xref: /openbsd-src/sys/netinet/tcp_input.c (revision daf88648c0e349d5c02e1504293082072c981640)
1 /*	$OpenBSD: tcp_input.c,v 1.200 2006/12/11 21:31:58 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 
79 #include <dev/rndvar.h>
80 
81 #include <net/if.h>
82 #include <net/route.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/ip.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/ip_var.h>
89 #include <netinet/tcp.h>
90 #include <netinet/tcp_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/tcpip.h>
95 #include <netinet/tcp_debug.h>
96 
97 struct	tcpiphdr tcp_saveti;
98 
99 #ifdef INET6
100 #include <netinet6/in6_var.h>
101 #include <netinet6/nd6.h>
102 
103 struct  tcpipv6hdr tcp_saveti6;
104 
105 /* for the packet header length in the mbuf */
106 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
107 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
108 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
109 #endif /* INET6 */
110 
111 int	tcprexmtthresh = 3;
112 int	tcptv_keep_init = TCPTV_KEEP_INIT;
113 
114 extern u_long sb_max;
115 
116 int tcp_rst_ppslim = 100;		/* 100pps */
117 int tcp_rst_ppslim_count = 0;
118 struct timeval tcp_rst_ppslim_last;
119 
120 int tcp_ackdrop_ppslim = 100;		/* 100pps */
121 int tcp_ackdrop_ppslim_count = 0;
122 struct timeval tcp_ackdrop_ppslim_last;
123 
124 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
125 
126 /* for modulo comparisons of timestamps */
127 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
128 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
129 
130 /* for TCP SACK comparisons */
131 #define	SEQ_MIN(a,b)	(SEQ_LT(a,b) ? (a) : (b))
132 #define	SEQ_MAX(a,b)	(SEQ_GT(a,b) ? (a) : (b))
133 
134 /*
135  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
136  */
137 #ifdef INET6
138 #define ND6_HINT(tp) \
139 do { \
140 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
141 	    tp->t_inpcb->inp_route6.ro_rt) { \
142 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \
143 	} \
144 } while (0)
145 #else
146 #define ND6_HINT(tp)
147 #endif
148 
149 #ifdef TCP_ECN
150 /*
151  * ECN (Explicit Congestion Notification) support based on RFC3168
152  * implementation note:
153  *   snd_last is used to track a recovery phase.
154  *   when cwnd is reduced, snd_last is set to snd_max.
155  *   while snd_last > snd_una, the sender is in a recovery phase and
156  *   its cwnd should not be reduced again.
157  *   snd_last follows snd_una when not in a recovery phase.
158  */
159 #endif
160 
161 /*
162  * Macro to compute ACK transmission behavior.  Delay the ACK unless
163  * we have already delayed an ACK (must send an ACK every two segments).
164  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
165  * option is enabled.
166  */
167 #define	TCP_SETUP_ACK(tp, tiflags) \
168 do { \
169 	if ((tp)->t_flags & TF_DELACK || \
170 	    (tcp_ack_on_push && (tiflags) & TH_PUSH)) \
171 		tp->t_flags |= TF_ACKNOW; \
172 	else \
173 		TCP_SET_DELACK(tp); \
174 } while (0)
175 
176 /*
177  * Insert segment ti into reassembly queue of tcp with
178  * control block tp.  Return TH_FIN if reassembly now includes
179  * a segment with FIN.  The macro form does the common case inline
180  * (segment is the next to be received on an established connection,
181  * and the queue is empty), avoiding linkage into and removal
182  * from the queue and repetition of various conversions.
183  * Set DELACK for segments received in order, but ack immediately
184  * when segments are out of order (so fast retransmit can work).
185  */
186 
187 int
188 tcp_reass(tp, th, m, tlen)
189 	struct tcpcb *tp;
190 	struct tcphdr *th;
191 	struct mbuf *m;
192 	int *tlen;
193 {
194 	struct tcpqent *p, *q, *nq, *tiqe;
195 	struct socket *so = tp->t_inpcb->inp_socket;
196 	int flags;
197 
198 	/*
199 	 * Call with th==0 after become established to
200 	 * force pre-ESTABLISHED data up to user socket.
201 	 */
202 	if (th == 0)
203 		goto present;
204 
205 	/*
206 	 * Allocate a new queue entry, before we throw away any data.
207 	 * If we can't, just drop the packet.  XXX
208 	 */
209 	tiqe = pool_get(&tcpqe_pool, PR_NOWAIT);
210 	if (tiqe == NULL) {
211 		tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead);
212 		if (tiqe != NULL && th->th_seq == tp->rcv_nxt) {
213 			/* Reuse last entry since new segment fills a hole */
214 			m_freem(tiqe->tcpqe_m);
215 			TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q);
216 		}
217 		if (tiqe == NULL || th->th_seq != tp->rcv_nxt) {
218 			/* Flush segment queue for this connection */
219 			tcp_freeq(tp);
220 			tcpstat.tcps_rcvmemdrop++;
221 			m_freem(m);
222 			return (0);
223 		}
224 	}
225 
226 	/*
227 	 * Find a segment which begins after this one does.
228 	 */
229 	for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL;
230 	    p = q, q = TAILQ_NEXT(q, tcpqe_q))
231 		if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq))
232 			break;
233 
234 	/*
235 	 * If there is a preceding segment, it may provide some of
236 	 * our data already.  If so, drop the data from the incoming
237 	 * segment.  If it provides all of our data, drop us.
238 	 */
239 	if (p != NULL) {
240 		struct tcphdr *phdr = p->tcpqe_tcp;
241 		int i;
242 
243 		/* conversion to int (in i) handles seq wraparound */
244 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
245 		if (i > 0) {
246 		        if (i >= *tlen) {
247 				tcpstat.tcps_rcvduppack++;
248 				tcpstat.tcps_rcvdupbyte += *tlen;
249 				m_freem(m);
250 				pool_put(&tcpqe_pool, tiqe);
251 				return (0);
252 			}
253 			m_adj(m, i);
254 			*tlen -= i;
255 			th->th_seq += i;
256 		}
257 	}
258 	tcpstat.tcps_rcvoopack++;
259 	tcpstat.tcps_rcvoobyte += *tlen;
260 
261 	/*
262 	 * While we overlap succeeding segments trim them or,
263 	 * if they are completely covered, dequeue them.
264 	 */
265 	for (; q != NULL; q = nq) {
266 		struct tcphdr *qhdr = q->tcpqe_tcp;
267 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
268 
269 		if (i <= 0)
270 			break;
271 		if (i < qhdr->th_reseqlen) {
272 			qhdr->th_seq += i;
273 			qhdr->th_reseqlen -= i;
274 			m_adj(q->tcpqe_m, i);
275 			break;
276 		}
277 		nq = TAILQ_NEXT(q, tcpqe_q);
278 		m_freem(q->tcpqe_m);
279 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
280 		pool_put(&tcpqe_pool, q);
281 	}
282 
283 	/* Insert the new segment queue entry into place. */
284 	tiqe->tcpqe_m = m;
285 	th->th_reseqlen = *tlen;
286 	tiqe->tcpqe_tcp = th;
287 	if (p == NULL) {
288 		TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q);
289 	} else {
290 		TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q);
291 	}
292 
293 present:
294 	/*
295 	 * Present data to user, advancing rcv_nxt through
296 	 * completed sequence space.
297 	 */
298 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
299 		return (0);
300 	q = TAILQ_FIRST(&tp->t_segq);
301 	if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt)
302 		return (0);
303 	if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen)
304 		return (0);
305 	do {
306 		tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen;
307 		flags = q->tcpqe_tcp->th_flags & TH_FIN;
308 
309 		nq = TAILQ_NEXT(q, tcpqe_q);
310 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
311 		ND6_HINT(tp);
312 		if (so->so_state & SS_CANTRCVMORE)
313 			m_freem(q->tcpqe_m);
314 		else
315 			sbappendstream(&so->so_rcv, q->tcpqe_m);
316 		pool_put(&tcpqe_pool, q);
317 		q = nq;
318 	} while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt);
319 	sorwakeup(so);
320 	return (flags);
321 }
322 
323 #ifdef INET6
324 int
325 tcp6_input(mp, offp, proto)
326 	struct mbuf **mp;
327 	int *offp, proto;
328 {
329 	struct mbuf *m = *mp;
330 
331 #if defined(NFAITH) && 0 < NFAITH
332 	if (m->m_pkthdr.rcvif) {
333 		if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
334 			/* XXX send icmp6 host/port unreach? */
335 			m_freem(m);
336 			return IPPROTO_DONE;
337 		}
338 	}
339 #endif
340 
341 	/*
342 	 * draft-itojun-ipv6-tcp-to-anycast
343 	 * better place to put this in?
344 	 */
345 	if (m->m_flags & M_ANYCAST6) {
346 		if (m->m_len >= sizeof(struct ip6_hdr)) {
347 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
348 			icmp6_error(m, ICMP6_DST_UNREACH,
349 				ICMP6_DST_UNREACH_ADDR,
350 				(caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
351 		} else
352 			m_freem(m);
353 		return IPPROTO_DONE;
354 	}
355 
356 	tcp_input(m, *offp, proto);
357 	return IPPROTO_DONE;
358 }
359 #endif
360 
361 /*
362  * TCP input routine, follows pages 65-76 of the
363  * protocol specification dated September, 1981 very closely.
364  */
365 void
366 tcp_input(struct mbuf *m, ...)
367 {
368 	struct ip *ip;
369 	struct inpcb *inp;
370 	u_int8_t *optp = NULL;
371 	int optlen = 0;
372 	int tlen, off;
373 	struct tcpcb *tp = 0;
374 	int tiflags;
375 	struct socket *so = NULL;
376 	int todrop, acked, ourfinisacked, needoutput = 0;
377 	int hdroptlen = 0;
378 	short ostate = 0;
379 	int iss = 0;
380 	u_long tiwin;
381 	struct tcp_opt_info opti;
382 	int iphlen;
383 	va_list ap;
384 	struct tcphdr *th;
385 #ifdef INET6
386 	struct ip6_hdr *ip6 = NULL;
387 #endif /* INET6 */
388 #ifdef IPSEC
389 	struct m_tag *mtag;
390 	struct tdb_ident *tdbi;
391 	struct tdb *tdb;
392 	int error, s;
393 #endif /* IPSEC */
394 	int af;
395 #ifdef TCP_ECN
396 	u_char iptos;
397 #endif
398 
399 	va_start(ap, m);
400 	iphlen = va_arg(ap, int);
401 	va_end(ap);
402 
403 	tcpstat.tcps_rcvtotal++;
404 
405 	opti.ts_present = 0;
406 	opti.maxseg = 0;
407 
408 	/*
409 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
410 	 * See below for AF specific multicast.
411 	 */
412 	if (m->m_flags & (M_BCAST|M_MCAST))
413 		goto drop;
414 
415 	/*
416 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
417 	 * TCP/IPv4.
418 	 */
419 	switch (mtod(m, struct ip *)->ip_v) {
420 #ifdef INET6
421 	case 6:
422 		af = AF_INET6;
423 		break;
424 #endif
425 	case 4:
426 		af = AF_INET;
427 		break;
428 	default:
429 		m_freem(m);
430 		return;	/*EAFNOSUPPORT*/
431 	}
432 
433 	/*
434 	 * Get IP and TCP header together in first mbuf.
435 	 * Note: IP leaves IP header in first mbuf.
436 	 */
437 	switch (af) {
438 	case AF_INET:
439 #ifdef DIAGNOSTIC
440 		if (iphlen < sizeof(struct ip)) {
441 			m_freem(m);
442 			return;
443 		}
444 #endif /* DIAGNOSTIC */
445 		break;
446 #ifdef INET6
447 	case AF_INET6:
448 #ifdef DIAGNOSTIC
449 		if (iphlen < sizeof(struct ip6_hdr)) {
450 			m_freem(m);
451 			return;
452 		}
453 #endif /* DIAGNOSTIC */
454 		break;
455 #endif
456 	default:
457 		m_freem(m);
458 		return;
459 	}
460 
461 	IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th));
462 	if (!th) {
463 		tcpstat.tcps_rcvshort++;
464 		return;
465 	}
466 
467 	tlen = m->m_pkthdr.len - iphlen;
468 	ip = NULL;
469 #ifdef INET6
470 	ip6 = NULL;
471 #endif
472 	switch (af) {
473 	case AF_INET:
474 		ip = mtod(m, struct ip *);
475 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
476 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
477 			goto drop;
478 #ifdef TCP_ECN
479 		/* save ip_tos before clearing it for checksum */
480 		iptos = ip->ip_tos;
481 #endif
482 		/*
483 		 * Checksum extended TCP header and data.
484 		 */
485 		if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) {
486 			if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) {
487 				tcpstat.tcps_inhwcsum++;
488 				tcpstat.tcps_rcvbadsum++;
489 				goto drop;
490 			}
491 			if (in4_cksum(m, IPPROTO_TCP, iphlen, tlen) != 0) {
492 				tcpstat.tcps_rcvbadsum++;
493 				goto drop;
494 			}
495 		} else {
496 			m->m_pkthdr.csum_flags &= ~M_TCP_CSUM_IN_OK;
497 			tcpstat.tcps_inhwcsum++;
498 		}
499 		break;
500 #ifdef INET6
501 	case AF_INET6:
502 		ip6 = mtod(m, struct ip6_hdr *);
503 #ifdef TCP_ECN
504 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
505 #endif
506 
507 		/* Be proactive about malicious use of IPv4 mapped address */
508 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
509 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
510 			/* XXX stat */
511 			goto drop;
512 		}
513 
514 		/*
515 		 * Be proactive about unspecified IPv6 address in source.
516 		 * As we use all-zero to indicate unbounded/unconnected pcb,
517 		 * unspecified IPv6 address can be used to confuse us.
518 		 *
519 		 * Note that packets with unspecified IPv6 destination is
520 		 * already dropped in ip6_input.
521 		 */
522 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
523 			/* XXX stat */
524 			goto drop;
525 		}
526 
527 		/* Discard packets to multicast */
528 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
529 			/* XXX stat */
530 			goto drop;
531 		}
532 
533 		/*
534 		 * Checksum extended TCP header and data.
535 		 */
536 		if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) {
537 			tcpstat.tcps_rcvbadsum++;
538 			goto drop;
539 		}
540 		break;
541 #endif
542 	}
543 
544 	/*
545 	 * Check that TCP offset makes sense,
546 	 * pull out TCP options and adjust length.		XXX
547 	 */
548 	off = th->th_off << 2;
549 	if (off < sizeof(struct tcphdr) || off > tlen) {
550 		tcpstat.tcps_rcvbadoff++;
551 		goto drop;
552 	}
553 	tlen -= off;
554 	if (off > sizeof(struct tcphdr)) {
555 		IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off);
556 		if (!th) {
557 			tcpstat.tcps_rcvshort++;
558 			return;
559 		}
560 		optlen = off - sizeof(struct tcphdr);
561 		optp = (u_int8_t *)(th + 1);
562 		/*
563 		 * Do quick retrieval of timestamp options ("options
564 		 * prediction?").  If timestamp is the only option and it's
565 		 * formatted as recommended in RFC 1323 appendix A, we
566 		 * quickly get the values now and not bother calling
567 		 * tcp_dooptions(), etc.
568 		 */
569 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
570 		     (optlen > TCPOLEN_TSTAMP_APPA &&
571 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
572 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
573 		     (th->th_flags & TH_SYN) == 0) {
574 			opti.ts_present = 1;
575 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
576 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
577 			optp = NULL;	/* we've parsed the options */
578 		}
579 	}
580 	tiflags = th->th_flags;
581 
582 	/*
583 	 * Convert TCP protocol specific fields to host format.
584 	 */
585 	NTOHL(th->th_seq);
586 	NTOHL(th->th_ack);
587 	NTOHS(th->th_win);
588 	NTOHS(th->th_urp);
589 
590 	/*
591 	 * Locate pcb for segment.
592 	 */
593 findpcb:
594 	switch (af) {
595 #ifdef INET6
596 	case AF_INET6:
597 		inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src, th->th_sport,
598 		    &ip6->ip6_dst, th->th_dport);
599 		break;
600 #endif
601 	case AF_INET:
602 		inp = in_pcbhashlookup(&tcbtable, ip->ip_src, th->th_sport,
603 		    ip->ip_dst, th->th_dport);
604 		break;
605 	}
606 	if (inp == 0) {
607 		int	inpl_flags = 0;
608 #if NPF > 0
609 		struct pf_mtag *t;
610 
611 		if ((t = pf_find_mtag(m)) != NULL &&
612 		    t->flags & PF_TAG_TRANSLATE_LOCALHOST)
613 			inpl_flags = INPLOOKUP_WILDCARD;
614 #endif
615 		++tcpstat.tcps_pcbhashmiss;
616 		switch (af) {
617 #ifdef INET6
618 		case AF_INET6:
619 			inp = in6_pcblookup_listen(&tcbtable,
620 			    &ip6->ip6_dst, th->th_dport, inpl_flags);
621 			break;
622 #endif /* INET6 */
623 		case AF_INET:
624 			inp = in_pcblookup_listen(&tcbtable,
625 			    ip->ip_dst, th->th_dport, inpl_flags);
626 			break;
627 		}
628 		/*
629 		 * If the state is CLOSED (i.e., TCB does not exist) then
630 		 * all data in the incoming segment is discarded.
631 		 * If the TCB exists but is in CLOSED state, it is embryonic,
632 		 * but should either do a listen or a connect soon.
