xref: /openbsd-src/sys/netinet/tcp_input.c (revision 850e275390052b330d93020bf619a739a3c277ac)
1 /*	$OpenBSD: tcp_input.c,v 1.221 2008/09/09 15:26:12 mpf Exp $	*/
2 /*	$NetBSD: tcp_input.c,v 1.23 1996/02/13 23:43:44 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)COPYRIGHT	1.1 (NRL) 17 January 1995
33  *
34  * NRL grants permission for redistribution and use in source and binary
35  * forms, with or without modification, of the software and documentation
36  * created at NRL provided that the following conditions are met:
37  *
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgements:
45  * 	This product includes software developed by the University of
46  * 	California, Berkeley and its contributors.
47  * 	This product includes software developed at the Information
48  * 	Technology Division, US Naval Research Laboratory.
49  * 4. Neither the name of the NRL nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
54  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
55  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
56  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
57  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
58  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
59  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
60  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
61  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
62  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
63  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
64  *
65  * The views and conclusions contained in the software and documentation
66  * are those of the authors and should not be interpreted as representing
67  * official policies, either expressed or implied, of the US Naval
68  * Research Laboratory (NRL).
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/mbuf.h>
74 #include <sys/protosw.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/kernel.h>
78 #include <sys/pool.h>
79 
80 #include <dev/rndvar.h>
81 
82 #include <net/if.h>
83 #include <net/route.h>
84 
85 #include <netinet/in.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/ip.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/tcp.h>
91 #include <netinet/tcp_fsm.h>
92 #include <netinet/tcp_seq.h>
93 #include <netinet/tcp_timer.h>
94 #include <netinet/tcp_var.h>
95 #include <netinet/tcpip.h>
96 #include <netinet/tcp_debug.h>
97 
98 #include "faith.h"
99 
100 #include "pf.h"
101 #if NPF > 0
102 #include <net/pfvar.h>
103 #endif
104 
105 struct	tcpiphdr tcp_saveti;
106 
107 int tcp_mss_adv(struct ifnet *, int);
108 
109 #ifdef INET6
110 #include <netinet6/in6_var.h>
111 #include <netinet6/nd6.h>
112 
113 struct  tcpipv6hdr tcp_saveti6;
114 
115 /* for the packet header length in the mbuf */
116 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
117 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
118 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
119 #endif /* INET6 */
120 
121 int	tcprexmtthresh = 3;
122 int	tcptv_keep_init = TCPTV_KEEP_INIT;
123 
124 extern u_long sb_max;
125 
126 int tcp_rst_ppslim = 100;		/* 100pps */
127 int tcp_rst_ppslim_count = 0;
128 struct timeval tcp_rst_ppslim_last;
129 
130 int tcp_ackdrop_ppslim = 100;		/* 100pps */
131 int tcp_ackdrop_ppslim_count = 0;
132 struct timeval tcp_ackdrop_ppslim_last;
133 
134 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
135 
136 /* for modulo comparisons of timestamps */
137 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
138 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
139 
140 /* for TCP SACK comparisons */
141 #define	SEQ_MIN(a,b)	(SEQ_LT(a,b) ? (a) : (b))
142 #define	SEQ_MAX(a,b)	(SEQ_GT(a,b) ? (a) : (b))
143 
144 /*
145  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
146  */
147 #ifdef INET6
148 #define ND6_HINT(tp) \
149 do { \
150 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
151 	    tp->t_inpcb->inp_route6.ro_rt) { \
152 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \
153 	} \
154 } while (0)
155 #else
156 #define ND6_HINT(tp)
157 #endif
158 
159 #ifdef TCP_ECN
160 /*
161  * ECN (Explicit Congestion Notification) support based on RFC3168
162  * implementation note:
163  *   snd_last is used to track a recovery phase.
164  *   when cwnd is reduced, snd_last is set to snd_max.
165  *   while snd_last > snd_una, the sender is in a recovery phase and
166  *   its cwnd should not be reduced again.
167  *   snd_last follows snd_una when not in a recovery phase.
168  */
169 #endif
170 
171 /*
172  * Macro to compute ACK transmission behavior.  Delay the ACK unless
173  * we have already delayed an ACK (must send an ACK every two segments).
174  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
175  * option is enabled.
176  */
177 #define	TCP_SETUP_ACK(tp, tiflags) \
178 do { \
179 	if ((tp)->t_flags & TF_DELACK || \
180 	    (tcp_ack_on_push && (tiflags) & TH_PUSH)) \
181 		tp->t_flags |= TF_ACKNOW; \
182 	else \
183 		TCP_SET_DELACK(tp); \
184 } while (0)
185 
186 /*
187  * Insert segment ti into reassembly queue of tcp with
188  * control block tp.  Return TH_FIN if reassembly now includes
189  * a segment with FIN.  The macro form does the common case inline
190  * (segment is the next to be received on an established connection,
191  * and the queue is empty), avoiding linkage into and removal
192  * from the queue and repetition of various conversions.
193  * Set DELACK for segments received in order, but ack immediately
194  * when segments are out of order (so fast retransmit can work).
195  */
196 
197 int
198 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen)
199 {
200 	struct tcpqent *p, *q, *nq, *tiqe;
201 	struct socket *so = tp->t_inpcb->inp_socket;
202 	int flags;
203 
204 	/*
205 	 * Call with th==0 after become established to
206 	 * force pre-ESTABLISHED data up to user socket.
207 	 */
208 	if (th == 0)
209 		goto present;
210 
211 	/*
212 	 * Allocate a new queue entry, before we throw away any data.
213 	 * If we can't, just drop the packet.  XXX
214 	 */
215 	tiqe = pool_get(&tcpqe_pool, PR_NOWAIT);
216 	if (tiqe == NULL) {
217 		tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead);
218 		if (tiqe != NULL && th->th_seq == tp->rcv_nxt) {
219 			/* Reuse last entry since new segment fills a hole */
220 			m_freem(tiqe->tcpqe_m);
221 			TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q);
222 		}
223 		if (tiqe == NULL || th->th_seq != tp->rcv_nxt) {
224 			/* Flush segment queue for this connection */
225 			tcp_freeq(tp);
226 			tcpstat.tcps_rcvmemdrop++;
227 			m_freem(m);
228 			return (0);
229 		}
230 	}
231 
232 	/*
233 	 * Find a segment which begins after this one does.
234 	 */
235 	for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL;
236 	    p = q, q = TAILQ_NEXT(q, tcpqe_q))
237 		if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq))
238 			break;
239 
240 	/*
241 	 * If there is a preceding segment, it may provide some of
242 	 * our data already.  If so, drop the data from the incoming
243 	 * segment.  If it provides all of our data, drop us.
244 	 */
245 	if (p != NULL) {
246 		struct tcphdr *phdr = p->tcpqe_tcp;
247 		int i;
248 
249 		/* conversion to int (in i) handles seq wraparound */
250 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
251 		if (i > 0) {
252 		        if (i >= *tlen) {
253 				tcpstat.tcps_rcvduppack++;
254 				tcpstat.tcps_rcvdupbyte += *tlen;
255 				m_freem(m);
256 				pool_put(&tcpqe_pool, tiqe);
257 				return (0);
258 			}
259 			m_adj(m, i);
260 			*tlen -= i;
261 			th->th_seq += i;
262 		}
263 	}
264 	tcpstat.tcps_rcvoopack++;
265 	tcpstat.tcps_rcvoobyte += *tlen;
266 
267 	/*
268 	 * While we overlap succeeding segments trim them or,
269 	 * if they are completely covered, dequeue them.
270 	 */
271 	for (; q != NULL; q = nq) {
272 		struct tcphdr *qhdr = q->tcpqe_tcp;
273 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
274 
275 		if (i <= 0)
276 			break;
277 		if (i < qhdr->th_reseqlen) {
278 			qhdr->th_seq += i;
279 			qhdr->th_reseqlen -= i;
280 			m_adj(q->tcpqe_m, i);
281 			break;
282 		}
283 		nq = TAILQ_NEXT(q, tcpqe_q);
284 		m_freem(q->tcpqe_m);
285 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
286 		pool_put(&tcpqe_pool, q);
287 	}
288 
289 	/* Insert the new segment queue entry into place. */
290 	tiqe->tcpqe_m = m;
291 	th->th_reseqlen = *tlen;
292 	tiqe->tcpqe_tcp = th;
293 	if (p == NULL) {
294 		TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q);
295 	} else {
296 		TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q);
297 	}
298 
299 present:
300 	/*
301 	 * Present data to user, advancing rcv_nxt through
302 	 * completed sequence space.
303 	 */
304 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
305 		return (0);
306 	q = TAILQ_FIRST(&tp->t_segq);
307 	if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt)
308 		return (0);
309 	if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen)
310 		return (0);
311 	do {
312 		tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen;
313 		flags = q->tcpqe_tcp->th_flags & TH_FIN;
314 
315 		nq = TAILQ_NEXT(q, tcpqe_q);
316 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
317 		ND6_HINT(tp);
318 		if (so->so_state & SS_CANTRCVMORE)
319 			m_freem(q->tcpqe_m);
320 		else
321 			sbappendstream(&so->so_rcv, q->tcpqe_m);
322 		pool_put(&tcpqe_pool, q);
323 		q = nq;
324 	} while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt);
325 	sorwakeup(so);
326 	return (flags);
327 }
328 
329 #ifdef INET6
330 int
331 tcp6_input(struct mbuf **mp, int *offp, int proto)
332 {
333 	struct mbuf *m = *mp;
334 
335 #if NFAITH > 0
336 	if (m->m_pkthdr.rcvif) {
337 		if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
338 			/* XXX send icmp6 host/port unreach? */
339 			m_freem(m);
340 			return IPPROTO_DONE;
341 		}
342 	}
343 #endif
344 
345 	/*
346 	 * draft-itojun-ipv6-tcp-to-anycast
347 	 * better place to put this in?
348 	 */
349 	if (m->m_flags & M_ANYCAST6) {
350 		if (m->m_len >= sizeof(struct ip6_hdr)) {
351 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
352 			icmp6_error(m, ICMP6_DST_UNREACH,
353 				ICMP6_DST_UNREACH_ADDR,
354 				(caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
355 		} else
356 			m_freem(m);
357 		return IPPROTO_DONE;
358 	}
359 
360 	tcp_input(m, *offp, proto);
361 	return IPPROTO_DONE;
362 }
363 #endif
364 
365 /*
366  * TCP input routine, follows pages 65-76 of the
367  * protocol specification dated September, 1981 very closely.
368  */
369 void
370 tcp_input(struct mbuf *m, ...)
371 {
372 	struct ip *ip;
373 	struct inpcb *inp = NULL;
374 	u_int8_t *optp = NULL;
375 	int optlen = 0;
376 	int tlen, off;
377 	struct tcpcb *tp = 0;
378 	int tiflags;
379 	struct socket *so = NULL;
380 	int todrop, acked, ourfinisacked, needoutput = 0;
381 	int hdroptlen = 0;
382 	short ostate = 0;
383 	tcp_seq iss, *reuse = NULL;
384 	u_long tiwin;
385 	struct tcp_opt_info opti;
386 	int iphlen;
387 	va_list ap;
388 	struct tcphdr *th;
389 #ifdef INET6
390 	struct ip6_hdr *ip6 = NULL;
391 #endif /* INET6 */
392 #ifdef IPSEC
393 	struct m_tag *mtag;
394 	struct tdb_ident *tdbi;
395 	struct tdb *tdb;
396 	int error, s;
397 #endif /* IPSEC */
398 	int af;
399 #ifdef TCP_ECN
400 	u_char iptos;
401 #endif
402 
403 	va_start(ap, m);
404 	iphlen = va_arg(ap, int);
405 	va_end(ap);
406 
407 	tcpstat.tcps_rcvtotal++;
408 
409 	opti.ts_present = 0;
410 	opti.maxseg = 0;
411 
412 	/*
413 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
414 	 * See below for AF specific multicast.
415 	 */
416 	if (m->m_flags & (M_BCAST|M_MCAST))
417 		goto drop;
418 
419 	/*
420 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
421 	 * TCP/IPv4.
422 	 */
423 	switch (mtod(m, struct ip *)->ip_v) {
424 #ifdef INET6
425 	case 6:
426 		af = AF_INET6;
427 		break;
428 #endif
429 	case 4:
430 		af = AF_INET;
431 		break;
432 	default:
433 		m_freem(m);
434 		return;	/*EAFNOSUPPORT*/
435 	}
436 
437 	/*
438 	 * Get IP and TCP header together in first mbuf.
439 	 * Note: IP leaves IP header in first mbuf.
440 	 */
441 	switch (af) {
442 	case AF_INET:
443 #ifdef DIAGNOSTIC
444 		if (iphlen < sizeof(struct ip)) {
445 			m_freem(m);
446 			return;
447 		}
448 #endif /* DIAGNOSTIC */
449 		break;
450 #ifdef INET6
451 	case AF_INET6:
452 #ifdef DIAGNOSTIC
453 		if (iphlen < sizeof(struct ip6_hdr)) {
454 			m_freem(m);
455 			return;
456 		}
457 #endif /* DIAGNOSTIC */
458 		break;
459 #endif
460 	default:
461 		m_freem(m);
462 		return;
463 	}
464 
465 	IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th));
466 	if (!th) {
467 		tcpstat.tcps_rcvshort++;
468 		return;
469 	}
470 
471 	tlen = m->m_pkthdr.len - iphlen;
472 	ip = NULL;
473 #ifdef INET6
474 	ip6 = NULL;
475 #endif
476 	switch (af) {
477 	case AF_INET:
478 		ip = mtod(m, struct ip *);
479 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
480 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
481 			goto drop;
482 #ifdef TCP_ECN
483 		/* save ip_tos before clearing it for checksum */
484 		iptos = ip->ip_tos;
485 #endif
486 		/*
487 		 * Checksum extended TCP header and data.
488 		 */
489 		if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) {
490 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) {
491 				tcpstat.tcps_inhwcsum++;
492 				tcpstat.tcps_rcvbadsum++;
493 				goto drop;
494 			}
495 			if (in4_cksum(m, IPPROTO_TCP, iphlen, tlen) != 0) {
496 				tcpstat.tcps_rcvbadsum++;
497 				goto drop;
498 			}
499 		} else {
500 			m->m_pkthdr.csum_flags &= ~M_TCP_CSUM_IN_OK;
501 			tcpstat.tcps_inhwcsum++;
502 		}
503 		break;
504 #ifdef INET6
505 	case AF_INET6:
506 		ip6 = mtod(m, struct ip6_hdr *);
507 #ifdef TCP_ECN
508 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
509 #endif
510 
511 		/* Be proactive about malicious use of IPv4 mapped address */
512 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
513 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
514 			/* XXX stat */
515 			goto drop;
516 		}
517 
518 		/*
519 		 * Be proactive about unspecified IPv6 address in source.
520 		 * As we use all-zero to indicate unbounded/unconnected pcb,
521 		 * unspecified IPv6 address can be used to confuse us.
522 		 *
523 		 * Note that packets with unspecified IPv6 destination is
524 		 * already dropped in ip6_input.
525 		 */
526 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
527 			/* XXX stat */
528 			goto drop;
529 		}
530 
531 		/* Discard packets to multicast */
532 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
533 			/* XXX stat */
534 			goto drop;
535 		}
536 
537 		/*
538 		 * Checksum extended TCP header and data.
539 		 */
540 		if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) {
541 			tcpstat.tcps_rcvbadsum++;
542 			goto drop;
543 		}
544 		break;
545 #endif
546 	}
547 
548 	/*
549 	 * Check that TCP offset makes sense,
550 	 * pull out TCP options and adjust length.		XXX
551 	 */
552 	off = th->th_off << 2;
553 	if (off < sizeof(struct tcphdr) || off > tlen) {
554 		tcpstat.tcps_rcvbadoff++;
555 		goto drop;
556 	}
557 	tlen -= off;
558 	if (off > sizeof(struct tcphdr)) {
559 		IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off);
560 		if (!th) {
561 			tcpstat.tcps_rcvshort++;
562 			return;
563 		}
564 		optlen = off - sizeof(struct tcphdr);
565 		optp = (u_int8_t *)(th + 1);
566 		/*
567 		 * Do quick retrieval of timestamp options ("options
568 		 * prediction?").  If timestamp is the only option and it's
569 		 * formatted as recommended in RFC 1323 appendix A, we
570 		 * quickly get the values now and not bother calling
571 		 * tcp_dooptions(), etc.
572 		 */
573 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
574 		     (optlen > TCPOLEN_TSTAMP_APPA &&
575 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
576 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
577 		     (th->th_flags & TH_SYN) == 0) {
578 			opti.ts_present = 1;
579 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
580 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
581 			optp = NULL;	/* we've parsed the options */
582 		}
583 	}
584 	tiflags = th->th_flags;
585 
586 	/*
587 	 * Convert TCP protocol specific fields to host format.
588 	 */
589 	NTOHL(th->th_seq);
590 	NTOHL(th->th_ack);
591 	NTOHS(th->th_win);
592 	NTOHS(th->th_urp);
593 
594 	/*
595 	 * Locate pcb for segment.
596 	 */
597 #if NPF > 0
598 	if (m->m_pkthdr.pf.statekey)
599 		inp = ((struct pf_state_key *)m->m_pkthdr.pf.statekey)->inp;
600 #endif
601 findpcb:
602 	if (inp == NULL) {
603 		switch (af) {
604 #ifdef INET6
605 		case AF_INET6:
606 			inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src,
607 			    th->th_sport, &ip6->ip6_dst, th->th_dport);
608 			break;
609 #endif
610 		case AF_INET:
611 			inp = in_pcbhashlookup(&tcbtable, ip->ip_src,
612 			    th->th_sport, ip->ip_dst, th->th_dport);
613 			break;
614 		}
615 #if NPF > 0
616 		if (m->m_pkthdr.pf.statekey && inp) {
617 			((struct pf_state_key *)m->m_pkthdr.pf.statekey)->inp =
618 			    inp;
619 			inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
620 		}
621 #endif
622 	}
623 	if (inp == NULL) {
624 		int	inpl_flags = 0;
625 		if (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST)
626 			inpl_flags = INPLOOKUP_WILDCARD;
627 		++tcpstat.tcps_pcbhashmiss;
628 		switch (af) {
629 #ifdef INET6
630 		case AF_INET6:
631 			inp = in6_pcblookup_listen(&tcbtable,
632 			    &ip6->ip6_dst, th->th_dport, inpl_flags, m);
633 			break;
634 #endif /* INET6 */
635 		case AF_INET:
636 			inp = in_pcblookup_listen(&tcbtable,
637 			    ip->ip_dst, th->th_dport, inpl_flags, m);
638 			break;
639 		}
640 		/*
641 		 * If the state is CLOSED (i.e., TCB does not exist) then
642 		 * all data in the incoming segment is discarded.
