xref: /freebsd/sys/netinet/tcp_input.c (revision 10534a60e8a62e6e70320bfa2e7fbe5ffe5ee4ba)
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  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
34  * $FreeBSD$
35  */
36 
37 #include "opt_ipfw.h"		/* for ipfw_fwd		*/
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.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/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/syslog.h>
52 
53 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
54 
55 #include <net/if.h>
56 #include <net/route.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_icmp.h>	/* for ICMP_BANDLIM		*/
62 #include <netinet/in_pcb.h>
63 #include <netinet/ip_var.h>
64 #include <netinet/icmp_var.h>	/* for ICMP_BANDLIM		*/
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcpip.h>
71 #ifdef TCPDEBUG
72 #include <netinet/tcp_debug.h>
73 static struct	tcpiphdr tcp_saveti;
74 #endif
75 
76 static int	tcprexmtthresh = 3;
77 tcp_seq	tcp_iss;
78 tcp_cc	tcp_ccgen;
79 
80 struct	tcpstat tcpstat;
81 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD,
82     &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
83 
84 static int log_in_vain = 0;
85 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
86     &log_in_vain, 0, "Log all incoming TCP connections");
87 
88 static int blackhole = 0;
89 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
90 	&blackhole, 0, "Do not send RST when dropping refused connections");
91 
92 int tcp_delack_enabled = 1;
93 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
94     &tcp_delack_enabled, 0,
95     "Delay ACK to try and piggyback it onto a data packet");
96 
97 #ifdef TCP_DROP_SYNFIN
98 static int drop_synfin = 0;
99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
100     &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
101 #endif
102 
103 #ifdef TCP_RESTRICT_RST
104 static int restrict_rst = 0;
105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, restrict_rst, CTLFLAG_RW,
106     &restrict_rst, 0, "Restrict RST emission");
107 #endif
108 
109 struct inpcbhead tcb;
110 struct inpcbinfo tcbinfo;
111 
112 static void	 tcp_dooptions __P((struct tcpcb *,
113 	    u_char *, int, struct tcpiphdr *, struct tcpopt *));
114 static void	 tcp_pulloutofband __P((struct socket *,
115 	    struct tcpiphdr *, struct mbuf *));
116 static int	 tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *));
117 static void	 tcp_xmit_timer __P((struct tcpcb *, int));
118 
119 
120 /*
121  * Insert segment ti into reassembly queue of tcp with
122  * control block tp.  Return TH_FIN if reassembly now includes
123  * a segment with FIN.  The macro form does the common case inline
124  * (segment is the next to be received on an established connection,
125  * and the queue is empty), avoiding linkage into and removal
126  * from the queue and repetition of various conversions.
127  * Set DELACK for segments received in order, but ack immediately
128  * when segments are out of order (so fast retransmit can work).
129  */
130 #define	TCP_REASS(tp, ti, m, so, flags) { \
131 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
132 	    (tp)->t_segq == NULL && \
133 	    (tp)->t_state == TCPS_ESTABLISHED) { \
134 		if (tcp_delack_enabled) \
135 			callout_reset(tp->tt_delack, tcp_delacktime, \
136 			    tcp_timer_delack, tp); \
137 		else \
138 			tp->t_flags |= TF_ACKNOW; \
139 		(tp)->rcv_nxt += (ti)->ti_len; \
140 		flags = (ti)->ti_flags & TH_FIN; \
141 		tcpstat.tcps_rcvpack++;\
142 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
143 		sbappend(&(so)->so_rcv, (m)); \
144 		sorwakeup(so); \
145 	} else { \
146 		(flags) = tcp_reass((tp), (ti), (m)); \
147 		tp->t_flags |= TF_ACKNOW; \
148 	} \
149 }
150 
151 static int
152 tcp_reass(tp, ti, m)
153 	register struct tcpcb *tp;
154 	register struct tcpiphdr *ti;
155 	struct mbuf *m;
156 {
157 	struct mbuf *q;
158 	struct mbuf *p;
159 	struct mbuf *nq;
160 	struct socket *so = tp->t_inpcb->inp_socket;
161 	int flags;
162 
163 #define GETTCP(m)	((struct tcpiphdr *)m->m_pkthdr.header)
164 
165 	/*
166 	 * Call with ti==0 after become established to
167 	 * force pre-ESTABLISHED data up to user socket.
168 	 */
169 	if (ti == 0)
170 		goto present;
171 
172 	m->m_pkthdr.header = ti;
173 
174 	/*
175 	 * Find a segment which begins after this one does.
176 	 */
177 	for (q = tp->t_segq, p = NULL; q; p = q, q = q->m_nextpkt)
178 		if (SEQ_GT(GETTCP(q)->ti_seq, ti->ti_seq))
179 			break;
180 
181 	/*
182 	 * If there is a preceding segment, it may provide some of
183 	 * our data already.  If so, drop the data from the incoming
184 	 * segment.  If it provides all of our data, drop us.
185 	 */
186 	if (p != NULL) {
187 		register int i;
188 		/* conversion to int (in i) handles seq wraparound */
189 		i = GETTCP(p)->ti_seq + GETTCP(p)->ti_len - ti->ti_seq;
190 		if (i > 0) {
191 			if (i >= ti->ti_len) {
192 				tcpstat.tcps_rcvduppack++;
193 				tcpstat.tcps_rcvdupbyte += ti->ti_len;
194 				m_freem(m);
195 				/*
196 				 * Try to present any queued data
197 				 * at the left window edge to the user.
198 				 * This is needed after the 3-WHS
199 				 * completes.
200 				 */
201 				goto present;	/* ??? */
202 			}
203 			m_adj(m, i);
204 			ti->ti_len -= i;
205 			ti->ti_seq += i;
206 		}
207 	}
208 	tcpstat.tcps_rcvoopack++;
209 	tcpstat.tcps_rcvoobyte += ti->ti_len;
210 
211 	/*
212 	 * While we overlap succeeding segments trim them or,
213 	 * if they are completely covered, dequeue them.
214 	 */
215 	while (q) {
216 		register int i = (ti->ti_seq + ti->ti_len) - GETTCP(q)->ti_seq;
217 		if (i <= 0)
218 			break;
219 		if (i < GETTCP(q)->ti_len) {
220 			GETTCP(q)->ti_seq += i;
221 			GETTCP(q)->ti_len -= i;
222 			m_adj(q, i);
223 			break;
224 		}
225 
226 		nq = q->m_nextpkt;
227 		if (p)
228 			p->m_nextpkt = nq;
229 		else
230 			tp->t_segq = nq;
231 		m_freem(q);
232 		q = nq;
233 	}
234 
235 	if (p == NULL) {
236 		m->m_nextpkt = tp->t_segq;
237 		tp->t_segq = m;
238 	} else {
239 		m->m_nextpkt = p->m_nextpkt;
240 		p->m_nextpkt = m;
241 	}
242 
243 present:
244 	/*
245 	 * Present data to user, advancing rcv_nxt through
246 	 * completed sequence space.
247 	 */
248 	if (!TCPS_HAVEESTABLISHED(tp->t_state))
249 		return (0);
250 	q = tp->t_segq;
251 	if (!q || GETTCP(q)->ti_seq != tp->rcv_nxt)
252 		return (0);
253 	do {
254 		tp->rcv_nxt += GETTCP(q)->ti_len;
255 		flags = GETTCP(q)->ti_flags & TH_FIN;
256 		nq = q->m_nextpkt;
257 		tp->t_segq = nq;
258 		q->m_nextpkt = NULL;
259 		if (so->so_state & SS_CANTRCVMORE)
260 			m_freem(q);
261 		else
262 			sbappend(&so->so_rcv, q);
263 		q = nq;
264 	} while (q && GETTCP(q)->ti_seq == tp->rcv_nxt);
265 	sorwakeup(so);
266 	return (flags);
267 
268 #undef GETTCP
269 }
270 
271 /*
272  * TCP input routine, follows pages 65-76 of the
273  * protocol specification dated September, 1981 very closely.
274  */
275 void
276 tcp_input(m, iphlen)
277 	register struct mbuf *m;
278 	int iphlen;
279 {
280 	register struct tcpiphdr *ti;
281 	register struct inpcb *inp;
282 	u_char *optp = NULL;
283 	int optlen = 0;
284 	int len, tlen, off;
285 	register struct tcpcb *tp = 0;
286 	register int tiflags;
287 	struct socket *so = 0;
288 	int todrop, acked, ourfinisacked, needoutput = 0;
289 	struct in_addr laddr;
290 	int dropsocket = 0;
291 	int iss = 0;
292 	u_long tiwin;
293 	struct tcpopt to;		/* options in this segment */
294 	struct rmxp_tao *taop;		/* pointer to our TAO cache entry */
295 	struct rmxp_tao	tao_noncached;	/* in case there's no cached entry */
296 #ifdef TCPDEBUG
297 	short ostate = 0;
298 #endif
299 
300 	bzero((char *)&to, sizeof(to));
301 
302 	tcpstat.tcps_rcvtotal++;
303 	/*
304 	 * Get IP and TCP header together in first mbuf.
305 	 * Note: IP leaves IP header in first mbuf.
306 	 */
307 	ti = mtod(m, struct tcpiphdr *);
308 	if (iphlen > sizeof (struct ip))
309 		ip_stripoptions(m, (struct mbuf *)0);
310 	if (m->m_len < sizeof (struct tcpiphdr)) {
311 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
312 			tcpstat.tcps_rcvshort++;
313 			return;
314 		}
315 		ti = mtod(m, struct tcpiphdr *);
316 	}
317 
318 	/*
319 	 * Checksum extended TCP header and data.
320 	 */
321 	tlen = ((struct ip *)ti)->ip_len;
322 	len = sizeof (struct ip) + tlen;
323 	bzero(ti->ti_x1, sizeof(ti->ti_x1));
324 	ti->ti_len = (u_short)tlen;
325 	HTONS(ti->ti_len);
326 	ti->ti_sum = in_cksum(m, len);
327 	if (ti->ti_sum) {
328 		tcpstat.tcps_rcvbadsum++;
329 		goto drop;
330 	}
331 
332 	/*
333 	 * Check that TCP offset makes sense,
334 	 * pull out TCP options and adjust length.		XXX
335 	 */
336 	off = ti->ti_off << 2;
337 	if (off < sizeof (struct tcphdr) || off > tlen) {
338 		tcpstat.tcps_rcvbadoff++;
339 		goto drop;
340 	}
341 	tlen -= off;
342 	ti->ti_len = tlen;
343 	if (off > sizeof (struct tcphdr)) {
344 		if (m->m_len < sizeof(struct ip) + off) {
345 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
346 				tcpstat.tcps_rcvshort++;
347 				return;
348 			}
349 			ti = mtod(m, struct tcpiphdr *);
350 		}
351 		optlen = off - sizeof (struct tcphdr);
352 		optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
353 	}
354 	tiflags = ti->ti_flags;
355 
356 #ifdef TCP_DROP_SYNFIN
357 	/*
358 	 * If the drop_synfin option is enabled, drop all packets with
359 	 * both the SYN and FIN bits set. This prevents e.g. nmap from
360 	 * identifying the TCP/IP stack.