633 		 */
634 		if (inp == 0) {
635 			++tcpstat.tcps_noport;
636 			goto dropwithreset_ratelim;
637 		}
638 	}
639 
640 	/* Check the minimum TTL for socket. */
641 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
642 		goto drop;
643 
644 	tp = intotcpcb(inp);
645 	if (tp == 0)
646 		goto dropwithreset_ratelim;
647 	if (tp->t_state == TCPS_CLOSED)
648 		goto drop;
649 
650 	/* Unscale the window into a 32-bit value. */
651 	if ((tiflags & TH_SYN) == 0)
652 		tiwin = th->th_win << tp->snd_scale;
653 	else
654 		tiwin = th->th_win;
655 
656 	so = inp->inp_socket;
657 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
658 		union syn_cache_sa src;
659 		union syn_cache_sa dst;
660 
661 		bzero(&src, sizeof(src));
662 		bzero(&dst, sizeof(dst));
663 		switch (af) {
664 #ifdef INET
665 		case AF_INET:
666 			src.sin.sin_len = sizeof(struct sockaddr_in);
667 			src.sin.sin_family = AF_INET;
668 			src.sin.sin_addr = ip->ip_src;
669 			src.sin.sin_port = th->th_sport;
670 
671 			dst.sin.sin_len = sizeof(struct sockaddr_in);
672 			dst.sin.sin_family = AF_INET;
673 			dst.sin.sin_addr = ip->ip_dst;
674 			dst.sin.sin_port = th->th_dport;
675 			break;
676 #endif
677 #ifdef INET6
678 		case AF_INET6:
679 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
680 			src.sin6.sin6_family = AF_INET6;
681 			src.sin6.sin6_addr = ip6->ip6_src;
682 			src.sin6.sin6_port = th->th_sport;
683 
684 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
685 			dst.sin6.sin6_family = AF_INET6;
686 			dst.sin6.sin6_addr = ip6->ip6_dst;
687 			dst.sin6.sin6_port = th->th_dport;
688 			break;
689 #endif /* INET6 */
690 		default:
691 			goto badsyn;	/*sanity*/
692 		}
693 
694 		if (so->so_options & SO_DEBUG) {
695 			ostate = tp->t_state;
696 			switch (af) {
697 #ifdef INET6
698 			case AF_INET6:
699 				bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6));
700 				bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th));
701 				break;
702 #endif
703 			case AF_INET:
704 				bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip));
705 				bcopy(th, &tcp_saveti.ti_t, sizeof(*th));
706 				break;
707 			}
708 		}
709 		if (so->so_options & SO_ACCEPTCONN) {
710 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
711 				if (tiflags & TH_RST) {
712 					syn_cache_reset(&src.sa, &dst.sa, th);
713 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
714 				    (TH_ACK|TH_SYN)) {
715 					/*
716 					 * Received a SYN,ACK.  This should
717 					 * never happen while we are in
718 					 * LISTEN.  Send an RST.
719 					 */
720 					goto badsyn;
721 				} else if (tiflags & TH_ACK) {
722 					so = syn_cache_get(&src.sa, &dst.sa,
723 						th, iphlen, tlen, so, m);
724 					if (so == NULL) {
725 						/*
726 						 * We don't have a SYN for
727 						 * this ACK; send an RST.
728 						 */
729 						goto badsyn;
730 					} else if (so ==
731 					    (struct socket *)(-1)) {
732 						/*
733 						 * We were unable to create
734 						 * the connection.  If the
735 						 * 3-way handshake was
736 						 * completed, and RST has
737 						 * been sent to the peer.
738 						 * Since the mbuf might be
739 						 * in use for the reply,
740 						 * do not free it.
741 						 */
742 						m = NULL;
743 					} else {
744 						/*
745 						 * We have created a
746 						 * full-blown connection.
747 						 */
748 						tp = NULL;
749 						inp = (struct inpcb *)so->so_pcb;
750 						tp = intotcpcb(inp);
751 						if (tp == NULL)
752 							goto badsyn;	/*XXX*/
753 
754 						/*
755 						 * Compute proper scaling
756 						 * value from buffer space
757 						 */
758 						tcp_rscale(tp, so->so_rcv.sb_hiwat);
759 						goto after_listen;
760 					}
761 				} else {
762 					/*
763 					 * None of RST, SYN or ACK was set.
764 					 * This is an invalid packet for a
765 					 * TCB in LISTEN state.  Send a RST.
766 					 */
767 					goto badsyn;
768 				}
769 			} else {
770 				/*
771 				 * Received a SYN.
772 				 */
773 #ifdef INET6
774 				/*
775 				 * If deprecated address is forbidden, we do
776 				 * not accept SYN to deprecated interface
777 				 * address to prevent any new inbound
778 				 * connection from getting established.
779 				 * When we do not accept SYN, we send a TCP
780 				 * RST, with deprecated source address (instead
781 				 * of dropping it).  We compromise it as it is
782 				 * much better for peer to send a RST, and
783 				 * RST will be the final packet for the
784 				 * exchange.
785 				 *
786 				 * If we do not forbid deprecated addresses, we
787 				 * accept the SYN packet.  RFC2462 does not
788 				 * suggest dropping SYN in this case.
789 				 * If we decipher RFC2462 5.5.4, it says like
790 				 * this:
791 				 * 1. use of deprecated addr with existing
792 				 *    communication is okay - "SHOULD continue
793 				 *    to be used"
794 				 * 2. use of it with new communication:
795 				 *   (2a) "SHOULD NOT be used if alternate
796 				 *        address with sufficient scope is
797 				 *        available"
798 				 *   (2b) nothing mentioned otherwise.
799 				 * Here we fall into (2b) case as we have no
800 				 * choice in our source address selection - we
801 				 * must obey the peer.
802 				 *
803 				 * The wording in RFC2462 is confusing, and
804 				 * there are multiple description text for
805 				 * deprecated address handling - worse, they
806 				 * are not exactly the same.  I believe 5.5.4
807 				 * is the best one, so we follow 5.5.4.
808 				 */
809 				if (ip6 && !ip6_use_deprecated) {
810 					struct in6_ifaddr *ia6;
811 
812 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
813 					    &ip6->ip6_dst)) &&
814 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
815 						tp = NULL;
816 						goto dropwithreset;
817 					}
818 				}
819 #endif
820 
821 				/*
822 				 * LISTEN socket received a SYN
823 				 * from itself?  This can't possibly
824 				 * be valid; drop the packet.
825 				 */
826 				if (th->th_dport == th->th_sport) {
827 					switch (af) {
828 #ifdef INET6
829 					case AF_INET6:
830 						if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src,
831 						    &ip6->ip6_dst)) {
832 							tcpstat.tcps_badsyn++;
833 							goto drop;
834 						}
835 						break;
836 #endif /* INET6 */
837 					case AF_INET:
838 						if (ip->ip_dst.s_addr == ip->ip_src.s_addr) {
839 							tcpstat.tcps_badsyn++;
840 							goto drop;
841 						}
842 						break;
843 					}
844 				}
845 
846 				/*
847 				 * SYN looks ok; create compressed TCP
848 				 * state for it.
849 				 */
850 				if (so->so_qlen <= so->so_qlimit &&
851 				    syn_cache_add(&src.sa, &dst.sa, th, iphlen,
852 						so, m, optp, optlen, &opti))
853 					m = NULL;
854 			}
855 			goto drop;
856 		}
857 	}
858 
859 after_listen:
860 #ifdef DIAGNOSTIC
861 	/*
862 	 * Should not happen now that all embryonic connections
863 	 * are handled with compressed state.
864 	 */
865 	if (tp->t_state == TCPS_LISTEN)
866 		panic("tcp_input: TCPS_LISTEN");
867 #endif
868 
869 #ifdef IPSEC
870 	/* Find most recent IPsec tag */
871 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
872         s = splnet();
873 	if (mtag != NULL) {
874 		tdbi = (struct tdb_ident *)(mtag + 1);
875 	        tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto);
876 	} else
877 		tdb = NULL;
878 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
879 	    tdb, inp);
880 	if (error) {
881 		splx(s);
882 		goto drop;
883 	}
884 
885 	/* Latch SA */
886 	if (inp->inp_tdb_in != tdb) {
887 		if (tdb) {
888 		        tdb_add_inp(tdb, inp, 1);
889 			if (inp->inp_ipo == NULL) {
890 				inp->inp_ipo = ipsec_add_policy(inp, af,
891 				    IPSP_DIRECTION_OUT);
892 				if (inp->inp_ipo == NULL) {
893 					splx(s);
894 					goto drop;
895 				}
896 			}
897 			if (inp->inp_ipo->ipo_dstid == NULL &&
898 			    tdb->tdb_srcid != NULL) {
899 				inp->inp_ipo->ipo_dstid = tdb->tdb_srcid;
900 				tdb->tdb_srcid->ref_count++;
901 			}
902 			if (inp->inp_ipsec_remotecred == NULL &&
903 			    tdb->tdb_remote_cred != NULL) {
904 				inp->inp_ipsec_remotecred =
905 				    tdb->tdb_remote_cred;
906 				tdb->tdb_remote_cred->ref_count++;
907 			}
908 			if (inp->inp_ipsec_remoteauth == NULL &&
909 			    tdb->tdb_remote_auth != NULL) {
910 				inp->inp_ipsec_remoteauth =
911 				    tdb->tdb_remote_auth;
912 				tdb->tdb_remote_auth->ref_count++;
913 			}
914 		} else { /* Just reset */
915 		        TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp,
916 				     inp_tdb_in_next);
917 			inp->inp_tdb_in = NULL;
918 		}
919 	}
920         splx(s);
921 #endif /* IPSEC */
922 
923 	/*
924 	 * Segment received on connection.
925 	 * Reset idle time and keep-alive timer.
926 	 */
927 	tp->t_rcvtime = tcp_now;
928 	if (TCPS_HAVEESTABLISHED(tp->t_state))
929 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
930 
931 #ifdef TCP_SACK
932 	if (tp->sack_enable)
933 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
934 #endif /* TCP_SACK */
935 
936 	/*
937 	 * Process options.
938 	 */
939 #ifdef TCP_SIGNATURE
940 	if (optp || (tp->t_flags & TF_SIGNATURE))
941 #else
942 	if (optp)
943 #endif
944 		if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti))
945 			goto drop;
946 
947 	if (opti.ts_present && opti.ts_ecr) {
948 		int rtt_test;
949 
950 		/* subtract out the tcp timestamp modulator */
951 		opti.ts_ecr -= tp->ts_modulate;
952 
953 		/* make sure ts_ecr is sensible */
954 		rtt_test = tcp_now - opti.ts_ecr;
955 		if (rtt_test < 0 || rtt_test > TCP_RTT_MAX)
956 			opti.ts_ecr = 0;
957 	}
958 
959 #ifdef TCP_ECN
960 	/* if congestion experienced, set ECE bit in subsequent packets. */
961 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
962 		tp->t_flags |= TF_RCVD_CE;
963 		tcpstat.tcps_ecn_rcvce++;
964 	}
965 #endif
966 	/*
967 	 * Header prediction: check for the two common cases
968 	 * of a uni-directional data xfer.  If the packet has
969 	 * no control flags, is in-sequence, the window didn't
970 	 * change and we're not retransmitting, it's a
971 	 * candidate.  If the length is zero and the ack moved
972 	 * forward, we're the sender side of the xfer.  Just
973 	 * free the data acked & wake any higher level process
974 	 * that was blocked waiting for space.  If the length
975 	 * is non-zero and the ack didn't move, we're the
976 	 * receiver side.  If we're getting packets in-order
977 	 * (the reassembly queue is empty), add the data to
978 	 * the socket buffer and note that we need a delayed ack.
979 	 */
980 	if (tp->t_state == TCPS_ESTABLISHED &&
981 #ifdef TCP_ECN
982 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
983 #else
984 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
985 #endif
986 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
987 	    th->th_seq == tp->rcv_nxt &&
988 	    tiwin && tiwin == tp->snd_wnd &&
989 	    tp->snd_nxt == tp->snd_max) {
990 
991 		/*
992 		 * If last ACK falls within this segment's sequence numbers,
993 		 *  record the timestamp.
994 		 * Fix from Braden, see Stevens p. 870
995 		 */
996 		if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
997 			tp->ts_recent_age = tcp_now;
998 			tp->ts_recent = opti.ts_val;
999 		}
1000 
1001 		if (tlen == 0) {
1002 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1003 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1004 			    tp->snd_cwnd >= tp->snd_wnd &&
1005 			    tp->t_dupacks == 0) {
1006 				/*
1007 				 * this is a pure ack for outstanding data.
1008 				 */
1009 				++tcpstat.tcps_predack;
1010 				if (opti.ts_present && opti.ts_ecr)
1011 					tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1012 				else if (tp->t_rtttime &&
1013 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1014 					tcp_xmit_timer(tp,
1015 					    tcp_now - tp->t_rtttime);
1016 				acked = th->th_ack - tp->snd_una;
1017 				tcpstat.tcps_rcvackpack++;
1018 				tcpstat.tcps_rcvackbyte += acked;
1019 				ND6_HINT(tp);
1020 				sbdrop(&so->so_snd, acked);
1021 
1022 				/*
1023 				 * If we had a pending ICMP message that
1024 				 * referres to data that have just been
1025 				 * acknowledged, disregard the recorded ICMP
1026 				 * message.
1027 				 */
1028 				if ((tp->t_flags & TF_PMTUD_PEND) &&
1029 				    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1030 					tp->t_flags &= ~TF_PMTUD_PEND;
1031 
1032 				/*
1033 				 * Keep track of the largest chunk of data
1034 				 * acknowledged since last PMTU update
1035 				 */
1036 				if (tp->t_pmtud_mss_acked < acked)
1037 					tp->t_pmtud_mss_acked = acked;
1038 
1039 				tp->snd_una = th->th_ack;
1040 #if defined(TCP_SACK) || defined(TCP_ECN)
1041 				/*
1042 				 * We want snd_last to track snd_una so
1043 				 * as to avoid sequence wraparound problems
1044 				 * for very large transfers.
1045 				 */
1046 #ifdef TCP_ECN
1047 				if (SEQ_GT(tp->snd_una, tp->snd_last))
1048 #endif
1049 				tp->snd_last = tp->snd_una;
1050 #endif /* TCP_SACK */
1051 #if defined(TCP_SACK) && defined(TCP_FACK)
1052 				tp->snd_fack = tp->snd_una;
1053 				tp->retran_data = 0;
1054 #endif /* TCP_FACK */
1055 				m_freem(m);
1056 
1057 				/*
1058 				 * If all outstanding data are acked, stop
1059 				 * retransmit timer, otherwise restart timer
1060 				 * using current (possibly backed-off) value.
1061 				 * If process is waiting for space,
1062 				 * wakeup/selwakeup/signal.  If data
1063 				 * are ready to send, let tcp_output
1064 				 * decide between more output or persist.
1065 				 */
1066 				if (tp->snd_una == tp->snd_max)
1067 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1068 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1069 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1070 
1071 				if (sb_notify(&so->so_snd))
1072 					sowwakeup(so);
1073 				if (so->so_snd.sb_cc)
1074 					(void) tcp_output(tp);
1075 				return;
1076 			}
1077 		} else if (th->th_ack == tp->snd_una &&
1078 		    TAILQ_EMPTY(&tp->t_segq) &&
1079 		    tlen <= sbspace(&so->so_rcv)) {
1080 			/*
1081 			 * This is a pure, in-sequence data packet
1082 			 * with nothing on the reassembly queue and
1083 			 * we have enough buffer space to take it.
1084 			 */
1085 #ifdef TCP_SACK
1086 			/* Clean receiver SACK report if present */
1087 			if (tp->sack_enable && tp->rcv_numsacks)
1088 				tcp_clean_sackreport(tp);
1089 #endif /* TCP_SACK */
1090 			++tcpstat.tcps_preddat;
1091 			tp->rcv_nxt += tlen;
1092 			tcpstat.tcps_rcvpack++;
1093 			tcpstat.tcps_rcvbyte += tlen;
1094 			ND6_HINT(tp);
1095 			/*
1096 			 * Drop TCP, IP headers and TCP options then add data
1097 			 * to socket buffer.
1098 			 */
1099 			if (so->so_state & SS_CANTRCVMORE)
1100 				m_freem(m);
1101 			else {
1102 				m_adj(m, iphlen + off);
1103 				sbappendstream(&so->so_rcv, m);
1104 			}
1105 			sorwakeup(so);
1106 			TCP_SETUP_ACK(tp, tiflags);
1107 			if (tp->t_flags & TF_ACKNOW)
1108 				(void) tcp_output(tp);
1109 			return;
1110 		}
1111 	}
1112 
1113 	/*
1114 	 * Compute mbuf offset to TCP data segment.
1115 	 */
1116 	hdroptlen = iphlen + off;
1117 
1118 	/*
1119 	 * Calculate amount of space in receive window,
1120 	 * and then do TCP input processing.
1121 	 * Receive window is amount of space in rcv queue,
1122 	 * but not less than advertised window.
1123 	 */
1124 	{ int win;
1125 
1126 	win = sbspace(&so->so_rcv);
1127 	if (win < 0)
1128 		win = 0;
1129 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1130 	}
1131 
1132 	switch (tp->t_state) {
1133 
1134 	/*
1135 	 * If the state is SYN_RECEIVED:
1136 	 * 	if seg contains SYN/ACK, send an RST.
1137 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1138 	 */
1139 
1140 	case TCPS_SYN_RECEIVED:
1141 		if (tiflags & TH_ACK) {
1142 			if (tiflags & TH_SYN) {
1143 				tcpstat.tcps_badsyn++;
1144 				goto dropwithreset;
1145 			}
1146 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1147 			    SEQ_GT(th->th_ack, tp->snd_max))
1148 				goto dropwithreset;
1149 		}
1150 		break;
1151 
1152 	/*
1153 	 * If the state is SYN_SENT:
1154 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1155 	 *	if seg contains a RST, then drop the connection.
1156 	 *	if seg does not contain SYN, then drop it.
1157 	 * Otherwise this is an acceptable SYN segment
1158 	 *	initialize tp->rcv_nxt and tp->irs
1159 	 *	if seg contains ack then advance tp->snd_una
1160 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1161 	 *	arrange for segment to be acked (eventually)
1162 	 *	continue processing rest of data/controls, beginning with URG
1163 	 */
1164 	case TCPS_SYN_SENT:
1165 		if ((tiflags & TH_ACK) &&
1166 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1167 		     SEQ_GT(th->th_ack, tp->snd_max)))
1168 			goto dropwithreset;
1169 		if (tiflags & TH_RST) {
1170 #ifdef TCP_ECN
1171 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1172 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1173 				goto drop;
1174 #endif
1175 			if (tiflags & TH_ACK)
1176 				tp = tcp_drop(tp, ECONNREFUSED);
1177 			goto drop;
1178 		}
1179 		if ((tiflags & TH_SYN) == 0)
1180 			goto drop;
1181 		if (tiflags & TH_ACK) {
1182 			tp->snd_una = th->th_ack;
1183 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1184 				tp->snd_nxt = tp->snd_una;
1185 		}
1186 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1187 		tp->irs = th->th_seq;
1188 		tcp_mss(tp, opti.maxseg);
1189 		/* Reset initial window to 1 segment for retransmit */
1190 		if (tp->t_rxtshift > 0)
1191 			tp->snd_cwnd = tp->t_maxseg;
1192 		tcp_rcvseqinit(tp);
1193 		tp->t_flags |= TF_ACKNOW;
1194 #ifdef TCP_SACK
1195                 /*
1196                  * If we've sent a SACK_PERMITTED option, and the peer
1197                  * also replied with one, then TF_SACK_PERMIT should have
1198                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1199                  */
1200 		if (tp->sack_enable)
1201 			tp->sack_enable = tp->t_flags & TF_SACK_PERMIT;
1202 #endif
1203 #ifdef TCP_ECN
1204 		/*
1205 		 * if ECE is set but CWR is not set for SYN-ACK, or
1206 		 * both ECE and CWR are set for simultaneous open,
1207 		 * peer is ECN capable.