643 		 * If the TCB exists but is in CLOSED state, it is embryonic,
644 		 * but should either do a listen or a connect soon.
645 		 */
646 		if (inp == 0) {
647 			++tcpstat.tcps_noport;
648 			goto dropwithreset_ratelim;
649 		}
650 	}
651 
652 	/* Check the minimum TTL for socket. */
653 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
654 		goto drop;
655 
656 	tp = intotcpcb(inp);
657 	if (tp == 0)
658 		goto dropwithreset_ratelim;
659 	if (tp->t_state == TCPS_CLOSED)
660 		goto drop;
661 
662 	/* Unscale the window into a 32-bit value. */
663 	if ((tiflags & TH_SYN) == 0)
664 		tiwin = th->th_win << tp->snd_scale;
665 	else
666 		tiwin = th->th_win;
667 
668 	so = inp->inp_socket;
669 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
670 		union syn_cache_sa src;
671 		union syn_cache_sa dst;
672 
673 		bzero(&src, sizeof(src));
674 		bzero(&dst, sizeof(dst));
675 		switch (af) {
676 #ifdef INET
677 		case AF_INET:
678 			src.sin.sin_len = sizeof(struct sockaddr_in);
679 			src.sin.sin_family = AF_INET;
680 			src.sin.sin_addr = ip->ip_src;
681 			src.sin.sin_port = th->th_sport;
682 
683 			dst.sin.sin_len = sizeof(struct sockaddr_in);
684 			dst.sin.sin_family = AF_INET;
685 			dst.sin.sin_addr = ip->ip_dst;
686 			dst.sin.sin_port = th->th_dport;
687 			break;
688 #endif
689 #ifdef INET6
690 		case AF_INET6:
691 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
692 			src.sin6.sin6_family = AF_INET6;
693 			src.sin6.sin6_addr = ip6->ip6_src;
694 			src.sin6.sin6_port = th->th_sport;
695 
696 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
697 			dst.sin6.sin6_family = AF_INET6;
698 			dst.sin6.sin6_addr = ip6->ip6_dst;
699 			dst.sin6.sin6_port = th->th_dport;
700 			break;
701 #endif /* INET6 */
702 		default:
703 			goto badsyn;	/*sanity*/
704 		}
705 
706 		if (so->so_options & SO_DEBUG) {
707 			ostate = tp->t_state;
708 			switch (af) {
709 #ifdef INET6
710 			case AF_INET6:
711 				bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6));
712 				bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th));
713 				break;
714 #endif
715 			case AF_INET:
716 				bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip));
717 				bcopy(th, &tcp_saveti.ti_t, sizeof(*th));
718 				break;
719 			}
720 		}
721 		if (so->so_options & SO_ACCEPTCONN) {
722 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
723 				if (tiflags & TH_RST) {
724 					syn_cache_reset(&src.sa, &dst.sa, th);
725 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
726 				    (TH_ACK|TH_SYN)) {
727 					/*
728 					 * Received a SYN,ACK.  This should
729 					 * never happen while we are in
730 					 * LISTEN.  Send an RST.
731 					 */
732 					goto badsyn;
733 				} else if (tiflags & TH_ACK) {
734 					so = syn_cache_get(&src.sa, &dst.sa,
735 						th, iphlen, tlen, so, m);
736 					if (so == NULL) {
737 						/*
738 						 * We don't have a SYN for
739 						 * this ACK; send an RST.
740 						 */
741 						goto badsyn;
742 					} else if (so ==
743 					    (struct socket *)(-1)) {
744 						/*
745 						 * We were unable to create
746 						 * the connection.  If the
747 						 * 3-way handshake was
748 						 * completed, and RST has
749 						 * been sent to the peer.
750 						 * Since the mbuf might be
751 						 * in use for the reply,
752 						 * do not free it.
753 						 */
754 						m = NULL;
755 					} else {
756 						/*
757 						 * We have created a
758 						 * full-blown connection.
759 						 */
760 						tp = NULL;
761 						inp = (struct inpcb *)so->so_pcb;
762 						tp = intotcpcb(inp);
763 						if (tp == NULL)
764 							goto badsyn;	/*XXX*/
765 
766 						/*
767 						 * Compute proper scaling
768 						 * value from buffer space
769 						 */
770 						tcp_rscale(tp, so->so_rcv.sb_hiwat);
771 						goto after_listen;
772 					}
773 				} else {
774 					/*
775 					 * None of RST, SYN or ACK was set.
776 					 * This is an invalid packet for a
777 					 * TCB in LISTEN state.  Send a RST.
778 					 */
779 					goto badsyn;
780 				}
781 			} else {
782 				/*
783 				 * Received a SYN.
784 				 */
785 #ifdef INET6
786 				/*
787 				 * If deprecated address is forbidden, we do
788 				 * not accept SYN to deprecated interface
789 				 * address to prevent any new inbound
790 				 * connection from getting established.
791 				 * When we do not accept SYN, we send a TCP
792 				 * RST, with deprecated source address (instead
793 				 * of dropping it).  We compromise it as it is
794 				 * much better for peer to send a RST, and
795 				 * RST will be the final packet for the
796 				 * exchange.
797 				 *
798 				 * If we do not forbid deprecated addresses, we
799 				 * accept the SYN packet.  RFC2462 does not
800 				 * suggest dropping SYN in this case.
801 				 * If we decipher RFC2462 5.5.4, it says like
802 				 * this:
803 				 * 1. use of deprecated addr with existing
804 				 *    communication is okay - "SHOULD continue
805 				 *    to be used"
806 				 * 2. use of it with new communication:
807 				 *   (2a) "SHOULD NOT be used if alternate
808 				 *        address with sufficient scope is
809 				 *        available"
810 				 *   (2b) nothing mentioned otherwise.
811 				 * Here we fall into (2b) case as we have no
812 				 * choice in our source address selection - we
813 				 * must obey the peer.
814 				 *
815 				 * The wording in RFC2462 is confusing, and
816 				 * there are multiple description text for
817 				 * deprecated address handling - worse, they
818 				 * are not exactly the same.  I believe 5.5.4
819 				 * is the best one, so we follow 5.5.4.
820 				 */
821 				if (ip6 && !ip6_use_deprecated) {
822 					struct in6_ifaddr *ia6;
823 
824 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
825 					    &ip6->ip6_dst)) &&
826 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
827 						tp = NULL;
828 						goto dropwithreset;
829 					}
830 				}
831 #endif
832 
833 				/*
834 				 * LISTEN socket received a SYN
835 				 * from itself?  This can't possibly
836 				 * be valid; drop the packet.
837 				 */
838 				if (th->th_dport == th->th_sport) {
839 					switch (af) {
840 #ifdef INET6
841 					case AF_INET6:
842 						if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src,
843 						    &ip6->ip6_dst)) {
844 							tcpstat.tcps_badsyn++;
845 							goto drop;
846 						}
847 						break;
848 #endif /* INET6 */
849 					case AF_INET:
850 						if (ip->ip_dst.s_addr == ip->ip_src.s_addr) {
851 							tcpstat.tcps_badsyn++;
852 							goto drop;
853 						}
854 						break;
855 					}
856 				}
857 
858 				/*
859 				 * SYN looks ok; create compressed TCP
860 				 * state for it.
861 				 */
862 				if (so->so_qlen <= so->so_qlimit &&
863 				    syn_cache_add(&src.sa, &dst.sa, th, iphlen,
864 				    so, m, optp, optlen, &opti, reuse))
865 					m = NULL;
866 			}
867 			goto drop;
868 		}
869 	}
870 
871 after_listen:
872 #ifdef DIAGNOSTIC
873 	/*
874 	 * Should not happen now that all embryonic connections
875 	 * are handled with compressed state.
876 	 */
877 	if (tp->t_state == TCPS_LISTEN)
878 		panic("tcp_input: TCPS_LISTEN");
879 #endif
880 
881 #if NPF > 0
882 		if (m->m_pkthdr.pf.statekey) {
883 			((struct pf_state_key *)m->m_pkthdr.pf.statekey)->inp =
884 			    inp;
885 			inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
886 		}
887 #endif
888 
889 #ifdef IPSEC
890 	/* Find most recent IPsec tag */
891 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
892         s = splnet();
893 	if (mtag != NULL) {
894 		tdbi = (struct tdb_ident *)(mtag + 1);
895 	        tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto);
896 	} else
897 		tdb = NULL;
898 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
899 	    tdb, inp);
900 	if (error) {
901 		splx(s);
902 		goto drop;
903 	}
904 
905 	/* Latch SA */
906 	if (inp->inp_tdb_in != tdb) {
907 		if (tdb) {
908 		        tdb_add_inp(tdb, inp, 1);
909 			if (inp->inp_ipo == NULL) {
910 				inp->inp_ipo = ipsec_add_policy(inp, af,
911 				    IPSP_DIRECTION_OUT);
912 				if (inp->inp_ipo == NULL) {
913 					splx(s);
914 					goto drop;
915 				}
916 			}
917 			if (inp->inp_ipo->ipo_dstid == NULL &&
918 			    tdb->tdb_srcid != NULL) {
919 				inp->inp_ipo->ipo_dstid = tdb->tdb_srcid;
920 				tdb->tdb_srcid->ref_count++;
921 			}
922 			if (inp->inp_ipsec_remotecred == NULL &&
923 			    tdb->tdb_remote_cred != NULL) {
924 				inp->inp_ipsec_remotecred =
925 				    tdb->tdb_remote_cred;
926 				tdb->tdb_remote_cred->ref_count++;
927 			}
928 			if (inp->inp_ipsec_remoteauth == NULL &&
929 			    tdb->tdb_remote_auth != NULL) {
930 				inp->inp_ipsec_remoteauth =
931 				    tdb->tdb_remote_auth;
932 				tdb->tdb_remote_auth->ref_count++;
933 			}
934 		} else { /* Just reset */
935 		        TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp,
936 				     inp_tdb_in_next);
937 			inp->inp_tdb_in = NULL;
938 		}
939 	}
940         splx(s);
941 #endif /* IPSEC */
942 
943 	/*
944 	 * Segment received on connection.
945 	 * Reset idle time and keep-alive timer.
946 	 */
947 	tp->t_rcvtime = tcp_now;
948 	if (TCPS_HAVEESTABLISHED(tp->t_state))
949 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
950 
951 #ifdef TCP_SACK
952 	if (tp->sack_enable)
953 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
954 #endif /* TCP_SACK */
955 
956 	/*
957 	 * Process options.
958 	 */
959 #ifdef TCP_SIGNATURE
960 	if (optp || (tp->t_flags & TF_SIGNATURE))
961 #else
962 	if (optp)
963 #endif
964 		if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti))
965 			goto drop;
966 
967 	if (opti.ts_present && opti.ts_ecr) {
968 		int rtt_test;
969 
970 		/* subtract out the tcp timestamp modulator */
971 		opti.ts_ecr -= tp->ts_modulate;
972 
973 		/* make sure ts_ecr is sensible */
974 		rtt_test = tcp_now - opti.ts_ecr;
975 		if (rtt_test < 0 || rtt_test > TCP_RTT_MAX)
976 			opti.ts_ecr = 0;
977 	}
978 
979 #ifdef TCP_ECN
980 	/* if congestion experienced, set ECE bit in subsequent packets. */
981 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
982 		tp->t_flags |= TF_RCVD_CE;
983 		tcpstat.tcps_ecn_rcvce++;
984 	}
985 #endif
986 	/*
987 	 * Header prediction: check for the two common cases
988 	 * of a uni-directional data xfer.  If the packet has
989 	 * no control flags, is in-sequence, the window didn't
990 	 * change and we're not retransmitting, it's a
991 	 * candidate.  If the length is zero and the ack moved
992 	 * forward, we're the sender side of the xfer.  Just
993 	 * free the data acked & wake any higher level process
994 	 * that was blocked waiting for space.  If the length
995 	 * is non-zero and the ack didn't move, we're the
996 	 * receiver side.  If we're getting packets in-order
997 	 * (the reassembly queue is empty), add the data to
998 	 * the socket buffer and note that we need a delayed ack.
999 	 */
1000 	if (tp->t_state == TCPS_ESTABLISHED &&
1001 #ifdef TCP_ECN
1002 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
1003 #else
1004 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1005 #endif
1006 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1007 	    th->th_seq == tp->rcv_nxt &&
1008 	    tiwin && tiwin == tp->snd_wnd &&
1009 	    tp->snd_nxt == tp->snd_max) {
1010 
1011 		/*
1012 		 * If last ACK falls within this segment's sequence numbers,
1013 		 *  record the timestamp.
1014 		 * Fix from Braden, see Stevens p. 870
1015 		 */
1016 		if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1017 			tp->ts_recent_age = tcp_now;
1018 			tp->ts_recent = opti.ts_val;
1019 		}
1020 
1021 		if (tlen == 0) {
1022 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1023 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1024 			    tp->snd_cwnd >= tp->snd_wnd &&
1025 			    tp->t_dupacks == 0) {
1026 				/*
1027 				 * this is a pure ack for outstanding data.
1028 				 */
1029 				++tcpstat.tcps_predack;
1030 				if (opti.ts_present && opti.ts_ecr)
1031 					tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1032 				else if (tp->t_rtttime &&
1033 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1034 					tcp_xmit_timer(tp,
1035 					    tcp_now - tp->t_rtttime);
1036 				acked = th->th_ack - tp->snd_una;
1037 				tcpstat.tcps_rcvackpack++;
1038 				tcpstat.tcps_rcvackbyte += acked;
1039 				ND6_HINT(tp);
1040 				sbdrop(&so->so_snd, acked);
1041 
1042 				/*
1043 				 * If we had a pending ICMP message that
1044 				 * referres to data that have just been
1045 				 * acknowledged, disregard the recorded ICMP
1046 				 * message.
1047 				 */
1048 				if ((tp->t_flags & TF_PMTUD_PEND) &&
1049 				    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1050 					tp->t_flags &= ~TF_PMTUD_PEND;
1051 
1052 				/*
1053 				 * Keep track of the largest chunk of data
1054 				 * acknowledged since last PMTU update
1055 				 */
1056 				if (tp->t_pmtud_mss_acked < acked)
1057 					tp->t_pmtud_mss_acked = acked;
1058 
1059 				tp->snd_una = th->th_ack;
1060 #if defined(TCP_SACK) || defined(TCP_ECN)
1061 				/*
1062 				 * We want snd_last to track snd_una so
1063 				 * as to avoid sequence wraparound problems
1064 				 * for very large transfers.
1065 				 */
1066 #ifdef TCP_ECN
1067 				if (SEQ_GT(tp->snd_una, tp->snd_last))
1068 #endif
1069 				tp->snd_last = tp->snd_una;
1070 #endif /* TCP_SACK */
1071 #if defined(TCP_SACK) && defined(TCP_FACK)
1072 				tp->snd_fack = tp->snd_una;
1073 				tp->retran_data = 0;
1074 #endif /* TCP_FACK */
1075 				m_freem(m);
1076 
1077 				/*
1078 				 * If all outstanding data are acked, stop
1079 				 * retransmit timer, otherwise restart timer
1080 				 * using current (possibly backed-off) value.
1081 				 * If process is waiting for space,
1082 				 * wakeup/selwakeup/signal.  If data
1083 				 * are ready to send, let tcp_output
1084 				 * decide between more output or persist.
1085 				 */
1086 				if (tp->snd_una == tp->snd_max)
1087 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1088 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1089 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1090 
1091 				if (sb_notify(&so->so_snd))
1092 					sowwakeup(so);
1093 				if (so->so_snd.sb_cc)
1094 					(void) tcp_output(tp);
1095 				return;
1096 			}
1097 		} else if (th->th_ack == tp->snd_una &&
1098 		    TAILQ_EMPTY(&tp->t_segq) &&
1099 		    tlen <= sbspace(&so->so_rcv)) {
1100 			/*
1101 			 * This is a pure, in-sequence data packet
1102 			 * with nothing on the reassembly queue and
1103 			 * we have enough buffer space to take it.
1104 			 */
1105 #ifdef TCP_SACK
1106 			/* Clean receiver SACK report if present */
1107 			if (tp->sack_enable && tp->rcv_numsacks)
1108 				tcp_clean_sackreport(tp);
1109 #endif /* TCP_SACK */
1110 			++tcpstat.tcps_preddat;
1111 			tp->rcv_nxt += tlen;
1112 			tcpstat.tcps_rcvpack++;
1113 			tcpstat.tcps_rcvbyte += tlen;
1114 			ND6_HINT(tp);
1115 			/*
1116 			 * Drop TCP, IP headers and TCP options then add data
1117 			 * to socket buffer.
1118 			 */
1119 			if (so->so_state & SS_CANTRCVMORE)
1120 				m_freem(m);
1121 			else {
1122 				m_adj(m, iphlen + off);
1123 				sbappendstream(&so->so_rcv, m);
1124 			}
1125 			sorwakeup(so);
1126 			TCP_SETUP_ACK(tp, tiflags);
1127 			if (tp->t_flags & TF_ACKNOW)
1128 				(void) tcp_output(tp);
1129 			return;
1130 		}
1131 	}
1132 
1133 	/*
1134 	 * Compute mbuf offset to TCP data segment.
1135 	 */
1136 	hdroptlen = iphlen + off;
1137 
1138 	/*
1139 	 * Calculate amount of space in receive window,
1140 	 * and then do TCP input processing.
1141 	 * Receive window is amount of space in rcv queue,
1142 	 * but not less than advertised window.
1143 	 */
1144 	{ int win;
1145 
1146 	win = sbspace(&so->so_rcv);
1147 	if (win < 0)
1148 		win = 0;
1149 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1150 	}
1151 
1152 	switch (tp->t_state) {
1153 
1154 	/*
1155 	 * If the state is SYN_RECEIVED:
1156 	 * 	if seg contains SYN/ACK, send an RST.
1157 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1158 	 */
1159 
1160 	case TCPS_SYN_RECEIVED:
1161 		if (tiflags & TH_ACK) {
1162 			if (tiflags & TH_SYN) {
1163 				tcpstat.tcps_badsyn++;
1164 				goto dropwithreset;
1165 			}
1166 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1167 			    SEQ_GT(th->th_ack, tp->snd_max))
1168 				goto dropwithreset;
1169 		}
1170 		break;
1171 
1172 	/*
1173 	 * If the state is SYN_SENT:
1174 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1175 	 *	if seg contains a RST, then drop the connection.
