xref: /freebsd/sys/netinet/tcp_input.c (revision 1f467eaaf82a9b863fc6264c7c8e6b40fffdb357)
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 static int tcp_insecure_rst = 0;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
136     &tcp_insecure_rst, 0,
137     "Follow the old (insecure) criteria for accepting RST packets.");
138 
139 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
140 	    "TCP Segment Reassembly Queue");
141 
142 static int tcp_reass_maxseg = 0;
143 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
144 	   &tcp_reass_maxseg, 0,
145 	   "Global maximum number of TCP Segments in Reassembly Queue");
146 
147 int tcp_reass_qsize = 0;
148 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
149 	   &tcp_reass_qsize, 0,
150 	   "Global number of TCP Segments currently in Reassembly Queue");
151 
152 static int tcp_reass_maxqlen = 48;
153 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW,
154 	   &tcp_reass_maxqlen, 0,
155 	   "Maximum number of TCP Segments per individual Reassembly Queue");
156 
157 static int tcp_reass_overflows = 0;
158 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
159 	   &tcp_reass_overflows, 0,
160 	   "Global number of TCP Segment Reassembly Queue Overflows");
161 
162 struct inpcbhead tcb;
163 #define	tcb6	tcb  /* for KAME src sync over BSD*'s */
164 struct inpcbinfo tcbinfo;
165 struct mtx	*tcbinfo_mtx;
166 
167 static void	 tcp_dooptions(struct tcpopt *, u_char *, int, int);
168 
169 static void	 tcp_pulloutofband(struct socket *,
170 		     struct tcphdr *, struct mbuf *, int);
171 static int	 tcp_reass(struct tcpcb *, struct tcphdr *, int *,
172 		     struct mbuf *);
173 static void	 tcp_xmit_timer(struct tcpcb *, int);
174 static void	 tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
175 static int	 tcp_timewait(struct tcptw *, struct tcpopt *,
176 		     struct tcphdr *, struct mbuf *, int);
177 
178 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
179 #ifdef INET6
180 #define ND6_HINT(tp) \
181 do { \
182 	if ((tp) && (tp)->t_inpcb && \
183 	    ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
184 		nd6_nud_hint(NULL, NULL, 0); \
185 } while (0)
186 #else
187 #define ND6_HINT(tp)
188 #endif
189 
190 /*
191  * Indicate whether this ack should be delayed.  We can delay the ack if
192  *	- there is no delayed ack timer in progress and
193  *	- our last ack wasn't a 0-sized window.  We never want to delay
194  *	  the ack that opens up a 0-sized window and
195  *		- delayed acks are enabled or
196  *		- this is a half-synchronized T/TCP connection.
197  */
198 #define DELAY_ACK(tp)							\
199 	((!callout_active(tp->tt_delack) &&				\
200 	    (tp->t_flags & TF_RXWIN0SENT) == 0) &&			\
201 	    (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
202 
203 /* Initialize TCP reassembly queue */
204 uma_zone_t	tcp_reass_zone;
205 void
206 tcp_reass_init()
207 {
208 	tcp_reass_maxseg = nmbclusters / 16;
209 	TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
210 	    &tcp_reass_maxseg);
211 	tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
212 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
213 	uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg);
214 }
215 
216 static int
217 tcp_reass(tp, th, tlenp, m)
218 	register struct tcpcb *tp;
219 	register struct tcphdr *th;
220 	int *tlenp;
221 	struct mbuf *m;
222 {
223 	struct tseg_qent *q;
224 	struct tseg_qent *p = NULL;
225 	struct tseg_qent *nq;
226 	struct tseg_qent *te = NULL;
227 	struct socket *so = tp->t_inpcb->inp_socket;
228 	int flags;
229 
230 	INP_LOCK_ASSERT(tp->t_inpcb);
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==NULL after become established to
240 	 * force pre-ESTABLISHED data up to user socket.
241 	 */
242 	if (th == NULL)
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 		*tlenp = 0;
259 		return (0);
260 	}
261 
262 	/*
263 	 * Allocate a new queue entry. If we can't, or hit the zone limit
264 	 * just drop the pkt.
265 	 */
266 	te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
267 	if (te == NULL) {
268 		tcpstat.tcps_rcvmemdrop++;
269 		m_freem(m);
270 		*tlenp = 0;
271 		return (0);
272 	}
273 	tp->t_segqlen++;
274 	tcp_reass_qsize++;
275 
276 	/*
277 	 * Find a segment which begins after this one does.
278 	 */
279 	LIST_FOREACH(q, &tp->t_segq, tqe_q) {
280 		if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
281 			break;
282 		p = q;
283 	}
284 
285 	/*
286 	 * If there is a preceding segment, it may provide some of
287 	 * our data already.  If so, drop the data from the incoming
288 	 * segment.  If it provides all of our data, drop us.
289 	 */
290 	if (p != NULL) {
291 		register int i;
292 		/* conversion to int (in i) handles seq wraparound */
293 		i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
294 		if (i > 0) {
295 			if (i >= *tlenp) {
296 				tcpstat.tcps_rcvduppack++;
297 				tcpstat.tcps_rcvdupbyte += *tlenp;
298 				m_freem(m);
299 				uma_zfree(tcp_reass_zone, te);
300 				tp->t_segqlen--;
301 				tcp_reass_qsize--;
302 				/*
303 				 * Try to present any queued data
304 				 * at the left window edge to the user.
305 				 * This is needed after the 3-WHS
306 				 * completes.
307 				 */
308 				goto present;	/* ??? */
309 			}
310 			m_adj(m, i);
311 			*tlenp -= i;
312 			th->th_seq += i;
313 		}
314 	}
315 	tcpstat.tcps_rcvoopack++;
316 	tcpstat.tcps_rcvoobyte += *tlenp;
317 
318 	/*
319 	 * While we overlap succeeding segments trim them or,
320 	 * if they are completely covered, dequeue them.
321 	 */
322 	while (q) {
323 		register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
324 		if (i <= 0)
325 			break;
326 		if (i < q->tqe_len) {
327 			q->tqe_th->th_seq += i;
328 			q->tqe_len -= i;
329 			m_adj(q->tqe_m, i);
330 			break;
331 		}
332 
333 		nq = LIST_NEXT(q, tqe_q);
334 		LIST_REMOVE(q, tqe_q);
335 		m_freem(q->tqe_m);
336 		uma_zfree(tcp_reass_zone, q);
337 		tp->t_segqlen--;
338 		tcp_reass_qsize--;
339 		q = nq;
340 	}
341 
342 	/* Insert the new segment queue entry into place. */
343 	te->tqe_m = m;
344 	te->tqe_th = th;
345 	te->tqe_len = *tlenp;
346 
347 	if (p == NULL) {
348 		LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
349 	} else {
350 		LIST_INSERT_AFTER(p, te, tqe_q);
351 	}
352 
353 present:
354 	/*
355 	 * Present data to user, advancing rcv_nxt through
356 	 * completed sequence space.
357 	 */
358 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
359 		return (0);
360 	q = LIST_FIRST(&tp->t_segq);
361 	if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
362 		return (0);
363 	SOCKBUF_LOCK(&so->so_rcv);
364 	do {
365 		tp->rcv_nxt += q->tqe_len;
366 		flags = q->tqe_th->th_flags & TH_FIN;
367 		nq = LIST_NEXT(q, tqe_q);
368 		LIST_REMOVE(q, tqe_q);
369 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
370 			m_freem(q->tqe_m);
371 		else
372 			sbappendstream_locked(&so->so_rcv, q->tqe_m);
373 		uma_zfree(tcp_reass_zone, q);
374 		tp->t_segqlen--;
375 		tcp_reass_qsize--;
376 		q = nq;
377 	} while (q && q->tqe_th->th_seq == tp->rcv_nxt);
378 	ND6_HINT(tp);
379 	sorwakeup_locked(so);
380 	return (flags);
381 }
382 
383 /*
384  * TCP input routine, follows pages 65-76 of the
385  * protocol specification dated September, 1981 very closely.
386  */
387 #ifdef INET6
388 int
389 tcp6_input(mp, offp, proto)
390 	struct mbuf **mp;
391 	int *offp, proto;
392 {
393 	register struct mbuf *m = *mp;
394 	struct in6_ifaddr *ia6;
395 
396 	IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
397 
398 	/*
399 	 * draft-itojun-ipv6-tcp-to-anycast
400 	 * better place to put this in?
401 	 */
402 	ia6 = ip6_getdstifaddr(m);
403 	if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
404 		struct ip6_hdr *ip6;
405 
406 		ip6 = mtod(m, struct ip6_hdr *);
407 		icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
408 			    (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
409 		return IPPROTO_DONE;
410 	}
411 
412 	tcp_input(m, *offp);
413 	return IPPROTO_DONE;
414 }
415 #endif
416 
417 void
418 tcp_input(m, off0)
419 	register struct mbuf *m;
420 	int off0;
421 {
422 	register struct tcphdr *th;
423 	register struct ip *ip = NULL;
424 	register struct ipovly *ipov;
425 	register struct inpcb *inp = NULL;
426 	u_char *optp = NULL;
427 	int optlen = 0;
428 	int len, tlen, off;
429 	int drop_hdrlen;
430 	register struct tcpcb *tp = 0;
431 	register int thflags;
432 	struct socket *so = 0;
433 	int todrop, acked, ourfinisacked, needoutput = 0;
434 	u_long tiwin;
435 	struct tcpopt to;		/* options in this segment */
436 	int headlocked = 0;
437 #ifdef IPFIREWALL_FORWARD
438 	struct m_tag *fwd_tag;
439 #endif
440 	int rstreason; /* For badport_bandlim accounting purposes */
441 
442 	struct ip6_hdr *ip6 = NULL;
443 #ifdef INET6
444 	int isipv6;
445 #else
446 	const int isipv6 = 0;
447 #endif
448 
449 #ifdef TCPDEBUG
450 	/*
451 	 * The size of tcp_saveipgen must be the size of the max ip header,
452 	 * now IPv6.
453 	 */
454 	u_char tcp_saveipgen[40];
455 	struct tcphdr tcp_savetcp;
456 	short ostate = 0;
457 #endif
458 
459 #ifdef INET6
460 	isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
461 #endif
462 	bzero((char *)&to, sizeof(to));
463 
464 	tcpstat.tcps_rcvtotal++;
465 
466 	if (isipv6) {
467 #ifdef INET6
468 		/* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
469 		ip6 = mtod(m, struct ip6_hdr *);
470 		tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
471 		if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
472 			tcpstat.tcps_rcvbadsum++;
473 			goto drop;
474 		}
475 		th = (struct tcphdr *)((caddr_t)ip6 + off0);
476 
477 		/*
478 		 * Be proactive about unspecified IPv6 address in source.
479 		 * As we use all-zero to indicate unbounded/unconnected pcb,
480 		 * unspecified IPv6 address can be used to confuse us.
481 		 *
482 		 * Note that packets with unspecified IPv6 destination is
483 		 * already dropped in ip6_input.
484 		 */
485 		if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
486 			/* XXX stat */
487 			goto drop;
488 		}
489 #else
490 		th = NULL;		/* XXX: avoid compiler warning */
491 #endif
492 	} else {
493 		/*
494 		 * Get IP and TCP header together in first mbuf.
495 		 * Note: IP leaves IP header in first mbuf.
496 		 */
497 		if (off0 > sizeof (struct ip)) {
498 			ip_stripoptions(m, (struct mbuf *)0);
499 			off0 = sizeof(struct ip);
500 		}
501 		if (m->m_len < sizeof (struct tcpiphdr)) {
502 			if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
503 				tcpstat.tcps_rcvshort++;
504 				return;
505 			}
506 		}
507 		ip = mtod(m, struct ip *);
508 		ipov = (struct ipovly *)ip;
509 		th = (struct tcphdr *)((caddr_t)ip + off0);
510 		tlen = ip->ip_len;
511 
512 		if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
513 			if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
514 				th->th_sum = m->m_pkthdr.csum_data;
515 			else
516 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
517 						ip->ip_dst.s_addr,
518 						htonl(m->m_pkthdr.csum_data +
519 							ip->ip_len +
520 							IPPROTO_TCP));
521 			th->th_sum ^= 0xffff;
522 #ifdef TCPDEBUG
523 			ipov->ih_len = (u_short)tlen;
524 			ipov->ih_len = htons(ipov->ih_len);
525 #endif
526 		} else {
527 			/*
528 			 * Checksum extended TCP header and data.
529 			 */
530 			len = sizeof (struct ip) + tlen;
531 			bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
532 			ipov->ih_len = (u_short)tlen;
533 			ipov->ih_len = htons(ipov->ih_len);
534 			th->th_sum = in_cksum(m, len);
535 		}
536 		if (th->th_sum) {
537 			tcpstat.tcps_rcvbadsum++;
538 			goto drop;
539 		}
540 #ifdef INET6
541 		/* Re-initialization for later version check */
542 		ip->ip_v = IPVERSION;
543 #endif
544 	}
545 
546 	/*
547 	 * Check that TCP offset makes sense,
548 	 * pull out TCP options and adjust length.		XXX
549 	 */
550 	off = th->th_off << 2;
551 	if (off < sizeof (struct tcphdr) || off > tlen) {
552 		tcpstat.tcps_rcvbadoff++;
553 		goto drop;
554 	}
555 	tlen -= off;	/* tlen is used instead of ti->ti_len */
556 	if (off > sizeof (struct tcphdr)) {
557 		if (isipv6) {
558 #ifdef INET6
559 			IP6_EXTHDR_CHECK(m, off0, off, );
560 			ip6 = mtod(m, struct ip6_hdr *);
561 			th = (struct tcphdr *)((caddr_t)ip6 + off0);
562 #endif
563 		} else {
564 			if (m->m_len < sizeof(struct ip) + off) {
565 				if ((m = m_pullup(m, sizeof (struct ip) + off))
566 						== 0) {
567 					tcpstat.tcps_rcvshort++;
568 					return;
569 				}
570 				ip = mtod(m, struct ip *);
571 				ipov = (struct ipovly *)ip;
572 				th = (struct tcphdr *)((caddr_t)ip + off0);
573 			}
574 		}
575 		optlen = off - sizeof (struct tcphdr);
576 		optp = (u_char *)(th + 1);
577 	}
578 	thflags = th->th_flags;
579 
580 #ifdef TCP_DROP_SYNFIN
581 	/*
582 	 * If the drop_synfin option is enabled, drop all packets with
583 	 * both the SYN and FIN bits set. This prevents e.g. nmap from
584 	 * identifying the TCP/IP stack.
585 	 *
586 	 * This is a violation of the TCP specification.
