xref: /freebsd/sys/netinet/tcp_input.c (revision 31489a9a2653e123121e8ca39b4be802013d2b50)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
30  * $FreeBSD$
31  */
32 
33 #include "opt_ipfw.h"		/* for ipfw_fwd		*/
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ipsec.h"
37 #include "opt_mac.h"
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 #include "opt_tcp_sack.h"
41 
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/mac.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h>		/* for proc0 declaration */
48 #include <sys/protosw.h>
49 #include <sys/signalvar.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/systm.h>
55 
56 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
57 
58 #include <vm/uma.h>
59 
60 #include <net/if.h>
61 #include <net/route.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_icmp.h>	/* for ICMP_BANDLIM		*/
69 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM		*/
70 #include <netinet/ip_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet/icmp6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet/tcp.h>
77 #include <netinet/tcp_fsm.h>
78 #include <netinet/tcp_seq.h>
79 #include <netinet/tcp_timer.h>
80 #include <netinet/tcp_var.h>
81 #include <netinet6/tcp6_var.h>
82 #include <netinet/tcpip.h>
83 #ifdef TCPDEBUG
84 #include <netinet/tcp_debug.h>
85 #endif /* TCPDEBUG */
86 
87 #ifdef FAST_IPSEC
88 #include <netipsec/ipsec.h>
89 #include <netipsec/ipsec6.h>
90 #endif /*FAST_IPSEC*/
91 
92 #ifdef IPSEC
93 #include <netinet6/ipsec.h>
94 #include <netinet6/ipsec6.h>
95 #include <netkey/key.h>
96 #endif /*IPSEC*/
97 
98 #include <machine/in_cksum.h>
99 
100 static const int tcprexmtthresh = 3;
101 
102 struct	tcpstat tcpstat;
103 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
104     &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
105 
106 static int log_in_vain = 0;
107 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
108     &log_in_vain, 0, "Log all incoming TCP connections");
109 
110 static int blackhole = 0;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
112 	&blackhole, 0, "Do not send RST when dropping refused connections");
113 
114 int tcp_delack_enabled = 1;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
116     &tcp_delack_enabled, 0,
117     "Delay ACK to try and piggyback it onto a data packet");
118 
119 #ifdef TCP_DROP_SYNFIN
120 static int drop_synfin = 0;
121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
122     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
123 #endif
124 
125 static int tcp_do_rfc3042 = 1;
126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
127     &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
128 
129 static int tcp_do_rfc3390 = 1;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
131     &tcp_do_rfc3390, 0,
132     "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
133 
134 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
135 	    "TCP Segment Reassembly Queue");
136 
137 static int tcp_reass_maxseg = 0;
138 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
139 	   &tcp_reass_maxseg, 0,
140 	   "Global maximum number of TCP Segments in Reassembly Queue");
141 
142 int tcp_reass_qsize = 0;
143 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
144 	   &tcp_reass_qsize, 0,
145 	   "Global number of TCP Segments currently in Reassembly Queue");
146 
147 static int tcp_reass_maxqlen = 48;
148 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
149 	   &tcp_reass_maxqlen, 0,
150 	   "Maximum number of TCP Segments per individual Reassembly Queue");
151 
152 static int tcp_reass_overflows = 0;
153 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
154 	   &tcp_reass_overflows, 0,
155 	   "Global number of TCP Segment Reassembly Queue Overflows");
156 
157 static int tcp_sack_recovery_initburst = 3;
158 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO,
159 	   initburst, CTLFLAG_RW,
160 	   &tcp_sack_recovery_initburst, 0,
161 	   "Initial Number of Rexmits when sack recovery is set up");
162 
163 struct inpcbhead tcb;
164 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
165 struct inpcbinfo tcbinfo;
166 struct mtx	*tcbinfo_mtx;
167 
168 static void	 tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *,
169 		     int, int, struct tcphdr *);
170 
171 static void	 tcp_pulloutofband(struct socket *,
172 		     struct tcphdr *, struct mbuf *, int);
173 static int	 tcp_reass(struct tcpcb *, struct tcphdr *, int *,
174 		     struct mbuf *);
175 static void	 tcp_xmit_timer(struct tcpcb *, int);
176 static void	 tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
177 static int	 tcp_timewait(struct tcptw *, struct tcpopt *,
178 		     struct tcphdr *, struct mbuf *, int);
179 
180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
181 #ifdef INET6
182 #define ND6_HINT(tp) \
183 do { \
184 	if ((tp) && (tp)->t_inpcb && \
185 	    ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
186 		nd6_nud_hint(NULL, NULL, 0); \
187 } while (0)
188 #else
189 #define ND6_HINT(tp)
190 #endif
191 
192 /*
193  * Indicate whether this ack should be delayed.  We can delay the ack if
194  *	- there is no delayed ack timer in progress and
195  *	- our last ack wasn't a 0-sized window.  We never want to delay
196  *	  the ack that opens up a 0-sized window and
197  *		- delayed acks are enabled or
198  *		- this is a half-synchronized T/TCP connection.
199  */
200 #define DELAY_ACK(tp)							\
201 	((!callout_active(tp->tt_delack) &&				\
202 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
203 	    (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
204 
205 /* Initialize TCP reassembly queue */
206 uma_zone_t	tcp_reass_zone;
207 void
208 tcp_reass_init()
209 {
210 	tcp_reass_maxseg = nmbclusters / 16;
211 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
212 	    &tcp_reass_maxseg);
213 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
214 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
215 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
216 }
217 
218 static int
219 tcp_reass(tp, th, tlenp, m)
220 	register struct tcpcb *tp;
221 	register struct tcphdr *th;
222 	int *tlenp;
223 	struct mbuf *m;
224 {
225 	struct tseg_qent *q;
226 	struct tseg_qent *p = NULL;
227 	struct tseg_qent *nq;
228 	struct tseg_qent *te = NULL;
229 	struct socket *so = tp->t_inpcb->inp_socket;
230 	int flags;
231 
232 	/*
233 	 * XXX: tcp_reass() is rather inefficient with its data structures
234 	 * and should be rewritten (see NetBSD for optimizations).  While
235 	 * doing that it should move to its own file tcp_reass.c.
236 	 */
237 
238 	/*
239 	 * Call with th==0 after become established to
240 	 * force pre-ESTABLISHED data up to user socket.
241 	 */
242 	if (th == 0)
243 		goto present;
244 
245 	/*
246 	 * Limit the number of segments in the reassembly queue to prevent
247 	 * holding on to too many segments (and thus running out of mbufs).
248 	 * Make sure to let the missing segment through which caused this
249 	 * queue.  Always keep one global queue entry spare to be able to
250 	 * process the missing segment.
251 	 */
252 	if (th->th_seq != tp->rcv_nxt &&
253 	    (tcp_reass_qsize + 1 >= tcp_reass_maxseg ||
254 	     tp->t_segqlen >= tcp_reass_maxqlen)) {
255 		tcp_reass_overflows++;
256 		tcpstat.tcps_rcvmemdrop++;
257 		m_freem(m);
258 		return (0);
259 	}
260 
261 	/*
262 	 * Allocate a new queue entry. If we can't, or hit the zone limit
263 	 * just drop the pkt.
264 	 */
265 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
266 	if (te == NULL) {
267 		tcpstat.tcps_rcvmemdrop++;
268 		m_freem(m);
269 		return (0);
270 	}
271 	tp->t_segqlen++;
272 	tcp_reass_qsize++;
273 
274 	/*
275 	 * Find a segment which begins after this one does.
276 	 */
277 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
278 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
279 			break;
280 		p = q;
281 	}
282 
283 	/*
284 	 * If there is a preceding segment, it may provide some of
285 	 * our data already.  If so, drop the data from the incoming
286 	 * segment.  If it provides all of our data, drop us.
287 	 */
288 	if (p != NULL) {
289 		register int i;
290 		/* conversion to int (in i) handles seq wraparound */
291 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
292 		if (i > 0) {
293 			if (i >= *tlenp) {
294 				tcpstat.tcps_rcvduppack++;
295 				tcpstat.tcps_rcvdupbyte += *tlenp;
296 				m_freem(m);
297 				uma_zfree(tcp_reass_zone, te);
298 				tp->t_segqlen--;
299 				tcp_reass_qsize--;
300 				/*
301 				 * Try to present any queued data
302 				 * at the left window edge to the user.
303 				 * This is needed after the 3-WHS
304 				 * completes.
305 				 */
306 				goto present;	/* ??? */
307 			}
308 			m_adj(m, i);
309 			*tlenp -= i;
310 			th->th_seq += i;
311 		}
312 	}
313 	tcpstat.tcps_rcvoopack++;
314 	tcpstat.tcps_rcvoobyte += *tlenp;
315 
316 	/*
317 	 * While we overlap succeeding segments trim them or,
318 	 * if they are completely covered, dequeue them.
319 	 */
320 	while (q) {
321 		register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
322 		if (i <= 0)
323 			break;
324 		if (i < q->tqe_len) {
325 			q->tqe_th->th_seq += i;
326 			q->tqe_len -= i;
327 			m_adj(q->tqe_m, i);
328 			break;
329 		}
330 
331 		nq = LIST_NEXT(q, tqe_q);
332 		LIST_REMOVE(q, tqe_q);
333 		m_freem(q->tqe_m);
334 		uma_zfree(tcp_reass_zone, q);
335 		tp->t_segqlen--;
336 		tcp_reass_qsize--;
337 		q = nq;
338 	}
339 
340 	/* Insert the new segment queue entry into place. */
341 	te->tqe_m = m;
342 	te->tqe_th = th;
343 	te->tqe_len = *tlenp;
344 
345 	if (p == NULL) {
346 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
347 	} else {
348 		LIST_INSERT_AFTER(p, te, tqe_q);
349 	}
350 
351 present:
352 	/*
353 	 * Present data to user, advancing rcv_nxt through
354 	 * completed sequence space.
355 	 */
356 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
357 		return (0);
358 	q = LIST_FIRST(&tp->t_segq);
359 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
360 		return (0);
361 	SOCKBUF_LOCK(&so->so_rcv);
362 	do {
363 		tp->rcv_nxt += q->tqe_len;
364 		flags = q->tqe_th->th_flags & TH_FIN;
365 		nq = LIST_NEXT(q, tqe_q);
366 		LIST_REMOVE(q, tqe_q);
367 		/* Unlocked read. */
368 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
369 			m_freem(q->tqe_m);
370 		else
371 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
372 		uma_zfree(tcp_reass_zone, q);
373 		tp->t_segqlen--;
374 		tcp_reass_qsize--;
375 		q = nq;
376 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
377 	ND6_HINT(tp);
378 	sorwakeup_locked(so);
379 	return (flags);
380 }
381 
382 /*
383  * TCP input routine, follows pages 65-76 of the
384  * protocol specification dated September, 1981 very closely.
385  */
386 #ifdef INET6
387 int
388 tcp6_input(mp, offp, proto)
389 	struct mbuf **mp;
390 	int *offp, proto;
391 {
392 	register struct mbuf *m = *mp;
393 	struct in6_ifaddr *ia6;
394 
395 	IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
396 
397 	/*
398 	 * draft-itojun-ipv6-tcp-to-anycast
399 	 * better place to put this in?
400 	 */
401 	ia6 = ip6_getdstifaddr(m);
402 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
403 		struct ip6_hdr *ip6;
404 
405 		ip6 = mtod(m, struct ip6_hdr *);
406 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
407 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
408 		return IPPROTO_DONE;
409 	}
410 
411 	tcp_input(m, *offp);
412 	return IPPROTO_DONE;
413 }
414 #endif
415 
416 void
417 tcp_input(m, off0)
418 	register struct mbuf *m;
419 	int off0;
420 {
421 	register struct tcphdr *th;
422 	register struct ip *ip = NULL;
423 	register struct ipovly *ipov;
424 	register struct inpcb *inp = NULL;
425 	u_char *optp = NULL;
426 	int optlen = 0;
427 	int len, tlen, off;
428 	int drop_hdrlen;
429 	register struct tcpcb *tp = 0;
430 	register int thflags;
431 	struct socket *so = 0;
432 	int todrop, acked, ourfinisacked, needoutput = 0;
433 	u_long tiwin;
434 	struct tcpopt to;		/* options in this segment */
435 	int headlocked = 0;
436 #ifdef IPFIREWALL_FORWARD
437 	struct m_tag *fwd_tag;
438 #endif
439 	int rstreason; /* For badport_bandlim accounting purposes */
440 
441 	struct ip6_hdr *ip6 = NULL;
442 #ifdef INET6
443 	int isipv6;
444 #else
445 	const int isipv6 = 0;
446 #endif
447 
448 #ifdef TCPDEBUG
449 	/*
450 	 * The size of tcp_saveipgen must be the size of the max ip header,
451 	 * now IPv6.
452 	 */
453 	u_char tcp_saveipgen[40];
454 	struct tcphdr tcp_savetcp;
455 	short ostate = 0;
456 #endif
457 
458 #ifdef INET6
459 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
460 #endif
461 	bzero((char *)&to, sizeof(to));
462 
463 	tcpstat.tcps_rcvtotal++;
464 
465 	if (isipv6) {
466 #ifdef INET6
467 		/* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
468 		ip6 = mtod(m, struct ip6_hdr *);
469 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
470 		if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
471 			tcpstat.tcps_rcvbadsum++;
472 			goto drop;
473 		}
474 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
475 
476 		/*
477 		 * Be proactive about unspecified IPv6 address in source.
478 		 * As we use all-zero to indicate unbounded/unconnected pcb,
479 		 * unspecified IPv6 address can be used to confuse us.
480 		 *
481 		 * Note that packets with unspecified IPv6 destination is
482 		 * already dropped in ip6_input.
483 		 */
484 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
485 			/* XXX stat */
486 			goto drop;
487 		}
488 #else
489 		th = NULL;		/* XXX: avoid compiler warning */
490 #endif
491 	} else {
492 		/*
493 		 * Get IP and TCP header together in first mbuf.
494 		 * Note: IP leaves IP header in first mbuf.
