xref: /openbsd-src/sys/netinet/tcp_input.c (revision ce7e0fc6a9d74d25b78fb6ad846387717f5172b6)
1 /*	$OpenBSD: tcp_input.c,v 1.111 2002/05/16 14:10:51 kjc 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. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)COPYRIGHT	1.1 (NRL) 17 January 1995
37  *
38  * NRL grants permission for redistribution and use in source and binary
39  * forms, with or without modification, of the software and documentation
40  * created at NRL provided that the following conditions are met:
41  *
42  * 1. Redistributions of source code must retain the above copyright
43  *    notice, this list of conditions and the following disclaimer.
44  * 2. Redistributions in binary form must reproduce the above copyright
45  *    notice, this list of conditions and the following disclaimer in the
46  *    documentation and/or other materials provided with the distribution.
47  * 3. All advertising materials mentioning features or use of this software
48  *    must display the following acknowledgements:
49  * 	This product includes software developed by the University of
50  * 	California, Berkeley and its contributors.
51  * 	This product includes software developed at the Information
52  * 	Technology Division, US Naval Research Laboratory.
53  * 4. Neither the name of the NRL nor the names of its contributors
54  *    may be used to endorse or promote products derived from this software
55  *    without specific prior written permission.
56  *
57  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
58  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
59  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
60  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
61  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
62  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
63  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
64  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
65  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
66  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
67  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
68  *
69  * The views and conclusions contained in the software and documentation
70  * are those of the authors and should not be interpreted as representing
71  * official policies, either expressed or implied, of the US Naval
72  * Research Laboratory (NRL).
73  */
74 
75 #ifndef TUBA_INCLUDE
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/mbuf.h>
79 #include <sys/protosw.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/kernel.h>
83 
84 #include <net/if.h>
85 #include <net/route.h>
86 
87 #include <netinet/in.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/ip.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/ip_var.h>
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #include <netinet/tcpip.h>
98 #include <netinet/tcp_debug.h>
99 
100 #ifdef INET6
101 #include <netinet6/in6_var.h>
102 #include <netinet6/nd6.h>
103 
104 struct	tcpiphdr tcp_saveti;
105 struct  tcpipv6hdr tcp_saveti6;
106 
107 /* for the packet header length in the mbuf */
108 #define M_PH_LEN(m)      (((struct mbuf *)(m))->m_pkthdr.len)
109 #define M_V6_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip6_hdr))
110 #define M_V4_LEN(m)      (M_PH_LEN(m) - sizeof(struct ip))
111 #endif /* INET6 */
112 
113 int	tcprexmtthresh = 3;
114 struct	tcpiphdr tcp_saveti;
115 int	tcptv_keep_init = TCPTV_KEEP_INIT;
116 
117 extern u_long sb_max;
118 
119 int tcp_rst_ppslim = 100;		/* 100pps */
120 int tcp_rst_ppslim_count = 0;
121 struct timeval tcp_rst_ppslim_last;
122 
123 #endif /* TUBA_INCLUDE */
124 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
125 
126 /* for modulo comparisons of timestamps */
127 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
128 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
129 
130 /*
131  * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint.
132  */
133 #ifdef INET6
134 #define ND6_HINT(tp) \
135 do { \
136 	if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \
137 	    tp->t_inpcb->inp_route6.ro_rt) { \
138 		nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \
139 	} \
140 } while (0)
141 #else
142 #define ND6_HINT(tp)
143 #endif
144 
145 #ifdef TCP_ECN
146 /*
147  * ECN (Explicit Congestion Notification) support based on RFC3168
148  * implementation note:
149  *   snd_last is used to track a recovery phase.
150  *   when cwnd is reduced, snd_last is set to snd_max.
151  *   while snd_last > snd_una, the sender is in a recovery phase and
152  *   its cwnd should not be reduced again.
153  *   snd_last follows snd_una when not in a recovery phase.
154  */
155 #endif
156 
157 /*
158  * Macro to compute ACK transmission behavior.  Delay the ACK unless
159  * we have already delayed an ACK (must send an ACK every two segments).
160  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
161  * option is enabled.
162  */
163 #define	TCP_SETUP_ACK(tp, tiflags) \
164 do { \
165 	if ((tp)->t_flags & TF_DELACK || \
166 	    (tcp_ack_on_push && (tiflags) & TH_PUSH)) \
167 		tp->t_flags |= TF_ACKNOW; \
168 	else \
169 		TCP_SET_DELACK(tp); \
170 } while (0)
171 
172 /*
173  * Insert segment ti into reassembly queue of tcp with
174  * control block tp.  Return TH_FIN if reassembly now includes
175  * a segment with FIN.  The macro form does the common case inline
176  * (segment is the next to be received on an established connection,
177  * and the queue is empty), avoiding linkage into and removal
178  * from the queue and repetition of various conversions.
179  * Set DELACK for segments received in order, but ack immediately
180  * when segments are out of order (so fast retransmit can work).
181  */
182 
183 #ifndef TUBA_INCLUDE
184 
185 int
186 tcp_reass(tp, th, m, tlen)
187 	struct tcpcb *tp;
188 	struct tcphdr *th;
189 	struct mbuf *m;
190 	int *tlen;
191 {
192 	struct ipqent *p, *q, *nq, *tiqe;
193 	struct socket *so = tp->t_inpcb->inp_socket;
194 	int flags;
195 
196 	/*
197 	 * Call with th==0 after become established to
198 	 * force pre-ESTABLISHED data up to user socket.
199 	 */
200 	if (th == 0)
201 		goto present;
202 
203 	/*
204 	 * Allocate a new queue entry, before we throw away any data.
205 	 * If we can't, just drop the packet.  XXX
206 	 */
207 	tiqe =  pool_get(&ipqent_pool, PR_NOWAIT);
208 	if (tiqe == NULL) {
209 		tcpstat.tcps_rcvmemdrop++;
210 		m_freem(m);
211 		return (0);
212 	}
213 
214 	/*
215 	 * Find a segment which begins after this one does.
216 	 */
217 	for (p = NULL, q = tp->segq.lh_first; q != NULL;
218 	    p = q, q = q->ipqe_q.le_next)
219 		if (SEQ_GT(q->ipqe_tcp->th_seq, th->th_seq))
220 			break;
221 
222 	/*
223 	 * If there is a preceding segment, it may provide some of
224 	 * our data already.  If so, drop the data from the incoming
225 	 * segment.  If it provides all of our data, drop us.
226 	 */
227 	if (p != NULL) {
228 		struct tcphdr *phdr = p->ipqe_tcp;
229 		int i;
230 
231 		/* conversion to int (in i) handles seq wraparound */
232 		i = phdr->th_seq + phdr->th_reseqlen - th->th_seq;
233 		if (i > 0) {
234 		        if (i >= *tlen) {
235 				tcpstat.tcps_rcvduppack++;
236 				tcpstat.tcps_rcvdupbyte += *tlen;
237 				m_freem(m);
238 				pool_put(&ipqent_pool, tiqe);
239 				return (0);
240 			}
241 			m_adj(m, i);
242 			*tlen -= i;
243 			th->th_seq += i;
244 		}
245 	}
246 	tcpstat.tcps_rcvoopack++;
247 	tcpstat.tcps_rcvoobyte += *tlen;
248 
249 	/*
250 	 * While we overlap succeeding segments trim them or,
251 	 * if they are completely covered, dequeue them.
252 	 */
253 	for (; q != NULL; q = nq) {
254 		struct tcphdr *qhdr = q->ipqe_tcp;
255 		int i = (th->th_seq + *tlen) - qhdr->th_seq;
256 
257 		if (i <= 0)
258 			break;
259 		if (i < qhdr->th_reseqlen) {
260 			qhdr->th_seq += i;
261 			qhdr->th_reseqlen -= i;
262 			m_adj(q->ipqe_m, i);
263 			break;
264 		}
265 		nq = q->ipqe_q.le_next;
266 		m_freem(q->ipqe_m);
267 		LIST_REMOVE(q, ipqe_q);
268 		pool_put(&ipqent_pool, q);
269 	}
270 
271 	/* Insert the new fragment queue entry into place. */
272 	tiqe->ipqe_m = m;
273 	th->th_reseqlen = *tlen;
274 	tiqe->ipqe_tcp = th;
275 	if (p == NULL) {
276 		LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
277 	} else {
278 		LIST_INSERT_AFTER(p, tiqe, ipqe_q);
279 	}
280 
281 present:
282 	/*
283 	 * Present data to user, advancing rcv_nxt through
284 	 * completed sequence space.
285 	 */
286 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
287 		return (0);
288 	q = tp->segq.lh_first;
289 	if (q == NULL || q->ipqe_tcp->th_seq != tp->rcv_nxt)
290 		return (0);
291 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->th_reseqlen)
292 		return (0);
293 	do {
294 		tp->rcv_nxt += q->ipqe_tcp->th_reseqlen;
295 		flags = q->ipqe_tcp->th_flags & TH_FIN;
296 
297 		nq = q->ipqe_q.le_next;
298 		LIST_REMOVE(q, ipqe_q);
299 		ND6_HINT(tp);
300 		if (so->so_state & SS_CANTRCVMORE)
301 			m_freem(q->ipqe_m);
302 		else
303 			sbappend(&so->so_rcv, q->ipqe_m);
304 		pool_put(&ipqent_pool, q);
305 		q = nq;
306 	} while (q != NULL && q->ipqe_tcp->th_seq == tp->rcv_nxt);
307 	sorwakeup(so);
308 	return (flags);
309 }
310 
311 /*
312  * First check for a port-specific bomb. We do not want to drop half-opens
313  * for other ports if this is the only port being bombed.  We only check
314  * the bottom 40 half open connections, to avoid wasting too much time.
315  *
316  * Or, otherwise it is more likely a generic syn bomb, so delete the oldest
317  * half-open connection.
318  */
319 void
320 tcpdropoldhalfopen(avoidtp, port)
321 	struct tcpcb *avoidtp;
322 	u_int16_t port;
323 {
324 	struct inpcb *inp;
325 	struct tcpcb *tp;
326 	int ncheck = 40;
327 	int s;
328 
329 	s = splnet();
330 	inp = tcbtable.inpt_queue.cqh_first;
331 	if (inp)						/* XXX */
332 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue && --ncheck;
333 	    inp = inp->inp_queue.cqe_prev) {
334 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
335 		    tp != avoidtp &&
336 		    tp->t_state == TCPS_SYN_RECEIVED &&
337 		    port == inp->inp_lport) {
338 			tcp_close(tp);
339 			goto done;
340 		}
341 	}
342 
343 	inp = tcbtable.inpt_queue.cqh_first;
344 	if (inp)						/* XXX */
345 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
346 	    inp = inp->inp_queue.cqe_prev) {
347 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
348 		    tp != avoidtp &&
349 		    tp->t_state == TCPS_SYN_RECEIVED) {
350 			tcp_close(tp);
351 			goto done;
352 		}
353 	}
354 done:
355 	splx(s);
356 }
357 
358 #ifdef INET6
359 int
360 tcp6_input(mp, offp, proto)
361 	struct mbuf **mp;
362 	int *offp, proto;
363 {
364 	struct mbuf *m = *mp;
365 
366 #if defined(NFAITH) && 0 < NFAITH
367 	if (m->m_pkthdr.rcvif) {
368 		if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) {
369 			/* XXX send icmp6 host/port unreach? */
370 			m_freem(m);
371 			return IPPROTO_DONE;
372 		}
373 	}
374 #endif
375 
376 	/*
377 	 * draft-itojun-ipv6-tcp-to-anycast
378 	 * better place to put this in?
379 	 */
380 	if (m->m_flags & M_ANYCAST6) {
381 		if (m->m_len >= sizeof(struct ip6_hdr)) {
382 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
383 			icmp6_error(m, ICMP6_DST_UNREACH,
384 				ICMP6_DST_UNREACH_ADDR,
385 				(caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
386 		} else
387 			m_freem(m);
388 		return IPPROTO_DONE;
389 	}
390 
391 	tcp_input(m, *offp, proto);
392 	return IPPROTO_DONE;
393 }
394 #endif
395 
396 /*
397  * TCP input routine, follows pages 65-76 of the
398  * protocol specification dated September, 1981 very closely.
399  */
400 void
401 tcp_input(struct mbuf *m, ...)
402 {
403 	struct ip *ip;
404 	struct inpcb *inp;
405 	caddr_t optp = NULL;
406 	int optlen = 0;
407 	int len, tlen, off;
408 	struct tcpcb *tp = 0;
409 	int tiflags;
410 	struct socket *so = NULL;
411 	int todrop, acked, ourfinisacked, needoutput = 0;
412 	int hdroptlen = 0;
413 	short ostate = 0;
414 	struct in_addr laddr;
415 	int dropsocket = 0;
416 	int iss = 0;
417 	u_long tiwin;
418 	u_int32_t ts_val, ts_ecr;
419 	int ts_present = 0;
420 	int iphlen;
421 	va_list ap;
422 	struct tcphdr *th;
423 #ifdef INET6
424 	struct in6_addr laddr6;
425 	struct ip6_hdr *ipv6 = NULL;
426 #endif /* INET6 */
427 #ifdef IPSEC
428 	struct m_tag *mtag;
429 	struct tdb_ident *tdbi;
430 	struct tdb *tdb;
431 	int error, s;
432 #endif /* IPSEC */
433 	int af;
434 #ifdef TCP_ECN
435 	u_char iptos;
436 #endif
437 
438 	va_start(ap, m);
439 	iphlen = va_arg(ap, int);
440 	va_end(ap);
441 
442 	tcpstat.tcps_rcvtotal++;
443 
444 	/*
445 	 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or
446 	 * TCP/IPv4.
447  	 */
448 	switch (mtod(m, struct ip *)->ip_v) {
449 #ifdef INET6
450 	case 6:
451 		af = AF_INET6;
452 		break;
453 #endif
454 	case 4:
455 		af = AF_INET;
456 		break;
457 	default:
458 		m_freem(m);
459 		return;	/*EAFNOSUPPORT*/
460 	}
461 
462 	/*
463 	 * Get IP and TCP header together in first mbuf.
464 	 * Note: IP leaves IP header in first mbuf.
