xref: /openbsd-src/sys/netinet/tcp_input.c (revision aa5e9e10509ffd51558f081f01cd78bfa3c4f2a5)
1 /*	$OpenBSD: tcp_input.c,v 1.262 2013/06/03 16:57:05 bluhm 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 "pf.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/mbuf.h>
76 #include <sys/protosw.h>
77 #include <sys/socket.h>
78 #include <sys/socketvar.h>
79 #include <sys/timeout.h>
80 #include <sys/kernel.h>
81 #include <sys/pool.h>
82 
83 #include <dev/rndvar.h>
84 
85 #include <net/if.h>
86 #include <net/route.h>
87 
88 #include <netinet/in.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/ip.h>
91 #include <netinet/in_pcb.h>
92 #include <netinet/ip_var.h>
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcpip.h>
99 #include <netinet/tcp_debug.h>
100 
101 #if NPF > 0
102 #include <net/pfvar.h>
103 #endif
104 
105 struct	tcpiphdr tcp_saveti;
106 
107 int tcp_mss_adv(struct ifnet *, int);
108 int tcp_flush_queue(struct tcpcb *);
109 
110 #ifdef INET6
111 #include <netinet6/in6_var.h>
112 #include <netinet6/nd6.h>
113 
114 struct  tcpipv6hdr tcp_saveti6;
115 
116 /* for the packet header length in the mbuf */
117 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
118 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
119 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
120 #endif /* INET6 */
121 
122 int	tcprexmtthresh = 3;
123 int	tcptv_keep_init = TCPTV_KEEP_INIT;
124 
125 int tcp_rst_ppslim = 100;		/* 100pps */
126 int tcp_rst_ppslim_count = 0;
127 struct timeval tcp_rst_ppslim_last;
128 
129 int tcp_ackdrop_ppslim = 100;		/* 100pps */
130 int tcp_ackdrop_ppslim_count = 0;
131 struct timeval tcp_ackdrop_ppslim_last;
132 
133 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
134 
135 /* for modulo comparisons of timestamps */
136 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
137 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
138 
139 /* for TCP SACK comparisons */
140 #define	SEQ_MIN(a,b)	(SEQ_LT(a,b) ? (a) : (b))
141 #define	SEQ_MAX(a,b)	(SEQ_GT(a,b) ? (a) : (b))
142 
143 /*
144  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
145  */
146 #ifdef INET6
147 #define ND6_HINT(tp) \
148 do { \
149 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
150 	    tp->t_inpcb->inp_route6.ro_rt) { \
151 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \
152 	} \
153 } while (0)
154 #else
155 #define ND6_HINT(tp)
156 #endif
157 
158 #ifdef TCP_ECN
159 /*
160  * ECN (Explicit Congestion Notification) support based on RFC3168
161  * implementation note:
162  *   snd_last is used to track a recovery phase.
163  *   when cwnd is reduced, snd_last is set to snd_max.
164  *   while snd_last > snd_una, the sender is in a recovery phase and
165  *   its cwnd should not be reduced again.
166  *   snd_last follows snd_una when not in a recovery phase.
167  */
168 #endif
169 
170 /*
171  * Macro to compute ACK transmission behavior.  Delay the ACK unless
172  * we have already delayed an ACK (must send an ACK every two segments).
173  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
174  * option is enabled or when the packet is coming from a loopback
175  * interface.
176  */
177 #define	TCP_SETUP_ACK(tp, tiflags, m) \
178 do { \
179 	if ((tp)->t_flags & TF_DELACK || \
180 	    (tcp_ack_on_push && (tiflags) & TH_PUSH) || \
181 	    (m && (m->m_flags & M_PKTHDR) && m->m_pkthdr.rcvif && \
182 	    (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK))) \
183 		tp->t_flags |= TF_ACKNOW; \
184 	else \
185 		TCP_SET_DELACK(tp); \
186 } while (0)
187 
188 void syn_cache_put(struct syn_cache *);
189 void syn_cache_rm(struct syn_cache *);
190 
191 /*
192  * Insert segment ti into reassembly queue of tcp with
193  * control block tp.  Return TH_FIN if reassembly now includes
194  * a segment with FIN.  The macro form does the common case inline
195  * (segment is the next to be received on an established connection,
196  * and the queue is empty), avoiding linkage into and removal
197  * from the queue and repetition of various conversions.
198  * Set DELACK for segments received in order, but ack immediately
199  * when segments are out of order (so fast retransmit can work).
200  */
201 
202 int
203 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen)
204 {
205 	struct tcpqent *p, *q, *nq, *tiqe;
206 
207 	/*
208 	 * Allocate a new queue entry, before we throw away any data.
209 	 * If we can't, just drop the packet.  XXX
210 	 */
211 	tiqe = pool_get(&tcpqe_pool, PR_NOWAIT);
212 	if (tiqe == NULL) {
213 		tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead);
214 		if (tiqe != NULL && th->th_seq == tp->rcv_nxt) {
215 			/* Reuse last entry since new segment fills a hole */
216 			m_freem(tiqe->tcpqe_m);
217 			TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q);
218 		}
219 		if (tiqe == NULL || th->th_seq != tp->rcv_nxt) {
220 			/* Flush segment queue for this connection */
221 			tcp_freeq(tp);
222 			tcpstat.tcps_rcvmemdrop++;
223 			m_freem(m);
224 			return (0);
225 		}
226 	}
227 
228 	/*
229 	 * Find a segment which begins after this one does.
230 	 */
231 	for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL;
232 	    p = q, q = TAILQ_NEXT(q, tcpqe_q))
233 		if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq))
234 			break;
235 
236 	/*
237 	 * If there is a preceding segment, it may provide some of
238 	 * our data already.  If so, drop the data from the incoming
239 	 * segment.  If it provides all of our data, drop us.
240 	 */
241 	if (p != NULL) {
242 		struct tcphdr *phdr = p->tcpqe_tcp;
243 		int i;
244 
245 		/* conversion to int (in i) handles seq wraparound */
246 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
247 		if (i > 0) {
248 		        if (i >= *tlen) {
249 				tcpstat.tcps_rcvduppack++;
250 				tcpstat.tcps_rcvdupbyte += *tlen;
251 				m_freem(m);
252 				pool_put(&tcpqe_pool, tiqe);
253 				return (0);
254 			}
255 			m_adj(m, i);
256 			*tlen -= i;
257 			th->th_seq += i;
258 		}
259 	}
260 	tcpstat.tcps_rcvoopack++;
261 	tcpstat.tcps_rcvoobyte += *tlen;
262 
263 	/*
264 	 * While we overlap succeeding segments trim them or,
265 	 * if they are completely covered, dequeue them.
266 	 */
267 	for (; q != NULL; q = nq) {
268 		struct tcphdr *qhdr = q->tcpqe_tcp;
269 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
270 
271 		if (i <= 0)
272 			break;
273 		if (i < qhdr->th_reseqlen) {
274 			qhdr->th_seq += i;
275 			qhdr->th_reseqlen -= i;
276 			m_adj(q->tcpqe_m, i);
277 			break;
278 		}
279 		nq = TAILQ_NEXT(q, tcpqe_q);
280 		m_freem(q->tcpqe_m);
281 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
282 		pool_put(&tcpqe_pool, q);
283 	}
284 
285 	/* Insert the new segment queue entry into place. */
286 	tiqe->tcpqe_m = m;
287 	th->th_reseqlen = *tlen;
288 	tiqe->tcpqe_tcp = th;
289 	if (p == NULL) {
290 		TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q);
291 	} else {
292 		TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q);
293 	}
294 
295 	if (th->th_seq != tp->rcv_nxt)
296 		return (0);
297 
298 	return (tcp_flush_queue(tp));
299 }
300 
301 int
302 tcp_flush_queue(struct tcpcb *tp)
303 {
304 	struct socket *so = tp->t_inpcb->inp_socket;
305 	struct tcpqent *q, *nq;
306 	int flags;
307 
308 	/*
309 	 * Present data to user, advancing rcv_nxt through
310 	 * completed sequence space.
311 	 */
312 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
313 		return (0);
314 	q = TAILQ_FIRST(&tp->t_segq);
315 	if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt)
316 		return (0);
317 	if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen)
318 		return (0);
319 	do {
320 		tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen;
321 		flags = q->tcpqe_tcp->th_flags & TH_FIN;
322 
323 		nq = TAILQ_NEXT(q, tcpqe_q);
324 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
325 		ND6_HINT(tp);
326 		if (so->so_state & SS_CANTRCVMORE)
327 			m_freem(q->tcpqe_m);
328 		else
329 			sbappendstream(&so->so_rcv, q->tcpqe_m);
330 		pool_put(&tcpqe_pool, q);
331 		q = nq;
332 	} while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt);
333 	tp->t_flags |= TF_BLOCKOUTPUT;
334 	sorwakeup(so);
335 	tp->t_flags &= ~TF_BLOCKOUTPUT;
336 	return (flags);
337 }
338 
339 #ifdef INET6
340 int
341 tcp6_input(struct mbuf **mp, int *offp, int proto)
342 {
343 	struct mbuf *m = *mp;
344 
345 	tcp_input(m, *offp, proto);
346 	return IPPROTO_DONE;
347 }
348 #endif
349 
350 /*
351  * TCP input routine, follows pages 65-76 of the
352  * protocol specification dated September, 1981 very closely.
353  */
354 void
355 tcp_input(struct mbuf *m, ...)
356 {
357 	struct ip *ip;
358 	struct inpcb *inp = NULL;
359 	u_int8_t *optp = NULL;
360 	int optlen = 0;
361 	int tlen, off;
362 	struct tcpcb *tp = NULL;
363 	int tiflags;
364 	struct socket *so = NULL;
365 	int todrop, acked, ourfinisacked;
366 	int hdroptlen = 0;
367 	short ostate = 0;
368 	tcp_seq iss, *reuse = NULL;
369 	u_long tiwin;
370 	struct tcp_opt_info opti;
371 	int iphlen;
372 	va_list ap;
373 	struct tcphdr *th;
374 #ifdef INET6
375 	struct ip6_hdr *ip6 = NULL;
376 #endif /* INET6 */
377 #ifdef IPSEC
378 	struct m_tag *mtag;
379 	struct tdb_ident *tdbi;
380 	struct tdb *tdb;
381 	int error, s;
382 #endif /* IPSEC */
383 	int af;
384 #ifdef TCP_ECN
385 	u_char iptos;
386 #endif
387 
388 	va_start(ap, m);
389 	iphlen = va_arg(ap, int);
390 	va_end(ap);
391 
392 	tcpstat.tcps_rcvtotal++;
393 
394 	opti.ts_present = 0;
395 	opti.maxseg = 0;
396 
397 	/*
398 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
399 	 * See below for AF specific multicast.
400 	 */
401 	if (m->m_flags & (M_BCAST|M_MCAST))
402 		goto drop;
403 
404 	/*
405 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
406 	 * TCP/IPv4.
407 	 */
408 	switch (mtod(m, struct ip *)->ip_v) {
409 #ifdef INET6
410 	case 6:
411 		af = AF_INET6;
412 		break;
413 #endif
414 	case 4:
415 		af = AF_INET;
416 		break;
417 	default:
418 		m_freem(m);
419 		return;	/*EAFNOSUPPORT*/
420 	}
421 
422 	/*
423 	 * Get IP and TCP header together in first mbuf.
424 	 * Note: IP leaves IP header in first mbuf.
425 	 */
426 	switch (af) {
427 	case AF_INET:
428 #ifdef DIAGNOSTIC
429 		if (iphlen < sizeof(struct ip)) {
430 			m_freem(m);
431 			return;
432 		}
433 #endif /* DIAGNOSTIC */
434 		break;
435 #ifdef INET6
436 	case AF_INET6:
437 #ifdef DIAGNOSTIC
438 		if (iphlen < sizeof(struct ip6_hdr)) {
439 			m_freem(m);
440 			return;
441 		}
442 #endif /* DIAGNOSTIC */
443 		break;
444 #endif
445 	default:
446 		m_freem(m);
447 		return;
448 	}
449 
450 	IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th));
451 	if (!th) {
452 		tcpstat.tcps_rcvshort++;
453 		return;
454 	}
455 
456 	tlen = m->m_pkthdr.len - iphlen;
457 	ip = NULL;
458 #ifdef INET6
459 	ip6 = NULL;
460 #endif
461 	switch (af) {
462 	case AF_INET:
463 		ip = mtod(m, struct ip *);
464 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
465 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif,
466 		    m->m_pkthdr.rdomain))
467 			goto drop;
468 #ifdef TCP_ECN
469 		/* save ip_tos before clearing it for checksum */
470 		iptos = ip->ip_tos;
471 #endif
472 		break;
473 #ifdef INET6
474 	case AF_INET6:
475 		ip6 = mtod(m, struct ip6_hdr *);
476 #ifdef TCP_ECN
477 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
478 #endif
479 
480 		/* Be proactive about malicious use of IPv4 mapped address */
481 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
482 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
483 			/* XXX stat */
484 			goto drop;
485 		}
486 
487 		/*
488 		 * Be proactive about unspecified IPv6 address in source.
489 		 * As we use all-zero to indicate unbounded/unconnected pcb,
490 		 * unspecified IPv6 address can be used to confuse us.
491 		 *
492 		 * Note that packets with unspecified IPv6 destination is
493 		 * already dropped in ip6_input.
494 		 */
495 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
496 			/* XXX stat */
497 			goto drop;
498 		}
499 
500 		/* Discard packets to multicast */
501 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
502 			/* XXX stat */
503 			goto drop;
504 		}
505 		break;
506 #endif
507 	}
508 
509 	/*
510 	 * Checksum extended TCP header and data.
511 	 */
512 	if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) {
513 		int sum;
514 
515 		if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) {
516 			tcpstat.tcps_inhwcsum++;
517 			tcpstat.tcps_rcvbadsum++;
518 			goto drop;
519 		}
520 		switch (af) {
521 		case AF_INET:
522 			sum = in4_cksum(m, IPPROTO_TCP, iphlen, tlen);
523 			break;
524 #ifdef INET6
525 		case AF_INET6:
526 			sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
527 			    tlen);
528 			break;
529 #endif
530 		}
531 		if (sum != 0) {
532 			tcpstat.tcps_rcvbadsum++;
533 			goto drop;
534 		}
535 	} else {
536 		m->m_pkthdr.csum_flags &= ~M_TCP_CSUM_IN_OK;
537 		tcpstat.tcps_inhwcsum++;
538 	}
539 
540 	/*
541 	 * Check that TCP offset makes sense,
542 	 * pull out TCP options and adjust length.		XXX
543 	 */
544 	off = th->th_off << 2;
545 	if (off < sizeof(struct tcphdr) || off > tlen) {
546 		tcpstat.tcps_rcvbadoff++;
547 		goto drop;
548 	}
549 	tlen -= off;
550 	if (off > sizeof(struct tcphdr)) {
551 		IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off);
552 		if (!th) {
553 			tcpstat.tcps_rcvshort++;
554 			return;
555 		}
556 		optlen = off - sizeof(struct tcphdr);
557 		optp = (u_int8_t *)(th + 1);
558 		/*
559 		 * Do quick retrieval of timestamp options ("options
560 		 * prediction?").  If timestamp is the only option and it's
561 		 * formatted as recommended in RFC 1323 appendix A, we
562 		 * quickly get the values now and not bother calling
563 		 * tcp_dooptions(), etc.
564 		 */
565 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
566 		     (optlen > TCPOLEN_TSTAMP_APPA &&
567 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
568 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
569 		     (th->th_flags & TH_SYN) == 0) {
570 			opti.ts_present = 1;
571 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
572 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
573 			optp = NULL;	/* we've parsed the options */
574 		}
575 	}
576 	tiflags = th->th_flags;
577 
578 	/*
579 	 * Convert TCP protocol specific fields to host format.
580 	 */
581 	NTOHL(th->th_seq);
582 	NTOHL(th->th_ack);
583 	NTOHS(th->th_win);
584 	NTOHS(th->th_urp);
585 
586 	/*
587 	 * Locate pcb for segment.
588 	 */
589 #if NPF > 0
590 	if (m->m_pkthdr.pf.statekey)
591 		inp = m->m_pkthdr.pf.statekey->inp;
592 #endif
593 findpcb:
594 	if (inp == NULL) {
595 		switch (af) {
596 #ifdef INET6
597 		case AF_INET6:
598 			inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src,
599 			    th->th_sport, &ip6->ip6_dst, th->th_dport);
600 			break;
601 #endif
602 		case AF_INET:
603 			inp = in_pcbhashlookup(&tcbtable, ip->ip_src,
604 			    th->th_sport, ip->ip_dst, th->th_dport,
605 			    m->m_pkthdr.rdomain);
606 			break;
607 		}
608 #if NPF > 0
609 		if (m->m_pkthdr.pf.statekey && inp) {
610 			m->m_pkthdr.pf.statekey->inp = inp;
611 			inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
612 		}
613 #endif
614 	}
615 	if (inp == NULL) {
616 		int	inpl_flags = 0;
617 		if (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST)
618 			inpl_flags = INPLOOKUP_WILDCARD;
619 		++tcpstat.tcps_pcbhashmiss;
620 		switch (af) {
621 #ifdef INET6
622 		case AF_INET6:
623 			inp = in6_pcblookup_listen(&tcbtable,
624 			    &ip6->ip6_dst, th->th_dport, inpl_flags, m);
625 			break;
626 #endif /* INET6 */
627 		case AF_INET:
628 			inp = in_pcblookup_listen(&tcbtable,
629 			    ip->ip_dst, th->th_dport, inpl_flags, m,
630 			    m->m_pkthdr.rdomain);
631 			break;
632 		}
633 		/*
634 		 * If the state is CLOSED (i.e., TCB does not exist) then
635 		 * all data in the incoming segment is discarded.
636 		 * If the TCB exists but is in CLOSED state, it is embryonic,
637 		 * but should either do a listen or a connect soon.
