xref: /openbsd-src/sys/netinet/tcp_input.c (revision 800a8ade7bb9667398282c53d4bb5e8ac8ee11be)
1 /*	$OpenBSD: tcp_input.c,v 1.306 2015/10/24 16:08:48 mpi 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 <net/if.h>
84 #include <net/if_var.h>
85 #include <net/route.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/ip.h>
89 #include <netinet/in_pcb.h>
90 #include <netinet/ip_var.h>
91 #include <netinet/tcp.h>
92 #include <netinet/tcp_fsm.h>
93 #include <netinet/tcp_seq.h>
94 #include <netinet/tcp_timer.h>
95 #include <netinet/tcp_var.h>
96 #include <netinet/tcpip.h>
97 #include <netinet/tcp_debug.h>
98 
99 #if NPF > 0
100 #include <net/pfvar.h>
101 #endif
102 
103 struct	tcpiphdr tcp_saveti;
104 
105 int tcp_mss_adv(struct mbuf *, int);
106 int tcp_flush_queue(struct tcpcb *);
107 
108 #ifdef INET6
109 #include <netinet6/in6_var.h>
110 #include <netinet6/nd6.h>
111 
112 struct  tcpipv6hdr tcp_saveti6;
113 
114 /* for the packet header length in the mbuf */
115 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
116 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
117 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
118 #endif /* INET6 */
119 
120 int	tcprexmtthresh = 3;
121 int	tcptv_keep_init = TCPTV_KEEP_INIT;
122 
123 int tcp_rst_ppslim = 100;		/* 100pps */
124 int tcp_rst_ppslim_count = 0;
125 struct timeval tcp_rst_ppslim_last;
126 
127 int tcp_ackdrop_ppslim = 100;		/* 100pps */
128 int tcp_ackdrop_ppslim_count = 0;
129 struct timeval tcp_ackdrop_ppslim_last;
130 
131 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
132 
133 /* for modulo comparisons of timestamps */
134 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
135 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
136 
137 /* for TCP SACK comparisons */
138 #define	SEQ_MIN(a,b)	(SEQ_LT(a,b) ? (a) : (b))
139 #define	SEQ_MAX(a,b)	(SEQ_GT(a,b) ? (a) : (b))
140 
141 /*
142  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
143  */
144 #ifdef INET6
145 #define ND6_HINT(tp) \
146 do { \
147 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
148 	    tp->t_inpcb->inp_route6.ro_rt) { \
149 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, \
150 		    tp->t_inpcb->inp_rtableid); \
151 	} \
152 } while (0)
153 #else
154 #define ND6_HINT(tp)
155 #endif
156 
157 #ifdef TCP_ECN
158 /*
159  * ECN (Explicit Congestion Notification) support based on RFC3168
160  * implementation note:
161  *   snd_last is used to track a recovery phase.
162  *   when cwnd is reduced, snd_last is set to snd_max.
163  *   while snd_last > snd_una, the sender is in a recovery phase and
164  *   its cwnd should not be reduced again.
165  *   snd_last follows snd_una when not in a recovery phase.
166  */
167 #endif
168 
169 /*
170  * Macro to compute ACK transmission behavior.  Delay the ACK unless
171  * we have already delayed an ACK (must send an ACK every two segments).
172  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
173  * option is enabled or when the packet is coming from a loopback
174  * interface.
175  */
176 #define	TCP_SETUP_ACK(tp, tiflags, m) \
177 do { \
178 	struct ifnet *ifp = NULL; \
179 	if (m && (m->m_flags & M_PKTHDR)) \
180 		ifp = if_get(m->m_pkthdr.ph_ifidx); \
181 	if ((tp)->t_flags & TF_DELACK || \
182 	    (tcp_ack_on_push && (tiflags) & TH_PUSH) || \
183 	    (ifp && (ifp->if_flags & IFF_LOOPBACK))) \
184 		tp->t_flags |= TF_ACKNOW; \
185 	else \
186 		TCP_SET_DELACK(tp); \
187 	if_put(ifp); \
188 } while (0)
189 
190 void syn_cache_put(struct syn_cache *);
191 void syn_cache_rm(struct syn_cache *);
192 
193 /*
194  * Insert segment ti into reassembly queue of tcp with
195  * control block tp.  Return TH_FIN if reassembly now includes
196  * a segment with FIN.  The macro form does the common case inline
197  * (segment is the next to be received on an established connection,
198  * and the queue is empty), avoiding linkage into and removal
199  * from the queue and repetition of various conversions.
200  * Set DELACK for segments received in order, but ack immediately
201  * when segments are out of order (so fast retransmit can work).
202  */
203 
204 int
205 tcp_reass(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m, int *tlen)
206 {
207 	struct tcpqent *p, *q, *nq, *tiqe;
208 
209 	/*
210 	 * Allocate a new queue entry, before we throw away any data.
211 	 * If we can't, just drop the packet.  XXX
212 	 */
213 	tiqe = pool_get(&tcpqe_pool, PR_NOWAIT);
214 	if (tiqe == NULL) {
215 		tiqe = TAILQ_LAST(&tp->t_segq, tcpqehead);
216 		if (tiqe != NULL && th->th_seq == tp->rcv_nxt) {
217 			/* Reuse last entry since new segment fills a hole */
218 			m_freem(tiqe->tcpqe_m);
219 			TAILQ_REMOVE(&tp->t_segq, tiqe, tcpqe_q);
220 		}
221 		if (tiqe == NULL || th->th_seq != tp->rcv_nxt) {
222 			/* Flush segment queue for this connection */
223 			tcp_freeq(tp);
224 			tcpstat.tcps_rcvmemdrop++;
225 			m_freem(m);
226 			return (0);
227 		}
228 	}
229 
230 	/*
231 	 * Find a segment which begins after this one does.
232 	 */
233 	for (p = NULL, q = TAILQ_FIRST(&tp->t_segq); q != NULL;
234 	    p = q, q = TAILQ_NEXT(q, tcpqe_q))
235 		if (SEQ_GT(q->tcpqe_tcp->th_seq, th->th_seq))
236 			break;
237 
238 	/*
239 	 * If there is a preceding segment, it may provide some of
240 	 * our data already.  If so, drop the data from the incoming
241 	 * segment.  If it provides all of our data, drop us.
242 	 */
243 	if (p != NULL) {
244 		struct tcphdr *phdr = p->tcpqe_tcp;
245 		int i;
246 
247 		/* conversion to int (in i) handles seq wraparound */
248 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
249 		if (i > 0) {
250 		        if (i >= *tlen) {
251 				tcpstat.tcps_rcvduppack++;
252 				tcpstat.tcps_rcvdupbyte += *tlen;
253 				m_freem(m);
254 				pool_put(&tcpqe_pool, tiqe);
255 				return (0);
256 			}
257 			m_adj(m, i);
258 			*tlen -= i;
259 			th->th_seq += i;
260 		}
261 	}
262 	tcpstat.tcps_rcvoopack++;
263 	tcpstat.tcps_rcvoobyte += *tlen;
264 
265 	/*
266 	 * While we overlap succeeding segments trim them or,
267 	 * if they are completely covered, dequeue them.
268 	 */
269 	for (; q != NULL; q = nq) {
270 		struct tcphdr *qhdr = q->tcpqe_tcp;
271 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
272 
273 		if (i <= 0)
274 			break;
275 		if (i < qhdr->th_reseqlen) {
276 			qhdr->th_seq += i;
277 			qhdr->th_reseqlen -= i;
278 			m_adj(q->tcpqe_m, i);
279 			break;
280 		}
281 		nq = TAILQ_NEXT(q, tcpqe_q);
282 		m_freem(q->tcpqe_m);
283 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
284 		pool_put(&tcpqe_pool, q);
285 	}
286 
287 	/* Insert the new segment queue entry into place. */
288 	tiqe->tcpqe_m = m;
289 	th->th_reseqlen = *tlen;
290 	tiqe->tcpqe_tcp = th;
291 	if (p == NULL) {
292 		TAILQ_INSERT_HEAD(&tp->t_segq, tiqe, tcpqe_q);
293 	} else {
294 		TAILQ_INSERT_AFTER(&tp->t_segq, p, tiqe, tcpqe_q);
295 	}
296 
297 	if (th->th_seq != tp->rcv_nxt)
298 		return (0);
299 
300 	return (tcp_flush_queue(tp));
301 }
302 
303 int
304 tcp_flush_queue(struct tcpcb *tp)
305 {
306 	struct socket *so = tp->t_inpcb->inp_socket;
307 	struct tcpqent *q, *nq;
308 	int flags;
309 
310 	/*
311 	 * Present data to user, advancing rcv_nxt through
312 	 * completed sequence space.
313 	 */
314 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
315 		return (0);
316 	q = TAILQ_FIRST(&tp->t_segq);
317 	if (q == NULL || q->tcpqe_tcp->th_seq != tp->rcv_nxt)
318 		return (0);
319 	if (tp->t_state == TCPS_SYN_RECEIVED && q->tcpqe_tcp->th_reseqlen)
320 		return (0);
321 	do {
322 		tp->rcv_nxt += q->tcpqe_tcp->th_reseqlen;
323 		flags = q->tcpqe_tcp->th_flags & TH_FIN;
324 
325 		nq = TAILQ_NEXT(q, tcpqe_q);
326 		TAILQ_REMOVE(&tp->t_segq, q, tcpqe_q);
327 		ND6_HINT(tp);
328 		if (so->so_state & SS_CANTRCVMORE)
329 			m_freem(q->tcpqe_m);
330 		else
331 			sbappendstream(&so->so_rcv, q->tcpqe_m);
332 		pool_put(&tcpqe_pool, q);
333 		q = nq;
334 	} while (q != NULL && q->tcpqe_tcp->th_seq == tp->rcv_nxt);
335 	tp->t_flags |= TF_BLOCKOUTPUT;
336 	sorwakeup(so);
337 	tp->t_flags &= ~TF_BLOCKOUTPUT;
338 	return (flags);
339 }
340 
341 #ifdef INET6
342 int
343 tcp6_input(struct mbuf **mp, int *offp, int proto)
344 {
345 	struct mbuf *m = *mp;
346 
347 	tcp_input(m, *offp, proto);
348 	return IPPROTO_DONE;
349 }
350 #endif
351 
352 /*
353  * TCP input routine, follows pages 65-76 of the
354  * protocol specification dated September, 1981 very closely.
355  */
356 void
357 tcp_input(struct mbuf *m, ...)
358 {
359 	struct ip *ip;
360 	struct inpcb *inp = NULL;
361 	u_int8_t *optp = NULL;
362 	int optlen = 0;
363 	int tlen, off;
364 	struct tcpcb *tp = NULL;
365 	int tiflags;
366 	struct socket *so = NULL;
367 	int todrop, acked, ourfinisacked;
368 	int hdroptlen = 0;
369 	short ostate = 0;
370 	tcp_seq iss, *reuse = NULL;
371 	u_long tiwin;
372 	struct tcp_opt_info opti;
373 	int iphlen;
374 	va_list ap;
375 	struct tcphdr *th;
376 #ifdef INET6
377 	struct ip6_hdr *ip6 = NULL;
378 #endif /* INET6 */
379 #ifdef IPSEC
380 	struct m_tag *mtag;
381 	struct tdb_ident *tdbi;
382 	struct tdb *tdb;
383 	int error;
384 #endif /* IPSEC */
385 	int af;
386 #ifdef TCP_ECN
387 	u_char iptos;
388 #endif
389 
390 	va_start(ap, m);
391 	iphlen = va_arg(ap, int);
392 	va_end(ap);
393 
394 	tcpstat.tcps_rcvtotal++;
395 
396 	opti.ts_present = 0;
397 	opti.maxseg = 0;
398 
399 	/*
400 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
401 	 */
402 	if (m->m_flags & (M_BCAST|M_MCAST))
403 		goto drop;
404 
405 	/*
406 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
407 	 * TCP/IPv4.
408 	 */
409 	switch (mtod(m, struct ip *)->ip_v) {
410 #ifdef INET6
411 	case 6:
412 		af = AF_INET6;
413 		break;
414 #endif
415 	case 4:
416 		af = AF_INET;
417 		break;
418 	default:
419 		m_freem(m);
420 		return;	/*EAFNOSUPPORT*/
421 	}
422 
423 	/*
424 	 * Get IP and TCP header together in first mbuf.
425 	 * Note: IP leaves IP header in first mbuf.
426 	 */
427 	switch (af) {
428 	case AF_INET:
429 #ifdef DIAGNOSTIC
430 		if (iphlen < sizeof(struct ip)) {
431 			m_freem(m);
432 			return;
433 		}
434 #endif /* DIAGNOSTIC */
435 		break;
436 #ifdef INET6
437 	case AF_INET6:
438 #ifdef DIAGNOSTIC
439 		if (iphlen < sizeof(struct ip6_hdr)) {
440 			m_freem(m);
441 			return;
442 		}
443 #endif /* DIAGNOSTIC */
444 		break;
445 #endif
446 	default:
447 		m_freem(m);
448 		return;
449 	}
450 
451 	IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th));
452 	if (!th) {
453 		tcpstat.tcps_rcvshort++;
454 		return;
455 	}
456 
457 	tlen = m->m_pkthdr.len - iphlen;
458 	ip = NULL;
459 #ifdef INET6
460 	ip6 = NULL;
461 #endif
462 	switch (af) {
463 	case AF_INET:
464 		ip = mtod(m, struct ip *);
465 #ifdef TCP_ECN
466 		/* save ip_tos before clearing it for checksum */
467 		iptos = ip->ip_tos;
468 #endif
469 		break;
470 #ifdef INET6
471 	case AF_INET6:
472 		ip6 = mtod(m, struct ip6_hdr *);
473 #ifdef TCP_ECN
474 		iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
475 #endif
476 
477 		/* Be proactive about malicious use of IPv4 mapped address */
478 		if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) ||
479 		    IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) {
480 			/* XXX stat */
481 			goto drop;
482 		}
483 
484 		/*
485 		 * Be proactive about unspecified IPv6 address in source.
486 		 * As we use all-zero to indicate unbounded/unconnected pcb,
487 		 * unspecified IPv6 address can be used to confuse us.
488 		 *
489 		 * Note that packets with unspecified IPv6 destination is
490 		 * already dropped in ip6_input.
491 		 */
492 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
493 			/* XXX stat */
494 			goto drop;
495 		}
496 
497 		/* Discard packets to multicast */
498 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) {
499 			/* XXX stat */
500 			goto drop;
501 		}
502 		break;
503 #endif
504 	}
505 
506 	/*
507 	 * Checksum extended TCP header and data.
508 	 */
509 	if ((m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_OK) == 0) {
510 		int sum;
511 
512 		if (m->m_pkthdr.csum_flags & M_TCP_CSUM_IN_BAD) {
513 			tcpstat.tcps_rcvbadsum++;
514 			goto drop;
515 		}
516 		tcpstat.tcps_inswcsum++;
517 		switch (af) {
518 		case AF_INET:
519 			sum = in4_cksum(m, IPPROTO_TCP, iphlen, tlen);
520 			break;
521 #ifdef INET6
522 		case AF_INET6:
523 			sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
524 			    tlen);
525 			break;
526 #endif
527 		}
528 		if (sum != 0) {
529 			tcpstat.tcps_rcvbadsum++;
530 			goto drop;
531 		}
532 	}
533 
534 	/*
535 	 * Check that TCP offset makes sense,
536 	 * pull out TCP options and adjust length.		XXX
537 	 */
538 	off = th->th_off << 2;
539 	if (off < sizeof(struct tcphdr) || off > tlen) {
540 		tcpstat.tcps_rcvbadoff++;
541 		goto drop;
542 	}
543 	tlen -= off;
544 	if (off > sizeof(struct tcphdr)) {
545 		IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off);
546 		if (!th) {
547 			tcpstat.tcps_rcvshort++;
548 			return;
549 		}
550 		optlen = off - sizeof(struct tcphdr);
551 		optp = (u_int8_t *)(th + 1);
552 		/*
553 		 * Do quick retrieval of timestamp options ("options
554 		 * prediction?").  If timestamp is the only option and it's
555 		 * formatted as recommended in RFC 1323 appendix A, we
556 		 * quickly get the values now and not bother calling
557 		 * tcp_dooptions(), etc.
558 		 */
559 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
560 		     (optlen > TCPOLEN_TSTAMP_APPA &&
561 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
562 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
563 		     (th->th_flags & TH_SYN) == 0) {
564 			opti.ts_present = 1;
565 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
566 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
567 			optp = NULL;	/* we've parsed the options */
568 		}
569 	}
570 	tiflags = th->th_flags;
571 
572 	/*
573 	 * Convert TCP protocol specific fields to host format.
574 	 */
575 	th->th_seq = ntohl(th->th_seq);
576 	th->th_ack = ntohl(th->th_ack);
577 	th->th_win = ntohs(th->th_win);
578 	th->th_urp = ntohs(th->th_urp);
579 
580 	/*
581 	 * Locate pcb for segment.
582 	 */
583 #if NPF > 0
584 	if (m->m_pkthdr.pf.statekey) {
585 		inp = m->m_pkthdr.pf.statekey->inp;
586 		if (inp && inp->inp_pf_sk)
587 			KASSERT(m->m_pkthdr.pf.statekey == inp->inp_pf_sk);
588 	}
589 #endif
590 findpcb:
591 	if (inp == NULL) {
592 		switch (af) {
593 #ifdef INET6
594 		case AF_INET6:
595 			inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src,
596 			    th->th_sport, &ip6->ip6_dst, th->th_dport,
597 			    m->m_pkthdr.ph_rtableid);
598 			break;
599 #endif
600 		case AF_INET:
601 			inp = in_pcbhashlookup(&tcbtable, ip->ip_src,
602 			    th->th_sport, ip->ip_dst, th->th_dport,
603 			    m->m_pkthdr.ph_rtableid);
604 			break;
605 		}
606 #if NPF > 0
607 		if (m->m_pkthdr.pf.statekey && inp) {
608 			m->m_pkthdr.pf.statekey->inp = inp;
609 			inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
610 		}
611 #endif
612 	}
613 	if (inp == NULL) {
614 		int	inpl_reverse = 0;
615 		if (m->m_pkthdr.pf.flags & PF_TAG_TRANSLATE_LOCALHOST)
616 			inpl_reverse = 1;
617 		++tcpstat.tcps_pcbhashmiss;
618 		switch (af) {
619 #ifdef INET6
620 		case AF_INET6:
621 			inp = in6_pcblookup_listen(&tcbtable,
622 			    &ip6->ip6_dst, th->th_dport, inpl_reverse, m,
623 			    m->m_pkthdr.ph_rtableid);
624 			break;
625 #endif /* INET6 */
626 		case AF_INET:
627 			inp = in_pcblookup_listen(&tcbtable,
628 			    ip->ip_dst, th->th_dport, inpl_reverse, m,
629 			    m->m_pkthdr.ph_rtableid);
630 			break;
631 		}
632 		/*
633 		 * If the state is CLOSED (i.e., TCB does not exist) then
634 		 * all data in the incoming segment is discarded.
635 		 * If the TCB exists but is in CLOSED state, it is embryonic,
636 		 * but should either do a listen or a connect soon.
