xref: /freebsd/sys/netinet/tcp_reass.c (revision a316b26e50bbed7cf655fbba726ab87d8ab7599d)
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  *	@(#)tcp_input.c	8.5 (Berkeley) 4/10/94
34  * $Id: tcp_input.c,v 1.9 1994/10/02 17:48:43 phk 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 #ifdef TCPDEBUG
62 #include <netinet/tcp_debug.h>
63 struct	tcpiphdr tcp_saveti;
64 #endif
65 
66 int	tcprexmtthresh = 3;
67 struct	inpcb *tcp_last_inpcb = &tcb;
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 		tp->t_flags |= TF_DELACK; \
92 		(tp)->rcv_nxt += (ti)->ti_len; \
93 		flags = (ti)->ti_flags & TH_FIN; \
94 		tcpstat.tcps_rcvpack++;\
95 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
96 		sbappend(&(so)->so_rcv, (m)); \
97 		sorwakeup(so); \
98 	} else { \
99 		(flags) = tcp_reass((tp), (ti), (m)); \
100 		tp->t_flags |= TF_ACKNOW; \
101 	} \
102 }
103 #ifndef TUBA_INCLUDE
104 
105 int
106 tcp_reass(tp, ti, m)
107 	register struct tcpcb *tp;
108 	register struct tcpiphdr *ti;
109 	struct mbuf *m;
110 {
111 	register struct tcpiphdr *q;
112 	struct socket *so = tp->t_inpcb->inp_socket;
113 	int flags;
114 
115 	/*
116 	 * Call with ti==0 after become established to
117 	 * force pre-ESTABLISHED data up to user socket.
118 	 */
119 	if (ti == 0)
120 		goto present;
121 
122 	/*
123 	 * Find a segment which begins after this one does.
124 	 */
125 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
126 	    q = (struct tcpiphdr *)q->ti_next)
127 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
128 			break;
129 
130 	/*
131 	 * If there is a preceding segment, it may provide some of
132 	 * our data already.  If so, drop the data from the incoming
133 	 * segment.  If it provides all of our data, drop us.
134 	 */
135 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
136 		register int i;
137 		q = (struct tcpiphdr *)q->ti_prev;
138 		/* conversion to int (in i) handles seq wraparound */
139 		i = q->ti_seq + q->ti_len - ti->ti_seq;
140 		if (i > 0) {
141 			if (i >= ti->ti_len) {
142 				tcpstat.tcps_rcvduppack++;
143 				tcpstat.tcps_rcvdupbyte += ti->ti_len;
144 				m_freem(m);
145 				return (0);
146 			}
147 			m_adj(m, i);
148 			ti->ti_len -= i;
149 			ti->ti_seq += i;
150 		}
151 		q = (struct tcpiphdr *)(q->ti_next);
152 	}
153 	tcpstat.tcps_rcvoopack++;
154 	tcpstat.tcps_rcvoobyte += ti->ti_len;
155 	REASS_MBUF(ti) = m;		/* XXX */
156 
157 	/*
158 	 * While we overlap succeeding segments trim them or,
159 	 * if they are completely covered, dequeue them.
160 	 */
161 	while (q != (struct tcpiphdr *)tp) {
162 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
163 		if (i <= 0)
164 			break;
165 		if (i < q->ti_len) {
166 			q->ti_seq += i;
167 			q->ti_len -= i;
168 			m_adj(REASS_MBUF(q), i);
169 			break;
170 		}
171 		q = (struct tcpiphdr *)q->ti_next;
172 		m = REASS_MBUF((struct tcpiphdr *)q->ti_prev);
173 		remque(q->ti_prev);
174 		m_freem(m);
175 	}
176 
177 	/*
178 	 * Stick new segment in its place.
179 	 */
180 	insque(ti, q->ti_prev);
181 
182 present:
183 	/*
184 	 * Present data to user, advancing rcv_nxt through
185 	 * completed sequence space.
186 	 */
187 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
188 		return (0);
189 	ti = tp->seg_next;
190 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
191 		return (0);
192 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
193 		return (0);
194 	do {
195 		tp->rcv_nxt += ti->ti_len;
196 		flags = ti->ti_flags & TH_FIN;
197 		remque(ti);
198 		m = REASS_MBUF(ti);
199 		ti = (struct tcpiphdr *)ti->ti_next;
200 		if (so->so_state & SS_CANTRCVMORE)
201 			m_freem(m);
202 		else
203 			sbappend(&so->so_rcv, m);
204 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
205 	sorwakeup(so);
206 	return (flags);
207 }
208 
209 /*
210  * TCP input routine, follows pages 65-76 of the
211  * protocol specification dated September, 1981 very closely.
212  */
213 void
214 tcp_input(m, iphlen)
215 	register struct mbuf *m;
216 	int iphlen;
217 {
218 	register struct tcpiphdr *ti;
219 	register struct inpcb *inp;
220 	caddr_t optp = NULL;
221 	int optlen = 0;
222 	int len, tlen, off;
223 	register struct tcpcb *tp = 0;
224 	register int tiflags;
225 	struct socket *so = 0;
226 	int todrop, acked, ourfinisacked, needoutput = 0;
227 	struct in_addr laddr;
228 	int dropsocket = 0;
229 	int iss = 0;
230 	u_long tiwin, ts_val, ts_ecr;
231 	int ts_present = 0;
232 #ifdef TCPDEBUG
233 	short ostate = 0;
234 #endif
235 
236 	tcpstat.tcps_rcvtotal++;
237 	/*
238 	 * Get IP and TCP header together in first mbuf.
239 	 * Note: IP leaves IP header in first mbuf.
240 	 */
241 	ti = mtod(m, struct tcpiphdr *);
242 	if (iphlen > sizeof (struct ip))
243 		ip_stripoptions(m, (struct mbuf *)0);
244 	if (m->m_len < sizeof (struct tcpiphdr)) {
245 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
246 			tcpstat.tcps_rcvshort++;
247 			return;
248 		}
249 		ti = mtod(m, struct tcpiphdr *);
250 	}
251 
252 	/*
253 	 * Checksum extended TCP header and data.
254 	 */
255 	tlen = ((struct ip *)ti)->ip_len;
256 	len = sizeof (struct ip) + tlen;
257 	ti->ti_next = ti->ti_prev = 0;
258 	ti->ti_x1 = 0;
259 	ti->ti_len = (u_short)tlen;
260 	HTONS(ti->ti_len);
261 	ti->ti_sum = in_cksum(m, len);
262 	if (ti->ti_sum) {
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 #ifdef TCPDEBUG
356 		if (so->so_options & SO_DEBUG) {
357 			ostate = tp->t_state;
358 			tcp_saveti = *ti;
359 		}
360 #endif
361 		if (so->so_options & SO_ACCEPTCONN) {
362 			so = sonewconn(so, 0);
363 			if (so == 0)
364 				goto drop;
365 			/*
366 			 * This is ugly, but ....
