xref: /csrg-svn/sys/netinet/tcp_input.c (revision 26824)
1 /*
2  * Copyright (c) 1982 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)tcp_input.c	6.22 (Berkeley) 03/12/86
7  */
8 
9 #include "param.h"
10 #include "systm.h"
11 #include "mbuf.h"
12 #include "protosw.h"
13 #include "socket.h"
14 #include "socketvar.h"
15 #include "errno.h"
16 
17 #include "../net/if.h"
18 #include "../net/route.h"
19 
20 #include "in.h"
21 #include "in_pcb.h"
22 #include "in_systm.h"
23 #include "ip.h"
24 #include "ip_var.h"
25 #include "tcp.h"
26 #include "tcp_fsm.h"
27 #include "tcp_seq.h"
28 #include "tcp_timer.h"
29 #include "tcp_var.h"
30 #include "tcpip.h"
31 #include "tcp_debug.h"
32 
33 int	tcpprintfs = 0;
34 int	tcpcksum = 1;
35 struct	tcpiphdr tcp_saveti;
36 extern	tcpnodelack;
37 
38 struct	tcpcb *tcp_newtcpcb();
39 
40 /*
41  * Insert segment ti into reassembly queue of tcp with
42  * control block tp.  Return TH_FIN if reassembly now includes
43  * a segment with FIN.  The macro form does the common case inline
44  * (segment is the next to be received on an established connection,
45  * and the queue is empty), avoiding linkage into and removal
46  * from the queue and repetition of various conversions.
47  */
48 #define	TCP_REASS(tp, ti, m, so, flags) { \
49 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
50 	    (tp)->seg_next == (struct tcpiphdr *)(tp) && \
51 	    (tp)->t_state == TCPS_ESTABLISHED) { \
52 		(tp)->rcv_nxt += (ti)->ti_len; \
53 		flags = (ti)->ti_flags & TH_FIN; \
54 		sbappend(&(so)->so_rcv, (m)); \
55 		sorwakeup(so); \
56 	} else \
57 		(flags) = tcp_reass((tp), (ti)); \
58 }
59 
60 tcp_reass(tp, ti)
61 	register struct tcpcb *tp;
62 	register struct tcpiphdr *ti;
63 {
64 	register struct tcpiphdr *q;
65 	struct socket *so = tp->t_inpcb->inp_socket;
66 	struct mbuf *m;
67 	int flags;
68 
69 	/*
70 	 * Call with ti==0 after become established to
71 	 * force pre-ESTABLISHED data up to user socket.
72 	 */
73 	if (ti == 0)
74 		goto present;
75 
76 	/*
77 	 * Find a segment which begins after this one does.
78 	 */
79 	for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
80 	    q = (struct tcpiphdr *)q->ti_next)
81 		if (SEQ_GT(q->ti_seq, ti->ti_seq))
82 			break;
83 
84 	/*
85 	 * If there is a preceding segment, it may provide some of
86 	 * our data already.  If so, drop the data from the incoming
87 	 * segment.  If it provides all of our data, drop us.
88 	 */
89 	if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
90 		register int i;
91 		q = (struct tcpiphdr *)q->ti_prev;
92 		/* conversion to int (in i) handles seq wraparound */
93 		i = q->ti_seq + q->ti_len - ti->ti_seq;
94 		if (i > 0) {
95 			if (i >= ti->ti_len)
96 				goto drop;
97 			m_adj(dtom(ti), i);
98 			ti->ti_len -= i;
99 			ti->ti_seq += i;
100 		}
101 		q = (struct tcpiphdr *)(q->ti_next);
102 	}
103 
104 	/*
105 	 * While we overlap succeeding segments trim them or,
106 	 * if they are completely covered, dequeue them.
107 	 */
108 	while (q != (struct tcpiphdr *)tp) {
109 		register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
110 		if (i <= 0)
111 			break;
112 		if (i < q->ti_len) {
113 			q->ti_seq += i;
114 			q->ti_len -= i;
115 			m_adj(dtom(q), i);
116 			break;
117 		}
118 		q = (struct tcpiphdr *)q->ti_next;
119 		m = dtom(q->ti_prev);
120 		remque(q->ti_prev);
121 		m_freem(m);
122 	}
123 
124 	/*
125 	 * Stick new segment in its place.
126 	 */
127 	insque(ti, q->ti_prev);
128 
129 present:
130 	/*
131 	 * Present data to user, advancing rcv_nxt through
132 	 * completed sequence space.