1208 		 */
1209 		if (tcp_do_ecn) {
1210 			if ((tiflags & (TH_ACK|TH_ECE|TH_CWR))
1211 			    == (TH_ACK|TH_ECE) ||
1212 			    (tiflags & (TH_ACK|TH_ECE|TH_CWR))
1213 			    == (TH_ECE|TH_CWR)) {
1214 				tp->t_flags |= TF_ECN_PERMIT;
1215 				tiflags &= ~(TH_ECE|TH_CWR);
1216 				tcpstat.tcps_ecn_accepts++;
1217 			}
1218 		}
1219 #endif
1220 
1221 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1222 			tcpstat.tcps_connects++;
1223 			soisconnected(so);
1224 			tp->t_state = TCPS_ESTABLISHED;
1225 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1226 			/* Do window scaling on this connection? */
1227 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1228 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1229 				tp->snd_scale = tp->requested_s_scale;
1230 				tp->rcv_scale = tp->request_r_scale;
1231 			}
1232 			tcp_reass_lock(tp);
1233 			(void) tcp_reass(tp, (struct tcphdr *)0,
1234 				(struct mbuf *)0, &tlen);
1235 			tcp_reass_unlock(tp);
1236 			/*
1237 			 * if we didn't have to retransmit the SYN,
1238 			 * use its rtt as our initial srtt & rtt var.
1239 			 */
1240 			if (tp->t_rtttime)
1241 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1242 			/*
1243 			 * Since new data was acked (the SYN), open the
1244 			 * congestion window by one MSS.  We do this
1245 			 * here, because we won't go through the normal
1246 			 * ACK processing below.  And since this is the
1247 			 * start of the connection, we know we are in
1248 			 * the exponential phase of slow-start.
1249 			 */
1250 			tp->snd_cwnd += tp->t_maxseg;
1251 		} else
1252 			tp->t_state = TCPS_SYN_RECEIVED;
1253 
1254 #if 0
1255 trimthenstep6:
1256 #endif
1257 		/*
1258 		 * Advance th->th_seq to correspond to first data byte.
1259 		 * If data, trim to stay within window,
1260 		 * dropping FIN if necessary.
1261 		 */
1262 		th->th_seq++;
1263 		if (tlen > tp->rcv_wnd) {
1264 			todrop = tlen - tp->rcv_wnd;
1265 			m_adj(m, -todrop);
1266 			tlen = tp->rcv_wnd;
1267 			tiflags &= ~TH_FIN;
1268 			tcpstat.tcps_rcvpackafterwin++;
1269 			tcpstat.tcps_rcvbyteafterwin += todrop;
1270 		}
1271 		tp->snd_wl1 = th->th_seq - 1;
1272 		tp->rcv_up = th->th_seq;
1273 		goto step6;
1274 	}
1275 
1276 	/*
1277 	 * States other than LISTEN or SYN_SENT.
1278 	 * First check timestamp, if present.
1279 	 * Then check that at least some bytes of segment are within
1280 	 * receive window.  If segment begins before rcv_nxt,
1281 	 * drop leading data (and SYN); if nothing left, just ack.
1282 	 *
1283 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1284 	 * and it's less than opti.ts_recent, drop it.
1285 	 */
1286 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1287 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1288 
1289 		/* Check to see if ts_recent is over 24 days old.  */
1290 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1291 			/*
1292 			 * Invalidate ts_recent.  If this segment updates
1293 			 * ts_recent, the age will be reset later and ts_recent
1294 			 * will get a valid value.  If it does not, setting
1295 			 * ts_recent to zero will at least satisfy the
1296 			 * requirement that zero be placed in the timestamp
1297 			 * echo reply when ts_recent isn't valid.  The
1298 			 * age isn't reset until we get a valid ts_recent
1299 			 * because we don't want out-of-order segments to be
1300 			 * dropped when ts_recent is old.
1301 			 */
1302 			tp->ts_recent = 0;
1303 		} else {
1304 			tcpstat.tcps_rcvduppack++;
1305 			tcpstat.tcps_rcvdupbyte += tlen;
1306 			tcpstat.tcps_pawsdrop++;
1307 			goto dropafterack;
1308 		}
1309 	}
1310 
1311 	todrop = tp->rcv_nxt - th->th_seq;
1312 	if (todrop > 0) {
1313 		if (tiflags & TH_SYN) {
1314 			tiflags &= ~TH_SYN;
1315 			th->th_seq++;
1316 			if (th->th_urp > 1)
1317 				th->th_urp--;
1318 			else
1319 				tiflags &= ~TH_URG;
1320 			todrop--;
1321 		}
1322 		if (todrop > tlen ||
1323 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1324 			/*
1325 			 * Any valid FIN must be to the left of the
1326 			 * window.  At this point, FIN must be a
1327 			 * duplicate or out-of-sequence, so drop it.
1328 			 */
1329 			tiflags &= ~TH_FIN;
1330 			/*
1331 			 * Send ACK to resynchronize, and drop any data,
1332 			 * but keep on processing for RST or ACK.
1333 			 */
1334 			tp->t_flags |= TF_ACKNOW;
1335 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1336 			tcpstat.tcps_rcvduppack++;
1337 		} else {
1338 			tcpstat.tcps_rcvpartduppack++;
1339 			tcpstat.tcps_rcvpartdupbyte += todrop;
1340 		}
1341 		hdroptlen += todrop;	/* drop from head afterwards */
1342 		th->th_seq += todrop;
1343 		tlen -= todrop;
1344 		if (th->th_urp > todrop)
1345 			th->th_urp -= todrop;
1346 		else {
1347 			tiflags &= ~TH_URG;
1348 			th->th_urp = 0;
1349 		}
1350 	}
1351 
1352 	/*
1353 	 * If new data are received on a connection after the
1354 	 * user processes are gone, then RST the other end.
1355 	 */
1356 	if ((so->so_state & SS_NOFDREF) &&
1357 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1358 		tp = tcp_close(tp);
1359 		tcpstat.tcps_rcvafterclose++;
1360 		goto dropwithreset;
1361 	}
1362 
1363 	/*
1364 	 * If segment ends after window, drop trailing data
1365 	 * (and PUSH and FIN); if nothing left, just ACK.
1366 	 */
1367 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1368 	if (todrop > 0) {
1369 		tcpstat.tcps_rcvpackafterwin++;
1370 		if (todrop >= tlen) {
1371 			tcpstat.tcps_rcvbyteafterwin += tlen;
1372 			/*
1373 			 * If a new connection request is received
1374 			 * while in TIME_WAIT, drop the old connection
1375 			 * and start over if the sequence numbers
1376 			 * are above the previous ones.
1377 			 */
1378 			if (tiflags & TH_SYN &&
1379 			    tp->t_state == TCPS_TIME_WAIT &&
1380 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1381 				iss = tp->snd_nxt + TCP_ISSINCR;
1382 				tp = tcp_close(tp);
1383 				goto findpcb;
1384 			}
1385 			/*
1386 			 * If window is closed can only take segments at
1387 			 * window edge, and have to drop data and PUSH from
1388 			 * incoming segments.  Continue processing, but
1389 			 * remember to ack.  Otherwise, drop segment
1390 			 * and ack.
1391 			 */
1392 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1393 				tp->t_flags |= TF_ACKNOW;
1394 				tcpstat.tcps_rcvwinprobe++;
1395 			} else
1396 				goto dropafterack;
1397 		} else
1398 			tcpstat.tcps_rcvbyteafterwin += todrop;
1399 		m_adj(m, -todrop);
1400 		tlen -= todrop;
1401 		tiflags &= ~(TH_PUSH|TH_FIN);
1402 	}
1403 
1404 	/*
1405 	 * If last ACK falls within this segment's sequence numbers,
1406 	 * record its timestamp if it's more recent.
1407 	 * Cf fix from Braden, see Stevens p. 870
1408 	 */
1409 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1410 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1411 		if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1412 		    ((tiflags & (TH_SYN|TH_FIN)) != 0)))
1413 			tp->ts_recent = opti.ts_val;
1414 		else
1415 			tp->ts_recent = 0;
1416 		tp->ts_recent_age = tcp_now;
1417 	}
1418 
1419 	/*
1420 	 * If the RST bit is set examine the state:
1421 	 *    SYN_RECEIVED STATE:
1422 	 *	If passive open, return to LISTEN state.
1423 	 *	If active open, inform user that connection was refused.
1424 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1425 	 *	Inform user that connection was reset, and close tcb.
1426 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1427 	 *	Close the tcb.
1428 	 */
1429 	if (tiflags & TH_RST) {
1430 		if (th->th_seq != tp->last_ack_sent &&
1431 		    th->th_seq != tp->rcv_nxt &&
1432 		    th->th_seq != (tp->rcv_nxt + 1))
1433 			goto drop;
1434 
1435 		switch (tp->t_state) {
1436 		case TCPS_SYN_RECEIVED:
1437 #ifdef TCP_ECN
1438 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1439 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1440 				goto drop;
1441 #endif
1442 			so->so_error = ECONNREFUSED;
1443 			goto close;
1444 
1445 		case TCPS_ESTABLISHED:
1446 		case TCPS_FIN_WAIT_1:
1447 		case TCPS_FIN_WAIT_2:
1448 		case TCPS_CLOSE_WAIT:
1449 			so->so_error = ECONNRESET;
1450 		close:
1451 			tp->t_state = TCPS_CLOSED;
1452 			tcpstat.tcps_drops++;
1453 			tp = tcp_close(tp);
1454 			goto drop;
1455 		case TCPS_CLOSING:
1456 		case TCPS_LAST_ACK:
1457 		case TCPS_TIME_WAIT:
1458 			tp = tcp_close(tp);
1459 			goto drop;
1460 		}
1461 	}
1462 
1463 	/*
1464 	 * If a SYN is in the window, then this is an
1465 	 * error and we ACK and drop the packet.
1466 	 */
1467 	if (tiflags & TH_SYN)
1468 		goto dropafterack_ratelim;
1469 
1470 	/*
1471 	 * If the ACK bit is off we drop the segment and return.
1472 	 */
1473 	if ((tiflags & TH_ACK) == 0) {
1474 		if (tp->t_flags & TF_ACKNOW)
1475 			goto dropafterack;
1476 		else
1477 			goto drop;
1478 	}
1479 
1480 	/*
1481 	 * Ack processing.
1482 	 */
1483 	switch (tp->t_state) {
1484 
1485 	/*
1486 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1487 	 * ESTABLISHED state and continue processing.
1488 	 * The ACK was checked above.
1489 	 */
1490 	case TCPS_SYN_RECEIVED:
1491 		tcpstat.tcps_connects++;
1492 		soisconnected(so);
1493 		tp->t_state = TCPS_ESTABLISHED;
1494 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1495 		/* Do window scaling? */
1496 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1497 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1498 			tp->snd_scale = tp->requested_s_scale;
1499 			tp->rcv_scale = tp->request_r_scale;
1500 		}
1501 		tcp_reass_lock(tp);
1502 		(void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0,
1503 				 &tlen);
1504 		tcp_reass_unlock(tp);
1505 		tp->snd_wl1 = th->th_seq - 1;
1506 		/* fall into ... */
1507 
1508 	/*
1509 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1510 	 * ACKs.  If the ack is in the range
1511 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1512 	 * then advance tp->snd_una to th->th_ack and drop
1513 	 * data from the retransmission queue.  If this ACK reflects
1514 	 * more up to date window information we update our window information.
1515 	 */
1516 	case TCPS_ESTABLISHED:
1517 	case TCPS_FIN_WAIT_1:
1518 	case TCPS_FIN_WAIT_2:
1519 	case TCPS_CLOSE_WAIT:
1520 	case TCPS_CLOSING:
1521 	case TCPS_LAST_ACK:
1522 	case TCPS_TIME_WAIT:
1523 #ifdef TCP_ECN
1524 		/*
1525 		 * if we receive ECE and are not already in recovery phase,
1526 		 * reduce cwnd by half but don't slow-start.
1527 		 * advance snd_last to snd_max not to reduce cwnd again
1528 		 * until all outstanding packets are acked.
1529 		 */
1530 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1531 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1532 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1533 				u_int win;
1534 
1535 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1536 				if (win > 1) {
1537 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1538 					tp->snd_cwnd = tp->snd_ssthresh;
1539 					tp->snd_last = tp->snd_max;
1540 					tp->t_flags |= TF_SEND_CWR;
1541 					tcpstat.tcps_cwr_ecn++;
1542 				}
1543 			}
1544 			tcpstat.tcps_ecn_rcvece++;
1545 		}
1546 		/*
1547 		 * if we receive CWR, we know that the peer has reduced
1548 		 * its congestion window.  stop sending ecn-echo.
1549 		 */
1550 		if ((tiflags & TH_CWR)) {
1551 			tp->t_flags &= ~TF_RCVD_CE;
1552 			tcpstat.tcps_ecn_rcvcwr++;
1553 		}
1554 #endif /* TCP_ECN */
1555 
1556 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1557 			/*
1558 			 * Duplicate/old ACK processing.
1559 			 * Increments t_dupacks:
1560 			 *	Pure duplicate (same seq/ack/window, no data)
1561 			 * Doesn't affect t_dupacks:
1562 			 *	Data packets.
1563 			 *	Normal window updates (window opens)
1564 			 * Resets t_dupacks:
1565 			 *	New data ACKed.
1566 			 *	Window shrinks
1567 			 *	Old ACK
1568 			 */
1569 			if (tlen) {
1570 				/* Drop very old ACKs unless th_seq matches */
1571 				if (th->th_seq != tp->rcv_nxt &&
1572 				   SEQ_LT(th->th_ack,
1573 				   tp->snd_una - tp->max_sndwnd)) {
1574 					tcpstat.tcps_rcvacktooold++;
1575 					goto drop;
1576 				}
1577 				break;
1578 			}
1579 			/*
1580 			 * If we get an old ACK, there is probably packet
1581 			 * reordering going on.  Be conservative and reset
1582 			 * t_dupacks so that we are less agressive in
1583 			 * doing a fast retransmit.
1584 			 */
1585 			if (th->th_ack != tp->snd_una) {
1586 				tp->t_dupacks = 0;
1587 				break;
1588 			}
1589 			if (tiwin == tp->snd_wnd) {
1590 				tcpstat.tcps_rcvdupack++;
1591 				/*
1592 				 * If we have outstanding data (other than
1593 				 * a window probe), this is a completely
1594 				 * duplicate ack (ie, window info didn't
1595 				 * change), the ack is the biggest we've
1596 				 * seen and we've seen exactly our rexmt
1597 				 * threshold of them, assume a packet
1598 				 * has been dropped and retransmit it.
1599 				 * Kludge snd_nxt & the congestion
1600 				 * window so we send only this one
1601 				 * packet.
1602 				 *
1603 				 * We know we're losing at the current
1604 				 * window size so do congestion avoidance
1605 				 * (set ssthresh to half the current window
1606 				 * and pull our congestion window back to
1607 				 * the new ssthresh).
1608 				 *
1609 				 * Dup acks mean that packets have left the
1610 				 * network (they're now cached at the receiver)
1611 				 * so bump cwnd by the amount in the receiver
1612 				 * to keep a constant cwnd packets in the
1613 				 * network.
1614 				 */
1615 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1616 					tp->t_dupacks = 0;
1617 #if defined(TCP_SACK) && defined(TCP_FACK)
1618 				/*
1619 				 * In FACK, can enter fast rec. if the receiver
1620 				 * reports a reass. queue longer than 3 segs.
1621 				 */
1622 				else if (++tp->t_dupacks == tcprexmtthresh ||
1623 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1624 				    tp->t_maxseg + tp->snd_una)) &&
1625 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1626 #else
1627 				else if (++tp->t_dupacks == tcprexmtthresh) {
1628 #endif /* TCP_FACK */
1629 					tcp_seq onxt = tp->snd_nxt;
1630 					u_long win =
1631 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1632 						2 / tp->t_maxseg;
1633 
1634 #if defined(TCP_SACK) || defined(TCP_ECN)
1635 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1636 					    	/*
1637 						 * False fast retx after
1638 						 * timeout.  Do not cut window.
1639 						 */
1640 						tp->t_dupacks = 0;
1641 						goto drop;
1642 					}
1643 #endif
1644 					if (win < 2)
1645 						win = 2;
1646 					tp->snd_ssthresh = win * tp->t_maxseg;
1647 #if defined(TCP_SACK)
1648 					tp->snd_last = tp->snd_max;
1649 #endif
1650 #ifdef TCP_SACK
1651                     			if (tp->sack_enable) {
1652 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1653 						tp->t_rtttime = 0;
1654 #ifdef TCP_ECN
1655 						tp->t_flags |= TF_SEND_CWR;
1656 #endif
1657 #if 1 /* TCP_ECN */
1658 						tcpstat.tcps_cwr_frecovery++;
1659 #endif
1660 						tcpstat.tcps_sack_recovery_episode++;
1661 #if defined(TCP_SACK) && defined(TCP_FACK)
1662 						tp->t_dupacks = tcprexmtthresh;
1663 						(void) tcp_output(tp);
1664 						/*
1665 						 * During FR, snd_cwnd is held
1666 						 * constant for FACK.
1667 						 */
1668 						tp->snd_cwnd = tp->snd_ssthresh;
1669 #else
1670 						/*
1671 						 * tcp_output() will send
1672 						 * oldest SACK-eligible rtx.