587 	 */
588 	if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
589 		goto drop;
590 #endif
591 
592 	/*
593 	 * Convert TCP protocol specific fields to host format.
594 	 */
595 	th->th_seq = ntohl(th->th_seq);
596 	th->th_ack = ntohl(th->th_ack);
597 	th->th_win = ntohs(th->th_win);
598 	th->th_urp = ntohs(th->th_urp);
599 
600 	/*
601 	 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
602 	 * until after ip6_savecontrol() is called and before other functions
603 	 * which don't want those proto headers.
604 	 * Because ip6_savecontrol() is going to parse the mbuf to
605 	 * search for data to be passed up to user-land, it wants mbuf
606 	 * parameters to be unchanged.
607 	 * XXX: the call of ip6_savecontrol() has been obsoleted based on
608 	 * latest version of the advanced API (20020110).
609 	 */
610 	drop_hdrlen = off0 + off;
611 
612 	/*
613 	 * Locate pcb for segment.
614 	 */
615 	INP_INFO_WLOCK(&tcbinfo);
616 	headlocked = 1;
617 findpcb:
618 	KASSERT(headlocked, ("tcp_input: findpcb: head not locked"));
619 #ifdef IPFIREWALL_FORWARD
620 	/* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
621 	fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
622 
623 	if (fwd_tag != NULL && isipv6 == 0) {	/* IPv6 support is not yet */
624 		struct sockaddr_in *next_hop;
625 
626 		next_hop = (struct sockaddr_in *)(fwd_tag+1);
627 		/*
628 		 * Transparently forwarded. Pretend to be the destination.
629 		 * already got one like this?
630 		 */
631 		inp = in_pcblookup_hash(&tcbinfo,
632 					ip->ip_src, th->th_sport,
633 					ip->ip_dst, th->th_dport,
634 					0, m->m_pkthdr.rcvif);
635 		if (!inp) {
636 			/* It's new.  Try to find the ambushing socket. */
637 			inp = in_pcblookup_hash(&tcbinfo,
638 						ip->ip_src, th->th_sport,
639 						next_hop->sin_addr,
640 						next_hop->sin_port ?
641 						    ntohs(next_hop->sin_port) :
642 						    th->th_dport,
643 						1, m->m_pkthdr.rcvif);
644 		}
645 		/* Remove the tag from the packet.  We don't need it anymore. */
646 		m_tag_delete(m, fwd_tag);
647 	} else {
648 #endif /* IPFIREWALL_FORWARD */
649 		if (isipv6) {
650 #ifdef INET6
651 			inp = in6_pcblookup_hash(&tcbinfo,
652 						 &ip6->ip6_src, th->th_sport,
653 						 &ip6->ip6_dst, th->th_dport,
654 						 1, m->m_pkthdr.rcvif);
655 #endif
656 		} else
657 			inp = in_pcblookup_hash(&tcbinfo,
658 						ip->ip_src, th->th_sport,
659 						ip->ip_dst, th->th_dport,
660 						1, m->m_pkthdr.rcvif);
661 #ifdef IPFIREWALL_FORWARD
662 	}
663 #endif /* IPFIREWALL_FORWARD */
664 
665 #if defined(IPSEC) || defined(FAST_IPSEC)
666 #ifdef INET6
667 	if (isipv6) {
668 		if (inp != NULL && ipsec6_in_reject(m, inp)) {
669 #ifdef IPSEC
670 			ipsec6stat.in_polvio++;
671 #endif
672 			goto drop;
673 		}
674 	} else
675 #endif /* INET6 */
676 	if (inp != NULL && ipsec4_in_reject(m, inp)) {
677 #ifdef IPSEC
678 		ipsecstat.in_polvio++;
679 #endif
680 		goto drop;
681 	}
682 #endif /*IPSEC || FAST_IPSEC*/
683 
684 	/*
685 	 * If the state is CLOSED (i.e., TCB does not exist) then
686 	 * all data in the incoming segment is discarded.
687 	 * If the TCB exists but is in CLOSED state, it is embryonic,
688 	 * but should either do a listen or a connect soon.
689 	 */
690 	if (inp == NULL) {
691 		if (log_in_vain) {
692 #ifdef INET6
693 			char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
694 #else
695 			char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"];
696 #endif
697 
698 			if (isipv6) {
699 #ifdef INET6
700 				strcpy(dbuf, "[");
701 				strcpy(sbuf, "[");
702 				strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
703 				strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
704 				strcat(dbuf, "]");
705 				strcat(sbuf, "]");
706 #endif
707 			} else {
708 				strcpy(dbuf, inet_ntoa(ip->ip_dst));
709 				strcpy(sbuf, inet_ntoa(ip->ip_src));
710 			}
711 			switch (log_in_vain) {
712 			case 1:
713 				if ((thflags & TH_SYN) == 0)
714 					break;
715 				/* FALLTHROUGH */
716 			case 2:
717 				log(LOG_INFO,
718 				    "Connection attempt to TCP %s:%d "
719 				    "from %s:%d flags:0x%02x\n",
720 				    dbuf, ntohs(th->th_dport), sbuf,
721 				    ntohs(th->th_sport), thflags);
722 				break;
723 			default:
724 				break;
725 			}
726 		}
727 		if (blackhole) {
728 			switch (blackhole) {
729 			case 1:
730 				if (thflags & TH_SYN)
731 					goto drop;
732 				break;
733 			case 2:
734 				goto drop;
735 			default:
736 				goto drop;
737 			}
738 		}
739 		rstreason = BANDLIM_RST_CLOSEDPORT;
740 		goto dropwithreset;
741 	}
742 	INP_LOCK(inp);
743 
744 	/* Check the minimum TTL for socket. */
745 	if (inp->inp_ip_minttl && inp->inp_ip_minttl > ip->ip_ttl)
746 		goto drop;
747 
748 	if (inp->inp_vflag & INP_TIMEWAIT) {
749 		/*
750 		 * The only option of relevance is TOF_CC, and only if
751 		 * present in a SYN segment.  See tcp_timewait().
752 		 */
753 		if (thflags & TH_SYN)
754 			tcp_dooptions(&to, optp, optlen, 1);
755 		if (tcp_timewait((struct tcptw *)inp->inp_ppcb,
756 		    &to, th, m, tlen))
757 			goto findpcb;
758 		/*
759 		 * tcp_timewait unlocks inp.
760 		 */
761 		INP_INFO_WUNLOCK(&tcbinfo);
762 		return;
763 	}
764 	tp = intotcpcb(inp);
765 	if (tp == 0) {
766 		INP_UNLOCK(inp);
767 		rstreason = BANDLIM_RST_CLOSEDPORT;
768 		goto dropwithreset;
769 	}
770 	if (tp->t_state == TCPS_CLOSED)
771 		goto drop;
772 
773 	/* Unscale the window into a 32-bit value. */
774 	if ((thflags & TH_SYN) == 0)
775 		tiwin = th->th_win << tp->snd_scale;
776 	else
777 		tiwin = th->th_win;
778 
779 #ifdef MAC
780 	INP_LOCK_ASSERT(inp);
781 	if (mac_check_inpcb_deliver(inp, m))
782 		goto drop;
783 #endif
784 	so = inp->inp_socket;
785 #ifdef TCPDEBUG
786 	if (so->so_options & SO_DEBUG) {
787 		ostate = tp->t_state;
788 		if (isipv6)
789 			bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
790 		else
791 			bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
792 		tcp_savetcp = *th;
793 	}
794 #endif
795 	if (so->so_options & SO_ACCEPTCONN) {
796 		struct in_conninfo inc;
797 
798 #ifdef INET6
799 		inc.inc_isipv6 = isipv6;
800 #endif
801 		if (isipv6) {
802 			inc.inc6_faddr = ip6->ip6_src;
803 			inc.inc6_laddr = ip6->ip6_dst;
804 		} else {
805 			inc.inc_faddr = ip->ip_src;
806 			inc.inc_laddr = ip->ip_dst;
807 		}
808 		inc.inc_fport = th->th_sport;
809 		inc.inc_lport = th->th_dport;
810 
811 	        /*
812 	         * If the state is LISTEN then ignore segment if it contains
813 		 * a RST.  If the segment contains an ACK then it is bad and
814 		 * send a RST.  If it does not contain a SYN then it is not
815 		 * interesting; drop it.
816 		 *
817 		 * If the state is SYN_RECEIVED (syncache) and seg contains
818 		 * an ACK, but not for our SYN/ACK, send a RST.  If the seg
819 		 * contains a RST, check the sequence number to see if it
820 		 * is a valid reset segment.
821 		 */
822 		if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
823 			if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
824 				if (!syncache_expand(&inc, th, &so, m)) {
825 					/*
826 					 * No syncache entry, or ACK was not
827 					 * for our SYN/ACK.  Send a RST.
828 					 */
829 					tcpstat.tcps_badsyn++;
830 					rstreason = BANDLIM_RST_OPENPORT;
831 					goto dropwithreset;
832 				}
833 				if (so == NULL) {
834 					/*
835 					 * Could not complete 3-way handshake,
836 					 * connection is being closed down, and
837 					 * syncache will free mbuf.
838 					 */
839 					INP_UNLOCK(inp);
840 					INP_INFO_WUNLOCK(&tcbinfo);
841 					return;
842 				}
843 				/*
844 				 * Socket is created in state SYN_RECEIVED.
845 				 * Continue processing segment.
846 				 */
847 				INP_UNLOCK(inp);
848 				inp = sotoinpcb(so);
849 				INP_LOCK(inp);
850 				tp = intotcpcb(inp);
851 				/*
852 				 * This is what would have happened in
853 				 * tcp_output() when the SYN,ACK was sent.
854 				 */
855 				tp->snd_up = tp->snd_una;
856 				tp->snd_max = tp->snd_nxt = tp->iss + 1;
857 				tp->last_ack_sent = tp->rcv_nxt;
858 				/*
859 				 * RFC1323: The window in SYN & SYN/ACK
860 				 * segments is never scaled.
861 				 */
862 				tp->snd_wnd = tiwin;	/* unscaled */
863 				goto after_listen;
864 			}
865 			if (thflags & TH_RST) {
866 				syncache_chkrst(&inc, th);
867 				goto drop;
868 			}
869 			if (thflags & TH_ACK) {
870 				syncache_badack(&inc);
871 				tcpstat.tcps_badsyn++;
872 				rstreason = BANDLIM_RST_OPENPORT;
873 				goto dropwithreset;
874 			}
875 			goto drop;
876 		}
877 
878 		/*
879 		 * Segment's flags are (SYN) or (SYN|FIN).
880 		 */
881 #ifdef INET6
882 		/*
883 		 * If deprecated address is forbidden,
884 		 * we do not accept SYN to deprecated interface
885 		 * address to prevent any new inbound connection from
886 		 * getting established.
887 		 * When we do not accept SYN, we send a TCP RST,
888 		 * with deprecated source address (instead of dropping
889 		 * it).  We compromise it as it is much better for peer
890 		 * to send a RST, and RST will be the final packet
891 		 * for the exchange.
892 		 *
893 		 * If we do not forbid deprecated addresses, we accept
894 		 * the SYN packet.  RFC2462 does not suggest dropping
895 		 * SYN in this case.
896 		 * If we decipher RFC2462 5.5.4, it says like this:
897 		 * 1. use of deprecated addr with existing
898 		 *    communication is okay - "SHOULD continue to be
899 		 *    used"
900 		 * 2. use of it with new communication:
901 		 *   (2a) "SHOULD NOT be used if alternate address
902 		 *        with sufficient scope is available"
903 		 *   (2b) nothing mentioned otherwise.
904 		 * Here we fall into (2b) case as we have no choice in
905 		 * our source address selection - we must obey the peer.
906 		 *
907 		 * The wording in RFC2462 is confusing, and there are
908 		 * multiple description text for deprecated address
909 		 * handling - worse, they are not exactly the same.
910 		 * I believe 5.5.4 is the best one, so we follow 5.5.4.
911 		 */
912 		if (isipv6 && !ip6_use_deprecated) {
913 			struct in6_ifaddr *ia6;
914 
915 			if ((ia6 = ip6_getdstifaddr(m)) &&
916 			    (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
917 				INP_UNLOCK(inp);
918 				tp = NULL;
919 				rstreason = BANDLIM_RST_OPENPORT;
920 				goto dropwithreset;
921 			}
922 		}
923 #endif
924 		/*
925 		 * If it is from this socket, drop it, it must be forged.
926 		 * Don't bother responding if the destination was a broadcast.
927 		 */
928 		if (th->th_dport == th->th_sport) {
929 			if (isipv6) {
930 				if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
931 						       &ip6->ip6_src))
932 					goto drop;
933 			} else {
934 				if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
935 					goto drop;
936 			}
937 		}
938 		/*
939 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
940 		 *
941 		 * Note that it is quite possible to receive unicast
942 		 * link-layer packets with a broadcast IP address. Use
943 		 * in_broadcast() to find them.
944 		 */
945 		if (m->m_flags & (M_BCAST|M_MCAST))
946 			goto drop;
947 		if (isipv6) {
948 			if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
949 			    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
950 				goto drop;
951 		} else {
952 			if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
953 			    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
954 			    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
955 			    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
956 				goto drop;
957 		}
958 		/*
959 		 * SYN appears to be valid; create compressed TCP state
960 		 * for syncache, or perform t/tcp connection.
961 		 */
962 		if (so->so_qlen <= so->so_qlimit) {
963 #ifdef TCPDEBUG
964 			if (so->so_options & SO_DEBUG)
965 				tcp_trace(TA_INPUT, ostate, tp,
966 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
967 #endif
968 			tcp_dooptions(&to, optp, optlen, 1);
969 			if (!syncache_add(&inc, &to, th, &so, m))
970 				goto drop;
971 			if (so == NULL) {
972 				/*
973 				 * Entry added to syncache, mbuf used to
974 				 * send SYN,ACK packet.
975 				 */
976 				KASSERT(headlocked, ("headlocked"));
977 				INP_UNLOCK(inp);
978 				INP_INFO_WUNLOCK(&tcbinfo);
979 				return;
980 			}
981 			/*
982 			 * Segment passed TAO tests.
983 			 */
984 			INP_UNLOCK(inp);
985 			inp = sotoinpcb(so);
986 			INP_LOCK(inp);
987 			tp = intotcpcb(inp);
988 			tp->snd_wnd = tiwin;
989 			tp->t_starttime = ticks;
990 			tp->t_state = TCPS_ESTABLISHED;
991 
992 			/*
993 			 * T/TCP logic:
994 			 * If there is a FIN or if there is data, then
995 			 * delay SYN,ACK(SYN) in the hope of piggy-backing
996 			 * it on a response segment.  Otherwise must send
997 			 * ACK now in case the other side is slow starting.
