't be concatenated.
574 		 */
575 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len)) {
576 			TCP_REASS_COUNTER_INCR(&tcp_reass_insert);
577 			break;
578 		}
579 
580 		/*
581 		 * We've received all the data in this segment before.
582 		 * mark it as a duplicate and return.
583 		 */
584 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
585 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
586 			tcps = TCP_STAT_GETREF();
587 			tcps[TCP_STAT_RCVDUPPACK]++;
588 			tcps[TCP_STAT_RCVDUPBYTE] += pkt_len;
589 			TCP_STAT_PUTREF();
590 			tcp_new_dsack(tp, pkt_seq, pkt_len);
591 			m_freem(m);
592 			if (tiqe != NULL) {
593 				tcpipqent_free(tiqe);
594 			}
595 			TCP_REASS_COUNTER_INCR(&tcp_reass_segdup);
596 			return (0);
597 		}
598 		/*
599 		 * Received segment completely overlaps this fragment
600 		 * so we drop the fragment (this keeps the temporal
601 		 * ordering of segments correct).
602 		 */
603 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
604 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
605 			rcvpartdupbyte += q->ipqe_len;
606 			m_freem(q->ipqe_m);
607 			TCP_REASS_COUNTER_INCR(&tcp_reass_fragdup);
608 			goto free_ipqe;
609 		}
610 		/*
611 		 * RX'ed segment extends past the end of the
612 		 * fragment.  Drop the overlapping bytes.  Then
613 		 * merge the fragment and segment then treat as
614 		 * a longer received packet.
615 		 */
616 		if (SEQ_LT(q->ipqe_seq, pkt_seq) &&
617 		    SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
618 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
619 #ifdef TCPREASS_DEBUG
620 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
621 			       tp, overlap,
622 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
623 #endif
624 			m_adj(m, overlap);
625 			rcvpartdupbyte += overlap;
626 			m_cat(q->ipre_mlast, m);
627 			TRAVERSE(q->ipre_mlast);
628 			m = q->ipqe_m;
629 			pkt_seq = q->ipqe_seq;
630 			pkt_len += q->ipqe_len - overlap;
631 			rcvoobyte -= overlap;
632 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlaptail);
633 			goto free_ipqe;
634 		}
635 		/*
636 		 * RX'ed segment extends past the front of the
637 		 * fragment.  Drop the overlapping bytes on the
638 		 * received packet.  The packet will then be
639 		 * contatentated with this fragment a bit later.
640 		 */
641 		if (SEQ_GT(q->ipqe_seq, pkt_seq) &&
642 		    SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
643 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
644 #ifdef TCPREASS_DEBUG
645 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
646 			       tp, overlap,
647 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
648 #endif
649 			m_adj(m, -overlap);
650 			pkt_len -= overlap;
651 			rcvpartdupbyte += overlap;
652 			TCP_REASS_COUNTER_INCR(&tcp_reass_overlapfront);
653 			rcvoobyte -= overlap;
654 		}
655 		/*
656 		 * If the received segment immediates precedes this
657 		 * fragment then tack the fragment onto this segment
658 		 * and reinsert the data.
659 		 */
660 		if (q->ipqe_seq == pkt_seq + pkt_len) {
661 #ifdef TCPREASS_DEBUG
662 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
663 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
664 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
665 #endif
666 			pkt_len += q->ipqe_len;
667 			pkt_flags |= q->ipqe_flags;
668 			m_cat(m, q->ipqe_m);
669 			TAILQ_REMOVE(&tp->segq, q, ipqe_q);
670 			TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
671 			tp->t_segqlen--;
672 			KASSERT(tp->t_segqlen >= 0);
673 			KASSERT(tp->t_segqlen != 0 ||
674 			    (TAILQ_EMPTY(&tp->segq) &&
675 			    TAILQ_EMPTY(&tp->timeq)));
676 			if (tiqe == NULL) {
677 				tiqe = q;
678 			} else {
679 				tcpipqent_free(q);
680 			}
681 			TCP_REASS_COUNTER_INCR(&tcp_reass_prepend);
682 			break;
683 		}
684 		/*
685 		 * If the fragment is before the segment, remember it.
686 		 * When this loop is terminated, p will contain the
687 		 * pointer to fragment that is right before the received
688 		 * segment.
689 		 */
690 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
691 			p = q;
692 
693 		continue;
694 
695 		/*
696 		 * This is a common operation.  It also will allow
697 		 * to save doing a malloc/free in most instances.
698 		 */
699 	  free_ipqe:
700 		TAILQ_REMOVE(&tp->segq, q, ipqe_q);
701 		TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
702 		tp->t_segqlen--;
703 		KASSERT(tp->t_segqlen >= 0);
704 		KASSERT(tp->t_segqlen != 0 ||
705 		    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
706 		if (tiqe == NULL) {
707 			tiqe = q;
708 		} else {
709 			tcpipqent_free(q);
710 		}
711 	}
712 
713 #ifdef TCP_REASS_COUNTERS
714 	if (count > 7)
715 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[0]);
716 	else if (count > 0)
717 		TCP_REASS_COUNTER_INCR(&tcp_reass_iteration[count]);
718 #endif
719 
720     insert_it:
721 
722 	/*
723 	 * Allocate a new queue entry since the received segment did not
724 	 * collapse onto any other out-of-order block; thus we are allocating
725 	 * a new block.  If it had collapsed, tiqe would not be NULL and
726 	 * we would be reusing it.
727 	 * XXX If we can't, just drop the packet.  XXX
728 	 */
729 	if (tiqe == NULL) {
730 		tiqe = tcpipqent_alloc();
731 		if (tiqe == NULL) {
732 			TCP_STATINC(TCP_STAT_RCVMEMDROP);
733 			m_freem(m);
734 			return (0);
735 		}
736 	}
737 
738 	/*
739 	 * Update the counters.
740 	 */
741 	tcps = TCP_STAT_GETREF();
742 	tcps[TCP_STAT_RCVOOPACK]++;
743 	tcps[TCP_STAT_RCVOOBYTE] += rcvoobyte;
744 	if (rcvpartdupbyte) {
745 	    tcps[TCP_STAT_RCVPARTDUPPACK]++;
746 	    tcps[TCP_STAT_RCVPARTDUPBYTE] += rcvpartdupbyte;
747 	}
748 	TCP_STAT_PUTREF();
749 
750 	/*
751 	 * Insert the new fragment queue entry into both queues.
752 	 */
753 	tiqe->ipqe_m = m;
754 	tiqe->ipre_mlast = m;
755 	tiqe->ipqe_seq = pkt_seq;
756 	tiqe->ipqe_len = pkt_len;
757 	tiqe->ipqe_flags = pkt_flags;
758 	if (p == NULL) {
759 		TAILQ_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
760 #ifdef TCPREASS_DEBUG
761 		if (tiqe->ipqe_seq != tp->rcv_nxt)
762 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
763 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
764 #endif
765 	} else {
766 		TAILQ_INSERT_AFTER(&tp->segq, p, tiqe, ipqe_q);
767 #ifdef TCPREASS_DEBUG
768 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
769 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
770 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
771 #endif
772 	}
773 	tp->t_segqlen++;
774 
775 skip_replacement:
776 
777 	TAILQ_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
778 
779 present:
780 	/*
781 	 * Present data to user, advancing rcv_nxt through
782 	 * completed sequence space.
783 	 */
784 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
785 		return (0);
786 	q = TAILQ_FIRST(&tp->segq);
787 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
788 		return (0);
789 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
790 		return (0);
791 
792 	tp->rcv_nxt += q->ipqe_len;
793 	pkt_flags = q->ipqe_flags & TH_FIN;
794 	nd6_hint(tp);
795 
796 	TAILQ_REMOVE(&tp->segq, q, ipqe_q);
797 	TAILQ_REMOVE(&tp->timeq, q, ipqe_timeq);
798 	tp->t_segqlen--;
799 	KASSERT(tp->t_segqlen >= 0);
800 	KASSERT(tp->t_segqlen != 0 ||
801 	    (TAILQ_EMPTY(&tp->segq) && TAILQ_EMPTY(&tp->timeq)));
802 	if (so->so_state & SS_CANTRCVMORE)
803 		m_freem(q->ipqe_m);
804 	else
805 		sbappendstream(&so->so_rcv, q->ipqe_m);
806 	tcpipqent_free(q);
807 	sorwakeup(so);
808 	return (pkt_flags);
809 }
810 
811 #ifdef INET6
812 int
813 tcp6_input(struct mbuf **mp, int *offp, int proto)
814 {
815 	struct mbuf *m = *mp;
816 
817 	/*
818 	 * draft-itojun-ipv6-tcp-to-anycast
819 	 * better place to put this in?
820 	 */
821 	if (m->m_flags & M_ANYCAST6) {
822 		struct ip6_hdr *ip6;
823 		if (m->m_len < sizeof(struct ip6_hdr)) {
824 			if ((m = m_pullup(m, sizeof(struct ip6_hdr))) == NULL) {
825 				TCP_STATINC(TCP_STAT_RCVSHORT);
826 				return IPPROTO_DONE;
827 			}
828 		}
829 		ip6 = mtod(m, struct ip6_hdr *);
830 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
831 		    (char *)&ip6->ip6_dst - (char *)ip6);
832 		return IPPROTO_DONE;
833 	}
834 
835 	tcp_input(m, *offp, proto);
836 	return IPPROTO_DONE;
837 }
838 #endif
839 
840 #ifdef INET
841 static void
842 tcp4_log_refused(const struct ip *ip, const struct tcphdr *th)
843 {
844 	char src[4*sizeof "123"];
845 	char dst[4*sizeof "123"];
846 
847 	if (ip) {
848 		strlcpy(src, inet_ntoa(ip->ip_src), sizeof(src));
849 		strlcpy(dst, inet_ntoa(ip->ip_dst), sizeof(dst));
850 	}
851 	else {
852 		strlcpy(src, "(unknown)", sizeof(src));
853 		strlcpy(dst, "(unknown)", sizeof(dst));
854 	}
855 	log(LOG_INFO,
856 	    "Connection attempt to TCP %s:%d from %s:%d\n",
857 	    dst, ntohs(th->th_dport),
858 	    src, ntohs(th->th_sport));
859 }
860 #endif
861 
862 #ifdef INET6
863 static void
864 tcp6_log_refused(const struct ip6_hdr *ip6, const struct tcphdr *th)
865 {
866 	char src[INET6_ADDRSTRLEN];
867 	char dst[INET6_ADDRSTRLEN];
868 
869 	if (ip6) {
870 		strlcpy(src, ip6_sprintf(&ip6->ip6_src), sizeof(src));
871 		strlcpy(dst, ip6_sprintf(&ip6->ip6_dst), sizeof(dst));
872 	}
873 	else {
874 		strlcpy(src, "(unknown v6)", sizeof(src));
875 		strlcpy(dst, "(unknown v6)", sizeof(dst));
876 	}
877 	log(LOG_INFO,
878 	    "Connection attempt to TCP [%s]:%d from [%s]:%d\n",
879 	    dst, ntohs(th->th_dport),
880 	    src, ntohs(th->th_sport));
881 }
882 #endif
883 
884 /*
885  * Checksum extended TCP header and data.
886  */
887 int
888 tcp_input_checksum(int af, struct mbuf *m, const struct tcphdr *th,
889     int toff, int off, int tlen)
890 {
891 
892 	/*
893 	 * XXX it's better to record and check if this mbuf is
894 	 * already checked.
895 	 */
896 
897 	switch (af) {
898 #ifdef INET
899 	case AF_INET:
900 		switch (m->m_pkthdr.csum_flags &
901 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv4) |
902 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
903 		case M_CSUM_TCPv4|M_CSUM_TCP_UDP_BAD:
904 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_bad);
905 			goto badcsum;
906 
907 		case M_CSUM_TCPv4|M_CSUM_DATA: {
908 			u_int32_t hw_csum = m->m_pkthdr.csum_data;
909 
910 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_data);
911 			if (m->m_pkthdr.csum_flags & M_CSUM_NO_PSEUDOHDR) {
912 				const struct ip *ip =
913 				    mtod(m, const struct ip *);
914 
915 				hw_csum = in_cksum_phdr(ip->ip_src.s_addr,
916 				    ip->ip_dst.s_addr,
917 				    htons(hw_csum + tlen + off + IPPROTO_TCP));
918 			}
919 			if ((hw_csum ^ 0xffff) != 0)
920 				goto badcsum;
921 			break;
922 		}
923 
924 		case M_CSUM_TCPv4:
925 			/* Checksum was okay. */
926 			TCP_CSUM_COUNTER_INCR(&tcp_hwcsum_ok);
927 			break;
928 
929 		default:
930 			/*
931 			 * Must compute it ourselves.  Maybe skip checksum
932 			 * on loopback interfaces.
933 			 */
934 			if (__predict_true(!(m->m_pkthdr.rcvif->if_flags &
935 					     IFF_LOOPBACK) ||
936 					   tcp_do_loopback_cksum)) {
937 				TCP_CSUM_COUNTER_INCR(&tcp_swcsum);
938 				if (in4_cksum(m, IPPROTO_TCP, toff,
939 					      tlen + off) != 0)
940 					goto badcsum;
941 			}
942 			break;
943 		}
944 		break;
945 #endif /* INET4 */
946 
947 #ifdef INET6
948 	case AF_INET6:
949 		switch (m->m_pkthdr.csum_flags &
950 			((m->m_pkthdr.rcvif->if_csum_flags_rx & M_CSUM_TCPv6) |
951 			 M_CSUM_TCP_UDP_BAD | M_CSUM_DATA)) {
952 		case M_CSUM_TCPv6|M_CSUM_TCP_UDP_BAD:
953 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_bad);
954 			goto badcsum;
955 
956 #if 0 /* notyet */
957 		case M_CSUM_TCPv6|M_CSUM_DATA:
958 #endif
959 
960 		case M_CSUM_TCPv6:
961 			/* Checksum was okay. */
962 			TCP_CSUM_COUNTER_INCR(&tcp6_hwcsum_ok);
963 			break;
964 
965 		default:
966 			/*
967 			 * Must compute it ourselves.  Maybe skip checksum
968 			 * on loopback interfaces.
969 			 */
970 			if (__predict_true((m->m_flags & M_LOOP) == 0 ||
971 			    tcp_do_loopback_cksum)) {
972 				TCP_CSUM_COUNTER_INCR(&tcp6_swcsum);
973 				if (in6_cksum(m, IPPROTO_TCP, toff,
974 				    tlen + off) != 0)
975 					goto badcsum;
976 			}
977 		}
978 		break;
979 #endif /* INET6 */
980 	}
981 
982 	return 0;
983 
984 badcsum:
985 	TCP_STATINC(TCP_STAT_RCVBADSUM);
986 	return -1;
987 }
988 
989 /*
990  * TCP input routine, follows pages 65-76 of RFC 793 very closely.
991  */
992 void
993 tcp_input(struct mbuf *m, ...)
994 {
995 	struct tcphdr *th;
996 	struct ip *ip;
997 	struct inpcb *inp;
998 #ifdef INET6
999 	struct ip6_hdr *ip6;
1000 	struct in6pcb *in6p;
1001 #endif
1002 	u_int8_t *optp = NULL;
1003 	int optlen = 0;
1004 	int len, tlen, toff, hdroptlen = 0;
1005 	struct tcpcb *tp = 0;
1006 	int tiflags;
1007 	struct socket *so = NULL;
1008 	int todrop, dupseg, acked, ourfinisacked, needoutput = 0;
1009 #ifdef TCP_DEBUG
1010 	short ostate = 0;
1011 #endif
1012 	u_long tiwin;
1013 	struct tcp_opt_info opti;
1014 	int off, iphlen;
1015 	va_list ap;
1016 	int af;		/* af on the wire */
1017 	struct mbuf *tcp_saveti = NULL;
1018 	uint32_t ts_rtt;
1019 	uint8_t iptos;
1020 	uint64_t *tcps;
1021 
1022 	MCLAIM(m, &tcp_rx_mowner);
1023 	va_start(ap, m);
1024 	toff = va_arg(ap, int);
1025 	(void)va_arg(ap, int);		/* ignore value, advance ap */
1026 	va_end(ap);
1027 
1028 	TCP_STATINC(TCP_STAT_RCVTOTAL);
1029 
1030 	memset(&opti, 0, sizeof(opti));
1031 	opti.ts_present = 0;
1032 	opti.maxseg = 0;
1033 
1034 	/*
1035 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN.
1036 	 *
1037 	 * TCP is, by definition, unicast, so we reject all
1038 	 * multicast outright.
1039 	 *
1040 	 * Note, there are additional src/dst address checks in
1041 	 * the AF-specific code below.
1042 	 */
1043 	if (m->m_flags & (M_BCAST|M_MCAST)) {
1044 		/* XXX stat */
1045 		goto drop;
1046 	}
1047 #ifdef INET6
1048 	if (m->m_flags & M_ANYCAST6) {
1049 		/* XXX stat */
1050 		goto drop;
1051 	}
1052 #endif
1053 
1054 	/*
1055 	 * Get IP and TCP header.
1056 	 * Note: IP leaves IP header in first mbuf.
1057 	 */
1058 	ip = mtod(m, struct ip *);
1059 #ifdef INET6
1060 	ip6 = NULL;
1061 #endif
1062 	switch (ip->ip_v) {
1063 #ifdef INET
1064 	case 4:
1065 		af = AF_INET;
1066 		iphlen = sizeof(struct ip);
1067 		ip = mtod(m, struct ip *);
1068 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1069 			sizeof(struct tcphdr));
1070 		if (th == NULL) {
1071 			TCP_STATINC(TCP_STAT_RCVSHORT);
1072 			return;
1073 		}
1074 		/* We do the checksum after PCB lookup... */
1075 		len = ntohs(ip->ip_len);
1076 		tlen = len - toff;
1077 		iptos = ip->ip_tos;
1078 		break;
1079 #endif
1080 #ifdef INET6
1081 	case 6:
1082 		ip = NULL;
1083 		iphlen = sizeof(struct ip6_hdr);
1084 		af = AF_INET6;
1085 		ip6 = mtod(m, struct ip6_hdr *);
1086 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff,
1087 			sizeof(struct tcphdr));
1088 		if (th == NULL) {
1089 			TCP_STATINC(TCP_STAT_RCVSHORT);
1090 			return;
1091 		}
1092 
1093 		/* Be proactive about malicious use of IPv4 mapped address */
1094 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
1095 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
1096 			/* XXX stat */
1097 			goto drop;
1098 		}
1099 
1100 		/*
1101 		 * Be proactive about unspecified IPv6 address in source.