361 	 *
362 	 * This is incompatible with RFC1644 extensions (T/TCP).
363 	 */
364 	if (drop_synfin && (tiflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
365 		goto drop;
366 #endif
367 
368 	/*
369 	 * Convert TCP protocol specific fields to host format.
370 	 */
371 	NTOHL(ti->ti_seq);
372 	NTOHL(ti->ti_ack);
373 	NTOHS(ti->ti_win);
374 	NTOHS(ti->ti_urp);
375 
376 	/*
377 	 * Drop TCP, IP headers and TCP options.
378 	 */
379 	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
380 	m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
381 
382 	/*
383 	 * Locate pcb for segment.
384 	 */
385 findpcb:
386 #ifdef IPFIREWALL_FORWARD
387 	if (ip_fw_fwd_addr != NULL) {
388 		/*
389 		 * Diverted. Pretend to be the destination.
390 		 * already got one like this?
391 		 */
392 		inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport,
393 			ti->ti_dst, ti->ti_dport, 0);
394 		if (!inp) {
395 			/*
396 			 * No, then it's new. Try find the ambushing socket
397 			 */
398 			if (!ip_fw_fwd_addr->sin_port) {
399 				inp = in_pcblookup_hash(&tcbinfo, ti->ti_src,
400 				    ti->ti_sport, ip_fw_fwd_addr->sin_addr,
401 				    ti->ti_dport, 1);
402 			} else {
403 				inp = in_pcblookup_hash(&tcbinfo,
404 				    ti->ti_src, ti->ti_sport,
405 	    			    ip_fw_fwd_addr->sin_addr,
406 				    ntohs(ip_fw_fwd_addr->sin_port), 1);
407 			}
408 		}
409 		ip_fw_fwd_addr = NULL;
410 	} else
411 #endif	/* IPFIREWALL_FORWARD */
412 
413 	inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport,
414 	    ti->ti_dst, ti->ti_dport, 1);
415 
416 	/*
417 	 * If the state is CLOSED (i.e., TCB does not exist) then
418 	 * all data in the incoming segment is discarded.
419 	 * If the TCB exists but is in CLOSED state, it is embryonic,
420 	 * but should either do a listen or a connect soon.
421 	 */
422 	if (inp == NULL) {
423 		if (log_in_vain) {
424 			char buf[4*sizeof "123"];
425 
426 			strcpy(buf, inet_ntoa(ti->ti_dst));
427 			switch (log_in_vain) {
428 			case 1:
429 				if(tiflags & TH_SYN)
430 					log(LOG_INFO,
431 			    		"Connection attempt to TCP %s:%d from %s:%d\n",
432 			    		buf, ntohs(ti->ti_dport),
433 					inet_ntoa(ti->ti_src),
434 			    		ntohs(ti->ti_sport));
435 				break;
436 			case 2:
437 				log(LOG_INFO,
438 			    	"Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n",
439 			    	buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src),
440 			    	ntohs(ti->ti_sport), tiflags);
441 				break;
442 			default:
443 				break;
444 			}
445 		}
446 #ifdef ICMP_BANDLIM
447 		if (badport_bandlim(1) < 0)
448 			goto drop;
449 #endif
450 		if (blackhole) {
451 			switch (blackhole) {
452 			case 1:
453 				if (tiflags & TH_SYN)
454 					goto drop;
455 				break;
456 			case 2:
457 				goto drop;
458 			default:
459 				goto drop;
460 			}
461 		}
462 		goto dropwithreset;
463 	}
464 	tp = intotcpcb(inp);
465 	if (tp == 0)
466 		goto dropwithreset;
467 	if (tp->t_state == TCPS_CLOSED)
468 		goto drop;
469 
470 	/* Unscale the window into a 32-bit value. */
471 	if ((tiflags & TH_SYN) == 0)
472 		tiwin = ti->ti_win << tp->snd_scale;
473 	else
474 		tiwin = ti->ti_win;
475 
476 	so = inp->inp_socket;
477 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
478 #ifdef TCPDEBUG
479 		if (so->so_options & SO_DEBUG) {
480 			ostate = tp->t_state;
481 			tcp_saveti = *ti;
482 		}
483 #endif
484 		if (so->so_options & SO_ACCEPTCONN) {
485 			register struct tcpcb *tp0 = tp;
486 			struct socket *so2;
487 			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
488 				/*
489 				 * Note: dropwithreset makes sure we don't
490 				 * send a RST in response to a RST.
491 				 */
492 				if (tiflags & TH_ACK) {
493 					tcpstat.tcps_badsyn++;
494 					goto dropwithreset;
495 				}
496 				goto drop;
497 			}
498 			so2 = sonewconn(so, 0);
499 			if (so2 == 0) {
500 				tcpstat.tcps_listendrop++;
501 				so2 = sodropablereq(so);
502 				if (so2) {
503 					tcp_drop(sototcpcb(so2), ETIMEDOUT);
504 					so2 = sonewconn(so, 0);
505 				}
506 				if (!so2)
507 					goto drop;
508 			}
509 			so = so2;
510 			/*
511 			 * This is ugly, but ....
512 			 *
513 			 * Mark socket as temporary until we're
514 			 * committed to keeping it.  The code at
515 			 * ``drop'' and ``dropwithreset'' check the
516 			 * flag dropsocket to see if the temporary
517 			 * socket created here should be discarded.
518 			 * We mark the socket as discardable until
519 			 * we're committed to it below in TCPS_LISTEN.
520 			 */
521 			dropsocket++;
522 			inp = (struct inpcb *)so->so_pcb;
523 			inp->inp_laddr = ti->ti_dst;
524 			inp->inp_lport = ti->ti_dport;
525 			if (in_pcbinshash(inp) != 0) {
526 				/*
527 				 * Undo the assignments above if we failed to
528 				 * put the PCB on the hash lists.
529 				 */
530 				inp->inp_laddr.s_addr = INADDR_ANY;
531 				inp->inp_lport = 0;
532 				goto drop;
533 			}
534 			inp->inp_options = ip_srcroute();
535 			tp = intotcpcb(inp);
536 			tp->t_state = TCPS_LISTEN;
537 			tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT);
538 
539 			/* Compute proper scaling value from buffer space */
540 			while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
541 			   TCP_MAXWIN << tp->request_r_scale <
542 			   so->so_rcv.sb_hiwat)
543 				tp->request_r_scale++;
544 		}
545 	}
546 
547 	/*
548 	 * Segment received on connection.
549 	 * Reset idle time and keep-alive timer.
550 	 */
551 	tp->t_rcvtime = ticks;
552 	if (TCPS_HAVEESTABLISHED(tp->t_state))
553 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
554 
555 	/*
556 	 * Process options if not in LISTEN state,
557 	 * else do it below (after getting remote address).
558 	 */
559 	if (tp->t_state != TCPS_LISTEN)
560 		tcp_dooptions(tp, optp, optlen, ti, &to);
561 
562 	/*
563 	 * Header prediction: check for the two common cases
564 	 * of a uni-directional data xfer.  If the packet has
565 	 * no control flags, is in-sequence, the window didn't
566 	 * change and we're not retransmitting, it's a
567 	 * candidate.  If the length is zero and the ack moved
568 	 * forward, we're the sender side of the xfer.  Just
569 	 * free the data acked & wake any higher level process
570 	 * that was blocked waiting for space.  If the length
571 	 * is non-zero and the ack didn't move, we're the
572 	 * receiver side.  If we're getting packets in-order
573 	 * (the reassembly queue is empty), add the data to
574 	 * the socket buffer and note that we need a delayed ack.
575 	 * Make sure that the hidden state-flags are also off.
576 	 * Since we check for TCPS_ESTABLISHED above, it can only
577 	 * be TH_NEEDSYN.
578 	 */
579 	if (tp->t_state == TCPS_ESTABLISHED &&
580 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
581 	    ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
582 	    ((to.to_flag & TOF_TS) == 0 ||
583 	     TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
584 	    /*
585 	     * Using the CC option is compulsory if once started:
586 	     *   the segment is OK if no T/TCP was negotiated or
587 	     *   if the segment has a CC option equal to CCrecv
588 	     */
589 	    ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
590 	     ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
591 	    ti->ti_seq == tp->rcv_nxt &&
592 	    tiwin && tiwin == tp->snd_wnd &&
593 	    tp->snd_nxt == tp->snd_max) {
594 
595 		/*
596 		 * If last ACK falls within this segment's sequence numbers,
597 		 * record the timestamp.
598 		 * NOTE that the test is modified according to the latest
599 		 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
600 		 */
601 		if ((to.to_flag & TOF_TS) != 0 &&
602 		   SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
603 			tp->ts_recent_age = ticks;
604 			tp->ts_recent = to.to_tsval;
605 		}
606 
607 		if (ti->ti_len == 0) {
608 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
609 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
610 			    tp->snd_cwnd >= tp->snd_wnd &&
611 			    tp->t_dupacks < tcprexmtthresh) {
612 				/*
613 				 * this is a pure ack for outstanding data.
614 				 */
615 				++tcpstat.tcps_predack;
616 				/*
617 				 * "bad retransmit" recovery
618 				 */
619 				if (tp->t_rxtshift == 1 &&
620 				    ticks < tp->t_badrxtwin) {
621 					tp->snd_cwnd = tp->snd_cwnd_prev;
622 					tp->snd_ssthresh =
623 					    tp->snd_ssthresh_prev;
624 					tp->snd_nxt = tp->snd_max;
625 					tp->t_badrxtwin = 0;
626 				}
627 				if ((to.to_flag & TOF_TS) != 0)
628 					tcp_xmit_timer(tp,
629 					    ticks - to.to_tsecr + 1);
630 				else if (tp->t_rtttime &&
631 					    SEQ_GT(ti->ti_ack, tp->t_rtseq))
632 					tcp_xmit_timer(tp, ticks - tp->t_rtttime);
633 				acked = ti->ti_ack - tp->snd_una;
634 				tcpstat.tcps_rcvackpack++;
635 				tcpstat.tcps_rcvackbyte += acked;
636 				sbdrop(&so->so_snd, acked);
637 				tp->snd_una = ti->ti_ack;
638 				m_freem(m);
639 
640 				/*
641 				 * If all outstanding data are acked, stop
642 				 * retransmit timer, otherwise restart timer
643 				 * using current (possibly backed-off) value.
644 				 * If process is waiting for space,
645 				 * wakeup/selwakeup/signal.  If data
646 				 * are ready to send, let tcp_output
647 				 * decide between more output or persist.