638 		 */
639 		if (inp == 0) {
640 			++tcpstat.tcps_noport;
641 			goto dropwithreset_ratelim;
642 		}
643 	}
644 
645 	/* Check the minimum TTL for socket. */
646 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
647 		goto drop;
648 
649 	tp = intotcpcb(inp);
650 	if (tp == 0)
651 		goto dropwithreset_ratelim;
652 	if (tp->t_state == TCPS_CLOSED)
653 		goto drop;
654 
655 	/* Unscale the window into a 32-bit value. */
656 	if ((tiflags & TH_SYN) == 0)
657 		tiwin = th->th_win << tp->snd_scale;
658 	else
659 		tiwin = th->th_win;
660 
661 	so = inp->inp_socket;
662 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
663 		union syn_cache_sa src;
664 		union syn_cache_sa dst;
665 
666 		bzero(&src, sizeof(src));
667 		bzero(&dst, sizeof(dst));
668 		switch (af) {
669 #ifdef INET
670 		case AF_INET:
671 			src.sin.sin_len = sizeof(struct sockaddr_in);
672 			src.sin.sin_family = AF_INET;
673 			src.sin.sin_addr = ip->ip_src;
674 			src.sin.sin_port = th->th_sport;
675 
676 			dst.sin.sin_len = sizeof(struct sockaddr_in);
677 			dst.sin.sin_family = AF_INET;
678 			dst.sin.sin_addr = ip->ip_dst;
679 			dst.sin.sin_port = th->th_dport;
680 			break;
681 #endif
682 #ifdef INET6
683 		case AF_INET6:
684 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
685 			src.sin6.sin6_family = AF_INET6;
686 			src.sin6.sin6_addr = ip6->ip6_src;
687 			src.sin6.sin6_port = th->th_sport;
688 
689 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
690 			dst.sin6.sin6_family = AF_INET6;
691 			dst.sin6.sin6_addr = ip6->ip6_dst;
692 			dst.sin6.sin6_port = th->th_dport;
693 			break;
694 #endif /* INET6 */
695 		default:
696 			goto badsyn;	/*sanity*/
697 		}
698 
699 		if (so->so_options & SO_DEBUG) {
700 			ostate = tp->t_state;
701 			switch (af) {
702 #ifdef INET6
703 			case AF_INET6:
704 				bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6));
705 				bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th));
706 				break;
707 #endif
708 			case AF_INET:
709 				bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip));
710 				bcopy(th, &tcp_saveti.ti_t, sizeof(*th));
711 				break;
712 			}
713 		}
714 		if (so->so_options & SO_ACCEPTCONN) {
715 			switch (tiflags & (TH_RST|TH_SYN|TH_ACK)) {
716 
717 			case TH_SYN|TH_ACK|TH_RST:
718 			case TH_SYN|TH_RST:
719 			case TH_ACK|TH_RST:
720 			case TH_RST:
721 				syn_cache_reset(&src.sa, &dst.sa, th,
722 				    inp->inp_rtableid);
723 				goto drop;
724 
725 			case TH_SYN|TH_ACK:
726 				/*
727 				 * Received a SYN,ACK.  This should
728 				 * never happen while we are in
729 				 * LISTEN.  Send an RST.
730 				 */
731 				goto badsyn;
732 
733 			case TH_ACK:
734 				so = syn_cache_get(&src.sa, &dst.sa,
735 					th, iphlen, tlen, so, m);
736 				if (so == NULL) {
737 					/*
738 					 * We don't have a SYN for
739 					 * this ACK; send an RST.
740 					 */
741 					goto badsyn;
742 				} else if (so == (struct socket *)(-1)) {
743 					/*
744 					 * We were unable to create
745 					 * the connection.  If the
746 					 * 3-way handshake was
747 					 * completed, and RST has
748 					 * been sent to the peer.
749 					 * Since the mbuf might be
750 					 * in use for the reply,
751 					 * do not free it.
752 					 */
753 					m = NULL;
754 					goto drop;
755 				} else {
756 					/*
757 					 * We have created a
758 					 * full-blown connection.
759 					 */
760 					tp = NULL;
761 					inp = sotoinpcb(so);
762 					tp = intotcpcb(inp);
763 					if (tp == NULL)
764 						goto badsyn;	/*XXX*/
765 
766 				}
767 				break;
768 
769 			default:
770 				/*
771 				 * None of RST, SYN or ACK was set.
772 				 * This is an invalid packet for a
773 				 * TCB in LISTEN state.  Send a RST.
774 				 */
775 				goto badsyn;
776 
777 			case TH_SYN:
778 				/*
779 				 * Received a SYN.
780 				 */
781 #ifdef INET6
782 				/*
783 				 * If deprecated address is forbidden, we do
784 				 * not accept SYN to deprecated interface
785 				 * address to prevent any new inbound
786 				 * connection from getting established.
787 				 * When we do not accept SYN, we send a TCP
788 				 * RST, with deprecated source address (instead
789 				 * of dropping it).  We compromise it as it is
790 				 * much better for peer to send a RST, and
791 				 * RST will be the final packet for the
792 				 * exchange.
793 				 *
794 				 * If we do not forbid deprecated addresses, we
795 				 * accept the SYN packet.  RFC2462 does not
796 				 * suggest dropping SYN in this case.
797 				 * If we decipher RFC2462 5.5.4, it says like
798 				 * this:
799 				 * 1. use of deprecated addr with existing
800 				 *    communication is okay - "SHOULD continue
801 				 *    to be used"
802 				 * 2. use of it with new communication:
803 				 *   (2a) "SHOULD NOT be used if alternate
804 				 *        address with sufficient scope is
805 				 *        available"
806 				 *   (2b) nothing mentioned otherwise.
807 				 * Here we fall into (2b) case as we have no
808 				 * choice in our source address selection - we
809 				 * must obey the peer.
810 				 *
811 				 * The wording in RFC2462 is confusing, and
812 				 * there are multiple description text for
813 				 * deprecated address handling - worse, they
814 				 * are not exactly the same.  I believe 5.5.4
815 				 * is the best one, so we follow 5.5.4.
816 				 */
817 				if (ip6 && !ip6_use_deprecated) {
818 					struct in6_ifaddr *ia6;
819 
820 					if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif,
821 					    &ip6->ip6_dst)) &&
822 					    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
823 						tp = NULL;
824 						goto dropwithreset;
825 					}
826 				}
827 #endif
828 
829 				/*
830 				 * LISTEN socket received a SYN
831 				 * from itself?  This can't possibly
832 				 * be valid; drop the packet.
833 				 */
834 				if (th->th_dport == th->th_sport) {
835 					switch (af) {
836 #ifdef INET6
837 					case AF_INET6:
838 						if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src,
839 						    &ip6->ip6_dst)) {
840 							tcpstat.tcps_badsyn++;
841 							goto drop;
842 						}
843 						break;
844 #endif /* INET6 */
845 					case AF_INET:
846 						if (ip->ip_dst.s_addr == ip->ip_src.s_addr) {
847 							tcpstat.tcps_badsyn++;
848 							goto drop;
849 						}
850 						break;
851 					}
852 				}
853 
854 				/*
855 				 * SYN looks ok; create compressed TCP
856 				 * state for it.
857 				 */
858 				if (so->so_qlen > so->so_qlimit ||
859 				    syn_cache_add(&src.sa, &dst.sa, th, iphlen,
860 				    so, m, optp, optlen, &opti, reuse) == -1)
861 					goto drop;
862 				return;
863 			}
864 		}
865 	}
866 
867 #ifdef DIAGNOSTIC
868 	/*
869 	 * Should not happen now that all embryonic connections
870 	 * are handled with compressed state.
871 	 */
872 	if (tp->t_state == TCPS_LISTEN)
873 		panic("tcp_input: TCPS_LISTEN");
874 #endif
875 
876 #if NPF > 0
877 	if (m->m_pkthdr.pf.statekey && !m->m_pkthdr.pf.statekey->inp &&
878 	    !inp->inp_pf_sk) {
879 		m->m_pkthdr.pf.statekey->inp = inp;
880 		inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
881 	}
882 	/* The statekey has finished finding the inp, it is no longer needed. */
883 	m->m_pkthdr.pf.statekey = NULL;
884 #endif
885 
886 #ifdef IPSEC
887 	/* Find most recent IPsec tag */
888 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
889         s = splnet();
890 	if (mtag != NULL) {
891 		tdbi = (struct tdb_ident *)(mtag + 1);
892 	        tdb = gettdb(tdbi->rdomain, tdbi->spi,
893 		    &tdbi->dst, tdbi->proto);
894 	} else
895 		tdb = NULL;
896 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
897 	    tdb, inp, 0);
898 	if (error) {
899 		splx(s);
900 		goto drop;
901 	}
902 
903 	/* Latch SA */
904 	if (inp->inp_tdb_in != tdb) {
905 		if (tdb) {
906 		        tdb_add_inp(tdb, inp, 1);
907 			if (inp->inp_ipo == NULL) {
908 				inp->inp_ipo = ipsec_add_policy(inp, af,
909 				    IPSP_DIRECTION_OUT);
910 				if (inp->inp_ipo == NULL) {
911 					splx(s);
912 					goto drop;
913 				}
914 			}
915 			if (inp->inp_ipo->ipo_dstid == NULL &&
916 			    tdb->tdb_srcid != NULL) {
917 				inp->inp_ipo->ipo_dstid = tdb->tdb_srcid;
918 				tdb->tdb_srcid->ref_count++;
919 			}
920 			if (inp->inp_ipsec_remotecred == NULL &&
921 			    tdb->tdb_remote_cred != NULL) {
922 				inp->inp_ipsec_remotecred =
923 				    tdb->tdb_remote_cred;
924 				tdb->tdb_remote_cred->ref_count++;
925 			}
926 			if (inp->inp_ipsec_remoteauth == NULL &&
927 			    tdb->tdb_remote_auth != NULL) {
928 				inp->inp_ipsec_remoteauth =
929 				    tdb->tdb_remote_auth;
930 				tdb->tdb_remote_auth->ref_count++;
931 			}
932 		} else { /* Just reset */
933 		        TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp,
934 				     inp_tdb_in_next);
935 			inp->inp_tdb_in = NULL;
936 		}
937 	}
938         splx(s);
939 #endif /* IPSEC */
940 
941 	/*
942 	 * Segment received on connection.
943 	 * Reset idle time and keep-alive timer.
944 	 */
945 	tp->t_rcvtime = tcp_now;
946 	if (TCPS_HAVEESTABLISHED(tp->t_state))
947 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
948 
949 #ifdef TCP_SACK
950 	if (tp->sack_enable)
951 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
952 #endif /* TCP_SACK */
953 
954 	/*
955 	 * Process options.
956 	 */
957 #ifdef TCP_SIGNATURE
958 	if (optp || (tp->t_flags & TF_SIGNATURE))
959 #else
960 	if (optp)
961 #endif
962 		if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti,
963 		    m->m_pkthdr.rdomain))
964 			goto drop;
965 
966 	if (opti.ts_present && opti.ts_ecr) {
967 		int rtt_test;
968 
969 		/* subtract out the tcp timestamp modulator */
970 		opti.ts_ecr -= tp->ts_modulate;
971 
972 		/* make sure ts_ecr is sensible */
973 		rtt_test = tcp_now - opti.ts_ecr;
974 		if (rtt_test < 0 || rtt_test > TCP_RTT_MAX)
975 			opti.ts_ecr = 0;
976 	}
977 
978 #ifdef TCP_ECN
979 	/* if congestion experienced, set ECE bit in subsequent packets. */
980 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
981 		tp->t_flags |= TF_RCVD_CE;
982 		tcpstat.tcps_ecn_rcvce++;
983 	}
984 #endif
985 	/*
986 	 * Header prediction: check for the two common cases
987 	 * of a uni-directional data xfer.  If the packet has
988 	 * no control flags, is in-sequence, the window didn't
989 	 * change and we're not retransmitting, it's a
990 	 * candidate.  If the length is zero and the ack moved
991 	 * forward, we're the sender side of the xfer.  Just
992 	 * free the data acked & wake any higher level process
993 	 * that was blocked waiting for space.  If the length
994 	 * is non-zero and the ack didn't move, we're the
995 	 * receiver side.  If we're getting packets in-order
996 	 * (the reassembly queue is empty), add the data to
997 	 * the socket buffer and note that we need a delayed ack.
998 	 */
999 	if (tp->t_state == TCPS_ESTABLISHED &&
1000 #ifdef TCP_ECN
1001 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
1002 #else
1003 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1004 #endif
1005 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
1006 	    th->th_seq == tp->rcv_nxt &&
1007 	    tiwin && tiwin == tp->snd_wnd &&
1008 	    tp->snd_nxt == tp->snd_max) {
1009 
1010 		/*
1011 		 * If last ACK falls within this segment's sequence numbers,
1012 		 *  record the timestamp.
1013 		 * Fix from Braden, see Stevens p. 870
1014 		 */
1015 		if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1016 			tp->ts_recent_age = tcp_now;
1017 			tp->ts_recent = opti.ts_val;
1018 		}
1019 
1020 		if (tlen == 0) {
1021 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1022 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1023 			    tp->snd_cwnd >= tp->snd_wnd &&
1024 			    tp->t_dupacks == 0) {
1025 				/*
1026 				 * this is a pure ack for outstanding data.
1027 				 */
1028 				++tcpstat.tcps_predack;
1029 				if (opti.ts_present && opti.ts_ecr)
1030 					tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1031 				else if (tp->t_rtttime &&
1032 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1033 					tcp_xmit_timer(tp,
1034 					    tcp_now - tp->t_rtttime);
1035 				acked = th->th_ack - tp->snd_una;
1036 				tcpstat.tcps_rcvackpack++;
1037 				tcpstat.tcps_rcvackbyte += acked;
1038 				ND6_HINT(tp);
1039 				sbdrop(&so->so_snd, acked);
1040 
1041 				/*
1042 				 * If we had a pending ICMP message that
1043 				 * referres to data that have just been
1044 				 * acknowledged, disregard the recorded ICMP
1045 				 * message.
1046 				 */
1047 				if ((tp->t_flags & TF_PMTUD_PEND) &&
1048 				    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1049 					tp->t_flags &= ~TF_PMTUD_PEND;
1050 
1051 				/*
1052 				 * Keep track of the largest chunk of data
1053 				 * acknowledged since last PMTU update
1054 				 */
1055 				if (tp->t_pmtud_mss_acked < acked)
1056 					tp->t_pmtud_mss_acked = acked;
1057 
1058 				tp->snd_una = th->th_ack;
1059 #if defined(TCP_SACK) || defined(TCP_ECN)
1060 				/*
1061 				 * We want snd_last to track snd_una so
1062 				 * as to avoid sequence wraparound problems
1063 				 * for very large transfers.
1064 				 */
1065 #ifdef TCP_ECN
1066 				if (SEQ_GT(tp->snd_una, tp->snd_last))
1067 #endif
1068 				tp->snd_last = tp->snd_una;
1069 #endif /* TCP_SACK */
1070 #if defined(TCP_SACK) && defined(TCP_FACK)
1071 				tp->snd_fack = tp->snd_una;
1072 				tp->retran_data = 0;
1073 #endif /* TCP_FACK */
1074 				m_freem(m);
1075 
1076 				/*
1077 				 * If all outstanding data are acked, stop
1078 				 * retransmit timer, otherwise restart timer
1079 				 * using current (possibly backed-off) value.
1080 				 * If process is waiting for space,
1081 				 * wakeup/selwakeup/signal.  If data
1082 				 * are ready to send, let tcp_output
1083 				 * decide between more output or persist.
1084 				 */
1085 				if (tp->snd_una == tp->snd_max)
1086 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1087 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1088 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1089 
1090 				tcp_update_sndspace(tp);
1091 				if (sb_notify(&so->so_snd)) {
1092 					tp->t_flags |= TF_BLOCKOUTPUT;
1093 					sowwakeup(so);
1094 					tp->t_flags &= ~TF_BLOCKOUTPUT;
1095 				}
1096 				if (so->so_snd.sb_cc ||
1097 				    tp->t_flags & TF_NEEDOUTPUT)
1098 					(void) tcp_output(tp);
1099 				return;
1100 			}
1101 		} else if (th->th_ack == tp->snd_una &&
1102 		    TAILQ_EMPTY(&tp->t_segq) &&
1103 		    tlen <= sbspace(&so->so_rcv)) {
1104 			/*
1105 			 * This is a pure, in-sequence data packet
1106 			 * with nothing on the reassembly queue and
1107 			 * we have enough buffer space to take it.
1108 			 */
1109 #ifdef TCP_SACK
1110 			/* Clean receiver SACK report if present */
1111 			if (tp->sack_enable && tp->rcv_numsacks)
1112 				tcp_clean_sackreport(tp);
1113 #endif /* TCP_SACK */
1114 			++tcpstat.tcps_preddat;
1115 			tp->rcv_nxt += tlen;
1116 			tcpstat.tcps_rcvpack++;
1117 			tcpstat.tcps_rcvbyte += tlen;
1118 			ND6_HINT(tp);
1119 
1120 			TCP_SETUP_ACK(tp, tiflags, m);
1121 			/*
1122 			 * Drop TCP, IP headers and TCP options then add data
1123 			 * to socket buffer.
1124 			 */
1125 			if (so->so_state & SS_CANTRCVMORE)
1126 				m_freem(m);
1127 			else {
1128 				if (opti.ts_present && opti.ts_ecr) {
1129 					if (tp->rfbuf_ts < opti.ts_ecr &&
1130 					    opti.ts_ecr - tp->rfbuf_ts < hz) {
1131 						tcp_update_rcvspace(tp);
1132 						/* Start over with next RTT. */
1133 						tp->rfbuf_cnt = 0;
1134 						tp->rfbuf_ts = 0;
1135 					} else
1136 						tp->rfbuf_cnt += tlen;
1137 				}
1138 				m_adj(m, iphlen + off);
1139 				sbappendstream(&so->so_rcv, m);
1140 			}
1141 			tp->t_flags |= TF_BLOCKOUTPUT;
1142 			sorwakeup(so);
1143 			tp->t_flags &= ~TF_BLOCKOUTPUT;
1144 			if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT))
1145 				(void) tcp_output(tp);
1146 			return;
1147 		}
1148 	}
1149 
1150 	/*
1151 	 * Compute mbuf offset to TCP data segment.
1152 	 */
1153 	hdroptlen = iphlen + off;
1154 
1155 	/*
1156 	 * Calculate amount of space in receive window,
1157 	 * and then do TCP input processing.
1158 	 * Receive window is amount of space in rcv queue,
1159 	 * but not less than advertised window.
1160 	 */
1161 	{ int win;
1162 
1163 	win = sbspace(&so->so_rcv);
1164 	if (win < 0)
1165 		win = 0;
1166 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1167 	}
1168 
1169 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1170 	tp->rfbuf_cnt = 0;
1171 	tp->rfbuf_ts = 0;
1172 
1173 	switch (tp->t_state) {
1174 
1175 	/*
1176 	 * If the state is SYN_RECEIVED:
1177 	 * 	if seg contains SYN/ACK, send an RST.
1178 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1179 	 */
1180 
1181 	case TCPS_SYN_RECEIVED:
1182 		if (tiflags & TH_ACK) {
1183 			if (tiflags & TH_SYN) {
1184 				tcpstat.tcps_badsyn++;
1185 				goto dropwithreset;
1186 			}
1187 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1188 			    SEQ_GT(th->th_ack, tp->snd_max))
1189 				goto dropwithreset;
1190 		}
1191 		break;
1192 
1193 	/*
1194 	 * If the state is SYN_SENT:
1195 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1196 	 *	if seg contains a RST, then drop the connection.
1197 	 *	if seg does not contain SYN, then drop it.
