xref: /netbsd-src/sys/netinet/tcp_input.c (revision d9158b13b5dfe46201430699a3f7a235ecf28df3)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
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  *	from: @(#)tcp_input.c	8.5 (Berkeley) 4/10/94
34  *	$Id: tcp_input.c,v 1.9 1994/05/13 06:06:39 mycroft Exp $
35  */
36 
37 #ifndef TUBA_INCLUDE
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/errno.h>
46 
47 #include <net/if.h>
48 #include <net/route.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip_var.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_fsm.h>
57 #include <netinet/tcp_seq.h>
58 #include <netinet/tcp_timer.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/tcpip.h>
61 #include <netinet/tcp_debug.h>
62 
63 int	tcprexmtthresh = 3;
64 struct	tcpiphdr tcp_saveti;
65 struct	inpcb *tcp_last_inpcb = &tcb;
66 
67 extern u_long sb_max;
68 
69 #endif /* TUBA_INCLUDE */
70 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
71 
72 /* for modulo comparisons of timestamps */
73 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
74 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
75 
76 
77 /*
78  * Insert segment ti into reassembly queue of tcp with
79  * control block tp.  Return TH_FIN if reassembly now includes
80  * a segment with FIN.  The macro form does the common case inline
81  * (segment is the next to be received on an established connection,
82  * and the queue is empty), avoiding linkage into and removal
83  * from the queue and repetition of various conversions.
84  * Set DELACK for segments received in order, but ack immediately
85  * when segments are out of order (so fast retransmit can work).
86  */
87 #define	TCP_REASS(tp, ti, m, so, flags) { \
88 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
89 	    (tp)->seg_next == (struct tcpiphdr *)(tp) && \
90 	    (tp)->t_state == TCPS_ESTABLISHED) { \
91 		if ((ti)->ti_flags & TH_PUSH) \
92 			tp->t_flags |= TF_ACKNOW; \
93 		else \
94 			tp->t_flags |= TF_DELACK; \
95 		(tp)->rcv_nxt += (ti)->ti_len; \
96 		flags = (ti)->ti_flags & TH_FIN; \
97 		tcpstat.tcps_rcvpack++;\
98 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
99 		sbappend(&(so)->so_rcv, (m)); \
100 		sorwakeup(so); \
101 	} else { \
102 		(flags) = tcp_reass((tp), (ti), (m)); \
103 		tp->t_flags |= TF_ACKNOW; \
104 	} \
105 }
106 #ifndef TUBA_INCLUDE
107 
108 int
109 tcp_reass(tp, ti, m)
110 	register struct tcpcb *tp;
111 	register struct tcpiphdr *ti;
112 	struct mbuf *m;
113 {
114 	register struct tcpiphdr *q;
115 	struct socket *so = tp->t_inpcb->inp_socket;
116 	int flags;
117 
118 	/*
119 	 * Call with ti==0 after become established to
120 	 * force pre-ESTABLISHED data up to user socket.
121 	 */
122 	if (ti == 0)
123 		goto present;
124 
125 	/*
126 	 * Find a segment which begins after this one does.
127 	 */
128 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
129 	    q = (struct tcpiphdr *)q->ti_next)
130 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
131 			break;
132 
133 	/*
134 	 * If there is a preceding segment, it may provide some of
135 	 * our data already.  If so, drop the data from the incoming
136 	 * segment.  If it provides all of our data, drop us.
137 	 */
138 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
139 		register int i;
140 		q = (struct tcpiphdr *)q->ti_prev;
141 		/* conversion to int (in i) handles seq wraparound */
142 		i = q->ti_seq + q->ti_len - ti->ti_seq;
143 		if (i > 0) {
144 			if (i >= ti->ti_len) {
145 				tcpstat.tcps_rcvduppack++;
146 				tcpstat.tcps_rcvdupbyte += ti->ti_len;
147 				m_freem(m);
148 				return (0);
149 			}
150 			m_adj(m, i);
151 			ti->ti_len -= i;
152 			ti->ti_seq += i;
153 		}
154 		q = (struct tcpiphdr *)(q->ti_next);
155 	}
156 	tcpstat.tcps_rcvoopack++;
157 	tcpstat.tcps_rcvoobyte += ti->ti_len;
158 	REASS_MBUF(ti) = m;		/* XXX */
159 
160 	/*
161 	 * While we overlap succeeding segments trim them or,
162 	 * if they are completely covered, dequeue them.
163 	 */
164 	while (q != (struct tcpiphdr *)tp) {
165 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
166 		if (i <= 0)
167 			break;
168 		if (i < q->ti_len) {
169 			q->ti_seq += i;
170 			q->ti_len -= i;
171 			m_adj(REASS_MBUF(q), i);
172 			break;
173 		}
174 		q = (struct tcpiphdr *)q->ti_next;
175 		m = REASS_MBUF((struct tcpiphdr *)q->ti_prev);
176 		remque(q->ti_prev);
177 		m_freem(m);
178 	}
179 
180 	/*
181 	 * Stick new segment in its place.
182 	 */
183 	insque(ti, q->ti_prev);
184 
185 present:
186 	/*
187 	 * Present data to user, advancing rcv_nxt through
188 	 * completed sequence space.
189 	 */
190 	if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
191 		return (0);
192 	ti = tp->seg_next;
193 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
194 		return (0);
195 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
196 		return (0);
197 	do {
198 		tp->rcv_nxt += ti->ti_len;
199 		flags = ti->ti_flags & TH_FIN;
200 		remque(ti);
201 		m = REASS_MBUF(ti);
202 		ti = (struct tcpiphdr *)ti->ti_next;
203 		if (so->so_state & SS_CANTRCVMORE)
204 			m_freem(m);
205 		else
206 			sbappend(&so->so_rcv, m);
207 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
208 	sorwakeup(so);
209 	return (flags);
210 }
211 
212 /*
213  * TCP input routine, follows pages 65-76 of the
214  * protocol specification dated September, 1981 very closely.
215  */
216 void
217 tcp_input(m, iphlen)
218 	register struct mbuf *m;
219 	int iphlen;
220 {
221 	register struct tcpiphdr *ti;
222 	register struct inpcb *inp;
223 	caddr_t optp = NULL;
224 	int optlen;
225 	int len, tlen, off;
226 	register struct tcpcb *tp = 0;
227 	register int tiflags;
228 	struct socket *so;
229 	int todrop, acked, ourfinisacked, needoutput = 0;
230 	short ostate;
231 	struct in_addr laddr;
232 	int dropsocket = 0;
233 	int iss = 0;
234 	u_long tiwin, ts_val, ts_ecr;
235 	int ts_present = 0;
236 
237 	tcpstat.tcps_rcvtotal++;
238 	/*
239 	 * Get IP and TCP header together in first mbuf.
240 	 * Note: IP leaves IP header in first mbuf.
241 	 */
242 	ti = mtod(m, struct tcpiphdr *);
243 	if (iphlen > sizeof (struct ip))
244 		ip_stripoptions(m, (struct mbuf *)0);
245 	if (m->m_len < sizeof (struct tcpiphdr)) {
246 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
247 			tcpstat.tcps_rcvshort++;
248 			return;
249 		}
250 		ti = mtod(m, struct tcpiphdr *);
251 	}
252 
253 	/*
254 	 * Checksum extended TCP header and data.