1176 	 *	if seg does not contain SYN, then drop it.
1177 	 * Otherwise this is an acceptable SYN segment
1178 	 *	initialize tp->rcv_nxt and tp->irs
1179 	 *	if seg contains ack then advance tp->snd_una
1180 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1181 	 *	arrange for segment to be acked (eventually)
1182 	 *	continue processing rest of data/controls, beginning with URG
1183 	 */
1184 	case TCPS_SYN_SENT:
1185 		if ((tiflags & TH_ACK) &&
1186 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1187 		     SEQ_GT(th->th_ack, tp->snd_max)))
1188 			goto dropwithreset;
1189 		if (tiflags & TH_RST) {
1190 #ifdef TCP_ECN
1191 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1192 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1193 				goto drop;
1194 #endif
1195 			if (tiflags & TH_ACK)
1196 				tp = tcp_drop(tp, ECONNREFUSED);
1197 			goto drop;
1198 		}
1199 		if ((tiflags & TH_SYN) == 0)
1200 			goto drop;
1201 		if (tiflags & TH_ACK) {
1202 			tp->snd_una = th->th_ack;
1203 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1204 				tp->snd_nxt = tp->snd_una;
1205 		}
1206 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1207 		tp->irs = th->th_seq;
1208 		tcp_mss(tp, opti.maxseg);
1209 		/* Reset initial window to 1 segment for retransmit */
1210 		if (tp->t_rxtshift > 0)
1211 			tp->snd_cwnd = tp->t_maxseg;
1212 		tcp_rcvseqinit(tp);
1213 		tp->t_flags |= TF_ACKNOW;
1214 #ifdef TCP_SACK
1215                 /*
1216                  * If we've sent a SACK_PERMITTED option, and the peer
1217                  * also replied with one, then TF_SACK_PERMIT should have
1218                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1219                  */
1220 		if (tp->sack_enable)
1221 			tp->sack_enable = tp->t_flags & TF_SACK_PERMIT;
1222 #endif
1223 #ifdef TCP_ECN
1224 		/*
1225 		 * if ECE is set but CWR is not set for SYN-ACK, or
1226 		 * both ECE and CWR are set for simultaneous open,
1227 		 * peer is ECN capable.
1228 		 */
1229 		if (tcp_do_ecn) {
1230 			if ((tiflags & (TH_ACK|TH_ECE|TH_CWR))
1231 			    == (TH_ACK|TH_ECE) ||
1232 			    (tiflags & (TH_ACK|TH_ECE|TH_CWR))
1233 			    == (TH_ECE|TH_CWR)) {
1234 				tp->t_flags |= TF_ECN_PERMIT;
1235 				tiflags &= ~(TH_ECE|TH_CWR);
1236 				tcpstat.tcps_ecn_accepts++;
1237 			}
1238 		}
1239 #endif
1240 
1241 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1242 			tcpstat.tcps_connects++;
1243 			soisconnected(so);
1244 			tp->t_state = TCPS_ESTABLISHED;
1245 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1246 			/* Do window scaling on this connection? */
1247 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1248 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1249 				tp->snd_scale = tp->requested_s_scale;
1250 				tp->rcv_scale = tp->request_r_scale;
1251 			}
1252 			(void) tcp_reass(tp, (struct tcphdr *)0,
1253 				(struct mbuf *)0, &tlen);
1254 			/*
1255 			 * if we didn't have to retransmit the SYN,
1256 			 * use its rtt as our initial srtt & rtt var.
1257 			 */
1258 			if (tp->t_rtttime)
1259 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1260 			/*
1261 			 * Since new data was acked (the SYN), open the
1262 			 * congestion window by one MSS.  We do this
1263 			 * here, because we won't go through the normal
1264 			 * ACK processing below.  And since this is the
1265 			 * start of the connection, we know we are in
1266 			 * the exponential phase of slow-start.
1267 			 */
1268 			tp->snd_cwnd += tp->t_maxseg;
1269 		} else
1270 			tp->t_state = TCPS_SYN_RECEIVED;
1271 
1272 #if 0
1273 trimthenstep6:
1274 #endif
1275 		/*
1276 		 * Advance th->th_seq to correspond to first data byte.
1277 		 * If data, trim to stay within window,
1278 		 * dropping FIN if necessary.
1279 		 */
1280 		th->th_seq++;
1281 		if (tlen > tp->rcv_wnd) {
1282 			todrop = tlen - tp->rcv_wnd;
1283 			m_adj(m, -todrop);
1284 			tlen = tp->rcv_wnd;
1285 			tiflags &= ~TH_FIN;
1286 			tcpstat.tcps_rcvpackafterwin++;
1287 			tcpstat.tcps_rcvbyteafterwin += todrop;
1288 		}
1289 		tp->snd_wl1 = th->th_seq - 1;
1290 		tp->rcv_up = th->th_seq;
1291 		goto step6;
1292 	/*
1293 	 * If a new connection request is received while in TIME_WAIT,
1294 	 * drop the old connection and start over if the if the
1295 	 * timestamp or the sequence numbers are above the previous
1296 	 * ones.
1297 	 */
1298 	case TCPS_TIME_WAIT:
1299 		if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) &&
1300 		    ((opti.ts_present &&
1301 		    TSTMP_LT(tp->ts_recent, opti.ts_val)) ||
1302 		    SEQ_GT(th->th_seq, tp->rcv_nxt))) {
1303 			/*
1304 			* Advance the iss by at least 32768, but
1305 			* clear the msb in order to make sure
1306 			* that SEG_LT(snd_nxt, iss).
1307 			*/
1308 			iss = tp->snd_nxt +
1309 			    ((arc4random() & 0x7fffffff) | 0x8000);
1310 			reuse = &iss;
1311 			tp = tcp_close(tp);
1312 			inp = NULL;
1313 			goto findpcb;
1314 		}
1315 	}
1316 
1317 	/*
1318 	 * States other than LISTEN or SYN_SENT.
1319 	 * First check timestamp, if present.
1320 	 * Then check that at least some bytes of segment are within
1321 	 * receive window.  If segment begins before rcv_nxt,
1322 	 * drop leading data (and SYN); if nothing left, just ack.
1323 	 *
1324 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1325 	 * and it's less than opti.ts_recent, drop it.
1326 	 */
1327 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1328 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1329 
1330 		/* Check to see if ts_recent is over 24 days old.  */
1331 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1332 			/*
1333 			 * Invalidate ts_recent.  If this segment updates
1334 			 * ts_recent, the age will be reset later and ts_recent
1335 			 * will get a valid value.  If it does not, setting
1336 			 * ts_recent to zero will at least satisfy the
1337 			 * requirement that zero be placed in the timestamp
1338 			 * echo reply when ts_recent isn't valid.  The
1339 			 * age isn't reset until we get a valid ts_recent
1340 			 * because we don't want out-of-order segments to be
1341 			 * dropped when ts_recent is old.
1342 			 */
1343 			tp->ts_recent = 0;
1344 		} else {
1345 			tcpstat.tcps_rcvduppack++;
1346 			tcpstat.tcps_rcvdupbyte += tlen;
1347 			tcpstat.tcps_pawsdrop++;
1348 			goto dropafterack;
1349 		}
1350 	}
1351 
1352 	todrop = tp->rcv_nxt - th->th_seq;
1353 	if (todrop > 0) {
1354 		if (tiflags & TH_SYN) {
1355 			tiflags &= ~TH_SYN;
1356 			th->th_seq++;
1357 			if (th->th_urp > 1)
1358 				th->th_urp--;
1359 			else
1360 				tiflags &= ~TH_URG;
1361 			todrop--;
1362 		}
1363 		if (todrop > tlen ||
1364 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1365 			/*
1366 			 * Any valid FIN must be to the left of the
1367 			 * window.  At this point, FIN must be a
1368 			 * duplicate or out-of-sequence, so drop it.
1369 			 */
1370 			tiflags &= ~TH_FIN;
1371 			/*
1372 			 * Send ACK to resynchronize, and drop any data,
1373 			 * but keep on processing for RST or ACK.
1374 			 */
1375 			tp->t_flags |= TF_ACKNOW;
1376 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1377 			tcpstat.tcps_rcvduppack++;
1378 		} else {
1379 			tcpstat.tcps_rcvpartduppack++;
1380 			tcpstat.tcps_rcvpartdupbyte += todrop;
1381 		}
1382 		hdroptlen += todrop;	/* drop from head afterwards */
1383 		th->th_seq += todrop;
1384 		tlen -= todrop;
1385 		if (th->th_urp > todrop)
1386 			th->th_urp -= todrop;
1387 		else {
1388 			tiflags &= ~TH_URG;
1389 			th->th_urp = 0;
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * If new data are received on a connection after the
1395 	 * user processes are gone, then RST the other end.
1396 	 */
1397 	if ((so->so_state & SS_NOFDREF) &&
1398 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1399 		tp = tcp_close(tp);
1400 		tcpstat.tcps_rcvafterclose++;
1401 		goto dropwithreset;
1402 	}
1403 
1404 	/*
1405 	 * If segment ends after window, drop trailing data
1406 	 * (and PUSH and FIN); if nothing left, just ACK.
1407 	 */
1408 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1409 	if (todrop > 0) {
1410 		tcpstat.tcps_rcvpackafterwin++;
1411 		if (todrop >= tlen) {
1412 			tcpstat.tcps_rcvbyteafterwin += tlen;
1413 			/*
1414 			 * If window is closed can only take segments at
1415 			 * window edge, and have to drop data and PUSH from
1416 			 * incoming segments.  Continue processing, but
1417 			 * remember to ack.  Otherwise, drop segment
1418 			 * and ack.
1419 			 */
1420 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1421 				tp->t_flags |= TF_ACKNOW;
1422 				tcpstat.tcps_rcvwinprobe++;
1423 			} else
1424 				goto dropafterack;
1425 		} else
1426 			tcpstat.tcps_rcvbyteafterwin += todrop;
1427 		m_adj(m, -todrop);
1428 		tlen -= todrop;
1429 		tiflags &= ~(TH_PUSH|TH_FIN);
1430 	}
1431 
1432 	/*
1433 	 * If last ACK falls within this segment's sequence numbers,
1434 	 * record its timestamp if it's more recent.
1435 	 * Cf fix from Braden, see Stevens p. 870
1436 	 */
1437 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1438 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1439 		if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1440 		    ((tiflags & (TH_SYN|TH_FIN)) != 0)))
1441 			tp->ts_recent = opti.ts_val;
1442 		else
1443 			tp->ts_recent = 0;
1444 		tp->ts_recent_age = tcp_now;
1445 	}
1446 
1447 	/*
1448 	 * If the RST bit is set examine the state:
1449 	 *    SYN_RECEIVED STATE:
1450 	 *	If passive open, return to LISTEN state.
1451 	 *	If active open, inform user that connection was refused.
1452 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1453 	 *	Inform user that connection was reset, and close tcb.
1454 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1455 	 *	Close the tcb.
1456 	 */
1457 	if (tiflags & TH_RST) {
1458 		if (th->th_seq != tp->last_ack_sent &&
1459 		    th->th_seq != tp->rcv_nxt &&
1460 		    th->th_seq != (tp->rcv_nxt + 1))
1461 			goto drop;
1462 
1463 		switch (tp->t_state) {
1464 		case TCPS_SYN_RECEIVED:
1465 #ifdef TCP_ECN
1466 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1467 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1468 				goto drop;
1469 #endif
1470 			so->so_error = ECONNREFUSED;
1471 			goto close;
1472 
1473 		case TCPS_ESTABLISHED:
1474 		case TCPS_FIN_WAIT_1:
1475 		case TCPS_FIN_WAIT_2:
1476 		case TCPS_CLOSE_WAIT:
1477 			so->so_error = ECONNRESET;
1478 		close:
1479 			tp->t_state = TCPS_CLOSED;
1480 			tcpstat.tcps_drops++;
1481 			tp = tcp_close(tp);
1482 			goto drop;
1483 		case TCPS_CLOSING:
1484 		case TCPS_LAST_ACK:
1485 		case TCPS_TIME_WAIT:
1486 			tp = tcp_close(tp);
1487 			goto drop;
1488 		}
1489 	}
1490 
1491 	/*
1492 	 * If a SYN is in the window, then this is an
1493 	 * error and we ACK and drop the packet.
1494 	 */
1495 	if (tiflags & TH_SYN)
1496 		goto dropafterack_ratelim;
1497 
1498 	/*
1499 	 * If the ACK bit is off we drop the segment and return.
1500 	 */
1501 	if ((tiflags & TH_ACK) == 0) {
1502 		if (tp->t_flags & TF_ACKNOW)
1503 			goto dropafterack;
1504 		else
1505 			goto drop;
1506 	}
1507 
1508 	/*
1509 	 * Ack processing.
1510 	 */
1511 	switch (tp->t_state) {
1512 
1513 	/*
1514 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1515 	 * ESTABLISHED state and continue processing.
1516 	 * The ACK was checked above.
1517 	 */
1518 	case TCPS_SYN_RECEIVED:
1519 		tcpstat.tcps_connects++;
1520 		soisconnected(so);
1521 		tp->t_state = TCPS_ESTABLISHED;
1522 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1523 		/* Do window scaling? */
1524 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1525 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1526 			tp->snd_scale = tp->requested_s_scale;
1527 			tp->rcv_scale = tp->request_r_scale;
1528 			tiwin = th->th_win << tp->snd_scale;
1529 		}
1530 		(void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0,
1531 				 &tlen);
1532 		tp->snd_wl1 = th->th_seq - 1;
1533 		/* fall into ... */
1534 
1535 	/*
1536 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1537 	 * ACKs.  If the ack is in the range
1538 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1539 	 * then advance tp->snd_una to th->th_ack and drop
1540 	 * data from the retransmission queue.  If this ACK reflects
1541 	 * more up to date window information we update our window information.
1542 	 */
1543 	case TCPS_ESTABLISHED:
1544 	case TCPS_FIN_WAIT_1:
1545 	case TCPS_FIN_WAIT_2:
1546 	case TCPS_CLOSE_WAIT:
1547 	case TCPS_CLOSING:
1548 	case TCPS_LAST_ACK:
1549 	case TCPS_TIME_WAIT:
1550 #ifdef TCP_ECN
1551 		/*
1552 		 * if we receive ECE and are not already in recovery phase,
1553 		 * reduce cwnd by half but don't slow-start.
1554 		 * advance snd_last to snd_max not to reduce cwnd again
1555 		 * until all outstanding packets are acked.
1556 		 */
1557 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1558 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1559 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1560 				u_int win;
1561 
1562 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1563 				if (win > 1) {
1564 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1565 					tp->snd_cwnd = tp->snd_ssthresh;
1566 					tp->snd_last = tp->snd_max;
1567 					tp->t_flags |= TF_SEND_CWR;
1568 					tcpstat.tcps_cwr_ecn++;
1569 				}
1570 			}
1571 			tcpstat.tcps_ecn_rcvece++;
1572 		}
1573 		/*
1574 		 * if we receive CWR, we know that the peer has reduced
1575 		 * its congestion window.  stop sending ecn-echo.
1576 		 */
1577 		if ((tiflags & TH_CWR)) {
1578 			tp->t_flags &= ~TF_RCVD_CE;
1579 			tcpstat.tcps_ecn_rcvcwr++;
1580 		}
1581 #endif /* TCP_ECN */
1582 
1583 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1584 			/*
1585 			 * Duplicate/old ACK processing.
1586 			 * Increments t_dupacks:
1587 			 *	Pure duplicate (same seq/ack/window, no data)
1588 			 * Doesn't affect t_dupacks:
1589 			 *	Data packets.
1590 			 *	Normal window updates (window opens)
1591 			 * Resets t_dupacks:
1592 			 *	New data ACKed.
1593 			 *	Window shrinks
1594 			 *	Old ACK
1595 			 */
1596 			if (tlen) {
1597 				/* Drop very old ACKs unless th_seq matches */
1598 				if (th->th_seq != tp->rcv_nxt &&
1599 				   SEQ_LT(th->th_ack,
1600 				   tp->snd_una - tp->max_sndwnd)) {
1601 					tcpstat.tcps_rcvacktooold++;
1602 					goto drop;
1603 				}
1604 				break;
1605 			}
1606 			/*
1607 			 * If we get an old ACK, there is probably packet
1608 			 * reordering going on.  Be conservative and reset
1609 			 * t_dupacks so that we are less aggressive in
1610 			 * doing a fast retransmit.
1611 			 */
1612 			if (th->th_ack != tp->snd_una) {
1613 				tp->t_dupacks = 0;
1614 				break;
1615 			}
1616 			if (tiwin == tp->snd_wnd) {
1617 				tcpstat.tcps_rcvdupack++;
1618 				/*
1619 				 * If we have outstanding data (other than
1620 				 * a window probe), this is a completely
1621 				 * duplicate ack (ie, window info didn't
1622 				 * change), the ack is the biggest we've
1623 				 * seen and we've seen exactly our rexmt
1624 				 * threshold of them, assume a packet
1625 				 * has been dropped and retransmit it.
1626 				 * Kludge snd_nxt & the congestion
1627 				 * window so we send only this one
1628 				 * packet.
1629 				 *
1630 				 * We know we're losing at the current
1631 				 * window size so do congestion avoidance
1632 				 * (set ssthresh to half the current window
1633 				 * and pull our congestion window back to
1634 				 * the new ssthresh).
1635 				 *
1636 				 * Dup acks mean that packets have left the
1637 				 * network (they're now cached at the receiver)
1638 				 * so bump cwnd by the amount in the receiver
1639 				 * to keep a constant cwnd packets in the
1640 				 * network.
1641 				 */
1642 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1643 					tp->t_dupacks = 0;
1644 #if defined(TCP_SACK) && defined(TCP_FACK)
1645 				/*
1646 				 * In FACK, can enter fast rec. if the receiver
1647 				 * reports a reass. queue longer than 3 segs.
1648 				 */
1649 				else if (++tp->t_dupacks == tcprexmtthresh ||
1650 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1651 				    tp->t_maxseg + tp->snd_una)) &&
1652 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1653 #else
1654 				else if (++tp->t_dupacks == tcprexmtthresh) {
1655 #endif /* TCP_FACK */
1656 					tcp_seq onxt = tp->snd_nxt;
1657 					u_long win =
1658 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1659 						2 / tp->t_maxseg;
1660 
1661 #if defined(TCP_SACK) || defined(TCP_ECN)
1662 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1663 					    	/*
1664 						 * False fast retx after
1665 						 * timeout.  Do not cut window.