998 			 */
999 			if (thflags & TH_FIN || tlen != 0)
1000 				tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1001 			else
1002 				tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1003 			tcpstat.tcps_connects++;
1004 			soisconnected(so);
1005 			goto trimthenstep6;
1006 		}
1007 		goto drop;
1008 	}
1009 after_listen:
1010 	KASSERT(headlocked, ("tcp_input: after_listen: head not locked"));
1011 	INP_LOCK_ASSERT(inp);
1012 
1013 	/* Syncache takes care of sockets in the listen state. */
1014 	KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN"));
1015 
1016 	/*
1017 	 * This is the second part of the MSS DoS prevention code (after
1018 	 * minmss on the sending side) and it deals with too many too small
1019 	 * tcp packets in a too short timeframe (1 second).
1020 	 *
1021 	 * For every full second we count the number of received packets
1022 	 * and bytes. If we get a lot of packets per second for this connection
1023 	 * (tcp_minmssoverload) we take a closer look at it and compute the
1024 	 * average packet size for the past second. If that is less than
1025 	 * tcp_minmss we get too many packets with very small payload which
1026 	 * is not good and burdens our system (and every packet generates
1027 	 * a wakeup to the process connected to our socket). We can reasonable
1028 	 * expect this to be small packet DoS attack to exhaust our CPU
1029 	 * cycles.
1030 	 *
1031 	 * Care has to be taken for the minimum packet overload value. This
1032 	 * value defines the minimum number of packets per second before we
1033 	 * start to worry. This must not be too low to avoid killing for
1034 	 * example interactive connections with many small packets like
1035 	 * telnet or SSH.
1036 	 *
1037 	 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1038 	 * this check.
1039 	 *
1040 	 * Account for packet if payload packet, skip over ACK, etc.
1041 	 */
1042 	if (tcp_minmss && tcp_minmssoverload &&
1043 	    tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1044 		if ((unsigned int)(tp->rcv_second - ticks) < hz) {
1045 			tp->rcv_pps++;
1046 			tp->rcv_byps += tlen + off;
1047 			if (tp->rcv_pps > tcp_minmssoverload) {
1048 				if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1049 					printf("too many small tcp packets from "
1050 					       "%s:%u, av. %lubyte/packet, "
1051 					       "dropping connection\n",
1052 #ifdef INET6
1053 						isipv6 ?
1054 						ip6_sprintf(&inp->inp_inc.inc6_faddr) :
1055 #endif
1056 						inet_ntoa(inp->inp_inc.inc_faddr),
1057 						inp->inp_inc.inc_fport,
1058 						tp->rcv_byps / tp->rcv_pps);
1059 					KASSERT(headlocked, ("tcp_input: "
1060 					    "after_listen: tcp_drop: head "
1061 					    "not locked"));
1062 					tp = tcp_drop(tp, ECONNRESET);
1063 					tcpstat.tcps_minmssdrops++;
1064 					goto drop;
1065 				}
1066 			}
1067 		} else {
1068 			tp->rcv_second = ticks + hz;
1069 			tp->rcv_pps = 1;
1070 			tp->rcv_byps = tlen + off;
1071 		}
1072 	}
1073 
1074 	/*
1075 	 * Segment received on connection.
1076 	 * Reset idle time and keep-alive timer.
1077 	 */
1078 	tp->t_rcvtime = ticks;
1079 	if (TCPS_HAVEESTABLISHED(tp->t_state))
1080 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1081 
1082 	/*
1083 	 * Process options only when we get SYN/ACK back. The SYN case
1084 	 * for incoming connections is handled in tcp_syncache.
1085 	 * XXX this is traditional behavior, may need to be cleaned up.
1086 	 */
1087 	tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
1088 	if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1089 		if (to.to_flags & TOF_SCALE) {
1090 			tp->t_flags |= TF_RCVD_SCALE;
1091 			tp->requested_s_scale = to.to_requested_s_scale;
1092 		}
1093 		if (to.to_flags & TOF_TS) {
1094 			tp->t_flags |= TF_RCVD_TSTMP;
1095 			tp->ts_recent = to.to_tsval;
1096 			tp->ts_recent_age = ticks;
1097 		}
1098 		if (to.to_flags & TOF_MSS)
1099 			tcp_mss(tp, to.to_mss);
1100 		if (tp->sack_enable) {
1101 			if (!(to.to_flags & TOF_SACK))
1102 				tp->sack_enable = 0;
1103 			else
1104 				tp->t_flags |= TF_SACK_PERMIT;
1105 		}
1106 
1107 	}
1108 
1109 	/*
1110 	 * Header prediction: check for the two common cases
1111 	 * of a uni-directional data xfer.  If the packet has
1112 	 * no control flags, is in-sequence, the window didn't
1113 	 * change and we're not retransmitting, it's a
1114 	 * candidate.  If the length is zero and the ack moved
1115 	 * forward, we're the sender side of the xfer.  Just
1116 	 * free the data acked & wake any higher level process
1117 	 * that was blocked waiting for space.  If the length
1118 	 * is non-zero and the ack didn't move, we're the
1119 	 * receiver side.  If we're getting packets in-order
1120 	 * (the reassembly queue is empty), add the data to
1121 	 * the socket buffer and note that we need a delayed ack.
1122 	 * Make sure that the hidden state-flags are also off.
1123 	 * Since we check for TCPS_ESTABLISHED above, it can only
1124 	 * be TH_NEEDSYN.
1125 	 */
1126 	if (tp->t_state == TCPS_ESTABLISHED &&
1127 	    (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1128 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1129 	    ((to.to_flags & TOF_TS) == 0 ||
1130 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1131 	     th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd &&
1132 	     tp->snd_nxt == tp->snd_max) {
1133 
1134 		/*
1135 		 * If last ACK falls within this segment's sequence numbers,
1136 		 * record the timestamp.
1137 		 * NOTE that the test is modified according to the latest
1138 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1139 		 */
1140 		if ((to.to_flags & TOF_TS) != 0 &&
1141 		    SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1142 			tp->ts_recent_age = ticks;
1143 			tp->ts_recent = to.to_tsval;
1144 		}
1145 
1146 		if (tlen == 0) {
1147 			if (SEQ_GT(th->th_ack, tp->snd_una) &&
1148 			    SEQ_LEQ(th->th_ack, tp->snd_max) &&
1149 			    tp->snd_cwnd >= tp->snd_wnd &&
1150 			    ((!tcp_do_newreno && !tp->sack_enable &&
1151 			      tp->t_dupacks < tcprexmtthresh) ||
1152 			     ((tcp_do_newreno || tp->sack_enable) &&
1153 			      !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 &&
1154 			      TAILQ_EMPTY(&tp->snd_holes)))) {
1155 				KASSERT(headlocked, ("headlocked"));
1156 				INP_INFO_WUNLOCK(&tcbinfo);
1157 				headlocked = 0;
1158 				/*
1159 				 * this is a pure ack for outstanding data.
1160 				 */
1161 				++tcpstat.tcps_predack;
1162 				/*
1163 				 * "bad retransmit" recovery
1164 				 */
1165 				if (tp->t_rxtshift == 1 &&
1166 				    ticks < tp->t_badrxtwin) {
1167 					++tcpstat.tcps_sndrexmitbad;
1168 					tp->snd_cwnd = tp->snd_cwnd_prev;
1169 					tp->snd_ssthresh =
1170 					    tp->snd_ssthresh_prev;
1171 					tp->snd_recover = tp->snd_recover_prev;
1172 					if (tp->t_flags & TF_WASFRECOVERY)
1173 					    ENTER_FASTRECOVERY(tp);
1174 					tp->snd_nxt = tp->snd_max;
1175 					tp->t_badrxtwin = 0;
1176 				}
1177 
1178 				/*
1179 				 * Recalculate the transmit timer / rtt.
1180 				 *
1181 				 * Some boxes send broken timestamp replies
1182 				 * during the SYN+ACK phase, ignore
1183 				 * timestamps of 0 or we could calculate a
1184 				 * huge RTT and blow up the retransmit timer.
1185 				 */
1186 				if ((to.to_flags & TOF_TS) != 0 &&
1187 				    to.to_tsecr) {
1188 					tcp_xmit_timer(tp,
1189 					    ticks - to.to_tsecr + 1);
1190 				} else if (tp->t_rtttime &&
1191 					    SEQ_GT(th->th_ack, tp->t_rtseq)) {
1192 					tcp_xmit_timer(tp,
1193 							ticks - tp->t_rtttime);
1194 				}
1195 				tcp_xmit_bandwidth_limit(tp, th->th_ack);
1196 				acked = th->th_ack - tp->snd_una;
1197 				tcpstat.tcps_rcvackpack++;
1198 				tcpstat.tcps_rcvackbyte += acked;
1199 				sbdrop(&so->so_snd, acked);
1200 				if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1201 				    SEQ_LEQ(th->th_ack, tp->snd_recover))
1202 					tp->snd_recover = th->th_ack - 1;
1203 				tp->snd_una = th->th_ack;
1204 				/*
1205 				 * pull snd_wl2 up to prevent seq wrap relative
1206 				 * to th_ack.
1207 				 */
1208 				tp->snd_wl2 = th->th_ack;
1209 				tp->t_dupacks = 0;
1210 				m_freem(m);
1211 				ND6_HINT(tp); /* some progress has been done */
1212 
1213 				/*
1214 				 * If all outstanding data are acked, stop
1215 				 * retransmit timer, otherwise restart timer
1216 				 * using current (possibly backed-off) value.
1217 				 * If process is waiting for space,
1218 				 * wakeup/selwakeup/signal.  If data
1219 				 * are ready to send, let tcp_output
1220 				 * decide between more output or persist.
1221 
1222 #ifdef TCPDEBUG
1223 				if (so->so_options & SO_DEBUG)
1224 					tcp_trace(TA_INPUT, ostate, tp,
1225 					    (void *)tcp_saveipgen,
1226 					    &tcp_savetcp, 0);
1227 #endif
1228 				 */
1229 				if (tp->snd_una == tp->snd_max)
1230 					callout_stop(tp->tt_rexmt);
1231 				else if (!callout_active(tp->tt_persist))
1232 					callout_reset(tp->tt_rexmt,
1233 						      tp->t_rxtcur,
1234 						      tcp_timer_rexmt, tp);
1235 
1236 				sowwakeup(so);
1237 				if (so->so_snd.sb_cc)
1238 					(void) tcp_output(tp);
1239 				goto check_delack;
1240 			}
1241 		} else if (th->th_ack == tp->snd_una &&
1242 		    LIST_EMPTY(&tp->t_segq) &&
1243 		    tlen <= sbspace(&so->so_rcv)) {
1244 			KASSERT(headlocked, ("headlocked"));
1245 			INP_INFO_WUNLOCK(&tcbinfo);
1246 			headlocked = 0;
1247 			/*
1248 			 * this is a pure, in-sequence data packet
1249 			 * with nothing on the reassembly queue and
1250 			 * we have enough buffer space to take it.
1251 			 */
1252 			/* Clean receiver SACK report if present */
1253 			if (tp->sack_enable && tp->rcv_numsacks)
1254 				tcp_clean_sackreport(tp);
1255 			++tcpstat.tcps_preddat;
1256 			tp->rcv_nxt += tlen;
1257 			/*
1258 			 * Pull snd_wl1 up to prevent seq wrap relative to
1259 			 * th_seq.
1260 			 */
1261 			tp->snd_wl1 = th->th_seq;
1262 			/*
1263 			 * Pull rcv_up up to prevent seq wrap relative to
1264 			 * rcv_nxt.
1265 			 */
1266 			tp->rcv_up = tp->rcv_nxt;
1267 			tcpstat.tcps_rcvpack++;
1268 			tcpstat.tcps_rcvbyte += tlen;
1269 			ND6_HINT(tp);	/* some progress has been done */
1270 			/*
1271 #ifdef TCPDEBUG
1272 			if (so->so_options & SO_DEBUG)
1273 				tcp_trace(TA_INPUT, ostate, tp,
1274 				    (void *)tcp_saveipgen, &tcp_savetcp, 0);
1275 #endif
1276 			 * Add data to socket buffer.
1277 			 */
1278 			SOCKBUF_LOCK(&so->so_rcv);
1279 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1280 				m_freem(m);
1281 			} else {
1282 				m_adj(m, drop_hdrlen);	/* delayed header drop */
1283 				sbappendstream_locked(&so->so_rcv, m);
1284 			}
1285 			sorwakeup_locked(so);
1286 			if (DELAY_ACK(tp)) {
1287 				tp->t_flags |= TF_DELACK;
1288 			} else {
1289 				tp->t_flags |= TF_ACKNOW;
1290 				tcp_output(tp);
1291 			}
1292 			goto check_delack;
1293 		}
1294 	}
1295 
1296 	/*
1297 	 * Calculate amount of space in receive window,
1298 	 * and then do TCP input processing.
1299 	 * Receive window is amount of space in rcv queue,
1300 	 * but not less than advertised window.
1301 	 */
1302 	{ int win;
1303 
1304 	win = sbspace(&so->so_rcv);
1305 	if (win < 0)
1306 		win = 0;
1307 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1308 	}
1309 
1310 	switch (tp->t_state) {
1311 
1312 	/*
1313 	 * If the state is SYN_RECEIVED:
1314 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
1315 	 */
1316 	case TCPS_SYN_RECEIVED:
1317 		if ((thflags & TH_ACK) &&
1318 		    (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1319 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1320 				rstreason = BANDLIM_RST_OPENPORT;
1321 				goto dropwithreset;
1322 		}
1323 		break;
1324 
1325 	/*
1326 	 * If the state is SYN_SENT:
1327 	 *	if seg contains an ACK, but not for our SYN, drop the input.
1328 	 *	if seg contains a RST, then drop the connection.
1329 	 *	if seg does not contain SYN, then drop it.