637 		 */
638 		if (inp == NULL) {
639 			++tcpstat.tcps_noport;
640 			goto dropwithreset_ratelim;
641 		}
642 	}
643 	KASSERT(sotoinpcb(inp->inp_socket) == inp);
644 	KASSERT(intotcpcb(inp) == NULL || intotcpcb(inp)->t_inpcb == inp);
645 
646 	/* Check the minimum TTL for socket. */
647 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
648 		goto drop;
649 
650 	tp = intotcpcb(inp);
651 	if (tp == NULL)
652 		goto dropwithreset_ratelim;
653 	if (tp->t_state == TCPS_CLOSED)
654 		goto drop;
655 
656 	/* Unscale the window into a 32-bit value. */
657 	if ((tiflags & TH_SYN) == 0)
658 		tiwin = th->th_win << tp->snd_scale;
659 	else
660 		tiwin = th->th_win;
661 
662 	so = inp->inp_socket;
663 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
664 		union syn_cache_sa src;
665 		union syn_cache_sa dst;
666 
667 		bzero(&src, sizeof(src));
668 		bzero(&dst, sizeof(dst));
669 		switch (af) {
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 #ifdef INET6
682 		case AF_INET6:
683 			src.sin6.sin6_len = sizeof(struct sockaddr_in6);
684 			src.sin6.sin6_family = AF_INET6;
685 			src.sin6.sin6_addr = ip6->ip6_src;
686 			src.sin6.sin6_port = th->th_sport;
687 
688 			dst.sin6.sin6_len = sizeof(struct sockaddr_in6);
689 			dst.sin6.sin6_family = AF_INET6;
690 			dst.sin6.sin6_addr = ip6->ip6_dst;
691 			dst.sin6.sin6_port = th->th_dport;
692 			break;
693 #endif /* INET6 */
694 		default:
695 			goto badsyn;	/*sanity*/
696 		}
697 
698 		if (so->so_options & SO_DEBUG) {
699 			ostate = tp->t_state;
700 			switch (af) {
701 #ifdef INET6
702 			case AF_INET6:
703 				bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6));
704 				bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th));
705 				break;
706 #endif
707 			case AF_INET:
708 				bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip));
709 				bcopy(th, &tcp_saveti.ti_t, sizeof(*th));
710 				break;
711 			}
712 		}
713 		if (so->so_options & SO_ACCEPTCONN) {
714 			switch (tiflags & (TH_RST|TH_SYN|TH_ACK)) {
715 
716 			case TH_SYN|TH_ACK|TH_RST:
717 			case TH_SYN|TH_RST:
718 			case TH_ACK|TH_RST:
719 			case TH_RST:
720 				syn_cache_reset(&src.sa, &dst.sa, th,
721 				    inp->inp_rtableid);
722 				goto drop;
723 
724 			case TH_SYN|TH_ACK:
725 				/*
726 				 * Received a SYN,ACK.  This should
727 				 * never happen while we are in
728 				 * LISTEN.  Send an RST.
729 				 */
730 				goto badsyn;
731 
732 			case TH_ACK:
733 				so = syn_cache_get(&src.sa, &dst.sa,
734 					th, iphlen, tlen, so, m);
735 				if (so == NULL) {
736 					/*
737 					 * We don't have a SYN for
738 					 * this ACK; send an RST.
739 					 */
740 					goto badsyn;
741 				} else if (so == (struct socket *)(-1)) {
742 					/*
743 					 * We were unable to create
744 					 * the connection.  If the
745 					 * 3-way handshake was
746 					 * completed, and RST has
747 					 * been sent to the peer.
748 					 * Since the mbuf might be
749 					 * in use for the reply,
750 					 * do not free it.
751 					 */
752 					m = NULL;
753 					goto drop;
754 				} else {
755 					/*
756 					 * We have created a
757 					 * full-blown connection.
758 					 */
759 					tp = NULL;
760 					inp = sotoinpcb(so);
761 					tp = intotcpcb(inp);
762 					if (tp == NULL)
763 						goto badsyn;	/*XXX*/
764 
765 				}
766 				break;
767 
768 			default:
769 				/*
770 				 * None of RST, SYN or ACK was set.
771 				 * This is an invalid packet for a
772 				 * TCB in LISTEN state.  Send a RST.
773 				 */
774 				goto badsyn;
775 
776 			case TH_SYN:
777 				/*
778 				 * Received a SYN.
779 				 */
780 #ifdef INET6
781 				/*
782 				 * If deprecated address is forbidden, we do
783 				 * not accept SYN to deprecated interface
784 				 * address to prevent any new inbound
785 				 * connection from getting established.
786 				 * When we do not accept SYN, we send a TCP
787 				 * RST, with deprecated source address (instead
788 				 * of dropping it).  We compromise it as it is
789 				 * much better for peer to send a RST, and
790 				 * RST will be the final packet for the
791 				 * exchange.
792 				 *
793 				 * If we do not forbid deprecated addresses, we
794 				 * accept the SYN packet.  RFC2462 does not
795 				 * suggest dropping SYN in this case.
796 				 * If we decipher RFC2462 5.5.4, it says like
797 				 * this:
798 				 * 1. use of deprecated addr with existing
799 				 *    communication is okay - "SHOULD continue
800 				 *    to be used"
801 				 * 2. use of it with new communication:
802 				 *   (2a) "SHOULD NOT be used if alternate
803 				 *        address with sufficient scope is
804 				 *        available"
805 				 *   (2b) nothing mentioned otherwise.
806 				 * Here we fall into (2b) case as we have no
807 				 * choice in our source address selection - we
808 				 * must obey the peer.
809 				 *
810 				 * The wording in RFC2462 is confusing, and
811 				 * there are multiple description text for
812 				 * deprecated address handling - worse, they
813 				 * are not exactly the same.  I believe 5.5.4
814 				 * is the best one, so we follow 5.5.4.
815 				 */
816 				if (ip6 && !ip6_use_deprecated) {
817 					struct in6_ifaddr *ia6;
818 					struct ifnet *ifp =
819 					    if_get(m->m_pkthdr.ph_ifidx);
820 
821 					if (ifp &&
822 					    (ia6 = in6ifa_ifpwithaddr(ifp,
823 					    &ip6->ip6_dst)) &&
824 					    (ia6->ia6_flags &
825 					    IN6_IFF_DEPRECATED)) {
826 						tp = NULL;
827 						if_put(ifp);
828 						goto dropwithreset;
829 					}
830 					if_put(ifp);
831 				}
832 #endif
833 
834 				/*
835 				 * LISTEN socket received a SYN
836 				 * from itself?  This can't possibly
837 				 * be valid; drop the packet.
838 				 */
839 				if (th->th_dport == th->th_sport) {
840 					switch (af) {
841 #ifdef INET6
842 					case AF_INET6:
843 						if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src,
844 						    &ip6->ip6_dst)) {
845 							tcpstat.tcps_badsyn++;
846 							goto drop;
847 						}
848 						break;
849 #endif /* INET6 */
850 					case AF_INET:
851 						if (ip->ip_dst.s_addr == ip->ip_src.s_addr) {
852 							tcpstat.tcps_badsyn++;
853 							goto drop;
854 						}
855 						break;
856 					}
857 				}
858 
859 				/*
860 				 * SYN looks ok; create compressed TCP
861 				 * state for it.
862 				 */
863 				if (so->so_qlen > so->so_qlimit ||
864 				    syn_cache_add(&src.sa, &dst.sa, th, iphlen,
865 				    so, m, optp, optlen, &opti, reuse) == -1) {
866 					tcpstat.tcps_dropsyn++;
867 					goto drop;
868 				}
869 				return;
870 			}
871 		}
872 	}
873 
874 #ifdef DIAGNOSTIC
875 	/*
876 	 * Should not happen now that all embryonic connections
877 	 * are handled with compressed state.
878 	 */
879 	if (tp->t_state == TCPS_LISTEN)
880 		panic("tcp_input: TCPS_LISTEN");
881 #endif
882 
883 #if NPF > 0
884 	if (m->m_pkthdr.pf.statekey && !m->m_pkthdr.pf.statekey->inp &&
885 	    !inp->inp_pf_sk) {
886 		m->m_pkthdr.pf.statekey->inp = inp;
887 		inp->inp_pf_sk = m->m_pkthdr.pf.statekey;
888 	}
889 	/* The statekey has finished finding the inp, it is no longer needed. */
890 	m->m_pkthdr.pf.statekey = NULL;
891 #endif
892 
893 #ifdef IPSEC
894 	/* Find most recent IPsec tag */
895 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
896 	if (mtag != NULL) {
897 		tdbi = (struct tdb_ident *)(mtag + 1);
898 	        tdb = gettdb(tdbi->rdomain, tdbi->spi,
899 		    &tdbi->dst, tdbi->proto);
900 	} else
901 		tdb = NULL;
902 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
903 	    tdb, inp, 0);
904 	if (error) {
905 		tcpstat.tcps_rcvnosec++;
906 		goto drop;
907 	}
908 #endif /* IPSEC */
909 
910 	/*
911 	 * Segment received on connection.
912 	 * Reset idle time and keep-alive timer.
913 	 */
914 	tp->t_rcvtime = tcp_now;
915 	if (TCPS_HAVEESTABLISHED(tp->t_state))
916 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
917 
918 #ifdef TCP_SACK
919 	if (tp->sack_enable)
920 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
921 #endif /* TCP_SACK */
922 
923 	/*
924 	 * Process options.
925 	 */
926 #ifdef TCP_SIGNATURE
927 	if (optp || (tp->t_flags & TF_SIGNATURE))
928 #else
929 	if (optp)
930 #endif
931 		if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti,
932 		    m->m_pkthdr.ph_rtableid))
933 			goto drop;
934 
935 	if (opti.ts_present && opti.ts_ecr) {
936 		int rtt_test;
937 
938 		/* subtract out the tcp timestamp modulator */
939 		opti.ts_ecr -= tp->ts_modulate;
940 
941 		/* make sure ts_ecr is sensible */
942 		rtt_test = tcp_now - opti.ts_ecr;
943 		if (rtt_test < 0 || rtt_test > TCP_RTT_MAX)
944 			opti.ts_ecr = 0;
945 	}
946 
947 #ifdef TCP_ECN
948 	/* if congestion experienced, set ECE bit in subsequent packets. */
949 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
950 		tp->t_flags |= TF_RCVD_CE;
951 		tcpstat.tcps_ecn_rcvce++;
952 	}
953 #endif
954 	/*
955 	 * Header prediction: check for the two common cases
956 	 * of a uni-directional data xfer.  If the packet has
957 	 * no control flags, is in-sequence, the window didn't
958 	 * change and we're not retransmitting, it's a
959 	 * candidate.  If the length is zero and the ack moved
960 	 * forward, we're the sender side of the xfer.  Just
961 	 * free the data acked & wake any higher level process
962 	 * that was blocked waiting for space.  If the length
963 	 * is non-zero and the ack didn't move, we're the
964 	 * receiver side.  If we're getting packets in-order
965 	 * (the reassembly queue is empty), add the data to
966 	 * the socket buffer and note that we need a delayed ack.
967 	 */
968 	if (tp->t_state == TCPS_ESTABLISHED &&
969 #ifdef TCP_ECN
970 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
971 #else
972 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
973 #endif
974 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
975 	    th->th_seq == tp->rcv_nxt &&
976 	    tiwin && tiwin == tp->snd_wnd &&
977 	    tp->snd_nxt == tp->snd_max) {
978 
979 		/*
980 		 * If last ACK falls within this segment's sequence numbers,
981 		 *  record the timestamp.
982 		 * Fix from Braden, see Stevens p. 870
983 		 */
984 		if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
985 			tp->ts_recent_age = tcp_now;
986 			tp->ts_recent = opti.ts_val;
987 		}
988 
989 		if (tlen == 0) {
990 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
991 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
992 			    tp->snd_cwnd >= tp->snd_wnd &&
993 			    tp->t_dupacks == 0) {
994 				/*
995 				 * this is a pure ack for outstanding data.
996 				 */
997 				++tcpstat.tcps_predack;
998 				if (opti.ts_present && opti.ts_ecr)
999 					tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1000 				else if (tp->t_rtttime &&
1001 				    SEQ_GT(th->th_ack, tp->t_rtseq))
1002 					tcp_xmit_timer(tp,
1003 					    tcp_now - tp->t_rtttime);
1004 				acked = th->th_ack - tp->snd_una;
1005 				tcpstat.tcps_rcvackpack++;
1006 				tcpstat.tcps_rcvackbyte += acked;
1007 				ND6_HINT(tp);
1008 				sbdrop(&so->so_snd, acked);
1009 
1010 				/*
1011 				 * If we had a pending ICMP message that
1012 				 * refers to data that have just been
1013 				 * acknowledged, disregard the recorded ICMP
1014 				 * message.
1015 				 */
1016 				if ((tp->t_flags & TF_PMTUD_PEND) &&
1017 				    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1018 					tp->t_flags &= ~TF_PMTUD_PEND;
1019 
1020 				/*
1021 				 * Keep track of the largest chunk of data
1022 				 * acknowledged since last PMTU update
1023 				 */
1024 				if (tp->t_pmtud_mss_acked < acked)
1025 					tp->t_pmtud_mss_acked = acked;
1026 
1027 				tp->snd_una = th->th_ack;
1028 #if defined(TCP_SACK) || defined(TCP_ECN)
1029 				/*
1030 				 * We want snd_last to track snd_una so
1031 				 * as to avoid sequence wraparound problems
1032 				 * for very large transfers.
1033 				 */
1034 #ifdef TCP_ECN
1035 				if (SEQ_GT(tp->snd_una, tp->snd_last))
1036 #endif
1037 				tp->snd_last = tp->snd_una;
1038 #endif /* TCP_SACK */
1039 #if defined(TCP_SACK) && defined(TCP_FACK)
1040 				tp->snd_fack = tp->snd_una;
1041 				tp->retran_data = 0;
1042 #endif /* TCP_FACK */
1043 				m_freem(m);
1044 
1045 				/*
1046 				 * If all outstanding data are acked, stop
1047 				 * retransmit timer, otherwise restart timer
1048 				 * using current (possibly backed-off) value.
1049 				 * If process is waiting for space,
1050 				 * wakeup/selwakeup/signal.  If data
1051 				 * are ready to send, let tcp_output
1052 				 * decide between more output or persist.
1053 				 */
1054 				if (tp->snd_una == tp->snd_max)
1055 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1056 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1057 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1058 
1059 				tcp_update_sndspace(tp);
1060 				if (sb_notify(&so->so_snd)) {
1061 					tp->t_flags |= TF_BLOCKOUTPUT;
1062 					sowwakeup(so);
1063 					tp->t_flags &= ~TF_BLOCKOUTPUT;
1064 				}
1065 				if (so->so_snd.sb_cc ||
1066 				    tp->t_flags & TF_NEEDOUTPUT)
1067 					(void) tcp_output(tp);
1068 				return;
1069 			}
1070 		} else if (th->th_ack == tp->snd_una &&
1071 		    TAILQ_EMPTY(&tp->t_segq) &&
1072 		    tlen <= sbspace(&so->so_rcv)) {
1073 			/*
1074 			 * This is a pure, in-sequence data packet
1075 			 * with nothing on the reassembly queue and
1076 			 * we have enough buffer space to take it.
1077 			 */
1078 #ifdef TCP_SACK
1079 			/* Clean receiver SACK report if present */
1080 			if (tp->sack_enable && tp->rcv_numsacks)
1081 				tcp_clean_sackreport(tp);
1082 #endif /* TCP_SACK */
1083 			++tcpstat.tcps_preddat;
1084 			tp->rcv_nxt += tlen;
1085 			tcpstat.tcps_rcvpack++;
1086 			tcpstat.tcps_rcvbyte += tlen;
1087 			ND6_HINT(tp);
1088 
1089 			TCP_SETUP_ACK(tp, tiflags, m);
1090 			/*
1091 			 * Drop TCP, IP headers and TCP options then add data
1092 			 * to socket buffer.
1093 			 */
1094 			if (so->so_state & SS_CANTRCVMORE)
1095 				m_freem(m);
1096 			else {
1097 				if (opti.ts_present && opti.ts_ecr) {
1098 					if (tp->rfbuf_ts < opti.ts_ecr &&
1099 					    opti.ts_ecr - tp->rfbuf_ts < hz) {
1100 						tcp_update_rcvspace(tp);
1101 						/* Start over with next RTT. */
1102 						tp->rfbuf_cnt = 0;
1103 						tp->rfbuf_ts = 0;
1104 					} else
1105 						tp->rfbuf_cnt += tlen;
1106 				}
1107 				m_adj(m, iphlen + off);
1108 				sbappendstream(&so->so_rcv, m);
1109 			}
1110 			tp->t_flags |= TF_BLOCKOUTPUT;
1111 			sorwakeup(so);
1112 			tp->t_flags &= ~TF_BLOCKOUTPUT;
1113 			if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT))
1114 				(void) tcp_output(tp);
1115 			return;
1116 		}
1117 	}
1118 
1119 	/*
1120 	 * Compute mbuf offset to TCP data segment.
1121 	 */
1122 	hdroptlen = iphlen + off;
1123 
1124 	/*
1125 	 * Calculate amount of space in receive window,
1126 	 * and then do TCP input processing.
1127 	 * Receive window is amount of space in rcv queue,
1128 	 * but not less than advertised window.
1129 	 */
1130 	{ int win;
1131 
1132 	win = sbspace(&so->so_rcv);
1133 	if (win < 0)
1134 		win = 0;
1135 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1136 	}
1137 
1138 	/* Reset receive buffer auto scaling when not in bulk receive mode. */
1139 	tp->rfbuf_cnt = 0;
1140 	tp->rfbuf_ts = 0;
1141 
1142 	switch (tp->t_state) {
1143 
1144 	/*
1145 	 * If the state is SYN_RECEIVED:
1146 	 * 	if seg contains SYN/ACK, send an RST.
1147 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1148 	 */
1149 
1150 	case TCPS_SYN_RECEIVED:
1151 		if (tiflags & TH_ACK) {
1152 			if (tiflags & TH_SYN) {
1153 				tcpstat.tcps_badsyn++;
1154 				goto dropwithreset;
1155 			}
1156 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1157 			    SEQ_GT(th->th_ack, tp->snd_max))
1158 				goto dropwithreset;
1159 		}
1160 		break;
1161 
1162 	/*
1163 	 * If the state is SYN_SENT:
1164 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1165 	 *	if seg contains a RST, then drop the connection.
1166 	 *	if seg does not contain SYN, then drop it.
1167 	 * Otherwise this is an acceptable SYN segment
1168 	 *	initialize tp->rcv_nxt and tp->irs
1169 	 *	if seg contains ack then advance tp->snd_una
1170 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1171 	 *	arrange for segment to be acked (eventually)
1172 	 *	continue processing rest of data/controls, beginning with URG
1173 	 */
1174 	case TCPS_SYN_SENT:
1175 		if ((tiflags & TH_ACK) &&
1176 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1177 		     SEQ_GT(th->th_ack, tp->snd_max)))
1178 			goto dropwithreset;
1179 		if (tiflags & TH_RST) {
1180 #ifdef TCP_ECN
1181 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1182 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1183 				goto drop;
1184 #endif
1185 			if (tiflags & TH_ACK)
1186 				tp = tcp_drop(tp, ECONNREFUSED);
1187 			goto drop;
1188 		}
1189 		if ((tiflags & TH_SYN) == 0)
1190 			goto drop;
1191 		if (tiflags & TH_ACK) {
1192 			tp->snd_una = th->th_ack;
1193 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1194 				tp->snd_nxt = tp->snd_una;
1195 		}
1196 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1197 		tp->irs = th->th_seq;
1198 		tcp_mss(tp, opti.maxseg);
1199 		/* Reset initial window to 1 segment for retransmit */
1200 		if (tp->t_rxtshift > 0)
1201 			tp->snd_cwnd = tp->t_maxseg;
1202 		tcp_rcvseqinit(tp);
1203 		tp->t_flags |= TF_ACKNOW;
1204 #ifdef TCP_SACK
1205                 /*
1206                  * If we've sent a SACK_PERMITTED option, and the peer
1207                  * also replied with one, then TF_SACK_PERMIT should have
1208                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1209                  */
1210 		if (tp->sack_enable)
1211 			tp->sack_enable = tp->t_flags & TF_SACK_PERMIT;
1212 #endif
1213 #ifdef TCP_ECN
1214 		/*
1215 		 * if ECE is set but CWR is not set for SYN-ACK, or
1216 		 * both ECE and CWR are set for simultaneous open,
1217 		 * peer is ECN capable.