1102 		 * As we use all-zero to indicate unbounded/unconnected pcb,
1103 		 * unspecified IPv6 address can be used to confuse us.
1104 		 *
1105 		 * Note that packets with unspecified IPv6 destination is
1106 		 * already dropped in ip6_input.
1107 		 */
1108 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
1109 			/* XXX stat */
1110 			goto drop;
1111 		}
1112 
1113 		/*
1114 		 * Make sure destination address is not multicast.
1115 		 * Source address checked in ip6_input().
1116 		 */
1117 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
1118 			/* XXX stat */
1119 			goto drop;
1120 		}
1121 
1122 		/* We do the checksum after PCB lookup... */
1123 		len = m->m_pkthdr.len;
1124 		tlen = len - toff;
1125 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
1126 		break;
1127 #endif
1128 	default:
1129 		m_freem(m);
1130 		return;
1131 	}
1132 
1133 	KASSERT(TCP_HDR_ALIGNED_P(th));
1134 
1135 	/*
1136 	 * Check that TCP offset makes sense,
1137 	 * pull out TCP options and adjust length.		XXX
1138 	 */
1139 	off = th->th_off << 2;
1140 	if (off < sizeof (struct tcphdr) || off > tlen) {
1141 		TCP_STATINC(TCP_STAT_RCVBADOFF);
1142 		goto drop;
1143 	}
1144 	tlen -= off;
1145 
1146 	/*
1147 	 * tcp_input() has been modified to use tlen to mean the TCP data
1148 	 * length throughout the function.  Other functions can use
1149 	 * m->m_pkthdr.len as the basis for calculating the TCP data length.
1150 	 * rja
1151 	 */
1152 
1153 	if (off > sizeof (struct tcphdr)) {
1154 		IP6_EXTHDR_GET(th, struct tcphdr *, m, toff, off);
1155 		if (th == NULL) {
1156 			TCP_STATINC(TCP_STAT_RCVSHORT);
1157 			return;
1158 		}
1159 		/*
1160 		 * NOTE: ip/ip6 will not be affected by m_pulldown()
1161 		 * (as they're before toff) and we don't need to update those.
1162 		 */
1163 		KASSERT(TCP_HDR_ALIGNED_P(th));
1164 		optlen = off - sizeof (struct tcphdr);
1165 		optp = ((u_int8_t *)th) + sizeof(struct tcphdr);
1166 		/*
1167 		 * Do quick retrieval of timestamp options ("options
1168 		 * prediction?").  If timestamp is the only option and it's
1169 		 * formatted as recommended in RFC 1323 appendix A, we
1170 		 * quickly get the values now and not bother calling
1171 		 * tcp_dooptions(), etc.
1172 		 */
1173 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
1174 		     (optlen > TCPOLEN_TSTAMP_APPA &&
1175 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
1176 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
1177 		     (th->th_flags & TH_SYN) == 0) {
1178 			opti.ts_present = 1;
1179 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
1180 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
1181 			optp = NULL;	/* we've parsed the options */
1182 		}
1183 	}
1184 	tiflags = th->th_flags;
1185 
1186 	/*
1187 	 * Locate pcb for segment.
1188 	 */
1189 findpcb:
1190 	inp = NULL;
1191 #ifdef INET6
1192 	in6p = NULL;
1193 #endif
1194 	switch (af) {
1195 #ifdef INET
1196 	case AF_INET:
1197 		inp = in_pcblookup_connect(&tcbtable, ip->ip_src, th->th_sport,
1198 		    ip->ip_dst, th->th_dport);
1199 		if (inp == 0) {
1200 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1201 			inp = in_pcblookup_bind(&tcbtable, ip->ip_dst, th->th_dport);
1202 		}
1203 #ifdef INET6
1204 		if (inp == 0) {
1205 			struct in6_addr s, d;
1206 
1207 			/* mapped addr case */
1208 			memset(&s, 0, sizeof(s));
1209 			s.s6_addr16[5] = htons(0xffff);
1210 			bcopy(&ip->ip_src, &s.s6_addr32[3], sizeof(ip->ip_src));
1211 			memset(&d, 0, sizeof(d));
1212 			d.s6_addr16[5] = htons(0xffff);
1213 			bcopy(&ip->ip_dst, &d.s6_addr32[3], sizeof(ip->ip_dst));
1214 			in6p = in6_pcblookup_connect(&tcbtable, &s,
1215 			    th->th_sport, &d, th->th_dport, 0);
1216 			if (in6p == 0) {
1217 				TCP_STATINC(TCP_STAT_PCBHASHMISS);
1218 				in6p = in6_pcblookup_bind(&tcbtable, &d,
1219 				    th->th_dport, 0);
1220 			}
1221 		}
1222 #endif
1223 #ifndef INET6
1224 		if (inp == 0)
1225 #else
1226 		if (inp == 0 && in6p == 0)
1227 #endif
1228 		{
1229 			TCP_STATINC(TCP_STAT_NOPORT);
1230 			if (tcp_log_refused &&
1231 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1232 				tcp4_log_refused(ip, th);
1233 			}
1234 			tcp_fields_to_host(th);
1235 			goto dropwithreset_ratelim;
1236 		}
1237 #if defined(IPSEC) || defined(FAST_IPSEC)
1238 		if (inp && (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0 &&
1239 		    ipsec4_in_reject(m, inp)) {
1240 			IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1241 			goto drop;
1242 		}
1243 #ifdef INET6
1244 		else if (in6p &&
1245 		    (in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1246 		    ipsec6_in_reject_so(m, in6p->in6p_socket)) {
1247 			IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1248 			goto drop;
1249 		}
1250 #endif
1251 #endif /*IPSEC*/
1252 		break;
1253 #endif /*INET*/
1254 #ifdef INET6
1255 	case AF_INET6:
1256 	    {
1257 		int faith;
1258 
1259 #if defined(NFAITH) && NFAITH > 0
1260 		faith = faithprefix(&ip6->ip6_dst);
1261 #else
1262 		faith = 0;
1263 #endif
1264 		in6p = in6_pcblookup_connect(&tcbtable, &ip6->ip6_src,
1265 		    th->th_sport, &ip6->ip6_dst, th->th_dport, faith);
1266 		if (in6p == NULL) {
1267 			TCP_STATINC(TCP_STAT_PCBHASHMISS);
1268 			in6p = in6_pcblookup_bind(&tcbtable, &ip6->ip6_dst,
1269 				th->th_dport, faith);
1270 		}
1271 		if (in6p == NULL) {
1272 			TCP_STATINC(TCP_STAT_NOPORT);
1273 			if (tcp_log_refused &&
1274 			    (tiflags & (TH_RST|TH_ACK|TH_SYN)) == TH_SYN) {
1275 				tcp6_log_refused(ip6, th);
1276 			}
1277 			tcp_fields_to_host(th);
1278 			goto dropwithreset_ratelim;
1279 		}
1280 #if defined(IPSEC) || defined(FAST_IPSEC)
1281 		if ((in6p->in6p_socket->so_options & SO_ACCEPTCONN) == 0 &&
1282 		    ipsec6_in_reject(m, in6p)) {
1283 			IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO);
1284 			goto drop;
1285 		}
1286 #endif /*IPSEC*/
1287 		break;
1288 	    }
1289 #endif
1290 	}
1291 
1292 	/*
1293 	 * If the state is CLOSED (i.e., TCB does not exist) then
1294 	 * all data in the incoming segment is discarded.
1295 	 * If the TCB exists but is in CLOSED state, it is embryonic,
1296 	 * but should either do a listen or a connect soon.
1297 	 */
1298 	tp = NULL;
1299 	so = NULL;
1300 	if (inp) {
1301 		tp = intotcpcb(inp);
1302 		so = inp->inp_socket;
1303 	}
1304 #ifdef INET6
1305 	else if (in6p) {
1306 		tp = in6totcpcb(in6p);
1307 		so = in6p->in6p_socket;
1308 	}
1309 #endif
1310 	if (tp == 0) {
1311 		tcp_fields_to_host(th);
1312 		goto dropwithreset_ratelim;
1313 	}
1314 	if (tp->t_state == TCPS_CLOSED)
1315 		goto drop;
1316 
1317 	KASSERT(so->so_lock == softnet_lock);
1318 	KASSERT(solocked(so));
1319 
1320 	/*
1321 	 * Checksum extended TCP header and data.
1322 	 */
1323 	if (tcp_input_checksum(af, m, th, toff, off, tlen))
1324 		goto badcsum;
1325 
1326 	tcp_fields_to_host(th);
1327 
1328 	/* Unscale the window into a 32-bit value. */
1329 	if ((tiflags & TH_SYN) == 0)
1330 		tiwin = th->th_win << tp->snd_scale;
1331 	else
1332 		tiwin = th->th_win;
1333 
1334 #ifdef INET6
1335 	/* save packet options if user wanted */
1336 	if (in6p && (in6p->in6p_flags & IN6P_CONTROLOPTS)) {
1337 		if (in6p->in6p_options) {
1338 			m_freem(in6p->in6p_options);
1339 			in6p->in6p_options = 0;
1340 		}
1341 		KASSERT(ip6 != NULL);
1342 		ip6_savecontrol(in6p, &in6p->in6p_options, ip6, m);
1343 	}
1344 #endif
1345 
1346 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
1347 		union syn_cache_sa src;
1348 		union syn_cache_sa dst;
1349 
1350 		memset(&src, 0, sizeof(src));
1351 		memset(&dst, 0, sizeof(dst));
1352 		switch (af) {
1353 #ifdef INET
1354 		case AF_INET:
1355 			src.sin.sin_len = sizeof(struct sockaddr_in);
1356 			src.sin.sin_family = AF_INET;
1357 			src.sin.sin_addr = ip->ip_src;
1358 			src.sin.sin_port = th->th_sport;
1359 
1360 			dst.sin.sin_len = sizeof(struct sockaddr_in);
1361 			dst.sin.sin_family = AF_INET;
1362 			dst.sin.sin_addr = ip->ip_dst;
1363 			dst.sin.sin_port = th->th_dport;
1364 			break;
1365 #endif
1366 #ifdef INET6
1367 		case AF_INET6:
1368 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
1369 			src.sin6.sin6_family = AF_INET6;
1370 			src.sin6.sin6_addr = ip6->ip6_src;
1371 			src.sin6.sin6_port = th->th_sport;
1372 
1373 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
1374 			dst.sin6.sin6_family = AF_INET6;
1375 			dst.sin6.sin6_addr = ip6->ip6_dst;
1376 			dst.sin6.sin6_port = th->th_dport;
1377 			break;
1378 #endif /* INET6 */
1379 		default:
1380 			goto badsyn;	/*sanity*/
1381 		}
1382 
1383 		if (so->so_options & SO_DEBUG) {
1384 #ifdef TCP_DEBUG
1385 			ostate = tp->t_state;
1386 #endif
1387 
1388 			tcp_saveti = NULL;
1389 			if (iphlen + sizeof(struct tcphdr) > MHLEN)
1390 				goto nosave;
1391 
1392 			if (m->m_len > iphlen && (m->m_flags & M_EXT) == 0) {
1393 				tcp_saveti = m_copym(m, 0, iphlen, M_DONTWAIT);
1394 				if (!tcp_saveti)
1395 					goto nosave;
1396 			} else {
1397 				MGETHDR(tcp_saveti, M_DONTWAIT, MT_HEADER);
1398 				if (!tcp_saveti)
1399 					goto nosave;
1400 				MCLAIM(m, &tcp_mowner);
1401 				tcp_saveti->m_len = iphlen;
1402 				m_copydata(m, 0, iphlen,
1403 				    mtod(tcp_saveti, void *));
1404 			}
1405 
1406 			if (M_TRAILINGSPACE(tcp_saveti) < sizeof(struct tcphdr)) {
1407 				m_freem(tcp_saveti);
1408 				tcp_saveti = NULL;
1409 			} else {
1410 				tcp_saveti->m_len += sizeof(struct tcphdr);
1411 				memcpy(mtod(tcp_saveti, char *) + iphlen, th,
1412 				    sizeof(struct tcphdr));
1413 			}
1414 	nosave:;
1415 		}
1416 		if (so->so_options & SO_ACCEPTCONN) {
1417 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1418 				if (tiflags & TH_RST) {
1419 					syn_cache_reset(&src.sa, &dst.sa, th);
1420 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
1421 				    (TH_ACK|TH_SYN)) {
1422 					/*
1423 					 * Received a SYN,ACK.  This should
1424 					 * never happen while we are in
1425 					 * LISTEN.  Send an RST.
1426 					 */
1427 					goto badsyn;
1428 				} else if (tiflags & TH_ACK) {
1429 					so = syn_cache_get(&src.sa, &dst.sa,
1430 						th, toff, tlen, so, m);
1431 					if (so == NULL) {
1432 						/*
1433 						 * We don't have a SYN for
1434 						 * this ACK; send an RST.
1435 						 */
1436 						goto badsyn;
1437 					} else if (so ==
1438 					    (struct socket *)(-1)) {
1439 						/*
1440 						 * We were unable to create
1441 						 * the connection.  If the
1442 						 * 3-way handshake was
1443 						 * completed, and RST has
1444 						 * been sent to the peer.
1445 						 * Since the mbuf might be
1446 						 * in use for the reply,
1447 						 * do not free it.
1448 						 */
1449 						m = NULL;
1450 					} else {
1451 						/*
1452 						 * We have created a
1453 						 * full-blown connection.
1454 						 */
1455 						tp = NULL;
1456 						inp = NULL;
1457 #ifdef INET6
1458 						in6p = NULL;
1459 #endif
1460 						switch (so->so_proto->pr_domain->dom_family) {
1461 #ifdef INET
1462 						case AF_INET:
1463 							inp = sotoinpcb(so);
1464 							tp = intotcpcb(inp);
1465 							break;
1466 #endif
1467 #ifdef INET6
1468 						case AF_INET6:
1469 							in6p = sotoin6pcb(so);
1470 							tp = in6totcpcb(in6p);
1471 							break;
1472 #endif
1473 						}
1474 						if (tp == NULL)
1475 							goto badsyn;	/*XXX*/
1476 						tiwin <<= tp->snd_scale;
1477 						goto after_listen;
1478 					}
1479 				} else {
1480 					/*
1481 					 * None of RST, SYN or ACK was set.
1482 					 * This is an invalid packet for a
1483 					 * TCB in LISTEN state.  Send a RST.
1484 					 */
1485 					goto badsyn;
1486 				}
1487 			} else {
1488 				/*
1489 				 * Received a SYN.
1490 				 *
1491 				 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1492 				 */
1493 				if (m->m_flags & (M_BCAST|M_MCAST))
1494 					goto drop;
1495 
1496 				switch (af) {
1497 #ifdef INET6
1498 				case AF_INET6:
1499 					if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
1500 						goto drop;
1501 					break;
1502 #endif /* INET6 */
1503 				case AF_INET:
1504 					if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1505 					    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1506 						goto drop;
1507 				break;
1508 				}
1509 
1510 #ifdef INET6
1511 				/*
1512 				 * If deprecated address is forbidden, we do
1513 				 * not accept SYN to deprecated interface
1514 				 * address to prevent any new inbound
1515 				 * connection from getting established.
1516 				 * When we do not accept SYN, we send a TCP
1517 				 * RST, with deprecated source address (instead
1518 				 * of dropping it).  We compromise it as it is
1519 				 * much better for peer to send a RST, and
1520 				 * RST will be the final packet for the
1521 				 * exchange.
1522 				 *
1523 				 * If we do not forbid deprecated addresses, we
1524 				 * accept the SYN packet.  RFC2462 does not
1525 				 * suggest dropping SYN in this case.
1526 				 * If we decipher RFC2462 5.5.4, it says like
1527 				 * this:
1528 				 * 1. use of deprecated addr with existing
1529 				 *    communication is okay - "SHOULD continue
1530 				 *    to be used"
1531 				 * 2. use of it with new communication:
1532 				 *   (2a) "SHOULD NOT be used if alternate
1533 				 *        address with sufficient scope is
1534 				 *        available"
1535 				 *   (2b) nothing mentioned otherwise.
1536 				 * Here we fall into (2b) case as we have no
1537 				 * choice in our source address selection - we
1538 				 * must obey the peer.
1539 				 *
1540 				 * The wording in RFC2462 is confusing, and
1541 				 * there are multiple description text for
1542 				 * deprecated address handling - worse, they
1543 				 * are not exactly the same.  I believe 5.5.4
1544 				 * is the best one, so we follow 5.5.4.
1545 				 */
1546 				if (af == AF_INET6 && !ip6_use_deprecated) {
1547 					struct in6_ifaddr *ia6;
1548 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
1549 					    &ip6->ip6_dst)) &&
1550 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1551 						tp = NULL;
1552 						goto dropwithreset;
1553 					}
1554 				}
1555 #endif
1556 
1557 #if defined(IPSEC) || defined(FAST_IPSEC)
1558 				switch (af) {
1559 #ifdef INET
1560 				case AF_INET:
1561 					if (ipsec4_in_reject_so(m, so)) {
1562 						IPSEC_STATINC(IPSEC_STAT_IN_POLVIO);
1563 						tp = NULL;
1564 						goto dropwithreset;
1565 					}
1566 					break;
1567 #endif
1568 #ifdef INET6
1569 				case AF_INET6:
1570 					if (ipsec6_in_reject_so(m, so)) {
1571 						IPSEC6_STATINC(IPSEC_STAT_IN_POLVIO);
1572 						tp = NULL;
1573 						goto dropwithreset;
1574 					}
1575 					break;
1576 #endif /*INET6*/
1577 				}
1578 #endif /*IPSEC*/
1579 
1580 				/*
1581 				 * LISTEN socket received a SYN
1582 				 * from itself?  This can't possibly
1583 				 * be valid; drop the packet.