367 			 *
368 			 * Mark socket as temporary until we're
369 			 * committed to keeping it.  The code at
370 			 * ``drop'' and ``dropwithreset'' check the
371 			 * flag dropsocket to see if the temporary
372 			 * socket created here should be discarded.
373 			 * We mark the socket as discardable until
374 			 * we're committed to it below in TCPS_LISTEN.
375 			 */
376 			dropsocket++;
377 			inp = (struct inpcb *)so->so_pcb;
378 			inp->inp_laddr = ti->ti_dst;
379 			inp->inp_lport = ti->ti_dport;
380 #if BSD>=43
381 			inp->inp_options = ip_srcroute();
382 #endif
383 			tp = intotcpcb(inp);
384 			tp->t_state = TCPS_LISTEN;
385 
386 			/* Compute proper scaling value from buffer space
387 			 */
388 			while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
389 			   TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
390 				tp->request_r_scale++;
391 		}
392 	}
393 
394 	/*
395 	 * Segment received on connection.
396 	 * Reset idle time and keep-alive timer.
397 	 */
398 	tp->t_idle = 0;
399 	tp->t_timer[TCPT_KEEP] = tcp_keepidle;
400 
401 	/*
402 	 * Process options if not in LISTEN state,
403 	 * else do it below (after getting remote address).
404 	 */
405 	if (optp && tp->t_state != TCPS_LISTEN)
406 		tcp_dooptions(tp, optp, optlen, ti,
407 			&ts_present, &ts_val, &ts_ecr);
408 
409 	/*
410 	 * Header prediction: check for the two common cases
411 	 * of a uni-directional data xfer.  If the packet has
412 	 * no control flags, is in-sequence, the window didn't
413 	 * change and we're not retransmitting, it's a
414 	 * candidate.  If the length is zero and the ack moved
415 	 * forward, we're the sender side of the xfer.  Just
416 	 * free the data acked & wake any higher level process
417 	 * that was blocked waiting for space.  If the length
418 	 * is non-zero and the ack didn't move, we're the
419 	 * receiver side.  If we're getting packets in-order
420 	 * (the reassembly queue is empty), add the data to
421 	 * the socket buffer and note that we need a delayed ack.
422 	 */
423 	if (tp->t_state == TCPS_ESTABLISHED &&
424 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
425 	    (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) &&
426 	    ti->ti_seq == tp->rcv_nxt &&
427 	    tiwin && tiwin == tp->snd_wnd &&
428 	    tp->snd_nxt == tp->snd_max) {
429 
430 		/*
431 		 * If last ACK falls within this segment's sequence numbers,
432 		 *  record the timestamp.
433 		 */
434 		if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
435 		   SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
436 			tp->ts_recent_age = tcp_now;
437 			tp->ts_recent = ts_val;
438 		}
439 
440 		if (ti->ti_len == 0) {
441 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
442 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
443 			    tp->snd_cwnd >= tp->snd_wnd) {
444 				/*
445 				 * this is a pure ack for outstanding data.
446 				 */
447 				++tcpstat.tcps_predack;
448 				if (ts_present)
449 					tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
450 				else if (tp->t_rtt &&
451 					    SEQ_GT(ti->ti_ack, tp->t_rtseq))
452 					tcp_xmit_timer(tp, tp->t_rtt);
453 				acked = ti->ti_ack - tp->snd_una;
454 				tcpstat.tcps_rcvackpack++;
455 				tcpstat.tcps_rcvackbyte += acked;
456 				sbdrop(&so->so_snd, acked);
457 				tp->snd_una = ti->ti_ack;
458 				m_freem(m);
459 
460 				/*
461 				 * If all outstanding data are acked, stop
462 				 * retransmit timer, otherwise restart timer
463 				 * using current (possibly backed-off) value.
464 				 * If process is waiting for space,
465 				 * wakeup/selwakeup/signal.  If data
466 				 * are ready to send, let tcp_output
467 				 * decide between more output or persist.
468 				 */
469 				if (tp->snd_una == tp->snd_max)
470 					tp->t_timer[TCPT_REXMT] = 0;
471 				else if (tp->t_timer[TCPT_PERSIST] == 0)
472 					tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
473 
474 				if (so->so_snd.sb_flags & SB_NOTIFY)
475 					sowwakeup(so);
476 				if (so->so_snd.sb_cc)
477 					(void) tcp_output(tp);
478 				return;
479 			}
480 		} else if (ti->ti_ack == tp->snd_una &&
481 		    tp->seg_next == (struct tcpiphdr *)tp &&
482 		    ti->ti_len <= sbspace(&so->so_rcv)) {
483 			/*
484 			 * this is a pure, in-sequence data packet
485 			 * with nothing on the reassembly queue and
486 			 * we have enough buffer space to take it.
487 			 */
488 			++tcpstat.tcps_preddat;
489 			tp->rcv_nxt += ti->ti_len;
490 			tcpstat.tcps_rcvpack++;
491 			tcpstat.tcps_rcvbyte += ti->ti_len;
492 			/*
493 			 * Drop TCP, IP headers and TCP options then add data
494 			 * to socket buffer.
495 			 */
496 			m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
497 			m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
498 			sbappend(&so->so_rcv, m);
499 			sorwakeup(so);
500 			/*
501 			 * If this is a small packet, then ACK now - with Nagel
502 			 *	congestion avoidance sender won't send more until
503 			 *	he gets an ACK.
504 			 */
505 			if ((unsigned)ti->ti_len < tp->t_maxseg) {
506 				tp->t_flags |= TF_ACKNOW;
507 				tcp_output(tp);
508 			} else {
509 				tp->t_flags |= TF_DELACK;
510 			}
511 			return;
512 		}
513 	}
514 
515 	/*
516 	 * Drop TCP, IP headers and TCP options.
517 	 */
518 	m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
519 	m->m_len  -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
520 
521 	/*
522 	 * Calculate amount of space in receive window,
523 	 * and then do TCP input processing.
524 	 * Receive window is amount of space in rcv queue,
525 	 * but not less than advertised window.
526 	 */
527 	{ int win;
528 
529 	win = sbspace(&so->so_rcv);
530 	if (win < 0)
531 		win = 0;
532 	tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
533 	}
534 
535 	switch (tp->t_state) {
536 
537 	/*
538 	 * If the state is LISTEN then ignore segment if it contains an RST.
539 	 * If the segment contains an ACK then it is bad and send a RST.
540 	 * If it does not contain a SYN then it is not interesting; drop it.
541 	 * Don't bother responding if the destination was a broadcast.
542 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
543 	 * tp->iss, and send a segment:
544 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
545 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
546 	 * Fill in remote peer address fields if not previously specified.
547 	 * Enter SYN_RECEIVED state, and process any other fields of this
548 	 * segment in this state.