465 	 */
466 	switch (af) {
467 	case AF_INET:
468 #ifdef DIAGNOSTIC
469 		if (iphlen < sizeof(struct ip)) {
470 			m_freem(m);
471 			return;
472 		}
473 #endif /* DIAGNOSTIC */
474 		if (iphlen > sizeof(struct ip)) {
475 #if 0	/*XXX*/
476 			ip_stripoptions(m, (struct mbuf *)0);
477 			iphlen = sizeof(struct ip);
478 #else
479 			m_freem(m);
480 			return;
481 #endif
482 		}
483 		break;
484 #ifdef INET6
485 	case AF_INET6:
486 #ifdef DIAGNOSTIC
487 		if (iphlen < sizeof(struct ip6_hdr)) {
488 			m_freem(m);
489 			return;
490 		}
491 #endif /* DIAGNOSTIC */
492 		if (iphlen > sizeof(struct ip6_hdr)) {
493 #if 0 /*XXX*/
494 			ipv6_stripoptions(m, iphlen);
495 			iphlen = sizeof(struct ip6_hdr);
496 #else
497 			m_freem(m);
498 			return;
499 #endif
500 		}
501 		break;
502 #endif
503 	default:
504 		m_freem(m);
505 		return;
506 	}
507 
508 	if (m->m_len < iphlen + sizeof(struct tcphdr)) {
509 		m = m_pullup2(m, iphlen + sizeof(struct tcphdr));
510 		if (m == NULL) {
511 			tcpstat.tcps_rcvshort++;
512 			return;
513 		}
514 	}
515 
516 	ip = NULL;
517 #ifdef INET6
518 	ipv6 = NULL;
519 #endif
520 	switch (af) {
521 	case AF_INET:
522 	    {
523 		struct tcpiphdr *ti;
524 
525 		ip = mtod(m, struct ip *);
526 #if 1
527 		tlen = m->m_pkthdr.len - iphlen;
528 #else
529 		tlen = ((struct ip *)ti)->ip_len;
530 #endif
531 		ti = mtod(m, struct tcpiphdr *);
532 
533 #ifdef TCP_ECN
534 		/* save ip_tos before clearing it for checksum */
535 		iptos = ip->ip_tos;
536 #endif
537 		/*
538 		 * Checksum extended TCP header and data.
539 		 */
540 		len = sizeof(struct ip) + tlen;
541 		bzero(ti->ti_x1, sizeof ti->ti_x1);
542 		ti->ti_len = (u_int16_t)tlen;
543 		HTONS(ti->ti_len);
544 		if ((m->m_pkthdr.csum & M_TCP_CSUM_IN_OK) == 0) {
545 			if (m->m_pkthdr.csum & M_TCP_CSUM_IN_BAD) {
546 				tcpstat.tcps_inhwcsum++;
547 				tcpstat.tcps_rcvbadsum++;
548 				goto drop;
549 			}
550 			if ((ti->ti_sum = in_cksum(m, len)) != 0) {
551 				tcpstat.tcps_rcvbadsum++;
552 				goto drop;
553 			}
554 		} else {
555 			m->m_pkthdr.csum &= ~M_TCP_CSUM_IN_OK;
556 			tcpstat.tcps_inhwcsum++;
557 		}
558 		break;
559 	    }
560 #ifdef INET6
561 	case AF_INET6:
562 		ipv6 = mtod(m, struct ip6_hdr *);
563 		tlen = m->m_pkthdr.len - iphlen;
564 #ifdef TCP_ECN
565 		iptos = (ntohl(ipv6->ip6_flow) >> 20) & 0xff;
566 #endif
567 
568 		/* Be proactive about malicious use of IPv4 mapped address */
569 		if (IN6_IS_ADDR_V4MAPPED(&ipv6->ip6_src) ||
570 		    IN6_IS_ADDR_V4MAPPED(&ipv6->ip6_dst)) {
571 			/* XXX stat */
572 			goto drop;
573 		}
574 
575 		/*
576 		 * Be proactive about unspecified IPv6 address in source.
577 		 * As we use all-zero to indicate unbounded/unconnected pcb,
578 		 * unspecified IPv6 address can be used to confuse us.
579 		 *
580 		 * Note that packets with unspecified IPv6 destination is
581 		 * already dropped in ip6_input.
582 		 */
583 		if (IN6_IS_ADDR_UNSPECIFIED(&ipv6->ip6_src)) {
584 			/* XXX stat */
585 			goto drop;
586 		}
587 
588 		/*
589 		 * Checksum extended TCP header and data.
590 		 */
591 		if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) {
592 			tcpstat.tcps_rcvbadsum++;
593 			goto drop;
594 		}
595 		break;
596 #endif
597 	}
598 #endif /* TUBA_INCLUDE */
599 
600 	th = (struct tcphdr *)(mtod(m, caddr_t) + iphlen);
601 
602 	/*
603 	 * Check that TCP offset makes sense,
604 	 * pull out TCP options and adjust length.		XXX
605 	 */
606 	off = th->th_off << 2;
607 	if (off < sizeof(struct tcphdr) || off > tlen) {
608 		tcpstat.tcps_rcvbadoff++;
609 		goto drop;
610 	}
611 	tlen -= off;
612 	if (off > sizeof(struct tcphdr)) {
613 		if (m->m_len < iphlen + off) {
614 			if ((m = m_pullup2(m, iphlen + off)) == NULL) {
615 				tcpstat.tcps_rcvshort++;
616 				return;
617 			}
618 			switch (af) {
619 			case AF_INET:
620 				ip = mtod(m, struct ip *);
621 				break;
622 #ifdef INET6
623 			case AF_INET6:
624 				ipv6 = mtod(m, struct ip6_hdr *);
625 				break;
626 #endif
627 			}
628 			th = (struct tcphdr *)(mtod(m, caddr_t) + iphlen);
629 		}
630 		optlen = off - sizeof(struct tcphdr);
631 		optp = mtod(m, caddr_t) + iphlen + sizeof(struct tcphdr);
632 		/*
633 		 * Do quick retrieval of timestamp options ("options
634 		 * prediction?").  If timestamp is the only option and it's
635 		 * formatted as recommended in RFC 1323 appendix A, we
636 		 * quickly get the values now and not bother calling
637 		 * tcp_dooptions(), etc.
638 		 */
639 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
640 		     (optlen > TCPOLEN_TSTAMP_APPA &&
641 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
642 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
643 		     (th->th_flags & TH_SYN) == 0) {
644 			ts_present = 1;
645 			ts_val = ntohl(*(u_int32_t *)(optp + 4));
646 			ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
647 			optp = NULL;	/* we've parsed the options */
648 		}
649 	}
650 	tiflags = th->th_flags;
651 
652 	/*
653 	 * Convert TCP protocol specific fields to host format.
654 	 */
655 	NTOHL(th->th_seq);
656 	NTOHL(th->th_ack);
657 	NTOHS(th->th_win);
658 	NTOHS(th->th_urp);
659 
660 	/*
661 	 * Locate pcb for segment.
662 	 */
663 findpcb:
664 	switch (af) {
665 #ifdef INET6
666 	case AF_INET6:
667 		inp = in6_pcbhashlookup(&tcbtable, &ipv6->ip6_src, th->th_sport,
668 		    &ipv6->ip6_dst, th->th_dport);
669 		break;
670 #endif
671 	case AF_INET:
672 		inp = in_pcbhashlookup(&tcbtable, ip->ip_src, th->th_sport,
673 		    ip->ip_dst, th->th_dport);
674 		break;
675 	}
676 	if (inp == 0) {
677 		++tcpstat.tcps_pcbhashmiss;
678 		switch (af) {
679 #ifdef INET6
680 		case AF_INET6:
681 			inp = in_pcblookup(&tcbtable, &ipv6->ip6_src,
682 			    th->th_sport, &ipv6->ip6_dst, th->th_dport,
683 			    INPLOOKUP_WILDCARD | INPLOOKUP_IPV6);
684 			break;
685 #endif /* INET6 */
686 		case AF_INET:
687 			inp = in_pcblookup(&tcbtable, &ip->ip_src, th->th_sport,
688 			    &ip->ip_dst, th->th_dport, INPLOOKUP_WILDCARD);
689 			break;
690 		}
691 		/*
692 		 * If the state is CLOSED (i.e., TCB does not exist) then
693 		 * all data in the incoming segment is discarded.
694 		 * If the TCB exists but is in CLOSED state, it is embryonic,
695 		 * but should either do a listen or a connect soon.
696 		 */
697 		if (inp == 0) {
698 			++tcpstat.tcps_noport;
699 			goto dropwithreset_ratelim;
700 		}
701 	}
702 
703 	tp = intotcpcb(inp);
704 	if (tp == 0)
705 		goto dropwithreset_ratelim;
706 	if (tp->t_state == TCPS_CLOSED)
707 		goto drop;
708 
709 	/* Unscale the window into a 32-bit value. */
710 	if ((tiflags & TH_SYN) == 0)
711 		tiwin = th->th_win << tp->snd_scale;
712 	else
713 		tiwin = th->th_win;
714 
715 	so = inp->inp_socket;
716 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
717 		if (so->so_options & SO_DEBUG) {
718 			ostate = tp->t_state;
719 			switch (af) {
720 #ifdef INET6
721 			case AF_INET6:
722 				tcp_saveti6 = *(mtod(m, struct tcpipv6hdr *));
723 				break;
724 #endif
725 			case AF_INET:
726 				tcp_saveti = *(mtod(m, struct tcpiphdr *));
727 				break;
728 			}
729 		}
730 		if (so->so_options & SO_ACCEPTCONN) {
731 			struct socket *so1;
732 
733 			so1 = sonewconn(so, 0);
734 			if (so1 == NULL) {
735 				tcpdropoldhalfopen(tp, th->th_dport);
736 				so1 = sonewconn(so, 0);
737 				if (so1 == NULL)
738 					goto drop;
739 			}
740 			so = so1;
741 			/*
742 			 * This is ugly, but ....
743 			 *
744 			 * Mark socket as temporary until we're
745 			 * committed to keeping it.  The code at
746 			 * ``drop'' and ``dropwithreset'' check the
747 			 * flag dropsocket to see if the temporary
748 			 * socket created here should be discarded.
749 			 * We mark the socket as discardable until
750 			 * we're committed to it below in TCPS_LISTEN.
751 			 */
752 			dropsocket++;
753 #ifdef IPSEC
754 			/*
755 			 * We need to copy the required security levels
756 			 * from the old pcb. Ditto for any other
757 			 * IPsec-related information.
758 			 */
759 			{
760 			  struct inpcb *newinp = (struct inpcb *)so->so_pcb;
761 			  bcopy(inp->inp_seclevel, newinp->inp_seclevel,
762 				sizeof(inp->inp_seclevel));
763 			  newinp->inp_secrequire = inp->inp_secrequire;
764 			  if (inp->inp_ipsec_localid != NULL) {
765 			  	newinp->inp_ipsec_localid = inp->inp_ipsec_localid;
766 				inp->inp_ipsec_localid->ref_count++;
767 			  }
768 			  if (inp->inp_ipsec_remoteid != NULL) {
769 			  	newinp->inp_ipsec_remoteid = inp->inp_ipsec_remoteid;
770 				inp->inp_ipsec_remoteid->ref_count++;
771 			  }
772 			  if (inp->inp_ipsec_localcred != NULL) {
773 			  	newinp->inp_ipsec_localcred = inp->inp_ipsec_localcred;
774 				inp->inp_ipsec_localcred->ref_count++;
775 			  }
776 			  if (inp->inp_ipsec_remotecred != NULL) {
777 			  	newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred;
778 				inp->inp_ipsec_remotecred->ref_count++;
779 			  }
780 			  if (inp->inp_ipsec_localauth != NULL) {
781 			  	newinp->inp_ipsec_localauth
782 				  = inp->inp_ipsec_localauth;
783 				inp->inp_ipsec_localauth->ref_count++;
784 			  }
785 			  if (inp->inp_ipsec_remoteauth != NULL) {
786 			  	newinp->inp_ipsec_remoteauth
787 				  = inp->inp_ipsec_remoteauth;
788 				inp->inp_ipsec_remoteauth->ref_count++;
789 			  }
790 			}
791 #endif /* IPSEC */
792 #ifdef INET6
793 			/*
794 			 * inp still has the OLD in_pcb stuff, set the
795 			 * v6-related flags on the new guy, too.   This is
796 			 * done particularly for the case where an AF_INET6
797 			 * socket is bound only to a port, and a v4 connection
798 			 * comes in on that port.
799 			 * we also copy the flowinfo from the original pcb
800 			 * to the new one.
801 			 */
802 			{
803 			  int flags = inp->inp_flags;
804 			  struct inpcb *oldinpcb = inp;
805 
806 			  inp = (struct inpcb *)so->so_pcb;
807 			  inp->inp_flags |= (flags & INP_IPV6);
808 			  if ((inp->inp_flags & INP_IPV6) != 0) {
809 			    inp->inp_ipv6.ip6_hlim =
810 			      oldinpcb->inp_ipv6.ip6_hlim;
811 			    inp->inp_ipv6.ip6_flow =
812 			      oldinpcb->inp_ipv6.ip6_flow;
813 			  }
814 			}
815 #else /* INET6 */
816 			inp = (struct inpcb *)so->so_pcb;
817 #endif /* INET6 */
818 			inp->inp_lport = th->th_dport;
819 			switch (af) {
820 #ifdef INET6
821 			case AF_INET6:
822 				inp->inp_laddr6 = ipv6->ip6_dst;
823 
824 				/*inp->inp_options = ip6_srcroute();*/ /* soon. */
825 				/*
826 				 * still need to tweak outbound options
827 				 * processing to include this mbuf in
828 				 * the right place and put the correct
829 				 * NextHdr values in the right places.
830 				 * XXX  rja
831 				 */
832 				break;
833 #endif /* INET6 */
834 			case AF_INET:
835 				inp->inp_laddr = ip->ip_dst;
836 				inp->inp_options = ip_srcroute();
837 				break;
838 			}
839 			in_pcbrehash(inp);
840 			tp = intotcpcb(inp);
841 			tp->t_state = TCPS_LISTEN;
842 
843 			/* Compute proper scaling value from buffer space */
844 			tcp_rscale(tp, so->so_rcv.sb_hiwat);
845 		}
846 	}
847 
848 #ifdef IPSEC
849 	/* Find most recent IPsec tag */
850 	mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL);
851         s = splnet();
852 	if (mtag != NULL) {
853 		tdbi = (struct tdb_ident *)(mtag + 1);
854 	        tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto);
855 	} else
856 		tdb = NULL;
857 	ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN,
858 	    tdb, inp);
859 
860 	/* Latch SA */
861 	if (inp->inp_tdb_in != tdb) {
862 		if (tdb) {
863 		        tdb_add_inp(tdb, inp, 1);
864 			if (inp->inp_ipsec_remoteid == NULL &&
865 			    tdb->tdb_srcid != NULL) {
866 				inp->inp_ipsec_remoteid = tdb->tdb_srcid;
867 				tdb->tdb_srcid->ref_count++;
868 			}
869 			if (inp->inp_ipsec_remotecred == NULL &&
870 			    tdb->tdb_remote_cred != NULL) {
871 				inp->inp_ipsec_remotecred =
872 				    tdb->tdb_remote_cred;
873 				tdb->tdb_remote_cred->ref_count++;
874 			}
875 			if (inp->inp_ipsec_remoteauth == NULL &&
876 			    tdb->tdb_remote_auth != NULL) {
877 				inp->inp_ipsec_remoteauth =
878 				    tdb->tdb_remote_auth;
879 				tdb->tdb_remote_auth->ref_count++;
880 			}
881 		} else { /* Just reset */
882 		        TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp,
883 				     inp_tdb_in_next);
884 			inp->inp_tdb_in = NULL;
885 		}
886 	}
887         splx(s);
888 
889 	/* Error or otherwise drop-packet indication */
890 	if (error)
891 		goto drop;
892 #endif /* IPSEC */
893 
894 	/*
895 	 * Segment received on connection.