495 		 */
496 		if (off0 > sizeof (struct ip)) {
497 			ip_stripoptions(m, (struct mbuf *)0);
498 			off0 = sizeof(struct ip);
499 		}
500 		if (m->m_len < sizeof (struct tcpiphdr)) {
501 			if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
502 				tcpstat.tcps_rcvshort++;
503 				return;
504 			}
505 		}
506 		ip = mtod(m, struct ip *);
507 		ipov = (struct ipovly *)ip;
508 		th = (struct tcphdr *)((caddr_t)ip + off0);
509 		tlen = ip->ip_len;
510 
511 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
512 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
513 				th->th_sum = m->m_pkthdr.csum_data;
514 			else
515 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
516 						ip->ip_dst.s_addr,
517 						htonl(m->m_pkthdr.csum_data +
518 							ip->ip_len +
519 							IPPROTO_TCP));
520 			th->th_sum ^= 0xffff;
521 #ifdef TCPDEBUG
522 			ipov->ih_len = (u_short)tlen;
523 			ipov->ih_len = htons(ipov->ih_len);
524 #endif
525 		} else {
526 			/*
527 			 * Checksum extended TCP header and data.
528 			 */
529 			len = sizeof (struct ip) + tlen;
530 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
531 			ipov->ih_len = (u_short)tlen;
532 			ipov->ih_len = htons(ipov->ih_len);
533 			th->th_sum = in_cksum(m, len);
534 		}
535 		if (th->th_sum) {
536 			tcpstat.tcps_rcvbadsum++;
537 			goto drop;
538 		}
539 #ifdef INET6
540 		/* Re-initialization for later version check */
541 		ip->ip_v = IPVERSION;
542 #endif
543 	}
544 
545 	/*
546 	 * Check that TCP offset makes sense,
547 	 * pull out TCP options and adjust length.		XXX
548 	 */
549 	off = th->th_off << 2;
550 	if (off < sizeof (struct tcphdr) || off > tlen) {
551 		tcpstat.tcps_rcvbadoff++;
552 		goto drop;
553 	}
554 	tlen -= off;	/* tlen is used instead of ti->ti_len */
555 	if (off > sizeof (struct tcphdr)) {
556 		if (isipv6) {
557 #ifdef INET6
558 			IP6_EXTHDR_CHECK(m, off0, off, );
559 			ip6 = mtod(m, struct ip6_hdr *);
560 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
561 #endif
562 		} else {
563 			if (m->m_len < sizeof(struct ip) + off) {
564 				if ((m = m_pullup(m, sizeof (struct ip) + off))
565 						== 0) {
566 					tcpstat.tcps_rcvshort++;
567 					return;
568 				}
569 				ip = mtod(m, struct ip *);
570 				ipov = (struct ipovly *)ip;
571 				th = (struct tcphdr *)((caddr_t)ip + off0);
572 			}
573 		}
574 		optlen = off - sizeof (struct tcphdr);
575 		optp = (u_char *)(th + 1);
576 	}
577 	thflags = th->th_flags;
578 
579 #ifdef TCP_DROP_SYNFIN
580 	/*
581 	 * If the drop_synfin option is enabled, drop all packets with
582 	 * both the SYN and FIN bits set. This prevents e.g. nmap from
583 	 * identifying the TCP/IP stack.
584 	 *
585 	 * This is a violation of the TCP specification.
586 	 */
587 	if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
588 		goto drop;
589 #endif
590 
591 	/*
592 	 * Convert TCP protocol specific fields to host format.
593 	 */
594 	th->th_seq = ntohl(th->th_seq);
595 	th->th_ack = ntohl(th->th_ack);
596 	th->th_win = ntohs(th->th_win);
597 	th->th_urp = ntohs(th->th_urp);
598 
599 	/*
600 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
601 	 * until after ip6_savecontrol() is called and before other functions
602 	 * which don't want those proto headers.
603 	 * Because ip6_savecontrol() is going to parse the mbuf to
604 	 * search for data to be passed up to user-land, it wants mbuf
605 	 * parameters to be unchanged.
606 	 * XXX: the call of ip6_savecontrol() has been obsoleted based on
607 	 * latest version of the advanced API (20020110).
608 	 */
609 	drop_hdrlen = off0 + off;
610 
611 	/*
612 	 * Locate pcb for segment.
613 	 */
614 	INP_INFO_WLOCK(&tcbinfo);
615 	headlocked = 1;
616 findpcb:
617 #ifdef IPFIREWALL_FORWARD
618 	/* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
619 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
620 
621 	if (fwd_tag != NULL && isipv6 == 0) {	/* IPv6 support is not yet */
622 		struct sockaddr_in *next_hop;
623 
624 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
625 		/*
626 		 * Transparently forwarded. Pretend to be the destination.
627 		 * already got one like this?
628 		 */
629 		inp = in_pcblookup_hash(&tcbinfo,
630 					ip->ip_src, th->th_sport,
631 					ip->ip_dst, th->th_dport,
632 					0, m->m_pkthdr.rcvif);
633 		if (!inp) {
634 			/* It's new.  Try to find the ambushing socket. */
635 			inp = in_pcblookup_hash(&tcbinfo,
636 						ip->ip_src, th->th_sport,
637 						next_hop->sin_addr,
638 						next_hop->sin_port ?
639 						    ntohs(next_hop->sin_port) :
640 						    th->th_dport,
641 						1, m->m_pkthdr.rcvif);
642 		}
643 		/* Remove the tag from the packet.  We don't need it anymore. */
644 		m_tag_delete(m, fwd_tag);
645 	} else {
646 #endif /* IPFIREWALL_FORWARD */
647 		if (isipv6) {
648 #ifdef INET6
649 			inp = in6_pcblookup_hash(&tcbinfo,
650 						 &ip6->ip6_src, th->th_sport,
651 						 &ip6->ip6_dst, th->th_dport,
652 						 1, m->m_pkthdr.rcvif);
653 #endif
654 		} else
655 			inp = in_pcblookup_hash(&tcbinfo,
656 						ip->ip_src, th->th_sport,
657 						ip->ip_dst, th->th_dport,
658 						1, m->m_pkthdr.rcvif);
659 #ifdef IPFIREWALL_FORWARD
660 	}
661 #endif /* IPFIREWALL_FORWARD */
662 
663 #if defined(IPSEC) || defined(FAST_IPSEC)
664 #ifdef INET6
665 	if (isipv6) {
666 		if (inp != NULL && ipsec6_in_reject(m, inp)) {
667 #ifdef IPSEC
668 			ipsec6stat.in_polvio++;
669 #endif
670 			goto drop;
671 		}
672 	} else
673 #endif /* INET6 */
674 	if (inp != NULL && ipsec4_in_reject(m, inp)) {
675 #ifdef IPSEC
676 		ipsecstat.in_polvio++;
677 #endif
678 		goto drop;
679 	}
680 #endif /*IPSEC || FAST_IPSEC*/
681 
682 	/*
683 	 * If the state is CLOSED (i.e., TCB does not exist) then
684 	 * all data in the incoming segment is discarded.
685 	 * If the TCB exists but is in CLOSED state, it is embryonic,
686 	 * but should either do a listen or a connect soon.
687 	 */
688 	if (inp == NULL) {
689 		if (log_in_vain) {
690 #ifdef INET6
691 			char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
692 #else
693 			char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
694 #endif
695 
696 			if (isipv6) {
697 #ifdef INET6
698 				strcpy(dbuf, "[");
699 				strcpy(sbuf, "[");
700 				strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
701 				strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
702 				strcat(dbuf, "]");
703 				strcat(sbuf, "]");
704 #endif
705 			} else {
706 				strcpy(dbuf, inet_ntoa(ip->ip_dst));
707 				strcpy(sbuf, inet_ntoa(ip->ip_src));
708 			}
709 			switch (log_in_vain) {
710 			case 1:
711 				if ((thflags & TH_SYN) == 0)
712 					break;
713 				/* FALLTHROUGH */
714 			case 2:
715 				log(LOG_INFO,
716 				    "Connection attempt to TCP %s:%d "
717 				    "from %s:%d flags:0x%02x\n",
718 				    dbuf, ntohs(th->th_dport), sbuf,
719 				    ntohs(th->th_sport), thflags);
720 				break;
721 			default:
722 				break;
723 			}
724 		}
725 		if (blackhole) {
726 			switch (blackhole) {
727 			case 1:
728 				if (thflags & TH_SYN)
729 					goto drop;
730 				break;
731 			case 2:
732 				goto drop;
733 			default:
734 				goto drop;
735 			}
736 		}
737 		rstreason = BANDLIM_RST_CLOSEDPORT;
738 		goto dropwithreset;
739 	}
740 	INP_LOCK(inp);
741 	if (inp->inp_vflag & INP_TIMEWAIT) {
742 		/*
743 		 * The only option of relevance is TOF_CC, and only if
744 		 * present in a SYN segment.  See tcp_timewait().
745 		 */
746 		if (thflags & TH_SYN)
747 			tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th);
748 		if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
749 		    &to, th, m, tlen))
750 			goto findpcb;
751 		/*
752 		 * tcp_timewait unlocks inp.
753 		 */
754 		INP_INFO_WUNLOCK(&tcbinfo);
755 		return;
756 	}
757 	tp = intotcpcb(inp);
758 	if (tp == 0) {
759 		INP_UNLOCK(inp);
760 		rstreason = BANDLIM_RST_CLOSEDPORT;
761 		goto dropwithreset;
762 	}
763 	if (tp->t_state == TCPS_CLOSED)
764 		goto drop;
765 
766 	/* Unscale the window into a 32-bit value. */
767 	if ((thflags & TH_SYN) == 0)
768 		tiwin = th->th_win << tp->snd_scale;
769 	else
770 		tiwin = th->th_win;
771 
772 #ifdef MAC
773 	INP_LOCK_ASSERT(inp);
774 	if (mac_check_inpcb_deliver(inp, m))
775 		goto drop;
776 #endif
777 	so = inp->inp_socket;
778 #ifdef TCPDEBUG
779 	if (so->so_options & SO_DEBUG) {
780 		ostate = tp->t_state;
781 		if (isipv6)
782 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
783 		else
784 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
785 		tcp_savetcp = *th;
786 	}
787 #endif
788 	if (so->so_options & SO_ACCEPTCONN) {
789 		struct in_conninfo inc;
790 
791 #ifdef INET6
792 		inc.inc_isipv6 = isipv6;
793 #endif
794 		if (isipv6) {
795 			inc.inc6_faddr = ip6->ip6_src;
796 			inc.inc6_laddr = ip6->ip6_dst;
797 		} else {
798 			inc.inc_faddr = ip->ip_src;
799 			inc.inc_laddr = ip->ip_dst;
800 		}
801 		inc.inc_fport = th->th_sport;
802 		inc.inc_lport = th->th_dport;
803 
804 	        /*
805 	         * If the state is LISTEN then ignore segment if it contains
806 		 * a RST.  If the segment contains an ACK then it is bad and
807 		 * send a RST.  If it does not contain a SYN then it is not
808 		 * interesting; drop it.
809 		 *
810 		 * If the state is SYN_RECEIVED (syncache) and seg contains
811 		 * an ACK, but not for our SYN/ACK, send a RST.  If the seg
812 		 * contains a RST, check the sequence number to see if it
813 		 * is a valid reset segment.
814 		 */
815 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
816 			if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
817 				if (!syncache_expand(&inc, th, &so, m)) {
818 					/*
819 					 * No syncache entry, or ACK was not
820 					 * for our SYN/ACK.  Send a RST.
821 					 */
822 					tcpstat.tcps_badsyn++;
823 					rstreason = BANDLIM_RST_OPENPORT;
824 					goto dropwithreset;
825 				}
826 				if (so == NULL) {
827 					/*
828 					 * Could not complete 3-way handshake,
829 					 * connection is being closed down, and
830 					 * syncache will free mbuf.
831 					 */
832 					INP_UNLOCK(inp);
833 					INP_INFO_WUNLOCK(&tcbinfo);
834 					return;
835 				}
836 				/*
837 				 * Socket is created in state SYN_RECEIVED.
838 				 * Continue processing segment.
839 				 */
840 				INP_UNLOCK(inp);
841 				inp = sotoinpcb(so);
842 				INP_LOCK(inp);
843 				tp = intotcpcb(inp);
844 				/*
845 				 * This is what would have happened in
846 				 * tcp_output() when the SYN,ACK was sent.
847 				 */
848 				tp->snd_up = tp->snd_una;
849 				tp->snd_max = tp->snd_nxt = tp->iss + 1;
850 				tp->last_ack_sent = tp->rcv_nxt;
851 				/*
852 				 * RFC1323: The window in SYN & SYN/ACK
853 				 * segments is never scaled.
854 				 */
855 				tp->snd_wnd = tiwin;	/* unscaled */
856 				goto after_listen;
857 			}
858 			if (thflags & TH_RST) {
859 				syncache_chkrst(&inc, th);
860 				goto drop;
861 			}
862 			if (thflags & TH_ACK) {
863 				syncache_badack(&inc);
864 				tcpstat.tcps_badsyn++;
865 				rstreason = BANDLIM_RST_OPENPORT;
866 				goto dropwithreset;
867 			}
868 			goto drop;
869 		}
870 
871 		/*
872 		 * Segment's flags are (SYN) or (SYN|FIN).
873 		 */
874 #ifdef INET6
875 		/*
876 		 * If deprecated address is forbidden,
877 		 * we do not accept SYN to deprecated interface
878 		 * address to prevent any new inbound connection from
879 		 * getting established.
880 		 * When we do not accept SYN, we send a TCP RST,
881 		 * with deprecated source address (instead of dropping
882 		 * it).  We compromise it as it is much better for peer
883 		 * to send a RST, and RST will be the final packet
884 		 * for the exchange.
885 		 *
886 		 * If we do not forbid deprecated addresses, we accept
887 		 * the SYN packet.  RFC2462 does not suggest dropping
888 		 * SYN in this case.
889 		 * If we decipher RFC2462 5.5.4, it says like this:
890 		 * 1. use of deprecated addr with existing
891 		 *    communication is okay - "SHOULD continue to be
892 		 *    used"
893 		 * 2. use of it with new communication:
894 		 *   (2a) "SHOULD NOT be used if alternate address
895 		 *        with sufficient scope is available"
896 		 *   (2b) nothing mentioned otherwise.