1218 		 */
1219 		if (tcp_do_ecn) {
1220 			switch (tiflags & (TH_ACK|TH_ECE|TH_CWR)) {
1221 			case TH_ACK|TH_ECE:
1222 			case TH_ECE|TH_CWR:
1223 				tp->t_flags |= TF_ECN_PERMIT;
1224 				tiflags &= ~(TH_ECE|TH_CWR);
1225 				tcpstat.tcps_ecn_accepts++;
1226 			}
1227 		}
1228 #endif
1229 
1230 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1231 			tcpstat.tcps_connects++;
1232 			soisconnected(so);
1233 			tp->t_state = TCPS_ESTABLISHED;
1234 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1235 			/* Do window scaling on this connection? */
1236 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1237 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1238 				tp->snd_scale = tp->requested_s_scale;
1239 				tp->rcv_scale = tp->request_r_scale;
1240 			}
1241 			tcp_flush_queue(tp);
1242 
1243 			/*
1244 			 * if we didn't have to retransmit the SYN,
1245 			 * use its rtt as our initial srtt & rtt var.
1246 			 */
1247 			if (tp->t_rtttime)
1248 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1249 			/*
1250 			 * Since new data was acked (the SYN), open the
1251 			 * congestion window by one MSS.  We do this
1252 			 * here, because we won't go through the normal
1253 			 * ACK processing below.  And since this is the
1254 			 * start of the connection, we know we are in
1255 			 * the exponential phase of slow-start.
1256 			 */
1257 			tp->snd_cwnd += tp->t_maxseg;
1258 		} else
1259 			tp->t_state = TCPS_SYN_RECEIVED;
1260 
1261 #if 0
1262 trimthenstep6:
1263 #endif
1264 		/*
1265 		 * Advance th->th_seq to correspond to first data byte.
1266 		 * If data, trim to stay within window,
1267 		 * dropping FIN if necessary.
1268 		 */
1269 		th->th_seq++;
1270 		if (tlen > tp->rcv_wnd) {
1271 			todrop = tlen - tp->rcv_wnd;
1272 			m_adj(m, -todrop);
1273 			tlen = tp->rcv_wnd;
1274 			tiflags &= ~TH_FIN;
1275 			tcpstat.tcps_rcvpackafterwin++;
1276 			tcpstat.tcps_rcvbyteafterwin += todrop;
1277 		}
1278 		tp->snd_wl1 = th->th_seq - 1;
1279 		tp->rcv_up = th->th_seq;
1280 		goto step6;
1281 	/*
1282 	 * If a new connection request is received while in TIME_WAIT,
1283 	 * drop the old connection and start over if the if the
1284 	 * timestamp or the sequence numbers are above the previous
1285 	 * ones.
1286 	 */
1287 	case TCPS_TIME_WAIT:
1288 		if (((tiflags & (TH_SYN|TH_ACK)) == TH_SYN) &&
1289 		    ((opti.ts_present &&
1290 		    TSTMP_LT(tp->ts_recent, opti.ts_val)) ||
1291 		    SEQ_GT(th->th_seq, tp->rcv_nxt))) {
1292 #if NPF > 0
1293 			/*
1294 			 * The socket will be recreated but the new state
1295 			 * has already been linked to the socket.  Remove the
1296 			 * link between old socket and new state.
1297 			 */
1298 			if (inp->inp_pf_sk) {
1299 				inp->inp_pf_sk->inp = NULL;
1300 				inp->inp_pf_sk = NULL;
1301 			}
1302 #endif
1303 			/*
1304 			* Advance the iss by at least 32768, but
1305 			* clear the msb in order to make sure
1306 			* that SEG_LT(snd_nxt, iss).
1307 			*/
1308 			iss = tp->snd_nxt +
1309 			    ((arc4random() & 0x7fffffff) | 0x8000);
1310 			reuse = &iss;
1311 			tp = tcp_close(tp);
1312 			inp = NULL;
1313 			goto findpcb;
1314 		}
1315 	}
1316 
1317 	/*
1318 	 * States other than LISTEN or SYN_SENT.
1319 	 * First check timestamp, if present.
1320 	 * Then check that at least some bytes of segment are within
1321 	 * receive window.  If segment begins before rcv_nxt,
1322 	 * drop leading data (and SYN); if nothing left, just ack.
1323 	 *
1324 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1325 	 * and it's less than opti.ts_recent, drop it.
1326 	 */
1327 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1328 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
1329 
1330 		/* Check to see if ts_recent is over 24 days old.  */
1331 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1332 			/*
1333 			 * Invalidate ts_recent.  If this segment updates
1334 			 * ts_recent, the age will be reset later and ts_recent
1335 			 * will get a valid value.  If it does not, setting
1336 			 * ts_recent to zero will at least satisfy the
1337 			 * requirement that zero be placed in the timestamp
1338 			 * echo reply when ts_recent isn't valid.  The
1339 			 * age isn't reset until we get a valid ts_recent
1340 			 * because we don't want out-of-order segments to be
1341 			 * dropped when ts_recent is old.
1342 			 */
1343 			tp->ts_recent = 0;
1344 		} else {
1345 			tcpstat.tcps_rcvduppack++;
1346 			tcpstat.tcps_rcvdupbyte += tlen;
1347 			tcpstat.tcps_pawsdrop++;
1348 			goto dropafterack;
1349 		}
1350 	}
1351 
1352 	todrop = tp->rcv_nxt - th->th_seq;
1353 	if (todrop > 0) {
1354 		if (tiflags & TH_SYN) {
1355 			tiflags &= ~TH_SYN;
1356 			th->th_seq++;
1357 			if (th->th_urp > 1)
1358 				th->th_urp--;
1359 			else
1360 				tiflags &= ~TH_URG;
1361 			todrop--;
1362 		}
1363 		if (todrop > tlen ||
1364 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1365 			/*
1366 			 * Any valid FIN must be to the left of the
1367 			 * window.  At this point, FIN must be a
1368 			 * duplicate or out-of-sequence, so drop it.
1369 			 */
1370 			tiflags &= ~TH_FIN;
1371 			/*
1372 			 * Send ACK to resynchronize, and drop any data,
1373 			 * but keep on processing for RST or ACK.
1374 			 */
1375 			tp->t_flags |= TF_ACKNOW;
1376 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1377 			tcpstat.tcps_rcvduppack++;
1378 		} else {
1379 			tcpstat.tcps_rcvpartduppack++;
1380 			tcpstat.tcps_rcvpartdupbyte += todrop;
1381 		}
1382 		hdroptlen += todrop;	/* drop from head afterwards */
1383 		th->th_seq += todrop;
1384 		tlen -= todrop;
1385 		if (th->th_urp > todrop)
1386 			th->th_urp -= todrop;
1387 		else {
1388 			tiflags &= ~TH_URG;
1389 			th->th_urp = 0;
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * If new data are received on a connection after the
1395 	 * user processes are gone, then RST the other end.
1396 	 */
1397 	if ((so->so_state & SS_NOFDREF) &&
1398 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1399 		tp = tcp_close(tp);
1400 		tcpstat.tcps_rcvafterclose++;
1401 		goto dropwithreset;
1402 	}
1403 
1404 	/*
1405 	 * If segment ends after window, drop trailing data
1406 	 * (and PUSH and FIN); if nothing left, just ACK.
1407 	 */
1408 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1409 	if (todrop > 0) {
1410 		tcpstat.tcps_rcvpackafterwin++;
1411 		if (todrop >= tlen) {
1412 			tcpstat.tcps_rcvbyteafterwin += tlen;
1413 			/*
1414 			 * If window is closed can only take segments at
1415 			 * window edge, and have to drop data and PUSH from
1416 			 * incoming segments.  Continue processing, but
1417 			 * remember to ack.  Otherwise, drop segment
1418 			 * and ack.
1419 			 */
1420 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1421 				tp->t_flags |= TF_ACKNOW;
1422 				tcpstat.tcps_rcvwinprobe++;
1423 			} else
1424 				goto dropafterack;
1425 		} else
1426 			tcpstat.tcps_rcvbyteafterwin += todrop;
1427 		m_adj(m, -todrop);
1428 		tlen -= todrop;
1429 		tiflags &= ~(TH_PUSH|TH_FIN);
1430 	}
1431 
1432 	/*
1433 	 * If last ACK falls within this segment's sequence numbers,
1434 	 * record its timestamp if it's more recent.
1435 	 * Cf fix from Braden, see Stevens p. 870
1436 	 */
1437 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1438 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1439 		if (SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1440 		    ((tiflags & (TH_SYN|TH_FIN)) != 0)))
1441 			tp->ts_recent = opti.ts_val;
1442 		else
1443 			tp->ts_recent = 0;
1444 		tp->ts_recent_age = tcp_now;
1445 	}
1446 
1447 	/*
1448 	 * If the RST bit is set examine the state:
1449 	 *    SYN_RECEIVED STATE:
1450 	 *	If passive open, return to LISTEN state.
1451 	 *	If active open, inform user that connection was refused.
1452 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1453 	 *	Inform user that connection was reset, and close tcb.
1454 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1455 	 *	Close the tcb.
1456 	 */
1457 	if (tiflags & TH_RST) {
1458 		if (th->th_seq != tp->last_ack_sent &&
1459 		    th->th_seq != tp->rcv_nxt &&
1460 		    th->th_seq != (tp->rcv_nxt + 1))
1461 			goto drop;
1462 
1463 		switch (tp->t_state) {
1464 		case TCPS_SYN_RECEIVED:
1465 #ifdef TCP_ECN
1466 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1467 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1468 				goto drop;
1469 #endif
1470 			so->so_error = ECONNREFUSED;
1471 			goto close;
1472 
1473 		case TCPS_ESTABLISHED:
1474 		case TCPS_FIN_WAIT_1:
1475 		case TCPS_FIN_WAIT_2:
1476 		case TCPS_CLOSE_WAIT:
1477 			so->so_error = ECONNRESET;
1478 		close:
1479 			tp->t_state = TCPS_CLOSED;
1480 			tcpstat.tcps_drops++;
1481 			tp = tcp_close(tp);
1482 			goto drop;
1483 		case TCPS_CLOSING:
1484 		case TCPS_LAST_ACK:
1485 		case TCPS_TIME_WAIT:
1486 			tp = tcp_close(tp);
1487 			goto drop;
1488 		}
1489 	}
1490 
1491 	/*
1492 	 * If a SYN is in the window, then this is an
1493 	 * error and we ACK and drop the packet.
1494 	 */
1495 	if (tiflags & TH_SYN)
1496 		goto dropafterack_ratelim;
1497 
1498 	/*
1499 	 * If the ACK bit is off we drop the segment and return.
1500 	 */
1501 	if ((tiflags & TH_ACK) == 0) {
1502 		if (tp->t_flags & TF_ACKNOW)
1503 			goto dropafterack;
1504 		else
1505 			goto drop;
1506 	}
1507 
1508 	/*
1509 	 * Ack processing.
1510 	 */
1511 	switch (tp->t_state) {
1512 
1513 	/*
1514 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1515 	 * ESTABLISHED state and continue processing.
1516 	 * The ACK was checked above.
1517 	 */
1518 	case TCPS_SYN_RECEIVED:
1519 		tcpstat.tcps_connects++;
1520 		soisconnected(so);
1521 		tp->t_state = TCPS_ESTABLISHED;
1522 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
1523 		/* Do window scaling? */
1524 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1525 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1526 			tp->snd_scale = tp->requested_s_scale;
1527 			tp->rcv_scale = tp->request_r_scale;
1528 			tiwin = th->th_win << tp->snd_scale;
1529 		}
1530 		tcp_flush_queue(tp);
1531 		tp->snd_wl1 = th->th_seq - 1;
1532 		/* fall into ... */
1533 
1534 	/*
1535 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1536 	 * ACKs.  If the ack is in the range
1537 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1538 	 * then advance tp->snd_una to th->th_ack and drop
1539 	 * data from the retransmission queue.  If this ACK reflects
1540 	 * more up to date window information we update our window information.
1541 	 */
1542 	case TCPS_ESTABLISHED:
1543 	case TCPS_FIN_WAIT_1:
1544 	case TCPS_FIN_WAIT_2:
1545 	case TCPS_CLOSE_WAIT:
1546 	case TCPS_CLOSING:
1547 	case TCPS_LAST_ACK:
1548 	case TCPS_TIME_WAIT:
1549 #ifdef TCP_ECN
1550 		/*
1551 		 * if we receive ECE and are not already in recovery phase,
1552 		 * reduce cwnd by half but don't slow-start.
1553 		 * advance snd_last to snd_max not to reduce cwnd again
1554 		 * until all outstanding packets are acked.
1555 		 */
1556 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1557 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1558 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1559 				u_int win;
1560 
1561 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1562 				if (win > 1) {
1563 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1564 					tp->snd_cwnd = tp->snd_ssthresh;
1565 					tp->snd_last = tp->snd_max;
1566 					tp->t_flags |= TF_SEND_CWR;
1567 					tcpstat.tcps_cwr_ecn++;
1568 				}
1569 			}
1570 			tcpstat.tcps_ecn_rcvece++;
1571 		}
1572 		/*
1573 		 * if we receive CWR, we know that the peer has reduced
1574 		 * its congestion window.  stop sending ecn-echo.
1575 		 */
1576 		if ((tiflags & TH_CWR)) {
1577 			tp->t_flags &= ~TF_RCVD_CE;
1578 			tcpstat.tcps_ecn_rcvcwr++;
1579 		}
1580 #endif /* TCP_ECN */
1581 
1582 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1583 			/*
1584 			 * Duplicate/old ACK processing.
1585 			 * Increments t_dupacks:
1586 			 *	Pure duplicate (same seq/ack/window, no data)
1587 			 * Doesn't affect t_dupacks:
1588 			 *	Data packets.
1589 			 *	Normal window updates (window opens)
1590 			 * Resets t_dupacks:
1591 			 *	New data ACKed.
1592 			 *	Window shrinks
1593 			 *	Old ACK
1594 			 */
1595 			if (tlen) {
1596 				/* Drop very old ACKs unless th_seq matches */
1597 				if (th->th_seq != tp->rcv_nxt &&
1598 				   SEQ_LT(th->th_ack,
1599 				   tp->snd_una - tp->max_sndwnd)) {
1600 					tcpstat.tcps_rcvacktooold++;
1601 					goto drop;
1602 				}
1603 				break;
1604 			}
1605 			/*
1606 			 * If we get an old ACK, there is probably packet
1607 			 * reordering going on.  Be conservative and reset
1608 			 * t_dupacks so that we are less aggressive in
1609 			 * doing a fast retransmit.
1610 			 */
1611 			if (th->th_ack != tp->snd_una) {
1612 				tp->t_dupacks = 0;
1613 				break;
1614 			}
1615 			if (tiwin == tp->snd_wnd) {
1616 				tcpstat.tcps_rcvdupack++;
1617 				/*
1618 				 * If we have outstanding data (other than
1619 				 * a window probe), this is a completely
1620 				 * duplicate ack (ie, window info didn't
1621 				 * change), the ack is the biggest we've
1622 				 * seen and we've seen exactly our rexmt
1623 				 * threshold of them, assume a packet
1624 				 * has been dropped and retransmit it.
1625 				 * Kludge snd_nxt & the congestion
1626 				 * window so we send only this one
1627 				 * packet.
1628 				 *
1629 				 * We know we're losing at the current
1630 				 * window size so do congestion avoidance
1631 				 * (set ssthresh to half the current window
1632 				 * and pull our congestion window back to
1633 				 * the new ssthresh).
1634 				 *
1635 				 * Dup acks mean that packets have left the
1636 				 * network (they're now cached at the receiver)
1637 				 * so bump cwnd by the amount in the receiver
1638 				 * to keep a constant cwnd packets in the
1639 				 * network.
1640 				 */
1641 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1642 					tp->t_dupacks = 0;
1643 #if defined(TCP_SACK) && defined(TCP_FACK)
1644 				/*
1645 				 * In FACK, can enter fast rec. if the receiver
1646 				 * reports a reass. queue longer than 3 segs.
1647 				 */
1648 				else if (++tp->t_dupacks == tcprexmtthresh ||
1649 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1650 				    tp->t_maxseg + tp->snd_una)) &&
1651 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1652 #else
1653 				else if (++tp->t_dupacks == tcprexmtthresh) {
1654 #endif /* TCP_FACK */
1655 					tcp_seq onxt = tp->snd_nxt;
1656 					u_long win =
1657 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1658 						2 / tp->t_maxseg;
1659 
1660 #if defined(TCP_SACK) || defined(TCP_ECN)
1661 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1662 						/*
1663 						 * False fast retx after
1664 						 * timeout.  Do not cut window.
1665 						 */
1666 						tp->t_dupacks = 0;
1667 						goto drop;
1668 					}
1669 #endif
1670 					if (win < 2)
1671 						win = 2;
1672 					tp->snd_ssthresh = win * tp->t_maxseg;
1673 #ifdef TCP_SACK
1674 					tp->snd_last = tp->snd_max;
1675 					if (tp->sack_enable) {
1676 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1677 						tp->t_rtttime = 0;
1678 #ifdef TCP_ECN
1679 						tp->t_flags |= TF_SEND_CWR;
1680 #endif
1681 						tcpstat.tcps_cwr_frecovery++;
1682 						tcpstat.tcps_sack_recovery_episode++;
1683 #if defined(TCP_SACK) && defined(TCP_FACK)
1684 						tp->t_dupacks = tcprexmtthresh;
1685 						(void) tcp_output(tp);
1686 						/*
1687 						 * During FR, snd_cwnd is held
1688 						 * constant for FACK.
1689 						 */
1690 						tp->snd_cwnd = tp->snd_ssthresh;
1691 #else
1692 						/*
1693 						 * tcp_output() will send
1694 						 * oldest SACK-eligible rtx.
1695 						 */
1696 						(void) tcp_output(tp);
1697 						tp->snd_cwnd = tp->snd_ssthresh+
1698 					           tp->t_maxseg * tp->t_dupacks;
1699 #endif /* TCP_FACK */
1700 						goto drop;
1701 					}
1702 #endif /* TCP_SACK */
1703 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1704 					tp->t_rtttime = 0;
1705 					tp->snd_nxt = th->th_ack;
1706 					tp->snd_cwnd = tp->t_maxseg;
1707 #ifdef TCP_ECN
1708 					tp->t_flags |= TF_SEND_CWR;
1709 #endif
1710 					tcpstat.tcps_cwr_frecovery++;
1711 					tcpstat.tcps_sndrexmitfast++;
1712 					(void) tcp_output(tp);
1713 
1714 					tp->snd_cwnd = tp->snd_ssthresh +
1715 					    tp->t_maxseg * tp->t_dupacks;
1716 					if (SEQ_GT(onxt, tp->snd_nxt))
1717 						tp->snd_nxt = onxt;
1718 					goto drop;
1719 				} else if (tp->t_dupacks > tcprexmtthresh) {
1720 #if defined(TCP_SACK) && defined(TCP_FACK)
1721 					/*
1722 					 * while (awnd < cwnd)
1723 					 *         sendsomething();
1724 					 */
1725 					if (tp->sack_enable) {
1726 						if (tp->snd_awnd < tp->snd_cwnd)
1727 							tcp_output(tp);
1728 						goto drop;
1729 					}
1730 #endif /* TCP_FACK */
1731 					tp->snd_cwnd += tp->t_maxseg;
1732 					(void) tcp_output(tp);
1733 					goto drop;
1734 				}
1735 			} else if (tiwin < tp->snd_wnd) {
1736 				/*
1737 				 * The window was retracted!  Previous dup
1738 				 * ACKs may have been due to packets arriving
1739 				 * after the shrunken window, not a missing
1740 				 * packet, so play it safe and reset t_dupacks
1741 				 */
1742 				tp->t_dupacks = 0;
1743 			}
1744 			break;
1745 		}
1746 		/*
1747 		 * If the congestion window was inflated to account
1748 		 * for the other side's cached packets, retract it.