896 	 * Reset idle time and keep-alive timer.
897 	 */
898 	tp->t_rcvtime = tcp_now;
899 	if (tp->t_state != TCPS_SYN_RECEIVED)
900 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
901 
902 #ifdef TCP_SACK
903 	if (!tp->sack_disable)
904 		tcp_del_sackholes(tp, th); /* Delete stale SACK holes */
905 #endif /* TCP_SACK */
906 
907 	/*
908 	 * Process options if not in LISTEN state,
909 	 * else do it below (after getting remote address).
910 	 */
911 	if (optp && tp->t_state != TCPS_LISTEN)
912 		tcp_dooptions(tp, optp, optlen, th,
913 			&ts_present, &ts_val, &ts_ecr);
914 
915 #ifdef TCP_SACK
916 	if (!tp->sack_disable) {
917 		tp->rcv_laststart = th->th_seq; /* last rec'vd segment*/
918 		tp->rcv_lastend = th->th_seq + tlen;
919 	}
920 #endif /* TCP_SACK */
921 #ifdef TCP_ECN
922 	/* if congestion experienced, set ECE bit in subsequent packets. */
923 	if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) {
924 		tp->t_flags |= TF_RCVD_CE;
925 		tcpstat.tcps_ecn_rcvce++;
926 	}
927 #endif
928 	/*
929 	 * Header prediction: check for the two common cases
930 	 * of a uni-directional data xfer.  If the packet has
931 	 * no control flags, is in-sequence, the window didn't
932 	 * change and we're not retransmitting, it's a
933 	 * candidate.  If the length is zero and the ack moved
934 	 * forward, we're the sender side of the xfer.  Just
935 	 * free the data acked & wake any higher level process
936 	 * that was blocked waiting for space.  If the length
937 	 * is non-zero and the ack didn't move, we're the
938 	 * receiver side.  If we're getting packets in-order
939 	 * (the reassembly queue is empty), add the data to
940 	 * the socket buffer and note that we need a delayed ack.
941 	 */
942 	if (tp->t_state == TCPS_ESTABLISHED &&
943 #ifdef TCP_ECN
944 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK &&
945 #else
946 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
947 #endif
948 	    (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) &&
949 	    th->th_seq == tp->rcv_nxt &&
950 	    tiwin && tiwin == tp->snd_wnd &&
951 	    tp->snd_nxt == tp->snd_max) {
952 
953 		/*
954 		 * If last ACK falls within this segment's sequence numbers,
955 		 *  record the timestamp.
956 		 * Fix from Braden, see Stevens p. 870
957 		 */
958 		if (ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
959 			tp->ts_recent_age = tcp_now;
960 			tp->ts_recent = ts_val;
961 		}
962 
963 		if (tlen == 0) {
964 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
965 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
966 			    tp->snd_cwnd >= tp->snd_wnd &&
967 			    tp->t_dupacks == 0) {
968 				/*
969 				 * this is a pure ack for outstanding data.
970 				 */
971 				++tcpstat.tcps_predack;
972 				if (ts_present)
973 					tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
974 				else if (tp->t_rtttime &&
975 					    SEQ_GT(th->th_ack, tp->t_rtseq))
976 					tcp_xmit_timer(tp,
977 					    tcp_now - tp->t_rtttime);
978 				acked = th->th_ack - tp->snd_una;
979 				tcpstat.tcps_rcvackpack++;
980 				tcpstat.tcps_rcvackbyte += acked;
981 				ND6_HINT(tp);
982 				sbdrop(&so->so_snd, acked);
983 				tp->snd_una = th->th_ack;
984 #if defined(TCP_SACK) || defined(TCP_ECN)
985 				/*
986 				 * We want snd_last to track snd_una so
987 				 * as to avoid sequence wraparound problems
988 				 * for very large transfers.
989 				 */
990 #ifdef TCP_ECN
991 				if (SEQ_GT(tp->snd_una, tp->snd_last))
992 #endif
993 				tp->snd_last = tp->snd_una;
994 #endif /* TCP_SACK */
995 #if defined(TCP_SACK) && defined(TCP_FACK)
996 				tp->snd_fack = tp->snd_una;
997 				tp->retran_data = 0;
998 #endif /* TCP_FACK */
999 				m_freem(m);
1000 
1001 				/*
1002 				 * If all outstanding data are acked, stop
1003 				 * retransmit timer, otherwise restart timer
1004 				 * using current (possibly backed-off) value.
1005 				 * If process is waiting for space,
1006 				 * wakeup/selwakeup/signal.  If data
1007 				 * are ready to send, let tcp_output
1008 				 * decide between more output or persist.
1009 				 */
1010 				if (tp->snd_una == tp->snd_max)
1011 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1012 				else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1013 					TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1014 
1015 				if (sb_notify(&so->so_snd))
1016 					sowwakeup(so);
1017 				if (so->so_snd.sb_cc)
1018 					(void) tcp_output(tp);
1019 				return;
1020 			}
1021 		} else if (th->th_ack == tp->snd_una &&
1022 		    tp->segq.lh_first == NULL &&
1023 		    tlen <= sbspace(&so->so_rcv)) {
1024 			/*
1025 			 * This is a pure, in-sequence data packet
1026 			 * with nothing on the reassembly queue and
1027 			 * we have enough buffer space to take it.
1028 			 */
1029 #ifdef TCP_SACK
1030 			/* Clean receiver SACK report if present */
1031 			if (!tp->sack_disable && tp->rcv_numsacks)
1032 				tcp_clean_sackreport(tp);
1033 #endif /* TCP_SACK */
1034 			++tcpstat.tcps_preddat;
1035 			tp->rcv_nxt += tlen;
1036 			tcpstat.tcps_rcvpack++;
1037 			tcpstat.tcps_rcvbyte += tlen;
1038 			ND6_HINT(tp);
1039 			/*
1040 			 * Drop TCP, IP headers and TCP options then add data
1041 			 * to socket buffer.
1042 			 */
1043 			m_adj(m, iphlen + off);
1044 			sbappend(&so->so_rcv, m);
1045 			sorwakeup(so);
1046 			TCP_SETUP_ACK(tp, tiflags);
1047 			if (tp->t_flags & TF_ACKNOW)
1048 				(void) tcp_output(tp);
1049 			return;
1050 		}
1051 	}
1052 
1053 	/*
1054 	 * Compute mbuf offset to TCP data segment.
1055 	 */
1056 	hdroptlen = iphlen + off;
1057 
1058 	/*
1059 	 * Calculate amount of space in receive window,
1060 	 * and then do TCP input processing.
1061 	 * Receive window is amount of space in rcv queue,
1062 	 * but not less than advertised window.
1063 	 */
1064 	{ int win;
1065 
1066 	win = sbspace(&so->so_rcv);
1067 	if (win < 0)
1068 		win = 0;
1069 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1070 	}
1071 
1072 	switch (tp->t_state) {
1073 
1074 	/*
1075 	 * If the state is LISTEN then ignore segment if it contains an RST.
1076 	 * If the segment contains an ACK then it is bad and send a RST.
1077 	 * If it does not contain a SYN then it is not interesting; drop it.
1078 	 * If it is from this socket, drop it, it must be forged.
1079 	 * Don't bother responding if the destination was a broadcast.
1080 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
1081 	 * tp->iss, and send a segment:
1082 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1083 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
1084 	 * Fill in remote peer address fields if not previously specified.
1085 	 * Enter SYN_RECEIVED state, and process any other fields of this
1086 	 * segment in this state.
1087 	 */
1088 	case TCPS_LISTEN: {
1089 		struct mbuf *am;
1090 		struct sockaddr_in *sin;
1091 #ifdef INET6
1092 		struct sockaddr_in6 *sin6;
1093 #endif /* INET6 */
1094 
1095 		if (tiflags & TH_RST)
1096 			goto drop;
1097 		if (tiflags & TH_ACK)
1098 			goto dropwithreset;
1099 		if ((tiflags & TH_SYN) == 0)
1100 			goto drop;
1101 		if (th->th_dport == th->th_sport) {
1102 			switch (af) {
1103 #ifdef INET6
1104 			case AF_INET6:
1105 				if (IN6_ARE_ADDR_EQUAL(&ipv6->ip6_src,
1106 				    &ipv6->ip6_dst))
1107 					goto drop;
1108 				break;
1109 #endif /* INET6 */
1110 			case AF_INET:
1111 				if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1112 					goto drop;
1113 				break;
1114 			}
1115 		}
1116 
1117 		/*
1118 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1119 		 */
1120 		if (m->m_flags & (M_BCAST|M_MCAST))
1121 			goto drop;
1122 		switch (af) {
1123 #ifdef INET6
1124 		case AF_INET6:
1125 			if (IN6_IS_ADDR_MULTICAST(&ipv6->ip6_dst))
1126 				goto drop;
1127 			break;
1128 #endif /* INET6 */
1129 		case AF_INET:
1130 			if (IN_MULTICAST(ip->ip_dst.s_addr) ||
1131 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1132 				goto drop;
1133 			break;
1134 		}
1135 		am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
1136 		if (am == NULL)
1137 			goto drop;
1138 		switch (af) {
1139 #ifdef INET6
1140 		case AF_INET6:
1141 			/*
1142 			 * This is probably the place to set the tp->pf value.
1143 			 * (Don't forget to do it in the v4 code as well!)
1144 			 *
1145 			 * Also, remember to blank out things like flowlabel, or
1146 			 * set flowlabel for accepted sockets in v6.
1147 			 *
1148 			 * FURTHERMORE, this is PROBABLY the place where the
1149 			 * whole business of key munging is set up for passive
1150 			 * connections.
1151 			 */
1152 			am->m_len = sizeof(struct sockaddr_in6);
1153 			sin6 = mtod(am, struct sockaddr_in6 *);
1154 			sin6->sin6_family = AF_INET6;
1155 			sin6->sin6_len = sizeof(struct sockaddr_in6);
1156 			sin6->sin6_addr = ipv6->ip6_src;
1157 			sin6->sin6_port = th->th_sport;
1158 			sin6->sin6_flowinfo = htonl(0x0fffffff) &
1159 				inp->inp_ipv6.ip6_flow;
1160 			laddr6 = inp->inp_laddr6;
1161 			if (IN6_IS_ADDR_UNSPECIFIED(&inp->inp_laddr6))
1162 				inp->inp_laddr6 = ipv6->ip6_dst;
1163 			/* This is a good optimization. */
1164 			if (in6_pcbconnect(inp, am)) {
1165 				inp->inp_laddr6 = laddr6;
1166 				(void) m_free(am);
1167 				goto drop;
1168 			}
1169 			break;
1170 #endif
1171 		case AF_INET:
1172 			/* drop IPv4 packet to AF_INET6 socket */
1173 			if (inp->inp_flags & INP_IPV6) {
1174 				(void) m_free(am);
1175 				goto drop;
1176 			}
1177 			am->m_len = sizeof(struct sockaddr_in);
1178 			sin = mtod(am, struct sockaddr_in *);
1179 			sin->sin_family = AF_INET;
1180 			sin->sin_len = sizeof(*sin);
1181 			sin->sin_addr = ip->ip_src;
1182 			sin->sin_port = th->th_sport;
1183 			bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
1184 			laddr = inp->inp_laddr;
1185 			if (inp->inp_laddr.s_addr == INADDR_ANY)
1186 				inp->inp_laddr = ip->ip_dst;
1187 			if (in_pcbconnect(inp, am)) {
1188 				inp->inp_laddr = laddr;
1189 				(void) m_free(am);
1190 				goto drop;
1191 			}
1192 			(void) m_free(am);
1193 			break;
1194 		}
1195 		tp->t_template = tcp_template(tp);
1196 		if (tp->t_template == 0) {
1197 			tp = tcp_drop(tp, ENOBUFS);
1198 			dropsocket = 0;		/* socket is already gone */
1199 			goto drop;
1200 		}
1201 		if (optp)
1202 			tcp_dooptions(tp, optp, optlen, th,
1203 				&ts_present, &ts_val, &ts_ecr);
1204 #ifdef TCP_SACK
1205 		/*
1206 		 * If peer did not send a SACK_PERMITTED option (i.e., if
1207 		 * tcp_dooptions() did not set TF_SACK_PERMIT), set
1208                  * sack_disable to 1 if it is currently 0.
1209                  */
1210                 if (!tp->sack_disable)
1211                         if ((tp->t_flags & TF_SACK_PERMIT) == 0)
1212                                 tp->sack_disable = 1;
1213 #endif
1214 
1215 		if (iss)
1216 			tp->iss = iss;
1217 		else {
1218 #ifdef TCP_COMPAT_42
1219 			tcp_iss += TCP_ISSINCR/2;
1220 			tp->iss = tcp_iss;
1221 #else /* TCP_COMPAT_42 */
1222 			tp->iss = tcp_rndiss_next();
1223 #endif /* !TCP_COMPAT_42 */
1224 		}
1225 		tp->irs = th->th_seq;
1226 		tcp_sendseqinit(tp);
1227 #if defined (TCP_SACK) || defined(TCP_ECN)
1228 		tp->snd_last = tp->snd_una;
1229 #endif /* TCP_SACK */
1230 #if defined(TCP_SACK) && defined(TCP_FACK)
1231 		tp->snd_fack = tp->snd_una;
1232 		tp->retran_data = 0;
1233 		tp->snd_awnd = 0;
1234 #endif /* TCP_FACK */
1235 #ifdef TCP_ECN
1236 		/*
1237 		 * if both ECE and CWR flag bits are set, peer is ECN capable.
1238 		 */
1239 		if (tcp_do_ecn &&
1240 		    (tiflags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) {
1241 			tp->t_flags |= TF_ECN_PERMIT;
1242 			tcpstat.tcps_ecn_accepts++;
1243 		}
1244 #endif
1245 		tcp_rcvseqinit(tp);
1246 		tp->t_flags |= TF_ACKNOW;
1247 		tp->t_state = TCPS_SYN_RECEIVED;
1248 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init);
1249 		dropsocket = 0;		/* committed to socket */
1250 		tcpstat.tcps_accepts++;
1251 		goto trimthenstep6;
1252 		}
1253 
1254 	/*
1255 	 * If the state is SYN_RECEIVED:
1256 	 * 	if seg contains SYN/ACK, send an RST.