648 				 */
649 				if (tp->snd_una == tp->snd_max)
650 					callout_stop(tp->tt_rexmt);
651 				else if (!callout_active(tp->tt_persist))
652 					callout_reset(tp->tt_rexmt,
653 						      tp->t_rxtcur,
654 						      tcp_timer_rexmt, tp);
655 
656 				sowwakeup(so);
657 				if (so->so_snd.sb_cc)
658 					(void) tcp_output(tp);
659 				return;
660 			}
661 		} else if (ti->ti_ack == tp->snd_una &&
662 		    tp->t_segq == NULL &&
663 		    ti->ti_len <= sbspace(&so->so_rcv)) {
664 			/*
665 			 * this is a pure, in-sequence data packet
666 			 * with nothing on the reassembly queue and
667 			 * we have enough buffer space to take it.
668 			 */
669 			++tcpstat.tcps_preddat;
670 			tp->rcv_nxt += ti->ti_len;
671 			tcpstat.tcps_rcvpack++;
672 			tcpstat.tcps_rcvbyte += ti->ti_len;
673 			/*
674 			 * Add data to socket buffer.
675 			 */
676 			sbappend(&so->so_rcv, m);
677 			sorwakeup(so);
678 			if (tcp_delack_enabled) {
679 	                        callout_reset(tp->tt_delack, tcp_delacktime,
680 	                            tcp_timer_delack, tp);
681 			} else {
682 				tp->t_flags |= TF_ACKNOW;
683 				tcp_output(tp);
684 			}
685 			return;
686 		}
687 	}
688 
689 	/*
690 	 * Calculate amount of space in receive window,
691 	 * and then do TCP input processing.
692 	 * Receive window is amount of space in rcv queue,
693 	 * but not less than advertised window.
694 	 */
695 	{ int win;
696 
697 	win = sbspace(&so->so_rcv);
698 	if (win < 0)
699 		win = 0;
700 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
701 	}
702 
703 	switch (tp->t_state) {
704 
705 	/*
706 	 * If the state is LISTEN then ignore segment if it contains an RST.
707 	 * If the segment contains an ACK then it is bad and send a RST.
708 	 * If it does not contain a SYN then it is not interesting; drop it.
709 	 * If it is from this socket, drop it, it must be forged.
710 	 * Don't bother responding if the destination was a broadcast.
711 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
712 	 * tp->iss, and send a segment:
713 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
714 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
715 	 * Fill in remote peer address fields if not previously specified.
716 	 * Enter SYN_RECEIVED state, and process any other fields of this
717 	 * segment in this state.
718 	 */
719 	case TCPS_LISTEN: {
720 		register struct sockaddr_in *sin;
721 
722 		if (tiflags & TH_RST)
723 			goto drop;
724 		if (tiflags & TH_ACK)
725 			goto dropwithreset;
726 		if ((tiflags & TH_SYN) == 0)
727 			goto drop;
728 		if ((ti->ti_dport == ti->ti_sport) &&
729 		    (ti->ti_dst.s_addr == ti->ti_src.s_addr))
730 			goto drop;
731 		/*
732 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
733 		 * in_broadcast() should never return true on a received
734 		 * packet with M_BCAST not set.
735 		 */
736 		if (m->m_flags & (M_BCAST|M_MCAST) ||
737 		    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
738 			goto drop;
739 		MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
740 		       M_NOWAIT);
741 		if (sin == NULL)
742 			goto drop;
743 		sin->sin_family = AF_INET;
744 		sin->sin_len = sizeof(*sin);
745 		sin->sin_addr = ti->ti_src;
746 		sin->sin_port = ti->ti_sport;
747 		bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
748 		laddr = inp->inp_laddr;
749 		if (inp->inp_laddr.s_addr == INADDR_ANY)
750 			inp->inp_laddr = ti->ti_dst;
751 		if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) {
752 			inp->inp_laddr = laddr;
753 			FREE(sin, M_SONAME);
754 			goto drop;
755 		}
756 		FREE(sin, M_SONAME);
757 		tp->t_template = tcp_template(tp);
758 		if (tp->t_template == 0) {
759 			tp = tcp_drop(tp, ENOBUFS);
760 			dropsocket = 0;		/* socket is already gone */
761 			goto drop;
762 		}
763 		if ((taop = tcp_gettaocache(inp)) == NULL) {
764 			taop = &tao_noncached;
765 			bzero(taop, sizeof(*taop));
766 		}
767 		tcp_dooptions(tp, optp, optlen, ti, &to);
768 		if (iss)
769 			tp->iss = iss;
770 		else
771 			tp->iss = tcp_iss;
772 		tcp_iss += TCP_ISSINCR/4;
773 		tp->irs = ti->ti_seq;
774 		tcp_sendseqinit(tp);
775 		tcp_rcvseqinit(tp);
776 		/*
777 		 * Initialization of the tcpcb for transaction;
778 		 *   set SND.WND = SEG.WND,
779 		 *   initialize CCsend and CCrecv.
780 		 */
781 		tp->snd_wnd = tiwin;	/* initial send-window */
782 		tp->cc_send = CC_INC(tcp_ccgen);
783 		tp->cc_recv = to.to_cc;
784 		/*
785 		 * Perform TAO test on incoming CC (SEG.CC) option, if any.
786 		 * - compare SEG.CC against cached CC from the same host,
787 		 *	if any.
788 		 * - if SEG.CC > chached value, SYN must be new and is accepted
789 		 *	immediately: save new CC in the cache, mark the socket
790 		 *	connected, enter ESTABLISHED state, turn on flag to
791 		 *	send a SYN in the next segment.
792 		 *	A virtual advertised window is set in rcv_adv to
793 		 *	initialize SWS prevention.  Then enter normal segment
794 		 *	processing: drop SYN, process data and FIN.
795 		 * - otherwise do a normal 3-way handshake.
796 		 */
797 		if ((to.to_flag & TOF_CC) != 0) {
798 		    if (((tp->t_flags & TF_NOPUSH) != 0) &&
799 			taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
800 
801 			taop->tao_cc = to.to_cc;
802 			tp->t_starttime = ticks;
803 			tp->t_state = TCPS_ESTABLISHED;
804 
805 			/*
806 			 * If there is a FIN, or if there is data and the
807 			 * connection is local, then delay SYN,ACK(SYN) in
808 			 * the hope of piggy-backing it on a response
809 			 * segment.  Otherwise must send ACK now in case
810 			 * the other side is slow starting.
811 			 */
812 			if (tcp_delack_enabled && ((tiflags & TH_FIN) ||
813 			    (ti->ti_len != 0 &&
814 			    in_localaddr(inp->inp_faddr)))) {
815                                 callout_reset(tp->tt_delack, tcp_delacktime,
816                                     tcp_timer_delack, tp);
817 				tp->t_flags |= TF_NEEDSYN;
818 			} else
819 				tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
820 
821 			/*
822 			 * Limit the `virtual advertised window' to TCP_MAXWIN
823 			 * here.  Even if we requested window scaling, it will
824 			 * become effective only later when our SYN is acked.
825 			 */
826 			tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
827 			tcpstat.tcps_connects++;
828 			soisconnected(so);
829 			callout_reset(tp->tt_keep, tcp_keepinit,
830 				      tcp_timer_keep, tp);
831 			dropsocket = 0;		/* committed to socket */
832 			tcpstat.tcps_accepts++;
833 			goto trimthenstep6;
834 		    }
835 		/* else do standard 3-way handshake */
836 		} else {
837 		    /*
838 		     * No CC option, but maybe CC.NEW:
839 		     *   invalidate cached value.
840 		     */
841 		     taop->tao_cc = 0;
842 		}
843 		/*
844 		 * TAO test failed or there was no CC option,
845 		 *    do a standard 3-way handshake.
846 		 */
847 		tp->t_flags |= TF_ACKNOW;
848 		tp->t_state = TCPS_SYN_RECEIVED;
849 		callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
850 		dropsocket = 0;		/* committed to socket */
851 		tcpstat.tcps_accepts++;
852 		goto trimthenstep6;
853 		}
854 
855 	/*
856 	 * If the state is SYN_RECEIVED:
857 	 *	if seg contains an ACK, but not for our SYN/ACK, send a RST.
858 	 */
859 	case TCPS_SYN_RECEIVED:
860 		if ((tiflags & TH_ACK) &&
861 		    (SEQ_LEQ(ti->ti_ack, tp->snd_una) ||
862 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
863 				goto dropwithreset;
864 		break;
865 
866 	/*
867 	 * If the state is SYN_SENT:
868 	 *	if seg contains an ACK, but not for our SYN, drop the input.
869 	 *	if seg contains a RST, then drop the connection.
870 	 *	if seg does not contain SYN, then drop it.
871 	 * Otherwise this is an acceptable SYN segment
872 	 *	initialize tp->rcv_nxt and tp->irs
873 	 *	if seg contains ack then advance tp->snd_una
874 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
875 	 *	arrange for segment to be acked (eventually)
876 	 *	continue processing rest of data/controls, beginning with URG
877 	 */
878 	case TCPS_SYN_SENT:
879 		if ((taop = tcp_gettaocache(inp)) == NULL) {
880 			taop = &tao_noncached;
881 			bzero(taop, sizeof(*taop));
882 		}
883 
884 		if ((tiflags & TH_ACK) &&
885 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
886 		     SEQ_GT(ti->ti_ack, tp->snd_max))) {
887 			/*
888 			 * If we have a cached CCsent for the remote host,
889 			 * hence we haven't just crashed and restarted,
890 			 * do not send a RST.  This may be a retransmission
891 			 * from the other side after our earlier ACK was lost.
892 			 * Our new SYN, when it arrives, will serve as the
893 			 * needed ACK.
894 			 */
895 			if (taop->tao_ccsent != 0)
896 				goto drop;
897 			else
898 				goto dropwithreset;
899 		}
900 		if (tiflags & TH_RST) {
901 			if (tiflags & TH_ACK)
902 				tp = tcp_drop(tp, ECONNREFUSED);
903 			goto drop;
904 		}
905 		if ((tiflags & TH_SYN) == 0)
906 			goto drop;
907 		tp->snd_wnd = ti->ti_win;	/* initial send window */
908 		tp->cc_recv = to.to_cc;		/* foreign CC */
909 
910 		tp->irs = ti->ti_seq;
911 		tcp_rcvseqinit(tp);
912 		if (tiflags & TH_ACK) {
913 			/*
914 			 * Our SYN was acked.  If segment contains CC.ECHO
915 			 * option, check it to make sure this segment really
916 			 * matches our SYN.  If not, just drop it as old
917 			 * duplicate, but send an RST if we're still playing
918 			 * by the old rules.  If no CC.ECHO option, make sure
919 			 * we don't get fooled into using T/TCP.