1330 	 * Otherwise this is an acceptable SYN segment
1331 	 *	initialize tp->rcv_nxt and tp->irs
1332 	 *	if seg contains ack then advance tp->snd_una
1333 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1334 	 *	arrange for segment to be acked (eventually)
1335 	 *	continue processing rest of data/controls, beginning with URG
1336 	 */
1337 	case TCPS_SYN_SENT:
1338 		if ((thflags & TH_ACK) &&
1339 		    (SEQ_LEQ(th->th_ack, tp->iss) ||
1340 		     SEQ_GT(th->th_ack, tp->snd_max))) {
1341 			rstreason = BANDLIM_UNLIMITED;
1342 			goto dropwithreset;
1343 		}
1344 		if (thflags & TH_RST) {
1345 			if (thflags & TH_ACK) {
1346 				KASSERT(headlocked, ("tcp_input: after_listen"
1347 				    ": tcp_drop.2: head not locked"));
1348 				tp = tcp_drop(tp, ECONNREFUSED);
1349 			}
1350 			goto drop;
1351 		}
1352 		if ((thflags & TH_SYN) == 0)
1353 			goto drop;
1354 		tp->snd_wnd = th->th_win;	/* initial send window */
1355 
1356 		tp->irs = th->th_seq;
1357 		tcp_rcvseqinit(tp);
1358 		if (thflags & TH_ACK) {
1359 			tcpstat.tcps_connects++;
1360 			soisconnected(so);
1361 #ifdef MAC
1362 			SOCK_LOCK(so);
1363 			mac_set_socket_peer_from_mbuf(m, so);
1364 			SOCK_UNLOCK(so);
1365 #endif
1366 			/* Do window scaling on this connection? */
1367 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1368 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1369 				tp->snd_scale = tp->requested_s_scale;
1370 				tp->rcv_scale = tp->request_r_scale;
1371 			}
1372 			tp->rcv_adv += tp->rcv_wnd;
1373 			tp->snd_una++;		/* SYN is acked */
1374 			/*
1375 			 * If there's data, delay ACK; if there's also a FIN
1376 			 * ACKNOW will be turned on later.
1377 			 */
1378 			if (DELAY_ACK(tp) && tlen != 0)
1379 				callout_reset(tp->tt_delack, tcp_delacktime,
1380 				    tcp_timer_delack, tp);
1381 			else
1382 				tp->t_flags |= TF_ACKNOW;
1383 			/*
1384 			 * Received <SYN,ACK> in SYN_SENT[*] state.
1385 			 * Transitions:
1386 			 *	SYN_SENT  --> ESTABLISHED
1387 			 *	SYN_SENT* --> FIN_WAIT_1
1388 			 */
1389 			tp->t_starttime = ticks;
1390 			if (tp->t_flags & TF_NEEDFIN) {
1391 				tp->t_state = TCPS_FIN_WAIT_1;
1392 				tp->t_flags &= ~TF_NEEDFIN;
1393 				thflags &= ~TH_SYN;
1394 			} else {
1395 				tp->t_state = TCPS_ESTABLISHED;
1396 				callout_reset(tp->tt_keep, tcp_keepidle,
1397 					      tcp_timer_keep, tp);
1398 			}
1399 		} else {
1400 			/*
1401 			 * Received initial SYN in SYN-SENT[*] state =>
1402 			 * simultaneous open.  If segment contains CC option
1403 			 * and there is a cached CC, apply TAO test.
1404 			 * If it succeeds, connection is * half-synchronized.
1405 			 * Otherwise, do 3-way handshake:
1406 			 *        SYN-SENT -> SYN-RECEIVED
1407 			 *        SYN-SENT* -> SYN-RECEIVED*
1408 			 * If there was no CC option, clear cached CC value.
1409 			 */
1410 			tp->t_flags |= TF_ACKNOW;
1411 			callout_stop(tp->tt_rexmt);
1412 			tp->t_state = TCPS_SYN_RECEIVED;
1413 		}
1414 
1415 trimthenstep6:
1416 		KASSERT(headlocked, ("tcp_input: trimthenstep6: head not "
1417 		    "locked"));
1418 		INP_LOCK_ASSERT(inp);
1419 
1420 		/*
1421 		 * Advance th->th_seq to correspond to first data byte.
1422 		 * If data, trim to stay within window,
1423 		 * dropping FIN if necessary.
1424 		 */
1425 		th->th_seq++;
1426 		if (tlen > tp->rcv_wnd) {
1427 			todrop = tlen - tp->rcv_wnd;
1428 			m_adj(m, -todrop);
1429 			tlen = tp->rcv_wnd;
1430 			thflags &= ~TH_FIN;
1431 			tcpstat.tcps_rcvpackafterwin++;
1432 			tcpstat.tcps_rcvbyteafterwin += todrop;
1433 		}
1434 		tp->snd_wl1 = th->th_seq - 1;
1435 		tp->rcv_up = th->th_seq;
1436 		/*
1437 		 * Client side of transaction: already sent SYN and data.
1438 		 * If the remote host used T/TCP to validate the SYN,
1439 		 * our data will be ACK'd; if so, enter normal data segment
1440 		 * processing in the middle of step 5, ack processing.
1441 		 * Otherwise, goto step 6.
1442 		 */
1443 		if (thflags & TH_ACK)
1444 			goto process_ACK;
1445 
1446 		goto step6;
1447 
1448 	/*
1449 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1450 	 *      do normal processing.
1451 	 *
1452 	 * NB: Leftover from RFC1644 T/TCP.  Cases to be reused later.
1453 	 */
1454 	case TCPS_LAST_ACK:
1455 	case TCPS_CLOSING:
1456 	case TCPS_TIME_WAIT:
1457 		KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1458 		break;  /* continue normal processing */
1459 	}
1460 
1461 	/*
1462 	 * States other than LISTEN or SYN_SENT.
1463 	 * First check the RST flag and sequence number since reset segments
1464 	 * are exempt from the timestamp and connection count tests.  This
1465 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1466 	 * below which allowed reset segments in half the sequence space
1467 	 * to fall though and be processed (which gives forged reset
1468 	 * segments with a random sequence number a 50 percent chance of
1469 	 * killing a connection).
1470 	 * Then check timestamp, if present.
1471 	 * Then check the connection count, if present.
1472 	 * Then check that at least some bytes of segment are within
1473 	 * receive window.  If segment begins before rcv_nxt,
1474 	 * drop leading data (and SYN); if nothing left, just ack.
1475 	 *
1476 	 *
1477 	 * If the RST bit is set, check the sequence number to see
1478 	 * if this is a valid reset segment.
1479 	 * RFC 793 page 37:
1480 	 *   In all states except SYN-SENT, all reset (RST) segments
1481 	 *   are validated by checking their SEQ-fields.  A reset is
1482 	 *   valid if its sequence number is in the window.
1483 	 * Note: this does not take into account delayed ACKs, so
1484 	 *   we should test against last_ack_sent instead of rcv_nxt.
1485 	 *   The sequence number in the reset segment is normally an
1486 	 *   echo of our outgoing acknowlegement numbers, but some hosts
1487 	 *   send a reset with the sequence number at the rightmost edge
1488 	 *   of our receive window, and we have to handle this case.
1489 	 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1490 	 *   that brute force RST attacks are possible.  To combat this,
1491 	 *   we use a much stricter check while in the ESTABLISHED state,
1492 	 *   only accepting RSTs where the sequence number is equal to
1493 	 *   last_ack_sent.  In all other states (the states in which a
1494 	 *   RST is more likely), the more permissive check is used.
1495 	 * If we have multiple segments in flight, the intial reset
1496 	 * segment sequence numbers will be to the left of last_ack_sent,
1497 	 * but they will eventually catch up.
1498 	 * In any case, it never made sense to trim reset segments to
1499 	 * fit the receive window since RFC 1122 says:
1500 	 *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1501 	 *
1502 	 *    A TCP SHOULD allow a received RST segment to include data.
1503 	 *
1504 	 *    DISCUSSION
1505 	 *         It has been suggested that a RST segment could contain
1506 	 *         ASCII text that encoded and explained the cause of the
1507 	 *         RST.  No standard has yet been established for such
1508 	 *         data.
1509 	 *
1510 	 * If the reset segment passes the sequence number test examine
1511 	 * the state:
1512 	 *    SYN_RECEIVED STATE:
1513 	 *	If passive open, return to LISTEN state.
1514 	 *	If active open, inform user that connection was refused.
1515 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1516 	 *	Inform user that connection was reset, and close tcb.
1517 	 *    CLOSING, LAST_ACK STATES:
1518 	 *	Close the tcb.
1519 	 *    TIME_WAIT STATE:
1520 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
1521 	 *      RFC 1337.
1522 	 */
1523 	if (thflags & TH_RST) {
1524 		if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1525 		    SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
1526 		    (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
1527 			switch (tp->t_state) {
1528 
1529 			case TCPS_SYN_RECEIVED:
1530 				so->so_error = ECONNREFUSED;
1531 				goto close;
1532 
1533 			case TCPS_ESTABLISHED:
1534 				if (tp->last_ack_sent != th->th_seq &&
1535 			 	    tcp_insecure_rst == 0) {
1536 					tcpstat.tcps_badrst++;
1537 					goto drop;
1538 				}
1539 			case TCPS_FIN_WAIT_1:
1540 			case TCPS_FIN_WAIT_2:
1541 			case TCPS_CLOSE_WAIT:
1542 				so->so_error = ECONNRESET;
1543 			close:
1544 				tp->t_state = TCPS_CLOSED;
1545 				tcpstat.tcps_drops++;
1546 				KASSERT(headlocked, ("tcp_input: "
1547 				    "trimthenstep6: tcp_close: head not "
1548 				    "locked"));
1549 				tp = tcp_close(tp);
1550 				break;
1551 
1552 			case TCPS_CLOSING:
1553 			case TCPS_LAST_ACK:
1554 				KASSERT(headlocked, ("trimthenstep6: "
1555 				    "tcp_close.2: head not locked"));
1556 				tp = tcp_close(tp);
1557 				break;
1558 
1559 			case TCPS_TIME_WAIT:
1560 				KASSERT(tp->t_state != TCPS_TIME_WAIT,
1561 				    ("timewait"));
1562 				break;
1563 			}
1564 		}
1565 		goto drop;
1566 	}
1567 
1568 	/*
1569 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1570 	 * and it's less than ts_recent, drop it.
1571 	 */
1572 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1573 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1574 
1575 		/* Check to see if ts_recent is over 24 days old.  */
1576 		if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1577 			/*
1578 			 * Invalidate ts_recent.  If this segment updates
1579 			 * ts_recent, the age will be reset later and ts_recent
1580 			 * will get a valid value.  If it does not, setting
1581 			 * ts_recent to zero will at least satisfy the
1582 			 * requirement that zero be placed in the timestamp
1583 			 * echo reply when ts_recent isn't valid.  The
1584 			 * age isn't reset until we get a valid ts_recent
1585 			 * because we don't want out-of-order segments to be
1586 			 * dropped when ts_recent is old.
1587 			 */
1588 			tp->ts_recent = 0;
1589 		} else {
1590 			tcpstat.tcps_rcvduppack++;
1591 			tcpstat.tcps_rcvdupbyte += tlen;
1592 			tcpstat.tcps_pawsdrop++;
1593 			if (tlen)
1594 				goto dropafterack;
1595 			goto drop;
1596 		}
1597 	}
1598 
1599 	/*
1600 	 * In the SYN-RECEIVED state, validate that the packet belongs to
1601 	 * this connection before trimming the data to fit the receive
1602 	 * window.  Check the sequence number versus IRS since we know
1603 	 * the sequence numbers haven't wrapped.  This is a partial fix
1604 	 * for the "LAND" DoS attack.
1605 	 */
1606 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1607 		rstreason = BANDLIM_RST_OPENPORT;
1608 		goto dropwithreset;
1609 	}
1610 
1611 	todrop = tp->rcv_nxt - th->th_seq;
1612 	if (todrop > 0) {
1613 		if (thflags & TH_SYN) {
1614 			thflags &= ~TH_SYN;
1615 			th->th_seq++;
1616 			if (th->th_urp > 1)
1617 				th->th_urp--;
1618 			else
1619 				thflags &= ~TH_URG;
1620 			todrop--;
1621 		}
1622 		/*
1623 		 * Following if statement from Stevens, vol. 2, p. 960.
1624 		 */
1625 		if (todrop > tlen
1626 		    || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1627 			/*
1628 			 * Any valid FIN must be to the left of the window.
1629 			 * At this point the FIN must be a duplicate or out
1630 			 * of sequence; drop it.
1631 			 */
1632 			thflags &= ~TH_FIN;
1633 
1634 			/*
1635 			 * Send an ACK to resynchronize and drop any data.
1636 			 * But keep on processing for RST or ACK.
1637 			 */
1638 			tp->t_flags |= TF_ACKNOW;
1639 			todrop = tlen;
1640 			tcpstat.tcps_rcvduppack++;
1641 			tcpstat.tcps_rcvdupbyte += todrop;
1642 		} else {
1643 			tcpstat.tcps_rcvpartduppack++;
1644 			tcpstat.tcps_rcvpartdupbyte += todrop;
1645 		}
1646 		drop_hdrlen += todrop;	/* drop from the top afterwards */
1647 		th->th_seq += todrop;
1648 		tlen -= todrop;
1649 		if (th->th_urp > todrop)
1650 			th->th_urp -= todrop;
1651 		else {
1652 			thflags &= ~TH_URG;
1653 			th->th_urp = 0;
1654 		}
1655 	}
1656 
1657 	/*
1658 	 * If new data are received on a connection after the
1659 	 * user processes are gone, then RST the other end.
1660 	 */
1661 	if ((so->so_state & SS_NOFDREF) &&
1662 	    tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1663 		KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not "
1664 		    "locked"));
1665 		tp = tcp_close(tp);
1666 		tcpstat.tcps_rcvafterclose++;
1667 		rstreason = BANDLIM_UNLIMITED;
1668 		goto dropwithreset;
1669 	}
1670 
1671 	/*
1672 	 * If segment ends after window, drop trailing data
1673 	 * (and PUSH and FIN); if nothing left, just ACK.
1674 	 */
1675 	todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
1676 	if (todrop > 0) {
1677 		tcpstat.tcps_rcvpackafterwin++;
1678 		if (todrop >= tlen) {
1679 			tcpstat.tcps_rcvbyteafterwin += tlen;
1680 			/*
1681 			 * If a new connection request is received
1682 			 * while in TIME_WAIT, drop the old connection
1683 			 * and start over if the sequence numbers
1684 			 * are above the previous ones.
1685 			 */
1686 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1687 			if (thflags & TH_SYN &&
1688 			    tp->t_state == TCPS_TIME_WAIT &&
1689 			    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1690 				KASSERT(headlocked, ("trimthenstep6: "
1691 				    "tcp_close.4: head not locked"));
1692 				tp = tcp_close(tp);
1693 				goto findpcb;
1694 			}
1695 			/*
1696 			 * If window is closed can only take segments at
1697 			 * window edge, and have to drop data and PUSH from
1698 			 * incoming segments.  Continue processing, but
1699 			 * remember to ack.  Otherwise, drop segment
1700 			 * and ack.
