xref: /netbsd-src/sys/netinet/tcp_input.c (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1 /*	$NetBSD: tcp_input.c,v 1.26 1996/09/15 18:11:09 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)tcp_input.c	8.5 (Berkeley) 4/10/94
36  */
37 
38 #ifndef TUBA_INCLUDE
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/errno.h>
47 
48 #include <net/if.h>
49 #include <net/route.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/ip_var.h>
56 #include <netinet/tcp.h>
57 #include <netinet/tcp_fsm.h>
58 #include <netinet/tcp_seq.h>
59 #include <netinet/tcp_timer.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/tcpip.h>
62 #include <netinet/tcp_debug.h>
63 
64 #include <machine/stdarg.h>
65 
66 int	tcprexmtthresh = 3;
67 struct	tcpiphdr tcp_saveti;
68 
69 extern u_long sb_max;
70 
71 #endif /* TUBA_INCLUDE */
72 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
73 
74 /* for modulo comparisons of timestamps */
75 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
76 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
77 
78 
79 /*
80  * Insert segment ti into reassembly queue of tcp with
81  * control block tp.  Return TH_FIN if reassembly now includes
82  * a segment with FIN.  The macro form does the common case inline
83  * (segment is the next to be received on an established connection,
84  * and the queue is empty), avoiding linkage into and removal
85  * from the queue and repetition of various conversions.
86  * Set DELACK for segments received in order, but ack immediately
87  * when segments are out of order (so fast retransmit can work).
88  */
89 #define	TCP_REASS(tp, ti, m, so, flags) { \
90 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
91 	    (tp)->segq.lh_first == NULL && \
92 	    (tp)->t_state == TCPS_ESTABLISHED) { \
93 		if ((ti)->ti_flags & TH_PUSH) \
94 			tp->t_flags |= TF_ACKNOW; \
95 		else \
96 			tp->t_flags |= TF_DELACK; \
97 		(tp)->rcv_nxt += (ti)->ti_len; \
98 		flags = (ti)->ti_flags & TH_FIN; \
99 		tcpstat.tcps_rcvpack++;\
100 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
101 		sbappend(&(so)->so_rcv, (m)); \
102 		sorwakeup(so); \
103 	} else { \
104 		(flags) = tcp_reass((tp), (ti), (m)); \
105 		tp->t_flags |= TF_ACKNOW; \
106 	} \
107 }
108 #ifndef TUBA_INCLUDE
109 
110 int
111 tcp_reass(tp, ti, m)
112 	register struct tcpcb *tp;
113 	register struct tcpiphdr *ti;
114 	struct mbuf *m;
115 {
116 	register struct ipqent *p, *q, *nq, *tiqe;
117 	struct socket *so = tp->t_inpcb->inp_socket;
118 	int flags;
119 
120 	/*
121 	 * Call with ti==0 after become established to
122 	 * force pre-ESTABLISHED data up to user socket.
123 	 */
124 	if (ti == 0)
125 		goto present;
126 
127 	/*
128 	 * Allocate a new queue entry, before we throw away any data.
129 	 * If we can't, just drop the packet.  XXX
130 	 */
131 	MALLOC(tiqe, struct ipqent *, sizeof (struct ipqent), M_IPQ, M_NOWAIT);
132 	if (tiqe == NULL) {
133 		tcpstat.tcps_rcvmemdrop++;
134 		m_freem(m);
135 		return (0);
136 	}
137 
138 	/*
139 	 * Find a segment which begins after this one does.
140 	 */
141 	for (p = NULL, q = tp->segq.lh_first; q != NULL;
142 	    p = q, q = q->ipqe_q.le_next)
143 		if (SEQ_GT(q->ipqe_tcp->ti_seq, ti->ti_seq))
144 			break;
145 
146 	/*
147 	 * If there is a preceding segment, it may provide some of
148 	 * our data already.  If so, drop the data from the incoming
149 	 * segment.  If it provides all of our data, drop us.
150 	 */
151 	if (p != NULL) {
152 		register struct tcpiphdr *phdr = p->ipqe_tcp;
153 		register int i;
154 
155 		/* conversion to int (in i) handles seq wraparound */
156 		i = phdr->ti_seq + phdr->ti_len - ti->ti_seq;
157 		if (i > 0) {
158 			if (i >= ti->ti_len) {
159 				tcpstat.tcps_rcvduppack++;
160 				tcpstat.tcps_rcvdupbyte += ti->ti_len;
161 				m_freem(m);
162 				FREE(tiqe, M_IPQ);
163 				return (0);
164 			}
165 			m_adj(m, i);
166 			ti->ti_len -= i;
167 			ti->ti_seq += i;
168 		}
169 	}
170 	tcpstat.tcps_rcvoopack++;
171 	tcpstat.tcps_rcvoobyte += ti->ti_len;
172 
173 	/*
174 	 * While we overlap succeeding segments trim them or,
175 	 * if they are completely covered, dequeue them.
176 	 */
177 	for (; q != NULL; q = nq) {
178 		register struct tcpiphdr *qhdr = q->ipqe_tcp;
179 		register int i = (ti->ti_seq + ti->ti_len) - qhdr->ti_seq;
180 
181 		if (i <= 0)
182 			break;
183 		if (i < qhdr->ti_len) {
184 			qhdr->ti_seq += i;
185 			qhdr->ti_len -= i;
186 			m_adj(q->ipqe_m, i);
187 			break;
188 		}
189 		nq = q->ipqe_q.le_next;
190 		m_freem(q->ipqe_m);
191 		LIST_REMOVE(q, ipqe_q);
192 		FREE(q, M_IPQ);
193 	}
194 
195 	/* Insert the new fragment queue entry into place. */
196 	tiqe->ipqe_m = m;
197 	tiqe->ipqe_tcp = ti;
198 	if (p == NULL) {
199 		LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
200 	} else {
201 		LIST_INSERT_AFTER(p, tiqe, ipqe_q);
202 	}
203 
204 present:
205 	/*
206 	 * Present data to user, advancing rcv_nxt through
207 	 * completed sequence space.
208 	 */
209 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
210 		return (0);
211 	q = tp->segq.lh_first;
212 	if (q == NULL || q->ipqe_tcp->ti_seq != tp->rcv_nxt)
213 		return (0);
214 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->ti_len)
215 		return (0);
216 	do {
217 		tp->rcv_nxt += q->ipqe_tcp->ti_len;
218 		flags = q->ipqe_tcp->ti_flags & TH_FIN;
219 
220 		nq = q->ipqe_q.le_next;
221 		LIST_REMOVE(q, ipqe_q);
222 		if (so->so_state & SS_CANTRCVMORE)
223 			m_freem(q->ipqe_m);
224 		else
225 			sbappend(&so->so_rcv, q->ipqe_m);
226 		FREE(q, M_IPQ);
227 		q = nq;
228 	} while (q != NULL && q->ipqe_tcp->ti_seq == tp->rcv_nxt);
229 	sorwakeup(so);
230 	return (flags);
231 }
232 
233 /*
234  * TCP input routine, follows pages 65-76 of the
235  * protocol specification dated September, 1981 very closely.
236  */
237 void
238 #if __STDC__
239 tcp_input(struct mbuf *m, ...)