920 			 */
921 			if (to.to_flag & TOF_CCECHO) {
922 				if (tp->cc_send != to.to_ccecho) {
923 					if (taop->tao_ccsent != 0)
924 						goto drop;
925 					else
926 						goto dropwithreset;
927 				}
928 			} else
929 				tp->t_flags &= ~TF_RCVD_CC;
930 			tcpstat.tcps_connects++;
931 			soisconnected(so);
932 			/* Do window scaling on this connection? */
933 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
934 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
935 				tp->snd_scale = tp->requested_s_scale;
936 				tp->rcv_scale = tp->request_r_scale;
937 			}
938 			/* Segment is acceptable, update cache if undefined. */
939 			if (taop->tao_ccsent == 0)
940 				taop->tao_ccsent = to.to_ccecho;
941 
942 			tp->rcv_adv += tp->rcv_wnd;
943 			tp->snd_una++;		/* SYN is acked */
944 			/*
945 			 * If there's data, delay ACK; if there's also a FIN
946 			 * ACKNOW will be turned on later.
947 			 */
948 			if (tcp_delack_enabled && ti->ti_len != 0)
949                                 callout_reset(tp->tt_delack, tcp_delacktime,
950                                     tcp_timer_delack, tp);
951 			else
952 				tp->t_flags |= TF_ACKNOW;
953 			/*
954 			 * Received <SYN,ACK> in SYN_SENT[*] state.
955 			 * Transitions:
956 			 *	SYN_SENT  --> ESTABLISHED
957 			 *	SYN_SENT* --> FIN_WAIT_1
958 			 */
959 			tp->t_starttime = ticks;
960 			if (tp->t_flags & TF_NEEDFIN) {
961 				tp->t_state = TCPS_FIN_WAIT_1;
962 				tp->t_flags &= ~TF_NEEDFIN;
963 				tiflags &= ~TH_SYN;
964 			} else {
965 				tp->t_state = TCPS_ESTABLISHED;
966 				callout_reset(tp->tt_keep, tcp_keepidle,
967 					      tcp_timer_keep, tp);
968 			}
969 		} else {
970 		/*
971 		 *  Received initial SYN in SYN-SENT[*] state => simul-
972 		 *  taneous open.  If segment contains CC option and there is
973 		 *  a cached CC, apply TAO test; if it succeeds, connection is
974 		 *  half-synchronized.  Otherwise, do 3-way handshake:
975 		 *        SYN-SENT -> SYN-RECEIVED
976 		 *        SYN-SENT* -> SYN-RECEIVED*
977 		 *  If there was no CC option, clear cached CC value.
978 		 */
979 			tp->t_flags |= TF_ACKNOW;
980 			callout_stop(tp->tt_rexmt);
981 			if (to.to_flag & TOF_CC) {
982 				if (taop->tao_cc != 0 &&
983 				    CC_GT(to.to_cc, taop->tao_cc)) {
984 					/*
985 					 * update cache and make transition:
986 					 *        SYN-SENT -> ESTABLISHED*
987 					 *        SYN-SENT* -> FIN-WAIT-1*
988 					 */
989 					taop->tao_cc = to.to_cc;
990 					tp->t_starttime = ticks;
991 					if (tp->t_flags & TF_NEEDFIN) {
992 						tp->t_state = TCPS_FIN_WAIT_1;
993 						tp->t_flags &= ~TF_NEEDFIN;
994 					} else {
995 						tp->t_state = TCPS_ESTABLISHED;
996 						callout_reset(tp->tt_keep,
997 							      tcp_keepidle,
998 							      tcp_timer_keep,
999 							      tp);
1000 					}
1001 					tp->t_flags |= TF_NEEDSYN;
1002 				} else
1003 					tp->t_state = TCPS_SYN_RECEIVED;
1004 			} else {
1005 				/* CC.NEW or no option => invalidate cache */
1006 				taop->tao_cc = 0;
1007 				tp->t_state = TCPS_SYN_RECEIVED;
1008 			}
1009 		}
1010 
1011 trimthenstep6:
1012 		/*
1013 		 * Advance ti->ti_seq to correspond to first data byte.
1014 		 * If data, trim to stay within window,
1015 		 * dropping FIN if necessary.
1016 		 */
1017 		ti->ti_seq++;
1018 		if (ti->ti_len > tp->rcv_wnd) {
1019 			todrop = ti->ti_len - tp->rcv_wnd;
1020 			m_adj(m, -todrop);
1021 			ti->ti_len = tp->rcv_wnd;
1022 			tiflags &= ~TH_FIN;
1023 			tcpstat.tcps_rcvpackafterwin++;
1024 			tcpstat.tcps_rcvbyteafterwin += todrop;
1025 		}
1026 		tp->snd_wl1 = ti->ti_seq - 1;
1027 		tp->rcv_up = ti->ti_seq;
1028 		/*
1029 		 *  Client side of transaction: already sent SYN and data.
1030 		 *  If the remote host used T/TCP to validate the SYN,
1031 		 *  our data will be ACK'd; if so, enter normal data segment
1032 		 *  processing in the middle of step 5, ack processing.
1033 		 *  Otherwise, goto step 6.
1034 		 */
1035  		if (tiflags & TH_ACK)
1036 			goto process_ACK;
1037 		goto step6;
1038 	/*
1039 	 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1040 	 *	if segment contains a SYN and CC [not CC.NEW] option:
1041 	 *              if state == TIME_WAIT and connection duration > MSL,
1042 	 *                  drop packet and send RST;
1043 	 *
1044 	 *		if SEG.CC > CCrecv then is new SYN, and can implicitly
1045 	 *		    ack the FIN (and data) in retransmission queue.
1046 	 *                  Complete close and delete TCPCB.  Then reprocess
1047 	 *                  segment, hoping to find new TCPCB in LISTEN state;
1048 	 *
1049 	 *		else must be old SYN; drop it.
1050 	 *      else do normal processing.
1051 	 */
1052 	case TCPS_LAST_ACK:
1053 	case TCPS_CLOSING:
1054 	case TCPS_TIME_WAIT:
1055 		if ((tiflags & TH_SYN) &&
1056 		    (to.to_flag & TOF_CC) && tp->cc_recv != 0) {
1057 			if (tp->t_state == TCPS_TIME_WAIT &&
1058 					(ticks - tp->t_starttime) > tcp_msl)
1059 				goto dropwithreset;
1060 			if (CC_GT(to.to_cc, tp->cc_recv)) {
1061 				tp = tcp_close(tp);
1062 				goto findpcb;
1063 			}
1064 			else
1065 				goto drop;
1066 		}
1067  		break;  /* continue normal processing */
1068 	}
1069 
1070 	/*
1071 	 * States other than LISTEN or SYN_SENT.
1072 	 * First check the RST flag and sequence number since reset segments
1073 	 * are exempt from the timestamp and connection count tests.  This
1074 	 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1075 	 * below which allowed reset segments in half the sequence space
1076 	 * to fall though and be processed (which gives forged reset
1077 	 * segments with a random sequence number a 50 percent chance of
1078 	 * killing a connection).
1079 	 * Then check timestamp, if present.
1080 	 * Then check the connection count, if present.
1081 	 * Then check that at least some bytes of segment are within
1082 	 * receive window.  If segment begins before rcv_nxt,
1083 	 * drop leading data (and SYN); if nothing left, just ack.
1084 	 *
1085 	 *
1086 	 * If the RST bit is set, check the sequence number to see
1087 	 * if this is a valid reset segment.
1088 	 * RFC 793 page 37:
1089 	 *   In all states except SYN-SENT, all reset (RST) segments
1090 	 *   are validated by checking their SEQ-fields.  A reset is
1091 	 *   valid if its sequence number is in the window.
1092 	 * Note: this does not take into account delayed ACKs, so
1093 	 *   we should test against last_ack_sent instead of rcv_nxt.
1094 	 *   Also, it does not make sense to allow reset segments with
1095 	 *   sequence numbers greater than last_ack_sent to be processed
1096 	 *   since these sequence numbers are just the acknowledgement
1097 	 *   numbers in our outgoing packets being echoed back at us,
1098 	 *   and these acknowledgement numbers are monotonically
1099 	 *   increasing.
1100 	 * If we have multiple segments in flight, the intial reset
1101 	 * segment sequence numbers will be to the left of last_ack_sent,
1102 	 * but they will eventually catch up.
1103 	 * In any case, it never made sense to trim reset segments to
1104 	 * fit the receive window since RFC 1122 says:
1105 	 *   4.2.2.12  RST Segment: RFC-793 Section 3.4
1106 	 *
1107 	 *    A TCP SHOULD allow a received RST segment to include data.
1108 	 *
1109 	 *    DISCUSSION
1110 	 *         It has been suggested that a RST segment could contain
1111 	 *         ASCII text that encoded and explained the cause of the
1112 	 *         RST.  No standard has yet been established for such
1113 	 *         data.
1114 	 *
1115 	 * If the reset segment passes the sequence number test examine
1116 	 * the state:
1117 	 *    SYN_RECEIVED STATE:
1118 	 *	If passive open, return to LISTEN state.
1119 	 *	If active open, inform user that connection was refused.
1120 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1121 	 *	Inform user that connection was reset, and close tcb.
1122 	 *    CLOSING, LAST_ACK STATES:
1123 	 *	Close the tcb.
1124 	 *    TIME_WAIT STATE:
1125 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
1126 	 *      RFC 1337.
1127 	 */
1128 	if (tiflags & TH_RST) {
1129 		if (tp->last_ack_sent == ti->ti_seq) {
1130 			switch (tp->t_state) {
1131 
1132 			case TCPS_SYN_RECEIVED:
1133 				so->so_error = ECONNREFUSED;
1134 				goto close;
1135 
1136 			case TCPS_ESTABLISHED:
1137 			case TCPS_FIN_WAIT_1:
1138 			case TCPS_FIN_WAIT_2:
1139 			case TCPS_CLOSE_WAIT:
1140 				so->so_error = ECONNRESET;
1141 			close:
1142 				tp->t_state = TCPS_CLOSED;
1143 				tcpstat.tcps_drops++;
1144 				tp = tcp_close(tp);
1145 				break;
1146 
1147 			case TCPS_CLOSING:
1148 			case TCPS_LAST_ACK:
1149 				tp = tcp_close(tp);
1150 				break;
1151 
1152 			case TCPS_TIME_WAIT:
1153 				break;
1154 			}
1155 		}
1156 		goto drop;
1157 	}
1158 
1159 	/*
1160 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1161 	 * and it's less than ts_recent, drop it.
1162 	 */
1163 	if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent &&
1164 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1165 
1166 		/* Check to see if ts_recent is over 24 days old.  */
1167 		if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1168 			/*
1169 			 * Invalidate ts_recent.  If this segment updates
1170 			 * ts_recent, the age will be reset later and ts_recent
1171 			 * will get a valid value.  If it does not, setting
1172 			 * ts_recent to zero will at least satisfy the
1173 			 * requirement that zero be placed in the timestamp
1174 			 * echo reply when ts_recent isn't valid.  The
1175 			 * age isn't reset until we get a valid ts_recent
1176 			 * because we don't want out-of-order segments to be
1177 			 * dropped when ts_recent is old.