1584 				 */
1585 				if (th->th_sport == th->th_dport) {
1586 					int i;
1587 
1588 					switch (af) {
1589 #ifdef INET
1590 					case AF_INET:
1591 						i = in_hosteq(ip->ip_src, ip->ip_dst);
1592 						break;
1593 #endif
1594 #ifdef INET6
1595 					case AF_INET6:
1596 						i = IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, &ip6->ip6_dst);
1597 						break;
1598 #endif
1599 					default:
1600 						i = 1;
1601 					}
1602 					if (i) {
1603 						TCP_STATINC(TCP_STAT_BADSYN);
1604 						goto drop;
1605 					}
1606 				}
1607 
1608 				/*
1609 				 * SYN looks ok; create compressed TCP
1610 				 * state for it.
1611 				 */
1612 				if (so->so_qlen <= so->so_qlimit &&
1613 				    syn_cache_add(&src.sa, &dst.sa, th, tlen,
1614 						so, m, optp, optlen, &opti))
1615 					m = NULL;
1616 			}
1617 			goto drop;
1618 		}
1619 	}
1620 
1621 after_listen:
1622 #ifdef DIAGNOSTIC
1623 	/*
1624 	 * Should not happen now that all embryonic connections
1625 	 * are handled with compressed state.
1626 	 */
1627 	if (tp->t_state == TCPS_LISTEN)
1628 		panic("tcp_input: TCPS_LISTEN");
1629 #endif
1630 
1631 	/*
1632 	 * Segment received on connection.
1633 	 * Reset idle time and keep-alive timer.
1634 	 */
1635 	tp->t_rcvtime = tcp_now;
1636 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1637 		TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1638 
1639 	/*
1640 	 * Process options.
1641 	 */
1642 #ifdef TCP_SIGNATURE
1643 	if (optp || (tp->t_flags & TF_SIGNATURE))
1644 #else
1645 	if (optp)
1646 #endif
1647 		if (tcp_dooptions(tp, optp, optlen, th, m, toff, &opti) < 0)
1648 			goto drop;
1649 
1650 	if (TCP_SACK_ENABLED(tp)) {
1651 		tcp_del_sackholes(tp, th);
1652 	}
1653 
1654 	if (TCP_ECN_ALLOWED(tp)) {
1655 		switch (iptos & IPTOS_ECN_MASK) {
1656 		case IPTOS_ECN_CE:
1657 			tp->t_flags |= TF_ECN_SND_ECE;
1658 			TCP_STATINC(TCP_STAT_ECN_CE);
1659 			break;
1660 		case IPTOS_ECN_ECT0:
1661 			TCP_STATINC(TCP_STAT_ECN_ECT);
1662 			break;
1663 		case IPTOS_ECN_ECT1:
1664 			/* XXX: ignore for now -- rpaulo */
1665 			break;
1666 		}
1667 
1668 		if (tiflags & TH_CWR)
1669 			tp->t_flags &= ~TF_ECN_SND_ECE;
1670 
1671 		/*
1672 		 * Congestion experienced.
1673 		 * Ignore if we are already trying to recover.
1674 		 */
1675 		if ((tiflags & TH_ECE) && SEQ_GEQ(tp->snd_una, tp->snd_recover))
1676 			tp->t_congctl->cong_exp(tp);
1677 	}
1678 
1679 	if (opti.ts_present && opti.ts_ecr) {
1680 		/*
1681 		 * Calculate the RTT from the returned time stamp and the
1682 		 * connection's time base.  If the time stamp is later than
1683 		 * the current time, or is extremely old, fall back to non-1323
1684 		 * RTT calculation.  Since ts_ecr is unsigned, we can test both
1685 		 * at the same time.
1686 		 */
1687 		ts_rtt = TCP_TIMESTAMP(tp) - opti.ts_ecr + 1;
1688 		if (ts_rtt > TCP_PAWS_IDLE)
1689 			ts_rtt = 0;
1690 	} else {
1691 		ts_rtt = 0;
1692 	}
1693 
1694 	/*
1695 	 * Header prediction: check for the two common cases
1696 	 * of a uni-directional data xfer.  If the packet has
1697 	 * no control flags, is in-sequence, the window didn't
1698 	 * change and we're not retransmitting, it's a
1699 	 * candidate.  If the length is zero and the ack moved
1700 	 * forward, we're the sender side of the xfer.  Just
1701 	 * free the data acked & wake any higher level process
1702 	 * that was blocked waiting for space.  If the length
1703 	 * is non-zero and the ack didn't move, we're the
1704 	 * receiver side.  If we're getting packets in-order
1705 	 * (the reassembly queue is empty), add the data to
1706 	 * the socket buffer and note that we need a delayed ack.
1707 	 */
1708 	if (tp->t_state == TCPS_ESTABLISHED &&
1709 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK))
1710 	        == TH_ACK &&
1711 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1712 	    th->th_seq == tp->rcv_nxt &&
1713 	    tiwin && tiwin == tp->snd_wnd &&
1714 	    tp->snd_nxt == tp->snd_max) {
1715 
1716 		/*
1717 		 * If last ACK falls within this segment's sequence numbers,
1718 		 * record the timestamp.
1719 		 * NOTE that the test is modified according to the latest
1720 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1721 		 *
1722 		 * note that we already know
1723 		 *	TSTMP_GEQ(opti.ts_val, tp->ts_recent)
1724 		 */
1725 		if (opti.ts_present &&
1726 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1727 			tp->ts_recent_age = tcp_now;
1728 			tp->ts_recent = opti.ts_val;
1729 		}
1730 
1731 		if (tlen == 0) {
1732 			/* Ack prediction. */
1733 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1734 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1735 			    tp->snd_cwnd >= tp->snd_wnd &&
1736 			    tp->t_partialacks < 0) {
1737 				/*
1738 				 * this is a pure ack for outstanding data.
1739 				 */
1740 				if (ts_rtt)
1741 					tcp_xmit_timer(tp, ts_rtt);
1742 				else if (tp->t_rtttime &&
1743 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1744 					tcp_xmit_timer(tp,
1745 					  tcp_now - tp->t_rtttime);
1746 				acked = th->th_ack - tp->snd_una;
1747 				tcps = TCP_STAT_GETREF();
1748 				tcps[TCP_STAT_PREDACK]++;
1749 				tcps[TCP_STAT_RCVACKPACK]++;
1750 				tcps[TCP_STAT_RCVACKBYTE] += acked;
1751 				TCP_STAT_PUTREF();
1752 				nd6_hint(tp);
1753 
1754 				if (acked > (tp->t_lastoff - tp->t_inoff))
1755 					tp->t_lastm = NULL;
1756 				sbdrop(&so->so_snd, acked);
1757 				tp->t_lastoff -= acked;
1758 
1759 				icmp_check(tp, th, acked);
1760 
1761 				tp->snd_una = th->th_ack;
1762 				tp->snd_fack = tp->snd_una;
1763 				if (SEQ_LT(tp->snd_high, tp->snd_una))
1764 					tp->snd_high = tp->snd_una;
1765 				m_freem(m);
1766 
1767 				/*
1768 				 * If all outstanding data are acked, stop
1769 				 * retransmit timer, otherwise restart timer
1770 				 * using current (possibly backed-off) value.
1771 				 * If process is waiting for space,
1772 				 * wakeup/selnotify/signal.  If data
1773 				 * are ready to send, let tcp_output
1774 				 * decide between more output or persist.
1775 				 */
1776 				if (tp->snd_una == tp->snd_max)
1777 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1778 				else if (TCP_TIMER_ISARMED(tp,
1779 				    TCPT_PERSIST) == 0)
1780 					TCP_TIMER_ARM(tp, TCPT_REXMT,
1781 					    tp->t_rxtcur);
1782 
1783 				sowwakeup(so);
1784 				if (so->so_snd.sb_cc)
1785 					(void) tcp_output(tp);
1786 				if (tcp_saveti)
1787 					m_freem(tcp_saveti);
1788 				return;
1789 			}
1790 		} else if (th->th_ack == tp->snd_una &&
1791 		    TAILQ_FIRST(&tp->segq) == NULL &&
1792 		    tlen <= sbspace(&so->so_rcv)) {
1793 			int newsize = 0;	/* automatic sockbuf scaling */
1794 
1795 			/*
1796 			 * this is a pure, in-sequence data packet
1797 			 * with nothing on the reassembly queue and
1798 			 * we have enough buffer space to take it.
1799 			 */
1800 			tp->rcv_nxt += tlen;
1801 			tcps = TCP_STAT_GETREF();
1802 			tcps[TCP_STAT_PREDDAT]++;
1803 			tcps[TCP_STAT_RCVPACK]++;
1804 			tcps[TCP_STAT_RCVBYTE] += tlen;
1805 			TCP_STAT_PUTREF();
1806 			nd6_hint(tp);
1807 
1808 		/*
1809 		 * Automatic sizing enables the performance of large buffers
1810 		 * and most of the efficiency of small ones by only allocating
1811 		 * space when it is needed.
1812 		 *
1813 		 * On the receive side the socket buffer memory is only rarely
1814 		 * used to any significant extent.  This allows us to be much
1815 		 * more aggressive in scaling the receive socket buffer.  For
1816 		 * the case that the buffer space is actually used to a large
1817 		 * extent and we run out of kernel memory we can simply drop
1818 		 * the new segments; TCP on the sender will just retransmit it
1819 		 * later.  Setting the buffer size too big may only consume too
1820 		 * much kernel memory if the application doesn't read() from
1821 		 * the socket or packet loss or reordering makes use of the
1822 		 * reassembly queue.
1823 		 *
1824 		 * The criteria to step up the receive buffer one notch are:
1825 		 *  1. the number of bytes received during the time it takes
1826 		 *     one timestamp to be reflected back to us (the RTT);
1827 		 *  2. received bytes per RTT is within seven eighth of the
1828 		 *     current socket buffer size;
1829 		 *  3. receive buffer size has not hit maximal automatic size;
1830 		 *
1831 		 * This algorithm does one step per RTT at most and only if
1832 		 * we receive a bulk stream w/o packet losses or reorderings.
1833 		 * Shrinking the buffer during idle times is not necessary as
1834 		 * it doesn't consume any memory when idle.
1835 		 *
1836 		 * TODO: Only step up if the application is actually serving
1837 		 * the buffer to better manage the socket buffer resources.
1838 		 */
1839 			if (tcp_do_autorcvbuf &&
1840 			    opti.ts_ecr &&
1841 			    (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1842 				if (opti.ts_ecr > tp->rfbuf_ts &&
1843 				    opti.ts_ecr - tp->rfbuf_ts < PR_SLOWHZ) {
1844 					if (tp->rfbuf_cnt >
1845 					    (so->so_rcv.sb_hiwat / 8 * 7) &&
1846 					    so->so_rcv.sb_hiwat <
1847 					    tcp_autorcvbuf_max) {
1848 						newsize =
1849 						    min(so->so_rcv.sb_hiwat +
1850 						    tcp_autorcvbuf_inc,
1851 						    tcp_autorcvbuf_max);
1852 					}
1853 					/* Start over with next RTT. */
1854 					tp->rfbuf_ts = 0;
1855 					tp->rfbuf_cnt = 0;
1856 				} else
1857 					tp->rfbuf_cnt += tlen;	/* add up */
1858 			}
1859 
1860 			/*
1861 			 * Drop TCP, IP headers and TCP options then add data
1862 			 * to socket buffer.
1863 			 */
1864 			if (so->so_state & SS_CANTRCVMORE)
1865 				m_freem(m);
1866 			else {
1867 				/*
1868 				 * Set new socket buffer size.
1869 				 * Give up when limit is reached.
1870 				 */
1871 				if (newsize)
1872 					if (!sbreserve(&so->so_rcv,
1873 					    newsize, so))
1874 						so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1875 				m_adj(m, toff + off);
1876 				sbappendstream(&so->so_rcv, m);
1877 			}
1878 			sorwakeup(so);
1879 			tcp_setup_ack(tp, th);
1880 			if (tp->t_flags & TF_ACKNOW)
1881 				(void) tcp_output(tp);
1882 			if (tcp_saveti)
1883 				m_freem(tcp_saveti);
1884 			return;
1885 		}
1886 	}
1887 
1888 	/*
1889 	 * Compute mbuf offset to TCP data segment.
1890 	 */
1891 	hdroptlen = toff + off;
1892 
1893 	/*
1894 	 * Calculate amount of space in receive window,
1895 	 * and then do TCP input processing.
1896 	 * Receive window is amount of space in rcv queue,
1897 	 * but not less than advertised window.
1898 	 */
1899 	{ int win;
1900 
1901 	win = sbspace(&so->so_rcv);
1902 	if (win < 0)
1903 		win = 0;
1904 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1905 	}
1906 
1907 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1908 	tp->rfbuf_ts = 0;
1909 	tp->rfbuf_cnt = 0;
1910 
1911 	switch (tp->t_state) {
1912 	/*
1913 	 * If the state is SYN_SENT:
1914 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1915 	 *	if seg contains a RST, then drop the connection.
1916 	 *	if seg does not contain SYN, then drop it.
1917 	 * Otherwise this is an acceptable SYN segment
1918 	 *	initialize tp->rcv_nxt and tp->irs
1919 	 *	if seg contains ack then advance tp->snd_una
1920 	 *	if seg contains a ECE and ECN support is enabled, the stream
1921 	 *	    is ECN capable.
1922 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1923 	 *	arrange for segment to be acked (eventually)
1924 	 *	continue processing rest of data/controls, beginning with URG
1925 	 */
1926 	case TCPS_SYN_SENT:
1927 		if ((tiflags & TH_ACK) &&
1928 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1929 		     SEQ_GT(th->th_ack, tp->snd_max)))
1930 			goto dropwithreset;
1931 		if (tiflags & TH_RST) {
1932 			if (tiflags & TH_ACK)
1933 				tp = tcp_drop(tp, ECONNREFUSED);
1934 			goto drop;
1935 		}
1936 		if ((tiflags & TH_SYN) == 0)
1937 			goto drop;
1938 		if (tiflags & TH_ACK) {
1939 			tp->snd_una = th->th_ack;
1940 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1941 				tp->snd_nxt = tp->snd_una;
1942 			if (SEQ_LT(tp->snd_high, tp->snd_una))
1943 				tp->snd_high = tp->snd_una;
1944 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1945 
1946 			if ((tiflags & TH_ECE) && tcp_do_ecn) {
1947 				tp->t_flags |= TF_ECN_PERMIT;
1948 				TCP_STATINC(TCP_STAT_ECN_SHS);
1949 			}
1950 
1951 		}
1952 		tp->irs = th->th_seq;
1953 		tcp_rcvseqinit(tp);
1954 		tp->t_flags |= TF_ACKNOW;
1955 		tcp_mss_from_peer(tp, opti.maxseg);
1956 
1957 		/*
1958 		 * Initialize the initial congestion window.  If we
1959 		 * had to retransmit the SYN, we must initialize cwnd
1960 		 * to 1 segment (i.e. the Loss Window).
1961 		 */
1962 		if (tp->t_flags & TF_SYN_REXMT)
1963 			tp->snd_cwnd = tp->t_peermss;
1964 		else {
1965 			int ss = tcp_init_win;
1966 #ifdef INET
1967 			if (inp != NULL && in_localaddr(inp->inp_faddr))
1968 				ss = tcp_init_win_local;
1969 #endif
1970 #ifdef INET6
1971 			if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
1972 				ss = tcp_init_win_local;
1973 #endif
1974 			tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
1975 		}
1976 
1977 		tcp_rmx_rtt(tp);
1978 		if (tiflags & TH_ACK) {
1979 			TCP_STATINC(TCP_STAT_CONNECTS);
1980 			soisconnected(so);
1981 			tcp_established(tp);
1982 			/* Do window scaling on this connection? */
1983 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1984 			    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1985 				tp->snd_scale = tp->requested_s_scale;
1986 				tp->rcv_scale = tp->request_r_scale;
1987 			}
1988 			TCP_REASS_LOCK(tp);
1989 			(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
1990 			TCP_REASS_UNLOCK(tp);
1991 			/*
1992 			 * if we didn't have to retransmit the SYN,
1993 			 * use its rtt as our initial srtt & rtt var.
1994 			 */
1995 			if (tp->t_rtttime)
1996 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1997 		} else
1998 			tp->t_state = TCPS_SYN_RECEIVED;
1999 
2000 		/*
2001 		 * Advance th->th_seq to correspond to first data byte.
2002 		 * If data, trim to stay within window,
2003 		 * dropping FIN if necessary.
2004 		 */
2005 		th->th_seq++;
2006 		if (tlen > tp->rcv_wnd) {
2007 			todrop = tlen - tp->rcv_wnd;
2008 			m_adj(m, -todrop);
2009 			tlen = tp->rcv_wnd;
2010 			tiflags &= ~TH_FIN;
2011 			tcps = TCP_STAT_GETREF();
2012 			tcps[TCP_STAT_RCVPACKAFTERWIN]++;
2013 			tcps[TCP_STAT_RCVBYTEAFTERWIN] += todrop;
2014 			TCP_STAT_PUTREF();
2015 		}
2016 		tp->snd_wl1 = th->th_seq - 1;
2017 		tp->rcv_up = th->th_seq;
2018 		goto step6;
2019 
2020 	/*
2021 	 * If the state is SYN_RECEIVED:
2022 	 *	If seg contains an ACK, but not for our SYN, drop the input
2023 	 *	and generate an RST.  See page 36, rfc793
2024 	 */
2025 	case TCPS_SYN_RECEIVED:
2026 		if ((tiflags & TH_ACK) &&
2027 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
2028 		     SEQ_GT(th->th_ack, tp->snd_max)))
2029 			goto dropwithreset;
2030 		break;
2031 	}
2032 
2033 	/*
2034 	 * States other than LISTEN or SYN_SENT.
2035 	 * First check timestamp, if present.
2036 	 * Then check that at least some bytes of segment are within
2037 	 * receive window.  If segment begins before rcv_nxt,
2038 	 * drop leading data (and SYN); if nothing left, just ack.