240 #else
241 tcp_input(m, va_alist)
242 	register struct mbuf *m;
243 #endif
244 {
245 	register struct tcpiphdr *ti;
246 	register struct inpcb *inp;
247 	caddr_t optp = NULL;
248 	int optlen = 0;
249 	int len, tlen, off;
250 	register struct tcpcb *tp = 0;
251 	register int tiflags;
252 	struct socket *so = NULL;
253 	int todrop, acked, ourfinisacked, needoutput = 0;
254 	short ostate = 0;
255 	struct in_addr laddr;
256 	int dropsocket = 0;
257 	int iss = 0;
258 	u_long tiwin;
259 	u_int32_t ts_val, ts_ecr;
260 	int ts_present = 0;
261 	int iphlen;
262 	va_list ap;
263 
264 	va_start(ap, m);
265 	iphlen = va_arg(ap, int);
266 	va_end(ap);
267 
268 	tcpstat.tcps_rcvtotal++;
269 	/*
270 	 * Get IP and TCP header together in first mbuf.
271 	 * Note: IP leaves IP header in first mbuf.
272 	 */
273 	ti = mtod(m, struct tcpiphdr *);
274 	if (iphlen > sizeof (struct ip))
275 		ip_stripoptions(m, (struct mbuf *)0);
276 	if (m->m_len < sizeof (struct tcpiphdr)) {
277 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
278 			tcpstat.tcps_rcvshort++;
279 			return;
280 		}
281 		ti = mtod(m, struct tcpiphdr *);
282 	}
283 
284 	/*
285 	 * Checksum extended TCP header and data.
286 	 */
287 	tlen = ((struct ip *)ti)->ip_len;
288 	len = sizeof (struct ip) + tlen;
289 	bzero(ti->ti_x1, sizeof ti->ti_x1);
290 	ti->ti_len = (u_int16_t)tlen;
291 	HTONS(ti->ti_len);
292 	if ((ti->ti_sum = in_cksum(m, len)) != 0) {
293 		tcpstat.tcps_rcvbadsum++;
294 		goto drop;
295 	}
296 #endif /* TUBA_INCLUDE */
297 
298 	/*
299 	 * Check that TCP offset makes sense,
300 	 * pull out TCP options and adjust length.		XXX
301 	 */
302 	off = ti->ti_off << 2;
303 	if (off < sizeof (struct tcphdr) || off > tlen) {
304 		tcpstat.tcps_rcvbadoff++;
305 		goto drop;
306 	}
307 	tlen -= off;
308 	ti->ti_len = tlen;
309 	if (off > sizeof (struct tcphdr)) {
310 		if (m->m_len < sizeof(struct ip) + off) {
311 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
312 				tcpstat.tcps_rcvshort++;
313 				return;
314 			}
315 			ti = mtod(m, struct tcpiphdr *);
316 		}
317 		optlen = off - sizeof (struct tcphdr);
318 		optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
319 		/*
320 		 * Do quick retrieval of timestamp options ("options
321 		 * prediction?").  If timestamp is the only option and it's
322 		 * formatted as recommended in RFC 1323 appendix A, we
323 		 * quickly get the values now and not bother calling
324 		 * tcp_dooptions(), etc.
325 		 */
326 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
327 		     (optlen > TCPOLEN_TSTAMP_APPA &&
328 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
329 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
330 		     (ti->ti_flags & TH_SYN) == 0) {
331 			ts_present = 1;
332 			ts_val = ntohl(*(u_int32_t *)(optp + 4));
333 			ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
334 			optp = NULL;	/* we've parsed the options */
335 		}
336 	}
337 	tiflags = ti->ti_flags;
338 
339 	/*
340 	 * Convert TCP protocol specific fields to host format.
341 	 */
342 	NTOHL(ti->ti_seq);
343 	NTOHL(ti->ti_ack);
344 	NTOHS(ti->ti_win);
345 	NTOHS(ti->ti_urp);
346 
347 	/*
348 	 * Locate pcb for segment.
349 	 */
350 findpcb:
351 	inp = in_pcblookup_connect(&tcbtable, ti->ti_src, ti->ti_sport,
352 	    ti->ti_dst, ti->ti_dport);
353 	if (inp == 0) {
354 		++tcpstat.tcps_pcbhashmiss;
355 		inp = in_pcblookup_bind(&tcbtable, ti->ti_dst, ti->ti_dport);
356 		if (inp == 0) {
357 			++tcpstat.tcps_noport;
358 			goto dropwithreset;
359 		}
360 	}
361 
362 	/*
363 	 * If the state is CLOSED (i.e., TCB does not exist) then
364 	 * all data in the incoming segment is discarded.
365 	 * If the TCB exists but is in CLOSED state, it is embryonic,
366 	 * but should either do a listen or a connect soon.
367 	 */
368 	tp = intotcpcb(inp);
369 	if (tp == 0)
370 		goto dropwithreset;
371 	if (tp->t_state == TCPS_CLOSED)
372 		goto drop;
373 
374 	/* Unscale the window into a 32-bit value. */
375 	if ((tiflags & TH_SYN) == 0)
376 		tiwin = ti->ti_win << tp->snd_scale;
377 	else
378 		tiwin = ti->ti_win;
379 
380 	so = inp->inp_socket;
381 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
382 		if (so->so_options & SO_DEBUG) {
383 			ostate = tp->t_state;
384 			tcp_saveti = *ti;
385 		}
386 		if (so->so_options & SO_ACCEPTCONN) {
387 			so = sonewconn(so, 0);
388 			if (so == 0)
389 				goto drop;
390 			/*
391 			 * This is ugly, but ....
392 			 *
393 			 * Mark socket as temporary until we're
394 			 * committed to keeping it.  The code at
395 			 * ``drop'' and ``dropwithreset'' check the
396 			 * flag dropsocket to see if the temporary
397 			 * socket created here should be discarded.
398 			 * We mark the socket as discardable until
399 			 * we're committed to it below in TCPS_LISTEN.
400 			 */
401 			dropsocket++;
402 			inp = (struct inpcb *)so->so_pcb;
403 			inp->inp_laddr = ti->ti_dst;
404 			inp->inp_lport = ti->ti_dport;
405 			in_pcbstate(inp, INP_BOUND);
406 #if BSD>=43
407 			inp->inp_options = ip_srcroute();
408 #endif
409 			tp = intotcpcb(inp);
410 			tp->t_state = TCPS_LISTEN;
411 
412 			/* Compute proper scaling value from buffer space
413 			 */
414 			while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
415 			   TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
416 				tp->request_r_scale++;
417 		}
418 	}
419 
420 	/*
421 	 * Segment received on connection.
422 	 * Reset idle time and keep-alive timer.
423 	 */
424 	tp->t_idle = 0;
425 	if (TCPS_HAVEESTABLISHED(tp->t_state))
426 		tp->t_timer[TCPT_KEEP] = tcp_keepidle;
427 
428 	/*
429 	 * Process options if not in LISTEN state,
430 	 * else do it below (after getting remote address).
431 	 */
432 	if (optp && tp->t_state != TCPS_LISTEN)
433 		tcp_dooptions(tp, optp, optlen, ti,
434 			&ts_present, &ts_val, &ts_ecr);
435 
436 	/*
437 	 * Header prediction: check for the two common cases
438 	 * of a uni-directional data xfer.  If the packet has
439 	 * no control flags, is in-sequence, the window didn't
440 	 * change and we're not retransmitting, it's a
441 	 * candidate.  If the length is zero and the ack moved
442 	 * forward, we're the sender side of the xfer.  Just
443 	 * free the data acked & wake any higher level process
444 	 * that was blocked waiting for space.  If the length
445 	 * is non-zero and the ack didn't move, we're the
446 	 * receiver side.  If we're getting packets in-order
447 	 * (the reassembly queue is empty), add the data to
448 	 * the socket buffer and note that we need a delayed ack.