1198 	 * Otherwise this is an acceptable SYN segment
1199 	 *	initialize tp->rcv_nxt and tp->irs
1200 	 *	if seg contains ack then advance tp->snd_una
1201 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1202 	 *	arrange for segment to be acked (eventually)
1203 	 *	continue processing rest of data/controls, beginning with URG
1204 	 */
1205 	case TCPS_SYN_SENT:
1206 		if ((tiflags & TH_ACK) &&
1207 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1208 		     SEQ_GT(th->th_ack, tp->snd_max)))
1209 			goto dropwithreset;
1210 		if (tiflags & TH_RST) {
1211 #ifdef TCP_ECN
1212 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1213 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1214 				goto drop;
1215 #endif
1216 			if (tiflags & TH_ACK)
1217 				tp = tcp_drop(tp, ECONNREFUSED);
1218 			goto drop;
1219 		}
1220 		if ((tiflags & TH_SYN) == 0)
1221 			goto drop;
1222 		if (tiflags & TH_ACK) {
1223 			tp->snd_una = th->th_ack;
1224 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1225 				tp->snd_nxt = tp->snd_una;
1226 		}
1227 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1228 		tp->irs = th->th_seq;
1229 		tcp_mss(tp, opti.maxseg);
1230 		/* Reset initial window to 1 segment for retransmit */
1231 		if (tp->t_rxtshift > 0)
1232 			tp->snd_cwnd = tp->t_maxseg;
1233 		tcp_rcvseqinit(tp);
1234 		tp->t_flags |= TF_ACKNOW;
1235 #ifdef TCP_SACK
1236                 /*
1237                  * If we've sent a SACK_PERMITTED option, and the peer
1238                  * also replied with one, then TF_SACK_PERMIT should have
1239                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1240                  */
1241 		if (tp->sack_enable)
1242 			tp->sack_enable = tp->t_flags & TF_SACK_PERMIT;
1243 #endif
1244 #ifdef TCP_ECN
1245 		/*
1246 		 * if ECE is set but CWR is not set for SYN-ACK, or
1247 		 * both ECE and CWR are set for simultaneous open,
1248 		 * peer is ECN capable.
1249 		 */
1250 		if (tcp_do_ecn) {
1251 			switch (tiflags & (TH_ACK|TH_ECE|TH_CWR)) {
1252 			case TH_ACK|TH_ECE:
1253 			case TH_ECE|TH_CWR:
1254 				tp->t_flags |= TF_ECN_PERMIT;
1255 				tiflags &= ~(TH_ECE|TH_CWR);
1256 				tcpstat.tcps_ecn_accepts++;
1257 			}
1258 		}
1259 #endif
1260 
1261 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1262 			tcpstat.tcps_connects++;
1263 			soisconnected(so);
1264 			tp->t_state = TCPS_ESTABLISHED;
1265 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1266 			/* Do window scaling on this connection? */
1267 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1268 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1269 				tp->snd_scale = tp->requested_s_scale;
1270 				tp->rcv_scale = tp->request_r_scale;
1271 			}
1272 			tcp_flush_queue(tp);
1273 
1274 			/*
1275 			 * if we didn't have to retransmit the SYN,
1276 			 * use its rtt as our initial srtt & rtt var.
1277 			 */
1278 			if (tp->t_rtttime)
1279 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1280 			/*
1281 			 * Since new data was acked (the SYN), open the
1282 			 * congestion window by one MSS.  We do this
1283 			 * here, because we won't go through the normal
1284 			 * ACK processing below.  And since this is the
1285 			 * start of the connection, we know we are in
1286 			 * the exponential phase of slow-start.
1287 			 */
1288 			tp->snd_cwnd += tp->t_maxseg;
1289 		} else
1290 			tp->t_state = TCPS_SYN_RECEIVED;
1291 
1292 #if 0
1293 trimthenstep6:
1294 #endif
1295 		/*
1296 		 * Advance th->th_seq to correspond to first data byte.
1297 		 * If data, trim to stay within window,
1298 		 * dropping FIN if necessary.
1299 		 */
1300 		th->th_seq++;
1301 		if (tlen > tp->rcv_wnd) {
1302 			todrop = tlen - tp->rcv_wnd;
1303 			m_adj(m, -todrop);
1304 			tlen = tp->rcv_wnd;
1305 			tiflags &= ~TH_FIN;
1306 			tcpstat.tcps_rcvpackafterwin++;
1307 			tcpstat.tcps_rcvbyteafterwin += todrop;
1308 		}
1309 		tp->snd_wl1 = th->th_seq - 1;
1310 		tp->rcv_up = th->th_seq;
1311 		goto step6;
1312 	/*
1313 	 * If a new connection request is received while in TIME_WAIT,
1314 	 * drop the old connection and start over if the if the
1315 	 * timestamp or the sequence numbers are above the previous
1316 	 * ones.
1317 	 */
1318 	case TCPS_TIME_WAIT:
1319 		if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) &&
1320 		    ((opti.ts_present &&
1321 		    TSTMP_LT(tp->ts_recent, opti.ts_val)) ||
1322 		    SEQ_GT(th->th_seq, tp->rcv_nxt))) {
1323 #if NPF > 0
1324 			/*
1325 			 * The socket will be recreated but the new state
1326 			 * has already been linked to the socket.  Remove the
1327 			 * link between old socket and new state.
1328 			 */
1329 			if (inp->inp_pf_sk) {
1330 				inp->inp_pf_sk->inp = NULL;
1331 				inp->inp_pf_sk = NULL;
1332 			}
1333 #endif
1334 			/*
1335 			* Advance the iss by at least 32768, but
1336 			* clear the msb in order to make sure
1337 			* that SEG_LT(snd_nxt, iss).
1338 			*/
1339 			iss = tp->snd_nxt +
1340 			    ((arc4random() & 0x7fffffff) | 0x8000);
1341 			reuse = &iss;
1342 			tp = tcp_close(tp);
1343 			inp = NULL;
1344 			goto findpcb;
1345 		}
1346 	}
1347 
1348 	/*
1349 	 * States other than LISTEN or SYN_SENT.
1350 	 * First check timestamp, if present.
1351 	 * Then check that at least some bytes of segment are within
1352 	 * receive window.  If segment begins before rcv_nxt,
1353 	 * drop leading data (and SYN); if nothing left, just ack.
1354 	 *
1355 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1356 	 * and it's less than opti.ts_recent, drop it.
1357 	 */
1358 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1359 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1360 
1361 		/* Check to see if ts_recent is over 24 days old.  */
1362 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1363 			/*
1364 			 * Invalidate ts_recent.  If this segment updates
1365 			 * ts_recent, the age will be reset later and ts_recent
1366 			 * will get a valid value.  If it does not, setting
1367 			 * ts_recent to zero will at least satisfy the
1368 			 * requirement that zero be placed in the timestamp
1369 			 * echo reply when ts_recent isn't valid.  The
1370 			 * age isn't reset until we get a valid ts_recent
1371 			 * because we don't want out-of-order segments to be
1372 			 * dropped when ts_recent is old.
1373 			 */
1374 			tp->ts_recent = 0;
1375 		} else {
1376 			tcpstat.tcps_rcvduppack++;
1377 			tcpstat.tcps_rcvdupbyte += tlen;
1378 			tcpstat.tcps_pawsdrop++;
1379 			goto dropafterack;
1380 		}
1381 	}
1382 
1383 	todrop = tp->rcv_nxt - th->th_seq;
1384 	if (todrop > 0) {
1385 		if (tiflags & TH_SYN) {
1386 			tiflags &= ~TH_SYN;
1387 			th->th_seq++;
1388 			if (th->th_urp > 1)
1389 				th->th_urp--;
1390 			else
1391 				tiflags &= ~TH_URG;
1392 			todrop--;
1393 		}
1394 		if (todrop > tlen ||
1395 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1396 			/*
1397 			 * Any valid FIN must be to the left of the
1398 			 * window.  At this point, FIN must be a
1399 			 * duplicate or out-of-sequence, so drop it.
1400 			 */
1401 			tiflags &= ~TH_FIN;
1402 			/*
1403 			 * Send ACK to resynchronize, and drop any data,
1404 			 * but keep on processing for RST or ACK.
1405 			 */
1406 			tp->t_flags |= TF_ACKNOW;
1407 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1408 			tcpstat.tcps_rcvduppack++;
1409 		} else {
1410 			tcpstat.tcps_rcvpartduppack++;
1411 			tcpstat.tcps_rcvpartdupbyte += todrop;
1412 		}
1413 		hdroptlen += todrop;	/* drop from head afterwards */
1414 		th->th_seq += todrop;
1415 		tlen -= todrop;
1416 		if (th->th_urp > todrop)
1417 			th->th_urp -= todrop;
1418 		else {
1419 			tiflags &= ~TH_URG;
1420 			th->th_urp = 0;
1421 		}
1422 	}
1423 
1424 	/*
1425 	 * If new data are received on a connection after the
1426 	 * user processes are gone, then RST the other end.
1427 	 */
1428 	if ((so->so_state & SS_NOFDREF) &&
1429 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1430 		tp = tcp_close(tp);
1431 		tcpstat.tcps_rcvafterclose++;
1432 		goto dropwithreset;
1433 	}
1434 
1435 	/*
1436 	 * If segment ends after window, drop trailing data
1437 	 * (and PUSH and FIN); if nothing left, just ACK.
1438 	 */
1439 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1440 	if (todrop > 0) {
1441 		tcpstat.tcps_rcvpackafterwin++;
1442 		if (todrop >= tlen) {
1443 			tcpstat.tcps_rcvbyteafterwin += tlen;
1444 			/*
1445 			 * If window is closed can only take segments at
1446 			 * window edge, and have to drop data and PUSH from
1447 			 * incoming segments.  Continue processing, but
1448 			 * remember to ack.  Otherwise, drop segment
1449 			 * and ack.
1450 			 */
1451 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1452 				tp->t_flags |= TF_ACKNOW;
1453 				tcpstat.tcps_rcvwinprobe++;
1454 			} else
1455 				goto dropafterack;
1456 		} else
1457 			tcpstat.tcps_rcvbyteafterwin += todrop;
1458 		m_adj(m, -todrop);
1459 		tlen -= todrop;
1460 		tiflags &= ~(TH_PUSH|TH_FIN);
1461 	}
1462 
1463 	/*
1464 	 * If last ACK falls within this segment's sequence numbers,
1465 	 * record its timestamp if it's more recent.
1466 	 * Cf fix from Braden, see Stevens p. 870
1467 	 */
1468 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1469 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1470 		if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1471 		    ((tiflags & (TH_SYN|TH_FIN)) != 0)))
1472 			tp->ts_recent = opti.ts_val;
1473 		else
1474 			tp->ts_recent = 0;
1475 		tp->ts_recent_age = tcp_now;
1476 	}
1477 
1478 	/*
1479 	 * If the RST bit is set examine the state:
1480 	 *    SYN_RECEIVED STATE:
1481 	 *	If passive open, return to LISTEN state.
1482 	 *	If active open, inform user that connection was refused.
1483 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1484 	 *	Inform user that connection was reset, and close tcb.
1485 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1486 	 *	Close the tcb.
1487 	 */
1488 	if (tiflags & TH_RST) {
1489 		if (th->th_seq != tp->last_ack_sent &&
1490 		    th->th_seq != tp->rcv_nxt &&
1491 		    th->th_seq != (tp->rcv_nxt + 1))
1492 			goto drop;
1493 
1494 		switch (tp->t_state) {
1495 		case TCPS_SYN_RECEIVED:
1496 #ifdef TCP_ECN
1497 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1498 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1499 				goto drop;
1500 #endif
1501 			so->so_error = ECONNREFUSED;
1502 			goto close;
1503 
1504 		case TCPS_ESTABLISHED:
1505 		case TCPS_FIN_WAIT_1:
1506 		case TCPS_FIN_WAIT_2:
1507 		case TCPS_CLOSE_WAIT:
1508 			so->so_error = ECONNRESET;
1509 		close:
1510 			tp->t_state = TCPS_CLOSED;
1511 			tcpstat.tcps_drops++;
1512 			tp = tcp_close(tp);
1513 			goto drop;
1514 		case TCPS_CLOSING:
1515 		case TCPS_LAST_ACK:
1516 		case TCPS_TIME_WAIT:
1517 			tp = tcp_close(tp);
1518 			goto drop;
1519 		}
1520 	}
1521 
1522 	/*
1523 	 * If a SYN is in the window, then this is an
1524 	 * error and we ACK and drop the packet.
1525 	 */
1526 	if (tiflags & TH_SYN)
1527 		goto dropafterack_ratelim;
1528 
1529 	/*
1530 	 * If the ACK bit is off we drop the segment and return.
1531 	 */
1532 	if ((tiflags & TH_ACK) == 0) {
1533 		if (tp->t_flags & TF_ACKNOW)
1534 			goto dropafterack;
1535 		else
1536 			goto drop;
1537 	}
1538 
1539 	/*
1540 	 * Ack processing.
1541 	 */
1542 	switch (tp->t_state) {
1543 
1544 	/*
1545 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1546 	 * ESTABLISHED state and continue processing.
1547 	 * The ACK was checked above.
1548 	 */
1549 	case TCPS_SYN_RECEIVED:
1550 		tcpstat.tcps_connects++;
1551 		soisconnected(so);
1552 		tp->t_state = TCPS_ESTABLISHED;
1553 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1554 		/* Do window scaling? */
1555 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1556 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1557 			tp->snd_scale = tp->requested_s_scale;
1558 			tp->rcv_scale = tp->request_r_scale;
1559 			tiwin = th->th_win << tp->snd_scale;
1560 		}
1561 		tcp_flush_queue(tp);
1562 		tp->snd_wl1 = th->th_seq - 1;
1563 		/* fall into ... */
1564 
1565 	/*
1566 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1567 	 * ACKs.  If the ack is in the range
1568 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1569 	 * then advance tp->snd_una to th->th_ack and drop
1570 	 * data from the retransmission queue.  If this ACK reflects
1571 	 * more up to date window information we update our window information.
1572 	 */
1573 	case TCPS_ESTABLISHED:
1574 	case TCPS_FIN_WAIT_1:
1575 	case TCPS_FIN_WAIT_2:
1576 	case TCPS_CLOSE_WAIT:
1577 	case TCPS_CLOSING:
1578 	case TCPS_LAST_ACK:
1579 	case TCPS_TIME_WAIT:
1580 #ifdef TCP_ECN
1581 		/*
1582 		 * if we receive ECE and are not already in recovery phase,
1583 		 * reduce cwnd by half but don't slow-start.
1584 		 * advance snd_last to snd_max not to reduce cwnd again
1585 		 * until all outstanding packets are acked.
1586 		 */
1587 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1588 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1589 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1590 				u_int win;
1591 
1592 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1593 				if (win > 1) {
1594 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1595 					tp->snd_cwnd = tp->snd_ssthresh;
1596 					tp->snd_last = tp->snd_max;
1597 					tp->t_flags |= TF_SEND_CWR;
1598 					tcpstat.tcps_cwr_ecn++;
1599 				}
1600 			}
1601 			tcpstat.tcps_ecn_rcvece++;
1602 		}
1603 		/*
1604 		 * if we receive CWR, we know that the peer has reduced
1605 		 * its congestion window.  stop sending ecn-echo.
1606 		 */
1607 		if ((tiflags & TH_CWR)) {
1608 			tp->t_flags &= ~TF_RCVD_CE;
1609 			tcpstat.tcps_ecn_rcvcwr++;
1610 		}
1611 #endif /* TCP_ECN */
1612 
1613 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1614 			/*
1615 			 * Duplicate/old ACK processing.
1616 			 * Increments t_dupacks:
1617 			 *	Pure duplicate (same seq/ack/window, no data)
1618 			 * Doesn't affect t_dupacks:
1619 			 *	Data packets.
1620 			 *	Normal window updates (window opens)
1621 			 * Resets t_dupacks:
1622 			 *	New data ACKed.
1623 			 *	Window shrinks
1624 			 *	Old ACK
1625 			 */
1626 			if (tlen) {
1627 				/* Drop very old ACKs unless th_seq matches */
1628 				if (th->th_seq != tp->rcv_nxt &&
1629 				   SEQ_LT(th->th_ack,
1630 				   tp->snd_una - tp->max_sndwnd)) {
1631 					tcpstat.tcps_rcvacktooold++;
1632 					goto drop;
1633 				}
1634 				break;
1635 			}
1636 			/*
1637 			 * If we get an old ACK, there is probably packet
1638 			 * reordering going on.  Be conservative and reset
1639 			 * t_dupacks so that we are less aggressive in
1640 			 * doing a fast retransmit.
1641 			 */
1642 			if (th->th_ack != tp->snd_una) {
1643 				tp->t_dupacks = 0;
1644 				break;
1645 			}
1646 			if (tiwin == tp->snd_wnd) {
1647 				tcpstat.tcps_rcvdupack++;
1648 				/*
1649 				 * If we have outstanding data (other than
1650 				 * a window probe), this is a completely
1651 				 * duplicate ack (ie, window info didn't
1652 				 * change), the ack is the biggest we've
1653 				 * seen and we've seen exactly our rexmt
1654 				 * threshold of them, assume a packet
1655 				 * has been dropped and retransmit it.
1656 				 * Kludge snd_nxt & the congestion
1657 				 * window so we send only this one
1658 				 * packet.
1659 				 *
1660 				 * We know we're losing at the current
1661 				 * window size so do congestion avoidance
1662 				 * (set ssthresh to half the current window
1663 				 * and pull our congestion window back to
1664 				 * the new ssthresh).
1665 				 *
1666 				 * Dup acks mean that packets have left the
1667 				 * network (they're now cached at the receiver)
1668 				 * so bump cwnd by the amount in the receiver
1669 				 * to keep a constant cwnd packets in the
1670 				 * network.
1671 				 */
1672 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1673 					tp->t_dupacks = 0;
1674 #if defined(TCP_SACK) && defined(TCP_FACK)
1675 				/*
1676 				 * In FACK, can enter fast rec. if the receiver
1677 				 * reports a reass. queue longer than 3 segs.
1678 				 */
1679 				else if (++tp->t_dupacks == tcprexmtthresh ||
1680 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1681 				    tp->t_maxseg + tp->snd_una)) &&
1682 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1683 #else
1684 				else if (++tp->t_dupacks == tcprexmtthresh) {
1685 #endif /* TCP_FACK */
1686 					tcp_seq onxt = tp->snd_nxt;
1687 					u_long win =
1688 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1689 						2 / tp->t_maxseg;
1690 
1691 #if defined(TCP_SACK) || defined(TCP_ECN)
1692 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1693 					    	/*
1694 						 * False fast retx after
1695 						 * timeout.  Do not cut window.
1696 						 */
1697 						tp->t_dupacks = 0;
1698 						goto drop;
1699 					}
1700 #endif
1701 					if (win < 2)
1702 						win = 2;
1703 					tp->snd_ssthresh = win * tp->t_maxseg;
1704 #ifdef TCP_SACK
1705 					tp->snd_last = tp->snd_max;
1706                     			if (tp->sack_enable) {
1707 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1708 						tp->t_rtttime = 0;
1709 #ifdef TCP_ECN
1710 						tp->t_flags |= TF_SEND_CWR;
1711 #endif
1712 						tcpstat.tcps_cwr_frecovery++;
1713 						tcpstat.tcps_sack_recovery_episode++;
1714 #if defined(TCP_SACK) && defined(TCP_FACK)
1715 						tp->t_dupacks = tcprexmtthresh;
1716 						(void) tcp_output(tp);
1717 						/*
1718 						 * During FR, snd_cwnd is held
1719 						 * constant for FACK.