1257 	 *	if seg contains an ACK, but not for our SYN/ACK, send an RST
1258   	 */
1259 
1260 	case TCPS_SYN_RECEIVED:
1261 		if (tiflags & TH_ACK) {
1262 			if (tiflags & TH_SYN) {
1263 				tcpstat.tcps_badsyn++;
1264 				goto dropwithreset;
1265 			}
1266 			if (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1267 			    SEQ_GT(th->th_ack, tp->snd_max))
1268 				goto dropwithreset;
1269 		}
1270 		break;
1271 
1272 	/*
1273 	 * If the state is SYN_SENT:
1274 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1275 	 *	if seg contains a RST, then drop the connection.
1276 	 *	if seg does not contain SYN, then drop it.
1277 	 * Otherwise this is an acceptable SYN segment
1278 	 *	initialize tp->rcv_nxt and tp->irs
1279 	 *	if seg contains ack then advance tp->snd_una
1280 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1281 	 *	arrange for segment to be acked (eventually)
1282 	 *	continue processing rest of data/controls, beginning with URG
1283 	 */
1284 	case TCPS_SYN_SENT:
1285 		if ((tiflags & TH_ACK) &&
1286 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1287 		     SEQ_GT(th->th_ack, tp->snd_max)))
1288 			goto dropwithreset;
1289 		if (tiflags & TH_RST) {
1290 #ifdef TCP_ECN
1291 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1292 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1293 				goto drop;
1294 #endif
1295 			if (tiflags & TH_ACK)
1296 				tp = tcp_drop(tp, ECONNREFUSED);
1297 			goto drop;
1298 		}
1299 		if ((tiflags & TH_SYN) == 0)
1300 			goto drop;
1301 		if (tiflags & TH_ACK) {
1302 			tp->snd_una = th->th_ack;
1303 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1304 				tp->snd_nxt = tp->snd_una;
1305 		}
1306 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1307 		tp->irs = th->th_seq;
1308 		tcp_rcvseqinit(tp);
1309 		tp->t_flags |= TF_ACKNOW;
1310 #ifdef TCP_SACK
1311                 /*
1312                  * If we've sent a SACK_PERMITTED option, and the peer
1313                  * also replied with one, then TF_SACK_PERMIT should have
1314                  * been set in tcp_dooptions().  If it was not, disable SACKs.
1315                  */
1316                 if (!tp->sack_disable)
1317                         if ((tp->t_flags & TF_SACK_PERMIT) == 0)
1318                                 tp->sack_disable = 1;
1319 #endif
1320 #ifdef TCP_ECN
1321 		/*
1322 		 * if ECE is set but CWR is not set for SYN-ACK, or
1323 		 * both ECE and CWR are set for simultaneous open,
1324 		 * peer is ECN capable.
1325 		 */
1326 		if (tcp_do_ecn) {
1327 			if ((tiflags & (TH_ACK|TH_ECE|TH_CWR))
1328 			    == (TH_ACK|TH_ECE) ||
1329 			    (tiflags & (TH_ACK|TH_ECE|TH_CWR))
1330 			    == (TH_ECE|TH_CWR)) {
1331 				tp->t_flags |= TF_ECN_PERMIT;
1332 				tiflags &= ~(TH_ECE|TH_CWR);
1333 				tcpstat.tcps_ecn_accepts++;
1334 			}
1335 		}
1336 #endif
1337 
1338 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
1339 			tcpstat.tcps_connects++;
1340 			soisconnected(so);
1341 			tp->t_state = TCPS_ESTABLISHED;
1342 			/* Do window scaling on this connection? */
1343 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1344 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1345 				tp->snd_scale = tp->requested_s_scale;
1346 				tp->rcv_scale = tp->request_r_scale;
1347 			}
1348 			(void) tcp_reass(tp, (struct tcphdr *)0,
1349 				(struct mbuf *)0, &tlen);
1350 			/*
1351 			 * if we didn't have to retransmit the SYN,
1352 			 * use its rtt as our initial srtt & rtt var.
1353 			 */
1354 			if (tp->t_rtttime)
1355 				tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1356 			/*
1357 			 * Since new data was acked (the SYN), open the
1358 			 * congestion window by one MSS.  We do this
1359 			 * here, because we won't go through the normal
1360 			 * ACK processing below.  And since this is the
1361 			 * start of the connection, we know we are in
1362 			 * the exponential phase of slow-start.
1363 			 */
1364 			tp->snd_cwnd += tp->t_maxseg;
1365 		} else
1366 			tp->t_state = TCPS_SYN_RECEIVED;
1367 
1368 trimthenstep6:
1369 		/*
1370 		 * Advance th->th_seq to correspond to first data byte.
1371 		 * If data, trim to stay within window,
1372 		 * dropping FIN if necessary.
1373 		 */
1374 		th->th_seq++;
1375 		if (tlen > tp->rcv_wnd) {
1376 			todrop = tlen - tp->rcv_wnd;
1377 			m_adj(m, -todrop);
1378 			tlen = tp->rcv_wnd;
1379 			tiflags &= ~TH_FIN;
1380 			tcpstat.tcps_rcvpackafterwin++;
1381 			tcpstat.tcps_rcvbyteafterwin += todrop;
1382 		}
1383 		tp->snd_wl1 = th->th_seq - 1;
1384 		tp->rcv_up = th->th_seq;
1385 		goto step6;
1386 	}
1387 
1388 	/*
1389 	 * States other than LISTEN or SYN_SENT.
1390 	 * First check timestamp, if present.
1391 	 * Then check that at least some bytes of segment are within
1392 	 * receive window.  If segment begins before rcv_nxt,
1393 	 * drop leading data (and SYN); if nothing left, just ack.
1394 	 *
1395 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1396 	 * and it's less than ts_recent, drop it.
1397 	 */
1398 	if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
1399 	    TSTMP_LT(ts_val, tp->ts_recent)) {
1400 
1401 		/* Check to see if ts_recent is over 24 days old.  */
1402 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1403 			/*
1404 			 * Invalidate ts_recent.  If this segment updates
1405 			 * ts_recent, the age will be reset later and ts_recent
1406 			 * will get a valid value.  If it does not, setting
1407 			 * ts_recent to zero will at least satisfy the
1408 			 * requirement that zero be placed in the timestamp
1409 			 * echo reply when ts_recent isn't valid.  The
1410 			 * age isn't reset until we get a valid ts_recent
1411 			 * because we don't want out-of-order segments to be
1412 			 * dropped when ts_recent is old.
1413 			 */
1414 			tp->ts_recent = 0;
1415 		} else {
1416 			tcpstat.tcps_rcvduppack++;
1417 			tcpstat.tcps_rcvdupbyte += tlen;
1418 			tcpstat.tcps_pawsdrop++;
1419 			goto dropafterack;
1420 		}
1421 	}
1422 
1423 	todrop = tp->rcv_nxt - th->th_seq;
1424 	if (todrop > 0) {
1425 		if (tiflags & TH_SYN) {
1426 			tiflags &= ~TH_SYN;
1427 			th->th_seq++;
1428 			if (th->th_urp > 1)
1429 				th->th_urp--;
1430 			else
1431 				tiflags &= ~TH_URG;
1432 			todrop--;
1433 		}
1434 		if (todrop >= tlen ||
1435 		    (todrop == tlen && (tiflags & TH_FIN) == 0)) {
1436 			/*
1437 			 * Any valid FIN must be to the left of the
1438 			 * window.  At this point, FIN must be a
1439 			 * duplicate or out-of-sequence, so drop it.
1440 			 */
1441 			tiflags &= ~TH_FIN;
1442 			/*
1443 			 * Send ACK to resynchronize, and drop any data,
1444 			 * but keep on processing for RST or ACK.
1445 			 */
1446 			tp->t_flags |= TF_ACKNOW;
1447 			tcpstat.tcps_rcvdupbyte += todrop = tlen;
1448 			tcpstat.tcps_rcvduppack++;
1449 		} else {
1450 			tcpstat.tcps_rcvpartduppack++;
1451 			tcpstat.tcps_rcvpartdupbyte += todrop;
1452 		}
1453 		hdroptlen += todrop;	/* drop from head afterwards */
1454 		th->th_seq += todrop;
1455 		tlen -= todrop;
1456 		if (th->th_urp > todrop)
1457 			th->th_urp -= todrop;
1458 		else {
1459 			tiflags &= ~TH_URG;
1460 			th->th_urp = 0;
1461 		}
1462 	}
1463 
1464 	/*
1465 	 * If new data are received on a connection after the
1466 	 * user processes are gone, then RST the other end.
1467 	 */
1468 	if ((so->so_state & SS_NOFDREF) &&
1469 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1470 		tp = tcp_close(tp);
1471 		tcpstat.tcps_rcvafterclose++;
1472 		goto dropwithreset;
1473 	}
1474 
1475 	/*
1476 	 * If segment ends after window, drop trailing data
1477 	 * (and PUSH and FIN); if nothing left, just ACK.
1478 	 */
1479 	todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1480 	if (todrop > 0) {
1481 		tcpstat.tcps_rcvpackafterwin++;
1482 		if (todrop >= tlen) {
1483 			tcpstat.tcps_rcvbyteafterwin += tlen;
1484 			/*
1485 			 * If a new connection request is received
1486 			 * while in TIME_WAIT, drop the old connection
1487 			 * and start over if the sequence numbers
1488 			 * are above the previous ones.
1489 			 */
1490 			if (tiflags & TH_SYN &&
1491 			    tp->t_state == TCPS_TIME_WAIT &&
1492 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1493 				iss = tp->snd_nxt + TCP_ISSINCR;
1494 				tp = tcp_close(tp);
1495 				goto findpcb;
1496 			}
1497 			/*
1498 			 * If window is closed can only take segments at
1499 			 * window edge, and have to drop data and PUSH from
1500 			 * incoming segments.  Continue processing, but
1501 			 * remember to ack.  Otherwise, drop segment
1502 			 * and ack.
1503 			 */
1504 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1505 				tp->t_flags |= TF_ACKNOW;
1506 				tcpstat.tcps_rcvwinprobe++;
1507 			} else
1508 				goto dropafterack;
1509 		} else
1510 			tcpstat.tcps_rcvbyteafterwin += todrop;
1511 		m_adj(m, -todrop);
1512 		tlen -= todrop;
1513 		tiflags &= ~(TH_PUSH|TH_FIN);
1514 	}
1515 
1516 	/*
1517 	 * If last ACK falls within this segment's sequence numbers,
1518 	 * record its timestamp.
1519 	 * Fix from Braden, see Stevens p. 870
1520 	 */
1521 	if (ts_present && TSTMP_GEQ(ts_val, tp->ts_recent) &&
1522 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1523 		tp->ts_recent_age = tcp_now;
1524 		tp->ts_recent = ts_val;
1525 	}
1526 
1527 	/*
1528 	 * If the RST bit is set examine the state:
1529 	 *    SYN_RECEIVED STATE:
1530 	 *	If passive open, return to LISTEN state.
1531 	 *	If active open, inform user that connection was refused.
1532 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1533 	 *	Inform user that connection was reset, and close tcb.
1534 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1535 	 *	Close the tcb.
1536 	 */
1537 	if (tiflags & TH_RST) {
1538 		if (th->th_seq != tp->last_ack_sent)
1539 			goto drop;
1540 
1541 		switch (tp->t_state) {
1542 		case TCPS_SYN_RECEIVED:
1543 #ifdef TCP_ECN
1544 			/* if ECN is enabled, fall back to non-ecn at rexmit */
1545 			if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
1546 				goto drop;
1547 #endif
1548 			so->so_error = ECONNREFUSED;
1549 			goto close;
1550 
1551 		case TCPS_ESTABLISHED:
1552 		case TCPS_FIN_WAIT_1:
1553 		case TCPS_FIN_WAIT_2:
1554 		case TCPS_CLOSE_WAIT:
1555 			so->so_error = ECONNRESET;
1556 		close:
1557 			tp->t_state = TCPS_CLOSED;
1558 			tcpstat.tcps_drops++;
1559 			tp = tcp_close(tp);
1560 			goto drop;
1561 		case TCPS_CLOSING:
1562 		case TCPS_LAST_ACK:
1563 		case TCPS_TIME_WAIT:
1564 			tp = tcp_close(tp);
1565 			goto drop;
1566 		}
1567 	}
1568 
1569 	/*
1570 	 * If a SYN is in the window, then this is an
1571 	 * error and we send an RST and drop the connection.
1572 	 */
1573 	if (tiflags & TH_SYN) {
1574 		tp = tcp_drop(tp, ECONNRESET);
1575 		goto dropwithreset;
1576 	}
1577 
1578 	/*
1579 	 * If the ACK bit is off we drop the segment and return.
1580 	 */
1581 	if ((tiflags & TH_ACK) == 0) {
1582 		if (tp->t_flags & TF_ACKNOW)
1583 			goto dropafterack;
1584 		else
1585 			goto drop;
1586 	}
1587 
1588 	/*
1589 	 * Ack processing.
1590 	 */
1591 	switch (tp->t_state) {
1592 
1593 	/*
1594 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1595 	 * ESTABLISHED state and continue processing.
1596 	 * The ACK was checked above.
1597 	 */
1598 	case TCPS_SYN_RECEIVED:
1599 		tcpstat.tcps_connects++;
1600 		soisconnected(so);
1601 		tp->t_state = TCPS_ESTABLISHED;
1602 		/* Do window scaling? */
1603 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1604 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1605 			tp->snd_scale = tp->requested_s_scale;
1606 			tp->rcv_scale = tp->request_r_scale;
1607 		}
1608 		(void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0,
1609 				 &tlen);
1610 		tp->snd_wl1 = th->th_seq - 1;
1611 		/* fall into ... */
1612 
1613 	/*
1614 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1615 	 * ACKs.  If the ack is in the range
1616 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1617 	 * then advance tp->snd_una to th->th_ack and drop
1618 	 * data from the retransmission queue.  If this ACK reflects
1619 	 * more up to date window information we update our window information.
1620 	 */
1621 	case TCPS_ESTABLISHED:
1622 	case TCPS_FIN_WAIT_1:
1623 	case TCPS_FIN_WAIT_2:
1624 	case TCPS_CLOSE_WAIT:
1625 	case TCPS_CLOSING:
1626 	case TCPS_LAST_ACK:
1627 	case TCPS_TIME_WAIT:
1628 #ifdef TCP_ECN
1629 		/*
1630 		 * if we receive ECE and are not already in recovery phase,
1631 		 * reduce cwnd by half but don't slow-start.
1632 		 * advance snd_last to snd_max not to reduce cwnd again
1633 		 * until all outstanding packets are acked.
1634 		 */
1635 		if (tcp_do_ecn && (tiflags & TH_ECE)) {
1636 			if ((tp->t_flags & TF_ECN_PERMIT) &&
1637 			    SEQ_GEQ(tp->snd_una, tp->snd_last)) {
1638 				u_int win;
1639 
1640 				win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg;
1641 				if (win > 1) {
1642 					tp->snd_ssthresh = win / 2 * tp->t_maxseg;
1643 					tp->snd_cwnd = tp->snd_ssthresh;
1644 					tp->snd_last = tp->snd_max;
1645 					tp->t_flags |= TF_SEND_CWR;
1646 					tcpstat.tcps_cwr_ecn++;
1647 				}
1648 			}
1649 			tcpstat.tcps_ecn_rcvece++;
1650 		}
1651 		/*
1652 		 * if we receive CWR, we know that the peer has reduced
1653 		 * its congestion window.  stop sending ecn-echo.
