xref: /openbsd-src/sys/netinet/tcp_input.c (revision 66ad965f4873a0970dea06fb53c307b8385e5a94)
1 /*	$OpenBSD: tcp_input.c,v 1.212 2008/02/20 11:24:02 markus 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 struct	tcpiphdr tcp_saveti;
99 
100 int tcp_mss_adv(struct ifnet *, int);
101 
102 #ifdef INET6
103 #include <netinet6/in6_var.h>
104 #include <netinet6/nd6.h>
105 
106 struct  tcpipv6hdr tcp_saveti6;
107 
108 /* for the packet header length in the mbuf */
109 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
110 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
111 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
112 #endif /* INET6 */
113 
114 int	tcprexmtthresh = 3;
115 int	tcptv_keep_init = TCPTV_KEEP_INIT;
116 
117 extern u_long sb_max;
118 
119 int tcp_rst_ppslim = 100;		/* 100pps */
120 int tcp_rst_ppslim_count = 0;
121 struct timeval tcp_rst_ppslim_last;
122 
123 int tcp_ackdrop_ppslim = 100;		/* 100pps */
124 int tcp_ackdrop_ppslim_count = 0;
125 struct timeval tcp_ackdrop_ppslim_last;
126 
127 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
128 
129 /* for modulo comparisons of timestamps */
130 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
131 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
132 
133 /* for TCP SACK comparisons */
134 #define	SEQ_MIN(a,b)	(SEQ_LT(a,b) ? (a) : (b))
135 #define	SEQ_MAX(a,b)	(SEQ_GT(a,b) ? (a) : (b))
136 
137 /*
138  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
139  */
140 #ifdef INET6
141 #define ND6_HINT(tp) \
142 do { \
143 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
144 	    tp->t_inpcb->inp_route6.ro_rt) { \
145 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \
146 	} \
147 } while (0)
148 #else
149 #define ND6_HINT(tp)
150 #endif
151 
152 #ifdef TCP_ECN
153 /*
154  * ECN (Explicit Congestion Notification) support based on RFC3168
155  * implementation note:
156  *   snd_last is used to track a recovery phase.
157  *   when cwnd is reduced, snd_last is set to snd_max.
158  *   while snd_last > snd_una, the sender is in a recovery phase and
159  *   its cwnd should not be reduced again.
160  *   snd_last follows snd_una when not in a recovery phase.
161  */
162 #endif
163 
164 /*
165  * Macro to compute ACK transmission behavior.  Delay the ACK unless
166  * we have already delayed an ACK (must send an ACK every two segments).
167  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
168  * option is enabled.
169  */
170 #define	TCP_SETUP_ACK(tp, tiflags) \
171 do { \
172 	if ((tp)->t_flags & TF_DELACK || \
173 	    (tcp_ack_on_push && (tiflags) & TH_PUSH)) \
174 		tp->t_flags |= TF_ACKNOW; \
175 	else \
176 		TCP_SET_DELACK(tp); \
177 } while (0)
178 
179 /*
180  * Insert segment ti into reassembly queue of tcp with
181  * control block tp.  Return TH_FIN if reassembly now includes
182  * a segment with FIN.  The macro form does the common case inline
183  * (segment is the next to be received on an established connection,
184  * and the queue is empty), avoiding linkage into and removal
185  * from the queue and repetition of various conversions.
186  * Set DELACK for segments received in order, but ack immediately
187  * when segments are out of order (so fast retransmit can work).
188  */
189 
190 int
191 tcp_reass(tp, th, m, tlen)
192 	struct tcpcb *tp;
193 	struct tcphdr *th;
194 	struct mbuf *m;
195 	int *tlen;
196 {
197 	struct tcpqent *p, *q, *nq, *tiqe;
198 	struct socket *so = tp->t_inpcb->inp_socket;
199 	int flags;
200 
201 	/*
202 	 * Call with th==0 after become established to
203 	 * force pre-ESTABLISHED data up to user socket.
204 	 */
205 	if (th == 0)
206 		goto present;
207 
208 	/*
209 	 * Allocate a new queue entry, before we throw away any data.
210 	 * If we can't, just drop the packet.  XXX
211 	 */
212 	tiqe = pool_get(&tcpqe_pool, PR_NOWAIT);
213 	if (tiqe == NULL) {
214 		tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead);
215 		if (tiqe != NULL && th->th_seq == tp->rcv_nxt) {
216 			/* Reuse last entry since new segment fills a hole */
217 			m_freem(tiqe->tcpqe_m);
218 			TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q);
219 		}
220 		if (tiqe == NULL || th->th_seq != tp->rcv_nxt) {
221 			/* Flush segment queue for this connection */
222 			tcp_freeq(tp);
223 			tcpstat.tcps_rcvmemdrop++;
224 			m_freem(m);
225 			return (0);
226 		}
227 	}
228 
229 	/*
230 	 * Find a segment which begins after this one does.
231 	 */
232 	for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL;
233 	    p = q, q = TAILQ_NEXT(q, tcpqe_q))
234 		if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq))
235 			break;
236 
237 	/*
238 	 * If there is a preceding segment, it may provide some of
239 	 * our data already.  If so, drop the data from the incoming
240 	 * segment.  If it provides all of our data, drop us.
241 	 */
242 	if (p != NULL) {
243 		struct tcphdr *phdr = p->tcpqe_tcp;
244 		int i;
245 
246 		/* conversion to int (in i) handles seq wraparound */
247 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
248 		if (i > 0) {
249 		        if (i >= *tlen) {
250 				tcpstat.tcps_rcvduppack++;
251 				tcpstat.tcps_rcvdupbyte += *tlen;
252 				m_freem(m);
253 				pool_put(&tcpqe_pool, tiqe);
254 				return (0);
255 			}
256 			m_adj(m, i);
257 			*tlen -= i;
258 			th->th_seq += i;
259 		}
260 	}
261 	tcpstat.tcps_rcvoopack++;
262 	tcpstat.tcps_rcvoobyte += *tlen;
263 
264 	/*
265 	 * While we overlap succeeding segments trim them or,
266 	 * if they are completely covered, dequeue them.
267 	 */
268 	for (; q != NULL; q = nq) {
269 		struct tcphdr *qhdr = q->tcpqe_tcp;
270 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
271 
272 		if (i <= 0)
273 			break;
274 		if (i < qhdr->th_reseqlen) {
275 			qhdr->th_seq += i;
276 			qhdr->th_reseqlen -= i;
277 			m_adj(q->tcpqe_m, i);
278 			break;
279 		}
280 		nq = TAILQ_NEXT(q, tcpqe_q);
281 		m_freem(q->tcpqe_m);
282 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
283 		pool_put(&tcpqe_pool, q);
284 	}
285 
286 	/* Insert the new segment queue entry into place. */
287 	tiqe->tcpqe_m = m;
288 	th->th_reseqlen = *tlen;
289 	tiqe->tcpqe_tcp = th;
290 	if (p == NULL) {
291 		TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q);
292 	} else {
293 		TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q);
294 	}
295 
296 present:
297 	/*
298 	 * Present data to user, advancing rcv_nxt through
299 	 * completed sequence space.
300 	 */
301 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
302 		return (0);
303 	q = TAILQ_FIRST(&tp->t_segq);
304 	if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt)
305 		return (0);
306 	if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen)
307 		return (0);
308 	do {
309 		tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen;
310 		flags = q->tcpqe_tcp->th_flags & TH_FIN;
311 
312 		nq = TAILQ_NEXT(q, tcpqe_q);
313 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
314 		ND6_HINT(tp);
315 		if (so->so_state & SS_CANTRCVMORE)
316 			m_freem(q->tcpqe_m);
317 		else
318 			sbappendstream(&so->so_rcv, q->tcpqe_m);
319 		pool_put(&tcpqe_pool, q);
320 		q = nq;
321 	} while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt);
322 	sorwakeup(so);
323 	return (flags);
324 }
325 
326 #ifdef INET6
327 int
328 tcp6_input(mp, offp, proto)
329 	struct mbuf **mp;
330 	int *offp, proto;
331 {
332 	struct mbuf *m = *mp;
333 
334 #if defined(NFAITH) && 0 < NFAITH
335 	if (m->m_pkthdr.rcvif) {
336 		if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
337 			/* XXX send icmp6 host/port unreach? */
338 			m_freem(m);
339 			return IPPROTO_DONE;
340 		}
341 	}
342 #endif
343 
344 	/*
345 	 * draft-itojun-ipv6-tcp-to-anycast
346 	 * better place to put this in?
347 	 */
348 	if (m->m_flags & M_ANYCAST6) {
349 		if (m->m_len >= sizeof(struct ip6_hdr)) {
350 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
351 			icmp6_error(m, ICMP6_DST_UNREACH,
352 				ICMP6_DST_UNREACH_ADDR,
353 				(caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
354 		} else
355 			m_freem(m);
356 		return IPPROTO_DONE;
357 	}
358 
359 	tcp_input(m, *offp, proto);
360 	return IPPROTO_DONE;
361 }
362 #endif
363 
364 /*
365  * TCP input routine, follows pages 65-76 of the
366  * protocol specification dated September, 1981 very closely.
367  */
368 void
369 tcp_input(struct mbuf *m, ...)
370 {
371 	struct ip *ip;
372 	struct inpcb *inp;
373 	u_int8_t *optp = NULL;
374 	int optlen = 0;
375 	int tlen, off;
376 	struct tcpcb *tp = 0;
377 	int tiflags;
378 	struct socket *so = NULL;
379 	int todrop, acked, ourfinisacked, needoutput = 0;
380 	int hdroptlen = 0;
381 	short ostate = 0;
382 	tcp_seq iss, *reuse = NULL;
383 	u_long tiwin;
384 	struct tcp_opt_info opti;
385 	int iphlen;
386 	va_list ap;
387 	struct tcphdr *th;
388 #ifdef INET6
389 	struct ip6_hdr *ip6 = NULL;
390 #endif /* INET6 */
391 #ifdef IPSEC
392 	struct m_tag *mtag;
393 	struct tdb_ident *tdbi;
394 	struct tdb *tdb;
395 	int error, s;
396 #endif /* IPSEC */
397 	int af;
398 #ifdef TCP_ECN
399 	u_char iptos;
400 #endif
401 
402 	va_start(ap, m);
403 	iphlen = va_arg(ap, int);
404 	va_end(ap);
405 
406 	tcpstat.tcps_rcvtotal++;
407 
408 	opti.ts_present = 0;
409 	opti.maxseg = 0;
410 
411 	/*
412 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
413 	 * See below for AF specific multicast.
414 	 */
415 	if (m->m_flags & (M_BCAST|M_MCAST))
416 		goto drop;
417 
418 	/*
419 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
420 	 * TCP/IPv4.
421 	 */
422 	switch (mtod(m, struct ip *)->ip_v) {
423 #ifdef INET6
424 	case 6:
425 		af = AF_INET6;
426 		break;
427 #endif
428 	case 4:
429 		af = AF_INET;
430 		break;
431 	default:
432 		m_freem(m);
433 		return;	/*EAFNOSUPPORT*/
434 	}
435 
436 	/*
437 	 * Get IP and TCP header together in first mbuf.
438 	 * Note: IP leaves IP header in first mbuf.
439 	 */
440 	switch (af) {
441 	case AF_INET:
442 #ifdef DIAGNOSTIC
443 		if (iphlen < sizeof(struct ip)) {
444 			m_freem(m);
445 			return;
446 		}
447 #endif /* DIAGNOSTIC */
448 		break;
449 #ifdef INET6
450 	case AF_INET6:
451 #ifdef DIAGNOSTIC
452 		if (iphlen < sizeof(struct ip6_hdr)) {
453 			m_freem(m);
454 			return;
455 		}
456 #endif /* DIAGNOSTIC */
457 		break;
458 #endif
459 	default:
460 		m_freem(m);
461 		return;
462 	}
463 
464 	IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th));
465 	if (!th) {
466 		tcpstat.tcps_rcvshort++;
467 		return;
468 	}
469 
470 	tlen = m->m_pkthdr.len - iphlen;
471 	ip = NULL;
472 #ifdef INET6
473 	ip6 = NULL;
474 #endif
475 	switch (af) {
476 	case AF_INET:
477 		ip = mtod(m, struct ip *);
478 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
479 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
480 			goto drop;
481 #ifdef TCP_ECN
482 		/* save ip_tos before clearing it for checksum */
483 		iptos = ip->ip_tos;
484 #endif
485 		/*
486 		 * Checksum extended TCP header and data.
487 		 */
488 		if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) {
489 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) {
490 				tcpstat.tcps_inhwcsum++;
491 				tcpstat.tcps_rcvbadsum++;
492 				goto drop;
493 			}
494 			if (in4_cksum(m, IPPROTO_TCP, iphlen, tlen) != 0) {
495 				tcpstat.tcps_rcvbadsum++;
496 				goto drop;
497 			}
498 		} else {
499 			m->m_pkthdr.csum_flags &= ~M_TCP_CSUM_IN_OK;
500 			tcpstat.tcps_inhwcsum++;
501 		}
502 		break;
503 #ifdef INET6
504 	case AF_INET6:
505 		ip6 = mtod(m, struct ip6_hdr *);
506 #ifdef TCP_ECN
507 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
508 #endif
509 
510 		/* Be proactive about malicious use of IPv4 mapped address */
511 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
512 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
513 			/* XXX stat */
514 			goto drop;
515 		}
516 
517 		/*
518 		 * Be proactive about unspecified IPv6 address in source.
519 		 * As we use all-zero to indicate unbounded/unconnected pcb,
520 		 * unspecified IPv6 address can be used to confuse us.
521 		 *
522 		 * Note that packets with unspecified IPv6 destination is
523 		 * already dropped in ip6_input.
524 		 */
525 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
526 			/* XXX stat */
527 			goto drop;
528 		}
529 
530 		/* Discard packets to multicast */
531 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
532 			/* XXX stat */
533 			goto drop;
534 		}
535 
536 		/*
537 		 * Checksum extended TCP header and data.
538 		 */
539 		if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) {
540 			tcpstat.tcps_rcvbadsum++;
541 			goto drop;
542 		}
543 		break;
544 #endif
545 	}
546 
547 	/*
548 	 * Check that TCP offset makes sense,
549 	 * pull out TCP options and adjust length.		XXX
550 	 */
551 	off = th->th_off << 2;
552 	if (off < sizeof(struct tcphdr) || off > tlen) {
553 		tcpstat.tcps_rcvbadoff++;
554 		goto drop;
555 	}
556 	tlen -= off;
557 	if (off > sizeof(struct tcphdr)) {
558 		IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off);
559 		if (!th) {
560 			tcpstat.tcps_rcvshort++;
561 			return;
562 		}
563 		optlen = off - sizeof(struct tcphdr);
564 		optp = (u_int8_t *)(th + 1);
565 		/*
566 		 * Do quick retrieval of timestamp options ("options
567 		 * prediction?").  If timestamp is the only option and it's
568 		 * formatted as recommended in RFC 1323 appendix A, we
569 		 * quickly get the values now and not bother calling
570 		 * tcp_dooptions(), etc.
571 		 */
572 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
573 		     (optlen > TCPOLEN_TSTAMP_APPA &&
574 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
575 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
576 		     (th->th_flags & TH_SYN) == 0) {
577 			opti.ts_present = 1;
578 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
579 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
580 			optp = NULL;	/* we've parsed the options */
581 		}
582 	}
583 	tiflags = th->th_flags;
584 
585 	/*
586 	 * Convert TCP protocol specific fields to host format.
587 	 */
588 	NTOHL(th->th_seq);
589 	NTOHL(th->th_ack);
590 	NTOHS(th->th_win);
591 	NTOHS(th->th_urp);
592 
593 	/*
594 	 * Locate pcb for segment.
595 	 */
596 findpcb:
597 	switch (af) {
598 #ifdef INET6
599 	case AF_INET6:
600 		inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src, th->th_sport,
601 		    &ip6->ip6_dst, th->th_dport);
602 		break;
603 #endif
604 	case AF_INET:
605 		inp = in_pcbhashlookup(&tcbtable, ip->ip_src, th->th_sport,
606 		    ip->ip_dst, th->th_dport);
607 		break;
608 	}
609 	if (inp == 0) {
610 		int	inpl_flags = 0;
611 		if (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST)
612 			inpl_flags = INPLOOKUP_WILDCARD;
613 		++tcpstat.tcps_pcbhashmiss;
614 		switch (af) {
615 #ifdef INET6
616 		case AF_INET6:
617 			inp = in6_pcblookup_listen(&tcbtable,
618 			    &ip6->ip6_dst, th->th_dport, inpl_flags);
619 			break;
620 #endif /* INET6 */
621 		case AF_INET:
622 			inp = in_pcblookup_listen(&tcbtable,
623 			    ip->ip_dst, th->th_dport, inpl_flags);
624 			break;
625 		}
626 		/*
627 		 * If the state is CLOSED (i.e., TCB does not exist) then
628 		 * all data in the incoming segment is discarded.
629 		 * If the TCB exists but is in CLOSED state, it is embryonic,
630 		 * but should either do a listen or a connect soon.
631 		 */
632 		if (inp == 0) {
633 			++tcpstat.tcps_noport;
634 			goto dropwithreset_ratelim;
635 		}
636 	}
637 
638 	/* Check the minimum TTL for socket. */
639 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
640 		goto drop;
641 
642 	tp = intotcpcb(inp);
643 	if (tp == 0)
644 		goto dropwithreset_ratelim;
645 	if (tp->t_state == TCPS_CLOSED)
646 		goto drop;
647 
648 	/* Unscale the window into a 32-bit value. */
649 	if ((tiflags & TH_SYN) == 0)
650 		tiwin = th->th_win << tp->snd_scale;
651 	else
652 		tiwin = th->th_win;
653 
654 	so = inp->inp_socket;
655 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
656 		union syn_cache_sa src;
657 		union syn_cache_sa dst;
658 
659 		bzero(&src, sizeof(src));
660 		bzero(&dst, sizeof(dst));
661 		switch (af) {
662 #ifdef INET
663 		case AF_INET:
664 			src.sin.sin_len = sizeof(struct sockaddr_in);
665 			src.sin.sin_family = AF_INET;
666 			src.sin.sin_addr = ip->ip_src;
667 			src.sin.sin_port = th->th_sport;
668 
669 			dst.sin.sin_len = sizeof(struct sockaddr_in);
670 			dst.sin.sin_family = AF_INET;
671 			dst.sin.sin_addr = ip->ip_dst;
672 			dst.sin.sin_port = th->th_dport;
673 			break;
674 #endif
675 #ifdef INET6
676 		case AF_INET6:
677 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
678 			src.sin6.sin6_family = AF_INET6;
679 			src.sin6.sin6_addr = ip6->ip6_src;
680 			src.sin6.sin6_port = th->th_sport;
681 
682 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
683 			dst.sin6.sin6_family = AF_INET6;
684 			dst.sin6.sin6_addr = ip6->ip6_dst;
685 			dst.sin6.sin6_port = th->th_dport;
686 			break;
687 #endif /* INET6 */
688 		default:
689 			goto badsyn;	/*sanity*/
690 		}
691 
692 		if (so->so_options & SO_DEBUG) {
693 			ostate = tp->t_state;
694 			switch (af) {
695 #ifdef INET6
696 			case AF_INET6:
697 				bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6));
698 				bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th));
699 				break;
700 #endif
701 			case AF_INET:
702 				bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip));
703 				bcopy(th, &tcp_saveti.ti_t, sizeof(*th));
704 				break;
705 			}
706 		}
707 		if (so->so_options & SO_ACCEPTCONN) {
708 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
709 				if (tiflags & TH_RST) {
710 					syn_cache_reset(&src.sa, &dst.sa, th);
711 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
712 				    (TH_ACK|TH_SYN)) {
713 					/*
714 					 * Received a SYN,ACK.  This should
715 					 * never happen while we are in
716 					 * LISTEN.  Send an RST.