1749 		 */
1750 #if defined(TCP_SACK)
1751 		if (tp->sack_enable) {
1752 			if (tp->t_dupacks >= tcprexmtthresh) {
1753 				/* Check for a partial ACK */
1754 				if (tcp_sack_partialack(tp, th)) {
1755 #if defined(TCP_SACK) && defined(TCP_FACK)
1756 					/* Force call to tcp_output */
1757 					if (tp->snd_awnd < tp->snd_cwnd)
1758 						tp->t_flags |= TF_NEEDOUTPUT;
1759 #else
1760 					tp->snd_cwnd += tp->t_maxseg;
1761 					tp->t_flags |= TF_NEEDOUTPUT;
1762 #endif /* TCP_FACK */
1763 				} else {
1764 					/* Out of fast recovery */
1765 					tp->snd_cwnd = tp->snd_ssthresh;
1766 					if (tcp_seq_subtract(tp->snd_max,
1767 					    th->th_ack) < tp->snd_ssthresh)
1768 						tp->snd_cwnd =
1769 						   tcp_seq_subtract(tp->snd_max,
1770 					           th->th_ack);
1771 					tp->t_dupacks = 0;
1772 #if defined(TCP_SACK) && defined(TCP_FACK)
1773 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1774 						tp->snd_fack = th->th_ack;
1775 #endif /* TCP_FACK */
1776 				}
1777 			}
1778 		} else {
1779 			if (tp->t_dupacks >= tcprexmtthresh &&
1780 			    !tcp_newreno(tp, th)) {
1781 				/* Out of fast recovery */
1782 				tp->snd_cwnd = tp->snd_ssthresh;
1783 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1784 				    tp->snd_ssthresh)
1785 					tp->snd_cwnd =
1786 					    tcp_seq_subtract(tp->snd_max,
1787 					    th->th_ack);
1788 				tp->t_dupacks = 0;
1789 			}
1790 		}
1791 		if (tp->t_dupacks < tcprexmtthresh)
1792 			tp->t_dupacks = 0;
1793 #else /* else no TCP_SACK */
1794 		if (tp->t_dupacks >= tcprexmtthresh &&
1795 		    tp->snd_cwnd > tp->snd_ssthresh)
1796 			tp->snd_cwnd = tp->snd_ssthresh;
1797 		tp->t_dupacks = 0;
1798 #endif
1799 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1800 			tcpstat.tcps_rcvacktoomuch++;
1801 			goto dropafterack_ratelim;
1802 		}
1803 		acked = th->th_ack - tp->snd_una;
1804 		tcpstat.tcps_rcvackpack++;
1805 		tcpstat.tcps_rcvackbyte += acked;
1806 
1807 		/*
1808 		 * If we have a timestamp reply, update smoothed
1809 		 * round trip time.  If no timestamp is present but
1810 		 * transmit timer is running and timed sequence
1811 		 * number was acked, update smoothed round trip time.
1812 		 * Since we now have an rtt measurement, cancel the
1813 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1814 		 * Recompute the initial retransmit timer.
1815 		 */
1816 		if (opti.ts_present && opti.ts_ecr)
1817 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr);
1818 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1819 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1820 
1821 		/*
1822 		 * If all outstanding data is acked, stop retransmit
1823 		 * timer and remember to restart (more output or persist).
1824 		 * If there is more data to be acked, restart retransmit
1825 		 * timer, using current (possibly backed-off) value.
1826 		 */
1827 		if (th->th_ack == tp->snd_max) {
1828 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1829 			tp->t_flags |= TF_NEEDOUTPUT;
1830 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1831 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1832 		/*
1833 		 * When new data is acked, open the congestion window.
1834 		 * If the window gives us less than ssthresh packets
1835 		 * in flight, open exponentially (maxseg per packet).
1836 		 * Otherwise open linearly: maxseg per window
1837 		 * (maxseg^2 / cwnd per packet).
1838 		 */
1839 		{
1840 		u_int cw = tp->snd_cwnd;
1841 		u_int incr = tp->t_maxseg;
1842 
1843 		if (cw > tp->snd_ssthresh)
1844 			incr = incr * incr / cw;
1845 #if defined (TCP_SACK)
1846 		if (tp->t_dupacks < tcprexmtthresh)
1847 #endif
1848 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1849 		}
1850 		ND6_HINT(tp);
1851 		if (acked > so->so_snd.sb_cc) {
1852 			tp->snd_wnd -= so->so_snd.sb_cc;
1853 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1854 			ourfinisacked = 1;
1855 		} else {
1856 			sbdrop(&so->so_snd, acked);
1857 			tp->snd_wnd -= acked;
1858 			ourfinisacked = 0;
1859 		}
1860 
1861 		tcp_update_sndspace(tp);
1862 		if (sb_notify(&so->so_snd)) {
1863 			tp->t_flags |= TF_BLOCKOUTPUT;
1864 			sowwakeup(so);
1865 			tp->t_flags &= ~TF_BLOCKOUTPUT;
1866 		}
1867 
1868 		/*
1869 		 * If we had a pending ICMP message that referred to data
1870 		 * that have just been acknowledged, disregard the recorded
1871 		 * ICMP message.
1872 		 */
1873 		if ((tp->t_flags & TF_PMTUD_PEND) &&
1874 		    SEQ_GT(th->th_ack, tp->t_pmtud_th_seq))
1875 			tp->t_flags &= ~TF_PMTUD_PEND;
1876 
1877 		/*
1878 		 * Keep track of the largest chunk of data acknowledged
1879 		 * since last PMTU update
1880 		 */
1881 		if (tp->t_pmtud_mss_acked < acked)
1882 			tp->t_pmtud_mss_acked = acked;
1883 
1884 		tp->snd_una = th->th_ack;
1885 #ifdef TCP_ECN
1886 		/* sync snd_last with snd_una */
1887 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1888 			tp->snd_last = tp->snd_una;
1889 #endif
1890 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1891 			tp->snd_nxt = tp->snd_una;
1892 #if defined (TCP_SACK) && defined (TCP_FACK)
1893 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1894 			tp->snd_fack = tp->snd_una;
1895 			/* Update snd_awnd for partial ACK
1896 			 * without any SACK blocks.
1897 			 */
1898 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1899 				tp->snd_fack) + tp->retran_data;
1900 		}
1901 #endif
1902 
1903 		switch (tp->t_state) {
1904 
1905 		/*
1906 		 * In FIN_WAIT_1 STATE in addition to the processing
1907 		 * for the ESTABLISHED state if our FIN is now acknowledged
1908 		 * then enter FIN_WAIT_2.
1909 		 */
1910 		case TCPS_FIN_WAIT_1:
1911 			if (ourfinisacked) {
1912 				/*
1913 				 * If we can't receive any more
1914 				 * data, then closing user can proceed.
1915 				 * Starting the timer is contrary to the
1916 				 * specification, but if we don't get a FIN
1917 				 * we'll hang forever.
1918 				 */
1919 				if (so->so_state & SS_CANTRCVMORE) {
1920 					soisdisconnected(so);
1921 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1922 				}
1923 				tp->t_state = TCPS_FIN_WAIT_2;
1924 			}
1925 			break;
1926 
1927 		/*
1928 		 * In CLOSING STATE in addition to the processing for
1929 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1930 		 * then enter the TIME-WAIT state, otherwise ignore
1931 		 * the segment.
1932 		 */
1933 		case TCPS_CLOSING:
1934 			if (ourfinisacked) {
1935 				tp->t_state = TCPS_TIME_WAIT;
1936 				tcp_canceltimers(tp);
1937 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1938 				soisdisconnected(so);
1939 			}
1940 			break;
1941 
1942 		/*
1943 		 * In LAST_ACK, we may still be waiting for data to drain
1944 		 * and/or to be acked, as well as for the ack of our FIN.
1945 		 * If our FIN is now acknowledged, delete the TCB,
1946 		 * enter the closed state and return.
1947 		 */
1948 		case TCPS_LAST_ACK:
1949 			if (ourfinisacked) {
1950 				tp = tcp_close(tp);
1951 				goto drop;
1952 			}
1953 			break;
1954 
1955 		/*
1956 		 * In TIME_WAIT state the only thing that should arrive
1957 		 * is a retransmission of the remote FIN.  Acknowledge
1958 		 * it and restart the finack timer.
1959 		 */
1960 		case TCPS_TIME_WAIT:
1961 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1962 			goto dropafterack;
1963 		}
1964 	}
1965 
1966 step6:
1967 	/*
1968 	 * Update window information.
1969 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1970 	 */
1971 	if ((tiflags & TH_ACK) &&
1972 	    (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq &&
1973 	    (SEQ_LT(tp->snd_wl2, th->th_ack) ||
1974 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
1975 		/* keep track of pure window updates */
1976 		if (tlen == 0 &&
1977 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
1978 			tcpstat.tcps_rcvwinupd++;
1979 		tp->snd_wnd = tiwin;
1980 		tp->snd_wl1 = th->th_seq;
1981 		tp->snd_wl2 = th->th_ack;
1982 		if (tp->snd_wnd > tp->max_sndwnd)
1983 			tp->max_sndwnd = tp->snd_wnd;
1984 		tp->t_flags |= TF_NEEDOUTPUT;
1985 	}
1986 
1987 	/*
1988 	 * Process segments with URG.
1989 	 */
1990 	if ((tiflags & TH_URG) && th->th_urp &&
1991 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1992 		/*
1993 		 * This is a kludge, but if we receive and accept
1994 		 * random urgent pointers, we'll crash in
1995 		 * soreceive.  It's hard to imagine someone
1996 		 * actually wanting to send this much urgent data.
1997 		 */
1998 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
1999 			th->th_urp = 0;			/* XXX */
2000 			tiflags &= ~TH_URG;		/* XXX */
2001 			goto dodata;			/* XXX */
2002 		}
2003 		/*
2004 		 * If this segment advances the known urgent pointer,
2005 		 * then mark the data stream.  This should not happen
2006 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2007 		 * a FIN has been received from the remote side.
2008 		 * In these states we ignore the URG.
2009 		 *
2010 		 * According to RFC961 (Assigned Protocols),
2011 		 * the urgent pointer points to the last octet
2012 		 * of urgent data.  We continue, however,
2013 		 * to consider it to indicate the first octet
2014 		 * of data past the urgent section as the original
2015 		 * spec states (in one of two places).
2016 		 */
2017 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2018 			tp->rcv_up = th->th_seq + th->th_urp;
2019 			so->so_oobmark = so->so_rcv.sb_cc +
2020 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2021 			if (so->so_oobmark == 0)
2022 				so->so_state |= SS_RCVATMARK;
2023 			sohasoutofband(so);
2024 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2025 		}
2026 		/*
2027 		 * Remove out of band data so doesn't get presented to user.
2028 		 * This can happen independent of advancing the URG pointer,
2029 		 * but if two URG's are pending at once, some out-of-band
2030 		 * data may creep in... ick.
2031 		 */
2032 		if (th->th_urp <= (u_int16_t) tlen &&
2033 		    (so->so_options & SO_OOBINLINE) == 0)
2034 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2035 	} else
2036 		/*
2037 		 * If no out of band data is expected,
2038 		 * pull receive urgent pointer along
2039 		 * with the receive window.
2040 		 */
2041 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2042 			tp->rcv_up = tp->rcv_nxt;
2043 dodata:							/* XXX */
2044 
2045 	/*
2046 	 * Process the segment text, merging it into the TCP sequencing queue,
2047 	 * and arranging for acknowledgment of receipt if necessary.
2048 	 * This process logically involves adjusting tp->rcv_wnd as data
2049 	 * is presented to the user (this happens in tcp_usrreq.c,
2050 	 * case PRU_RCVD).  If a FIN has already been received on this
2051 	 * connection then we just ignore the text.
2052 	 */
2053 	if ((tlen || (tiflags & TH_FIN)) &&
2054 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2055 #ifdef TCP_SACK
2056 		tcp_seq laststart = th->th_seq;
2057 		tcp_seq lastend = th->th_seq + tlen;
2058 #endif
2059 		if (th->th_seq == tp->rcv_nxt && TAILQ_EMPTY(&tp->t_segq) &&
2060 		    tp->t_state == TCPS_ESTABLISHED) {
2061 			TCP_SETUP_ACK(tp, tiflags, m);
2062 			tp->rcv_nxt += tlen;
2063 			tiflags = th->th_flags & TH_FIN;
2064 			tcpstat.tcps_rcvpack++;
2065 			tcpstat.tcps_rcvbyte += tlen;
2066 			ND6_HINT(tp);
2067 			if (so->so_state & SS_CANTRCVMORE)
2068 				m_freem(m);
2069 			else {
2070 				m_adj(m, hdroptlen);
2071 				sbappendstream(&so->so_rcv, m);
2072 			}
2073 			tp->t_flags |= TF_BLOCKOUTPUT;
2074 			sorwakeup(so);
2075 			tp->t_flags &= ~TF_BLOCKOUTPUT;
2076 		} else {
2077 			m_adj(m, hdroptlen);
2078 			tiflags = tcp_reass(tp, th, m, &tlen);
2079 			tp->t_flags |= TF_ACKNOW;
2080 		}
2081 #ifdef TCP_SACK
2082 		if (tp->sack_enable)
2083 			tcp_update_sack_list(tp, laststart, lastend);
2084 #endif
2085 
2086 		/*
2087 		 * variable len never referenced again in modern BSD,
2088 		 * so why bother computing it ??
2089 		 */
2090 #if 0
2091 		/*
2092 		 * Note the amount of data that peer has sent into
2093 		 * our window, in order to estimate the sender's
2094 		 * buffer size.
2095 		 */
2096 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2097 #endif /* 0 */
2098 	} else {
2099 		m_freem(m);
2100 		tiflags &= ~TH_FIN;
2101 	}
2102 
2103 	/*
2104 	 * If FIN is received ACK the FIN and let the user know
2105 	 * that the connection is closing.  Ignore a FIN received before
2106 	 * the connection is fully established.
2107 	 */
2108 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2109 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2110 			socantrcvmore(so);
2111 			tp->t_flags |= TF_ACKNOW;
2112 			tp->rcv_nxt++;
2113 		}
2114 		switch (tp->t_state) {
2115 
2116 		/*
2117 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2118 		 */
2119 		case TCPS_ESTABLISHED:
2120 			tp->t_state = TCPS_CLOSE_WAIT;
2121 			break;
2122 
2123 		/*
2124 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2125 		 * enter the CLOSING state.
2126 		 */
2127 		case TCPS_FIN_WAIT_1:
2128 			tp->t_state = TCPS_CLOSING;
2129 			break;
2130 
2131 		/*
2132 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2133 		 * starting the time-wait timer, turning off the other
2134 		 * standard timers.
2135 		 */
2136 		case TCPS_FIN_WAIT_2:
2137 			tp->t_state = TCPS_TIME_WAIT;
2138 			tcp_canceltimers(tp);
2139 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2140 			soisdisconnected(so);
2141 			break;
2142 
2143 		/*
2144 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2145 		 */
2146 		case TCPS_TIME_WAIT:
2147 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2148 			break;
2149 		}
2150 	}
2151 	if (so->so_options & SO_DEBUG) {
2152 		switch (tp->pf) {
2153 #ifdef INET6
2154 		case PF_INET6:
2155 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2156 			    0, tlen);
2157 			break;
2158 #endif /* INET6 */
2159 		case PF_INET:
2160 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2161 			    0, tlen);
2162 			break;
2163 		}
2164 	}
2165 
2166 	/*
2167 	 * Return any desired output.
2168 	 */
2169 	if (tp->t_flags & (TF_ACKNOW|TF_NEEDOUTPUT))
2170 		(void) tcp_output(tp);
2171 	return;
2172 
2173 badsyn:
2174 	/*
2175 	 * Received a bad SYN.  Increment counters and dropwithreset.
2176 	 */
2177 	tcpstat.tcps_badsyn++;
2178 	tp = NULL;
2179 	goto dropwithreset;
2180 
2181 dropafterack_ratelim:
2182 	if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count,
2183 	    tcp_ackdrop_ppslim) == 0) {
2184 		/* XXX stat */
2185 		goto drop;
2186 	}
2187 	/* ...fall into dropafterack... */
2188 
2189 dropafterack:
2190 	/*
2191 	 * Generate an ACK dropping incoming segment if it occupies
2192 	 * sequence space, where the ACK reflects our state.
2193 	 */
2194 	if (tiflags & TH_RST)
2195 		goto drop;
2196 	m_freem(m);
2197 	tp->t_flags |= TF_ACKNOW;
2198 	(void) tcp_output(tp);
2199 	return;
2200 
2201 dropwithreset_ratelim:
2202 	/*
2203 	 * We may want to rate-limit RSTs in certain situations,
2204 	 * particularly if we are sending an RST in response to
2205 	 * an attempt to connect to or otherwise communicate with
2206 	 * a port for which we have no socket.
2207 	 */
2208 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2209 	    tcp_rst_ppslim) == 0) {
2210 		/* XXX stat */
2211 		goto drop;
2212 	}
2213 	/* ...fall into dropwithreset... */
2214 
2215 dropwithreset:
2216 	/*
2217 	 * Generate a RST, dropping incoming segment.
2218 	 * Make ACK acceptable to originator of segment.
2219 	 * Don't bother to respond to RST.
2220 	 */
2221 	if (tiflags & TH_RST)
2222 		goto drop;
2223 	if (tiflags & TH_ACK) {
2224 		tcp_respond(tp, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack,
2225 		    TH_RST, m->m_pkthdr.ph_rtableid);
2226 	} else {
2227 		if (tiflags & TH_SYN)
2228 			tlen++;
2229 		tcp_respond(tp, mtod(m, caddr_t), th, th->th_seq + tlen,
2230 		    (tcp_seq)0, TH_RST|TH_ACK, m->m_pkthdr.ph_rtableid);
2231 	}
2232 	m_freem(m);
2233 	return;
2234 
2235 drop:
2236 	/*
2237 	 * Drop space held by incoming segment and return.