449 	 */
450 	if (tp->t_state == TCPS_ESTABLISHED &&
451 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
452 	    (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) &&
453 	    ti->ti_seq == tp->rcv_nxt &&
454 	    tiwin && tiwin == tp->snd_wnd &&
455 	    tp->snd_nxt == tp->snd_max) {
456 
457 		/*
458 		 * If last ACK falls within this segment's sequence numbers,
459 		 *  record the timestamp.
460 		 */
461 		if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
462 		   SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
463 			tp->ts_recent_age = tcp_now;
464 			tp->ts_recent = ts_val;
465 		}
466 
467 		if (ti->ti_len == 0) {
468 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
469 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
470 			    tp->snd_cwnd >= tp->snd_wnd &&
471 			    tp->t_dupacks < tcprexmtthresh) {
472 				/*
473 				 * this is a pure ack for outstanding data.
474 				 */
475 				++tcpstat.tcps_predack;
476 				if (ts_present)
477 					tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
478 				else if (tp->t_rtt &&
479 					    SEQ_GT(ti->ti_ack, tp->t_rtseq))
480 					tcp_xmit_timer(tp, tp->t_rtt);
481 				acked = ti->ti_ack - tp->snd_una;
482 				tcpstat.tcps_rcvackpack++;
483 				tcpstat.tcps_rcvackbyte += acked;
484 				sbdrop(&so->so_snd, acked);
485 				tp->snd_una = ti->ti_ack;
486 				m_freem(m);
487 
488 				/*
489 				 * If all outstanding data are acked, stop
490 				 * retransmit timer, otherwise restart timer
491 				 * using current (possibly backed-off) value.
492 				 * If process is waiting for space,
493 				 * wakeup/selwakeup/signal.  If data
494 				 * are ready to send, let tcp_output
495 				 * decide between more output or persist.
496 				 */
497 				if (tp->snd_una == tp->snd_max)
498 					tp->t_timer[TCPT_REXMT] = 0;
499 				else if (tp->t_timer[TCPT_PERSIST] == 0)
500 					tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
501 
502 				if (sb_notify(&so->so_snd))
503 					sowwakeup(so);
504 				if (so->so_snd.sb_cc)
505 					(void) tcp_output(tp);
506 				return;
507 			}
508 		} else if (ti->ti_ack == tp->snd_una &&
509 		    tp->segq.lh_first == NULL &&
510 		    ti->ti_len <= sbspace(&so->so_rcv)) {
511 			/*
512 			 * this is a pure, in-sequence data packet
513 			 * with nothing on the reassembly queue and
514 			 * we have enough buffer space to take it.
515 			 */
516 			++tcpstat.tcps_preddat;
517 			tp->rcv_nxt += ti->ti_len;
518 			tcpstat.tcps_rcvpack++;
519 			tcpstat.tcps_rcvbyte += ti->ti_len;
520 			/*
521 			 * Drop TCP, IP headers and TCP options then add data
522 			 * to socket buffer.
523 			 */
524 			m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
525 			m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
526 			sbappend(&so->so_rcv, m);
527 			sorwakeup(so);
528 			if (ti->ti_flags & TH_PUSH)
529 				tp->t_flags |= TF_ACKNOW;
530 			else
531 				tp->t_flags |= TF_DELACK;
532 			return;
533 		}
534 	}
535 
536 	/*
537 	 * Drop TCP, IP headers and TCP options.
538 	 */
539 	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
540 	m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
541 
542 	/*
543 	 * Calculate amount of space in receive window,
544 	 * and then do TCP input processing.
545 	 * Receive window is amount of space in rcv queue,
546 	 * but not less than advertised window.
547 	 */
548 	{ int win;
549 
550 	win = sbspace(&so->so_rcv);
551 	if (win < 0)
552 		win = 0;
553 	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
554 	}
555 
556 	switch (tp->t_state) {
557 
558 	/*
559 	 * If the state is LISTEN then ignore segment if it contains an RST.
560 	 * If the segment contains an ACK then it is bad and send a RST.
561 	 * If it does not contain a SYN then it is not interesting; drop it.
562 	 * Don't bother responding if the destination was a broadcast.
563 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
564 	 * tp->iss, and send a segment:
565 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
566 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
567 	 * Fill in remote peer address fields if not previously specified.
568 	 * Enter SYN_RECEIVED state, and process any other fields of this
569 	 * segment in this state.
570 	 */
571 	case TCPS_LISTEN: {
572 		struct mbuf *am;
573 		register struct sockaddr_in *sin;
574 
575 		if (tiflags & TH_RST)
576 			goto drop;
577 		if (tiflags & TH_ACK)
578 			goto dropwithreset;
579 		if ((tiflags & TH_SYN) == 0)
580 			goto drop;
581 		/*
582 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
583 		 * in_broadcast() should never return true on a received
584 		 * packet with M_BCAST not set.
585 		 */
586 		if (m->m_flags & (M_BCAST|M_MCAST) ||
587 		    IN_MULTICAST(ti->ti_dst.s_addr))
588 			goto drop;
589 		am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
590 		if (am == NULL)
591 			goto drop;
592 		am->m_len = sizeof (struct sockaddr_in);
593 		sin = mtod(am, struct sockaddr_in *);
594 		sin->sin_family = AF_INET;
595 		sin->sin_len = sizeof(*sin);
596 		sin->sin_addr = ti->ti_src;
597 		sin->sin_port = ti->ti_sport;
598 		bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
599 		laddr = inp->inp_laddr;
600 		if (in_nullhost(laddr))
601 			inp->inp_laddr = ti->ti_dst;
602 		if (in_pcbconnect(inp, am)) {
603 			inp->inp_laddr = laddr;
604 			(void) m_free(am);
605 			goto drop;
606 		}
607 		(void) m_free(am);
608 		tp->t_template = tcp_template(tp);
609 		if (tp->t_template == 0) {
610 			tp = tcp_drop(tp, ENOBUFS);
611 			dropsocket = 0;		/* socket is already gone */
612 			goto drop;
613 		}
614 		if (optp)
615 			tcp_dooptions(tp, optp, optlen, ti,
616 				&ts_present, &ts_val, &ts_ecr);
617 		if (iss)
618 			tp->iss = iss;
619 		else
620 			tp->iss = tcp_iss;
621 		tcp_iss += TCP_ISSINCR/2;
622 		tp->irs = ti->ti_seq;
623 		tcp_sendseqinit(tp);
624 		tcp_rcvseqinit(tp);
625 		tp->t_flags |= TF_ACKNOW;
626 		tp->t_state = TCPS_SYN_RECEIVED;
627 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
628 		dropsocket = 0;		/* committed to socket */
629 		tcpstat.tcps_accepts++;
630 		goto trimthenstep6;
631 		}
632 
633 	/*
634 	 * If the state is SYN_SENT:
635 	 *	if seg contains an ACK, but not for our SYN, drop the input.
636 	 *	if seg contains a RST, then drop the connection.
637 	 *	if seg does not contain SYN, then drop it.