1720 						 */
1721 						tp->snd_cwnd = tp->snd_ssthresh;
1722 #else
1723 						/*
1724 						 * tcp_output() will send
1725 						 * oldest SACK-eligible rtx.
1726 						 */
1727 						(void) tcp_output(tp);
1728 						tp->snd_cwnd = tp->snd_ssthresh+
1729 					           tp->t_maxseg * tp->t_dupacks;
1730 #endif /* TCP_FACK */
1731 						goto drop;
1732 					}
1733 #endif /* TCP_SACK */
1734 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1735 					tp->t_rtttime = 0;
1736 					tp->snd_nxt = th->th_ack;
1737 					tp->snd_cwnd = tp->t_maxseg;
1738 #ifdef TCP_ECN
1739 					tp->t_flags |= TF_SEND_CWR;
1740 #endif
1741 					tcpstat.tcps_cwr_frecovery++;
1742 					tcpstat.tcps_sndrexmitfast++;
1743 					(void) tcp_output(tp);
1744 
1745 					tp->snd_cwnd = tp->snd_ssthresh +
1746 					    tp->t_maxseg * tp->t_dupacks;
1747 					if (SEQ_GT(onxt, tp->snd_nxt))
1748 						tp->snd_nxt = onxt;
1749 					goto drop;
1750 				} else if (tp->t_dupacks > tcprexmtthresh) {
1751 #if defined(TCP_SACK) && defined(TCP_FACK)
1752 					/*
1753 					 * while (awnd < cwnd)
1754 					 *         sendsomething();
1755 					 */
1756 					if (tp->sack_enable) {
1757 						if (tp->snd_awnd < tp->snd_cwnd)
1758 							tcp_output(tp);
1759 						goto drop;
1760 					}
1761 #endif /* TCP_FACK */
1762 					tp->snd_cwnd += tp->t_maxseg;
1763 					(void) tcp_output(tp);
1764 					goto drop;
1765 				}
1766 			} else if (tiwin < tp->snd_wnd) {
1767 				/*
1768 				 * The window was retracted!  Previous dup
1769 				 * ACKs may have been due to packets arriving
1770 				 * after the shrunken window, not a missing
1771 				 * packet, so play it safe and reset t_dupacks
1772 				 */
1773 				tp->t_dupacks = 0;
1774 			}
1775 			break;
1776 		}
1777 		/*
1778 		 * If the congestion window was inflated to account
1779 		 * for the other side's cached packets, retract it.
1780 		 */
1781 #if defined(TCP_SACK)
1782 		if (tp->sack_enable) {
1783 			if (tp->t_dupacks >= tcprexmtthresh) {
1784 				/* Check for a partial ACK */
1785 				if (tcp_sack_partialack(tp, th)) {
1786 #if defined(TCP_SACK) && defined(TCP_FACK)
1787 					/* Force call to tcp_output */
1788 					if (tp->snd_awnd < tp->snd_cwnd)
1789 						tp->t_flags |= TF_NEEDOUTPUT;
1790 #else
1791 					tp->snd_cwnd += tp->t_maxseg;
1792 					tp->t_flags |= TF_NEEDOUTPUT;
1793 #endif /* TCP_FACK */
1794 				} else {
1795 					/* Out of fast recovery */
1796 					tp->snd_cwnd = tp->snd_ssthresh;
1797 					if (tcp_seq_subtract(tp->snd_max,
1798 					    th->th_ack) < tp->snd_ssthresh)
1799 						tp->snd_cwnd =
1800 						   tcp_seq_subtract(tp->snd_max,
1801 					           th->th_ack);
1802 					tp->t_dupacks = 0;
1803 #if defined(TCP_SACK) && defined(TCP_FACK)
1804 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1805 						tp->snd_fack = th->th_ack;
1806 #endif /* TCP_FACK */
1807 				}
1808 			}
1809 		} else {
1810 			if (tp->t_dupacks >= tcprexmtthresh &&
1811 			    !tcp_newreno(tp, th)) {
1812 				/* Out of fast recovery */
1813 				tp->snd_cwnd = tp->snd_ssthresh;
1814 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1815 			  	    tp->snd_ssthresh)
1816 					tp->snd_cwnd =
1817 					    tcp_seq_subtract(tp->snd_max,
1818 					    th->th_ack);
1819 				tp->t_dupacks = 0;
1820 			}
1821 		}
1822 		if (tp->t_dupacks < tcprexmtthresh)
1823 			tp->t_dupacks = 0;
1824 #else /* else no TCP_SACK */
1825 		if (tp->t_dupacks >= tcprexmtthresh &&
1826 		    tp->snd_cwnd > tp->snd_ssthresh)
1827 			tp->snd_cwnd = tp->snd_ssthresh;
1828 		tp->t_dupacks = 0;
1829 #endif
1830 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1831 			tcpstat.tcps_rcvacktoomuch++;
1832 			goto dropafterack_ratelim;
1833 		}
1834 		acked = th->th_ack - tp->snd_una;
1835 		tcpstat.tcps_rcvackpack++;
1836 		tcpstat.tcps_rcvackbyte += acked;
1837 
1838 		/*
1839 		 * If we have a timestamp reply, update smoothed
1840 		 * round trip time.  If no timestamp is present but
1841 		 * transmit timer is running and timed sequence
1842 		 * number was acked, update smoothed round trip time.
1843 		 * Since we now have an rtt measurement, cancel the
1844 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1845 		 * Recompute the initial retransmit timer.
1846 		 */
1847 		if (opti.ts_present && opti.ts_ecr)
1848 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1849 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1850 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1851 
1852 		/*
1853 		 * If all outstanding data is acked, stop retransmit
1854 		 * timer and remember to restart (more output or persist).
1855 		 * If there is more data to be acked, restart retransmit
1856 		 * timer, using current (possibly backed-off) value.
1857 		 */
1858 		if (th->th_ack == tp->snd_max) {
1859 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1860 			tp->t_flags |= TF_NEEDOUTPUT;
1861 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1862 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1863 		/*
1864 		 * When new data is acked, open the congestion window.
1865 		 * If the window gives us less than ssthresh packets
1866 		 * in flight, open exponentially (maxseg per packet).
1867 		 * Otherwise open linearly: maxseg per window
1868 		 * (maxseg^2 / cwnd per packet).
1869 		 */
1870 		{
1871 		u_int cw = tp->snd_cwnd;
1872 		u_int incr = tp->t_maxseg;
1873 
1874 		if (cw > tp->snd_ssthresh)
1875 			incr = incr * incr / cw;
1876 #if defined (TCP_SACK)
1877 		if (tp->t_dupacks < tcprexmtthresh)
1878 #endif
1879 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1880 		}
1881 		ND6_HINT(tp);
1882 		if (acked > so->so_snd.sb_cc) {
1883 			tp->snd_wnd -= so->so_snd.sb_cc;
1884 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1885 			ourfinisacked = 1;
1886 		} else {
1887 			sbdrop(&so->so_snd, acked);
1888 			tp->snd_wnd -= acked;
1889 			ourfinisacked = 0;
1890 		}
1891 
1892 		tcp_update_sndspace(tp);
1893 		if (sb_notify(&so->so_snd)) {
1894 			tp->t_flags |= TF_BLOCKOUTPUT;
1895 			sowwakeup(so);
1896 			tp->t_flags &= ~TF_BLOCKOUTPUT;
1897 		}
1898 
1899 		/*
1900 		 * If we had a pending ICMP message that referred to data
1901 		 * that have just been acknowledged, disregard the recorded
1902 		 * ICMP message.
1903 		 */
1904 		if ((tp->t_flags & TF_PMTUD_PEND) &&
1905 		    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1906 			tp->t_flags &= ~TF_PMTUD_PEND;
1907 
1908 		/*
1909 		 * Keep track of the largest chunk of data acknowledged
1910 		 * since last PMTU update
1911 		 */
1912 		if (tp->t_pmtud_mss_acked < acked)
1913 			tp->t_pmtud_mss_acked = acked;
1914 
1915 		tp->snd_una = th->th_ack;
1916 #ifdef TCP_ECN
1917 		/* sync snd_last with snd_una */
1918 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1919 			tp->snd_last = tp->snd_una;
1920 #endif
1921 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1922 			tp->snd_nxt = tp->snd_una;
1923 #if defined (TCP_SACK) && defined (TCP_FACK)
1924 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1925 			tp->snd_fack = tp->snd_una;
1926 			/* Update snd_awnd for partial ACK
1927 			 * without any SACK blocks.
1928 			 */
1929 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1930 				tp->snd_fack) + tp->retran_data;
1931 		}
1932 #endif
1933 
1934 		switch (tp->t_state) {
1935 
1936 		/*
1937 		 * In FIN_WAIT_1 STATE in addition to the processing
1938 		 * for the ESTABLISHED state if our FIN is now acknowledged
1939 		 * then enter FIN_WAIT_2.
1940 		 */
1941 		case TCPS_FIN_WAIT_1:
1942 			if (ourfinisacked) {
1943 				/*
1944 				 * If we can't receive any more
1945 				 * data, then closing user can proceed.
1946 				 * Starting the timer is contrary to the
1947 				 * specification, but if we don't get a FIN
1948 				 * we'll hang forever.
1949 				 */
1950 				if (so->so_state & SS_CANTRCVMORE) {
1951 					soisdisconnected(so);
1952 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1953 				}
1954 				tp->t_state = TCPS_FIN_WAIT_2;
1955 			}
1956 			break;
1957 
1958 		/*
1959 		 * In CLOSING STATE in addition to the processing for
1960 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1961 		 * then enter the TIME-WAIT state, otherwise ignore
1962 		 * the segment.
1963 		 */
1964 		case TCPS_CLOSING:
1965 			if (ourfinisacked) {
1966 				tp->t_state = TCPS_TIME_WAIT;
1967 				tcp_canceltimers(tp);
1968 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1969 				soisdisconnected(so);
1970 			}
1971 			break;
1972 
1973 		/*
1974 		 * In LAST_ACK, we may still be waiting for data to drain
1975 		 * and/or to be acked, as well as for the ack of our FIN.
1976 		 * If our FIN is now acknowledged, delete the TCB,
1977 		 * enter the closed state and return.
1978 		 */
1979 		case TCPS_LAST_ACK:
1980 			if (ourfinisacked) {
1981 				tp = tcp_close(tp);
1982 				goto drop;
1983 			}
1984 			break;
1985 
1986 		/*
1987 		 * In TIME_WAIT state the only thing that should arrive
1988 		 * is a retransmission of the remote FIN.  Acknowledge
1989 		 * it and restart the finack timer.
1990 		 */
1991 		case TCPS_TIME_WAIT:
1992 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1993 			goto dropafterack;
1994 		}
1995 	}
1996 
1997 step6:
1998 	/*
1999 	 * Update window information.
2000 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2001 	 */
2002 	if ((tiflags & TH_ACK) &&
2003 	    (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq &&
2004 	    (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2005 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2006 		/* keep track of pure window updates */
2007 		if (tlen == 0 &&
2008 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2009 			tcpstat.tcps_rcvwinupd++;
2010 		tp->snd_wnd = tiwin;
2011 		tp->snd_wl1 = th->th_seq;
2012 		tp->snd_wl2 = th->th_ack;
2013 		if (tp->snd_wnd > tp->max_sndwnd)
2014 			tp->max_sndwnd = tp->snd_wnd;
2015 		tp->t_flags |= TF_NEEDOUTPUT;
2016 	}
2017 
2018 	/*
2019 	 * Process segments with URG.
2020 	 */
2021 	if ((tiflags & TH_URG) && th->th_urp &&
2022 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2023 		/*
2024 		 * This is a kludge, but if we receive and accept
2025 		 * random urgent pointers, we'll crash in
2026 		 * soreceive.  It's hard to imagine someone
2027 		 * actually wanting to send this much urgent data.
2028 		 */
2029 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2030 			th->th_urp = 0;			/* XXX */
2031 			tiflags &= ~TH_URG;		/* XXX */
2032 			goto dodata;			/* XXX */
2033 		}
2034 		/*
2035 		 * If this segment advances the known urgent pointer,
2036 		 * then mark the data stream.  This should not happen
2037 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2038 		 * a FIN has been received from the remote side.
2039 		 * In these states we ignore the URG.
2040 		 *
2041 		 * According to RFC961 (Assigned Protocols),
2042 		 * the urgent pointer points to the last octet
2043 		 * of urgent data.  We continue, however,
2044 		 * to consider it to indicate the first octet
2045 		 * of data past the urgent section as the original
2046 		 * spec states (in one of two places).
2047 		 */
2048 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2049 			tp->rcv_up = th->th_seq + th->th_urp;
2050 			so->so_oobmark = so->so_rcv.sb_cc +
2051 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2052 			if (so->so_oobmark == 0)
2053 				so->so_state |= SS_RCVATMARK;
2054 			sohasoutofband(so);
2055 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2056 		}
2057 		/*
2058 		 * Remove out of band data so doesn't get presented to user.
2059 		 * This can happen independent of advancing the URG pointer,
2060 		 * but if two URG's are pending at once, some out-of-band
2061 		 * data may creep in... ick.
2062 		 */
2063 		if (th->th_urp <= (u_int16_t) tlen
2064 #ifdef SO_OOBINLINE
2065 		     && (so->so_options & SO_OOBINLINE) == 0
2066 #endif
2067 		     )
2068 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2069 	} else
2070 		/*
2071 		 * If no out of band data is expected,
2072 		 * pull receive urgent pointer along
2073 		 * with the receive window.
2074 		 */
2075 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2076 			tp->rcv_up = tp->rcv_nxt;
2077 dodata:							/* XXX */
2078 
2079 	/*
2080 	 * Process the segment text, merging it into the TCP sequencing queue,
2081 	 * and arranging for acknowledgment of receipt if necessary.
2082 	 * This process logically involves adjusting tp->rcv_wnd as data
2083 	 * is presented to the user (this happens in tcp_usrreq.c,
2084 	 * case PRU_RCVD).  If a FIN has already been received on this
2085 	 * connection then we just ignore the text.
2086 	 */
2087 	if ((tlen || (tiflags & TH_FIN)) &&
2088 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2089 #ifdef TCP_SACK
2090 		tcp_seq laststart = th->th_seq;
2091 		tcp_seq lastend = th->th_seq + tlen;
2092 #endif
2093 		if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) &&
2094 		    tp->t_state == TCPS_ESTABLISHED) {
2095 			TCP_SETUP_ACK(tp, tiflags, m);
2096 			tp->rcv_nxt += tlen;
2097 			tiflags = th->th_flags & TH_FIN;
2098 			tcpstat.tcps_rcvpack++;
2099 			tcpstat.tcps_rcvbyte += tlen;
2100 			ND6_HINT(tp);
2101 			if (so->so_state & SS_CANTRCVMORE)
2102 				m_freem(m);
2103 			else {
2104 				m_adj(m, hdroptlen);
2105 				sbappendstream(&so->so_rcv, m);
2106 			}
2107 			tp->t_flags |= TF_BLOCKOUTPUT;
2108 			sorwakeup(so);
2109 			tp->t_flags &= ~TF_BLOCKOUTPUT;
2110 		} else {
2111 			m_adj(m, hdroptlen);
2112 			tiflags = tcp_reass(tp, th, m, &tlen);
2113 			tp->t_flags |= TF_ACKNOW;
2114 		}
2115 #ifdef TCP_SACK
2116 		if (tp->sack_enable)
2117 			tcp_update_sack_list(tp, laststart, lastend);
2118 #endif
2119 
2120 		/*
2121 		 * variable len never referenced again in modern BSD,
2122 		 * so why bother computing it ??
2123 		 */
2124 #if 0
2125 		/*
2126 		 * Note the amount of data that peer has sent into
2127 		 * our window, in order to estimate the sender's
2128 		 * buffer size.
2129 		 */
2130 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2131 #endif /* 0 */
2132 	} else {
2133 		m_freem(m);
2134 		tiflags &= ~TH_FIN;
2135 	}
2136 
2137 	/*
2138 	 * If FIN is received ACK the FIN and let the user know
2139 	 * that the connection is closing.  Ignore a FIN received before
2140 	 * the connection is fully established.
2141 	 */
2142 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2143 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2144 			socantrcvmore(so);
2145 			tp->t_flags |= TF_ACKNOW;
2146 			tp->rcv_nxt++;
2147 		}
2148 		switch (tp->t_state) {
2149 
2150 		/*
2151 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2152 		 */
2153 		case TCPS_ESTABLISHED:
2154 			tp->t_state = TCPS_CLOSE_WAIT;
2155 			break;
2156 
2157 		/*
2158 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2159 		 * enter the CLOSING state.
2160 		 */
2161 		case TCPS_FIN_WAIT_1:
2162 			tp->t_state = TCPS_CLOSING;
2163 			break;
2164 
2165 		/*
2166 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2167 		 * starting the time-wait timer, turning off the other
2168 		 * standard timers.
2169 		 */
2170 		case TCPS_FIN_WAIT_2:
2171 			tp->t_state = TCPS_TIME_WAIT;
2172 			tcp_canceltimers(tp);
2173 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2174 			soisdisconnected(so);
2175 			break;
2176 
2177 		/*
2178 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2179 		 */
2180 		case TCPS_TIME_WAIT:
2181 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2182 			break;
2183 		}
2184 	}
2185 	if (so->so_options & SO_DEBUG) {
2186 		switch (tp->pf) {
2187 #ifdef INET6
2188 		case PF_INET6:
2189 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2190 			    0, tlen);
2191 			break;
2192 #endif /* INET6 */
2193 		case PF_INET:
2194 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2195 			    0, tlen);
2196 			break;
2197 		}
2198 	}
2199 
2200 	/*
2201 	 * Return any desired output.
2202 	 */
2203 	if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT))
2204 		(void) tcp_output(tp);
2205 	return;
2206 
2207 badsyn:
2208 	/*
2209 	 * Received a bad SYN.  Increment counters and dropwithreset.
2210 	 */
2211 	tcpstat.tcps_badsyn++;
2212 	tp = NULL;
2213 	goto dropwithreset;
2214 
2215 dropafterack_ratelim:
2216 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2217 	    tcp_ackdrop_ppslim) == 0) {
2218 		/* XXX stat */
2219 		goto drop;
2220 	}
2221 	/* ...fall into dropafterack... */
2222 
2223 dropafterack:
2224 	/*
2225 	 * Generate an ACK dropping incoming segment if it occupies
2226 	 * sequence space, where the ACK reflects our state.
2227 	 */
2228 	if (tiflags & TH_RST)
2229 		goto drop;
2230 	m_freem(m);
2231 	tp->t_flags |= TF_ACKNOW;
2232 	(void) tcp_output(tp);
2233 	return;
2234 
2235 dropwithreset_ratelim:
2236 	/*
2237 	 * We may want to rate-limit RSTs in certain situations,
2238 	 * particularly if we are sending an RST in response to
2239 	 * an attempt to connect to or otherwise communicate with
2240 	 * a port for which we have no socket.