255 	 */
256 	tlen = ((struct ip *)ti)->ip_len;
257 	len = sizeof (struct ip) + tlen;
258 	ti->ti_next = ti->ti_prev = 0;
259 	ti->ti_x1 = 0;
260 	ti->ti_len = (u_short)tlen;
261 	HTONS(ti->ti_len);
262 	if (ti->ti_sum = in_cksum(m, len)) {
263 		tcpstat.tcps_rcvbadsum++;
264 		goto drop;
265 	}
266 #endif /* TUBA_INCLUDE */
267 
268 	/*
269 	 * Check that TCP offset makes sense,
270 	 * pull out TCP options and adjust length.		XXX
271 	 */
272 	off = ti->ti_off << 2;
273 	if (off < sizeof (struct tcphdr) || off > tlen) {
274 		tcpstat.tcps_rcvbadoff++;
275 		goto drop;
276 	}
277 	tlen -= off;
278 	ti->ti_len = tlen;
279 	if (off > sizeof (struct tcphdr)) {
280 		if (m->m_len < sizeof(struct ip) + off) {
281 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
282 				tcpstat.tcps_rcvshort++;
283 				return;
284 			}
285 			ti = mtod(m, struct tcpiphdr *);
286 		}
287 		optlen = off - sizeof (struct tcphdr);
288 		optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
289 		/*
290 		 * Do quick retrieval of timestamp options ("options
291 		 * prediction?").  If timestamp is the only option and it's
292 		 * formatted as recommended in RFC 1323 appendix A, we
293 		 * quickly get the values now and not bother calling
294 		 * tcp_dooptions(), etc.
295 		 */
296 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
297 		     (optlen > TCPOLEN_TSTAMP_APPA &&
298 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
299 		     *(u_long *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
300 		     (ti->ti_flags & TH_SYN) == 0) {
301 			ts_present = 1;
302 			ts_val = ntohl(*(u_long *)(optp + 4));
303 			ts_ecr = ntohl(*(u_long *)(optp + 8));
304 			optp = NULL;	/* we've parsed the options */
305 		}
306 	}
307 	tiflags = ti->ti_flags;
308 
309 	/*
310 	 * Convert TCP protocol specific fields to host format.
311 	 */
312 	NTOHL(ti->ti_seq);
313 	NTOHL(ti->ti_ack);
314 	NTOHS(ti->ti_win);
315 	NTOHS(ti->ti_urp);
316 
317 	/*
318 	 * Locate pcb for segment.
319 	 */
320 findpcb:
321 	inp = tcp_last_inpcb;
322 	if (inp->inp_lport != ti->ti_dport ||
323 	    inp->inp_fport != ti->ti_sport ||
324 	    inp->inp_faddr.s_addr != ti->ti_src.s_addr ||
325 	    inp->inp_laddr.s_addr != ti->ti_dst.s_addr) {
326 		inp = in_pcblookup(&tcb, ti->ti_src, ti->ti_sport,
327 		    ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD);
328 		if (inp)
329 			tcp_last_inpcb = inp;
330 		++tcpstat.tcps_pcbcachemiss;
331 	}
332 
333 	/*
334 	 * If the state is CLOSED (i.e., TCB does not exist) then
335 	 * all data in the incoming segment is discarded.
336 	 * If the TCB exists but is in CLOSED state, it is embryonic,
337 	 * but should either do a listen or a connect soon.
338 	 */
339 	if (inp == 0)
340 		goto dropwithreset;
341 	tp = intotcpcb(inp);
342 	if (tp == 0)
343 		goto dropwithreset;
344 	if (tp->t_state == TCPS_CLOSED)
345 		goto drop;
346 
347 	/* Unscale the window into a 32-bit value. */
348 	if ((tiflags & TH_SYN) == 0)
349 		tiwin = ti->ti_win << tp->snd_scale;
350 	else
351 		tiwin = ti->ti_win;
352 
353 	so = inp->inp_socket;
354 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
355 		if (so->so_options & SO_DEBUG) {
356 			ostate = tp->t_state;
357 			tcp_saveti = *ti;
358 		}
359 		if (so->so_options & SO_ACCEPTCONN) {
360 			so = sonewconn(so, 0);
361 			if (so == 0)
362 				goto drop;
363 			/*
364 			 * This is ugly, but ....
365 			 *
366 			 * Mark socket as temporary until we're
367 			 * committed to keeping it.  The code at
368 			 * ``drop'' and ``dropwithreset'' check the
369 			 * flag dropsocket to see if the temporary
370 			 * socket created here should be discarded.
371 			 * We mark the socket as discardable until
372 			 * we're committed to it below in TCPS_LISTEN.
373 			 */
374 			dropsocket++;
375 			inp = (struct inpcb *)so->so_pcb;
376 			inp->inp_laddr = ti->ti_dst;
377 			inp->inp_lport = ti->ti_dport;
378 #if BSD>=43
379 			inp->inp_options = ip_srcroute();
380 #endif
381 			tp = intotcpcb(inp);
382 			tp->t_state = TCPS_LISTEN;
383 
384 			/* Compute proper scaling value from buffer space
385 			 */
386 			while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
387 			   TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
388 				tp->request_r_scale++;
389 		}
390 	}
391 
392 	/*
393 	 * Segment received on connection.
394 	 * Reset idle time and keep-alive timer.
395 	 */
396 	tp->t_idle = 0;
397 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
398 
399 	/*
400 	 * Process options if not in LISTEN state,
401 	 * else do it below (after getting remote address).
402 	 */
403 	if (optp && tp->t_state != TCPS_LISTEN)
404 		tcp_dooptions(tp, optp, optlen, ti,
405 			&ts_present, &ts_val, &ts_ecr);
406 
407 	/*
408 	 * Header prediction: check for the two common cases
409 	 * of a uni-directional data xfer.  If the packet has
410 	 * no control flags, is in-sequence, the window didn't
411 	 * change and we're not retransmitting, it's a
412 	 * candidate.  If the length is zero and the ack moved
413 	 * forward, we're the sender side of the xfer.  Just
414 	 * free the data acked & wake any higher level process
415 	 * that was blocked waiting for space.  If the length
416 	 * is non-zero and the ack didn't move, we're the
417 	 * receiver side.  If we're getting packets in-order
418 	 * (the reassembly queue is empty), add the data to
419 	 * the socket buffer and note that we need a delayed ack.
420 	 */
421 	if (tp->t_state == TCPS_ESTABLISHED &&
422 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
423 	    (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) &&
424 	    ti->ti_seq == tp->rcv_nxt &&
425 	    tiwin && tiwin == tp->snd_wnd &&
426 	    tp->snd_nxt == tp->snd_max) {
427 
428 		/*
429 		 * If last ACK falls within this segment's sequence numbers,
430 		 *  record the timestamp.
431 		 */
432 		if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
433 		   SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
434 			tp->ts_recent_age = tcp_now;
435 			tp->ts_recent = ts_val;
436 		}
437 
438 		if (ti->ti_len == 0) {
439 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
440 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
441 			    tp->snd_cwnd >= tp->snd_wnd) {
442 				/*
443 				 * this is a pure ack for outstanding data.