549 	 */
550 	case TCPS_LISTEN: {
551 		struct mbuf *am;
552 		register struct sockaddr_in *sin;
553 
554 		if (tiflags & TH_RST)
555 			goto drop;
556 		if (tiflags & TH_ACK)
557 			goto dropwithreset;
558 		if ((tiflags & TH_SYN) == 0)
559 			goto drop;
560 		/*
561 		 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
562 		 * in_broadcast() should never return true on a received
563 		 * packet with M_BCAST not set.
564 		 */
565 		if (m->m_flags & (M_BCAST|M_MCAST) ||
566 		    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
567 			goto drop;
568 		am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
569 		if (am == NULL)
570 			goto drop;
571 		am->m_len = sizeof (struct sockaddr_in);
572 		sin = mtod(am, struct sockaddr_in *);
573 		sin->sin_family = AF_INET;
574 		sin->sin_len = sizeof(*sin);
575 		sin->sin_addr = ti->ti_src;
576 		sin->sin_port = ti->ti_sport;
577 		bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
578 		laddr = inp->inp_laddr;
579 		if (inp->inp_laddr.s_addr == INADDR_ANY)
580 			inp->inp_laddr = ti->ti_dst;
581 		if (in_pcbconnect(inp, am)) {
582 			inp->inp_laddr = laddr;
583 			(void) m_free(am);
584 			goto drop;
585 		}
586 		(void) m_free(am);
587 		tp->t_template = tcp_template(tp);
588 		if (tp->t_template == 0) {
589 			tp = tcp_drop(tp, ENOBUFS);
590 			dropsocket = 0;		/* socket is already gone */
591 			goto drop;
592 		}
593 		if (optp)
594 			tcp_dooptions(tp, optp, optlen, ti,
595 				&ts_present, &ts_val, &ts_ecr);
596 		if (iss)
597 			tp->iss = iss;
598 		else
599 			tp->iss = tcp_iss;
600 		tcp_iss += TCP_ISSINCR/2;
601 		tp->irs = ti->ti_seq;
602 		tcp_sendseqinit(tp);
603 		tcp_rcvseqinit(tp);
604 		tp->t_flags |= TF_ACKNOW;
605 		tp->t_state = TCPS_SYN_RECEIVED;
606 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
607 		dropsocket = 0;		/* committed to socket */
608 		tcpstat.tcps_accepts++;
609 		goto trimthenstep6;
610 		}
611 
612 	/*
613 	 * If the state is SYN_SENT:
614 	 *	if seg contains an ACK, but not for our SYN, drop the input.
615 	 *	if seg contains a RST, then drop the connection.
616 	 *	if seg does not contain SYN, then drop it.
617 	 * Otherwise this is an acceptable SYN segment
618 	 *	initialize tp->rcv_nxt and tp->irs
619 	 *	if seg contains ack then advance tp->snd_una
620 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
621 	 *	arrange for segment to be acked (eventually)
622 	 *	continue processing rest of data/controls, beginning with URG
623 	 */
624 	case TCPS_SYN_SENT:
625 		if ((tiflags & TH_ACK) &&
626 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
627 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
628 			goto dropwithreset;
629 		if (tiflags & TH_RST) {
630 			if (tiflags & TH_ACK)
631 				tp = tcp_drop(tp, ECONNREFUSED);
632 			goto drop;
633 		}
634 		if ((tiflags & TH_SYN) == 0)
635 			goto drop;
636 		if (tiflags & TH_ACK) {
637 			tp->snd_una = ti->ti_ack;
638 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
639 				tp->snd_nxt = tp->snd_una;
640 		}
641 		tp->t_timer[TCPT_REXMT] = 0;
642 		tp->irs = ti->ti_seq;
643 		tcp_rcvseqinit(tp);
644 		tp->t_flags |= TF_ACKNOW;
645 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
646 			tcpstat.tcps_connects++;
647 			soisconnected(so);
648 			tp->t_state = TCPS_ESTABLISHED;
649 			/* Do window scaling on this connection? */
650 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
651 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
652 				tp->snd_scale = tp->requested_s_scale;
653 				tp->rcv_scale = tp->request_r_scale;
654 			}
655 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
656 				(struct mbuf *)0);
657 			/*
658 			 * if we didn't have to retransmit the SYN,
659 			 * use its rtt as our initial srtt & rtt var.
660 			 */
661 			if (tp->t_rtt)
662 				tcp_xmit_timer(tp, tp->t_rtt);
663 		} else
664 			tp->t_state = TCPS_SYN_RECEIVED;
665 
666 trimthenstep6:
667 		/*
668 		 * Advance ti->ti_seq to correspond to first data byte.
669 		 * If data, trim to stay within window,
670 		 * dropping FIN if necessary.
671 		 */
672 		ti->ti_seq++;
673 		if (ti->ti_len > tp->rcv_wnd) {
674 			todrop = ti->ti_len - tp->rcv_wnd;
675 			m_adj(m, -todrop);
676 			ti->ti_len = tp->rcv_wnd;
677 			tiflags &= ~TH_FIN;
678 			tcpstat.tcps_rcvpackafterwin++;
679 			tcpstat.tcps_rcvbyteafterwin += todrop;
680 		}
681 		tp->snd_wl1 = ti->ti_seq - 1;
682 		tp->rcv_up = ti->ti_seq;
683 		goto step6;
684 	}
685 
686 	/*
687 	 * States other than LISTEN or SYN_SENT.
688 	 * First check timestamp, if present.
689 	 * Then check that at least some bytes of segment are within
690 	 * receive window.  If segment begins before rcv_nxt,
691 	 * drop leading data (and SYN); if nothing left, just ack.
692 	 *
693 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
694 	 * and it's less than ts_recent, drop it.
695 	 */
696 	if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
697 	    TSTMP_LT(ts_val, tp->ts_recent)) {
698 
699 		/* Check to see if ts_recent is over 24 days old.  */
700 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
701 			/*
702 			 * Invalidate ts_recent.  If this segment updates
703 			 * ts_recent, the age will be reset later and ts_recent
704 			 * will get a valid value.  If it does not, setting
705 			 * ts_recent to zero will at least satisfy the
706 			 * requirement that zero be placed in the timestamp
707 			 * echo reply when ts_recent isn't valid.  The
708 			 * age isn't reset until we get a valid ts_recent
709 			 * because we don't want out-of-order segments to be
710 			 * dropped when ts_recent is old.
711 			 */
712 			tp->ts_recent = 0;
713 		} else {
714 			tcpstat.tcps_rcvduppack++;
715 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
716 			tcpstat.tcps_pawsdrop++;
717 			goto dropafterack;
718 		}
719 	}
720 
721 	todrop = tp->rcv_nxt - ti->ti_seq;
722 	if (todrop > 0) {
723 		if (tiflags & TH_SYN) {
724 			tiflags &= ~TH_SYN;
725 			ti->ti_seq++;
726 			if (ti->ti_urp > 1)
727 				ti->ti_urp--;
728 			else
729 				tiflags &= ~TH_URG;
730 			todrop--;
731 		}
732 		if (todrop >= ti->ti_len) {
733 			tcpstat.tcps_rcvduppack++;
734 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
735 			/*
736 			 * If segment is just one to the left of the window,
737 			 * check two special cases:
738 			 * 1. Don't toss RST in response to 4.2-style keepalive.