638 	 * Otherwise this is an acceptable SYN segment
639 	 *	initialize tp->rcv_nxt and tp->irs
640 	 *	if seg contains ack then advance tp->snd_una
641 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
642 	 *	arrange for segment to be acked (eventually)
643 	 *	continue processing rest of data/controls, beginning with URG
644 	 */
645 	case TCPS_SYN_SENT:
646 		if ((tiflags & TH_ACK) &&
647 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
648 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
649 			goto dropwithreset;
650 		if (tiflags & TH_RST) {
651 			if (tiflags & TH_ACK)
652 				tp = tcp_drop(tp, ECONNREFUSED);
653 			goto drop;
654 		}
655 		if ((tiflags & TH_SYN) == 0)
656 			goto drop;
657 		if (tiflags & TH_ACK) {
658 			tp->snd_una = ti->ti_ack;
659 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
660 				tp->snd_nxt = tp->snd_una;
661 		}
662 		tp->t_timer[TCPT_REXMT] = 0;
663 		tp->irs = ti->ti_seq;
664 		tcp_rcvseqinit(tp);
665 		tp->t_flags |= TF_ACKNOW;
666 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
667 			tcpstat.tcps_connects++;
668 			soisconnected(so);
669 			tp->t_state = TCPS_ESTABLISHED;
670 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
671 			/* Do window scaling on this connection? */
672 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
673 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
674 				tp->snd_scale = tp->requested_s_scale;
675 				tp->rcv_scale = tp->request_r_scale;
676 			}
677 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
678 				(struct mbuf *)0);
679 			/*
680 			 * if we didn't have to retransmit the SYN,
681 			 * use its rtt as our initial srtt & rtt var.
682 			 */
683 			if (tp->t_rtt)
684 				tcp_xmit_timer(tp, tp->t_rtt);
685 		} else
686 			tp->t_state = TCPS_SYN_RECEIVED;
687 
688 trimthenstep6:
689 		/*
690 		 * Advance ti->ti_seq to correspond to first data byte.
691 		 * If data, trim to stay within window,
692 		 * dropping FIN if necessary.
693 		 */
694 		ti->ti_seq++;
695 		if (ti->ti_len > tp->rcv_wnd) {
696 			todrop = ti->ti_len - tp->rcv_wnd;
697 			m_adj(m, -todrop);
698 			ti->ti_len = tp->rcv_wnd;
699 			tiflags &= ~TH_FIN;
700 			tcpstat.tcps_rcvpackafterwin++;
701 			tcpstat.tcps_rcvbyteafterwin += todrop;
702 		}
703 		tp->snd_wl1 = ti->ti_seq - 1;
704 		tp->rcv_up = ti->ti_seq;
705 		goto step6;
706 	}
707 
708 	/*
709 	 * States other than LISTEN or SYN_SENT.
710 	 * First check timestamp, if present.
711 	 * Then check that at least some bytes of segment are within
712 	 * receive window.  If segment begins before rcv_nxt,
713 	 * drop leading data (and SYN); if nothing left, just ack.
714 	 *
715 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
716 	 * and it's less than ts_recent, drop it.
717 	 */
718 	if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
719 	    TSTMP_LT(ts_val, tp->ts_recent)) {
720 
721 		/* Check to see if ts_recent is over 24 days old.  */
722 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
723 			/*
724 			 * Invalidate ts_recent.  If this segment updates
725 			 * ts_recent, the age will be reset later and ts_recent
726 			 * will get a valid value.  If it does not, setting
727 			 * ts_recent to zero will at least satisfy the
728 			 * requirement that zero be placed in the timestamp
729 			 * echo reply when ts_recent isn't valid.  The
730 			 * age isn't reset until we get a valid ts_recent
731 			 * because we don't want out-of-order segments to be
732 			 * dropped when ts_recent is old.
733 			 */
734 			tp->ts_recent = 0;
735 		} else {
736 			tcpstat.tcps_rcvduppack++;
737 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
738 			tcpstat.tcps_pawsdrop++;
739 			goto dropafterack;
740 		}
741 	}
742 
743 	todrop = tp->rcv_nxt - ti->ti_seq;
744 	if (todrop > 0) {
745 		if (tiflags & TH_SYN) {
746 			tiflags &= ~TH_SYN;
747 			ti->ti_seq++;
748 			if (ti->ti_urp > 1)
749 				ti->ti_urp--;
750 			else {
751 				tiflags &= ~TH_URG;
752 				ti->ti_urp = 0;
753 			}
754 			todrop--;
755 		}
756 		if (todrop >= ti->ti_len) {
757 			/*
758 			 * Any valid FIN must be to the left of the
759 			 * window.  At this point, FIN must be a
760 			 * duplicate or out-of-sequence, so drop it.
761 			 */
762 			tiflags &= ~TH_FIN;
763 			/*
764 			 * Send ACK to resynchronize, and drop any data,
765 			 * but keep on processing for RST or ACK.
766 			 */
767 			tp->t_flags |= TF_ACKNOW;
768 			tcpstat.tcps_rcvdupbyte += todrop = ti->ti_len;
769 			tcpstat.tcps_rcvduppack++;
770 		} else {
771 			tcpstat.tcps_rcvpartduppack++;
772 			tcpstat.tcps_rcvpartdupbyte += todrop;
773 		}
774 		m_adj(m, todrop);
775 		ti->ti_seq += todrop;
776 		ti->ti_len -= todrop;
777 		if (ti->ti_urp > todrop)
778 			ti->ti_urp -= todrop;
779 		else {
780 			tiflags &= ~TH_URG;
781 			ti->ti_urp = 0;
782 		}
783 	}
784 
785 	/*
786 	 * If new data are received on a connection after the
787 	 * user processes are gone, then RST the other end.
788 	 */
789 	if ((so->so_state & SS_NOFDREF) &&
790 	    tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
791 		tp = tcp_close(tp);
792 		tcpstat.tcps_rcvafterclose++;
793 		goto dropwithreset;
794 	}
795 
796 	/*
797 	 * If segment ends after window, drop trailing data
798 	 * (and PUSH and FIN); if nothing left, just ACK.
799 	 */
800 	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
801 	if (todrop > 0) {
802 		tcpstat.tcps_rcvpackafterwin++;
803 		if (todrop >= ti->ti_len) {
804 			tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
805 			/*
806 			 * If a new connection request is received
807 			 * while in TIME_WAIT, drop the old connection
808 			 * and start over if the sequence numbers
809 			 * are above the previous ones.
810 			 */
811 			if (tiflags & TH_SYN &&
812 			    tp->t_state == TCPS_TIME_WAIT &&
813 			    SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
814 				iss = tp->rcv_nxt + TCP_ISSINCR;
815 				tp = tcp_close(tp);
816 				goto findpcb;
817 			}
818 			/*
819 			 * If window is closed can only take segments at
820 			 * window edge, and have to drop data and PUSH from
821 			 * incoming segments.  Continue processing, but
822 			 * remember to ack.  Otherwise, drop segment
823 			 * and ack.
824 			 */
825 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
826 				tp->t_flags |= TF_ACKNOW;
827 				tcpstat.tcps_rcvwinprobe++;
828 			} else
829 				goto dropafterack;
830 		} else
831 			tcpstat.tcps_rcvbyteafterwin += todrop;
832 		m_adj(m, -todrop);
833 		ti->ti_len -= todrop;
834 		tiflags &= ~(TH_PUSH|TH_FIN);
835 	}
836 
837 	/*
838 	 * If last ACK falls within this segment's sequence numbers,
839 	 * record its timestamp.
840 	 */
841 	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
842 	    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
843 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
844 		tp->ts_recent_age = tcp_now;
845 		tp->ts_recent = ts_val;
846 	}
847 
848 	/*
849 	 * If the RST bit is set examine the state:
850 	 *    SYN_RECEIVED STATE:
851 	 *	If passive open, return to LISTEN state.