1673 						 */
1674 						(void) tcp_output(tp);
1675 						tp->snd_cwnd = tp->snd_ssthresh+
1676 					           tp->t_maxseg * tp->t_dupacks;
1677 #endif /* TCP_FACK */
1678 						goto drop;
1679 					}
1680 #endif /* TCP_SACK */
1681 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1682 					tp->t_rtttime = 0;
1683 					tp->snd_nxt = th->th_ack;
1684 					tp->snd_cwnd = tp->t_maxseg;
1685 #ifdef TCP_ECN
1686 					tp->t_flags |= TF_SEND_CWR;
1687 #endif
1688 #if 1 /* TCP_ECN */
1689 					tcpstat.tcps_cwr_frecovery++;
1690 #endif
1691 					tcpstat.tcps_sndrexmitfast++;
1692 					(void) tcp_output(tp);
1693 
1694 					tp->snd_cwnd = tp->snd_ssthresh +
1695 					    tp->t_maxseg * tp->t_dupacks;
1696 					if (SEQ_GT(onxt, tp->snd_nxt))
1697 						tp->snd_nxt = onxt;
1698 					goto drop;
1699 				} else if (tp->t_dupacks > tcprexmtthresh) {
1700 #if defined(TCP_SACK) && defined(TCP_FACK)
1701 					/*
1702 					 * while (awnd < cwnd)
1703 					 *         sendsomething();
1704 					 */
1705 					if (tp->sack_enable) {
1706 						if (tp->snd_awnd < tp->snd_cwnd)
1707 							tcp_output(tp);
1708 						goto drop;
1709 					}
1710 #endif /* TCP_FACK */
1711 					tp->snd_cwnd += tp->t_maxseg;
1712 					(void) tcp_output(tp);
1713 					goto drop;
1714 				}
1715 			} else if (tiwin < tp->snd_wnd) {
1716 				/*
1717 				 * The window was retracted!  Previous dup
1718 				 * ACKs may have been due to packets arriving
1719 				 * after the shrunken window, not a missing
1720 				 * packet, so play it safe and reset t_dupacks
1721 				 */
1722 				tp->t_dupacks = 0;
1723 			}
1724 			break;
1725 		}
1726 		/*
1727 		 * If the congestion window was inflated to account
1728 		 * for the other side's cached packets, retract it.
1729 		 */
1730 #if defined(TCP_SACK)
1731 		if (tp->sack_enable) {
1732 			if (tp->t_dupacks >= tcprexmtthresh) {
1733 				/* Check for a partial ACK */
1734 				if (tcp_sack_partialack(tp, th)) {
1735 #if defined(TCP_SACK) && defined(TCP_FACK)
1736 					/* Force call to tcp_output */
1737 					if (tp->snd_awnd < tp->snd_cwnd)
1738 						needoutput = 1;
1739 #else
1740 					tp->snd_cwnd += tp->t_maxseg;
1741 					needoutput = 1;
1742 #endif /* TCP_FACK */
1743 				} else {
1744 					/* Out of fast recovery */
1745 					tp->snd_cwnd = tp->snd_ssthresh;
1746 					if (tcp_seq_subtract(tp->snd_max,
1747 					    th->th_ack) < tp->snd_ssthresh)
1748 						tp->snd_cwnd =
1749 						   tcp_seq_subtract(tp->snd_max,
1750 					           th->th_ack);
1751 					tp->t_dupacks = 0;
1752 #if defined(TCP_SACK) && defined(TCP_FACK)
1753 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1754 						tp->snd_fack = th->th_ack;
1755 #endif /* TCP_FACK */
1756 				}
1757 			}
1758 		} else {
1759 			if (tp->t_dupacks >= tcprexmtthresh &&
1760 			    !tcp_newreno(tp, th)) {
1761 				/* Out of fast recovery */
1762 				tp->snd_cwnd = tp->snd_ssthresh;
1763 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1764 			  	    tp->snd_ssthresh)
1765 					tp->snd_cwnd =
1766 					    tcp_seq_subtract(tp->snd_max,
1767 					    th->th_ack);
1768 				tp->t_dupacks = 0;
1769 			}
1770 		}
1771 		if (tp->t_dupacks < tcprexmtthresh)
1772 			tp->t_dupacks = 0;
1773 #else /* else no TCP_SACK */
1774 		if (tp->t_dupacks >= tcprexmtthresh &&
1775 		    tp->snd_cwnd > tp->snd_ssthresh)
1776 			tp->snd_cwnd = tp->snd_ssthresh;
1777 		tp->t_dupacks = 0;
1778 #endif
1779 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1780 			tcpstat.tcps_rcvacktoomuch++;
1781 			goto dropafterack_ratelim;
1782 		}
1783 		acked = th->th_ack - tp->snd_una;
1784 		tcpstat.tcps_rcvackpack++;
1785 		tcpstat.tcps_rcvackbyte += acked;
1786 
1787 		/*
1788 		 * If we have a timestamp reply, update smoothed
1789 		 * round trip time.  If no timestamp is present but
1790 		 * transmit timer is running and timed sequence
1791 		 * number was acked, update smoothed round trip time.
1792 		 * Since we now have an rtt measurement, cancel the
1793 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1794 		 * Recompute the initial retransmit timer.
1795 		 */
1796 		if (opti.ts_present && opti.ts_ecr)
1797 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1798 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1799 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1800 
1801 		/*
1802 		 * If all outstanding data is acked, stop retransmit
1803 		 * timer and remember to restart (more output or persist).
1804 		 * If there is more data to be acked, restart retransmit
1805 		 * timer, using current (possibly backed-off) value.
1806 		 */
1807 		if (th->th_ack == tp->snd_max) {
1808 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1809 			needoutput = 1;
1810 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1811 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1812 		/*
1813 		 * When new data is acked, open the congestion window.
1814 		 * If the window gives us less than ssthresh packets
1815 		 * in flight, open exponentially (maxseg per packet).
1816 		 * Otherwise open linearly: maxseg per window
1817 		 * (maxseg^2 / cwnd per packet).
1818 		 */
1819 		{
1820 		u_int cw = tp->snd_cwnd;
1821 		u_int incr = tp->t_maxseg;
1822 
1823 		if (cw > tp->snd_ssthresh)
1824 			incr = incr * incr / cw;
1825 #if defined (TCP_SACK)
1826 		if (tp->t_dupacks < tcprexmtthresh)
1827 #endif
1828 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1829 		}
1830 		ND6_HINT(tp);
1831 		if (acked > so->so_snd.sb_cc) {
1832 			tp->snd_wnd -= so->so_snd.sb_cc;
1833 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1834 			ourfinisacked = 1;
1835 		} else {
1836 			sbdrop(&so->so_snd, acked);
1837 			tp->snd_wnd -= acked;
1838 			ourfinisacked = 0;
1839 		}
1840 		if (sb_notify(&so->so_snd))
1841 			sowwakeup(so);
1842 
1843 		/*
1844 		 * If we had a pending ICMP message that referred to data
1845 		 * that have just been acknowledged, disregard the recorded
1846 		 * ICMP message.
1847 		 */
1848 		if ((tp->t_flags & TF_PMTUD_PEND) &&
1849 		    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1850 			tp->t_flags &= ~TF_PMTUD_PEND;
1851 
1852 		/*
1853 		 * Keep track of the largest chunk of data acknowledged
1854 		 * since last PMTU update
1855 		 */
1856 		if (tp->t_pmtud_mss_acked < acked)
1857 		    tp->t_pmtud_mss_acked = acked;
1858 
1859 		tp->snd_una = th->th_ack;
1860 #ifdef TCP_ECN
1861 		/* sync snd_last with snd_una */
1862 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1863 			tp->snd_last = tp->snd_una;
1864 #endif
1865 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1866 			tp->snd_nxt = tp->snd_una;
1867 #if defined (TCP_SACK) && defined (TCP_FACK)
1868 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1869 			tp->snd_fack = tp->snd_una;
1870 			/* Update snd_awnd for partial ACK
1871 			 * without any SACK blocks.
1872 			 */
1873 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1874 				tp->snd_fack) + tp->retran_data;
1875 		}
1876 #endif
1877 
1878 		switch (tp->t_state) {
1879 
1880 		/*
1881 		 * In FIN_WAIT_1 STATE in addition to the processing
1882 		 * for the ESTABLISHED state if our FIN is now acknowledged
1883 		 * then enter FIN_WAIT_2.
1884 		 */
1885 		case TCPS_FIN_WAIT_1:
1886 			if (ourfinisacked) {
1887 				/*
1888 				 * If we can't receive any more
1889 				 * data, then closing user can proceed.
1890 				 * Starting the timer is contrary to the
1891 				 * specification, but if we don't get a FIN
1892 				 * we'll hang forever.
1893 				 */
1894 				if (so->so_state & SS_CANTRCVMORE) {
1895 					soisdisconnected(so);
1896 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1897 				}
1898 				tp->t_state = TCPS_FIN_WAIT_2;
1899 			}
1900 			break;
1901 
1902 		/*
1903 		 * In CLOSING STATE in addition to the processing for
1904 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1905 		 * then enter the TIME-WAIT state, otherwise ignore
1906 		 * the segment.
1907 		 */
1908 		case TCPS_CLOSING:
1909 			if (ourfinisacked) {
1910 				tp->t_state = TCPS_TIME_WAIT;
1911 				tcp_canceltimers(tp);
1912 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1913 				soisdisconnected(so);
1914 			}
1915 			break;
1916 
1917 		/*
1918 		 * In LAST_ACK, we may still be waiting for data to drain
1919 		 * and/or to be acked, as well as for the ack of our FIN.
1920 		 * If our FIN is now acknowledged, delete the TCB,
1921 		 * enter the closed state and return.
1922 		 */
1923 		case TCPS_LAST_ACK:
1924 			if (ourfinisacked) {
1925 				tp = tcp_close(tp);
1926 				goto drop;
1927 			}
1928 			break;
1929 
1930 		/*
1931 		 * In TIME_WAIT state the only thing that should arrive
1932 		 * is a retransmission of the remote FIN.  Acknowledge
1933 		 * it and restart the finack timer.
1934 		 */
1935 		case TCPS_TIME_WAIT:
1936 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1937 			goto dropafterack;
1938 		}
1939 	}
1940 
1941 step6:
1942 	/*
1943 	 * Update window information.
1944 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1945 	 */
1946 	if ((tiflags & TH_ACK) &&
1947 	    (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq &&
1948 	    (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1949 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1950 		/* keep track of pure window updates */
1951 		if (tlen == 0 &&
1952 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1953 			tcpstat.tcps_rcvwinupd++;
1954 		tp->snd_wnd = tiwin;
1955 		tp->snd_wl1 = th->th_seq;
1956 		tp->snd_wl2 = th->th_ack;
1957 		if (tp->snd_wnd > tp->max_sndwnd)
1958 			tp->max_sndwnd = tp->snd_wnd;
1959 		needoutput = 1;
1960 	}
1961 
1962 	/*
1963 	 * Process segments with URG.
1964 	 */
1965 	if ((tiflags & TH_URG) && th->th_urp &&
1966 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1967 		/*
1968 		 * This is a kludge, but if we receive and accept
1969 		 * random urgent pointers, we'll crash in
1970 		 * soreceive.  It's hard to imagine someone
1971 		 * actually wanting to send this much urgent data.
1972 		 */
1973 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
1974 			th->th_urp = 0;			/* XXX */
1975 			tiflags &= ~TH_URG;		/* XXX */
1976 			goto dodata;			/* XXX */
1977 		}
1978 		/*
1979 		 * If this segment advances the known urgent pointer,
1980 		 * then mark the data stream.  This should not happen
1981 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1982 		 * a FIN has been received from the remote side.
1983 		 * In these states we ignore the URG.
1984 		 *
1985 		 * According to RFC961 (Assigned Protocols),
1986 		 * the urgent pointer points to the last octet
1987 		 * of urgent data.  We continue, however,
1988 		 * to consider it to indicate the first octet
1989 		 * of data past the urgent section as the original
1990 		 * spec states (in one of two places).
1991 		 */
1992 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
1993 			tp->rcv_up = th->th_seq + th->th_urp;
1994 			so->so_oobmark = so->so_rcv.sb_cc +
1995 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1996 			if (so->so_oobmark == 0)
1997 				so->so_state |= SS_RCVATMARK;
1998 			sohasoutofband(so);
1999 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2000 		}
2001 		/*
2002 		 * Remove out of band data so doesn't get presented to user.
2003 		 * This can happen independent of advancing the URG pointer,
2004 		 * but if two URG's are pending at once, some out-of-band
2005 		 * data may creep in... ick.
2006 		 */
2007 		if (th->th_urp <= (u_int16_t) tlen
2008 #ifdef SO_OOBINLINE
2009 		     && (so->so_options & SO_OOBINLINE) == 0
2010 #endif
2011 		     )
2012 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2013 	} else
2014 		/*
2015 		 * If no out of band data is expected,
2016 		 * pull receive urgent pointer along
2017 		 * with the receive window.
2018 		 */
2019 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2020 			tp->rcv_up = tp->rcv_nxt;
2021 dodata:							/* XXX */
2022 
2023 	/*
2024 	 * Process the segment text, merging it into the TCP sequencing queue,
2025 	 * and arranging for acknowledgment of receipt if necessary.
2026 	 * This process logically involves adjusting tp->rcv_wnd as data
2027 	 * is presented to the user (this happens in tcp_usrreq.c,
2028 	 * case PRU_RCVD).  If a FIN has already been received on this
2029 	 * connection then we just ignore the text.
2030 	 */
2031 	if ((tlen || (tiflags & TH_FIN)) &&
2032 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2033 #ifdef TCP_SACK
2034 		tcp_seq laststart = th->th_seq;
2035 		tcp_seq lastend = th->th_seq + tlen;
2036 #endif
2037 		tcp_reass_lock(tp);
2038 		if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) &&
2039 		    tp->t_state == TCPS_ESTABLISHED) {
2040 			tcp_reass_unlock(tp);
2041 			TCP_SETUP_ACK(tp, tiflags);
2042 			tp->rcv_nxt += tlen;
2043 			tiflags = th->th_flags & TH_FIN;
2044 			tcpstat.tcps_rcvpack++;
2045 			tcpstat.tcps_rcvbyte += tlen;
2046 			ND6_HINT(tp);
2047 			if (so->so_state & SS_CANTRCVMORE)
2048 				m_freem(m);
2049 			else {
2050 				m_adj(m, hdroptlen);
2051 				sbappendstream(&so->so_rcv, m);
2052 			}
2053 			sorwakeup(so);
2054 		} else {
2055 			m_adj(m, hdroptlen);
2056 			tiflags = tcp_reass(tp, th, m, &tlen);
2057 			tcp_reass_unlock(tp);
2058 			tp->t_flags |= TF_ACKNOW;
2059 		}
2060 #ifdef TCP_SACK
2061 		if (tp->sack_enable)
2062 			tcp_update_sack_list(tp, laststart, lastend);
2063 #endif
2064 
2065 		/*
2066 		 * variable len never referenced again in modern BSD,
2067 		 * so why bother computing it ??
2068 		 */
2069 #if 0
2070 		/*
2071 		 * Note the amount of data that peer has sent into
2072 		 * our window, in order to estimate the sender's
2073 		 * buffer size.
2074 		 */
2075 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2076 #endif /* 0 */
2077 	} else {
2078 		m_freem(m);
2079 		tiflags &= ~TH_FIN;
2080 	}
2081 
2082 	/*
2083 	 * If FIN is received ACK the FIN and let the user know
2084 	 * that the connection is closing.  Ignore a FIN received before
2085 	 * the connection is fully established.
2086 	 */
2087 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2088 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2089 			socantrcvmore(so);
2090 			tp->t_flags |= TF_ACKNOW;
2091 			tp->rcv_nxt++;
2092 		}
2093 		switch (tp->t_state) {
2094 
2095 		/*
2096 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2097 		 */
2098 		case TCPS_ESTABLISHED:
2099 			tp->t_state = TCPS_CLOSE_WAIT;
2100 			break;
2101 
2102 		/*
2103 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2104 		 * enter the CLOSING state.
2105 		 */
2106 		case TCPS_FIN_WAIT_1:
2107 			tp->t_state = TCPS_CLOSING;
2108 			break;
2109 
2110 		/*
2111 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2112 		 * starting the time-wait timer, turning off the other
2113 		 * standard timers.
2114 		 */
2115 		case TCPS_FIN_WAIT_2:
2116 			tp->t_state = TCPS_TIME_WAIT;
2117 			tcp_canceltimers(tp);
2118 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2119 			soisdisconnected(so);
2120 			break;
2121 
2122 		/*
2123 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2124 		 */
2125 		case TCPS_TIME_WAIT:
2126 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2127 			break;
2128 		}
2129 	}
2130 	if (so->so_options & SO_DEBUG) {
2131 		switch (tp->pf) {
2132 #ifdef INET6
2133 		case PF_INET6:
2134 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2135 			    0, tlen);
2136 			break;
2137 #endif /* INET6 */
2138 		case PF_INET:
2139 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2140 			    0, tlen);
2141 			break;
2142 		}
2143 	}
2144 
2145 	/*
2146 	 * Return any desired output.
2147 	 */
2148 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2149 		(void) tcp_output(tp);
2150 	}
2151 	return;
2152 
2153 badsyn:
2154 	/*
2155 	 * Received a bad SYN.  Increment counters and dropwithreset.
2156 	 */
2157 	tcpstat.tcps_badsyn++;
2158 	tp = NULL;
2159 	goto dropwithreset;
2160 
2161 dropafterack_ratelim:
2162 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2163 	    tcp_ackdrop_ppslim) == 0) {
2164 		/* XXX stat */
2165 		goto drop;
2166 	}
2167 	/* ...fall into dropafterack... */
2168 
2169 dropafterack:
2170 	/*
2171 	 * Generate an ACK dropping incoming segment if it occupies
2172 	 * sequence space, where the ACK reflects our state.
2173 	 */
2174 	if (tiflags & TH_RST)
2175 		goto drop;
2176 	m_freem(m);
2177 	tp->t_flags |= TF_ACKNOW;
2178 	(void) tcp_output(tp);
2179 	return;
2180 
2181 dropwithreset_ratelim:
2182 	/*
2183 	 * We may want to rate-limit RSTs in certain situations,
2184 	 * particularly if we are sending an RST in response to
2185 	 * an attempt to connect to or otherwise communicate with
2186 	 * a port for which we have no socket.
2187 	 */
2188 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2189 	    tcp_rst_ppslim) == 0) {
2190 		/* XXX stat */
2191 		goto drop;
2192 	}
2193 	/* ...fall into dropwithreset... */
2194 
2195 dropwithreset:
2196 	/*
2197 	 * Generate a RST, dropping incoming segment.
2198 	 * Make ACK acceptable to originator of segment.
2199 	 * Don't bother to respond to RST.
2200 	 */
2201 	if (tiflags & TH_RST)
2202 		goto drop;
2203 	if (tiflags & TH_ACK) {
2204 		tcp_respond(tp, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack,
2205 		    TH_RST);
2206 	} else {
2207 		if (tiflags & TH_SYN)
2208 			tlen++;
2209 		tcp_respond(tp, mtod(m, caddr_t), m, th->th_seq + tlen,
2210 		    (tcp_seq)0, TH_RST|TH_ACK);
2211 	}
2212 	return;
2213 
2214 drop:
2215 	/*
2216 	 * Drop space held by incoming segment and return.