444 				 */
445 				++tcpstat.tcps_predack;
446 				if (ts_present)
447 					tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
448 				else if (tp->t_rtt &&
449 					    SEQ_GT(ti->ti_ack, tp->t_rtseq))
450 					tcp_xmit_timer(tp, tp->t_rtt);
451 				acked = ti->ti_ack - tp->snd_una;
452 				tcpstat.tcps_rcvackpack++;
453 				tcpstat.tcps_rcvackbyte += acked;
454 				sbdrop(&so->so_snd, acked);
455 				tp->snd_una = ti->ti_ack;
456 				m_freem(m);
457 
458 				/*
459 				 * If all outstanding data are acked, stop
460 				 * retransmit timer, otherwise restart timer
461 				 * using current (possibly backed-off) value.
462 				 * If process is waiting for space,
463 				 * wakeup/selwakeup/signal.  If data
464 				 * are ready to send, let tcp_output
465 				 * decide between more output or persist.
466 				 */
467 				if (tp->snd_una == tp->snd_max)
468 					tp->t_timer[TCPT_REXMT] = 0;
469 				else if (tp->t_timer[TCPT_PERSIST] == 0)
470 					tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
471 
472 				if (so->so_snd.sb_flags & SB_NOTIFY)
473 					sowwakeup(so);
474 				if (so->so_snd.sb_cc)
475 					(void) tcp_output(tp);
476 				return;
477 			}
478 		} else if (ti->ti_ack == tp->snd_una &&
479 		    tp->seg_next == (struct tcpiphdr *)tp &&
480 		    ti->ti_len <= sbspace(&so->so_rcv)) {
481 			/*
482 			 * this is a pure, in-sequence data packet
483 			 * with nothing on the reassembly queue and
484 			 * we have enough buffer space to take it.
485 			 */
486 			++tcpstat.tcps_preddat;
487 			tp->rcv_nxt += ti->ti_len;
488 			tcpstat.tcps_rcvpack++;
489 			tcpstat.tcps_rcvbyte += ti->ti_len;
490 			/*
491 			 * Drop TCP, IP headers and TCP options then add data
492 			 * to socket buffer.
493 			 */
494 			m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
495 			m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
496 			sbappend(&so->so_rcv, m);
497 			sorwakeup(so);
498 			if (ti->ti_flags & TH_PUSH)
499 				tp->t_flags |= TF_ACKNOW;
500 			else
501 				tp->t_flags |= TF_DELACK;
502 			return;
503 		}
504 	}
505 
506 	/*
507 	 * Drop TCP, IP headers and TCP options.
508 	 */
509 	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
510 	m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
511 
512 	/*
513 	 * Calculate amount of space in receive window,
514 	 * and then do TCP input processing.
515 	 * Receive window is amount of space in rcv queue,
516 	 * but not less than advertised window.
517 	 */
518 	{ int win;
519 
520 	win = sbspace(&so->so_rcv);
521 	if (win < 0)
522 		win = 0;
523 	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
524 	}
525 
526 	switch (tp->t_state) {
527 
528 	/*
529 	 * If the state is LISTEN then ignore segment if it contains an RST.
530 	 * If the segment contains an ACK then it is bad and send a RST.
531 	 * If it does not contain a SYN then it is not interesting; drop it.
532 	 * Don't bother responding if the destination was a broadcast.
533 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
534 	 * tp->iss, and send a segment:
535 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
536 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
537 	 * Fill in remote peer address fields if not previously specified.
538 	 * Enter SYN_RECEIVED state, and process any other fields of this
539 	 * segment in this state.
540 	 */
541 	case TCPS_LISTEN: {
542 		struct mbuf *am;
543 		register struct sockaddr_in *sin;
544 
545 		if (tiflags & TH_RST)
546 			goto drop;
547 		if (tiflags & TH_ACK)
548 			goto dropwithreset;
549 		if ((tiflags & TH_SYN) == 0)
550 			goto drop;
551 		/*
552 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
553 		 * in_broadcast() should never return true on a received
554 		 * packet with M_BCAST not set.
555 		 */
556 		if (m->m_flags & (M_BCAST|M_MCAST) ||
557 		    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
558 			goto drop;
559 		am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
560 		if (am == NULL)
561 			goto drop;
562 		am->m_len = sizeof (struct sockaddr_in);
563 		sin = mtod(am, struct sockaddr_in *);
564 		sin->sin_family = AF_INET;
565 		sin->sin_len = sizeof(*sin);
566 		sin->sin_addr = ti->ti_src;
567 		sin->sin_port = ti->ti_sport;
568 		bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
569 		laddr = inp->inp_laddr;
570 		if (inp->inp_laddr.s_addr == INADDR_ANY)
571 			inp->inp_laddr = ti->ti_dst;
572 		if (in_pcbconnect(inp, am)) {
573 			inp->inp_laddr = laddr;
574 			(void) m_free(am);
575 			goto drop;
576 		}
577 		(void) m_free(am);
578 		tp->t_template = tcp_template(tp);
579 		if (tp->t_template == 0) {
580 			tp = tcp_drop(tp, ENOBUFS);
581 			dropsocket = 0;		/* socket is already gone */
582 			goto drop;
583 		}
584 		if (optp)
585 			tcp_dooptions(tp, optp, optlen, ti,
586 				&ts_present, &ts_val, &ts_ecr);
587 		if (iss)
588 			tp->iss = iss;
589 		else
590 			tp->iss = tcp_iss;
591 		tcp_iss += TCP_ISSINCR/2;
592 		tp->irs = ti->ti_seq;
593 		tcp_sendseqinit(tp);
594 		tcp_rcvseqinit(tp);
595 		tp->t_flags |= TF_ACKNOW;
596 		tp->t_state = TCPS_SYN_RECEIVED;
597 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
598 		dropsocket = 0;		/* committed to socket */
599 		tcpstat.tcps_accepts++;
600 		goto trimthenstep6;
601 		}
602 
603 	/*
604 	 * If the state is SYN_SENT:
605 	 *	if seg contains an ACK, but not for our SYN, drop the input.
606 	 *	if seg contains a RST, then drop the connection.
607 	 *	if seg does not contain SYN, then drop it.
608 	 * Otherwise this is an acceptable SYN segment
609 	 *	initialize tp->rcv_nxt and tp->irs
610 	 *	if seg contains ack then advance tp->snd_una
611 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
612 	 *	arrange for segment to be acked (eventually)
613 	 *	continue processing rest of data/controls, beginning with URG
614 	 */
615 	case TCPS_SYN_SENT:
616 		if ((tiflags & TH_ACK) &&
617 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
618 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
619 			goto dropwithreset;
620 		if (tiflags & TH_RST) {
621 			if (tiflags & TH_ACK)
622 				tp = tcp_drop(tp, ECONNREFUSED);
623 			goto drop;
624 		}
625 		if ((tiflags & TH_SYN) == 0)
626 			goto drop;
627 		if (tiflags & TH_ACK) {
628 			tp->snd_una = ti->ti_ack;
629 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
630 				tp->snd_nxt = tp->snd_una;
631 		}
632 		tp->t_timer[TCPT_REXMT] = 0;
633 		tp->irs = ti->ti_seq;
634 		tcp_rcvseqinit(tp);
635 		tp->t_flags |= TF_ACKNOW;
636 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
637 			tcpstat.tcps_connects++;
638 			soisconnected(so);
639 			tp->t_state = TCPS_ESTABLISHED;
640 			/* Do window scaling on this connection? */
641 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
642 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
643 				tp->snd_scale = tp->requested_s_scale;
644 				tp->rcv_scale = tp->request_r_scale;
645 			}
646 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
647 				(struct mbuf *)0);
648 			/*
649 			 * if we didn't have to retransmit the SYN,
650 			 * use its rtt as our initial srtt & rtt var.