133 	 */
134 	if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
135 		return (0);
136 	ti = tp->seg_next;
137 	if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
138 		return (0);
139 	if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
140 		return (0);
141 	do {
142 		tp->rcv_nxt += ti->ti_len;
143 		flags = ti->ti_flags & TH_FIN;
144 		remque(ti);
145 		m = dtom(ti);
146 		ti = (struct tcpiphdr *)ti->ti_next;
147 		if (so->so_state & SS_CANTRCVMORE)
148 			m_freem(m);
149 		else
150 			sbappend(&so->so_rcv, m);
151 	} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
152 	sorwakeup(so);
153 	return (flags);
154 drop:
155 	m_freem(dtom(ti));
156 	return (0);
157 }
158 
159 /*
160  * TCP input routine, follows pages 65-76 of the
161  * protocol specification dated September, 1981 very closely.
162  */
163 tcp_input(m0)
164 	struct mbuf *m0;
165 {
166 	register struct tcpiphdr *ti;
167 	struct inpcb *inp;
168 	register struct mbuf *m;
169 	struct mbuf *om = 0;
170 	int len, tlen, off;
171 	register struct tcpcb *tp = 0;
172 	register int tiflags;
173 	struct socket *so;
174 	int todrop, acked, needoutput = 0;
175 	short ostate;
176 	struct in_addr laddr;
177 	int dropsocket = 0;
178 
179 	/*
180 	 * Get IP and TCP header together in first mbuf.
181 	 * Note: IP leaves IP header in first mbuf.
182 	 */
183 	m = m0;
184 	ti = mtod(m, struct tcpiphdr *);
185 	if (((struct ip *)ti)->ip_hl > (sizeof (struct ip) >> 2))
186 		ip_stripoptions((struct ip *)ti, (struct mbuf *)0);
187 	if (m->m_off > MMAXOFF || m->m_len < sizeof (struct tcpiphdr)) {
188 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
189 			tcpstat.tcps_hdrops++;
190 			return;
191 		}
192 		ti = mtod(m, struct tcpiphdr *);
193 	}
194 
195 	/*
196 	 * Checksum extended TCP header and data.
197 	 */
198 	tlen = ((struct ip *)ti)->ip_len;
199 	len = sizeof (struct ip) + tlen;
200 	if (tcpcksum) {
201 		ti->ti_next = ti->ti_prev = 0;
202 		ti->ti_x1 = 0;
203 		ti->ti_len = (u_short)tlen;
204 		ti->ti_len = htons((u_short)ti->ti_len);
205 		if (ti->ti_sum = in_cksum(m, len)) {
206 			if (tcpprintfs)
207 				printf("tcp sum: src %x\n", ti->ti_src);
208 			tcpstat.tcps_badsum++;
209 			goto drop;
210 		}
211 	}
212 
213 	/*
214 	 * Check that TCP offset makes sense,
215 	 * pull out TCP options and adjust length.
216 	 */
217 	off = ti->ti_off << 2;
218 	if (off < sizeof (struct tcphdr) || off > tlen) {
219 		if (tcpprintfs)
220 			printf("tcp off: src %x off %d\n", ti->ti_src, off);
221 		tcpstat.tcps_badoff++;
222 		goto drop;
223 	}
224 	tlen -= off;
225 	ti->ti_len = tlen;
226 	if (off > sizeof (struct tcphdr)) {
227 		if (m->m_len < sizeof(struct ip) + off) {
228 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
229 				tcpstat.tcps_hdrops++;
230 				return;
231 			}
232 			ti = mtod(m, struct tcpiphdr *);
233 		}
234 		om = m_get(M_DONTWAIT, MT_DATA);
235 		if (om == 0)
236 			goto drop;
237 		om->m_len = off - sizeof (struct tcphdr);
238 		{ caddr_t op = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
239 		  bcopy(op, mtod(om, caddr_t), (unsigned)om->m_len);
240 		  m->m_len -= om->m_len;
241 		  bcopy(op+om->m_len, op,
242 		   (unsigned)(m->m_len-sizeof (struct tcpiphdr)));
243 		}
244 	}
245 	tiflags = ti->ti_flags;
246 
247 	/*
248 	 * Drop TCP and IP headers; TCP options were dropped above.
249 	 */
250 	m->m_off += sizeof(struct tcpiphdr);
251 	m->m_len -= sizeof(struct tcpiphdr);
252 
253 	/*
254 	 * Convert TCP protocol specific fields to host format.
255 	 */
256 	ti->ti_seq = ntohl(ti->ti_seq);
257 	ti->ti_ack = ntohl(ti->ti_ack);
258 	ti->ti_win = ntohs(ti->ti_win);
259 	ti->ti_urp = ntohs(ti->ti_urp);
260 
261 	/*
262 	 * Locate pcb for segment.
263 	 */
264 	inp = in_pcblookup
265 		(&tcb, ti->ti_src, ti->ti_sport, ti->ti_dst, ti->ti_dport,
266 		INPLOOKUP_WILDCARD);
267 
268 	/*
269 	 * If the state is CLOSED (i.e., TCB does not exist) then
270 	 * all data in the incoming segment is discarded.