2238 	 */
2239 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2240 		switch (tp->pf) {
2241 #ifdef INET6
2242 		case PF_INET6:
2243 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2244 			    0, tlen);
2245 			break;
2246 #endif /* INET6 */
2247 		case PF_INET:
2248 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2249 			    0, tlen);
2250 			break;
2251 		}
2252 	}
2253 
2254 	m_freem(m);
2255 	return;
2256 }
2257 
2258 int
2259 tcp_dooptions(struct tcpcb *tp, u_char *cp, int cnt, struct tcphdr *th,
2260     struct mbuf *m, int iphlen, struct tcp_opt_info *oi,
2261     u_int rtableid)
2262 {
2263 	u_int16_t mss = 0;
2264 	int opt, optlen;
2265 #ifdef TCP_SIGNATURE
2266 	caddr_t sigp = NULL;
2267 	struct tdb *tdb = NULL;
2268 #endif /* TCP_SIGNATURE */
2269 
2270 	for (; cp && cnt > 0; cnt -= optlen, cp += optlen) {
2271 		opt = cp[0];
2272 		if (opt == TCPOPT_EOL)
2273 			break;
2274 		if (opt == TCPOPT_NOP)
2275 			optlen = 1;
2276 		else {
2277 			if (cnt < 2)
2278 				break;
2279 			optlen = cp[1];
2280 			if (optlen < 2 || optlen > cnt)
2281 				break;
2282 		}
2283 		switch (opt) {
2284 
2285 		default:
2286 			continue;
2287 
2288 		case TCPOPT_MAXSEG:
2289 			if (optlen != TCPOLEN_MAXSEG)
2290 				continue;
2291 			if (!(th->th_flags & TH_SYN))
2292 				continue;
2293 			if (TCPS_HAVERCVDSYN(tp->t_state))
2294 				continue;
2295 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2296 			mss = ntohs(mss);
2297 			oi->maxseg = mss;
2298 			break;
2299 
2300 		case TCPOPT_WINDOW:
2301 			if (optlen != TCPOLEN_WINDOW)
2302 				continue;
2303 			if (!(th->th_flags & TH_SYN))
2304 				continue;
2305 			if (TCPS_HAVERCVDSYN(tp->t_state))
2306 				continue;
2307 			tp->t_flags |= TF_RCVD_SCALE;
2308 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2309 			break;
2310 
2311 		case TCPOPT_TIMESTAMP:
2312 			if (optlen != TCPOLEN_TIMESTAMP)
2313 				continue;
2314 			oi->ts_present = 1;
2315 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
2316 			oi->ts_val = ntohl(oi->ts_val);
2317 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
2318 			oi->ts_ecr = ntohl(oi->ts_ecr);
2319 
2320 			if (!(th->th_flags & TH_SYN))
2321 				continue;
2322 			if (TCPS_HAVERCVDSYN(tp->t_state))
2323 				continue;
2324 			/*
2325 			 * A timestamp received in a SYN makes
2326 			 * it ok to send timestamp requests and replies.
2327 			 */
2328 			tp->t_flags |= TF_RCVD_TSTMP;
2329 			tp->ts_recent = oi->ts_val;
2330 			tp->ts_recent_age = tcp_now;
2331 			break;
2332 
2333 #ifdef TCP_SACK
2334 		case TCPOPT_SACK_PERMITTED:
2335 			if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED)
2336 				continue;
2337 			if (!(th->th_flags & TH_SYN))
2338 				continue;
2339 			if (TCPS_HAVERCVDSYN(tp->t_state))
2340 				continue;
2341 			/* MUST only be set on SYN */
2342 			tp->t_flags |= TF_SACK_PERMIT;
2343 			break;
2344 		case TCPOPT_SACK:
2345 			tcp_sack_option(tp, th, cp, optlen);
2346 			break;
2347 #endif
2348 #ifdef TCP_SIGNATURE
2349 		case TCPOPT_SIGNATURE:
2350 			if (optlen != TCPOLEN_SIGNATURE)
2351 				continue;
2352 
2353 			if (sigp && timingsafe_bcmp(sigp, cp + 2, 16))
2354 				return (-1);
2355 
2356 			sigp = cp + 2;
2357 			break;
2358 #endif /* TCP_SIGNATURE */
2359 		}
2360 	}
2361 
2362 #ifdef TCP_SIGNATURE
2363 	if (tp->t_flags & TF_SIGNATURE) {
2364 		union sockaddr_union src, dst;
2365 
2366 		memset(&src, 0, sizeof(union sockaddr_union));
2367 		memset(&dst, 0, sizeof(union sockaddr_union));
2368 
2369 		switch (tp->pf) {
2370 		case 0:
2371 		case AF_INET:
2372 			src.sa.sa_len = sizeof(struct sockaddr_in);
2373 			src.sa.sa_family = AF_INET;
2374 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
2375 			dst.sa.sa_len = sizeof(struct sockaddr_in);
2376 			dst.sa.sa_family = AF_INET;
2377 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
2378 			break;
2379 #ifdef INET6
2380 		case AF_INET6:
2381 			src.sa.sa_len = sizeof(struct sockaddr_in6);
2382 			src.sa.sa_family = AF_INET6;
2383 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
2384 			dst.sa.sa_len = sizeof(struct sockaddr_in6);
2385 			dst.sa.sa_family = AF_INET6;
2386 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
2387 			break;
2388 #endif /* INET6 */
2389 		}
2390 
2391 		tdb = gettdbbysrcdst(rtable_l2(rtableid),
2392 		    0, &src, &dst, IPPROTO_TCP);
2393 
2394 		/*
2395 		 * We don't have an SA for this peer, so we turn off
2396 		 * TF_SIGNATURE on the listen socket
2397 		 */
2398 		if (tdb == NULL && tp->t_state == TCPS_LISTEN)
2399 			tp->t_flags &= ~TF_SIGNATURE;
2400 
2401 	}
2402 
2403 	if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) {
2404 		tcpstat.tcps_rcvbadsig++;
2405 		return (-1);
2406 	}
2407 
2408 	if (sigp) {
2409 		char sig[16];
2410 
2411 		if (tdb == NULL) {
2412 			tcpstat.tcps_rcvbadsig++;
2413 			return (-1);
2414 		}
2415 
2416 		if (tcp_signature(tdb, tp->pf, m, th, iphlen, 1, sig) < 0)
2417 			return (-1);
2418 
2419 		if (timingsafe_bcmp(sig, sigp, 16)) {
2420 			tcpstat.tcps_rcvbadsig++;
2421 			return (-1);
2422 		}
2423 
2424 		tcpstat.tcps_rcvgoodsig++;
2425 	}
2426 #endif /* TCP_SIGNATURE */
2427 
2428 	return (0);
2429 }
2430 
2431 #if defined(TCP_SACK)
2432 u_long
2433 tcp_seq_subtract(u_long a, u_long b)
2434 {
2435 	return ((long)(a - b));
2436 }
2437 #endif
2438 
2439 
2440 #ifdef TCP_SACK
2441 /*
2442  * This function is called upon receipt of new valid data (while not in header
2443  * prediction mode), and it updates the ordered list of sacks.
2444  */
2445 void
2446 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_laststart,
2447     tcp_seq rcv_lastend)
2448 {
2449 	/*
2450 	 * First reported block MUST be the most recent one.  Subsequent
2451 	 * blocks SHOULD be in the order in which they arrived at the
2452 	 * receiver.  These two conditions make the implementation fully
2453 	 * compliant with RFC 2018.
2454 	 */
2455 	int i, j = 0, count = 0, lastpos = -1;
2456 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2457 
2458 	/* First clean up current list of sacks */
2459 	for (i = 0; i < tp->rcv_numsacks; i++) {
2460 		sack = tp->sackblks[i];
2461 		if (sack.start == 0 && sack.end == 0) {
2462 			count++; /* count = number of blocks to be discarded */
2463 			continue;
2464 		}
2465 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2466 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2467 			count++;
2468 		} else {
2469 			temp[j].start = tp->sackblks[i].start;
2470 			temp[j++].end = tp->sackblks[i].end;
2471 		}
2472 	}
2473 	tp->rcv_numsacks -= count;
2474 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2475 		tcp_clean_sackreport(tp);
2476 		if (SEQ_LT(tp->rcv_nxt, rcv_laststart)) {
2477 			/* ==> need first sack block */
2478 			tp->sackblks[0].start = rcv_laststart;
2479 			tp->sackblks[0].end = rcv_lastend;
2480 			tp->rcv_numsacks = 1;
2481 		}
2482 		return;
2483 	}
2484 	/* Otherwise, sack blocks are already present. */
2485 	for (i = 0; i < tp->rcv_numsacks; i++)
2486 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2487 	if (SEQ_GEQ(tp->rcv_nxt, rcv_lastend))
2488 		return;     /* sack list remains unchanged */
2489 	/*
2490 	 * From here, segment just received should be (part of) the 1st sack.
2491 	 * Go through list, possibly coalescing sack block entries.
2492 	 */
2493 	firstsack.start = rcv_laststart;
2494 	firstsack.end = rcv_lastend;
2495 	for (i = 0; i < tp->rcv_numsacks; i++) {
2496 		sack = tp->sackblks[i];
2497 		if (SEQ_LT(sack.end, firstsack.start) ||
2498 		    SEQ_GT(sack.start, firstsack.end))
2499 			continue; /* no overlap */
2500 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2501 			/*
2502 			 * identical block; delete it here since we will
2503 			 * move it to the front of the list.
2504 			 */
2505 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2506 			lastpos = i;    /* last posn with a zero entry */
2507 			continue;
2508 		}
2509 		if (SEQ_LEQ(sack.start, firstsack.start))
2510 			firstsack.start = sack.start; /* merge blocks */
2511 		if (SEQ_GEQ(sack.end, firstsack.end))
2512 			firstsack.end = sack.end;     /* merge blocks */
2513 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2514 		lastpos = i;    /* last posn with a zero entry */
2515 	}
2516 	if (lastpos != -1) {    /* at least one merge */
2517 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2518 			sack = tp->sackblks[i];
2519 			if (sack.start == 0 && sack.end == 0)
2520 				continue;
2521 			temp[j++] = sack;
2522 		}
2523 		tp->rcv_numsacks = j; /* including first blk (added later) */
2524 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2525 			tp->sackblks[i] = temp[i];
2526 	} else {        /* no merges -- shift sacks by 1 */
2527 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2528 			tp->rcv_numsacks++;
2529 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2530 			tp->sackblks[i] = tp->sackblks[i-1];
2531 	}
2532 	tp->sackblks[0] = firstsack;
2533 	return;
2534 }
2535 
2536 /*
2537  * Process the TCP SACK option.  tp->snd_holes is an ordered list
2538  * of holes (oldest to newest, in terms of the sequence space).
2539  */
2540 void
2541 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2542 {
2543 	int tmp_olen;
2544 	u_char *tmp_cp;
2545 	struct sackhole *cur, *p, *temp;
2546 
2547 	if (!tp->sack_enable)
2548 		return;
2549 	/* SACK without ACK doesn't make sense. */
2550 	if ((th->th_flags & TH_ACK) == 0)
2551 	       return;
2552 	/* Make sure the ACK on this segment is in [snd_una, snd_max]. */
2553 	if (SEQ_LT(th->th_ack, tp->snd_una) ||
2554 	    SEQ_GT(th->th_ack, tp->snd_max))
2555 		return;
2556 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2557 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2558 		return;
2559 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2560 	tmp_cp = cp + 2;
2561 	tmp_olen = optlen - 2;
2562 	tcpstat.tcps_sack_rcv_opts++;
2563 	if (tp->snd_numholes < 0)
2564 		tp->snd_numholes = 0;
2565 	if (tp->t_maxseg == 0)
2566 		panic("tcp_sack_option"); /* Should never happen */
2567 	while (tmp_olen > 0) {
2568 		struct sackblk sack;
2569 
2570 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2571 		sack.start = ntohl(sack.start);
2572 		bcopy(tmp_cp + sizeof(tcp_seq),
2573 		    (char *) &(sack.end), sizeof(tcp_seq));
2574 		sack.end = ntohl(sack.end);
2575 		tmp_olen -= TCPOLEN_SACK;
2576 		tmp_cp += TCPOLEN_SACK;
2577 		if (SEQ_LEQ(sack.end, sack.start))
2578 			continue; /* bad SACK fields */
2579 		if (SEQ_LEQ(sack.end, tp->snd_una))
2580 			continue; /* old block */
2581 #if defined(TCP_SACK) && defined(TCP_FACK)
2582 		/* Updates snd_fack.  */
2583 		if (SEQ_GT(sack.end, tp->snd_fack))
2584 			tp->snd_fack = sack.end;
2585 #endif /* TCP_FACK */
2586 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2587 			if (SEQ_LT(sack.start, th->th_ack))
2588 				continue;
2589 		}
2590 		if (SEQ_GT(sack.end, tp->snd_max))
2591 			continue;
2592 		if (tp->snd_holes == NULL) { /* first hole */
2593 			tp->snd_holes = (struct sackhole *)
2594 			    pool_get(&sackhl_pool, PR_NOWAIT);
2595 			if (tp->snd_holes == NULL) {
2596 				/* ENOBUFS, so ignore SACKed block for now*/
2597 				goto done;
2598 			}
2599 			cur = tp->snd_holes;
2600 			cur->start = th->th_ack;
2601 			cur->end = sack.start;
2602 			cur->rxmit = cur->start;
2603 			cur->next = NULL;
2604 			tp->snd_numholes = 1;
2605 			tp->rcv_lastsack = sack.end;
2606 			/*
2607 			 * dups is at least one.  If more data has been
2608 			 * SACKed, it can be greater than one.
2609 			 */
2610 			cur->dups = min(tcprexmtthresh,
2611 			    ((sack.end - cur->end)/tp->t_maxseg));
2612 			if (cur->dups < 1)
2613 				cur->dups = 1;
2614 			continue; /* with next sack block */
2615 		}
2616 		/* Go thru list of holes:  p = previous,  cur = current */
2617 		p = cur = tp->snd_holes;
2618 		while (cur) {
2619 			if (SEQ_LEQ(sack.end, cur->start))
2620 				/* SACKs data before the current hole */
2621 				break; /* no use going through more holes */
2622 			if (SEQ_GEQ(sack.start, cur->end)) {
2623 				/* SACKs data beyond the current hole */
2624 				cur->dups++;
2625 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2626 				    tcprexmtthresh)
2627 					cur->dups = tcprexmtthresh;
2628 				p = cur;
2629 				cur = cur->next;
2630 				continue;
2631 			}
2632 			if (SEQ_LEQ(sack.start, cur->start)) {
2633 				/* Data acks at least the beginning of hole */
2634 #if defined(TCP_SACK) && defined(TCP_FACK)
2635 				if (SEQ_GT(sack.end, cur->rxmit))
2636 					tp->retran_data -=
2637 					    tcp_seq_subtract(cur->rxmit,
2638 					    cur->start);
2639 				else
2640 					tp->retran_data -=
2641 					    tcp_seq_subtract(sack.end,
2642 					    cur->start);
2643 #endif /* TCP_FACK */
2644 				if (SEQ_GEQ(sack.end, cur->end)) {
2645 					/* Acks entire hole, so delete hole */
2646 					if (p != cur) {
2647 						p->next = cur->next;
2648 						pool_put(&sackhl_pool, cur);
2649 						cur = p->next;
2650 					} else {
2651 						cur = cur->next;
2652 						pool_put(&sackhl_pool, p);
2653 						p = cur;
2654 						tp->snd_holes = p;
2655 					}
2656 					tp->snd_numholes--;
2657 					continue;
2658 				}
2659 				/* otherwise, move start of hole forward */
2660 				cur->start = sack.end;
2661 				cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
2662 				p = cur;
2663 				cur = cur->next;
2664 				continue;
2665 			}
2666 			/* move end of hole backward */
2667 			if (SEQ_GEQ(sack.end, cur->end)) {
2668 #if defined(TCP_SACK) && defined(TCP_FACK)
2669 				if (SEQ_GT(cur->rxmit, sack.start))
2670 					tp->retran_data -=
2671 					    tcp_seq_subtract(cur->rxmit,
2672 					    sack.start);
2673 #endif /* TCP_FACK */
2674 				cur->end = sack.start;
2675 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2676 				cur->dups++;
2677 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2678 				    tcprexmtthresh)
2679 					cur->dups = tcprexmtthresh;
2680 				p = cur;
2681 				cur = cur->next;
2682 				continue;
2683 			}
2684 			if (SEQ_LT(cur->start, sack.start) &&
2685 			    SEQ_GT(cur->end, sack.end)) {
2686 				/*
2687 				 * ACKs some data in middle of a hole; need to
2688 				 * split current hole
2689 				 */
2690 				temp = (struct sackhole *)
2691 				    pool_get(&sackhl_pool, PR_NOWAIT);
2692 				if (temp == NULL)
2693 					goto done; /* ENOBUFS */
2694 #if defined(TCP_SACK) && defined(TCP_FACK)
2695 				if (SEQ_GT(cur->rxmit, sack.end))
2696 					tp->retran_data -=
2697 					    tcp_seq_subtract(sack.end,
2698 					    sack.start);
2699 				else if (SEQ_GT(cur->rxmit, sack.start))
2700 					tp->retran_data -=
2701 					    tcp_seq_subtract(cur->rxmit,
2702 					    sack.start);
2703 #endif /* TCP_FACK */
2704 				temp->next = cur->next;
2705 				temp->start = sack.end;
2706 				temp->end = cur->end;
2707 				temp->dups = cur->dups;
2708 				temp->rxmit = SEQ_MAX(cur->rxmit, temp->start);
2709 				cur->end = sack.start;
2710 				cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
2711 				cur->dups++;
2712 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2713 					tcprexmtthresh)
2714 					cur->dups = tcprexmtthresh;
2715 				cur->next = temp;
2716 				p = temp;
2717 				cur = p->next;
2718 				tp->snd_numholes++;
2719 			}
2720 		}
2721 		/* At this point, p points to the last hole on the list */
2722 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2723 			/*
2724 			 * Need to append new hole at end.
2725 			 * Last hole is p (and it's not NULL).
2726 			 */
2727 			temp = (struct sackhole *)
2728 			    pool_get(&sackhl_pool, PR_NOWAIT);
2729 			if (temp == NULL)
2730 				goto done; /* ENOBUFS */
2731 			temp->start = tp->rcv_lastsack;
2732 			temp->end = sack.start;
2733 			temp->dups = min(tcprexmtthresh,
2734 			    ((sack.end - sack.start)/tp->t_maxseg));
2735 			if (temp->dups < 1)
2736 				temp->dups = 1;
2737 			temp->rxmit = temp->start;
2738 			temp->next = 0;
2739 			p->next = temp;
2740 			tp->rcv_lastsack = sack.end;
2741 			tp->snd_numholes++;
2742 		}
2743 	}
2744 done:
2745 #if defined(TCP_SACK) && defined(TCP_FACK)
2746 	/*
2747 	 * Update retran_data and snd_awnd.  Go through the list of
2748 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2749 	 */
2750 	tp->retran_data = 0;
2751 	cur = tp->snd_holes;
2752 	while (cur) {
2753 		tp->retran_data += cur->rxmit - cur->start;
2754 		cur = cur->next;
2755 	}
2756 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2757 	    tp->retran_data;
2758 #endif /* TCP_FACK */
2759 
2760 	return;
2761 }
2762 
2763 /*
2764  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2765  * it is completely acked; otherwise, tcp_sack_option(), called from
2766  * tcp_dooptions(), will fix up the hole.
2767  */
2768 void
2769 tcp_del_sackholes(struct tcpcb *tp, struct tcphdr *th)
2770 {
2771 	if (tp->sack_enable && tp->t_state != TCPS_LISTEN) {
2772 		/* max because this could be an older ack just arrived */
2773 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2774 			th->th_ack : tp->snd_una;
2775 		struct sackhole *cur = tp->snd_holes;
2776 		struct sackhole *prev;
2777 		while (cur)
2778 			if (SEQ_LEQ(cur->end, lastack)) {
2779 				prev = cur;
2780 				cur = cur->next;
2781 				pool_put(&sackhl_pool, prev);
2782 				tp->snd_numholes--;
2783 			} else if (SEQ_LT(cur->start, lastack)) {
2784 				cur->start = lastack;
2785 				if (SEQ_LT(cur->rxmit, cur->start))
2786 					cur->rxmit = cur->start;
2787 				break;
2788 			} else
2789 				break;
2790 		tp->snd_holes = cur;
2791 	}
2792 }
2793 
2794 /*
2795  * Delete all receiver-side SACK information.