1701 			 */
1702 			if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1703 				tp->t_flags |= TF_ACKNOW;
1704 				tcpstat.tcps_rcvwinprobe++;
1705 			} else
1706 				goto dropafterack;
1707 		} else
1708 			tcpstat.tcps_rcvbyteafterwin += todrop;
1709 		m_adj(m, -todrop);
1710 		tlen -= todrop;
1711 		thflags &= ~(TH_PUSH|TH_FIN);
1712 	}
1713 
1714 	/*
1715 	 * If last ACK falls within this segment's sequence numbers,
1716 	 * record its timestamp.
1717 	 * NOTE:
1718 	 * 1) That the test incorporates suggestions from the latest
1719 	 *    proposal of the tcplw@cray.com list (Braden 1993/04/26).
1720 	 * 2) That updating only on newer timestamps interferes with
1721 	 *    our earlier PAWS tests, so this check should be solely
1722 	 *    predicated on the sequence space of this segment.
1723 	 * 3) That we modify the segment boundary check to be
1724 	 *        Last.ACK.Sent <= SEG.SEQ + SEG.Len
1725 	 *    instead of RFC1323's
1726 	 *        Last.ACK.Sent < SEG.SEQ + SEG.Len,
1727 	 *    This modified check allows us to overcome RFC1323's
1728 	 *    limitations as described in Stevens TCP/IP Illustrated
1729 	 *    Vol. 2 p.869. In such cases, we can still calculate the
1730 	 *    RTT correctly when RCV.NXT == Last.ACK.Sent.
1731 	 */
1732 	if ((to.to_flags & TOF_TS) != 0 &&
1733 	    SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1734 	    SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1735 		((thflags & (TH_SYN|TH_FIN)) != 0))) {
1736 		tp->ts_recent_age = ticks;
1737 		tp->ts_recent = to.to_tsval;
1738 	}
1739 
1740 	/*
1741 	 * If a SYN is in the window, then this is an
1742 	 * error and we send an RST and drop the connection.
1743 	 */
1744 	if (thflags & TH_SYN) {
1745 		KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: "
1746 		    "head not locked"));
1747 		tp = tcp_drop(tp, ECONNRESET);
1748 		rstreason = BANDLIM_UNLIMITED;
1749 		goto drop;
1750 	}
1751 
1752 	/*
1753 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1754 	 * flag is on (half-synchronized state), then queue data for
1755 	 * later processing; else drop segment and return.
1756 	 */
1757 	if ((thflags & TH_ACK) == 0) {
1758 		if (tp->t_state == TCPS_SYN_RECEIVED ||
1759 		    (tp->t_flags & TF_NEEDSYN))
1760 			goto step6;
1761 		else
1762 			goto drop;
1763 	}
1764 
1765 	/*
1766 	 * Ack processing.
1767 	 */
1768 	switch (tp->t_state) {
1769 
1770 	/*
1771 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1772 	 * ESTABLISHED state and continue processing.
1773 	 * The ACK was checked above.
1774 	 */
1775 	case TCPS_SYN_RECEIVED:
1776 
1777 		tcpstat.tcps_connects++;
1778 		soisconnected(so);
1779 		/* Do window scaling? */
1780 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1781 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1782 			tp->snd_scale = tp->requested_s_scale;
1783 			tp->rcv_scale = tp->request_r_scale;
1784 		}
1785 		/*
1786 		 * Make transitions:
1787 		 *      SYN-RECEIVED  -> ESTABLISHED
1788 		 *      SYN-RECEIVED* -> FIN-WAIT-1
1789 		 */
1790 		tp->t_starttime = ticks;
1791 		if (tp->t_flags & TF_NEEDFIN) {
1792 			tp->t_state = TCPS_FIN_WAIT_1;
1793 			tp->t_flags &= ~TF_NEEDFIN;
1794 		} else {
1795 			tp->t_state = TCPS_ESTABLISHED;
1796 			callout_reset(tp->tt_keep, tcp_keepidle,
1797 				      tcp_timer_keep, tp);
1798 		}
1799 		/*
1800 		 * If segment contains data or ACK, will call tcp_reass()
1801 		 * later; if not, do so now to pass queued data to user.
1802 		 */
1803 		if (tlen == 0 && (thflags & TH_FIN) == 0)
1804 			(void) tcp_reass(tp, (struct tcphdr *)0, 0,
1805 			    (struct mbuf *)0);
1806 		tp->snd_wl1 = th->th_seq - 1;
1807 		/* FALLTHROUGH */
1808 
1809 	/*
1810 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1811 	 * ACKs.  If the ack is in the range
1812 	 *	tp->snd_una < th->th_ack <= tp->snd_max
1813 	 * then advance tp->snd_una to th->th_ack and drop
1814 	 * data from the retransmission queue.  If this ACK reflects
1815 	 * more up to date window information we update our window information.
1816 	 */
1817 	case TCPS_ESTABLISHED:
1818 	case TCPS_FIN_WAIT_1:
1819 	case TCPS_FIN_WAIT_2:
1820 	case TCPS_CLOSE_WAIT:
1821 	case TCPS_CLOSING:
1822 	case TCPS_LAST_ACK:
1823 	case TCPS_TIME_WAIT:
1824 		KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
1825 		if (SEQ_GT(th->th_ack, tp->snd_max)) {
1826 			tcpstat.tcps_rcvacktoomuch++;
1827 			goto dropafterack;
1828 		}
1829 		if (tp->sack_enable &&
1830 		    (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes)))
1831 			tcp_sack_doack(tp, &to, th->th_ack);
1832 		if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1833 			if (tlen == 0 && tiwin == tp->snd_wnd) {
1834 				tcpstat.tcps_rcvdupack++;
1835 				/*
1836 				 * If we have outstanding data (other than
1837 				 * a window probe), this is a completely
1838 				 * duplicate ack (ie, window info didn't
1839 				 * change), the ack is the biggest we've
1840 				 * seen and we've seen exactly our rexmt
1841 				 * threshhold of them, assume a packet
1842 				 * has been dropped and retransmit it.
1843 				 * Kludge snd_nxt & the congestion
1844 				 * window so we send only this one
1845 				 * packet.
1846 				 *
1847 				 * We know we're losing at the current
1848 				 * window size so do congestion avoidance
1849 				 * (set ssthresh to half the current window
1850 				 * and pull our congestion window back to
1851 				 * the new ssthresh).
1852 				 *
1853 				 * Dup acks mean that packets have left the
1854 				 * network (they're now cached at the receiver)
1855 				 * so bump cwnd by the amount in the receiver
1856 				 * to keep a constant cwnd packets in the
1857 				 * network.
1858 				 */
1859 				if (!callout_active(tp->tt_rexmt) ||
1860 				    th->th_ack != tp->snd_una)
1861 					tp->t_dupacks = 0;
1862 				else if (++tp->t_dupacks > tcprexmtthresh ||
1863 					 ((tcp_do_newreno || tp->sack_enable) &&
1864 					  IN_FASTRECOVERY(tp))) {
1865                                         if (tp->sack_enable && IN_FASTRECOVERY(tp)) {
1866 						int awnd;
1867 
1868 						/*
1869 						 * Compute the amount of data in flight first.
1870 						 * We can inject new data into the pipe iff
1871 						 * we have less than 1/2 the original window's
1872 						 * worth of data in flight.
1873 						 */
1874 						awnd = (tp->snd_nxt - tp->snd_fack) +
1875 							tp->sackhint.sack_bytes_rexmit;
1876 						if (awnd < tp->snd_ssthresh) {
1877 							tp->snd_cwnd += tp->t_maxseg;
1878 							if (tp->snd_cwnd > tp->snd_ssthresh)
1879 								tp->snd_cwnd = tp->snd_ssthresh;
1880 						}
1881 					} else
1882 						tp->snd_cwnd += tp->t_maxseg;
1883 					(void) tcp_output(tp);
1884 					goto drop;
1885 				} else if (tp->t_dupacks == tcprexmtthresh) {
1886 					tcp_seq onxt = tp->snd_nxt;
1887 					u_int win;
1888 
1889 					/*
1890 					 * If we're doing sack, check to
1891 					 * see if we're already in sack
1892 					 * recovery. If we're not doing sack,
1893 					 * check to see if we're in newreno
1894 					 * recovery.
1895 					 */
1896 					if (tp->sack_enable) {
1897 						if (IN_FASTRECOVERY(tp)) {
1898 							tp->t_dupacks = 0;
1899 							break;
1900 						}
1901 					} else if (tcp_do_newreno) {
1902 						if (SEQ_LEQ(th->th_ack,
1903 						    tp->snd_recover)) {
1904 							tp->t_dupacks = 0;
1905 							break;
1906 						}
1907 					}
1908 					win = min(tp->snd_wnd, tp->snd_cwnd) /
1909 					    2 / tp->t_maxseg;
1910 					if (win < 2)
1911 						win = 2;
1912 					tp->snd_ssthresh = win * tp->t_maxseg;
1913 					ENTER_FASTRECOVERY(tp);
1914 					tp->snd_recover = tp->snd_max;
1915 					callout_stop(tp->tt_rexmt);
1916 					tp->t_rtttime = 0;
1917 					if (tp->sack_enable) {
1918 						tcpstat.tcps_sack_recovery_episode++;
1919 						tp->sack_newdata = tp->snd_nxt;
1920 						tp->snd_cwnd = tp->t_maxseg;
1921 						(void) tcp_output(tp);
1922 						goto drop;
1923 					}
1924 					tp->snd_nxt = th->th_ack;
1925 					tp->snd_cwnd = tp->t_maxseg;
1926 					(void) tcp_output(tp);
1927 					KASSERT(tp->snd_limited <= 2,
1928 					    ("tp->snd_limited too big"));
1929 					tp->snd_cwnd = tp->snd_ssthresh +
1930 					     tp->t_maxseg *
1931 					     (tp->t_dupacks - tp->snd_limited);
1932 					if (SEQ_GT(onxt, tp->snd_nxt))
1933 						tp->snd_nxt = onxt;
1934 					goto drop;
1935 				} else if (tcp_do_rfc3042) {
1936 					u_long oldcwnd = tp->snd_cwnd;
1937 					tcp_seq oldsndmax = tp->snd_max;
1938 					u_int sent;
1939 
1940 					KASSERT(tp->t_dupacks == 1 ||
1941 					    tp->t_dupacks == 2,
1942 					    ("dupacks not 1 or 2"));
1943 					if (tp->t_dupacks == 1)
1944 						tp->snd_limited = 0;
1945 					tp->snd_cwnd =
1946 					    (tp->snd_nxt - tp->snd_una) +
1947 					    (tp->t_dupacks - tp->snd_limited) *
1948 					    tp->t_maxseg;
1949 					(void) tcp_output(tp);
1950 					sent = tp->snd_max - oldsndmax;
1951 					if (sent > tp->t_maxseg) {
1952 						KASSERT((tp->t_dupacks == 2 &&
1953 						    tp->snd_limited == 0) ||
1954 						   (sent == tp->t_maxseg + 1 &&
1955 						    tp->t_flags & TF_SENTFIN),
1956 						    ("sent too much"));
1957 						tp->snd_limited = 2;
1958 					} else if (sent > 0)
1959 						++tp->snd_limited;
1960 					tp->snd_cwnd = oldcwnd;
1961 					goto drop;
1962 				}
1963 			} else
1964 				tp->t_dupacks = 0;
1965 			break;
1966 		}
1967 
1968 		KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
1969 
1970 		/*
1971 		 * If the congestion window was inflated to account
1972 		 * for the other side's cached packets, retract it.
1973 		 */
1974 		if (tcp_do_newreno || tp->sack_enable) {
1975 			if (IN_FASTRECOVERY(tp)) {
1976 				if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1977 					if (tp->sack_enable)
1978 						tcp_sack_partialack(tp, th);
1979 					else
1980 						tcp_newreno_partial_ack(tp, th);
1981 				} else {
1982 					/*
1983 					 * Out of fast recovery.
1984 					 * Window inflation should have left us
1985 					 * with approximately snd_ssthresh
1986 					 * outstanding data.
1987 					 * But in case we would be inclined to
1988 					 * send a burst, better to do it via
1989 					 * the slow start mechanism.
1990 					 */
1991 					if (SEQ_GT(th->th_ack +
1992 							tp->snd_ssthresh,
1993 						   tp->snd_max))
1994 						tp->snd_cwnd = tp->snd_max -
1995 								th->th_ack +
1996 								tp->t_maxseg;
1997 					else
1998 						tp->snd_cwnd = tp->snd_ssthresh;
1999 				}
2000 			}
2001 		} else {
2002 			if (tp->t_dupacks >= tcprexmtthresh &&
2003 			    tp->snd_cwnd > tp->snd_ssthresh)
2004 				tp->snd_cwnd = tp->snd_ssthresh;
2005 		}
2006 		tp->t_dupacks = 0;
2007 		/*
2008 		 * If we reach this point, ACK is not a duplicate,
2009 		 *     i.e., it ACKs something we sent.
2010 		 */
2011 		if (tp->t_flags & TF_NEEDSYN) {
2012 			/*
2013 			 * T/TCP: Connection was half-synchronized, and our
2014 			 * SYN has been ACK'd (so connection is now fully
2015 			 * synchronized).  Go to non-starred state,
2016 			 * increment snd_una for ACK of SYN, and check if
2017 			 * we can do window scaling.
2018 			 */
2019 			tp->t_flags &= ~TF_NEEDSYN;
2020 			tp->snd_una++;
2021 			/* Do window scaling? */
2022 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2023 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
2024 				tp->snd_scale = tp->requested_s_scale;
2025 				tp->rcv_scale = tp->request_r_scale;
2026 			}
2027 		}
2028 
2029 process_ACK:
2030 		KASSERT(headlocked, ("tcp_input: process_ACK: head not "
2031 		    "locked"));
2032 		INP_LOCK_ASSERT(inp);
2033 
2034 		acked = th->th_ack - tp->snd_una;
2035 		tcpstat.tcps_rcvackpack++;
2036 		tcpstat.tcps_rcvackbyte += acked;
2037 
2038 		/*
2039 		 * If we just performed our first retransmit, and the ACK
2040 		 * arrives within our recovery window, then it was a mistake
2041 		 * to do the retransmit in the first place.  Recover our
2042 		 * original cwnd and ssthresh, and proceed to transmit where
2043 		 * we left off.