271 	 */
272 	if (inp == 0)
273 		goto dropwithreset;
274 	tp = intotcpcb(inp);
275 	if (tp == 0)
276 		goto dropwithreset;
277 	so = inp->inp_socket;
278 	if (so->so_options & SO_DEBUG) {
279 		ostate = tp->t_state;
280 		tcp_saveti = *ti;
281 	}
282 	if (so->so_options & SO_ACCEPTCONN) {
283 		so = sonewconn(so);
284 		if (so == 0)
285 			goto drop;
286 		/*
287 		 * This is ugly, but ....
288 		 *
289 		 * Mark socket as temporary until we're
290 		 * committed to keeping it.  The code at
291 		 * ``drop'' and ``dropwithreset'' check the
292 		 * flag dropsocket to see if the temporary
293 		 * socket created here should be discarded.
294 		 * We mark the socket as discardable until
295 		 * we're committed to it below in TCPS_LISTEN.
296 		 */
297 		dropsocket++;
298 		inp = (struct inpcb *)so->so_pcb;
299 		inp->inp_laddr = ti->ti_dst;
300 		inp->inp_lport = ti->ti_dport;
301 		inp->inp_options = ip_srcroute();
302 		tp = intotcpcb(inp);
303 		tp->t_state = TCPS_LISTEN;
304 	}
305 
306 	/*
307 	 * Segment received on connection.
308 	 * Reset idle time and keep-alive timer.
309 	 */
310 	tp->t_idle = 0;
311 	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
312 
313 	/*
314 	 * Process options if not in LISTEN state,
315 	 * else do it below (after getting remote address).
316 	 */
317 	if (om && tp->t_state != TCPS_LISTEN) {
318 		tcp_dooptions(tp, om, ti);
319 		om = 0;
320 	}
321 
322 	/*
323 	 * Calculate amount of space in receive window,
324 	 * and then do TCP input processing.
325 	 * Receive window is amount of space in rcv queue,
326 	 * but not less than advertised window.
327 	 */
328 	{ int win;
329 
330 	win = sbspace(&so->so_rcv);
331 	if (win < 0)
332 		win = 0;
333 	tp->rcv_wnd = MAX(win, (int)(tp->rcv_adv - tp->rcv_nxt));
334 	}
335 
336 	switch (tp->t_state) {
337 
338 	/*
339 	 * If the state is LISTEN then ignore segment if it contains an RST.
340 	 * If the segment contains an ACK then it is bad and send a RST.
341 	 * If it does not contain a SYN then it is not interesting; drop it.
342 	 * Don't bother responding if the destination was a broadcast.
343 	 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
344 	 * tp->iss, and send a segment:
345 	 *     <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
346 	 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
347 	 * Fill in remote peer address fields if not previously specified.
348 	 * Enter SYN_RECEIVED state, and process any other fields of this
349 	 * segment in this state.
350 	 */
351 	case TCPS_LISTEN: {
352 		struct mbuf *am;
353 		register struct sockaddr_in *sin;
354 
355 		if (tiflags & TH_RST)
356 			goto drop;
357 		if (tiflags & TH_ACK)
358 			goto dropwithreset;
359 		if ((tiflags & TH_SYN) == 0)
360 			goto drop;
361 		if (in_broadcast(ti->ti_dst))
362 			goto drop;
363 		am = m_get(M_DONTWAIT, MT_SONAME);
364 		if (am == NULL)
365 			goto drop;
366 		am->m_len = sizeof (struct sockaddr_in);
367 		sin = mtod(am, struct sockaddr_in *);
368 		sin->sin_family = AF_INET;
369 		sin->sin_addr = ti->ti_src;
370 		sin->sin_port = ti->ti_sport;
371 		laddr = inp->inp_laddr;
372 		if (inp->inp_laddr.s_addr == INADDR_ANY)
373 			inp->inp_laddr = ti->ti_dst;
374 		if (in_pcbconnect(inp, am)) {
375 			inp->inp_laddr = laddr;
376 			(void) m_free(am);
377 			goto drop;
378 		}
379 		(void) m_free(am);
380 		tp->t_template = tcp_template(tp);
381 		if (tp->t_template == 0) {
382 			tp = tcp_drop(tp, ENOBUFS);
383 			dropsocket = 0;		/* socket is already gone */
384 			goto drop;
385 		}
386 		if (om) {
387 			tcp_dooptions(tp, om, ti);
388 			om = 0;
389 		}
390 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
391 		tp->irs = ti->ti_seq;
392 		tcp_sendseqinit(tp);
393 		tcp_rcvseqinit(tp);
394 		tp->t_flags |= TF_ACKNOW;
395 		tp->t_state = TCPS_SYN_RECEIVED;
396 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
397 		dropsocket = 0;		/* committed to socket */
398 		goto trimthenstep6;
399 		}
400 
401 	/*
402 	 * If the state is SYN_SENT:
403 	 *	if seg contains an ACK, but not for our SYN, drop the input.