2039 	 *
2040 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2041 	 * and it's less than ts_recent, drop it.
2042 	 */
2043 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
2044 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
2045 
2046 		/* Check to see if ts_recent is over 24 days old.  */
2047 		if (tcp_now - tp->ts_recent_age > TCP_PAWS_IDLE) {
2048 			/*
2049 			 * Invalidate ts_recent.  If this segment updates
2050 			 * ts_recent, the age will be reset later and ts_recent
2051 			 * will get a valid value.  If it does not, setting
2052 			 * ts_recent to zero will at least satisfy the
2053 			 * requirement that zero be placed in the timestamp
2054 			 * echo reply when ts_recent isn't valid.  The
2055 			 * age isn't reset until we get a valid ts_recent
2056 			 * because we don't want out-of-order segments to be
2057 			 * dropped when ts_recent is old.
2058 			 */
2059 			tp->ts_recent = 0;
2060 		} else {
2061 			tcps = TCP_STAT_GETREF();
2062 			tcps[TCP_STAT_RCVDUPPACK]++;
2063 			tcps[TCP_STAT_RCVDUPBYTE] += tlen;
2064 			tcps[TCP_STAT_PAWSDROP]++;
2065 			TCP_STAT_PUTREF();
2066 			tcp_new_dsack(tp, th->th_seq, tlen);
2067 			goto dropafterack;
2068 		}
2069 	}
2070 
2071 	todrop = tp->rcv_nxt - th->th_seq;
2072 	dupseg = false;
2073 	if (todrop > 0) {
2074 		if (tiflags & TH_SYN) {
2075 			tiflags &= ~TH_SYN;
2076 			th->th_seq++;
2077 			if (th->th_urp > 1)
2078 				th->th_urp--;
2079 			else {
2080 				tiflags &= ~TH_URG;
2081 				th->th_urp = 0;
2082 			}
2083 			todrop--;
2084 		}
2085 		if (todrop > tlen ||
2086 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
2087 			/*
2088 			 * Any valid FIN or RST must be to the left of the
2089 			 * window.  At this point the FIN or RST must be a
2090 			 * duplicate or out of sequence; drop it.
2091 			 */
2092 			if (tiflags & TH_RST)
2093 				goto drop;
2094 			tiflags &= ~(TH_FIN|TH_RST);
2095 			/*
2096 			 * Send an ACK to resynchronize and drop any data.
2097 			 * But keep on processing for RST or ACK.
2098 			 */
2099 			tp->t_flags |= TF_ACKNOW;
2100 			todrop = tlen;
2101 			dupseg = true;
2102 			tcps = TCP_STAT_GETREF();
2103 			tcps[TCP_STAT_RCVDUPPACK]++;
2104 			tcps[TCP_STAT_RCVDUPBYTE] += todrop;
2105 			TCP_STAT_PUTREF();
2106 		} else if ((tiflags & TH_RST) &&
2107 			   th->th_seq != tp->last_ack_sent) {
2108 			/*
2109 			 * Test for reset before adjusting the sequence
2110 			 * number for overlapping data.
2111 			 */
2112 			goto dropafterack_ratelim;
2113 		} else {
2114 			tcps = TCP_STAT_GETREF();
2115 			tcps[TCP_STAT_RCVPARTDUPPACK]++;
2116 			tcps[TCP_STAT_RCVPARTDUPBYTE] += todrop;
2117 			TCP_STAT_PUTREF();
2118 		}
2119 		tcp_new_dsack(tp, th->th_seq, todrop);
2120 		hdroptlen += todrop;	/*drop from head afterwards*/
2121 		th->th_seq += todrop;
2122 		tlen -= todrop;
2123 		if (th->th_urp > todrop)
2124 			th->th_urp -= todrop;
2125 		else {
2126 			tiflags &= ~TH_URG;
2127 			th->th_urp = 0;
2128 		}
2129 	}
2130 
2131 	/*
2132 	 * If new data are received on a connection after the
2133 	 * user processes are gone, then RST the other end.
2134 	 */
2135 	if ((so->so_state & SS_NOFDREF) &&
2136 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2137 		tp = tcp_close(tp);
2138 		TCP_STATINC(TCP_STAT_RCVAFTERCLOSE);
2139 		goto dropwithreset;
2140 	}
2141 
2142 	/*
2143 	 * If segment ends after window, drop trailing data
2144 	 * (and PUSH and FIN); if nothing left, just ACK.
2145 	 */
2146 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2147 	if (todrop > 0) {
2148 		TCP_STATINC(TCP_STAT_RCVPACKAFTERWIN);
2149 		if (todrop >= tlen) {
2150 			/*
2151 			 * The segment actually starts after the window.
2152 			 * th->th_seq + tlen - tp->rcv_nxt - tp->rcv_wnd >= tlen
2153 			 * th->th_seq - tp->rcv_nxt - tp->rcv_wnd >= 0
2154 			 * th->th_seq >= tp->rcv_nxt + tp->rcv_wnd
2155 			 */
2156 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, tlen);
2157 			/*
2158 			 * If a new connection request is received
2159 			 * while in TIME_WAIT, drop the old connection
2160 			 * and start over if the sequence numbers
2161 			 * are above the previous ones.
2162 			 *
2163 			 * NOTE: We will checksum the packet again, and
2164 			 * so we need to put the header fields back into
2165 			 * network order!
2166 			 * XXX This kind of sucks, but we don't expect
2167 			 * XXX this to happen very often, so maybe it
2168 			 * XXX doesn't matter so much.
2169 			 */
2170 			if (tiflags & TH_SYN &&
2171 			    tp->t_state == TCPS_TIME_WAIT &&
2172 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2173 				tp = tcp_close(tp);
2174 				tcp_fields_to_net(th);
2175 				goto findpcb;
2176 			}
2177 			/*
2178 			 * If window is closed can only take segments at
2179 			 * window edge, and have to drop data and PUSH from
2180 			 * incoming segments.  Continue processing, but
2181 			 * remember to ack.  Otherwise, drop segment
2182 			 * and (if not RST) ack.
2183 			 */
2184 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2185 				tp->t_flags |= TF_ACKNOW;
2186 				TCP_STATINC(TCP_STAT_RCVWINPROBE);
2187 			} else
2188 				goto dropafterack;
2189 		} else
2190 			TCP_STATADD(TCP_STAT_RCVBYTEAFTERWIN, todrop);
2191 		m_adj(m, -todrop);
2192 		tlen -= todrop;
2193 		tiflags &= ~(TH_PUSH|TH_FIN);
2194 	}
2195 
2196 	/*
2197 	 * If last ACK falls within this segment's sequence numbers,
2198 	 *  record the timestamp.
2199 	 * NOTE:
2200 	 * 1) That the test incorporates suggestions from the latest
2201 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
2202 	 * 2) That updating only on newer timestamps interferes with
2203 	 *    our earlier PAWS tests, so this check should be solely
2204 	 *    predicated on the sequence space of this segment.
2205 	 * 3) That we modify the segment boundary check to be
2206 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
2207 	 *    instead of RFC1323's
2208 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
2209 	 *    This modified check allows us to overcome RFC1323's
2210 	 *    limitations as described in Stevens TCP/IP Illustrated
2211 	 *    Vol. 2 p.869. In such cases, we can still calculate the
2212 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
2213 	 */
2214 	if (opti.ts_present &&
2215 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2216 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2217 		    ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
2218 		tp->ts_recent_age = tcp_now;
2219 		tp->ts_recent = opti.ts_val;
2220 	}
2221 
2222 	/*
2223 	 * If the RST bit is set examine the state:
2224 	 *    SYN_RECEIVED STATE:
2225 	 *	If passive open, return to LISTEN state.
2226 	 *	If active open, inform user that connection was refused.
2227 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
2228 	 *	Inform user that connection was reset, and close tcb.
2229 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
2230 	 *	Close the tcb.
2231 	 */
2232 	if (tiflags & TH_RST) {
2233 		if (th->th_seq != tp->last_ack_sent)
2234 			goto dropafterack_ratelim;
2235 
2236 		switch (tp->t_state) {
2237 		case TCPS_SYN_RECEIVED:
2238 			so->so_error = ECONNREFUSED;
2239 			goto close;
2240 
2241 		case TCPS_ESTABLISHED:
2242 		case TCPS_FIN_WAIT_1:
2243 		case TCPS_FIN_WAIT_2:
2244 		case TCPS_CLOSE_WAIT:
2245 			so->so_error = ECONNRESET;
2246 		close:
2247 			tp->t_state = TCPS_CLOSED;
2248 			TCP_STATINC(TCP_STAT_DROPS);
2249 			tp = tcp_close(tp);
2250 			goto drop;
2251 
2252 		case TCPS_CLOSING:
2253 		case TCPS_LAST_ACK:
2254 		case TCPS_TIME_WAIT:
2255 			tp = tcp_close(tp);
2256 			goto drop;
2257 		}
2258 	}
2259 
2260 	/*
2261 	 * Since we've covered the SYN-SENT and SYN-RECEIVED states above
2262 	 * we must be in a synchronized state.  RFC791 states (under RST
2263 	 * generation) that any unacceptable segment (an out-of-order SYN
2264 	 * qualifies) received in a synchronized state must elicit only an
2265 	 * empty acknowledgment segment ... and the connection remains in
2266 	 * the same state.
2267 	 */
2268 	if (tiflags & TH_SYN) {
2269 		if (tp->rcv_nxt == th->th_seq) {
2270 			tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack - 1,
2271 			    TH_ACK);
2272 			if (tcp_saveti)
2273 				m_freem(tcp_saveti);
2274 			return;
2275 		}
2276 
2277 		goto dropafterack_ratelim;
2278 	}
2279 
2280 	/*
2281 	 * If the ACK bit is off we drop the segment and return.
2282 	 */
2283 	if ((tiflags & TH_ACK) == 0) {
2284 		if (tp->t_flags & TF_ACKNOW)
2285 			goto dropafterack;
2286 		else
2287 			goto drop;
2288 	}
2289 
2290 	/*
2291 	 * Ack processing.
2292 	 */
2293 	switch (tp->t_state) {
2294 
2295 	/*
2296 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
2297 	 * ESTABLISHED state and continue processing, otherwise
2298 	 * send an RST.
2299 	 */
2300 	case TCPS_SYN_RECEIVED:
2301 		if (SEQ_GT(tp->snd_una, th->th_ack) ||
2302 		    SEQ_GT(th->th_ack, tp->snd_max))
2303 			goto dropwithreset;
2304 		TCP_STATINC(TCP_STAT_CONNECTS);
2305 		soisconnected(so);
2306 		tcp_established(tp);
2307 		/* Do window scaling? */
2308 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2309 		    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2310 			tp->snd_scale = tp->requested_s_scale;
2311 			tp->rcv_scale = tp->request_r_scale;
2312 		}
2313 		TCP_REASS_LOCK(tp);
2314 		(void) tcp_reass(tp, NULL, (struct mbuf *)0, &tlen);
2315 		TCP_REASS_UNLOCK(tp);
2316 		tp->snd_wl1 = th->th_seq - 1;
2317 		/* fall into ... */
2318 
2319 	/*
2320 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2321 	 * ACKs.  If the ack is in the range
2322 	 *	tp->snd_una < th->th_ack <= tp->snd_max
2323 	 * then advance tp->snd_una to th->th_ack and drop
2324 	 * data from the retransmission queue.  If this ACK reflects
2325 	 * more up to date window information we update our window information.
2326 	 */
2327 	case TCPS_ESTABLISHED:
2328 	case TCPS_FIN_WAIT_1:
2329 	case TCPS_FIN_WAIT_2:
2330 	case TCPS_CLOSE_WAIT:
2331 	case TCPS_CLOSING:
2332 	case TCPS_LAST_ACK:
2333 	case TCPS_TIME_WAIT:
2334 
2335 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2336 			if (tlen == 0 && !dupseg && tiwin == tp->snd_wnd) {
2337 				TCP_STATINC(TCP_STAT_RCVDUPPACK);
2338 				/*
2339 				 * If we have outstanding data (other than
2340 				 * a window probe), this is a completely
2341 				 * duplicate ack (ie, window info didn't
2342 				 * change), the ack is the biggest we've
2343 				 * seen and we've seen exactly our rexmt
2344 				 * threshhold of them, assume a packet
2345 				 * has been dropped and retransmit it.
2346 				 * Kludge snd_nxt & the congestion
2347 				 * window so we send only this one
2348 				 * packet.
2349 				 */
2350 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
2351 				    th->th_ack != tp->snd_una)
2352 					tp->t_dupacks = 0;
2353 				else if (tp->t_partialacks < 0 &&
2354 					 (++tp->t_dupacks == tcprexmtthresh ||
2355 					 TCP_FACK_FASTRECOV(tp))) {
2356 					/*
2357 					 * Do the fast retransmit, and adjust
2358 					 * congestion control paramenters.
2359 					 */
2360 					if (tp->t_congctl->fast_retransmit(tp, th)) {
2361 						/* False fast retransmit */
2362 						break;
2363 					} else
2364 						goto drop;
2365 				} else if (tp->t_dupacks > tcprexmtthresh) {
2366 					tp->snd_cwnd += tp->t_segsz;
2367 					(void) tcp_output(tp);
2368 					goto drop;
2369 				}
2370 			} else {
2371 				/*
2372 				 * If the ack appears to be very old, only
2373 				 * allow data that is in-sequence.  This
2374 				 * makes it somewhat more difficult to insert
2375 				 * forged data by guessing sequence numbers.
2376 				 * Sent an ack to try to update the send
2377 				 * sequence number on the other side.
2378 				 */
2379 				if (tlen && th->th_seq != tp->rcv_nxt &&
2380 				    SEQ_LT(th->th_ack,
2381 				    tp->snd_una - tp->max_sndwnd))
2382 					goto dropafterack;
2383 			}
2384 			break;
2385 		}
2386 		/*
2387 		 * If the congestion window was inflated to account
2388 		 * for the other side's cached packets, retract it.
2389 		 */
2390 		/* XXX: make SACK have his own congestion control
2391 		 * struct -- rpaulo */
2392 		if (TCP_SACK_ENABLED(tp))
2393 			tcp_sack_newack(tp, th);
2394 		else
2395 			tp->t_congctl->fast_retransmit_newack(tp, th);
2396 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
2397 			TCP_STATINC(TCP_STAT_RCVACKTOOMUCH);
2398 			goto dropafterack;
2399 		}
2400 		acked = th->th_ack - tp->snd_una;
2401 		tcps = TCP_STAT_GETREF();
2402 		tcps[TCP_STAT_RCVACKPACK]++;
2403 		tcps[TCP_STAT_RCVACKBYTE] += acked;
2404 		TCP_STAT_PUTREF();
2405 
2406 		/*
2407 		 * If we have a timestamp reply, update smoothed
2408 		 * round trip time.  If no timestamp is present but
2409 		 * transmit timer is running and timed sequence
2410 		 * number was acked, update smoothed round trip time.
2411 		 * Since we now have an rtt measurement, cancel the
2412 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2413 		 * Recompute the initial retransmit timer.
2414 		 */
2415 		if (ts_rtt)
2416 			tcp_xmit_timer(tp, ts_rtt);
2417 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2418 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
2419 
2420 		/*
2421 		 * If all outstanding data is acked, stop retransmit
2422 		 * timer and remember to restart (more output or persist).
2423 		 * If there is more data to be acked, restart retransmit
2424 		 * timer, using current (possibly backed-off) value.
2425 		 */
2426 		if (th->th_ack == tp->snd_max) {
2427 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
2428 			needoutput = 1;
2429 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
2430 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2431 
2432 		/*
2433 		 * New data has been acked, adjust the congestion window.
2434 		 */
2435 		tp->t_congctl->newack(tp, th);
2436 
2437 		nd6_hint(tp);
2438 		if (acked > so->so_snd.sb_cc) {
2439 			tp->snd_wnd -= so->so_snd.sb_cc;
2440 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2441 			ourfinisacked = 1;
2442 		} else {
2443 			if (acked > (tp->t_lastoff - tp->t_inoff))
2444 				tp->t_lastm = NULL;
2445 			sbdrop(&so->so_snd, acked);
2446 			tp->t_lastoff -= acked;
2447 			tp->snd_wnd -= acked;
2448 			ourfinisacked = 0;
2449 		}
2450 		sowwakeup(so);
2451 
2452 		icmp_check(tp, th, acked);
2453 
2454 		tp->snd_una = th->th_ack;
2455 		if (SEQ_GT(tp->snd_una, tp->snd_fack))
2456 			tp->snd_fack = tp->snd_una;
2457 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2458 			tp->snd_nxt = tp->snd_una;
2459 		if (SEQ_LT(tp->snd_high, tp->snd_una))
2460 			tp->snd_high = tp->snd_una;
2461 
2462 		switch (tp->t_state) {
2463 
2464 		/*
2465 		 * In FIN_WAIT_1 STATE in addition to the processing
2466 		 * for the ESTABLISHED state if our FIN is now acknowledged
2467 		 * then enter FIN_WAIT_2.
2468 		 */
2469 		case TCPS_FIN_WAIT_1:
2470 			if (ourfinisacked) {
2471 				/*
2472 				 * If we can't receive any more
2473 				 * data, then closing user can proceed.
2474 				 * Starting the timer is contrary to the
2475 				 * specification, but if we don't get a FIN
2476 				 * we'll hang forever.
2477 				 */
2478 				if (so->so_state & SS_CANTRCVMORE) {
2479 					soisdisconnected(so);
2480 					if (tp->t_maxidle > 0)
2481 						TCP_TIMER_ARM(tp, TCPT_2MSL,
2482 						    tp->t_maxidle);
2483 				}
2484 				tp->t_state = TCPS_FIN_WAIT_2;
2485 			}
2486 			break;
2487 
2488 	 	/*
2489 		 * In CLOSING STATE in addition to the processing for
2490 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2491 		 * then enter the TIME-WAIT state, otherwise ignore
2492 		 * the segment.