717 					 */
718 					goto badsyn;
719 				} else if (tiflags & TH_ACK) {
720 					so = syn_cache_get(&src.sa, &dst.sa,
721 						th, iphlen, tlen, so, m);
722 					if (so == NULL) {
723 						/*
724 						 * We don't have a SYN for
725 						 * this ACK; send an RST.
726 						 */
727 						goto badsyn;
728 					} else if (so ==
729 					    (struct socket *)(-1)) {
730 						/*
731 						 * We were unable to create
732 						 * the connection.  If the
733 						 * 3-way handshake was
734 						 * completed, and RST has
735 						 * been sent to the peer.
736 						 * Since the mbuf might be
737 						 * in use for the reply,
738 						 * do not free it.
739 						 */
740 						m = NULL;
741 					} else {
742 						/*
743 						 * We have created a
744 						 * full-blown connection.
745 						 */
746 						tp = NULL;
747 						inp = (struct inpcb *)so->so_pcb;
748 						tp = intotcpcb(inp);
749 						if (tp == NULL)
750 							goto badsyn;	/*XXX*/
751 
752 						/*
753 						 * Compute proper scaling
754 						 * value from buffer space
755 						 */
756 						tcp_rscale(tp, so->so_rcv.sb_hiwat);
757 						goto after_listen;
758 					}
759 				} else {
760 					/*
761 					 * None of RST, SYN or ACK was set.
762 					 * This is an invalid packet for a
763 					 * TCB in LISTEN state.  Send a RST.
764 					 */
765 					goto badsyn;
766 				}
767 			} else {
768 				/*
769 				 * Received a SYN.
770 				 */
771 #ifdef INET6
772 				/*
773 				 * If deprecated address is forbidden, we do
774 				 * not accept SYN to deprecated interface
775 				 * address to prevent any new inbound
776 				 * connection from getting established.
777 				 * When we do not accept SYN, we send a TCP
778 				 * RST, with deprecated source address (instead
779 				 * of dropping it).  We compromise it as it is
780 				 * much better for peer to send a RST, and
781 				 * RST will be the final packet for the
782 				 * exchange.
783 				 *
784 				 * If we do not forbid deprecated addresses, we
785 				 * accept the SYN packet.  RFC2462 does not
786 				 * suggest dropping SYN in this case.
787 				 * If we decipher RFC2462 5.5.4, it says like
788 				 * this:
789 				 * 1. use of deprecated addr with existing
790 				 *    communication is okay - "SHOULD continue
791 				 *    to be used"
792 				 * 2. use of it with new communication:
793 				 *   (2a) "SHOULD NOT be used if alternate
794 				 *        address with sufficient scope is
795 				 *        available"
796 				 *   (2b) nothing mentioned otherwise.
797 				 * Here we fall into (2b) case as we have no
798 				 * choice in our source address selection - we
799 				 * must obey the peer.
800 				 *
801 				 * The wording in RFC2462 is confusing, and
802 				 * there are multiple description text for
803 				 * deprecated address handling - worse, they
804 				 * are not exactly the same.  I believe 5.5.4
805 				 * is the best one, so we follow 5.5.4.
806 				 */
807 				if (ip6 && !ip6_use_deprecated) {
808 					struct in6_ifaddr *ia6;
809 
810 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
811 					    &ip6->ip6_dst)) &&
812 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
813 						tp = NULL;
814 						goto dropwithreset;
815 					}
816 				}
817 #endif
818 
819 				/*
820 				 * LISTEN socket received a SYN
821 				 * from itself?  This can't possibly
822 				 * be valid; drop the packet.
823 				 */
824 				if (th->th_dport == th->th_sport) {
825 					switch (af) {
826 #ifdef INET6
827 					case AF_INET6:
828 						if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src,
829 						    &ip6->ip6_dst)) {
830 							tcpstat.tcps_badsyn++;
831 							goto drop;
832 						}
833 						break;
834 #endif /* INET6 */
835 					case AF_INET:
836 						if (ip->ip_dst.s_addr == ip->ip_src.s_addr) {
837 							tcpstat.tcps_badsyn++;
838 							goto drop;
839 						}
840 						break;
841 					}
842 				}
843 
844 				/*
845 				 * SYN looks ok; create compressed TCP
846 				 * state for it.
847 				 */
848 				if (so->so_qlen <= so->so_qlimit &&
849 				    syn_cache_add(&src.sa, &dst.sa, th, iphlen,
850 				    so, m, optp, optlen, &opti, reuse))
851 					m = NULL;
852 			}
853 			goto drop;
854 		}
855 	}
856 
857 after_listen:
858 #ifdef DIAGNOSTIC
859 	/*
860 	 * Should not happen now that all embryonic connections
861 	 * are handled with compressed state.
862 	 */
863 	if (tp->t_state == TCPS_LISTEN)
864 		panic("tcp_input: TCPS_LISTEN");
865 #endif
866 
867 #ifdef IPSEC
868 	/* Find most recent IPsec tag */
869 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
870         s = splnet();
871 	if (mtag != NULL) {
872 		tdbi = (struct tdb_ident *)(mtag + 1);
873 	        tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto);
874 	} else
875 		tdb = NULL;
876 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
877 	    tdb, inp);
878 	if (error) {
879 		splx(s);
880 		goto drop;
881 	}
882 
883 	/* Latch SA */
884 	if (inp->inp_tdb_in != tdb) {
885 		if (tdb) {
886 		        tdb_add_inp(tdb, inp, 1);
887 			if (inp->inp_ipo == NULL) {
888 				inp->inp_ipo = ipsec_add_policy(inp, af,
889 				    IPSP_DIRECTION_OUT);
890 				if (inp->inp_ipo == NULL) {
891 					splx(s);
892 					goto drop;
893 				}
894 			}
895 			if (inp->inp_ipo->ipo_dstid == NULL &&
896 			    tdb->tdb_srcid != NULL) {
897 				inp->inp_ipo->ipo_dstid = tdb->tdb_srcid;
898 				tdb->tdb_srcid->ref_count++;
899 			}
900 			if (inp->inp_ipsec_remotecred == NULL &&
901 			    tdb->tdb_remote_cred != NULL) {
902 				inp->inp_ipsec_remotecred =
903 				    tdb->tdb_remote_cred;
904 				tdb->tdb_remote_cred->ref_count++;
905 			}
906 			if (inp->inp_ipsec_remoteauth == NULL &&
907 			    tdb->tdb_remote_auth != NULL) {
908 				inp->inp_ipsec_remoteauth =
909 				    tdb->tdb_remote_auth;
910 				tdb->tdb_remote_auth->ref_count++;
911 			}
912 		} else { /* Just reset */
913 		        TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp,
914 				     inp_tdb_in_next);
915 			inp->inp_tdb_in = NULL;
916 		}
917 	}
918         splx(s);
919 #endif /* IPSEC */
920 
921 	/*
922 	 * Segment received on connection.
923 	 * Reset idle time and keep-alive timer.
924 	 */
925 	tp->t_rcvtime = tcp_now;
926 	if (TCPS_HAVEESTABLISHED(tp->t_state))
927 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
928 
929 #ifdef TCP_SACK
930 	if (tp->sack_enable)
931 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
932 #endif /* TCP_SACK */
933 
934 	/*
935 	 * Process options.
936 	 */
937 #ifdef TCP_SIGNATURE
938 	if (optp || (tp->t_flags & TF_SIGNATURE))
939 #else
940 	if (optp)
941 #endif
942 		if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti))
943 			goto drop;
944 
945 	if (opti.ts_present && opti.ts_ecr) {
946 		int rtt_test;
947 
948 		/* subtract out the tcp timestamp modulator */
949 		opti.ts_ecr -= tp->ts_modulate;
950 
951 		/* make sure ts_ecr is sensible */
952 		rtt_test = tcp_now - opti.ts_ecr;
953 		if (rtt_test < 0 || rtt_test > TCP_RTT_MAX)
954 			opti.ts_ecr = 0;
955 	}
956 
957 #ifdef TCP_ECN
958 	/* if congestion experienced, set ECE bit in subsequent packets. */
959 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
960 		tp->t_flags |= TF_RCVD_CE;
961 		tcpstat.tcps_ecn_rcvce++;
962 	}
963 #endif
964 	/*
965 	 * Header prediction: check for the two common cases
966 	 * of a uni-directional data xfer.  If the packet has
967 	 * no control flags, is in-sequence, the window didn't
968 	 * change and we're not retransmitting, it's a
969 	 * candidate.  If the length is zero and the ack moved
970 	 * forward, we're the sender side of the xfer.  Just
971 	 * free the data acked & wake any higher level process
972 	 * that was blocked waiting for space.  If the length
973 	 * is non-zero and the ack didn't move, we're the
974 	 * receiver side.  If we're getting packets in-order
975 	 * (the reassembly queue is empty), add the data to
976 	 * the socket buffer and note that we need a delayed ack.
977 	 */
978 	if (tp->t_state == TCPS_ESTABLISHED &&
979 #ifdef TCP_ECN
980 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
981 #else
982 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
983 #endif
984 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
985 	    th->th_seq == tp->rcv_nxt &&
986 	    tiwin && tiwin == tp->snd_wnd &&
987 	    tp->snd_nxt == tp->snd_max) {
988 
989 		/*
990 		 * If last ACK falls within this segment's sequence numbers,
991 		 *  record the timestamp.
992 		 * Fix from Braden, see Stevens p. 870
993 		 */
994 		if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
995 			tp->ts_recent_age = tcp_now;
996 			tp->ts_recent = opti.ts_val;
997 		}
998 
999 		if (tlen == 0) {
1000 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1001 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1002 			    tp->snd_cwnd >= tp->snd_wnd &&
1003 			    tp->t_dupacks == 0) {
1004 				/*
1005 				 * this is a pure ack for outstanding data.
1006 				 */
1007 				++tcpstat.tcps_predack;
1008 				if (opti.ts_present && opti.ts_ecr)
1009 					tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1010 				else if (tp->t_rtttime &&
1011 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1012 					tcp_xmit_timer(tp,
1013 					    tcp_now - tp->t_rtttime);
1014 				acked = th->th_ack - tp->snd_una;
1015 				tcpstat.tcps_rcvackpack++;
1016 				tcpstat.tcps_rcvackbyte += acked;
1017 				ND6_HINT(tp);
1018 				sbdrop(&so->so_snd, acked);
1019 
1020 				/*
1021 				 * If we had a pending ICMP message that
1022 				 * referres to data that have just been
1023 				 * acknowledged, disregard the recorded ICMP
1024 				 * message.
1025 				 */
1026 				if ((tp->t_flags & TF_PMTUD_PEND) &&
1027 				    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1028 					tp->t_flags &= ~TF_PMTUD_PEND;
1029 
1030 				/*
1031 				 * Keep track of the largest chunk of data
1032 				 * acknowledged since last PMTU update
1033 				 */
1034 				if (tp->t_pmtud_mss_acked < acked)
1035 					tp->t_pmtud_mss_acked = acked;
1036 
1037 				tp->snd_una = th->th_ack;
1038 #if defined(TCP_SACK) || defined(TCP_ECN)
1039 				/*
1040 				 * We want snd_last to track snd_una so
1041 				 * as to avoid sequence wraparound problems
1042 				 * for very large transfers.
1043 				 */
1044 #ifdef TCP_ECN
1045 				if (SEQ_GT(tp->snd_una, tp->snd_last))
1046 #endif
1047 				tp->snd_last = tp->snd_una;
1048 #endif /* TCP_SACK */
1049 #if defined(TCP_SACK) && defined(TCP_FACK)
1050 				tp->snd_fack = tp->snd_una;
1051 				tp->retran_data = 0;
1052 #endif /* TCP_FACK */
1053 				m_freem(m);
1054 
1055 				/*
1056 				 * If all outstanding data are acked, stop
1057 				 * retransmit timer, otherwise restart timer
1058 				 * using current (possibly backed-off) value.
1059 				 * If process is waiting for space,
1060 				 * wakeup/selwakeup/signal.  If data
1061 				 * are ready to send, let tcp_output
1062 				 * decide between more output or persist.
1063 				 */
1064 				if (tp->snd_una == tp->snd_max)
1065 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1066 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1067 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1068 
1069 				if (sb_notify(&so->so_snd))
1070 					sowwakeup(so);
1071 				if (so->so_snd.sb_cc)
1072 					(void) tcp_output(tp);
1073 				return;
1074 			}
1075 		} else if (th->th_ack == tp->snd_una &&
1076 		    TAILQ_EMPTY(&tp->t_segq) &&
1077 		    tlen <= sbspace(&so->so_rcv)) {
1078 			/*
1079 			 * This is a pure, in-sequence data packet
1080 			 * with nothing on the reassembly queue and
1081 			 * we have enough buffer space to take it.
1082 			 */
1083 #ifdef TCP_SACK
1084 			/* Clean receiver SACK report if present */
1085 			if (tp->sack_enable && tp->rcv_numsacks)
1086 				tcp_clean_sackreport(tp);
1087 #endif /* TCP_SACK */
1088 			++tcpstat.tcps_preddat;
1089 			tp->rcv_nxt += tlen;
1090 			tcpstat.tcps_rcvpack++;
1091 			tcpstat.tcps_rcvbyte += tlen;
1092 			ND6_HINT(tp);
1093 			/*
1094 			 * Drop TCP, IP headers and TCP options then add data
1095 			 * to socket buffer.
1096 			 */
1097 			if (so->so_state & SS_CANTRCVMORE)
1098 				m_freem(m);
1099 			else {
1100 				m_adj(m, iphlen + off);
1101 				sbappendstream(&so->so_rcv, m);
1102 			}
1103 			sorwakeup(so);
1104 			TCP_SETUP_ACK(tp, tiflags);
1105 			if (tp->t_flags & TF_ACKNOW)
1106 				(void) tcp_output(tp);
1107 			return;
1108 		}
1109 	}
1110 
1111 	/*
1112 	 * Compute mbuf offset to TCP data segment.
1113 	 */
1114 	hdroptlen = iphlen + off;
1115 
1116 	/*
1117 	 * Calculate amount of space in receive window,
1118 	 * and then do TCP input processing.
1119 	 * Receive window is amount of space in rcv queue,
1120 	 * but not less than advertised window.
1121 	 */
1122 	{ int win;
1123 
1124 	win = sbspace(&so->so_rcv);
1125 	if (win < 0)
1126 		win = 0;
1127 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1128 	}
1129 
1130 	switch (tp->t_state) {
1131 
1132 	/*
1133 	 * If the state is SYN_RECEIVED:
1134 	 * 	if seg contains SYN/ACK, send an RST.
1135 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1136 	 */
1137 
1138 	case TCPS_SYN_RECEIVED:
1139 		if (tiflags & TH_ACK) {
1140 			if (tiflags & TH_SYN) {
1141 				tcpstat.tcps_badsyn++;
1142 				goto dropwithreset;
1143 			}
1144 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1145 			    SEQ_GT(th->th_ack, tp->snd_max))
1146 				goto dropwithreset;
1147 		}
1148 		break;
1149 
1150 	/*
1151 	 * If the state is SYN_SENT:
1152 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1153 	 *	if seg contains a RST, then drop the connection.
1154 	 *	if seg does not contain SYN, then drop it.
1155 	 * Otherwise this is an acceptable SYN segment
1156 	 *	initialize tp->rcv_nxt and tp->irs
1157 	 *	if seg contains ack then advance tp->snd_una
1158 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1159 	 *	arrange for segment to be acked (eventually)
1160 	 *	continue processing rest of data/controls, beginning with URG
1161 	 */
1162 	case TCPS_SYN_SENT:
1163 		if ((tiflags & TH_ACK) &&
1164 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1165 		     SEQ_GT(th->th_ack, tp->snd_max)))
1166 			goto dropwithreset;
1167 		if (tiflags & TH_RST) {
1168 #ifdef TCP_ECN
1169 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1170 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1171 				goto drop;
1172 #endif
1173 			if (tiflags & TH_ACK)
1174 				tp = tcp_drop(tp, ECONNREFUSED);
1175 			goto drop;
1176 		}
1177 		if ((tiflags & TH_SYN) == 0)
1178 			goto drop;
1179 		if (tiflags & TH_ACK) {
1180 			tp->snd_una = th->th_ack;
1181 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1182 				tp->snd_nxt = tp->snd_una;
1183 		}
1184 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1185 		tp->irs = th->th_seq;
1186 		tcp_mss(tp, opti.maxseg);
1187 		/* Reset initial window to 1 segment for retransmit */
1188 		if (tp->t_rxtshift > 0)
1189 			tp->snd_cwnd = tp->t_maxseg;
1190 		tcp_rcvseqinit(tp);
1191 		tp->t_flags |= TF_ACKNOW;
1192 #ifdef TCP_SACK
1193                 /*
1194                  * If we've sent a SACK_PERMITTED option, and the peer
1195                  * also replied with one, then TF_SACK_PERMIT should have
1196                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1197                  */
1198 		if (tp->sack_enable)
1199 			tp->sack_enable = tp->t_flags & TF_SACK_PERMIT;
1200 #endif
1201 #ifdef TCP_ECN
1202 		/*
1203 		 * if ECE is set but CWR is not set for SYN-ACK, or
1204 		 * both ECE and CWR are set for simultaneous open,
1205 		 * peer is ECN capable.