1654 		 */
1655 		if ((tiflags & TH_CWR)) {
1656 			tp->t_flags &= ~TF_RCVD_CE;
1657 			tcpstat.tcps_ecn_rcvcwr++;
1658 		}
1659 #endif /* TCP_ECN */
1660 
1661 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1662 			/*
1663 			 * Duplicate/old ACK processing.
1664 			 * Increments t_dupacks:
1665 			 *	Pure duplicate (same seq/ack/window, no data)
1666 			 * Doesn't affect t_dupacks:
1667 			 *	Data packets.
1668 			 *	Normal window updates (window opens)
1669 			 * Resets t_dupacks:
1670 			 *	New data ACKed.
1671 			 *	Window shrinks
1672 			 *	Old ACK
1673 			 */
1674 			if (tlen)
1675 				break;
1676 			/*
1677 			 * If we get an old ACK, there is probably packet
1678 			 * reordering going on.  Be conservative and reset
1679 			 * t_dupacks so that we are less agressive in
1680 			 * doing a fast retransmit.
1681 			 */
1682 			if (th->th_ack != tp->snd_una) {
1683 				tp->t_dupacks = 0;
1684 				break;
1685 			}
1686 			if (tiwin == tp->snd_wnd) {
1687 				tcpstat.tcps_rcvdupack++;
1688 				/*
1689 				 * If we have outstanding data (other than
1690 				 * a window probe), this is a completely
1691 				 * duplicate ack (ie, window info didn't
1692 				 * change), the ack is the biggest we've
1693 				 * seen and we've seen exactly our rexmt
1694 				 * threshhold of them, assume a packet
1695 				 * has been dropped and retransmit it.
1696 				 * Kludge snd_nxt & the congestion
1697 				 * window so we send only this one
1698 				 * packet.
1699 				 *
1700 				 * We know we're losing at the current
1701 				 * window size so do congestion avoidance
1702 				 * (set ssthresh to half the current window
1703 				 * and pull our congestion window back to
1704 				 * the new ssthresh).
1705 				 *
1706 				 * Dup acks mean that packets have left the
1707 				 * network (they're now cached at the receiver)
1708 				 * so bump cwnd by the amount in the receiver
1709 				 * to keep a constant cwnd packets in the
1710 				 * network.
1711 				 */
1712 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0)
1713 					tp->t_dupacks = 0;
1714 #if defined(TCP_SACK) && defined(TCP_FACK)
1715 				/*
1716 				 * In FACK, can enter fast rec. if the receiver
1717 				 * reports a reass. queue longer than 3 segs.
1718 				 */
1719 				else if (++tp->t_dupacks == tcprexmtthresh ||
1720 				    ((SEQ_GT(tp->snd_fack, tcprexmtthresh *
1721 				    tp->t_maxseg + tp->snd_una)) &&
1722 				    SEQ_GT(tp->snd_una, tp->snd_last))) {
1723 #else
1724 				else if (++tp->t_dupacks == tcprexmtthresh) {
1725 #endif /* TCP_FACK */
1726 					tcp_seq onxt = tp->snd_nxt;
1727 					u_long win =
1728 					    ulmin(tp->snd_wnd, tp->snd_cwnd) /
1729 						2 / tp->t_maxseg;
1730 
1731 #if defined(TCP_SACK) || defined(TCP_ECN)
1732 					if (SEQ_LT(th->th_ack, tp->snd_last)){
1733 					    	/*
1734 						 * False fast retx after
1735 						 * timeout.  Do not cut window.
1736 						 */
1737 						tp->t_dupacks = 0;
1738 						goto drop;
1739 					}
1740 #endif
1741 					if (win < 2)
1742 						win = 2;
1743 					tp->snd_ssthresh = win * tp->t_maxseg;
1744 #if defined(TCP_SACK)
1745 					tp->snd_last = tp->snd_max;
1746 #endif
1747 #ifdef TCP_SACK
1748                     			if (!tp->sack_disable) {
1749 						TCP_TIMER_DISARM(tp, TCPT_REXMT);
1750 						tp->t_rtttime = 0;
1751 #ifdef TCP_ECN
1752 						tp->t_flags |= TF_SEND_CWR;
1753 #endif
1754 #if 1 /* TCP_ECN */
1755 						tcpstat.tcps_cwr_frecovery++;
1756 #endif
1757 						tcpstat.tcps_sndrexmitfast++;
1758 #if defined(TCP_SACK) && defined(TCP_FACK)
1759 						tp->t_dupacks = tcprexmtthresh;
1760 						(void) tcp_output(tp);
1761 						/*
1762 						 * During FR, snd_cwnd is held
1763 						 * constant for FACK.
1764 						 */
1765 						tp->snd_cwnd = tp->snd_ssthresh;
1766 #else
1767 						/*
1768 						 * tcp_output() will send
1769 						 * oldest SACK-eligible rtx.
1770 						 */
1771 						(void) tcp_output(tp);
1772 						tp->snd_cwnd = tp->snd_ssthresh+
1773 					           tp->t_maxseg * tp->t_dupacks;
1774 #endif /* TCP_FACK */
1775 						goto drop;
1776 					}
1777 #endif /* TCP_SACK */
1778 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1779 					tp->t_rtttime = 0;
1780 					tp->snd_nxt = th->th_ack;
1781 					tp->snd_cwnd = tp->t_maxseg;
1782 #ifdef TCP_ECN
1783 					tp->t_flags |= TF_SEND_CWR;
1784 #endif
1785 #if 1 /* TCP_ECN */
1786 					tcpstat.tcps_cwr_frecovery++;
1787 #endif
1788 					tcpstat.tcps_sndrexmitfast++;
1789 					(void) tcp_output(tp);
1790 
1791 					tp->snd_cwnd = tp->snd_ssthresh +
1792 					    tp->t_maxseg * tp->t_dupacks;
1793 					if (SEQ_GT(onxt, tp->snd_nxt))
1794 						tp->snd_nxt = onxt;
1795 					goto drop;
1796 				} else if (tp->t_dupacks > tcprexmtthresh) {
1797 #if defined(TCP_SACK) && defined(TCP_FACK)
1798 					/*
1799 					 * while (awnd < cwnd)
1800 					 *         sendsomething();
1801 					 */
1802 					if (!tp->sack_disable) {
1803 						if (tp->snd_awnd < tp->snd_cwnd)
1804 							tcp_output(tp);
1805 						goto drop;
1806 					}
1807 #endif /* TCP_FACK */
1808 					tp->snd_cwnd += tp->t_maxseg;
1809 					(void) tcp_output(tp);
1810 					goto drop;
1811 				}
1812 			} else if (tiwin < tp->snd_wnd) {
1813 				/*
1814 				 * The window was retracted!  Previous dup
1815 				 * ACKs may have been due to packets arriving
1816 				 * after the shrunken window, not a missing
1817 				 * packet, so play it safe and reset t_dupacks
1818 				 */
1819 				tp->t_dupacks = 0;
1820 			}
1821 			break;
1822 		}
1823 		/*
1824 		 * If the congestion window was inflated to account
1825 		 * for the other side's cached packets, retract it.
1826 		 */
1827 #if defined(TCP_SACK)
1828 		if (!tp->sack_disable) {
1829 			if (tp->t_dupacks >= tcprexmtthresh) {
1830 				/* Check for a partial ACK */
1831 				if (tcp_sack_partialack(tp, th)) {
1832 #if defined(TCP_SACK) && defined(TCP_FACK)
1833 					/* Force call to tcp_output */
1834 					if (tp->snd_awnd < tp->snd_cwnd)
1835 						needoutput = 1;
1836 #else
1837 					tp->snd_cwnd += tp->t_maxseg;
1838 					needoutput = 1;
1839 #endif /* TCP_FACK */
1840 				} else {
1841 					/* Out of fast recovery */
1842 					tp->snd_cwnd = tp->snd_ssthresh;
1843 					if (tcp_seq_subtract(tp->snd_max,
1844 					    th->th_ack) < tp->snd_ssthresh)
1845 						tp->snd_cwnd =
1846 						   tcp_seq_subtract(tp->snd_max,
1847 					           th->th_ack);
1848 					tp->t_dupacks = 0;
1849 #if defined(TCP_SACK) && defined(TCP_FACK)
1850 					if (SEQ_GT(th->th_ack, tp->snd_fack))
1851 						tp->snd_fack = th->th_ack;
1852 #endif /* TCP_FACK */
1853 				}
1854 			}
1855 		} else {
1856 			if (tp->t_dupacks >= tcprexmtthresh &&
1857 			    !tcp_newreno(tp, th)) {
1858 				/* Out of fast recovery */
1859 				tp->snd_cwnd = tp->snd_ssthresh;
1860 				if (tcp_seq_subtract(tp->snd_max, th->th_ack) <
1861 			  	    tp->snd_ssthresh)
1862 					tp->snd_cwnd =
1863 					    tcp_seq_subtract(tp->snd_max,
1864 					    th->th_ack);
1865 				tp->t_dupacks = 0;
1866 			}
1867 		}
1868 		if (tp->t_dupacks < tcprexmtthresh)
1869 			tp->t_dupacks = 0;
1870 #else /* else no TCP_SACK */
1871 		if (tp->t_dupacks >= tcprexmtthresh &&
1872 		    tp->snd_cwnd > tp->snd_ssthresh)
1873 			tp->snd_cwnd = tp->snd_ssthresh;
1874 		tp->t_dupacks = 0;
1875 #endif
1876 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1877 			tcpstat.tcps_rcvacktoomuch++;
1878 			goto dropafterack;
1879 		}
1880 		acked = th->th_ack - tp->snd_una;
1881 		tcpstat.tcps_rcvackpack++;
1882 		tcpstat.tcps_rcvackbyte += acked;
1883 
1884 		/*
1885 		 * If we have a timestamp reply, update smoothed
1886 		 * round trip time.  If no timestamp is present but
1887 		 * transmit timer is running and timed sequence
1888 		 * number was acked, update smoothed round trip time.
1889 		 * Since we now have an rtt measurement, cancel the
1890 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1891 		 * Recompute the initial retransmit timer.
1892 		 */
1893 		if (ts_present)
1894 			tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1895 		else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
1896 			tcp_xmit_timer(tp, tcp_now - tp->t_rtttime);
1897 
1898 		/*
1899 		 * If all outstanding data is acked, stop retransmit
1900 		 * timer and remember to restart (more output or persist).
1901 		 * If there is more data to be acked, restart retransmit
1902 		 * timer, using current (possibly backed-off) value.
1903 		 */
1904 		if (th->th_ack == tp->snd_max) {
1905 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1906 			needoutput = 1;
1907 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1908 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1909 		/*
1910 		 * When new data is acked, open the congestion window.
1911 		 * If the window gives us less than ssthresh packets
1912 		 * in flight, open exponentially (maxseg per packet).
1913 		 * Otherwise open linearly: maxseg per window
1914 		 * (maxseg^2 / cwnd per packet).
1915 		 */
1916 		{
1917 		u_int cw = tp->snd_cwnd;
1918 		u_int incr = tp->t_maxseg;
1919 
1920 		if (cw > tp->snd_ssthresh)
1921 			incr = incr * incr / cw;
1922 #if defined (TCP_SACK)
1923 		if (tp->t_dupacks < tcprexmtthresh)
1924 #endif
1925 		tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1926 		}
1927 		ND6_HINT(tp);
1928 		if (acked > so->so_snd.sb_cc) {
1929 			tp->snd_wnd -= so->so_snd.sb_cc;
1930 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1931 			ourfinisacked = 1;
1932 		} else {
1933 			sbdrop(&so->so_snd, acked);
1934 			tp->snd_wnd -= acked;
1935 			ourfinisacked = 0;
1936 		}
1937 		if (sb_notify(&so->so_snd))
1938 			sowwakeup(so);
1939 		tp->snd_una = th->th_ack;
1940 #ifdef TCP_ECN
1941 		/* sync snd_last with snd_una */
1942 		if (SEQ_GT(tp->snd_una, tp->snd_last))
1943 			tp->snd_last = tp->snd_una;
1944 #endif
1945 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1946 			tp->snd_nxt = tp->snd_una;
1947 #if defined (TCP_SACK) && defined (TCP_FACK)
1948 		if (SEQ_GT(tp->snd_una, tp->snd_fack)) {
1949 			tp->snd_fack = tp->snd_una;
1950 			/* Update snd_awnd for partial ACK
1951 			 * without any SACK blocks.
1952 			 */
1953 			tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt,
1954 				tp->snd_fack) + tp->retran_data;
1955 		}
1956 #endif
1957 
1958 		switch (tp->t_state) {
1959 
1960 		/*
1961 		 * In FIN_WAIT_1 STATE in addition to the processing
1962 		 * for the ESTABLISHED state if our FIN is now acknowledged
1963 		 * then enter FIN_WAIT_2.
1964 		 */
1965 		case TCPS_FIN_WAIT_1:
1966 			if (ourfinisacked) {
1967 				/*
1968 				 * If we can't receive any more
1969 				 * data, then closing user can proceed.
1970 				 * Starting the timer is contrary to the
1971 				 * specification, but if we don't get a FIN
1972 				 * we'll hang forever.
1973 				 */
1974 				if (so->so_state & SS_CANTRCVMORE) {
1975 					soisdisconnected(so);
1976 					TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle);
1977 				}
1978 				tp->t_state = TCPS_FIN_WAIT_2;
1979 			}
1980 			break;
1981 
1982 		/*
1983 		 * In CLOSING STATE in addition to the processing for
1984 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1985 		 * then enter the TIME-WAIT state, otherwise ignore
1986 		 * the segment.
1987 		 */
1988 		case TCPS_CLOSING:
1989 			if (ourfinisacked) {
1990 				tp->t_state = TCPS_TIME_WAIT;
1991 				tcp_canceltimers(tp);
1992 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1993 				soisdisconnected(so);
1994 			}
1995 			break;
1996 
1997 		/*
1998 		 * In LAST_ACK, we may still be waiting for data to drain
1999 		 * and/or to be acked, as well as for the ack of our FIN.
2000 		 * If our FIN is now acknowledged, delete the TCB,
2001 		 * enter the closed state and return.
2002 		 */
2003 		case TCPS_LAST_ACK:
2004 			if (ourfinisacked) {
2005 				tp = tcp_close(tp);
2006 				goto drop;
2007 			}
2008 			break;
2009 
2010 		/*
2011 		 * In TIME_WAIT state the only thing that should arrive
2012 		 * is a retransmission of the remote FIN.  Acknowledge
2013 		 * it and restart the finack timer.
2014 		 */
2015 		case TCPS_TIME_WAIT:
2016 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2017 			goto dropafterack;
2018 		}
2019 	}
2020 
2021 step6:
2022 	/*
2023 	 * Update window information.