1178 			 */
1179 			tp->ts_recent = 0;
1180 		} else {
1181 			tcpstat.tcps_rcvduppack++;
1182 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
1183 			tcpstat.tcps_pawsdrop++;
1184 			goto dropafterack;
1185 		}
1186 	}
1187 
1188 	/*
1189 	 * T/TCP mechanism
1190 	 *   If T/TCP was negotiated and the segment doesn't have CC,
1191 	 *   or if its CC is wrong then drop the segment.
1192 	 *   RST segments do not have to comply with this.
1193 	 */
1194 	if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1195 	    ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1196  		goto dropafterack;
1197 
1198 	/*
1199 	 * In the SYN-RECEIVED state, validate that the packet belongs to
1200 	 * this connection before trimming the data to fit the receive
1201 	 * window.  Check the sequence number versus IRS since we know
1202 	 * the sequence numbers haven't wrapped.  This is a partial fix
1203 	 * for the "LAND" DoS attack.
1204 	 */
1205 	if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(ti->ti_seq, tp->irs))
1206 		goto dropwithreset;
1207 
1208 	todrop = tp->rcv_nxt - ti->ti_seq;
1209 	if (todrop > 0) {
1210 		if (tiflags & TH_SYN) {
1211 			tiflags &= ~TH_SYN;
1212 			ti->ti_seq++;
1213 			if (ti->ti_urp > 1)
1214 				ti->ti_urp--;
1215 			else
1216 				tiflags &= ~TH_URG;
1217 			todrop--;
1218 		}
1219 		/*
1220 		 * Following if statement from Stevens, vol. 2, p. 960.
1221 		 */
1222 		if (todrop > ti->ti_len
1223 		    || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
1224 			/*
1225 			 * Any valid FIN must be to the left of the window.
1226 			 * At this point the FIN must be a duplicate or out
1227 			 * of sequence; drop it.
1228 			 */
1229 			tiflags &= ~TH_FIN;
1230 
1231 			/*
1232 			 * Send an ACK to resynchronize and drop any data.
1233 			 * But keep on processing for RST or ACK.
1234 			 */
1235 			tp->t_flags |= TF_ACKNOW;
1236 			todrop = ti->ti_len;
1237 			tcpstat.tcps_rcvduppack++;
1238 			tcpstat.tcps_rcvdupbyte += todrop;
1239 		} else {
1240 			tcpstat.tcps_rcvpartduppack++;
1241 			tcpstat.tcps_rcvpartdupbyte += todrop;
1242 		}
1243 		m_adj(m, todrop);
1244 		ti->ti_seq += todrop;
1245 		ti->ti_len -= todrop;
1246 		if (ti->ti_urp > todrop)
1247 			ti->ti_urp -= todrop;
1248 		else {
1249 			tiflags &= ~TH_URG;
1250 			ti->ti_urp = 0;
1251 		}
1252 	}
1253 
1254 	/*
1255 	 * If new data are received on a connection after the
1256 	 * user processes are gone, then RST the other end.
1257 	 */
1258 	if ((so->so_state & SS_NOFDREF) &&
1259 	    tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
1260 		tp = tcp_close(tp);
1261 		tcpstat.tcps_rcvafterclose++;
1262 		goto dropwithreset;
1263 	}
1264 
1265 	/*
1266 	 * If segment ends after window, drop trailing data
1267 	 * (and PUSH and FIN); if nothing left, just ACK.
1268 	 */
1269 	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1270 	if (todrop > 0) {
1271 		tcpstat.tcps_rcvpackafterwin++;
1272 		if (todrop >= ti->ti_len) {
1273 			tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1274 			/*
1275 			 * If a new connection request is received
1276 			 * while in TIME_WAIT, drop the old connection
1277 			 * and start over if the sequence numbers
1278 			 * are above the previous ones.
1279 			 */
1280 			if (tiflags & TH_SYN &&
1281 			    tp->t_state == TCPS_TIME_WAIT &&
1282 			    SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1283 				iss = tp->snd_nxt + TCP_ISSINCR;
1284 				tp = tcp_close(tp);
1285 				goto findpcb;
1286 			}
1287 			/*
1288 			 * If window is closed can only take segments at
1289 			 * window edge, and have to drop data and PUSH from
1290 			 * incoming segments.  Continue processing, but
1291 			 * remember to ack.  Otherwise, drop segment
1292 			 * and ack.
1293 			 */
1294 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1295 				tp->t_flags |= TF_ACKNOW;
1296 				tcpstat.tcps_rcvwinprobe++;
1297 			} else
1298 				goto dropafterack;
1299 		} else
1300 			tcpstat.tcps_rcvbyteafterwin += todrop;
1301 		m_adj(m, -todrop);
1302 		ti->ti_len -= todrop;
1303 		tiflags &= ~(TH_PUSH|TH_FIN);
1304 	}
1305 
1306 	/*
1307 	 * If last ACK falls within this segment's sequence numbers,
1308 	 * record its timestamp.
1309 	 * NOTE that the test is modified according to the latest
1310 	 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1311 	 */
1312 	if ((to.to_flag & TOF_TS) != 0 &&
1313 	    SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
1314 		tp->ts_recent_age = ticks;
1315 		tp->ts_recent = to.to_tsval;
1316 	}
1317 
1318 	/*
1319 	 * If a SYN is in the window, then this is an
1320 	 * error and we send an RST and drop the connection.
1321 	 */
1322 	if (tiflags & TH_SYN) {
1323 		tp = tcp_drop(tp, ECONNRESET);
1324 		goto dropwithreset;
1325 	}
1326 
1327 	/*
1328 	 * If the ACK bit is off:  if in SYN-RECEIVED state or SENDSYN
1329 	 * flag is on (half-synchronized state), then queue data for
1330 	 * later processing; else drop segment and return.
1331 	 */
1332 	if ((tiflags & TH_ACK) == 0) {
1333 		if (tp->t_state == TCPS_SYN_RECEIVED ||
1334 		    (tp->t_flags & TF_NEEDSYN))
1335 			goto step6;
1336 		else
1337 			goto drop;
1338 	}
1339 
1340 	/*
1341 	 * Ack processing.
1342 	 */
1343 	switch (tp->t_state) {
1344 
1345 	/*
1346 	 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1347 	 * ESTABLISHED state and continue processing.
1348 	 * The ACK was checked above.
1349 	 */
1350 	case TCPS_SYN_RECEIVED:
1351 
1352 		tcpstat.tcps_connects++;
1353 		soisconnected(so);
1354 		/* Do window scaling? */
1355 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1356 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1357 			tp->snd_scale = tp->requested_s_scale;
1358 			tp->rcv_scale = tp->request_r_scale;
1359 		}
1360 		/*
1361 		 * Upon successful completion of 3-way handshake,
1362 		 * update cache.CC if it was undefined, pass any queued
1363 		 * data to the user, and advance state appropriately.
1364 		 */
1365 		if ((taop = tcp_gettaocache(inp)) != NULL &&
1366 		    taop->tao_cc == 0)
1367 			taop->tao_cc = tp->cc_recv;
1368 
1369 		/*
1370 		 * Make transitions:
1371 		 *      SYN-RECEIVED  -> ESTABLISHED
1372 		 *      SYN-RECEIVED* -> FIN-WAIT-1
1373 		 */
1374 		tp->t_starttime = ticks;
1375 		if (tp->t_flags & TF_NEEDFIN) {
1376 			tp->t_state = TCPS_FIN_WAIT_1;
1377 			tp->t_flags &= ~TF_NEEDFIN;
1378 		} else {
1379 			tp->t_state = TCPS_ESTABLISHED;
1380 			callout_reset(tp->tt_keep, tcp_keepidle,
1381 				      tcp_timer_keep, tp);
1382 		}
1383 		/*
1384 		 * If segment contains data or ACK, will call tcp_reass()
1385 		 * later; if not, do so now to pass queued data to user.
1386 		 */
1387 		if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0)
1388 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
1389 			    (struct mbuf *)0);
1390 		tp->snd_wl1 = ti->ti_seq - 1;
1391 		/* fall into ... */
1392 
1393 	/*
1394 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1395 	 * ACKs.  If the ack is in the range
1396 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
1397 	 * then advance tp->snd_una to ti->ti_ack and drop
1398 	 * data from the retransmission queue.  If this ACK reflects
1399 	 * more up to date window information we update our window information.
1400 	 */
1401 	case TCPS_ESTABLISHED:
1402 	case TCPS_FIN_WAIT_1:
1403 	case TCPS_FIN_WAIT_2:
1404 	case TCPS_CLOSE_WAIT:
1405 	case TCPS_CLOSING:
1406 	case TCPS_LAST_ACK:
1407 	case TCPS_TIME_WAIT:
1408 
1409 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1410 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1411 				tcpstat.tcps_rcvdupack++;
1412 				/*
1413 				 * If we have outstanding data (other than
1414 				 * a window probe), this is a completely
1415 				 * duplicate ack (ie, window info didn't
1416 				 * change), the ack is the biggest we've
1417 				 * seen and we've seen exactly our rexmt
1418 				 * threshhold of them, assume a packet
1419 				 * has been dropped and retransmit it.
1420 				 * Kludge snd_nxt & the congestion
1421 				 * window so we send only this one
1422 				 * packet.
1423 				 *
1424 				 * We know we're losing at the current
1425 				 * window size so do congestion avoidance
1426 				 * (set ssthresh to half the current window
1427 				 * and pull our congestion window back to
1428 				 * the new ssthresh).
1429 				 *
1430 				 * Dup acks mean that packets have left the
1431 				 * network (they're now cached at the receiver)
1432 				 * so bump cwnd by the amount in the receiver
1433 				 * to keep a constant cwnd packets in the
1434 				 * network.
1435 				 */
1436 				if (!callout_active(tp->tt_rexmt) ||
1437 				    ti->ti_ack != tp->snd_una)
1438 					tp->t_dupacks = 0;
1439 				else if (++tp->t_dupacks == tcprexmtthresh) {
1440 					tcp_seq onxt = tp->snd_nxt;
1441 					u_int win =
1442 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1443 						tp->t_maxseg;
1444 
1445 					if (win < 2)
1446 						win = 2;
1447 					tp->snd_ssthresh = win * tp->t_maxseg;
1448 					callout_stop(tp->tt_rexmt);
1449 					tp->t_rtttime = 0;
1450 					tp->snd_nxt = ti->ti_ack;
1451 					tp->snd_cwnd = tp->t_maxseg;
1452 					(void) tcp_output(tp);
1453 					tp->snd_cwnd = tp->snd_ssthresh +
1454 					       tp->t_maxseg * tp->t_dupacks;
1455 					if (SEQ_GT(onxt, tp->snd_nxt))
1456 						tp->snd_nxt = onxt;
1457 					goto drop;
1458 				} else if (tp->t_dupacks > tcprexmtthresh) {
1459 					tp->snd_cwnd += tp->t_maxseg;
1460 					(void) tcp_output(tp);
1461 					goto drop;
1462 				}
1463 			} else
1464 				tp->t_dupacks = 0;
1465 			break;
1466 		}
1467 		/*
1468 		 * If the congestion window was inflated to account
1469 		 * for the other side's cached packets, retract it.