1206 		 */
1207 		if (tcp_do_ecn) {
1208 			if ((tiflags & (TH_ACK|TH_ECE|TH_CWR))
1209 			    == (TH_ACK|TH_ECE) ||
1210 			    (tiflags & (TH_ACK|TH_ECE|TH_CWR))
1211 			    == (TH_ECE|TH_CWR)) {
1212 				tp->t_flags |= TF_ECN_PERMIT;
1213 				tiflags &= ~(TH_ECE|TH_CWR);
1214 				tcpstat.tcps_ecn_accepts++;
1215 			}
1216 		}
1217 #endif
1218 
1219 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1220 			tcpstat.tcps_connects++;
1221 			soisconnected(so);
1222 			tp->t_state = TCPS_ESTABLISHED;
1223 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1224 			/* Do window scaling on this connection? */
1225 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1226 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1227 				tp->snd_scale = tp->requested_s_scale;
1228 				tp->rcv_scale = tp->request_r_scale;
1229 			}
1230 			tcp_reass_lock(tp);
1231 			(void) tcp_reass(tp, (struct tcphdr *)0,
1232 				(struct mbuf *)0, &tlen);
1233 			tcp_reass_unlock(tp);
1234 			/*
1235 			 * if we didn't have to retransmit the SYN,
1236 			 * use its rtt as our initial srtt & rtt var.
1237 			 */
1238 			if (tp->t_rtttime)
1239 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1240 			/*
1241 			 * Since new data was acked (the SYN), open the
1242 			 * congestion window by one MSS.  We do this
1243 			 * here, because we won't go through the normal
1244 			 * ACK processing below.  And since this is the
1245 			 * start of the connection, we know we are in
1246 			 * the exponential phase of slow-start.
1247 			 */
1248 			tp->snd_cwnd += tp->t_maxseg;
1249 		} else
1250 			tp->t_state = TCPS_SYN_RECEIVED;
1251 
1252 #if 0
1253 trimthenstep6:
1254 #endif
1255 		/*
1256 		 * Advance th->th_seq to correspond to first data byte.
1257 		 * If data, trim to stay within window,
1258 		 * dropping FIN if necessary.
1259 		 */
1260 		th->th_seq++;
1261 		if (tlen > tp->rcv_wnd) {
1262 			todrop = tlen - tp->rcv_wnd;
1263 			m_adj(m, -todrop);
1264 			tlen = tp->rcv_wnd;
1265 			tiflags &= ~TH_FIN;
1266 			tcpstat.tcps_rcvpackafterwin++;
1267 			tcpstat.tcps_rcvbyteafterwin += todrop;
1268 		}
1269 		tp->snd_wl1 = th->th_seq - 1;
1270 		tp->rcv_up = th->th_seq;
1271 		goto step6;
1272 	/*
1273 	 * If a new connection request is received while in TIME_WAIT,
1274 	 * drop the old connection and start over if the if the
1275 	 * timestamp or the sequence numbers are above the previous
1276 	 * ones.
1277 	 */
1278 	case TCPS_TIME_WAIT:
1279 		if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) &&
1280 		    ((opti.ts_present &&
1281 		    TSTMP_LT(tp->ts_recent, opti.ts_val)) ||
1282 		    SEQ_GT(th->th_seq, tp->rcv_nxt))) {
1283 			/*
1284 			* Advance the iss by at least 32768, but
1285 			* clear the msb in order to make sure
1286 			* that SEG_LT(snd_nxt, iss).
1287 			*/
1288 			iss = tp->snd_nxt +
1289 			    ((arc4random() & 0x7fffffff) | 0x8000);
1290 			reuse = &iss;
1291 			tp = tcp_close(tp);
1292 			goto findpcb;
1293 		}
1294 	}
1295 
1296 	/*
1297 	 * States other than LISTEN or SYN_SENT.
1298 	 * First check timestamp, if present.
1299 	 * Then check that at least some bytes of segment are within
1300 	 * receive window.  If segment begins before rcv_nxt,
1301 	 * drop leading data (and SYN); if nothing left, just ack.
1302 	 *
1303 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1304 	 * and it's less than opti.ts_recent, drop it.
1305 	 */
1306 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1307 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1308 
1309 		/* Check to see if ts_recent is over 24 days old.  */
1310 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1311 			/*
1312 			 * Invalidate ts_recent.  If this segment updates
1313 			 * ts_recent, the age will be reset later and ts_recent
1314 			 * will get a valid value.  If it does not, setting
1315 			 * ts_recent to zero will at least satisfy the
1316 			 * requirement that zero be placed in the timestamp
1317 			 * echo reply when ts_recent isn't valid.  The
1318 			 * age isn't reset until we get a valid ts_recent
1319 			 * because we don't want out-of-order segments to be
1320 			 * dropped when ts_recent is old.
1321 			 */
1322 			tp->ts_recent = 0;
1323 		} else {
1324 			tcpstat.tcps_rcvduppack++;
1325 			tcpstat.tcps_rcvdupbyte += tlen;
1326 			tcpstat.tcps_pawsdrop++;
1327 			goto dropafterack;
1328 		}
1329 	}
1330 
1331 	todrop = tp->rcv_nxt - th->th_seq;
1332 	if (todrop > 0) {
1333 		if (tiflags & TH_SYN) {
1334 			tiflags &= ~TH_SYN;
1335 			th->th_seq++;
1336 			if (th->th_urp > 1)
1337 				th->th_urp--;
1338 			else
1339 				tiflags &= ~TH_URG;
1340 			todrop--;
1341 		}
1342 		if (todrop > tlen ||
1343 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1344 			/*
1345 			 * Any valid FIN must be to the left of the
1346 			 * window.  At this point, FIN must be a
1347 			 * duplicate or out-of-sequence, so drop it.
1348 			 */
1349 			tiflags &= ~TH_FIN;
1350 			/*
1351 			 * Send ACK to resynchronize, and drop any data,
1352 			 * but keep on processing for RST or ACK.
1353 			 */
1354 			tp->t_flags |= TF_ACKNOW;
1355 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1356 			tcpstat.tcps_rcvduppack++;
1357 		} else {
1358 			tcpstat.tcps_rcvpartduppack++;
1359 			tcpstat.tcps_rcvpartdupbyte += todrop;
1360 		}
1361 		hdroptlen += todrop;	/* drop from head afterwards */
1362 		th->th_seq += todrop;
1363 		tlen -= todrop;
1364 		if (th->th_urp > todrop)
1365 			th->th_urp -= todrop;
1366 		else {
1367 			tiflags &= ~TH_URG;
1368 			th->th_urp = 0;
1369 		}
1370 	}
1371 
1372 	/*
1373 	 * If new data are received on a connection after the
1374 	 * user processes are gone, then RST the other end.
1375 	 */
1376 	if ((so->so_state & SS_NOFDREF) &&
1377 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1378 		tp = tcp_close(tp);
1379 		tcpstat.tcps_rcvafterclose++;
1380 		goto dropwithreset;
1381 	}
1382 
1383 	/*
1384 	 * If segment ends after window, drop trailing data
1385 	 * (and PUSH and FIN); if nothing left, just ACK.
1386 	 */
1387 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1388 	if (todrop > 0) {
1389 		tcpstat.tcps_rcvpackafterwin++;
1390 		if (todrop >= tlen) {
1391 			tcpstat.tcps_rcvbyteafterwin += tlen;
1392 			/*
1393 			 * If window is closed can only take segments at
1394 			 * window edge, and have to drop data and PUSH from
1395 			 * incoming segments.  Continue processing, but
1396 			 * remember to ack.  Otherwise, drop segment
1397 			 * and ack.
1398 			 */
1399 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1400 				tp->t_flags |= TF_ACKNOW;
1401 				tcpstat.tcps_rcvwinprobe++;
1402 			} else
1403 				goto dropafterack;
1404 		} else
1405 			tcpstat.tcps_rcvbyteafterwin += todrop;
1406 		m_adj(m, -todrop);
1407 		tlen -= todrop;
1408 		tiflags &= ~(TH_PUSH|TH_FIN);
1409 	}
1410 
1411 	/*
1412 	 * If last ACK falls within this segment's sequence numbers,
1413 	 * record its timestamp if it's more recent.
1414 	 * Cf fix from Braden, see Stevens p. 870
1415 	 */
1416 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1417 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1418 		if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1419 		    ((tiflags & (TH_SYN|TH_FIN)) != 0)))
1420 			tp->ts_recent = opti.ts_val;
1421 		else
1422 			tp->ts_recent = 0;
1423 		tp->ts_recent_age = tcp_now;
1424 	}
1425 
1426 	/*
1427 	 * If the RST bit is set examine the state:
1428 	 *    SYN_RECEIVED STATE:
1429 	 *	If passive open, return to LISTEN state.
1430 	 *	If active open, inform user that connection was refused.
1431 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1432 	 *	Inform user that connection was reset, and close tcb.
1433 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1434 	 *	Close the tcb.
1435 	 */
1436 	if (tiflags & TH_RST) {
1437 		if (th->th_seq != tp->last_ack_sent &&
1438 		    th->th_seq != tp->rcv_nxt &&
1439 		    th->th_seq != (tp->rcv_nxt + 1))
1440 			goto drop;
1441 
1442 		switch (tp->t_state) {
1443 		case TCPS_SYN_RECEIVED:
1444 #ifdef TCP_ECN
1445 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1446 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1447 				goto drop;
1448 #endif
1449 			so->so_error = ECONNREFUSED;
1450 			goto close;
1451 
1452 		case TCPS_ESTABLISHED:
1453 		case TCPS_FIN_WAIT_1:
1454 		case TCPS_FIN_WAIT_2:
1455 		case TCPS_CLOSE_WAIT:
1456 			so->so_error = ECONNRESET;
1457 		close:
1458 			tp->t_state = TCPS_CLOSED;
1459 			tcpstat.tcps_drops++;
1460 			tp = tcp_close(tp);
1461 			goto drop;
1462 		case TCPS_CLOSING:
1463 		case TCPS_LAST_ACK:
1464 		case TCPS_TIME_WAIT:
1465 			tp = tcp_close(tp);
1466 			goto drop;
1467 		}
1468 	}
1469 
1470 	/*
1471 	 * If a SYN is in the window, then this is an
1472 	 * error and we ACK and drop the packet.
1473 	 */
1474 	if (tiflags & TH_SYN)
1475 		goto dropafterack_ratelim;
1476 
1477 	/*
1478 	 * If the ACK bit is off we drop the segment and return.
1479 	 */
1480 	if ((tiflags & TH_ACK) == 0) {
1481 		if (tp->t_flags & TF_ACKNOW)
1482 			goto dropafterack;
1483 		else
1484 			goto drop;
1485 	}
1486 
1487 	/*
1488 	 * Ack processing.
1489 	 */
1490 	switch (tp->t_state) {
1491 
1492 	/*
1493 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1494 	 * ESTABLISHED state and continue processing.
1495 	 * The ACK was checked above.
1496 	 */
1497 	case TCPS_SYN_RECEIVED:
1498 		tcpstat.tcps_connects++;
1499 		soisconnected(so);
1500 		tp->t_state = TCPS_ESTABLISHED;
1501 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1502 		/* Do window scaling? */
1503 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1504 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1505 			tp->snd_scale = tp->requested_s_scale;
1506 			tp->rcv_scale = tp->request_r_scale;
1507 			tiwin = th->th_win << tp->snd_scale;
1508 		}
1509 		tcp_reass_lock(tp);
1510 		(void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0,
1511 				 &tlen);
1512 		tcp_reass_unlock(tp);
1513 		tp->snd_wl1 = th->th_seq - 1;
1514 		/* fall into ... */
1515 
1516 	/*
1517 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1518 	 * ACKs.  If the ack is in the range
1519 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1520 	 * then advance tp->snd_una to th->th_ack and drop
1521 	 * data from the retransmission queue.  If this ACK reflects
1522 	 * more up to date window information we update our window information.
1523 	 */
1524 	case TCPS_ESTABLISHED:
1525 	case TCPS_FIN_WAIT_1:
1526 	case TCPS_FIN_WAIT_2:
1527 	case TCPS_CLOSE_WAIT:
1528 	case TCPS_CLOSING:
1529 	case TCPS_LAST_ACK:
1530 	case TCPS_TIME_WAIT:
1531 #ifdef TCP_ECN
1532 		/*
1533 		 * if we receive ECE and are not already in recovery phase,
1534 		 * reduce cwnd by half but don't slow-start.
1535 		 * advance snd_last to snd_max not to reduce cwnd again
1536 		 * until all outstanding packets are acked.
1537 		 */
1538 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1539 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1540 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1541 				u_int win;
1542 
1543 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1544 				if (win > 1) {
1545 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1546 					tp->snd_cwnd = tp->snd_ssthresh;
1547 					tp->snd_last = tp->snd_max;
1548 					tp->t_flags |= TF_SEND_CWR;
1549 					tcpstat.tcps_cwr_ecn++;
1550 				}
1551 			}
1552 			tcpstat.tcps_ecn_rcvece++;
1553 		}
1554 		/*
1555 		 * if we receive CWR, we know that the peer has reduced
1556 		 * its congestion window.  stop sending ecn-echo.
1557 		 */
1558 		if ((tiflags & TH_CWR)) {
1559 			tp->t_flags &= ~TF_RCVD_CE;
1560 			tcpstat.tcps_ecn_rcvcwr++;
1561 		}
1562 #endif /* TCP_ECN */
1563 
1564 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1565 			/*
1566 			 * Duplicate/old ACK processing.
1567 			 * Increments t_dupacks:
1568 			 *	Pure duplicate (same seq/ack/window, no data)
1569 			 * Doesn't affect t_dupacks:
1570 			 *	Data packets.
1571 			 *	Normal window updates (window opens)
1572 			 * Resets t_dupacks:
1573 			 *	New data ACKed.
1574 			 *	Window shrinks
1575 			 *	Old ACK
1576 			 */
1577 			if (tlen) {
1578 				/* Drop very old ACKs unless th_seq matches */
1579 				if (th->th_seq != tp->rcv_nxt &&
1580 				   SEQ_LT(th->th_ack,
1581 				   tp->snd_una - tp->max_sndwnd)) {
1582 					tcpstat.tcps_rcvacktooold++;
1583 					goto drop;
1584 				}
1585 				break;
1586 			}
1587 			/*
1588 			 * If we get an old ACK, there is probably packet
1589 			 * reordering going on.  Be conservative and reset
1590 			 * t_dupacks so that we are less aggressive in
1591 			 * doing a fast retransmit.
1592 			 */
1593 			if (th->th_ack != tp->snd_una) {
1594 				tp->t_dupacks = 0;
1595 				break;
1596 			}
1597 			if (tiwin == tp->snd_wnd) {
1598 				tcpstat.tcps_rcvdupack++;
1599 				/*
1600 				 * If we have outstanding data (other than
1601 				 * a window probe), this is a completely
1602 				 * duplicate ack (ie, window info didn't
1603 				 * change), the ack is the biggest we've
1604 				 * seen and we've seen exactly our rexmt
1605 				 * threshold of them, assume a packet
1606 				 * has been dropped and retransmit it.
1607 				 * Kludge snd_nxt & the congestion
1608 				 * window so we send only this one
1609 				 * packet.
1610 				 *
1611 				 * We know we're losing at the current
1612 				 * window size so do congestion avoidance
1613 				 * (set ssthresh to half the current window
1614 				 * and pull our congestion window back to
1615 				 * the new ssthresh).
1616 				 *
1617 				 * Dup acks mean that packets have left the
1618 				 * network (they're now cached at the receiver)
1619 				 * so bump cwnd by the amount in the receiver
1620 				 * to keep a constant cwnd packets in the
1621 				 * network.
1622 				 */
1623 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1624 					tp->t_dupacks = 0;
1625 #if defined(TCP_SACK) && defined(TCP_FACK)
1626 				/*
1627 				 * In FACK, can enter fast rec. if the receiver
1628 				 * reports a reass. queue longer than 3 segs.
1629 				 */
1630 				else if (++tp->t_dupacks == tcprexmtthresh ||
1631 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1632 				    tp->t_maxseg + tp->snd_una)) &&
1633 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1634 #else
1635 				else if (++tp->t_dupacks == tcprexmtthresh) {
1636 #endif /* TCP_FACK */
1637 					tcp_seq onxt = tp->snd_nxt;
1638 					u_long win =
1639 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1640 						2 / tp->t_maxseg;
1641 
1642 #if defined(TCP_SACK) || defined(TCP_ECN)
1643 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1644 					    	/*
1645 						 * False fast retx after
1646 						 * timeout.  Do not cut window.
1647 						 */
1648 						tp->t_dupacks = 0;
1649 						goto drop;
1650 					}
1651 #endif
1652 					if (win < 2)
1653 						win = 2;
1654 					tp->snd_ssthresh = win * tp->t_maxseg;
1655 #if defined(TCP_SACK)
1656 					tp->snd_last = tp->snd_max;
1657 #endif
1658 #ifdef TCP_SACK
1659                     			if (tp->sack_enable) {
1660 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1661 						tp->t_rtttime = 0;
1662 #ifdef TCP_ECN
1663 						tp->t_flags |= TF_SEND_CWR;
1664 #endif
1665 #if 1 /* TCP_ECN */
1666 						tcpstat.tcps_cwr_frecovery++;
1667 #endif
1668 						tcpstat.tcps_sack_recovery_episode++;
1669 #if defined(TCP_SACK) && defined(TCP_FACK)
1670 						tp->t_dupacks = tcprexmtthresh;
1671 						(void) tcp_output(tp);
1672 						/*
1673 						 * During FR, snd_cwnd is held
1674 						 * constant for FACK.
1675 						 */
1676 						tp->snd_cwnd = tp->snd_ssthresh;
1677 #else
1678 						/*
1679 						 * tcp_output() will send
1680 						 * oldest SACK-eligible rtx.