739 			 * 2. If the only thing to drop is a FIN, we can drop
740 			 *    it, but check the ACK or we will get into FIN
741 			 *    wars if our FINs crossed (both CLOSING).
742 			 * In either case, send ACK to resynchronize,
743 			 * but keep on processing for RST or ACK.
744 			 */
745 			if ((tiflags & TH_FIN && todrop == ti->ti_len + 1)
746 #ifdef TCP_COMPAT_42
747 			  || (tiflags & TH_RST && ti->ti_seq == tp->rcv_nxt - 1)
748 #endif
749 			   ) {
750 				todrop = ti->ti_len;
751 				tiflags &= ~TH_FIN;
752 				tp->t_flags |= TF_ACKNOW;
753 			} else {
754 				/*
755 				 * Handle the case when a bound socket connects
756 				 * to itself. Allow packets with a SYN and
757 				 * an ACK to continue with the processing.
758 				 */
759 				if (todrop != 0 || (tiflags & TH_ACK) == 0)
760 					goto dropafterack;
761 			}
762 		} else {
763 			tcpstat.tcps_rcvpartduppack++;
764 			tcpstat.tcps_rcvpartdupbyte += todrop;
765 		}
766 		m_adj(m, todrop);
767 		ti->ti_seq += todrop;
768 		ti->ti_len -= todrop;
769 		if (ti->ti_urp > todrop)
770 			ti->ti_urp -= todrop;
771 		else {
772 			tiflags &= ~TH_URG;
773 			ti->ti_urp = 0;
774 		}
775 	}
776 
777 	/*
778 	 * If new data are received on a connection after the
779 	 * user processes are gone, then RST the other end.
780 	 */
781 	if ((so->so_state & SS_NOFDREF) &&
782 	    tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
783 		tp = tcp_close(tp);
784 		tcpstat.tcps_rcvafterclose++;
785 		goto dropwithreset;
786 	}
787 
788 	/*
789 	 * If segment ends after window, drop trailing data
790 	 * (and PUSH and FIN); if nothing left, just ACK.
791 	 */
792 	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
793 	if (todrop > 0) {
794 		tcpstat.tcps_rcvpackafterwin++;
795 		if (todrop >= ti->ti_len) {
796 			tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
797 			/*
798 			 * If a new connection request is received
799 			 * while in TIME_WAIT, drop the old connection
800 			 * and start over if the sequence numbers
801 			 * are above the previous ones.
802 			 */
803 			if (tiflags & TH_SYN &&
804 			    tp->t_state == TCPS_TIME_WAIT &&
805 			    SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
806 				iss = tp->rcv_nxt + TCP_ISSINCR;
807 				tp = tcp_close(tp);
808 				goto findpcb;
809 			}
810 			/*
811 			 * If window is closed can only take segments at
812 			 * window edge, and have to drop data and PUSH from
813 			 * incoming segments.  Continue processing, but
814 			 * remember to ack.  Otherwise, drop segment
815 			 * and ack.
816 			 */
817 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
818 				tp->t_flags |= TF_ACKNOW;
819 				tcpstat.tcps_rcvwinprobe++;
820 			} else
821 				goto dropafterack;
822 		} else
823 			tcpstat.tcps_rcvbyteafterwin += todrop;
824 		m_adj(m, -todrop);
825 		ti->ti_len -= todrop;
826 		tiflags &= ~(TH_PUSH|TH_FIN);
827 	}
828 
829 	/*
830 	 * If last ACK falls within this segment's sequence numbers,
831 	 * record its timestamp.
832 	 */
833 	if (ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
834 	    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
835 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
836 		tp->ts_recent_age = tcp_now;
837 		tp->ts_recent = ts_val;
838 	}
839 
840 	/*
841 	 * If the RST bit is set examine the state:
842 	 *    SYN_RECEIVED STATE:
843 	 *	If passive open, return to LISTEN state.
844 	 *	If active open, inform user that connection was refused.
845 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
846 	 *	Inform user that connection was reset, and close tcb.
847 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
848 	 *	Close the tcb.
849 	 */
850 	if (tiflags&TH_RST) switch (tp->t_state) {
851 
852 	case TCPS_SYN_RECEIVED:
853 		so->so_error = ECONNREFUSED;
854 		goto close;
855 
856 	case TCPS_ESTABLISHED:
857 	case TCPS_FIN_WAIT_1:
858 	case TCPS_FIN_WAIT_2:
859 	case TCPS_CLOSE_WAIT:
860 		so->so_error = ECONNRESET;
861 	close:
862 		tp->t_state = TCPS_CLOSED;
863 		tcpstat.tcps_drops++;
864 		tp = tcp_close(tp);
865 		goto drop;
866 
867 	case TCPS_CLOSING:
868 	case TCPS_LAST_ACK:
869 	case TCPS_TIME_WAIT:
870 		tp = tcp_close(tp);
871 		goto drop;
872 	}
873 
874 	/*
875 	 * If a SYN is in the window, then this is an
876 	 * error and we send an RST and drop the connection.
877 	 */
878 	if (tiflags & TH_SYN) {
879 		tp = tcp_drop(tp, ECONNRESET);
880 		goto dropwithreset;
881 	}
882 
883 	/*
884 	 * If the ACK bit is off we drop the segment and return.
885 	 */
886 	if ((tiflags & TH_ACK) == 0)
887 		goto drop;
888 
889 	/*
890 	 * Ack processing.
891 	 */
892 	switch (tp->t_state) {
893 
894 	/*
895 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
896 	 * ESTABLISHED state and continue processing, otherwise
897 	 * send an RST.
898 	 */
899 	case TCPS_SYN_RECEIVED:
900 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
901 		    SEQ_GT(ti->ti_ack, tp->snd_max))
902 			goto dropwithreset;
903 		tcpstat.tcps_connects++;
904 		soisconnected(so);
905 		tp->t_state = TCPS_ESTABLISHED;
906 		/* Do window scaling? */
907 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
908 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
909 			tp->snd_scale = tp->requested_s_scale;
910 			tp->rcv_scale = tp->request_r_scale;
911 		}
912 		(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
913 		tp->snd_wl1 = ti->ti_seq - 1;
914 		/* fall into ... */
915 
916 	/*
917 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
918 	 * ACKs.  If the ack is in the range
919 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
920 	 * then advance tp->snd_una to ti->ti_ack and drop
921 	 * data from the retransmission queue.  If this ACK reflects
922 	 * more up to date window information we update our window information.