2796  */
2797 void
2798 tcp_clean_sackreport(struct tcpcb *tp)
2799 {
2800 	int i;
2801 
2802 	tp->rcv_numsacks = 0;
2803 	for (i = 0; i < MAX_SACK_BLKS; i++)
2804 		tp->sackblks[i].start = tp->sackblks[i].end=0;
2805 
2806 }
2807 
2808 /*
2809  * Checks for partial ack.  If partial ack arrives, turn off retransmission
2810  * timer, deflate the window, do not clear tp->t_dupacks, and return 1.
2811  * If the ack advances at least to tp->snd_last, return 0.
2812  */
2813 int
2814 tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th)
2815 {
2816 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2817 		/* Turn off retx. timer (will start again next segment) */
2818 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2819 		tp->t_rtttime = 0;
2820 #ifndef TCP_FACK
2821 		/*
2822 		 * Partial window deflation.  This statement relies on the
2823 		 * fact that tp->snd_una has not been updated yet.  In FACK
2824 		 * hold snd_cwnd constant during fast recovery.
2825 		 */
2826 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2827 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2828 			tp->snd_cwnd += tp->t_maxseg;
2829 		} else
2830 			tp->snd_cwnd = tp->t_maxseg;
2831 #endif
2832 		return (1);
2833 	}
2834 	return (0);
2835 }
2836 #endif /* TCP_SACK */
2837 
2838 /*
2839  * Pull out of band byte out of a segment so
2840  * it doesn't appear in the user's data queue.
2841  * It is still reflected in the segment length for
2842  * sequencing purposes.
2843  */
2844 void
2845 tcp_pulloutofband(struct socket *so, u_int urgent, struct mbuf *m, int off)
2846 {
2847         int cnt = off + urgent - 1;
2848 
2849 	while (cnt >= 0) {
2850 		if (m->m_len > cnt) {
2851 			char *cp = mtod(m, caddr_t) + cnt;
2852 			struct tcpcb *tp = sototcpcb(so);
2853 
2854 			tp->t_iobc = *cp;
2855 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2856 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2857 			m->m_len--;
2858 			return;
2859 		}
2860 		cnt -= m->m_len;
2861 		m = m->m_next;
2862 		if (m == NULL)
2863 			break;
2864 	}
2865 	panic("tcp_pulloutofband");
2866 }
2867 
2868 /*
2869  * Collect new round-trip time estimate
2870  * and update averages and current timeout.
2871  */
2872 void
2873 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2874 {
2875 	short delta;
2876 	short rttmin;
2877 
2878 	if (rtt < 0)
2879 		rtt = 0;
2880 	else if (rtt > TCP_RTT_MAX)
2881 		rtt = TCP_RTT_MAX;
2882 
2883 	tcpstat.tcps_rttupdated++;
2884 	if (tp->t_srtt != 0) {
2885 		/*
2886 		 * delta is fixed point with 2 (TCP_RTT_BASE_SHIFT) bits
2887 		 * after the binary point (scaled by 4), whereas
2888 		 * srtt is stored as fixed point with 5 bits after the
2889 		 * binary point (i.e., scaled by 32).  The following magic
2890 		 * is equivalent to the smoothing algorithm in rfc793 with
2891 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2892 		 * point).
2893 		 */
2894 		delta = (rtt << TCP_RTT_BASE_SHIFT) -
2895 		    (tp->t_srtt >> TCP_RTT_SHIFT);
2896 		if ((tp->t_srtt += delta) <= 0)
2897 			tp->t_srtt = 1 << TCP_RTT_BASE_SHIFT;
2898 		/*
2899 		 * We accumulate a smoothed rtt variance (actually, a
2900 		 * smoothed mean difference), then set the retransmit
2901 		 * timer to smoothed rtt + 4 times the smoothed variance.
2902 		 * rttvar is stored as fixed point with 4 bits after the
2903 		 * binary point (scaled by 16).  The following is
2904 		 * equivalent to rfc793 smoothing with an alpha of .75
2905 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2906 		 * rfc793's wired-in beta.
2907 		 */
2908 		if (delta < 0)
2909 			delta = -delta;
2910 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2911 		if ((tp->t_rttvar += delta) <= 0)
2912 			tp->t_rttvar = 1 << TCP_RTT_BASE_SHIFT;
2913 	} else {
2914 		/*
2915 		 * No rtt measurement yet - use the unsmoothed rtt.
2916 		 * Set the variance to half the rtt (so our first
2917 		 * retransmit happens at 3*rtt).
2918 		 */
2919 		tp->t_srtt = (rtt + 1) << (TCP_RTT_SHIFT + TCP_RTT_BASE_SHIFT);
2920 		tp->t_rttvar = (rtt + 1) <<
2921 		    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
2922 	}
2923 	tp->t_rtttime = 0;
2924 	tp->t_rxtshift = 0;
2925 
2926 	/*
2927 	 * the retransmit should happen at rtt + 4 * rttvar.
2928 	 * Because of the way we do the smoothing, srtt and rttvar
2929 	 * will each average +1/2 tick of bias.  When we compute
2930 	 * the retransmit timer, we want 1/2 tick of rounding and
2931 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2932 	 * firing of the timer.  The bias will give us exactly the
2933 	 * 1.5 tick we need.  But, because the bias is
2934 	 * statistical, we have to test that we don't drop below
2935 	 * the minimum feasible timer (which is 2 ticks).
2936 	 */
2937 	rttmin = min(max(rtt + 2, tp->t_rttmin), TCPTV_REXMTMAX);
2938 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2939 
2940 	/*
2941 	 * We received an ack for a packet that wasn't retransmitted;
2942 	 * it is probably safe to discard any error indications we've
2943 	 * received recently.  This isn't quite right, but close enough
2944 	 * for now (a route might have failed after we sent a segment,
2945 	 * and the return path might not be symmetrical).
2946 	 */
2947 	tp->t_softerror = 0;
2948 }
2949 
2950 /*
2951  * Determine a reasonable value for maxseg size.
2952  * If the route is known, check route for mtu.
2953  * If none, use an mss that can be handled on the outgoing
2954  * interface without forcing IP to fragment; if bigger than
2955  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2956  * to utilize large mbufs.  If no route is found, route has no mtu,
2957  * or the destination isn't local, use a default, hopefully conservative
2958  * size (usually 512 or the default IP max size, but no more than the mtu
2959  * of the interface), as we can't discover anything about intervening
2960  * gateways or networks.  We also initialize the congestion/slow start
2961  * window to be a single segment if the destination isn't local.
2962  * While looking at the routing entry, we also initialize other path-dependent
2963  * parameters from pre-set or cached values in the routing entry.
2964  *
2965  * Also take into account the space needed for options that we
2966  * send regularly.  Make maxseg shorter by that amount to assure
2967  * that we can send maxseg amount of data even when the options
2968  * are present.  Store the upper limit of the length of options plus
2969  * data in maxopd.
2970  *
2971  * NOTE: offer == -1 indicates that the maxseg size changed due to
2972  * Path MTU discovery.
2973  */
2974 int
2975 tcp_mss(struct tcpcb *tp, int offer)
2976 {
2977 	struct rtentry *rt;
2978 	struct ifnet *ifp;
2979 	int mss, mssopt;
2980 	int iphlen;
2981 	struct inpcb *inp;
2982 
2983 	inp = tp->t_inpcb;
2984 
2985 	mssopt = mss = tcp_mssdflt;
2986 
2987 	rt = in_pcbrtentry(inp);
2988 
2989 	if (rt == NULL)
2990 		goto out;
2991 
2992 	ifp = rt->rt_ifp;
2993 
2994 	switch (tp->pf) {
2995 #ifdef INET6
2996 	case AF_INET6:
2997 		iphlen = sizeof(struct ip6_hdr);
2998 		break;
2999 #endif
3000 	case AF_INET:
3001 		iphlen = sizeof(struct ip);
3002 		break;
3003 	default:
3004 		/* the family does not support path MTU discovery */
3005 		goto out;
3006 	}
3007 
3008 	/*
3009 	 * if there's an mtu associated with the route and we support
3010 	 * path MTU discovery for the underlying protocol family, use it.
3011 	 */
3012 	if (rt->rt_rmx.rmx_mtu) {
3013 		/*
3014 		 * One may wish to lower MSS to take into account options,
3015 		 * especially security-related options.
3016 		 */
3017 		if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) {
3018 			/*
3019 			 * RFC2460 section 5, last paragraph: if path MTU is
3020 			 * smaller than 1280, use 1280 as packet size and
3021 			 * attach fragment header.
3022 			 */
3023 			mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) -
3024 			    sizeof(struct tcphdr);
3025 		} else {
3026 			mss = rt->rt_rmx.rmx_mtu - iphlen -
3027 			    sizeof(struct tcphdr);
3028 		}
3029 	} else if (!ifp) {
3030 		/*
3031 		 * ifp may be null and rmx_mtu may be zero in certain
3032 		 * v6 cases (e.g., if ND wasn't able to resolve the
3033 		 * destination host.
3034 		 */
3035 		goto out;
3036 	} else if (ifp->if_flags & IFF_LOOPBACK) {
3037 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3038 	} else if (tp->pf == AF_INET) {
3039 		if (ip_mtudisc)
3040 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3041 	}
3042 #ifdef INET6
3043 	else if (tp->pf == AF_INET6) {
3044 		/*
3045 		 * for IPv6, path MTU discovery is always turned on,
3046 		 * or the node must use packet size <= 1280.
3047 		 */
3048 		mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr);
3049 	}
3050 #endif /* INET6 */
3051 
3052 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3053 	if (offer != -1) {
3054 #ifndef INET6
3055 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3056 #else
3057 		if (tp->pf == AF_INET6)
3058 			mssopt = IN6_LINKMTU(ifp) - iphlen -
3059 			    sizeof(struct tcphdr);
3060 		else
3061 			mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3062 #endif
3063 
3064 		mssopt = max(tcp_mssdflt, mssopt);
3065 	}
3066 
3067  out:
3068 	/*
3069 	 * The current mss, t_maxseg, is initialized to the default value.
3070 	 * If we compute a smaller value, reduce the current mss.
3071 	 * If we compute a larger value, return it for use in sending
3072 	 * a max seg size option, but don't store it for use
3073 	 * unless we received an offer at least that large from peer.
3074 	 *
3075 	 * However, do not accept offers lower than the minimum of
3076 	 * the interface MTU and 216.
3077 	 */
3078 	if (offer > 0)
3079 		tp->t_peermss = offer;
3080 	if (tp->t_peermss)
3081 		mss = min(mss, max(tp->t_peermss, 216));
3082 
3083 	/* sanity - at least max opt. space */
3084 	mss = max(mss, 64);
3085 
3086 	/*
3087 	 * maxopd stores the maximum length of data AND options
3088 	 * in a segment; maxseg is the amount of data in a normal
3089 	 * segment.  We need to store this value (maxopd) apart
3090 	 * from maxseg, because now every segment carries options
3091 	 * and thus we normally have somewhat less data in segments.
3092 	 */
3093 	tp->t_maxopd = mss;
3094 
3095 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3096 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3097 		mss -= TCPOLEN_TSTAMP_APPA;
3098 #ifdef TCP_SIGNATURE
3099 	if (tp->t_flags & TF_SIGNATURE)
3100 		mss -= TCPOLEN_SIGLEN;
3101 #endif
3102 
3103 	if (offer == -1) {
3104 		/* mss changed due to Path MTU discovery */
3105 		tp->t_flags &= ~TF_PMTUD_PEND;
3106 		tp->t_pmtud_mtu_sent = 0;
3107 		tp->t_pmtud_mss_acked = 0;
3108 		if (mss < tp->t_maxseg) {
3109 			/*
3110 			 * Follow suggestion in RFC 2414 to reduce the
3111 			 * congestion window by the ratio of the old
3112 			 * segment size to the new segment size.
3113 			 */
3114 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3115 					     mss, mss);
3116 		}
3117 	} else if (tcp_do_rfc3390 == 2) {
3118 		/* increase initial window  */
3119 		tp->snd_cwnd = ulmin(10 * mss, ulmax(2 * mss, 14600));
3120 	} else if (tcp_do_rfc3390) {
3121 		/* increase initial window  */
3122 		tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380));
3123 	} else
3124 		tp->snd_cwnd = mss;
3125 
3126 	tp->t_maxseg = mss;
3127 
3128 	return (offer != -1 ? mssopt : mss);
3129 }
3130 
3131 u_int
3132 tcp_hdrsz(struct tcpcb *tp)
3133 {
3134 	u_int hlen;
3135 
3136 	switch (tp->pf) {
3137 #ifdef INET6
3138 	case AF_INET6:
3139 		hlen = sizeof(struct ip6_hdr);
3140 		break;
3141 #endif
3142 	case AF_INET:
3143 		hlen = sizeof(struct ip);
3144 		break;
3145 	default:
3146 		hlen = 0;
3147 		break;
3148 	}
3149 	hlen += sizeof(struct tcphdr);
3150 
3151 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3152 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3153 		hlen += TCPOLEN_TSTAMP_APPA;
3154 #ifdef TCP_SIGNATURE
3155 	if (tp->t_flags & TF_SIGNATURE)
3156 		hlen += TCPOLEN_SIGLEN;
3157 #endif
3158 	return (hlen);
3159 }
3160 
3161 /*
3162  * Set connection variables based on the effective MSS.
3163  * We are passed the TCPCB for the actual connection.  If we
3164  * are the server, we are called by the compressed state engine
3165  * when the 3-way handshake is complete.  If we are the client,
3166  * we are called when we receive the SYN,ACK from the server.
3167  *
3168  * NOTE: The t_maxseg value must be initialized in the TCPCB
3169  * before this routine is called!
3170  */
3171 void
3172 tcp_mss_update(struct tcpcb *tp)
3173 {
3174 	int mss;
3175 	u_long bufsize;
3176 	struct rtentry *rt;
3177 	struct socket *so;
3178 
3179 	so = tp->t_inpcb->inp_socket;
3180 	mss = tp->t_maxseg;
3181 
3182 	rt = in_pcbrtentry(tp->t_inpcb);
3183 
3184 	if (rt == NULL)
3185 		return;
3186 
3187 	bufsize = so->so_snd.sb_hiwat;
3188 	if (bufsize < mss) {
3189 		mss = bufsize;
3190 		/* Update t_maxseg and t_maxopd */
3191 		tcp_mss(tp, mss);
3192 	} else {
3193 		bufsize = roundup(bufsize, mss);
3194 		if (bufsize > sb_max)
3195 			bufsize = sb_max;
3196 		(void)sbreserve(&so->so_snd, bufsize);
3197 	}
3198 
3199 	bufsize = so->so_rcv.sb_hiwat;
3200 	if (bufsize > mss) {
3201 		bufsize = roundup(bufsize, mss);
3202 		if (bufsize > sb_max)
3203 			bufsize = sb_max;
3204 		(void)sbreserve(&so->so_rcv, bufsize);
3205 	}
3206 
3207 }
3208 
3209 #if defined (TCP_SACK)
3210 /*
3211  * Checks for partial ack.  If partial ack arrives, force the retransmission
3212  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3213  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3214  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3215  */
3216 int
3217 tcp_newreno(struct tcpcb *tp, struct tcphdr *th)
3218 {
3219 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3220 		/*
3221 		 * snd_una has not been updated and the socket send buffer
3222 		 * not yet drained of the acked data, so we have to leave
3223 		 * snd_una as it was to get the correct data offset in
3224 		 * tcp_output().
3225 		 */
3226 		tcp_seq onxt = tp->snd_nxt;
3227 		u_long  ocwnd = tp->snd_cwnd;
3228 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3229 		tp->t_rtttime = 0;
3230 		tp->snd_nxt = th->th_ack;
3231 		/*
3232 		 * Set snd_cwnd to one segment beyond acknowledged offset
3233 		 * (tp->snd_una not yet updated when this function is called)
3234 		 */
3235 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3236 		(void) tcp_output(tp);
3237 		tp->snd_cwnd = ocwnd;
3238 		if (SEQ_GT(onxt, tp->snd_nxt))
3239 			tp->snd_nxt = onxt;
3240 		/*
3241 		 * Partial window deflation.  Relies on fact that tp->snd_una
3242 		 * not updated yet.
3243 		 */
3244 		if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3245 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
3246 		else
3247 			tp->snd_cwnd = 0;
3248 		tp->snd_cwnd += tp->t_maxseg;
3249 
3250 		return 1;
3251 	}
3252 	return 0;
3253 }
3254 #endif /* TCP_SACK */
3255 
3256 int
3257 tcp_mss_adv(struct mbuf *m, int af)
3258 {
3259 	int mss = 0;
3260 	int iphlen;
3261 	struct ifnet *ifp = NULL;
3262 
3263 	if (m && (m->m_flags & M_PKTHDR))
3264 		ifp = if_get(m->m_pkthdr.ph_ifidx);
3265 
3266 	switch (af) {
3267 	case AF_INET:
3268 		if (ifp != NULL)
3269 			mss = ifp->if_mtu;
3270 		iphlen = sizeof(struct ip);
3271 		break;
3272 #ifdef INET6
3273 	case AF_INET6:
3274 		if (ifp != NULL)
3275 			mss = IN6_LINKMTU(ifp);
3276 		iphlen = sizeof(struct ip6_hdr);
3277 		break;
3278 #endif
3279 	default:
3280 		unhandled_af(af);
3281 	}
3282 	if_put(ifp);
3283 	mss = mss - iphlen - sizeof(struct tcphdr);
3284 	return (max(mss, tcp_mssdflt));
3285 }
3286 
3287 /*
3288  * TCP compressed state engine.  Currently used to hold compressed
3289  * state for SYN_RECEIVED.
3290  */
3291 
3292 /* syn hash parameters */
3293 #define	TCP_SYN_HASH_SIZE	293
3294 #define	TCP_SYN_BUCKET_SIZE	35
3295 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
3296 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
3297 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
3298 int	tcp_syn_cache_count;
3299 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
3300 u_int32_t tcp_syn_hash[5];
3301 
3302 #define SYN_HASH(sa, sp, dp) \
3303 	(((sa)->s_addr ^ tcp_syn_hash[0]) *				\
3304 	(((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp))) ^ tcp_syn_hash[4]))
3305 #ifndef INET6
3306 #define	SYN_HASHALL(hash, src, dst) \
3307 do {									\
3308 	hash = SYN_HASH(&satosin(src)->sin_addr,			\
3309 		satosin(src)->sin_port,					\
3310 		satosin(dst)->sin_port);				\
3311 } while (/*CONSTCOND*/ 0)
3312 #else
3313 #define SYN_HASH6(sa, sp, dp) \
3314 	(((sa)->s6_addr32[0] ^ tcp_syn_hash[0]) *			\
3315 	((sa)->s6_addr32[1] ^ tcp_syn_hash[1]) *			\
3316 	((sa)->s6_addr32[2] ^ tcp_syn_hash[2]) *			\
3317 	((sa)->s6_addr32[3] ^ tcp_syn_hash[3]) *			\
3318 	(((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp))) ^ tcp_syn_hash[4]))
3319 
3320 #define SYN_HASHALL(hash, src, dst) \
3321 do {									\
3322 	switch ((src)->sa_family) {					\
3323 	case AF_INET:							\
3324 		hash = SYN_HASH(&satosin(src)->sin_addr,		\
3325 			satosin(src)->sin_port,				\
3326 			satosin(dst)->sin_port);			\
3327 		break;							\
3328 	case AF_INET6:							\
3329 		hash = SYN_HASH6(&satosin6(src)->sin6_addr,		\
3330 			satosin6(src)->sin6_port,			\
3331 			satosin6(dst)->sin6_port);			\
3332 		break;							\
3333 	default:							\
3334 		hash = 0;						\
3335 	}								\
3336 } while (/*CONSTCOND*/0)
3337 #endif /* INET6 */
3338 
3339 void
3340 syn_cache_rm(struct syn_cache *sc)
3341 {
3342 	sc->sc_flags |= SCF_DEAD;
3343 	TAILQ_REMOVE(&tcp_syn_cache[sc->sc_bucketidx].sch_bucket,
3344 	    sc, sc_bucketq);
3345 	sc->sc_tp = NULL;
3346 	LIST_REMOVE(sc, sc_tpq);
3347 	tcp_syn_cache[sc->sc_bucketidx].sch_length--;
3348 	timeout_del(&sc->sc_timer);
3349 	tcp_syn_cache_count--;
3350 }
3351 
3352 void
3353 syn_cache_put(struct syn_cache *sc)
3354 {
3355 	if (sc->sc_ipopts)
3356 		(void) m_free(sc->sc_ipopts);
3357 	if (sc->sc_route4.ro_rt != NULL) {
3358 		rtfree(sc->sc_route4.ro_rt);
3359 		sc->sc_route4.ro_rt = NULL;
3360 	}
3361 	timeout_set(&sc->sc_timer, syn_cache_reaper, sc);
3362 	timeout_add(&sc->sc_timer, 0);
3363 }
3364 
3365 struct pool syn_cache_pool;
3366 
3367 /*
3368  * We don't estimate RTT with SYNs, so each packet starts with the default
3369  * RTT and each timer step has a fixed timeout value.