651 			 */
652 			if (tp->t_rtt)
653 				tcp_xmit_timer(tp, tp->t_rtt);
654 		} else
655 			tp->t_state = TCPS_SYN_RECEIVED;
656 
657 trimthenstep6:
658 		/*
659 		 * Advance ti->ti_seq to correspond to first data byte.
660 		 * If data, trim to stay within window,
661 		 * dropping FIN if necessary.
662 		 */
663 		ti->ti_seq++;
664 		if (ti->ti_len > tp->rcv_wnd) {
665 			todrop = ti->ti_len - tp->rcv_wnd;
666 			m_adj(m, -todrop);
667 			ti->ti_len = tp->rcv_wnd;
668 			tiflags &= ~TH_FIN;
669 			tcpstat.tcps_rcvpackafterwin++;
670 			tcpstat.tcps_rcvbyteafterwin += todrop;
671 		}
672 		tp->snd_wl1 = ti->ti_seq - 1;
673 		tp->rcv_up = ti->ti_seq;
674 		goto step6;
675 	}
676 
677 	/*
678 	 * States other than LISTEN or SYN_SENT.
679 	 * First check timestamp, if present.
680 	 * Then check that at least some bytes of segment are within
681 	 * receive window.  If segment begins before rcv_nxt,
682 	 * drop leading data (and SYN); if nothing left, just ack.
683 	 *
684 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
685 	 * and it's less than ts_recent, drop it.
686 	 */
687 	if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
688 	    TSTMP_LT(ts_val, tp->ts_recent)) {
689 
690 		/* Check to see if ts_recent is over 24 days old.  */
691 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
692 			/*
693 			 * Invalidate ts_recent.  If this segment updates
694 			 * ts_recent, the age will be reset later and ts_recent
695 			 * will get a valid value.  If it does not, setting
696 			 * ts_recent to zero will at least satisfy the
697 			 * requirement that zero be placed in the timestamp
698 			 * echo reply when ts_recent isn't valid.  The
699 			 * age isn't reset until we get a valid ts_recent
700 			 * because we don't want out-of-order segments to be
701 			 * dropped when ts_recent is old.
702 			 */
703 			tp->ts_recent = 0;
704 		} else {
705 			tcpstat.tcps_rcvduppack++;
706 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
707 			tcpstat.tcps_pawsdrop++;
708 			goto dropafterack;
709 		}
710 	}
711 
712 	todrop = tp->rcv_nxt - ti->ti_seq;
713 	if (todrop > 0) {
714 		if (tiflags & TH_SYN) {
715 			tiflags &= ~TH_SYN;
716 			ti->ti_seq++;
717 			if (ti->ti_urp > 1)
718 				ti->ti_urp--;
719 			else
720 				tiflags &= ~TH_URG;
721 			todrop--;
722 		}
723 		if (todrop >= ti->ti_len) {
724 			/*
725 			 * Any valid FIN must be to the left of the
726 			 * window.  At this point, FIN must be a
727 			 * duplicate or out-of-sequence, so drop it.
728 			 */
729 			tiflags &= ~TH_FIN;
730 			/*
731 			 * Send ACK to resynchronize, and drop any data,
732 			 * but keep on processing for RST or ACK.
733 			 */
734 			tp->t_flags |= TF_ACKNOW;
735 			tcpstat.tcps_rcvdupbyte += todrop = ti->ti_len;
736 			tcpstat.tcps_rcvduppack++;
737 		} else {
738 			tcpstat.tcps_rcvpartduppack++;
739 			tcpstat.tcps_rcvpartdupbyte += todrop;
740 		}
741 		m_adj(m, todrop);
742 		ti->ti_seq += todrop;
743 		ti->ti_len -= todrop;
744 		if (ti->ti_urp > todrop)
745 			ti->ti_urp -= todrop;
746 		else {
747 			tiflags &= ~TH_URG;
748 			ti->ti_urp = 0;
749 		}
750 	}
751 
752 	/*
753 	 * If new data are received on a connection after the
754 	 * user processes are gone, then RST the other end.
755 	 */
756 	if ((so->so_state & SS_NOFDREF) &&
757 	    tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
758 		tp = tcp_close(tp);
759 		tcpstat.tcps_rcvafterclose++;
760 		goto dropwithreset;
761 	}
762 
763 	/*
764 	 * If segment ends after window, drop trailing data
765 	 * (and PUSH and FIN); if nothing left, just ACK.
766 	 */
767 	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
768 	if (todrop > 0) {
769 		tcpstat.tcps_rcvpackafterwin++;
770 		if (todrop >= ti->ti_len) {
771 			tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
772 			/*
773 			 * If a new connection request is received
774 			 * while in TIME_WAIT, drop the old connection
775 			 * and start over if the sequence numbers
776 			 * are above the previous ones.
777 			 */
778 			if (tiflags & TH_SYN &&
779 			    tp->t_state == TCPS_TIME_WAIT &&
780 			    SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
781 				iss = tp->rcv_nxt + TCP_ISSINCR;
782 				tp = tcp_close(tp);
783 				goto findpcb;
784 			}
785 			/*
786 			 * If window is closed can only take segments at
787 			 * window edge, and have to drop data and PUSH from
788 			 * incoming segments.  Continue processing, but
789 			 * remember to ack.  Otherwise, drop segment
790 			 * and ack.
791 			 */
792 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
793 				tp->t_flags |= TF_ACKNOW;
794 				tcpstat.tcps_rcvwinprobe++;
795 			} else
796 				goto dropafterack;
797 		} else
798 			tcpstat.tcps_rcvbyteafterwin += todrop;
799 		m_adj(m, -todrop);
800 		ti->ti_len -= todrop;
801 		tiflags &= ~(TH_PUSH|TH_FIN);
802 	}
803 
804 	/*
805 	 * If last ACK falls within this segment's sequence numbers,
806 	 * record its timestamp.
807 	 */
808 	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
809 	    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
810 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
811 		tp->ts_recent_age = tcp_now;
812 		tp->ts_recent = ts_val;
813 	}
814 
815 	/*
816 	 * If the RST bit is set examine the state:
817 	 *    SYN_RECEIVED STATE:
818 	 *	If passive open, return to LISTEN state.
819 	 *	If active open, inform user that connection was refused.
820 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
821 	 *	Inform user that connection was reset, and close tcb.
822 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
823 	 *	Close the tcb.