852 	 *	If active open, inform user that connection was refused.
853 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
854 	 *	Inform user that connection was reset, and close tcb.
855 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
856 	 *	Close the tcb.
857 	 */
858 	if (tiflags&TH_RST) switch (tp->t_state) {
859 
860 	case TCPS_SYN_RECEIVED:
861 		so->so_error = ECONNREFUSED;
862 		goto close;
863 
864 	case TCPS_ESTABLISHED:
865 	case TCPS_FIN_WAIT_1:
866 	case TCPS_FIN_WAIT_2:
867 	case TCPS_CLOSE_WAIT:
868 		so->so_error = ECONNRESET;
869 	close:
870 		tp->t_state = TCPS_CLOSED;
871 		tcpstat.tcps_drops++;
872 		tp = tcp_close(tp);
873 		goto drop;
874 
875 	case TCPS_CLOSING:
876 	case TCPS_LAST_ACK:
877 	case TCPS_TIME_WAIT:
878 		tp = tcp_close(tp);
879 		goto drop;
880 	}
881 
882 	/*
883 	 * If a SYN is in the window, then this is an
884 	 * error and we send an RST and drop the connection.
885 	 */
886 	if (tiflags & TH_SYN) {
887 		tp = tcp_drop(tp, ECONNRESET);
888 		goto dropwithreset;
889 	}
890 
891 	/*
892 	 * If the ACK bit is off we drop the segment and return.
893 	 */
894 	if ((tiflags & TH_ACK) == 0)
895 		goto drop;
896 
897 	/*
898 	 * Ack processing.
899 	 */
900 	switch (tp->t_state) {
901 
902 	/*
903 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
904 	 * ESTABLISHED state and continue processing, otherwise
905 	 * send an RST.
906 	 */
907 	case TCPS_SYN_RECEIVED:
908 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
909 		    SEQ_GT(ti->ti_ack, tp->snd_max))
910 			goto dropwithreset;
911 		tcpstat.tcps_connects++;
912 		soisconnected(so);
913 		tp->t_state = TCPS_ESTABLISHED;
914 		tp->t_timer[TCPT_KEEP] = tcp_keepidle;
915 		/* Do window scaling? */
916 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
917 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
918 			tp->snd_scale = tp->requested_s_scale;
919 			tp->rcv_scale = tp->request_r_scale;
920 		}
921 		(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
922 		tp->snd_wl1 = ti->ti_seq - 1;
923 		/* fall into ... */
924 
925 	/*
926 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
927 	 * ACKs.  If the ack is in the range
928 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
929 	 * then advance tp->snd_una to ti->ti_ack and drop
930 	 * data from the retransmission queue.  If this ACK reflects
931 	 * more up to date window information we update our window information.
932 	 */
933 	case TCPS_ESTABLISHED:
934 	case TCPS_FIN_WAIT_1:
935 	case TCPS_FIN_WAIT_2:
936 	case TCPS_CLOSE_WAIT:
937 	case TCPS_CLOSING:
938 	case TCPS_LAST_ACK:
939 	case TCPS_TIME_WAIT:
940 
941 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
942 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
943 				tcpstat.tcps_rcvdupack++;
944 				/*
945 				 * If we have outstanding data (other than
946 				 * a window probe), this is a completely
947 				 * duplicate ack (ie, window info didn't
948 				 * change), the ack is the biggest we've
949 				 * seen and we've seen exactly our rexmt
950 				 * threshhold of them, assume a packet
951 				 * has been dropped and retransmit it.
952 				 * Kludge snd_nxt & the congestion
953 				 * window so we send only this one
954 				 * packet.
955 				 *
956 				 * We know we're losing at the current
957 				 * window size so do congestion avoidance
958 				 * (set ssthresh to half the current window
959 				 * and pull our congestion window back to
960 				 * the new ssthresh).
961 				 *
962 				 * Dup acks mean that packets have left the
963 				 * network (they're now cached at the receiver)
964 				 * so bump cwnd by the amount in the receiver
965 				 * to keep a constant cwnd packets in the
966 				 * network.
967 				 */
968 				if (tp->t_timer[TCPT_REXMT] == 0 ||
969 				    ti->ti_ack != tp->snd_una)
970 					tp->t_dupacks = 0;
971 				else if (++tp->t_dupacks == tcprexmtthresh) {
972 					tcp_seq onxt = tp->snd_nxt;
973 					u_int win =
974 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
975 						tp->t_maxseg;
976 
977 					if (win < 2)
978 						win = 2;
979 					tp->snd_ssthresh = win * tp->t_maxseg;
980 					tp->t_timer[TCPT_REXMT] = 0;
981 					tp->t_rtt = 0;
982 					tp->snd_nxt = ti->ti_ack;
983 					tp->snd_cwnd = tp->t_maxseg;
984 					(void) tcp_output(tp);
985 					tp->snd_cwnd = tp->snd_ssthresh +
986 					       tp->t_maxseg * tp->t_dupacks;
987 					if (SEQ_GT(onxt, tp->snd_nxt))
988 						tp->snd_nxt = onxt;
989 					goto drop;
990 				} else if (tp->t_dupacks > tcprexmtthresh) {
991 					tp->snd_cwnd += tp->t_maxseg;
992 					(void) tcp_output(tp);
993 					goto drop;
994 				}
995 			} else
996 				tp->t_dupacks = 0;
997 			break;
998 		}
999 		/*
1000 		 * If the congestion window was inflated to account
1001 		 * for the other side's cached packets, retract it.
1002 		 */
1003 		if (tp->t_dupacks >= tcprexmtthresh &&
1004 		    tp->snd_cwnd > tp->snd_ssthresh)
1005 			tp->snd_cwnd = tp->snd_ssthresh;
1006 		tp->t_dupacks = 0;
1007 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1008 			tcpstat.tcps_rcvacktoomuch++;
1009 			goto dropafterack;
1010 		}
1011 		acked = ti->ti_ack - tp->snd_una;
1012 		tcpstat.tcps_rcvackpack++;
1013 		tcpstat.tcps_rcvackbyte += acked;
1014 
1015 		/*
1016 		 * If we have a timestamp reply, update smoothed
1017 		 * round trip time.  If no timestamp is present but
1018 		 * transmit timer is running and timed sequence
1019 		 * number was acked, update smoothed round trip time.
1020 		 * Since we now have an rtt measurement, cancel the
1021 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1022 		 * Recompute the initial retransmit timer.
1023 		 */
1024 		if (ts_present)
1025 			tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1026 		else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1027 			tcp_xmit_timer(tp,tp->t_rtt);
1028 
1029 		/*
1030 		 * If all outstanding data is acked, stop retransmit
1031 		 * timer and remember to restart (more output or persist).
1032 		 * If there is more data to be acked, restart retransmit
1033 		 * timer, using current (possibly backed-off) value.
1034 		 */
1035 		if (ti->ti_ack == tp->snd_max) {
1036 			tp->t_timer[TCPT_REXMT] = 0;
1037 			needoutput = 1;
1038 		} else if (tp->t_timer[TCPT_PERSIST] == 0)
1039 			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1040 		/*
1041 		 * When new data is acked, open the congestion window.
1042 		 * If the window gives us less than ssthresh packets
1043 		 * in flight, open exponentially (maxseg per packet).
1044 		 * Otherwise open linearly: maxseg per window
1045 		 * (maxseg^2 / cwnd per packet), plus a constant
1046 		 * fraction of a packet (maxseg/8) to help larger windows
1047 		 * open quickly enough.