1666 						 */
1667 						tp->t_dupacks = 0;
1668 						goto drop;
1669 					}
1670 #endif
1671 					if (win < 2)
1672 						win = 2;
1673 					tp->snd_ssthresh = win * tp->t_maxseg;
1674 #ifdef TCP_SACK
1675 					tp->snd_last = tp->snd_max;
1676                     			if (tp->sack_enable) {
1677 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1678 						tp->t_rtttime = 0;
1679 #ifdef TCP_ECN
1680 						tp->t_flags |= TF_SEND_CWR;
1681 #endif
1682 						tcpstat.tcps_cwr_frecovery++;
1683 						tcpstat.tcps_sack_recovery_episode++;
1684 #if defined(TCP_SACK) && defined(TCP_FACK)
1685 						tp->t_dupacks = tcprexmtthresh;
1686 						(void) tcp_output(tp);
1687 						/*
1688 						 * During FR, snd_cwnd is held
1689 						 * constant for FACK.
1690 						 */
1691 						tp->snd_cwnd = tp->snd_ssthresh;
1692 #else
1693 						/*
1694 						 * tcp_output() will send
1695 						 * oldest SACK-eligible rtx.
1696 						 */
1697 						(void) tcp_output(tp);
1698 						tp->snd_cwnd = tp->snd_ssthresh+
1699 					           tp->t_maxseg * tp->t_dupacks;
1700 #endif /* TCP_FACK */
1701 						goto drop;
1702 					}
1703 #endif /* TCP_SACK */
1704 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1705 					tp->t_rtttime = 0;
1706 					tp->snd_nxt = th->th_ack;
1707 					tp->snd_cwnd = tp->t_maxseg;
1708 #ifdef TCP_ECN
1709 					tp->t_flags |= TF_SEND_CWR;
1710 #endif
1711 					tcpstat.tcps_cwr_frecovery++;
1712 					tcpstat.tcps_sndrexmitfast++;
1713 					(void) tcp_output(tp);
1714 
1715 					tp->snd_cwnd = tp->snd_ssthresh +
1716 					    tp->t_maxseg * tp->t_dupacks;
1717 					if (SEQ_GT(onxt, tp->snd_nxt))
1718 						tp->snd_nxt = onxt;
1719 					goto drop;
1720 				} else if (tp->t_dupacks > tcprexmtthresh) {
1721 #if defined(TCP_SACK) && defined(TCP_FACK)
1722 					/*
1723 					 * while (awnd < cwnd)
1724 					 *         sendsomething();
1725 					 */
1726 					if (tp->sack_enable) {
1727 						if (tp->snd_awnd < tp->snd_cwnd)
1728 							tcp_output(tp);
1729 						goto drop;
1730 					}
1731 #endif /* TCP_FACK */
1732 					tp->snd_cwnd += tp->t_maxseg;
1733 					(void) tcp_output(tp);
1734 					goto drop;
1735 				}
1736 			} else if (tiwin < tp->snd_wnd) {
1737 				/*
1738 				 * The window was retracted!  Previous dup
1739 				 * ACKs may have been due to packets arriving
1740 				 * after the shrunken window, not a missing
1741 				 * packet, so play it safe and reset t_dupacks
1742 				 */
1743 				tp->t_dupacks = 0;
1744 			}
1745 			break;
1746 		}
1747 		/*
1748 		 * If the congestion window was inflated to account
1749 		 * for the other side's cached packets, retract it.
1750 		 */
1751 #if defined(TCP_SACK)
1752 		if (tp->sack_enable) {
1753 			if (tp->t_dupacks >= tcprexmtthresh) {
1754 				/* Check for a partial ACK */
1755 				if (tcp_sack_partialack(tp, th)) {
1756 #if defined(TCP_SACK) && defined(TCP_FACK)
1757 					/* Force call to tcp_output */
1758 					if (tp->snd_awnd < tp->snd_cwnd)
1759 						needoutput = 1;
1760 #else
1761 					tp->snd_cwnd += tp->t_maxseg;
1762 					needoutput = 1;
1763 #endif /* TCP_FACK */
1764 				} else {
1765 					/* Out of fast recovery */
1766 					tp->snd_cwnd = tp->snd_ssthresh;
1767 					if (tcp_seq_subtract(tp->snd_max,
1768 					    th->th_ack) < tp->snd_ssthresh)
1769 						tp->snd_cwnd =
1770 						   tcp_seq_subtract(tp->snd_max,
1771 					           th->th_ack);
1772 					tp->t_dupacks = 0;
1773 #if defined(TCP_SACK) && defined(TCP_FACK)
1774 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1775 						tp->snd_fack = th->th_ack;
1776 #endif /* TCP_FACK */
1777 				}
1778 			}
1779 		} else {
1780 			if (tp->t_dupacks >= tcprexmtthresh &&
1781 			    !tcp_newreno(tp, th)) {
1782 				/* Out of fast recovery */
1783 				tp->snd_cwnd = tp->snd_ssthresh;
1784 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1785 			  	    tp->snd_ssthresh)
1786 					tp->snd_cwnd =
1787 					    tcp_seq_subtract(tp->snd_max,
1788 					    th->th_ack);
1789 				tp->t_dupacks = 0;
1790 			}
1791 		}
1792 		if (tp->t_dupacks < tcprexmtthresh)
1793 			tp->t_dupacks = 0;
1794 #else /* else no TCP_SACK */
1795 		if (tp->t_dupacks >= tcprexmtthresh &&
1796 		    tp->snd_cwnd > tp->snd_ssthresh)
1797 			tp->snd_cwnd = tp->snd_ssthresh;
1798 		tp->t_dupacks = 0;
1799 #endif
1800 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1801 			tcpstat.tcps_rcvacktoomuch++;
1802 			goto dropafterack_ratelim;
1803 		}
1804 		acked = th->th_ack - tp->snd_una;
1805 		tcpstat.tcps_rcvackpack++;
1806 		tcpstat.tcps_rcvackbyte += acked;
1807 
1808 		/*
1809 		 * If we have a timestamp reply, update smoothed
1810 		 * round trip time.  If no timestamp is present but
1811 		 * transmit timer is running and timed sequence
1812 		 * number was acked, update smoothed round trip time.
1813 		 * Since we now have an rtt measurement, cancel the
1814 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1815 		 * Recompute the initial retransmit timer.
1816 		 */
1817 		if (opti.ts_present && opti.ts_ecr)
1818 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1819 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1820 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1821 
1822 		/*
1823 		 * If all outstanding data is acked, stop retransmit
1824 		 * timer and remember to restart (more output or persist).
1825 		 * If there is more data to be acked, restart retransmit
1826 		 * timer, using current (possibly backed-off) value.
1827 		 */
1828 		if (th->th_ack == tp->snd_max) {
1829 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1830 			needoutput = 1;
1831 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1832 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1833 		/*
1834 		 * When new data is acked, open the congestion window.
1835 		 * If the window gives us less than ssthresh packets
1836 		 * in flight, open exponentially (maxseg per packet).
1837 		 * Otherwise open linearly: maxseg per window
1838 		 * (maxseg^2 / cwnd per packet).
1839 		 */
1840 		{
1841 		u_int cw = tp->snd_cwnd;
1842 		u_int incr = tp->t_maxseg;
1843 
1844 		if (cw > tp->snd_ssthresh)
1845 			incr = incr * incr / cw;
1846 #if defined (TCP_SACK)
1847 		if (tp->t_dupacks < tcprexmtthresh)
1848 #endif
1849 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1850 		}
1851 		ND6_HINT(tp);
1852 		if (acked > so->so_snd.sb_cc) {
1853 			tp->snd_wnd -= so->so_snd.sb_cc;
1854 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1855 			ourfinisacked = 1;
1856 		} else {
1857 			sbdrop(&so->so_snd, acked);
1858 			tp->snd_wnd -= acked;
1859 			ourfinisacked = 0;
1860 		}
1861 		if (sb_notify(&so->so_snd))
1862 			sowwakeup(so);
1863 
1864 		/*
1865 		 * If we had a pending ICMP message that referred to data
1866 		 * that have just been acknowledged, disregard the recorded
1867 		 * ICMP message.
1868 		 */
1869 		if ((tp->t_flags & TF_PMTUD_PEND) &&
1870 		    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1871 			tp->t_flags &= ~TF_PMTUD_PEND;
1872 
1873 		/*
1874 		 * Keep track of the largest chunk of data acknowledged
1875 		 * since last PMTU update
1876 		 */
1877 		if (tp->t_pmtud_mss_acked < acked)
1878 			tp->t_pmtud_mss_acked = acked;
1879 
1880 		tp->snd_una = th->th_ack;
1881 #ifdef TCP_ECN
1882 		/* sync snd_last with snd_una */
1883 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1884 			tp->snd_last = tp->snd_una;
1885 #endif
1886 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1887 			tp->snd_nxt = tp->snd_una;
1888 #if defined (TCP_SACK) && defined (TCP_FACK)
1889 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1890 			tp->snd_fack = tp->snd_una;
1891 			/* Update snd_awnd for partial ACK
1892 			 * without any SACK blocks.
1893 			 */
1894 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1895 				tp->snd_fack) + tp->retran_data;
1896 		}
1897 #endif
1898 
1899 		switch (tp->t_state) {
1900 
1901 		/*
1902 		 * In FIN_WAIT_1 STATE in addition to the processing
1903 		 * for the ESTABLISHED state if our FIN is now acknowledged
1904 		 * then enter FIN_WAIT_2.
1905 		 */
1906 		case TCPS_FIN_WAIT_1:
1907 			if (ourfinisacked) {
1908 				/*
1909 				 * If we can't receive any more
1910 				 * data, then closing user can proceed.
1911 				 * Starting the timer is contrary to the
1912 				 * specification, but if we don't get a FIN
1913 				 * we'll hang forever.
1914 				 */
1915 				if (so->so_state & SS_CANTRCVMORE) {
1916 					soisdisconnected(so);
1917 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1918 				}
1919 				tp->t_state = TCPS_FIN_WAIT_2;
1920 			}
1921 			break;
1922 
1923 		/*
1924 		 * In CLOSING STATE in addition to the processing for
1925 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1926 		 * then enter the TIME-WAIT state, otherwise ignore
1927 		 * the segment.
1928 		 */
1929 		case TCPS_CLOSING:
1930 			if (ourfinisacked) {
1931 				tp->t_state = TCPS_TIME_WAIT;
1932 				tcp_canceltimers(tp);
1933 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1934 				soisdisconnected(so);
1935 			}
1936 			break;
1937 
1938 		/*
1939 		 * In LAST_ACK, we may still be waiting for data to drain
1940 		 * and/or to be acked, as well as for the ack of our FIN.
1941 		 * If our FIN is now acknowledged, delete the TCB,
1942 		 * enter the closed state and return.
1943 		 */
1944 		case TCPS_LAST_ACK:
1945 			if (ourfinisacked) {
1946 				tp = tcp_close(tp);
1947 				goto drop;
1948 			}
1949 			break;
1950 
1951 		/*
1952 		 * In TIME_WAIT state the only thing that should arrive
1953 		 * is a retransmission of the remote FIN.  Acknowledge
1954 		 * it and restart the finack timer.
1955 		 */
1956 		case TCPS_TIME_WAIT:
1957 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1958 			goto dropafterack;
1959 		}
1960 	}
1961 
1962 step6:
1963 	/*
1964 	 * Update window information.
1965 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1966 	 */
1967 	if ((tiflags & TH_ACK) &&
1968 	    (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq &&
1969 	    (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1970 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1971 		/* keep track of pure window updates */
1972 		if (tlen == 0 &&
1973 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1974 			tcpstat.tcps_rcvwinupd++;
1975 		tp->snd_wnd = tiwin;
1976 		tp->snd_wl1 = th->th_seq;
1977 		tp->snd_wl2 = th->th_ack;
1978 		if (tp->snd_wnd > tp->max_sndwnd)
1979 			tp->max_sndwnd = tp->snd_wnd;
1980 		needoutput = 1;
1981 	}
1982 
1983 	/*
1984 	 * Process segments with URG.
1985 	 */
1986 	if ((tiflags & TH_URG) && th->th_urp &&
1987 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1988 		/*
1989 		 * This is a kludge, but if we receive and accept
1990 		 * random urgent pointers, we'll crash in
1991 		 * soreceive.  It's hard to imagine someone
1992 		 * actually wanting to send this much urgent data.
1993 		 */
1994 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
1995 			th->th_urp = 0;			/* XXX */
1996 			tiflags &= ~TH_URG;		/* XXX */
1997 			goto dodata;			/* XXX */
1998 		}
1999 		/*
2000 		 * If this segment advances the known urgent pointer,
2001 		 * then mark the data stream.  This should not happen
2002 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2003 		 * a FIN has been received from the remote side.
2004 		 * In these states we ignore the URG.
2005 		 *
2006 		 * According to RFC961 (Assigned Protocols),
2007 		 * the urgent pointer points to the last octet
2008 		 * of urgent data.  We continue, however,
2009 		 * to consider it to indicate the first octet
2010 		 * of data past the urgent section as the original
2011 		 * spec states (in one of two places).
2012 		 */
2013 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2014 			tp->rcv_up = th->th_seq + th->th_urp;
2015 			so->so_oobmark = so->so_rcv.sb_cc +
2016 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2017 			if (so->so_oobmark == 0)
2018 				so->so_state |= SS_RCVATMARK;
2019 			sohasoutofband(so);
2020 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2021 		}
2022 		/*
2023 		 * Remove out of band data so doesn't get presented to user.
2024 		 * This can happen independent of advancing the URG pointer,
2025 		 * but if two URG's are pending at once, some out-of-band
2026 		 * data may creep in... ick.
2027 		 */
2028 		if (th->th_urp <= (u_int16_t) tlen
2029 #ifdef SO_OOBINLINE
2030 		     && (so->so_options & SO_OOBINLINE) == 0
2031 #endif
2032 		     )
2033 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2034 	} else
2035 		/*
2036 		 * If no out of band data is expected,
2037 		 * pull receive urgent pointer along
2038 		 * with the receive window.
2039 		 */
2040 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2041 			tp->rcv_up = tp->rcv_nxt;
2042 dodata:							/* XXX */
2043 
2044 	/*
2045 	 * Process the segment text, merging it into the TCP sequencing queue,
2046 	 * and arranging for acknowledgment of receipt if necessary.
2047 	 * This process logically involves adjusting tp->rcv_wnd as data
2048 	 * is presented to the user (this happens in tcp_usrreq.c,
2049 	 * case PRU_RCVD).  If a FIN has already been received on this
2050 	 * connection then we just ignore the text.
2051 	 */
2052 	if ((tlen || (tiflags & TH_FIN)) &&
2053 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2054 #ifdef TCP_SACK
2055 		tcp_seq laststart = th->th_seq;
2056 		tcp_seq lastend = th->th_seq + tlen;
2057 #endif
2058 		if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) &&
2059 		    tp->t_state == TCPS_ESTABLISHED) {
2060 			TCP_SETUP_ACK(tp, tiflags);
2061 			tp->rcv_nxt += tlen;
2062 			tiflags = th->th_flags & TH_FIN;
2063 			tcpstat.tcps_rcvpack++;
2064 			tcpstat.tcps_rcvbyte += tlen;
2065 			ND6_HINT(tp);
2066 			if (so->so_state & SS_CANTRCVMORE)
2067 				m_freem(m);
2068 			else {
2069 				m_adj(m, hdroptlen);
2070 				sbappendstream(&so->so_rcv, m);
2071 			}
2072 			sorwakeup(so);
2073 		} else {
2074 			m_adj(m, hdroptlen);
2075 			tiflags = tcp_reass(tp, th, m, &tlen);
2076 			tp->t_flags |= TF_ACKNOW;
2077 		}
2078 #ifdef TCP_SACK
2079 		if (tp->sack_enable)
2080 			tcp_update_sack_list(tp, laststart, lastend);
2081 #endif
2082 
2083 		/*
2084 		 * variable len never referenced again in modern BSD,
2085 		 * so why bother computing it ??
2086 		 */
2087 #if 0
2088 		/*
2089 		 * Note the amount of data that peer has sent into
2090 		 * our window, in order to estimate the sender's
2091 		 * buffer size.
2092 		 */
2093 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2094 #endif /* 0 */
2095 	} else {
2096 		m_freem(m);
2097 		tiflags &= ~TH_FIN;
2098 	}
2099 
2100 	/*
2101 	 * If FIN is received ACK the FIN and let the user know
2102 	 * that the connection is closing.  Ignore a FIN received before
2103 	 * the connection is fully established.
2104 	 */
2105 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2106 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2107 			socantrcvmore(so);
2108 			tp->t_flags |= TF_ACKNOW;
2109 			tp->rcv_nxt++;
2110 		}
2111 		switch (tp->t_state) {
2112 
2113 		/*
2114 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2115 		 */
2116 		case TCPS_ESTABLISHED:
2117 			tp->t_state = TCPS_CLOSE_WAIT;
2118 			break;
2119 
2120 		/*
2121 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2122 		 * enter the CLOSING state.
2123 		 */
2124 		case TCPS_FIN_WAIT_1:
2125 			tp->t_state = TCPS_CLOSING;
2126 			break;
2127 
2128 		/*
2129 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2130 		 * starting the time-wait timer, turning off the other
2131 		 * standard timers.
2132 		 */
2133 		case TCPS_FIN_WAIT_2:
2134 			tp->t_state = TCPS_TIME_WAIT;
2135 			tcp_canceltimers(tp);
2136 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2137 			soisdisconnected(so);
2138 			break;
2139 
2140 		/*
2141 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2142 		 */
2143 		case TCPS_TIME_WAIT:
2144 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2145 			break;
2146 		}
2147 	}
2148 	if (so->so_options & SO_DEBUG) {
2149 		switch (tp->pf) {
2150 #ifdef INET6
2151 		case PF_INET6:
2152 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2153 			    0, tlen);
2154 			break;
2155 #endif /* INET6 */
2156 		case PF_INET:
2157 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2158 			    0, tlen);
2159 			break;
2160 		}
2161 	}
2162 
2163 	/*
2164 	 * Return any desired output.
2165 	 */
2166 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2167 		(void) tcp_output(tp);
2168 	}
2169 	return;
2170 
2171 badsyn:
2172 	/*
2173 	 * Received a bad SYN.  Increment counters and dropwithreset.