824 	 */
825 	if (tiflags&TH_RST) switch (tp->t_state) {
826 
827 	case TCPS_SYN_RECEIVED:
828 		so->so_error = ECONNREFUSED;
829 		goto close;
830 
831 	case TCPS_ESTABLISHED:
832 	case TCPS_FIN_WAIT_1:
833 	case TCPS_FIN_WAIT_2:
834 	case TCPS_CLOSE_WAIT:
835 		so->so_error = ECONNRESET;
836 	close:
837 		tp->t_state = TCPS_CLOSED;
838 		tcpstat.tcps_drops++;
839 		tp = tcp_close(tp);
840 		goto drop;
841 
842 	case TCPS_CLOSING:
843 	case TCPS_LAST_ACK:
844 	case TCPS_TIME_WAIT:
845 		tp = tcp_close(tp);
846 		goto drop;
847 	}
848 
849 	/*
850 	 * If a SYN is in the window, then this is an
851 	 * error and we send an RST and drop the connection.
852 	 */
853 	if (tiflags & TH_SYN) {
854 		tp = tcp_drop(tp, ECONNRESET);
855 		goto dropwithreset;
856 	}
857 
858 	/*
859 	 * If the ACK bit is off we drop the segment and return.
860 	 */
861 	if ((tiflags & TH_ACK) == 0)
862 		goto drop;
863 
864 	/*
865 	 * Ack processing.
866 	 */
867 	switch (tp->t_state) {
868 
869 	/*
870 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
871 	 * ESTABLISHED state and continue processing, otherwise
872 	 * send an RST.
873 	 */
874 	case TCPS_SYN_RECEIVED:
875 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
876 		    SEQ_GT(ti->ti_ack, tp->snd_max))
877 			goto dropwithreset;
878 		tcpstat.tcps_connects++;
879 		soisconnected(so);
880 		tp->t_state = TCPS_ESTABLISHED;
881 		/* Do window scaling? */
882 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
883 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
884 			tp->snd_scale = tp->requested_s_scale;
885 			tp->rcv_scale = tp->request_r_scale;
886 		}
887 		(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
888 		tp->snd_wl1 = ti->ti_seq - 1;
889 		/* fall into ... */
890 
891 	/*
892 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
893 	 * ACKs.  If the ack is in the range
894 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
895 	 * then advance tp->snd_una to ti->ti_ack and drop
896 	 * data from the retransmission queue.  If this ACK reflects
897 	 * more up to date window information we update our window information.
898 	 */
899 	case TCPS_ESTABLISHED:
900 	case TCPS_FIN_WAIT_1:
901 	case TCPS_FIN_WAIT_2:
902 	case TCPS_CLOSE_WAIT:
903 	case TCPS_CLOSING:
904 	case TCPS_LAST_ACK:
905 	case TCPS_TIME_WAIT:
906 
907 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
908 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
909 				tcpstat.tcps_rcvdupack++;
910 				/*
911 				 * If we have outstanding data (other than
912 				 * a window probe), this is a completely
913 				 * duplicate ack (ie, window info didn't
914 				 * change), the ack is the biggest we've
915 				 * seen and we've seen exactly our rexmt
916 				 * threshhold of them, assume a packet
917 				 * has been dropped and retransmit it.
918 				 * Kludge snd_nxt & the congestion
919 				 * window so we send only this one
920 				 * packet.
921 				 *
922 				 * We know we're losing at the current
923 				 * window size so do congestion avoidance
924 				 * (set ssthresh to half the current window
925 				 * and pull our congestion window back to
926 				 * the new ssthresh).
927 				 *
928 				 * Dup acks mean that packets have left the
929 				 * network (they're now cached at the receiver)
930 				 * so bump cwnd by the amount in the receiver
931 				 * to keep a constant cwnd packets in the
932 				 * network.
933 				 */
934 				if (tp->t_timer[TCPT_REXMT] == 0 ||
935 				    ti->ti_ack != tp->snd_una)
936 					tp->t_dupacks = 0;
937 				else if (++tp->t_dupacks == tcprexmtthresh) {
938 					tcp_seq onxt = tp->snd_nxt;
939 					u_int win =
940 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
941 						tp->t_maxseg;
942 
943 					if (win < 2)
944 						win = 2;
945 					tp->snd_ssthresh = win * tp->t_maxseg;
946 					tp->t_timer[TCPT_REXMT] = 0;
947 					tp->t_rtt = 0;
948 					tp->snd_nxt = ti->ti_ack;
949 					tp->snd_cwnd = tp->t_maxseg;
950 					(void) tcp_output(tp);
951 					tp->snd_cwnd = tp->snd_ssthresh +
952 					       tp->t_maxseg * tp->t_dupacks;
953 					if (SEQ_GT(onxt, tp->snd_nxt))
954 						tp->snd_nxt = onxt;
955 					goto drop;
956 				} else if (tp->t_dupacks > tcprexmtthresh) {
957 					tp->snd_cwnd += tp->t_maxseg;
958 					(void) tcp_output(tp);
959 					goto drop;
960 				}
961 			} else
962 				tp->t_dupacks = 0;
963 			break;
964 		}
965 		/*
966 		 * If the congestion window was inflated to account
967 		 * for the other side's cached packets, retract it.
968 		 */
969 		if (tp->t_dupacks > tcprexmtthresh &&
970 		    tp->snd_cwnd > tp->snd_ssthresh)
971 			tp->snd_cwnd = tp->snd_ssthresh;
972 		tp->t_dupacks = 0;
973 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
974 			tcpstat.tcps_rcvacktoomuch++;
975 			goto dropafterack;
976 		}
977 		acked = ti->ti_ack - tp->snd_una;
978 		tcpstat.tcps_rcvackpack++;
979 		tcpstat.tcps_rcvackbyte += acked;
980 
981 		/*
982 		 * If we have a timestamp reply, update smoothed
983 		 * round trip time.  If no timestamp is present but
984 		 * transmit timer is running and timed sequence
985 		 * number was acked, update smoothed round trip time.
986 		 * Since we now have an rtt measurement, cancel the
987 		 * timer backoff (cf., Phil Karn's retransmit alg.).
988 		 * Recompute the initial retransmit timer.
989 		 */
990 		if (ts_present)
991 			tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
992 		else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
993 			tcp_xmit_timer(tp,tp->t_rtt);
994 
995 		/*
996 		 * If all outstanding data is acked, stop retransmit
997 		 * timer and remember to restart (more output or persist).
998 		 * If there is more data to be acked, restart retransmit
999 		 * timer, using current (possibly backed-off) value.
1000 		 */
1001 		if (ti->ti_ack == tp->snd_max) {
1002 			tp->t_timer[TCPT_REXMT] = 0;
1003 			needoutput = 1;
1004 		} else if (tp->t_timer[TCPT_PERSIST] == 0)
1005 			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1006 		/*
1007 		 * When new data is acked, open the congestion window.
1008 		 * If the window gives us less than ssthresh packets
1009 		 * in flight, open exponentially (maxseg per packet).
1010 		 * Otherwise open linearly: maxseg per window
1011 		 * (maxseg^2 / cwnd per packet), plus a constant
1012 		 * fraction of a packet (maxseg/8) to help larger windows
1013 		 * open quickly enough.