897 		 * Here we fall into (2b) case as we have no choice in
898 		 * our source address selection - we must obey the peer.
899 		 *
900 		 * The wording in RFC2462 is confusing, and there are
901 		 * multiple description text for deprecated address
902 		 * handling - worse, they are not exactly the same.
903 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
904 		 */
905 		if (isipv6 && !ip6_use_deprecated) {
906 			struct in6_ifaddr *ia6;
907 
908 			if ((ia6 = ip6_getdstifaddr(m)) &&
909 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
910 				INP_UNLOCK(inp);
911 				tp = NULL;
912 				rstreason = BANDLIM_RST_OPENPORT;
913 				goto dropwithreset;
914 			}
915 		}
916 #endif
917 		/*
918 		 * If it is from this socket, drop it, it must be forged.
919 		 * Don't bother responding if the destination was a broadcast.
920 		 */
921 		if (th->th_dport == th->th_sport) {
922 			if (isipv6) {
923 				if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
924 						       &ip6->ip6_src))
925 					goto drop;
926 			} else {
927 				if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
928 					goto drop;
929 			}
930 		}
931 		/*
932 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
933 		 *
934 		 * Note that it is quite possible to receive unicast
935 		 * link-layer packets with a broadcast IP address. Use
936 		 * in_broadcast() to find them.
937 		 */
938 		if (m->m_flags & (M_BCAST|M_MCAST))
939 			goto drop;
940 		if (isipv6) {
941 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
942 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
943 				goto drop;
944 		} else {
945 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
946 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
947 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
948 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
949 				goto drop;
950 		}
951 		/*
952 		 * SYN appears to be valid; create compressed TCP state
953 		 * for syncache, or perform t/tcp connection.
954 		 */
955 		if (so->so_qlen <= so->so_qlimit) {
956 #ifdef TCPDEBUG
957 			if (so->so_options & SO_DEBUG)
958 				tcp_trace(TA_INPUT, ostate, tp,
959 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
960 #endif
961 			tcp_dooptions(tp, &to, optp, optlen, 1, th);
962 			if (!syncache_add(&inc, &to, th, &so, m))
963 				goto drop;
964 			if (so == NULL) {
965 				/*
966 				 * Entry added to syncache, mbuf used to
967 				 * send SYN,ACK packet.
968 				 */
969 				KASSERT(headlocked, ("headlocked"));
970 				INP_UNLOCK(inp);
971 				INP_INFO_WUNLOCK(&tcbinfo);
972 				return;
973 			}
974 			/*
975 			 * Segment passed TAO tests.
976 			 */
977 			INP_UNLOCK(inp);
978 			inp = sotoinpcb(so);
979 			INP_LOCK(inp);
980 			tp = intotcpcb(inp);
981 			tp->snd_wnd = tiwin;
982 			tp->t_starttime = ticks;
983 			tp->t_state = TCPS_ESTABLISHED;
984 
985 			/*
986 			 * T/TCP logic:
987 			 * If there is a FIN or if there is data, then
988 			 * delay SYN,ACK(SYN) in the hope of piggy-backing
989 			 * it on a response segment.  Otherwise must send
990 			 * ACK now in case the other side is slow starting.
991 			 */
992 			if (thflags & TH_FIN || tlen != 0)
993 				tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
994 			else
995 				tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
996 			tcpstat.tcps_connects++;
997 			soisconnected(so);
998 			goto trimthenstep6;
999 		}
1000 		goto drop;
1001 	}
1002 after_listen:
1003 	KASSERT(headlocked, ("tcp_input(): after_listen head is not locked"));
1004 	INP_LOCK_ASSERT(inp);
1005 
1006 	/* XXX temp debugging */
1007 	/* should not happen - syncache should pick up these connections */
1008 	if (tp->t_state == TCPS_LISTEN)
1009 		panic("tcp_input: TCPS_LISTEN");
1010 
1011 	/*
1012 	 * This is the second part of the MSS DoS prevention code (after
1013 	 * minmss on the sending side) and it deals with too many too small
1014 	 * tcp packets in a too short timeframe (1 second).
1015 	 *
1016 	 * For every full second we count the number of received packets
1017 	 * and bytes. If we get a lot of packets per second for this connection
1018 	 * (tcp_minmssoverload) we take a closer look at it and compute the
1019 	 * average packet size for the past second. If that is less than
1020 	 * tcp_minmss we get too many packets with very small payload which
1021 	 * is not good and burdens our system (and every packet generates
1022 	 * a wakeup to the process connected to our socket). We can reasonable
1023 	 * expect this to be small packet DoS attack to exhaust our CPU
1024 	 * cycles.
1025 	 *
1026 	 * Care has to be taken for the minimum packet overload value. This
1027 	 * value defines the minimum number of packets per second before we
1028 	 * start to worry. This must not be too low to avoid killing for
1029 	 * example interactive connections with many small packets like
1030 	 * telnet or SSH.
1031 	 *
1032 	 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1033 	 * this check.
1034 	 *
1035 	 * Account for packet if payload packet, skip over ACK, etc.
1036 	 */
1037 	if (tcp_minmss && tcp_minmssoverload &&
1038 	    tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1039 		if (tp->rcv_second > ticks) {
1040 			tp->rcv_pps++;
1041 			tp->rcv_byps += tlen + off;
1042 			if (tp->rcv_pps > tcp_minmssoverload) {
1043 				if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1044 					printf("too many small tcp packets from "
1045 					       "%s:%u, av. %lubyte/packet, "
1046 					       "dropping connection\n",
1047 #ifdef INET6
1048 						isipv6 ?
1049 						ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1050 #endif
1051 						inet_ntoa(inp->inp_inc.inc_faddr),
1052 						inp->inp_inc.inc_fport,
1053 						tp->rcv_byps / tp->rcv_pps);
1054 					tp = tcp_drop(tp, ECONNRESET);
1055 					tcpstat.tcps_minmssdrops++;
1056 					goto drop;
1057 				}
1058 			}
1059 		} else {
1060 			tp->rcv_second = ticks + hz;
1061 			tp->rcv_pps = 1;
1062 			tp->rcv_byps = tlen + off;
1063 		}
1064 	}
1065 
1066 	/*
1067 	 * Segment received on connection.
1068 	 * Reset idle time and keep-alive timer.
1069 	 */
1070 	tp->t_rcvtime = ticks;
1071 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1072 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1073 
1074 	/*
1075 	 * Process options only when we get SYN/ACK back. The SYN case
1076 	 * for incoming connections is handled in tcp_syncache.
1077 	 * XXX this is traditional behavior, may need to be cleaned up.
1078 	 */
1079 	tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th);
1080 	if (thflags & TH_SYN) {
1081 		if (to.to_flags & TOF_SCALE) {
1082 			tp->t_flags |= TF_RCVD_SCALE;
1083 			tp->requested_s_scale = to.to_requested_s_scale;
1084 		}
1085 		if (to.to_flags & TOF_TS) {
1086 			tp->t_flags |= TF_RCVD_TSTMP;
1087 			tp->ts_recent = to.to_tsval;
1088 			tp->ts_recent_age = ticks;
1089 		}
1090 		if (to.to_flags & TOF_MSS)
1091 			tcp_mss(tp, to.to_mss);
1092 		if (tp->sack_enable) {
1093 			if (!(to.to_flags & TOF_SACK))
1094 				tp->sack_enable = 0;
1095 			else
1096 				tp->t_flags |= TF_SACK_PERMIT;
1097 		}
1098 
1099 	}
1100 
1101 	if (tp->sack_enable) {
1102 		/* Delete stale (cumulatively acked) SACK holes */
1103 		tcp_del_sackholes(tp, th);
1104 		tp->rcv_laststart = th->th_seq; /* last recv'd segment*/
1105 		tp->rcv_lastend = th->th_seq + tlen;
1106 	}
1107 
1108 	/*
1109 	 * Header prediction: check for the two common cases
1110 	 * of a uni-directional data xfer.  If the packet has
1111 	 * no control flags, is in-sequence, the window didn't
1112 	 * change and we're not retransmitting, it's a
1113 	 * candidate.  If the length is zero and the ack moved
1114 	 * forward, we're the sender side of the xfer.  Just
1115 	 * free the data acked & wake any higher level process
1116 	 * that was blocked waiting for space.  If the length
1117 	 * is non-zero and the ack didn't move, we're the
1118 	 * receiver side.  If we're getting packets in-order
1119 	 * (the reassembly queue is empty), add the data to
1120 	 * the socket buffer and note that we need a delayed ack.
1121 	 * Make sure that the hidden state-flags are also off.
1122 	 * Since we check for TCPS_ESTABLISHED above, it can only
1123 	 * be TH_NEEDSYN.
1124 	 */
1125 	if (tp->t_state == TCPS_ESTABLISHED &&
1126 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1127 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1128 	    ((to.to_flags & TOF_TS) == 0 ||
1129 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1130 	     th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1131 	     tp->snd_nxt == tp->snd_max) {
1132 
1133 		/*
1134 		 * If last ACK falls within this segment's sequence numbers,
1135 		 * record the timestamp.
1136 		 * NOTE that the test is modified according to the latest
1137 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1138 		 */
1139 		if ((to.to_flags & TOF_TS) != 0 &&
1140 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1141 			tp->ts_recent_age = ticks;
1142 			tp->ts_recent = to.to_tsval;
1143 		}
1144 
1145 		if (tlen == 0) {
1146 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1147 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1148 			    tp->snd_cwnd >= tp->snd_wnd &&
1149 			    ((!tcp_do_newreno && !tp->sack_enable &&
1150 			      tp->t_dupacks < tcprexmtthresh) ||
1151 			     ((tcp_do_newreno || tp->sack_enable) &&
1152 			      !IN_FASTRECOVERY(tp)))) {
1153 				KASSERT(headlocked, ("headlocked"));
1154 				/*
1155 				 * this is a pure ack for outstanding data.
1156 				 */
1157 				++tcpstat.tcps_predack;
1158 				/*
1159 				 * "bad retransmit" recovery
1160 				 */
1161 				if (tp->t_rxtshift == 1 &&
1162 				    ticks < tp->t_badrxtwin) {
1163 					++tcpstat.tcps_sndrexmitbad;
1164 					tp->snd_cwnd = tp->snd_cwnd_prev;
1165 					tp->snd_ssthresh =
1166 					    tp->snd_ssthresh_prev;
1167 					tp->snd_recover = tp->snd_recover_prev;
1168 					if (tp->t_flags & TF_WASFRECOVERY)
1169 					    ENTER_FASTRECOVERY(tp);
1170 					tp->snd_nxt = tp->snd_max;
1171 					tp->t_badrxtwin = 0;
1172 				}
1173 
1174 				/*
1175 				 * Recalculate the transmit timer / rtt.
1176 				 *
1177 				 * Some boxes send broken timestamp replies
1178 				 * during the SYN+ACK phase, ignore
1179 				 * timestamps of 0 or we could calculate a
1180 				 * huge RTT and blow up the retransmit timer.
1181 				 */
1182 				if ((to.to_flags & TOF_TS) != 0 &&
1183 				    to.to_tsecr) {
1184 					tcp_xmit_timer(tp,
1185 					    ticks - to.to_tsecr + 1);
1186 				} else if (tp->t_rtttime &&
1187 					    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1188 					tcp_xmit_timer(tp,
1189 							ticks - tp->t_rtttime);
1190 				}
1191 				tcp_xmit_bandwidth_limit(tp, th->th_ack);
1192 				acked = th->th_ack - tp->snd_una;
1193 				tcpstat.tcps_rcvackpack++;
1194 				tcpstat.tcps_rcvackbyte += acked;
1195 				sbdrop(&so->so_snd, acked);
1196 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1197 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1198 					tp->snd_recover = th->th_ack - 1;
1199 				tp->snd_una = th->th_ack;
1200 				/*
1201 				 * pull snd_wl2 up to prevent seq wrap relative
1202 				 * to th_ack.
1203 				 */
1204 				tp->snd_wl2 = th->th_ack;
1205 				tp->t_dupacks = 0;
1206 				m_freem(m);
1207 				ND6_HINT(tp); /* some progress has been done */
1208 
1209 				/*
1210 				 * If all outstanding data are acked, stop
1211 				 * retransmit timer, otherwise restart timer
1212 				 * using current (possibly backed-off) value.
1213 				 * If process is waiting for space,
1214 				 * wakeup/selwakeup/signal.  If data
1215 				 * are ready to send, let tcp_output
1216 				 * decide between more output or persist.
1217 
1218 #ifdef TCPDEBUG
1219 				if (so->so_options & SO_DEBUG)
1220 					tcp_trace(TA_INPUT, ostate, tp,
1221 					    (void *)tcp_saveipgen,
1222 					    &tcp_savetcp, 0);
1223 #endif
1224 				 */
1225 				if (tp->snd_una == tp->snd_max)
1226 					callout_stop(tp->tt_rexmt);
1227 				else if (!callout_active(tp->tt_persist))
1228 					callout_reset(tp->tt_rexmt,
1229 						      tp->t_rxtcur,
1230 						      tcp_timer_rexmt, tp);
1231 
1232 				sowwakeup(so);
1233 				if (so->so_snd.sb_cc)
1234 					(void) tcp_output(tp);
1235 				goto check_delack;
1236 			}
1237 		} else if (th->th_ack == tp->snd_una &&
1238 		    LIST_EMPTY(&tp->t_segq) &&
1239 		    tlen <= sbspace(&so->so_rcv)) {
1240 			KASSERT(headlocked, ("headlocked"));
1241 			/*
1242 			 * this is a pure, in-sequence data packet
1243 			 * with nothing on the reassembly queue and
1244 			 * we have enough buffer space to take it.
1245 			 */
1246 			/* Clean receiver SACK report if present */
1247 			if (tp->sack_enable && tp->rcv_numsacks)
1248 				tcp_clean_sackreport(tp);
1249 			++tcpstat.tcps_preddat;
1250 			tp->rcv_nxt += tlen;
1251 			/*
1252 			 * Pull snd_wl1 up to prevent seq wrap relative to
1253 			 * th_seq.