1470 		 */
1471 		if (tp->t_dupacks >= tcprexmtthresh &&
1472 		    tp->snd_cwnd > tp->snd_ssthresh)
1473 			tp->snd_cwnd = tp->snd_ssthresh;
1474 		tp->t_dupacks = 0;
1475 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1476 			tcpstat.tcps_rcvacktoomuch++;
1477 			goto dropafterack;
1478 		}
1479 		/*
1480 		 *  If we reach this point, ACK is not a duplicate,
1481 		 *     i.e., it ACKs something we sent.
1482 		 */
1483 		if (tp->t_flags & TF_NEEDSYN) {
1484 			/*
1485 			 * T/TCP: Connection was half-synchronized, and our
1486 			 * SYN has been ACK'd (so connection is now fully
1487 			 * synchronized).  Go to non-starred state,
1488 			 * increment snd_una for ACK of SYN, and check if
1489 			 * we can do window scaling.
1490 			 */
1491 			tp->t_flags &= ~TF_NEEDSYN;
1492 			tp->snd_una++;
1493 			/* Do window scaling? */
1494 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1495 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1496 				tp->snd_scale = tp->requested_s_scale;
1497 				tp->rcv_scale = tp->request_r_scale;
1498 			}
1499 		}
1500 
1501 process_ACK:
1502 		acked = ti->ti_ack - tp->snd_una;
1503 		tcpstat.tcps_rcvackpack++;
1504 		tcpstat.tcps_rcvackbyte += acked;
1505 
1506 		/*
1507 		 * If we just performed our first retransmit, and the ACK
1508 		 * arrives within our recovery window, then it was a mistake
1509 		 * to do the retransmit in the first place.  Recover our
1510 		 * original cwnd and ssthresh, and proceed to transmit where
1511 		 * we left off.
1512 		 */
1513 		if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
1514 			tp->snd_cwnd = tp->snd_cwnd_prev;
1515 			tp->snd_ssthresh = tp->snd_ssthresh_prev;
1516 			tp->snd_nxt = tp->snd_max;
1517 			tp->t_badrxtwin = 0;	/* XXX probably not required */
1518 		}
1519 
1520 		/*
1521 		 * If we have a timestamp reply, update smoothed
1522 		 * round trip time.  If no timestamp is present but
1523 		 * transmit timer is running and timed sequence
1524 		 * number was acked, update smoothed round trip time.
1525 		 * Since we now have an rtt measurement, cancel the
1526 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1527 		 * Recompute the initial retransmit timer.
1528 		 */
1529 		if (to.to_flag & TOF_TS)
1530 			tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1531 		else if (tp->t_rtttime && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1532 			tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1533 
1534 		/*
1535 		 * If all outstanding data is acked, stop retransmit
1536 		 * timer and remember to restart (more output or persist).
1537 		 * If there is more data to be acked, restart retransmit
1538 		 * timer, using current (possibly backed-off) value.
1539 		 */
1540 		if (ti->ti_ack == tp->snd_max) {
1541 			callout_stop(tp->tt_rexmt);
1542 			needoutput = 1;
1543 		} else if (!callout_active(tp->tt_persist))
1544 			callout_reset(tp->tt_rexmt, tp->t_rxtcur,
1545 				      tcp_timer_rexmt, tp);
1546 
1547 		/*
1548 		 * If no data (only SYN) was ACK'd,
1549 		 *    skip rest of ACK processing.
1550 		 */
1551 		if (acked == 0)
1552 			goto step6;
1553 
1554 		/*
1555 		 * When new data is acked, open the congestion window.
1556 		 * If the window gives us less than ssthresh packets
1557 		 * in flight, open exponentially (maxseg per packet).
1558 		 * Otherwise open linearly: maxseg per window
1559 		 * (maxseg^2 / cwnd per packet).
1560 		 */
1561 		{
1562 		register u_int cw = tp->snd_cwnd;
1563 		register u_int incr = tp->t_maxseg;
1564 
1565 		if (cw > tp->snd_ssthresh)
1566 			incr = incr * incr / cw;
1567 		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale);
1568 		}
1569 		if (acked > so->so_snd.sb_cc) {
1570 			tp->snd_wnd -= so->so_snd.sb_cc;
1571 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1572 			ourfinisacked = 1;
1573 		} else {
1574 			sbdrop(&so->so_snd, acked);
1575 			tp->snd_wnd -= acked;
1576 			ourfinisacked = 0;
1577 		}
1578 		sowwakeup(so);
1579 		tp->snd_una = ti->ti_ack;
1580 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1581 			tp->snd_nxt = tp->snd_una;
1582 
1583 		switch (tp->t_state) {
1584 
1585 		/*
1586 		 * In FIN_WAIT_1 STATE in addition to the processing
1587 		 * for the ESTABLISHED state if our FIN is now acknowledged
1588 		 * then enter FIN_WAIT_2.
1589 		 */
1590 		case TCPS_FIN_WAIT_1:
1591 			if (ourfinisacked) {
1592 				/*
1593 				 * If we can't receive any more
1594 				 * data, then closing user can proceed.
1595 				 * Starting the timer is contrary to the
1596 				 * specification, but if we don't get a FIN
1597 				 * we'll hang forever.
1598 				 */
1599 				if (so->so_state & SS_CANTRCVMORE) {
1600 					soisdisconnected(so);
1601 					callout_reset(tp->tt_2msl, tcp_maxidle,
1602 						      tcp_timer_2msl, tp);
1603 				}
1604 				tp->t_state = TCPS_FIN_WAIT_2;
1605 			}
1606 			break;
1607 
1608 	 	/*
1609 		 * In CLOSING STATE in addition to the processing for
1610 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1611 		 * then enter the TIME-WAIT state, otherwise ignore
1612 		 * the segment.
1613 		 */
1614 		case TCPS_CLOSING:
1615 			if (ourfinisacked) {
1616 				tp->t_state = TCPS_TIME_WAIT;
1617 				tcp_canceltimers(tp);
1618 				/* Shorten TIME_WAIT [RFC-1644, p.28] */
1619 				if (tp->cc_recv != 0 &&
1620 				    (ticks - tp->t_starttime) < tcp_msl)
1621 					callout_reset(tp->tt_2msl,
1622 						      tp->t_rxtcur *
1623 						      TCPTV_TWTRUNC,
1624 						      tcp_timer_2msl, tp);
1625 				else
1626 					callout_reset(tp->tt_2msl, 2 * tcp_msl,
1627 						      tcp_timer_2msl, tp);
1628 				soisdisconnected(so);
1629 			}
1630 			break;
1631 
1632 		/*
1633 		 * In LAST_ACK, we may still be waiting for data to drain
1634 		 * and/or to be acked, as well as for the ack of our FIN.
1635 		 * If our FIN is now acknowledged, delete the TCB,
1636 		 * enter the closed state and return.
1637 		 */
1638 		case TCPS_LAST_ACK:
1639 			if (ourfinisacked) {
1640 				tp = tcp_close(tp);
1641 				goto drop;
1642 			}
1643 			break;
1644 
1645 		/*
1646 		 * In TIME_WAIT state the only thing that should arrive
1647 		 * is a retransmission of the remote FIN.  Acknowledge
1648 		 * it and restart the finack timer.
1649 		 */
1650 		case TCPS_TIME_WAIT:
1651 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
1652 				      tcp_timer_2msl, tp);
1653 			goto dropafterack;
1654 		}
1655 	}
1656 
1657 step6:
1658 	/*
1659 	 * Update window information.
1660 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1661 	 */
1662 	if ((tiflags & TH_ACK) &&
1663 	    (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1664 	    (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1665 	     (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1666 		/* keep track of pure window updates */
1667 		if (ti->ti_len == 0 &&
1668 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1669 			tcpstat.tcps_rcvwinupd++;
1670 		tp->snd_wnd = tiwin;
1671 		tp->snd_wl1 = ti->ti_seq;
1672 		tp->snd_wl2 = ti->ti_ack;
1673 		if (tp->snd_wnd > tp->max_sndwnd)
1674 			tp->max_sndwnd = tp->snd_wnd;
1675 		needoutput = 1;
1676 	}
1677 
1678 	/*
1679 	 * Process segments with URG.
1680 	 */
1681 	if ((tiflags & TH_URG) && ti->ti_urp &&
1682 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1683 		/*
1684 		 * This is a kludge, but if we receive and accept
1685 		 * random urgent pointers, we'll crash in
1686 		 * soreceive.  It's hard to imagine someone
1687 		 * actually wanting to send this much urgent data.
1688 		 */
1689 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1690 			ti->ti_urp = 0;			/* XXX */
1691 			tiflags &= ~TH_URG;		/* XXX */
1692 			goto dodata;			/* XXX */
1693 		}
1694 		/*
1695 		 * If this segment advances the known urgent pointer,
1696 		 * then mark the data stream.  This should not happen
1697 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1698 		 * a FIN has been received from the remote side.
1699 		 * In these states we ignore the URG.
1700 		 *
1701 		 * According to RFC961 (Assigned Protocols),
1702 		 * the urgent pointer points to the last octet
1703 		 * of urgent data.  We continue, however,
1704 		 * to consider it to indicate the first octet
1705 		 * of data past the urgent section as the original
1706 		 * spec states (in one of two places).
1707 		 */
1708 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1709 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1710 			so->so_oobmark = so->so_rcv.sb_cc +
1711 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1712 			if (so->so_oobmark == 0)
1713 				so->so_state |= SS_RCVATMARK;
1714 			sohasoutofband(so);
1715 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1716 		}
1717 		/*
1718 		 * Remove out of band data so doesn't get presented to user.
1719 		 * This can happen independent of advancing the URG pointer,
1720 		 * but if two URG's are pending at once, some out-of-band
1721 		 * data may creep in... ick.
1722 		 */
1723 		if (ti->ti_urp <= (u_long)ti->ti_len
1724 #ifdef SO_OOBINLINE
1725 		     && (so->so_options & SO_OOBINLINE) == 0
1726 #endif
1727 		     )
1728 			tcp_pulloutofband(so, ti, m);
1729 	} else
1730 		/*
1731 		 * If no out of band data is expected,
1732 		 * pull receive urgent pointer along
1733 		 * with the receive window.
1734 		 */
1735 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1736 			tp->rcv_up = tp->rcv_nxt;
1737 dodata:							/* XXX */
1738 
1739 	/*
1740 	 * Process the segment text, merging it into the TCP sequencing queue,
1741 	 * and arranging for acknowledgment of receipt if necessary.
1742 	 * This process logically involves adjusting tp->rcv_wnd as data
1743 	 * is presented to the user (this happens in tcp_usrreq.c,
1744 	 * case PRU_RCVD).  If a FIN has already been received on this
1745 	 * connection then we just ignore the text.
1746 	 */
1747 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
1748 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1749 		TCP_REASS(tp, ti, m, so, tiflags);
1750 		/*
1751 		 * Note the amount of data that peer has sent into
1752 		 * our window, in order to estimate the sender's
1753 		 * buffer size.