2241 	 */
2242 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2243 	    tcp_rst_ppslim) == 0) {
2244 		/* XXX stat */
2245 		goto drop;
2246 	}
2247 	/* ...fall into dropwithreset... */
2248 
2249 dropwithreset:
2250 	/*
2251 	 * Generate a RST, dropping incoming segment.
2252 	 * Make ACK acceptable to originator of segment.
2253 	 * Don't bother to respond to RST.
2254 	 */
2255 	if (tiflags & TH_RST)
2256 		goto drop;
2257 	if (tiflags & TH_ACK) {
2258 		tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack,
2259 		    TH_RST, m->m_pkthdr.rdomain);
2260 	} else {
2261 		if (tiflags & TH_SYN)
2262 			tlen++;
2263 		tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen,
2264 		    (tcp_seq)0, TH_RST|TH_ACK, m->m_pkthdr.rdomain);
2265 	}
2266 	m_freem(m);
2267 	return;
2268 
2269 drop:
2270 	/*
2271 	 * Drop space held by incoming segment and return.
2272 	 */
2273 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2274 		switch (tp->pf) {
2275 #ifdef INET6
2276 		case PF_INET6:
2277 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2278 			    0, tlen);
2279 			break;
2280 #endif /* INET6 */
2281 		case PF_INET:
2282 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2283 			    0, tlen);
2284 			break;
2285 		}
2286 	}
2287 
2288 	m_freem(m);
2289 	return;
2290 }
2291 
2292 int
2293 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th,
2294     struct mbuf *m, int iphlen, struct tcp_opt_info *oi,
2295     u_int rtableid)
2296 {
2297 	u_int16_t mss = 0;
2298 	int opt, optlen;
2299 #ifdef TCP_SIGNATURE
2300 	caddr_t sigp = NULL;
2301 	struct tdb *tdb = NULL;
2302 #endif /* TCP_SIGNATURE */
2303 
2304 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2305 		opt = cp[0];
2306 		if (opt == TCPOPT_EOL)
2307 			break;
2308 		if (opt == TCPOPT_NOP)
2309 			optlen = 1;
2310 		else {
2311 			if (cnt < 2)
2312 				break;
2313 			optlen = cp[1];
2314 			if (optlen < 2 || optlen > cnt)
2315 				break;
2316 		}
2317 		switch (opt) {
2318 
2319 		default:
2320 			continue;
2321 
2322 		case TCPOPT_MAXSEG:
2323 			if (optlen != TCPOLEN_MAXSEG)
2324 				continue;
2325 			if (!(th->th_flags & TH_SYN))
2326 				continue;
2327 			if (TCPS_HAVERCVDSYN(tp->t_state))
2328 				continue;
2329 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2330 			NTOHS(mss);
2331 			oi->maxseg = mss;
2332 			break;
2333 
2334 		case TCPOPT_WINDOW:
2335 			if (optlen != TCPOLEN_WINDOW)
2336 				continue;
2337 			if (!(th->th_flags & TH_SYN))
2338 				continue;
2339 			if (TCPS_HAVERCVDSYN(tp->t_state))
2340 				continue;
2341 			tp->t_flags |= TF_RCVD_SCALE;
2342 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2343 			break;
2344 
2345 		case TCPOPT_TIMESTAMP:
2346 			if (optlen != TCPOLEN_TIMESTAMP)
2347 				continue;
2348 			oi->ts_present = 1;
2349 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2350 			NTOHL(oi->ts_val);
2351 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2352 			NTOHL(oi->ts_ecr);
2353 
2354 			if (!(th->th_flags & TH_SYN))
2355 				continue;
2356 			if (TCPS_HAVERCVDSYN(tp->t_state))
2357 				continue;
2358 			/*
2359 			 * A timestamp received in a SYN makes
2360 			 * it ok to send timestamp requests and replies.
2361 			 */
2362 			tp->t_flags |= TF_RCVD_TSTMP;
2363 			tp->ts_recent = oi->ts_val;
2364 			tp->ts_recent_age = tcp_now;
2365 			break;
2366 
2367 #ifdef TCP_SACK
2368 		case TCPOPT_SACK_PERMITTED:
2369 			if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED)
2370 				continue;
2371 			if (!(th->th_flags & TH_SYN))
2372 				continue;
2373 			if (TCPS_HAVERCVDSYN(tp->t_state))
2374 				continue;
2375 			/* MUST only be set on SYN */
2376 			tp->t_flags |= TF_SACK_PERMIT;
2377 			break;
2378 		case TCPOPT_SACK:
2379 			tcp_sack_option(tp, th, cp, optlen);
2380 			break;
2381 #endif
2382 #ifdef TCP_SIGNATURE
2383 		case TCPOPT_SIGNATURE:
2384 			if (optlen != TCPOLEN_SIGNATURE)
2385 				continue;
2386 
2387 			if (sigp && timingsafe_bcmp(sigp, cp + 2, 16))
2388 				return (-1);
2389 
2390 			sigp = cp + 2;
2391 			break;
2392 #endif /* TCP_SIGNATURE */
2393 		}
2394 	}
2395 
2396 #ifdef TCP_SIGNATURE
2397 	if (tp->t_flags & TF_SIGNATURE) {
2398 		union sockaddr_union src, dst;
2399 
2400 		memset(&src, 0, sizeof(union sockaddr_union));
2401 		memset(&dst, 0, sizeof(union sockaddr_union));
2402 
2403 		switch (tp->pf) {
2404 		case 0:
2405 #ifdef INET
2406 		case AF_INET:
2407 			src.sa.sa_len = sizeof(struct sockaddr_in);
2408 			src.sa.sa_family = AF_INET;
2409 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
2410 			dst.sa.sa_len = sizeof(struct sockaddr_in);
2411 			dst.sa.sa_family = AF_INET;
2412 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
2413 			break;
2414 #endif
2415 #ifdef INET6
2416 		case AF_INET6:
2417 			src.sa.sa_len = sizeof(struct sockaddr_in6);
2418 			src.sa.sa_family = AF_INET6;
2419 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
2420 			dst.sa.sa_len = sizeof(struct sockaddr_in6);
2421 			dst.sa.sa_family = AF_INET6;
2422 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
2423 			break;
2424 #endif /* INET6 */
2425 		}
2426 
2427 		tdb = gettdbbysrcdst(rtable_l2(rtableid),
2428 		    0, &src, &dst, IPPROTO_TCP);
2429 
2430 		/*
2431 		 * We don't have an SA for this peer, so we turn off
2432 		 * TF_SIGNATURE on the listen socket
2433 		 */
2434 		if (tdb == NULL && tp->t_state == TCPS_LISTEN)
2435 			tp->t_flags &= ~TF_SIGNATURE;
2436 
2437 	}
2438 
2439 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
2440 		tcpstat.tcps_rcvbadsig++;
2441 		return (-1);
2442 	}
2443 
2444 	if (sigp) {
2445 		char sig[16];
2446 
2447 		if (tdb == NULL) {
2448 			tcpstat.tcps_rcvbadsig++;
2449 			return (-1);
2450 		}
2451 
2452 		if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0)
2453 			return (-1);
2454 
2455 		if (timingsafe_bcmp(sig, sigp, 16)) {
2456 			tcpstat.tcps_rcvbadsig++;
2457 			return (-1);
2458 		}
2459 
2460 		tcpstat.tcps_rcvgoodsig++;
2461 	}
2462 #endif /* TCP_SIGNATURE */
2463 
2464 	return (0);
2465 }
2466 
2467 #if defined(TCP_SACK)
2468 u_long
2469 tcp_seq_subtract(u_long a, u_long b)
2470 {
2471 	return ((long)(a - b));
2472 }
2473 #endif
2474 
2475 
2476 #ifdef TCP_SACK
2477 /*
2478  * This function is called upon receipt of new valid data (while not in header
2479  * prediction mode), and it updates the ordered list of sacks.
2480  */
2481 void
2482 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart,
2483     tcp_seq rcv_lastend)
2484 {
2485 	/*
2486 	 * First reported block MUST be the most recent one.  Subsequent
2487 	 * blocks SHOULD be in the order in which they arrived at the
2488 	 * receiver.  These two conditions make the implementation fully
2489 	 * compliant with RFC 2018.
2490 	 */
2491 	int i, j = 0, count = 0, lastpos = -1;
2492 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2493 
2494 	/* First clean up current list of sacks */
2495 	for (i = 0; i < tp->rcv_numsacks; i++) {
2496 		sack = tp->sackblks[i];
2497 		if (sack.start == 0 && sack.end == 0) {
2498 			count++; /* count = number of blocks to be discarded */
2499 			continue;
2500 		}
2501 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2502 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2503 			count++;
2504 		} else {
2505 			temp[j].start = tp->sackblks[i].start;
2506 			temp[j++].end = tp->sackblks[i].end;
2507 		}
2508 	}
2509 	tp->rcv_numsacks -= count;
2510 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2511 		tcp_clean_sackreport(tp);
2512 		if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) {
2513 			/* ==> need first sack block */
2514 			tp->sackblks[0].start = rcv_laststart;
2515 			tp->sackblks[0].end = rcv_lastend;
2516 			tp->rcv_numsacks = 1;
2517 		}
2518 		return;
2519 	}
2520 	/* Otherwise, sack blocks are already present. */
2521 	for (i = 0; i < tp->rcv_numsacks; i++)
2522 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2523 	if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend))
2524 		return;     /* sack list remains unchanged */
2525 	/*
2526 	 * From here, segment just received should be (part of) the 1st sack.
2527 	 * Go through list, possibly coalescing sack block entries.
2528 	 */
2529 	firstsack.start = rcv_laststart;
2530 	firstsack.end = rcv_lastend;
2531 	for (i = 0; i < tp->rcv_numsacks; i++) {
2532 		sack = tp->sackblks[i];
2533 		if (SEQ_LT(sack.end, firstsack.start) ||
2534 		    SEQ_GT(sack.start, firstsack.end))
2535 			continue; /* no overlap */
2536 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2537 			/*
2538 			 * identical block; delete it here since we will
2539 			 * move it to the front of the list.
2540 			 */
2541 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2542 			lastpos = i;    /* last posn with a zero entry */
2543 			continue;
2544 		}
2545 		if (SEQ_LEQ(sack.start, firstsack.start))
2546 			firstsack.start = sack.start; /* merge blocks */
2547 		if (SEQ_GEQ(sack.end, firstsack.end))
2548 			firstsack.end = sack.end;     /* merge blocks */
2549 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2550 		lastpos = i;    /* last posn with a zero entry */
2551 	}
2552 	if (lastpos != -1) {    /* at least one merge */
2553 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2554 			sack = tp->sackblks[i];
2555 			if (sack.start == 0 && sack.end == 0)
2556 				continue;
2557 			temp[j++] = sack;
2558 		}
2559 		tp->rcv_numsacks = j; /* including first blk (added later) */
2560 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2561 			tp->sackblks[i] = temp[i];
2562 	} else {        /* no merges -- shift sacks by 1 */
2563 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2564 			tp->rcv_numsacks++;
2565 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2566 			tp->sackblks[i] = tp->sackblks[i-1];
2567 	}
2568 	tp->sackblks[0] = firstsack;
2569 	return;
2570 }
2571 
2572 /*
2573  * Process the TCP SACK option.  tp->snd_holes is an ordered list
2574  * of holes (oldest to newest, in terms of the sequence space).
2575  */
2576 void
2577 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2578 {
2579 	int tmp_olen;
2580 	u_char *tmp_cp;
2581 	struct sackhole *cur, *p, *temp;
2582 
2583 	if (!tp->sack_enable)
2584 		return;
2585 	/* SACK without ACK doesn't make sense. */
2586 	if ((th->th_flags & TH_ACK) == 0)
2587 	       return;
2588 	/* Make sure the ACK on this segment is in [snd_una, snd_max]. */
2589 	if (SEQ_LT(th->th_ack, tp->snd_una) ||
2590 	    SEQ_GT(th->th_ack, tp->snd_max))
2591 		return;
2592 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2593 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2594 		return;
2595 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2596 	tmp_cp = cp + 2;
2597 	tmp_olen = optlen - 2;
2598 	tcpstat.tcps_sack_rcv_opts++;
2599 	if (tp->snd_numholes < 0)
2600 		tp->snd_numholes = 0;
2601 	if (tp->t_maxseg == 0)
2602 		panic("tcp_sack_option"); /* Should never happen */
2603 	while (tmp_olen > 0) {
2604 		struct sackblk sack;
2605 
2606 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2607 		NTOHL(sack.start);
2608 		bcopy(tmp_cp + sizeof(tcp_seq),
2609 		    (char *) &(sack.end), sizeof(tcp_seq));
2610 		NTOHL(sack.end);
2611 		tmp_olen -= TCPOLEN_SACK;
2612 		tmp_cp += TCPOLEN_SACK;
2613 		if (SEQ_LEQ(sack.end, sack.start))
2614 			continue; /* bad SACK fields */
2615 		if (SEQ_LEQ(sack.end, tp->snd_una))
2616 			continue; /* old block */
2617 #if defined(TCP_SACK) && defined(TCP_FACK)
2618 		/* Updates snd_fack.  */
2619 		if (SEQ_GT(sack.end, tp->snd_fack))
2620 			tp->snd_fack = sack.end;
2621 #endif /* TCP_FACK */
2622 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2623 			if (SEQ_LT(sack.start, th->th_ack))
2624 				continue;
2625 		}
2626 		if (SEQ_GT(sack.end, tp->snd_max))
2627 			continue;
2628 		if (tp->snd_holes == NULL) { /* first hole */
2629 			tp->snd_holes = (struct sackhole *)
2630 			    pool_get(&sackhl_pool, PR_NOWAIT);
2631 			if (tp->snd_holes == NULL) {
2632 				/* ENOBUFS, so ignore SACKed block for now*/
2633 				goto done;
2634 			}
2635 			cur = tp->snd_holes;
2636 			cur->start = th->th_ack;
2637 			cur->end = sack.start;
2638 			cur->rxmit = cur->start;
2639 			cur->next = NULL;
2640 			tp->snd_numholes = 1;
2641 			tp->rcv_lastsack = sack.end;
2642 			/*
2643 			 * dups is at least one.  If more data has been
2644 			 * SACKed, it can be greater than one.
2645 			 */
2646 			cur->dups = min(tcprexmtthresh,
2647 			    ((sack.end - cur->end)/tp->t_maxseg));
2648 			if (cur->dups < 1)
2649 				cur->dups = 1;
2650 			continue; /* with next sack block */
2651 		}
2652 		/* Go thru list of holes:  p = previous,  cur = current */
2653 		p = cur = tp->snd_holes;
2654 		while (cur) {
2655 			if (SEQ_LEQ(sack.end, cur->start))
2656 				/* SACKs data before the current hole */
2657 				break; /* no use going through more holes */
2658 			if (SEQ_GEQ(sack.start, cur->end)) {
2659 				/* SACKs data beyond the current hole */
2660 				cur->dups++;
2661 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2662 				    tcprexmtthresh)
2663 					cur->dups = tcprexmtthresh;
2664 				p = cur;
2665 				cur = cur->next;
2666 				continue;
2667 			}
2668 			if (SEQ_LEQ(sack.start, cur->start)) {
2669 				/* Data acks at least the beginning of hole */
2670 #if defined(TCP_SACK) && defined(TCP_FACK)
2671 				if (SEQ_GT(sack.end, cur->rxmit))
2672 					tp->retran_data -=
2673 				    	    tcp_seq_subtract(cur->rxmit,
2674 					    cur->start);
2675 				else
2676 					tp->retran_data -=
2677 					    tcp_seq_subtract(sack.end,
2678 					    cur->start);
2679 #endif /* TCP_FACK */
2680 				if (SEQ_GEQ(sack.end, cur->end)) {
2681 					/* Acks entire hole, so delete hole */
2682 					if (p != cur) {
2683 						p->next = cur->next;
2684 						pool_put(&sackhl_pool, cur);
2685 						cur = p->next;
2686 					} else {
2687 						cur = cur->next;
2688 						pool_put(&sackhl_pool, p);
2689 						p = cur;
2690 						tp->snd_holes = p;
2691 					}
2692 					tp->snd_numholes--;
2693 					continue;
2694 				}
2695 				/* otherwise, move start of hole forward */
2696 				cur->start = sack.end;
2697 				cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
2698 				p = cur;
2699 				cur = cur->next;
2700 				continue;
2701 			}
2702 			/* move end of hole backward */
2703 			if (SEQ_GEQ(sack.end, cur->end)) {
2704 #if defined(TCP_SACK) && defined(TCP_FACK)
2705 				if (SEQ_GT(cur->rxmit, sack.start))
2706 					tp->retran_data -=
2707 					    tcp_seq_subtract(cur->rxmit,
2708 					    sack.start);
2709 #endif /* TCP_FACK */
2710 				cur->end = sack.start;
2711 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2712 				cur->dups++;
2713 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2714 				    tcprexmtthresh)
2715 					cur->dups = tcprexmtthresh;
2716 				p = cur;
2717 				cur = cur->next;
2718 				continue;
2719 			}
2720 			if (SEQ_LT(cur->start, sack.start) &&
2721 			    SEQ_GT(cur->end, sack.end)) {
2722 				/*
2723 				 * ACKs some data in middle of a hole; need to
2724 				 * split current hole
2725 				 */
2726 				temp = (struct sackhole *)
2727 				    pool_get(&sackhl_pool, PR_NOWAIT);
2728 				if (temp == NULL)
2729 					goto done; /* ENOBUFS */
2730 #if defined(TCP_SACK) && defined(TCP_FACK)
2731 				if (SEQ_GT(cur->rxmit, sack.end))
2732 					tp->retran_data -=
2733 					    tcp_seq_subtract(sack.end,
2734 					    sack.start);
2735 				else if (SEQ_GT(cur->rxmit, sack.start))
2736 					tp->retran_data -=
2737 					    tcp_seq_subtract(cur->rxmit,
2738 					    sack.start);
2739 #endif /* TCP_FACK */
2740 				temp->next = cur->next;
2741 				temp->start = sack.end;
2742 				temp->end = cur->end;
2743 				temp->dups = cur->dups;
2744 				temp->rxmit = SEQ_MAX(cur->rxmit, temp->start);
2745 				cur->end = sack.start;
2746 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2747 				cur->dups++;
2748 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2749 					tcprexmtthresh)
2750 					cur->dups = tcprexmtthresh;
2751 				cur->next = temp;
2752 				p = temp;
2753 				cur = p->next;
2754 				tp->snd_numholes++;
2755 			}
2756 		}
2757 		/* At this point, p points to the last hole on the list */
2758 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2759 			/*
2760 			 * Need to append new hole at end.
2761 			 * Last hole is p (and it's not NULL).