1254 			 */
1255 			tp->snd_wl1 = th->th_seq;
1256 			/*
1257 			 * Pull rcv_up up to prevent seq wrap relative to
1258 			 * rcv_nxt.
1259 			 */
1260 			tp->rcv_up = tp->rcv_nxt;
1261 			tcpstat.tcps_rcvpack++;
1262 			tcpstat.tcps_rcvbyte += tlen;
1263 			ND6_HINT(tp);	/* some progress has been done */
1264 			/*
1265 #ifdef TCPDEBUG
1266 			if (so->so_options & SO_DEBUG)
1267 				tcp_trace(TA_INPUT, ostate, tp,
1268 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1269 #endif
1270 			 * Add data to socket buffer.
1271 			 */
1272 			/* Unlocked read. */
1273 			SOCKBUF_LOCK(&so->so_rcv);
1274 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1275 				m_freem(m);
1276 			} else {
1277 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1278 				sbappendstream_locked(&so->so_rcv, m);
1279 			}
1280 			sorwakeup_locked(so);
1281 			if (DELAY_ACK(tp)) {
1282 				tp->t_flags |= TF_DELACK;
1283 			} else {
1284 				tp->t_flags |= TF_ACKNOW;
1285 				tcp_output(tp);
1286 			}
1287 			goto check_delack;
1288 		}
1289 	}
1290 
1291 	/*
1292 	 * Calculate amount of space in receive window,
1293 	 * and then do TCP input processing.
1294 	 * Receive window is amount of space in rcv queue,
1295 	 * but not less than advertised window.
1296 	 */
1297 	{ int win;
1298 
1299 	win = sbspace(&so->so_rcv);
1300 	if (win < 0)
1301 		win = 0;
1302 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1303 	}
1304 
1305 	switch (tp->t_state) {
1306 
1307 	/*
1308 	 * If the state is SYN_RECEIVED:
1309 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1310 	 */
1311 	case TCPS_SYN_RECEIVED:
1312 		if ((thflags & TH_ACK) &&
1313 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1314 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1315 				rstreason = BANDLIM_RST_OPENPORT;
1316 				goto dropwithreset;
1317 		}
1318 		break;
1319 
1320 	/*
1321 	 * If the state is SYN_SENT:
1322 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1323 	 *	if seg contains a RST, then drop the connection.
1324 	 *	if seg does not contain SYN, then drop it.
1325 	 * Otherwise this is an acceptable SYN segment
1326 	 *	initialize tp->rcv_nxt and tp->irs
1327 	 *	if seg contains ack then advance tp->snd_una
1328 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1329 	 *	arrange for segment to be acked (eventually)
1330 	 *	continue processing rest of data/controls, beginning with URG
1331 	 */
1332 	case TCPS_SYN_SENT:
1333 		if ((thflags & TH_ACK) &&
1334 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1335 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1336 			rstreason = BANDLIM_UNLIMITED;
1337 			goto dropwithreset;
1338 		}
1339 		if (thflags & TH_RST) {
1340 			if (thflags & TH_ACK)
1341 				tp = tcp_drop(tp, ECONNREFUSED);
1342 			goto drop;
1343 		}
1344 		if ((thflags & TH_SYN) == 0)
1345 			goto drop;
1346 		tp->snd_wnd = th->th_win;	/* initial send window */
1347 
1348 		tp->irs = th->th_seq;
1349 		tcp_rcvseqinit(tp);
1350 		if (thflags & TH_ACK) {
1351 			tcpstat.tcps_connects++;
1352 			soisconnected(so);
1353 #ifdef MAC
1354 			SOCK_LOCK(so);
1355 			mac_set_socket_peer_from_mbuf(m, so);
1356 			SOCK_UNLOCK(so);
1357 #endif
1358 			/* Do window scaling on this connection? */
1359 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1360 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1361 				tp->snd_scale = tp->requested_s_scale;
1362 				tp->rcv_scale = tp->request_r_scale;
1363 			}
1364 			tp->rcv_adv += tp->rcv_wnd;
1365 			tp->snd_una++;		/* SYN is acked */
1366 			/*
1367 			 * If there's data, delay ACK; if there's also a FIN
1368 			 * ACKNOW will be turned on later.
1369 			 */
1370 			if (DELAY_ACK(tp) && tlen != 0)
1371 				callout_reset(tp->tt_delack, tcp_delacktime,
1372 				    tcp_timer_delack, tp);
1373 			else
1374 				tp->t_flags |= TF_ACKNOW;
1375 			/*
1376 			 * Received <SYN,ACK> in SYN_SENT[*] state.
1377 			 * Transitions:
1378 			 *	SYN_SENT  --> ESTABLISHED
1379 			 *	SYN_SENT* --> FIN_WAIT_1
1380 			 */
1381 			tp->t_starttime = ticks;
1382 			if (tp->t_flags & TF_NEEDFIN) {
1383 				tp->t_state = TCPS_FIN_WAIT_1;
1384 				tp->t_flags &= ~TF_NEEDFIN;
1385 				thflags &= ~TH_SYN;
1386 			} else {
1387 				tp->t_state = TCPS_ESTABLISHED;
1388 				callout_reset(tp->tt_keep, tcp_keepidle,
1389 					      tcp_timer_keep, tp);
1390 			}
1391 		} else {
1392 			/*
1393 			 * Received initial SYN in SYN-SENT[*] state =>
1394 			 * simultaneous open.  If segment contains CC option
1395 			 * and there is a cached CC, apply TAO test.
1396 			 * If it succeeds, connection is * half-synchronized.
1397 			 * Otherwise, do 3-way handshake:
1398 			 *        SYN-SENT -> SYN-RECEIVED
1399 			 *        SYN-SENT* -> SYN-RECEIVED*
1400 			 * If there was no CC option, clear cached CC value.
1401 			 */
1402 			tp->t_flags |= TF_ACKNOW;
1403 			callout_stop(tp->tt_rexmt);
1404 			tp->t_state = TCPS_SYN_RECEIVED;
1405 		}
1406 
1407 trimthenstep6:
1408 		KASSERT(headlocked,
1409 		    ("tcp_input(): trimthenstep6 head is not locked"));
1410 		INP_LOCK_ASSERT(inp);
1411 
1412 		/*
1413 		 * Advance th->th_seq to correspond to first data byte.
1414 		 * If data, trim to stay within window,
1415 		 * dropping FIN if necessary.
1416 		 */
1417 		th->th_seq++;
1418 		if (tlen > tp->rcv_wnd) {
1419 			todrop = tlen - tp->rcv_wnd;
1420 			m_adj(m, -todrop);
1421 			tlen = tp->rcv_wnd;
1422 			thflags &= ~TH_FIN;
1423 			tcpstat.tcps_rcvpackafterwin++;
1424 			tcpstat.tcps_rcvbyteafterwin += todrop;
1425 		}
1426 		tp->snd_wl1 = th->th_seq - 1;
1427 		tp->rcv_up = th->th_seq;
1428 		/*
1429 		 * Client side of transaction: already sent SYN and data.
1430 		 * If the remote host used T/TCP to validate the SYN,
1431 		 * our data will be ACK'd; if so, enter normal data segment
1432 		 * processing in the middle of step 5, ack processing.
1433 		 * Otherwise, goto step 6.
1434 		 */
1435 		if (thflags & TH_ACK)
1436 			goto process_ACK;
1437 
1438 		goto step6;
1439 
1440 	/*
1441 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1442 	 *      do normal processing.
1443 	 *
1444 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
1445 	 */
1446 	case TCPS_LAST_ACK:
1447 	case TCPS_CLOSING:
1448 	case TCPS_TIME_WAIT:
1449 		KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1450 		break;  /* continue normal processing */
1451 	}
1452 
1453 	/*
1454 	 * States other than LISTEN or SYN_SENT.
1455 	 * First check the RST flag and sequence number since reset segments
1456 	 * are exempt from the timestamp and connection count tests.  This
1457 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1458 	 * below which allowed reset segments in half the sequence space
1459 	 * to fall though and be processed (which gives forged reset
1460 	 * segments with a random sequence number a 50 percent chance of
1461 	 * killing a connection).
1462 	 * Then check timestamp, if present.
1463 	 * Then check the connection count, if present.
1464 	 * Then check that at least some bytes of segment are within
1465 	 * receive window.  If segment begins before rcv_nxt,
1466 	 * drop leading data (and SYN); if nothing left, just ack.
1467 	 *
1468 	 *
1469 	 * If the RST bit is set, check the sequence number to see
1470 	 * if this is a valid reset segment.
1471 	 * RFC 793 page 37:
1472 	 *   In all states except SYN-SENT, all reset (RST) segments
1473 	 *   are validated by checking their SEQ-fields.  A reset is
1474 	 *   valid if its sequence number is in the window.
1475 	 * Note: this does not take into account delayed ACKs, so
1476 	 *   we should test against last_ack_sent instead of rcv_nxt.
1477 	 *   The sequence number in the reset segment is normally an
1478 	 *   echo of our outgoing acknowlegement numbers, but some hosts
1479 	 *   send a reset with the sequence number at the rightmost edge
1480 	 *   of our receive window, and we have to handle this case.
1481 	 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1482 	 *   that brute force RST attacks are possible.  To combat this,
1483 	 *   we use a much stricter check while in the ESTABLISHED state,
1484 	 *   only accepting RSTs where the sequence number is equal to
1485 	 *   last_ack_sent.  In all other states (the states in which a
1486 	 *   RST is more likely), the more permissive check is used.
1487 	 * If we have multiple segments in flight, the intial reset
1488 	 * segment sequence numbers will be to the left of last_ack_sent,
1489 	 * but they will eventually catch up.
1490 	 * In any case, it never made sense to trim reset segments to
1491 	 * fit the receive window since RFC 1122 says:
1492 	 *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1493 	 *
1494 	 *    A TCP SHOULD allow a received RST segment to include data.
1495 	 *
1496 	 *    DISCUSSION
1497 	 *         It has been suggested that a RST segment could contain
1498 	 *         ASCII text that encoded and explained the cause of the
1499 	 *         RST.  No standard has yet been established for such
1500 	 *         data.
1501 	 *
1502 	 * If the reset segment passes the sequence number test examine
1503 	 * the state:
1504 	 *    SYN_RECEIVED STATE:
1505 	 *	If passive open, return to LISTEN state.
1506 	 *	If active open, inform user that connection was refused.
1507 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1508 	 *	Inform user that connection was reset, and close tcb.
1509 	 *    CLOSING, LAST_ACK STATES:
1510 	 *	Close the tcb.
1511 	 *    TIME_WAIT STATE:
1512 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
1513 	 *      RFC 1337.
1514 	 */
1515 	if (thflags & TH_RST) {
1516 		if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1517 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1518 			switch (tp->t_state) {
1519 
1520 			case TCPS_SYN_RECEIVED:
1521 				so->so_error = ECONNREFUSED;
1522 				goto close;
1523 
1524 			case TCPS_ESTABLISHED:
1525 				if (tp->last_ack_sent != th->th_seq) {
1526 					tcpstat.tcps_badrst++;
1527 					goto drop;
1528 				}
1529 			case TCPS_FIN_WAIT_1:
1530 			case TCPS_FIN_WAIT_2:
1531 			case TCPS_CLOSE_WAIT:
1532 				so->so_error = ECONNRESET;
1533 			close:
1534 				tp->t_state = TCPS_CLOSED;
1535 				tcpstat.tcps_drops++;
1536 				tp = tcp_close(tp);
1537 				break;
1538 
1539 			case TCPS_CLOSING:
1540 			case TCPS_LAST_ACK:
1541 				tp = tcp_close(tp);
1542 				break;
1543 
1544 			case TCPS_TIME_WAIT:
1545 				KASSERT(tp->t_state != TCPS_TIME_WAIT,
1546 				    ("timewait"));
1547 				break;
1548 			}
1549 		}
1550 		goto drop;
1551 	}
1552 
1553 	/*
1554 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1555 	 * and it's less than ts_recent, drop it.
1556 	 */
1557 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1558 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1559 
1560 		/* Check to see if ts_recent is over 24 days old.  */
1561 		if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1562 			/*
1563 			 * Invalidate ts_recent.  If this segment updates
1564 			 * ts_recent, the age will be reset later and ts_recent
1565 			 * will get a valid value.  If it does not, setting
1566 			 * ts_recent to zero will at least satisfy the
1567 			 * requirement that zero be placed in the timestamp
1568 			 * echo reply when ts_recent isn't valid.  The
1569 			 * age isn't reset until we get a valid ts_recent
1570 			 * because we don't want out-of-order segments to be
1571 			 * dropped when ts_recent is old.
1572 			 */
1573 			tp->ts_recent = 0;
1574 		} else {
1575 			tcpstat.tcps_rcvduppack++;
1576 			tcpstat.tcps_rcvdupbyte += tlen;
1577 			tcpstat.tcps_pawsdrop++;
1578 			if (tlen)
1579 				goto dropafterack;
1580 			goto drop;
1581 		}
1582 	}
1583 
1584 	/*
1585 	 * In the SYN-RECEIVED state, validate that the packet belongs to
1586 	 * this connection before trimming the data to fit the receive
1587 	 * window.  Check the sequence number versus IRS since we know
1588 	 * the sequence numbers haven't wrapped.  This is a partial fix
1589 	 * for the "LAND" DoS attack.
1590 	 */
1591 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1592 		rstreason = BANDLIM_RST_OPENPORT;
1593 		goto dropwithreset;
1594 	}
1595 
1596 	todrop = tp->rcv_nxt - th->th_seq;
1597 	if (todrop > 0) {
1598 		if (thflags & TH_SYN) {
1599 			thflags &= ~TH_SYN;
1600 			th->th_seq++;
1601 			if (th->th_urp > 1)
1602 				th->th_urp--;
1603 			else
1604 				thflags &= ~TH_URG;
1605 			todrop--;
1606 		}
1607 		/*
1608 		 * Following if statement from Stevens, vol. 2, p. 960.
1609 		 */
1610 		if (todrop > tlen
1611 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1612 			/*
1613 			 * Any valid FIN must be to the left of the window.
1614 			 * At this point the FIN must be a duplicate or out
1615 			 * of sequence; drop it.