2174 	 */
2175 	tcpstat.tcps_badsyn++;
2176 	tp = NULL;
2177 	goto dropwithreset;
2178 
2179 dropafterack_ratelim:
2180 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2181 	    tcp_ackdrop_ppslim) == 0) {
2182 		/* XXX stat */
2183 		goto drop;
2184 	}
2185 	/* ...fall into dropafterack... */
2186 
2187 dropafterack:
2188 	/*
2189 	 * Generate an ACK dropping incoming segment if it occupies
2190 	 * sequence space, where the ACK reflects our state.
2191 	 */
2192 	if (tiflags & TH_RST)
2193 		goto drop;
2194 	m_freem(m);
2195 	tp->t_flags |= TF_ACKNOW;
2196 	(void) tcp_output(tp);
2197 	return;
2198 
2199 dropwithreset_ratelim:
2200 	/*
2201 	 * We may want to rate-limit RSTs in certain situations,
2202 	 * particularly if we are sending an RST in response to
2203 	 * an attempt to connect to or otherwise communicate with
2204 	 * a port for which we have no socket.
2205 	 */
2206 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2207 	    tcp_rst_ppslim) == 0) {
2208 		/* XXX stat */
2209 		goto drop;
2210 	}
2211 	/* ...fall into dropwithreset... */
2212 
2213 dropwithreset:
2214 	/*
2215 	 * Generate a RST, dropping incoming segment.
2216 	 * Make ACK acceptable to originator of segment.
2217 	 * Don't bother to respond to RST.
2218 	 */
2219 	if (tiflags & TH_RST)
2220 		goto drop;
2221 	if (tiflags & TH_ACK) {
2222 		tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack,
2223 		    TH_RST);
2224 	} else {
2225 		if (tiflags & TH_SYN)
2226 			tlen++;
2227 		tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen,
2228 		    (tcp_seq)0, TH_RST|TH_ACK);
2229 	}
2230 	m_freem(m);
2231 	return;
2232 
2233 drop:
2234 	/*
2235 	 * Drop space held by incoming segment and return.
2236 	 */
2237 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2238 		switch (tp->pf) {
2239 #ifdef INET6
2240 		case PF_INET6:
2241 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2242 			    0, tlen);
2243 			break;
2244 #endif /* INET6 */
2245 		case PF_INET:
2246 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2247 			    0, tlen);
2248 			break;
2249 		}
2250 	}
2251 
2252 	m_freem(m);
2253 	return;
2254 }
2255 
2256 int
2257 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th,
2258     struct mbuf *m, int iphlen, struct tcp_opt_info *oi)
2259 {
2260 	u_int16_t mss = 0;
2261 	int opt, optlen;
2262 #ifdef TCP_SIGNATURE
2263 	caddr_t sigp = NULL;
2264 	struct tdb *tdb = NULL;
2265 #endif /* TCP_SIGNATURE */
2266 
2267 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2268 		opt = cp[0];
2269 		if (opt == TCPOPT_EOL)
2270 			break;
2271 		if (opt == TCPOPT_NOP)
2272 			optlen = 1;
2273 		else {
2274 			if (cnt < 2)
2275 				break;
2276 			optlen = cp[1];
2277 			if (optlen < 2 || optlen > cnt)
2278 				break;
2279 		}
2280 		switch (opt) {
2281 
2282 		default:
2283 			continue;
2284 
2285 		case TCPOPT_MAXSEG:
2286 			if (optlen != TCPOLEN_MAXSEG)
2287 				continue;
2288 			if (!(th->th_flags & TH_SYN))
2289 				continue;
2290 			if (TCPS_HAVERCVDSYN(tp->t_state))
2291 				continue;
2292 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2293 			NTOHS(mss);
2294 			oi->maxseg = mss;
2295 			break;
2296 
2297 		case TCPOPT_WINDOW:
2298 			if (optlen != TCPOLEN_WINDOW)
2299 				continue;
2300 			if (!(th->th_flags & TH_SYN))
2301 				continue;
2302 			if (TCPS_HAVERCVDSYN(tp->t_state))
2303 				continue;
2304 			tp->t_flags |= TF_RCVD_SCALE;
2305 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2306 			break;
2307 
2308 		case TCPOPT_TIMESTAMP:
2309 			if (optlen != TCPOLEN_TIMESTAMP)
2310 				continue;
2311 			oi->ts_present = 1;
2312 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2313 			NTOHL(oi->ts_val);
2314 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2315 			NTOHL(oi->ts_ecr);
2316 
2317 			if (!(th->th_flags & TH_SYN))
2318 				continue;
2319 			if (TCPS_HAVERCVDSYN(tp->t_state))
2320 				continue;
2321 			/*
2322 			 * A timestamp received in a SYN makes
2323 			 * it ok to send timestamp requests and replies.
2324 			 */
2325 			tp->t_flags |= TF_RCVD_TSTMP;
2326 			tp->ts_recent = oi->ts_val;
2327 			tp->ts_recent_age = tcp_now;
2328 			break;
2329 
2330 #ifdef TCP_SACK
2331 		case TCPOPT_SACK_PERMITTED:
2332 			if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED)
2333 				continue;
2334 			if (!(th->th_flags & TH_SYN))
2335 				continue;
2336 			if (TCPS_HAVERCVDSYN(tp->t_state))
2337 				continue;
2338 			/* MUST only be set on SYN */
2339 			tp->t_flags |= TF_SACK_PERMIT;
2340 			break;
2341 		case TCPOPT_SACK:
2342 			tcp_sack_option(tp, th, cp, optlen);
2343 			break;
2344 #endif
2345 #ifdef TCP_SIGNATURE
2346 		case TCPOPT_SIGNATURE:
2347 			if (optlen != TCPOLEN_SIGNATURE)
2348 				continue;
2349 
2350 			if (sigp && bcmp(sigp, cp + 2, 16))
2351 				return (-1);
2352 
2353 			sigp = cp + 2;
2354 			break;
2355 #endif /* TCP_SIGNATURE */
2356 		}
2357 	}
2358 
2359 #ifdef TCP_SIGNATURE
2360 	if (tp->t_flags & TF_SIGNATURE) {
2361 		union sockaddr_union src, dst;
2362 
2363 		memset(&src, 0, sizeof(union sockaddr_union));
2364 		memset(&dst, 0, sizeof(union sockaddr_union));
2365 
2366 		switch (tp->pf) {
2367 		case 0:
2368 #ifdef INET
2369 		case AF_INET:
2370 			src.sa.sa_len = sizeof(struct sockaddr_in);
2371 			src.sa.sa_family = AF_INET;
2372 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
2373 			dst.sa.sa_len = sizeof(struct sockaddr_in);
2374 			dst.sa.sa_family = AF_INET;
2375 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
2376 			break;
2377 #endif
2378 #ifdef INET6
2379 		case AF_INET6:
2380 			src.sa.sa_len = sizeof(struct sockaddr_in6);
2381 			src.sa.sa_family = AF_INET6;
2382 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
2383 			dst.sa.sa_len = sizeof(struct sockaddr_in6);
2384 			dst.sa.sa_family = AF_INET6;
2385 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
2386 			break;
2387 #endif /* INET6 */
2388 		}
2389 
2390 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
2391 
2392 		/*
2393 		 * We don't have an SA for this peer, so we turn off
2394 		 * TF_SIGNATURE on the listen socket
2395 		 */
2396 		if (tdb == NULL && tp->t_state == TCPS_LISTEN)
2397 			tp->t_flags &= ~TF_SIGNATURE;
2398 
2399 	}
2400 
2401 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
2402 		tcpstat.tcps_rcvbadsig++;
2403 		return (-1);
2404 	}
2405 
2406 	if (sigp) {
2407 		char sig[16];
2408 
2409 		if (tdb == NULL) {
2410 			tcpstat.tcps_rcvbadsig++;
2411 			return (-1);
2412 		}
2413 
2414 		if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0)
2415 			return (-1);
2416 
2417 		if (bcmp(sig, sigp, 16)) {
2418 			tcpstat.tcps_rcvbadsig++;
2419 			return (-1);
2420 		}
2421 
2422 		tcpstat.tcps_rcvgoodsig++;
2423 	}
2424 #endif /* TCP_SIGNATURE */
2425 
2426 	return (0);
2427 }
2428 
2429 #if defined(TCP_SACK)
2430 u_long
2431 tcp_seq_subtract(u_long a, u_long b)
2432 {
2433 	return ((long)(a - b));
2434 }
2435 #endif
2436 
2437 
2438 #ifdef TCP_SACK
2439 /*
2440  * This function is called upon receipt of new valid data (while not in header
2441  * prediction mode), and it updates the ordered list of sacks.
2442  */
2443 void
2444 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart,
2445     tcp_seq rcv_lastend)
2446 {
2447 	/*
2448 	 * First reported block MUST be the most recent one.  Subsequent
2449 	 * blocks SHOULD be in the order in which they arrived at the
2450 	 * receiver.  These two conditions make the implementation fully
2451 	 * compliant with RFC 2018.
2452 	 */
2453 	int i, j = 0, count = 0, lastpos = -1;
2454 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2455 
2456 	/* First clean up current list of sacks */
2457 	for (i = 0; i < tp->rcv_numsacks; i++) {
2458 		sack = tp->sackblks[i];
2459 		if (sack.start == 0 && sack.end == 0) {
2460 			count++; /* count = number of blocks to be discarded */
2461 			continue;
2462 		}
2463 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2464 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2465 			count++;
2466 		} else {
2467 			temp[j].start = tp->sackblks[i].start;
2468 			temp[j++].end = tp->sackblks[i].end;
2469 		}
2470 	}
2471 	tp->rcv_numsacks -= count;
2472 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2473 		tcp_clean_sackreport(tp);
2474 		if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) {
2475 			/* ==> need first sack block */
2476 			tp->sackblks[0].start = rcv_laststart;
2477 			tp->sackblks[0].end = rcv_lastend;
2478 			tp->rcv_numsacks = 1;
2479 		}
2480 		return;
2481 	}
2482 	/* Otherwise, sack blocks are already present. */
2483 	for (i = 0; i < tp->rcv_numsacks; i++)
2484 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2485 	if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend))
2486 		return;     /* sack list remains unchanged */
2487 	/*
2488 	 * From here, segment just received should be (part of) the 1st sack.
2489 	 * Go through list, possibly coalescing sack block entries.
2490 	 */
2491 	firstsack.start = rcv_laststart;
2492 	firstsack.end = rcv_lastend;
2493 	for (i = 0; i < tp->rcv_numsacks; i++) {
2494 		sack = tp->sackblks[i];
2495 		if (SEQ_LT(sack.end, firstsack.start) ||
2496 		    SEQ_GT(sack.start, firstsack.end))
2497 			continue; /* no overlap */
2498 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2499 			/*
2500 			 * identical block; delete it here since we will
2501 			 * move it to the front of the list.
2502 			 */
2503 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2504 			lastpos = i;    /* last posn with a zero entry */
2505 			continue;
2506 		}
2507 		if (SEQ_LEQ(sack.start, firstsack.start))
2508 			firstsack.start = sack.start; /* merge blocks */
2509 		if (SEQ_GEQ(sack.end, firstsack.end))
2510 			firstsack.end = sack.end;     /* merge blocks */
2511 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2512 		lastpos = i;    /* last posn with a zero entry */
2513 	}
2514 	if (lastpos != -1) {    /* at least one merge */
2515 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2516 			sack = tp->sackblks[i];
2517 			if (sack.start == 0 && sack.end == 0)
2518 				continue;
2519 			temp[j++] = sack;
2520 		}
2521 		tp->rcv_numsacks = j; /* including first blk (added later) */
2522 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2523 			tp->sackblks[i] = temp[i];
2524 	} else {        /* no merges -- shift sacks by 1 */
2525 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2526 			tp->rcv_numsacks++;
2527 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2528 			tp->sackblks[i] = tp->sackblks[i-1];
2529 	}
2530 	tp->sackblks[0] = firstsack;
2531 	return;
2532 }
2533 
2534 /*
2535  * Process the TCP SACK option.  tp->snd_holes is an ordered list
2536  * of holes (oldest to newest, in terms of the sequence space).
2537  */
2538 void
2539 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2540 {
2541 	int tmp_olen;
2542 	u_char *tmp_cp;
2543 	struct sackhole *cur, *p, *temp;
2544 
2545 	if (!tp->sack_enable)
2546 		return;
2547 	/* SACK without ACK doesn't make sense. */
2548 	if ((th->th_flags & TH_ACK) == 0)
2549 	       return;
2550 	/* Make sure the ACK on this segment is in [snd_una, snd_max]. */
2551 	if (SEQ_LT(th->th_ack, tp->snd_una) ||
2552 	    SEQ_GT(th->th_ack, tp->snd_max))
2553 		return;
2554 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2555 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2556 		return;
2557 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2558 	tmp_cp = cp + 2;
2559 	tmp_olen = optlen - 2;
2560 	tcpstat.tcps_sack_rcv_opts++;
2561 	if (tp->snd_numholes < 0)
2562 		tp->snd_numholes = 0;
2563 	if (tp->t_maxseg == 0)
2564 		panic("tcp_sack_option"); /* Should never happen */
2565 	while (tmp_olen > 0) {
2566 		struct sackblk sack;
2567 
2568 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2569 		NTOHL(sack.start);
2570 		bcopy(tmp_cp + sizeof(tcp_seq),
2571 		    (char *) &(sack.end), sizeof(tcp_seq));
2572 		NTOHL(sack.end);
2573 		tmp_olen -= TCPOLEN_SACK;
2574 		tmp_cp += TCPOLEN_SACK;
2575 		if (SEQ_LEQ(sack.end, sack.start))
2576 			continue; /* bad SACK fields */
2577 		if (SEQ_LEQ(sack.end, tp->snd_una))
2578 			continue; /* old block */
2579 #if defined(TCP_SACK) && defined(TCP_FACK)
2580 		/* Updates snd_fack.  */
2581 		if (SEQ_GT(sack.end, tp->snd_fack))
2582 			tp->snd_fack = sack.end;
2583 #endif /* TCP_FACK */
2584 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2585 			if (SEQ_LT(sack.start, th->th_ack))
2586 				continue;
2587 		}
2588 		if (SEQ_GT(sack.end, tp->snd_max))
2589 			continue;
2590 		if (tp->snd_holes == NULL) { /* first hole */
2591 			tp->snd_holes = (struct sackhole *)
2592 			    pool_get(&sackhl_pool, PR_NOWAIT);
2593 			if (tp->snd_holes == NULL) {
2594 				/* ENOBUFS, so ignore SACKed block for now*/
2595 				goto done;
2596 			}
2597 			cur = tp->snd_holes;
2598 			cur->start = th->th_ack;
2599 			cur->end = sack.start;
2600 			cur->rxmit = cur->start;
2601 			cur->next = NULL;
2602 			tp->snd_numholes = 1;
2603 			tp->rcv_lastsack = sack.end;
2604 			/*
2605 			 * dups is at least one.  If more data has been
2606 			 * SACKed, it can be greater than one.
2607 			 */
2608 			cur->dups = min(tcprexmtthresh,
2609 			    ((sack.end - cur->end)/tp->t_maxseg));
2610 			if (cur->dups < 1)
2611 				cur->dups = 1;
2612 			continue; /* with next sack block */
2613 		}
2614 		/* Go thru list of holes:  p = previous,  cur = current */
2615 		p = cur = tp->snd_holes;
2616 		while (cur) {
2617 			if (SEQ_LEQ(sack.end, cur->start))
2618 				/* SACKs data before the current hole */
2619 				break; /* no use going through more holes */
2620 			if (SEQ_GEQ(sack.start, cur->end)) {
2621 				/* SACKs data beyond the current hole */
2622 				cur->dups++;
2623 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2624 				    tcprexmtthresh)
2625 					cur->dups = tcprexmtthresh;
2626 				p = cur;
2627 				cur = cur->next;
2628 				continue;
2629 			}
2630 			if (SEQ_LEQ(sack.start, cur->start)) {
2631 				/* Data acks at least the beginning of hole */
2632 #if defined(TCP_SACK) && defined(TCP_FACK)
2633 				if (SEQ_GT(sack.end, cur->rxmit))
2634 					tp->retran_data -=
2635 				    	    tcp_seq_subtract(cur->rxmit,
2636 					    cur->start);
2637 				else
2638 					tp->retran_data -=
2639 					    tcp_seq_subtract(sack.end,
2640 					    cur->start);
2641 #endif /* TCP_FACK */
2642 				if (SEQ_GEQ(sack.end, cur->end)) {
2643 					/* Acks entire hole, so delete hole */
2644 					if (p != cur) {
2645 						p->next = cur->next;
2646 						pool_put(&sackhl_pool, cur);
2647 						cur = p->next;
2648 					} else {
2649 						cur = cur->next;
2650 						pool_put(&sackhl_pool, p);
2651 						p = cur;
2652 						tp->snd_holes = p;
2653 					}
2654 					tp->snd_numholes--;
2655 					continue;
2656 				}
2657 				/* otherwise, move start of hole forward */
2658 				cur->start = sack.end;
2659 				cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
2660 				p = cur;
2661 				cur = cur->next;
2662 				continue;
2663 			}
2664 			/* move end of hole backward */
2665 			if (SEQ_GEQ(sack.end, cur->end)) {
2666 #if defined(TCP_SACK) && defined(TCP_FACK)
2667 				if (SEQ_GT(cur->rxmit, sack.start))
2668 					tp->retran_data -=
2669 					    tcp_seq_subtract(cur->rxmit,
2670 					    sack.start);
2671 #endif /* TCP_FACK */
2672 				cur->end = sack.start;
2673 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2674 				cur->dups++;
2675 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2676 				    tcprexmtthresh)
2677 					cur->dups = tcprexmtthresh;
2678 				p = cur;
2679 				cur = cur->next;
2680 				continue;
2681 			}
2682 			if (SEQ_LT(cur->start, sack.start) &&
2683 			    SEQ_GT(cur->end, sack.end)) {
2684 				/*
2685 				 * ACKs some data in middle of a hole; need to
2686 				 * split current hole
2687 				 */
2688 				temp = (struct sackhole *)
2689 				    pool_get(&sackhl_pool, PR_NOWAIT);
2690 				if (temp == NULL)
2691 					goto done; /* ENOBUFS */
2692 #if defined(TCP_SACK) && defined(TCP_FACK)
2693 				if (SEQ_GT(cur->rxmit, sack.end))
2694 					tp->retran_data -=
2695 					    tcp_seq_subtract(sack.end,
2696 					    sack.start);
2697 				else if (SEQ_GT(cur->rxmit, sack.start))
2698 					tp->retran_data -=
2699 					    tcp_seq_subtract(cur->rxmit,
2700 					    sack.start);
2701 #endif /* TCP_FACK */
2702 				temp->next = cur->next;
2703 				temp->start = sack.end;
2704 				temp->end = cur->end;
2705 				temp->dups = cur->dups;
2706 				temp->rxmit = SEQ_MAX(cur->rxmit, temp->start);
2707 				cur->end = sack.start;
2708 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2709 				cur->dups++;
2710 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2711 					tcprexmtthresh)
2712 					cur->dups = tcprexmtthresh;
2713 				cur->next = temp;
2714 				p = temp;
2715 				cur = p->next;
2716 				tp->snd_numholes++;
2717 			}
2718 		}
2719 		/* At this point, p points to the last hole on the list */
2720 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2721 			/*
2722 			 * Need to append new hole at end.