1754 		 */
1755 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1756 	} else {
1757 		m_freem(m);
1758 		tiflags &= ~TH_FIN;
1759 	}
1760 
1761 	/*
1762 	 * If FIN is received ACK the FIN and let the user know
1763 	 * that the connection is closing.
1764 	 */
1765 	if (tiflags & TH_FIN) {
1766 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1767 			socantrcvmore(so);
1768 			/*
1769 			 *  If connection is half-synchronized
1770 			 *  (ie NEEDSYN flag on) then delay ACK,
1771 			 *  so it may be piggybacked when SYN is sent.
1772 			 *  Otherwise, since we received a FIN then no
1773 			 *  more input can be expected, send ACK now.
1774 			 */
1775 			if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN))
1776                                 callout_reset(tp->tt_delack, tcp_delacktime,
1777                                     tcp_timer_delack, tp);
1778 			else
1779 				tp->t_flags |= TF_ACKNOW;
1780 			tp->rcv_nxt++;
1781 		}
1782 		switch (tp->t_state) {
1783 
1784 	 	/*
1785 		 * In SYN_RECEIVED and ESTABLISHED STATES
1786 		 * enter the CLOSE_WAIT state.
1787 		 */
1788 		case TCPS_SYN_RECEIVED:
1789 			tp->t_starttime = ticks;
1790 			/*FALLTHROUGH*/
1791 		case TCPS_ESTABLISHED:
1792 			tp->t_state = TCPS_CLOSE_WAIT;
1793 			break;
1794 
1795 	 	/*
1796 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1797 		 * enter the CLOSING state.
1798 		 */
1799 		case TCPS_FIN_WAIT_1:
1800 			tp->t_state = TCPS_CLOSING;
1801 			break;
1802 
1803 	 	/*
1804 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1805 		 * starting the time-wait timer, turning off the other
1806 		 * standard timers.
1807 		 */
1808 		case TCPS_FIN_WAIT_2:
1809 			tp->t_state = TCPS_TIME_WAIT;
1810 			tcp_canceltimers(tp);
1811 			/* Shorten TIME_WAIT [RFC-1644, p.28] */
1812 			if (tp->cc_recv != 0 &&
1813 			    (ticks - tp->t_starttime) < tcp_msl) {
1814 				callout_reset(tp->tt_2msl,
1815 					      tp->t_rxtcur * TCPTV_TWTRUNC,
1816 					      tcp_timer_2msl, tp);
1817 				/* For transaction client, force ACK now. */
1818 				tp->t_flags |= TF_ACKNOW;
1819 			}
1820 			else
1821 				callout_reset(tp->tt_2msl, 2 * tcp_msl,
1822 					      tcp_timer_2msl, tp);
1823 			soisdisconnected(so);
1824 			break;
1825 
1826 		/*
1827 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1828 		 */
1829 		case TCPS_TIME_WAIT:
1830 			callout_reset(tp->tt_2msl, 2 * tcp_msl,
1831 				      tcp_timer_2msl, tp);
1832 			break;
1833 		}
1834 	}
1835 #ifdef TCPDEBUG
1836 	if (so->so_options & SO_DEBUG)
1837 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1838 #endif
1839 
1840 	/*
1841 	 * Return any desired output.
1842 	 */
1843 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1844 		(void) tcp_output(tp);
1845 	return;
1846 
1847 dropafterack:
1848 	/*
1849 	 * Generate an ACK dropping incoming segment if it occupies
1850 	 * sequence space, where the ACK reflects our state.
1851 	 *
1852 	 * We can now skip the test for the RST flag since all
1853 	 * paths to this code happen after packets containing
1854 	 * RST have been dropped.
1855 	 *
1856 	 * In the SYN-RECEIVED state, don't send an ACK unless the
1857 	 * segment we received passes the SYN-RECEIVED ACK test.
1858 	 * If it fails send a RST.  This breaks the loop in the
1859 	 * "LAND" DoS attack, and also prevents an ACK storm
1860 	 * between two listening ports that have been sent forged
1861 	 * SYN segments, each with the source address of the other.
1862 	 */
1863 	if (tp->t_state == TCPS_SYN_RECEIVED && (tiflags & TH_ACK) &&
1864 	    (SEQ_GT(tp->snd_una, ti->ti_ack) ||
1865 	     SEQ_GT(ti->ti_ack, tp->snd_max)) )
1866 		goto dropwithreset;
1867 #ifdef TCPDEBUG
1868 	if (so->so_options & SO_DEBUG)
1869 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1870 #endif
1871 	m_freem(m);
1872 	tp->t_flags |= TF_ACKNOW;
1873 	(void) tcp_output(tp);
1874 	return;
1875 
1876 dropwithreset:
1877 #ifdef TCP_RESTRICT_RST
1878 	if (restrict_rst)
1879 		goto drop;
1880 #endif
1881 	/*
1882 	 * Generate a RST, dropping incoming segment.
1883 	 * Make ACK acceptable to originator of segment.
1884 	 * Don't bother to respond if destination was broadcast/multicast.
1885 	 */
1886 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1887 	    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1888 		goto drop;
1889 #ifdef TCPDEBUG
1890 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1891 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1892 #endif
1893 	if (tiflags & TH_ACK)
1894 		tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1895 	else {
1896 		if (tiflags & TH_SYN)
1897 			ti->ti_len++;
1898 		tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1899 		    TH_RST|TH_ACK);
1900 	}
1901 	/* destroy temporarily created socket */
1902 	if (dropsocket)
1903 		(void) soabort(so);
1904 	return;
1905 
1906 drop:
1907 	/*
1908 	 * Drop space held by incoming segment and return.
1909 	 */
1910 #ifdef TCPDEBUG
1911 	if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1912 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1913 #endif
1914 	m_freem(m);
1915 	/* destroy temporarily created socket */
1916 	if (dropsocket)
1917 		(void) soabort(so);
1918 	return;
1919 }
1920 
1921 static void
1922 tcp_dooptions(tp, cp, cnt, ti, to)
1923 	struct tcpcb *tp;
1924 	u_char *cp;
1925 	int cnt;
1926 	struct tcpiphdr *ti;
1927 	struct tcpopt *to;
1928 {
1929 	u_short mss = 0;
1930 	int opt, optlen;
1931 
1932 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1933 		opt = cp[0];
1934 		if (opt == TCPOPT_EOL)
1935 			break;
1936 		if (opt == TCPOPT_NOP)
1937 			optlen = 1;
1938 		else {
1939 			optlen = cp[1];
1940 			if (optlen <= 0)
1941 				break;
1942 		}
1943 		switch (opt) {
1944 
1945 		default:
1946 			continue;
1947 
1948 		case TCPOPT_MAXSEG:
1949 			if (optlen != TCPOLEN_MAXSEG)
1950 				continue;
1951 			if (!(ti->ti_flags & TH_SYN))
1952 				continue;
1953 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
1954 			NTOHS(mss);
1955 			break;
1956 
1957 		case TCPOPT_WINDOW:
1958 			if (optlen != TCPOLEN_WINDOW)
1959 				continue;
1960 			if (!(ti->ti_flags & TH_SYN))
1961 				continue;
1962 			tp->t_flags |= TF_RCVD_SCALE;
1963 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1964 			break;
1965 
1966 		case TCPOPT_TIMESTAMP:
1967 			if (optlen != TCPOLEN_TIMESTAMP)
1968 				continue;
1969 			to->to_flag |= TOF_TS;
1970 			bcopy((char *)cp + 2,
1971 			    (char *)&to->to_tsval, sizeof(to->to_tsval));
1972 			NTOHL(to->to_tsval);
1973 			bcopy((char *)cp + 6,
1974 			    (char *)&to->to_tsecr, sizeof(to->to_tsecr));
1975 			NTOHL(to->to_tsecr);
1976 
1977 			/*
1978 			 * A timestamp received in a SYN makes
1979 			 * it ok to send timestamp requests and replies.
1980 			 */
1981 			if (ti->ti_flags & TH_SYN) {
1982 				tp->t_flags |= TF_RCVD_TSTMP;
1983 				tp->ts_recent = to->to_tsval;
1984 				tp->ts_recent_age = ticks;
1985 			}
1986 			break;
1987 		case TCPOPT_CC:
1988 			if (optlen != TCPOLEN_CC)
1989 				continue;
1990 			to->to_flag |= TOF_CC;
1991 			bcopy((char *)cp + 2,
1992 			    (char *)&to->to_cc, sizeof(to->to_cc));
1993 			NTOHL(to->to_cc);
1994 			/*
1995 			 * A CC or CC.new option received in a SYN makes
1996 			 * it ok to send CC in subsequent segments.
1997 			 */
1998 			if (ti->ti_flags & TH_SYN)
1999 				tp->t_flags |= TF_RCVD_CC;
2000 			break;
2001 		case TCPOPT_CCNEW:
2002 			if (optlen != TCPOLEN_CC)
2003 				continue;
2004 			if (!(ti->ti_flags & TH_SYN))
2005 				continue;
2006 			to->to_flag |= TOF_CCNEW;
2007 			bcopy((char *)cp + 2,
2008 			    (char *)&to->to_cc, sizeof(to->to_cc));
2009 			NTOHL(to->to_cc);
2010 			/*
2011 			 * A CC or CC.new option received in a SYN makes
2012 			 * it ok to send CC in subsequent segments.
2013 			 */
2014 			tp->t_flags |= TF_RCVD_CC;
2015 			break;
2016 		case TCPOPT_CCECHO:
2017 			if (optlen != TCPOLEN_CC)
2018 				continue;
2019 			if (!(ti->ti_flags & TH_SYN))
2020 				continue;
2021 			to->to_flag |= TOF_CCECHO;
2022 			bcopy((char *)cp + 2,
2023 			    (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2024 			NTOHL(to->to_ccecho);
2025 			break;
2026 		}
2027 	}
2028 	if (ti->ti_flags & TH_SYN)
2029 		tcp_mss(tp, mss);	/* sets t_maxseg */
2030 }
2031 
2032 /*
2033  * Pull out of band byte out of a segment so
2034  * it doesn't appear in the user's data queue.
2035  * It is still reflected in the segment length for
2036  * sequencing purposes.
2037  */
2038 static void
2039 tcp_pulloutofband(so, ti, m)
2040 	struct socket *so;
2041 	struct tcpiphdr *ti;
2042 	register struct mbuf *m;
2043 {
2044 	int cnt = ti->ti_urp - 1;
2045 
2046 	while (cnt >= 0) {
2047 		if (m->m_len > cnt) {
2048 			char *cp = mtod(m, caddr_t) + cnt;
2049 			struct tcpcb *tp = sototcpcb(so);
2050 
2051 			tp->t_iobc = *cp;
2052 			tp->t_oobflags |= TCPOOB_HAVEDATA;
2053 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2054 			m->m_len--;
2055 			return;
2056 		}
2057 		cnt -= m->m_len;
2058 		m = m->m_next;
2059 		if (m == 0)
2060 			break;
2061 	}
2062 	panic("tcp_pulloutofband");
2063 }
2064 
2065 /*
2066  * Collect new round-trip time estimate
2067  * and update averages and current timeout.