1681 						 */
1682 						(void) tcp_output(tp);
1683 						tp->snd_cwnd = tp->snd_ssthresh+
1684 					           tp->t_maxseg * tp->t_dupacks;
1685 #endif /* TCP_FACK */
1686 						goto drop;
1687 					}
1688 #endif /* TCP_SACK */
1689 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1690 					tp->t_rtttime = 0;
1691 					tp->snd_nxt = th->th_ack;
1692 					tp->snd_cwnd = tp->t_maxseg;
1693 #ifdef TCP_ECN
1694 					tp->t_flags |= TF_SEND_CWR;
1695 #endif
1696 #if 1 /* TCP_ECN */
1697 					tcpstat.tcps_cwr_frecovery++;
1698 #endif
1699 					tcpstat.tcps_sndrexmitfast++;
1700 					(void) tcp_output(tp);
1701 
1702 					tp->snd_cwnd = tp->snd_ssthresh +
1703 					    tp->t_maxseg * tp->t_dupacks;
1704 					if (SEQ_GT(onxt, tp->snd_nxt))
1705 						tp->snd_nxt = onxt;
1706 					goto drop;
1707 				} else if (tp->t_dupacks > tcprexmtthresh) {
1708 #if defined(TCP_SACK) && defined(TCP_FACK)
1709 					/*
1710 					 * while (awnd < cwnd)
1711 					 *         sendsomething();
1712 					 */
1713 					if (tp->sack_enable) {
1714 						if (tp->snd_awnd < tp->snd_cwnd)
1715 							tcp_output(tp);
1716 						goto drop;
1717 					}
1718 #endif /* TCP_FACK */
1719 					tp->snd_cwnd += tp->t_maxseg;
1720 					(void) tcp_output(tp);
1721 					goto drop;
1722 				}
1723 			} else if (tiwin < tp->snd_wnd) {
1724 				/*
1725 				 * The window was retracted!  Previous dup
1726 				 * ACKs may have been due to packets arriving
1727 				 * after the shrunken window, not a missing
1728 				 * packet, so play it safe and reset t_dupacks
1729 				 */
1730 				tp->t_dupacks = 0;
1731 			}
1732 			break;
1733 		}
1734 		/*
1735 		 * If the congestion window was inflated to account
1736 		 * for the other side's cached packets, retract it.
1737 		 */
1738 #if defined(TCP_SACK)
1739 		if (tp->sack_enable) {
1740 			if (tp->t_dupacks >= tcprexmtthresh) {
1741 				/* Check for a partial ACK */
1742 				if (tcp_sack_partialack(tp, th)) {
1743 #if defined(TCP_SACK) && defined(TCP_FACK)
1744 					/* Force call to tcp_output */
1745 					if (tp->snd_awnd < tp->snd_cwnd)
1746 						needoutput = 1;
1747 #else
1748 					tp->snd_cwnd += tp->t_maxseg;
1749 					needoutput = 1;
1750 #endif /* TCP_FACK */
1751 				} else {
1752 					/* Out of fast recovery */
1753 					tp->snd_cwnd = tp->snd_ssthresh;
1754 					if (tcp_seq_subtract(tp->snd_max,
1755 					    th->th_ack) < tp->snd_ssthresh)
1756 						tp->snd_cwnd =
1757 						   tcp_seq_subtract(tp->snd_max,
1758 					           th->th_ack);
1759 					tp->t_dupacks = 0;
1760 #if defined(TCP_SACK) && defined(TCP_FACK)
1761 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1762 						tp->snd_fack = th->th_ack;
1763 #endif /* TCP_FACK */
1764 				}
1765 			}
1766 		} else {
1767 			if (tp->t_dupacks >= tcprexmtthresh &&
1768 			    !tcp_newreno(tp, th)) {
1769 				/* Out of fast recovery */
1770 				tp->snd_cwnd = tp->snd_ssthresh;
1771 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1772 			  	    tp->snd_ssthresh)
1773 					tp->snd_cwnd =
1774 					    tcp_seq_subtract(tp->snd_max,
1775 					    th->th_ack);
1776 				tp->t_dupacks = 0;
1777 			}
1778 		}
1779 		if (tp->t_dupacks < tcprexmtthresh)
1780 			tp->t_dupacks = 0;
1781 #else /* else no TCP_SACK */
1782 		if (tp->t_dupacks >= tcprexmtthresh &&
1783 		    tp->snd_cwnd > tp->snd_ssthresh)
1784 			tp->snd_cwnd = tp->snd_ssthresh;
1785 		tp->t_dupacks = 0;
1786 #endif
1787 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1788 			tcpstat.tcps_rcvacktoomuch++;
1789 			goto dropafterack_ratelim;
1790 		}
1791 		acked = th->th_ack - tp->snd_una;
1792 		tcpstat.tcps_rcvackpack++;
1793 		tcpstat.tcps_rcvackbyte += acked;
1794 
1795 		/*
1796 		 * If we have a timestamp reply, update smoothed
1797 		 * round trip time.  If no timestamp is present but
1798 		 * transmit timer is running and timed sequence
1799 		 * number was acked, update smoothed round trip time.
1800 		 * Since we now have an rtt measurement, cancel the
1801 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1802 		 * Recompute the initial retransmit timer.
1803 		 */
1804 		if (opti.ts_present && opti.ts_ecr)
1805 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1806 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1807 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1808 
1809 		/*
1810 		 * If all outstanding data is acked, stop retransmit
1811 		 * timer and remember to restart (more output or persist).
1812 		 * If there is more data to be acked, restart retransmit
1813 		 * timer, using current (possibly backed-off) value.
1814 		 */
1815 		if (th->th_ack == tp->snd_max) {
1816 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1817 			needoutput = 1;
1818 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1819 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1820 		/*
1821 		 * When new data is acked, open the congestion window.
1822 		 * If the window gives us less than ssthresh packets
1823 		 * in flight, open exponentially (maxseg per packet).
1824 		 * Otherwise open linearly: maxseg per window
1825 		 * (maxseg^2 / cwnd per packet).
1826 		 */
1827 		{
1828 		u_int cw = tp->snd_cwnd;
1829 		u_int incr = tp->t_maxseg;
1830 
1831 		if (cw > tp->snd_ssthresh)
1832 			incr = incr * incr / cw;
1833 #if defined (TCP_SACK)
1834 		if (tp->t_dupacks < tcprexmtthresh)
1835 #endif
1836 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1837 		}
1838 		ND6_HINT(tp);
1839 		if (acked > so->so_snd.sb_cc) {
1840 			tp->snd_wnd -= so->so_snd.sb_cc;
1841 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1842 			ourfinisacked = 1;
1843 		} else {
1844 			sbdrop(&so->so_snd, acked);
1845 			tp->snd_wnd -= acked;
1846 			ourfinisacked = 0;
1847 		}
1848 		if (sb_notify(&so->so_snd))
1849 			sowwakeup(so);
1850 
1851 		/*
1852 		 * If we had a pending ICMP message that referred to data
1853 		 * that have just been acknowledged, disregard the recorded
1854 		 * ICMP message.
1855 		 */
1856 		if ((tp->t_flags & TF_PMTUD_PEND) &&
1857 		    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1858 			tp->t_flags &= ~TF_PMTUD_PEND;
1859 
1860 		/*
1861 		 * Keep track of the largest chunk of data acknowledged
1862 		 * since last PMTU update
1863 		 */
1864 		if (tp->t_pmtud_mss_acked < acked)
1865 			tp->t_pmtud_mss_acked = acked;
1866 
1867 		tp->snd_una = th->th_ack;
1868 #ifdef TCP_ECN
1869 		/* sync snd_last with snd_una */
1870 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1871 			tp->snd_last = tp->snd_una;
1872 #endif
1873 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1874 			tp->snd_nxt = tp->snd_una;
1875 #if defined (TCP_SACK) && defined (TCP_FACK)
1876 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1877 			tp->snd_fack = tp->snd_una;
1878 			/* Update snd_awnd for partial ACK
1879 			 * without any SACK blocks.
1880 			 */
1881 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1882 				tp->snd_fack) + tp->retran_data;
1883 		}
1884 #endif
1885 
1886 		switch (tp->t_state) {
1887 
1888 		/*
1889 		 * In FIN_WAIT_1 STATE in addition to the processing
1890 		 * for the ESTABLISHED state if our FIN is now acknowledged
1891 		 * then enter FIN_WAIT_2.
1892 		 */
1893 		case TCPS_FIN_WAIT_1:
1894 			if (ourfinisacked) {
1895 				/*
1896 				 * If we can't receive any more
1897 				 * data, then closing user can proceed.
1898 				 * Starting the timer is contrary to the
1899 				 * specification, but if we don't get a FIN
1900 				 * we'll hang forever.
1901 				 */
1902 				if (so->so_state & SS_CANTRCVMORE) {
1903 					soisdisconnected(so);
1904 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1905 				}
1906 				tp->t_state = TCPS_FIN_WAIT_2;
1907 			}
1908 			break;
1909 
1910 		/*
1911 		 * In CLOSING STATE in addition to the processing for
1912 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1913 		 * then enter the TIME-WAIT state, otherwise ignore
1914 		 * the segment.
1915 		 */
1916 		case TCPS_CLOSING:
1917 			if (ourfinisacked) {
1918 				tp->t_state = TCPS_TIME_WAIT;
1919 				tcp_canceltimers(tp);
1920 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1921 				soisdisconnected(so);
1922 			}
1923 			break;
1924 
1925 		/*
1926 		 * In LAST_ACK, we may still be waiting for data to drain
1927 		 * and/or to be acked, as well as for the ack of our FIN.
1928 		 * If our FIN is now acknowledged, delete the TCB,
1929 		 * enter the closed state and return.
1930 		 */
1931 		case TCPS_LAST_ACK:
1932 			if (ourfinisacked) {
1933 				tp = tcp_close(tp);
1934 				goto drop;
1935 			}
1936 			break;
1937 
1938 		/*
1939 		 * In TIME_WAIT state the only thing that should arrive
1940 		 * is a retransmission of the remote FIN.  Acknowledge
1941 		 * it and restart the finack timer.
1942 		 */
1943 		case TCPS_TIME_WAIT:
1944 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1945 			goto dropafterack;
1946 		}
1947 	}
1948 
1949 step6:
1950 	/*
1951 	 * Update window information.
1952 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1953 	 */
1954 	if ((tiflags & TH_ACK) &&
1955 	    (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq &&
1956 	    (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1957 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1958 		/* keep track of pure window updates */
1959 		if (tlen == 0 &&
1960 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1961 			tcpstat.tcps_rcvwinupd++;
1962 		tp->snd_wnd = tiwin;
1963 		tp->snd_wl1 = th->th_seq;
1964 		tp->snd_wl2 = th->th_ack;
1965 		if (tp->snd_wnd > tp->max_sndwnd)
1966 			tp->max_sndwnd = tp->snd_wnd;
1967 		needoutput = 1;
1968 	}
1969 
1970 	/*
1971 	 * Process segments with URG.
1972 	 */
1973 	if ((tiflags & TH_URG) && th->th_urp &&
1974 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1975 		/*
1976 		 * This is a kludge, but if we receive and accept
1977 		 * random urgent pointers, we'll crash in
1978 		 * soreceive.  It's hard to imagine someone
1979 		 * actually wanting to send this much urgent data.
1980 		 */
1981 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
1982 			th->th_urp = 0;			/* XXX */
1983 			tiflags &= ~TH_URG;		/* XXX */
1984 			goto dodata;			/* XXX */
1985 		}
1986 		/*
1987 		 * If this segment advances the known urgent pointer,
1988 		 * then mark the data stream.  This should not happen
1989 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1990 		 * a FIN has been received from the remote side.
1991 		 * In these states we ignore the URG.
1992 		 *
1993 		 * According to RFC961 (Assigned Protocols),
1994 		 * the urgent pointer points to the last octet
1995 		 * of urgent data.  We continue, however,
1996 		 * to consider it to indicate the first octet
1997 		 * of data past the urgent section as the original
1998 		 * spec states (in one of two places).
1999 		 */
2000 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2001 			tp->rcv_up = th->th_seq + th->th_urp;
2002 			so->so_oobmark = so->so_rcv.sb_cc +
2003 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2004 			if (so->so_oobmark == 0)
2005 				so->so_state |= SS_RCVATMARK;
2006 			sohasoutofband(so);
2007 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2008 		}
2009 		/*
2010 		 * Remove out of band data so doesn't get presented to user.
2011 		 * This can happen independent of advancing the URG pointer,
2012 		 * but if two URG's are pending at once, some out-of-band
2013 		 * data may creep in... ick.
2014 		 */
2015 		if (th->th_urp <= (u_int16_t) tlen
2016 #ifdef SO_OOBINLINE
2017 		     && (so->so_options & SO_OOBINLINE) == 0
2018 #endif
2019 		     )
2020 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2021 	} else
2022 		/*
2023 		 * If no out of band data is expected,
2024 		 * pull receive urgent pointer along
2025 		 * with the receive window.
2026 		 */
2027 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2028 			tp->rcv_up = tp->rcv_nxt;
2029 dodata:							/* XXX */
2030 
2031 	/*
2032 	 * Process the segment text, merging it into the TCP sequencing queue,
2033 	 * and arranging for acknowledgment of receipt if necessary.
2034 	 * This process logically involves adjusting tp->rcv_wnd as data
2035 	 * is presented to the user (this happens in tcp_usrreq.c,
2036 	 * case PRU_RCVD).  If a FIN has already been received on this
2037 	 * connection then we just ignore the text.
2038 	 */
2039 	if ((tlen || (tiflags & TH_FIN)) &&
2040 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2041 #ifdef TCP_SACK
2042 		tcp_seq laststart = th->th_seq;
2043 		tcp_seq lastend = th->th_seq + tlen;
2044 #endif
2045 		tcp_reass_lock(tp);
2046 		if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) &&
2047 		    tp->t_state == TCPS_ESTABLISHED) {
2048 			tcp_reass_unlock(tp);
2049 			TCP_SETUP_ACK(tp, tiflags);
2050 			tp->rcv_nxt += tlen;
2051 			tiflags = th->th_flags & TH_FIN;
2052 			tcpstat.tcps_rcvpack++;
2053 			tcpstat.tcps_rcvbyte += tlen;
2054 			ND6_HINT(tp);
2055 			if (so->so_state & SS_CANTRCVMORE)
2056 				m_freem(m);
2057 			else {
2058 				m_adj(m, hdroptlen);
2059 				sbappendstream(&so->so_rcv, m);
2060 			}
2061 			sorwakeup(so);
2062 		} else {
2063 			m_adj(m, hdroptlen);
2064 			tiflags = tcp_reass(tp, th, m, &tlen);
2065 			tcp_reass_unlock(tp);
2066 			tp->t_flags |= TF_ACKNOW;
2067 		}
2068 #ifdef TCP_SACK
2069 		if (tp->sack_enable)
2070 			tcp_update_sack_list(tp, laststart, lastend);
2071 #endif
2072 
2073 		/*
2074 		 * variable len never referenced again in modern BSD,
2075 		 * so why bother computing it ??
2076 		 */
2077 #if 0
2078 		/*
2079 		 * Note the amount of data that peer has sent into
2080 		 * our window, in order to estimate the sender's
2081 		 * buffer size.
2082 		 */
2083 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2084 #endif /* 0 */
2085 	} else {
2086 		m_freem(m);
2087 		tiflags &= ~TH_FIN;
2088 	}
2089 
2090 	/*
2091 	 * If FIN is received ACK the FIN and let the user know
2092 	 * that the connection is closing.  Ignore a FIN received before
2093 	 * the connection is fully established.
2094 	 */
2095 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2096 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2097 			socantrcvmore(so);
2098 			tp->t_flags |= TF_ACKNOW;
2099 			tp->rcv_nxt++;
2100 		}
2101 		switch (tp->t_state) {
2102 
2103 		/*
2104 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2105 		 */
2106 		case TCPS_ESTABLISHED:
2107 			tp->t_state = TCPS_CLOSE_WAIT;
2108 			break;
2109 
2110 		/*
2111 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2112 		 * enter the CLOSING state.
2113 		 */
2114 		case TCPS_FIN_WAIT_1:
2115 			tp->t_state = TCPS_CLOSING;
2116 			break;
2117 
2118 		/*
2119 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2120 		 * starting the time-wait timer, turning off the other
2121 		 * standard timers.
2122 		 */
2123 		case TCPS_FIN_WAIT_2:
2124 			tp->t_state = TCPS_TIME_WAIT;
2125 			tcp_canceltimers(tp);
2126 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2127 			soisdisconnected(so);
2128 			break;
2129 
2130 		/*
2131 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2132 		 */
2133 		case TCPS_TIME_WAIT:
2134 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2135 			break;
2136 		}
2137 	}
2138 	if (so->so_options & SO_DEBUG) {
2139 		switch (tp->pf) {
2140 #ifdef INET6
2141 		case PF_INET6:
2142 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2143 			    0, tlen);
2144 			break;
2145 #endif /* INET6 */
2146 		case PF_INET:
2147 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2148 			    0, tlen);
2149 			break;
2150 		}
2151 	}
2152 
2153 	/*
2154 	 * Return any desired output.
2155 	 */
2156 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2157 		(void) tcp_output(tp);
2158 	}
2159 	return;
2160 
2161 badsyn:
2162 	/*
2163 	 * Received a bad SYN.  Increment counters and dropwithreset.
2164 	 */
2165 	tcpstat.tcps_badsyn++;
2166 	tp = NULL;
2167 	goto dropwithreset;
2168 
2169 dropafterack_ratelim:
2170 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2171 	    tcp_ackdrop_ppslim) == 0) {
2172 		/* XXX stat */
2173 		goto drop;
2174 	}
2175 	/* ...fall into dropafterack... */
2176 
2177 dropafterack:
2178 	/*
2179 	 * Generate an ACK dropping incoming segment if it occupies
2180 	 * sequence space, where the ACK reflects our state.
2181 	 */
2182 	if (tiflags & TH_RST)
2183 		goto drop;
2184 	m_freem(m);
2185 	tp->t_flags |= TF_ACKNOW;
2186 	(void) tcp_output(tp);
2187 	return;
2188 
2189 dropwithreset_ratelim:
2190 	/*
2191 	 * We may want to rate-limit RSTs in certain situations,
2192 	 * particularly if we are sending an RST in response to
2193 	 * an attempt to connect to or otherwise communicate with
2194 	 * a port for which we have no socket.
2195 	 */
2196 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2197 	    tcp_rst_ppslim) == 0) {
2198 		/* XXX stat */
2199 		goto drop;
2200 	}
2201 	/* ...fall into dropwithreset... */
2202 
2203 dropwithreset:
2204 	/*
2205 	 * Generate a RST, dropping incoming segment.
2206 	 * Make ACK acceptable to originator of segment.
2207 	 * Don't bother to respond to RST.
2208 	 */
2209 	if (tiflags & TH_RST)
2210 		goto drop;
2211 	if (tiflags & TH_ACK) {
2212 		tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack,
2213 		    TH_RST);
2214 	} else {
2215 		if (tiflags & TH_SYN)
2216 			tlen++;
2217 		tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen,
2218 		    (tcp_seq)0, TH_RST|TH_ACK);
2219 	}
2220 	m_freem(m);
2221 	return;
2222 
2223 drop:
2224 	/*
2225 	 * Drop space held by incoming segment and return.