2044 		 */
2045 		if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2046 			++tcpstat.tcps_sndrexmitbad;
2047 			tp->snd_cwnd = tp->snd_cwnd_prev;
2048 			tp->snd_ssthresh = tp->snd_ssthresh_prev;
2049 			tp->snd_recover = tp->snd_recover_prev;
2050 			if (tp->t_flags & TF_WASFRECOVERY)
2051 				ENTER_FASTRECOVERY(tp);
2052 			tp->snd_nxt = tp->snd_max;
2053 			tp->t_badrxtwin = 0;	/* XXX probably not required */
2054 		}
2055 
2056 		/*
2057 		 * If we have a timestamp reply, update smoothed
2058 		 * round trip time.  If no timestamp is present but
2059 		 * transmit timer is running and timed sequence
2060 		 * number was acked, update smoothed round trip time.
2061 		 * Since we now have an rtt measurement, cancel the
2062 		 * timer backoff (cf., Phil Karn's retransmit alg.).
2063 		 * Recompute the initial retransmit timer.
2064 		 *
2065 		 * Some boxes send broken timestamp replies
2066 		 * during the SYN+ACK phase, ignore
2067 		 * timestamps of 0 or we could calculate a
2068 		 * huge RTT and blow up the retransmit timer.
2069 		 */
2070 		if ((to.to_flags & TOF_TS) != 0 &&
2071 		    to.to_tsecr) {
2072 			tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2073 		} else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2074 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2075 		}
2076 		tcp_xmit_bandwidth_limit(tp, th->th_ack);
2077 
2078 		/*
2079 		 * If all outstanding data is acked, stop retransmit
2080 		 * timer and remember to restart (more output or persist).
2081 		 * If there is more data to be acked, restart retransmit
2082 		 * timer, using current (possibly backed-off) value.
2083 		 */
2084 		if (th->th_ack == tp->snd_max) {
2085 			callout_stop(tp->tt_rexmt);
2086 			needoutput = 1;
2087 		} else if (!callout_active(tp->tt_persist))
2088 			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2089 				      tcp_timer_rexmt, tp);
2090 
2091 		/*
2092 		 * If no data (only SYN) was ACK'd,
2093 		 *    skip rest of ACK processing.
2094 		 */
2095 		if (acked == 0)
2096 			goto step6;
2097 
2098 		/*
2099 		 * When new data is acked, open the congestion window.
2100 		 * If the window gives us less than ssthresh packets
2101 		 * in flight, open exponentially (maxseg per packet).
2102 		 * Otherwise open linearly: maxseg per window
2103 		 * (maxseg^2 / cwnd per packet).
2104 		 */
2105 		if ((!tcp_do_newreno && !tp->sack_enable) ||
2106 		    !IN_FASTRECOVERY(tp)) {
2107 			register u_int cw = tp->snd_cwnd;
2108 			register u_int incr = tp->t_maxseg;
2109 			if (cw > tp->snd_ssthresh)
2110 				incr = incr * incr / cw;
2111 			tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2112 		}
2113 		SOCKBUF_LOCK(&so->so_snd);
2114 		if (acked > so->so_snd.sb_cc) {
2115 			tp->snd_wnd -= so->so_snd.sb_cc;
2116 			sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2117 			ourfinisacked = 1;
2118 		} else {
2119 			sbdrop_locked(&so->so_snd, acked);
2120 			tp->snd_wnd -= acked;
2121 			ourfinisacked = 0;
2122 		}
2123 		sowwakeup_locked(so);
2124 		/* detect una wraparound */
2125 		if ((tcp_do_newreno || tp->sack_enable) &&
2126 		    !IN_FASTRECOVERY(tp) &&
2127 		    SEQ_GT(tp->snd_una, tp->snd_recover) &&
2128 		    SEQ_LEQ(th->th_ack, tp->snd_recover))
2129 			tp->snd_recover = th->th_ack - 1;
2130 		if ((tcp_do_newreno || tp->sack_enable) &&
2131 		    IN_FASTRECOVERY(tp) &&
2132 		    SEQ_GEQ(th->th_ack, tp->snd_recover))
2133 			EXIT_FASTRECOVERY(tp);
2134 		tp->snd_una = th->th_ack;
2135 		if (tp->sack_enable) {
2136 			if (SEQ_GT(tp->snd_una, tp->snd_recover))
2137 				tp->snd_recover = tp->snd_una;
2138 		}
2139 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2140 			tp->snd_nxt = tp->snd_una;
2141 
2142 		switch (tp->t_state) {
2143 
2144 		/*
2145 		 * In FIN_WAIT_1 STATE in addition to the processing
2146 		 * for the ESTABLISHED state if our FIN is now acknowledged
2147 		 * then enter FIN_WAIT_2.
2148 		 */
2149 		case TCPS_FIN_WAIT_1:
2150 			if (ourfinisacked) {
2151 				/*
2152 				 * If we can't receive any more
2153 				 * data, then closing user can proceed.
2154 				 * Starting the timer is contrary to the
2155 				 * specification, but if we don't get a FIN
2156 				 * we'll hang forever.
2157 				 */
2158 		/* XXXjl
2159 		 * we should release the tp also, and use a
2160 		 * compressed state.
2161 		 */
2162 				if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2163 					soisdisconnected(so);
2164 					callout_reset(tp->tt_2msl, tcp_maxidle,
2165 						      tcp_timer_2msl, tp);
2166 				}
2167 				tp->t_state = TCPS_FIN_WAIT_2;
2168 			}
2169 			break;
2170 
2171 		/*
2172 		 * In CLOSING STATE in addition to the processing for
2173 		 * the ESTABLISHED state if the ACK acknowledges our FIN
2174 		 * then enter the TIME-WAIT state, otherwise ignore
2175 		 * the segment.
2176 		 */
2177 		case TCPS_CLOSING:
2178 			if (ourfinisacked) {
2179 				KASSERT(headlocked, ("tcp_input: process_ACK: "
2180 				    "head not locked"));
2181 				tcp_twstart(tp);
2182 				INP_INFO_WUNLOCK(&tcbinfo);
2183 				m_freem(m);
2184 				return;
2185 			}
2186 			break;
2187 
2188 		/*
2189 		 * In LAST_ACK, we may still be waiting for data to drain
2190 		 * and/or to be acked, as well as for the ack of our FIN.
2191 		 * If our FIN is now acknowledged, delete the TCB,
2192 		 * enter the closed state and return.
2193 		 */
2194 		case TCPS_LAST_ACK:
2195 			if (ourfinisacked) {
2196 				KASSERT(headlocked, ("tcp_input: process_ACK:"
2197 				    " tcp_close: head not locked"));
2198 				tp = tcp_close(tp);
2199 				goto drop;
2200 			}
2201 			break;
2202 
2203 		/*
2204 		 * In TIME_WAIT state the only thing that should arrive
2205 		 * is a retransmission of the remote FIN.  Acknowledge
2206 		 * it and restart the finack timer.
2207 		 */
2208 		case TCPS_TIME_WAIT:
2209 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2210 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
2211 				      tcp_timer_2msl, tp);
2212 			goto dropafterack;
2213 		}
2214 	}
2215 
2216 step6:
2217 	KASSERT(headlocked, ("tcp_input: step6: head not locked"));
2218 	INP_LOCK_ASSERT(inp);
2219 
2220 	/*
2221 	 * Update window information.
2222 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2223 	 */
2224 	if ((thflags & TH_ACK) &&
2225 	    (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2226 	    (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2227 	     (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2228 		/* keep track of pure window updates */
2229 		if (tlen == 0 &&
2230 		    tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2231 			tcpstat.tcps_rcvwinupd++;
2232 		tp->snd_wnd = tiwin;
2233 		tp->snd_wl1 = th->th_seq;
2234 		tp->snd_wl2 = th->th_ack;
2235 		if (tp->snd_wnd > tp->max_sndwnd)
2236 			tp->max_sndwnd = tp->snd_wnd;
2237 		needoutput = 1;
2238 	}
2239 
2240 	/*
2241 	 * Process segments with URG.
2242 	 */
2243 	if ((thflags & TH_URG) && th->th_urp &&
2244 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2245 		/*
2246 		 * This is a kludge, but if we receive and accept
2247 		 * random urgent pointers, we'll crash in
2248 		 * soreceive.  It's hard to imagine someone
2249 		 * actually wanting to send this much urgent data.
2250 		 */
2251 		SOCKBUF_LOCK(&so->so_rcv);
2252 		if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2253 			th->th_urp = 0;			/* XXX */
2254 			thflags &= ~TH_URG;		/* XXX */
2255 			SOCKBUF_UNLOCK(&so->so_rcv);	/* XXX */
2256 			goto dodata;			/* XXX */
2257 		}
2258 		/*
2259 		 * If this segment advances the known urgent pointer,
2260 		 * then mark the data stream.  This should not happen
2261 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2262 		 * a FIN has been received from the remote side.
2263 		 * In these states we ignore the URG.
2264 		 *
2265 		 * According to RFC961 (Assigned Protocols),
2266 		 * the urgent pointer points to the last octet
2267 		 * of urgent data.  We continue, however,
2268 		 * to consider it to indicate the first octet
2269 		 * of data past the urgent section as the original
2270 		 * spec states (in one of two places).
2271 		 */
2272 		if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2273 			tp->rcv_up = th->th_seq + th->th_urp;
2274 			so->so_oobmark = so->so_rcv.sb_cc +
2275 			    (tp->rcv_up - tp->rcv_nxt) - 1;
2276 			if (so->so_oobmark == 0)
2277 				so->so_rcv.sb_state |= SBS_RCVATMARK;
2278 			sohasoutofband(so);
2279 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2280 		}
2281 		SOCKBUF_UNLOCK(&so->so_rcv);
2282 		/*
2283 		 * Remove out of band data so doesn't get presented to user.
2284 		 * This can happen independent of advancing the URG pointer,
2285 		 * but if two URG's are pending at once, some out-of-band
2286 		 * data may creep in... ick.
2287 		 */
2288 		if (th->th_urp <= (u_long)tlen &&
2289 		    !(so->so_options & SO_OOBINLINE)) {
2290 			/* hdr drop is delayed */
2291 			tcp_pulloutofband(so, th, m, drop_hdrlen);
2292 		}
2293 	} else {
2294 		/*
2295 		 * If no out of band data is expected,
2296 		 * pull receive urgent pointer along
2297 		 * with the receive window.
2298 		 */
2299 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2300 			tp->rcv_up = tp->rcv_nxt;
2301 	}
2302 dodata:							/* XXX */
2303 	KASSERT(headlocked, ("tcp_input: dodata: head not locked"));
2304 	INP_LOCK_ASSERT(inp);
2305 
2306 	/*
2307 	 * Process the segment text, merging it into the TCP sequencing queue,
2308 	 * and arranging for acknowledgment of receipt if necessary.
2309 	 * This process logically involves adjusting tp->rcv_wnd as data
2310 	 * is presented to the user (this happens in tcp_usrreq.c,
2311 	 * case PRU_RCVD).  If a FIN has already been received on this
2312 	 * connection then we just ignore the text.
2313 	 */
2314 	if ((tlen || (thflags & TH_FIN)) &&
2315 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2316 		tcp_seq save_start = th->th_seq;
2317 		tcp_seq save_end = th->th_seq + tlen;
2318 		m_adj(m, drop_hdrlen);	/* delayed header drop */
2319 		/*
2320 		 * Insert segment which includes th into TCP reassembly queue
2321 		 * with control block tp.  Set thflags to whether reassembly now
2322 		 * includes a segment with FIN.  This handles the common case
2323 		 * inline (segment is the next to be received on an established
2324 		 * connection, and the queue is empty), avoiding linkage into
2325 		 * and removal from the queue and repetition of various
2326 		 * conversions.
2327 		 * Set DELACK for segments received in order, but ack
2328 		 * immediately when segments are out of order (so
2329 		 * fast retransmit can work).
2330 		 */
2331 		if (th->th_seq == tp->rcv_nxt &&
2332 		    LIST_EMPTY(&tp->t_segq) &&
2333 		    TCPS_HAVEESTABLISHED(tp->t_state)) {
2334 			if (DELAY_ACK(tp))
2335 				tp->t_flags |= TF_DELACK;
2336 			else
2337 				tp->t_flags |= TF_ACKNOW;
2338 			tp->rcv_nxt += tlen;
2339 			thflags = th->th_flags & TH_FIN;
2340 			tcpstat.tcps_rcvpack++;
2341 			tcpstat.tcps_rcvbyte += tlen;
2342 			ND6_HINT(tp);
2343 			SOCKBUF_LOCK(&so->so_rcv);
2344 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2345 				m_freem(m);
2346 			else
2347 				sbappendstream_locked(&so->so_rcv, m);
2348 			sorwakeup_locked(so);
2349 		} else {
2350 			thflags = tcp_reass(tp, th, &tlen, m);
2351 			tp->t_flags |= TF_ACKNOW;
2352 		}
2353 		if (tlen > 0 && tp->sack_enable)
2354 			tcp_update_sack_list(tp, save_start, save_end);
2355 		/*
2356 		 * Note the amount of data that peer has sent into
2357 		 * our window, in order to estimate the sender's
2358 		 * buffer size.
2359 		 */
2360 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2361 	} else {
2362 		m_freem(m);
2363 		thflags &= ~TH_FIN;
2364 	}
2365 
2366 	/*
2367 	 * If FIN is received ACK the FIN and let the user know
2368 	 * that the connection is closing.
2369 	 */
2370 	if (thflags & TH_FIN) {
2371 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2372 			socantrcvmore(so);
2373 			/*
2374 			 * If connection is half-synchronized
2375 			 * (ie NEEDSYN flag on) then delay ACK,
2376 			 * so it may be piggybacked when SYN is sent.
2377 			 * Otherwise, since we received a FIN then no
2378 			 * more input can be expected, send ACK now.
2379 			 */
2380 			if (tp->t_flags & TF_NEEDSYN)
2381 				tp->t_flags |= TF_DELACK;
2382 			else
2383 				tp->t_flags |= TF_ACKNOW;
2384 			tp->rcv_nxt++;
2385 		}
2386 		switch (tp->t_state) {
2387 
2388 		/*
2389 		 * In SYN_RECEIVED and ESTABLISHED STATES
2390 		 * enter the CLOSE_WAIT state.
2391 		 */
2392 		case TCPS_SYN_RECEIVED:
2393 			tp->t_starttime = ticks;
2394 			/*FALLTHROUGH*/
2395 		case TCPS_ESTABLISHED:
2396 			tp->t_state = TCPS_CLOSE_WAIT;
2397 			break;
2398 
2399 		/*
2400 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2401 		 * enter the CLOSING state.
2402 		 */
2403 		case TCPS_FIN_WAIT_1:
2404 			tp->t_state = TCPS_CLOSING;
2405 			break;
2406 
2407 		/*
2408 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2409 		 * starting the time-wait timer, turning off the other
2410 		 * standard timers.