2068  */
2069 static void
2070 tcp_xmit_timer(tp, rtt)
2071 	register struct tcpcb *tp;
2072 	int rtt;
2073 {
2074 	register int delta;
2075 
2076 	tcpstat.tcps_rttupdated++;
2077 	tp->t_rttupdated++;
2078 	if (tp->t_srtt != 0) {
2079 		/*
2080 		 * srtt is stored as fixed point with 5 bits after the
2081 		 * binary point (i.e., scaled by 8).  The following magic
2082 		 * is equivalent to the smoothing algorithm in rfc793 with
2083 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2084 		 * point).  Adjust rtt to origin 0.
2085 		 */
2086 		delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2087 			- (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2088 
2089 		if ((tp->t_srtt += delta) <= 0)
2090 			tp->t_srtt = 1;
2091 
2092 		/*
2093 		 * We accumulate a smoothed rtt variance (actually, a
2094 		 * smoothed mean difference), then set the retransmit
2095 		 * timer to smoothed rtt + 4 times the smoothed variance.
2096 		 * rttvar is stored as fixed point with 4 bits after the
2097 		 * binary point (scaled by 16).  The following is
2098 		 * equivalent to rfc793 smoothing with an alpha of .75
2099 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
2100 		 * rfc793's wired-in beta.
2101 		 */
2102 		if (delta < 0)
2103 			delta = -delta;
2104 		delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2105 		if ((tp->t_rttvar += delta) <= 0)
2106 			tp->t_rttvar = 1;
2107 	} else {
2108 		/*
2109 		 * No rtt measurement yet - use the unsmoothed rtt.
2110 		 * Set the variance to half the rtt (so our first
2111 		 * retransmit happens at 3*rtt).
2112 		 */
2113 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
2114 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2115 	}
2116 	tp->t_rtttime = 0;
2117 	tp->t_rxtshift = 0;
2118 
2119 	/*
2120 	 * the retransmit should happen at rtt + 4 * rttvar.
2121 	 * Because of the way we do the smoothing, srtt and rttvar
2122 	 * will each average +1/2 tick of bias.  When we compute
2123 	 * the retransmit timer, we want 1/2 tick of rounding and
2124 	 * 1 extra tick because of +-1/2 tick uncertainty in the
2125 	 * firing of the timer.  The bias will give us exactly the
2126 	 * 1.5 tick we need.  But, because the bias is
2127 	 * statistical, we have to test that we don't drop below
2128 	 * the minimum feasible timer (which is 2 ticks).
2129 	 */
2130 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2131 		      max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2132 
2133 	/*
2134 	 * We received an ack for a packet that wasn't retransmitted;
2135 	 * it is probably safe to discard any error indications we've
2136 	 * received recently.  This isn't quite right, but close enough
2137 	 * for now (a route might have failed after we sent a segment,
2138 	 * and the return path might not be symmetrical).
2139 	 */
2140 	tp->t_softerror = 0;
2141 }
2142 
2143 /*
2144  * Determine a reasonable value for maxseg size.
2145  * If the route is known, check route for mtu.
2146  * If none, use an mss that can be handled on the outgoing
2147  * interface without forcing IP to fragment; if bigger than
2148  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2149  * to utilize large mbufs.  If no route is found, route has no mtu,
2150  * or the destination isn't local, use a default, hopefully conservative
2151  * size (usually 512 or the default IP max size, but no more than the mtu
2152  * of the interface), as we can't discover anything about intervening
2153  * gateways or networks.  We also initialize the congestion/slow start
2154  * window to be a single segment if the destination isn't local.
2155  * While looking at the routing entry, we also initialize other path-dependent
2156  * parameters from pre-set or cached values in the routing entry.
2157  *
2158  * Also take into account the space needed for options that we
2159  * send regularly.  Make maxseg shorter by that amount to assure
2160  * that we can send maxseg amount of data even when the options
2161  * are present.  Store the upper limit of the length of options plus
2162  * data in maxopd.
2163  *
2164  * NOTE that this routine is only called when we process an incoming
2165  * segment, for outgoing segments only tcp_mssopt is called.
2166  *
2167  * In case of T/TCP, we call this routine during implicit connection
2168  * setup as well (offer = -1), to initialize maxseg from the cached
2169  * MSS of our peer.
2170  */
2171 void
2172 tcp_mss(tp, offer)
2173 	struct tcpcb *tp;
2174 	int offer;
2175 {
2176 	register struct rtentry *rt;
2177 	struct ifnet *ifp;
2178 	register int rtt, mss;
2179 	u_long bufsize;
2180 	struct inpcb *inp;
2181 	struct socket *so;
2182 	struct rmxp_tao *taop;
2183 	int origoffer = offer;
2184 
2185 	inp = tp->t_inpcb;
2186 	if ((rt = tcp_rtlookup(inp)) == NULL) {
2187 		tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2188 		return;
2189 	}
2190 	ifp = rt->rt_ifp;
2191 	so = inp->inp_socket;
2192 
2193 	taop = rmx_taop(rt->rt_rmx);
2194 	/*
2195 	 * Offer == -1 means that we didn't receive SYN yet,
2196 	 * use cached value in that case;
2197 	 */
2198 	if (offer == -1)
2199 		offer = taop->tao_mssopt;
2200 	/*
2201 	 * Offer == 0 means that there was no MSS on the SYN segment,
2202 	 * in this case we use tcp_mssdflt.
2203 	 */
2204 	if (offer == 0)
2205 		offer = tcp_mssdflt;
2206 	else
2207 		/*
2208 		 * Sanity check: make sure that maxopd will be large
2209 		 * enough to allow some data on segments even is the
2210 		 * all the option space is used (40bytes).  Otherwise
2211 		 * funny things may happen in tcp_output.
2212 		 */
2213 		offer = max(offer, 64);
2214 	taop->tao_mssopt = offer;
2215 
2216 	/*
2217 	 * While we're here, check if there's an initial rtt
2218 	 * or rttvar.  Convert from the route-table units
2219 	 * to scaled multiples of the slow timeout timer.
2220 	 */
2221 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2222 		/*
2223 		 * XXX the lock bit for RTT indicates that the value
2224 		 * is also a minimum value; this is subject to time.
2225 		 */
2226 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
2227 			tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2228 		tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2229 		tcpstat.tcps_usedrtt++;
2230 		if (rt->rt_rmx.rmx_rttvar) {
2231 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2232 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2233 			tcpstat.tcps_usedrttvar++;
2234 		} else {
2235 			/* default variation is +- 1 rtt */
2236 			tp->t_rttvar =
2237 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2238 		}
2239 		TCPT_RANGESET(tp->t_rxtcur,
2240 			      ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2241 			      tp->t_rttmin, TCPTV_REXMTMAX);
2242 	}
2243 	/*
2244 	 * if there's an mtu associated with the route, use it
2245 	 */
2246 	if (rt->rt_rmx.rmx_mtu)
2247 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
2248 	else
2249 	{
2250 		mss = ifp->if_mtu - sizeof(struct tcpiphdr);
2251 		if (!in_localaddr(inp->inp_faddr))
2252 			mss = min(mss, tcp_mssdflt);
2253 	}
2254 	mss = min(mss, offer);
2255 	/*
2256 	 * maxopd stores the maximum length of data AND options
2257 	 * in a segment; maxseg is the amount of data in a normal
2258 	 * segment.  We need to store this value (maxopd) apart
2259 	 * from maxseg, because now every segment carries options
2260 	 * and thus we normally have somewhat less data in segments.
2261 	 */
2262 	tp->t_maxopd = mss;
2263 
2264 	/*
2265 	 * In case of T/TCP, origoffer==-1 indicates, that no segments
2266 	 * were received yet.  In this case we just guess, otherwise
2267 	 * we do the same as before T/TCP.
2268 	 */
2269  	if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2270 	    (origoffer == -1 ||
2271 	     (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2272 		mss -= TCPOLEN_TSTAMP_APPA;
2273  	if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2274 	    (origoffer == -1 ||
2275 	     (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2276 		mss -= TCPOLEN_CC_APPA;
2277 
2278 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
2279 		if (mss > MCLBYTES)
2280 			mss &= ~(MCLBYTES-1);
2281 #else
2282 		if (mss > MCLBYTES)
2283 			mss = mss / MCLBYTES * MCLBYTES;
2284 #endif
2285 	/*
2286 	 * If there's a pipesize, change the socket buffer
2287 	 * to that size.  Make the socket buffers an integral
2288 	 * number of mss units; if the mss is larger than
2289 	 * the socket buffer, decrease the mss.
2290 	 */
2291 #ifdef RTV_SPIPE
2292 	if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2293 #endif
2294 		bufsize = so->so_snd.sb_hiwat;
2295 	if (bufsize < mss)
2296 		mss = bufsize;
2297 	else {
2298 		bufsize = roundup(bufsize, mss);
2299 		if (bufsize > sb_max)
2300 			bufsize = sb_max;
2301 		(void)sbreserve(&so->so_snd, bufsize, so, NULL);
2302 	}
2303 	tp->t_maxseg = mss;
2304 
2305 #ifdef RTV_RPIPE
2306 	if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2307 #endif
2308 		bufsize = so->so_rcv.sb_hiwat;
2309 	if (bufsize > mss) {
2310 		bufsize = roundup(bufsize, mss);
2311 		if (bufsize > sb_max)
2312 			bufsize = sb_max;
2313 		(void)sbreserve(&so->so_rcv, bufsize, so, NULL);
2314 	}
2315 
2316 	/*
2317 	 * Set the slow-start flight size depending on whether this
2318 	 * is a local network or not.
2319 	 */
2320 	if (in_localaddr(inp->inp_faddr))
2321 		tp->snd_cwnd = mss * ss_fltsz_local;
2322 	else
2323 		tp->snd_cwnd = mss * ss_fltsz;
2324 
2325 	if (rt->rt_rmx.rmx_ssthresh) {
2326 		/*
2327 		 * There's some sort of gateway or interface
2328 		 * buffer limit on the path.  Use this to set
2329 		 * the slow start threshhold, but set the
2330 		 * threshold to no less than 2*mss.
2331 		 */
2332 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2333 		tcpstat.tcps_usedssthresh++;
2334 	}
2335 }
2336 
2337 /*
2338  * Determine the MSS option to send on an outgoing SYN.
2339  */
2340 int
2341 tcp_mssopt(tp)
2342 	struct tcpcb *tp;
2343 {
2344 	struct rtentry *rt;
2345 
2346 	rt = tcp_rtlookup(tp->t_inpcb);
2347 	if (rt == NULL)
2348 		return tcp_mssdflt;
2349 
2350 	return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
2351 }
2352