2217 	 */
2218 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2219 		switch (tp->pf) {
2220 #ifdef INET6
2221 		case PF_INET6:
2222 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2223 			    0, tlen);
2224 			break;
2225 #endif /* INET6 */
2226 		case PF_INET:
2227 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2228 			    0, tlen);
2229 			break;
2230 		}
2231 	}
2232 
2233 	m_freem(m);
2234 	return;
2235 }
2236 
2237 int
2238 tcp_dooptions(tp, cp, cnt, th, m, iphlen, oi)
2239 	struct tcpcb *tp;
2240 	u_char *cp;
2241 	int cnt;
2242 	struct tcphdr *th;
2243 	struct mbuf *m;
2244 	int iphlen;
2245 	struct tcp_opt_info *oi;
2246 {
2247 	u_int16_t mss = 0;
2248 	int opt, optlen;
2249 #ifdef TCP_SIGNATURE
2250 	caddr_t sigp = NULL;
2251 	struct tdb *tdb = NULL;
2252 #endif /* TCP_SIGNATURE */
2253 
2254 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2255 		opt = cp[0];
2256 		if (opt == TCPOPT_EOL)
2257 			break;
2258 		if (opt == TCPOPT_NOP)
2259 			optlen = 1;
2260 		else {
2261 			if (cnt < 2)
2262 				break;
2263 			optlen = cp[1];
2264 			if (optlen < 2 || optlen > cnt)
2265 				break;
2266 		}
2267 		switch (opt) {
2268 
2269 		default:
2270 			continue;
2271 
2272 		case TCPOPT_MAXSEG:
2273 			if (optlen != TCPOLEN_MAXSEG)
2274 				continue;
2275 			if (!(th->th_flags & TH_SYN))
2276 				continue;
2277 			if (TCPS_HAVERCVDSYN(tp->t_state))
2278 				continue;
2279 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2280 			NTOHS(mss);
2281 			oi->maxseg = mss;
2282 			break;
2283 
2284 		case TCPOPT_WINDOW:
2285 			if (optlen != TCPOLEN_WINDOW)
2286 				continue;
2287 			if (!(th->th_flags & TH_SYN))
2288 				continue;
2289 			if (TCPS_HAVERCVDSYN(tp->t_state))
2290 				continue;
2291 			tp->t_flags |= TF_RCVD_SCALE;
2292 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2293 			break;
2294 
2295 		case TCPOPT_TIMESTAMP:
2296 			if (optlen != TCPOLEN_TIMESTAMP)
2297 				continue;
2298 			oi->ts_present = 1;
2299 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2300 			NTOHL(oi->ts_val);
2301 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2302 			NTOHL(oi->ts_ecr);
2303 
2304 			if (!(th->th_flags & TH_SYN))
2305 				continue;
2306 			if (TCPS_HAVERCVDSYN(tp->t_state))
2307 				continue;
2308 			/*
2309 			 * A timestamp received in a SYN makes
2310 			 * it ok to send timestamp requests and replies.
2311 			 */
2312 			tp->t_flags |= TF_RCVD_TSTMP;
2313 			tp->ts_recent = oi->ts_val;
2314 			tp->ts_recent_age = tcp_now;
2315 			break;
2316 
2317 #ifdef TCP_SACK
2318 		case TCPOPT_SACK_PERMITTED:
2319 			if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED)
2320 				continue;
2321 			if (!(th->th_flags & TH_SYN))
2322 				continue;
2323 			if (TCPS_HAVERCVDSYN(tp->t_state))
2324 				continue;
2325 			/* MUST only be set on SYN */
2326 			tp->t_flags |= TF_SACK_PERMIT;
2327 			break;
2328 		case TCPOPT_SACK:
2329 			tcp_sack_option(tp, th, cp, optlen);
2330 			break;
2331 #endif
2332 #ifdef TCP_SIGNATURE
2333 		case TCPOPT_SIGNATURE:
2334 			if (optlen != TCPOLEN_SIGNATURE)
2335 				continue;
2336 
2337 			if (sigp && bcmp(sigp, cp + 2, 16))
2338 				return (-1);
2339 
2340 			sigp = cp + 2;
2341 			break;
2342 #endif /* TCP_SIGNATURE */
2343 		}
2344 	}
2345 
2346 #ifdef TCP_SIGNATURE
2347 	if (tp->t_flags & TF_SIGNATURE) {
2348 		union sockaddr_union src, dst;
2349 
2350 		memset(&src, 0, sizeof(union sockaddr_union));
2351 		memset(&dst, 0, sizeof(union sockaddr_union));
2352 
2353 		switch (tp->pf) {
2354 		case 0:
2355 #ifdef INET
2356 		case AF_INET:
2357 			src.sa.sa_len = sizeof(struct sockaddr_in);
2358 			src.sa.sa_family = AF_INET;
2359 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
2360 			dst.sa.sa_len = sizeof(struct sockaddr_in);
2361 			dst.sa.sa_family = AF_INET;
2362 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
2363 			break;
2364 #endif
2365 #ifdef INET6
2366 		case AF_INET6:
2367 			src.sa.sa_len = sizeof(struct sockaddr_in6);
2368 			src.sa.sa_family = AF_INET6;
2369 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
2370 			dst.sa.sa_len = sizeof(struct sockaddr_in6);
2371 			dst.sa.sa_family = AF_INET6;
2372 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
2373 			break;
2374 #endif /* INET6 */
2375 		}
2376 
2377 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
2378 
2379 		/*
2380 		 * We don't have an SA for this peer, so we turn off
2381 		 * TF_SIGNATURE on the listen socket
2382 		 */
2383 		if (tdb == NULL && tp->t_state == TCPS_LISTEN)
2384 			tp->t_flags &= ~TF_SIGNATURE;
2385 
2386 	}
2387 
2388 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
2389 		tcpstat.tcps_rcvbadsig++;
2390 		return (-1);
2391 	}
2392 
2393 	if (sigp) {
2394 		char sig[16];
2395 
2396 		if (tdb == NULL) {
2397 			tcpstat.tcps_rcvbadsig++;
2398 			return (-1);
2399 		}
2400 
2401 		if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0)
2402 			return (-1);
2403 
2404 		if (bcmp(sig, sigp, 16)) {
2405 			tcpstat.tcps_rcvbadsig++;
2406 			return (-1);
2407 		}
2408 
2409 		tcpstat.tcps_rcvgoodsig++;
2410 	}
2411 #endif /* TCP_SIGNATURE */
2412 
2413 	return (0);
2414 }
2415 
2416 #if defined(TCP_SACK)
2417 u_long
2418 tcp_seq_subtract(a, b)
2419 	u_long a, b;
2420 {
2421 	return ((long)(a - b));
2422 }
2423 #endif
2424 
2425 
2426 #ifdef TCP_SACK
2427 /*
2428  * This function is called upon receipt of new valid data (while not in header
2429  * prediction mode), and it updates the ordered list of sacks.
2430  */
2431 void
2432 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart,
2433     tcp_seq rcv_lastend)
2434 {
2435 	/*
2436 	 * First reported block MUST be the most recent one.  Subsequent
2437 	 * blocks SHOULD be in the order in which they arrived at the
2438 	 * receiver.  These two conditions make the implementation fully
2439 	 * compliant with RFC 2018.
2440 	 */
2441 	int i, j = 0, count = 0, lastpos = -1;
2442 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2443 
2444 	/* First clean up current list of sacks */
2445 	for (i = 0; i < tp->rcv_numsacks; i++) {
2446 		sack = tp->sackblks[i];
2447 		if (sack.start == 0 && sack.end == 0) {
2448 			count++; /* count = number of blocks to be discarded */
2449 			continue;
2450 		}
2451 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2452 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2453 			count++;
2454 		} else {
2455 			temp[j].start = tp->sackblks[i].start;
2456 			temp[j++].end = tp->sackblks[i].end;
2457 		}
2458 	}
2459 	tp->rcv_numsacks -= count;
2460 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2461 		tcp_clean_sackreport(tp);
2462 		if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) {
2463 			/* ==> need first sack block */
2464 			tp->sackblks[0].start = rcv_laststart;
2465 			tp->sackblks[0].end = rcv_lastend;
2466 			tp->rcv_numsacks = 1;
2467 		}
2468 		return;
2469 	}
2470 	/* Otherwise, sack blocks are already present. */
2471 	for (i = 0; i < tp->rcv_numsacks; i++)
2472 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2473 	if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend))
2474 		return;     /* sack list remains unchanged */
2475 	/*
2476 	 * From here, segment just received should be (part of) the 1st sack.
2477 	 * Go through list, possibly coalescing sack block entries.
2478 	 */
2479 	firstsack.start = rcv_laststart;
2480 	firstsack.end = rcv_lastend;
2481 	for (i = 0; i < tp->rcv_numsacks; i++) {
2482 		sack = tp->sackblks[i];
2483 		if (SEQ_LT(sack.end, firstsack.start) ||
2484 		    SEQ_GT(sack.start, firstsack.end))
2485 			continue; /* no overlap */
2486 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2487 			/*
2488 			 * identical block; delete it here since we will
2489 			 * move it to the front of the list.
2490 			 */
2491 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2492 			lastpos = i;    /* last posn with a zero entry */
2493 			continue;
2494 		}
2495 		if (SEQ_LEQ(sack.start, firstsack.start))
2496 			firstsack.start = sack.start; /* merge blocks */
2497 		if (SEQ_GEQ(sack.end, firstsack.end))
2498 			firstsack.end = sack.end;     /* merge blocks */
2499 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2500 		lastpos = i;    /* last posn with a zero entry */
2501 	}
2502 	if (lastpos != -1) {    /* at least one merge */
2503 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2504 			sack = tp->sackblks[i];
2505 			if (sack.start == 0 && sack.end == 0)
2506 				continue;
2507 			temp[j++] = sack;
2508 		}
2509 		tp->rcv_numsacks = j; /* including first blk (added later) */
2510 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2511 			tp->sackblks[i] = temp[i];
2512 	} else {        /* no merges -- shift sacks by 1 */
2513 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2514 			tp->rcv_numsacks++;
2515 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2516 			tp->sackblks[i] = tp->sackblks[i-1];
2517 	}
2518 	tp->sackblks[0] = firstsack;
2519 	return;
2520 }
2521 
2522 /*
2523  * Process the TCP SACK option.  tp->snd_holes is an ordered list
2524  * of holes (oldest to newest, in terms of the sequence space).
2525  */
2526 void
2527 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2528 {
2529 	int tmp_olen;
2530 	u_char *tmp_cp;
2531 	struct sackhole *cur, *p, *temp;
2532 
2533 	if (!tp->sack_enable)
2534 		return;
2535 	/* SACK without ACK doesn't make sense. */
2536 	if ((th->th_flags & TH_ACK) == 0)
2537 	       return;
2538 	/* Make sure the ACK on this segment is in [snd_una, snd_max]. */
2539 	if (SEQ_LT(th->th_ack, tp->snd_una) ||
2540 	    SEQ_GT(th->th_ack, tp->snd_max))
2541 		return;
2542 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2543 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2544 		return;
2545 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2546 	tmp_cp = cp + 2;
2547 	tmp_olen = optlen - 2;
2548 	tcpstat.tcps_sack_rcv_opts++;
2549 	if (tp->snd_numholes < 0)
2550 		tp->snd_numholes = 0;
2551 	if (tp->t_maxseg == 0)
2552 		panic("tcp_sack_option"); /* Should never happen */
2553 	while (tmp_olen > 0) {
2554 		struct sackblk sack;
2555 
2556 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2557 		NTOHL(sack.start);
2558 		bcopy(tmp_cp + sizeof(tcp_seq),
2559 		    (char *) &(sack.end), sizeof(tcp_seq));
2560 		NTOHL(sack.end);
2561 		tmp_olen -= TCPOLEN_SACK;
2562 		tmp_cp += TCPOLEN_SACK;
2563 		if (SEQ_LEQ(sack.end, sack.start))
2564 			continue; /* bad SACK fields */
2565 		if (SEQ_LEQ(sack.end, tp->snd_una))
2566 			continue; /* old block */
2567 #if defined(TCP_SACK) && defined(TCP_FACK)
2568 		/* Updates snd_fack.  */
2569 		if (SEQ_GT(sack.end, tp->snd_fack))
2570 			tp->snd_fack = sack.end;
2571 #endif /* TCP_FACK */
2572 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2573 			if (SEQ_LT(sack.start, th->th_ack))
2574 				continue;
2575 		}
2576 		if (SEQ_GT(sack.end, tp->snd_max))
2577 			continue;
2578 		if (tp->snd_holes == NULL) { /* first hole */
2579 			tp->snd_holes = (struct sackhole *)
2580 			    pool_get(&sackhl_pool, PR_NOWAIT);
2581 			if (tp->snd_holes == NULL) {
2582 				/* ENOBUFS, so ignore SACKed block for now*/
2583 				goto done;
2584 			}
2585 			cur = tp->snd_holes;
2586 			cur->start = th->th_ack;
2587 			cur->end = sack.start;
2588 			cur->rxmit = cur->start;
2589 			cur->next = NULL;
2590 			tp->snd_numholes = 1;
2591 			tp->rcv_lastsack = sack.end;
2592 			/*
2593 			 * dups is at least one.  If more data has been
2594 			 * SACKed, it can be greater than one.
2595 			 */
2596 			cur->dups = min(tcprexmtthresh,
2597 			    ((sack.end - cur->end)/tp->t_maxseg));
2598 			if (cur->dups < 1)
2599 				cur->dups = 1;
2600 			continue; /* with next sack block */
2601 		}
2602 		/* Go thru list of holes:  p = previous,  cur = current */
2603 		p = cur = tp->snd_holes;
2604 		while (cur) {
2605 			if (SEQ_LEQ(sack.end, cur->start))
2606 				/* SACKs data before the current hole */
2607 				break; /* no use going through more holes */
2608 			if (SEQ_GEQ(sack.start, cur->end)) {
2609 				/* SACKs data beyond the current hole */
2610 				cur->dups++;
2611 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2612 				    tcprexmtthresh)
2613 					cur->dups = tcprexmtthresh;
2614 				p = cur;
2615 				cur = cur->next;
2616 				continue;
2617 			}
2618 			if (SEQ_LEQ(sack.start, cur->start)) {
2619 				/* Data acks at least the beginning of hole */
2620 #if defined(TCP_SACK) && defined(TCP_FACK)
2621 				if (SEQ_GT(sack.end, cur->rxmit))
2622 					tp->retran_data -=
2623 				    	    tcp_seq_subtract(cur->rxmit,
2624 					    cur->start);
2625 				else
2626 					tp->retran_data -=
2627 					    tcp_seq_subtract(sack.end,
2628 					    cur->start);
2629 #endif /* TCP_FACK */
2630 				if (SEQ_GEQ(sack.end, cur->end)) {
2631 					/* Acks entire hole, so delete hole */
2632 					if (p != cur) {
2633 						p->next = cur->next;
2634 						pool_put(&sackhl_pool, cur);
2635 						cur = p->next;
2636 					} else {
2637 						cur = cur->next;
2638 						pool_put(&sackhl_pool, p);
2639 						p = cur;
2640 						tp->snd_holes = p;
2641 					}
2642 					tp->snd_numholes--;
2643 					continue;
2644 				}
2645 				/* otherwise, move start of hole forward */
2646 				cur->start = sack.end;
2647 				cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
2648 				p = cur;
2649 				cur = cur->next;
2650 				continue;
2651 			}
2652 			/* move end of hole backward */
2653 			if (SEQ_GEQ(sack.end, cur->end)) {
2654 #if defined(TCP_SACK) && defined(TCP_FACK)
2655 				if (SEQ_GT(cur->rxmit, sack.start))
2656 					tp->retran_data -=
2657 					    tcp_seq_subtract(cur->rxmit,
2658 					    sack.start);
2659 #endif /* TCP_FACK */
2660 				cur->end = sack.start;
2661 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2662 				cur->dups++;
2663 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2664 				    tcprexmtthresh)
2665 					cur->dups = tcprexmtthresh;
2666 				p = cur;
2667 				cur = cur->next;
2668 				continue;
2669 			}
2670 			if (SEQ_LT(cur->start, sack.start) &&
2671 			    SEQ_GT(cur->end, sack.end)) {
2672 				/*
2673 				 * ACKs some data in middle of a hole; need to
2674 				 * split current hole
2675 				 */
2676 				temp = (struct sackhole *)
2677 				    pool_get(&sackhl_pool, PR_NOWAIT);
2678 				if (temp == NULL)
2679 					goto done; /* ENOBUFS */
2680 #if defined(TCP_SACK) && defined(TCP_FACK)
2681 				if (SEQ_GT(cur->rxmit, sack.end))
2682 					tp->retran_data -=
2683 					    tcp_seq_subtract(sack.end,
2684 					    sack.start);
2685 				else if (SEQ_GT(cur->rxmit, sack.start))
2686 					tp->retran_data -=
2687 					    tcp_seq_subtract(cur->rxmit,
2688 					    sack.start);
2689 #endif /* TCP_FACK */
2690 				temp->next = cur->next;
2691 				temp->start = sack.end;
2692 				temp->end = cur->end;
2693 				temp->dups = cur->dups;
2694 				temp->rxmit = SEQ_MAX(cur->rxmit, temp->start);
2695 				cur->end = sack.start;
2696 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2697 				cur->dups++;
2698 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2699 					tcprexmtthresh)
2700 					cur->dups = tcprexmtthresh;
2701 				cur->next = temp;
2702 				p = temp;
2703 				cur = p->next;
2704 				tp->snd_numholes++;
2705 			}
2706 		}
2707 		/* At this point, p points to the last hole on the list */
2708 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2709 			/*
2710 			 * Need to append new hole at end.
2711 			 * Last hole is p (and it's not NULL).
2712 			 */
2713 			temp = (struct sackhole *)
2714 			    pool_get(&sackhl_pool, PR_NOWAIT);
2715 			if (temp == NULL)
2716 				goto done; /* ENOBUFS */
2717 			temp->start = tp->rcv_lastsack;
2718 			temp->end = sack.start;
2719 			temp->dups = min(tcprexmtthresh,
2720 			    ((sack.end - sack.start)/tp->t_maxseg));
2721 			if (temp->dups < 1)
2722 				temp->dups = 1;
2723 			temp->rxmit = temp->start;
2724 			temp->next = 0;
2725 			p->next = temp;
2726 			tp->rcv_lastsack = sack.end;
2727 			tp->snd_numholes++;
2728 		}
2729 	}
2730 done:
2731 #if defined(TCP_SACK) && defined(TCP_FACK)
2732 	/*
2733 	 * Update retran_data and snd_awnd.  Go through the list of
2734 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2735 	 */
2736 	tp->retran_data = 0;
2737 	cur = tp->snd_holes;
2738 	while (cur) {
2739 		tp->retran_data += cur->rxmit - cur->start;
2740 		cur = cur->next;
2741 	}
2742 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2743 	    tp->retran_data;
2744 #endif /* TCP_FACK */
2745 
2746 	return;
2747 }
2748 
2749 /*
2750  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2751  * it is completely acked; otherwise, tcp_sack_option(), called from
2752  * tcp_dooptions(), will fix up the hole.