2226 	 */
2227 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2228 		switch (tp->pf) {
2229 #ifdef INET6
2230 		case PF_INET6:
2231 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2232 			    0, tlen);
2233 			break;
2234 #endif /* INET6 */
2235 		case PF_INET:
2236 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2237 			    0, tlen);
2238 			break;
2239 		}
2240 	}
2241 
2242 	m_freem(m);
2243 	return;
2244 }
2245 
2246 int
2247 tcp_dooptions(tp, cp, cnt, th, m, iphlen, oi)
2248 	struct tcpcb *tp;
2249 	u_char *cp;
2250 	int cnt;
2251 	struct tcphdr *th;
2252 	struct mbuf *m;
2253 	int iphlen;
2254 	struct tcp_opt_info *oi;
2255 {
2256 	u_int16_t mss = 0;
2257 	int opt, optlen;
2258 #ifdef TCP_SIGNATURE
2259 	caddr_t sigp = NULL;
2260 	struct tdb *tdb = NULL;
2261 #endif /* TCP_SIGNATURE */
2262 
2263 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2264 		opt = cp[0];
2265 		if (opt == TCPOPT_EOL)
2266 			break;
2267 		if (opt == TCPOPT_NOP)
2268 			optlen = 1;
2269 		else {
2270 			if (cnt < 2)
2271 				break;
2272 			optlen = cp[1];
2273 			if (optlen < 2 || optlen > cnt)
2274 				break;
2275 		}
2276 		switch (opt) {
2277 
2278 		default:
2279 			continue;
2280 
2281 		case TCPOPT_MAXSEG:
2282 			if (optlen != TCPOLEN_MAXSEG)
2283 				continue;
2284 			if (!(th->th_flags & TH_SYN))
2285 				continue;
2286 			if (TCPS_HAVERCVDSYN(tp->t_state))
2287 				continue;
2288 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2289 			NTOHS(mss);
2290 			oi->maxseg = mss;
2291 			break;
2292 
2293 		case TCPOPT_WINDOW:
2294 			if (optlen != TCPOLEN_WINDOW)
2295 				continue;
2296 			if (!(th->th_flags & TH_SYN))
2297 				continue;
2298 			if (TCPS_HAVERCVDSYN(tp->t_state))
2299 				continue;
2300 			tp->t_flags |= TF_RCVD_SCALE;
2301 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2302 			break;
2303 
2304 		case TCPOPT_TIMESTAMP:
2305 			if (optlen != TCPOLEN_TIMESTAMP)
2306 				continue;
2307 			oi->ts_present = 1;
2308 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2309 			NTOHL(oi->ts_val);
2310 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2311 			NTOHL(oi->ts_ecr);
2312 
2313 			if (!(th->th_flags & TH_SYN))
2314 				continue;
2315 			if (TCPS_HAVERCVDSYN(tp->t_state))
2316 				continue;
2317 			/*
2318 			 * A timestamp received in a SYN makes
2319 			 * it ok to send timestamp requests and replies.
2320 			 */
2321 			tp->t_flags |= TF_RCVD_TSTMP;
2322 			tp->ts_recent = oi->ts_val;
2323 			tp->ts_recent_age = tcp_now;
2324 			break;
2325 
2326 #ifdef TCP_SACK
2327 		case TCPOPT_SACK_PERMITTED:
2328 			if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED)
2329 				continue;
2330 			if (!(th->th_flags & TH_SYN))
2331 				continue;
2332 			if (TCPS_HAVERCVDSYN(tp->t_state))
2333 				continue;
2334 			/* MUST only be set on SYN */
2335 			tp->t_flags |= TF_SACK_PERMIT;
2336 			break;
2337 		case TCPOPT_SACK:
2338 			tcp_sack_option(tp, th, cp, optlen);
2339 			break;
2340 #endif
2341 #ifdef TCP_SIGNATURE
2342 		case TCPOPT_SIGNATURE:
2343 			if (optlen != TCPOLEN_SIGNATURE)
2344 				continue;
2345 
2346 			if (sigp && bcmp(sigp, cp + 2, 16))
2347 				return (-1);
2348 
2349 			sigp = cp + 2;
2350 			break;
2351 #endif /* TCP_SIGNATURE */
2352 		}
2353 	}
2354 
2355 #ifdef TCP_SIGNATURE
2356 	if (tp->t_flags & TF_SIGNATURE) {
2357 		union sockaddr_union src, dst;
2358 
2359 		memset(&src, 0, sizeof(union sockaddr_union));
2360 		memset(&dst, 0, sizeof(union sockaddr_union));
2361 
2362 		switch (tp->pf) {
2363 		case 0:
2364 #ifdef INET
2365 		case AF_INET:
2366 			src.sa.sa_len = sizeof(struct sockaddr_in);
2367 			src.sa.sa_family = AF_INET;
2368 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
2369 			dst.sa.sa_len = sizeof(struct sockaddr_in);
2370 			dst.sa.sa_family = AF_INET;
2371 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
2372 			break;
2373 #endif
2374 #ifdef INET6
2375 		case AF_INET6:
2376 			src.sa.sa_len = sizeof(struct sockaddr_in6);
2377 			src.sa.sa_family = AF_INET6;
2378 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
2379 			dst.sa.sa_len = sizeof(struct sockaddr_in6);
2380 			dst.sa.sa_family = AF_INET6;
2381 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
2382 			break;
2383 #endif /* INET6 */
2384 		}
2385 
2386 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
2387 
2388 		/*
2389 		 * We don't have an SA for this peer, so we turn off
2390 		 * TF_SIGNATURE on the listen socket
2391 		 */
2392 		if (tdb == NULL && tp->t_state == TCPS_LISTEN)
2393 			tp->t_flags &= ~TF_SIGNATURE;
2394 
2395 	}
2396 
2397 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
2398 		tcpstat.tcps_rcvbadsig++;
2399 		return (-1);
2400 	}
2401 
2402 	if (sigp) {
2403 		char sig[16];
2404 
2405 		if (tdb == NULL) {
2406 			tcpstat.tcps_rcvbadsig++;
2407 			return (-1);
2408 		}
2409 
2410 		if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0)
2411 			return (-1);
2412 
2413 		if (bcmp(sig, sigp, 16)) {
2414 			tcpstat.tcps_rcvbadsig++;
2415 			return (-1);
2416 		}
2417 
2418 		tcpstat.tcps_rcvgoodsig++;
2419 	}
2420 #endif /* TCP_SIGNATURE */
2421 
2422 	return (0);
2423 }
2424 
2425 #if defined(TCP_SACK)
2426 u_long
2427 tcp_seq_subtract(a, b)
2428 	u_long a, b;
2429 {
2430 	return ((long)(a - b));
2431 }
2432 #endif
2433 
2434 
2435 #ifdef TCP_SACK
2436 /*
2437  * This function is called upon receipt of new valid data (while not in header
2438  * prediction mode), and it updates the ordered list of sacks.
2439  */
2440 void
2441 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart,
2442     tcp_seq rcv_lastend)
2443 {
2444 	/*
2445 	 * First reported block MUST be the most recent one.  Subsequent
2446 	 * blocks SHOULD be in the order in which they arrived at the
2447 	 * receiver.  These two conditions make the implementation fully
2448 	 * compliant with RFC 2018.
2449 	 */
2450 	int i, j = 0, count = 0, lastpos = -1;
2451 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2452 
2453 	/* First clean up current list of sacks */
2454 	for (i = 0; i < tp->rcv_numsacks; i++) {
2455 		sack = tp->sackblks[i];
2456 		if (sack.start == 0 && sack.end == 0) {
2457 			count++; /* count = number of blocks to be discarded */
2458 			continue;
2459 		}
2460 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2461 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2462 			count++;
2463 		} else {
2464 			temp[j].start = tp->sackblks[i].start;
2465 			temp[j++].end = tp->sackblks[i].end;
2466 		}
2467 	}
2468 	tp->rcv_numsacks -= count;
2469 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2470 		tcp_clean_sackreport(tp);
2471 		if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) {
2472 			/* ==> need first sack block */
2473 			tp->sackblks[0].start = rcv_laststart;
2474 			tp->sackblks[0].end = rcv_lastend;
2475 			tp->rcv_numsacks = 1;
2476 		}
2477 		return;
2478 	}
2479 	/* Otherwise, sack blocks are already present. */
2480 	for (i = 0; i < tp->rcv_numsacks; i++)
2481 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2482 	if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend))
2483 		return;     /* sack list remains unchanged */
2484 	/*
2485 	 * From here, segment just received should be (part of) the 1st sack.
2486 	 * Go through list, possibly coalescing sack block entries.
2487 	 */
2488 	firstsack.start = rcv_laststart;
2489 	firstsack.end = rcv_lastend;
2490 	for (i = 0; i < tp->rcv_numsacks; i++) {
2491 		sack = tp->sackblks[i];
2492 		if (SEQ_LT(sack.end, firstsack.start) ||
2493 		    SEQ_GT(sack.start, firstsack.end))
2494 			continue; /* no overlap */
2495 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2496 			/*
2497 			 * identical block; delete it here since we will
2498 			 * move it to the front of the list.
2499 			 */
2500 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2501 			lastpos = i;    /* last posn with a zero entry */
2502 			continue;
2503 		}
2504 		if (SEQ_LEQ(sack.start, firstsack.start))
2505 			firstsack.start = sack.start; /* merge blocks */
2506 		if (SEQ_GEQ(sack.end, firstsack.end))
2507 			firstsack.end = sack.end;     /* merge blocks */
2508 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2509 		lastpos = i;    /* last posn with a zero entry */
2510 	}
2511 	if (lastpos != -1) {    /* at least one merge */
2512 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2513 			sack = tp->sackblks[i];
2514 			if (sack.start == 0 && sack.end == 0)
2515 				continue;
2516 			temp[j++] = sack;
2517 		}
2518 		tp->rcv_numsacks = j; /* including first blk (added later) */
2519 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2520 			tp->sackblks[i] = temp[i];
2521 	} else {        /* no merges -- shift sacks by 1 */
2522 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2523 			tp->rcv_numsacks++;
2524 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2525 			tp->sackblks[i] = tp->sackblks[i-1];
2526 	}
2527 	tp->sackblks[0] = firstsack;
2528 	return;
2529 }
2530 
2531 /*
2532  * Process the TCP SACK option.  tp->snd_holes is an ordered list
2533  * of holes (oldest to newest, in terms of the sequence space).
2534  */
2535 void
2536 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2537 {
2538 	int tmp_olen;
2539 	u_char *tmp_cp;
2540 	struct sackhole *cur, *p, *temp;
2541 
2542 	if (!tp->sack_enable)
2543 		return;
2544 	/* SACK without ACK doesn't make sense. */
2545 	if ((th->th_flags & TH_ACK) == 0)
2546 	       return;
2547 	/* Make sure the ACK on this segment is in [snd_una, snd_max]. */
2548 	if (SEQ_LT(th->th_ack, tp->snd_una) ||
2549 	    SEQ_GT(th->th_ack, tp->snd_max))
2550 		return;
2551 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2552 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2553 		return;
2554 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2555 	tmp_cp = cp + 2;
2556 	tmp_olen = optlen - 2;
2557 	tcpstat.tcps_sack_rcv_opts++;
2558 	if (tp->snd_numholes < 0)
2559 		tp->snd_numholes = 0;
2560 	if (tp->t_maxseg == 0)
2561 		panic("tcp_sack_option"); /* Should never happen */
2562 	while (tmp_olen > 0) {
2563 		struct sackblk sack;
2564 
2565 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2566 		NTOHL(sack.start);
2567 		bcopy(tmp_cp + sizeof(tcp_seq),
2568 		    (char *) &(sack.end), sizeof(tcp_seq));
2569 		NTOHL(sack.end);
2570 		tmp_olen -= TCPOLEN_SACK;
2571 		tmp_cp += TCPOLEN_SACK;
2572 		if (SEQ_LEQ(sack.end, sack.start))
2573 			continue; /* bad SACK fields */
2574 		if (SEQ_LEQ(sack.end, tp->snd_una))
2575 			continue; /* old block */
2576 #if defined(TCP_SACK) && defined(TCP_FACK)
2577 		/* Updates snd_fack.  */
2578 		if (SEQ_GT(sack.end, tp->snd_fack))
2579 			tp->snd_fack = sack.end;
2580 #endif /* TCP_FACK */
2581 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2582 			if (SEQ_LT(sack.start, th->th_ack))
2583 				continue;
2584 		}
2585 		if (SEQ_GT(sack.end, tp->snd_max))
2586 			continue;
2587 		if (tp->snd_holes == NULL) { /* first hole */
2588 			tp->snd_holes = (struct sackhole *)
2589 			    pool_get(&sackhl_pool, PR_NOWAIT);
2590 			if (tp->snd_holes == NULL) {
2591 				/* ENOBUFS, so ignore SACKed block for now*/
2592 				goto done;
2593 			}
2594 			cur = tp->snd_holes;
2595 			cur->start = th->th_ack;
2596 			cur->end = sack.start;
2597 			cur->rxmit = cur->start;
2598 			cur->next = NULL;
2599 			tp->snd_numholes = 1;
2600 			tp->rcv_lastsack = sack.end;
2601 			/*
2602 			 * dups is at least one.  If more data has been
2603 			 * SACKed, it can be greater than one.
2604 			 */
2605 			cur->dups = min(tcprexmtthresh,
2606 			    ((sack.end - cur->end)/tp->t_maxseg));
2607 			if (cur->dups < 1)
2608 				cur->dups = 1;
2609 			continue; /* with next sack block */
2610 		}
2611 		/* Go thru list of holes:  p = previous,  cur = current */
2612 		p = cur = tp->snd_holes;
2613 		while (cur) {
2614 			if (SEQ_LEQ(sack.end, cur->start))
2615 				/* SACKs data before the current hole */
2616 				break; /* no use going through more holes */
2617 			if (SEQ_GEQ(sack.start, cur->end)) {
2618 				/* SACKs data beyond the current hole */
2619 				cur->dups++;
2620 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2621 				    tcprexmtthresh)
2622 					cur->dups = tcprexmtthresh;
2623 				p = cur;
2624 				cur = cur->next;
2625 				continue;
2626 			}
2627 			if (SEQ_LEQ(sack.start, cur->start)) {
2628 				/* Data acks at least the beginning of hole */
2629 #if defined(TCP_SACK) && defined(TCP_FACK)
2630 				if (SEQ_GT(sack.end, cur->rxmit))
2631 					tp->retran_data -=
2632 				    	    tcp_seq_subtract(cur->rxmit,
2633 					    cur->start);
2634 				else
2635 					tp->retran_data -=
2636 					    tcp_seq_subtract(sack.end,
2637 					    cur->start);
2638 #endif /* TCP_FACK */
2639 				if (SEQ_GEQ(sack.end, cur->end)) {
2640 					/* Acks entire hole, so delete hole */
2641 					if (p != cur) {
2642 						p->next = cur->next;
2643 						pool_put(&sackhl_pool, cur);
2644 						cur = p->next;
2645 					} else {
2646 						cur = cur->next;
2647 						pool_put(&sackhl_pool, p);
2648 						p = cur;
2649 						tp->snd_holes = p;
2650 					}
2651 					tp->snd_numholes--;
2652 					continue;
2653 				}
2654 				/* otherwise, move start of hole forward */
2655 				cur->start = sack.end;
2656 				cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
2657 				p = cur;
2658 				cur = cur->next;
2659 				continue;
2660 			}
2661 			/* move end of hole backward */
2662 			if (SEQ_GEQ(sack.end, cur->end)) {
2663 #if defined(TCP_SACK) && defined(TCP_FACK)
2664 				if (SEQ_GT(cur->rxmit, sack.start))
2665 					tp->retran_data -=
2666 					    tcp_seq_subtract(cur->rxmit,
2667 					    sack.start);
2668 #endif /* TCP_FACK */
2669 				cur->end = sack.start;
2670 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2671 				cur->dups++;
2672 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2673 				    tcprexmtthresh)
2674 					cur->dups = tcprexmtthresh;
2675 				p = cur;
2676 				cur = cur->next;
2677 				continue;
2678 			}
2679 			if (SEQ_LT(cur->start, sack.start) &&
2680 			    SEQ_GT(cur->end, sack.end)) {
2681 				/*
2682 				 * ACKs some data in middle of a hole; need to
2683 				 * split current hole
2684 				 */
2685 				temp = (struct sackhole *)
2686 				    pool_get(&sackhl_pool, PR_NOWAIT);
2687 				if (temp == NULL)
2688 					goto done; /* ENOBUFS */
2689 #if defined(TCP_SACK) && defined(TCP_FACK)
2690 				if (SEQ_GT(cur->rxmit, sack.end))
2691 					tp->retran_data -=
2692 					    tcp_seq_subtract(sack.end,
2693 					    sack.start);
2694 				else if (SEQ_GT(cur->rxmit, sack.start))
2695 					tp->retran_data -=
2696 					    tcp_seq_subtract(cur->rxmit,
2697 					    sack.start);
2698 #endif /* TCP_FACK */
2699 				temp->next = cur->next;
2700 				temp->start = sack.end;
2701 				temp->end = cur->end;
2702 				temp->dups = cur->dups;
2703 				temp->rxmit = SEQ_MAX(cur->rxmit, temp->start);
2704 				cur->end = sack.start;
2705 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2706 				cur->dups++;
2707 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2708 					tcprexmtthresh)
2709 					cur->dups = tcprexmtthresh;
2710 				cur->next = temp;
2711 				p = temp;
2712 				cur = p->next;
2713 				tp->snd_numholes++;
2714 			}
2715 		}
2716 		/* At this point, p points to the last hole on the list */
2717 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2718 			/*
2719 			 * Need to append new hole at end.
2720 			 * Last hole is p (and it's not NULL).