1048 		 */
1049 		{
1050 		register u_int cw = tp->snd_cwnd;
1051 		register u_int incr = tp->t_maxseg;
1052 
1053 		if (cw > tp->snd_ssthresh)
1054 			incr = incr * incr / cw;
1055 		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1056 		}
1057 		if (acked > so->so_snd.sb_cc) {
1058 			tp->snd_wnd -= so->so_snd.sb_cc;
1059 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1060 			ourfinisacked = 1;
1061 		} else {
1062 			sbdrop(&so->so_snd, acked);
1063 			tp->snd_wnd -= acked;
1064 			ourfinisacked = 0;
1065 		}
1066 		if (sb_notify(&so->so_snd))
1067 			sowwakeup(so);
1068 		tp->snd_una = ti->ti_ack;
1069 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1070 			tp->snd_nxt = tp->snd_una;
1071 
1072 		switch (tp->t_state) {
1073 
1074 		/*
1075 		 * In FIN_WAIT_1 STATE in addition to the processing
1076 		 * for the ESTABLISHED state if our FIN is now acknowledged
1077 		 * then enter FIN_WAIT_2.
1078 		 */
1079 		case TCPS_FIN_WAIT_1:
1080 			if (ourfinisacked) {
1081 				/*
1082 				 * If we can't receive any more
1083 				 * data, then closing user can proceed.
1084 				 * Starting the timer is contrary to the
1085 				 * specification, but if we don't get a FIN
1086 				 * we'll hang forever.
1087 				 */
1088 				if (so->so_state & SS_CANTRCVMORE) {
1089 					soisdisconnected(so);
1090 					tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1091 				}
1092 				tp->t_state = TCPS_FIN_WAIT_2;
1093 			}
1094 			break;
1095 
1096 	 	/*
1097 		 * In CLOSING STATE in addition to the processing for
1098 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1099 		 * then enter the TIME-WAIT state, otherwise ignore
1100 		 * the segment.
1101 		 */
1102 		case TCPS_CLOSING:
1103 			if (ourfinisacked) {
1104 				tp->t_state = TCPS_TIME_WAIT;
1105 				tcp_canceltimers(tp);
1106 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1107 				soisdisconnected(so);
1108 			}
1109 			break;
1110 
1111 		/*
1112 		 * In LAST_ACK, we may still be waiting for data to drain
1113 		 * and/or to be acked, as well as for the ack of our FIN.
1114 		 * If our FIN is now acknowledged, delete the TCB,
1115 		 * enter the closed state and return.
1116 		 */
1117 		case TCPS_LAST_ACK:
1118 			if (ourfinisacked) {
1119 				tp = tcp_close(tp);
1120 				goto drop;
1121 			}
1122 			break;
1123 
1124 		/*
1125 		 * In TIME_WAIT state the only thing that should arrive
1126 		 * is a retransmission of the remote FIN.  Acknowledge
1127 		 * it and restart the finack timer.
1128 		 */
1129 		case TCPS_TIME_WAIT:
1130 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1131 			goto dropafterack;
1132 		}
1133 	}
1134 
1135 step6:
1136 	/*
1137 	 * Update window information.
1138 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1139 	 */
1140 	if (((tiflags & TH_ACK) && SEQ_LT(tp->snd_wl1, ti->ti_seq)) ||
1141 	    (tp->snd_wl1 == ti->ti_seq && SEQ_LT(tp->snd_wl2, ti->ti_ack)) ||
1142 	    (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)) {
1143 		/* keep track of pure window updates */
1144 		if (ti->ti_len == 0 &&
1145 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1146 			tcpstat.tcps_rcvwinupd++;
1147 		tp->snd_wnd = tiwin;
1148 		tp->snd_wl1 = ti->ti_seq;
1149 		tp->snd_wl2 = ti->ti_ack;
1150 		if (tp->snd_wnd > tp->max_sndwnd)
1151 			tp->max_sndwnd = tp->snd_wnd;
1152 		needoutput = 1;
1153 	}
1154 
1155 	/*
1156 	 * Process segments with URG.
1157 	 */
1158 	if ((tiflags & TH_URG) && ti->ti_urp &&
1159 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1160 		/*
1161 		 * This is a kludge, but if we receive and accept
1162 		 * random urgent pointers, we'll crash in
1163 		 * soreceive.  It's hard to imagine someone
1164 		 * actually wanting to send this much urgent data.
1165 		 */
1166 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1167 			ti->ti_urp = 0;			/* XXX */
1168 			tiflags &= ~TH_URG;		/* XXX */
1169 			goto dodata;			/* XXX */
1170 		}
1171 		/*
1172 		 * If this segment advances the known urgent pointer,
1173 		 * then mark the data stream.  This should not happen
1174 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1175 		 * a FIN has been received from the remote side.
1176 		 * In these states we ignore the URG.
1177 		 *
1178 		 * According to RFC961 (Assigned Protocols),
1179 		 * the urgent pointer points to the last octet
1180 		 * of urgent data.  We continue, however,
1181 		 * to consider it to indicate the first octet
1182 		 * of data past the urgent section as the original
1183 		 * spec states (in one of two places).
1184 		 */
1185 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1186 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1187 			so->so_oobmark = so->so_rcv.sb_cc +
1188 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1189 			if (so->so_oobmark == 0)
1190 				so->so_state |= SS_RCVATMARK;
1191 			sohasoutofband(so);
1192 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1193 		}
1194 		/*
1195 		 * Remove out of band data so doesn't get presented to user.
1196 		 * This can happen independent of advancing the URG pointer,
1197 		 * but if two URG's are pending at once, some out-of-band
1198 		 * data may creep in... ick.
1199 		 */
1200 		if (ti->ti_urp <= (u_int16_t) ti->ti_len
1201 #ifdef SO_OOBINLINE
1202 		     && (so->so_options & SO_OOBINLINE) == 0
1203 #endif
1204 		     )
1205 			tcp_pulloutofband(so, ti, m);
1206 	} else
1207 		/*
1208 		 * If no out of band data is expected,
1209 		 * pull receive urgent pointer along
1210 		 * with the receive window.
1211 		 */
1212 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1213 			tp->rcv_up = tp->rcv_nxt;
1214 dodata:							/* XXX */
1215 
1216 	/*
1217 	 * Process the segment text, merging it into the TCP sequencing queue,
1218 	 * and arranging for acknowledgment of receipt if necessary.
1219 	 * This process logically involves adjusting tp->rcv_wnd as data
1220 	 * is presented to the user (this happens in tcp_usrreq.c,
1221 	 * case PRU_RCVD).  If a FIN has already been received on this
1222 	 * connection then we just ignore the text.
1223 	 */
1224 	if ((ti->ti_len || (tiflags & TH_FIN)) &&
1225 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1226 		TCP_REASS(tp, ti, m, so, tiflags);
1227 		/*
1228 		 * Note the amount of data that peer has sent into
1229 		 * our window, in order to estimate the sender's
1230 		 * buffer size.
1231 		 */
1232 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1233 	} else {
1234 		m_freem(m);
1235 		tiflags &= ~TH_FIN;
1236 	}
1237 
1238 	/*
1239 	 * If FIN is received ACK the FIN and let the user know
1240 	 * that the connection is closing.  Ignore a FIN received before
1241 	 * the connection is fully established.