2723 			 * Last hole is p (and it's not NULL).
2724 			 */
2725 			temp = (struct sackhole *)
2726 			    pool_get(&sackhl_pool, PR_NOWAIT);
2727 			if (temp == NULL)
2728 				goto done; /* ENOBUFS */
2729 			temp->start = tp->rcv_lastsack;
2730 			temp->end = sack.start;
2731 			temp->dups = min(tcprexmtthresh,
2732 			    ((sack.end - sack.start)/tp->t_maxseg));
2733 			if (temp->dups < 1)
2734 				temp->dups = 1;
2735 			temp->rxmit = temp->start;
2736 			temp->next = 0;
2737 			p->next = temp;
2738 			tp->rcv_lastsack = sack.end;
2739 			tp->snd_numholes++;
2740 		}
2741 	}
2742 done:
2743 #if defined(TCP_SACK) && defined(TCP_FACK)
2744 	/*
2745 	 * Update retran_data and snd_awnd.  Go through the list of
2746 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2747 	 */
2748 	tp->retran_data = 0;
2749 	cur = tp->snd_holes;
2750 	while (cur) {
2751 		tp->retran_data += cur->rxmit - cur->start;
2752 		cur = cur->next;
2753 	}
2754 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2755 	    tp->retran_data;
2756 #endif /* TCP_FACK */
2757 
2758 	return;
2759 }
2760 
2761 /*
2762  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2763  * it is completely acked; otherwise, tcp_sack_option(), called from
2764  * tcp_dooptions(), will fix up the hole.
2765  */
2766 void
2767 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th)
2768 {
2769 	if (tp->sack_enable && tp->t_state != TCPS_LISTEN) {
2770 		/* max because this could be an older ack just arrived */
2771 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2772 			th->th_ack : tp->snd_una;
2773 		struct sackhole *cur = tp->snd_holes;
2774 		struct sackhole *prev;
2775 		while (cur)
2776 			if (SEQ_LEQ(cur->end, lastack)) {
2777 				prev = cur;
2778 				cur = cur->next;
2779 				pool_put(&sackhl_pool, prev);
2780 				tp->snd_numholes--;
2781 			} else if (SEQ_LT(cur->start, lastack)) {
2782 				cur->start = lastack;
2783 				if (SEQ_LT(cur->rxmit, cur->start))
2784 					cur->rxmit = cur->start;
2785 				break;
2786 			} else
2787 				break;
2788 		tp->snd_holes = cur;
2789 	}
2790 }
2791 
2792 /*
2793  * Delete all receiver-side SACK information.
2794  */
2795 void
2796 tcp_clean_sackreport(struct tcpcb *tp)
2797 {
2798 	int i;
2799 
2800 	tp->rcv_numsacks = 0;
2801 	for (i = 0; i < MAX_SACK_BLKS; i++)
2802 		tp->sackblks[i].start = tp->sackblks[i].end=0;
2803 
2804 }
2805 
2806 /*
2807  * Checks for partial ack.  If partial ack arrives, turn off retransmission
2808  * timer, deflate the window, do not clear tp->t_dupacks, and return 1.
2809  * If the ack advances at least to tp->snd_last, return 0.
2810  */
2811 int
2812 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th)
2813 {
2814 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2815 		/* Turn off retx. timer (will start again next segment) */
2816 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2817 		tp->t_rtttime = 0;
2818 #ifndef TCP_FACK
2819 		/*
2820 		 * Partial window deflation.  This statement relies on the
2821 		 * fact that tp->snd_una has not been updated yet.  In FACK
2822 		 * hold snd_cwnd constant during fast recovery.
2823 		 */
2824 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2825 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2826 			tp->snd_cwnd += tp->t_maxseg;
2827 		} else
2828 			tp->snd_cwnd = tp->t_maxseg;
2829 #endif
2830 		return (1);
2831 	}
2832 	return (0);
2833 }
2834 #endif /* TCP_SACK */
2835 
2836 /*
2837  * Pull out of band byte out of a segment so
2838  * it doesn't appear in the user's data queue.
2839  * It is still reflected in the segment length for
2840  * sequencing purposes.
2841  */
2842 void
2843 tcp_pulloutofband(struct socket *so, u_int urgent, struct mbuf *m, int off)
2844 {
2845         int cnt = off + urgent - 1;
2846 
2847 	while (cnt >= 0) {
2848 		if (m->m_len > cnt) {
2849 			char *cp = mtod(m, caddr_t) + cnt;
2850 			struct tcpcb *tp = sototcpcb(so);
2851 
2852 			tp->t_iobc = *cp;
2853 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2854 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2855 			m->m_len--;
2856 			return;
2857 		}
2858 		cnt -= m->m_len;
2859 		m = m->m_next;
2860 		if (m == 0)
2861 			break;
2862 	}
2863 	panic("tcp_pulloutofband");
2864 }
2865 
2866 /*
2867  * Collect new round-trip time estimate
2868  * and update averages and current timeout.
2869  */
2870 void
2871 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2872 {
2873 	short delta;
2874 	short rttmin;
2875 
2876 	if (rtt < 0)
2877 		rtt = 0;
2878 	else if (rtt > TCP_RTT_MAX)
2879 		rtt = TCP_RTT_MAX;
2880 
2881 	tcpstat.tcps_rttupdated++;
2882 	if (tp->t_srtt != 0) {
2883 		/*
2884 		 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits
2885 		 * after the binary point (scaled by 4), whereas
2886 		 * srtt is stored as fixed point with 5 bits after the
2887 		 * binary point (i.e., scaled by 32).  The following magic
2888 		 * is equivalent to the smoothing algorithm in rfc793 with
2889 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2890 		 * point).
2891 		 */
2892 		delta = (rtt << TCP_RTT_BASE_SHIFT) -
2893 		    (tp->t_srtt >> TCP_RTT_SHIFT);
2894 		if ((tp->t_srtt += delta) <= 0)
2895 			tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT;
2896 		/*
2897 		 * We accumulate a smoothed rtt variance (actually, a
2898 		 * smoothed mean difference), then set the retransmit
2899 		 * timer to smoothed rtt + 4 times the smoothed variance.
2900 		 * rttvar is stored as fixed point with 4 bits after the
2901 		 * binary point (scaled by 16).  The following is
2902 		 * equivalent to rfc793 smoothing with an alpha of .75
2903 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2904 		 * rfc793's wired-in beta.
2905 		 */
2906 		if (delta < 0)
2907 			delta = -delta;
2908 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2909 		if ((tp->t_rttvar += delta) <= 0)
2910 			tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT;
2911 	} else {
2912 		/*
2913 		 * No rtt measurement yet - use the unsmoothed rtt.
2914 		 * Set the variance to half the rtt (so our first
2915 		 * retransmit happens at 3*rtt).
2916 		 */
2917 		tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
2918 		tp->t_rttvar = (rtt + 1) <<
2919 		    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
2920 	}
2921 	tp->t_rtttime = 0;
2922 	tp->t_rxtshift = 0;
2923 
2924 	/*
2925 	 * the retransmit should happen at rtt + 4 * rttvar.
2926 	 * Because of the way we do the smoothing, srtt and rttvar
2927 	 * will each average +1/2 tick of bias.  When we compute
2928 	 * the retransmit timer, we want 1/2 tick of rounding and
2929 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2930 	 * firing of the timer.  The bias will give us exactly the
2931 	 * 1.5 tick we need.  But, because the bias is
2932 	 * statistical, we have to test that we don't drop below
2933 	 * the minimum feasible timer (which is 2 ticks).
2934 	 */
2935 	rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX);
2936 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2937 
2938 	/*
2939 	 * We received an ack for a packet that wasn't retransmitted;
2940 	 * it is probably safe to discard any error indications we've
2941 	 * received recently.  This isn't quite right, but close enough
2942 	 * for now (a route might have failed after we sent a segment,
2943 	 * and the return path might not be symmetrical).
2944 	 */
2945 	tp->t_softerror = 0;
2946 }
2947 
2948 /*
2949  * Determine a reasonable value for maxseg size.
2950  * If the route is known, check route for mtu.
2951  * If none, use an mss that can be handled on the outgoing
2952  * interface without forcing IP to fragment; if bigger than
2953  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2954  * to utilize large mbufs.  If no route is found, route has no mtu,
2955  * or the destination isn't local, use a default, hopefully conservative
2956  * size (usually 512 or the default IP max size, but no more than the mtu
2957  * of the interface), as we can't discover anything about intervening
2958  * gateways or networks.  We also initialize the congestion/slow start
2959  * window to be a single segment if the destination isn't local.
2960  * While looking at the routing entry, we also initialize other path-dependent
2961  * parameters from pre-set or cached values in the routing entry.
2962  *
2963  * Also take into account the space needed for options that we
2964  * send regularly.  Make maxseg shorter by that amount to assure
2965  * that we can send maxseg amount of data even when the options
2966  * are present.  Store the upper limit of the length of options plus
2967  * data in maxopd.
2968  *
2969  * NOTE: offer == -1 indicates that the maxseg size changed due to
2970  * Path MTU discovery.
2971  */
2972 int
2973 tcp_mss(struct tcpcb *tp, int offer)
2974 {
2975 	struct rtentry *rt;
2976 	struct ifnet *ifp;
2977 	int mss, mssopt;
2978 	int iphlen;
2979 	struct inpcb *inp;
2980 
2981 	inp = tp->t_inpcb;
2982 
2983 	mssopt = mss = tcp_mssdflt;
2984 
2985 	rt = in_pcbrtentry(inp);
2986 
2987 	if (rt == NULL)
2988 		goto out;
2989 
2990 	ifp = rt->rt_ifp;
2991 
2992 	switch (tp->pf) {
2993 #ifdef INET6
2994 	case AF_INET6:
2995 		iphlen = sizeof(struct ip6_hdr);
2996 		break;
2997 #endif
2998 	case AF_INET:
2999 		iphlen = sizeof(struct ip);
3000 		break;
3001 	default:
3002 		/* the family does not support path MTU discovery */
3003 		goto out;
3004 	}
3005 
3006 #ifdef RTV_MTU
3007 	/*
3008 	 * if there's an mtu associated with the route and we support
3009 	 * path MTU discovery for the underlying protocol family, use it.
3010 	 */
3011 	if (rt->rt_rmx.rmx_mtu) {
3012 		/*
3013 		 * One may wish to lower MSS to take into account options,
3014 		 * especially security-related options.
3015 		 */
3016 		if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) {
3017 			/*
3018 			 * RFC2460 section 5, last paragraph: if path MTU is
3019 			 * smaller than 1280, use 1280 as packet size and
3020 			 * attach fragment header.
3021 			 */
3022 			mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) -
3023 			    sizeof(struct tcphdr);
3024 		} else
3025 			mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr);
3026 	} else
3027 #endif /* RTV_MTU */
3028 	if (!ifp)
3029 		/*
3030 		 * ifp may be null and rmx_mtu may be zero in certain
3031 		 * v6 cases (e.g., if ND wasn't able to resolve the
3032 		 * destination host.
3033 		 */
3034 		goto out;
3035 	else if (ifp->if_flags & IFF_LOOPBACK)
3036 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3037 	else if (tp->pf == AF_INET) {
3038 		if (ip_mtudisc)
3039 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3040 		else if (inp && in_localaddr(inp->inp_faddr))
3041 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3042 	}
3043 #ifdef INET6
3044 	else if (tp->pf == AF_INET6) {
3045 		/*
3046 		 * for IPv6, path MTU discovery is always turned on,
3047 		 * or the node must use packet size <= 1280.
3048 		 */
3049 		mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr);
3050 	}
3051 #endif /* INET6 */
3052 
3053 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3054 	if (offer != -1) {
3055 #ifndef INET6
3056 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3057 #else
3058 		if (tp->pf == AF_INET6)
3059 			mssopt = IN6_LINKMTU(ifp) - iphlen -
3060 			    sizeof(struct tcphdr);
3061 		else
3062 			mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3063 #endif
3064 
3065 		mssopt = max(tcp_mssdflt, mssopt);
3066 	}
3067 
3068  out:
3069 	/*
3070 	 * The current mss, t_maxseg, is initialized to the default value.
3071 	 * If we compute a smaller value, reduce the current mss.
3072 	 * If we compute a larger value, return it for use in sending
3073 	 * a max seg size option, but don't store it for use
3074 	 * unless we received an offer at least that large from peer.
3075 	 *
3076 	 * However, do not accept offers lower than the minimum of
3077 	 * the interface MTU and 216.
3078 	 */
3079 	if (offer > 0)
3080 		tp->t_peermss = offer;
3081 	if (tp->t_peermss)
3082 		mss = min(mss, max(tp->t_peermss, 216));
3083 
3084 	/* sanity - at least max opt. space */
3085 	mss = max(mss, 64);
3086 
3087 	/*
3088 	 * maxopd stores the maximum length of data AND options
3089 	 * in a segment; maxseg is the amount of data in a normal
3090 	 * segment.  We need to store this value (maxopd) apart
3091 	 * from maxseg, because now every segment carries options
3092 	 * and thus we normally have somewhat less data in segments.
3093 	 */
3094 	tp->t_maxopd = mss;
3095 
3096 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3097 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3098 		mss -= TCPOLEN_TSTAMP_APPA;
3099 #ifdef TCP_SIGNATURE
3100 	if (tp->t_flags & TF_SIGNATURE)
3101 		mss -= TCPOLEN_SIGLEN;
3102 #endif
3103 
3104 	if (offer == -1) {
3105 		/* mss changed due to Path MTU discovery */
3106 		tp->t_flags &= ~TF_PMTUD_PEND;
3107 		tp->t_pmtud_mtu_sent = 0;
3108 		tp->t_pmtud_mss_acked = 0;
3109 		if (mss < tp->t_maxseg) {
3110 			/*
3111 			 * Follow suggestion in RFC 2414 to reduce the
3112 			 * congestion window by the ratio of the old
3113 			 * segment size to the new segment size.
3114 			 */
3115 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3116 					     mss, mss);
3117 		}
3118 	} else if (tcp_do_rfc3390) {
3119 		/* increase initial window  */
3120 		tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380));
3121 	} else
3122 		tp->snd_cwnd = mss;
3123 
3124 	tp->t_maxseg = mss;
3125 
3126 	return (offer != -1 ? mssopt : mss);
3127 }
3128 
3129 u_int
3130 tcp_hdrsz(struct tcpcb *tp)
3131 {
3132 	u_int hlen;
3133 
3134 	switch (tp->pf) {
3135 #ifdef INET6
3136 	case AF_INET6:
3137 		hlen = sizeof(struct ip6_hdr);
3138 		break;
3139 #endif
3140 	case AF_INET:
3141 		hlen = sizeof(struct ip);
3142 		break;
3143 	default:
3144 		hlen = 0;
3145 		break;
3146 	}
3147 	hlen += sizeof(struct tcphdr);
3148 
3149 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3150 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3151 		hlen += TCPOLEN_TSTAMP_APPA;
3152 #ifdef TCP_SIGNATURE
3153 	if (tp->t_flags & TF_SIGNATURE)
3154 		hlen += TCPOLEN_SIGLEN;
3155 #endif
3156 	return (hlen);
3157 }
3158 
3159 /*
3160  * Set connection variables based on the effective MSS.
3161  * We are passed the TCPCB for the actual connection.  If we
3162  * are the server, we are called by the compressed state engine
3163  * when the 3-way handshake is complete.  If we are the client,
3164  * we are called when we receive the SYN,ACK from the server.
3165  *
3166  * NOTE: The t_maxseg value must be initialized in the TCPCB
3167  * before this routine is called!
3168  */
3169 void
3170 tcp_mss_update(struct tcpcb *tp)
3171 {
3172 	int mss;
3173 	u_long bufsize;
3174 	struct rtentry *rt;
3175 	struct socket *so;
3176 
3177 	so = tp->t_inpcb->inp_socket;
3178 	mss = tp->t_maxseg;
3179 
3180 	rt = in_pcbrtentry(tp->t_inpcb);
3181 
3182 	if (rt == NULL)
3183 		return;
3184 
3185 	bufsize = so->so_snd.sb_hiwat;
3186 	if (bufsize < mss) {
3187 		mss = bufsize;
3188 		/* Update t_maxseg and t_maxopd */
3189 		tcp_mss(tp, mss);
3190 	} else {
3191 		bufsize = roundup(bufsize, mss);
3192 		if (bufsize > sb_max)
3193 			bufsize = sb_max;
3194 		(void)sbreserve(&so->so_snd, bufsize);
3195 	}
3196 
3197 	bufsize = so->so_rcv.sb_hiwat;
3198 	if (bufsize > mss) {
3199 		bufsize = roundup(bufsize, mss);
3200 		if (bufsize > sb_max)
3201 			bufsize = sb_max;
3202 		(void)sbreserve(&so->so_rcv, bufsize);
3203 	}
3204 
3205 }
3206 
3207 #if defined (TCP_SACK)
3208 /*
3209  * Checks for partial ack.  If partial ack arrives, force the retransmission
3210  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3211  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3212  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3213  */
3214 int
3215 tcp_newreno(struct tcpcb *tp, struct tcphdr *th)
3216 {
3217 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3218 		/*
3219 		 * snd_una has not been updated and the socket send buffer
3220 		 * not yet drained of the acked data, so we have to leave
3221 		 * snd_una as it was to get the correct data offset in
3222 		 * tcp_output().
3223 		 */
3224 		tcp_seq onxt = tp->snd_nxt;
3225 		u_long  ocwnd = tp->snd_cwnd;
3226 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3227 		tp->t_rtttime = 0;
3228 		tp->snd_nxt = th->th_ack;
3229 		/*
3230 		 * Set snd_cwnd to one segment beyond acknowledged offset
3231 		 * (tp->snd_una not yet updated when this function is called)
3232 		 */
3233 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3234 		(void) tcp_output(tp);
3235 		tp->snd_cwnd = ocwnd;
3236 		if (SEQ_GT(onxt, tp->snd_nxt))
3237 			tp->snd_nxt = onxt;
3238 		/*
3239 		 * Partial window deflation.  Relies on fact that tp->snd_una
3240 		 * not updated yet.