1616 			 */
1617 			thflags &= ~TH_FIN;
1618 
1619 			/*
1620 			 * Send an ACK to resynchronize and drop any data.
1621 			 * But keep on processing for RST or ACK.
1622 			 */
1623 			tp->t_flags |= TF_ACKNOW;
1624 			todrop = tlen;
1625 			tcpstat.tcps_rcvduppack++;
1626 			tcpstat.tcps_rcvdupbyte += todrop;
1627 		} else {
1628 			tcpstat.tcps_rcvpartduppack++;
1629 			tcpstat.tcps_rcvpartdupbyte += todrop;
1630 		}
1631 		drop_hdrlen += todrop;	/* drop from the top afterwards */
1632 		th->th_seq += todrop;
1633 		tlen -= todrop;
1634 		if (th->th_urp > todrop)
1635 			th->th_urp -= todrop;
1636 		else {
1637 			thflags &= ~TH_URG;
1638 			th->th_urp = 0;
1639 		}
1640 	}
1641 
1642 	/*
1643 	 * If new data are received on a connection after the
1644 	 * user processes are gone, then RST the other end.
1645 	 */
1646 	if ((so->so_state & SS_NOFDREF) &&
1647 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1648 		tp = tcp_close(tp);
1649 		tcpstat.tcps_rcvafterclose++;
1650 		rstreason = BANDLIM_UNLIMITED;
1651 		goto dropwithreset;
1652 	}
1653 
1654 	/*
1655 	 * If segment ends after window, drop trailing data
1656 	 * (and PUSH and FIN); if nothing left, just ACK.
1657 	 */
1658 	todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1659 	if (todrop > 0) {
1660 		tcpstat.tcps_rcvpackafterwin++;
1661 		if (todrop >= tlen) {
1662 			tcpstat.tcps_rcvbyteafterwin += tlen;
1663 			/*
1664 			 * If a new connection request is received
1665 			 * while in TIME_WAIT, drop the old connection
1666 			 * and start over if the sequence numbers
1667 			 * are above the previous ones.
1668 			 */
1669 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1670 			if (thflags & TH_SYN &&
1671 			    tp->t_state == TCPS_TIME_WAIT &&
1672 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1673 				tp = tcp_close(tp);
1674 				goto findpcb;
1675 			}
1676 			/*
1677 			 * If window is closed can only take segments at
1678 			 * window edge, and have to drop data and PUSH from
1679 			 * incoming segments.  Continue processing, but
1680 			 * remember to ack.  Otherwise, drop segment
1681 			 * and ack.
1682 			 */
1683 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1684 				tp->t_flags |= TF_ACKNOW;
1685 				tcpstat.tcps_rcvwinprobe++;
1686 			} else
1687 				goto dropafterack;
1688 		} else
1689 			tcpstat.tcps_rcvbyteafterwin += todrop;
1690 		m_adj(m, -todrop);
1691 		tlen -= todrop;
1692 		thflags &= ~(TH_PUSH|TH_FIN);
1693 	}
1694 
1695 	/*
1696 	 * If last ACK falls within this segment's sequence numbers,
1697 	 * record its timestamp.
1698 	 * NOTE that the test is modified according to the latest
1699 	 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1700 	 */
1701 	if ((to.to_flags & TOF_TS) != 0 &&
1702 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1703 		tp->ts_recent_age = ticks;
1704 		tp->ts_recent = to.to_tsval;
1705 	}
1706 
1707 	/*
1708 	 * If a SYN is in the window, then this is an
1709 	 * error and we send an RST and drop the connection.
1710 	 */
1711 	if (thflags & TH_SYN) {
1712 		tp = tcp_drop(tp, ECONNRESET);
1713 		rstreason = BANDLIM_UNLIMITED;
1714 		goto drop;
1715 	}
1716 
1717 	/*
1718 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1719 	 * flag is on (half-synchronized state), then queue data for
1720 	 * later processing; else drop segment and return.
1721 	 */
1722 	if ((thflags & TH_ACK) == 0) {
1723 		if (tp->t_state == TCPS_SYN_RECEIVED ||
1724 		    (tp->t_flags & TF_NEEDSYN))
1725 			goto step6;
1726 		else
1727 			goto drop;
1728 	}
1729 
1730 	/*
1731 	 * Ack processing.
1732 	 */
1733 	switch (tp->t_state) {
1734 
1735 	/*
1736 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1737 	 * ESTABLISHED state and continue processing.
1738 	 * The ACK was checked above.
1739 	 */
1740 	case TCPS_SYN_RECEIVED:
1741 
1742 		tcpstat.tcps_connects++;
1743 		soisconnected(so);
1744 		/* Do window scaling? */
1745 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1746 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1747 			tp->snd_scale = tp->requested_s_scale;
1748 			tp->rcv_scale = tp->request_r_scale;
1749 		}
1750 		/*
1751 		 * Make transitions:
1752 		 *      SYN-RECEIVED  -> ESTABLISHED
1753 		 *      SYN-RECEIVED* -> FIN-WAIT-1
1754 		 */
1755 		tp->t_starttime = ticks;
1756 		if (tp->t_flags & TF_NEEDFIN) {
1757 			tp->t_state = TCPS_FIN_WAIT_1;
1758 			tp->t_flags &= ~TF_NEEDFIN;
1759 		} else {
1760 			tp->t_state = TCPS_ESTABLISHED;
1761 			callout_reset(tp->tt_keep, tcp_keepidle,
1762 				      tcp_timer_keep, tp);
1763 		}
1764 		/*
1765 		 * If segment contains data or ACK, will call tcp_reass()
1766 		 * later; if not, do so now to pass queued data to user.
1767 		 */
1768 		if (tlen == 0 && (thflags & TH_FIN) == 0)
1769 			(void) tcp_reass(tp, (struct tcphdr *)0, 0,
1770 			    (struct mbuf *)0);
1771 		tp->snd_wl1 = th->th_seq - 1;
1772 		/* FALLTHROUGH */
1773 
1774 	/*
1775 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1776 	 * ACKs.  If the ack is in the range
1777 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1778 	 * then advance tp->snd_una to th->th_ack and drop
1779 	 * data from the retransmission queue.  If this ACK reflects
1780 	 * more up to date window information we update our window information.
1781 	 */
1782 	case TCPS_ESTABLISHED:
1783 	case TCPS_FIN_WAIT_1:
1784 	case TCPS_FIN_WAIT_2:
1785 	case TCPS_CLOSE_WAIT:
1786 	case TCPS_CLOSING:
1787 	case TCPS_LAST_ACK:
1788 	case TCPS_TIME_WAIT:
1789 		KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1790 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1791 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1792 				tcpstat.tcps_rcvdupack++;
1793 				/*
1794 				 * If we have outstanding data (other than
1795 				 * a window probe), this is a completely
1796 				 * duplicate ack (ie, window info didn't
1797 				 * change), the ack is the biggest we've
1798 				 * seen and we've seen exactly our rexmt
1799 				 * threshhold of them, assume a packet
1800 				 * has been dropped and retransmit it.
1801 				 * Kludge snd_nxt & the congestion
1802 				 * window so we send only this one
1803 				 * packet.
1804 				 *
1805 				 * We know we're losing at the current
1806 				 * window size so do congestion avoidance
1807 				 * (set ssthresh to half the current window
1808 				 * and pull our congestion window back to
1809 				 * the new ssthresh).
1810 				 *
1811 				 * Dup acks mean that packets have left the
1812 				 * network (they're now cached at the receiver)
1813 				 * so bump cwnd by the amount in the receiver
1814 				 * to keep a constant cwnd packets in the
1815 				 * network.
1816 				 */
1817 				if (!callout_active(tp->tt_rexmt) ||
1818 				    th->th_ack != tp->snd_una)
1819 					tp->t_dupacks = 0;
1820 				else if (++tp->t_dupacks > tcprexmtthresh ||
1821 					 ((tcp_do_newreno || tp->sack_enable) &&
1822 					  IN_FASTRECOVERY(tp))) {
1823 					tp->snd_cwnd += tp->t_maxseg;
1824 					(void) tcp_output(tp);
1825 					goto drop;
1826 				} else if (tp->t_dupacks == tcprexmtthresh) {
1827 					tcp_seq onxt = tp->snd_nxt;
1828 					u_int win;
1829 
1830 					/*
1831 					 * If we're doing sack, check to
1832 					 * see if we're already in sack
1833 					 * recovery. If we're not doing sack,
1834 					 * check to see if we're in newreno
1835 					 * recovery.
1836 					 */
1837 					if (tp->sack_enable) {
1838 						if (IN_FASTRECOVERY(tp)) {
1839 							tp->t_dupacks = 0;
1840 							break;
1841 						}
1842 					} else if (tcp_do_newreno) {
1843 						if (SEQ_LEQ(th->th_ack,
1844 						    tp->snd_recover)) {
1845 							tp->t_dupacks = 0;
1846 							break;
1847 						}
1848 					}
1849 					win = min(tp->snd_wnd, tp->snd_cwnd) /
1850 					    2 / tp->t_maxseg;
1851 					if (win < 2)
1852 						win = 2;
1853 					tp->snd_ssthresh = win * tp->t_maxseg;
1854 					ENTER_FASTRECOVERY(tp);
1855 					tp->snd_recover = tp->snd_max;
1856 					callout_stop(tp->tt_rexmt);
1857 					tp->t_rtttime = 0;
1858 					if (tp->sack_enable) {
1859 						tcpstat.tcps_sack_recovery_episode++;
1860 						tp->sack_newdata = tp->snd_nxt;
1861 						tp->snd_cwnd =
1862 							tp->t_maxseg * tcp_sack_recovery_initburst;
1863 						(void) tcp_output(tp);
1864 						tp->snd_cwnd +=
1865 						    tp->snd_ssthresh;
1866 						goto drop;
1867 					}
1868 
1869 					tp->snd_nxt = th->th_ack;
1870 					tp->snd_cwnd = tp->t_maxseg;
1871 					(void) tcp_output(tp);
1872 					KASSERT(tp->snd_limited <= 2,
1873 					    ("tp->snd_limited too big"));
1874 					tp->snd_cwnd = tp->snd_ssthresh +
1875 					     tp->t_maxseg *
1876 					     (tp->t_dupacks - tp->snd_limited);
1877 					if (SEQ_GT(onxt, tp->snd_nxt))
1878 						tp->snd_nxt = onxt;
1879 					goto drop;
1880 				} else if (tcp_do_rfc3042) {
1881 					u_long oldcwnd = tp->snd_cwnd;
1882 					tcp_seq oldsndmax = tp->snd_max;
1883 					u_int sent;
1884 
1885 					KASSERT(tp->t_dupacks == 1 ||
1886 					    tp->t_dupacks == 2,
1887 					    ("dupacks not 1 or 2"));
1888 					if (tp->t_dupacks == 1)
1889 						tp->snd_limited = 0;
1890 					tp->snd_cwnd =
1891 					    (tp->snd_nxt - tp->snd_una) +
1892 					    (tp->t_dupacks - tp->snd_limited) *
1893 					    tp->t_maxseg;
1894 					(void) tcp_output(tp);
1895 					sent = tp->snd_max - oldsndmax;
1896 					if (sent > tp->t_maxseg) {
1897 						KASSERT((tp->t_dupacks == 2 &&
1898 						    tp->snd_limited == 0) ||
1899 						   (sent == tp->t_maxseg + 1 &&
1900 						    tp->t_flags & TF_SENTFIN),
1901 						    ("sent too much"));
1902 						tp->snd_limited = 2;
1903 					} else if (sent > 0)
1904 						++tp->snd_limited;
1905 					tp->snd_cwnd = oldcwnd;
1906 					goto drop;
1907 				}
1908 			} else
1909 				tp->t_dupacks = 0;
1910 			break;
1911 		}
1912 
1913 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1914 
1915 		/*
1916 		 * If the congestion window was inflated to account
1917 		 * for the other side's cached packets, retract it.
1918 		 */
1919 		if (tcp_do_newreno || tp->sack_enable) {
1920 			if (IN_FASTRECOVERY(tp)) {
1921 				if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1922 					if (tp->sack_enable)
1923 						tcp_sack_partialack(tp, th);
1924 					else
1925 						tcp_newreno_partial_ack(tp, th);
1926 				} else {
1927 					/*
1928 					 * Out of fast recovery.
1929 					 * Window inflation should have left us
1930 					 * with approximately snd_ssthresh
1931 					 * outstanding data.
1932 					 * But in case we would be inclined to
1933 					 * send a burst, better to do it via
1934 					 * the slow start mechanism.
1935 					 */
1936 					if (SEQ_GT(th->th_ack +
1937 							tp->snd_ssthresh,
1938 						   tp->snd_max))
1939 						tp->snd_cwnd = tp->snd_max -
1940 								th->th_ack +
1941 								tp->t_maxseg;
1942 					else
1943 						tp->snd_cwnd = tp->snd_ssthresh;
1944 				}
1945 			}
1946 		} else {
1947 			if (tp->t_dupacks >= tcprexmtthresh &&
1948 			    tp->snd_cwnd > tp->snd_ssthresh)
1949 				tp->snd_cwnd = tp->snd_ssthresh;
1950 		}
1951 		tp->t_dupacks = 0;
1952 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1953 			tcpstat.tcps_rcvacktoomuch++;
1954 			goto dropafterack;
1955 		}
1956 		/*
1957 		 * If we reach this point, ACK is not a duplicate,
1958 		 *     i.e., it ACKs something we sent.
1959 		 */
1960 		if (tp->t_flags & TF_NEEDSYN) {
1961 			/*
1962 			 * T/TCP: Connection was half-synchronized, and our
1963 			 * SYN has been ACK'd (so connection is now fully
1964 			 * synchronized).  Go to non-starred state,
1965 			 * increment snd_una for ACK of SYN, and check if
1966 			 * we can do window scaling.