404 	 *	if seg contains a RST, then drop the connection.
405 	 *	if seg does not contain SYN, then drop it.
406 	 * Otherwise this is an acceptable SYN segment
407 	 *	initialize tp->rcv_nxt and tp->irs
408 	 *	if seg contains ack then advance tp->snd_una
409 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
410 	 *	arrange for segment to be acked (eventually)
411 	 *	continue processing rest of data/controls, beginning with URG
412 	 */
413 	case TCPS_SYN_SENT:
414 		if ((tiflags & TH_ACK) &&
415 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
416 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
417 			goto dropwithreset;
418 		if (tiflags & TH_RST) {
419 			if (tiflags & TH_ACK)
420 				tp = tcp_drop(tp, ECONNREFUSED);
421 			goto drop;
422 		}
423 		if ((tiflags & TH_SYN) == 0)
424 			goto drop;
425 		tp->snd_una = ti->ti_ack;
426 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
427 			tp->snd_nxt = tp->snd_una;
428 		tp->t_timer[TCPT_REXMT] = 0;
429 		tp->irs = ti->ti_seq;
430 		tcp_rcvseqinit(tp);
431 		tp->t_flags |= TF_ACKNOW;
432 		if (SEQ_GT(tp->snd_una, tp->iss)) {
433 			soisconnected(so);
434 			tp->t_state = TCPS_ESTABLISHED;
435 			tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp));
436 			(void) tcp_reass(tp, (struct tcpiphdr *)0);
437 		} else
438 			tp->t_state = TCPS_SYN_RECEIVED;
439 		goto trimthenstep6;
440 
441 trimthenstep6:
442 		/*
443 		 * Advance ti->ti_seq to correspond to first data byte.
444 		 * If data, trim to stay within window,
445 		 * dropping FIN if necessary.
446 		 */
447 		ti->ti_seq++;
448 		if (ti->ti_len > tp->rcv_wnd) {
449 			todrop = ti->ti_len - tp->rcv_wnd;
450 			m_adj(m, -todrop);
451 			ti->ti_len = tp->rcv_wnd;
452 			tiflags &= ~TH_FIN;
453 		}
454 		tp->snd_wl1 = ti->ti_seq - 1;
455 		tp->rcv_up = ti->ti_seq;
456 		goto step6;
457 	}
458 
459 	/*
460 	 * If data is received on a connection after the
461 	 * user processes are gone, then RST the other end.
462 	 */
463 	if ((so->so_state & SS_NOFDREF) && tp->t_state > TCPS_CLOSE_WAIT &&
464 	    ti->ti_len) {
465 		tp = tcp_close(tp);
466 		goto dropwithreset;
467 	}
468 
469 	/*
470 	 * States other than LISTEN or SYN_SENT.
471 	 * First check that at least some bytes of segment are within
472 	 * receive window.
473 	 */
474 	if (tp->rcv_wnd == 0) {
475 		/*
476 		 * If window is closed can only take segments at
477 		 * window edge, and have to drop data and PUSH from
478 		 * incoming segments.
479 		 */
480 		if (tp->rcv_nxt != ti->ti_seq)
481 			goto dropafterack;
482 		if (ti->ti_len > 0) {
483 			m_adj(m, ti->ti_len);
484 			ti->ti_len = 0;
485 			tiflags &= ~(TH_PUSH|TH_FIN);
486 		}
487 	} else {
488 		/*
489 		 * If segment begins before rcv_nxt, drop leading
490 		 * data (and SYN); if nothing left, just ack.
491 		 */
492 		todrop = tp->rcv_nxt - ti->ti_seq;
493 		if (todrop > 0) {
494 			if (tiflags & TH_SYN) {
495 				tiflags &= ~TH_SYN;
496 				ti->ti_seq++;
497 				if (ti->ti_urp > 1)
498 					ti->ti_urp--;
499 				else
500 					tiflags &= ~TH_URG;
501 				todrop--;
502 			}
503 			if (todrop > ti->ti_len ||
504 			    todrop == ti->ti_len && (tiflags&TH_FIN) == 0)
505 				goto dropafterack;
506 			m_adj(m, todrop);
507 			ti->ti_seq += todrop;
508 			ti->ti_len -= todrop;
509 			if (ti->ti_urp > todrop)
510 				ti->ti_urp -= todrop;
511 			else {
512 				tiflags &= ~TH_URG;
513 				ti->ti_urp = 0;
514 			}
515 		}
516 		/*
517 		 * If segment ends after window, drop trailing data
518 		 * (and PUSH and FIN); if nothing left, just ACK.