2024 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2025 	 */
2026 	if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2027 	    (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) ||
2028 	    (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) {
2029 		/* keep track of pure window updates */
2030 		if (tlen == 0 &&
2031 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2032 			tcpstat.tcps_rcvwinupd++;
2033 		tp->snd_wnd = tiwin;
2034 		tp->snd_wl1 = th->th_seq;
2035 		tp->snd_wl2 = th->th_ack;
2036 		if (tp->snd_wnd > tp->max_sndwnd)
2037 			tp->max_sndwnd = tp->snd_wnd;
2038 		needoutput = 1;
2039 	}
2040 
2041 	/*
2042 	 * Process segments with URG.
2043 	 */
2044 	if ((tiflags & TH_URG) && th->th_urp &&
2045 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2046 		/*
2047 		 * This is a kludge, but if we receive and accept
2048 		 * random urgent pointers, we'll crash in
2049 		 * soreceive.  It's hard to imagine someone
2050 		 * actually wanting to send this much urgent data.
2051 		 */
2052 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2053 			th->th_urp = 0;			/* XXX */
2054 			tiflags &= ~TH_URG;		/* XXX */
2055 			goto dodata;			/* XXX */
2056 		}
2057 		/*
2058 		 * If this segment advances the known urgent pointer,
2059 		 * then mark the data stream.  This should not happen
2060 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2061 		 * a FIN has been received from the remote side.
2062 		 * In these states we ignore the URG.
2063 		 *
2064 		 * According to RFC961 (Assigned Protocols),
2065 		 * the urgent pointer points to the last octet
2066 		 * of urgent data.  We continue, however,
2067 		 * to consider it to indicate the first octet
2068 		 * of data past the urgent section as the original
2069 		 * spec states (in one of two places).
2070 		 */
2071 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2072 			tp->rcv_up = th->th_seq + th->th_urp;
2073 			so->so_oobmark = so->so_rcv.sb_cc +
2074 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2075 			if (so->so_oobmark == 0)
2076 				so->so_state |= SS_RCVATMARK;
2077 			sohasoutofband(so);
2078 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2079 		}
2080 		/*
2081 		 * Remove out of band data so doesn't get presented to user.
2082 		 * This can happen independent of advancing the URG pointer,
2083 		 * but if two URG's are pending at once, some out-of-band
2084 		 * data may creep in... ick.
2085 		 */
2086 		if (th->th_urp <= (u_int16_t) tlen
2087 #ifdef SO_OOBINLINE
2088 		     && (so->so_options & SO_OOBINLINE) == 0
2089 #endif
2090 		     )
2091 		        tcp_pulloutofband(so, th->th_urp, m, hdroptlen);
2092 	} else
2093 		/*
2094 		 * If no out of band data is expected,
2095 		 * pull receive urgent pointer along
2096 		 * with the receive window.
2097 		 */
2098 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2099 			tp->rcv_up = tp->rcv_nxt;
2100 dodata:							/* XXX */
2101 
2102 	/*
2103 	 * Process the segment text, merging it into the TCP sequencing queue,
2104 	 * and arranging for acknowledgment of receipt if necessary.
2105 	 * This process logically involves adjusting tp->rcv_wnd as data
2106 	 * is presented to the user (this happens in tcp_usrreq.c,
2107 	 * case PRU_RCVD).  If a FIN has already been received on this
2108 	 * connection then we just ignore the text.
2109 	 */
2110 	if ((tlen || (tiflags & TH_FIN)) &&
2111 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2112 		if (th->th_seq == tp->rcv_nxt && tp->segq.lh_first == NULL &&
2113 		    tp->t_state == TCPS_ESTABLISHED) {
2114 			TCP_SETUP_ACK(tp, tiflags);
2115 			tp->rcv_nxt += tlen;
2116 			tiflags = th->th_flags & TH_FIN;
2117 			tcpstat.tcps_rcvpack++;
2118 			tcpstat.tcps_rcvbyte += tlen;
2119 			ND6_HINT(tp);
2120 			m_adj(m, hdroptlen);
2121 			sbappend(&so->so_rcv, m);
2122 			sorwakeup(so);
2123 		} else {
2124 			m_adj(m, hdroptlen);
2125 			tiflags = tcp_reass(tp, th, m, &tlen);
2126 			tp->t_flags |= TF_ACKNOW;
2127 		}
2128 #ifdef TCP_SACK
2129 		if (!tp->sack_disable)
2130 			tcp_update_sack_list(tp);
2131 #endif
2132 
2133 		/*
2134 		 * variable len never referenced again in modern BSD,
2135 		 * so why bother computing it ??
2136 		 */
2137 #if 0
2138 		/*
2139 		 * Note the amount of data that peer has sent into
2140 		 * our window, in order to estimate the sender's
2141 		 * buffer size.
2142 		 */
2143 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2144 #endif /* 0 */
2145 	} else {
2146 		m_freem(m);
2147 		tiflags &= ~TH_FIN;
2148 	}
2149 
2150 	/*
2151 	 * If FIN is received ACK the FIN and let the user know
2152 	 * that the connection is closing.  Ignore a FIN received before
2153 	 * the connection is fully established.
2154 	 */
2155 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
2156 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2157 			socantrcvmore(so);
2158 			tp->t_flags |= TF_ACKNOW;
2159 			tp->rcv_nxt++;
2160 		}
2161 		switch (tp->t_state) {
2162 
2163 		/*
2164 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
2165 		 */
2166 		case TCPS_ESTABLISHED:
2167 			tp->t_state = TCPS_CLOSE_WAIT;
2168 			break;
2169 
2170 		/*
2171 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2172 		 * enter the CLOSING state.
2173 		 */
2174 		case TCPS_FIN_WAIT_1:
2175 			tp->t_state = TCPS_CLOSING;
2176 			break;
2177 
2178 		/*
2179 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2180 		 * starting the time-wait timer, turning off the other
2181 		 * standard timers.
2182 		 */
2183 		case TCPS_FIN_WAIT_2:
2184 			tp->t_state = TCPS_TIME_WAIT;
2185 			tcp_canceltimers(tp);
2186 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2187 			soisdisconnected(so);
2188 			break;
2189 
2190 		/*
2191 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2192 		 */
2193 		case TCPS_TIME_WAIT:
2194 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
2195 			break;
2196 		}
2197 	}
2198 	if (so->so_options & SO_DEBUG) {
2199 		switch (tp->pf == PF_INET6) {
2200 #ifdef INET6
2201 		case PF_INET6:
2202 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6,
2203 			    0, tlen);
2204 			break;
2205 #endif /* INET6 */
2206 		case PF_INET:
2207 			tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti,
2208 			    0, tlen);
2209 			break;
2210 		}
2211 	}
2212 
2213 	/*
2214 	 * Return any desired output.
2215 	 */
2216 	if (needoutput || (tp->t_flags & TF_ACKNOW)) {
2217 		(void) tcp_output(tp);
2218 	}
2219 	return;
2220 
2221 dropafterack:
2222 	/*
2223 	 * Generate an ACK dropping incoming segment if it occupies
2224 	 * sequence space, where the ACK reflects our state.
2225 	 */
2226 	if (tiflags & TH_RST)
2227 		goto drop;
2228 	m_freem(m);
2229 	tp->t_flags |= TF_ACKNOW;
2230 	(void) tcp_output(tp);
2231 	return;
2232 
2233 dropwithreset_ratelim:
2234 	/*
2235 	 * We may want to rate-limit RSTs in certain situations,
2236 	 * particularly if we are sending an RST in response to
2237 	 * an attempt to connect to or otherwise communicate with
2238 	 * a port for which we have no socket.
2239 	 */
2240 	if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count,
2241 	    tcp_rst_ppslim) == 0) {
2242 		/* XXX stat */
2243 		goto drop;
2244 	}
2245 	/* ...fall into dropwithreset... */
2246 
2247 dropwithreset:
2248 	/*
2249 	 * Generate a RST, dropping incoming segment.
2250 	 * Make ACK acceptable to originator of segment.
2251 	 * Don't bother to respond if destination was broadcast/multicast.
2252 	 */
2253 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2254 		goto drop;
2255 	switch (af) {
2256 #ifdef INET6
2257 	case AF_INET6:
2258 		/* For following calls to tcp_respond */
2259 		if (IN6_IS_ADDR_MULTICAST(&ipv6->ip6_dst))
2260 			goto drop;
2261 		break;
2262 #endif /* INET6 */
2263 	case AF_INET:
2264 		if (IN_MULTICAST(ip->ip_dst.s_addr) ||
2265 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2266 			goto drop;
2267 	}
2268 	if (tiflags & TH_ACK) {
2269 		tcp_respond(tp, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack,
2270 		    TH_RST);
2271 	} else {
2272 		if (tiflags & TH_SYN)
2273 			tlen++;
2274 		tcp_respond(tp, mtod(m, caddr_t), m, th->th_seq + tlen,
2275 		    (tcp_seq)0, TH_RST|TH_ACK);
2276 	}
2277 	/* destroy temporarily created socket */
2278 	if (dropsocket)
2279 		(void) soabort(so);
2280 	return;
2281 
2282 drop:
2283 	/*
2284 	 * Drop space held by incoming segment and return.
2285 	 */
2286 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
2287 		switch (tp->pf) {
2288 #ifdef INET6
2289 		case PF_INET6:
2290 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6,
2291 			    0, tlen);
2292 			break;
2293 #endif /* INET6 */
2294 		case PF_INET:
2295 			tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti,
2296 			    0, tlen);
2297 			break;
2298 		}
2299 	}
2300 
2301 	m_freem(m);
2302 	/* destroy temporarily created socket */
2303 	if (dropsocket)
2304 		(void) soabort(so);
2305 	return;
2306 #ifndef TUBA_INCLUDE
2307 }
2308 
2309 void
2310 tcp_dooptions(tp, cp, cnt, th, ts_present, ts_val, ts_ecr)
2311 	struct tcpcb *tp;
2312 	u_char *cp;
2313 	int cnt;
2314 	struct tcphdr *th;
2315 	int *ts_present;
2316 	u_int32_t *ts_val, *ts_ecr;
2317 {
2318 	u_int16_t mss = 0;
2319 	int opt, optlen;
2320 
2321 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2322 		opt = cp[0];
2323 		if (opt == TCPOPT_EOL)
2324 			break;
2325 		if (opt == TCPOPT_NOP)
2326 			optlen = 1;
2327 		else {
2328 			if (cnt < 2)
2329 				break;
2330 			optlen = cp[1];
2331 			if (optlen < 2 || optlen > cnt)
2332 				break;
2333 		}
2334 		switch (opt) {
2335 
2336 		default:
2337 			continue;
2338 
2339 		case TCPOPT_MAXSEG:
2340 			if (optlen != TCPOLEN_MAXSEG)
2341 				continue;
2342 			if (!(th->th_flags & TH_SYN))
2343 				continue;
2344 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2345 			NTOHS(mss);
2346 			break;
2347 
2348 		case TCPOPT_WINDOW:
2349 			if (optlen != TCPOLEN_WINDOW)
2350 				continue;
2351 			if (!(th->th_flags & TH_SYN))
2352 				continue;
2353 			tp->t_flags |= TF_RCVD_SCALE;
2354 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2355 			break;
2356 
2357 		case TCPOPT_TIMESTAMP:
2358 			if (optlen != TCPOLEN_TIMESTAMP)
2359 				continue;
2360 			*ts_present = 1;
2361 			bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val));
2362 			NTOHL(*ts_val);
2363 			bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr));
2364 			NTOHL(*ts_ecr);
2365 
2366 			/*
2367 			 * A timestamp received in a SYN makes
2368 			 * it ok to send timestamp requests and replies.
2369 			 */
2370 			if (th->th_flags & TH_SYN) {
2371 				tp->t_flags |= TF_RCVD_TSTMP;
2372 				tp->ts_recent = *ts_val;
2373 				tp->ts_recent_age = tcp_now;
2374 			}
2375 			break;
2376 
2377 #ifdef TCP_SACK
2378 		case TCPOPT_SACK_PERMITTED:
2379 			if (tp->sack_disable || optlen!=TCPOLEN_SACK_PERMITTED)
2380 				continue;
2381 			if (th->th_flags & TH_SYN)
2382 				/* MUST only be set on SYN */
2383 				tp->t_flags |= TF_SACK_PERMIT;
2384 			break;
2385 		case TCPOPT_SACK:
2386 			if (tcp_sack_option(tp, th, cp, optlen))
2387 				continue;
2388 			break;
2389 #endif
2390 		}
2391 	}
2392 	/* Update t_maxopd and t_maxseg after all options are processed */
2393 	if (th->th_flags & TH_SYN) {
2394 		(void) tcp_mss(tp, mss);	/* sets t_maxseg */
2395 
2396 		if (mss)
2397 			tcp_mss_update(tp);
2398 	}
2399 }
2400 
2401 #if defined(TCP_SACK)
2402 u_long
2403 tcp_seq_subtract(a, b)
2404 	u_long a, b;
2405 {
2406 	return ((long)(a - b));
2407 }
2408 #endif
2409 
2410 
2411 #ifdef TCP_SACK
2412 /*
2413  * This function is called upon receipt of new valid data (while not in header
2414  * prediction mode), and it updates the ordered list of sacks.
2415  */
2416 void
2417 tcp_update_sack_list(tp)
2418 	struct tcpcb *tp;
2419 {
2420 	/*
2421 	 * First reported block MUST be the most recent one.  Subsequent
2422 	 * blocks SHOULD be in the order in which they arrived at the
2423 	 * receiver.  These two conditions make the implementation fully
2424 	 * compliant with RFC 2018.
2425 	 */
2426 	int i, j = 0, count = 0, lastpos = -1;
2427 	struct sackblk sack, firstsack, temp[MAX_SACK_BLKS];
2428 
2429 	/* First clean up current list of sacks */
2430 	for (i = 0; i < tp->rcv_numsacks; i++) {
2431 		sack = tp->sackblks[i];
2432 		if (sack.start == 0 && sack.end == 0) {
2433 			count++; /* count = number of blocks to be discarded */
2434 			continue;
2435 		}
2436 		if (SEQ_LEQ(sack.end, tp->rcv_nxt)) {
2437 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2438 			count++;
2439 		} else {
2440 			temp[j].start = tp->sackblks[i].start;
2441 			temp[j++].end = tp->sackblks[i].end;
2442 		}
2443 	}
2444 	tp->rcv_numsacks -= count;
2445 	if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */
2446 		tcp_clean_sackreport(tp);
2447 		if (SEQ_LT(tp->rcv_nxt, tp->rcv_laststart)) {
2448 			/* ==> need first sack block */
2449 			tp->sackblks[0].start = tp->rcv_laststart;
2450 			tp->sackblks[0].end = tp->rcv_lastend;
2451 			tp->rcv_numsacks = 1;
2452 		}
2453 		return;
2454 	}
2455 	/* Otherwise, sack blocks are already present. */
2456 	for (i = 0; i < tp->rcv_numsacks; i++)
2457 		tp->sackblks[i] = temp[i]; /* first copy back sack list */
2458 	if (SEQ_GEQ(tp->rcv_nxt, tp->rcv_lastend))
2459 		return;     /* sack list remains unchanged */
2460 	/*
2461 	 * From here, segment just received should be (part of) the 1st sack.