1014 		 */
1015 		{
1016 		register u_int cw = tp->snd_cwnd;
1017 		register u_int incr = tp->t_maxseg;
1018 
1019 		if (cw > tp->snd_ssthresh)
1020 			incr = incr * incr / cw + incr / 8;
1021 		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1022 		}
1023 		if (acked > so->so_snd.sb_cc) {
1024 			tp->snd_wnd -= so->so_snd.sb_cc;
1025 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1026 			ourfinisacked = 1;
1027 		} else {
1028 			sbdrop(&so->so_snd, acked);
1029 			tp->snd_wnd -= acked;
1030 			ourfinisacked = 0;
1031 		}
1032 		if (so->so_snd.sb_flags & SB_NOTIFY)
1033 			sowwakeup(so);
1034 		tp->snd_una = ti->ti_ack;
1035 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1036 			tp->snd_nxt = tp->snd_una;
1037 
1038 		switch (tp->t_state) {
1039 
1040 		/*
1041 		 * In FIN_WAIT_1 STATE in addition to the processing
1042 		 * for the ESTABLISHED state if our FIN is now acknowledged
1043 		 * then enter FIN_WAIT_2.
1044 		 */
1045 		case TCPS_FIN_WAIT_1:
1046 			if (ourfinisacked) {
1047 				/*
1048 				 * If we can't receive any more
1049 				 * data, then closing user can proceed.
1050 				 * Starting the timer is contrary to the
1051 				 * specification, but if we don't get a FIN
1052 				 * we'll hang forever.
1053 				 */
1054 				if (so->so_state & SS_CANTRCVMORE) {
1055 					soisdisconnected(so);
1056 					tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1057 				}
1058 				tp->t_state = TCPS_FIN_WAIT_2;
1059 			}
1060 			break;
1061 
1062 	 	/*
1063 		 * In CLOSING STATE in addition to the processing for
1064 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1065 		 * then enter the TIME-WAIT state, otherwise ignore
1066 		 * the segment.
1067 		 */
1068 		case TCPS_CLOSING:
1069 			if (ourfinisacked) {
1070 				tp->t_state = TCPS_TIME_WAIT;
1071 				tcp_canceltimers(tp);
1072 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1073 				soisdisconnected(so);
1074 			}
1075 			break;
1076 
1077 		/*
1078 		 * In LAST_ACK, we may still be waiting for data to drain
1079 		 * and/or to be acked, as well as for the ack of our FIN.
1080 		 * If our FIN is now acknowledged, delete the TCB,
1081 		 * enter the closed state and return.
1082 		 */
1083 		case TCPS_LAST_ACK:
1084 			if (ourfinisacked) {
1085 				tp = tcp_close(tp);
1086 				goto drop;
1087 			}
1088 			break;
1089 
1090 		/*
1091 		 * In TIME_WAIT state the only thing that should arrive
1092 		 * is a retransmission of the remote FIN.  Acknowledge
1093 		 * it and restart the finack timer.
1094 		 */
1095 		case TCPS_TIME_WAIT:
1096 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1097 			goto dropafterack;
1098 		}
1099 	}
1100 
1101 step6:
1102 	/*
1103 	 * Update window information.
1104 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1105 	 */
1106 	if ((tiflags & TH_ACK) &&
1107 	    (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq &&
1108 	    (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1109 	     tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))) {
1110 		/* keep track of pure window updates */
1111 		if (ti->ti_len == 0 &&
1112 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1113 			tcpstat.tcps_rcvwinupd++;
1114 		tp->snd_wnd = tiwin;
1115 		tp->snd_wl1 = ti->ti_seq;
1116 		tp->snd_wl2 = ti->ti_ack;
1117 		if (tp->snd_wnd > tp->max_sndwnd)
1118 			tp->max_sndwnd = tp->snd_wnd;
1119 		needoutput = 1;
1120 	}
1121 
1122 	/*
1123 	 * Process segments with URG.
1124 	 */
1125 	if ((tiflags & TH_URG) && ti->ti_urp &&
1126 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1127 		/*
1128 		 * This is a kludge, but if we receive and accept
1129 		 * random urgent pointers, we'll crash in
1130 		 * soreceive.  It's hard to imagine someone
1131 		 * actually wanting to send this much urgent data.
1132 		 */
1133 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1134 			ti->ti_urp = 0;			/* XXX */
1135 			tiflags &= ~TH_URG;		/* XXX */
1136 			goto dodata;			/* XXX */
1137 		}
1138 		/*
1139 		 * If this segment advances the known urgent pointer,
1140 		 * then mark the data stream.  This should not happen
1141 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1142 		 * a FIN has been received from the remote side.
1143 		 * In these states we ignore the URG.
1144 		 *
1145 		 * According to RFC961 (Assigned Protocols),
1146 		 * the urgent pointer points to the last octet
1147 		 * of urgent data.  We continue, however,
1148 		 * to consider it to indicate the first octet
1149 		 * of data past the urgent section as the original
1150 		 * spec states (in one of two places).
1151 		 */
1152 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1153 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1154 			so->so_oobmark = so->so_rcv.sb_cc +
1155 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1156 			if (so->so_oobmark == 0)
1157 				so->so_state |= SS_RCVATMARK;
1158 			sohasoutofband(so);
1159 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1160 		}
1161 		/*
1162 		 * Remove out of band data so doesn't get presented to user.
1163 		 * This can happen independent of advancing the URG pointer,
1164 		 * but if two URG's are pending at once, some out-of-band
1165 		 * data may creep in... ick.
1166 		 */
1167 		if (ti->ti_urp <= ti->ti_len
1168 #ifdef SO_OOBINLINE
1169 		     && (so->so_options & SO_OOBINLINE) == 0
1170 #endif
1171 		     )
1172 			tcp_pulloutofband(so, ti, m);
1173 	} else
1174 		/*
1175 		 * If no out of band data is expected,
1176 		 * pull receive urgent pointer along
1177 		 * with the receive window.
1178 		 */
1179 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1180 			tp->rcv_up = tp->rcv_nxt;
1181 dodata:							/* XXX */
1182 
1183 	/*
1184 	 * Process the segment text, merging it into the TCP sequencing queue,
1185 	 * and arranging for acknowledgment of receipt if necessary.
1186 	 * This process logically involves adjusting tp->rcv_wnd as data
1187 	 * is presented to the user (this happens in tcp_usrreq.c,
1188 	 * case PRU_RCVD).  If a FIN has already been received on this
1189 	 * connection then we just ignore the text.
1190 	 */
1191 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
1192 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1193 		TCP_REASS(tp, ti, m, so, tiflags);
1194 		/*
1195 		 * Note the amount of data that peer has sent into
1196 		 * our window, in order to estimate the sender's
1197 		 * buffer size.
1198 		 */
1199 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1200 	} else {
1201 		m_freem(m);
1202 		tiflags &= ~TH_FIN;
1203 	}
1204 
1205 	/*
1206 	 * If FIN is received ACK the FIN and let the user know
1207 	 * that the connection is closing.
1208 	 */
1209 	if (tiflags & TH_FIN) {
1210 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1211 			socantrcvmore(so);
1212 			tp->t_flags |= TF_ACKNOW;
1213 			tp->rcv_nxt++;
1214 		}
1215 		switch (tp->t_state) {
1216 
1217 	 	/*
1218 		 * In SYN_RECEIVED and ESTABLISHED STATES
1219 		 * enter the CLOSE_WAIT state.