519 		 */
520 		todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
521 		if (todrop > 0) {
522 			if (todrop >= ti->ti_len)
523 				goto dropafterack;
524 			m_adj(m, -todrop);
525 			ti->ti_len -= todrop;
526 			tiflags &= ~(TH_PUSH|TH_FIN);
527 		}
528 	}
529 
530 	/*
531 	 * If the RST bit is set examine the state:
532 	 *    SYN_RECEIVED STATE:
533 	 *	If passive open, return to LISTEN state.
534 	 *	If active open, inform user that connection was refused.
535 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
536 	 *	Inform user that connection was reset, and close tcb.
537 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
538 	 *	Close the tcb.
539 	 */
540 	if (tiflags&TH_RST) switch (tp->t_state) {
541 
542 	case TCPS_SYN_RECEIVED:
543 		tp = tcp_drop(tp, ECONNREFUSED);
544 		goto drop;
545 
546 	case TCPS_ESTABLISHED:
547 	case TCPS_FIN_WAIT_1:
548 	case TCPS_FIN_WAIT_2:
549 	case TCPS_CLOSE_WAIT:
550 		tp = tcp_drop(tp, ECONNRESET);
551 		goto drop;
552 
553 	case TCPS_CLOSING:
554 	case TCPS_LAST_ACK:
555 	case TCPS_TIME_WAIT:
556 		tp = tcp_close(tp);
557 		goto drop;
558 	}
559 
560 	/*
561 	 * If a SYN is in the window, then this is an
562 	 * error and we send an RST and drop the connection.
563 	 */
564 	if (tiflags & TH_SYN) {
565 		tp = tcp_drop(tp, ECONNRESET);
566 		goto dropwithreset;
567 	}
568 
569 	/*
570 	 * If the ACK bit is off we drop the segment and return.
571 	 */
572 	if ((tiflags & TH_ACK) == 0)
573 		goto drop;
574 
575 	/*
576 	 * Ack processing.
577 	 */
578 	switch (tp->t_state) {
579 
580 	/*
581 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
582 	 * ESTABLISHED state and continue processing, othewise
583 	 * send an RST.
584 	 */
585 	case TCPS_SYN_RECEIVED:
586 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
587 		    SEQ_GT(ti->ti_ack, tp->snd_max))
588 			goto dropwithreset;
589 		tp->snd_una++;			/* SYN acked */
590 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
591 			tp->snd_nxt = tp->snd_una;
592 		tp->t_timer[TCPT_REXMT] = 0;
593 		soisconnected(so);
594 		tp->t_state = TCPS_ESTABLISHED;
595 		tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp));
596 		(void) tcp_reass(tp, (struct tcpiphdr *)0);
597 		tp->snd_wl1 = ti->ti_seq - 1;
598 		/* fall into ... */
599 
600 	/*
601 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
602 	 * ACKs.  If the ack is in the range
603 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
604 	 * then advance tp->snd_una to ti->ti_ack and drop
605 	 * data from the retransmission queue.  If this ACK reflects
606 	 * more up to date window information we update our window information.
607 	 */
608 	case TCPS_ESTABLISHED:
609 	case TCPS_FIN_WAIT_1:
610 	case TCPS_FIN_WAIT_2:
611 	case TCPS_CLOSE_WAIT:
612 	case TCPS_CLOSING:
613 	case TCPS_LAST_ACK:
614 	case TCPS_TIME_WAIT:
615 #define	ourfinisacked	(acked > 0)
616 
617 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una))
618 			break;
619 		if (SEQ_GT(ti->ti_ack, tp->snd_max))
620 			goto dropafterack;
621 		acked = ti->ti_ack - tp->snd_una;
622 
623 		/*
624 		 * If transmit timer is running and timed sequence
625 		 * number was acked, update smoothed round trip time.
626 		 */
627 		if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq)) {
628 			if (tp->t_srtt == 0)
629 				tp->t_srtt = tp->t_rtt;
630 			else
631 				tp->t_srtt =
632 				    tcp_alpha * tp->t_srtt +
633 				    (1 - tcp_alpha) * tp->t_rtt;
634 			tp->t_rtt = 0;
635 		}
636 
637 		/*
638 		 * If all outstanding data is acked, stop retransmit
639 		 * timer and remember to restart (more output or persist).
640 		 * If there is more data to be acked, restart retransmit
641 		 * timer.
642 		 */
643 		if (ti->ti_ack == tp->snd_max) {
644 			tp->t_timer[TCPT_REXMT] = 0;
645 			needoutput = 1;
646 		} else if (tp->t_timer[TCPT_PERSIST] == 0) {
647 			TCPT_RANGESET(tp->t_timer[TCPT_REXMT],
648 			    tcp_beta * tp->t_srtt, TCPTV_MIN, TCPTV_MAX);
649 			tp->t_rxtshift = 0;
650 		}
651 		/*
652 		 * When new data is acked, open the congestion window a bit.