923 	 */
924 	case TCPS_ESTABLISHED:
925 	case TCPS_FIN_WAIT_1:
926 	case TCPS_FIN_WAIT_2:
927 	case TCPS_CLOSE_WAIT:
928 	case TCPS_CLOSING:
929 	case TCPS_LAST_ACK:
930 	case TCPS_TIME_WAIT:
931 
932 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
933 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
934 				tcpstat.tcps_rcvdupack++;
935 				/*
936 				 * If we have outstanding data (other than
937 				 * a window probe), this is a completely
938 				 * duplicate ack (ie, window info didn't
939 				 * change), the ack is the biggest we've
940 				 * seen and we've seen exactly our rexmt
941 				 * threshhold of them, assume a packet
942 				 * has been dropped and retransmit it.
943 				 * Kludge snd_nxt & the congestion
944 				 * window so we send only this one
945 				 * packet.
946 				 *
947 				 * We know we're losing at the current
948 				 * window size so do congestion avoidance
949 				 * (set ssthresh to half the current window
950 				 * and pull our congestion window back to
951 				 * the new ssthresh).
952 				 *
953 				 * Dup acks mean that packets have left the
954 				 * network (they're now cached at the receiver)
955 				 * so bump cwnd by the amount in the receiver
956 				 * to keep a constant cwnd packets in the
957 				 * network.
958 				 */
959 				if (tp->t_timer[TCPT_REXMT] == 0 ||
960 				    ti->ti_ack != tp->snd_una)
961 					tp->t_dupacks = 0;
962 				else if (++tp->t_dupacks == tcprexmtthresh) {
963 					tcp_seq onxt = tp->snd_nxt;
964 					u_int win =
965 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
966 						tp->t_maxseg;
967 
968 					if (win < 2)
969 						win = 2;
970 					tp->snd_ssthresh = win * tp->t_maxseg;
971 					tp->t_timer[TCPT_REXMT] = 0;
972 					tp->t_rtt = 0;
973 					tp->snd_nxt = ti->ti_ack;
974 					tp->snd_cwnd = tp->t_maxseg;
975 					(void) tcp_output(tp);
976 					tp->snd_cwnd = tp->snd_ssthresh +
977 					       tp->t_maxseg * tp->t_dupacks;
978 					if (SEQ_GT(onxt, tp->snd_nxt))
979 						tp->snd_nxt = onxt;
980 					goto drop;
981 				} else if (tp->t_dupacks > tcprexmtthresh) {
982 					tp->snd_cwnd += tp->t_maxseg;
983 					(void) tcp_output(tp);
984 					goto drop;
985 				}
986 			} else
987 				tp->t_dupacks = 0;
988 			break;
989 		}
990 		/*
991 		 * If the congestion window was inflated to account
992 		 * for the other side's cached packets, retract it.
993 		 */
994 		if (tp->t_dupacks > tcprexmtthresh &&
995 		    tp->snd_cwnd > tp->snd_ssthresh)
996 			tp->snd_cwnd = tp->snd_ssthresh;
997 		tp->t_dupacks = 0;
998 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
999 			tcpstat.tcps_rcvacktoomuch++;
1000 			goto dropafterack;
1001 		}
1002 		acked = ti->ti_ack - tp->snd_una;
1003 		tcpstat.tcps_rcvackpack++;
1004 		tcpstat.tcps_rcvackbyte += acked;
1005 
1006 		/*
1007 		 * If we have a timestamp reply, update smoothed
1008 		 * round trip time.  If no timestamp is present but
1009 		 * transmit timer is running and timed sequence
1010 		 * number was acked, update smoothed round trip time.
1011 		 * Since we now have an rtt measurement, cancel the
1012 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1013 		 * Recompute the initial retransmit timer.
1014 		 */
1015 		if (ts_present)
1016 			tcp_xmit_timer(tp, tcp_now-ts_ecr+1);
1017 		else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1018 			tcp_xmit_timer(tp,tp->t_rtt);
1019 
1020 		/*
1021 		 * If all outstanding data is acked, stop retransmit
1022 		 * timer and remember to restart (more output or persist).
1023 		 * If there is more data to be acked, restart retransmit
1024 		 * timer, using current (possibly backed-off) value.
1025 		 */
1026 		if (ti->ti_ack == tp->snd_max) {
1027 			tp->t_timer[TCPT_REXMT] = 0;
1028 			needoutput = 1;
1029 		} else if (tp->t_timer[TCPT_PERSIST] == 0)
1030 			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1031 		/*
1032 		 * When new data is acked, open the congestion window.
1033 		 * If the window gives us less than ssthresh packets
1034 		 * in flight, open exponentially (maxseg per packet).
1035 		 * Otherwise open linearly: maxseg per window
1036 		 * (maxseg^2 / cwnd per packet).
1037 		 */
1038 		{
1039 		register u_int cw = tp->snd_cwnd;
1040 		register u_int incr = tp->t_maxseg;
1041 
1042 		if (cw > tp->snd_ssthresh)
1043 			incr = incr * incr / cw;
1044 		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1045 		}
1046 		if (acked > so->so_snd.sb_cc) {
1047 			tp->snd_wnd -= so->so_snd.sb_cc;
1048 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1049 			ourfinisacked = 1;
1050 		} else {
1051 			sbdrop(&so->so_snd, acked);
1052 			tp->snd_wnd -= acked;
1053 			ourfinisacked = 0;
1054 		}
1055 		if (so->so_snd.sb_flags & SB_NOTIFY)
1056 			sowwakeup(so);
1057 		tp->snd_una = ti->ti_ack;
1058 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1059 			tp->snd_nxt = tp->snd_una;
1060 
1061 		switch (tp->t_state) {
1062 
1063 		/*
1064 		 * In FIN_WAIT_1 STATE in addition to the processing
1065 		 * for the ESTABLISHED state if our FIN is now acknowledged
1066 		 * then enter FIN_WAIT_2.
1067 		 */
1068 		case TCPS_FIN_WAIT_1:
1069 			if (ourfinisacked) {
1070 				/*
1071 				 * If we can't receive any more
1072 				 * data, then closing user can proceed.
1073 				 * Starting the timer is contrary to the
1074 				 * specification, but if we don't get a FIN
1075 				 * we'll hang forever.
1076 				 */
1077 				if (so->so_state & SS_CANTRCVMORE) {
1078 					soisdisconnected(so);
1079 					tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1080 				}
1081 				tp->t_state = TCPS_FIN_WAIT_2;
1082 			}
1083 			break;
1084 
1085 	 	/*
1086 		 * In CLOSING STATE in addition to the processing for
1087 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1088 		 * then enter the TIME-WAIT state, otherwise ignore
1089 		 * the segment.