2762 			 */
2763 			temp = (struct sackhole *)
2764 			    pool_get(&sackhl_pool, PR_NOWAIT);
2765 			if (temp == NULL)
2766 				goto done; /* ENOBUFS */
2767 			temp->start = tp->rcv_lastsack;
2768 			temp->end = sack.start;
2769 			temp->dups = min(tcprexmtthresh,
2770 			    ((sack.end - sack.start)/tp->t_maxseg));
2771 			if (temp->dups < 1)
2772 				temp->dups = 1;
2773 			temp->rxmit = temp->start;
2774 			temp->next = 0;
2775 			p->next = temp;
2776 			tp->rcv_lastsack = sack.end;
2777 			tp->snd_numholes++;
2778 		}
2779 	}
2780 done:
2781 #if defined(TCP_SACK) && defined(TCP_FACK)
2782 	/*
2783 	 * Update retran_data and snd_awnd.  Go through the list of
2784 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2785 	 */
2786 	tp->retran_data = 0;
2787 	cur = tp->snd_holes;
2788 	while (cur) {
2789 		tp->retran_data += cur->rxmit - cur->start;
2790 		cur = cur->next;
2791 	}
2792 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2793 	    tp->retran_data;
2794 #endif /* TCP_FACK */
2795 
2796 	return;
2797 }
2798 
2799 /*
2800  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2801  * it is completely acked; otherwise, tcp_sack_option(), called from
2802  * tcp_dooptions(), will fix up the hole.
2803  */
2804 void
2805 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th)
2806 {
2807 	if (tp->sack_enable && tp->t_state != TCPS_LISTEN) {
2808 		/* max because this could be an older ack just arrived */
2809 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2810 			th->th_ack : tp->snd_una;
2811 		struct sackhole *cur = tp->snd_holes;
2812 		struct sackhole *prev;
2813 		while (cur)
2814 			if (SEQ_LEQ(cur->end, lastack)) {
2815 				prev = cur;
2816 				cur = cur->next;
2817 				pool_put(&sackhl_pool, prev);
2818 				tp->snd_numholes--;
2819 			} else if (SEQ_LT(cur->start, lastack)) {
2820 				cur->start = lastack;
2821 				if (SEQ_LT(cur->rxmit, cur->start))
2822 					cur->rxmit = cur->start;
2823 				break;
2824 			} else
2825 				break;
2826 		tp->snd_holes = cur;
2827 	}
2828 }
2829 
2830 /*
2831  * Delete all receiver-side SACK information.
2832  */
2833 void
2834 tcp_clean_sackreport(struct tcpcb *tp)
2835 {
2836 	int i;
2837 
2838 	tp->rcv_numsacks = 0;
2839 	for (i = 0; i < MAX_SACK_BLKS; i++)
2840 		tp->sackblks[i].start = tp->sackblks[i].end=0;
2841 
2842 }
2843 
2844 /*
2845  * Checks for partial ack.  If partial ack arrives, turn off retransmission
2846  * timer, deflate the window, do not clear tp->t_dupacks, and return 1.
2847  * If the ack advances at least to tp->snd_last, return 0.
2848  */
2849 int
2850 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th)
2851 {
2852 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2853 		/* Turn off retx. timer (will start again next segment) */
2854 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2855 		tp->t_rtttime = 0;
2856 #ifndef TCP_FACK
2857 		/*
2858 		 * Partial window deflation.  This statement relies on the
2859 		 * fact that tp->snd_una has not been updated yet.  In FACK
2860 		 * hold snd_cwnd constant during fast recovery.
2861 		 */
2862 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2863 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2864 			tp->snd_cwnd += tp->t_maxseg;
2865 		} else
2866 			tp->snd_cwnd = tp->t_maxseg;
2867 #endif
2868 		return (1);
2869 	}
2870 	return (0);
2871 }
2872 #endif /* TCP_SACK */
2873 
2874 /*
2875  * Pull out of band byte out of a segment so
2876  * it doesn't appear in the user's data queue.
2877  * It is still reflected in the segment length for
2878  * sequencing purposes.
2879  */
2880 void
2881 tcp_pulloutofband(struct socket *so, u_int urgent, struct mbuf *m, int off)
2882 {
2883         int cnt = off + urgent - 1;
2884 
2885 	while (cnt >= 0) {
2886 		if (m->m_len > cnt) {
2887 			char *cp = mtod(m, caddr_t) + cnt;
2888 			struct tcpcb *tp = sototcpcb(so);
2889 
2890 			tp->t_iobc = *cp;
2891 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2892 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2893 			m->m_len--;
2894 			return;
2895 		}
2896 		cnt -= m->m_len;
2897 		m = m->m_next;
2898 		if (m == 0)
2899 			break;
2900 	}
2901 	panic("tcp_pulloutofband");
2902 }
2903 
2904 /*
2905  * Collect new round-trip time estimate
2906  * and update averages and current timeout.
2907  */
2908 void
2909 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2910 {
2911 	short delta;
2912 	short rttmin;
2913 
2914 	if (rtt < 0)
2915 		rtt = 0;
2916 	else if (rtt > TCP_RTT_MAX)
2917 		rtt = TCP_RTT_MAX;
2918 
2919 	tcpstat.tcps_rttupdated++;
2920 	if (tp->t_srtt != 0) {
2921 		/*
2922 		 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits
2923 		 * after the binary point (scaled by 4), whereas
2924 		 * srtt is stored as fixed point with 5 bits after the
2925 		 * binary point (i.e., scaled by 32).  The following magic
2926 		 * is equivalent to the smoothing algorithm in rfc793 with
2927 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2928 		 * point).
2929 		 */
2930 		delta = (rtt << TCP_RTT_BASE_SHIFT) -
2931 		    (tp->t_srtt >> TCP_RTT_SHIFT);
2932 		if ((tp->t_srtt += delta) <= 0)
2933 			tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT;
2934 		/*
2935 		 * We accumulate a smoothed rtt variance (actually, a
2936 		 * smoothed mean difference), then set the retransmit
2937 		 * timer to smoothed rtt + 4 times the smoothed variance.
2938 		 * rttvar is stored as fixed point with 4 bits after the
2939 		 * binary point (scaled by 16).  The following is
2940 		 * equivalent to rfc793 smoothing with an alpha of .75
2941 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2942 		 * rfc793's wired-in beta.
2943 		 */
2944 		if (delta < 0)
2945 			delta = -delta;
2946 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2947 		if ((tp->t_rttvar += delta) <= 0)
2948 			tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT;
2949 	} else {
2950 		/*
2951 		 * No rtt measurement yet - use the unsmoothed rtt.
2952 		 * Set the variance to half the rtt (so our first
2953 		 * retransmit happens at 3*rtt).
2954 		 */
2955 		tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
2956 		tp->t_rttvar = (rtt + 1) <<
2957 		    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
2958 	}
2959 	tp->t_rtttime = 0;
2960 	tp->t_rxtshift = 0;
2961 
2962 	/*
2963 	 * the retransmit should happen at rtt + 4 * rttvar.
2964 	 * Because of the way we do the smoothing, srtt and rttvar
2965 	 * will each average +1/2 tick of bias.  When we compute
2966 	 * the retransmit timer, we want 1/2 tick of rounding and
2967 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2968 	 * firing of the timer.  The bias will give us exactly the
2969 	 * 1.5 tick we need.  But, because the bias is
2970 	 * statistical, we have to test that we don't drop below
2971 	 * the minimum feasible timer (which is 2 ticks).
2972 	 */
2973 	rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX);
2974 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2975 
2976 	/*
2977 	 * We received an ack for a packet that wasn't retransmitted;
2978 	 * it is probably safe to discard any error indications we've
2979 	 * received recently.  This isn't quite right, but close enough
2980 	 * for now (a route might have failed after we sent a segment,
2981 	 * and the return path might not be symmetrical).
2982 	 */
2983 	tp->t_softerror = 0;
2984 }
2985 
2986 /*
2987  * Determine a reasonable value for maxseg size.
2988  * If the route is known, check route for mtu.
2989  * If none, use an mss that can be handled on the outgoing
2990  * interface without forcing IP to fragment; if bigger than
2991  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2992  * to utilize large mbufs.  If no route is found, route has no mtu,
2993  * or the destination isn't local, use a default, hopefully conservative
2994  * size (usually 512 or the default IP max size, but no more than the mtu
2995  * of the interface), as we can't discover anything about intervening
2996  * gateways or networks.  We also initialize the congestion/slow start
2997  * window to be a single segment if the destination isn't local.
2998  * While looking at the routing entry, we also initialize other path-dependent
2999  * parameters from pre-set or cached values in the routing entry.
3000  *
3001  * Also take into account the space needed for options that we
3002  * send regularly.  Make maxseg shorter by that amount to assure
3003  * that we can send maxseg amount of data even when the options
3004  * are present.  Store the upper limit of the length of options plus
3005  * data in maxopd.
3006  *
3007  * NOTE: offer == -1 indicates that the maxseg size changed due to
3008  * Path MTU discovery.
3009  */
3010 int
3011 tcp_mss(struct tcpcb *tp, int offer)
3012 {
3013 	struct rtentry *rt;
3014 	struct ifnet *ifp;
3015 	int mss, mssopt;
3016 	int iphlen;
3017 	struct inpcb *inp;
3018 
3019 	inp = tp->t_inpcb;
3020 
3021 	mssopt = mss = tcp_mssdflt;
3022 
3023 	rt = in_pcbrtentry(inp);
3024 
3025 	if (rt == NULL)
3026 		goto out;
3027 
3028 	ifp = rt->rt_ifp;
3029 
3030 	switch (tp->pf) {
3031 #ifdef INET6
3032 	case AF_INET6:
3033 		iphlen = sizeof(struct ip6_hdr);
3034 		break;
3035 #endif
3036 	case AF_INET:
3037 		iphlen = sizeof(struct ip);
3038 		break;
3039 	default:
3040 		/* the family does not support path MTU discovery */
3041 		goto out;
3042 	}
3043 
3044 #ifdef RTV_MTU
3045 	/*
3046 	 * if there's an mtu associated with the route and we support
3047 	 * path MTU discovery for the underlying protocol family, use it.
3048 	 */
3049 	if (rt->rt_rmx.rmx_mtu) {
3050 		/*
3051 		 * One may wish to lower MSS to take into account options,
3052 		 * especially security-related options.
3053 		 */
3054 		if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) {
3055 			/*
3056 			 * RFC2460 section 5, last paragraph: if path MTU is
3057 			 * smaller than 1280, use 1280 as packet size and
3058 			 * attach fragment header.
3059 			 */
3060 			mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) -
3061 			    sizeof(struct tcphdr);
3062 		} else
3063 			mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr);
3064 	} else
3065 #endif /* RTV_MTU */
3066 	if (!ifp)
3067 		/*
3068 		 * ifp may be null and rmx_mtu may be zero in certain
3069 		 * v6 cases (e.g., if ND wasn't able to resolve the
3070 		 * destination host.
3071 		 */
3072 		goto out;
3073 	else if (ifp->if_flags & IFF_LOOPBACK)
3074 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3075 	else if (tp->pf == AF_INET) {
3076 		if (ip_mtudisc)
3077 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3078 		else if (inp && in_localaddr(inp->inp_faddr, inp->inp_rtableid))
3079 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3080 	}
3081 #ifdef INET6
3082 	else if (tp->pf == AF_INET6) {
3083 		/*
3084 		 * for IPv6, path MTU discovery is always turned on,
3085 		 * or the node must use packet size <= 1280.
3086 		 */
3087 		mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr);
3088 	}
3089 #endif /* INET6 */
3090 
3091 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3092 	if (offer != -1) {
3093 #ifndef INET6
3094 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3095 #else
3096 		if (tp->pf == AF_INET6)
3097 			mssopt = IN6_LINKMTU(ifp) - iphlen -
3098 			    sizeof(struct tcphdr);
3099 		else
3100 			mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3101 #endif
3102 
3103 		mssopt = max(tcp_mssdflt, mssopt);
3104 	}
3105 
3106  out:
3107 	/*
3108 	 * The current mss, t_maxseg, is initialized to the default value.
3109 	 * If we compute a smaller value, reduce the current mss.
3110 	 * If we compute a larger value, return it for use in sending
3111 	 * a max seg size option, but don't store it for use
3112 	 * unless we received an offer at least that large from peer.
3113 	 *
3114 	 * However, do not accept offers lower than the minimum of
3115 	 * the interface MTU and 216.
3116 	 */
3117 	if (offer > 0)
3118 		tp->t_peermss = offer;
3119 	if (tp->t_peermss)
3120 		mss = min(mss, max(tp->t_peermss, 216));
3121 
3122 	/* sanity - at least max opt. space */
3123 	mss = max(mss, 64);
3124 
3125 	/*
3126 	 * maxopd stores the maximum length of data AND options
3127 	 * in a segment; maxseg is the amount of data in a normal
3128 	 * segment.  We need to store this value (maxopd) apart
3129 	 * from maxseg, because now every segment carries options
3130 	 * and thus we normally have somewhat less data in segments.
3131 	 */
3132 	tp->t_maxopd = mss;
3133 
3134 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3135 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3136 		mss -= TCPOLEN_TSTAMP_APPA;
3137 #ifdef TCP_SIGNATURE
3138 	if (tp->t_flags & TF_SIGNATURE)
3139 		mss -= TCPOLEN_SIGLEN;
3140 #endif
3141 
3142 	if (offer == -1) {
3143 		/* mss changed due to Path MTU discovery */
3144 		tp->t_flags &= ~TF_PMTUD_PEND;
3145 		tp->t_pmtud_mtu_sent = 0;
3146 		tp->t_pmtud_mss_acked = 0;
3147 		if (mss < tp->t_maxseg) {
3148 			/*
3149 			 * Follow suggestion in RFC 2414 to reduce the
3150 			 * congestion window by the ratio of the old
3151 			 * segment size to the new segment size.
3152 			 */
3153 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3154 					     mss, mss);
3155 		}
3156 	} else if (tcp_do_rfc3390 == 2) {
3157 		/* increase initial window  */
3158 		tp->snd_cwnd = ulmin(10 * mss, ulmax(2 * mss, 14600));
3159 	} else if (tcp_do_rfc3390) {
3160 		/* increase initial window  */
3161 		tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380));
3162 	} else
3163 		tp->snd_cwnd = mss;
3164 
3165 	tp->t_maxseg = mss;
3166 
3167 	return (offer != -1 ? mssopt : mss);
3168 }
3169 
3170 u_int
3171 tcp_hdrsz(struct tcpcb *tp)
3172 {
3173 	u_int hlen;
3174 
3175 	switch (tp->pf) {
3176 #ifdef INET6
3177 	case AF_INET6:
3178 		hlen = sizeof(struct ip6_hdr);
3179 		break;
3180 #endif
3181 	case AF_INET:
3182 		hlen = sizeof(struct ip);
3183 		break;
3184 	default:
3185 		hlen = 0;
3186 		break;
3187 	}
3188 	hlen += sizeof(struct tcphdr);
3189 
3190 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3191 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3192 		hlen += TCPOLEN_TSTAMP_APPA;
3193 #ifdef TCP_SIGNATURE
3194 	if (tp->t_flags & TF_SIGNATURE)
3195 		hlen += TCPOLEN_SIGLEN;
3196 #endif
3197 	return (hlen);
3198 }
3199 
3200 /*
3201  * Set connection variables based on the effective MSS.
3202  * We are passed the TCPCB for the actual connection.  If we
3203  * are the server, we are called by the compressed state engine
3204  * when the 3-way handshake is complete.  If we are the client,
3205  * we are called when we receive the SYN,ACK from the server.
3206  *
3207  * NOTE: The t_maxseg value must be initialized in the TCPCB
3208  * before this routine is called!
3209  */
3210 void
3211 tcp_mss_update(struct tcpcb *tp)
3212 {
3213 	int mss;
3214 	u_long bufsize;
3215 	struct rtentry *rt;
3216 	struct socket *so;
3217 
3218 	so = tp->t_inpcb->inp_socket;
3219 	mss = tp->t_maxseg;
3220 
3221 	rt = in_pcbrtentry(tp->t_inpcb);
3222 
3223 	if (rt == NULL)
3224 		return;
3225 
3226 	bufsize = so->so_snd.sb_hiwat;
3227 	if (bufsize < mss) {
3228 		mss = bufsize;
3229 		/* Update t_maxseg and t_maxopd */
3230 		tcp_mss(tp, mss);
3231 	} else {
3232 		bufsize = roundup(bufsize, mss);
3233 		if (bufsize > sb_max)
3234 			bufsize = sb_max;
3235 		(void)sbreserve(&so->so_snd, bufsize);
3236 	}
3237 
3238 	bufsize = so->so_rcv.sb_hiwat;
3239 	if (bufsize > mss) {
3240 		bufsize = roundup(bufsize, mss);
3241 		if (bufsize > sb_max)
3242 			bufsize = sb_max;
3243 		(void)sbreserve(&so->so_rcv, bufsize);
3244 	}
3245 
3246 }
3247 
3248 #if defined (TCP_SACK)
3249 /*
3250  * Checks for partial ack.  If partial ack arrives, force the retransmission
3251  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3252  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3253  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3254  */
3255 int
3256 tcp_newreno(struct tcpcb *tp, struct tcphdr *th)
3257 {
3258 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3259 		/*
3260 		 * snd_una has not been updated and the socket send buffer
3261 		 * not yet drained of the acked data, so we have to leave
3262 		 * snd_una as it was to get the correct data offset in
3263 		 * tcp_output().
3264 		 */
3265 		tcp_seq onxt = tp->snd_nxt;
3266 		u_long  ocwnd = tp->snd_cwnd;
3267 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3268 		tp->t_rtttime = 0;
3269 		tp->snd_nxt = th->th_ack;
3270 		/*
3271 		 * Set snd_cwnd to one segment beyond acknowledged offset
3272 		 * (tp->snd_una not yet updated when this function is called)
3273 		 */
3274 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3275 		(void) tcp_output(tp);
3276 		tp->snd_cwnd = ocwnd;
3277 		if (SEQ_GT(onxt, tp->snd_nxt))
3278 			tp->snd_nxt = onxt;
3279 		/*
3280 		 * Partial window deflation.  Relies on fact that tp->snd_una
3281 		 * not updated yet.
3282 		 */
3283 		if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3284 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
3285 		else
3286 			tp->snd_cwnd = 0;
3287 		tp->snd_cwnd += tp->t_maxseg;
3288 
3289 		return 1;
3290 	}
3291 	return 0;
3292 }
3293 #endif /* TCP_SACK */
3294 
3295 int
3296 tcp_mss_adv(struct ifnet *ifp, int af)
3297 {
3298 	int mss = 0;
3299 	int iphlen;
3300 
3301 	switch (af) {
3302 	case AF_INET:
3303 		if (ifp != NULL)
3304 			mss = ifp->if_mtu;
3305 		iphlen = sizeof(struct ip);
3306 		break;
3307 #ifdef INET6
3308 	case AF_INET6:
3309 		if (ifp != NULL)
3310 			mss = IN6_LINKMTU(ifp);
3311 		iphlen = sizeof(struct ip6_hdr);
3312 		break;
3313 #endif
3314 	}
3315 	mss = mss - iphlen - sizeof(struct tcphdr);
3316 	return (max(mss, tcp_mssdflt));
3317 }
3318 
3319 /*
3320  * TCP compressed state engine.  Currently used to hold compressed
3321  * state for SYN_RECEIVED.