653 		 */
654 		if (acked > 0)
655 			tp->snd_cwnd = MIN(11 * tp->snd_cwnd / 10, 65535);
656 		if (acked > so->so_snd.sb_cc) {
657 			tp->snd_wnd -= so->so_snd.sb_cc;
658 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
659 		} else {
660 			sbdrop(&so->so_snd, acked);
661 			tp->snd_wnd -= acked;
662 			acked = 0;
663 		}
664 		if ((so->so_snd.sb_flags & SB_WAIT) || so->so_snd.sb_sel)
665 			sowwakeup(so);
666 		tp->snd_una = ti->ti_ack;
667 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
668 			tp->snd_nxt = tp->snd_una;
669 
670 		switch (tp->t_state) {
671 
672 		/*
673 		 * In FIN_WAIT_1 STATE in addition to the processing
674 		 * for the ESTABLISHED state if our FIN is now acknowledged
675 		 * then enter FIN_WAIT_2.
676 		 */
677 		case TCPS_FIN_WAIT_1:
678 			if (ourfinisacked) {
679 				/*
680 				 * If we can't receive any more
681 				 * data, then closing user can proceed.
682 				 * Starting the timer is contrary to the
683 				 * specification, but if we don't get a FIN
684 				 * we'll hang forever.
685 				 */
686 				if (so->so_state & SS_CANTRCVMORE) {
687 					soisdisconnected(so);
688 					tp->t_timer[TCPT_2MSL] = TCPTV_MAXIDLE;
689 				}
690 				tp->t_state = TCPS_FIN_WAIT_2;
691 			}
692 			break;
693 
694 	 	/*
695 		 * In CLOSING STATE in addition to the processing for
696 		 * the ESTABLISHED state if the ACK acknowledges our FIN
697 		 * then enter the TIME-WAIT state, otherwise ignore
698 		 * the segment.
699 		 */
700 		case TCPS_CLOSING:
701 			if (ourfinisacked) {
702 				tp->t_state = TCPS_TIME_WAIT;
703 				tcp_canceltimers(tp);
704 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
705 				soisdisconnected(so);
706 			}
707 			break;
708 
709 		/*
710 		 * The only thing that can arrive in  LAST_ACK state
711 		 * is an acknowledgment of our FIN.  If our FIN is now
712 		 * acknowledged, delete the TCB, enter the closed state
713 		 * and return.
714 		 */
715 		case TCPS_LAST_ACK:
716 			if (ourfinisacked)
717 				tp = tcp_close(tp);
718 			goto drop;
719 
720 		/*
721 		 * In TIME_WAIT state the only thing that should arrive
722 		 * is a retransmission of the remote FIN.  Acknowledge
723 		 * it and restart the finack timer.
724 		 */
725 		case TCPS_TIME_WAIT:
726 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
727 			goto dropafterack;
728 		}
729 #undef ourfinisacked
730 	}
731 
732 step6:
733 	/*
734 	 * Update window information.
735 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
736 	 */
737 	if ((tiflags & TH_ACK) &&
738 	    (SEQ_LT(tp->snd_wl1, ti->ti_seq) || tp->snd_wl1 == ti->ti_seq &&
739 	    (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
740 	     tp->snd_wl2 == ti->ti_ack && ti->ti_win > tp->snd_wnd))) {
741 		tp->snd_wnd = ti->ti_win;
742 		tp->snd_wl1 = ti->ti_seq;
743 		tp->snd_wl2 = ti->ti_ack;
744 		if (tp->snd_wnd > tp->max_sndwnd)
745 			tp->max_sndwnd = tp->snd_wnd;
746 		needoutput = 1;
747 	}
748 
749 	/*
750 	 * Process segments with URG.
751 	 */
752 	if ((tiflags & TH_URG) && ti->ti_urp &&
753 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
754 		/*
755 		 * This is a kludge, but if we receive and accept
756 		 * random urgent pointers, we'll crash in
757 		 * soreceive.  It's hard to imagine someone
758 		 * actually wanting to send this much urgent data.
759 		 */
760 		if (ti->ti_urp + so->so_rcv.sb_cc > SB_MAX) {
761 			ti->ti_urp = 0;			/* XXX */
762 			tiflags &= ~TH_URG;		/* XXX */
763 			goto dodata;			/* XXX */
764 		}
765 		/*
766 		 * If this segment advances the known urgent pointer,
767 		 * then mark the data stream.  This should not happen
768 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
769 		 * a FIN has been received from the remote side.
770 		 * In these states we ignore the URG.
771 		 */
772 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
773 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
774 			so->so_oobmark = so->so_rcv.sb_cc +
775 			    (tp->rcv_up - tp->rcv_nxt) - 1;
776 			if (so->so_oobmark == 0)
777 				so->so_state |= SS_RCVATMARK;
778 			sohasoutofband(so);
779 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
780 		}
781 		/*
782 		 * Remove out of band data so doesn't get presented to user.