3241 		 */
3242 		if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3243 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
3244 		else
3245 			tp->snd_cwnd = 0;
3246 		tp->snd_cwnd += tp->t_maxseg;
3247 
3248 		return 1;
3249 	}
3250 	return 0;
3251 }
3252 #endif /* TCP_SACK */
3253 
3254 int
3255 tcp_mss_adv(struct ifnet *ifp, int af)
3256 {
3257 	int mss = 0;
3258 	int iphlen;
3259 
3260 	switch (af) {
3261 	case AF_INET:
3262 		if (ifp != NULL)
3263 			mss = ifp->if_mtu;
3264 		iphlen = sizeof(struct ip);
3265 		break;
3266 #ifdef INET6
3267 	case AF_INET6:
3268 		if (ifp != NULL)
3269 			mss = IN6_LINKMTU(ifp);
3270 		iphlen = sizeof(struct ip6_hdr);
3271 		break;
3272 #endif
3273 	}
3274 	mss = mss - iphlen - sizeof(struct tcphdr);
3275 	return (max(mss, tcp_mssdflt));
3276 }
3277 
3278 /*
3279  * TCP compressed state engine.  Currently used to hold compressed
3280  * state for SYN_RECEIVED.
3281  */
3282 
3283 u_long	syn_cache_count;
3284 u_int32_t syn_hash1, syn_hash2;
3285 
3286 #define SYN_HASH(sa, sp, dp) \
3287 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3288 				     ((u_int32_t)(sp)))^syn_hash2)))
3289 #ifndef INET6
3290 #define	SYN_HASHALL(hash, src, dst) \
3291 do {									\
3292 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
3293 		((struct sockaddr_in *)(src))->sin_port,		\
3294 		((struct sockaddr_in *)(dst))->sin_port);		\
3295 } while (/*CONSTCOND*/ 0)
3296 #else
3297 #define SYN_HASH6(sa, sp, dp) \
3298 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3299 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3300 	 & 0x7fffffff)
3301 
3302 #define SYN_HASHALL(hash, src, dst) \
3303 do {									\
3304 	switch ((src)->sa_family) {					\
3305 	case AF_INET:							\
3306 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
3307 			((struct sockaddr_in *)(src))->sin_port,	\
3308 			((struct sockaddr_in *)(dst))->sin_port);	\
3309 		break;							\
3310 	case AF_INET6:							\
3311 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
3312 			((struct sockaddr_in6 *)(src))->sin6_port,	\
3313 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
3314 		break;							\
3315 	default:							\
3316 		hash = 0;						\
3317 	}								\
3318 } while (/*CONSTCOND*/0)
3319 #endif /* INET6 */
3320 
3321 #define	SYN_CACHE_RM(sc)						\
3322 do {									\
3323 	(sc)->sc_flags |= SCF_DEAD;					\
3324 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
3325 	    (sc), sc_bucketq);						\
3326 	(sc)->sc_tp = NULL;						\
3327 	LIST_REMOVE((sc), sc_tpq);					\
3328 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
3329 	timeout_del(&(sc)->sc_timer);					\
3330 	syn_cache_count--;						\
3331 } while (/*CONSTCOND*/0)
3332 
3333 #define	SYN_CACHE_PUT(sc)						\
3334 do {									\
3335 	if ((sc)->sc_ipopts)						\
3336 		(void) m_free((sc)->sc_ipopts);				\
3337 	if ((sc)->sc_route4.ro_rt != NULL)				\
3338 		RTFREE((sc)->sc_route4.ro_rt);				\
3339 	timeout_set(&(sc)->sc_timer, syn_cache_reaper, (sc));		\
3340 	timeout_add(&(sc)->sc_timer, 0);				\
3341 } while (/*CONSTCOND*/0)
3342 
3343 struct pool syn_cache_pool;
3344 
3345 /*
3346  * We don't estimate RTT with SYNs, so each packet starts with the default
3347  * RTT and each timer step has a fixed timeout value.
3348  */
3349 #define	SYN_CACHE_TIMER_ARM(sc)						\
3350 do {									\
3351 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3352 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3353 	    TCPTV_REXMTMAX);						\
3354 	if (!timeout_initialized(&(sc)->sc_timer))			\
3355 		timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc));	\
3356 	timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \
3357 } while (/*CONSTCOND*/0)
3358 
3359 #define	SYN_CACHE_TIMESTAMP(sc)	tcp_now + (sc)->sc_modulate
3360 
3361 void
3362 syn_cache_init()
3363 {
3364 	int i;
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 	/* Initialize the syn cache pool. */
3371 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3372 	    "synpl", NULL);
3373 }
3374 
3375 void
3376 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3377 {
3378 	struct syn_cache_head *scp;
3379 	struct syn_cache *sc2;
3380 	int s;
3381 
3382 	/*
3383 	 * If there are no entries in the hash table, reinitialize
3384 	 * the hash secrets.
3385 	 */
3386 	if (syn_cache_count == 0) {
3387 		syn_hash1 = arc4random();
3388 		syn_hash2 = arc4random();
3389 	}
3390 
3391 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3392 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3393 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3394 
3395 	/*
3396 	 * Make sure that we don't overflow the per-bucket
3397 	 * limit or the total cache size limit.
3398 	 */
3399 	s = splsoftnet();
3400 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3401 		tcpstat.tcps_sc_bucketoverflow++;
3402 		/*
3403 		 * The bucket is full.  Toss the oldest element in the
3404 		 * bucket.  This will be the first entry in the bucket.
3405 		 */
3406 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3407 #ifdef DIAGNOSTIC
3408 		/*
3409 		 * This should never happen; we should always find an
3410 		 * entry in our bucket.
3411 		 */
3412 		if (sc2 == NULL)
3413 			panic("syn_cache_insert: bucketoverflow: impossible");
3414 #endif
3415 		SYN_CACHE_RM(sc2);
3416 		SYN_CACHE_PUT(sc2);
3417 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3418 		struct syn_cache_head *scp2, *sce;
3419 
3420 		tcpstat.tcps_sc_overflowed++;
3421 		/*
3422 		 * The cache is full.  Toss the oldest entry in the
3423 		 * first non-empty bucket we can find.
3424 		 *
3425 		 * XXX We would really like to toss the oldest
3426 		 * entry in the cache, but we hope that this
3427 		 * condition doesn't happen very often.
3428 		 */
3429 		scp2 = scp;
3430 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3431 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3432 			for (++scp2; scp2 != scp; scp2++) {
3433 				if (scp2 >= sce)
3434 					scp2 = &tcp_syn_cache[0];
3435 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3436 					break;
3437 			}
3438 #ifdef DIAGNOSTIC
3439 			/*
3440 			 * This should never happen; we should always find a
3441 			 * non-empty bucket.
3442 			 */
3443 			if (scp2 == scp)
3444 				panic("syn_cache_insert: cacheoverflow: "
3445 				    "impossible");
3446 #endif
3447 		}
3448 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3449 		SYN_CACHE_RM(sc2);
3450 		SYN_CACHE_PUT(sc2);
3451 	}
3452 
3453 	/*
3454 	 * Initialize the entry's timer.
3455 	 */
3456 	sc->sc_rxttot = 0;
3457 	sc->sc_rxtshift = 0;
3458 	SYN_CACHE_TIMER_ARM(sc);
3459 
3460 	/* Link it from tcpcb entry */
3461 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3462 
3463 	/* Put it into the bucket. */
3464 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3465 	scp->sch_length++;
3466 	syn_cache_count++;
3467 
3468 	tcpstat.tcps_sc_added++;
3469 	splx(s);
3470 }
3471 
3472 /*
3473  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3474  * If we have retransmitted an entry the maximum number of times, expire
3475  * that entry.
3476  */
3477 void
3478 syn_cache_timer(void *arg)
3479 {
3480 	struct syn_cache *sc = arg;
3481 	int s;
3482 
3483 	s = splsoftnet();
3484 	if (sc->sc_flags & SCF_DEAD) {
3485 		splx(s);
3486 		return;
3487 	}
3488 
3489 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3490 		/* Drop it -- too many retransmissions. */
3491 		goto dropit;
3492 	}
3493 
3494 	/*
3495 	 * Compute the total amount of time this entry has
3496 	 * been on a queue.  If this entry has been on longer
3497 	 * than the keep alive timer would allow, expire it.
3498 	 */
3499 	sc->sc_rxttot += sc->sc_rxtcur;
3500 	if (sc->sc_rxttot >= tcptv_keep_init)
3501 		goto dropit;
3502 
3503 	tcpstat.tcps_sc_retransmitted++;
3504 	(void) syn_cache_respond(sc, NULL);
3505 
3506 	/* Advance the timer back-off. */
3507 	sc->sc_rxtshift++;
3508 	SYN_CACHE_TIMER_ARM(sc);
3509 
3510 	splx(s);
3511 	return;
3512 
3513  dropit:
3514 	tcpstat.tcps_sc_timed_out++;
3515 	SYN_CACHE_RM(sc);
3516 	SYN_CACHE_PUT(sc);
3517 	splx(s);
3518 }
3519 
3520 void
3521 syn_cache_reaper(void *arg)
3522 {
3523 	struct syn_cache *sc = arg;
3524 	int s;
3525 
3526 	s = splsoftnet();
3527 	pool_put(&syn_cache_pool, (sc));
3528 	splx(s);
3529 	return;
3530 }
3531 
3532 /*
3533  * Remove syn cache created by the specified tcb entry,
3534  * because this does not make sense to keep them
3535  * (if there's no tcb entry, syn cache entry will never be used)
3536  */
3537 void
3538 syn_cache_cleanup(struct tcpcb *tp)
3539 {
3540 	struct syn_cache *sc, *nsc;
3541 	int s;
3542 
3543 	s = splsoftnet();
3544 
3545 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3546 		nsc = LIST_NEXT(sc, sc_tpq);
3547 
3548 #ifdef DIAGNOSTIC
3549 		if (sc->sc_tp != tp)
3550 			panic("invalid sc_tp in syn_cache_cleanup");
3551 #endif
3552 		SYN_CACHE_RM(sc);
3553 		SYN_CACHE_PUT(sc);
3554 	}
3555 	/* just for safety */
3556 	LIST_INIT(&tp->t_sc);
3557 
3558 	splx(s);
3559 }
3560 
3561 /*
3562  * Find an entry in the syn cache.
3563  */
3564 struct syn_cache *
3565 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst,
3566     struct syn_cache_head **headp)
3567 {
3568 	struct syn_cache *sc;
3569 	struct syn_cache_head *scp;
3570 	u_int32_t hash;
3571 	int s;
3572 
3573 	SYN_HASHALL(hash, src, dst);
3574 
3575 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3576 	*headp = scp;
3577 	s = splsoftnet();
3578 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3579 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3580 		if (sc->sc_hash != hash)
3581 			continue;
3582 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3583 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3584 			splx(s);
3585 			return (sc);
3586 		}
3587 	}
3588 	splx(s);
3589 	return (NULL);
3590 }
3591 
3592 /*
3593  * This function gets called when we receive an ACK for a
3594  * socket in the LISTEN state.  We look up the connection
3595  * in the syn cache, and if its there, we pull it out of
3596  * the cache and turn it into a full-blown connection in
3597  * the SYN-RECEIVED state.
3598  *
3599  * The return values may not be immediately obvious, and their effects
3600  * can be subtle, so here they are:
3601  *
3602  *	NULL	SYN was not found in cache; caller should drop the
3603  *		packet and send an RST.
3604  *
3605  *	-1	We were unable to create the new connection, and are
3606  *		aborting it.  An ACK,RST is being sent to the peer
3607  *		(unless we got screwey sequence numbners; see below),
3608  *		because the 3-way handshake has been completed.  Caller
3609  *		should not free the mbuf, since we may be using it.  If
3610  *		we are not, we will free it.
3611  *
3612  *	Otherwise, the return value is a pointer to the new socket
3613  *	associated with the connection.
3614  */
3615 struct socket *
3616 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3617     u_int hlen, u_int tlen, struct socket *so, struct mbuf *m)
3618 {
3619 	struct syn_cache *sc;
3620 	struct syn_cache_head *scp;
3621 	struct inpcb *inp = NULL;
3622 	struct tcpcb *tp = 0;
3623 	struct mbuf *am;
3624 	int s;
3625 	struct socket *oso;
3626 
3627 	s = splsoftnet();
3628 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3629 		splx(s);
3630 		return (NULL);
3631 	}
3632 
3633 	/*
3634 	 * Verify the sequence and ack numbers.  Try getting the correct
3635 	 * response again.
3636 	 */
3637 	if ((th->th_ack != sc->sc_iss + 1) ||
3638 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3639 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3640 		(void) syn_cache_respond(sc, m);
3641 		splx(s);
3642 		return ((struct socket *)(-1));
3643 	}
3644 
3645 	/* Remove this cache entry */
3646 	SYN_CACHE_RM(sc);
3647 	splx(s);
3648 
3649 	/*
3650 	 * Ok, create the full blown connection, and set things up
3651 	 * as they would have been set up if we had created the
3652 	 * connection when the SYN arrived.  If we can't create
3653 	 * the connection, abort it.
3654 	 */
3655 	oso = so;
3656 	so = sonewconn(so, SS_ISCONNECTED);
3657 	if (so == NULL)
3658 		goto resetandabort;
3659 
3660 	inp = sotoinpcb(oso);
3661 #ifdef IPSEC
3662 	/*
3663 	 * We need to copy the required security levels
3664 	 * from the old pcb. Ditto for any other
3665 	 * IPsec-related information.
3666 	 */
3667 	{
3668 	  struct inpcb *newinp = (struct inpcb *)so->so_pcb;
3669 	  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
3670 		sizeof(inp->inp_seclevel));
3671 	  newinp->inp_secrequire = inp->inp_secrequire;
3672 	  if (inp->inp_ipo != NULL) {
3673 		  newinp->inp_ipo = inp->inp_ipo;
3674 		  inp->inp_ipo->ipo_ref_count++;
3675 	  }
3676 	  if (inp->inp_ipsec_remotecred != NULL) {
3677 		  newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred;
3678 		  inp->inp_ipsec_remotecred->ref_count++;
3679 	  }
3680 	  if (inp->inp_ipsec_remoteauth != NULL) {
3681 		  newinp->inp_ipsec_remoteauth
3682 		      = inp->inp_ipsec_remoteauth;
3683 		  inp->inp_ipsec_remoteauth->ref_count++;
3684 	  }
3685 	}
3686 #endif /* IPSEC */
3687 #ifdef INET6
3688 	/*
3689 	 * inp still has the OLD in_pcb stuff, set the
3690 	 * v6-related flags on the new guy, too.
3691 	 */
3692 	{
3693 	  int flags = inp->inp_flags;
3694 	  struct inpcb *oldinpcb = inp;
3695 
3696 	  inp = (struct inpcb *)so->so_pcb;
3697 	  inp->inp_flags |= (flags & INP_IPV6);
3698 	  if ((inp->inp_flags & INP_IPV6) != 0) {
3699 	    inp->inp_ipv6.ip6_hlim =
3700 	      oldinpcb->inp_ipv6.ip6_hlim;
3701 	  }
3702 	}
3703 #else /* INET6 */
3704 	inp = (struct inpcb *)so->so_pcb;
3705 #endif /* INET6 */
3706 
3707 	inp->inp_lport = th->th_dport;
3708 	switch (src->sa_family) {
3709 #ifdef INET6
3710 	case AF_INET6:
3711 		inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr;
3712 		break;
3713 #endif /* INET6 */
3714 	case AF_INET:
3715 
3716 		inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3717 		inp->inp_options = ip_srcroute();
3718 		if (inp->inp_options == NULL) {
3719 			inp->inp_options = sc->sc_ipopts;
3720 			sc->sc_ipopts = NULL;
3721 		}
3722 		break;
3723 	}
3724 	in_pcbrehash(inp);
3725 
3726 	/*
3727 	 * Give the new socket our cached route reference.
3728 	 */
3729 	if (src->sa_family == AF_INET)
3730 		inp->inp_route = sc->sc_route4;         /* struct assignment */
3731 #ifdef INET6
3732 	else
3733 		inp->inp_route6 = sc->sc_route6;
3734 #endif
3735 	sc->sc_route4.ro_rt = NULL;
3736 
3737 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3738 	if (am == NULL)
3739 		goto resetandabort;
3740 	am->m_len = src->sa_len;
3741 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3742 
3743 	switch (src->sa_family) {
3744 	case AF_INET:
3745 		/* drop IPv4 packet to AF_INET6 socket */
3746 		if (inp->inp_flags & INP_IPV6) {
3747 			(void) m_free(am);
3748 			goto resetandabort;
3749 		}
3750 		if (in_pcbconnect(inp, am)) {
3751 			(void) m_free(am);
3752 			goto resetandabort;
3753 		}
3754 		break;
3755 #ifdef INET6
3756 	case AF_INET6:
3757 		if (in6_pcbconnect(inp, am)) {
3758 			(void) m_free(am);
3759 			goto resetandabort;
3760 		}
3761 		break;
3762 #endif
3763 	}
3764 	(void) m_free(am);
3765 
3766 	tp = intotcpcb(inp);
3767 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3768 	if (sc->sc_request_r_scale != 15) {
3769 		tp->requested_s_scale = sc->sc_requested_s_scale;
3770 		tp->request_r_scale = sc->sc_request_r_scale;
3771 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3772 	}
3773 	if (sc->sc_flags & SCF_TIMESTAMP)
3774 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3775 
3776 	tp->t_template = tcp_template(tp);
3777 	if (tp->t_template == 0) {
3778 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3779 		so = NULL;
3780 		m_freem(m);
3781 		goto abort;
3782 	}
3783 #ifdef TCP_SACK
3784 	tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT;
3785 #endif
3786 
3787 	tp->ts_modulate = sc->sc_modulate;
3788 	tp->iss = sc->sc_iss;
3789 	tp->irs = sc->sc_irs;
3790 	tcp_sendseqinit(tp);
3791 #if defined (TCP_SACK) || defined(TCP_ECN)
3792 	tp->snd_last = tp->snd_una;
3793 #endif /* TCP_SACK */
3794 #if defined(TCP_SACK) && defined(TCP_FACK)
3795 	tp->snd_fack = tp->snd_una;
3796 	tp->retran_data = 0;
3797 	tp->snd_awnd = 0;
3798 #endif /* TCP_FACK */
3799 #ifdef TCP_ECN
3800 	if (sc->sc_flags & SCF_ECN_PERMIT) {
3801 		tp->t_flags |= TF_ECN_PERMIT;
3802 		tcpstat.tcps_ecn_accepts++;
3803 	}
3804 #endif
3805 #ifdef TCP_SACK
3806 	if (sc->sc_flags & SCF_SACK_PERMIT)
3807 		tp->t_flags |= TF_SACK_PERMIT;
3808 #endif
3809 #ifdef TCP_SIGNATURE
3810 	if (sc->sc_flags & SCF_SIGNATURE)
3811 		tp->t_flags |= TF_SIGNATURE;
3812 #endif
3813 	tcp_rcvseqinit(tp);
3814 	tp->t_state = TCPS_SYN_RECEIVED;
3815 	tp->t_rcvtime = tcp_now;
3816 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
3817 	tcpstat.tcps_accepts++;
3818 
3819 	tcp_mss(tp, sc->sc_peermaxseg);	 /* sets t_maxseg */
3820 	if (sc->sc_peermaxseg)
3821 		tcp_mss_update(tp);
3822 	/* Reset initial window to 1 segment for retransmit */
3823 	if (sc->sc_rxtshift > 0)
3824 		tp->snd_cwnd = tp->t_maxseg;
3825 	tp->snd_wl1 = sc->sc_irs;
3826 	tp->rcv_up = sc->sc_irs + 1;
3827 
3828 	/*
3829 	 * This is what whould have happened in tcp_output() when
3830 	 * the SYN,ACK was sent.