2721 			 */
2722 			temp = (struct sackhole *)
2723 			    pool_get(&sackhl_pool, PR_NOWAIT);
2724 			if (temp == NULL)
2725 				goto done; /* ENOBUFS */
2726 			temp->start = tp->rcv_lastsack;
2727 			temp->end = sack.start;
2728 			temp->dups = min(tcprexmtthresh,
2729 			    ((sack.end - sack.start)/tp->t_maxseg));
2730 			if (temp->dups < 1)
2731 				temp->dups = 1;
2732 			temp->rxmit = temp->start;
2733 			temp->next = 0;
2734 			p->next = temp;
2735 			tp->rcv_lastsack = sack.end;
2736 			tp->snd_numholes++;
2737 		}
2738 	}
2739 done:
2740 #if defined(TCP_SACK) && defined(TCP_FACK)
2741 	/*
2742 	 * Update retran_data and snd_awnd.  Go through the list of
2743 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2744 	 */
2745 	tp->retran_data = 0;
2746 	cur = tp->snd_holes;
2747 	while (cur) {
2748 		tp->retran_data += cur->rxmit - cur->start;
2749 		cur = cur->next;
2750 	}
2751 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2752 	    tp->retran_data;
2753 #endif /* TCP_FACK */
2754 
2755 	return;
2756 }
2757 
2758 /*
2759  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2760  * it is completely acked; otherwise, tcp_sack_option(), called from
2761  * tcp_dooptions(), will fix up the hole.
2762  */
2763 void
2764 tcp_del_sackholes(tp, th)
2765 	struct tcpcb *tp;
2766 	struct tcphdr *th;
2767 {
2768 	if (tp->sack_enable && tp->t_state != TCPS_LISTEN) {
2769 		/* max because this could be an older ack just arrived */
2770 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2771 			th->th_ack : tp->snd_una;
2772 		struct sackhole *cur = tp->snd_holes;
2773 		struct sackhole *prev;
2774 		while (cur)
2775 			if (SEQ_LEQ(cur->end, lastack)) {
2776 				prev = cur;
2777 				cur = cur->next;
2778 				pool_put(&sackhl_pool, prev);
2779 				tp->snd_numholes--;
2780 			} else if (SEQ_LT(cur->start, lastack)) {
2781 				cur->start = lastack;
2782 				if (SEQ_LT(cur->rxmit, cur->start))
2783 					cur->rxmit = cur->start;
2784 				break;
2785 			} else
2786 				break;
2787 		tp->snd_holes = cur;
2788 	}
2789 }
2790 
2791 /*
2792  * Delete all receiver-side SACK information.
2793  */
2794 void
2795 tcp_clean_sackreport(tp)
2796 	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(tp, th)
2813 	struct tcpcb *tp;
2814 	struct tcphdr *th;
2815 {
2816 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2817 		/* Turn off retx. timer (will start again next segment) */
2818 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2819 		tp->t_rtttime = 0;
2820 #ifndef TCP_FACK
2821 		/*
2822 		 * Partial window deflation.  This statement relies on the
2823 		 * fact that tp->snd_una has not been updated yet.  In FACK
2824 		 * hold snd_cwnd constant during fast recovery.
2825 		 */
2826 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2827 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2828 			tp->snd_cwnd += tp->t_maxseg;
2829 		} else
2830 			tp->snd_cwnd = tp->t_maxseg;
2831 #endif
2832 		return (1);
2833 	}
2834 	return (0);
2835 }
2836 #endif /* TCP_SACK */
2837 
2838 /*
2839  * Pull out of band byte out of a segment so
2840  * it doesn't appear in the user's data queue.
2841  * It is still reflected in the segment length for
2842  * sequencing purposes.
2843  */
2844 void
2845 tcp_pulloutofband(so, urgent, m, off)
2846 	struct socket *so;
2847 	u_int urgent;
2848 	struct mbuf *m;
2849 	int off;
2850 {
2851         int cnt = off + urgent - 1;
2852 
2853 	while (cnt >= 0) {
2854 		if (m->m_len > cnt) {
2855 			char *cp = mtod(m, caddr_t) + cnt;
2856 			struct tcpcb *tp = sototcpcb(so);
2857 
2858 			tp->t_iobc = *cp;
2859 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2860 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2861 			m->m_len--;
2862 			return;
2863 		}
2864 		cnt -= m->m_len;
2865 		m = m->m_next;
2866 		if (m == 0)
2867 			break;
2868 	}
2869 	panic("tcp_pulloutofband");
2870 }
2871 
2872 /*
2873  * Collect new round-trip time estimate
2874  * and update averages and current timeout.
2875  */
2876 void
2877 tcp_xmit_timer(tp, rtt)
2878 	struct tcpcb *tp;
2879 	short rtt;
2880 {
2881 	short delta;
2882 	short rttmin;
2883 
2884 	if (rtt < 0)
2885 		rtt = 0;
2886 	else if (rtt > TCP_RTT_MAX)
2887 		rtt = TCP_RTT_MAX;
2888 
2889 	tcpstat.tcps_rttupdated++;
2890 	if (tp->t_srtt != 0) {
2891 		/*
2892 		 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits
2893 		 * after the binary point (scaled by 4), whereas
2894 		 * srtt is stored as fixed point with 5 bits after the
2895 		 * binary point (i.e., scaled by 32).  The following magic
2896 		 * is equivalent to the smoothing algorithm in rfc793 with
2897 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2898 		 * point).
2899 		 */
2900 		delta = (rtt << TCP_RTT_BASE_SHIFT) -
2901 		    (tp->t_srtt >> TCP_RTT_SHIFT);
2902 		if ((tp->t_srtt += delta) <= 0)
2903 			tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT;
2904 		/*
2905 		 * We accumulate a smoothed rtt variance (actually, a
2906 		 * smoothed mean difference), then set the retransmit
2907 		 * timer to smoothed rtt + 4 times the smoothed variance.
2908 		 * rttvar is stored as fixed point with 4 bits after the
2909 		 * binary point (scaled by 16).  The following is
2910 		 * equivalent to rfc793 smoothing with an alpha of .75
2911 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2912 		 * rfc793's wired-in beta.
2913 		 */
2914 		if (delta < 0)
2915 			delta = -delta;
2916 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2917 		if ((tp->t_rttvar += delta) <= 0)
2918 			tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT;
2919 	} else {
2920 		/*
2921 		 * No rtt measurement yet - use the unsmoothed rtt.
2922 		 * Set the variance to half the rtt (so our first
2923 		 * retransmit happens at 3*rtt).
2924 		 */
2925 		tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
2926 		tp->t_rttvar = (rtt + 1) <<
2927 		    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
2928 	}
2929 	tp->t_rtttime = 0;
2930 	tp->t_rxtshift = 0;
2931 
2932 	/*
2933 	 * the retransmit should happen at rtt + 4 * rttvar.
2934 	 * Because of the way we do the smoothing, srtt and rttvar
2935 	 * will each average +1/2 tick of bias.  When we compute
2936 	 * the retransmit timer, we want 1/2 tick of rounding and
2937 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2938 	 * firing of the timer.  The bias will give us exactly the
2939 	 * 1.5 tick we need.  But, because the bias is
2940 	 * statistical, we have to test that we don't drop below
2941 	 * the minimum feasible timer (which is 2 ticks).
2942 	 */
2943 	rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX);
2944 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2945 
2946 	/*
2947 	 * We received an ack for a packet that wasn't retransmitted;
2948 	 * it is probably safe to discard any error indications we've
2949 	 * received recently.  This isn't quite right, but close enough
2950 	 * for now (a route might have failed after we sent a segment,
2951 	 * and the return path might not be symmetrical).
2952 	 */
2953 	tp->t_softerror = 0;
2954 }
2955 
2956 /*
2957  * Determine a reasonable value for maxseg size.
2958  * If the route is known, check route for mtu.
2959  * If none, use an mss that can be handled on the outgoing
2960  * interface without forcing IP to fragment; if bigger than
2961  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2962  * to utilize large mbufs.  If no route is found, route has no mtu,
2963  * or the destination isn't local, use a default, hopefully conservative
2964  * size (usually 512 or the default IP max size, but no more than the mtu
2965  * of the interface), as we can't discover anything about intervening
2966  * gateways or networks.  We also initialize the congestion/slow start
2967  * window to be a single segment if the destination isn't local.
2968  * While looking at the routing entry, we also initialize other path-dependent
2969  * parameters from pre-set or cached values in the routing entry.
2970  *
2971  * Also take into account the space needed for options that we
2972  * send regularly.  Make maxseg shorter by that amount to assure
2973  * that we can send maxseg amount of data even when the options
2974  * are present.  Store the upper limit of the length of options plus
2975  * data in maxopd.
2976  *
2977  * NOTE: offer == -1 indicates that the maxseg size changed due to
2978  * Path MTU discovery.
2979  */
2980 int
2981 tcp_mss(tp, offer)
2982 	struct tcpcb *tp;
2983 	int offer;
2984 {
2985 	struct rtentry *rt;
2986 	struct ifnet *ifp;
2987 	int mss, mssopt;
2988 	int iphlen;
2989 	struct inpcb *inp;
2990 
2991 	inp = tp->t_inpcb;
2992 
2993 	mssopt = mss = tcp_mssdflt;
2994 
2995 	rt = in_pcbrtentry(inp);
2996 
2997 	if (rt == NULL)
2998 		goto out;
2999 
3000 	ifp = rt->rt_ifp;
3001 
3002 	switch (tp->pf) {
3003 #ifdef INET6
3004 	case AF_INET6:
3005 		iphlen = sizeof(struct ip6_hdr);
3006 		break;
3007 #endif
3008 	case AF_INET:
3009 		iphlen = sizeof(struct ip);
3010 		break;
3011 	default:
3012 		/* the family does not support path MTU discovery */
3013 		goto out;
3014 	}
3015 
3016 #ifdef RTV_MTU
3017 	/*
3018 	 * if there's an mtu associated with the route and we support
3019 	 * path MTU discovery for the underlying protocol family, use it.
3020 	 */
3021 	if (rt->rt_rmx.rmx_mtu) {
3022 		/*
3023 		 * One may wish to lower MSS to take into account options,
3024 		 * especially security-related options.
3025 		 */
3026 		if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) {
3027 			/*
3028 			 * RFC2460 section 5, last paragraph: if path MTU is
3029 			 * smaller than 1280, use 1280 as packet size and
3030 			 * attach fragment header.
3031 			 */
3032 			mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) -
3033 			    sizeof(struct tcphdr);
3034 		} else
3035 			mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr);
3036 	} else
3037 #endif /* RTV_MTU */
3038 	if (!ifp)
3039 		/*
3040 		 * ifp may be null and rmx_mtu may be zero in certain
3041 		 * v6 cases (e.g., if ND wasn't able to resolve the
3042 		 * destination host.
3043 		 */
3044 		goto out;
3045 	else if (ifp->if_flags & IFF_LOOPBACK)
3046 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3047 	else if (tp->pf == AF_INET) {
3048 		if (ip_mtudisc)
3049 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3050 		else if (inp && in_localaddr(inp->inp_faddr))
3051 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3052 	}
3053 #ifdef INET6
3054 	else if (tp->pf == AF_INET6) {
3055 		/*
3056 		 * for IPv6, path MTU discovery is always turned on,
3057 		 * or the node must use packet size <= 1280.
3058 		 */
3059 		mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr);
3060 	}
3061 #endif /* INET6 */
3062 
3063 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3064 	if (offer != -1) {
3065 #ifndef INET6
3066 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3067 #else
3068 		if (tp->pf == AF_INET6)
3069 			mssopt = IN6_LINKMTU(ifp) - iphlen -
3070 			    sizeof(struct tcphdr);
3071 		else
3072 			mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3073 #endif
3074 
3075 		mssopt = max(tcp_mssdflt, mssopt);
3076 	}
3077 
3078  out:
3079 	/*
3080 	 * The current mss, t_maxseg, is initialized to the default value.
3081 	 * If we compute a smaller value, reduce the current mss.
3082 	 * If we compute a larger value, return it for use in sending
3083 	 * a max seg size option, but don't store it for use
3084 	 * unless we received an offer at least that large from peer.
3085 	 *
3086 	 * However, do not accept offers lower than the minimum of
3087 	 * the interface MTU and 216.
3088 	 */
3089 	if (offer > 0)
3090 		tp->t_peermss = offer;
3091 	if (tp->t_peermss)
3092 		mss = min(mss, max(tp->t_peermss, 216));
3093 
3094 	/* sanity - at least max opt. space */
3095 	mss = max(mss, 64);
3096 
3097 	/*
3098 	 * maxopd stores the maximum length of data AND options
3099 	 * in a segment; maxseg is the amount of data in a normal
3100 	 * segment.  We need to store this value (maxopd) apart
3101 	 * from maxseg, because now every segment carries options
3102 	 * and thus we normally have somewhat less data in segments.
3103 	 */
3104 	tp->t_maxopd = mss;
3105 
3106 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3107 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3108 		mss -= TCPOLEN_TSTAMP_APPA;
3109 #ifdef TCP_SIGNATURE
3110 	if (tp->t_flags & TF_SIGNATURE)
3111 		mss -= TCPOLEN_SIGLEN;
3112 #endif
3113 
3114 	if (offer == -1) {
3115 		/* mss changed due to Path MTU discovery */
3116 		tp->t_flags &= ~TF_PMTUD_PEND;
3117 		tp->t_pmtud_mtu_sent = 0;
3118 		tp->t_pmtud_mss_acked = 0;
3119 		if (mss < tp->t_maxseg) {
3120 			/*
3121 			 * Follow suggestion in RFC 2414 to reduce the
3122 			 * congestion window by the ratio of the old
3123 			 * segment size to the new segment size.
3124 			 */
3125 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3126 					     mss, mss);
3127 		}
3128 	} else if (tcp_do_rfc3390) {
3129 		/* increase initial window  */
3130 		tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380));
3131 	} else
3132 		tp->snd_cwnd = mss;
3133 
3134 	tp->t_maxseg = mss;
3135 
3136 	return (offer != -1 ? mssopt : mss);
3137 }
3138 
3139 u_int
3140 tcp_hdrsz(struct tcpcb *tp)
3141 {
3142 	u_int hlen;
3143 
3144 	switch (tp->pf) {
3145 #ifdef INET6
3146 	case AF_INET6:
3147 		hlen = sizeof(struct ip6_hdr);
3148 		break;
3149 #endif
3150 	case AF_INET:
3151 		hlen = sizeof(struct ip);
3152 		break;
3153 	default:
3154 		hlen = 0;
3155 		break;
3156 	}
3157 	hlen += sizeof(struct tcphdr);
3158 
3159 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3160 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3161 		hlen += TCPOLEN_TSTAMP_APPA;
3162 #ifdef TCP_SIGNATURE
3163 	if (tp->t_flags & TF_SIGNATURE)
3164 		hlen += TCPOLEN_SIGLEN;
3165 #endif
3166 	return (hlen);
3167 }
3168 
3169 /*
3170  * Set connection variables based on the effective MSS.
3171  * We are passed the TCPCB for the actual connection.  If we
3172  * are the server, we are called by the compressed state engine
3173  * when the 3-way handshake is complete.  If we are the client,
3174  * we are called when we receive the SYN,ACK from the server.
3175  *
3176  * NOTE: The t_maxseg value must be initialized in the TCPCB
3177  * before this routine is called!
3178  */
3179 void
3180 tcp_mss_update(tp)
3181 	struct tcpcb *tp;
3182 {
3183 	int mss;
3184 	u_long bufsize;
3185 	struct rtentry *rt;
3186 	struct socket *so;
3187 
3188 	so = tp->t_inpcb->inp_socket;
3189 	mss = tp->t_maxseg;
3190 
3191 	rt = in_pcbrtentry(tp->t_inpcb);
3192 
3193 	if (rt == NULL)
3194 		return;
3195 
3196 	bufsize = so->so_snd.sb_hiwat;
3197 	if (bufsize < mss) {
3198 		mss = bufsize;
3199 		/* Update t_maxseg and t_maxopd */
3200 		tcp_mss(tp, mss);
3201 	} else {
3202 		bufsize = roundup(bufsize, mss);
3203 		if (bufsize > sb_max)
3204 			bufsize = sb_max;
3205 		(void)sbreserve(&so->so_snd, bufsize);
3206 	}
3207 
3208 	bufsize = so->so_rcv.sb_hiwat;
3209 	if (bufsize > mss) {
3210 		bufsize = roundup(bufsize, mss);
3211 		if (bufsize > sb_max)
3212 			bufsize = sb_max;
3213 		(void)sbreserve(&so->so_rcv, bufsize);
3214 	}
3215 
3216 }
3217 
3218 #if defined (TCP_SACK)
3219 /*
3220  * Checks for partial ack.  If partial ack arrives, force the retransmission
3221  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3222  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3223  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3224  */
3225 int
3226 tcp_newreno(tp, th)
3227 	struct tcpcb *tp;
3228 	struct tcphdr *th;
3229 {
3230 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3231 		/*
3232 		 * snd_una has not been updated and the socket send buffer
3233 		 * not yet drained of the acked data, so we have to leave
3234 		 * snd_una as it was to get the correct data offset in
3235 		 * tcp_output().
3236 		 */
3237 		tcp_seq onxt = tp->snd_nxt;
3238 		u_long  ocwnd = tp->snd_cwnd;
3239 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3240 		tp->t_rtttime = 0;
3241 		tp->snd_nxt = th->th_ack;
3242 		/*
3243 		 * Set snd_cwnd to one segment beyond acknowledged offset
3244 		 * (tp->snd_una not yet updated when this function is called)
3245 		 */
3246 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3247 		(void) tcp_output(tp);
3248 		tp->snd_cwnd = ocwnd;
3249 		if (SEQ_GT(onxt, tp->snd_nxt))
3250 			tp->snd_nxt = onxt;
3251 		/*
3252 		 * Partial window deflation.  Relies on fact that tp->snd_una
3253 		 * not updated yet.
3254 		 */
3255 		if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3256 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
3257 		else
3258 			tp->snd_cwnd = 0;
3259 		tp->snd_cwnd += tp->t_maxseg;
3260 
3261 		return 1;
3262 	}
3263 	return 0;
3264 }
3265 #endif /* TCP_SACK */
3266 
3267 int
3268 tcp_mss_adv(struct ifnet *ifp, int af)
3269 {
3270 	int mss = 0;
3271 	int iphlen;
3272 
3273 	switch (af) {
3274 	case AF_INET:
3275 		if (ifp != NULL)
3276 			mss = ifp->if_mtu;
3277 		iphlen = sizeof(struct ip);
3278 		break;
3279 #ifdef INET6
3280 	case AF_INET6:
3281 		if (ifp != NULL)
3282 			mss = IN6_LINKMTU(ifp);
3283 		iphlen = sizeof(struct ip6_hdr);
3284 		break;
3285 #endif
3286 	}
3287 	mss = mss - iphlen - sizeof(struct tcphdr);
3288 	return (max(mss, tcp_mssdflt));
3289 }
3290 
3291 /*
3292  * TCP compressed state engine.  Currently used to hold compressed
3293  * state for SYN_RECEIVED.