783 		 * This can happen independent of advancing the URG pointer,
784 		 * but if two URG's are pending at once, some out-of-band
785 		 * data may creep in... ick.
786 		 */
787 		if (ti->ti_urp <= ti->ti_len)
788 			tcp_pulloutofband(so, ti);
789 	} else
790 		/*
791 		 * If no out of band data is expected,
792 		 * pull receive urgent pointer along
793 		 * with the receive window.
794 		 */
795 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
796 			tp->rcv_up = tp->rcv_nxt;
797 dodata:							/* XXX */
798 
799 	/*
800 	 * Process the segment text, merging it into the TCP sequencing queue,
801 	 * and arranging for acknowledgment of receipt if necessary.
802 	 * This process logically involves adjusting tp->rcv_wnd as data
803 	 * is presented to the user (this happens in tcp_usrreq.c,
804 	 * case PRU_RCVD).  If a FIN has already been received on this
805 	 * connection then we just ignore the text.
806 	 */
807 	if ((ti->ti_len || (tiflags&TH_FIN)) &&
808 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
809 		TCP_REASS(tp, ti, m, so, tiflags);
810 		if (tcpnodelack == 0)
811 			tp->t_flags |= TF_DELACK;
812 		else
813 			tp->t_flags |= TF_ACKNOW;
814 		/*
815 		 * Note the amount of data that peer has sent into
816 		 * our window, in order to estimate the sender's
817 		 * buffer size.
818 		 */
819 		len = so->so_rcv.sb_hiwat - (tp->rcv_nxt - tp->rcv_adv);
820 		if (len > tp->max_rcvd)
821 			tp->max_rcvd = len;
822 	} else {
823 		m_freem(m);
824 		tiflags &= ~TH_FIN;
825 	}
826 
827 	/*
828 	 * If FIN is received ACK the FIN and let the user know
829 	 * that the connection is closing.
830 	 */
831 	if (tiflags & TH_FIN) {
832 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
833 			socantrcvmore(so);
834 			tp->t_flags |= TF_ACKNOW;
835 			tp->rcv_nxt++;
836 		}
837 		switch (tp->t_state) {
838 
839 	 	/*
840 		 * In SYN_RECEIVED and ESTABLISHED STATES
841 		 * enter the CLOSE_WAIT state.
842 		 */
843 		case TCPS_SYN_RECEIVED:
844 		case TCPS_ESTABLISHED:
845 			tp->t_state = TCPS_CLOSE_WAIT;
846 			break;
847 
848 	 	/*
849 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
850 		 * enter the CLOSING state.
851 		 */
852 		case TCPS_FIN_WAIT_1:
853 			tp->t_state = TCPS_CLOSING;
854 			break;
855 
856 	 	/*
857 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
858 		 * starting the time-wait timer, turning off the other
859 		 * standard timers.
860 		 */
861 		case TCPS_FIN_WAIT_2:
862 			tp->t_state = TCPS_TIME_WAIT;
863 			tcp_canceltimers(tp);
864 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
865 			soisdisconnected(so);
866 			break;
867 
868 		/*
869 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
870 		 */
871 		case TCPS_TIME_WAIT:
872 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
873 			break;
874 		}
875 	}
876 	if (so->so_options & SO_DEBUG)
877 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
878 
879 	/*
880 	 * Return any desired output.
881 	 */
882 	if (needoutput || (tp->t_flags & TF_ACKNOW))
883 		(void) tcp_output(tp);
884 	return;
885 
886 dropafterack:
887 	/*
888 	 * Generate an ACK dropping incoming segment if it occupies
889 	 * sequence space, where the ACK reflects our state.
890 	 */
891 	if (tiflags & TH_RST)
892 		goto drop;
893 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
894 		tcp_trace(TA_RESPOND, ostate, tp, &tcp_saveti, 0);
895 	tcp_respond(tp, ti, tp->rcv_nxt, tp->snd_nxt, TH_ACK);
896 	return;
897 
898 dropwithreset:
899 	if (om) {
900 		(void) m_free(om);
901 		om = 0;
902 	}
903 	/*
904 	 * Generate a RST, dropping incoming segment.
905 	 * Make ACK acceptable to originator of segment.
906 	 * Don't bother to respond if destination was broadcast.
907 	 */
908 	if ((tiflags & TH_RST) || in_broadcast(ti->ti_dst))
909 		goto drop;
910 	if (tiflags & TH_ACK)
911 		tcp_respond(tp, ti, (tcp_seq)0, ti->ti_ack, TH_RST);
912 	else {
913 		if (tiflags & TH_SYN)
914 			ti->ti_len++;
915 		tcp_respond(tp, ti, ti->ti_seq+ti->ti_len, (tcp_seq)0,
916 		    TH_RST|TH_ACK);
917 	}
918 	/* destroy temporarily created socket */
919 	if (dropsocket)
920 		(void) soabort(so);
921 	return;
922 
923 drop:
924 	if (om)
925 		(void) m_free(om);
926 	/*
927 	 * Drop space held by incoming segment and return.