3322  */
3323 
3324 u_long	syn_cache_count;
3325 u_int32_t syn_hash1, syn_hash2;
3326 
3327 #define SYN_HASH(sa, sp, dp) \
3328 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
3329 				     ((u_int32_t)(sp)))^syn_hash2)))
3330 #ifndef INET6
3331 #define	SYN_HASHALL(hash, src, dst) \
3332 do {									\
3333 	hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr,	\
3334 		((struct sockaddr_in *)(src))->sin_port,		\
3335 		((struct sockaddr_in *)(dst))->sin_port);		\
3336 } while (/*CONSTCOND*/ 0)
3337 #else
3338 #define SYN_HASH6(sa, sp, dp) \
3339 	((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \
3340 	  (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \
3341 	 & 0x7fffffff)
3342 
3343 #define SYN_HASHALL(hash, src, dst) \
3344 do {									\
3345 	switch ((src)->sa_family) {					\
3346 	case AF_INET:							\
3347 		hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \
3348 			((struct sockaddr_in *)(src))->sin_port,	\
3349 			((struct sockaddr_in *)(dst))->sin_port);	\
3350 		break;							\
3351 	case AF_INET6:							\
3352 		hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \
3353 			((struct sockaddr_in6 *)(src))->sin6_port,	\
3354 			((struct sockaddr_in6 *)(dst))->sin6_port);	\
3355 		break;							\
3356 	default:							\
3357 		hash = 0;						\
3358 	}								\
3359 } while (/*CONSTCOND*/0)
3360 #endif /* INET6 */
3361 
3362 void
3363 syn_cache_rm(struct syn_cache *sc)
3364 {
3365 	sc->sc_flags |= SCF_DEAD;
3366 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3367 	    sc, sc_bucketq);
3368 	sc->sc_tp = NULL;
3369 	LIST_REMOVE(sc, sc_tpq);
3370 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3371 	timeout_del(&sc->sc_timer);
3372 	syn_cache_count--;
3373 }
3374 
3375 void
3376 syn_cache_put(struct syn_cache *sc)
3377 {
3378 	if (sc->sc_ipopts)
3379 		(void) m_free(sc->sc_ipopts);
3380 	if (sc->sc_route4.ro_rt != NULL)
3381 		RTFREE(sc->sc_route4.ro_rt);
3382 	timeout_set(&sc->sc_timer, syn_cache_reaper, sc);
3383 	timeout_add(&sc->sc_timer, 0);
3384 }
3385 
3386 struct pool syn_cache_pool;
3387 
3388 /*
3389  * We don't estimate RTT with SYNs, so each packet starts with the default
3390  * RTT and each timer step has a fixed timeout value.
3391  */
3392 #define	SYN_CACHE_TIMER_ARM(sc)						\
3393 do {									\
3394 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3395 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3396 	    TCPTV_REXMTMAX);						\
3397 	if (!timeout_initialized(&(sc)->sc_timer))			\
3398 		timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc));	\
3399 	timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \
3400 } while (/*CONSTCOND*/0)
3401 
3402 #define	SYN_CACHE_TIMESTAMP(sc)	tcp_now + (sc)->sc_modulate
3403 
3404 void
3405 syn_cache_init()
3406 {
3407 	int i;
3408 
3409 	/* Initialize the hash buckets. */
3410 	for (i = 0; i < tcp_syn_cache_size; i++)
3411 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3412 
3413 	/* Initialize the syn cache pool. */
3414 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3415 	    "synpl", NULL);
3416 }
3417 
3418 void
3419 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3420 {
3421 	struct syn_cache_head *scp;
3422 	struct syn_cache *sc2;
3423 	int s;
3424 
3425 	/*
3426 	 * If there are no entries in the hash table, reinitialize
3427 	 * the hash secrets.
3428 	 */
3429 	if (syn_cache_count == 0) {
3430 		syn_hash1 = arc4random();
3431 		syn_hash2 = arc4random();
3432 	}
3433 
3434 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3435 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3436 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3437 
3438 	/*
3439 	 * Make sure that we don't overflow the per-bucket
3440 	 * limit or the total cache size limit.
3441 	 */
3442 	s = splsoftnet();
3443 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3444 		tcpstat.tcps_sc_bucketoverflow++;
3445 		/*
3446 		 * The bucket is full.  Toss the oldest element in the
3447 		 * bucket.  This will be the first entry in the bucket.
3448 		 */
3449 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3450 #ifdef DIAGNOSTIC
3451 		/*
3452 		 * This should never happen; we should always find an
3453 		 * entry in our bucket.
3454 		 */
3455 		if (sc2 == NULL)
3456 			panic("syn_cache_insert: bucketoverflow: impossible");
3457 #endif
3458 		syn_cache_rm(sc2);
3459 		syn_cache_put(sc2);
3460 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
3461 		struct syn_cache_head *scp2, *sce;
3462 
3463 		tcpstat.tcps_sc_overflowed++;
3464 		/*
3465 		 * The cache is full.  Toss the oldest entry in the
3466 		 * first non-empty bucket we can find.
3467 		 *
3468 		 * XXX We would really like to toss the oldest
3469 		 * entry in the cache, but we hope that this
3470 		 * condition doesn't happen very often.
3471 		 */
3472 		scp2 = scp;
3473 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3474 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3475 			for (++scp2; scp2 != scp; scp2++) {
3476 				if (scp2 >= sce)
3477 					scp2 = &tcp_syn_cache[0];
3478 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3479 					break;
3480 			}
3481 #ifdef DIAGNOSTIC
3482 			/*
3483 			 * This should never happen; we should always find a
3484 			 * non-empty bucket.
3485 			 */
3486 			if (scp2 == scp)
3487 				panic("syn_cache_insert: cacheoverflow: "
3488 				    "impossible");
3489 #endif
3490 		}
3491 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3492 		syn_cache_rm(sc2);
3493 		syn_cache_put(sc2);
3494 	}
3495 
3496 	/*
3497 	 * Initialize the entry's timer.
3498 	 */
3499 	sc->sc_rxttot = 0;
3500 	sc->sc_rxtshift = 0;
3501 	SYN_CACHE_TIMER_ARM(sc);
3502 
3503 	/* Link it from tcpcb entry */
3504 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3505 
3506 	/* Put it into the bucket. */
3507 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3508 	scp->sch_length++;
3509 	syn_cache_count++;
3510 
3511 	tcpstat.tcps_sc_added++;
3512 	splx(s);
3513 }
3514 
3515 /*
3516  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3517  * If we have retransmitted an entry the maximum number of times, expire
3518  * that entry.
3519  */
3520 void
3521 syn_cache_timer(void *arg)
3522 {
3523 	struct syn_cache *sc = arg;
3524 	int s;
3525 
3526 	s = splsoftnet();
3527 	if (sc->sc_flags & SCF_DEAD) {
3528 		splx(s);
3529 		return;
3530 	}
3531 
3532 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3533 		/* Drop it -- too many retransmissions. */
3534 		goto dropit;
3535 	}
3536 
3537 	/*
3538 	 * Compute the total amount of time this entry has
3539 	 * been on a queue.  If this entry has been on longer
3540 	 * than the keep alive timer would allow, expire it.
3541 	 */
3542 	sc->sc_rxttot += sc->sc_rxtcur;
3543 	if (sc->sc_rxttot >= tcptv_keep_init)
3544 		goto dropit;
3545 
3546 	tcpstat.tcps_sc_retransmitted++;
3547 	(void) syn_cache_respond(sc, NULL);
3548 
3549 	/* Advance the timer back-off. */
3550 	sc->sc_rxtshift++;
3551 	SYN_CACHE_TIMER_ARM(sc);
3552 
3553 	splx(s);
3554 	return;
3555 
3556  dropit:
3557 	tcpstat.tcps_sc_timed_out++;
3558 	syn_cache_rm(sc);
3559 	syn_cache_put(sc);
3560 	splx(s);
3561 }
3562 
3563 void
3564 syn_cache_reaper(void *arg)
3565 {
3566 	struct syn_cache *sc = arg;
3567 	int s;
3568 
3569 	s = splsoftnet();
3570 	pool_put(&syn_cache_pool, (sc));
3571 	splx(s);
3572 	return;
3573 }
3574 
3575 /*
3576  * Remove syn cache created by the specified tcb entry,
3577  * because this does not make sense to keep them
3578  * (if there's no tcb entry, syn cache entry will never be used)
3579  */
3580 void
3581 syn_cache_cleanup(struct tcpcb *tp)
3582 {
3583 	struct syn_cache *sc, *nsc;
3584 	int s;
3585 
3586 	s = splsoftnet();
3587 
3588 	for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) {
3589 		nsc = LIST_NEXT(sc, sc_tpq);
3590 
3591 #ifdef DIAGNOSTIC
3592 		if (sc->sc_tp != tp)
3593 			panic("invalid sc_tp in syn_cache_cleanup");
3594 #endif
3595 		syn_cache_rm(sc);
3596 		syn_cache_put(sc);
3597 	}
3598 	/* just for safety */
3599 	LIST_INIT(&tp->t_sc);
3600 
3601 	splx(s);
3602 }
3603 
3604 /*
3605  * Find an entry in the syn cache.
3606  */
3607 struct syn_cache *
3608 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst,
3609     struct syn_cache_head **headp, u_int rtableid)
3610 {
3611 	struct syn_cache *sc;
3612 	struct syn_cache_head *scp;
3613 	u_int32_t hash;
3614 	int s;
3615 
3616 	SYN_HASHALL(hash, src, dst);
3617 
3618 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3619 	*headp = scp;
3620 	s = splsoftnet();
3621 	for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL;
3622 	     sc = TAILQ_NEXT(sc, sc_bucketq)) {
3623 		if (sc->sc_hash != hash)
3624 			continue;
3625 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3626 		    !bcmp(&sc->sc_dst, dst, dst->sa_len) &&
3627 		    rtable_l2(rtableid) == rtable_l2(sc->sc_rtableid)) {
3628 			splx(s);
3629 			return (sc);
3630 		}
3631 	}
3632 	splx(s);
3633 	return (NULL);
3634 }
3635 
3636 /*
3637  * This function gets called when we receive an ACK for a
3638  * socket in the LISTEN state.  We look up the connection
3639  * in the syn cache, and if its there, we pull it out of
3640  * the cache and turn it into a full-blown connection in
3641  * the SYN-RECEIVED state.
3642  *
3643  * The return values may not be immediately obvious, and their effects
3644  * can be subtle, so here they are:
3645  *
3646  *	NULL	SYN was not found in cache; caller should drop the
3647  *		packet and send an RST.
3648  *
3649  *	-1	We were unable to create the new connection, and are
3650  *		aborting it.  An ACK,RST is being sent to the peer
3651  *		(unless we got screwey sequence numbners; see below),
3652  *		because the 3-way handshake has been completed.  Caller
3653  *		should not free the mbuf, since we may be using it.  If
3654  *		we are not, we will free it.
3655  *
3656  *	Otherwise, the return value is a pointer to the new socket
3657  *	associated with the connection.
3658  */
3659 struct socket *
3660 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3661     u_int hlen, u_int tlen, struct socket *so, struct mbuf *m)
3662 {
3663 	struct syn_cache *sc;
3664 	struct syn_cache_head *scp;
3665 	struct inpcb *inp = NULL;
3666 	struct tcpcb *tp = NULL;
3667 	struct mbuf *am;
3668 	int s;
3669 	struct socket *oso;
3670 #if NPF > 0
3671 	struct pf_divert *divert = NULL;
3672 #endif
3673 
3674 	s = splsoftnet();
3675 	if ((sc = syn_cache_lookup(src, dst, &scp,
3676 	    sotoinpcb(so)->inp_rtableid)) == NULL) {
3677 		splx(s);
3678 		return (NULL);
3679 	}
3680 
3681 	/*
3682 	 * Verify the sequence and ack numbers.  Try getting the correct
3683 	 * response again.
3684 	 */
3685 	if ((th->th_ack != sc->sc_iss + 1) ||
3686 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3687 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3688 		(void) syn_cache_respond(sc, m);
3689 		splx(s);
3690 		return ((struct socket *)(-1));
3691 	}
3692 
3693 	/* Remove this cache entry */
3694 	syn_cache_rm(sc);
3695 	splx(s);
3696 
3697 	/*
3698 	 * Ok, create the full blown connection, and set things up
3699 	 * as they would have been set up if we had created the
3700 	 * connection when the SYN arrived.  If we can't create
3701 	 * the connection, abort it.
3702 	 */
3703 	oso = so;
3704 	so = sonewconn(so, SS_ISCONNECTED);
3705 	if (so == NULL)
3706 		goto resetandabort;
3707 
3708 	inp = sotoinpcb(oso);
3709 
3710 #ifdef IPSEC
3711 	/*
3712 	 * We need to copy the required security levels
3713 	 * from the old pcb. Ditto for any other
3714 	 * IPsec-related information.
3715 	 */
3716 	{
3717 	  struct inpcb *newinp = sotoinpcb(so);
3718 	  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
3719 		sizeof(inp->inp_seclevel));
3720 	  newinp->inp_secrequire = inp->inp_secrequire;
3721 	  if (inp->inp_ipo != NULL) {
3722 		  newinp->inp_ipo = inp->inp_ipo;
3723 		  inp->inp_ipo->ipo_ref_count++;
3724 	  }
3725 	  if (inp->inp_ipsec_remotecred != NULL) {
3726 		  newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred;
3727 		  inp->inp_ipsec_remotecred->ref_count++;
3728 	  }
3729 	  if (inp->inp_ipsec_remoteauth != NULL) {
3730 		  newinp->inp_ipsec_remoteauth
3731 		      = inp->inp_ipsec_remoteauth;
3732 		  inp->inp_ipsec_remoteauth->ref_count++;
3733 	  }
3734 	}
3735 #endif /* IPSEC */
3736 #ifdef INET6
3737 	/*
3738 	 * inp still has the OLD in_pcb stuff, set the
3739 	 * v6-related flags on the new guy, too.
3740 	 */
3741 	{
3742 	  int flags = inp->inp_flags;
3743 	  struct inpcb *oldinpcb = inp;
3744 
3745 	  inp = sotoinpcb(so);
3746 	  inp->inp_flags |= (flags & INP_IPV6);
3747 	  if ((inp->inp_flags & INP_IPV6) != 0) {
3748 	    inp->inp_ipv6.ip6_hlim =
3749 	      oldinpcb->inp_ipv6.ip6_hlim;
3750 	  }
3751 	}
3752 #else /* INET6 */
3753 	inp = sotoinpcb(so);
3754 #endif /* INET6 */
3755 
3756 #if NPF > 0
3757 	if (m && m->m_pkthdr.pf.flags & PF_TAG_DIVERTED &&
3758 	    (divert = pf_find_divert(m)) != NULL)
3759 		inp->inp_rtableid = divert->rdomain;
3760 	else
3761 #endif
3762 	/* inherit rtable from listening socket */
3763 	inp->inp_rtableid = sc->sc_rtableid;
3764 
3765 	inp->inp_lport = th->th_dport;
3766 	switch (src->sa_family) {
3767 #ifdef INET6
3768 	case AF_INET6:
3769 		inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr;
3770 		break;
3771 #endif /* INET6 */
3772 	case AF_INET:
3773 
3774 		inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr;
3775 		inp->inp_options = ip_srcroute();
3776 		if (inp->inp_options == NULL) {
3777 			inp->inp_options = sc->sc_ipopts;
3778 			sc->sc_ipopts = NULL;
3779 		}
3780 		break;
3781 	}
3782 	in_pcbrehash(inp);
3783 
3784 	/*
3785 	 * Give the new socket our cached route reference.
3786 	 */
3787 	if (src->sa_family == AF_INET)
3788 		inp->inp_route = sc->sc_route4;         /* struct assignment */
3789 #ifdef INET6
3790 	else
3791 		inp->inp_route6 = sc->sc_route6;
3792 #endif
3793 	sc->sc_route4.ro_rt = NULL;
3794 
3795 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3796 	if (am == NULL)
3797 		goto resetandabort;
3798 	am->m_len = src->sa_len;
3799 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3800 
3801 	switch (src->sa_family) {
3802 	case AF_INET:
3803 		/* drop IPv4 packet to AF_INET6 socket */
3804 		if (inp->inp_flags & INP_IPV6) {
3805 			(void) m_free(am);
3806 			goto resetandabort;
3807 		}
3808 		if (in_pcbconnect(inp, am)) {
3809 			(void) m_free(am);
3810 			goto resetandabort;
3811 		}
3812 		break;
3813 #ifdef INET6
3814 	case AF_INET6:
3815 		if (in6_pcbconnect(inp, am)) {
3816 			(void) m_free(am);
3817 			goto resetandabort;
3818 		}
3819 		break;
3820 #endif
3821 	}
3822 	(void) m_free(am);
3823 
3824 	tp = intotcpcb(inp);
3825 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3826 	if (sc->sc_request_r_scale != 15) {
3827 		tp->requested_s_scale = sc->sc_requested_s_scale;
3828 		tp->request_r_scale = sc->sc_request_r_scale;
3829 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3830 	}
3831 	if (sc->sc_flags & SCF_TIMESTAMP)
3832 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3833 
3834 	tp->t_template = tcp_template(tp);
3835 	if (tp->t_template == 0) {
3836 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3837 		so = NULL;
3838 		m_freem(m);
3839 		goto abort;
3840 	}
3841 #ifdef TCP_SACK
3842 	tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT;
3843 #endif
3844 
3845 	tp->ts_modulate = sc->sc_modulate;
3846 	tp->ts_recent = sc->sc_timestamp;
3847 	tp->iss = sc->sc_iss;
3848 	tp->irs = sc->sc_irs;
3849 	tcp_sendseqinit(tp);
3850 #if defined (TCP_SACK) || defined(TCP_ECN)
3851 	tp->snd_last = tp->snd_una;
3852 #endif /* TCP_SACK */
3853 #if defined(TCP_SACK) && defined(TCP_FACK)
3854 	tp->snd_fack = tp->snd_una;
3855 	tp->retran_data = 0;
3856 	tp->snd_awnd = 0;
3857 #endif /* TCP_FACK */
3858 #ifdef TCP_ECN
3859 	if (sc->sc_flags & SCF_ECN_PERMIT) {
3860 		tp->t_flags |= TF_ECN_PERMIT;
3861 		tcpstat.tcps_ecn_accepts++;
3862 	}
3863 #endif
3864 #ifdef TCP_SACK
3865 	if (sc->sc_flags & SCF_SACK_PERMIT)
3866 		tp->t_flags |= TF_SACK_PERMIT;
3867 #endif
3868 #ifdef TCP_SIGNATURE
3869 	if (sc->sc_flags & SCF_SIGNATURE)
3870 		tp->t_flags |= TF_SIGNATURE;
3871 #endif
3872 	tcp_rcvseqinit(tp);
3873 	tp->t_state = TCPS_SYN_RECEIVED;
3874 	tp->t_rcvtime = tcp_now;
3875 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
3876 	tcpstat.tcps_accepts++;
3877 
3878 	tcp_mss(tp, sc->sc_peermaxseg);	 /* sets t_maxseg */
3879 	if (sc->sc_peermaxseg)
3880 		tcp_mss_update(tp);
3881 	/* Reset initial window to 1 segment for retransmit */
3882 	if (sc->sc_rxtshift > 0)
3883 		tp->snd_cwnd = tp->t_maxseg;
3884 	tp->snd_wl1 = sc->sc_irs;
3885 	tp->rcv_up = sc->sc_irs + 1;
3886 
3887 	/*
3888 	 * This is what whould have happened in tcp_output() when
3889 	 * the SYN,ACK was sent.