1090 		 */
1091 		case TCPS_CLOSING:
1092 			if (ourfinisacked) {
1093 				tp->t_state = TCPS_TIME_WAIT;
1094 				tcp_canceltimers(tp);
1095 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1096 				soisdisconnected(so);
1097 			}
1098 			break;
1099 
1100 		/*
1101 		 * In LAST_ACK, we may still be waiting for data to drain
1102 		 * and/or to be acked, as well as for the ack of our FIN.
1103 		 * If our FIN is now acknowledged, delete the TCB,
1104 		 * enter the closed state and return.
1105 		 */
1106 		case TCPS_LAST_ACK:
1107 			if (ourfinisacked) {
1108 				tp = tcp_close(tp);
1109 				goto drop;
1110 			}
1111 			break;
1112 
1113 		/*
1114 		 * In TIME_WAIT state the only thing that should arrive
1115 		 * is a retransmission of the remote FIN.  Acknowledge
1116 		 * it and restart the finack timer.
1117 		 */
1118 		case TCPS_TIME_WAIT:
1119 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1120 			goto dropafterack;
1121 		}
1122 	}
1123 
1124 step6:
1125 	/*
1126 	 * Update window information.
1127 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1128 	 */
1129 	if ((tiflags & TH_ACK) &&
1130 	    (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1131 	    (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1132 	     (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1133 		/* keep track of pure window updates */
1134 		if (ti->ti_len == 0 &&
1135 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1136 			tcpstat.tcps_rcvwinupd++;
1137 		tp->snd_wnd = tiwin;
1138 		tp->snd_wl1 = ti->ti_seq;
1139 		tp->snd_wl2 = ti->ti_ack;
1140 		if (tp->snd_wnd > tp->max_sndwnd)
1141 			tp->max_sndwnd = tp->snd_wnd;
1142 		needoutput = 1;
1143 	}
1144 
1145 	/*
1146 	 * Process segments with URG.
1147 	 */
1148 	if ((tiflags & TH_URG) && ti->ti_urp &&
1149 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1150 		/*
1151 		 * This is a kludge, but if we receive and accept
1152 		 * random urgent pointers, we'll crash in
1153 		 * soreceive.  It's hard to imagine someone
1154 		 * actually wanting to send this much urgent data.
1155 		 */
1156 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1157 			ti->ti_urp = 0;			/* XXX */
1158 			tiflags &= ~TH_URG;		/* XXX */
1159 			goto dodata;			/* XXX */
1160 		}
1161 		/*
1162 		 * If this segment advances the known urgent pointer,
1163 		 * then mark the data stream.  This should not happen
1164 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1165 		 * a FIN has been received from the remote side.
1166 		 * In these states we ignore the URG.
1167 		 *
1168 		 * According to RFC961 (Assigned Protocols),
1169 		 * the urgent pointer points to the last octet
1170 		 * of urgent data.  We continue, however,
1171 		 * to consider it to indicate the first octet
1172 		 * of data past the urgent section as the original
1173 		 * spec states (in one of two places).
1174 		 */
1175 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1176 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1177 			so->so_oobmark = so->so_rcv.sb_cc +
1178 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1179 			if (so->so_oobmark == 0)
1180 				so->so_state |= SS_RCVATMARK;
1181 			sohasoutofband(so);
1182 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1183 		}
1184 		/*
1185 		 * Remove out of band data so doesn't get presented to user.
1186 		 * This can happen independent of advancing the URG pointer,
1187 		 * but if two URG's are pending at once, some out-of-band
1188 		 * data may creep in... ick.
1189 		 */
1190 		if (ti->ti_urp <= (u_long)ti->ti_len
1191 #ifdef SO_OOBINLINE
1192 		     && (so->so_options & SO_OOBINLINE) == 0
1193 #endif
1194 		     )
1195 			tcp_pulloutofband(so, ti, m);
1196 	} else
1197 		/*
1198 		 * If no out of band data is expected,
1199 		 * pull receive urgent pointer along
1200 		 * with the receive window.
1201 		 */
1202 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1203 			tp->rcv_up = tp->rcv_nxt;
1204 dodata:							/* XXX */
1205 
1206 	/*
1207 	 * Process the segment text, merging it into the TCP sequencing queue,
1208 	 * and arranging for acknowledgment of receipt if necessary.
1209 	 * This process logically involves adjusting tp->rcv_wnd as data
1210 	 * is presented to the user (this happens in tcp_usrreq.c,
1211 	 * case PRU_RCVD).  If a FIN has already been received on this
1212 	 * connection then we just ignore the text.
1213 	 */
1214 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
1215 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1216 		TCP_REASS(tp, ti, m, so, tiflags);
1217 		/*
1218 		 * Note the amount of data that peer has sent into
1219 		 * our window, in order to estimate the sender's
1220 		 * buffer size.
1221 		 */
1222 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1223 	} else {
1224 		m_freem(m);
1225 		tiflags &= ~TH_FIN;
1226 	}
1227 
1228 	/*
1229 	 * If FIN is received ACK the FIN and let the user know
1230 	 * that the connection is closing.
1231 	 */
1232 	if (tiflags & TH_FIN) {
1233 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1234 			socantrcvmore(so);
1235 			tp->t_flags |= TF_ACKNOW;
1236 			tp->rcv_nxt++;
1237 		}
1238 		switch (tp->t_state) {
1239 
1240 	 	/*
1241 		 * In SYN_RECEIVED and ESTABLISHED STATES
1242 		 * enter the CLOSE_WAIT state.
1243 		 */
1244 		case TCPS_SYN_RECEIVED:
1245 		case TCPS_ESTABLISHED:
1246 			tp->t_state = TCPS_CLOSE_WAIT;
1247 			break;
1248 
1249 	 	/*
1250 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1251 		 * enter the CLOSING state.
1252 		 */
1253 		case TCPS_FIN_WAIT_1:
1254 			tp->t_state = TCPS_CLOSING;
1255 			break;
1256 
1257 	 	/*
1258 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1259 		 * starting the time-wait timer, turning off the other
1260 		 * standard timers.
1261 		 */
1262 		case TCPS_FIN_WAIT_2:
1263 			tp->t_state = TCPS_TIME_WAIT;
1264 			tcp_canceltimers(tp);
1265 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1266 			soisdisconnected(so);
1267 			break;
1268 
1269 		/*
1270 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1271 		 */
1272 		case TCPS_TIME_WAIT:
1273 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1274 			break;
1275 		}
1276 	}
1277 #ifdef TCPDEBUG
1278 	if (so->so_options & SO_DEBUG)
1279 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1280 #endif
1281 
1282 	/*
1283 	 * If this is a small packet, then ACK now - with Nagel
1284 	 *      congestion avoidance sender won't send more until
1285 	 *      he gets an ACK.
1286 	 */
1287 	if (ti->ti_len && ((unsigned)ti->ti_len < tp->t_maxseg))
1288 		tp->t_flags |= TF_ACKNOW;
1289 
1290 	/*
1291 	 * Return any desired output.