2411 		 */
2412 		case TCPS_FIN_WAIT_2:
2413 			KASSERT(headlocked == 1, ("tcp_input: dodata: "
2414 			    "TCP_FIN_WAIT_2: head not locked"));
2415 			tcp_twstart(tp);
2416 			INP_INFO_WUNLOCK(&tcbinfo);
2417 			return;
2418 
2419 		/*
2420 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2421 		 */
2422 		case TCPS_TIME_WAIT:
2423 			KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait"));
2424 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
2425 				      tcp_timer_2msl, tp);
2426 			break;
2427 		}
2428 	}
2429 	INP_INFO_WUNLOCK(&tcbinfo);
2430 	headlocked = 0;
2431 #ifdef TCPDEBUG
2432 	if (so->so_options & SO_DEBUG)
2433 		tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2434 			  &tcp_savetcp, 0);
2435 #endif
2436 
2437 	/*
2438 	 * Return any desired output.
2439 	 */
2440 	if (needoutput || (tp->t_flags & TF_ACKNOW))
2441 		(void) tcp_output(tp);
2442 
2443 check_delack:
2444 	KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked"));
2445 	INP_LOCK_ASSERT(inp);
2446 	if (tp->t_flags & TF_DELACK) {
2447 		tp->t_flags &= ~TF_DELACK;
2448 		callout_reset(tp->tt_delack, tcp_delacktime,
2449 		    tcp_timer_delack, tp);
2450 	}
2451 	INP_UNLOCK(inp);
2452 	return;
2453 
2454 dropafterack:
2455 	KASSERT(headlocked, ("tcp_input: dropafterack: head not locked"));
2456 	/*
2457 	 * Generate an ACK dropping incoming segment if it occupies
2458 	 * sequence space, where the ACK reflects our state.
2459 	 *
2460 	 * We can now skip the test for the RST flag since all
2461 	 * paths to this code happen after packets containing
2462 	 * RST have been dropped.
2463 	 *
2464 	 * In the SYN-RECEIVED state, don't send an ACK unless the
2465 	 * segment we received passes the SYN-RECEIVED ACK test.
2466 	 * If it fails send a RST.  This breaks the loop in the
2467 	 * "LAND" DoS attack, and also prevents an ACK storm
2468 	 * between two listening ports that have been sent forged
2469 	 * SYN segments, each with the source address of the other.
2470 	 */
2471 	if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2472 	    (SEQ_GT(tp->snd_una, th->th_ack) ||
2473 	     SEQ_GT(th->th_ack, tp->snd_max)) ) {
2474 		rstreason = BANDLIM_RST_OPENPORT;
2475 		goto dropwithreset;
2476 	}
2477 #ifdef TCPDEBUG
2478 	if (so->so_options & SO_DEBUG)
2479 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2480 			  &tcp_savetcp, 0);
2481 #endif
2482 	KASSERT(headlocked, ("headlocked should be 1"));
2483 	INP_INFO_WUNLOCK(&tcbinfo);
2484 	tp->t_flags |= TF_ACKNOW;
2485 	(void) tcp_output(tp);
2486 	INP_UNLOCK(inp);
2487 	m_freem(m);
2488 	return;
2489 
2490 dropwithreset:
2491 	KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked"));
2492 	/*
2493 	 * Generate a RST, dropping incoming segment.
2494 	 * Make ACK acceptable to originator of segment.
2495 	 * Don't bother to respond if destination was broadcast/multicast.
2496 	 */
2497 	if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2498 		goto drop;
2499 	if (isipv6) {
2500 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2501 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2502 			goto drop;
2503 	} else {
2504 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2505 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2506 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2507 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2508 			goto drop;
2509 	}
2510 	/* IPv6 anycast check is done at tcp6_input() */
2511 
2512 	/*
2513 	 * Perform bandwidth limiting.
2514 	 */
2515 	if (badport_bandlim(rstreason) < 0)
2516 		goto drop;
2517 
2518 #ifdef TCPDEBUG
2519 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2520 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2521 			  &tcp_savetcp, 0);
2522 #endif
2523 
2524 	if (thflags & TH_ACK)
2525 		/* mtod() below is safe as long as hdr dropping is delayed */
2526 		tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2527 			    TH_RST);
2528 	else {
2529 		if (thflags & TH_SYN)
2530 			tlen++;
2531 		/* mtod() below is safe as long as hdr dropping is delayed */
2532 		tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2533 			    (tcp_seq)0, TH_RST|TH_ACK);
2534 	}
2535 
2536 	if (tp)
2537 		INP_UNLOCK(inp);
2538 	if (headlocked)
2539 		INP_INFO_WUNLOCK(&tcbinfo);
2540 	return;
2541 
2542 drop:
2543 	/*
2544 	 * Drop space held by incoming segment and return.
2545 	 */
2546 #ifdef TCPDEBUG
2547 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2548 		tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2549 			  &tcp_savetcp, 0);
2550 #endif
2551 	if (tp)
2552 		INP_UNLOCK(inp);
2553 	if (headlocked)
2554 		INP_INFO_WUNLOCK(&tcbinfo);
2555 	m_freem(m);
2556 	return;
2557 }
2558 
2559 /*
2560  * Parse TCP options and place in tcpopt.
2561  */
2562 static void
2563 tcp_dooptions(to, cp, cnt, is_syn)
2564 	struct tcpopt *to;
2565 	u_char *cp;
2566 	int cnt;
2567 	int is_syn;
2568 {
2569 	int opt, optlen;
2570 
2571 	to->to_flags = 0;
2572 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
2573 		opt = cp[0];
2574 		if (opt == TCPOPT_EOL)
2575 			break;
2576 		if (opt == TCPOPT_NOP)
2577 			optlen = 1;
2578 		else {
2579 			if (cnt < 2)
2580 				break;
2581 			optlen = cp[1];
2582 			if (optlen < 2 || optlen > cnt)
2583 				break;
2584 		}
2585 		switch (opt) {
2586 		case TCPOPT_MAXSEG:
2587 			if (optlen != TCPOLEN_MAXSEG)
2588 				continue;
2589 			if (!is_syn)
2590 				continue;
2591 			to->to_flags |= TOF_MSS;
2592 			bcopy((char *)cp + 2,
2593 			    (char *)&to->to_mss, sizeof(to->to_mss));
2594 			to->to_mss = ntohs(to->to_mss);
2595 			break;
2596 		case TCPOPT_WINDOW:
2597 			if (optlen != TCPOLEN_WINDOW)
2598 				continue;
2599 			if (! is_syn)
2600 				continue;
2601 			to->to_flags |= TOF_SCALE;
2602 			to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2603 			break;
2604 		case TCPOPT_TIMESTAMP:
2605 			if (optlen != TCPOLEN_TIMESTAMP)
2606 				continue;
2607 			to->to_flags |= TOF_TS;
2608 			bcopy((char *)cp + 2,
2609 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
2610 			to->to_tsval = ntohl(to->to_tsval);
2611 			bcopy((char *)cp + 6,
2612 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2613 			to->to_tsecr = ntohl(to->to_tsecr);
2614 			/*
2615 			 * If echoed timestamp is later than the current time,
2616 			 * fall back to non RFC1323 RTT calculation.
2617 			 */
2618 			if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks))
2619 				to->to_tsecr = 0;
2620 			break;
2621 #ifdef TCP_SIGNATURE
2622 		/*
2623 		 * XXX In order to reply to a host which has set the
2624 		 * TCP_SIGNATURE option in its initial SYN, we have to
2625 		 * record the fact that the option was observed here
2626 		 * for the syncache code to perform the correct response.
2627 		 */
2628 		case TCPOPT_SIGNATURE:
2629 			if (optlen != TCPOLEN_SIGNATURE)
2630 				continue;
2631 			to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2632 			break;
2633 #endif
2634 		case TCPOPT_SACK_PERMITTED:
2635 			if (!tcp_do_sack ||
2636 			    optlen != TCPOLEN_SACK_PERMITTED)
2637 				continue;
2638 			if (is_syn) {
2639 				/* MUST only be set on SYN */
2640 				to->to_flags |= TOF_SACK;
2641 			}
2642 			break;
2643 		case TCPOPT_SACK:
2644 			if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2645 				continue;
2646 			to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2647 			to->to_sacks = cp + 2;
2648 			tcpstat.tcps_sack_rcv_blocks++;
2649 			break;
2650 		default:
2651 			continue;
2652 		}
2653 	}
2654 }
2655 
2656 /*
2657  * Pull out of band byte out of a segment so
2658  * it doesn't appear in the user's data queue.
2659  * It is still reflected in the segment length for
2660  * sequencing purposes.
2661  */
2662 static void
2663 tcp_pulloutofband(so, th, m, off)
2664 	struct socket *so;
2665 	struct tcphdr *th;
2666 	register struct mbuf *m;
2667 	int off;		/* delayed to be droped hdrlen */
2668 {
2669 	int cnt = off + th->th_urp - 1;
2670 
2671 	while (cnt >= 0) {
2672 		if (m->m_len > cnt) {
2673 			char *cp = mtod(m, caddr_t) + cnt;
2674 			struct tcpcb *tp = sototcpcb(so);
2675 
2676 			tp->t_iobc = *cp;
2677 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2678 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2679 			m->m_len--;
2680 			if (m->m_flags & M_PKTHDR)
2681 				m->m_pkthdr.len--;
2682 			return;
2683 		}
2684 		cnt -= m->m_len;
2685 		m = m->m_next;
2686 		if (m == 0)
2687 			break;
2688 	}
2689 	panic("tcp_pulloutofband");
2690 }
2691 
2692 /*
2693  * Collect new round-trip time estimate
2694  * and update averages and current timeout.
2695  */
2696 static void
2697 tcp_xmit_timer(tp, rtt)
2698 	register struct tcpcb *tp;
2699 	int rtt;
2700 {
2701 	register int delta;
2702 
2703 	INP_LOCK_ASSERT(tp->t_inpcb);
2704 
2705 	tcpstat.tcps_rttupdated++;
2706 	tp->t_rttupdated++;
2707 	if (tp->t_srtt != 0) {
2708 		/*
2709 		 * srtt is stored as fixed point with 5 bits after the
2710 		 * binary point (i.e., scaled by 8).  The following magic
2711 		 * is equivalent to the smoothing algorithm in rfc793 with
2712 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2713 		 * point).  Adjust rtt to origin 0.
2714 		 */
2715 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2716 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2717 
2718 		if ((tp->t_srtt += delta) <= 0)
2719 			tp->t_srtt = 1;
2720 
2721 		/*
2722 		 * We accumulate a smoothed rtt variance (actually, a
2723 		 * smoothed mean difference), then set the retransmit
2724 		 * timer to smoothed rtt + 4 times the smoothed variance.
2725 		 * rttvar is stored as fixed point with 4 bits after the
2726 		 * binary point (scaled by 16).  The following is
2727 		 * equivalent to rfc793 smoothing with an alpha of .75
2728 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2729 		 * rfc793's wired-in beta.
2730 		 */
2731 		if (delta < 0)
2732 			delta = -delta;
2733 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2734 		if ((tp->t_rttvar += delta) <= 0)
2735 			tp->t_rttvar = 1;
2736 		if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2737 		    tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2738 	} else {
2739 		/*
2740 		 * No rtt measurement yet - use the unsmoothed rtt.
2741 		 * Set the variance to half the rtt (so our first
2742 		 * retransmit happens at 3*rtt).
2743 		 */
2744 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
2745 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2746 		tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2747 	}
2748 	tp->t_rtttime = 0;
2749 	tp->t_rxtshift = 0;
2750 
2751 	/*
2752 	 * the retransmit should happen at rtt + 4 * rttvar.
2753 	 * Because of the way we do the smoothing, srtt and rttvar
2754 	 * will each average +1/2 tick of bias.  When we compute
2755 	 * the retransmit timer, we want 1/2 tick of rounding and
2756 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2757 	 * firing of the timer.  The bias will give us exactly the
2758 	 * 1.5 tick we need.  But, because the bias is
2759 	 * statistical, we have to test that we don't drop below
2760 	 * the minimum feasible timer (which is 2 ticks).
2761 	 */
2762 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2763 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2764 
2765 	/*
2766 	 * We received an ack for a packet that wasn't retransmitted;
2767 	 * it is probably safe to discard any error indications we've
2768 	 * received recently.  This isn't quite right, but close enough
2769 	 * for now (a route might have failed after we sent a segment,
2770 	 * and the return path might not be symmetrical).
2771 	 */
2772 	tp->t_softerror = 0;
2773 }
2774 
2775 /*
2776  * Determine a reasonable value for maxseg size.
2777  * If the route is known, check route for mtu.
2778  * If none, use an mss that can be handled on the outgoing
2779  * interface without forcing IP to fragment; if bigger than
2780  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2781  * to utilize large mbufs.  If no route is found, route has no mtu,
2782  * or the destination isn't local, use a default, hopefully conservative
2783  * size (usually 512 or the default IP max size, but no more than the mtu
2784  * of the interface), as we can't discover anything about intervening
2785  * gateways or networks.  We also initialize the congestion/slow start
2786  * window to be a single segment if the destination isn't local.
2787  * While looking at the routing entry, we also initialize other path-dependent
2788  * parameters from pre-set or cached values in the routing entry.
2789  *
2790  * Also take into account the space needed for options that we
2791  * send regularly.  Make maxseg shorter by that amount to assure
2792  * that we can send maxseg amount of data even when the options
2793  * are present.  Store the upper limit of the length of options plus
2794  * data in maxopd.
2795  *
2796  *
2797  * In case of T/TCP, we call this routine during implicit connection
2798  * setup as well (offer = -1), to initialize maxseg from the cached
2799  * MSS of our peer.
2800  *
2801  * NOTE that this routine is only called when we process an incoming
2802  * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2803  */
2804 void
2805 tcp_mss(tp, offer)
2806 	struct tcpcb *tp;
2807 	int offer;
2808 {
2809 	int rtt, mss;
2810 	u_long bufsize;
2811 	u_long maxmtu;
2812 	struct inpcb *inp = tp->t_inpcb;
2813 	struct socket *so;
2814 	struct hc_metrics_lite metrics;
2815 	int origoffer = offer;
2816 #ifdef INET6
2817 	int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2818 	size_t min_protoh = isipv6 ?