1242 	 */
1243 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
1244 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1245 			socantrcvmore(so);
1246 			tp->t_flags |= TF_ACKNOW;
1247 			tp->rcv_nxt++;
1248 		}
1249 		switch (tp->t_state) {
1250 
1251 	 	/*
1252 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
1253 		 */
1254 		case TCPS_ESTABLISHED:
1255 			tp->t_state = TCPS_CLOSE_WAIT;
1256 			break;
1257 
1258 	 	/*
1259 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1260 		 * enter the CLOSING state.
1261 		 */
1262 		case TCPS_FIN_WAIT_1:
1263 			tp->t_state = TCPS_CLOSING;
1264 			break;
1265 
1266 	 	/*
1267 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1268 		 * starting the time-wait timer, turning off the other
1269 		 * standard timers.
1270 		 */
1271 		case TCPS_FIN_WAIT_2:
1272 			tp->t_state = TCPS_TIME_WAIT;
1273 			tcp_canceltimers(tp);
1274 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1275 			soisdisconnected(so);
1276 			break;
1277 
1278 		/*
1279 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1280 		 */
1281 		case TCPS_TIME_WAIT:
1282 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1283 			break;
1284 		}
1285 	}
1286 	if (so->so_options & SO_DEBUG)
1287 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1288 
1289 	/*
1290 	 * Return any desired output.
1291 	 */
1292 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1293 		(void) tcp_output(tp);
1294 	return;
1295 
1296 dropafterack:
1297 	/*
1298 	 * Generate an ACK dropping incoming segment if it occupies
1299 	 * sequence space, where the ACK reflects our state.
1300 	 */
1301 	if (tiflags & TH_RST)
1302 		goto drop;
1303 	m_freem(m);
1304 	tp->t_flags |= TF_ACKNOW;
1305 	(void) tcp_output(tp);
1306 	return;
1307 
1308 dropwithreset:
1309 	/*
1310 	 * Generate a RST, dropping incoming segment.
1311 	 * Make ACK acceptable to originator of segment.
1312 	 * Don't bother to respond if destination was broadcast/multicast.
1313 	 */
1314 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1315 	    IN_MULTICAST(ti->ti_dst.s_addr))
1316 		goto drop;
1317 	if (tiflags & TH_ACK)
1318 		tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1319 	else {
1320 		if (tiflags & TH_SYN)
1321 			ti->ti_len++;
1322 		tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1323 		    TH_RST|TH_ACK);
1324 	}
1325 	/* destroy temporarily created socket */
1326 	if (dropsocket)
1327 		(void) soabort(so);
1328 	return;
1329 
1330 drop:
1331 	/*
1332 	 * Drop space held by incoming segment and return.
1333 	 */
1334 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1335 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1336 	m_freem(m);
1337 	/* destroy temporarily created socket */
1338 	if (dropsocket)
1339 		(void) soabort(so);
1340 	return;
1341 #ifndef TUBA_INCLUDE
1342 }
1343 
1344 void
1345 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr)
1346 	struct tcpcb *tp;
1347 	u_char *cp;
1348 	int cnt;
1349 	struct tcpiphdr *ti;
1350 	int *ts_present;
1351 	u_int32_t *ts_val, *ts_ecr;
1352 {
1353 	u_int16_t mss;
1354 	int opt, optlen;
1355 
1356 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1357 		opt = cp[0];
1358 		if (opt == TCPOPT_EOL)
1359 			break;
1360 		if (opt == TCPOPT_NOP)
1361 			optlen = 1;
1362 		else {
1363 			optlen = cp[1];
1364 			if (optlen <= 0)
1365 				break;
1366 		}
1367 		switch (opt) {
1368 
1369 		default:
1370 			continue;
1371 
1372 		case TCPOPT_MAXSEG:
1373 			if (optlen != TCPOLEN_MAXSEG)
1374 				continue;
1375 			if (!(ti->ti_flags & TH_SYN))
1376 				continue;
1377 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
1378 			NTOHS(mss);
1379 			(void) tcp_mss(tp, mss);	/* sets t_maxseg */
1380 			break;
1381 
1382 		case TCPOPT_WINDOW:
1383 			if (optlen != TCPOLEN_WINDOW)
1384 				continue;
1385 			if (!(ti->ti_flags & TH_SYN))
1386 				continue;
1387 			tp->t_flags |= TF_RCVD_SCALE;
1388 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1389 			break;
1390 
1391 		case TCPOPT_TIMESTAMP:
1392 			if (optlen != TCPOLEN_TIMESTAMP)
1393 				continue;
1394 			*ts_present = 1;
1395 			bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val));
1396 			NTOHL(*ts_val);
1397 			bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr));
1398 			NTOHL(*ts_ecr);
1399 
1400 			/*
1401 			 * A timestamp received in a SYN makes
1402 			 * it ok to send timestamp requests and replies.
1403 			 */
1404 			if (ti->ti_flags & TH_SYN) {
1405 				tp->t_flags |= TF_RCVD_TSTMP;
1406 				tp->ts_recent = *ts_val;
1407 				tp->ts_recent_age = tcp_now;
1408 			}
1409 			break;
1410 		}
1411 	}
1412 }
1413 
1414 /*
1415  * Pull out of band byte out of a segment so
1416  * it doesn't appear in the user's data queue.
1417  * It is still reflected in the segment length for
1418  * sequencing purposes.
1419  */
1420 void
1421 tcp_pulloutofband(so, ti, m)
1422 	struct socket *so;
1423 	struct tcpiphdr *ti;
1424 	register struct mbuf *m;
1425 {
1426 	int cnt = ti->ti_urp - 1;
1427 
1428 	while (cnt >= 0) {
1429 		if (m->m_len > cnt) {
1430 			char *cp = mtod(m, caddr_t) + cnt;
1431 			struct tcpcb *tp = sototcpcb(so);
1432 
1433 			tp->t_iobc = *cp;
1434 			tp->t_oobflags |= TCPOOB_HAVEDATA;
1435 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1436 			m->m_len--;
1437 			return;
1438 		}
1439 		cnt -= m->m_len;
1440 		m = m->m_next;
1441 		if (m == 0)
1442 			break;
1443 	}
1444 	panic("tcp_pulloutofband");
1445 }
1446 
1447 /*
1448  * Collect new round-trip time estimate
1449  * and update averages and current timeout.
1450  */
1451 void
1452 tcp_xmit_timer(tp, rtt)
1453 	register struct tcpcb *tp;
1454 	short rtt;
1455 {
1456 	register short delta;
1457 
1458 	tcpstat.tcps_rttupdated++;
1459 	--rtt;
1460 	if (tp->t_srtt != 0) {
1461 		/*
1462 		 * srtt is stored as fixed point with 3 bits after the
1463 		 * binary point (i.e., scaled by 8).  The following magic
1464 		 * is equivalent to the smoothing algorithm in rfc793 with
1465 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1466 		 * point).  Adjust rtt to origin 0.
1467 		 */
1468 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
1469 		if ((tp->t_srtt += delta) <= 0)
1470 			tp->t_srtt = 1;
1471 		/*
1472 		 * We accumulate a smoothed rtt variance (actually, a
1473 		 * smoothed mean difference), then set the retransmit
1474 		 * timer to smoothed rtt + 4 times the smoothed variance.
1475 		 * rttvar is stored as fixed point with 2 bits after the
1476 		 * binary point (scaled by 4).  The following is
1477 		 * equivalent to rfc793 smoothing with an alpha of .75
1478 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1479 		 * rfc793's wired-in beta.