1292 	 */
1293 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1294 		(void) tcp_output(tp);
1295 	return;
1296 
1297 dropafterack:
1298 	/*
1299 	 * Generate an ACK dropping incoming segment if it occupies
1300 	 * sequence space, where the ACK reflects our state.
1301 	 */
1302 	if (tiflags & TH_RST)
1303 		goto drop;
1304 	m_freem(m);
1305 	tp->t_flags |= TF_ACKNOW;
1306 	(void) tcp_output(tp);
1307 	return;
1308 
1309 dropwithreset:
1310 	/*
1311 	 * Generate a RST, dropping incoming segment.
1312 	 * Make ACK acceptable to originator of segment.
1313 	 * Don't bother to respond if destination was broadcast/multicast.
1314 	 */
1315 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1316 	    IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1317 		goto drop;
1318 	if (tiflags & TH_ACK)
1319 		tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1320 	else {
1321 		if (tiflags & TH_SYN)
1322 			ti->ti_len++;
1323 		tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1324 		    TH_RST|TH_ACK);
1325 	}
1326 	/* destroy temporarily created socket */
1327 	if (dropsocket)
1328 		(void) soabort(so);
1329 	return;
1330 
1331 drop:
1332 	/*
1333 	 * Drop space held by incoming segment and return.
1334 	 */
1335 #ifdef TCPDEBUG
1336 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1337 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1338 #endif
1339 	m_freem(m);
1340 	/* destroy temporarily created socket */
1341 	if (dropsocket)
1342 		(void) soabort(so);
1343 	return;
1344 #ifndef TUBA_INCLUDE
1345 }
1346 
1347 void
1348 tcp_dooptions(tp, cp, cnt, ti, ts_present, ts_val, ts_ecr)
1349 	struct tcpcb *tp;
1350 	u_char *cp;
1351 	int cnt;
1352 	struct tcpiphdr *ti;
1353 	int *ts_present;
1354 	u_long *ts_val, *ts_ecr;
1355 {
1356 	u_short mss;
1357 	int opt, optlen;
1358 
1359 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1360 		opt = cp[0];
1361 		if (opt == TCPOPT_EOL)
1362 			break;
1363 		if (opt == TCPOPT_NOP)
1364 			optlen = 1;
1365 		else {
1366 			optlen = cp[1];
1367 			if (optlen <= 0)
1368 				break;
1369 		}
1370 		switch (opt) {
1371 
1372 		default:
1373 			continue;
1374 
1375 		case TCPOPT_MAXSEG:
1376 			if (optlen != TCPOLEN_MAXSEG)
1377 				continue;
1378 			if (!(ti->ti_flags & TH_SYN))
1379 				continue;
1380 			bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
1381 			NTOHS(mss);
1382 			(void) tcp_mss(tp, mss);	/* sets t_maxseg */
1383 			break;
1384 
1385 		case TCPOPT_WINDOW:
1386 			if (optlen != TCPOLEN_WINDOW)
1387 				continue;
1388 			if (!(ti->ti_flags & TH_SYN))
1389 				continue;
1390 			tp->t_flags |= TF_RCVD_SCALE;
1391 			tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1392 			break;
1393 
1394 		case TCPOPT_TIMESTAMP:
1395 			if (optlen != TCPOLEN_TIMESTAMP)
1396 				continue;
1397 			*ts_present = 1;
1398 			bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val));
1399 			NTOHL(*ts_val);
1400 			bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr));
1401 			NTOHL(*ts_ecr);
1402 
1403 			/*
1404 			 * A timestamp received in a SYN makes
1405 			 * it ok to send timestamp requests and replies.
1406 			 */
1407 			if (ti->ti_flags & TH_SYN) {
1408 				tp->t_flags |= TF_RCVD_TSTMP;
1409 				tp->ts_recent = *ts_val;
1410 				tp->ts_recent_age = tcp_now;
1411 			}
1412 			break;
1413 		}
1414 	}
1415 }
1416 
1417 /*
1418  * Pull out of band byte out of a segment so
1419  * it doesn't appear in the user's data queue.
1420  * It is still reflected in the segment length for
1421  * sequencing purposes.
1422  */
1423 void
1424 tcp_pulloutofband(so, ti, m)
1425 	struct socket *so;
1426 	struct tcpiphdr *ti;
1427 	register struct mbuf *m;
1428 {
1429 	int cnt = ti->ti_urp - 1;
1430 
1431 	while (cnt >= 0) {
1432 		if (m->m_len > cnt) {
1433 			char *cp = mtod(m, caddr_t) + cnt;
1434 			struct tcpcb *tp = sototcpcb(so);
1435 
1436 			tp->t_iobc = *cp;
1437 			tp->t_oobflags |= TCPOOB_HAVEDATA;
1438 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1439 			m->m_len--;
1440 			return;
1441 		}
1442 		cnt -= m->m_len;
1443 		m = m->m_next;
1444 		if (m == 0)
1445 			break;
1446 	}
1447 	panic("tcp_pulloutofband");
1448 }
1449 
1450 /*
1451  * Collect new round-trip time estimate
1452  * and update averages and current timeout.
1453  */
1454 void
1455 tcp_xmit_timer(tp, rtt)
1456 	register struct tcpcb *tp;
1457 	short rtt;
1458 {
1459 	register short delta;
1460 
1461 	tcpstat.tcps_rttupdated++;
1462 	if (tp->t_srtt != 0) {
1463 		/*
1464 		 * srtt is stored as fixed point with 3 bits after the
1465 		 * binary point (i.e., scaled by 8).  The following magic
1466 		 * is equivalent to the smoothing algorithm in rfc793 with
1467 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1468 		 * point).  Adjust rtt to origin 0.
1469 		 */
1470 		delta = rtt - 1 - (tp->t_srtt >> TCP_RTT_SHIFT);
1471 		if ((tp->t_srtt += delta) <= 0)
1472 			tp->t_srtt = 1;
1473 		/*
1474 		 * We accumulate a smoothed rtt variance (actually, a
1475 		 * smoothed mean difference), then set the retransmit
1476 		 * timer to smoothed rtt + 4 times the smoothed variance.
1477 		 * rttvar is stored as fixed point with 2 bits after the
1478 		 * binary point (scaled by 4).  The following is
1479 		 * equivalent to rfc793 smoothing with an alpha of .75
1480 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1481 		 * rfc793's wired-in beta.
1482 		 */
1483 		if (delta < 0)
1484 			delta = -delta;
1485 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1486 		if ((tp->t_rttvar += delta) <= 0)
1487 			tp->t_rttvar = 1;
1488 	} else {
1489 		/*
1490 		 * No rtt measurement yet - use the unsmoothed rtt.
1491 		 * Set the variance to half the rtt (so our first
1492 		 * retransmit happens at 3*rtt).
1493 		 */
1494 		tp->t_srtt = rtt << TCP_RTT_SHIFT;
1495 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
1496 	}
1497 	tp->t_rtt = 0;
1498 	tp->t_rxtshift = 0;
1499 
1500 	/*
1501 	 * the retransmit should happen at rtt + 4 * rttvar.