2493 		 */
2494 		case TCPS_CLOSING:
2495 			if (ourfinisacked) {
2496 				tp->t_state = TCPS_TIME_WAIT;
2497 				tcp_canceltimers(tp);
2498 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2499 				soisdisconnected(so);
2500 			}
2501 			break;
2502 
2503 		/*
2504 		 * In LAST_ACK, we may still be waiting for data to drain
2505 		 * and/or to be acked, as well as for the ack of our FIN.
2506 		 * If our FIN is now acknowledged, delete the TCB,
2507 		 * enter the closed state and return.
2508 		 */
2509 		case TCPS_LAST_ACK:
2510 			if (ourfinisacked) {
2511 				tp = tcp_close(tp);
2512 				goto drop;
2513 			}
2514 			break;
2515 
2516 		/*
2517 		 * In TIME_WAIT state the only thing that should arrive
2518 		 * is a retransmission of the remote FIN.  Acknowledge
2519 		 * it and restart the finack timer.
2520 		 */
2521 		case TCPS_TIME_WAIT:
2522 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2523 			goto dropafterack;
2524 		}
2525 	}
2526 
2527 step6:
2528 	/*
2529 	 * Update window information.
2530 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2531 	 */
2532 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2533 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2534 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2535 		/* keep track of pure window updates */
2536 		if (tlen == 0 &&
2537 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2538 			TCP_STATINC(TCP_STAT_RCVWINUPD);
2539 		tp->snd_wnd = tiwin;
2540 		tp->snd_wl1 = th->th_seq;
2541 		tp->snd_wl2 = th->th_ack;
2542 		if (tp->snd_wnd > tp->max_sndwnd)
2543 			tp->max_sndwnd = tp->snd_wnd;
2544 		needoutput = 1;
2545 	}
2546 
2547 	/*
2548 	 * Process segments with URG.
2549 	 */
2550 	if ((tiflags & TH_URG) && th->th_urp &&
2551 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2552 		/*
2553 		 * This is a kludge, but if we receive and accept
2554 		 * random urgent pointers, we'll crash in
2555 		 * soreceive.  It's hard to imagine someone
2556 		 * actually wanting to send this much urgent data.
2557 		 */
2558 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2559 			th->th_urp = 0;			/* XXX */
2560 			tiflags &= ~TH_URG;		/* XXX */
2561 			goto dodata;			/* XXX */
2562 		}
2563 		/*
2564 		 * If this segment advances the known urgent pointer,
2565 		 * then mark the data stream.  This should not happen
2566 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2567 		 * a FIN has been received from the remote side.
2568 		 * In these states we ignore the URG.
2569 		 *
2570 		 * According to RFC961 (Assigned Protocols),
2571 		 * the urgent pointer points to the last octet
2572 		 * of urgent data.  We continue, however,
2573 		 * to consider it to indicate the first octet
2574 		 * of data past the urgent section as the original
2575 		 * spec states (in one of two places).
2576 		 */
2577 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2578 			tp->rcv_up = th->th_seq + th->th_urp;
2579 			so->so_oobmark = so->so_rcv.sb_cc +
2580 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2581 			if (so->so_oobmark == 0)
2582 				so->so_state |= SS_RCVATMARK;
2583 			sohasoutofband(so);
2584 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2585 		}
2586 		/*
2587 		 * Remove out of band data so doesn't get presented to user.
2588 		 * This can happen independent of advancing the URG pointer,
2589 		 * but if two URG's are pending at once, some out-of-band
2590 		 * data may creep in... ick.
2591 		 */
2592 		if (th->th_urp <= (u_int16_t) tlen
2593 #ifdef SO_OOBINLINE
2594 		     && (so->so_options & SO_OOBINLINE) == 0
2595 #endif
2596 		     )
2597 			tcp_pulloutofband(so, th, m, hdroptlen);
2598 	} else
2599 		/*
2600 		 * If no out of band data is expected,
2601 		 * pull receive urgent pointer along
2602 		 * with the receive window.
2603 		 */
2604 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2605 			tp->rcv_up = tp->rcv_nxt;
2606 dodata:							/* XXX */
2607 
2608 	/*
2609 	 * Process the segment text, merging it into the TCP sequencing queue,
2610 	 * and arranging for acknowledgement of receipt if necessary.
2611 	 * This process logically involves adjusting tp->rcv_wnd as data
2612 	 * is presented to the user (this happens in tcp_usrreq.c,
2613 	 * case PRU_RCVD).  If a FIN has already been received on this
2614 	 * connection then we just ignore the text.
2615 	 */
2616 	if ((tlen || (tiflags & TH_FIN)) &&
2617 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2618 		/*
2619 		 * Insert segment ti into reassembly queue of tcp with
2620 		 * control block tp.  Return TH_FIN if reassembly now includes
2621 		 * a segment with FIN.  The macro form does the common case
2622 		 * inline (segment is the next to be received on an
2623 		 * established connection, and the queue is empty),
2624 		 * avoiding linkage into and removal from the queue and
2625 		 * repetition of various conversions.
2626 		 * Set DELACK for segments received in order, but ack
2627 		 * immediately when segments are out of order
2628 		 * (so fast retransmit can work).
2629 		 */
2630 		/* NOTE: this was TCP_REASS() macro, but used only once */
2631 		TCP_REASS_LOCK(tp);
2632 		if (th->th_seq == tp->rcv_nxt &&
2633 		    TAILQ_FIRST(&tp->segq) == NULL &&
2634 		    tp->t_state == TCPS_ESTABLISHED) {
2635 			tcp_setup_ack(tp, th);
2636 			tp->rcv_nxt += tlen;
2637 			tiflags = th->th_flags & TH_FIN;
2638 			tcps = TCP_STAT_GETREF();
2639 			tcps[TCP_STAT_RCVPACK]++;
2640 			tcps[TCP_STAT_RCVBYTE] += tlen;
2641 			TCP_STAT_PUTREF();
2642 			nd6_hint(tp);
2643 			if (so->so_state & SS_CANTRCVMORE)
2644 				m_freem(m);
2645 			else {
2646 				m_adj(m, hdroptlen);
2647 				sbappendstream(&(so)->so_rcv, m);
2648 			}
2649 			TCP_REASS_UNLOCK(tp);
2650 			sorwakeup(so);
2651 		} else {
2652 			m_adj(m, hdroptlen);
2653 			tiflags = tcp_reass(tp, th, m, &tlen);
2654 			tp->t_flags |= TF_ACKNOW;
2655 			TCP_REASS_UNLOCK(tp);
2656 		}
2657 
2658 		/*
2659 		 * Note the amount of data that peer has sent into
2660 		 * our window, in order to estimate the sender's
2661 		 * buffer size.
2662 		 */
2663 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2664 	} else {
2665 		m_freem(m);
2666 		m = NULL;
2667 		tiflags &= ~TH_FIN;
2668 	}
2669 
2670 	/*
2671 	 * If FIN is received ACK the FIN and let the user know
2672 	 * that the connection is closing.  Ignore a FIN received before
2673 	 * the connection is fully established.
2674 	 */
2675 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2676 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2677 			socantrcvmore(so);
2678 			tp->t_flags |= TF_ACKNOW;
2679 			tp->rcv_nxt++;
2680 		}
2681 		switch (tp->t_state) {
2682 
2683 	 	/*
2684 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2685 		 */
2686 		case TCPS_ESTABLISHED:
2687 			tp->t_state = TCPS_CLOSE_WAIT;
2688 			break;
2689 
2690 	 	/*
2691 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2692 		 * enter the CLOSING state.
2693 		 */
2694 		case TCPS_FIN_WAIT_1:
2695 			tp->t_state = TCPS_CLOSING;
2696 			break;
2697 
2698 	 	/*
2699 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2700 		 * starting the time-wait timer, turning off the other
2701 		 * standard timers.
2702 		 */
2703 		case TCPS_FIN_WAIT_2:
2704 			tp->t_state = TCPS_TIME_WAIT;
2705 			tcp_canceltimers(tp);
2706 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2707 			soisdisconnected(so);
2708 			break;
2709 
2710 		/*
2711 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2712 		 */
2713 		case TCPS_TIME_WAIT:
2714 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2715 			break;
2716 		}
2717 	}
2718 #ifdef TCP_DEBUG
2719 	if (so->so_options & SO_DEBUG)
2720 		tcp_trace(TA_INPUT, ostate, tp, tcp_saveti, 0);
2721 #endif
2722 
2723 	/*
2724 	 * Return any desired output.
2725 	 */
2726 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2727 		(void) tcp_output(tp);
2728 	}
2729 	if (tcp_saveti)
2730 		m_freem(tcp_saveti);
2731 	return;
2732 
2733 badsyn:
2734 	/*
2735 	 * Received a bad SYN.  Increment counters and dropwithreset.
2736 	 */
2737 	TCP_STATINC(TCP_STAT_BADSYN);
2738 	tp = NULL;
2739 	goto dropwithreset;
2740 
2741 dropafterack:
2742 	/*
2743 	 * Generate an ACK dropping incoming segment if it occupies
2744 	 * sequence space, where the ACK reflects our state.
2745 	 */
2746 	if (tiflags & TH_RST)
2747 		goto drop;
2748 	goto dropafterack2;
2749 
2750 dropafterack_ratelim:
2751 	/*
2752 	 * We may want to rate-limit ACKs against SYN/RST attack.
2753 	 */
2754 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2755 	    tcp_ackdrop_ppslim) == 0) {
2756 		/* XXX stat */
2757 		goto drop;
2758 	}
2759 	/* ...fall into dropafterack2... */
2760 
2761 dropafterack2:
2762 	m_freem(m);
2763 	tp->t_flags |= TF_ACKNOW;
2764 	(void) tcp_output(tp);
2765 	if (tcp_saveti)
2766 		m_freem(tcp_saveti);
2767 	return;
2768 
2769 dropwithreset_ratelim:
2770 	/*
2771 	 * We may want to rate-limit RSTs in certain situations,
2772 	 * particularly if we are sending an RST in response to
2773 	 * an attempt to connect to or otherwise communicate with
2774 	 * a port for which we have no socket.
2775 	 */
2776 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2777 	    tcp_rst_ppslim) == 0) {
2778 		/* XXX stat */
2779 		goto drop;
2780 	}
2781 	/* ...fall into dropwithreset... */
2782 
2783 dropwithreset:
2784 	/*
2785 	 * Generate a RST, dropping incoming segment.
2786 	 * Make ACK acceptable to originator of segment.
2787 	 */
2788 	if (tiflags & TH_RST)
2789 		goto drop;
2790 
2791 	switch (af) {
2792 #ifdef INET6
2793 	case AF_INET6:
2794 		/* For following calls to tcp_respond */
2795 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst))
2796 			goto drop;
2797 		break;
2798 #endif /* INET6 */
2799 	case AF_INET:
2800 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2801 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2802 			goto drop;
2803 	}
2804 
2805 	if (tiflags & TH_ACK)
2806 		(void)tcp_respond(tp, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
2807 	else {
2808 		if (tiflags & TH_SYN)
2809 			tlen++;
2810 		(void)tcp_respond(tp, m, m, th, th->th_seq + tlen, (tcp_seq)0,
2811 		    TH_RST|TH_ACK);
2812 	}
2813 	if (tcp_saveti)
2814 		m_freem(tcp_saveti);
2815 	return;
2816 
2817 badcsum:
2818 drop:
2819 	/*
2820 	 * Drop space held by incoming segment and return.
2821 	 */
2822 	if (tp) {
2823 		if (tp->t_inpcb)
2824 			so = tp->t_inpcb->inp_socket;
2825 #ifdef INET6
2826 		else if (tp->t_in6pcb)
2827 			so = tp->t_in6pcb->in6p_socket;
2828 #endif
2829 		else
2830 			so = NULL;
2831 #ifdef TCP_DEBUG
2832 		if (so && (so->so_options & SO_DEBUG) != 0)
2833 			tcp_trace(TA_DROP, ostate, tp, tcp_saveti, 0);
2834 #endif
2835 	}
2836 	if (tcp_saveti)
2837 		m_freem(tcp_saveti);
2838 	m_freem(m);
2839 	return;
2840 }
2841 
2842 #ifdef TCP_SIGNATURE
2843 int
2844 tcp_signature_apply(void *fstate, void *data, u_int len)
2845 {
2846 
2847 	MD5Update(fstate, (u_char *)data, len);
2848 	return (0);
2849 }
2850 
2851 struct secasvar *
2852 tcp_signature_getsav(struct mbuf *m, struct tcphdr *th)
2853 {
2854 	struct secasvar *sav;
2855 #ifdef FAST_IPSEC
2856 	union sockaddr_union dst;
2857 #endif
2858 	struct ip *ip;
2859 	struct ip6_hdr *ip6;
2860 
2861 	ip = mtod(m, struct ip *);
2862 	switch (ip->ip_v) {
2863 	case 4:
2864 		ip = mtod(m, struct ip *);
2865 		ip6 = NULL;
2866 		break;
2867 	case 6:
2868 		ip = NULL;
2869 		ip6 = mtod(m, struct ip6_hdr *);
2870 		break;
2871 	default:
2872 		return (NULL);
2873 	}
2874 
2875 #ifdef FAST_IPSEC
2876 	/* Extract the destination from the IP header in the mbuf. */
2877 	memset(&dst, 0, sizeof(union sockaddr_union));
2878 	if (ip !=NULL) {
2879 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2880 		dst.sa.sa_family = AF_INET;
2881 		dst.sin.sin_addr = ip->ip_dst;
2882 	} else {
2883 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2884 		dst.sa.sa_family = AF_INET6;
2885 		dst.sin6.sin6_addr = ip6->ip6_dst;
2886 	}
2887 
2888 	/*
2889 	 * Look up an SADB entry which matches the address of the peer.
2890 	 */
2891 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2892 #else
2893 	if (ip)
2894 		sav = key_allocsa(AF_INET, (void *)&ip->ip_src,
2895 		    (void *)&ip->ip_dst, IPPROTO_TCP,
2896 		    htonl(TCP_SIG_SPI), 0, 0);
2897 	else
2898 		sav = key_allocsa(AF_INET6, (void *)&ip6->ip6_src,
2899 		    (void *)&ip6->ip6_dst, IPPROTO_TCP,
2900 		    htonl(TCP_SIG_SPI), 0, 0);
2901 #endif
2902 
2903 	return (sav);	/* freesav must be performed by caller */
2904 }
2905 
2906 int
2907 tcp_signature(struct mbuf *m, struct tcphdr *th, int thoff,
2908     struct secasvar *sav, char *sig)
2909 {
2910 	MD5_CTX ctx;
2911 	struct ip *ip;
2912 	struct ipovly *ipovly;
2913 	struct ip6_hdr *ip6;
2914 	struct ippseudo ippseudo;
2915 	struct ip6_hdr_pseudo ip6pseudo;
2916 	struct tcphdr th0;
2917 	int l, tcphdrlen;
2918 
2919 	if (sav == NULL)
2920 		return (-1);
2921 
2922 	tcphdrlen = th->th_off * 4;
2923 
2924 	switch (mtod(m, struct ip *)->ip_v) {
2925 	case 4:
2926 		ip = mtod(m, struct ip *);
2927 		ip6 = NULL;
2928 		break;
2929 	case 6:
2930 		ip = NULL;
2931 		ip6 = mtod(m, struct ip6_hdr *);
2932 		break;
2933 	default:
2934 		return (-1);
2935 	}
2936 
2937 	MD5Init(&ctx);
2938 
2939 	if (ip) {
2940 		memset(&ippseudo, 0, sizeof(ippseudo));
2941 		ipovly = (struct ipovly *)ip;
2942 		ippseudo.ippseudo_src = ipovly->ih_src;
2943 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2944 		ippseudo.ippseudo_pad = 0;
2945 		ippseudo.ippseudo_p = IPPROTO_TCP;
2946 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - thoff);
2947 		MD5Update(&ctx, (char *)&ippseudo, sizeof(ippseudo));
2948 	} else {
2949 		memset(&ip6pseudo, 0, sizeof(ip6pseudo));
2950 		ip6pseudo.ip6ph_src = ip6->ip6_src;
2951 		in6_clearscope(&ip6pseudo.ip6ph_src);
2952 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
2953 		in6_clearscope(&ip6pseudo.ip6ph_dst);
2954 		ip6pseudo.ip6ph_len = htons(m->m_pkthdr.len - thoff);
2955 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
2956 		MD5Update(&ctx, (char *)&ip6pseudo, sizeof(ip6pseudo));
2957 	}
2958 
2959 	th0 = *th;
2960 	th0.th_sum = 0;
2961 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
2962 
2963 	l = m->m_pkthdr.len - thoff - tcphdrlen;
2964 	if (l > 0)
2965 		m_apply(m, thoff + tcphdrlen,
2966 		    m->m_pkthdr.len - thoff - tcphdrlen,
2967 		    tcp_signature_apply, &ctx);
2968 
2969 	MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2970 	MD5Final(sig, &ctx);
2971 
2972 	return (0);
2973 }
2974 #endif
2975 
2976 static int
2977 tcp_dooptions(struct tcpcb *tp, const u_char *cp, int cnt,
2978     struct tcphdr *th,
2979     struct mbuf *m, int toff, struct tcp_opt_info *oi)
2980 {
2981 	u_int16_t mss;
2982 	int opt, optlen = 0;
2983 #ifdef TCP_SIGNATURE
2984 	void *sigp = NULL;
2985 	char sigbuf[TCP_SIGLEN];
2986 	struct secasvar *sav = NULL;
2987 #endif
2988 
2989 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2990 		opt = cp[0];
2991 		if (opt == TCPOPT_EOL)
2992 			break;
2993 		if (opt == TCPOPT_NOP)
2994 			optlen = 1;
2995 		else {
2996 			if (cnt < 2)
2997 				break;
2998 			optlen = cp[1];
2999 			if (optlen < 2 || optlen > cnt)
3000 				break;
3001 		}
3002 		switch (opt) {
3003 
3004 		default:
3005 			continue;
3006 
3007 		case TCPOPT_MAXSEG:
3008 			if (optlen != TCPOLEN_MAXSEG)
3009 				continue;
3010 			if (!(th->th_flags & TH_SYN))
3011 				continue;
3012 			if (TCPS_HAVERCVDSYN(tp->t_state))
3013 				continue;
3014 			bcopy(cp + 2, &mss, sizeof(mss));
3015 			oi->maxseg = ntohs(mss);
3016 			break;
3017 
3018 		case TCPOPT_WINDOW:
3019 			if (optlen != TCPOLEN_WINDOW)
3020 				continue;
3021 			if (!(th->th_flags & TH_SYN))
3022 				continue;
3023 			if (TCPS_HAVERCVDSYN(tp->t_state))
3024 				continue;
3025 			tp->t_flags |= TF_RCVD_SCALE;
3026 			tp->requested_s_scale = cp[2];
3027 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
3028 #if 0	/*XXX*/
3029 				char *p;
3030 
3031 				if (ip)
3032 					p = ntohl(ip->ip_src);
3033 #ifdef INET6
3034 				else if (ip6)
3035 					p = ip6_sprintf(&ip6->ip6_src);
3036 #endif
3037 				else
3038 					p = "(unknown)";
3039 				log(LOG_ERR, "TCP: invalid wscale %d from %s, "
3040 				    "assuming %d\n",
3041 				    tp->requested_s_scale, p,
3042 				    TCP_MAX_WINSHIFT);
3043 #else
3044 				log(LOG_ERR, "TCP: invalid wscale %d, "
3045 				    "assuming %d\n",
3046 				    tp->requested_s_scale,
3047 				    TCP_MAX_WINSHIFT);
3048 #endif
3049 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
3050 			}
3051 			break;
3052 
3053 		case TCPOPT_TIMESTAMP:
3054 			if (optlen != TCPOLEN_TIMESTAMP)
3055 				continue;
3056 			oi->ts_present = 1;
3057 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
3058 			NTOHL(oi->ts_val);
3059 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
3060 			NTOHL(oi->ts_ecr);
3061 
3062 			if (!(th->th_flags & TH_SYN))
3063 				continue;
3064 			if (TCPS_HAVERCVDSYN(tp->t_state))
3065 				continue;
3066 			/*
3067 			 * A timestamp received in a SYN makes
3068 			 * it ok to send timestamp requests and replies.