3370  */
3371 #define	SYN_CACHE_TIMER_ARM(sc)						\
3372 do {									\
3373 	TCPT_RANGESET((sc)->sc_rxtcur,					\
3374 	    TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN,	\
3375 	    TCPTV_REXMTMAX);						\
3376 	if (!timeout_initialized(&(sc)->sc_timer))			\
3377 		timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc));	\
3378 	timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \
3379 } while (/*CONSTCOND*/0)
3380 
3381 #define	SYN_CACHE_TIMESTAMP(sc)	tcp_now + (sc)->sc_modulate
3382 
3383 void
3384 syn_cache_init()
3385 {
3386 	int i;
3387 
3388 	/* Initialize the hash buckets. */
3389 	for (i = 0; i < tcp_syn_cache_size; i++)
3390 		TAILQ_INIT(&tcp_syn_cache[i].sch_bucket);
3391 
3392 	/* Initialize the syn cache pool. */
3393 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
3394 	    "syncache", NULL);
3395 	pool_setipl(&syn_cache_pool, IPL_SOFTNET);
3396 }
3397 
3398 void
3399 syn_cache_insert(struct syn_cache *sc, struct tcpcb *tp)
3400 {
3401 	struct syn_cache_head *scp;
3402 	struct syn_cache *sc2;
3403 	int s;
3404 
3405 	/*
3406 	 * If there are no entries in the hash table, reinitialize
3407 	 * the hash secrets.
3408 	 */
3409 	if (tcp_syn_cache_count == 0)
3410 		arc4random_buf(tcp_syn_hash, sizeof(tcp_syn_hash));
3411 
3412 	SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa);
3413 	sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size;
3414 	scp = &tcp_syn_cache[sc->sc_bucketidx];
3415 
3416 	/*
3417 	 * Make sure that we don't overflow the per-bucket
3418 	 * limit or the total cache size limit.
3419 	 */
3420 	s = splsoftnet();
3421 	if (scp->sch_length >= tcp_syn_bucket_limit) {
3422 		tcpstat.tcps_sc_bucketoverflow++;
3423 		/*
3424 		 * The bucket is full.  Toss the oldest element in the
3425 		 * bucket.  This will be the first entry in the bucket.
3426 		 */
3427 		sc2 = TAILQ_FIRST(&scp->sch_bucket);
3428 #ifdef DIAGNOSTIC
3429 		/*
3430 		 * This should never happen; we should always find an
3431 		 * entry in our bucket.
3432 		 */
3433 		if (sc2 == NULL)
3434 			panic("syn_cache_insert: bucketoverflow: impossible");
3435 #endif
3436 		syn_cache_rm(sc2);
3437 		syn_cache_put(sc2);
3438 	} else if (tcp_syn_cache_count >= tcp_syn_cache_limit) {
3439 		struct syn_cache_head *scp2, *sce;
3440 
3441 		tcpstat.tcps_sc_overflowed++;
3442 		/*
3443 		 * The cache is full.  Toss the oldest entry in the
3444 		 * first non-empty bucket we can find.
3445 		 *
3446 		 * XXX We would really like to toss the oldest
3447 		 * entry in the cache, but we hope that this
3448 		 * condition doesn't happen very often.
3449 		 */
3450 		scp2 = scp;
3451 		if (TAILQ_EMPTY(&scp2->sch_bucket)) {
3452 			sce = &tcp_syn_cache[tcp_syn_cache_size];
3453 			for (++scp2; scp2 != scp; scp2++) {
3454 				if (scp2 >= sce)
3455 					scp2 = &tcp_syn_cache[0];
3456 				if (! TAILQ_EMPTY(&scp2->sch_bucket))
3457 					break;
3458 			}
3459 #ifdef DIAGNOSTIC
3460 			/*
3461 			 * This should never happen; we should always find a
3462 			 * non-empty bucket.
3463 			 */
3464 			if (scp2 == scp)
3465 				panic("syn_cache_insert: cacheoverflow: "
3466 				    "impossible");
3467 #endif
3468 		}
3469 		sc2 = TAILQ_FIRST(&scp2->sch_bucket);
3470 		syn_cache_rm(sc2);
3471 		syn_cache_put(sc2);
3472 	}
3473 
3474 	/*
3475 	 * Initialize the entry's timer.
3476 	 */
3477 	sc->sc_rxttot = 0;
3478 	sc->sc_rxtshift = 0;
3479 	SYN_CACHE_TIMER_ARM(sc);
3480 
3481 	/* Link it from tcpcb entry */
3482 	LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq);
3483 
3484 	/* Put it into the bucket. */
3485 	TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq);
3486 	scp->sch_length++;
3487 	tcp_syn_cache_count++;
3488 
3489 	tcpstat.tcps_sc_added++;
3490 	splx(s);
3491 }
3492 
3493 /*
3494  * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted.
3495  * If we have retransmitted an entry the maximum number of times, expire
3496  * that entry.
3497  */
3498 void
3499 syn_cache_timer(void *arg)
3500 {
3501 	struct syn_cache *sc = arg;
3502 	int s;
3503 
3504 	s = splsoftnet();
3505 	if (sc->sc_flags & SCF_DEAD) {
3506 		splx(s);
3507 		return;
3508 	}
3509 
3510 	if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) {
3511 		/* Drop it -- too many retransmissions. */
3512 		goto dropit;
3513 	}
3514 
3515 	/*
3516 	 * Compute the total amount of time this entry has
3517 	 * been on a queue.  If this entry has been on longer
3518 	 * than the keep alive timer would allow, expire it.
3519 	 */
3520 	sc->sc_rxttot += sc->sc_rxtcur;
3521 	if (sc->sc_rxttot >= tcptv_keep_init)
3522 		goto dropit;
3523 
3524 	tcpstat.tcps_sc_retransmitted++;
3525 	(void) syn_cache_respond(sc, NULL);
3526 
3527 	/* Advance the timer back-off. */
3528 	sc->sc_rxtshift++;
3529 	SYN_CACHE_TIMER_ARM(sc);
3530 
3531 	splx(s);
3532 	return;
3533 
3534  dropit:
3535 	tcpstat.tcps_sc_timed_out++;
3536 	syn_cache_rm(sc);
3537 	syn_cache_put(sc);
3538 	splx(s);
3539 }
3540 
3541 void
3542 syn_cache_reaper(void *arg)
3543 {
3544 	struct syn_cache *sc = arg;
3545 
3546 	pool_put(&syn_cache_pool, (sc));
3547 	return;
3548 }
3549 
3550 /*
3551  * Remove syn cache created by the specified tcb entry,
3552  * because this does not make sense to keep them
3553  * (if there's no tcb entry, syn cache entry will never be used)
3554  */
3555 void
3556 syn_cache_cleanup(struct tcpcb *tp)
3557 {
3558 	struct syn_cache *sc, *nsc;
3559 	int s;
3560 
3561 	s = splsoftnet();
3562 
3563 	LIST_FOREACH_SAFE(sc, &tp->t_sc, sc_tpq, nsc) {
3564 #ifdef DIAGNOSTIC
3565 		if (sc->sc_tp != tp)
3566 			panic("invalid sc_tp in syn_cache_cleanup");
3567 #endif
3568 		syn_cache_rm(sc);
3569 		syn_cache_put(sc);
3570 	}
3571 	/* just for safety */
3572 	LIST_INIT(&tp->t_sc);
3573 
3574 	splx(s);
3575 }
3576 
3577 /*
3578  * Find an entry in the syn cache.
3579  */
3580 struct syn_cache *
3581 syn_cache_lookup(struct sockaddr *src, struct sockaddr *dst,
3582     struct syn_cache_head **headp, u_int rtableid)
3583 {
3584 	struct syn_cache *sc;
3585 	struct syn_cache_head *scp;
3586 	u_int32_t hash;
3587 
3588 	splsoftassert(IPL_SOFTNET);
3589 
3590 	if (tcp_syn_cache_count == 0)
3591 		return (NULL);
3592 
3593 	SYN_HASHALL(hash, src, dst);
3594 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
3595 	*headp = scp;
3596 	TAILQ_FOREACH(sc, &scp->sch_bucket, sc_bucketq) {
3597 		if (sc->sc_hash != hash)
3598 			continue;
3599 		if (!bcmp(&sc->sc_src, src, src->sa_len) &&
3600 		    !bcmp(&sc->sc_dst, dst, dst->sa_len) &&
3601 		    rtable_l2(rtableid) == rtable_l2(sc->sc_rtableid))
3602 			return (sc);
3603 	}
3604 	return (NULL);
3605 }
3606 
3607 /*
3608  * This function gets called when we receive an ACK for a
3609  * socket in the LISTEN state.  We look up the connection
3610  * in the syn cache, and if its there, we pull it out of
3611  * the cache and turn it into a full-blown connection in
3612  * the SYN-RECEIVED state.
3613  *
3614  * The return values may not be immediately obvious, and their effects
3615  * can be subtle, so here they are:
3616  *
3617  *	NULL	SYN was not found in cache; caller should drop the
3618  *		packet and send an RST.
3619  *
3620  *	-1	We were unable to create the new connection, and are
3621  *		aborting it.  An ACK,RST is being sent to the peer
3622  *		(unless we got screwey sequence numbners; see below),
3623  *		because the 3-way handshake has been completed.  Caller
3624  *		should not free the mbuf, since we may be using it.  If
3625  *		we are not, we will free it.
3626  *
3627  *	Otherwise, the return value is a pointer to the new socket
3628  *	associated with the connection.
3629  */
3630 struct socket *
3631 syn_cache_get(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3632     u_int hlen, u_int tlen, struct socket *so, struct mbuf *m)
3633 {
3634 	struct syn_cache *sc;
3635 	struct syn_cache_head *scp;
3636 	struct inpcb *inp = NULL;
3637 	struct tcpcb *tp = NULL;
3638 	struct mbuf *am;
3639 	int s;
3640 	struct socket *oso;
3641 #if NPF > 0
3642 	struct pf_divert *divert = NULL;
3643 #endif
3644 
3645 	s = splsoftnet();
3646 	if ((sc = syn_cache_lookup(src, dst, &scp,
3647 	    sotoinpcb(so)->inp_rtableid)) == NULL) {
3648 		splx(s);
3649 		return (NULL);
3650 	}
3651 
3652 	/*
3653 	 * Verify the sequence and ack numbers.  Try getting the correct
3654 	 * response again.
3655 	 */
3656 	if ((th->th_ack != sc->sc_iss + 1) ||
3657 	    SEQ_LEQ(th->th_seq, sc->sc_irs) ||
3658 	    SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) {
3659 		(void) syn_cache_respond(sc, m);
3660 		splx(s);
3661 		return ((struct socket *)(-1));
3662 	}
3663 
3664 	/* Remove this cache entry */
3665 	syn_cache_rm(sc);
3666 	splx(s);
3667 
3668 	/*
3669 	 * Ok, create the full blown connection, and set things up
3670 	 * as they would have been set up if we had created the
3671 	 * connection when the SYN arrived.  If we can't create
3672 	 * the connection, abort it.
3673 	 */
3674 	oso = so;
3675 	so = sonewconn(so, SS_ISCONNECTED);
3676 	if (so == NULL)
3677 		goto resetandabort;
3678 
3679 	inp = sotoinpcb(oso);
3680 
3681 #ifdef IPSEC
3682 	/*
3683 	 * We need to copy the required security levels
3684 	 * from the old pcb. Ditto for any other
3685 	 * IPsec-related information.
3686 	 */
3687 	{
3688 	  struct inpcb *newinp = sotoinpcb(so);
3689 	  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
3690 		sizeof(inp->inp_seclevel));
3691 	}
3692 #endif /* IPSEC */
3693 #ifdef INET6
3694 	/*
3695 	 * inp still has the OLD in_pcb stuff, set the
3696 	 * v6-related flags on the new guy, too.
3697 	 */
3698 	{
3699 	  int flags = inp->inp_flags;
3700 	  struct inpcb *oldinpcb = inp;
3701 
3702 	  inp = sotoinpcb(so);
3703 	  inp->inp_flags |= (flags & INP_IPV6);
3704 	  if ((inp->inp_flags & INP_IPV6) != 0) {
3705 	    inp->inp_ipv6.ip6_hlim =
3706 	      oldinpcb->inp_ipv6.ip6_hlim;
3707 	  }
3708 	}
3709 #else /* INET6 */
3710 	inp = sotoinpcb(so);
3711 #endif /* INET6 */
3712 
3713 #if NPF > 0
3714 	if (m && m->m_pkthdr.pf.flags & PF_TAG_DIVERTED &&
3715 	    (divert = pf_find_divert(m)) != NULL)
3716 		inp->inp_rtableid = divert->rdomain;
3717 	else
3718 #endif
3719 	/* inherit rtable from listening socket */
3720 	inp->inp_rtableid = sc->sc_rtableid;
3721 
3722 	inp->inp_lport = th->th_dport;
3723 	switch (src->sa_family) {
3724 #ifdef INET6
3725 	case AF_INET6:
3726 		inp->inp_laddr6 = satosin6(dst)->sin6_addr;
3727 		break;
3728 #endif /* INET6 */
3729 	case AF_INET:
3730 		inp->inp_laddr = satosin(dst)->sin_addr;
3731 		inp->inp_options = ip_srcroute(m);
3732 		if (inp->inp_options == NULL) {
3733 			inp->inp_options = sc->sc_ipopts;
3734 			sc->sc_ipopts = NULL;
3735 		}
3736 		break;
3737 	}
3738 	in_pcbrehash(inp);
3739 
3740 	/*
3741 	 * Give the new socket our cached route reference.
3742 	 */
3743 	if (src->sa_family == AF_INET)
3744 		inp->inp_route = sc->sc_route4;         /* struct assignment */
3745 #ifdef INET6
3746 	else
3747 		inp->inp_route6 = sc->sc_route6;
3748 #endif
3749 	sc->sc_route4.ro_rt = NULL;
3750 
3751 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
3752 	if (am == NULL)
3753 		goto resetandabort;
3754 	am->m_len = src->sa_len;
3755 	bcopy(src, mtod(am, caddr_t), src->sa_len);
3756 
3757 	switch (src->sa_family) {
3758 	case AF_INET:
3759 		/* drop IPv4 packet to AF_INET6 socket */
3760 		if (inp->inp_flags & INP_IPV6) {
3761 			(void) m_free(am);
3762 			goto resetandabort;
3763 		}
3764 		if (in_pcbconnect(inp, am)) {
3765 			(void) m_free(am);
3766 			goto resetandabort;
3767 		}
3768 		break;
3769 #ifdef INET6
3770 	case AF_INET6:
3771 		if (in6_pcbconnect(inp, am)) {
3772 			(void) m_free(am);
3773 			goto resetandabort;
3774 		}
3775 		break;
3776 #endif
3777 	}
3778 	(void) m_free(am);
3779 
3780 	tp = intotcpcb(inp);
3781 	tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY;
3782 	if (sc->sc_request_r_scale != 15) {
3783 		tp->requested_s_scale = sc->sc_requested_s_scale;
3784 		tp->request_r_scale = sc->sc_request_r_scale;
3785 		tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE;
3786 	}
3787 	if (sc->sc_flags & SCF_TIMESTAMP)
3788 		tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP;
3789 
3790 	tp->t_template = tcp_template(tp);
3791 	if (tp->t_template == 0) {
3792 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
3793 		so = NULL;
3794 		m_freem(m);
3795 		goto abort;
3796 	}
3797 #ifdef TCP_SACK
3798 	tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT;
3799 #endif
3800 
3801 	tp->ts_modulate = sc->sc_modulate;
3802 	tp->ts_recent = sc->sc_timestamp;
3803 	tp->iss = sc->sc_iss;
3804 	tp->irs = sc->sc_irs;
3805 	tcp_sendseqinit(tp);
3806 #if defined (TCP_SACK) || defined(TCP_ECN)
3807 	tp->snd_last = tp->snd_una;
3808 #endif /* TCP_SACK */
3809 #if defined(TCP_SACK) && defined(TCP_FACK)
3810 	tp->snd_fack = tp->snd_una;
3811 	tp->retran_data = 0;
3812 	tp->snd_awnd = 0;
3813 #endif /* TCP_FACK */
3814 #ifdef TCP_ECN
3815 	if (sc->sc_flags & SCF_ECN_PERMIT) {
3816 		tp->t_flags |= TF_ECN_PERMIT;
3817 		tcpstat.tcps_ecn_accepts++;
3818 	}
3819 #endif
3820 #ifdef TCP_SACK
3821 	if (sc->sc_flags & SCF_SACK_PERMIT)
3822 		tp->t_flags |= TF_SACK_PERMIT;
3823 #endif
3824 #ifdef TCP_SIGNATURE
3825 	if (sc->sc_flags & SCF_SIGNATURE)
3826 		tp->t_flags |= TF_SIGNATURE;
3827 #endif
3828 	tcp_rcvseqinit(tp);
3829 	tp->t_state = TCPS_SYN_RECEIVED;
3830 	tp->t_rcvtime = tcp_now;
3831 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
3832 	tcpstat.tcps_accepts++;
3833 
3834 	tcp_mss(tp, sc->sc_peermaxseg);	 /* sets t_maxseg */
3835 	if (sc->sc_peermaxseg)
3836 		tcp_mss_update(tp);
3837 	/* Reset initial window to 1 segment for retransmit */
3838 	if (sc->sc_rxtshift > 0)
3839 		tp->snd_cwnd = tp->t_maxseg;
3840 	tp->snd_wl1 = sc->sc_irs;
3841 	tp->rcv_up = sc->sc_irs + 1;
3842 
3843 	/*
3844 	 * This is what whould have happened in tcp_output() when
3845 	 * the SYN,ACK was sent.
3846 	 */
3847 	tp->snd_up = tp->snd_una;
3848 	tp->snd_max = tp->snd_nxt = tp->iss+1;
3849 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
3850 	if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv))
3851 		tp->rcv_adv = tp->rcv_nxt + sc->sc_win;
3852 	tp->last_ack_sent = tp->rcv_nxt;
3853 
3854 	tcpstat.tcps_sc_completed++;
3855 	syn_cache_put(sc);
3856 	return (so);
3857 
3858 resetandabort:
3859 	tcp_respond(NULL, mtod(m, caddr_t), th, (tcp_seq)0, th->th_ack, TH_RST,
3860 	    m->m_pkthdr.ph_rtableid);
3861 	m_freem(m);
3862 abort:
3863 	if (so != NULL)
3864 		(void) soabort(so);
3865 	syn_cache_put(sc);
3866 	tcpstat.tcps_sc_aborted++;
3867 	return ((struct socket *)(-1));
3868 }
3869 
3870 /*
3871  * This function is called when we get a RST for a
3872  * non-existent connection, so that we can see if the
3873  * connection is in the syn cache.  If it is, zap it.