1220 		 */
1221 		case TCPS_SYN_RECEIVED:
1222 		case TCPS_ESTABLISHED:
1223 			tp->t_state = TCPS_CLOSE_WAIT;
1224 			break;
1225 
1226 	 	/*
1227 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1228 		 * enter the CLOSING state.
1229 		 */
1230 		case TCPS_FIN_WAIT_1:
1231 			tp->t_state = TCPS_CLOSING;
1232 			break;
1233 
1234 	 	/*
1235 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1236 		 * starting the time-wait timer, turning off the other
1237 		 * standard timers.
1238 		 */
1239 		case TCPS_FIN_WAIT_2:
1240 			tp->t_state = TCPS_TIME_WAIT;
1241 			tcp_canceltimers(tp);
1242 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1243 			soisdisconnected(so);
1244 			break;
1245 
1246 		/*
1247 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1248 		 */
1249 		case TCPS_TIME_WAIT:
1250 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1251 			break;
1252 		}
1253 	}
1254 	if (so->so_options & SO_DEBUG)
1255 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1256 
1257 	/*
1258 	 * Return any desired output.
1259 	 */
1260 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1261 		(void) tcp_output(tp);
1262 	return;
1263 
1264 dropafterack:
1265 	/*
1266 	 * Generate an ACK dropping incoming segment if it occupies
1267 	 * sequence space, where the ACK reflects our state.
1268 	 */
1269 	if (tiflags & TH_RST)
1270 		goto drop;
1271 	m_freem(m);
1272 	tp->t_flags |= TF_ACKNOW;
1273 	(void) tcp_output(tp);
1274 	return;
1275 
1276 dropwithreset:
1277 	/*
1278 	 * Generate a RST, dropping incoming segment.
1279 	 * Make ACK acceptable to originator of segment.
1280 	 * Don't bother to respond if destination was broadcast/multicast.
1281 	 */
1282 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1283 	    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1284 		goto drop;
1285 	if (tiflags & TH_ACK)
1286 		tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1287 	else {
1288 		if (tiflags & TH_SYN)
1289 			ti->ti_len++;
1290 		tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1291 		    TH_RST|TH_ACK);
1292 	}
1293 	/* destroy temporarily created socket */
1294 	if (dropsocket)
1295 		(void) soabort(so);
1296 	return;
1297 
1298 drop:
1299 	/*
1300 	 * Drop space held by incoming segment and return.
1301 	 */
1302 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1303 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1304 	m_freem(m);
1305 	/* destroy temporarily created socket */
1306 	if (dropsocket)
1307 		(void) soabort(so);
1308 	return;
1309 #ifndef TUBA_INCLUDE
1310 }
1311 
1312 void
1313 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr)
1314 	struct tcpcb *tp;
1315 	u_char *cp;
1316 	int cnt;
1317 	struct tcpiphdr *ti;
1318 	int *ts_present;
1319 	u_long *ts_val, *ts_ecr;
1320 {
1321 	u_short mss;
1322 	int opt, optlen;
1323 
1324 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1325 		opt = cp[0];
1326 		if (opt == TCPOPT_EOL)
1327 			break;
1328 		if (opt == TCPOPT_NOP)
1329 			optlen = 1;
1330 		else {
1331 			optlen = cp[1];
1332 			if (optlen <= 0)
1333 				break;
1334 		}
1335 		switch (opt) {
1336 
1337 		default:
1338 			continue;
1339 
1340 		case TCPOPT_MAXSEG:
1341 			if (optlen != TCPOLEN_MAXSEG)
1342 				continue;
1343 			if (!(ti->ti_flags & TH_SYN))
1344 				continue;
1345 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
1346 			NTOHS(mss);
1347 			(void) tcp_mss(tp, mss);	/* sets t_maxseg */
1348 			break;
1349 
1350 		case TCPOPT_WINDOW:
1351 			if (optlen != TCPOLEN_WINDOW)
1352 				continue;
1353 			if (!(ti->ti_flags & TH_SYN))
1354 				continue;
1355 			tp->t_flags |= TF_RCVD_SCALE;
1356 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1357 			break;
1358 
1359 		case TCPOPT_TIMESTAMP:
1360 			if (optlen != TCPOLEN_TIMESTAMP)
1361 				continue;
1362 			*ts_present = 1;
1363 			bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val));
1364 			NTOHL(*ts_val);
1365 			bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr));
1366 			NTOHL(*ts_ecr);
1367 
1368 			/*
1369 			 * A timestamp received in a SYN makes
1370 			 * it ok to send timestamp requests and replies.
1371 			 */
1372 			if (ti->ti_flags & TH_SYN) {
1373 				tp->t_flags |= TF_RCVD_TSTMP;
1374 				tp->ts_recent = *ts_val;
1375 				tp->ts_recent_age = tcp_now;
1376 			}
1377 			break;
1378 		}
1379 	}
1380 }
1381 
1382 /*
1383  * Pull out of band byte out of a segment so
1384  * it doesn't appear in the user's data queue.
1385  * It is still reflected in the segment length for
1386  * sequencing purposes.
1387  */
1388 void
1389 tcp_pulloutofband(so, ti, m)
1390 	struct socket *so;
1391 	struct tcpiphdr *ti;
1392 	register struct mbuf *m;
1393 {
1394 	int cnt = ti->ti_urp - 1;
1395 
1396 	while (cnt >= 0) {
1397 		if (m->m_len > cnt) {
1398 			char *cp = mtod(m, caddr_t) + cnt;
1399 			struct tcpcb *tp = sototcpcb(so);
1400 
1401 			tp->t_iobc = *cp;
1402 			tp->t_oobflags |= TCPOOB_HAVEDATA;
1403 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1404 			m->m_len--;
1405 			return;
1406 		}
1407 		cnt -= m->m_len;
1408 		m = m->m_next;
1409 		if (m == 0)
1410 			break;
1411 	}
1412 	panic("tcp_pulloutofband");
1413 }
1414 
1415 /*
1416  * Collect new round-trip time estimate
1417  * and update averages and current timeout.
1418  */
1419 void
1420 tcp_xmit_timer(tp, rtt)
1421 	register struct tcpcb *tp;
1422 	short rtt;
1423 {
1424 	register short delta;
1425 
1426 	tcpstat.tcps_rttupdated++;
1427 	if (tp->t_srtt != 0) {
1428 		/*
1429 		 * srtt is stored as fixed point with 3 bits after the
1430 		 * binary point (i.e., scaled by 8).  The following magic
1431 		 * is equivalent to the smoothing algorithm in rfc793 with
1432 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1433 		 * point).  Adjust rtt to origin 0.
1434 		 */
1435 		delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1436 		if ((tp->t_srtt += delta) <= 0)
1437 			tp->t_srtt = 1;
1438 		/*
1439 		 * We accumulate a smoothed rtt variance (actually, a
1440 		 * smoothed mean difference), then set the retransmit
1441 		 * timer to smoothed rtt + 4 times the smoothed variance.
1442 		 * rttvar is stored as fixed point with 2 bits after the
1443 		 * binary point (scaled by 4).  The following is
1444 		 * equivalent to rfc793 smoothing with an alpha of .75
1445 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1446 		 * rfc793's wired-in beta.