2819 			    sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2820 			    sizeof (struct tcpiphdr);
2821 #else
2822 	const size_t min_protoh = sizeof(struct tcpiphdr);
2823 #endif
2824 
2825 	/* initialize */
2826 #ifdef INET6
2827 	if (isipv6) {
2828 		maxmtu = tcp_maxmtu6(&inp->inp_inc);
2829 		tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2830 	} else
2831 #endif
2832 	{
2833 		maxmtu = tcp_maxmtu(&inp->inp_inc);
2834 		tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2835 	}
2836 	so = inp->inp_socket;
2837 
2838 	/*
2839 	 * no route to sender, stay with default mss and return
2840 	 */
2841 	if (maxmtu == 0)
2842 		return;
2843 
2844 	/* what have we got? */
2845 	switch (offer) {
2846 		case 0:
2847 			/*
2848 			 * Offer == 0 means that there was no MSS on the SYN
2849 			 * segment, in this case we use tcp_mssdflt.
2850 			 */
2851 			offer =
2852 #ifdef INET6
2853 				isipv6 ? tcp_v6mssdflt :
2854 #endif
2855 				tcp_mssdflt;
2856 			break;
2857 
2858 		case -1:
2859 			/*
2860 			 * Offer == -1 means that we didn't receive SYN yet.
2861 			 */
2862 			/* FALLTHROUGH */
2863 
2864 		default:
2865 			/*
2866 			 * Prevent DoS attack with too small MSS. Round up
2867 			 * to at least minmss.
2868 			 */
2869 			offer = max(offer, tcp_minmss);
2870 			/*
2871 			 * Sanity check: make sure that maxopd will be large
2872 			 * enough to allow some data on segments even if the
2873 			 * all the option space is used (40bytes).  Otherwise
2874 			 * funny things may happen in tcp_output.
2875 			 */
2876 			offer = max(offer, 64);
2877 	}
2878 
2879 	/*
2880 	 * rmx information is now retrieved from tcp_hostcache
2881 	 */
2882 	tcp_hc_get(&inp->inp_inc, &metrics);
2883 
2884 	/*
2885 	 * if there's a discovered mtu int tcp hostcache, use it
2886 	 * else, use the link mtu.
2887 	 */
2888 	if (metrics.rmx_mtu)
2889 		mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2890 	else {
2891 #ifdef INET6
2892 		if (isipv6) {
2893 			mss = maxmtu - min_protoh;
2894 			if (!path_mtu_discovery &&
2895 			    !in6_localaddr(&inp->in6p_faddr))
2896 				mss = min(mss, tcp_v6mssdflt);
2897 		} else
2898 #endif
2899 		{
2900 			mss = maxmtu - min_protoh;
2901 			if (!path_mtu_discovery &&
2902 			    !in_localaddr(inp->inp_faddr))
2903 				mss = min(mss, tcp_mssdflt);
2904 		}
2905 	}
2906 	mss = min(mss, offer);
2907 
2908 	/*
2909 	 * maxopd stores the maximum length of data AND options
2910 	 * in a segment; maxseg is the amount of data in a normal
2911 	 * segment.  We need to store this value (maxopd) apart
2912 	 * from maxseg, because now every segment carries options
2913 	 * and thus we normally have somewhat less data in segments.
2914 	 */
2915 	tp->t_maxopd = mss;
2916 
2917 	/*
2918 	 * origoffer==-1 indicates, that no segments were received yet.
2919 	 * In this case we just guess.
2920 	 */
2921 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2922 	    (origoffer == -1 ||
2923 	     (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2924 		mss -= TCPOLEN_TSTAMP_APPA;
2925 	tp->t_maxseg = mss;
2926 
2927 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
2928 		if (mss > MCLBYTES)
2929 			mss &= ~(MCLBYTES-1);
2930 #else
2931 		if (mss > MCLBYTES)
2932 			mss = mss / MCLBYTES * MCLBYTES;
2933 #endif
2934 	tp->t_maxseg = mss;
2935 
2936 	/*
2937 	 * If there's a pipesize, change the socket buffer to that size,
2938 	 * don't change if sb_hiwat is different than default (then it
2939 	 * has been changed on purpose with setsockopt).
2940 	 * Make the socket buffers an integral number of mss units;
2941 	 * if the mss is larger than the socket buffer, decrease the mss.
2942 	 */
2943 	SOCKBUF_LOCK(&so->so_snd);
2944 	if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2945 		bufsize = metrics.rmx_sendpipe;
2946 	else
2947 		bufsize = so->so_snd.sb_hiwat;
2948 	if (bufsize < mss)
2949 		mss = bufsize;
2950 	else {
2951 		bufsize = roundup(bufsize, mss);
2952 		if (bufsize > sb_max)
2953 			bufsize = sb_max;
2954 		if (bufsize > so->so_snd.sb_hiwat)
2955 			(void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
2956 	}
2957 	SOCKBUF_UNLOCK(&so->so_snd);
2958 	tp->t_maxseg = mss;
2959 
2960 	SOCKBUF_LOCK(&so->so_rcv);
2961 	if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2962 		bufsize = metrics.rmx_recvpipe;
2963 	else
2964 		bufsize = so->so_rcv.sb_hiwat;
2965 	if (bufsize > mss) {
2966 		bufsize = roundup(bufsize, mss);
2967 		if (bufsize > sb_max)
2968 			bufsize = sb_max;
2969 		if (bufsize > so->so_rcv.sb_hiwat)
2970 			(void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
2971 	}
2972 	SOCKBUF_UNLOCK(&so->so_rcv);
2973 	/*
2974 	 * While we're here, check the others too
2975 	 */
2976 	if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2977 		tp->t_srtt = rtt;
2978 		tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2979 		tcpstat.tcps_usedrtt++;
2980 		if (metrics.rmx_rttvar) {
2981 			tp->t_rttvar = metrics.rmx_rttvar;
2982 			tcpstat.tcps_usedrttvar++;
2983 		} else {
2984 			/* default variation is +- 1 rtt */
2985 			tp->t_rttvar =
2986 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2987 		}
2988 		TCPT_RANGESET(tp->t_rxtcur,
2989 			      ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2990 			      tp->t_rttmin, TCPTV_REXMTMAX);
2991 	}
2992 	if (metrics.rmx_ssthresh) {
2993 		/*
2994 		 * There's some sort of gateway or interface
2995 		 * buffer limit on the path.  Use this to set
2996 		 * the slow start threshhold, but set the
2997 		 * threshold to no less than 2*mss.
2998 		 */
2999 		tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3000 		tcpstat.tcps_usedssthresh++;
3001 	}
3002 	if (metrics.rmx_bandwidth)
3003 		tp->snd_bandwidth = metrics.rmx_bandwidth;
3004 
3005 	/*
3006 	 * Set the slow-start flight size depending on whether this
3007 	 * is a local network or not.
3008 	 *
3009 	 * Extend this so we cache the cwnd too and retrieve it here.
3010 	 * Make cwnd even bigger than RFC3390 suggests but only if we
3011 	 * have previous experience with the remote host. Be careful
3012 	 * not make cwnd bigger than remote receive window or our own
3013 	 * send socket buffer. Maybe put some additional upper bound
3014 	 * on the retrieved cwnd. Should do incremental updates to
3015 	 * hostcache when cwnd collapses so next connection doesn't
3016 	 * overloads the path again.
3017 	 *
3018 	 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3019 	 * We currently check only in syncache_socket for that.
3020 	 */
3021 #define TCP_METRICS_CWND
3022 #ifdef TCP_METRICS_CWND
3023 	if (metrics.rmx_cwnd)
3024 		tp->snd_cwnd = max(mss,
3025 				min(metrics.rmx_cwnd / 2,
3026 				 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3027 	else
3028 #endif
3029 	if (tcp_do_rfc3390)
3030 		tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3031 #ifdef INET6
3032 	else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3033 		 (!isipv6 && in_localaddr(inp->inp_faddr)))
3034 #else
3035 	else if (in_localaddr(inp->inp_faddr))
3036 #endif
3037 		tp->snd_cwnd = mss * ss_fltsz_local;
3038 	else
3039 		tp->snd_cwnd = mss * ss_fltsz;
3040 }
3041 
3042 /*
3043  * Determine the MSS option to send on an outgoing SYN.
3044  */
3045 int
3046 tcp_mssopt(inc)
3047 	struct in_conninfo *inc;
3048 {
3049 	int mss = 0;
3050 	u_long maxmtu = 0;
3051 	u_long thcmtu = 0;
3052 	size_t min_protoh;
3053 #ifdef INET6
3054 	int isipv6 = inc->inc_isipv6 ? 1 : 0;
3055 #endif
3056 
3057 	KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3058 
3059 #ifdef INET6
3060 	if (isipv6) {
3061 		mss = tcp_v6mssdflt;
3062 		maxmtu = tcp_maxmtu6(inc);
3063 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3064 		min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3065 	} else
3066 #endif
3067 	{
3068 		mss = tcp_mssdflt;
3069 		maxmtu = tcp_maxmtu(inc);
3070 		thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3071 		min_protoh = sizeof(struct tcpiphdr);
3072 	}
3073 	if (maxmtu && thcmtu)
3074 		mss = min(maxmtu, thcmtu) - min_protoh;
3075 	else if (maxmtu || thcmtu)
3076 		mss = max(maxmtu, thcmtu) - min_protoh;
3077 
3078 	return (mss);
3079 }
3080 
3081 
3082 /*
3083  * On a partial ack arrives, force the retransmission of the
3084  * next unacknowledged segment.  Do not clear tp->t_dupacks.
3085  * By setting snd_nxt to ti_ack, this forces retransmission timer to
3086  * be started again.
3087  */
3088 static void
3089 tcp_newreno_partial_ack(tp, th)
3090 	struct tcpcb *tp;
3091 	struct tcphdr *th;
3092 {
3093 	tcp_seq onxt = tp->snd_nxt;
3094 	u_long  ocwnd = tp->snd_cwnd;
3095 
3096 	callout_stop(tp->tt_rexmt);
3097 	tp->t_rtttime = 0;
3098 	tp->snd_nxt = th->th_ack;
3099 	/*
3100 	 * Set snd_cwnd to one segment beyond acknowledged offset.
3101 	 * (tp->snd_una has not yet been updated when this function is called.)
3102 	 */
3103 	tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3104 	tp->t_flags |= TF_ACKNOW;
3105 	(void) tcp_output(tp);
3106 	tp->snd_cwnd = ocwnd;
3107 	if (SEQ_GT(onxt, tp->snd_nxt))
3108 		tp->snd_nxt = onxt;
3109 	/*
3110 	 * Partial window deflation.  Relies on fact that tp->snd_una
3111 	 * not updated yet.
3112 	 */
3113 	if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3114 		tp->snd_cwnd -= th->th_ack - tp->snd_una;
3115 	else
3116 		tp->snd_cwnd = 0;
3117 	tp->snd_cwnd += tp->t_maxseg;
3118 }
3119 
3120 /*
3121  * Returns 1 if the TIME_WAIT state was killed and we should start over,
3122  * looking for a pcb in the listen state.  Returns 0 otherwise.
3123  */
3124 static int
3125 tcp_timewait(tw, to, th, m, tlen)
3126 	struct tcptw *tw;
3127 	struct tcpopt *to;
3128 	struct tcphdr *th;
3129 	struct mbuf *m;
3130 	int tlen;
3131 {
3132 	int thflags;
3133 	tcp_seq seq;
3134 #ifdef INET6
3135 	int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
3136 #else
3137 	const int isipv6 = 0;
3138 #endif
3139 
3140 	/* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */
3141 	INP_INFO_WLOCK_ASSERT(&tcbinfo);
3142 	INP_LOCK_ASSERT(tw->tw_inpcb);
3143 
3144 	thflags = th->th_flags;
3145 
3146 	/*
3147 	 * NOTE: for FIN_WAIT_2 (to be added later),
3148 	 * must validate sequence number before accepting RST
3149 	 */
3150 
3151 	/*
3152 	 * If the segment contains RST:
3153 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
3154 	 *      RFC 1337.
3155 	 */
3156 	if (thflags & TH_RST)
3157 		goto drop;
3158 
3159 #if 0
3160 /* PAWS not needed at the moment */
3161 	/*
3162 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
3163 	 * and it's less than ts_recent, drop it.
3164 	 */
3165 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
3166 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
3167 		if ((thflags & TH_ACK) == 0)
3168 			goto drop;
3169 		goto ack;
3170 	}
3171 	/*
3172 	 * ts_recent is never updated because we never accept new segments.
3173 	 */
3174 #endif
3175 
3176 	/*
3177 	 * If a new connection request is received
3178 	 * while in TIME_WAIT, drop the old connection
3179 	 * and start over if the sequence numbers
3180 	 * are above the previous ones.
3181 	 */
3182 	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
3183 		(void) tcp_twclose(tw, 0);
3184 		return (1);
3185 	}
3186 
3187 	/*
3188 	 * Drop the the segment if it does not contain an ACK.
3189 	 */
3190 	if ((thflags & TH_ACK) == 0)
3191 		goto drop;
3192 
3193 	/*
3194 	 * Reset the 2MSL timer if this is a duplicate FIN.
3195 	 */
3196 	if (thflags & TH_FIN) {
3197 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
3198 		if (seq + 1 == tw->rcv_nxt)
3199 			tcp_timer_2msl_reset(tw, 2 * tcp_msl);
3200 	}
3201 
3202 	/*
3203 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
3204 	 */
3205 	if (thflags != TH_ACK || tlen != 0 ||
3206 	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
3207 		tcp_twrespond(tw, TH_ACK);
3208 	goto drop;
3209 
3210 	/*
3211 	 * Generate a RST, dropping incoming segment.
3212 	 * Make ACK acceptable to originator of segment.
3213 	 * Don't bother to respond if destination was broadcast/multicast.
3214 	 */
3215 	if (m->m_flags & (M_BCAST|M_MCAST))
3216 		goto drop;
3217 	if (isipv6) {
3218 		struct ip6_hdr *ip6;
3219 
3220 		/* IPv6 anycast check is done at tcp6_input() */
3221 		ip6 = mtod(m, struct ip6_hdr *);
3222 		if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3223 		    IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3224 			goto drop;
3225 	} else {
3226 		struct ip *ip;
3227 
3228 		ip = mtod(m, struct ip *);
3229 		if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3230 		    IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3231 		    ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3232 		    in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3233 			goto drop;
3234 	}
3235 	if (thflags & TH_ACK) {
3236 		tcp_respond(NULL,
3237 		    mtod(m, void *), th, m, 0, th->th_ack, TH_RST);
3238 	} else {
3239 		seq = th->th_seq + (thflags & TH_SYN ? 1 : 0);
3240 		tcp_respond(NULL,
3241 		    mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK);
3242 	}
3243 	INP_UNLOCK(tw->tw_inpcb);
3244 	return (0);
3245 
3246 drop:
3247 	INP_UNLOCK(tw->tw_inpcb);
3248 	m_freem(m);
3249 	return (0);
3250 }
3251