3069 			 */
3070 			tp->t_flags |= TF_RCVD_TSTMP;
3071 			tp->ts_recent = oi->ts_val;
3072 			tp->ts_recent_age = tcp_now;
3073                         break;
3074 
3075 		case TCPOPT_SACK_PERMITTED:
3076 			if (optlen != TCPOLEN_SACK_PERMITTED)
3077 				continue;
3078 			if (!(th->th_flags & TH_SYN))
3079 				continue;
3080 			if (TCPS_HAVERCVDSYN(tp->t_state))
3081 				continue;
3082 			if (tcp_do_sack) {
3083 				tp->t_flags |= TF_SACK_PERMIT;
3084 				tp->t_flags |= TF_WILL_SACK;
3085 			}
3086 			break;
3087 
3088 		case TCPOPT_SACK:
3089 			tcp_sack_option(tp, th, cp, optlen);
3090 			break;
3091 #ifdef TCP_SIGNATURE
3092 		case TCPOPT_SIGNATURE:
3093 			if (optlen != TCPOLEN_SIGNATURE)
3094 				continue;
3095 			if (sigp && memcmp(sigp, cp + 2, TCP_SIGLEN))
3096 				return (-1);
3097 
3098 			sigp = sigbuf;
3099 			memcpy(sigbuf, cp + 2, TCP_SIGLEN);
3100 			tp->t_flags |= TF_SIGNATURE;
3101 			break;
3102 #endif
3103 		}
3104 	}
3105 
3106 #ifdef TCP_SIGNATURE
3107 	if (tp->t_flags & TF_SIGNATURE) {
3108 
3109 		sav = tcp_signature_getsav(m, th);
3110 
3111 		if (sav == NULL && tp->t_state == TCPS_LISTEN)
3112 			return (-1);
3113 	}
3114 
3115 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
3116 		if (sav == NULL)
3117 			return (-1);
3118 #ifdef FAST_IPSEC
3119 		KEY_FREESAV(&sav);
3120 #else
3121 		key_freesav(sav);
3122 #endif
3123 		return (-1);
3124 	}
3125 
3126 	if (sigp) {
3127 		char sig[TCP_SIGLEN];
3128 
3129 		tcp_fields_to_net(th);
3130 		if (tcp_signature(m, th, toff, sav, sig) < 0) {
3131 			tcp_fields_to_host(th);
3132 			if (sav == NULL)
3133 				return (-1);
3134 #ifdef FAST_IPSEC
3135 			KEY_FREESAV(&sav);
3136 #else
3137 			key_freesav(sav);
3138 #endif
3139 			return (-1);
3140 		}
3141 		tcp_fields_to_host(th);
3142 
3143 		if (memcmp(sig, sigp, TCP_SIGLEN)) {
3144 			TCP_STATINC(TCP_STAT_BADSIG);
3145 			if (sav == NULL)
3146 				return (-1);
3147 #ifdef FAST_IPSEC
3148 			KEY_FREESAV(&sav);
3149 #else
3150 			key_freesav(sav);
3151 #endif
3152 			return (-1);
3153 		} else
3154 			TCP_STATINC(TCP_STAT_GOODSIG);
3155 
3156 		key_sa_recordxfer(sav, m);
3157 #ifdef FAST_IPSEC
3158 		KEY_FREESAV(&sav);
3159 #else
3160 		key_freesav(sav);
3161 #endif
3162 	}
3163 #endif
3164 
3165 	return (0);
3166 }
3167 
3168 /*
3169  * Pull out of band byte out of a segment so
3170  * it doesn't appear in the user's data queue.
3171  * It is still reflected in the segment length for
3172  * sequencing purposes.
3173  */
3174 void
3175 tcp_pulloutofband(struct socket *so, struct tcphdr *th,
3176     struct mbuf *m, int off)
3177 {
3178 	int cnt = off + th->th_urp - 1;
3179 
3180 	while (cnt >= 0) {
3181 		if (m->m_len > cnt) {
3182 			char *cp = mtod(m, char *) + cnt;
3183 			struct tcpcb *tp = sototcpcb(so);
3184 
3185 			tp->t_iobc = *cp;
3186 			tp->t_oobflags |= TCPOOB_HAVEDATA;
3187 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3188 			m->m_len--;
3189 			return;
3190 		}
3191 		cnt -= m->m_len;
3192 		m = m->m_next;
3193 		if (m == 0)
3194 			break;
3195 	}
3196 	panic("tcp_pulloutofband");
3197 }
3198 
3199 /*
3200  * Collect new round-trip time estimate
3201  * and update averages and current timeout.
3202  */
3203 void
3204 tcp_xmit_timer(struct tcpcb *tp, uint32_t rtt)
3205 {
3206 	int32_t delta;
3207 
3208 	TCP_STATINC(TCP_STAT_RTTUPDATED);
3209 	if (tp->t_srtt != 0) {
3210 		/*
3211 		 * srtt is stored as fixed point with 3 bits after the
3212 		 * binary point (i.e., scaled by 8).  The following magic
3213 		 * is equivalent to the smoothing algorithm in rfc793 with
3214 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3215 		 * point).  Adjust rtt to origin 0.
3216 		 */
3217 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
3218 		if ((tp->t_srtt += delta) <= 0)
3219 			tp->t_srtt = 1 << 2;
3220 		/*
3221 		 * We accumulate a smoothed rtt variance (actually, a
3222 		 * smoothed mean difference), then set the retransmit
3223 		 * timer to smoothed rtt + 4 times the smoothed variance.
3224 		 * rttvar is stored as fixed point with 2 bits after the
3225 		 * binary point (scaled by 4).  The following is
3226 		 * equivalent to rfc793 smoothing with an alpha of .75
3227 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
3228 		 * rfc793's wired-in beta.
3229 		 */
3230 		if (delta < 0)
3231 			delta = -delta;
3232 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
3233 		if ((tp->t_rttvar += delta) <= 0)
3234 			tp->t_rttvar = 1 << 2;
3235 	} else {
3236 		/*
3237 		 * No rtt measurement yet - use the unsmoothed rtt.
3238 		 * Set the variance to half the rtt (so our first
3239 		 * retransmit happens at 3*rtt).
3240 		 */
3241 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
3242 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
3243 	}
3244 	tp->t_rtttime = 0;
3245 	tp->t_rxtshift = 0;
3246 
3247 	/*
3248 	 * the retransmit should happen at rtt + 4 * rttvar.
3249 	 * Because of the way we do the smoothing, srtt and rttvar
3250 	 * will each average +1/2 tick of bias.  When we compute
3251 	 * the retransmit timer, we want 1/2 tick of rounding and
3252 	 * 1 extra tick because of +-1/2 tick uncertainty in the
3253 	 * firing of the timer.  The bias will give us exactly the
3254 	 * 1.5 tick we need.  But, because the bias is
3255 	 * statistical, we have to test that we don't drop below
3256 	 * the minimum feasible timer (which is 2 ticks).
3257 	 */
3258 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3259 	    max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3260 
3261 	/*
3262 	 * We received an ack for a packet that wasn't retransmitted;
3263 	 * it is probably safe to discard any error indications we've
3264 	 * received recently.  This isn't quite right, but close enough
3265 	 * for now (a route might have failed after we sent a segment,
3266 	 * and the return path might not be symmetrical).
3267 	 */
3268 	tp->t_softerror = 0;
3269 }
3270 
3271 
3272 /*
3273  * TCP compressed state engine.  Currently used to hold compressed
3274  * state for SYN_RECEIVED.
3275  */
3276 
3277 u_long	syn_cache_count;
3278 u_int32_t syn_hash1, syn_hash2;
3279 
3280 #define SYN_HASH(sa, sp, dp) \
3281 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3282 				     ((u_int32_t)(sp)))^syn_hash2)))
3283 #ifndef INET6
3284 #define	SYN_HASHALL(hash, src, dst) \
3285 do {									\
3286 	hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr,	\
3287 		((const struct sockaddr_in *)(src))->sin_port,		\
3288 		((const struct sockaddr_in *)(dst))->sin_port);		\
3289 } while (/*CONSTCOND*/ 0)
3290 #else
3291 #define SYN_HASH6(sa, sp, dp) \
3292 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3293 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3294 	 & 0x7fffffff)
3295 
3296 #define SYN_HASHALL(hash, src, dst) \
3297 do {									\
3298 	switch ((src)->sa_family) {					\
3299 	case AF_INET:							\
3300 		hash = SYN_HASH(&((const struct sockaddr_in *)(src))->sin_addr, \
3301 			((const struct sockaddr_in *)(src))->sin_port,	\
3302 			((const struct sockaddr_in *)(dst))->sin_port);	\
3303 		break;							\
3304 	case AF_INET6:							\
3305 		hash = SYN_HASH6(&((const struct sockaddr_in6 *)(src))->sin6_addr, \
3306 			((const struct sockaddr_in6 *)(src))->sin6_port,	\
3307 			((const struct sockaddr_in6 *)(dst))->sin6_port);	\
3308 		break;							\
3309 	default:							\
3310 		hash = 0;						\
3311 	}								\
3312 } while (/*CONSTCOND*/0)
3313 #endif /* INET6 */
3314 
3315 static struct pool syn_cache_pool;
3316 
3317 /*
3318  * We don't estimate RTT with SYNs, so each packet starts with the default
3319  * RTT and each timer step has a fixed timeout value.
3320  */
3321 #define	SYN_CACHE_TIMER_ARM(sc)						\
3322 do {									\
3323 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3324 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3325 	    TCPTV_REXMTMAX);						\
3326 	callout_reset(&(sc)->sc_timer,					\
3327 	    (sc)->sc_rxtcur * (hz / PR_SLOWHZ), syn_cache_timer, (sc));	\
3328 } while (/*CONSTCOND*/0)
3329 
3330 #define	SYN_CACHE_TIMESTAMP(sc)	(tcp_now - (sc)->sc_timebase)
3331 
3332 static inline void
3333 syn_cache_rm(struct syn_cache *sc)
3334 {
3335 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3336 	    sc, sc_bucketq);
3337 	sc->sc_tp = NULL;
3338 	LIST_REMOVE(sc, sc_tpq);
3339 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3340 	callout_stop(&sc->sc_timer);
3341 	syn_cache_count--;
3342 }
3343 
3344 static inline void
3345 syn_cache_put(struct syn_cache *sc)
3346 {
3347 	if (sc->sc_ipopts)
3348 		(void) m_free(sc->sc_ipopts);
3349 	rtcache_free(&sc->sc_route);
3350 	if (callout_invoking(&sc->sc_timer))
3351 		sc->sc_flags |= SCF_DEAD;
3352 	else {
3353 		callout_destroy(&sc->sc_timer);
3354 		pool_put(&syn_cache_pool, sc);
3355 	}
3356 }
3357 
3358 void
3359 syn_cache_init(void)
3360 {
3361 	int i;
3362 
3363 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3364 	    "synpl", NULL, IPL_SOFTNET);
3365 
3366 	/* Initialize the hash buckets. */
3367 	for (i = 0; i < tcp_syn_cache_size; i++)
3368 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3369 }
3370 
3371 void
3372 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3373 {
3374 	struct syn_cache_head *scp;
3375 	struct syn_cache *sc2;
3376 	int s;
3377 
3378 	/*
3379 	 * If there are no entries in the hash table, reinitialize
3380 	 * the hash secrets.
3381 	 */
3382 	if (syn_cache_count == 0) {
3383 		syn_hash1 = arc4random();
3384 		syn_hash2 = arc4random();
3385 	}
3386 
3387 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3388 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3389 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3390 
3391 	/*
3392 	 * Make sure that we don't overflow the per-bucket
3393 	 * limit or the total cache size limit.
3394 	 */
3395 	s = splsoftnet();
3396 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3397 		TCP_STATINC(TCP_STAT_SC_BUCKETOVERFLOW);
3398 		/*
3399 		 * The bucket is full.  Toss the oldest element in the
3400 		 * bucket.  This will be the first entry in the bucket.
3401 		 */
3402 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3403 #ifdef DIAGNOSTIC
3404 		/*
3405 		 * This should never happen; we should always find an
3406 		 * entry in our bucket.
3407 		 */
3408 		if (sc2 == NULL)
3409 			panic("syn_cache_insert: bucketoverflow: impossible");
3410 #endif
3411 		syn_cache_rm(sc2);
3412 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3413 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3414 		struct syn_cache_head *scp2, *sce;
3415 
3416 		TCP_STATINC(TCP_STAT_SC_OVERFLOWED);
3417 		/*
3418 		 * The cache is full.  Toss the oldest entry in the
3419 		 * first non-empty bucket we can find.
3420 		 *
3421 		 * XXX We would really like to toss the oldest
3422 		 * entry in the cache, but we hope that this
3423 		 * condition doesn't happen very often.
3424 		 */
3425 		scp2 = scp;
3426 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3427 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3428 			for (++scp2; scp2 != scp; scp2++) {
3429 				if (scp2 >= sce)
3430 					scp2 = &tcp_syn_cache[0];
3431 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3432 					break;
3433 			}
3434 #ifdef DIAGNOSTIC
3435 			/*
3436 			 * This should never happen; we should always find a
3437 			 * non-empty bucket.
3438 			 */
3439 			if (scp2 == scp)
3440 				panic("syn_cache_insert: cacheoverflow: "
3441 				    "impossible");
3442 #endif
3443 		}
3444 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3445 		syn_cache_rm(sc2);
3446 		syn_cache_put(sc2);	/* calls pool_put but see spl above */
3447 	}
3448 
3449 	/*
3450 	 * Initialize the entry's timer.
3451 	 */
3452 	sc->sc_rxttot = 0;
3453 	sc->sc_rxtshift = 0;
3454 	SYN_CACHE_TIMER_ARM(sc);
3455 
3456 	/* Link it from tcpcb entry */
3457 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3458 
3459 	/* Put it into the bucket. */
3460 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3461 	scp->sch_length++;
3462 	syn_cache_count++;
3463 
3464 	TCP_STATINC(TCP_STAT_SC_ADDED);
3465 	splx(s);
3466 }
3467 
3468 /*
3469  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3470  * If we have retransmitted an entry the maximum number of times, expire
3471  * that entry.
3472  */
3473 void
3474 syn_cache_timer(void *arg)
3475 {
3476 	struct syn_cache *sc = arg;
3477 
3478 	mutex_enter(softnet_lock);
3479 	KERNEL_LOCK(1, NULL);
3480 	callout_ack(&sc->sc_timer);
3481 
3482 	if (__predict_false(sc->sc_flags & SCF_DEAD)) {
3483 		TCP_STATINC(TCP_STAT_SC_DELAYED_FREE);
3484 		callout_destroy(&sc->sc_timer);
3485 		pool_put(&syn_cache_pool, sc);
3486 		KERNEL_UNLOCK_ONE(NULL);
3487 		mutex_exit(softnet_lock);
3488 		return;
3489 	}
3490 
3491 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3492 		/* Drop it -- too many retransmissions. */
3493 		goto dropit;
3494 	}
3495 
3496 	/*
3497 	 * Compute the total amount of time this entry has
3498 	 * been on a queue.  If this entry has been on longer
3499 	 * than the keep alive timer would allow, expire it.
3500 	 */
3501 	sc->sc_rxttot += sc->sc_rxtcur;
3502 	if (sc->sc_rxttot >= tcp_keepinit)
3503 		goto dropit;
3504 
3505 	TCP_STATINC(TCP_STAT_SC_RETRANSMITTED);
3506 	(void) syn_cache_respond(sc, NULL);
3507 
3508 	/* Advance the timer back-off. */
3509 	sc->sc_rxtshift++;
3510 	SYN_CACHE_TIMER_ARM(sc);
3511 
3512 	KERNEL_UNLOCK_ONE(NULL);
3513 	mutex_exit(softnet_lock);
3514 	return;
3515 
3516  dropit:
3517 	TCP_STATINC(TCP_STAT_SC_TIMED_OUT);
3518 	syn_cache_rm(sc);
3519 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3520 	KERNEL_UNLOCK_ONE(NULL);
3521 	mutex_exit(softnet_lock);
3522 }
3523 
3524 /*
3525  * Remove syn cache created by the specified tcb entry,
3526  * because this does not make sense to keep them
3527  * (if there's no tcb entry, syn cache entry will never be used)
3528  */
3529 void
3530 syn_cache_cleanup(struct tcpcb *tp)
3531 {
3532 	struct syn_cache *sc, *nsc;
3533 	int s;
3534 
3535 	s = splsoftnet();
3536 
3537 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3538 		nsc = LIST_NEXT(sc, sc_tpq);
3539 
3540 #ifdef DIAGNOSTIC
3541 		if (sc->sc_tp != tp)
3542 			panic("invalid sc_tp in syn_cache_cleanup");
3543 #endif
3544 		syn_cache_rm(sc);
3545 		syn_cache_put(sc);	/* calls pool_put but see spl above */
3546 	}
3547 	/* just for safety */
3548 	LIST_INIT(&tp->t_sc);
3549 
3550 	splx(s);
3551 }
3552 
3553 /*
3554  * Find an entry in the syn cache.