1967 			 */
1968 			tp->t_flags &= ~TF_NEEDSYN;
1969 			tp->snd_una++;
1970 			/* Do window scaling? */
1971 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1972 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1973 				tp->snd_scale = tp->requested_s_scale;
1974 				tp->rcv_scale = tp->request_r_scale;
1975 			}
1976 		}
1977 
1978 process_ACK:
1979 		KASSERT(headlocked,
1980 		    ("tcp_input(): process_ACK head is not locked"));
1981 		INP_LOCK_ASSERT(inp);
1982 
1983 		acked = th->th_ack - tp->snd_una;
1984 		tcpstat.tcps_rcvackpack++;
1985 		tcpstat.tcps_rcvackbyte += acked;
1986 
1987 		/*
1988 		 * If we just performed our first retransmit, and the ACK
1989 		 * arrives within our recovery window, then it was a mistake
1990 		 * to do the retransmit in the first place.  Recover our
1991 		 * original cwnd and ssthresh, and proceed to transmit where
1992 		 * we left off.
1993 		 */
1994 		if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1995 			++tcpstat.tcps_sndrexmitbad;
1996 			tp->snd_cwnd = tp->snd_cwnd_prev;
1997 			tp->snd_ssthresh = tp->snd_ssthresh_prev;
1998 			tp->snd_recover = tp->snd_recover_prev;
1999 			if (tp->t_flags & TF_WASFRECOVERY)
2000 				ENTER_FASTRECOVERY(tp);
2001 			tp->snd_nxt = tp->snd_max;
2002 			tp->t_badrxtwin = 0;	/* XXX probably not required */
2003 		}
2004 
2005 		/*
2006 		 * If we have a timestamp reply, update smoothed
2007 		 * round trip time.  If no timestamp is present but
2008 		 * transmit timer is running and timed sequence
2009 		 * number was acked, update smoothed round trip time.
2010 		 * Since we now have an rtt measurement, cancel the
2011 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2012 		 * Recompute the initial retransmit timer.
2013 		 *
2014 		 * Some boxes send broken timestamp replies
2015 		 * during the SYN+ACK phase, ignore
2016 		 * timestamps of 0 or we could calculate a
2017 		 * huge RTT and blow up the retransmit timer.
2018 		 */
2019 		if ((to.to_flags & TOF_TS) != 0 &&
2020 		    to.to_tsecr) {
2021 			tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2022 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2023 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2024 		}
2025 		tcp_xmit_bandwidth_limit(tp, th->th_ack);
2026 
2027 		/*
2028 		 * If all outstanding data is acked, stop retransmit
2029 		 * timer and remember to restart (more output or persist).
2030 		 * If there is more data to be acked, restart retransmit
2031 		 * timer, using current (possibly backed-off) value.
2032 		 */
2033 		if (th->th_ack == tp->snd_max) {
2034 			callout_stop(tp->tt_rexmt);
2035 			needoutput = 1;
2036 		} else if (!callout_active(tp->tt_persist))
2037 			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2038 				      tcp_timer_rexmt, tp);
2039 
2040 		/*
2041 		 * If no data (only SYN) was ACK'd,
2042 		 *    skip rest of ACK processing.
2043 		 */
2044 		if (acked == 0)
2045 			goto step6;
2046 
2047 		/*
2048 		 * When new data is acked, open the congestion window.
2049 		 * If the window gives us less than ssthresh packets
2050 		 * in flight, open exponentially (maxseg per packet).
2051 		 * Otherwise open linearly: maxseg per window
2052 		 * (maxseg^2 / cwnd per packet).
2053 		 */
2054 		if ((!tcp_do_newreno && !tp->sack_enable) ||
2055 		    !IN_FASTRECOVERY(tp)) {
2056 			register u_int cw = tp->snd_cwnd;
2057 			register u_int incr = tp->t_maxseg;
2058 			if (cw > tp->snd_ssthresh)
2059 				incr = incr * incr / cw;
2060 			tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2061 		}
2062 		SOCKBUF_LOCK(&so->so_snd);
2063 		if (acked > so->so_snd.sb_cc) {
2064 			tp->snd_wnd -= so->so_snd.sb_cc;
2065 			sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2066 			ourfinisacked = 1;
2067 		} else {
2068 			sbdrop_locked(&so->so_snd, acked);
2069 			tp->snd_wnd -= acked;
2070 			ourfinisacked = 0;
2071 		}
2072 		sowwakeup_locked(so);
2073 		/* detect una wraparound */
2074 		if ((tcp_do_newreno || tp->sack_enable) &&
2075 		    !IN_FASTRECOVERY(tp) &&
2076 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2077 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2078 			tp->snd_recover = th->th_ack - 1;
2079 		if ((tcp_do_newreno || tp->sack_enable) &&
2080 		    IN_FASTRECOVERY(tp) &&
2081 		    SEQ_GEQ(th->th_ack, tp->snd_recover))
2082 			EXIT_FASTRECOVERY(tp);
2083 		tp->snd_una = th->th_ack;
2084 		if (tp->sack_enable) {
2085 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2086 				tp->snd_recover = tp->snd_una;
2087 		}
2088 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2089 			tp->snd_nxt = tp->snd_una;
2090 
2091 		switch (tp->t_state) {
2092 
2093 		/*
2094 		 * In FIN_WAIT_1 STATE in addition to the processing
2095 		 * for the ESTABLISHED state if our FIN is now acknowledged
2096 		 * then enter FIN_WAIT_2.
2097 		 */
2098 		case TCPS_FIN_WAIT_1:
2099 			if (ourfinisacked) {
2100 				/*
2101 				 * If we can't receive any more
2102 				 * data, then closing user can proceed.
2103 				 * Starting the timer is contrary to the
2104 				 * specification, but if we don't get a FIN
2105 				 * we'll hang forever.
2106 				 */
2107 		/* XXXjl
2108 		 * we should release the tp also, and use a
2109 		 * compressed state.
2110 		 */
2111 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2112 					soisdisconnected(so);
2113 					callout_reset(tp->tt_2msl, tcp_maxidle,
2114 						      tcp_timer_2msl, tp);
2115 				}
2116 				tp->t_state = TCPS_FIN_WAIT_2;
2117 			}
2118 			break;
2119 
2120 		/*
2121 		 * In CLOSING STATE in addition to the processing for
2122 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2123 		 * then enter the TIME-WAIT state, otherwise ignore
2124 		 * the segment.
2125 		 */
2126 		case TCPS_CLOSING:
2127 			if (ourfinisacked) {
2128 				KASSERT(headlocked, ("headlocked"));
2129 				tcp_twstart(tp);
2130 				INP_INFO_WUNLOCK(&tcbinfo);
2131 				m_freem(m);
2132 				return;
2133 			}
2134 			break;
2135 
2136 		/*
2137 		 * In LAST_ACK, we may still be waiting for data to drain
2138 		 * and/or to be acked, as well as for the ack of our FIN.
2139 		 * If our FIN is now acknowledged, delete the TCB,
2140 		 * enter the closed state and return.
2141 		 */
2142 		case TCPS_LAST_ACK:
2143 			if (ourfinisacked) {
2144 				tp = tcp_close(tp);
2145 				goto drop;
2146 			}
2147 			break;
2148 
2149 		/*
2150 		 * In TIME_WAIT state the only thing that should arrive
2151 		 * is a retransmission of the remote FIN.  Acknowledge
2152 		 * it and restart the finack timer.
2153 		 */
2154 		case TCPS_TIME_WAIT:
2155 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2156 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
2157 				      tcp_timer_2msl, tp);
2158 			goto dropafterack;
2159 		}
2160 	}
2161 
2162 step6:
2163 	KASSERT(headlocked, ("tcp_input(): step6 head is not locked"));
2164 	INP_LOCK_ASSERT(inp);
2165 
2166 	/*
2167 	 * Update window information.
2168 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2169 	 */
2170 	if ((thflags & TH_ACK) &&
2171 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2172 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2173 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2174 		/* keep track of pure window updates */
2175 		if (tlen == 0 &&
2176 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2177 			tcpstat.tcps_rcvwinupd++;
2178 		tp->snd_wnd = tiwin;
2179 		tp->snd_wl1 = th->th_seq;
2180 		tp->snd_wl2 = th->th_ack;
2181 		if (tp->snd_wnd > tp->max_sndwnd)
2182 			tp->max_sndwnd = tp->snd_wnd;
2183 		needoutput = 1;
2184 	}
2185 
2186 	/*
2187 	 * Process segments with URG.
2188 	 */
2189 	if ((thflags & TH_URG) && th->th_urp &&
2190 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2191 		/*
2192 		 * This is a kludge, but if we receive and accept
2193 		 * random urgent pointers, we'll crash in
2194 		 * soreceive.  It's hard to imagine someone
2195 		 * actually wanting to send this much urgent data.
2196 		 */
2197 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2198 			th->th_urp = 0;			/* XXX */
2199 			thflags &= ~TH_URG;		/* XXX */
2200 			goto dodata;			/* XXX */
2201 		}
2202 		/*
2203 		 * If this segment advances the known urgent pointer,
2204 		 * then mark the data stream.  This should not happen
2205 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2206 		 * a FIN has been received from the remote side.
2207 		 * In these states we ignore the URG.
2208 		 *
2209 		 * According to RFC961 (Assigned Protocols),
2210 		 * the urgent pointer points to the last octet
2211 		 * of urgent data.  We continue, however,
2212 		 * to consider it to indicate the first octet
2213 		 * of data past the urgent section as the original
2214 		 * spec states (in one of two places).
2215 		 */
2216 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2217 			tp->rcv_up = th->th_seq + th->th_urp;
2218 			SOCKBUF_LOCK(&so->so_rcv);
2219 			so->so_oobmark = so->so_rcv.sb_cc +
2220 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2221 			if (so->so_oobmark == 0)
2222 				so->so_rcv.sb_state |= SBS_RCVATMARK;
2223 			SOCKBUF_UNLOCK(&so->so_rcv);
2224 			sohasoutofband(so);
2225 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2226 		}
2227 		/*
2228 		 * Remove out of band data so doesn't get presented to user.
2229 		 * This can happen independent of advancing the URG pointer,
2230 		 * but if two URG's are pending at once, some out-of-band
2231 		 * data may creep in... ick.
2232 		 */
2233 		if (th->th_urp <= (u_long)tlen &&
2234 		    !(so->so_options & SO_OOBINLINE)) {
2235 			/* hdr drop is delayed */
2236 			tcp_pulloutofband(so, th, m, drop_hdrlen);
2237 		}
2238 	} else {
2239 		/*
2240 		 * If no out of band data is expected,
2241 		 * pull receive urgent pointer along
2242 		 * with the receive window.
2243 		 */
2244 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2245 			tp->rcv_up = tp->rcv_nxt;
2246 	}
2247 dodata:							/* XXX */
2248 	KASSERT(headlocked, ("tcp_input(): dodata head is not locked"));
2249 	INP_LOCK_ASSERT(inp);
2250 
2251 	/*
2252 	 * Process the segment text, merging it into the TCP sequencing queue,
2253 	 * and arranging for acknowledgment of receipt if necessary.
2254 	 * This process logically involves adjusting tp->rcv_wnd as data
2255 	 * is presented to the user (this happens in tcp_usrreq.c,
2256 	 * case PRU_RCVD).  If a FIN has already been received on this
2257 	 * connection then we just ignore the text.
2258 	 */
2259 	if ((tlen || (thflags & TH_FIN)) &&
2260 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2261 		m_adj(m, drop_hdrlen);	/* delayed header drop */
2262 		/*
2263 		 * Insert segment which includes th into TCP reassembly queue
2264 		 * with control block tp.  Set thflags to whether reassembly now
2265 		 * includes a segment with FIN.  This handles the common case
2266 		 * inline (segment is the next to be received on an established
2267 		 * connection, and the queue is empty), avoiding linkage into
2268 		 * and removal from the queue and repetition of various
2269 		 * conversions.
2270 		 * Set DELACK for segments received in order, but ack
2271 		 * immediately when segments are out of order (so
2272 		 * fast retransmit can work).
2273 		 */
2274 		if (th->th_seq == tp->rcv_nxt &&
2275 		    LIST_EMPTY(&tp->t_segq) &&
2276 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
2277 			if (DELAY_ACK(tp))
2278 				tp->t_flags |= TF_DELACK;
2279 			else
2280 				tp->t_flags |= TF_ACKNOW;
2281 			tp->rcv_nxt += tlen;
2282 			thflags = th->th_flags & TH_FIN;
2283 			tcpstat.tcps_rcvpack++;
2284 			tcpstat.tcps_rcvbyte += tlen;
2285 			ND6_HINT(tp);
2286 			/* Unlocked read. */
2287 			SOCKBUF_LOCK(&so->so_rcv);
2288 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2289 				m_freem(m);
2290 			else
2291 				sbappendstream_locked(&so->so_rcv, m);
2292 			sorwakeup_locked(so);
2293 		} else {
2294 			thflags = tcp_reass(tp, th, &tlen, m);
2295 			tp->t_flags |= TF_ACKNOW;
2296 		}
2297 			if (tp->sack_enable)
2298 				tcp_update_sack_list(tp);
2299 		/*
2300 		 * Note the amount of data that peer has sent into
2301 		 * our window, in order to estimate the sender's
2302 		 * buffer size.
2303 		 */
2304 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2305 	} else {
2306 		m_freem(m);
2307 		thflags &= ~TH_FIN;
2308 	}
2309 
2310 	/*
2311 	 * If FIN is received ACK the FIN and let the user know
2312 	 * that the connection is closing.
2313 	 */
2314 	if (thflags & TH_FIN) {
2315 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2316 			socantrcvmore(so);
2317 			/*
2318 			 * If connection is half-synchronized
2319 			 * (ie NEEDSYN flag on) then delay ACK,
2320 			 * so it may be piggybacked when SYN is sent.
2321 			 * Otherwise, since we received a FIN then no
2322 			 * more input can be expected, send ACK now.
2323 			 */
2324 			if (tp->t_flags & TF_NEEDSYN)
2325 				tp->t_flags |= TF_DELACK;
2326 			else
2327 				tp->t_flags |= TF_ACKNOW;
2328 			tp->rcv_nxt++;
2329 		}
2330 		switch (tp->t_state) {
2331 
2332 		/*
2333 		 * In SYN_RECEIVED and ESTABLISHED STATES
2334 		 * enter the CLOSE_WAIT state.