1480 		 */
1481 		if (delta < 0)
1482 			delta = -delta;
1483 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1484 		if ((tp->t_rttvar += delta) <= 0)
1485 			tp->t_rttvar = 1;
1486 	} else {
1487 		/*
1488 		 * No rtt measurement yet - use the unsmoothed rtt.
1489 		 * Set the variance to half the rtt (so our first
1490 		 * retransmit happens at 3*rtt).
1491 		 */
1492 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
1493 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
1494 	}
1495 	tp->t_rtt = 0;
1496 	tp->t_rxtshift = 0;
1497 
1498 	/*
1499 	 * the retransmit should happen at rtt + 4 * rttvar.
1500 	 * Because of the way we do the smoothing, srtt and rttvar
1501 	 * will each average +1/2 tick of bias.  When we compute
1502 	 * the retransmit timer, we want 1/2 tick of rounding and
1503 	 * 1 extra tick because of +-1/2 tick uncertainty in the
1504 	 * firing of the timer.  The bias will give us exactly the
1505 	 * 1.5 tick we need.  But, because the bias is
1506 	 * statistical, we have to test that we don't drop below
1507 	 * the minimum feasible timer (which is 2 ticks).
1508 	 */
1509 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1510 	    rtt + 2, TCPTV_REXMTMAX);
1511 
1512 	/*
1513 	 * We received an ack for a packet that wasn't retransmitted;
1514 	 * it is probably safe to discard any error indications we've
1515 	 * received recently.  This isn't quite right, but close enough
1516 	 * for now (a route might have failed after we sent a segment,
1517 	 * and the return path might not be symmetrical).
1518 	 */
1519 	tp->t_softerror = 0;
1520 }
1521 
1522 /*
1523  * Determine a reasonable value for maxseg size.
1524  * If the route is known, check route for mtu.
1525  * If none, use an mss that can be handled on the outgoing
1526  * interface without forcing IP to fragment; if bigger than
1527  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1528  * to utilize large mbufs.  If no route is found, route has no mtu,
1529  * or the destination isn't local, use a default, hopefully conservative
1530  * size (usually 512 or the default IP max size, but no more than the mtu
1531  * of the interface), as we can't discover anything about intervening
1532  * gateways or networks.  We also initialize the congestion/slow start
1533  * window to be a single segment if the destination isn't local.
1534  * While looking at the routing entry, we also initialize other path-dependent
1535  * parameters from pre-set or cached values in the routing entry.
1536  */
1537 int
1538 tcp_mss(tp, offer)
1539 	register struct tcpcb *tp;
1540 	u_int offer;
1541 {
1542 	struct route *ro;
1543 	register struct rtentry *rt;
1544 	struct ifnet *ifp;
1545 	register int rtt, mss;
1546 	u_long bufsize;
1547 	struct inpcb *inp;
1548 	struct socket *so;
1549 	extern int tcp_mssdflt;
1550 
1551 	inp = tp->t_inpcb;
1552 	ro = &inp->inp_route;
1553 
1554 	if ((rt = ro->ro_rt) == (struct rtentry *)0) {
1555 		/* No route yet, so try to acquire one */
1556 		if (!in_nullhost(inp->inp_faddr)) {
1557 			ro->ro_dst.sa_family = AF_INET;
1558 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1559 			satosin(&ro->ro_dst)->sin_addr = inp->inp_faddr;
1560 			rtalloc(ro);
1561 		}
1562 		if ((rt = ro->ro_rt) == (struct rtentry *)0)
1563 			return (tcp_mssdflt);
1564 	}
1565 	ifp = rt->rt_ifp;
1566 	so = inp->inp_socket;
1567 
1568 #ifdef RTV_MTU	/* if route characteristics exist ... */
1569 	/*
1570 	 * While we're here, check if there's an initial rtt
1571 	 * or rttvar.  Convert from the route-table units
1572 	 * to scaled multiples of the slow timeout timer.
1573 	 */
1574 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1575 		/*
1576 		 * XXX the lock bit for MTU indicates that the value
1577 		 * is also a minimum value; this is subject to time.
1578 		 */
1579 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1580 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
1581 		tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
1582 		if (rt->rt_rmx.rmx_rttvar)
1583 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1584 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
1585 		else
1586 			/* default variation is +- 1 rtt */
1587 			tp->t_rttvar =
1588 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1589 		TCPT_RANGESET((long) tp->t_rxtcur,
1590 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
1591 		    tp->t_rttmin, TCPTV_REXMTMAX);
1592 	}
1593 	/*
1594 	 * if there's an mtu associated with the route, use it
1595 	 */
1596 	if (rt->rt_rmx.rmx_mtu)
1597 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
1598 	else
1599 #endif /* RTV_MTU */
1600 	{
1601 		mss = ifp->if_mtu - sizeof(struct tcpiphdr);
1602 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1603 		if (mss > MCLBYTES)
1604 			mss &= ~(MCLBYTES-1);
1605 #else
1606 		if (mss > MCLBYTES)
1607 			mss = mss / MCLBYTES * MCLBYTES;
1608 #endif
1609 		if (!in_localaddr(inp->inp_faddr))
1610 			mss = min(mss, tcp_mssdflt);
1611 	}
1612 	/*
1613 	 * The current mss, t_maxseg, is initialized to the default value.
1614 	 * If we compute a smaller value, reduce the current mss.
1615 	 * If we compute a larger value, return it for use in sending
1616 	 * a max seg size option, but don't store it for use
1617 	 * unless we received an offer at least that large from peer.
1618 	 * However, do not accept offers under 32 bytes.
1619 	 */
1620 	if (offer)
1621 		mss = min(mss, offer);
1622 	mss = max(mss, 32);		/* sanity */
1623 	if (mss < tp->t_maxseg || offer != 0) {
1624 		/*
1625 		 * If there's a pipesize, change the socket buffer
1626 		 * to that size.  Make the socket buffers an integral
1627 		 * number of mss units; if the mss is larger than
1628 		 * the socket buffer, decrease the mss.
1629 		 */
1630 #ifdef RTV_SPIPE
1631 		if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
1632 #endif
1633 			bufsize = so->so_snd.sb_hiwat;
1634 		if (bufsize < mss)
1635 			mss = bufsize;
1636 		else {
1637 			bufsize = roundup(bufsize, mss);
1638 			if (bufsize > sb_max)
1639 				bufsize = sb_max;
1640 			(void)sbreserve(&so->so_snd, bufsize);
1641 		}
1642 		tp->t_maxseg = mss;
1643 
1644 #ifdef RTV_RPIPE
1645 		if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
1646 #endif
1647 			bufsize = so->so_rcv.sb_hiwat;
1648 		if (bufsize > mss) {
1649 			bufsize = roundup(bufsize, mss);
1650 			if (bufsize > sb_max)
1651 				bufsize = sb_max;
1652 			(void)sbreserve(&so->so_rcv, bufsize);
1653 		}
1654 	}
1655 	tp->snd_cwnd = mss;
1656 
1657 #ifdef RTV_SSTHRESH
1658 	if (rt->rt_rmx.rmx_ssthresh) {
1659 		/*
1660 		 * There's some sort of gateway or interface
1661 		 * buffer limit on the path.  Use this to set
1662 		 * the slow start threshhold, but set the
1663 		 * threshold to no less than 2*mss.
1664 		 */
1665 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1666 	}
1667 #endif /* RTV_MTU */
1668 	return (mss);
1669 }
1670 #endif /* TUBA_INCLUDE */
1671