2753  */
2754 void
2755 tcp_del_sackholes(tp, th)
2756 	struct tcpcb *tp;
2757 	struct tcphdr *th;
2758 {
2759 	if (tp->sack_enable && tp->t_state != TCPS_LISTEN) {
2760 		/* max because this could be an older ack just arrived */
2761 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2762 			th->th_ack : tp->snd_una;
2763 		struct sackhole *cur = tp->snd_holes;
2764 		struct sackhole *prev;
2765 		while (cur)
2766 			if (SEQ_LEQ(cur->end, lastack)) {
2767 				prev = cur;
2768 				cur = cur->next;
2769 				pool_put(&sackhl_pool, prev);
2770 				tp->snd_numholes--;
2771 			} else if (SEQ_LT(cur->start, lastack)) {
2772 				cur->start = lastack;
2773 				if (SEQ_LT(cur->rxmit, cur->start))
2774 					cur->rxmit = cur->start;
2775 				break;
2776 			} else
2777 				break;
2778 		tp->snd_holes = cur;
2779 	}
2780 }
2781 
2782 /*
2783  * Delete all receiver-side SACK information.
2784  */
2785 void
2786 tcp_clean_sackreport(tp)
2787 	struct tcpcb *tp;
2788 {
2789 	int i;
2790 
2791 	tp->rcv_numsacks = 0;
2792 	for (i = 0; i < MAX_SACK_BLKS; i++)
2793 		tp->sackblks[i].start = tp->sackblks[i].end=0;
2794 
2795 }
2796 
2797 /*
2798  * Checks for partial ack.  If partial ack arrives, turn off retransmission
2799  * timer, deflate the window, do not clear tp->t_dupacks, and return 1.
2800  * If the ack advances at least to tp->snd_last, return 0.
2801  */
2802 int
2803 tcp_sack_partialack(tp, th)
2804 	struct tcpcb *tp;
2805 	struct tcphdr *th;
2806 {
2807 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2808 		/* Turn off retx. timer (will start again next segment) */
2809 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2810 		tp->t_rtttime = 0;
2811 #ifndef TCP_FACK
2812 		/*
2813 		 * Partial window deflation.  This statement relies on the
2814 		 * fact that tp->snd_una has not been updated yet.  In FACK
2815 		 * hold snd_cwnd constant during fast recovery.
2816 		 */
2817 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2818 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2819 			tp->snd_cwnd += tp->t_maxseg;
2820 		} else
2821 			tp->snd_cwnd = tp->t_maxseg;
2822 #endif
2823 		return (1);
2824 	}
2825 	return (0);
2826 }
2827 #endif /* TCP_SACK */
2828 
2829 /*
2830  * Pull out of band byte out of a segment so
2831  * it doesn't appear in the user's data queue.
2832  * It is still reflected in the segment length for
2833  * sequencing purposes.
2834  */
2835 void
2836 tcp_pulloutofband(so, urgent, m, off)
2837 	struct socket *so;
2838 	u_int urgent;
2839 	struct mbuf *m;
2840 	int off;
2841 {
2842         int cnt = off + urgent - 1;
2843 
2844 	while (cnt >= 0) {
2845 		if (m->m_len > cnt) {
2846 			char *cp = mtod(m, caddr_t) + cnt;
2847 			struct tcpcb *tp = sototcpcb(so);
2848 
2849 			tp->t_iobc = *cp;
2850 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2851 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2852 			m->m_len--;
2853 			return;
2854 		}
2855 		cnt -= m->m_len;
2856 		m = m->m_next;
2857 		if (m == 0)
2858 			break;
2859 	}
2860 	panic("tcp_pulloutofband");
2861 }
2862 
2863 /*
2864  * Collect new round-trip time estimate
2865  * and update averages and current timeout.
2866  */
2867 void
2868 tcp_xmit_timer(tp, rtt)
2869 	struct tcpcb *tp;
2870 	short rtt;
2871 {
2872 	short delta;
2873 	short rttmin;
2874 
2875 	if (rtt < 0)
2876 		rtt = 0;
2877 	else if (rtt > TCP_RTT_MAX)
2878 		rtt = TCP_RTT_MAX;
2879 
2880 	tcpstat.tcps_rttupdated++;
2881 	if (tp->t_srtt != 0) {
2882 		/*
2883 		 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits
2884 		 * after the binary point (scaled by 4), whereas
2885 		 * srtt is stored as fixed point with 5 bits after the
2886 		 * binary point (i.e., scaled by 32).  The following magic
2887 		 * is equivalent to the smoothing algorithm in rfc793 with
2888 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2889 		 * point).
2890 		 */
2891 		delta = (rtt << TCP_RTT_BASE_SHIFT) -
2892 		    (tp->t_srtt >> TCP_RTT_SHIFT);
2893 		if ((tp->t_srtt += delta) <= 0)
2894 			tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT;
2895 		/*
2896 		 * We accumulate a smoothed rtt variance (actually, a
2897 		 * smoothed mean difference), then set the retransmit
2898 		 * timer to smoothed rtt + 4 times the smoothed variance.
2899 		 * rttvar is stored as fixed point with 4 bits after the
2900 		 * binary point (scaled by 16).  The following is
2901 		 * equivalent to rfc793 smoothing with an alpha of .75
2902 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2903 		 * rfc793's wired-in beta.
2904 		 */
2905 		if (delta < 0)
2906 			delta = -delta;
2907 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2908 		if ((tp->t_rttvar += delta) <= 0)
2909 			tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT;
2910 	} else {
2911 		/*
2912 		 * No rtt measurement yet - use the unsmoothed rtt.
2913 		 * Set the variance to half the rtt (so our first
2914 		 * retransmit happens at 3*rtt).
2915 		 */
2916 		tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
2917 		tp->t_rttvar = (rtt + 1) <<
2918 		    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
2919 	}
2920 	tp->t_rtttime = 0;
2921 	tp->t_rxtshift = 0;
2922 
2923 	/*
2924 	 * the retransmit should happen at rtt + 4 * rttvar.
2925 	 * Because of the way we do the smoothing, srtt and rttvar
2926 	 * will each average +1/2 tick of bias.  When we compute
2927 	 * the retransmit timer, we want 1/2 tick of rounding and
2928 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2929 	 * firing of the timer.  The bias will give us exactly the
2930 	 * 1.5 tick we need.  But, because the bias is
2931 	 * statistical, we have to test that we don't drop below
2932 	 * the minimum feasible timer (which is 2 ticks).
2933 	 */
2934 	rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX);
2935 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2936 
2937 	/*
2938 	 * We received an ack for a packet that wasn't retransmitted;
2939 	 * it is probably safe to discard any error indications we've
2940 	 * received recently.  This isn't quite right, but close enough
2941 	 * for now (a route might have failed after we sent a segment,
2942 	 * and the return path might not be symmetrical).
2943 	 */
2944 	tp->t_softerror = 0;
2945 }
2946 
2947 /*
2948  * Determine a reasonable value for maxseg size.
2949  * If the route is known, check route for mtu.
2950  * If none, use an mss that can be handled on the outgoing
2951  * interface without forcing IP to fragment; if bigger than
2952  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2953  * to utilize large mbufs.  If no route is found, route has no mtu,
2954  * or the destination isn't local, use a default, hopefully conservative
2955  * size (usually 512 or the default IP max size, but no more than the mtu
2956  * of the interface), as we can't discover anything about intervening
2957  * gateways or networks.  We also initialize the congestion/slow start
2958  * window to be a single segment if the destination isn't local.
2959  * While looking at the routing entry, we also initialize other path-dependent
2960  * parameters from pre-set or cached values in the routing entry.
2961  *
2962  * Also take into account the space needed for options that we
2963  * send regularly.  Make maxseg shorter by that amount to assure
2964  * that we can send maxseg amount of data even when the options
2965  * are present.  Store the upper limit of the length of options plus
2966  * data in maxopd.
2967  *
2968  * NOTE: offer == -1 indicates that the maxseg size changed due to
2969  * Path MTU discovery.
2970  */
2971 int
2972 tcp_mss(tp, offer)
2973 	struct tcpcb *tp;
2974 	int offer;
2975 {
2976 	struct rtentry *rt;
2977 	struct ifnet *ifp;
2978 	int mss, mssopt;
2979 	int iphlen;
2980 	struct inpcb *inp;
2981 
2982 	inp = tp->t_inpcb;
2983 
2984 	mssopt = mss = tcp_mssdflt;
2985 
2986 	rt = in_pcbrtentry(inp);
2987 
2988 	if (rt == NULL)
2989 		goto out;
2990 
2991 	ifp = rt->rt_ifp;
2992 
2993 	switch (tp->pf) {
2994 #ifdef INET6
2995 	case AF_INET6:
2996 		iphlen = sizeof(struct ip6_hdr);
2997 		break;
2998 #endif
2999 	case AF_INET:
3000 		iphlen = sizeof(struct ip);
3001 		break;
3002 	default:
3003 		/* the family does not support path MTU discovery */
3004 		goto out;
3005 	}
3006 
3007 #ifdef RTV_MTU
3008 	/*
3009 	 * if there's an mtu associated with the route and we support
3010 	 * path MTU discovery for the underlying protocol family, use it.
3011 	 */
3012 	if (rt->rt_rmx.rmx_mtu) {
3013 		/*
3014 		 * One may wish to lower MSS to take into account options,
3015 		 * especially security-related options.
3016 		 */
3017 		if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) {
3018 			/*
3019 			 * RFC2460 section 5, last paragraph: if path MTU is
3020 			 * smaller than 1280, use 1280 as packet size and
3021 			 * attach fragment header.
3022 			 */
3023 			mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) -
3024 			    sizeof(struct tcphdr);
3025 		} else
3026 			mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr);
3027 	} else
3028 #endif /* RTV_MTU */
3029 	if (!ifp)
3030 		/*
3031 		 * ifp may be null and rmx_mtu may be zero in certain
3032 		 * v6 cases (e.g., if ND wasn't able to resolve the
3033 		 * destination host.
3034 		 */
3035 		goto out;
3036 	else if (ifp->if_flags & IFF_LOOPBACK)
3037 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3038 	else if (tp->pf == AF_INET) {
3039 		if (ip_mtudisc)
3040 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3041 		else if (inp && in_localaddr(inp->inp_faddr))
3042 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3043 	}
3044 #ifdef INET6
3045 	else if (tp->pf == AF_INET6) {
3046 		/*
3047 		 * for IPv6, path MTU discovery is always turned on,
3048 		 * or the node must use packet size <= 1280.
3049 		 */
3050 		mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr);
3051 	}
3052 #endif /* INET6 */
3053 
3054 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3055 	if (offer != -1) {
3056 #ifndef INET6
3057 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3058 #else
3059 		if (tp->pf == AF_INET6)
3060 			mssopt = IN6_LINKMTU(ifp) - iphlen -
3061 			    sizeof(struct tcphdr);
3062 		else
3063 			mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3064 #endif
3065 
3066 		mssopt = max(tcp_mssdflt, mssopt);
3067 	}
3068 
3069  out:
3070 	/*
3071 	 * The current mss, t_maxseg, is initialized to the default value.
3072 	 * If we compute a smaller value, reduce the current mss.
3073 	 * If we compute a larger value, return it for use in sending
3074 	 * a max seg size option, but don't store it for use
3075 	 * unless we received an offer at least that large from peer.
3076 	 *
3077 	 * However, do not accept offers lower than the minimum of
3078 	 * the interface MTU and 216.
3079 	 */
3080 	if (offer > 0)
3081 		tp->t_peermss = offer;
3082 	if (tp->t_peermss)
3083 		mss = min(mss, max(tp->t_peermss, 216));
3084 
3085 	/* sanity - at least max opt. space */
3086 	mss = max(mss, 64);
3087 
3088 	/*
3089 	 * maxopd stores the maximum length of data AND options
3090 	 * in a segment; maxseg is the amount of data in a normal
3091 	 * segment.  We need to store this value (maxopd) apart
3092 	 * from maxseg, because now every segment carries options
3093 	 * and thus we normally have somewhat less data in segments.
3094 	 */
3095 	tp->t_maxopd = mss;
3096 
3097 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3098 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3099 		mss -= TCPOLEN_TSTAMP_APPA;
3100 #ifdef TCP_SIGNATURE
3101 	if (tp->t_flags & TF_SIGNATURE)
3102 		mss -= TCPOLEN_SIGLEN;
3103 #endif
3104 
3105 	if (offer == -1) {
3106 		/* mss changed due to Path MTU discovery */
3107 		tp->t_flags &= ~TF_PMTUD_PEND;
3108 		tp->t_pmtud_mtu_sent = 0;
3109 		tp->t_pmtud_mss_acked = 0;
3110 		if (mss < tp->t_maxseg) {
3111 			/*
3112 			 * Follow suggestion in RFC 2414 to reduce the
3113 			 * congestion window by the ratio of the old
3114 			 * segment size to the new segment size.
3115 			 */
3116 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3117 					     mss, mss);
3118 		}
3119 	} else if (tcp_do_rfc3390) {
3120 		/* increase initial window  */
3121 		tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380));
3122 	} else
3123 		tp->snd_cwnd = mss;
3124 
3125 	tp->t_maxseg = mss;
3126 
3127 	return (offer != -1 ? mssopt : mss);
3128 }
3129 
3130 u_int
3131 tcp_hdrsz(struct tcpcb *tp)
3132 {
3133 	u_int hlen;
3134 
3135 	switch (tp->pf) {
3136 #ifdef INET6
3137 	case AF_INET6:
3138 		hlen = sizeof(struct ip6_hdr);
3139 		break;
3140 #endif
3141 	case AF_INET:
3142 		hlen = sizeof(struct ip);
3143 		break;
3144 	default:
3145 		hlen = 0;
3146 		break;
3147 	}
3148 	hlen += sizeof(struct tcphdr);
3149 
3150 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3151 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3152 		hlen += TCPOLEN_TSTAMP_APPA;
3153 #ifdef TCP_SIGNATURE
3154 	if (tp->t_flags & TF_SIGNATURE)
3155 		hlen += TCPOLEN_SIGLEN;
3156 #endif
3157 	return (hlen);
3158 }
3159 
3160 /*
3161  * Set connection variables based on the effective MSS.
3162  * We are passed the TCPCB for the actual connection.  If we
3163  * are the server, we are called by the compressed state engine
3164  * when the 3-way handshake is complete.  If we are the client,
3165  * we are called when we receive the SYN,ACK from the server.
3166  *
3167  * NOTE: The t_maxseg value must be initialized in the TCPCB
3168  * before this routine is called!
3169  */
3170 void
3171 tcp_mss_update(tp)
3172 	struct tcpcb *tp;
3173 {
3174 	int mss;
3175 	u_long bufsize;
3176 	struct rtentry *rt;
3177 	struct socket *so;
3178 
3179 	so = tp->t_inpcb->inp_socket;
3180 	mss = tp->t_maxseg;
3181 
3182 	rt = in_pcbrtentry(tp->t_inpcb);
3183 
3184 	if (rt == NULL)
3185 		return;
3186 
3187 	bufsize = so->so_snd.sb_hiwat;
3188 	if (bufsize < mss) {
3189 		mss = bufsize;
3190 		/* Update t_maxseg and t_maxopd */
3191 		tcp_mss(tp, mss);
3192 	} else {
3193 		bufsize = roundup(bufsize, mss);
3194 		if (bufsize > sb_max)
3195 			bufsize = sb_max;
3196 		(void)sbreserve(&so->so_snd, bufsize);
3197 	}
3198 
3199 	bufsize = so->so_rcv.sb_hiwat;
3200 	if (bufsize > mss) {
3201 		bufsize = roundup(bufsize, mss);
3202 		if (bufsize > sb_max)
3203 			bufsize = sb_max;
3204 		(void)sbreserve(&so->so_rcv, bufsize);
3205 	}
3206 
3207 }
3208 
3209 #if defined (TCP_SACK)
3210 /*
3211  * Checks for partial ack.  If partial ack arrives, force the retransmission
3212  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3213  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3214  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3215  */
3216 int
3217 tcp_newreno(tp, th)
3218 	struct tcpcb *tp;
3219 	struct tcphdr *th;
3220 {
3221 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3222 		/*
3223 		 * snd_una has not been updated and the socket send buffer
3224 		 * not yet drained of the acked data, so we have to leave
3225 		 * snd_una as it was to get the correct data offset in
3226 		 * tcp_output().
3227 		 */
3228 		tcp_seq onxt = tp->snd_nxt;
3229 		u_long  ocwnd = tp->snd_cwnd;
3230 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3231 		tp->t_rtttime = 0;
3232 		tp->snd_nxt = th->th_ack;
3233 		/*
3234 		 * Set snd_cwnd to one segment beyond acknowledged offset
3235 		 * (tp->snd_una not yet updated when this function is called)
3236 		 */
3237 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3238 		(void) tcp_output(tp);
3239 		tp->snd_cwnd = ocwnd;
3240 		if (SEQ_GT(onxt, tp->snd_nxt))
3241 			tp->snd_nxt = onxt;
3242 		/*
3243 		 * Partial window deflation.  Relies on fact that tp->snd_una
3244 		 * not updated yet.
3245 		 */
3246 		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3247 		return 1;
3248 	}
3249 	return 0;
3250 }
3251 #endif /* TCP_SACK */
3252 
3253 static int
3254 tcp_mss_adv(struct ifnet *ifp, int af)
3255 {
3256 	int mss = 0;
3257 	int iphlen;
3258 
3259 	switch (af) {
3260 	case AF_INET:
3261 		if (ifp != NULL)
3262 			mss = ifp->if_mtu;
3263 		iphlen = sizeof(struct ip);
3264 		break;
3265 #ifdef INET6
3266 	case AF_INET6:
3267 		if (ifp != NULL)
3268 			mss = IN6_LINKMTU(ifp);
3269 		iphlen = sizeof(struct ip6_hdr);
3270 		break;
3271 #endif
3272 	}
3273 	mss = mss - iphlen - sizeof(struct tcphdr);
3274 	return (max(mss, tcp_mssdflt));
3275 }
3276 
3277 /*
3278  * TCP compressed state engine.  Currently used to hold compressed
3279  * state for SYN_RECEIVED.