3874  */
3875 
3876 void
3877 syn_cache_reset(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3878     u_int rtableid)
3879 {
3880 	struct syn_cache *sc;
3881 	struct syn_cache_head *scp;
3882 	int s = splsoftnet();
3883 
3884 	if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) {
3885 		splx(s);
3886 		return;
3887 	}
3888 	if (SEQ_LT(th->th_seq, sc->sc_irs) ||
3889 	    SEQ_GT(th->th_seq, sc->sc_irs+1)) {
3890 		splx(s);
3891 		return;
3892 	}
3893 	syn_cache_rm(sc);
3894 	splx(s);
3895 	tcpstat.tcps_sc_reset++;
3896 	syn_cache_put(sc);
3897 }
3898 
3899 void
3900 syn_cache_unreach(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3901     u_int rtableid)
3902 {
3903 	struct syn_cache *sc;
3904 	struct syn_cache_head *scp;
3905 	int s;
3906 
3907 	s = splsoftnet();
3908 	if ((sc = syn_cache_lookup(src, dst, &scp, rtableid)) == NULL) {
3909 		splx(s);
3910 		return;
3911 	}
3912 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
3913 	if (ntohl (th->th_seq) != sc->sc_iss) {
3914 		splx(s);
3915 		return;
3916 	}
3917 
3918 	/*
3919 	 * If we've retransmitted 3 times and this is our second error,
3920 	 * we remove the entry.  Otherwise, we allow it to continue on.
3921 	 * This prevents us from incorrectly nuking an entry during a
3922 	 * spurious network outage.
3923 	 *
3924 	 * See tcp_notify().
3925 	 */
3926 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) {
3927 		sc->sc_flags |= SCF_UNREACH;
3928 		splx(s);
3929 		return;
3930 	}
3931 
3932 	syn_cache_rm(sc);
3933 	splx(s);
3934 	tcpstat.tcps_sc_unreach++;
3935 	syn_cache_put(sc);
3936 }
3937 
3938 /*
3939  * Given a LISTEN socket and an inbound SYN request, add
3940  * this to the syn cache, and send back a segment:
3941  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
3942  * to the source.
3943  *
3944  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
3945  * Doing so would require that we hold onto the data and deliver it
3946  * to the application.  However, if we are the target of a SYN-flood
3947  * DoS attack, an attacker could send data which would eventually
3948  * consume all available buffer space if it were ACKed.  By not ACKing
3949  * the data, we avoid this DoS scenario.
3950  */
3951 
3952 int
3953 syn_cache_add(struct sockaddr *src, struct sockaddr *dst, struct tcphdr *th,
3954     u_int iphlen, struct socket *so, struct mbuf *m, u_char *optp, int optlen,
3955     struct tcp_opt_info *oi, tcp_seq *issp)
3956 {
3957 	struct tcpcb tb, *tp;
3958 	long win;
3959 	struct syn_cache *sc;
3960 	struct syn_cache_head *scp;
3961 	struct mbuf *ipopts;
3962 
3963 	tp = sototcpcb(so);
3964 
3965 	/*
3966 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
3967 	 *
3968 	 * Note this check is performed in tcp_input() very early on.
3969 	 */
3970 
3971 	/*
3972 	 * Initialize some local state.
3973 	 */
3974 	win = sbspace(&so->so_rcv);
3975 	if (win > TCP_MAXWIN)
3976 		win = TCP_MAXWIN;
3977 
3978 	bzero(&tb, sizeof(tb));
3979 #ifdef TCP_SIGNATURE
3980 	if (optp || (tp->t_flags & TF_SIGNATURE)) {
3981 #else
3982 	if (optp) {
3983 #endif
3984 		tb.pf = tp->pf;
3985 #ifdef TCP_SACK
3986 		tb.sack_enable = tp->sack_enable;
3987 #endif
3988 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
3989 #ifdef TCP_SIGNATURE
3990 		if (tp->t_flags & TF_SIGNATURE)
3991 			tb.t_flags |= TF_SIGNATURE;
3992 #endif
3993 		tb.t_state = TCPS_LISTEN;
3994 		if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi,
3995 		    sotoinpcb(so)->inp_rtableid))
3996 			return (-1);
3997 	}
3998 
3999 	switch (src->sa_family) {
4000 	case AF_INET:
4001 		/*
4002 		 * Remember the IP options, if any.
4003 		 */
4004 		ipopts = ip_srcroute(m);
4005 		break;
4006 	default:
4007 		ipopts = NULL;
4008 	}
4009 
4010 	/*
4011 	 * See if we already have an entry for this connection.
4012 	 * If we do, resend the SYN,ACK.  We do not count this
4013 	 * as a retransmission (XXX though maybe we should).
4014 	 */
4015 	if ((sc = syn_cache_lookup(src, dst, &scp, sotoinpcb(so)->inp_rtableid))
4016 	    != NULL) {
4017 		tcpstat.tcps_sc_dupesyn++;
4018 		if (ipopts) {
4019 			/*
4020 			 * If we were remembering a previous source route,
4021 			 * forget it and use the new one we've been given.
4022 			 */
4023 			if (sc->sc_ipopts)
4024 				(void) m_free(sc->sc_ipopts);
4025 			sc->sc_ipopts = ipopts;
4026 		}
4027 		sc->sc_timestamp = tb.ts_recent;
4028 		if (syn_cache_respond(sc, m) == 0) {
4029 			tcpstat.tcps_sndacks++;
4030 			tcpstat.tcps_sndtotal++;
4031 		}
4032 		return (0);
4033 	}
4034 
4035 	sc = pool_get(&syn_cache_pool, PR_NOWAIT|PR_ZERO);
4036 	if (sc == NULL) {
4037 		if (ipopts)
4038 			(void) m_free(ipopts);
4039 		return (-1);
4040 	}
4041 
4042 	/*
4043 	 * Fill in the cache, and put the necessary IP and TCP
4044 	 * options into the reply.
4045 	 */
4046 	bcopy(src, &sc->sc_src, src->sa_len);
4047 	bcopy(dst, &sc->sc_dst, dst->sa_len);
4048 	sc->sc_rtableid = sotoinpcb(so)->inp_rtableid;
4049 	sc->sc_flags = 0;
4050 	sc->sc_ipopts = ipopts;
4051 	sc->sc_irs = th->th_seq;
4052 
4053 	sc->sc_iss = issp ? *issp : arc4random();
4054 	sc->sc_peermaxseg = oi->maxseg;
4055 	sc->sc_ourmaxseg = tcp_mss_adv(m, sc->sc_src.sa.sa_family);
4056 	sc->sc_win = win;
4057 	sc->sc_timestamp = tb.ts_recent;
4058 	if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) ==
4059 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
4060 		sc->sc_flags |= SCF_TIMESTAMP;
4061 		sc->sc_modulate = arc4random();
4062 	}
4063 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
4064 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
4065 		sc->sc_requested_s_scale = tb.requested_s_scale;
4066 		sc->sc_request_r_scale = 0;
4067 		/*
4068 		 * Pick the smallest possible scaling factor that
4069 		 * will still allow us to scale up to sb_max.
4070 		 *
4071 		 * We do this because there are broken firewalls that
4072 		 * will corrupt the window scale option, leading to
4073 		 * the other endpoint believing that our advertised
4074 		 * window is unscaled.  At scale factors larger than
4075 		 * 5 the unscaled window will drop below 1500 bytes,
4076 		 * leading to serious problems when traversing these
4077 		 * broken firewalls.
4078 		 *
4079 		 * With the default sbmax of 256K, a scale factor
4080 		 * of 3 will be chosen by this algorithm.  Those who
4081 		 * choose a larger sbmax should watch out
4082 		 * for the compatiblity problems mentioned above.
4083 		 *
4084 		 * RFC1323: The Window field in a SYN (i.e., a <SYN>
4085 		 * or <SYN,ACK>) segment itself is never scaled.
4086 		 */
4087 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
4088 		    (TCP_MAXWIN << sc->sc_request_r_scale) < sb_max)
4089 			sc->sc_request_r_scale++;
4090 	} else {
4091 		sc->sc_requested_s_scale = 15;
4092 		sc->sc_request_r_scale = 15;
4093 	}
4094 #ifdef TCP_ECN
4095 	/*
4096 	 * if both ECE and CWR flag bits are set, peer is ECN capable.
4097 	 */
4098 	if (tcp_do_ecn &&
4099 	    (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR))
4100 		sc->sc_flags |= SCF_ECN_PERMIT;
4101 #endif
4102 #ifdef TCP_SACK
4103 	/*
4104 	 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option
4105 	 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT).
4106 	 */
4107 	if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT))
4108 		sc->sc_flags |= SCF_SACK_PERMIT;
4109 #endif
4110 #ifdef TCP_SIGNATURE
4111 	if (tb.t_flags & TF_SIGNATURE)
4112 		sc->sc_flags |= SCF_SIGNATURE;
4113 #endif
4114 	sc->sc_tp = tp;
4115 	if (syn_cache_respond(sc, m) == 0) {
4116 		syn_cache_insert(sc, tp);
4117 		tcpstat.tcps_sndacks++;
4118 		tcpstat.tcps_sndtotal++;
4119 	} else {
4120 		syn_cache_put(sc);
4121 		tcpstat.tcps_sc_dropped++;
4122 	}
4123 
4124 	return (0);
4125 }
4126 
4127 int
4128 syn_cache_respond(struct syn_cache *sc, struct mbuf *m)
4129 {
4130 	struct route *ro;
4131 	u_int8_t *optp;
4132 	int optlen, error;
4133 	u_int16_t tlen;
4134 	struct ip *ip = NULL;
4135 #ifdef INET6
4136 	struct ip6_hdr *ip6 = NULL;
4137 #endif
4138 	struct tcphdr *th;
4139 	u_int hlen;
4140 	struct inpcb *inp;
4141 
4142 	switch (sc->sc_src.sa.sa_family) {
4143 	case AF_INET:
4144 		hlen = sizeof(struct ip);
4145 		ro = &sc->sc_route4;
4146 		break;
4147 #ifdef INET6
4148 	case AF_INET6:
4149 		hlen = sizeof(struct ip6_hdr);
4150 		ro = (struct route *)&sc->sc_route6;
4151 		break;
4152 #endif
4153 	default:
4154 		m_freem(m);
4155 		return (EAFNOSUPPORT);
4156 	}
4157 
4158 	/* Compute the size of the TCP options. */
4159 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
4160 #ifdef TCP_SACK
4161 	    ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) +
4162 #endif
4163 #ifdef TCP_SIGNATURE
4164 	    ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) +
4165 #endif
4166 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
4167 
4168 	tlen = hlen + sizeof(struct tcphdr) + optlen;
4169 
4170 	/*
4171 	 * Create the IP+TCP header from scratch.
4172 	 */
4173 	m_freem(m);
4174 #ifdef DIAGNOSTIC
4175 	if (max_linkhdr + tlen > MCLBYTES)
4176 		return (ENOBUFS);
4177 #endif
4178 	MGETHDR(m, M_DONTWAIT, MT_DATA);
4179 	if (m && max_linkhdr + tlen > MHLEN) {
4180 		MCLGET(m, M_DONTWAIT);
4181 		if ((m->m_flags & M_EXT) == 0) {
4182 			m_freem(m);
4183 			m = NULL;
4184 		}
4185 	}
4186 	if (m == NULL)
4187 		return (ENOBUFS);
4188 
4189 	/* Fixup the mbuf. */
4190 	m->m_data += max_linkhdr;
4191 	m->m_len = m->m_pkthdr.len = tlen;
4192 	m->m_pkthdr.ph_ifidx = 0;
4193 	m->m_pkthdr.ph_rtableid = sc->sc_rtableid;
4194 	memset(mtod(m, u_char *), 0, tlen);
4195 
4196 	switch (sc->sc_src.sa.sa_family) {
4197 	case AF_INET:
4198 		ip = mtod(m, struct ip *);
4199 		ip->ip_dst = sc->sc_src.sin.sin_addr;
4200 		ip->ip_src = sc->sc_dst.sin.sin_addr;
4201 		ip->ip_p = IPPROTO_TCP;
4202 		th = (struct tcphdr *)(ip + 1);
4203 		th->th_dport = sc->sc_src.sin.sin_port;
4204 		th->th_sport = sc->sc_dst.sin.sin_port;
4205 		break;
4206 #ifdef INET6
4207 	case AF_INET6:
4208 		ip6 = mtod(m, struct ip6_hdr *);
4209 		ip6->ip6_dst = sc->sc_src.sin6.sin6_addr;
4210 		ip6->ip6_src = sc->sc_dst.sin6.sin6_addr;
4211 		ip6->ip6_nxt = IPPROTO_TCP;
4212 		/* ip6_plen will be updated in ip6_output() */
4213 		th = (struct tcphdr *)(ip6 + 1);
4214 		th->th_dport = sc->sc_src.sin6.sin6_port;
4215 		th->th_sport = sc->sc_dst.sin6.sin6_port;
4216 		break;
4217 #endif
4218 	default:
4219 		unhandled_af(sc->sc_src.sa.sa_family);
4220 	}
4221 
4222 	th->th_seq = htonl(sc->sc_iss);
4223 	th->th_ack = htonl(sc->sc_irs + 1);
4224 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
4225 	th->th_flags = TH_SYN|TH_ACK;
4226 #ifdef TCP_ECN
4227 	/* Set ECE for SYN-ACK if peer supports ECN. */
4228 	if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT))
4229 		th->th_flags |= TH_ECE;
4230 #endif
4231 	th->th_win = htons(sc->sc_win);
4232 	/* th_sum already 0 */
4233 	/* th_urp already 0 */
4234 
4235 	/* Tack on the TCP options. */
4236 	optp = (u_int8_t *)(th + 1);
4237 	*optp++ = TCPOPT_MAXSEG;
4238 	*optp++ = 4;
4239 	*optp++ = (sc->sc_ourmaxseg >> 8) & 0xff;
4240 	*optp++ = sc->sc_ourmaxseg & 0xff;
4241 
4242 #ifdef TCP_SACK
4243 	/* Include SACK_PERMIT_HDR option if peer has already done so. */
4244 	if (sc->sc_flags & SCF_SACK_PERMIT) {
4245 		*((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR);
4246 		optp += 4;
4247 	}
4248 #endif
4249 
4250 	if (sc->sc_request_r_scale != 15) {
4251 		*((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 |
4252 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
4253 		    sc->sc_request_r_scale);
4254 		optp += 4;
4255 	}
4256 
4257 	if (sc->sc_flags & SCF_TIMESTAMP) {
4258 		u_int32_t *lp = (u_int32_t *)(optp);
4259 		/* Form timestamp option as shown in appendix A of RFC 1323. */
4260 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
4261 		*lp++ = htonl(SYN_CACHE_TIMESTAMP(sc));
4262 		*lp   = htonl(sc->sc_timestamp);
4263 		optp += TCPOLEN_TSTAMP_APPA;
4264 	}
4265 
4266 #ifdef TCP_SIGNATURE
4267 	if (sc->sc_flags & SCF_SIGNATURE) {
4268 		union sockaddr_union src, dst;
4269 		struct tdb *tdb;
4270 
4271 		bzero(&src, sizeof(union sockaddr_union));
4272 		bzero(&dst, sizeof(union sockaddr_union));
4273 		src.sa.sa_len = sc->sc_src.sa.sa_len;
4274 		src.sa.sa_family = sc->sc_src.sa.sa_family;
4275 		dst.sa.sa_len = sc->sc_dst.sa.sa_len;
4276 		dst.sa.sa_family = sc->sc_dst.sa.sa_family;
4277 
4278 		switch (sc->sc_src.sa.sa_family) {
4279 		case 0:	/*default to PF_INET*/
4280 		case AF_INET:
4281 			src.sin.sin_addr = mtod(m, struct ip *)->ip_src;
4282 			dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst;
4283 			break;
4284 #ifdef INET6
4285 		case AF_INET6:
4286 			src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src;
4287 			dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst;
4288 			break;
4289 #endif /* INET6 */
4290 		}
4291 
4292 		tdb = gettdbbysrcdst(rtable_l2(sc->sc_rtableid),
4293 		    0, &src, &dst, IPPROTO_TCP);
4294 		if (tdb == NULL) {
4295 			m_freem(m);
4296 			return (EPERM);
4297 		}
4298 
4299 		/* Send signature option */
4300 		*(optp++) = TCPOPT_SIGNATURE;
4301 		*(optp++) = TCPOLEN_SIGNATURE;
4302 
4303 		if (tcp_signature(tdb, sc->sc_src.sa.sa_family, m, th,
4304 		    hlen, 0, optp) < 0) {
4305 			m_freem(m);
4306 			return (EINVAL);
4307 		}
4308 		optp += 16;
4309 
4310 		/* Pad options list to the next 32 bit boundary and
4311 		 * terminate it.
4312 		 */
4313 		*optp++ = TCPOPT_NOP;
4314 		*optp++ = TCPOPT_EOL;
4315 	}
4316 #endif /* TCP_SIGNATURE */
4317 
4318 	/* Compute the packet's checksum. */
4319 	switch (sc->sc_src.sa.sa_family) {
4320 	case AF_INET:
4321 		ip->ip_len = htons(tlen - hlen);
4322 		th->th_sum = 0;
4323 		th->th_sum = in_cksum(m, tlen);
4324 		break;
4325 #ifdef INET6
4326 	case AF_INET6:
4327 		ip6->ip6_plen = htons(tlen - hlen);
4328 		th->th_sum = 0;
4329 		th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen);
4330 		break;
4331 #endif
4332 	}
4333 
4334 	/* use IPsec policy and ttl from listening socket, on SYN ACK */
4335 	inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL;
4336 
4337 	/*
4338 	 * Fill in some straggling IP bits.  Note the stack expects
4339 	 * ip_len to be in host order, for convenience.
4340 	 */
4341 	switch (sc->sc_src.sa.sa_family) {
4342 	case AF_INET:
4343 		ip->ip_len = htons(tlen);
4344 		ip->ip_ttl = inp ? inp->inp_ip.ip_ttl : ip_defttl;
4345 		if (inp != NULL)
4346 			ip->ip_tos = inp->inp_ip.ip_tos;
4347 		break;
4348 #ifdef INET6
4349 	case AF_INET6:
4350 		ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
4351 		ip6->ip6_vfc |= IPV6_VERSION;
4352 		ip6->ip6_plen = htons(tlen - hlen);
4353 		/* ip6_hlim will be initialized afterwards */
4354 		/* leave flowlabel = 0, it is legal and require no state mgmt */
4355 		break;
4356 #endif
4357 	}
4358 
4359 	switch (sc->sc_src.sa.sa_family) {
4360 	case AF_INET:
4361 		error = ip_output(m, sc->sc_ipopts, ro,
4362 		    (ip_mtudisc ? IP_MTUDISC : 0),  NULL, inp, 0);
4363 		break;
4364 #ifdef INET6
4365 	case AF_INET6:
4366 		ip6->ip6_hlim = in6_selecthlim(NULL);
4367 
4368 		error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0,
4369 		    NULL, NULL);
4370 		break;
4371 #endif
4372 	default:
4373 		error = EAFNOSUPPORT;
4374 		break;
4375 	}
4376 	return (error);
4377 }
4378