3831 	 */
3832 	tp->snd_up = tp->snd_una;
3833 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3834 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3835 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3836 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3837 	tp->last_ack_sent = tp->rcv_nxt;
3838 
3839 	tcpstat.tcps_sc_completed++;
3840 	SYN_CACHE_PUT(sc);
3841 	return (so);
3842 
3843 resetandabort:
3844 	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST);
3845 	m_freem(m);
3846 abort:
3847 	if (so != NULL)
3848 		(void) soabort(so);
3849 	SYN_CACHE_PUT(sc);
3850 	tcpstat.tcps_sc_aborted++;
3851 	return ((struct socket *)(-1));
3852 }
3853 
3854 /*
3855  * This function is called when we get a RST for a
3856  * non-existent connection, so that we can see if the
3857  * connection is in the syn cache.  If it is, zap it.
3858  */
3859 
3860 void
3861 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3862 {
3863 	struct syn_cache *sc;
3864 	struct syn_cache_head *scp;
3865 	int s = splsoftnet();
3866 
3867 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3868 		splx(s);
3869 		return;
3870 	}
3871 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3872 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3873 		splx(s);
3874 		return;
3875 	}
3876 	SYN_CACHE_RM(sc);
3877 	splx(s);
3878 	tcpstat.tcps_sc_reset++;
3879 	SYN_CACHE_PUT(sc);
3880 }
3881 
3882 void
3883 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th)
3884 {
3885 	struct syn_cache *sc;
3886 	struct syn_cache_head *scp;
3887 	int s;
3888 
3889 	s = splsoftnet();
3890 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3891 		splx(s);
3892 		return;
3893 	}
3894 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3895 	if (ntohl (th->th_seq) != sc->sc_iss) {
3896 		splx(s);
3897 		return;
3898 	}
3899 
3900 	/*
3901 	 * If we've retransmitted 3 times and this is our second error,
3902 	 * we remove the entry.  Otherwise, we allow it to continue on.
3903 	 * This prevents us from incorrectly nuking an entry during a
3904 	 * spurious network outage.
3905 	 *
3906 	 * See tcp_notify().
3907 	 */
3908 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3909 		sc->sc_flags |= SCF_UNREACH;
3910 		splx(s);
3911 		return;
3912 	}
3913 
3914 	SYN_CACHE_RM(sc);
3915 	splx(s);
3916 	tcpstat.tcps_sc_unreach++;
3917 	SYN_CACHE_PUT(sc);
3918 }
3919 
3920 /*
3921  * Given a LISTEN socket and an inbound SYN request, add
3922  * this to the syn cache, and send back a segment:
3923  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3924  * to the source.
3925  *
3926  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3927  * Doing so would require that we hold onto the data and deliver it
3928  * to the application.  However, if we are the target of a SYN-flood
3929  * DoS attack, an attacker could send data which would eventually
3930  * consume all available buffer space if it were ACKed.  By not ACKing
3931  * the data, we avoid this DoS scenario.
3932  */
3933 
3934 int
3935 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3936     u_int iphlen, struct socket *so, struct mbuf *m, u_char *optp, int optlen,
3937     struct tcp_opt_info *oi, tcp_seq *issp)
3938 {
3939 	struct tcpcb tb, *tp;
3940 	long win;
3941 	struct syn_cache *sc;
3942 	struct syn_cache_head *scp;
3943 	struct mbuf *ipopts;
3944 
3945 	tp = sototcpcb(so);
3946 
3947 	/*
3948 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3949 	 *
3950 	 * Note this check is performed in tcp_input() very early on.
3951 	 */
3952 
3953 	/*
3954 	 * Initialize some local state.
3955 	 */
3956 	win = sbspace(&so->so_rcv);
3957 	if (win > TCP_MAXWIN)
3958 		win = TCP_MAXWIN;
3959 
3960 #ifdef TCP_SIGNATURE
3961 	if (optp || (tp->t_flags & TF_SIGNATURE)) {
3962 #else
3963 	if (optp) {
3964 #endif
3965 		tb.pf = tp->pf;
3966 #ifdef TCP_SACK
3967 		tb.sack_enable = tp->sack_enable;
3968 #endif
3969 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3970 #ifdef TCP_SIGNATURE
3971 		if (tp->t_flags & TF_SIGNATURE)
3972 			tb.t_flags |= TF_SIGNATURE;
3973 #endif
3974 		tb.t_state = TCPS_LISTEN;
3975 		if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi))
3976 			return (0);
3977 	} else
3978 		tb.t_flags = 0;
3979 
3980 	switch (src->sa_family) {
3981 #ifdef INET
3982 	case AF_INET:
3983 		/*
3984 		 * Remember the IP options, if any.
3985 		 */
3986 		ipopts = ip_srcroute();
3987 		break;
3988 #endif
3989 	default:
3990 		ipopts = NULL;
3991 	}
3992 
3993 	/*
3994 	 * See if we already have an entry for this connection.
3995 	 * If we do, resend the SYN,ACK.  We do not count this
3996 	 * as a retransmission (XXX though maybe we should).
3997 	 */
3998 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
3999 		tcpstat.tcps_sc_dupesyn++;
4000 		if (ipopts) {
4001 			/*
4002 			 * If we were remembering a previous source route,
4003 			 * forget it and use the new one we've been given.
4004 			 */
4005 			if (sc->sc_ipopts)
4006 				(void) m_free(sc->sc_ipopts);
4007 			sc->sc_ipopts = ipopts;
4008 		}
4009 		sc->sc_timestamp = tb.ts_recent;
4010 		if (syn_cache_respond(sc, m) == 0) {
4011 			tcpstat.tcps_sndacks++;
4012 			tcpstat.tcps_sndtotal++;
4013 		}
4014 		return (1);
4015 	}
4016 
4017 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4018 	if (sc == NULL) {
4019 		if (ipopts)
4020 			(void) m_free(ipopts);
4021 		return (0);
4022 	}
4023 
4024 	/*
4025 	 * Fill in the cache, and put the necessary IP and TCP
4026 	 * options into the reply.
4027 	 */
4028 	bzero(sc, sizeof(struct syn_cache));
4029 	bzero(&sc->sc_timer, sizeof(sc->sc_timer));
4030 	bcopy(src, &sc->sc_src, src->sa_len);
4031 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4032 	sc->sc_flags = 0;
4033 	sc->sc_ipopts = ipopts;
4034 	sc->sc_irs = th->th_seq;
4035 
4036 	sc->sc_iss = issp ? *issp : arc4random();
4037 	sc->sc_peermaxseg = oi->maxseg;
4038 	sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ?
4039 	    m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family);
4040 	sc->sc_win = win;
4041 	sc->sc_timestamp = tb.ts_recent;
4042 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4043 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
4044 		sc->sc_flags |= SCF_TIMESTAMP;
4045 		sc->sc_modulate = arc4random();
4046 	}
4047 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4048 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4049 		sc->sc_requested_s_scale = tb.requested_s_scale;
4050 		sc->sc_request_r_scale = 0;
4051 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4052 		    TCP_MAXWIN << sc->sc_request_r_scale <
4053 		    so->so_rcv.sb_hiwat)
4054 			sc->sc_request_r_scale++;
4055 	} else {
4056 		sc->sc_requested_s_scale = 15;
4057 		sc->sc_request_r_scale = 15;
4058 	}
4059 #ifdef TCP_ECN
4060 	/*
4061 	 * if both ECE and CWR flag bits are set, peer is ECN capable.
4062 	 */
4063 	if (tcp_do_ecn &&
4064 	    (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR))
4065 		sc->sc_flags |= SCF_ECN_PERMIT;
4066 #endif
4067 #ifdef TCP_SACK
4068 	/*
4069 	 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option
4070 	 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT).
4071 	 */
4072 	if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT))
4073 		sc->sc_flags |= SCF_SACK_PERMIT;
4074 #endif
4075 #ifdef TCP_SIGNATURE
4076 	if (tb.t_flags & TF_SIGNATURE)
4077 		sc->sc_flags |= SCF_SIGNATURE;
4078 #endif
4079 	sc->sc_tp = tp;
4080 	if (syn_cache_respond(sc, m) == 0) {
4081 		syn_cache_insert(sc, tp);
4082 		tcpstat.tcps_sndacks++;
4083 		tcpstat.tcps_sndtotal++;
4084 	} else {
4085 		SYN_CACHE_PUT(sc);
4086 		tcpstat.tcps_sc_dropped++;
4087 	}
4088 	return (1);
4089 }
4090 
4091 int
4092 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4093 {
4094 	struct route *ro;
4095 	u_int8_t *optp;
4096 	int optlen, error;
4097 	u_int16_t tlen;
4098 	struct ip *ip = NULL;
4099 #ifdef INET6
4100 	struct ip6_hdr *ip6 = NULL;
4101 #endif
4102 	struct tcphdr *th;
4103 	u_int hlen;
4104 	struct inpcb *inp;
4105 
4106 	switch (sc->sc_src.sa.sa_family) {
4107 	case AF_INET:
4108 		hlen = sizeof(struct ip);
4109 		ro = &sc->sc_route4;
4110 		break;
4111 #ifdef INET6
4112 	case AF_INET6:
4113 		hlen = sizeof(struct ip6_hdr);
4114 		ro = (struct route *)&sc->sc_route6;
4115 		break;
4116 #endif
4117 	default:
4118 		if (m)
4119 			m_freem(m);
4120 		return (EAFNOSUPPORT);
4121 	}
4122 
4123 	/* Compute the size of the TCP options. */
4124 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4125 #ifdef TCP_SACK
4126 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) +
4127 #endif
4128 #ifdef TCP_SIGNATURE
4129 	    ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) +
4130 #endif
4131 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4132 
4133 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4134 
4135 	/*
4136 	 * Create the IP+TCP header from scratch.
4137 	 */
4138 	if (m)
4139 		m_freem(m);
4140 #ifdef DIAGNOSTIC
4141 	if (max_linkhdr + tlen > MCLBYTES)
4142 		return (ENOBUFS);
4143 #endif
4144 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4145 	if (m && max_linkhdr + tlen > MHLEN) {
4146 		MCLGET(m, M_DONTWAIT);
4147 		if ((m->m_flags & M_EXT) == 0) {
4148 			m_freem(m);
4149 			m = NULL;
4150 		}
4151 	}
4152 	if (m == NULL)
4153 		return (ENOBUFS);
4154 
4155 	/* Fixup the mbuf. */
4156 	m->m_data += max_linkhdr;
4157 	m->m_len = m->m_pkthdr.len = tlen;
4158 	m->m_pkthdr.rcvif = NULL;
4159 	memset(mtod(m, u_char *), 0, tlen);
4160 
4161 	switch (sc->sc_src.sa.sa_family) {
4162 	case AF_INET:
4163 		ip = mtod(m, struct ip *);
4164 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4165 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4166 		ip->ip_p = IPPROTO_TCP;
4167 		th = (struct tcphdr *)(ip + 1);
4168 		th->th_dport = sc->sc_src.sin.sin_port;
4169 		th->th_sport = sc->sc_dst.sin.sin_port;
4170 		break;
4171 #ifdef INET6
4172 	case AF_INET6:
4173 		ip6 = mtod(m, struct ip6_hdr *);
4174 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4175 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4176 		ip6->ip6_nxt = IPPROTO_TCP;
4177 		/* ip6_plen will be updated in ip6_output() */
4178 		th = (struct tcphdr *)(ip6 + 1);
4179 		th->th_dport = sc->sc_src.sin6.sin6_port;
4180 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4181 		break;
4182 #endif
4183 	default:
4184 		th = NULL;
4185 	}
4186 
4187 	th->th_seq = htonl(sc->sc_iss);
4188 	th->th_ack = htonl(sc->sc_irs + 1);
4189 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4190 	th->th_flags = TH_SYN|TH_ACK;
4191 #ifdef TCP_ECN
4192 	/* Set ECE for SYN-ACK if peer supports ECN. */
4193 	if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT))
4194 		th->th_flags |= TH_ECE;
4195 #endif
4196 	th->th_win = htons(sc->sc_win);
4197 	/* th_sum already 0 */
4198 	/* th_urp already 0 */
4199 
4200 	/* Tack on the TCP options. */
4201 	optp = (u_int8_t *)(th + 1);
4202 	*optp++ = TCPOPT_MAXSEG;
4203 	*optp++ = 4;
4204 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4205 	*optp++ = sc->sc_ourmaxseg & 0xff;
4206 
4207 #ifdef TCP_SACK
4208 	/* Include SACK_PERMIT_HDR option if peer has already done so. */
4209 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4210 		*((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR);
4211 		optp += 4;
4212 	}
4213 #endif
4214 
4215 	if (sc->sc_request_r_scale != 15) {
4216 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4217 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4218 		    sc->sc_request_r_scale);
4219 		optp += 4;
4220 	}
4221 
4222 	if (sc->sc_flags & SCF_TIMESTAMP) {
4223 		u_int32_t *lp = (u_int32_t *)(optp);
4224 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4225 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4226 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4227 		*lp   = htonl(sc->sc_timestamp);
4228 		optp += TCPOLEN_TSTAMP_APPA;
4229 	}
4230 
4231 #ifdef TCP_SIGNATURE
4232 	if (sc->sc_flags & SCF_SIGNATURE) {
4233 		union sockaddr_union src, dst;
4234 		struct tdb *tdb;
4235 
4236 		bzero(&src, sizeof(union sockaddr_union));
4237 		bzero(&dst, sizeof(union sockaddr_union));
4238 		src.sa.sa_len = sc->sc_src.sa.sa_len;
4239 		src.sa.sa_family = sc->sc_src.sa.sa_family;
4240 		dst.sa.sa_len = sc->sc_dst.sa.sa_len;
4241 		dst.sa.sa_family = sc->sc_dst.sa.sa_family;
4242 
4243 		switch (sc->sc_src.sa.sa_family) {
4244 		case 0:	/*default to PF_INET*/
4245 #ifdef INET
4246 		case AF_INET:
4247 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
4248 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
4249 			break;
4250 #endif /* INET */
4251 #ifdef INET6
4252 		case AF_INET6:
4253 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
4254 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
4255 			break;
4256 #endif /* INET6 */
4257 		}
4258 
4259 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
4260 		if (tdb == NULL) {
4261 			if (m)
4262 				m_freem(m);
4263 			return (EPERM);
4264 		}
4265 
4266 		/* Send signature option */
4267 		*(optp++) = TCPOPT_SIGNATURE;
4268 		*(optp++) = TCPOLEN_SIGNATURE;
4269 
4270 		if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th,
4271 		    hlen, 0, optp) < 0) {
4272 			if (m)
4273 				m_freem(m);
4274 			return (EINVAL);
4275 		}
4276 		optp += 16;
4277 
4278 		/* Pad options list to the next 32 bit boundary and
4279 		 * terminate it.
4280 		 */
4281 		*optp++ = TCPOPT_NOP;
4282 		*optp++ = TCPOPT_EOL;
4283 	}
4284 #endif /* TCP_SIGNATURE */
4285 
4286 	/* Compute the packet's checksum. */
4287 	switch (sc->sc_src.sa.sa_family) {
4288 	case AF_INET:
4289 		ip->ip_len = htons(tlen - hlen);
4290 		th->th_sum = 0;
4291 		th->th_sum = in_cksum(m, tlen);
4292 		break;
4293 #ifdef INET6
4294 	case AF_INET6:
4295 		ip6->ip6_plen = htons(tlen - hlen);
4296 		th->th_sum = 0;
4297 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4298 		break;
4299 #endif
4300 	}
4301 
4302 	/* use IPsec policy and ttl from listening socket, on SYN ACK */
4303 	inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL;
4304 
4305 	/*
4306 	 * Fill in some straggling IP bits.  Note the stack expects
4307 	 * ip_len to be in host order, for convenience.
4308 	 */
4309 	switch (sc->sc_src.sa.sa_family) {
4310 #ifdef INET
4311 	case AF_INET:
4312 		ip->ip_len = htons(tlen);
4313 		ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl;
4314 		/* XXX tos? */
4315 		break;
4316 #endif
4317 #ifdef INET6
4318 	case AF_INET6:
4319 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4320 		ip6->ip6_vfc |= IPV6_VERSION;
4321 		ip6->ip6_plen = htons(tlen - hlen);
4322 		/* ip6_hlim will be initialized afterwards */
4323 		/* leave flowlabel = 0, it is legal and require no state mgmt */
4324 		break;
4325 #endif
4326 	}
4327 
4328 	switch (sc->sc_src.sa.sa_family) {
4329 #ifdef INET
4330 	case AF_INET:
4331 		error = ip_output(m, sc->sc_ipopts, ro,
4332 		    (ip_mtudisc ? IP_MTUDISC : 0),
4333 		    (struct ip_moptions *)NULL, inp);
4334 		break;
4335 #endif
4336 #ifdef INET6
4337 	case AF_INET6:
4338 		ip6->ip6_hlim = in6_selecthlim(NULL,
4339 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4340 
4341 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4342 			(struct ip6_moptions *)0, NULL, NULL);
4343 		break;
4344 #endif
4345 	default:
4346 		error = EAFNOSUPPORT;
4347 		break;
4348 	}
4349 	return (error);
4350 }
4351