928 	 */
929 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
930 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
931 	m_freem(m);
932 	/* destroy temporarily created socket */
933 	if (dropsocket)
934 		(void) soabort(so);
935 	return;
936 }
937 
938 tcp_dooptions(tp, om, ti)
939 	struct tcpcb *tp;
940 	struct mbuf *om;
941 	struct tcpiphdr *ti;
942 {
943 	register u_char *cp;
944 	int opt, optlen, cnt;
945 
946 	cp = mtod(om, u_char *);
947 	cnt = om->m_len;
948 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
949 		opt = cp[0];
950 		if (opt == TCPOPT_EOL)
951 			break;
952 		if (opt == TCPOPT_NOP)
953 			optlen = 1;
954 		else {
955 			optlen = cp[1];
956 			if (optlen <= 0)
957 				break;
958 		}
959 		switch (opt) {
960 
961 		default:
962 			break;
963 
964 		case TCPOPT_MAXSEG:
965 			if (optlen != 4)
966 				continue;
967 			if (!(ti->ti_flags & TH_SYN))
968 				continue;
969 			tp->t_maxseg = *(u_short *)(cp + 2);
970 			tp->t_maxseg = ntohs((u_short)tp->t_maxseg);
971 			tp->t_maxseg = MIN(tp->t_maxseg, tcp_mss(tp));
972 			break;
973 		}
974 	}
975 	(void) m_free(om);
976 }
977 
978 /*
979  * Pull out of band byte out of a segment so
980  * it doesn't appear in the user's data queue.
981  * It is still reflected in the segment length for
982  * sequencing purposes.
983  */
984 tcp_pulloutofband(so, ti)
985 	struct socket *so;
986 	struct tcpiphdr *ti;
987 {
988 	register struct mbuf *m;
989 	int cnt = ti->ti_urp - 1;
990 
991 	m = dtom(ti);
992 	while (cnt >= 0) {
993 		if (m->m_len > cnt) {
994 			char *cp = mtod(m, caddr_t) + cnt;
995 			struct tcpcb *tp = sototcpcb(so);
996 
997 			tp->t_iobc = *cp;
998 			tp->t_oobflags |= TCPOOB_HAVEDATA;
999 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1000 			m->m_len--;
1001 			return;
1002 		}
1003 		cnt -= m->m_len;
1004 		m = m->m_next;
1005 		if (m == 0)
1006 			break;
1007 	}
1008 	panic("tcp_pulloutofband");
1009 }
1010 
1011 /*
1012  *  Determine a reasonable value for maxseg size.
1013  *  If the route is known, use one that can be handled
1014  *  on the given interface without forcing IP to fragment.
1015  *  If bigger than a page (CLBYTES), round down to nearest pagesize
1016  *  to utilize pagesize mbufs.
1017  *  If interface pointer is unavailable, or the destination isn't local,
1018  *  use a conservative size (512 or the default IP max size, but no more
1019  *  than the mtu of the interface through which we route),
1020  *  as we can't discover anything about intervening gateways or networks.
1021  *
1022  *  This is ugly, and doesn't belong at this level, but has to happen somehow.
1023  */
1024 tcp_mss(tp)
1025 	register struct tcpcb *tp;
1026 {
1027 	struct route *ro;
1028 	struct ifnet *ifp;
1029 	int mss;
1030 	struct inpcb *inp;
1031 
1032 	inp = tp->t_inpcb;
1033 	ro = &inp->inp_route;
1034 	if ((ro->ro_rt == (struct rtentry *)0) ||
1035 	    (ifp = ro->ro_rt->rt_ifp) == (struct ifnet *)0) {
1036 		/* No route yet, so try to acquire one */
1037 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1038 			ro->ro_dst.sa_family = AF_INET;
1039 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1040 				inp->inp_faddr;
1041 			rtalloc(ro);
1042 		}
1043 		if ((ro->ro_rt == 0) || (ifp = ro->ro_rt->rt_ifp) == 0)
1044 			return (TCP_MSS);
1045 	}
1046 
1047 	mss = ifp->if_mtu - sizeof(struct tcpiphdr);
1048 #if	(CLBYTES & (CLBYTES - 1)) == 0
1049 	if (mss > CLBYTES)
1050 		mss &= ~(CLBYTES-1);
1051 #else
1052 	if (mss > CLBYTES)
1053 		mss = mss / CLBYTES * CLBYTES;
1054 #endif
1055 	if (in_localaddr(inp->inp_faddr))
1056 		return (mss);
1057 	return (MIN(mss, TCP_MSS));
1058 }
1059