3280  */
3281 
3282 u_long	syn_cache_count;
3283 u_int32_t syn_hash1, syn_hash2;
3284 
3285 #define SYN_HASH(sa, sp, dp) \
3286 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3287 				     ((u_int32_t)(sp)))^syn_hash2)))
3288 #ifndef INET6
3289 #define	SYN_HASHALL(hash, src, dst) \
3290 do {									\
3291 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
3292 		((struct sockaddr_in *)(src))->sin_port,		\
3293 		((struct sockaddr_in *)(dst))->sin_port);		\
3294 } while (/*CONSTCOND*/ 0)
3295 #else
3296 #define SYN_HASH6(sa, sp, dp) \
3297 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3298 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3299 	 & 0x7fffffff)
3300 
3301 #define SYN_HASHALL(hash, src, dst) \
3302 do {									\
3303 	switch ((src)->sa_family) {					\
3304 	case AF_INET:							\
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 		break;							\
3309 	case AF_INET6:							\
3310 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
3311 			((struct sockaddr_in6 *)(src))->sin6_port,	\
3312 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
3313 		break;							\
3314 	default:							\
3315 		hash = 0;						\
3316 	}								\
3317 } while (/*CONSTCOND*/0)
3318 #endif /* INET6 */
3319 
3320 #define	SYN_CACHE_RM(sc)						\
3321 do {									\
3322 	(sc)->sc_flags |= SCF_DEAD;					\
3323 	TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket,	\
3324 	    (sc), sc_bucketq);						\
3325 	(sc)->sc_tp = NULL;						\
3326 	LIST_REMOVE((sc), sc_tpq);					\
3327 	tcp_syn_cache[(sc)->sc_bucketidx].sch_length--;			\
3328 	timeout_del(&(sc)->sc_timer);					\
3329 	syn_cache_count--;						\
3330 } while (/*CONSTCOND*/0)
3331 
3332 #define	SYN_CACHE_PUT(sc)						\
3333 do {									\
3334 	if ((sc)->sc_ipopts)						\
3335 		(void) m_free((sc)->sc_ipopts);				\
3336 	if ((sc)->sc_route4.ro_rt != NULL)				\
3337 		RTFREE((sc)->sc_route4.ro_rt);				\
3338 	timeout_set(&(sc)->sc_timer, syn_cache_reaper, (sc));		\
3339 	timeout_add(&(sc)->sc_timer, 0);				\
3340 } while (/*CONSTCOND*/0)
3341 
3342 struct pool syn_cache_pool;
3343 
3344 /*
3345  * We don't estimate RTT with SYNs, so each packet starts with the default
3346  * RTT and each timer step has a fixed timeout value.
3347  */
3348 #define	SYN_CACHE_TIMER_ARM(sc)						\
3349 do {									\
3350 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3351 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3352 	    TCPTV_REXMTMAX);						\
3353 	if (!timeout_initialized(&(sc)->sc_timer))			\
3354 		timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc));	\
3355 	timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \
3356 } while (/*CONSTCOND*/0)
3357 
3358 #define	SYN_CACHE_TIMESTAMP(sc)	tcp_now + (sc)->sc_modulate
3359 
3360 void
3361 syn_cache_init()
3362 {
3363 	int i;
3364 
3365 	/* Initialize the hash buckets. */
3366 	for (i = 0; i < tcp_syn_cache_size; i++)
3367 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3368 
3369 	/* Initialize the syn cache pool. */
3370 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3371 	    "synpl", NULL);
3372 }
3373 
3374 void
3375 syn_cache_insert(sc, tp)
3376 	struct syn_cache *sc;
3377 	struct tcpcb *tp;
3378 {
3379 	struct syn_cache_head *scp;
3380 	struct syn_cache *sc2;
3381 	int s;
3382 
3383 	/*
3384 	 * If there are no entries in the hash table, reinitialize
3385 	 * the hash secrets.
3386 	 */
3387 	if (syn_cache_count == 0) {
3388 		syn_hash1 = arc4random();
3389 		syn_hash2 = arc4random();
3390 	}
3391 
3392 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3393 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3394 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3395 
3396 	/*
3397 	 * Make sure that we don't overflow the per-bucket
3398 	 * limit or the total cache size limit.
3399 	 */
3400 	s = splsoftnet();
3401 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3402 		tcpstat.tcps_sc_bucketoverflow++;
3403 		/*
3404 		 * The bucket is full.  Toss the oldest element in the
3405 		 * bucket.  This will be the first entry in the bucket.
3406 		 */
3407 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3408 #ifdef DIAGNOSTIC
3409 		/*
3410 		 * This should never happen; we should always find an
3411 		 * entry in our bucket.
3412 		 */
3413 		if (sc2 == NULL)
3414 			panic("syn_cache_insert: bucketoverflow: impossible");
3415 #endif
3416 		SYN_CACHE_RM(sc2);
3417 		SYN_CACHE_PUT(sc2);
3418 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3419 		struct syn_cache_head *scp2, *sce;
3420 
3421 		tcpstat.tcps_sc_overflowed++;
3422 		/*
3423 		 * The cache is full.  Toss the oldest entry in the
3424 		 * first non-empty bucket we can find.
3425 		 *
3426 		 * XXX We would really like to toss the oldest
3427 		 * entry in the cache, but we hope that this
3428 		 * condition doesn't happen very often.
3429 		 */
3430 		scp2 = scp;
3431 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3432 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3433 			for (++scp2; scp2 != scp; scp2++) {
3434 				if (scp2 >= sce)
3435 					scp2 = &tcp_syn_cache[0];
3436 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3437 					break;
3438 			}
3439 #ifdef DIAGNOSTIC
3440 			/*
3441 			 * This should never happen; we should always find a
3442 			 * non-empty bucket.
3443 			 */
3444 			if (scp2 == scp)
3445 				panic("syn_cache_insert: cacheoverflow: "
3446 				    "impossible");
3447 #endif
3448 		}
3449 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3450 		SYN_CACHE_RM(sc2);
3451 		SYN_CACHE_PUT(sc2);
3452 	}
3453 
3454 	/*
3455 	 * Initialize the entry's timer.
3456 	 */
3457 	sc->sc_rxttot = 0;
3458 	sc->sc_rxtshift = 0;
3459 	SYN_CACHE_TIMER_ARM(sc);
3460 
3461 	/* Link it from tcpcb entry */
3462 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3463 
3464 	/* Put it into the bucket. */
3465 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3466 	scp->sch_length++;
3467 	syn_cache_count++;
3468 
3469 	tcpstat.tcps_sc_added++;
3470 	splx(s);
3471 }
3472 
3473 /*
3474  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3475  * If we have retransmitted an entry the maximum number of times, expire
3476  * that entry.
3477  */
3478 void
3479 syn_cache_timer(void *arg)
3480 {
3481 	struct syn_cache *sc = arg;
3482 	int s;
3483 
3484 	s = splsoftnet();
3485 	if (sc->sc_flags & SCF_DEAD) {
3486 		splx(s);
3487 		return;
3488 	}
3489 
3490 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3491 		/* Drop it -- too many retransmissions. */
3492 		goto dropit;
3493 	}
3494 
3495 	/*
3496 	 * Compute the total amount of time this entry has
3497 	 * been on a queue.  If this entry has been on longer
3498 	 * than the keep alive timer would allow, expire it.
3499 	 */
3500 	sc->sc_rxttot += sc->sc_rxtcur;
3501 	if (sc->sc_rxttot >= tcptv_keep_init)
3502 		goto dropit;
3503 
3504 	tcpstat.tcps_sc_retransmitted++;
3505 	(void) syn_cache_respond(sc, NULL);
3506 
3507 	/* Advance the timer back-off. */
3508 	sc->sc_rxtshift++;
3509 	SYN_CACHE_TIMER_ARM(sc);
3510 
3511 	splx(s);
3512 	return;
3513 
3514  dropit:
3515 	tcpstat.tcps_sc_timed_out++;
3516 	SYN_CACHE_RM(sc);
3517 	SYN_CACHE_PUT(sc);
3518 	splx(s);
3519 }
3520 
3521 void
3522 syn_cache_reaper(void *arg)
3523 {
3524 	struct syn_cache *sc = arg;
3525 	int s;
3526 
3527 	s = splsoftnet();
3528 	pool_put(&syn_cache_pool, (sc));
3529 	splx(s);
3530 	return;
3531 }
3532 
3533 /*
3534  * Remove syn cache created by the specified tcb entry,
3535  * because this does not make sense to keep them
3536  * (if there's no tcb entry, syn cache entry will never be used)
3537  */
3538 void
3539 syn_cache_cleanup(tp)
3540 	struct tcpcb *tp;
3541 {
3542 	struct syn_cache *sc, *nsc;
3543 	int s;
3544 
3545 	s = splsoftnet();
3546 
3547 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3548 		nsc = LIST_NEXT(sc, sc_tpq);
3549 
3550 #ifdef DIAGNOSTIC
3551 		if (sc->sc_tp != tp)
3552 			panic("invalid sc_tp in syn_cache_cleanup");
3553 #endif
3554 		SYN_CACHE_RM(sc);
3555 		SYN_CACHE_PUT(sc);
3556 	}
3557 	/* just for safety */
3558 	LIST_INIT(&tp->t_sc);
3559 
3560 	splx(s);
3561 }
3562 
3563 /*
3564  * Find an entry in the syn cache.
3565  */
3566 struct syn_cache *
3567 syn_cache_lookup(src, dst, headp)
3568 	struct sockaddr *src;
3569 	struct sockaddr *dst;
3570 	struct syn_cache_head **headp;
3571 {
3572 	struct syn_cache *sc;
3573 	struct syn_cache_head *scp;
3574 	u_int32_t hash;
3575 	int s;
3576 
3577 	SYN_HASHALL(hash, src, dst);
3578 
3579 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3580 	*headp = scp;
3581 	s = splsoftnet();
3582 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3583 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3584 		if (sc->sc_hash != hash)
3585 			continue;
3586 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3587 		    !bcmp(&sc->sc_dst, dst, dst->sa_len)) {
3588 			splx(s);
3589 			return (sc);
3590 		}
3591 	}
3592 	splx(s);
3593 	return (NULL);
3594 }
3595 
3596 /*
3597  * This function gets called when we receive an ACK for a
3598  * socket in the LISTEN state.  We look up the connection
3599  * in the syn cache, and if its there, we pull it out of
3600  * the cache and turn it into a full-blown connection in
3601  * the SYN-RECEIVED state.
3602  *
3603  * The return values may not be immediately obvious, and their effects
3604  * can be subtle, so here they are:
3605  *
3606  *	NULL	SYN was not found in cache; caller should drop the
3607  *		packet and send an RST.
3608  *
3609  *	-1	We were unable to create the new connection, and are
3610  *		aborting it.  An ACK,RST is being sent to the peer
3611  *		(unless we got screwey sequence numbners; see below),
3612  *		because the 3-way handshake has been completed.  Caller
3613  *		should not free the mbuf, since we may be using it.  If
3614  *		we are not, we will free it.
3615  *
3616  *	Otherwise, the return value is a pointer to the new socket
3617  *	associated with the connection.
3618  */
3619 struct socket *
3620 syn_cache_get(src, dst, th, hlen, tlen, so, m)
3621 	struct sockaddr *src;
3622 	struct sockaddr *dst;
3623 	struct tcphdr *th;
3624 	unsigned int hlen, tlen;
3625 	struct socket *so;
3626 	struct mbuf *m;
3627 {
3628 	struct syn_cache *sc;
3629 	struct syn_cache_head *scp;
3630 	struct inpcb *inp = NULL;
3631 	struct tcpcb *tp = 0;
3632 	struct mbuf *am;
3633 	int s;
3634 	struct socket *oso;
3635 
3636 	s = splsoftnet();
3637 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3638 		splx(s);
3639 		return (NULL);
3640 	}
3641 
3642 	/*
3643 	 * Verify the sequence and ack numbers.  Try getting the correct
3644 	 * response again.
3645 	 */
3646 	if ((th->th_ack != sc->sc_iss + 1) ||
3647 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3648 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3649 		(void) syn_cache_respond(sc, m);
3650 		splx(s);
3651 		return ((struct socket *)(-1));
3652 	}
3653 
3654 	/* Remove this cache entry */
3655 	SYN_CACHE_RM(sc);
3656 	splx(s);
3657 
3658 	/*
3659 	 * Ok, create the full blown connection, and set things up
3660 	 * as they would have been set up if we had created the
3661 	 * connection when the SYN arrived.  If we can't create
3662 	 * the connection, abort it.
3663 	 */
3664 	oso = so;
3665 	so = sonewconn(so, SS_ISCONNECTED);
3666 	if (so == NULL)
3667 		goto resetandabort;
3668 
3669 	inp = sotoinpcb(oso);
3670 #ifdef IPSEC
3671 	/*
3672 	 * We need to copy the required security levels
3673 	 * from the old pcb. Ditto for any other
3674 	 * IPsec-related information.
3675 	 */
3676 	{
3677 	  struct inpcb *newinp = (struct inpcb *)so->so_pcb;
3678 	  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
3679 		sizeof(inp->inp_seclevel));
3680 	  newinp->inp_secrequire = inp->inp_secrequire;
3681 	  if (inp->inp_ipo != NULL) {
3682 		  newinp->inp_ipo = inp->inp_ipo;
3683 		  inp->inp_ipo->ipo_ref_count++;
3684 	  }
3685 	  if (inp->inp_ipsec_remotecred != NULL) {
3686 		  newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred;
3687 		  inp->inp_ipsec_remotecred->ref_count++;
3688 	  }
3689 	  if (inp->inp_ipsec_remoteauth != NULL) {
3690 		  newinp->inp_ipsec_remoteauth
3691 		      = inp->inp_ipsec_remoteauth;
3692 		  inp->inp_ipsec_remoteauth->ref_count++;
3693 	  }
3694 	}
3695 #endif /* IPSEC */
3696 #ifdef INET6
3697 	/*
3698 	 * inp still has the OLD in_pcb stuff, set the
3699 	 * v6-related flags on the new guy, too.
3700 	 */
3701 	{
3702 	  int flags = inp->inp_flags;
3703 	  struct inpcb *oldinpcb = inp;
3704 
3705 	  inp = (struct inpcb *)so->so_pcb;
3706 	  inp->inp_flags |= (flags & INP_IPV6);
3707 	  if ((inp->inp_flags & INP_IPV6) != 0) {
3708 	    inp->inp_ipv6.ip6_hlim =
3709 	      oldinpcb->inp_ipv6.ip6_hlim;
3710 	  }
3711 	}
3712 #else /* INET6 */
3713 	inp = (struct inpcb *)so->so_pcb;
3714 #endif /* INET6 */
3715 
3716 	inp->inp_lport = th->th_dport;
3717 	switch (src->sa_family) {
3718 #ifdef INET6
3719 	case AF_INET6:
3720 		inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr;
3721 		break;
3722 #endif /* INET6 */
3723 	case AF_INET:
3724 
3725 		inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3726 		inp->inp_options = ip_srcroute();
3727 		if (inp->inp_options == NULL) {
3728 			inp->inp_options = sc->sc_ipopts;
3729 			sc->sc_ipopts = NULL;
3730 		}
3731 		break;
3732 	}
3733 	in_pcbrehash(inp);
3734 
3735 	/*
3736 	 * Give the new socket our cached route reference.
3737 	 */
3738 	if (src->sa_family == AF_INET)
3739 		inp->inp_route = sc->sc_route4;         /* struct assignment */
3740 #ifdef INET6
3741 	else
3742 		inp->inp_route6 = sc->sc_route6;
3743 #endif
3744 	sc->sc_route4.ro_rt = NULL;
3745 
3746 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3747 	if (am == NULL)
3748 		goto resetandabort;
3749 	am->m_len = src->sa_len;
3750 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3751 
3752 	switch (src->sa_family) {
3753 	case AF_INET:
3754 		/* drop IPv4 packet to AF_INET6 socket */
3755 		if (inp->inp_flags & INP_IPV6) {
3756 			(void) m_free(am);
3757 			goto resetandabort;
3758 		}
3759 		if (in_pcbconnect(inp, am)) {
3760 			(void) m_free(am);
3761 			goto resetandabort;
3762 		}
3763 		break;
3764 #ifdef INET6
3765 	case AF_INET6:
3766 		if (in6_pcbconnect(inp, am)) {
3767 			(void) m_free(am);
3768 			goto resetandabort;
3769 		}
3770 		break;
3771 #endif
3772 	}
3773 	(void) m_free(am);
3774 
3775 	tp = intotcpcb(inp);
3776 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3777 	if (sc->sc_request_r_scale != 15) {
3778 		tp->requested_s_scale = sc->sc_requested_s_scale;
3779 		tp->request_r_scale = sc->sc_request_r_scale;
3780 		tp->snd_scale = sc->sc_requested_s_scale;
3781 		tp->rcv_scale = sc->sc_request_r_scale;
3782 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3783 	}
3784 	if (sc->sc_flags & SCF_TIMESTAMP)
3785 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3786 
3787 	tp->t_template = tcp_template(tp);
3788 	if (tp->t_template == 0) {
3789 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3790 		so = NULL;
3791 		m_freem(m);
3792 		goto abort;
3793 	}
3794 #ifdef TCP_SACK
3795 	tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT;
3796 #endif
3797 
3798 	tp->ts_modulate = sc->sc_modulate;
3799 	tp->iss = sc->sc_iss;
3800 	tp->irs = sc->sc_irs;
3801 	tcp_sendseqinit(tp);
3802 #if defined (TCP_SACK) || defined(TCP_ECN)
3803 	tp->snd_last = tp->snd_una;
3804 #endif /* TCP_SACK */
3805 #if defined(TCP_SACK) && defined(TCP_FACK)
3806 	tp->snd_fack = tp->snd_una;
3807 	tp->retran_data = 0;
3808 	tp->snd_awnd = 0;
3809 #endif /* TCP_FACK */
3810 #ifdef TCP_ECN
3811 	if (sc->sc_flags & SCF_ECN_PERMIT) {
3812 		tp->t_flags |= TF_ECN_PERMIT;
3813 		tcpstat.tcps_ecn_accepts++;
3814 	}
3815 #endif
3816 #ifdef TCP_SACK
3817 	if (sc->sc_flags & SCF_SACK_PERMIT)
3818 		tp->t_flags |= TF_SACK_PERMIT;
3819 #endif
3820 #ifdef TCP_SIGNATURE
3821 	if (sc->sc_flags & SCF_SIGNATURE)
3822 		tp->t_flags |= TF_SIGNATURE;
3823 #endif
3824 	tcp_rcvseqinit(tp);
3825 	tp->t_state = TCPS_SYN_RECEIVED;
3826 	tp->t_rcvtime = tcp_now;
3827 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
3828 	tcpstat.tcps_accepts++;
3829 
3830 	tcp_mss(tp, sc->sc_peermaxseg);	 /* sets t_maxseg */
3831 	if (sc->sc_peermaxseg)
3832 		tcp_mss_update(tp);
3833 	/* Reset initial window to 1 segment for retransmit */
3834 	if (sc->sc_rxtshift > 0)
3835 		tp->snd_cwnd = tp->t_maxseg;
3836 	tp->snd_wl1 = sc->sc_irs;
3837 	tp->rcv_up = sc->sc_irs + 1;
3838 
3839 	/*
3840 	 * This is what whould have happened in tcp_output() when
3841 	 * the SYN,ACK was sent.