3555  */
3556 struct syn_cache *
3557 syn_cache_lookup(const struct sockaddr *src, const struct sockaddr *dst,
3558     struct syn_cache_head **headp)
3559 {
3560 	struct syn_cache *sc;
3561 	struct syn_cache_head *scp;
3562 	u_int32_t hash;
3563 	int s;
3564 
3565 	SYN_HASHALL(hash, src, dst);
3566 
3567 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3568 	*headp = scp;
3569 	s = splsoftnet();
3570 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3571 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3572 		if (sc->sc_hash != hash)
3573 			continue;
3574 		if (!memcmp(&sc->sc_src, src, src->sa_len) &&
3575 		    !memcmp(&sc->sc_dst, dst, dst->sa_len)) {
3576 			splx(s);
3577 			return (sc);
3578 		}
3579 	}
3580 	splx(s);
3581 	return (NULL);
3582 }
3583 
3584 /*
3585  * This function gets called when we receive an ACK for a
3586  * socket in the LISTEN state.  We look up the connection
3587  * in the syn cache, and if its there, we pull it out of
3588  * the cache and turn it into a full-blown connection in
3589  * the SYN-RECEIVED state.
3590  *
3591  * The return values may not be immediately obvious, and their effects
3592  * can be subtle, so here they are:
3593  *
3594  *	NULL	SYN was not found in cache; caller should drop the
3595  *		packet and send an RST.
3596  *
3597  *	-1	We were unable to create the new connection, and are
3598  *		aborting it.  An ACK,RST is being sent to the peer
3599  *		(unless we got screwey sequence numbners; see below),
3600  *		because the 3-way handshake has been completed.  Caller
3601  *		should not free the mbuf, since we may be using it.  If
3602  *		we are not, we will free it.
3603  *
3604  *	Otherwise, the return value is a pointer to the new socket
3605  *	associated with the connection.
3606  */
3607 struct socket *
3608 syn_cache_get(struct sockaddr *src, struct sockaddr *dst,
3609     struct tcphdr *th, unsigned int hlen, unsigned int tlen,
3610     struct socket *so, struct mbuf *m)
3611 {
3612 	struct syn_cache *sc;
3613 	struct syn_cache_head *scp;
3614 	struct inpcb *inp = NULL;
3615 #ifdef INET6
3616 	struct in6pcb *in6p = NULL;
3617 #endif
3618 	struct tcpcb *tp = 0;
3619 	struct mbuf *am;
3620 	int s;
3621 	struct socket *oso;
3622 
3623 	s = splsoftnet();
3624 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3625 		splx(s);
3626 		return (NULL);
3627 	}
3628 
3629 	/*
3630 	 * Verify the sequence and ack numbers.  Try getting the correct
3631 	 * response again.
3632 	 */
3633 	if ((th->th_ack != sc->sc_iss + 1) ||
3634 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3635 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3636 		(void) syn_cache_respond(sc, m);
3637 		splx(s);
3638 		return ((struct socket *)(-1));
3639 	}
3640 
3641 	/* Remove this cache entry */
3642 	syn_cache_rm(sc);
3643 	splx(s);
3644 
3645 	/*
3646 	 * Ok, create the full blown connection, and set things up
3647 	 * as they would have been set up if we had created the
3648 	 * connection when the SYN arrived.  If we can't create
3649 	 * the connection, abort it.
3650 	 */
3651 	/*
3652 	 * inp still has the OLD in_pcb stuff, set the
3653 	 * v6-related flags on the new guy, too.   This is
3654 	 * done particularly for the case where an AF_INET6
3655 	 * socket is bound only to a port, and a v4 connection
3656 	 * comes in on that port.
3657 	 * we also copy the flowinfo from the original pcb
3658 	 * to the new one.
3659 	 */
3660 	oso = so;
3661 	so = sonewconn(so, SS_ISCONNECTED);
3662 	if (so == NULL)
3663 		goto resetandabort;
3664 
3665 	switch (so->so_proto->pr_domain->dom_family) {
3666 #ifdef INET
3667 	case AF_INET:
3668 		inp = sotoinpcb(so);
3669 		break;
3670 #endif
3671 #ifdef INET6
3672 	case AF_INET6:
3673 		in6p = sotoin6pcb(so);
3674 		break;
3675 #endif
3676 	}
3677 	switch (src->sa_family) {
3678 #ifdef INET
3679 	case AF_INET:
3680 		if (inp) {
3681 			inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3682 			inp->inp_lport = ((struct sockaddr_in *)dst)->sin_port;
3683 			inp->inp_options = ip_srcroute();
3684 			in_pcbstate(inp, INP_BOUND);
3685 			if (inp->inp_options == NULL) {
3686 				inp->inp_options = sc->sc_ipopts;
3687 				sc->sc_ipopts = NULL;
3688 			}
3689 		}
3690 #ifdef INET6
3691 		else if (in6p) {
3692 			/* IPv4 packet to AF_INET6 socket */
3693 			memset(&in6p->in6p_laddr, 0, sizeof(in6p->in6p_laddr));
3694 			in6p->in6p_laddr.s6_addr16[5] = htons(0xffff);
3695 			bcopy(&((struct sockaddr_in *)dst)->sin_addr,
3696 				&in6p->in6p_laddr.s6_addr32[3],
3697 				sizeof(((struct sockaddr_in *)dst)->sin_addr));
3698 			in6p->in6p_lport = ((struct sockaddr_in *)dst)->sin_port;
3699 			in6totcpcb(in6p)->t_family = AF_INET;
3700 			if (sotoin6pcb(oso)->in6p_flags & IN6P_IPV6_V6ONLY)
3701 				in6p->in6p_flags |= IN6P_IPV6_V6ONLY;
3702 			else
3703 				in6p->in6p_flags &= ~IN6P_IPV6_V6ONLY;
3704 			in6_pcbstate(in6p, IN6P_BOUND);
3705 		}
3706 #endif
3707 		break;
3708 #endif
3709 #ifdef INET6
3710 	case AF_INET6:
3711 		if (in6p) {
3712 			in6p->in6p_laddr = ((struct sockaddr_in6 *)dst)->sin6_addr;
3713 			in6p->in6p_lport = ((struct sockaddr_in6 *)dst)->sin6_port;
3714 			in6_pcbstate(in6p, IN6P_BOUND);
3715 		}
3716 		break;
3717 #endif
3718 	}
3719 #ifdef INET6
3720 	if (in6p && in6totcpcb(in6p)->t_family == AF_INET6 && sotoinpcb(oso)) {
3721 		struct in6pcb *oin6p = sotoin6pcb(oso);
3722 		/* inherit socket options from the listening socket */
3723 		in6p->in6p_flags |= (oin6p->in6p_flags & IN6P_CONTROLOPTS);
3724 		if (in6p->in6p_flags & IN6P_CONTROLOPTS) {
3725 			m_freem(in6p->in6p_options);
3726 			in6p->in6p_options = 0;
3727 		}
3728 		ip6_savecontrol(in6p, &in6p->in6p_options,
3729 			mtod(m, struct ip6_hdr *), m);
3730 	}
3731 #endif
3732 
3733 #if defined(IPSEC) || defined(FAST_IPSEC)
3734 	/*
3735 	 * we make a copy of policy, instead of sharing the policy,
3736 	 * for better behavior in terms of SA lookup and dead SA removal.
3737 	 */
3738 	if (inp) {
3739 		/* copy old policy into new socket's */
3740 		if (ipsec_copy_pcbpolicy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
3741 			printf("tcp_input: could not copy policy\n");
3742 	}
3743 #ifdef INET6
3744 	else if (in6p) {
3745 		/* copy old policy into new socket's */
3746 		if (ipsec_copy_pcbpolicy(sotoin6pcb(oso)->in6p_sp,
3747 		    in6p->in6p_sp))
3748 			printf("tcp_input: could not copy policy\n");
3749 	}
3750 #endif
3751 #endif
3752 
3753 	/*
3754 	 * Give the new socket our cached route reference.
3755 	 */
3756 	if (inp) {
3757 		rtcache_copy(&inp->inp_route, &sc->sc_route);
3758 		rtcache_free(&sc->sc_route);
3759 	}
3760 #ifdef INET6
3761 	else {
3762 		rtcache_copy(&in6p->in6p_route, &sc->sc_route);
3763 		rtcache_free(&sc->sc_route);
3764 	}
3765 #endif
3766 
3767 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3768 	if (am == NULL)
3769 		goto resetandabort;
3770 	MCLAIM(am, &tcp_mowner);
3771 	am->m_len = src->sa_len;
3772 	bcopy(src, mtod(am, void *), src->sa_len);
3773 	if (inp) {
3774 		if (in_pcbconnect(inp, am, &lwp0)) {
3775 			(void) m_free(am);
3776 			goto resetandabort;
3777 		}
3778 	}
3779 #ifdef INET6
3780 	else if (in6p) {
3781 		if (src->sa_family == AF_INET) {
3782 			/* IPv4 packet to AF_INET6 socket */
3783 			struct sockaddr_in6 *sin6;
3784 			sin6 = mtod(am, struct sockaddr_in6 *);
3785 			am->m_len = sizeof(*sin6);
3786 			memset(sin6, 0, sizeof(*sin6));
3787 			sin6->sin6_family = AF_INET6;
3788 			sin6->sin6_len = sizeof(*sin6);
3789 			sin6->sin6_port = ((struct sockaddr_in *)src)->sin_port;
3790 			sin6->sin6_addr.s6_addr16[5] = htons(0xffff);
3791 			bcopy(&((struct sockaddr_in *)src)->sin_addr,
3792 				&sin6->sin6_addr.s6_addr32[3],
3793 				sizeof(sin6->sin6_addr.s6_addr32[3]));
3794 		}
3795 		if (in6_pcbconnect(in6p, am, NULL)) {
3796 			(void) m_free(am);
3797 			goto resetandabort;
3798 		}
3799 	}
3800 #endif
3801 	else {
3802 		(void) m_free(am);
3803 		goto resetandabort;
3804 	}
3805 	(void) m_free(am);
3806 
3807 	if (inp)
3808 		tp = intotcpcb(inp);
3809 #ifdef INET6
3810 	else if (in6p)
3811 		tp = in6totcpcb(in6p);
3812 #endif
3813 	else
3814 		tp = NULL;
3815 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3816 	if (sc->sc_request_r_scale != 15) {
3817 		tp->requested_s_scale = sc->sc_requested_s_scale;
3818 		tp->request_r_scale = sc->sc_request_r_scale;
3819 		tp->snd_scale = sc->sc_requested_s_scale;
3820 		tp->rcv_scale = sc->sc_request_r_scale;
3821 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3822 	}
3823 	if (sc->sc_flags & SCF_TIMESTAMP)
3824 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3825 	tp->ts_timebase = sc->sc_timebase;
3826 
3827 	tp->t_template = tcp_template(tp);
3828 	if (tp->t_template == 0) {
3829 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3830 		so = NULL;
3831 		m_freem(m);
3832 		goto abort;
3833 	}
3834 
3835 	tp->iss = sc->sc_iss;
3836 	tp->irs = sc->sc_irs;
3837 	tcp_sendseqinit(tp);
3838 	tcp_rcvseqinit(tp);
3839 	tp->t_state = TCPS_SYN_RECEIVED;
3840 	TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepinit);
3841 	TCP_STATINC(TCP_STAT_ACCEPTS);
3842 
3843 	if ((sc->sc_flags & SCF_SACK_PERMIT) && tcp_do_sack)
3844 		tp->t_flags |= TF_WILL_SACK;
3845 
3846 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tcp_do_ecn)
3847 		tp->t_flags |= TF_ECN_PERMIT;
3848 
3849 #ifdef TCP_SIGNATURE
3850 	if (sc->sc_flags & SCF_SIGNATURE)
3851 		tp->t_flags |= TF_SIGNATURE;
3852 #endif
3853 
3854 	/* Initialize tp->t_ourmss before we deal with the peer's! */
3855 	tp->t_ourmss = sc->sc_ourmaxseg;
3856 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
3857 
3858 	/*
3859 	 * Initialize the initial congestion window.  If we
3860 	 * had to retransmit the SYN,ACK, we must initialize cwnd
3861 	 * to 1 segment (i.e. the Loss Window).
3862 	 */
3863 	if (sc->sc_rxtshift)
3864 		tp->snd_cwnd = tp->t_peermss;
3865 	else {
3866 		int ss = tcp_init_win;
3867 #ifdef INET
3868 		if (inp != NULL && in_localaddr(inp->inp_faddr))
3869 			ss = tcp_init_win_local;
3870 #endif
3871 #ifdef INET6
3872 		if (in6p != NULL && in6_localaddr(&in6p->in6p_faddr))
3873 			ss = tcp_init_win_local;
3874 #endif
3875 		tp->snd_cwnd = TCP_INITIAL_WINDOW(ss, tp->t_peermss);
3876 	}
3877 
3878 	tcp_rmx_rtt(tp);
3879 	tp->snd_wl1 = sc->sc_irs;
3880 	tp->rcv_up = sc->sc_irs + 1;
3881 
3882 	/*
3883 	 * This is what whould have happened in tcp_output() when
3884 	 * the SYN,ACK was sent.
3885 	 */
3886 	tp->snd_up = tp->snd_una;
3887 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3888 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3889 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3890 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3891 	tp->last_ack_sent = tp->rcv_nxt;
3892 	tp->t_partialacks = -1;
3893 	tp->t_dupacks = 0;
3894 
3895 	TCP_STATINC(TCP_STAT_SC_COMPLETED);
3896 	s = splsoftnet();
3897 	syn_cache_put(sc);
3898 	splx(s);
3899 	return (so);
3900 
3901 resetandabort:
3902 	(void)tcp_respond(NULL, m, m, th, (tcp_seq)0, th->th_ack, TH_RST);
3903 abort:
3904 	if (so != NULL) {
3905 		(void) soqremque(so, 1);
3906 		(void) soabort(so);
3907 		mutex_enter(softnet_lock);
3908 	}
3909 	s = splsoftnet();
3910 	syn_cache_put(sc);
3911 	splx(s);
3912 	TCP_STATINC(TCP_STAT_SC_ABORTED);
3913 	return ((struct socket *)(-1));
3914 }
3915 
3916 /*
3917  * This function is called when we get a RST for a
3918  * non-existent connection, so that we can see if the
3919  * connection is in the syn cache.  If it is, zap it.
3920  */
3921 
3922 void
3923 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3924 {
3925 	struct syn_cache *sc;
3926 	struct syn_cache_head *scp;
3927 	int s = splsoftnet();
3928 
3929 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3930 		splx(s);
3931 		return;
3932 	}
3933 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3934 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3935 		splx(s);
3936 		return;
3937 	}
3938 	syn_cache_rm(sc);
3939 	TCP_STATINC(TCP_STAT_SC_RESET);
3940 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3941 	splx(s);
3942 }
3943 
3944 void
3945 syn_cache_unreach(const struct sockaddr *src, const struct sockaddr *dst,
3946     struct tcphdr *th)
3947 {
3948 	struct syn_cache *sc;
3949 	struct syn_cache_head *scp;
3950 	int s;
3951 
3952 	s = splsoftnet();
3953 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3954 		splx(s);
3955 		return;
3956 	}
3957 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3958 	if (ntohl (th->th_seq) != sc->sc_iss) {
3959 		splx(s);
3960 		return;
3961 	}
3962 
3963 	/*
3964 	 * If we've retransmitted 3 times and this is our second error,
3965 	 * we remove the entry.  Otherwise, we allow it to continue on.
3966 	 * This prevents us from incorrectly nuking an entry during a
3967 	 * spurious network outage.
3968 	 *
3969 	 * See tcp_notify().
3970 	 */
3971 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3972 		sc->sc_flags |= SCF_UNREACH;
3973 		splx(s);
3974 		return;
3975 	}
3976 
3977 	syn_cache_rm(sc);
3978 	TCP_STATINC(TCP_STAT_SC_UNREACH);
3979 	syn_cache_put(sc);	/* calls pool_put but see spl above */
3980 	splx(s);
3981 }
3982 
3983 /*
3984  * Given a LISTEN socket and an inbound SYN request, add
3985  * this to the syn cache, and send back a segment:
3986  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3987  * to the source.
3988  *
3989  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3990  * Doing so would require that we hold onto the data and deliver it
3991  * to the application.  However, if we are the target of a SYN-flood
3992  * DoS attack, an attacker could send data which would eventually
3993  * consume all available buffer space if it were ACKed.  By not ACKing
3994  * the data, we avoid this DoS scenario.
3995  */
3996 
3997 int
3998 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3999     unsigned int hlen, struct socket *so, struct mbuf *m, u_char *optp,
4000     int optlen, struct tcp_opt_info *oi)
4001 {
4002 	struct tcpcb tb, *tp;
4003 	long win;
4004 	struct syn_cache *sc;
4005 	struct syn_cache_head *scp;
4006 	struct mbuf *ipopts;
4007 	struct tcp_opt_info opti;
4008 	int s;
4009 
4010 	tp = sototcpcb(so);
4011 
4012 	memset(&opti, 0, sizeof(opti));
4013 
4014 	/*
4015 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
4016 	 *
4017 	 * Note this check is performed in tcp_input() very early on.