3294  */
3295 
3296 u_long	syn_cache_count;
3297 u_int32_t syn_hash1, syn_hash2;
3298 
3299 #define SYN_HASH(sa, sp, dp) \
3300 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3301 				     ((u_int32_t)(sp)))^syn_hash2)))
3302 #ifndef INET6
3303 #define	SYN_HASHALL(hash, src, dst) \
3304 do {									\
3305 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
3306 		((struct sockaddr_in *)(src))->sin_port,		\
3307 		((struct sockaddr_in *)(dst))->sin_port);		\
3308 } while (/*CONSTCOND*/ 0)
3309 #else
3310 #define SYN_HASH6(sa, sp, dp) \
3311 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3312 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3313 	 & 0x7fffffff)
3314 
3315 #define SYN_HASHALL(hash, src, dst) \
3316 do {									\
3317 	switch ((src)->sa_family) {					\
3318 	case AF_INET:							\
3319 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
3320 			((struct sockaddr_in *)(src))->sin_port,	\
3321 			((struct sockaddr_in *)(dst))->sin_port);	\
3322 		break;							\
3323 	case AF_INET6:							\
3324 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
3325 			((struct sockaddr_in6 *)(src))->sin6_port,	\
3326 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
3327 		break;							\
3328 	default:							\
3329 		hash = 0;						\
3330 	}								\
3331 } while (/*CONSTCOND*/0)
3332 #endif /* INET6 */
3333 
3334 #define	SYN_CACHE_RM(sc)						\
3335 do {									\
3336 	(sc)->sc_flags |= SCF_DEAD;					\
3337 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
3338 	    (sc), sc_bucketq);						\
3339 	(sc)->sc_tp = NULL;						\
3340 	LIST_REMOVE((sc), sc_tpq);					\
3341 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
3342 	timeout_del(&(sc)->sc_timer);					\
3343 	syn_cache_count--;						\
3344 } while (/*CONSTCOND*/0)
3345 
3346 #define	SYN_CACHE_PUT(sc)						\
3347 do {									\
3348 	if ((sc)->sc_ipopts)						\
3349 		(void) m_free((sc)->sc_ipopts);				\
3350 	if ((sc)->sc_route4.ro_rt != NULL)				\
3351 		RTFREE((sc)->sc_route4.ro_rt);				\
3352 	timeout_set(&(sc)->sc_timer, syn_cache_reaper, (sc));		\
3353 	timeout_add(&(sc)->sc_timer, 0);				\
3354 } while (/*CONSTCOND*/0)
3355 
3356 struct pool syn_cache_pool;
3357 
3358 /*
3359  * We don't estimate RTT with SYNs, so each packet starts with the default
3360  * RTT and each timer step has a fixed timeout value.
3361  */
3362 #define	SYN_CACHE_TIMER_ARM(sc)						\
3363 do {									\
3364 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3365 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3366 	    TCPTV_REXMTMAX);						\
3367 	if (!timeout_initialized(&(sc)->sc_timer))			\
3368 		timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc));	\
3369 	timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \
3370 } while (/*CONSTCOND*/0)
3371 
3372 #define	SYN_CACHE_TIMESTAMP(sc)	tcp_now + (sc)->sc_modulate
3373 
3374 void
3375 syn_cache_init()
3376 {
3377 	int i;
3378 
3379 	/* Initialize the hash buckets. */
3380 	for (i = 0; i < tcp_syn_cache_size; i++)
3381 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3382 
3383 	/* Initialize the syn cache pool. */
3384 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3385 	    "synpl", NULL);
3386 }
3387 
3388 void
3389 syn_cache_insert(sc, tp)
3390 	struct syn_cache *sc;
3391 	struct tcpcb *tp;
3392 {
3393 	struct syn_cache_head *scp;
3394 	struct syn_cache *sc2;
3395 	int s;
3396 
3397 	/*
3398 	 * If there are no entries in the hash table, reinitialize
3399 	 * the hash secrets.
3400 	 */
3401 	if (syn_cache_count == 0) {
3402 		syn_hash1 = arc4random();
3403 		syn_hash2 = arc4random();
3404 	}
3405 
3406 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3407 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3408 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3409 
3410 	/*
3411 	 * Make sure that we don't overflow the per-bucket
3412 	 * limit or the total cache size limit.
3413 	 */
3414 	s = splsoftnet();
3415 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3416 		tcpstat.tcps_sc_bucketoverflow++;
3417 		/*
3418 		 * The bucket is full.  Toss the oldest element in the
3419 		 * bucket.  This will be the first entry in the bucket.
3420 		 */
3421 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3422 #ifdef DIAGNOSTIC
3423 		/*
3424 		 * This should never happen; we should always find an
3425 		 * entry in our bucket.
3426 		 */
3427 		if (sc2 == NULL)
3428 			panic("syn_cache_insert: bucketoverflow: impossible");
3429 #endif
3430 		SYN_CACHE_RM(sc2);
3431 		SYN_CACHE_PUT(sc2);
3432 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3433 		struct syn_cache_head *scp2, *sce;
3434 
3435 		tcpstat.tcps_sc_overflowed++;
3436 		/*
3437 		 * The cache is full.  Toss the oldest entry in the
3438 		 * first non-empty bucket we can find.
3439 		 *
3440 		 * XXX We would really like to toss the oldest
3441 		 * entry in the cache, but we hope that this
3442 		 * condition doesn't happen very often.
3443 		 */
3444 		scp2 = scp;
3445 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3446 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3447 			for (++scp2; scp2 != scp; scp2++) {
3448 				if (scp2 >= sce)
3449 					scp2 = &tcp_syn_cache[0];
3450 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3451 					break;
3452 			}
3453 #ifdef DIAGNOSTIC
3454 			/*
3455 			 * This should never happen; we should always find a
3456 			 * non-empty bucket.
3457 			 */
3458 			if (scp2 == scp)
3459 				panic("syn_cache_insert: cacheoverflow: "
3460 				    "impossible");
3461 #endif
3462 		}
3463 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3464 		SYN_CACHE_RM(sc2);
3465 		SYN_CACHE_PUT(sc2);
3466 	}
3467 
3468 	/*
3469 	 * Initialize the entry's timer.
3470 	 */
3471 	sc->sc_rxttot = 0;
3472 	sc->sc_rxtshift = 0;
3473 	SYN_CACHE_TIMER_ARM(sc);
3474 
3475 	/* Link it from tcpcb entry */
3476 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3477 
3478 	/* Put it into the bucket. */
3479 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3480 	scp->sch_length++;
3481 	syn_cache_count++;
3482 
3483 	tcpstat.tcps_sc_added++;
3484 	splx(s);
3485 }
3486 
3487 /*
3488  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3489  * If we have retransmitted an entry the maximum number of times, expire
3490  * that entry.
3491  */
3492 void
3493 syn_cache_timer(void *arg)
3494 {
3495 	struct syn_cache *sc = arg;
3496 	int s;
3497 
3498 	s = splsoftnet();
3499 	if (sc->sc_flags & SCF_DEAD) {
3500 		splx(s);
3501 		return;
3502 	}
3503 
3504 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3505 		/* Drop it -- too many retransmissions. */
3506 		goto dropit;
3507 	}
3508 
3509 	/*
3510 	 * Compute the total amount of time this entry has
3511 	 * been on a queue.  If this entry has been on longer
3512 	 * than the keep alive timer would allow, expire it.
3513 	 */
3514 	sc->sc_rxttot += sc->sc_rxtcur;
3515 	if (sc->sc_rxttot >= tcptv_keep_init)
3516 		goto dropit;
3517 
3518 	tcpstat.tcps_sc_retransmitted++;
3519 	(void) syn_cache_respond(sc, NULL);
3520 
3521 	/* Advance the timer back-off. */
3522 	sc->sc_rxtshift++;
3523 	SYN_CACHE_TIMER_ARM(sc);
3524 
3525 	splx(s);
3526 	return;
3527 
3528  dropit:
3529 	tcpstat.tcps_sc_timed_out++;
3530 	SYN_CACHE_RM(sc);
3531 	SYN_CACHE_PUT(sc);
3532 	splx(s);
3533 }
3534 
3535 void
3536 syn_cache_reaper(void *arg)
3537 {
3538 	struct syn_cache *sc = arg;
3539 	int s;
3540 
3541 	s = splsoftnet();
3542 	pool_put(&syn_cache_pool, (sc));
3543 	splx(s);
3544 	return;
3545 }
3546 
3547 /*
3548  * Remove syn cache created by the specified tcb entry,
3549  * because this does not make sense to keep them
3550  * (if there's no tcb entry, syn cache entry will never be used)
3551  */
3552 void
3553 syn_cache_cleanup(tp)
3554 	struct tcpcb *tp;
3555 {
3556 	struct syn_cache *sc, *nsc;
3557 	int s;
3558 
3559 	s = splsoftnet();
3560 
3561 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3562 		nsc = LIST_NEXT(sc, sc_tpq);
3563 
3564 #ifdef DIAGNOSTIC
3565 		if (sc->sc_tp != tp)
3566 			panic("invalid sc_tp in syn_cache_cleanup");
3567 #endif
3568 		SYN_CACHE_RM(sc);
3569 		SYN_CACHE_PUT(sc);
3570 	}
3571 	/* just for safety */
3572 	LIST_INIT(&tp->t_sc);
3573 
3574 	splx(s);
3575 }
3576 
3577 /*
3578  * Find an entry in the syn cache.
3579  */
3580 struct syn_cache *
3581 syn_cache_lookup(src, dst, headp)
3582 	struct sockaddr *src;
3583 	struct sockaddr *dst;
3584 	struct syn_cache_head **headp;
3585 {
3586 	struct syn_cache *sc;
3587 	struct syn_cache_head *scp;
3588 	u_int32_t hash;
3589 	int s;
3590 
3591 	SYN_HASHALL(hash, src, dst);
3592 
3593 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3594 	*headp = scp;
3595 	s = splsoftnet();
3596 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3597 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3598 		if (sc->sc_hash != hash)
3599 			continue;
3600 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3601 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3602 			splx(s);
3603 			return (sc);
3604 		}
3605 	}
3606 	splx(s);
3607 	return (NULL);
3608 }
3609 
3610 /*
3611  * This function gets called when we receive an ACK for a
3612  * socket in the LISTEN state.  We look up the connection
3613  * in the syn cache, and if its there, we pull it out of
3614  * the cache and turn it into a full-blown connection in
3615  * the SYN-RECEIVED state.
3616  *
3617  * The return values may not be immediately obvious, and their effects
3618  * can be subtle, so here they are:
3619  *
3620  *	NULL	SYN was not found in cache; caller should drop the
3621  *		packet and send an RST.
3622  *
3623  *	-1	We were unable to create the new connection, and are
3624  *		aborting it.  An ACK,RST is being sent to the peer
3625  *		(unless we got screwey sequence numbners; see below),
3626  *		because the 3-way handshake has been completed.  Caller
3627  *		should not free the mbuf, since we may be using it.  If
3628  *		we are not, we will free it.
3629  *
3630  *	Otherwise, the return value is a pointer to the new socket
3631  *	associated with the connection.
3632  */
3633 struct socket *
3634 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3635 	struct sockaddr *src;
3636 	struct sockaddr *dst;
3637 	struct tcphdr *th;
3638 	unsigned int hlen, tlen;
3639 	struct socket *so;
3640 	struct mbuf *m;
3641 {
3642 	struct syn_cache *sc;
3643 	struct syn_cache_head *scp;
3644 	struct inpcb *inp = NULL;
3645 	struct tcpcb *tp = 0;
3646 	struct mbuf *am;
3647 	int s;
3648 	struct socket *oso;
3649 
3650 	s = splsoftnet();
3651 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3652 		splx(s);
3653 		return (NULL);
3654 	}
3655 
3656 	/*
3657 	 * Verify the sequence and ack numbers.  Try getting the correct
3658 	 * response again.
3659 	 */
3660 	if ((th->th_ack != sc->sc_iss + 1) ||
3661 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3662 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3663 		(void) syn_cache_respond(sc, m);
3664 		splx(s);
3665 		return ((struct socket *)(-1));
3666 	}
3667 
3668 	/* Remove this cache entry */
3669 	SYN_CACHE_RM(sc);
3670 	splx(s);
3671 
3672 	/*
3673 	 * Ok, create the full blown connection, and set things up
3674 	 * as they would have been set up if we had created the
3675 	 * connection when the SYN arrived.  If we can't create
3676 	 * the connection, abort it.
3677 	 */
3678 	oso = so;
3679 	so = sonewconn(so, SS_ISCONNECTED);
3680 	if (so == NULL)
3681 		goto resetandabort;
3682 
3683 	inp = sotoinpcb(oso);
3684 #ifdef IPSEC
3685 	/*
3686 	 * We need to copy the required security levels
3687 	 * from the old pcb. Ditto for any other
3688 	 * IPsec-related information.
3689 	 */
3690 	{
3691 	  struct inpcb *newinp = (struct inpcb *)so->so_pcb;
3692 	  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
3693 		sizeof(inp->inp_seclevel));
3694 	  newinp->inp_secrequire = inp->inp_secrequire;
3695 	  if (inp->inp_ipo != NULL) {
3696 		  newinp->inp_ipo = inp->inp_ipo;
3697 		  inp->inp_ipo->ipo_ref_count++;
3698 	  }
3699 	  if (inp->inp_ipsec_remotecred != NULL) {
3700 		  newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred;
3701 		  inp->inp_ipsec_remotecred->ref_count++;
3702 	  }
3703 	  if (inp->inp_ipsec_remoteauth != NULL) {
3704 		  newinp->inp_ipsec_remoteauth
3705 		      = inp->inp_ipsec_remoteauth;
3706 		  inp->inp_ipsec_remoteauth->ref_count++;
3707 	  }
3708 	}
3709 #endif /* IPSEC */
3710 #ifdef INET6
3711 	/*
3712 	 * inp still has the OLD in_pcb stuff, set the
3713 	 * v6-related flags on the new guy, too.
3714 	 */
3715 	{
3716 	  int flags = inp->inp_flags;
3717 	  struct inpcb *oldinpcb = inp;
3718 
3719 	  inp = (struct inpcb *)so->so_pcb;
3720 	  inp->inp_flags |= (flags & INP_IPV6);
3721 	  if ((inp->inp_flags & INP_IPV6) != 0) {
3722 	    inp->inp_ipv6.ip6_hlim =
3723 	      oldinpcb->inp_ipv6.ip6_hlim;
3724 	  }
3725 	}
3726 #else /* INET6 */
3727 	inp = (struct inpcb *)so->so_pcb;
3728 #endif /* INET6 */
3729 
3730 	inp->inp_lport = th->th_dport;
3731 	switch (src->sa_family) {
3732 #ifdef INET6
3733 	case AF_INET6:
3734 		inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr;
3735 		break;
3736 #endif /* INET6 */
3737 	case AF_INET:
3738 
3739 		inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3740 		inp->inp_options = ip_srcroute();
3741 		if (inp->inp_options == NULL) {
3742 			inp->inp_options = sc->sc_ipopts;
3743 			sc->sc_ipopts = NULL;
3744 		}
3745 		break;
3746 	}
3747 	in_pcbrehash(inp);
3748 
3749 	/*
3750 	 * Give the new socket our cached route reference.
3751 	 */
3752 	if (src->sa_family == AF_INET)
3753 		inp->inp_route = sc->sc_route4;         /* struct assignment */
3754 #ifdef INET6
3755 	else
3756 		inp->inp_route6 = sc->sc_route6;
3757 #endif
3758 	sc->sc_route4.ro_rt = NULL;
3759 
3760 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3761 	if (am == NULL)
3762 		goto resetandabort;
3763 	am->m_len = src->sa_len;
3764 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3765 
3766 	switch (src->sa_family) {
3767 	case AF_INET:
3768 		/* drop IPv4 packet to AF_INET6 socket */
3769 		if (inp->inp_flags & INP_IPV6) {
3770 			(void) m_free(am);
3771 			goto resetandabort;
3772 		}
3773 		if (in_pcbconnect(inp, am)) {
3774 			(void) m_free(am);
3775 			goto resetandabort;
3776 		}
3777 		break;
3778 #ifdef INET6
3779 	case AF_INET6:
3780 		if (in6_pcbconnect(inp, am)) {
3781 			(void) m_free(am);
3782 			goto resetandabort;
3783 		}
3784 		break;
3785 #endif
3786 	}
3787 	(void) m_free(am);
3788 
3789 	tp = intotcpcb(inp);
3790 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3791 	if (sc->sc_request_r_scale != 15) {
3792 		tp->requested_s_scale = sc->sc_requested_s_scale;
3793 		tp->request_r_scale = sc->sc_request_r_scale;
3794 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3795 	}
3796 	if (sc->sc_flags & SCF_TIMESTAMP)
3797 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3798 
3799 	tp->t_template = tcp_template(tp);
3800 	if (tp->t_template == 0) {
3801 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3802 		so = NULL;
3803 		m_freem(m);
3804 		goto abort;
3805 	}
3806 #ifdef TCP_SACK
3807 	tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT;
3808 #endif
3809 
3810 	tp->ts_modulate = sc->sc_modulate;
3811 	tp->iss = sc->sc_iss;
3812 	tp->irs = sc->sc_irs;
3813 	tcp_sendseqinit(tp);
3814 #if defined (TCP_SACK) || defined(TCP_ECN)
3815 	tp->snd_last = tp->snd_una;
3816 #endif /* TCP_SACK */
3817 #if defined(TCP_SACK) && defined(TCP_FACK)
3818 	tp->snd_fack = tp->snd_una;
3819 	tp->retran_data = 0;
3820 	tp->snd_awnd = 0;
3821 #endif /* TCP_FACK */
3822 #ifdef TCP_ECN
3823 	if (sc->sc_flags & SCF_ECN_PERMIT) {
3824 		tp->t_flags |= TF_ECN_PERMIT;
3825 		tcpstat.tcps_ecn_accepts++;
3826 	}
3827 #endif
3828 #ifdef TCP_SACK
3829 	if (sc->sc_flags & SCF_SACK_PERMIT)
3830 		tp->t_flags |= TF_SACK_PERMIT;
3831 #endif
3832 #ifdef TCP_SIGNATURE
3833 	if (sc->sc_flags & SCF_SIGNATURE)
3834 		tp->t_flags |= TF_SIGNATURE;
3835 #endif
3836 	tcp_rcvseqinit(tp);
3837 	tp->t_state = TCPS_SYN_RECEIVED;
3838 	tp->t_rcvtime = tcp_now;
3839 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
3840 	tcpstat.tcps_accepts++;
3841 
3842 	tcp_mss(tp, sc->sc_peermaxseg);	 /* sets t_maxseg */
3843 	if (sc->sc_peermaxseg)
3844 		tcp_mss_update(tp);
3845 	/* Reset initial window to 1 segment for retransmit */
3846 	if (sc->sc_rxtshift > 0)
3847 		tp->snd_cwnd = tp->t_maxseg;
3848 	tp->snd_wl1 = sc->sc_irs;
3849 	tp->rcv_up = sc->sc_irs + 1;
3850 
3851 	/*
3852 	 * This is what whould have happened in tcp_output() when
3853 	 * the SYN,ACK was sent.