1447 		 */
1448 		if (delta < 0)
1449 			delta = -delta;
1450 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1451 		if ((tp->t_rttvar += delta) <= 0)
1452 			tp->t_rttvar = 1;
1453 	} else {
1454 		/*
1455 		 * No rtt measurement yet - use the unsmoothed rtt.
1456 		 * Set the variance to half the rtt (so our first
1457 		 * retransmit happens at 3*rtt).
1458 		 */
1459 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
1460 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1461 	}
1462 	tp->t_rtt = 0;
1463 	tp->t_rxtshift = 0;
1464 
1465 	/*
1466 	 * the retransmit should happen at rtt + 4 * rttvar.
1467 	 * Because of the way we do the smoothing, srtt and rttvar
1468 	 * will each average +1/2 tick of bias.  When we compute
1469 	 * the retransmit timer, we want 1/2 tick of rounding and
1470 	 * 1 extra tick because of +-1/2 tick uncertainty in the
1471 	 * firing of the timer.  The bias will give us exactly the
1472 	 * 1.5 tick we need.  But, because the bias is
1473 	 * statistical, we have to test that we don't drop below
1474 	 * the minimum feasible timer (which is 2 ticks).
1475 	 */
1476 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1477 	    tp->t_rttmin, TCPTV_REXMTMAX);
1478 
1479 	/*
1480 	 * We received an ack for a packet that wasn't retransmitted;
1481 	 * it is probably safe to discard any error indications we've
1482 	 * received recently.  This isn't quite right, but close enough
1483 	 * for now (a route might have failed after we sent a segment,
1484 	 * and the return path might not be symmetrical).
1485 	 */
1486 	tp->t_softerror = 0;
1487 }
1488 
1489 /*
1490  * Determine a reasonable value for maxseg size.
1491  * If the route is known, check route for mtu.
1492  * If none, use an mss that can be handled on the outgoing
1493  * interface without forcing IP to fragment; if bigger than
1494  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1495  * to utilize large mbufs.  If no route is found, route has no mtu,
1496  * or the destination isn't local, use a default, hopefully conservative
1497  * size (usually 512 or the default IP max size, but no more than the mtu
1498  * of the interface), as we can't discover anything about intervening
1499  * gateways or networks.  We also initialize the congestion/slow start
1500  * window to be a single segment if the destination isn't local.
1501  * While looking at the routing entry, we also initialize other path-dependent
1502  * parameters from pre-set or cached values in the routing entry.
1503  */
1504 int
1505 tcp_mss(tp, offer)
1506 	register struct tcpcb *tp;
1507 	u_int offer;
1508 {
1509 	struct route *ro;
1510 	register struct rtentry *rt;
1511 	struct ifnet *ifp;
1512 	register int rtt, mss;
1513 	u_long bufsize;
1514 	struct inpcb *inp;
1515 	struct socket *so;
1516 	extern int tcp_mssdflt;
1517 
1518 	inp = tp->t_inpcb;
1519 	ro = &inp->inp_route;
1520 
1521 	if ((rt = ro->ro_rt) == (struct rtentry *)0) {
1522 		/* No route yet, so try to acquire one */
1523 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1524 			ro->ro_dst.sa_family = AF_INET;
1525 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1526 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1527 				inp->inp_faddr;
1528 			rtalloc(ro);
1529 		}
1530 		if ((rt = ro->ro_rt) == (struct rtentry *)0)
1531 			return (tcp_mssdflt);
1532 	}
1533 	ifp = rt->rt_ifp;
1534 	so = inp->inp_socket;
1535 
1536 #ifdef RTV_MTU	/* if route characteristics exist ... */
1537 	/*
1538 	 * While we're here, check if there's an initial rtt
1539 	 * or rttvar.  Convert from the route-table units
1540 	 * to scaled multiples of the slow timeout timer.
1541 	 */
1542 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1543 		/*
1544 		 * XXX the lock bit for MTU indicates that the value
1545 		 * is also a minimum value; this is subject to time.
1546 		 */
1547 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1548 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
1549 		tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
1550 		if (rt->rt_rmx.rmx_rttvar)
1551 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1552 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
1553 		else
1554 			/* default variation is +- 1 rtt */
1555 			tp->t_rttvar =
1556 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1557 		TCPT_RANGESET(tp->t_rxtcur,
1558 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
1559 		    tp->t_rttmin, TCPTV_REXMTMAX);
1560 	}
1561 	/*
1562 	 * if there's an mtu associated with the route, use it
1563 	 */
1564 	if (rt->rt_rmx.rmx_mtu)
1565 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
1566 	else
1567 #endif /* RTV_MTU */
1568 	{
1569 		mss = ifp->if_mtu - sizeof(struct tcpiphdr);
1570 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1571 		if (mss > MCLBYTES)
1572 			mss &= ~(MCLBYTES-1);
1573 #else
1574 		if (mss > MCLBYTES)
1575 			mss = mss / MCLBYTES * MCLBYTES;
1576 #endif
1577 		if (!in_localaddr(inp->inp_faddr))
1578 			mss = min(mss, tcp_mssdflt);
1579 	}
1580 	/*
1581 	 * The current mss, t_maxseg, is initialized to the default value.
1582 	 * If we compute a smaller value, reduce the current mss.
1583 	 * If we compute a larger value, return it for use in sending
1584 	 * a max seg size option, but don't store it for use
1585 	 * unless we received an offer at least that large from peer.
1586 	 * However, do not accept offers under 32 bytes.
1587 	 */
1588 	if (offer)
1589 		mss = min(mss, offer);
1590 	mss = max(mss, 32);		/* sanity */
1591 	if (mss < tp->t_maxseg || offer != 0) {
1592 		/*
1593 		 * If there's a pipesize, change the socket buffer
1594 		 * to that size.  Make the socket buffers an integral
1595 		 * number of mss units; if the mss is larger than
1596 		 * the socket buffer, decrease the mss.
1597 		 */
1598 #ifdef RTV_SPIPE
1599 		if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
1600 #endif
1601 			bufsize = so->so_snd.sb_hiwat;
1602 		if (bufsize < mss)
1603 			mss = bufsize;
1604 		else {
1605 			bufsize = roundup(bufsize, mss);
1606 			if (bufsize > sb_max)
1607 				bufsize = sb_max;
1608 			(void)sbreserve(&so->so_snd, bufsize);
1609 		}
1610 		tp->t_maxseg = mss;
1611 
1612 #ifdef RTV_RPIPE
1613 		if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
1614 #endif
1615 			bufsize = so->so_rcv.sb_hiwat;
1616 		if (bufsize > mss) {
1617 			bufsize = roundup(bufsize, mss);
1618 			if (bufsize > sb_max)
1619 				bufsize = sb_max;
1620 			(void)sbreserve(&so->so_rcv, bufsize);
1621 		}
1622 	}
1623 	tp->snd_cwnd = mss;
1624 
1625 #ifdef RTV_SSTHRESH
1626 	if (rt->rt_rmx.rmx_ssthresh) {
1627 		/*
1628 		 * There's some sort of gateway or interface
1629 		 * buffer limit on the path.  Use this to set
1630 		 * the slow start threshhold, but set the
1631 		 * threshold to no less than 2*mss.
1632 		 */
1633 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1634 	}
1635 #endif /* RTV_MTU */
1636 	return (mss);
1637 }
1638 #endif /* TUBA_INCLUDE */
1639