3890 	 */
3891 	tp->snd_up = tp->snd_una;
3892 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3893 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3894 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3895 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3896 	tp->last_ack_sent = tp->rcv_nxt;
3897 
3898 	tcpstat.tcps_sc_completed++;
3899 	syn_cache_put(sc);
3900 	return (so);
3901 
3902 resetandabort:
3903 	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST,
3904 	    m->m_pkthdr.rdomain);
3905 	m_freem(m);
3906 abort:
3907 	if (so != NULL)
3908 		(void) soabort(so);
3909 	syn_cache_put(sc);
3910 	tcpstat.tcps_sc_aborted++;
3911 	return ((struct socket *)(-1));
3912 }
3913 
3914 /*
3915  * This function is called when we get a RST for a
3916  * non-existent connection, so that we can see if the
3917  * connection is in the syn cache.  If it is, zap it.
3918  */
3919 
3920 void
3921 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3922     u_int rtableid)
3923 {
3924 	struct syn_cache *sc;
3925 	struct syn_cache_head *scp;
3926 	int s = splsoftnet();
3927 
3928 	if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) {
3929 		splx(s);
3930 		return;
3931 	}
3932 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3933 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3934 		splx(s);
3935 		return;
3936 	}
3937 	syn_cache_rm(sc);
3938 	splx(s);
3939 	tcpstat.tcps_sc_reset++;
3940 	syn_cache_put(sc);
3941 }
3942 
3943 void
3944 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3945     u_int rtableid)
3946 {
3947 	struct syn_cache *sc;
3948 	struct syn_cache_head *scp;
3949 	int s;
3950 
3951 	s = splsoftnet();
3952 	if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) {
3953 		splx(s);
3954 		return;
3955 	}
3956 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3957 	if (ntohl (th->th_seq) != sc->sc_iss) {
3958 		splx(s);
3959 		return;
3960 	}
3961 
3962 	/*
3963 	 * If we've retransmitted 3 times and this is our second error,
3964 	 * we remove the entry.  Otherwise, we allow it to continue on.
3965 	 * This prevents us from incorrectly nuking an entry during a
3966 	 * spurious network outage.
3967 	 *
3968 	 * See tcp_notify().
3969 	 */
3970 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3971 		sc->sc_flags |= SCF_UNREACH;
3972 		splx(s);
3973 		return;
3974 	}
3975 
3976 	syn_cache_rm(sc);
3977 	splx(s);
3978 	tcpstat.tcps_sc_unreach++;
3979 	syn_cache_put(sc);
3980 }
3981 
3982 /*
3983  * Given a LISTEN socket and an inbound SYN request, add
3984  * this to the syn cache, and send back a segment:
3985  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3986  * to the source.
3987  *
3988  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3989  * Doing so would require that we hold onto the data and deliver it
3990  * to the application.  However, if we are the target of a SYN-flood
3991  * DoS attack, an attacker could send data which would eventually
3992  * consume all available buffer space if it were ACKed.  By not ACKing
3993  * the data, we avoid this DoS scenario.
3994  */
3995 
3996 int
3997 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3998     u_int iphlen, struct socket *so, struct mbuf *m, u_char *optp, int optlen,
3999     struct tcp_opt_info *oi, tcp_seq *issp)
4000 {
4001 	struct tcpcb tb, *tp;
4002 	long win;
4003 	struct syn_cache *sc;
4004 	struct syn_cache_head *scp;
4005 	struct mbuf *ipopts;
4006 
4007 	tp = sototcpcb(so);
4008 
4009 	/*
4010 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
4011 	 *
4012 	 * Note this check is performed in tcp_input() very early on.
4013 	 */
4014 
4015 	/*
4016 	 * Initialize some local state.
4017 	 */
4018 	win = sbspace(&so->so_rcv);
4019 	if (win > TCP_MAXWIN)
4020 		win = TCP_MAXWIN;
4021 
4022 #ifdef TCP_SIGNATURE
4023 	if (optp || (tp->t_flags & TF_SIGNATURE)) {
4024 #else
4025 	if (optp) {
4026 #endif
4027 		bzero(&tb, sizeof(tb));
4028 		tb.pf = tp->pf;
4029 #ifdef TCP_SACK
4030 		tb.sack_enable = tp->sack_enable;
4031 #endif
4032 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
4033 #ifdef TCP_SIGNATURE
4034 		if (tp->t_flags & TF_SIGNATURE)
4035 			tb.t_flags |= TF_SIGNATURE;
4036 #endif
4037 		tb.t_state = TCPS_LISTEN;
4038 		if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi,
4039 		    sotoinpcb(so)->inp_rtableid))
4040 			return (-1);
4041 	} else
4042 		tb.t_flags = 0;
4043 
4044 	switch (src->sa_family) {
4045 #ifdef INET
4046 	case AF_INET:
4047 		/*
4048 		 * Remember the IP options, if any.
4049 		 */
4050 		ipopts = ip_srcroute();
4051 		break;
4052 #endif
4053 	default:
4054 		ipopts = NULL;
4055 	}
4056 
4057 	/*
4058 	 * See if we already have an entry for this connection.
4059 	 * If we do, resend the SYN,ACK.  We do not count this
4060 	 * as a retransmission (XXX though maybe we should).
4061 	 */
4062 	if ((sc = syn_cache_lookup(src, dst, &scp, sotoinpcb(so)->inp_rtableid))
4063 	    != NULL) {
4064 		tcpstat.tcps_sc_dupesyn++;
4065 		if (ipopts) {
4066 			/*
4067 			 * If we were remembering a previous source route,
4068 			 * forget it and use the new one we've been given.
4069 			 */
4070 			if (sc->sc_ipopts)
4071 				(void) m_free(sc->sc_ipopts);
4072 			sc->sc_ipopts = ipopts;
4073 		}
4074 		sc->sc_timestamp = tb.ts_recent;
4075 		if (syn_cache_respond(sc, m) == 0) {
4076 			tcpstat.tcps_sndacks++;
4077 			tcpstat.tcps_sndtotal++;
4078 		}
4079 		return (0);
4080 	}
4081 
4082 	sc = pool_get(&syn_cache_pool, PR_NOWAIT|PR_ZERO);
4083 	if (sc == NULL) {
4084 		if (ipopts)
4085 			(void) m_free(ipopts);
4086 		return (-1);
4087 	}
4088 
4089 	/*
4090 	 * Fill in the cache, and put the necessary IP and TCP
4091 	 * options into the reply.
4092 	 */
4093 	bcopy(src, &sc->sc_src, src->sa_len);
4094 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4095 	sc->sc_rtableid = sotoinpcb(so)->inp_rtableid;
4096 	sc->sc_flags = 0;
4097 	sc->sc_ipopts = ipopts;
4098 	sc->sc_irs = th->th_seq;
4099 
4100 	sc->sc_iss = issp ? *issp : arc4random();
4101 	sc->sc_peermaxseg = oi->maxseg;
4102 	sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ?
4103 	    m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family);
4104 	sc->sc_win = win;
4105 	sc->sc_timestamp = tb.ts_recent;
4106 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4107 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
4108 		sc->sc_flags |= SCF_TIMESTAMP;
4109 		sc->sc_modulate = arc4random();
4110 	}
4111 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4112 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4113 		sc->sc_requested_s_scale = tb.requested_s_scale;
4114 		sc->sc_request_r_scale = 0;
4115 		/*
4116 		 * Pick the smallest possible scaling factor that
4117 		 * will still allow us to scale up to sb_max.
4118 		 *
4119 		 * We do this because there are broken firewalls that
4120 		 * will corrupt the window scale option, leading to
4121 		 * the other endpoint believing that our advertised
4122 		 * window is unscaled.  At scale factors larger than
4123 		 * 5 the unscaled window will drop below 1500 bytes,
4124 		 * leading to serious problems when traversing these
4125 		 * broken firewalls.
4126 		 *
4127 		 * With the default sbmax of 256K, a scale factor
4128 		 * of 3 will be chosen by this algorithm.  Those who
4129 		 * choose a larger sbmax should watch out
4130 		 * for the compatiblity problems mentioned above.
4131 		 *
4132 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4133 		 * or <SYN,ACK>) segment itself is never scaled.
4134 		 */
4135 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4136 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4137 			sc->sc_request_r_scale++;
4138 	} else {
4139 		sc->sc_requested_s_scale = 15;
4140 		sc->sc_request_r_scale = 15;
4141 	}
4142 #ifdef TCP_ECN
4143 	/*
4144 	 * if both ECE and CWR flag bits are set, peer is ECN capable.
4145 	 */
4146 	if (tcp_do_ecn &&
4147 	    (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR))
4148 		sc->sc_flags |= SCF_ECN_PERMIT;
4149 #endif
4150 #ifdef TCP_SACK
4151 	/*
4152 	 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option
4153 	 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT).
4154 	 */
4155 	if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT))
4156 		sc->sc_flags |= SCF_SACK_PERMIT;
4157 #endif
4158 #ifdef TCP_SIGNATURE
4159 	if (tb.t_flags & TF_SIGNATURE)
4160 		sc->sc_flags |= SCF_SIGNATURE;
4161 #endif
4162 	sc->sc_tp = tp;
4163 	if (syn_cache_respond(sc, m) == 0) {
4164 		syn_cache_insert(sc, tp);
4165 		tcpstat.tcps_sndacks++;
4166 		tcpstat.tcps_sndtotal++;
4167 	} else {
4168 		syn_cache_put(sc);
4169 		tcpstat.tcps_sc_dropped++;
4170 	}
4171 
4172 	return (0);
4173 }
4174 
4175 int
4176 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4177 {
4178 	struct route *ro;
4179 	u_int8_t *optp;
4180 	int optlen, error;
4181 	u_int16_t tlen;
4182 	struct ip *ip = NULL;
4183 #ifdef INET6
4184 	struct ip6_hdr *ip6 = NULL;
4185 #endif
4186 	struct tcphdr *th;
4187 	u_int hlen;
4188 	struct inpcb *inp;
4189 
4190 	switch (sc->sc_src.sa.sa_family) {
4191 	case AF_INET:
4192 		hlen = sizeof(struct ip);
4193 		ro = &sc->sc_route4;
4194 		break;
4195 #ifdef INET6
4196 	case AF_INET6:
4197 		hlen = sizeof(struct ip6_hdr);
4198 		ro = (struct route *)&sc->sc_route6;
4199 		break;
4200 #endif
4201 	default:
4202 		if (m)
4203 			m_freem(m);
4204 		return (EAFNOSUPPORT);
4205 	}
4206 
4207 	/* Compute the size of the TCP options. */
4208 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4209 #ifdef TCP_SACK
4210 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) +
4211 #endif
4212 #ifdef TCP_SIGNATURE
4213 	    ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) +
4214 #endif
4215 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4216 
4217 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4218 
4219 	/*
4220 	 * Create the IP+TCP header from scratch.
4221 	 */
4222 	if (m)
4223 		m_freem(m);
4224 #ifdef DIAGNOSTIC
4225 	if (max_linkhdr + tlen > MCLBYTES)
4226 		return (ENOBUFS);
4227 #endif
4228 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4229 	if (m && max_linkhdr + tlen > MHLEN) {
4230 		MCLGET(m, M_DONTWAIT);
4231 		if ((m->m_flags & M_EXT) == 0) {
4232 			m_freem(m);
4233 			m = NULL;
4234 		}
4235 	}
4236 	if (m == NULL)
4237 		return (ENOBUFS);
4238 
4239 	/* Fixup the mbuf. */
4240 	m->m_data += max_linkhdr;
4241 	m->m_len = m->m_pkthdr.len = tlen;
4242 	m->m_pkthdr.rcvif = NULL;
4243 	m->m_pkthdr.rdomain = sc->sc_rtableid;
4244 	memset(mtod(m, u_char *), 0, tlen);
4245 
4246 	switch (sc->sc_src.sa.sa_family) {
4247 	case AF_INET:
4248 		ip = mtod(m, struct ip *);
4249 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4250 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4251 		ip->ip_p = IPPROTO_TCP;
4252 		th = (struct tcphdr *)(ip + 1);
4253 		th->th_dport = sc->sc_src.sin.sin_port;
4254 		th->th_sport = sc->sc_dst.sin.sin_port;
4255 		break;
4256 #ifdef INET6
4257 	case AF_INET6:
4258 		ip6 = mtod(m, struct ip6_hdr *);
4259 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4260 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4261 		ip6->ip6_nxt = IPPROTO_TCP;
4262 		/* ip6_plen will be updated in ip6_output() */
4263 		th = (struct tcphdr *)(ip6 + 1);
4264 		th->th_dport = sc->sc_src.sin6.sin6_port;
4265 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4266 		break;
4267 #endif
4268 	default:
4269 		th = NULL;
4270 	}
4271 
4272 	th->th_seq = htonl(sc->sc_iss);
4273 	th->th_ack = htonl(sc->sc_irs + 1);
4274 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4275 	th->th_flags = TH_SYN|TH_ACK;
4276 #ifdef TCP_ECN
4277 	/* Set ECE for SYN-ACK if peer supports ECN. */
4278 	if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT))
4279 		th->th_flags |= TH_ECE;
4280 #endif
4281 	th->th_win = htons(sc->sc_win);
4282 	/* th_sum already 0 */
4283 	/* th_urp already 0 */
4284 
4285 	/* Tack on the TCP options. */
4286 	optp = (u_int8_t *)(th + 1);
4287 	*optp++ = TCPOPT_MAXSEG;
4288 	*optp++ = 4;
4289 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4290 	*optp++ = sc->sc_ourmaxseg & 0xff;
4291 
4292 #ifdef TCP_SACK
4293 	/* Include SACK_PERMIT_HDR option if peer has already done so. */
4294 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4295 		*((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR);
4296 		optp += 4;
4297 	}
4298 #endif
4299 
4300 	if (sc->sc_request_r_scale != 15) {
4301 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4302 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4303 		    sc->sc_request_r_scale);
4304 		optp += 4;
4305 	}
4306 
4307 	if (sc->sc_flags & SCF_TIMESTAMP) {
4308 		u_int32_t *lp = (u_int32_t *)(optp);
4309 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4310 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4311 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4312 		*lp   = htonl(sc->sc_timestamp);
4313 		optp += TCPOLEN_TSTAMP_APPA;
4314 	}
4315 
4316 #ifdef TCP_SIGNATURE
4317 	if (sc->sc_flags & SCF_SIGNATURE) {
4318 		union sockaddr_union src, dst;
4319 		struct tdb *tdb;
4320 
4321 		bzero(&src, sizeof(union sockaddr_union));
4322 		bzero(&dst, sizeof(union sockaddr_union));
4323 		src.sa.sa_len = sc->sc_src.sa.sa_len;
4324 		src.sa.sa_family = sc->sc_src.sa.sa_family;
4325 		dst.sa.sa_len = sc->sc_dst.sa.sa_len;
4326 		dst.sa.sa_family = sc->sc_dst.sa.sa_family;
4327 
4328 		switch (sc->sc_src.sa.sa_family) {
4329 		case 0:	/*default to PF_INET*/
4330 #ifdef INET
4331 		case AF_INET:
4332 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
4333 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
4334 			break;
4335 #endif /* INET */
4336 #ifdef INET6
4337 		case AF_INET6:
4338 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
4339 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
4340 			break;
4341 #endif /* INET6 */
4342 		}
4343 
4344 		tdb = gettdbbysrcdst(rtable_l2(sc->sc_rtableid),
4345 		    0, &src, &dst, IPPROTO_TCP);
4346 		if (tdb == NULL) {
4347 			if (m)
4348 				m_freem(m);
4349 			return (EPERM);
4350 		}
4351 
4352 		/* Send signature option */
4353 		*(optp++) = TCPOPT_SIGNATURE;
4354 		*(optp++) = TCPOLEN_SIGNATURE;
4355 
4356 		if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th,
4357 		    hlen, 0, optp) < 0) {
4358 			if (m)
4359 				m_freem(m);
4360 			return (EINVAL);
4361 		}
4362 		optp += 16;
4363 
4364 		/* Pad options list to the next 32 bit boundary and
4365 		 * terminate it.
4366 		 */
4367 		*optp++ = TCPOPT_NOP;
4368 		*optp++ = TCPOPT_EOL;
4369 	}
4370 #endif /* TCP_SIGNATURE */
4371 
4372 	/* Compute the packet's checksum. */
4373 	switch (sc->sc_src.sa.sa_family) {
4374 	case AF_INET:
4375 		ip->ip_len = htons(tlen - hlen);
4376 		th->th_sum = 0;
4377 		th->th_sum = in_cksum(m, tlen);
4378 		break;
4379 #ifdef INET6
4380 	case AF_INET6:
4381 		ip6->ip6_plen = htons(tlen - hlen);
4382 		th->th_sum = 0;
4383 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4384 		break;
4385 #endif
4386 	}
4387 
4388 	/* use IPsec policy and ttl from listening socket, on SYN ACK */
4389 	inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL;
4390 
4391 	/*
4392 	 * Fill in some straggling IP bits.  Note the stack expects
4393 	 * ip_len to be in host order, for convenience.
4394 	 */
4395 	switch (sc->sc_src.sa.sa_family) {
4396 #ifdef INET
4397 	case AF_INET:
4398 		ip->ip_len = htons(tlen);
4399 		ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl;
4400 		if (inp != NULL)
4401 			ip->ip_tos = inp->inp_ip.ip_tos;
4402 		break;
4403 #endif
4404 #ifdef INET6
4405 	case AF_INET6:
4406 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4407 		ip6->ip6_vfc |= IPV6_VERSION;
4408 		ip6->ip6_plen = htons(tlen - hlen);
4409 		/* ip6_hlim will be initialized afterwards */
4410 		/* leave flowlabel = 0, it is legal and require no state mgmt */
4411 		break;
4412 #endif
4413 	}
4414 
4415 	switch (sc->sc_src.sa.sa_family) {
4416 #ifdef INET
4417 	case AF_INET:
4418 		error = ip_output(m, sc->sc_ipopts, ro,
4419 		    (ip_mtudisc ? IP_MTUDISC : 0),
4420 		    (struct ip_moptions *)NULL, inp);
4421 		break;
4422 #endif
4423 #ifdef INET6
4424 	case AF_INET6:
4425 		ip6->ip6_hlim = in6_selecthlim(NULL,
4426 				ro->ro_rt ? ro->ro_rt->rt_ifp : NULL);
4427 
4428 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4429 			(struct ip6_moptions *)0, NULL, NULL);
4430 		break;
4431 #endif
4432 	default:
4433 		error = EAFNOSUPPORT;
4434 		break;
4435 	}
4436 	return (error);
4437 }
4438