1502 	 * Because of the way we do the smoothing, srtt and rttvar
1503 	 * will each average +1/2 tick of bias.  When we compute
1504 	 * the retransmit timer, we want 1/2 tick of rounding and
1505 	 * 1 extra tick because of +-1/2 tick uncertainty in the
1506 	 * firing of the timer.  The bias will give us exactly the
1507 	 * 1.5 tick we need.  But, because the bias is
1508 	 * statistical, we have to test that we don't drop below
1509 	 * the minimum feasible timer (which is 2 ticks).
1510 	 */
1511 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1512 	    tp->t_rttmin, TCPTV_REXMTMAX);
1513 
1514 	/*
1515 	 * We received an ack for a packet that wasn't retransmitted;
1516 	 * it is probably safe to discard any error indications we've
1517 	 * received recently.  This isn't quite right, but close enough
1518 	 * for now (a route might have failed after we sent a segment,
1519 	 * and the return path might not be symmetrical).
1520 	 */
1521 	tp->t_softerror = 0;
1522 }
1523 
1524 /*
1525  * Determine a reasonable value for maxseg size.
1526  * If the route is known, check route for mtu.
1527  * If none, use an mss that can be handled on the outgoing
1528  * interface without forcing IP to fragment; if bigger than
1529  * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
1530  * to utilize large mbufs.  If no route is found, route has no mtu,
1531  * or the destination isn't local, use a default, hopefully conservative
1532  * size (usually 512 or the default IP max size, but no more than the mtu
1533  * of the interface), as we can't discover anything about intervening
1534  * gateways or networks.  We also initialize the congestion/slow start
1535  * window to be a single segment if the destination isn't local.
1536  * While looking at the routing entry, we also initialize other path-dependent
1537  * parameters from pre-set or cached values in the routing entry.
1538  */
1539 int
1540 tcp_mss(tp, offer)
1541 	register struct tcpcb *tp;
1542 	u_int offer;
1543 {
1544 	struct route *ro;
1545 	register struct rtentry *rt;
1546 	struct ifnet *ifp;
1547 	register int rtt, mss;
1548 	u_long bufsize;
1549 	struct inpcb *inp;
1550 	struct socket *so;
1551 	extern int tcp_mssdflt;
1552 
1553 	inp = tp->t_inpcb;
1554 	ro = &inp->inp_route;
1555 
1556 	if ((rt = ro->ro_rt) == (struct rtentry *)0) {
1557 		/* No route yet, so try to acquire one */
1558 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1559 			ro->ro_dst.sa_family = AF_INET;
1560 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1561 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1562 				inp->inp_faddr;
1563 			rtalloc(ro);
1564 		}
1565 		if ((rt = ro->ro_rt) == (struct rtentry *)0)
1566 			return (tcp_mssdflt);
1567 	}
1568 	ifp = rt->rt_ifp;
1569 	so = inp->inp_socket;
1570 
1571 #ifdef RTV_MTU	/* if route characteristics exist ... */
1572 	/*
1573 	 * While we're here, check if there's an initial rtt
1574 	 * or rttvar.  Convert from the route-table units
1575 	 * to scaled multiples of the slow timeout timer.
1576 	 */
1577 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1578 		/*
1579 		 * XXX the lock bit for MTU indicates that the value
1580 		 * is also a minimum value; this is subject to time.
1581 		 */
1582 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
1583 			tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
1584 		tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
1585 		if (rt->rt_rmx.rmx_rttvar)
1586 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1587 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
1588 		else
1589 			/* default variation is +- 1 rtt */
1590 			tp->t_rttvar =
1591 			    tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1592 		TCPT_RANGESET(tp->t_rxtcur,
1593 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
1594 		    tp->t_rttmin, TCPTV_REXMTMAX);
1595 	}
1596 	/*
1597 	 * if there's an mtu associated with the route, use it
1598 	 */
1599 	if (rt->rt_rmx.rmx_mtu)
1600 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
1601 	else
1602 #endif /* RTV_MTU */
1603 	{
1604 		mss = ifp->if_mtu - sizeof(struct tcpiphdr);
1605 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1606 		if (mss > MCLBYTES)
1607 			mss &= ~(MCLBYTES-1);
1608 #else
1609 		if (mss > MCLBYTES)
1610 			mss = mss / MCLBYTES * MCLBYTES;
1611 #endif
1612 		if (!in_localaddr(inp->inp_faddr))
1613 			mss = min(mss, tcp_mssdflt);
1614 	}
1615 	/*
1616 	 * The current mss, t_maxseg, is initialized to the default value.
1617 	 * If we compute a smaller value, reduce the current mss.
1618 	 * If we compute a larger value, return it for use in sending
1619 	 * a max seg size option, but don't store it for use
1620 	 * unless we received an offer at least that large from peer.
1621 	 * However, do not accept offers under 32 bytes.
1622 	 */
1623 	if (offer)
1624 		mss = min(mss, offer);
1625 	mss = max(mss, 32);		/* sanity */
1626 	if (mss < tp->t_maxseg || offer != 0) {
1627 		/*
1628 		 * If there's a pipesize, change the socket buffer
1629 		 * to that size.  Make the socket buffers an integral
1630 		 * number of mss units; if the mss is larger than
1631 		 * the socket buffer, decrease the mss.
1632 		 */
1633 #ifdef RTV_SPIPE
1634 		if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
1635 #endif
1636 			bufsize = so->so_snd.sb_hiwat;
1637 		if (bufsize < mss)
1638 			mss = bufsize;
1639 		else {
1640 			bufsize = roundup(bufsize, mss);
1641 			if (bufsize > sb_max)
1642 				bufsize = sb_max;
1643 			(void)sbreserve(&so->so_snd, bufsize);
1644 		}
1645 		tp->t_maxseg = mss;
1646 
1647 #ifdef RTV_RPIPE
1648 		if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
1649 #endif
1650 			bufsize = so->so_rcv.sb_hiwat;
1651 		if (bufsize > mss) {
1652 			bufsize = roundup(bufsize, mss);
1653 			if (bufsize > sb_max)
1654 				bufsize = sb_max;
1655 			(void)sbreserve(&so->so_rcv, bufsize);
1656 		}
1657 	}
1658 	tp->snd_cwnd = mss;
1659 
1660 #ifdef RTV_SSTHRESH
1661 	if (rt->rt_rmx.rmx_ssthresh) {
1662 		/*
1663 		 * There's some sort of gateway or interface
1664 		 * buffer limit on the path.  Use this to set
1665 		 * the slow start threshhold, but set the
1666 		 * threshold to no less than 2*mss.
1667 		 */
1668 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
1669 	}
1670 #endif /* RTV_MTU */
1671 	return (mss);
1672 }
1673 #endif /* TUBA_INCLUDE */
1674