2462 	 * Go through list, possibly coalescing sack block entries.
2463 	 */
2464 	firstsack.start = tp->rcv_laststart;
2465 	firstsack.end = tp->rcv_lastend;
2466 	for (i = 0; i < tp->rcv_numsacks; i++) {
2467 		sack = tp->sackblks[i];
2468 		if (SEQ_LT(sack.end, firstsack.start) ||
2469 		    SEQ_GT(sack.start, firstsack.end))
2470 			continue; /* no overlap */
2471 		if (sack.start == firstsack.start && sack.end == firstsack.end){
2472 			/*
2473 			 * identical block; delete it here since we will
2474 			 * move it to the front of the list.
2475 			 */
2476 			tp->sackblks[i].start = tp->sackblks[i].end = 0;
2477 			lastpos = i;    /* last posn with a zero entry */
2478 			continue;
2479 		}
2480 		if (SEQ_LEQ(sack.start, firstsack.start))
2481 			firstsack.start = sack.start; /* merge blocks */
2482 		if (SEQ_GEQ(sack.end, firstsack.end))
2483 			firstsack.end = sack.end;     /* merge blocks */
2484 		tp->sackblks[i].start = tp->sackblks[i].end = 0;
2485 		lastpos = i;    /* last posn with a zero entry */
2486 	}
2487 	if (lastpos != -1) {    /* at least one merge */
2488 		for (i = 0, j = 1; i < tp->rcv_numsacks; i++) {
2489 			sack = tp->sackblks[i];
2490 			if (sack.start == 0 && sack.end == 0)
2491 				continue;
2492 			temp[j++] = sack;
2493 		}
2494 		tp->rcv_numsacks = j; /* including first blk (added later) */
2495 		for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */
2496 			tp->sackblks[i] = temp[i];
2497 	} else {        /* no merges -- shift sacks by 1 */
2498 		if (tp->rcv_numsacks < MAX_SACK_BLKS)
2499 			tp->rcv_numsacks++;
2500 		for (i = tp->rcv_numsacks-1; i > 0; i--)
2501 			tp->sackblks[i] = tp->sackblks[i-1];
2502 	}
2503 	tp->sackblks[0] = firstsack;
2504 	return;
2505 }
2506 
2507 /*
2508  * Process the TCP SACK option.  Returns 1 if tcp_dooptions() should continue,
2509  * and 0 otherwise, if the option was fine.  tp->snd_holes is an ordered list
2510  * of holes (oldest to newest, in terms of the sequence space).
2511  */
2512 int
2513 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen)
2514 {
2515 	int tmp_olen;
2516 	u_char *tmp_cp;
2517 	struct sackhole *cur, *p, *temp;
2518 
2519 	if (tp->sack_disable)
2520 		return (1);
2521 
2522 	/* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */
2523 	if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2524 		return (1);
2525 	tmp_cp = cp + 2;
2526 	tmp_olen = optlen - 2;
2527 	if (tp->snd_numholes < 0)
2528 		tp->snd_numholes = 0;
2529 	if (tp->t_maxseg == 0)
2530 		panic("tcp_sack_option"); /* Should never happen */
2531 	while (tmp_olen > 0) {
2532 		struct sackblk sack;
2533 
2534 		bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq));
2535 		NTOHL(sack.start);
2536 		bcopy(tmp_cp + sizeof(tcp_seq),
2537 		    (char *) &(sack.end), sizeof(tcp_seq));
2538 		NTOHL(sack.end);
2539 		tmp_olen -= TCPOLEN_SACK;
2540 		tmp_cp += TCPOLEN_SACK;
2541 		if (SEQ_LEQ(sack.end, sack.start))
2542 			continue; /* bad SACK fields */
2543 		if (SEQ_LEQ(sack.end, tp->snd_una))
2544 			continue; /* old block */
2545 #if defined(TCP_SACK) && defined(TCP_FACK)
2546 		/* Updates snd_fack.  */
2547 		if (SEQ_GT(sack.end, tp->snd_fack))
2548 			tp->snd_fack = sack.end;
2549 #endif /* TCP_FACK */
2550 		if (SEQ_GT(th->th_ack, tp->snd_una)) {
2551 			if (SEQ_LT(sack.start, th->th_ack))
2552 				continue;
2553 		}
2554 		if (SEQ_GT(sack.end, tp->snd_max))
2555 			continue;
2556 		if (tp->snd_holes == NULL) { /* first hole */
2557 			tp->snd_holes = (struct sackhole *)
2558 			    pool_get(&sackhl_pool, PR_NOWAIT);
2559 			if (tp->snd_holes == NULL) {
2560 				/* ENOBUFS, so ignore SACKed block for now*/
2561 				continue;
2562 			}
2563 			cur = tp->snd_holes;
2564 			cur->start = th->th_ack;
2565 			cur->end = sack.start;
2566 			cur->rxmit = cur->start;
2567 			cur->next = NULL;
2568 			tp->snd_numholes = 1;
2569 			tp->rcv_lastsack = sack.end;
2570 			/*
2571 			 * dups is at least one.  If more data has been
2572 			 * SACKed, it can be greater than one.
2573 			 */
2574 			cur->dups = min(tcprexmtthresh,
2575 			    ((sack.end - cur->end)/tp->t_maxseg));
2576 			if (cur->dups < 1)
2577 				cur->dups = 1;
2578 			continue; /* with next sack block */
2579 		}
2580 		/* Go thru list of holes:  p = previous,  cur = current */
2581 		p = cur = tp->snd_holes;
2582 		while (cur) {
2583 			if (SEQ_LEQ(sack.end, cur->start))
2584 				/* SACKs data before the current hole */
2585 				break; /* no use going through more holes */
2586 			if (SEQ_GEQ(sack.start, cur->end)) {
2587 				/* SACKs data beyond the current hole */
2588 				cur->dups++;
2589 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2590 				    tcprexmtthresh)
2591 					cur->dups = tcprexmtthresh;
2592 				p = cur;
2593 				cur = cur->next;
2594 				continue;
2595 			}
2596 			if (SEQ_LEQ(sack.start, cur->start)) {
2597 				/* Data acks at least the beginning of hole */
2598 #if defined(TCP_SACK) && defined(TCP_FACK)
2599 				if (SEQ_GT(sack.end, cur->rxmit))
2600 					tp->retran_data -=
2601 				    	    tcp_seq_subtract(cur->rxmit,
2602 					    cur->start);
2603 				else
2604 					tp->retran_data -=
2605 					    tcp_seq_subtract(sack.end,
2606 					    cur->start);
2607 #endif /* TCP_FACK */
2608 				if (SEQ_GEQ(sack.end, cur->end)) {
2609 					/* Acks entire hole, so delete hole */
2610 					if (p != cur) {
2611 						p->next = cur->next;
2612 						pool_put(&sackhl_pool, cur);
2613 						cur = p->next;
2614 					} else {
2615 						cur = cur->next;
2616 						pool_put(&sackhl_pool, p);
2617 						p = cur;
2618 						tp->snd_holes = p;
2619 					}
2620 					tp->snd_numholes--;
2621 					continue;
2622 				}
2623 				/* otherwise, move start of hole forward */
2624 				cur->start = sack.end;
2625 				cur->rxmit = max (cur->rxmit, cur->start);
2626 				p = cur;
2627 				cur = cur->next;
2628 				continue;
2629 			}
2630 			/* move end of hole backward */
2631 			if (SEQ_GEQ(sack.end, cur->end)) {
2632 #if defined(TCP_SACK) && defined(TCP_FACK)
2633 				if (SEQ_GT(cur->rxmit, sack.start))
2634 					tp->retran_data -=
2635 					    tcp_seq_subtract(cur->rxmit,
2636 					    sack.start);
2637 #endif /* TCP_FACK */
2638 				cur->end = sack.start;
2639 				cur->rxmit = min(cur->rxmit, cur->end);
2640 				cur->dups++;
2641 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2642 				    tcprexmtthresh)
2643 					cur->dups = tcprexmtthresh;
2644 				p = cur;
2645 				cur = cur->next;
2646 				continue;
2647 			}
2648 			if (SEQ_LT(cur->start, sack.start) &&
2649 			    SEQ_GT(cur->end, sack.end)) {
2650 				/*
2651 				 * ACKs some data in middle of a hole; need to
2652 				 * split current hole
2653 				 */
2654 				temp = (struct sackhole *)
2655 				    pool_get(&sackhl_pool, PR_NOWAIT);
2656 				if (temp == NULL)
2657 					continue; /* ENOBUFS */
2658 #if defined(TCP_SACK) && defined(TCP_FACK)
2659 				if (SEQ_GT(cur->rxmit, sack.end))
2660 					tp->retran_data -=
2661 					    tcp_seq_subtract(sack.end,
2662 					    sack.start);
2663 				else if (SEQ_GT(cur->rxmit, sack.start))
2664 					tp->retran_data -=
2665 					    tcp_seq_subtract(cur->rxmit,
2666 					    sack.start);
2667 #endif /* TCP_FACK */
2668 				temp->next = cur->next;
2669 				temp->start = sack.end;
2670 				temp->end = cur->end;
2671 				temp->dups = cur->dups;
2672 				temp->rxmit = max(cur->rxmit, temp->start);
2673 				cur->end = sack.start;
2674 				cur->rxmit = min(cur->rxmit, cur->end);
2675 				cur->dups++;
2676 				if (((sack.end - cur->end)/tp->t_maxseg) >=
2677 					tcprexmtthresh)
2678 					cur->dups = tcprexmtthresh;
2679 				cur->next = temp;
2680 				p = temp;
2681 				cur = p->next;
2682 				tp->snd_numholes++;
2683 			}
2684 		}
2685 		/* At this point, p points to the last hole on the list */
2686 		if (SEQ_LT(tp->rcv_lastsack, sack.start)) {
2687 			/*
2688 			 * Need to append new hole at end.
2689 			 * Last hole is p (and it's not NULL).
2690 			 */
2691 			temp = (struct sackhole *)
2692 			    pool_get(&sackhl_pool, PR_NOWAIT);
2693 			if (temp == NULL)
2694 				continue; /* ENOBUFS */
2695 			temp->start = tp->rcv_lastsack;
2696 			temp->end = sack.start;
2697 			temp->dups = min(tcprexmtthresh,
2698 			    ((sack.end - sack.start)/tp->t_maxseg));
2699 			if (temp->dups < 1)
2700 				temp->dups = 1;
2701 			temp->rxmit = temp->start;
2702 			temp->next = 0;
2703 			p->next = temp;
2704 			tp->rcv_lastsack = sack.end;
2705 			tp->snd_numholes++;
2706 		}
2707 	}
2708 #if defined(TCP_SACK) && defined(TCP_FACK)
2709 	/*
2710 	 * Update retran_data and snd_awnd.  Go through the list of
2711 	 * holes.   Increment retran_data by (hole->rxmit - hole->start).
2712 	 */
2713 	tp->retran_data = 0;
2714 	cur = tp->snd_holes;
2715 	while (cur) {
2716 		tp->retran_data += cur->rxmit - cur->start;
2717 		cur = cur->next;
2718 	}
2719 	tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) +
2720 	    tp->retran_data;
2721 #endif /* TCP_FACK */
2722 
2723 	return (0);
2724 }
2725 
2726 /*
2727  * Delete stale (i.e, cumulatively ack'd) holes.  Hole is deleted only if
2728  * it is completely acked; otherwise, tcp_sack_option(), called from
2729  * tcp_dooptions(), will fix up the hole.
2730  */
2731 void
2732 tcp_del_sackholes(tp, th)
2733 	struct tcpcb *tp;
2734 	struct tcphdr *th;
2735 {
2736 	if (!tp->sack_disable && tp->t_state != TCPS_LISTEN) {
2737 		/* max because this could be an older ack just arrived */
2738 		tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ?
2739 			th->th_ack : tp->snd_una;
2740 		struct sackhole *cur = tp->snd_holes;
2741 		struct sackhole *prev;
2742 		while (cur)
2743 			if (SEQ_LEQ(cur->end, lastack)) {
2744 				prev = cur;
2745 				cur = cur->next;
2746 				pool_put(&sackhl_pool, prev);
2747 				tp->snd_numholes--;
2748 			} else if (SEQ_LT(cur->start, lastack)) {
2749 				cur->start = lastack;
2750 				if (SEQ_LT(cur->rxmit, cur->start))
2751 					cur->rxmit = cur->start;
2752 				break;
2753 			} else
2754 				break;
2755 		tp->snd_holes = cur;
2756 	}
2757 }
2758 
2759 /*
2760  * Delete all receiver-side SACK information.
2761  */
2762 void
2763 tcp_clean_sackreport(tp)
2764 	struct tcpcb *tp;
2765 {
2766 	int i;
2767 
2768 	tp->rcv_numsacks = 0;
2769 	for (i = 0; i < MAX_SACK_BLKS; i++)
2770 		tp->sackblks[i].start = tp->sackblks[i].end=0;
2771 
2772 }
2773 
2774 /*
2775  * Checks for partial ack.  If partial ack arrives, turn off retransmission
2776  * timer, deflate the window, do not clear tp->t_dupacks, and return 1.
2777  * If the ack advances at least to tp->snd_last, return 0.
2778  */
2779 int
2780 tcp_sack_partialack(tp, th)
2781 	struct tcpcb *tp;
2782 	struct tcphdr *th;
2783 {
2784 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
2785 		/* Turn off retx. timer (will start again next segment) */
2786 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
2787 		tp->t_rtttime = 0;
2788 #ifndef TCP_FACK
2789 		/*
2790 		 * Partial window deflation.  This statement relies on the
2791 		 * fact that tp->snd_una has not been updated yet.  In FACK
2792 		 * hold snd_cwnd constant during fast recovery.
2793 		 */
2794 		if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) {
2795 			tp->snd_cwnd -= th->th_ack - tp->snd_una;
2796 			tp->snd_cwnd += tp->t_maxseg;
2797 		} else
2798 			tp->snd_cwnd = tp->t_maxseg;
2799 #endif
2800 		return (1);
2801 	}
2802 	return (0);
2803 }
2804 #endif /* TCP_SACK */
2805 
2806 /*
2807  * Pull out of band byte out of a segment so
2808  * it doesn't appear in the user's data queue.
2809  * It is still reflected in the segment length for
2810  * sequencing purposes.