4018 	 */
4019 
4020 	/*
4021 	 * Initialize some local state.
4022 	 */
4023 	win = sbspace(&so->so_rcv);
4024 	if (win > TCP_MAXWIN)
4025 		win = TCP_MAXWIN;
4026 
4027 	switch (src->sa_family) {
4028 #ifdef INET
4029 	case AF_INET:
4030 		/*
4031 		 * Remember the IP options, if any.
4032 		 */
4033 		ipopts = ip_srcroute();
4034 		break;
4035 #endif
4036 	default:
4037 		ipopts = NULL;
4038 	}
4039 
4040 #ifdef TCP_SIGNATURE
4041 	if (optp || (tp->t_flags & TF_SIGNATURE))
4042 #else
4043 	if (optp)
4044 #endif
4045 	{
4046 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4047 #ifdef TCP_SIGNATURE
4048 		tb.t_flags |= (tp->t_flags & TF_SIGNATURE);
4049 #endif
4050 		tb.t_state = TCPS_LISTEN;
4051 		if (tcp_dooptions(&tb, optp, optlen, th, m, m->m_pkthdr.len -
4052 		    sizeof(struct tcphdr) - optlen - hlen, oi) < 0)
4053 			return (0);
4054 	} else
4055 		tb.t_flags = 0;
4056 
4057 	/*
4058 	 * See if we already have an entry for this connection.
4059 	 * If we do, resend the SYN,ACK.  We do not count this
4060 	 * as a retransmission (XXX though maybe we should).
4061 	 */
4062 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4063 		TCP_STATINC(TCP_STAT_SC_DUPESYN);
4064 		if (ipopts) {
4065 			/*
4066 			 * If we were remembering a previous source route,
4067 			 * forget it and use the new one we've been given.
4068 			 */
4069 			if (sc->sc_ipopts)
4070 				(void) m_free(sc->sc_ipopts);
4071 			sc->sc_ipopts = ipopts;
4072 		}
4073 		sc->sc_timestamp = tb.ts_recent;
4074 		if (syn_cache_respond(sc, m) == 0) {
4075 			uint64_t *tcps = TCP_STAT_GETREF();
4076 			tcps[TCP_STAT_SNDACKS]++;
4077 			tcps[TCP_STAT_SNDTOTAL]++;
4078 			TCP_STAT_PUTREF();
4079 		}
4080 		return (1);
4081 	}
4082 
4083 	s = splsoftnet();
4084 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4085 	splx(s);
4086 	if (sc == NULL) {
4087 		if (ipopts)
4088 			(void) m_free(ipopts);
4089 		return (0);
4090 	}
4091 
4092 	/*
4093 	 * Fill in the cache, and put the necessary IP and TCP
4094 	 * options into the reply.
4095 	 */
4096 	memset(sc, 0, sizeof(struct syn_cache));
4097 	callout_init(&sc->sc_timer, CALLOUT_MPSAFE);
4098 	bcopy(src, &sc->sc_src, src->sa_len);
4099 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4100 	sc->sc_flags = 0;
4101 	sc->sc_ipopts = ipopts;
4102 	sc->sc_irs = th->th_seq;
4103 	switch (src->sa_family) {
4104 #ifdef INET
4105 	case AF_INET:
4106 	    {
4107 		struct sockaddr_in *srcin = (void *) src;
4108 		struct sockaddr_in *dstin = (void *) dst;
4109 
4110 		sc->sc_iss = tcp_new_iss1(&dstin->sin_addr,
4111 		    &srcin->sin_addr, dstin->sin_port,
4112 		    srcin->sin_port, sizeof(dstin->sin_addr), 0);
4113 		break;
4114 	    }
4115 #endif /* INET */
4116 #ifdef INET6
4117 	case AF_INET6:
4118 	    {
4119 		struct sockaddr_in6 *srcin6 = (void *) src;
4120 		struct sockaddr_in6 *dstin6 = (void *) dst;
4121 
4122 		sc->sc_iss = tcp_new_iss1(&dstin6->sin6_addr,
4123 		    &srcin6->sin6_addr, dstin6->sin6_port,
4124 		    srcin6->sin6_port, sizeof(dstin6->sin6_addr), 0);
4125 		break;
4126 	    }
4127 #endif /* INET6 */
4128 	}
4129 	sc->sc_peermaxseg = oi->maxseg;
4130 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
4131 						m->m_pkthdr.rcvif : NULL,
4132 						sc->sc_src.sa.sa_family);
4133 	sc->sc_win = win;
4134 	sc->sc_timebase = tcp_now - 1;	/* see tcp_newtcpcb() */
4135 	sc->sc_timestamp = tb.ts_recent;
4136 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4137 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP))
4138 		sc->sc_flags |= SCF_TIMESTAMP;
4139 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4140 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4141 		sc->sc_requested_s_scale = tb.requested_s_scale;
4142 		sc->sc_request_r_scale = 0;
4143 		/*
4144 		 * Pick the smallest possible scaling factor that
4145 		 * will still allow us to scale up to sb_max.
4146 		 *
4147 		 * We do this because there are broken firewalls that
4148 		 * will corrupt the window scale option, leading to
4149 		 * the other endpoint believing that our advertised
4150 		 * window is unscaled.  At scale factors larger than
4151 		 * 5 the unscaled window will drop below 1500 bytes,
4152 		 * leading to serious problems when traversing these
4153 		 * broken firewalls.
4154 		 *
4155 		 * With the default sbmax of 256K, a scale factor
4156 		 * of 3 will be chosen by this algorithm.  Those who
4157 		 * choose a larger sbmax should watch out
4158 		 * for the compatiblity problems mentioned above.
4159 		 *
4160 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4161 		 * or <SYN,ACK>) segment itself is never scaled.
4162 		 */
4163 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4164 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4165 			sc->sc_request_r_scale++;
4166 	} else {
4167 		sc->sc_requested_s_scale = 15;
4168 		sc->sc_request_r_scale = 15;
4169 	}
4170 	if ((tb.t_flags & TF_SACK_PERMIT) && tcp_do_sack)
4171 		sc->sc_flags |= SCF_SACK_PERMIT;
4172 
4173 	/*
4174 	 * ECN setup packet recieved.
4175 	 */
4176 	if ((th->th_flags & (TH_ECE|TH_CWR)) && tcp_do_ecn)
4177 		sc->sc_flags |= SCF_ECN_PERMIT;
4178 
4179 #ifdef TCP_SIGNATURE
4180 	if (tb.t_flags & TF_SIGNATURE)
4181 		sc->sc_flags |= SCF_SIGNATURE;
4182 #endif
4183 	sc->sc_tp = tp;
4184 	if (syn_cache_respond(sc, m) == 0) {
4185 		uint64_t *tcps = TCP_STAT_GETREF();
4186 		tcps[TCP_STAT_SNDACKS]++;
4187 		tcps[TCP_STAT_SNDTOTAL]++;
4188 		TCP_STAT_PUTREF();
4189 		syn_cache_insert(sc, tp);
4190 	} else {
4191 		s = splsoftnet();
4192 		syn_cache_put(sc);
4193 		splx(s);
4194 		TCP_STATINC(TCP_STAT_SC_DROPPED);
4195 	}
4196 	return (1);
4197 }
4198 
4199 int
4200 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4201 {
4202 #ifdef INET6
4203 	struct rtentry *rt;
4204 #endif
4205 	struct route *ro;
4206 	u_int8_t *optp;
4207 	int optlen, error;
4208 	u_int16_t tlen;
4209 	struct ip *ip = NULL;
4210 #ifdef INET6
4211 	struct ip6_hdr *ip6 = NULL;
4212 #endif
4213 	struct tcpcb *tp = NULL;
4214 	struct tcphdr *th;
4215 	u_int hlen;
4216 	struct socket *so;
4217 
4218 	ro = &sc->sc_route;
4219 	switch (sc->sc_src.sa.sa_family) {
4220 	case AF_INET:
4221 		hlen = sizeof(struct ip);
4222 		break;
4223 #ifdef INET6
4224 	case AF_INET6:
4225 		hlen = sizeof(struct ip6_hdr);
4226 		break;
4227 #endif
4228 	default:
4229 		if (m)
4230 			m_freem(m);
4231 		return (EAFNOSUPPORT);
4232 	}
4233 
4234 	/* Compute the size of the TCP options. */
4235 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4236 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? (TCPOLEN_SACK_PERMITTED + 2) : 0) +
4237 #ifdef TCP_SIGNATURE
4238 	    ((sc->sc_flags & SCF_SIGNATURE) ? (TCPOLEN_SIGNATURE + 2) : 0) +
4239 #endif
4240 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4241 
4242 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4243 
4244 	/*
4245 	 * Create the IP+TCP header from scratch.
4246 	 */
4247 	if (m)
4248 		m_freem(m);
4249 #ifdef DIAGNOSTIC
4250 	if (max_linkhdr + tlen > MCLBYTES)
4251 		return (ENOBUFS);
4252 #endif
4253 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4254 	if (m && tlen > MHLEN) {
4255 		MCLGET(m, M_DONTWAIT);
4256 		if ((m->m_flags & M_EXT) == 0) {
4257 			m_freem(m);
4258 			m = NULL;
4259 		}
4260 	}
4261 	if (m == NULL)
4262 		return (ENOBUFS);
4263 	MCLAIM(m, &tcp_tx_mowner);
4264 
4265 	/* Fixup the mbuf. */
4266 	m->m_data += max_linkhdr;
4267 	m->m_len = m->m_pkthdr.len = tlen;
4268 	if (sc->sc_tp) {
4269 		tp = sc->sc_tp;
4270 		if (tp->t_inpcb)
4271 			so = tp->t_inpcb->inp_socket;
4272 #ifdef INET6
4273 		else if (tp->t_in6pcb)
4274 			so = tp->t_in6pcb->in6p_socket;
4275 #endif
4276 		else
4277 			so = NULL;
4278 	} else
4279 		so = NULL;
4280 	m->m_pkthdr.rcvif = NULL;
4281 	memset(mtod(m, u_char *), 0, tlen);
4282 
4283 	switch (sc->sc_src.sa.sa_family) {
4284 	case AF_INET:
4285 		ip = mtod(m, struct ip *);
4286 		ip->ip_v = 4;
4287 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4288 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4289 		ip->ip_p = IPPROTO_TCP;
4290 		th = (struct tcphdr *)(ip + 1);
4291 		th->th_dport = sc->sc_src.sin.sin_port;
4292 		th->th_sport = sc->sc_dst.sin.sin_port;
4293 		break;
4294 #ifdef INET6
4295 	case AF_INET6:
4296 		ip6 = mtod(m, struct ip6_hdr *);
4297 		ip6->ip6_vfc = IPV6_VERSION;
4298 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4299 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4300 		ip6->ip6_nxt = IPPROTO_TCP;
4301 		/* ip6_plen will be updated in ip6_output() */
4302 		th = (struct tcphdr *)(ip6 + 1);
4303 		th->th_dport = sc->sc_src.sin6.sin6_port;
4304 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4305 		break;
4306 #endif
4307 	default:
4308 		th = NULL;
4309 	}
4310 
4311 	th->th_seq = htonl(sc->sc_iss);
4312 	th->th_ack = htonl(sc->sc_irs + 1);
4313 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4314 	th->th_flags = TH_SYN|TH_ACK;
4315 	th->th_win = htons(sc->sc_win);
4316 	/* th_sum already 0 */
4317 	/* th_urp already 0 */
4318 
4319 	/* Tack on the TCP options. */
4320 	optp = (u_int8_t *)(th + 1);
4321 	*optp++ = TCPOPT_MAXSEG;
4322 	*optp++ = 4;
4323 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4324 	*optp++ = sc->sc_ourmaxseg & 0xff;
4325 
4326 	if (sc->sc_request_r_scale != 15) {
4327 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4328 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4329 		    sc->sc_request_r_scale);
4330 		optp += 4;
4331 	}
4332 
4333 	if (sc->sc_flags & SCF_TIMESTAMP) {
4334 		u_int32_t *lp = (u_int32_t *)(optp);
4335 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4336 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4337 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4338 		*lp   = htonl(sc->sc_timestamp);
4339 		optp += TCPOLEN_TSTAMP_APPA;
4340 	}
4341 
4342 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4343 		u_int8_t *p = optp;
4344 
4345 		/* Let the peer know that we will SACK. */
4346 		p[0] = TCPOPT_SACK_PERMITTED;
4347 		p[1] = 2;
4348 		p[2] = TCPOPT_NOP;
4349 		p[3] = TCPOPT_NOP;
4350 		optp += 4;
4351 	}
4352 
4353 	/*
4354 	 * Send ECN SYN-ACK setup packet.
4355 	 * Routes can be asymetric, so, even if we receive a packet
4356 	 * with ECE and CWR set, we must not assume no one will block
4357 	 * the ECE packet we are about to send.
4358 	 */
4359 	if ((sc->sc_flags & SCF_ECN_PERMIT) && tp &&
4360 	    SEQ_GEQ(tp->snd_nxt, tp->snd_max)) {
4361 		th->th_flags |= TH_ECE;
4362 		TCP_STATINC(TCP_STAT_ECN_SHS);
4363 
4364 		/*
4365 		 * draft-ietf-tcpm-ecnsyn-00.txt
4366 		 *
4367 		 * "[...] a TCP node MAY respond to an ECN-setup
4368 		 * SYN packet by setting ECT in the responding
4369 		 * ECN-setup SYN/ACK packet, indicating to routers
4370 		 * that the SYN/ACK packet is ECN-Capable.
4371 		 * This allows a congested router along the path
4372 		 * to mark the packet instead of dropping the
4373 		 * packet as an indication of congestion."
4374 		 *
4375 		 * "[...] There can be a great benefit in setting
4376 		 * an ECN-capable codepoint in SYN/ACK packets [...]
4377 		 * Congestion is  most likely to occur in
4378 		 * the server-to-client direction.  As a result,
4379 		 * setting an ECN-capable codepoint in SYN/ACK
4380 		 * packets can reduce the occurence of three-second
4381 		 * retransmit timeouts resulting from the drop
4382 		 * of SYN/ACK packets."
4383 		 *
4384 		 * Page 4 and 6, January 2006.
4385 		 */
4386 
4387 		switch (sc->sc_src.sa.sa_family) {
4388 #ifdef INET
4389 		case AF_INET:
4390 			ip->ip_tos |= IPTOS_ECN_ECT0;
4391 			break;
4392 #endif
4393 #ifdef INET6
4394 		case AF_INET6:
4395 			ip6->ip6_flow |= htonl(IPTOS_ECN_ECT0 << 20);
4396 			break;
4397 #endif
4398 		}
4399 		TCP_STATINC(TCP_STAT_ECN_ECT);
4400 	}
4401 
4402 #ifdef TCP_SIGNATURE
4403 	if (sc->sc_flags & SCF_SIGNATURE) {
4404 		struct secasvar *sav;
4405 		u_int8_t *sigp;
4406 
4407 		sav = tcp_signature_getsav(m, th);
4408 
4409 		if (sav == NULL) {
4410 			if (m)
4411 				m_freem(m);
4412 			return (EPERM);
4413 		}
4414 
4415 		*optp++ = TCPOPT_SIGNATURE;
4416 		*optp++ = TCPOLEN_SIGNATURE;
4417 		sigp = optp;
4418 		memset(optp, 0, TCP_SIGLEN);
4419 		optp += TCP_SIGLEN;
4420 		*optp++ = TCPOPT_NOP;
4421 		*optp++ = TCPOPT_EOL;
4422 
4423 		(void)tcp_signature(m, th, hlen, sav, sigp);
4424 
4425 		key_sa_recordxfer(sav, m);
4426 #ifdef FAST_IPSEC
4427 		KEY_FREESAV(&sav);
4428 #else
4429 		key_freesav(sav);
4430 #endif
4431 	}
4432 #endif
4433 
4434 	/* Compute the packet's checksum. */
4435 	switch (sc->sc_src.sa.sa_family) {
4436 	case AF_INET:
4437 		ip->ip_len = htons(tlen - hlen);
4438 		th->th_sum = 0;
4439 		th->th_sum = in4_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4440 		break;
4441 #ifdef INET6
4442 	case AF_INET6:
4443 		ip6->ip6_plen = htons(tlen - hlen);
4444 		th->th_sum = 0;
4445 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4446 		break;
4447 #endif
4448 	}
4449 
4450 	/*
4451 	 * Fill in some straggling IP bits.  Note the stack expects
4452 	 * ip_len to be in host order, for convenience.
4453 	 */
4454 	switch (sc->sc_src.sa.sa_family) {
4455 #ifdef INET
4456 	case AF_INET:
4457 		ip->ip_len = htons(tlen);
4458 		ip->ip_ttl = ip_defttl;
4459 		/* XXX tos? */
4460 		break;
4461 #endif
4462 #ifdef INET6
4463 	case AF_INET6:
4464 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4465 		ip6->ip6_vfc |= IPV6_VERSION;
4466 		ip6->ip6_plen = htons(tlen - hlen);
4467 		/* ip6_hlim will be initialized afterwards */
4468 		/* XXX flowlabel? */
4469 		break;
4470 #endif
4471 	}
4472 
4473 	/* XXX use IPsec policy on listening socket, on SYN ACK */
4474 	tp = sc->sc_tp;
4475 
4476 	switch (sc->sc_src.sa.sa_family) {
4477 #ifdef INET
4478 	case AF_INET:
4479 		error = ip_output(m, sc->sc_ipopts, ro,
4480 		    (ip_mtudisc ? IP_MTUDISC : 0),
4481 		    (struct ip_moptions *)NULL, so);
4482 		break;
4483 #endif
4484 #ifdef INET6
4485 	case AF_INET6:
4486 		ip6->ip6_hlim = in6_selecthlim(NULL,
4487 				(rt = rtcache_validate(ro)) != NULL ? rt->rt_ifp
4488 				                                    : NULL);
4489 
4490 		error = ip6_output(m, NULL /*XXX*/, ro, 0, NULL, so, NULL);
4491 		break;
4492 #endif
4493 	default:
4494 		error = EAFNOSUPPORT;
4495 		break;
4496 	}
4497 	return (error);
4498 }
4499