3842 	 */
3843 	tp->snd_up = tp->snd_una;
3844 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3845 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3846 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3847 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3848 	tp->last_ack_sent = tp->rcv_nxt;
3849 
3850 	tcpstat.tcps_sc_completed++;
3851 	SYN_CACHE_PUT(sc);
3852 	return (so);
3853 
3854 resetandabort:
3855 	tcp_respond(NULL, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack, TH_RST);
3856 abort:
3857 	if (so != NULL)
3858 		(void) soabort(so);
3859 	SYN_CACHE_PUT(sc);
3860 	tcpstat.tcps_sc_aborted++;
3861 	return ((struct socket *)(-1));
3862 }
3863 
3864 /*
3865  * This function is called when we get a RST for a
3866  * non-existent connection, so that we can see if the
3867  * connection is in the syn cache.  If it is, zap it.
3868  */
3869 
3870 void
3871 syn_cache_reset(src, dst, th)
3872 	struct sockaddr *src;
3873 	struct sockaddr *dst;
3874 	struct tcphdr *th;
3875 {
3876 	struct syn_cache *sc;
3877 	struct syn_cache_head *scp;
3878 	int s = splsoftnet();
3879 
3880 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3881 		splx(s);
3882 		return;
3883 	}
3884 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3885 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3886 		splx(s);
3887 		return;
3888 	}
3889 	SYN_CACHE_RM(sc);
3890 	splx(s);
3891 	tcpstat.tcps_sc_reset++;
3892 	SYN_CACHE_PUT(sc);
3893 }
3894 
3895 void
3896 syn_cache_unreach(src, dst, th)
3897 	struct sockaddr *src;
3898 	struct sockaddr *dst;
3899 	struct tcphdr *th;
3900 {
3901 	struct syn_cache *sc;
3902 	struct syn_cache_head *scp;
3903 	int s;
3904 
3905 	s = splsoftnet();
3906 	if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) {
3907 		splx(s);
3908 		return;
3909 	}
3910 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3911 	if (ntohl (th->th_seq) != sc->sc_iss) {
3912 		splx(s);
3913 		return;
3914 	}
3915 
3916 	/*
3917 	 * If we've retransmitted 3 times and this is our second error,
3918 	 * we remove the entry.  Otherwise, we allow it to continue on.
3919 	 * This prevents us from incorrectly nuking an entry during a
3920 	 * spurious network outage.
3921 	 *
3922 	 * See tcp_notify().
3923 	 */
3924 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3925 		sc->sc_flags |= SCF_UNREACH;
3926 		splx(s);
3927 		return;
3928 	}
3929 
3930 	SYN_CACHE_RM(sc);
3931 	splx(s);
3932 	tcpstat.tcps_sc_unreach++;
3933 	SYN_CACHE_PUT(sc);
3934 }
3935 
3936 /*
3937  * Given a LISTEN socket and an inbound SYN request, add
3938  * this to the syn cache, and send back a segment:
3939  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3940  * to the source.
3941  *
3942  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3943  * Doing so would require that we hold onto the data and deliver it
3944  * to the application.  However, if we are the target of a SYN-flood
3945  * DoS attack, an attacker could send data which would eventually
3946  * consume all available buffer space if it were ACKed.  By not ACKing
3947  * the data, we avoid this DoS scenario.
3948  */
3949 
3950 int
3951 syn_cache_add(src, dst, th, iphlen, so, m, optp, optlen, oi)
3952 	struct sockaddr *src;
3953 	struct sockaddr *dst;
3954 	struct tcphdr *th;
3955 	unsigned int iphlen;
3956 	struct socket *so;
3957 	struct mbuf *m;
3958 	u_char *optp;
3959 	int optlen;
3960 	struct tcp_opt_info *oi;
3961 {
3962 	struct tcpcb tb, *tp;
3963 	long win;
3964 	struct syn_cache *sc;
3965 	struct syn_cache_head *scp;
3966 	struct mbuf *ipopts;
3967 
3968 	tp = sototcpcb(so);
3969 
3970 	/*
3971 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3972 	 *
3973 	 * Note this check is performed in tcp_input() very early on.
3974 	 */
3975 
3976 	/*
3977 	 * Initialize some local state.
3978 	 */
3979 	win = sbspace(&so->so_rcv);
3980 	if (win > TCP_MAXWIN)
3981 		win = TCP_MAXWIN;
3982 
3983 #ifdef TCP_SIGNATURE
3984 	if (optp || (tp->t_flags & TF_SIGNATURE)) {
3985 #else
3986 	if (optp) {
3987 #endif
3988 		tb.pf = tp->pf;
3989 #ifdef TCP_SACK
3990 		tb.sack_enable = tp->sack_enable;
3991 #endif
3992 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3993 #ifdef TCP_SIGNATURE
3994 		if (tp->t_flags & TF_SIGNATURE)
3995 			tb.t_flags |= TF_SIGNATURE;
3996 #endif
3997 		tb.t_state = TCPS_LISTEN;
3998 		if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi))
3999 			return (0);
4000 	} else
4001 		tb.t_flags = 0;
4002 
4003 	switch (src->sa_family) {
4004 #ifdef INET
4005 	case AF_INET:
4006 		/*
4007 		 * Remember the IP options, if any.
4008 		 */
4009 		ipopts = ip_srcroute();
4010 		break;
4011 #endif
4012 	default:
4013 		ipopts = NULL;
4014 	}
4015 
4016 	/*
4017 	 * See if we already have an entry for this connection.
4018 	 * If we do, resend the SYN,ACK.  We do not count this
4019 	 * as a retransmission (XXX though maybe we should).
4020 	 */
4021 	if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) {
4022 		tcpstat.tcps_sc_dupesyn++;
4023 		if (ipopts) {
4024 			/*
4025 			 * If we were remembering a previous source route,
4026 			 * forget it and use the new one we've been given.
4027 			 */
4028 			if (sc->sc_ipopts)
4029 				(void) m_free(sc->sc_ipopts);
4030 			sc->sc_ipopts = ipopts;
4031 		}
4032 		sc->sc_timestamp = tb.ts_recent;
4033 		if (syn_cache_respond(sc, m) == 0) {
4034 			tcpstat.tcps_sndacks++;
4035 			tcpstat.tcps_sndtotal++;
4036 		}
4037 		return (1);
4038 	}
4039 
4040 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
4041 	if (sc == NULL) {
4042 		if (ipopts)
4043 			(void) m_free(ipopts);
4044 		return (0);
4045 	}
4046 
4047 	/*
4048 	 * Fill in the cache, and put the necessary IP and TCP
4049 	 * options into the reply.
4050 	 */
4051 	bzero(sc, sizeof(struct syn_cache));
4052 	bzero(&sc->sc_timer, sizeof(sc->sc_timer));
4053 	bcopy(src, &sc->sc_src, src->sa_len);
4054 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4055 	sc->sc_flags = 0;
4056 	sc->sc_ipopts = ipopts;
4057 	sc->sc_irs = th->th_seq;
4058 
4059 #ifdef TCP_COMPAT_42
4060 	tcp_iss += TCP_ISSINCR/2;
4061 	sc->sc_iss = tcp_iss;
4062 #else
4063 	sc->sc_iss = tcp_rndiss_next();
4064 #endif
4065 	sc->sc_peermaxseg = oi->maxseg;
4066 	sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ?
4067 	    m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family);
4068 	sc->sc_win = win;
4069 	sc->sc_timestamp = tb.ts_recent;
4070 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4071 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
4072 		sc->sc_flags |= SCF_TIMESTAMP;
4073 		sc->sc_modulate = arc4random();
4074 	}
4075 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4076 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4077 		sc->sc_requested_s_scale = tb.requested_s_scale;
4078 		sc->sc_request_r_scale = 0;
4079 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4080 		    TCP_MAXWIN << sc->sc_request_r_scale <
4081 		    so->so_rcv.sb_hiwat)
4082 			sc->sc_request_r_scale++;
4083 	} else {
4084 		sc->sc_requested_s_scale = 15;
4085 		sc->sc_request_r_scale = 15;
4086 	}
4087 #ifdef TCP_ECN
4088 	/*
4089 	 * if both ECE and CWR flag bits are set, peer is ECN capable.
4090 	 */
4091 	if (tcp_do_ecn &&
4092 	    (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR))
4093 		sc->sc_flags |= SCF_ECN_PERMIT;
4094 #endif
4095 #ifdef TCP_SACK
4096 	/*
4097 	 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option
4098 	 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT).
4099 	 */
4100 	if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT))
4101 		sc->sc_flags |= SCF_SACK_PERMIT;
4102 #endif
4103 #ifdef TCP_SIGNATURE
4104 	if (tb.t_flags & TF_SIGNATURE)
4105 		sc->sc_flags |= SCF_SIGNATURE;
4106 #endif
4107 	sc->sc_tp = tp;
4108 	if (syn_cache_respond(sc, m) == 0) {
4109 		syn_cache_insert(sc, tp);
4110 		tcpstat.tcps_sndacks++;
4111 		tcpstat.tcps_sndtotal++;
4112 	} else {
4113 		SYN_CACHE_PUT(sc);
4114 		tcpstat.tcps_sc_dropped++;
4115 	}
4116 	return (1);
4117 }
4118 
4119 int
4120 syn_cache_respond(sc, m)
4121 	struct syn_cache *sc;
4122 	struct mbuf *m;
4123 {
4124 	struct route *ro;
4125 	u_int8_t *optp;
4126 	int optlen, error;
4127 	u_int16_t tlen;
4128 	struct ip *ip = NULL;
4129 #ifdef INET6
4130 	struct ip6_hdr *ip6 = NULL;
4131 #endif
4132 	struct tcphdr *th;
4133 	u_int hlen;
4134 	struct inpcb *inp;
4135 
4136 	switch (sc->sc_src.sa.sa_family) {
4137 	case AF_INET:
4138 		hlen = sizeof(struct ip);
4139 		ro = &sc->sc_route4;
4140 		break;
4141 #ifdef INET6
4142 	case AF_INET6:
4143 		hlen = sizeof(struct ip6_hdr);
4144 		ro = (struct route *)&sc->sc_route6;
4145 		break;
4146 #endif
4147 	default:
4148 		if (m)
4149 			m_freem(m);
4150 		return (EAFNOSUPPORT);
4151 	}
4152 
4153 	/* Compute the size of the TCP options. */
4154 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4155 #ifdef TCP_SACK
4156 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) +
4157 #endif
4158 #ifdef TCP_SIGNATURE
4159 	    ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) +
4160 #endif
4161 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4162 
4163 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4164 
4165 	/*
4166 	 * Create the IP+TCP header from scratch.
4167 	 */
4168 	if (m)
4169 		m_freem(m);
4170 #ifdef DIAGNOSTIC
4171 	if (max_linkhdr + tlen > MCLBYTES)
4172 		return (ENOBUFS);
4173 #endif
4174 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4175 	if (m && max_linkhdr + tlen > MHLEN) {
4176 		MCLGET(m, M_DONTWAIT);
4177 		if ((m->m_flags & M_EXT) == 0) {
4178 			m_freem(m);
4179 			m = NULL;
4180 		}
4181 	}
4182 	if (m == NULL)
4183 		return (ENOBUFS);
4184 
4185 	/* Fixup the mbuf. */
4186 	m->m_data += max_linkhdr;
4187 	m->m_len = m->m_pkthdr.len = tlen;
4188 	m->m_pkthdr.rcvif = NULL;
4189 	memset(mtod(m, u_char *), 0, tlen);
4190 
4191 	switch (sc->sc_src.sa.sa_family) {
4192 	case AF_INET:
4193 		ip = mtod(m, struct ip *);
4194 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4195 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4196 		ip->ip_p = IPPROTO_TCP;
4197 		th = (struct tcphdr *)(ip + 1);
4198 		th->th_dport = sc->sc_src.sin.sin_port;
4199 		th->th_sport = sc->sc_dst.sin.sin_port;
4200 		break;
4201 #ifdef INET6
4202 	case AF_INET6:
4203 		ip6 = mtod(m, struct ip6_hdr *);
4204 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4205 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4206 		ip6->ip6_nxt = IPPROTO_TCP;
4207 		/* ip6_plen will be updated in ip6_output() */
4208 		th = (struct tcphdr *)(ip6 + 1);
4209 		th->th_dport = sc->sc_src.sin6.sin6_port;
4210 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4211 		break;
4212 #endif
4213 	default:
4214 		th = NULL;
4215 	}
4216 
4217 	th->th_seq = htonl(sc->sc_iss);
4218 	th->th_ack = htonl(sc->sc_irs + 1);
4219 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4220 	th->th_flags = TH_SYN|TH_ACK;
4221 #ifdef TCP_ECN
4222 	/* Set ECE for SYN-ACK if peer supports ECN. */
4223 	if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT))
4224 		th->th_flags |= TH_ECE;
4225 #endif
4226 	th->th_win = htons(sc->sc_win);
4227 	/* th_sum already 0 */
4228 	/* th_urp already 0 */
4229 
4230 	/* Tack on the TCP options. */
4231 	optp = (u_int8_t *)(th + 1);
4232 	*optp++ = TCPOPT_MAXSEG;
4233 	*optp++ = 4;
4234 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4235 	*optp++ = sc->sc_ourmaxseg & 0xff;
4236 
4237 #ifdef TCP_SACK
4238 	/* Include SACK_PERMIT_HDR option if peer has already done so. */
4239 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4240 		*((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR);
4241 		optp += 4;
4242 	}
4243 #endif
4244 
4245 	if (sc->sc_request_r_scale != 15) {
4246 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4247 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4248 		    sc->sc_request_r_scale);
4249 		optp += 4;
4250 	}
4251 
4252 	if (sc->sc_flags & SCF_TIMESTAMP) {
4253 		u_int32_t *lp = (u_int32_t *)(optp);
4254 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4255 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4256 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4257 		*lp   = htonl(sc->sc_timestamp);
4258 		optp += TCPOLEN_TSTAMP_APPA;
4259 	}
4260 
4261 #ifdef TCP_SIGNATURE
4262 	if (sc->sc_flags & SCF_SIGNATURE) {
4263 		union sockaddr_union src, dst;
4264 		struct tdb *tdb;
4265 
4266 		bzero(&src, sizeof(union sockaddr_union));
4267 		bzero(&dst, sizeof(union sockaddr_union));
4268 		src.sa.sa_len = sc->sc_src.sa.sa_len;
4269 		src.sa.sa_family = sc->sc_src.sa.sa_family;
4270 		dst.sa.sa_len = sc->sc_dst.sa.sa_len;
4271 		dst.sa.sa_family = sc->sc_dst.sa.sa_family;
4272 
4273 		switch (sc->sc_src.sa.sa_family) {
4274 		case 0:	/*default to PF_INET*/
4275 #ifdef INET
4276 		case AF_INET:
4277 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
4278 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
4279 			break;
4280 #endif /* INET */
4281 #ifdef INET6
4282 		case AF_INET6:
4283 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
4284 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
4285 			break;
4286 #endif /* INET6 */
4287 		}
4288 
4289 		tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP);
4290 		if (tdb == NULL) {
4291 			if (m)
4292 				m_freem(m);
4293 			return (EPERM);
4294 		}
4295 
4296 		/* Send signature option */
4297 		*(optp++) = TCPOPT_SIGNATURE;
4298 		*(optp++) = TCPOLEN_SIGNATURE;
4299 
4300 		if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th,
4301 		    hlen, 0, optp) < 0) {
4302 			if (m)
4303 				m_freem(m);
4304 			return (EINVAL);
4305 		}
4306 		optp += 16;
4307 
4308 		/* Pad options list to the next 32 bit boundary and
4309 		 * terminate it.
4310 		 */
4311 		*optp++ = TCPOPT_NOP;
4312 		*optp++ = TCPOPT_EOL;
4313 	}
4314 #endif /* TCP_SIGNATURE */
4315 
4316 	/* Compute the packet's checksum. */
4317 	switch (sc->sc_src.sa.sa_family) {
4318 	case AF_INET:
4319 		ip->ip_len = htons(tlen - hlen);
4320 		th->th_sum = 0;
4321 		th->th_sum = in_cksum(m, tlen);
4322 		break;
4323 #ifdef INET6
4324 	case AF_INET6:
4325 		ip6->ip6_plen = htons(tlen - hlen);
4326 		th->th_sum = 0;
4327 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4328 		break;
4329 #endif
4330 	}
4331 
4332 	/* use IPsec policy and ttl from listening socket, on SYN ACK */
4333 	inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL;
4334 
4335 	/*
4336 	 * Fill in some straggling IP bits.  Note the stack expects
4337 	 * ip_len to be in host order, for convenience.
4338 	 */
4339 	switch (sc->sc_src.sa.sa_family) {
4340 #ifdef INET
4341 	case AF_INET:
4342 		ip->ip_len = htons(tlen);
4343 		ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl;
4344 		/* XXX tos? */
4345 		break;
4346 #endif
4347 #ifdef INET6
4348 	case AF_INET6:
4349 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4350 		ip6->ip6_vfc |= IPV6_VERSION;
4351 		ip6->ip6_plen = htons(tlen - hlen);
4352 		/* ip6_hlim will be initialized afterwards */
4353 		/* leave flowlabel = 0, it is legal and require no state mgmt */
4354 		break;
4355 #endif
4356 	}
4357 
4358 	switch (sc->sc_src.sa.sa_family) {
4359 #ifdef INET
4360 	case AF_INET:
4361 		error = ip_output(m, sc->sc_ipopts, ro,
4362 		    (ip_mtudisc ? IP_MTUDISC : 0),
4363 		    (struct ip_moptions *)NULL, inp);
4364 		break;
4365 #endif
4366 #ifdef INET6
4367 	case AF_INET6:
4368 		ip6->ip6_hlim = in6_selecthlim(NULL,
4369 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4370 
4371 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4372 			(struct ip6_moptions *)0, NULL);
4373 		break;
4374 #endif
4375 	default:
4376 		error = EAFNOSUPPORT;
4377 		break;
4378 	}
4379 	return (error);
4380 }
4381