2811  */
2812 void
2813 tcp_pulloutofband(so, urgent, m, off)
2814 	struct socket *so;
2815 	u_int urgent;
2816 	struct mbuf *m;
2817 	int off;
2818 {
2819         int cnt = off + urgent - 1;
2820 
2821 	while (cnt >= 0) {
2822 		if (m->m_len > cnt) {
2823 			char *cp = mtod(m, caddr_t) + cnt;
2824 			struct tcpcb *tp = sototcpcb(so);
2825 
2826 			tp->t_iobc = *cp;
2827 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2828 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2829 			m->m_len--;
2830 			return;
2831 		}
2832 		cnt -= m->m_len;
2833 		m = m->m_next;
2834 		if (m == 0)
2835 			break;
2836 	}
2837 	panic("tcp_pulloutofband");
2838 }
2839 
2840 /*
2841  * Collect new round-trip time estimate
2842  * and update averages and current timeout.
2843  */
2844 void
2845 tcp_xmit_timer(tp, rtt)
2846 	struct tcpcb *tp;
2847 	short rtt;
2848 {
2849 	short delta;
2850 	short rttmin;
2851 
2852 	tcpstat.tcps_rttupdated++;
2853 	--rtt;
2854 	if (tp->t_srtt != 0) {
2855 		/*
2856 		 * srtt is stored as fixed point with 3 bits after the
2857 		 * binary point (i.e., scaled by 8).  The following magic
2858 		 * is equivalent to the smoothing algorithm in rfc793 with
2859 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2860 		 * point).  Adjust rtt to origin 0.
2861 		 */
2862 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
2863 		if ((tp->t_srtt += delta) <= 0)
2864 			tp->t_srtt = 1;
2865 		/*
2866 		 * We accumulate a smoothed rtt variance (actually, a
2867 		 * smoothed mean difference), then set the retransmit
2868 		 * timer to smoothed rtt + 4 times the smoothed variance.
2869 		 * rttvar is stored as fixed point with 2 bits after the
2870 		 * binary point (scaled by 4).  The following is
2871 		 * equivalent to rfc793 smoothing with an alpha of .75
2872 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2873 		 * rfc793's wired-in beta.
2874 		 */
2875 		if (delta < 0)
2876 			delta = -delta;
2877 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
2878 		if ((tp->t_rttvar += delta) <= 0)
2879 			tp->t_rttvar = 1;
2880 	} else {
2881 		/*
2882 		 * No rtt measurement yet - use the unsmoothed rtt.
2883 		 * Set the variance to half the rtt (so our first
2884 		 * retransmit happens at 3*rtt).
2885 		 */
2886 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
2887 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
2888 	}
2889 	tp->t_rtttime = 0;
2890 	tp->t_rxtshift = 0;
2891 
2892 	/*
2893 	 * the retransmit should happen at rtt + 4 * rttvar.
2894 	 * Because of the way we do the smoothing, srtt and rttvar
2895 	 * will each average +1/2 tick of bias.  When we compute
2896 	 * the retransmit timer, we want 1/2 tick of rounding and
2897 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2898 	 * firing of the timer.  The bias will give us exactly the
2899 	 * 1.5 tick we need.  But, because the bias is
2900 	 * statistical, we have to test that we don't drop below
2901 	 * the minimum feasible timer (which is 2 ticks).
2902 	 */
2903 	if (tp->t_rttmin > rtt + 2)
2904 		rttmin = tp->t_rttmin;
2905 	else
2906 		rttmin = rtt + 2;
2907 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
2908 
2909 	/*
2910 	 * We received an ack for a packet that wasn't retransmitted;
2911 	 * it is probably safe to discard any error indications we've
2912 	 * received recently.  This isn't quite right, but close enough
2913 	 * for now (a route might have failed after we sent a segment,
2914 	 * and the return path might not be symmetrical).
2915 	 */
2916 	tp->t_softerror = 0;
2917 }
2918 
2919 /*
2920  * Determine a reasonable value for maxseg size.
2921  * If the route is known, check route for mtu.
2922  * If none, use an mss that can be handled on the outgoing
2923  * interface without forcing IP to fragment; if bigger than
2924  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2925  * to utilize large mbufs.  If no route is found, route has no mtu,
2926  * or the destination isn't local, use a default, hopefully conservative
2927  * size (usually 512 or the default IP max size, but no more than the mtu
2928  * of the interface), as we can't discover anything about intervening
2929  * gateways or networks.  We also initialize the congestion/slow start
2930  * window to be a single segment if the destination isn't local.
2931  * While looking at the routing entry, we also initialize other path-dependent
2932  * parameters from pre-set or cached values in the routing entry.
2933  *
2934  * Also take into account the space needed for options that we
2935  * send regularly.  Make maxseg shorter by that amount to assure
2936  * that we can send maxseg amount of data even when the options
2937  * are present.  Store the upper limit of the length of options plus
2938  * data in maxopd.
2939  *
2940  * NOTE: offer == -1 indicates that the maxseg size changed due to
2941  * Path MTU discovery.
2942  */
2943 int
2944 tcp_mss(tp, offer)
2945 	struct tcpcb *tp;
2946 	int offer;
2947 {
2948 	struct rtentry *rt;
2949 	struct ifnet *ifp;
2950 	int mss, mssopt;
2951 	int iphlen;
2952 	int is_ipv6 = 0;
2953 	struct inpcb *inp;
2954 
2955 	inp = tp->t_inpcb;
2956 
2957 	mssopt = mss = tcp_mssdflt;
2958 
2959 	rt = in_pcbrtentry(inp);
2960 
2961 	if (rt == NULL)
2962 		goto out;
2963 
2964 	ifp = rt->rt_ifp;
2965 
2966 	switch (tp->pf) {
2967 #ifdef INET6
2968 	case AF_INET6:
2969 		iphlen = sizeof(struct ip6_hdr);
2970 		is_ipv6 = 1;
2971 		break;
2972 #endif
2973 	case AF_INET:
2974 		iphlen = sizeof(struct ip);
2975 		break;
2976 	default:
2977 		/* the family does not support path MTU discovery */
2978 		goto out;
2979 	}
2980 
2981 #ifdef RTV_MTU
2982 	/*
2983 	 * if there's an mtu associated with the route and we support
2984 	 * path MTU discovery for the underlying protocol family, use it.
2985 	 */
2986 	if (rt->rt_rmx.rmx_mtu) {
2987 		/*
2988 		 * One may wish to lower MSS to take into account options,
2989 		 * especially security-related options.
2990 		 */
2991 		mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr);
2992 	} else
2993 #endif /* RTV_MTU */
2994 	if (!ifp)
2995 		/*
2996 		 * ifp may be null and rmx_mtu may be zero in certain
2997 		 * v6 cases (e.g., if ND wasn't able to resolve the
2998 		 * destination host.
2999 		 */
3000 		goto out;
3001 	else if (ifp->if_flags & IFF_LOOPBACK)
3002 		mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3003 	else if (!is_ipv6) {
3004 		if (ip_mtudisc)
3005 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3006 		else if (inp && in_localaddr(inp->inp_faddr))
3007 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3008 	}
3009 #ifdef INET6
3010 	else if (is_ipv6) {
3011 		if (inp && IN6_IS_ADDR_V4MAPPED(&inp->inp_faddr6)) {
3012 			/* mapped addr case */
3013 			struct in_addr d;
3014 			bcopy(&inp->inp_faddr6.s6_addr32[3], &d, sizeof(d));
3015 			if (ip_mtudisc || in_localaddr(d))
3016 				mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3017 		} else {
3018 			/*
3019 			 * for IPv6, path MTU discovery is always turned on,
3020 			 * or the node must use packet size <= 1280.
3021 			 */
3022 			mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3023 		}
3024 	}
3025 #endif /* INET6 */
3026 
3027 	/* Calculate the value that we offer in TCPOPT_MAXSEG */
3028 	if (offer != -1) {
3029 		mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr);
3030 		mssopt = max(tcp_mssdflt, mssopt);
3031 	}
3032 
3033  out:
3034 	/*
3035 	 * The current mss, t_maxseg, is initialized to the default value.
3036 	 * If we compute a smaller value, reduce the current mss.
3037 	 * If we compute a larger value, return it for use in sending
3038 	 * a max seg size option, but don't store it for use
3039 	 * unless we received an offer at least that large from peer.
3040 	 * However, do not accept offers under 64 bytes.
3041 	 */
3042 	if (offer > 0)
3043 		tp->t_peermss = offer;
3044 	if (tp->t_peermss)
3045 		mss = min(mss, tp->t_peermss);
3046 	mss = max(mss, 64);		/* sanity - at least max opt. space */
3047 
3048 	/*
3049 	 * maxopd stores the maximum length of data AND options
3050 	 * in a segment; maxseg is the amount of data in a normal
3051 	 * segment.  We need to store this value (maxopd) apart
3052 	 * from maxseg, because now every segment carries options
3053 	 * and thus we normally have somewhat less data in segments.
3054 	 */
3055 	tp->t_maxopd = mss;
3056 
3057  	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3058 	    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
3059 		mss -= TCPOLEN_TSTAMP_APPA;
3060 
3061 	if (offer == -1) {
3062 		/* mss changed due to Path MTU discovery */
3063 		if (mss < tp->t_maxseg) {
3064 			/*
3065 			 * Follow suggestion in RFC 2414 to reduce the
3066 			 * congestion window by the ratio of the old
3067 			 * segment size to the new segment size.
3068 			 */
3069 			tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) *
3070 					     mss, mss);
3071 		}
3072 	} else
3073 		tp->snd_cwnd = mss;
3074 
3075 	tp->t_maxseg = mss;
3076 
3077 	return (offer != -1 ? mssopt : mss);
3078 }
3079 
3080 /*
3081  * Set connection variables based on the effective MSS.
3082  * We are passed the TCPCB for the actual connection.  If we
3083  * are the server, we are called by the compressed state engine
3084  * when the 3-way handshake is complete.  If we are the client,
3085  * we are called when we receive the SYN,ACK from the server.
3086  *
3087  * NOTE: The t_maxseg value must be initialized in the TCPCB
3088  * before this routine is called!
3089  */
3090 void
3091 tcp_mss_update(tp)
3092 	struct tcpcb *tp;
3093 {
3094 	int mss, rtt;
3095 	u_long bufsize;
3096 	struct rtentry *rt;
3097 	struct socket *so;
3098 
3099 	so = tp->t_inpcb->inp_socket;
3100 	mss = tp->t_maxseg;
3101 
3102 	rt = in_pcbrtentry(tp->t_inpcb);
3103 
3104 	if (rt == NULL)
3105 		return;
3106 
3107 #ifdef RTV_MTU	/* if route characteristics exist ... */
3108 	/*
3109 	 * While we're here, check if there's an initial rtt
3110 	 * or rttvar.  Convert from the route-table units
3111 	 * to scaled multiples of the slow timeout timer.
3112 	 */
3113 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3114 		/*
3115 		 * XXX the lock bit for MTU indicates that the value
3116 		 * is also a minimum value; this is subject to time.
3117 		 */
3118 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
3119 			TCPT_RANGESET(tp->t_rttmin,
3120 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
3121 			    TCPTV_MIN, TCPTV_REXMTMAX);
3122 		tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
3123 		if (rt->rt_rmx.rmx_rttvar)
3124 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3125 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
3126 		else
3127 			/* default variation is +- 1 rtt */
3128 			tp->t_rttvar =
3129 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3130 		TCPT_RANGESET((long) tp->t_rxtcur,
3131 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3132 		    tp->t_rttmin, TCPTV_REXMTMAX);
3133 	}
3134 #endif
3135 
3136 	/*
3137 	 * If there's a pipesize, change the socket buffer
3138 	 * to that size.  Make the socket buffers an integral
3139 	 * number of mss units; if the mss is larger than
3140 	 * the socket buffer, decrease the mss.
3141 	 */
3142 #ifdef RTV_SPIPE
3143 	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3144 #endif
3145 		bufsize = so->so_snd.sb_hiwat;
3146 	if (bufsize < mss) {
3147 		mss = bufsize;
3148 		/* Update t_maxseg and t_maxopd */
3149 		tcp_mss(tp, mss);
3150 	} else {
3151 		bufsize = roundup(bufsize, mss);
3152 		if (bufsize > sb_max)
3153 			bufsize = sb_max;
3154 		(void)sbreserve(&so->so_snd, bufsize);
3155 	}
3156 
3157 #ifdef RTV_RPIPE
3158 	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3159 #endif
3160 		bufsize = so->so_rcv.sb_hiwat;
3161 	if (bufsize > mss) {
3162 		bufsize = roundup(bufsize, mss);
3163 		if (bufsize > sb_max)
3164 			bufsize = sb_max;
3165 		(void)sbreserve(&so->so_rcv, bufsize);
3166 #ifdef RTV_RPIPE
3167 		if (rt->rt_rmx.rmx_recvpipe > 0)
3168 			tcp_rscale(tp, so->so_rcv.sb_hiwat);
3169 #endif
3170 	}
3171 
3172 #ifdef RTV_SSTHRESH
3173 	if (rt->rt_rmx.rmx_ssthresh) {
3174 		/*
3175 		 * There's some sort of gateway or interface
3176 		 * buffer limit on the path.  Use this to set
3177 		 * the slow start threshhold, but set the
3178 		 * threshold to no less than 2*mss.
3179 		 */
3180 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3181 	}
3182 #endif /* RTV_MTU */
3183 }
3184 #endif /* TUBA_INCLUDE */
3185 
3186 #if defined (TCP_SACK)
3187 /*
3188  * Checks for partial ack.  If partial ack arrives, force the retransmission
3189  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3190  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
3191  * be started again.  If the ack advances at least to tp->snd_last, return 0.
3192  */
3193 int
3194 tcp_newreno(tp, th)
3195 	struct tcpcb *tp;
3196 	struct tcphdr *th;
3197 {
3198 	if (SEQ_LT(th->th_ack, tp->snd_last)) {
3199 		/*
3200 		 * snd_una has not been updated and the socket send buffer
3201 		 * not yet drained of the acked data, so we have to leave
3202 		 * snd_una as it was to get the correct data offset in
3203 		 * tcp_output().
3204 		 */
3205 		tcp_seq onxt = tp->snd_nxt;
3206 		u_long  ocwnd = tp->snd_cwnd;
3207 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
3208 		tp->t_rtttime = 0;
3209 		tp->snd_nxt = th->th_ack;
3210 		/*
3211 		 * Set snd_cwnd to one segment beyond acknowledged offset
3212 		 * (tp->snd_una not yet updated when this function is called)
3213 		 */
3214 		tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3215 		(void) tcp_output(tp);
3216 		tp->snd_cwnd = ocwnd;
3217 		if (SEQ_GT(onxt, tp->snd_nxt))
3218 			tp->snd_nxt = onxt;
3219 		/*
3220 		 * Partial window deflation.  Relies on fact that tp->snd_una
3221 		 * not updated yet.
3222 		 */
3223 		tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3224 		return 1;
3225 	}
3226 	return 0;
3227 }
3228 #endif /* TCP_SACK */
3229