3854 	 */
3855 	tp->snd_up = tp->snd_una;
3856 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3857 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3858 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3859 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3860 	tp->last_ack_sent = tp->rcv_nxt;
3861 
3862 	tcpstat.tcps_sc_completed++;
3863 	SYN_CACHE_PUT(sc);
3864 	return (so);
3865 
3866 resetandabort:
3867 	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST);
3868 	m_freem(m);
3869 abort:
3870 	if (so != NULL)
3871 		(void) soabort(so);
3872 	SYN_CACHE_PUT(sc);
3873 	tcpstat.tcps_sc_aborted++;
3874 	return ((struct socket *)(-1));
3875 }
3876 
3877 /*
3878  * This function is called when we get a RST for a
3879  * non-existent connection, so that we can see if the
3880  * connection is in the syn cache.  If it is, zap it.
3881  */
3882 
3883 void
3884 syn_cache_reset(src, dst, th)
3885 	struct sockaddr *src;
3886 	struct sockaddr *dst;
3887 	struct tcphdr *th;
3888 {
3889 	struct syn_cache *sc;
3890 	struct syn_cache_head *scp;
3891 	int s = splsoftnet();
3892 
3893 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3894 		splx(s);
3895 		return;
3896 	}
3897 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3898 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3899 		splx(s);
3900 		return;
3901 	}
3902 	SYN_CACHE_RM(sc);
3903 	splx(s);
3904 	tcpstat.tcps_sc_reset++;
3905 	SYN_CACHE_PUT(sc);
3906 }
3907 
3908 void
3909 syn_cache_unreach(src, dst, th)
3910 	struct sockaddr *src;
3911 	struct sockaddr *dst;
3912 	struct tcphdr *th;
3913 {
3914 	struct syn_cache *sc;
3915 	struct syn_cache_head *scp;
3916 	int s;
3917 
3918 	s = splsoftnet();
3919 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3920 		splx(s);
3921 		return;
3922 	}
3923 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3924 	if (ntohl (th->th_seq) != sc->sc_iss) {
3925 		splx(s);
3926 		return;
3927 	}
3928 
3929 	/*
3930 	 * If we've retransmitted 3 times and this is our second error,
3931 	 * we remove the entry.  Otherwise, we allow it to continue on.
3932 	 * This prevents us from incorrectly nuking an entry during a
3933 	 * spurious network outage.
3934 	 *
3935 	 * See tcp_notify().
3936 	 */
3937 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3938 		sc->sc_flags |= SCF_UNREACH;
3939 		splx(s);
3940 		return;
3941 	}
3942 
3943 	SYN_CACHE_RM(sc);
3944 	splx(s);
3945 	tcpstat.tcps_sc_unreach++;
3946 	SYN_CACHE_PUT(sc);
3947 }
3948 
3949 /*
3950  * Given a LISTEN socket and an inbound SYN request, add
3951  * this to the syn cache, and send back a segment:
3952  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3953  * to the source.
3954  *
3955  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3956  * Doing so would require that we hold onto the data and deliver it
3957  * to the application.  However, if we are the target of a SYN-flood
3958  * DoS attack, an attacker could send data which would eventually
3959  * consume all available buffer space if it were ACKed.  By not ACKing
3960  * the data, we avoid this DoS scenario.
3961  */
3962 
3963 int
3964 syn_cache_add(src, dst, th, iphlen, so, m, optp, optlen, oi, issp)
3965 	struct sockaddr *src;
3966 	struct sockaddr *dst;
3967 	struct tcphdr *th;
3968 	unsigned int iphlen;
3969 	struct socket *so;
3970 	struct mbuf *m;
3971 	u_char *optp;
3972 	int optlen;
3973 	struct tcp_opt_info *oi;
3974 	tcp_seq *issp;
3975 {
3976 	struct tcpcb tb, *tp;
3977 	long win;
3978 	struct syn_cache *sc;
3979 	struct syn_cache_head *scp;
3980 	struct mbuf *ipopts;
3981 
3982 	tp = sototcpcb(so);
3983 
3984 	/*
3985 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3986 	 *
3987 	 * Note this check is performed in tcp_input() very early on.
3988 	 */
3989 
3990 	/*
3991 	 * Initialize some local state.
3992 	 */
3993 	win = sbspace(&so->so_rcv);
3994 	if (win > TCP_MAXWIN)
3995 		win = TCP_MAXWIN;
3996 
3997 #ifdef TCP_SIGNATURE
3998 	if (optp || (tp->t_flags & TF_SIGNATURE)) {
3999 #else
4000 	if (optp) {
4001 #endif
4002 		tb.pf = tp->pf;
4003 #ifdef TCP_SACK
4004 		tb.sack_enable = tp->sack_enable;
4005 #endif
4006 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4007 #ifdef TCP_SIGNATURE
4008 		if (tp->t_flags & TF_SIGNATURE)
4009 			tb.t_flags |= TF_SIGNATURE;
4010 #endif
4011 		tb.t_state = TCPS_LISTEN;
4012 		if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi))
4013 			return (0);
4014 	} else
4015 		tb.t_flags = 0;
4016 
4017 	switch (src->sa_family) {
4018 #ifdef INET
4019 	case AF_INET:
4020 		/*
4021 		 * Remember the IP options, if any.
4022 		 */
4023 		ipopts = ip_srcroute();
4024 		break;
4025 #endif
4026 	default:
4027 		ipopts = NULL;
4028 	}
4029 
4030 	/*
4031 	 * See if we already have an entry for this connection.
4032 	 * If we do, resend the SYN,ACK.  We do not count this
4033 	 * as a retransmission (XXX though maybe we should).
4034 	 */
4035 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4036 		tcpstat.tcps_sc_dupesyn++;
4037 		if (ipopts) {
4038 			/*
4039 			 * If we were remembering a previous source route,
4040 			 * forget it and use the new one we've been given.
4041 			 */
4042 			if (sc->sc_ipopts)
4043 				(void) m_free(sc->sc_ipopts);
4044 			sc->sc_ipopts = ipopts;
4045 		}
4046 		sc->sc_timestamp = tb.ts_recent;
4047 		if (syn_cache_respond(sc, m) == 0) {
4048 			tcpstat.tcps_sndacks++;
4049 			tcpstat.tcps_sndtotal++;
4050 		}
4051 		return (1);
4052 	}
4053 
4054 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4055 	if (sc == NULL) {
4056 		if (ipopts)
4057 			(void) m_free(ipopts);
4058 		return (0);
4059 	}
4060 
4061 	/*
4062 	 * Fill in the cache, and put the necessary IP and TCP
4063 	 * options into the reply.
4064 	 */
4065 	bzero(sc, sizeof(struct syn_cache));
4066 	bzero(&sc->sc_timer, sizeof(sc->sc_timer));
4067 	bcopy(src, &sc->sc_src, src->sa_len);
4068 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4069 	sc->sc_flags = 0;
4070 	sc->sc_ipopts = ipopts;
4071 	sc->sc_irs = th->th_seq;
4072 
4073 	sc->sc_iss = issp ? *issp : arc4random();
4074 	sc->sc_peermaxseg = oi->maxseg;
4075 	sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ?
4076 	    m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family);
4077 	sc->sc_win = win;
4078 	sc->sc_timestamp = tb.ts_recent;
4079 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4080 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
4081 		sc->sc_flags |= SCF_TIMESTAMP;
4082 		sc->sc_modulate = arc4random();
4083 	}
4084 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4085 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4086 		sc->sc_requested_s_scale = tb.requested_s_scale;
4087 		sc->sc_request_r_scale = 0;
4088 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4089 		    TCP_MAXWIN << sc->sc_request_r_scale <
4090 		    so->so_rcv.sb_hiwat)
4091 			sc->sc_request_r_scale++;
4092 	} else {
4093 		sc->sc_requested_s_scale = 15;
4094 		sc->sc_request_r_scale = 15;
4095 	}
4096 #ifdef TCP_ECN
4097 	/*
4098 	 * if both ECE and CWR flag bits are set, peer is ECN capable.
4099 	 */
4100 	if (tcp_do_ecn &&
4101 	    (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR))
4102 		sc->sc_flags |= SCF_ECN_PERMIT;
4103 #endif
4104 #ifdef TCP_SACK
4105 	/*
4106 	 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option
4107 	 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT).
4108 	 */
4109 	if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT))
4110 		sc->sc_flags |= SCF_SACK_PERMIT;
4111 #endif
4112 #ifdef TCP_SIGNATURE
4113 	if (tb.t_flags & TF_SIGNATURE)
4114 		sc->sc_flags |= SCF_SIGNATURE;
4115 #endif
4116 	sc->sc_tp = tp;
4117 	if (syn_cache_respond(sc, m) == 0) {
4118 		syn_cache_insert(sc, tp);
4119 		tcpstat.tcps_sndacks++;
4120 		tcpstat.tcps_sndtotal++;
4121 	} else {
4122 		SYN_CACHE_PUT(sc);
4123 		tcpstat.tcps_sc_dropped++;
4124 	}
4125 	return (1);
4126 }
4127 
4128 int
4129 syn_cache_respond(sc, m)
4130 	struct syn_cache *sc;
4131 	struct mbuf *m;
4132 {
4133 	struct route *ro;
4134 	u_int8_t *optp;
4135 	int optlen, error;
4136 	u_int16_t tlen;
4137 	struct ip *ip = NULL;
4138 #ifdef INET6
4139 	struct ip6_hdr *ip6 = NULL;
4140 #endif
4141 	struct tcphdr *th;
4142 	u_int hlen;
4143 	struct inpcb *inp;
4144 
4145 	switch (sc->sc_src.sa.sa_family) {
4146 	case AF_INET:
4147 		hlen = sizeof(struct ip);
4148 		ro = &sc->sc_route4;
4149 		break;
4150 #ifdef INET6
4151 	case AF_INET6:
4152 		hlen = sizeof(struct ip6_hdr);
4153 		ro = (struct route *)&sc->sc_route6;
4154 		break;
4155 #endif
4156 	default:
4157 		if (m)
4158 			m_freem(m);
4159 		return (EAFNOSUPPORT);
4160 	}
4161 
4162 	/* Compute the size of the TCP options. */
4163 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4164 #ifdef TCP_SACK
4165 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) +
4166 #endif
4167 #ifdef TCP_SIGNATURE
4168 	    ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) +
4169 #endif
4170 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4171 
4172 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4173 
4174 	/*
4175 	 * Create the IP+TCP header from scratch.
4176 	 */
4177 	if (m)
4178 		m_freem(m);
4179 #ifdef DIAGNOSTIC
4180 	if (max_linkhdr + tlen > MCLBYTES)
4181 		return (ENOBUFS);
4182 #endif
4183 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4184 	if (m && max_linkhdr + tlen > MHLEN) {
4185 		MCLGET(m, M_DONTWAIT);
4186 		if ((m->m_flags & M_EXT) == 0) {
4187 			m_freem(m);
4188 			m = NULL;
4189 		}
4190 	}
4191 	if (m == NULL)
4192 		return (ENOBUFS);
4193 
4194 	/* Fixup the mbuf. */
4195 	m->m_data += max_linkhdr;
4196 	m->m_len = m->m_pkthdr.len = tlen;
4197 	m->m_pkthdr.rcvif = NULL;
4198 	memset(mtod(m, u_char *), 0, tlen);
4199 
4200 	switch (sc->sc_src.sa.sa_family) {
4201 	case AF_INET:
4202 		ip = mtod(m, struct ip *);
4203 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4204 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4205 		ip->ip_p = IPPROTO_TCP;
4206 		th = (struct tcphdr *)(ip + 1);
4207 		th->th_dport = sc->sc_src.sin.sin_port;
4208 		th->th_sport = sc->sc_dst.sin.sin_port;
4209 		break;
4210 #ifdef INET6
4211 	case AF_INET6:
4212 		ip6 = mtod(m, struct ip6_hdr *);
4213 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4214 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4215 		ip6->ip6_nxt = IPPROTO_TCP;
4216 		/* ip6_plen will be updated in ip6_output() */
4217 		th = (struct tcphdr *)(ip6 + 1);
4218 		th->th_dport = sc->sc_src.sin6.sin6_port;
4219 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4220 		break;
4221 #endif
4222 	default:
4223 		th = NULL;
4224 	}
4225 
4226 	th->th_seq = htonl(sc->sc_iss);
4227 	th->th_ack = htonl(sc->sc_irs + 1);
4228 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4229 	th->th_flags = TH_SYN|TH_ACK;
4230 #ifdef TCP_ECN
4231 	/* Set ECE for SYN-ACK if peer supports ECN. */
4232 	if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT))
4233 		th->th_flags |= TH_ECE;
4234 #endif
4235 	th->th_win = htons(sc->sc_win);
4236 	/* th_sum already 0 */
4237 	/* th_urp already 0 */
4238 
4239 	/* Tack on the TCP options. */
4240 	optp = (u_int8_t *)(th + 1);
4241 	*optp++ = TCPOPT_MAXSEG;
4242 	*optp++ = 4;
4243 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4244 	*optp++ = sc->sc_ourmaxseg & 0xff;
4245 
4246 #ifdef TCP_SACK
4247 	/* Include SACK_PERMIT_HDR option if peer has already done so. */
4248 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4249 		*((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR);
4250 		optp += 4;
4251 	}
4252 #endif
4253 
4254 	if (sc->sc_request_r_scale != 15) {
4255 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4256 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4257 		    sc->sc_request_r_scale);
4258 		optp += 4;
4259 	}
4260 
4261 	if (sc->sc_flags & SCF_TIMESTAMP) {
4262 		u_int32_t *lp = (u_int32_t *)(optp);
4263 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4264 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4265 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4266 		*lp   = htonl(sc->sc_timestamp);
4267 		optp += TCPOLEN_TSTAMP_APPA;
4268 	}
4269 
4270 #ifdef TCP_SIGNATURE
4271 	if (sc->sc_flags & SCF_SIGNATURE) {
4272 		union sockaddr_union src, dst;
4273 		struct tdb *tdb;
4274 
4275 		bzero(&src, sizeof(union sockaddr_union));
4276 		bzero(&dst, sizeof(union sockaddr_union));
4277 		src.sa.sa_len = sc->sc_src.sa.sa_len;
4278 		src.sa.sa_family = sc->sc_src.sa.sa_family;
4279 		dst.sa.sa_len = sc->sc_dst.sa.sa_len;
4280 		dst.sa.sa_family = sc->sc_dst.sa.sa_family;
4281 
4282 		switch (sc->sc_src.sa.sa_family) {
4283 		case 0:	/*default to PF_INET*/
4284 #ifdef INET
4285 		case AF_INET:
4286 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
4287 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
4288 			break;
4289 #endif /* INET */
4290 #ifdef INET6
4291 		case AF_INET6:
4292 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
4293 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
4294 			break;
4295 #endif /* INET6 */
4296 		}
4297 
4298 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
4299 		if (tdb == NULL) {
4300 			if (m)
4301 				m_freem(m);
4302 			return (EPERM);
4303 		}
4304 
4305 		/* Send signature option */
4306 		*(optp++) = TCPOPT_SIGNATURE;
4307 		*(optp++) = TCPOLEN_SIGNATURE;
4308 
4309 		if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th,
4310 		    hlen, 0, optp) < 0) {
4311 			if (m)
4312 				m_freem(m);
4313 			return (EINVAL);
4314 		}
4315 		optp += 16;
4316 
4317 		/* Pad options list to the next 32 bit boundary and
4318 		 * terminate it.
4319 		 */
4320 		*optp++ = TCPOPT_NOP;
4321 		*optp++ = TCPOPT_EOL;
4322 	}
4323 #endif /* TCP_SIGNATURE */
4324 
4325 	/* Compute the packet's checksum. */
4326 	switch (sc->sc_src.sa.sa_family) {
4327 	case AF_INET:
4328 		ip->ip_len = htons(tlen - hlen);
4329 		th->th_sum = 0;
4330 		th->th_sum = in_cksum(m, tlen);
4331 		break;
4332 #ifdef INET6
4333 	case AF_INET6:
4334 		ip6->ip6_plen = htons(tlen - hlen);
4335 		th->th_sum = 0;
4336 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4337 		break;
4338 #endif
4339 	}
4340 
4341 	/* use IPsec policy and ttl from listening socket, on SYN ACK */
4342 	inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL;
4343 
4344 	/*
4345 	 * Fill in some straggling IP bits.  Note the stack expects
4346 	 * ip_len to be in host order, for convenience.
4347 	 */
4348 	switch (sc->sc_src.sa.sa_family) {
4349 #ifdef INET
4350 	case AF_INET:
4351 		ip->ip_len = htons(tlen);
4352 		ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl;
4353 		/* XXX tos? */
4354 		break;
4355 #endif
4356 #ifdef INET6
4357 	case AF_INET6:
4358 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4359 		ip6->ip6_vfc |= IPV6_VERSION;
4360 		ip6->ip6_plen = htons(tlen - hlen);
4361 		/* ip6_hlim will be initialized afterwards */
4362 		/* leave flowlabel = 0, it is legal and require no state mgmt */
4363 		break;
4364 #endif
4365 	}
4366 
4367 	switch (sc->sc_src.sa.sa_family) {
4368 #ifdef INET
4369 	case AF_INET:
4370 		error = ip_output(m, sc->sc_ipopts, ro,
4371 		    (ip_mtudisc ? IP_MTUDISC : 0),
4372 		    (struct ip_moptions *)NULL, inp);
4373 		break;
4374 #endif
4375 #ifdef INET6
4376 	case AF_INET6:
4377 		ip6->ip6_hlim = in6_selecthlim(NULL,
4378 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4379 
4380 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4381 			(struct ip6_moptions *)0, NULL, NULL);
4382 		break;
4383 #endif
4384 	default:
4385 		error = EAFNOSUPPORT;
4386 		break;
4387 	}
4388 	return (error);
4389 }
4390