2335 		 */
2336 		case TCPS_SYN_RECEIVED:
2337 			tp->t_starttime = ticks;
2338 			/*FALLTHROUGH*/
2339 		case TCPS_ESTABLISHED:
2340 			tp->t_state = TCPS_CLOSE_WAIT;
2341 			break;
2342 
2343 		/*
2344 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2345 		 * enter the CLOSING state.
2346 		 */
2347 		case TCPS_FIN_WAIT_1:
2348 			tp->t_state = TCPS_CLOSING;
2349 			break;
2350 
2351 		/*
2352 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2353 		 * starting the time-wait timer, turning off the other
2354 		 * standard timers.
2355 		 */
2356 		case TCPS_FIN_WAIT_2:
2357 			KASSERT(headlocked == 1, ("headlocked should be 1"));
2358 			tcp_twstart(tp);
2359 			INP_INFO_WUNLOCK(&tcbinfo);
2360 			return;
2361 
2362 		/*
2363 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2364 		 */
2365 		case TCPS_TIME_WAIT:
2366 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2367 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
2368 				      tcp_timer_2msl, tp);
2369 			break;
2370 		}
2371 	}
2372 #ifdef TCPDEBUG
2373 	if (so->so_options & SO_DEBUG)
2374 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2375 			  &tcp_savetcp, 0);
2376 #endif
2377 
2378 	/*
2379 	 * Return any desired output.
2380 	 */
2381 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2382 		(void) tcp_output(tp);
2383 
2384 check_delack:
2385 	KASSERT(headlocked == 1, ("headlocked should be 1"));
2386 	INP_LOCK_ASSERT(inp);
2387 	if (tp->t_flags & TF_DELACK) {
2388 		tp->t_flags &= ~TF_DELACK;
2389 		callout_reset(tp->tt_delack, tcp_delacktime,
2390 		    tcp_timer_delack, tp);
2391 	}
2392 	INP_UNLOCK(inp);
2393 	INP_INFO_WUNLOCK(&tcbinfo);
2394 	return;
2395 
2396 dropafterack:
2397 	/*
2398 	 * Generate an ACK dropping incoming segment if it occupies
2399 	 * sequence space, where the ACK reflects our state.
2400 	 *
2401 	 * We can now skip the test for the RST flag since all
2402 	 * paths to this code happen after packets containing
2403 	 * RST have been dropped.
2404 	 *
2405 	 * In the SYN-RECEIVED state, don't send an ACK unless the
2406 	 * segment we received passes the SYN-RECEIVED ACK test.
2407 	 * If it fails send a RST.  This breaks the loop in the
2408 	 * "LAND" DoS attack, and also prevents an ACK storm
2409 	 * between two listening ports that have been sent forged
2410 	 * SYN segments, each with the source address of the other.
2411 	 */
2412 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2413 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
2414 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
2415 		rstreason = BANDLIM_RST_OPENPORT;
2416 		goto dropwithreset;
2417 	}
2418 #ifdef TCPDEBUG
2419 	if (so->so_options & SO_DEBUG)
2420 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2421 			  &tcp_savetcp, 0);
2422 #endif
2423 	KASSERT(headlocked, ("headlocked should be 1"));
2424 	tp->t_flags |= TF_ACKNOW;
2425 	(void) tcp_output(tp);
2426 	INP_UNLOCK(inp);
2427 	INP_INFO_WUNLOCK(&tcbinfo);
2428 	m_freem(m);
2429 	return;
2430 
2431 dropwithreset:
2432 	/*
2433 	 * Generate a RST, dropping incoming segment.
2434 	 * Make ACK acceptable to originator of segment.
2435 	 * Don't bother to respond if destination was broadcast/multicast.
2436 	 */
2437 	if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2438 		goto drop;
2439 	if (isipv6) {
2440 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2441 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2442 			goto drop;
2443 	} else {
2444 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2445 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2446 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2447 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2448 			goto drop;
2449 	}
2450 	/* IPv6 anycast check is done at tcp6_input() */
2451 
2452 	/*
2453 	 * Perform bandwidth limiting.
2454 	 */
2455 	if (badport_bandlim(rstreason) < 0)
2456 		goto drop;
2457 
2458 #ifdef TCPDEBUG
2459 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2460 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2461 			  &tcp_savetcp, 0);
2462 #endif
2463 
2464 	if (thflags & TH_ACK)
2465 		/* mtod() below is safe as long as hdr dropping is delayed */
2466 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2467 			    TH_RST);
2468 	else {
2469 		if (thflags & TH_SYN)
2470 			tlen++;
2471 		/* mtod() below is safe as long as hdr dropping is delayed */
2472 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2473 			    (tcp_seq)0, TH_RST|TH_ACK);
2474 	}
2475 
2476 	if (tp)
2477 		INP_UNLOCK(inp);
2478 	if (headlocked)
2479 		INP_INFO_WUNLOCK(&tcbinfo);
2480 	return;
2481 
2482 drop:
2483 	/*
2484 	 * Drop space held by incoming segment and return.
2485 	 */
2486 #ifdef TCPDEBUG
2487 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2488 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2489 			  &tcp_savetcp, 0);
2490 #endif
2491 	if (tp)
2492 		INP_UNLOCK(inp);
2493 	if (headlocked)
2494 		INP_INFO_WUNLOCK(&tcbinfo);
2495 	m_freem(m);
2496 	return;
2497 }
2498 
2499 /*
2500  * Parse TCP options and place in tcpopt.
2501  */
2502 static void
2503 tcp_dooptions(tp, to, cp, cnt, is_syn, th)
2504 	struct tcpcb *tp;
2505 	struct tcpopt *to;
2506 	u_char *cp;
2507 	int cnt;
2508 	int is_syn;
2509 	struct tcphdr *th;
2510 {
2511 	int opt, optlen;
2512 
2513 	to->to_flags = 0;
2514 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2515 		opt = cp[0];
2516 		if (opt == TCPOPT_EOL)
2517 			break;
2518 		if (opt == TCPOPT_NOP)
2519 			optlen = 1;
2520 		else {
2521 			if (cnt < 2)
2522 				break;
2523 			optlen = cp[1];
2524 			if (optlen < 2 || optlen > cnt)
2525 				break;
2526 		}
2527 		switch (opt) {
2528 		case TCPOPT_MAXSEG:
2529 			if (optlen != TCPOLEN_MAXSEG)
2530 				continue;
2531 			if (!is_syn)
2532 				continue;
2533 			to->to_flags |= TOF_MSS;
2534 			bcopy((char *)cp + 2,
2535 			    (char *)&to->to_mss, sizeof(to->to_mss));
2536 			to->to_mss = ntohs(to->to_mss);
2537 			break;
2538 		case TCPOPT_WINDOW:
2539 			if (optlen != TCPOLEN_WINDOW)
2540 				continue;
2541 			if (! is_syn)
2542 				continue;
2543 			to->to_flags |= TOF_SCALE;
2544 			to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2545 			break;
2546 		case TCPOPT_TIMESTAMP:
2547 			if (optlen != TCPOLEN_TIMESTAMP)
2548 				continue;
2549 			to->to_flags |= TOF_TS;
2550 			bcopy((char *)cp + 2,
2551 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
2552 			to->to_tsval = ntohl(to->to_tsval);
2553 			bcopy((char *)cp + 6,
2554 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2555 			to->to_tsecr = ntohl(to->to_tsecr);
2556 			break;
2557 #ifdef TCP_SIGNATURE
2558 		/*
2559 		 * XXX In order to reply to a host which has set the
2560 		 * TCP_SIGNATURE option in its initial SYN, we have to
2561 		 * record the fact that the option was observed here
2562 		 * for the syncache code to perform the correct response.
2563 		 */
2564 		case TCPOPT_SIGNATURE:
2565 			if (optlen != TCPOLEN_SIGNATURE)
2566 				continue;
2567 			to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2568 			break;
2569 #endif
2570 		case TCPOPT_SACK_PERMITTED:
2571 			if (!tcp_do_sack ||
2572 			    optlen != TCPOLEN_SACK_PERMITTED)
2573 				continue;
2574 			if (is_syn) {
2575 				/* MUST only be set on SYN */
2576 				to->to_flags |= TOF_SACK;
2577 			}
2578 			break;
2579 
2580 		case TCPOPT_SACK:
2581 			if (!tp || tcp_sack_option(tp, th, cp, optlen))
2582 				continue;
2583 			break;
2584 		default:
2585 			continue;
2586 		}
2587 	}
2588 }
2589 
2590 /*
2591  * Pull out of band byte out of a segment so
2592  * it doesn't appear in the user's data queue.
2593  * It is still reflected in the segment length for
2594  * sequencing purposes.
2595  */
2596 static void
2597 tcp_pulloutofband(so, th, m, off)
2598 	struct socket *so;
2599 	struct tcphdr *th;
2600 	register struct mbuf *m;
2601 	int off;		/* delayed to be droped hdrlen */
2602 {
2603 	int cnt = off + th->th_urp - 1;
2604 
2605 	while (cnt >= 0) {
2606 		if (m->m_len > cnt) {
2607 			char *cp = mtod(m, caddr_t) + cnt;
2608 			struct tcpcb *tp = sototcpcb(so);
2609 
2610 			tp->t_iobc = *cp;
2611 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2612 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2613 			m->m_len--;
2614 			if (m->m_flags & M_PKTHDR)
2615 				m->m_pkthdr.len--;
2616 			return;
2617 		}
2618 		cnt -= m->m_len;
2619 		m = m->m_next;
2620 		if (m == 0)
2621 			break;
2622 	}
2623 	panic("tcp_pulloutofband");
2624 }
2625 
2626 /*
2627  * Collect new round-trip time estimate
2628  * and update averages and current timeout.
2629  */
2630 static void
2631 tcp_xmit_timer(tp, rtt)
2632 	register struct tcpcb *tp;
2633 	int rtt;
2634 {
2635 	register int delta;
2636 
2637 	tcpstat.tcps_rttupdated++;
2638 	tp->t_rttupdated++;
2639 	if (tp->t_srtt != 0) {
2640 		/*
2641 		 * srtt is stored as fixed point with 5 bits after the
2642 		 * binary point (i.e., scaled by 8).  The following magic
2643 		 * is equivalent to the smoothing algorithm in rfc793 with
2644 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2645 		 * point).  Adjust rtt to origin 0.
2646 		 */
2647 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2648 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2649 
2650 		if ((tp->t_srtt += delta) <= 0)
2651 			tp->t_srtt = 1;
2652 
2653 		/*
2654 		 * We accumulate a smoothed rtt variance (actually, a
2655 		 * smoothed mean difference), then set the retransmit
2656 		 * timer to smoothed rtt + 4 times the smoothed variance.
2657 		 * rttvar is stored as fixed point with 4 bits after the
2658 		 * binary point (scaled by 16).  The following is
2659 		 * equivalent to rfc793 smoothing with an alpha of .75
2660 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2661 		 * rfc793's wired-in beta.
2662 		 */
2663 		if (delta < 0)
2664 			delta = -delta;
2665 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2666 		if ((tp->t_rttvar += delta) <= 0)
2667 			tp->t_rttvar = 1;
2668 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2669 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2670 	} else {
2671 		/*
2672 		 * No rtt measurement yet - use the unsmoothed rtt.
2673 		 * Set the variance to half the rtt (so our first
2674 		 * retransmit happens at 3*rtt).
2675 		 */
2676 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
2677 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2678 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2679 	}
2680 	tp->t_rtttime = 0;
2681 	tp->t_rxtshift = 0;
2682 
2683 	/*
2684 	 * the retransmit should happen at rtt + 4 * rttvar.
2685 	 * Because of the way we do the smoothing, srtt and rttvar
2686 	 * will each average +1/2 tick of bias.  When we compute
2687 	 * the retransmit timer, we want 1/2 tick of rounding and
2688 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2689 	 * firing of the timer.  The bias will give us exactly the
2690 	 * 1.5 tick we need.  But, because the bias is
2691 	 * statistical, we have to test that we don't drop below
2692 	 * the minimum feasible timer (which is 2 ticks).
2693 	 */
2694 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2695 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2696 
2697 	/*
2698 	 * We received an ack for a packet that wasn't retransmitted;
2699 	 * it is probably safe to discard any error indications we've
2700 	 * received recently.  This isn't quite right, but close enough
2701 	 * for now (a route might have failed after we sent a segment,
2702 	 * and the return path might not be symmetrical).
2703 	 */
2704 	tp->t_softerror = 0;
2705 }
2706 
2707 /*
2708  * Determine a reasonable value for maxseg size.
2709  * If the route is known, check route for mtu.
2710  * If none, use an mss that can be handled on the outgoing
2711  * interface without forcing IP to fragment; if bigger than
2712  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2713  * to utilize large mbufs.  If no route is found, route has no mtu,
2714  * or the destination isn't local, use a default, hopefully conservative
2715  * size (usually 512 or the default IP max size, but no more than the mtu
2716  * of the interface), as we can't discover anything about intervening
2717  * gateways or networks.  We also initialize the congestion/slow start
2718  * window to be a single segment if the destination isn't local.
2719  * While looking at the routing entry, we also initialize other path-dependent
2720  * parameters from pre-set or cached values in the routing entry.
2721  *
2722  * Also take into account the space needed for options that we
2723  * send regularly.  Make maxseg shorter by that amount to assure
2724  * that we can send maxseg amount of data even when the options
2725  * are present.  Store the upper limit of the length of options plus
2726  * data in maxopd.
2727  *
2728  *
2729  * In case of T/TCP, we call this routine during implicit connection
2730  * setup as well (offer = -1), to initialize maxseg from the cached
2731  * MSS of our peer.
2732  *
2733  * NOTE that this routine is only called when we process an incoming
2734  * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2735  */
2736 void
2737 tcp_mss(tp, offer)
2738 	struct tcpcb *tp;
2739 	int offer;
2740 {
2741 	int rtt, mss;
2742 	u_long bufsize;
2743 	u_long maxmtu;
2744 	struct inpcb *inp = tp->t_inpcb;
2745 	struct socket *so;
2746 	struct hc_metrics_lite metrics;
2747 	int origoffer = offer;
2748 #ifdef INET6
2749 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2750 	size_t min_protoh = isipv6 ?
2751 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2752 			    sizeof (struct tcpiphdr);
2753 #else
2754 	const size_t min_protoh = sizeof(struct tcpiphdr);
2755 #endif
2756 
2757 	/* initialize */
2758 #ifdef INET6
2759 	if (isipv6) {
2760 		maxmtu = tcp_maxmtu6(&inp->inp_inc);
2761 		tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2762 	} else
2763 #endif
2764 	{
2765 		maxmtu = tcp_maxmtu(&inp->inp_inc);
2766 		tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2767 	}
2768 	so = inp->inp_socket;
2769 
2770 	/*
2771 	 * no route to sender, stay with default mss and return
2772 	 */
2773 	if (maxmtu == 0)
2774 		return;
2775 
2776 	/* what have we got? */
2777 	switch (offer) {
2778 		case 0:
2779 			/*
2780 			 * Offer == 0 means that there was no MSS on the SYN
2781 			 * segment, in this case we use tcp_mssdflt.
2782 			 */
2783 			offer =
2784 #ifdef INET6
2785 				isipv6 ? tcp_v6mssdflt :
2786 #endif
2787 				tcp_mssdflt;
2788 			break;
2789 
2790 		case -1:
2791 			/*
2792 			 * Offer == -1 means that we didn't receive SYN yet.
2793 			 */
2794 			/* FALLTHROUGH */
2795 
2796 		default:
2797 			/*
2798 			 * Prevent DoS attack with too small MSS. Round up
2799 			 * to at least minmss.
2800 			 */
2801 			offer = max(offer, tcp_minmss);
2802 			/*
2803 			 * Sanity check: make sure that maxopd will be large
2804 			 * enough to allow some data on segments even if the
2805 			 * all the option space is used (40bytes).  Otherwise
2806 			 * funny things may happen in tcp_output.
2807 			 */
2808 			offer = max(offer, 64);
2809 	}
2810 
2811 	/*
2812 	 * rmx information is now retrieved from tcp_hostcache
2813 	 */
2814 	tcp_hc_get(&inp->inp_inc, &metrics);
2815 
2816 	/*
2817 	 * if there's a discovered mtu int tcp hostcache, use it
2818 	 * else, use the link mtu.
2819 	 */
2820 	if (metrics.rmx_mtu)
2821 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2822 	else {
2823 #ifdef INET6
2824 		if (isipv6) {
2825 			mss = maxmtu - min_protoh;
2826 			if (!path_mtu_discovery &&
2827 			    !in6_localaddr(&inp->in6p_faddr))
2828 				mss = min(mss, tcp_v6mssdflt);
2829 		} else
2830 #endif
2831 		{
2832 			mss = maxmtu - min_protoh;
2833 			if (!path_mtu_discovery &&
2834 			    !in_localaddr(inp->inp_faddr))
2835 				mss = min(mss, tcp_mssdflt);
2836 		}
2837 	}
2838 	mss = min(mss, offer);
2839 
2840 	/*
2841 	 * maxopd stores the maximum length of data AND options
2842 	 * in a segment; maxseg is the amount of data in a normal
2843 	 * segment.  We need to store this value (maxopd) apart
2844 	 * from maxseg, because now every segment carries options
2845 	 * and thus we normally have somewhat less data in segments.
2846 	 */
2847 	tp->t_maxopd = mss;
2848 
2849 	/*
2850 	 * origoffer==-1 indicates, that no segments were received yet.
2851 	 * In this case we just guess.
2852 	 */
2853 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2854 	    (origoffer == -1 ||
2855 	     (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2856 		mss -= TCPOLEN_TSTAMP_APPA;
2857 	tp->t_maxseg = mss;
2858 
2859 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
2860 		if (mss > MCLBYTES)
2861 			mss &= ~(MCLBYTES-1);
2862 #else
2863 		if (mss > MCLBYTES)
2864 			mss = mss / MCLBYTES * MCLBYTES;
2865 #endif
2866 	tp->t_maxseg = mss;
2867 
2868 	/*
2869 	 * If there's a pipesize, change the socket buffer to that size,
2870 	 * don't change if sb_hiwat is different than default (then it
2871 	 * has been changed on purpose with setsockopt).
2872 	 * Make the socket buffers an integral number of mss units;
2873 	 * if the mss is larger than the socket buffer, decrease the mss.
2874 	 */
2875 	SOCKBUF_LOCK(&so->so_snd);
2876 	if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2877 		bufsize = metrics.rmx_sendpipe;
2878 	else
2879 		bufsize = so->so_snd.sb_hiwat;
2880 	if (bufsize < mss)
2881 		mss = bufsize;
2882 	else {
2883 		bufsize = roundup(bufsize, mss);
2884 		if (bufsize > sb_max)
2885 			bufsize = sb_max;
2886 		if (bufsize > so->so_snd.sb_hiwat)
2887 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
2888 	}
2889 	SOCKBUF_UNLOCK(&so->so_snd);
2890 	tp->t_maxseg = mss;
2891 
2892 	SOCKBUF_LOCK(&so->so_rcv);
2893 	if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2894 		bufsize = metrics.rmx_recvpipe;
2895 	else
2896 		bufsize = so->so_rcv.sb_hiwat;
2897 	if (bufsize > mss) {
2898 		bufsize = roundup(bufsize, mss);
2899 		if (bufsize > sb_max)
2900 			bufsize = sb_max;
2901 		if (bufsize > so->so_rcv.sb_hiwat)
2902 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
2903 	}
2904 	SOCKBUF_UNLOCK(&so->so_rcv);
2905 	/*
2906 	 * While we're here, check the others too
2907 	 */
2908 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2909 		tp->t_srtt = rtt;
2910 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2911 		tcpstat.tcps_usedrtt++;
2912 		if (metrics.rmx_rttvar) {
2913 			tp->t_rttvar = metrics.rmx_rttvar;
2914 			tcpstat.tcps_usedrttvar++;
2915 		} else {
2916 			/* default variation is +- 1 rtt */
2917 			tp->t_rttvar =
2918 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2919 		}
2920 		TCPT_RANGESET(tp->t_rxtcur,
2921 			      ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2922 			      tp->t_rttmin, TCPTV_REXMTMAX);
2923 	}
2924 	if (metrics.rmx_ssthresh) {
2925 		/*
2926 		 * There's some sort of gateway or interface
2927 		 * buffer limit on the path.  Use this to set
2928 		 * the slow start threshhold, but set the
2929 		 * threshold to no less than 2*mss.
2930 		 */
2931 		tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
2932 		tcpstat.tcps_usedssthresh++;
2933 	}
2934 	if (metrics.rmx_bandwidth)
2935 		tp->snd_bandwidth = metrics.rmx_bandwidth;
2936 
2937 	/*
2938 	 * Set the slow-start flight size depending on whether this
2939 	 * is a local network or not.
2940 	 *
2941 	 * Extend this so we cache the cwnd too and retrieve it here.
2942 	 * Make cwnd even bigger than RFC3390 suggests but only if we
2943 	 * have previous experience with the remote host. Be careful
2944 	 * not make cwnd bigger than remote receive window or our own
2945 	 * send socket buffer. Maybe put some additional upper bound
2946 	 * on the retrieved cwnd. Should do incremental updates to
2947 	 * hostcache when cwnd collapses so next connection doesn't
2948 	 * overloads the path again.
2949 	 *
2950 	 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
2951 	 * We currently check only in syncache_socket for that.
2952 	 */
2953 #define TCP_METRICS_CWND
2954 #ifdef TCP_METRICS_CWND
2955 	if (metrics.rmx_cwnd)
2956 		tp->snd_cwnd = max(mss,
2957 				min(metrics.rmx_cwnd / 2,
2958 				 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
2959 	else
2960 #endif
2961 	if (tcp_do_rfc3390)
2962 		tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
2963 #ifdef INET6
2964 	else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
2965 		 (!isipv6 && in_localaddr(inp->inp_faddr)))
2966 #else
2967 	else if (in_localaddr(inp->inp_faddr))
2968 #endif
2969 		tp->snd_cwnd = mss * ss_fltsz_local;
2970 	else
2971 		tp->snd_cwnd = mss * ss_fltsz;
2972 }
2973 
2974 /*
2975  * Determine the MSS option to send on an outgoing SYN.
2976  */
2977 int
2978 tcp_mssopt(inc)
2979 	struct in_conninfo *inc;
2980 {
2981 	int mss = 0;
2982 	u_long maxmtu = 0;
2983 	u_long thcmtu = 0;
2984 	size_t min_protoh;
2985 #ifdef INET6
2986 	int isipv6 = inc->inc_isipv6 ? 1 : 0;
2987 #endif
2988 
2989 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
2990 
2991 #ifdef INET6
2992 	if (isipv6) {
2993 		mss = tcp_v6mssdflt;
2994 		maxmtu = tcp_maxmtu6(inc);
2995 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
2996 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2997 	} else
2998 #endif
2999 	{
3000 		mss = tcp_mssdflt;
3001 		maxmtu = tcp_maxmtu(inc);
3002 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3003 		min_protoh = sizeof(struct tcpiphdr);
3004 	}
3005 	if (maxmtu && thcmtu)
3006 		mss = min(maxmtu, thcmtu) - min_protoh;
3007 	else if (maxmtu || thcmtu)
3008 		mss = max(maxmtu, thcmtu) - min_protoh;
3009 
3010 	return (mss);
3011 }
3012 
3013 
3014 /*
3015  * On a partial ack arrives, force the retransmission of the
3016  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3017  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3018  * be started again.
3019  */
3020 static void
3021 tcp_newreno_partial_ack(tp, th)
3022 	struct tcpcb *tp;
3023 	struct tcphdr *th;
3024 {
3025 	tcp_seq onxt = tp->snd_nxt;
3026 	u_long  ocwnd = tp->snd_cwnd;
3027 
3028 	callout_stop(tp->tt_rexmt);
3029 	tp->t_rtttime = 0;
3030 	tp->snd_nxt = th->th_ack;
3031 	/*
3032 	 * Set snd_cwnd to one segment beyond acknowledged offset.
3033 	 * (tp->snd_una has not yet been updated when this function is called.)
3034 	 */
3035 	tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3036 	tp->t_flags |= TF_ACKNOW;
3037 	(void) tcp_output(tp);
3038 	tp->snd_cwnd = ocwnd;
3039 	if (SEQ_GT(onxt, tp->snd_nxt))
3040 		tp->snd_nxt = onxt;
3041 	/*
3042 	 * Partial window deflation.  Relies on fact that tp->snd_una
3043 	 * not updated yet.
3044 	 */
3045 	tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3046 }
3047 
3048 /*
3049  * Returns 1 if the TIME_WAIT state was killed and we should start over,
3050  * looking for a pcb in the listen state.  Returns 0 otherwise.
3051  */
3052 static int
3053 tcp_timewait(tw, to, th, m, tlen)
3054 	struct tcptw *tw;
3055 	struct tcpopt *to;
3056 	struct tcphdr *th;
3057 	struct mbuf *m;
3058 	int tlen;
3059 {
3060 	int thflags;
3061 	tcp_seq seq;
3062 #ifdef INET6
3063 	int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3064 #else
3065 	const int isipv6 = 0;
3066 #endif
3067 
3068 	thflags = th->th_flags;
3069 
3070 	/*
3071 	 * NOTE: for FIN_WAIT_2 (to be added later),
3072 	 * must validate sequence number before accepting RST
3073 	 */
3074 
3075 	/*
3076 	 * If the segment contains RST:
3077 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
3078 	 *      RFC 1337.
3079 	 */
3080 	if (thflags & TH_RST)
3081 		goto drop;
3082 
3083 #if 0
3084 /* PAWS not needed at the moment */
3085 	/*
3086 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3087 	 * and it's less than ts_recent, drop it.
3088 	 */
3089 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3090 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3091 		if ((thflags & TH_ACK) == 0)
3092 			goto drop;
3093 		goto ack;
3094 	}
3095 	/*
3096 	 * ts_recent is never updated because we never accept new segments.
3097 	 */
3098 #endif
3099 
3100 	/*
3101 	 * If a new connection request is received
3102 	 * while in TIME_WAIT, drop the old connection
3103 	 * and start over if the sequence numbers
3104 	 * are above the previous ones.
3105 	 */
3106 	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3107 		(void) tcp_twclose(tw, 0);
3108 		return (1);
3109 	}
3110 
3111 	/*
3112 	 * Drop the the segment if it does not contain an ACK.
3113 	 */
3114 	if ((thflags & TH_ACK) == 0)
3115 		goto drop;
3116 
3117 	/*
3118 	 * Reset the 2MSL timer if this is a duplicate FIN.
3119 	 */
3120 	if (thflags & TH_FIN) {
3121 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3122 		if (seq + 1 == tw->rcv_nxt)
3123 			tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3124 	}
3125 
3126 	/*
3127 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
3128 	 */
3129 	if (thflags != TH_ACK || tlen != 0 ||
3130 	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3131 		tcp_twrespond(tw, TH_ACK);
3132 	goto drop;
3133 
3134 	/*
3135 	 * Generate a RST, dropping incoming segment.
3136 	 * Make ACK acceptable to originator of segment.
3137 	 * Don't bother to respond if destination was broadcast/multicast.
3138 	 */
3139 	if (m->m_flags & (M_BCAST|M_MCAST))
3140 		goto drop;
3141 	if (isipv6) {
3142 		struct ip6_hdr *ip6;
3143 
3144 		/* IPv6 anycast check is done at tcp6_input() */
3145 		ip6 = mtod(m, struct ip6_hdr *);
3146 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3147 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3148 			goto drop;
3149 	} else {
3150 		struct ip *ip;
3151 
3152 		ip = mtod(m, struct ip *);
3153 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3154 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3155 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3156 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3157 			goto drop;
3158 	}
3159 	if (thflags & TH_ACK) {
3160 		tcp_respond(NULL,
3161 		    mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3162 	} else {
3163 		seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3164 		tcp_respond(NULL,
3165 		    mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3166 	}
3167 	INP_UNLOCK(tw->tw_inpcb);
3168 	return (0);
3169 
3170 drop:
3171 	INP_UNLOCK(tw->tw_inpcb);
3172 	m_freem(m);
3173 	return (0);
3174 }
3175