xref: /netbsd-src/sys/netinet/tcp_input.c (revision 4d7e773266e3c3f48566c86c0ad52d51c6454fd1)
1 /*	$NetBSD: tcp_input.c,v 1.33 1997/10/10 01:51:07 explorer 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 /*
39  *	TODO list for SYN cache stuff:
40  *
41  *	(a) The definition of "struct syn_cache" says:
42  *
43  *		This structure should not exceeed 32 bytes.
44  *
45  *	    but it's 40 bytes on the Alpha.  Can reduce memory use one
46  *	    of two ways:
47  *
48  *		(1) Use a dynamically-sized hash table, and handle
49  *		    collisions by rehashing.  Then sc_next is unnecessary.
50  *
51  *		(2) Allocate syn_cache structures in pages (or some other
52  *		    large chunk).  This would probably be desirable for
53  *		    maintaining locality of reference anyway.
54  *
55  *		    If you do this, you can change sc_next to a page/index
56  *		    value, and make it a 32-bit (or maybe even 16-bit)
57  *		    integer, thus partly obviating the need for the previous
58  *		    hack.
59  *
60  *	    It's also worth noting this this is necessary for IPv6, as well,
61  *	    where we use 32 bytes just for the IP addresses, so eliminating
62  *	    wastage is going to become more important.  (BTW, has anyone
63  *	    integreated these changes with one fo the IPv6 status that are
64  *	    available?)
65  *
66  *	(b) Find room for a "state" field, which is needed to keep a
67  *	    compressed state for TIME_WAIT TCBs.  It's been noted already
68  *	    that this is fairly important for very high-volume web and
69  *	    mail servers, which use a large number of short-lived
70  *	    connections.
71  */
72 
73 #ifndef TUBA_INCLUDE
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/protosw.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/errno.h>
82 
83 #include <net/if.h>
84 #include <net/route.h>
85 
86 #include <netinet/in.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/ip.h>
89 #include <netinet/in_pcb.h>
90 #include <netinet/ip_var.h>
91 #include <netinet/tcp.h>
92 #include <netinet/tcp_fsm.h>
93 #include <netinet/tcp_seq.h>
94 #include <netinet/tcp_timer.h>
95 #include <netinet/tcp_var.h>
96 #include <netinet/tcpip.h>
97 #include <netinet/tcp_debug.h>
98 
99 #include <machine/stdarg.h>
100 
101 int	tcprexmtthresh = 3;
102 struct	tcpiphdr tcp_saveti;
103 
104 extern u_long sb_max;
105 
106 #endif /* TUBA_INCLUDE */
107 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
108 
109 /* for modulo comparisons of timestamps */
110 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
111 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
112 
113 /*
114  * Insert segment ti into reassembly queue of tcp with
115  * control block tp.  Return TH_FIN if reassembly now includes
116  * a segment with FIN.  The macro form does the common case inline
117  * (segment is the next to be received on an established connection,
118  * and the queue is empty), avoiding linkage into and removal
119  * from the queue and repetition of various conversions.
120  * Set DELACK for segments received in order, but ack immediately
121  * when segments are out of order (so fast retransmit can work).
122  */
123 #define	TCP_REASS(tp, ti, m, so, flags) { \
124 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
125 	    (tp)->segq.lh_first == NULL && \
126 	    (tp)->t_state == TCPS_ESTABLISHED) { \
127 		if ((ti)->ti_flags & TH_PUSH) \
128 			tp->t_flags |= TF_ACKNOW; \
129 		else \
130 			tp->t_flags |= TF_DELACK; \
131 		(tp)->rcv_nxt += (ti)->ti_len; \
132 		flags = (ti)->ti_flags & TH_FIN; \
133 		tcpstat.tcps_rcvpack++;\
134 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
135 		sbappend(&(so)->so_rcv, (m)); \
136 		sorwakeup(so); \
137 	} else { \
138 		(flags) = tcp_reass((tp), (ti), (m)); \
139 		tp->t_flags |= TF_ACKNOW; \
140 	} \
141 }
142 #ifndef TUBA_INCLUDE
143 
144 int
145 tcp_reass(tp, ti, m)
146 	register struct tcpcb *tp;
147 	register struct tcpiphdr *ti;
148 	struct mbuf *m;
149 {
150 	register struct ipqent *p, *q, *nq, *tiqe;
151 	struct socket *so = tp->t_inpcb->inp_socket;
152 	int flags;
153 
154 	/*
155 	 * Call with ti==0 after become established to
156 	 * force pre-ESTABLISHED data up to user socket.
157 	 */
158 	if (ti == 0)
159 		goto present;
160 
161 	/*
162 	 * Allocate a new queue entry, before we throw away any data.
163 	 * If we can't, just drop the packet.  XXX
164 	 */
165 	MALLOC(tiqe, struct ipqent *, sizeof (struct ipqent), M_IPQ, M_NOWAIT);
166 	if (tiqe == NULL) {
167 		tcpstat.tcps_rcvmemdrop++;
168 		m_freem(m);
169 		return (0);
170 	}
171 
172 	/*
173 	 * Find a segment which begins after this one does.
174 	 */
175 	for (p = NULL, q = tp->segq.lh_first; q != NULL;
176 	    p = q, q = q->ipqe_q.le_next)
177 		if (SEQ_GT(q->ipqe_tcp->ti_seq, ti->ti_seq))
178 			break;
179 
180 	/*
181 	 * If there is a preceding segment, it may provide some of
182 	 * our data already.  If so, drop the data from the incoming
183 	 * segment.  If it provides all of our data, drop us.
184 	 */
185 	if (p != NULL) {
186 		register struct tcpiphdr *phdr = p->ipqe_tcp;
187 		register int i;
188 
189 		/* conversion to int (in i) handles seq wraparound */
190 		i = phdr->ti_seq + phdr->ti_len - ti->ti_seq;
191 		if (i > 0) {
192 			if (i >= ti->ti_len) {
193 				tcpstat.tcps_rcvduppack++;
194 				tcpstat.tcps_rcvdupbyte += ti->ti_len;
195 				m_freem(m);
196 				FREE(tiqe, M_IPQ);
197 				return (0);
198 			}
199 			m_adj(m, i);
200 			ti->ti_len -= i;
201 			ti->ti_seq += i;
202 		}
203 	}
204 	tcpstat.tcps_rcvoopack++;
205 	tcpstat.tcps_rcvoobyte += ti->ti_len;
206 
207 	/*
208 	 * While we overlap succeeding segments trim them or,
209 	 * if they are completely covered, dequeue them.
210 	 */
211 	for (; q != NULL; q = nq) {
212 		register struct tcpiphdr *qhdr = q->ipqe_tcp;
213 		register int i = (ti->ti_seq + ti->ti_len) - qhdr->ti_seq;
214 
215 		if (i <= 0)
216 			break;
217 		if (i < qhdr->ti_len) {
218 			qhdr->ti_seq += i;
219 			qhdr->ti_len -= i;
220 			m_adj(q->ipqe_m, i);
221 			break;
222 		}
223 		nq = q->ipqe_q.le_next;
224 		m_freem(q->ipqe_m);
225 		LIST_REMOVE(q, ipqe_q);
226 		FREE(q, M_IPQ);
227 	}
228 
229 	/* Insert the new fragment queue entry into place. */
230 	tiqe->ipqe_m = m;
231 	tiqe->ipqe_tcp = ti;
232 	if (p == NULL) {
233 		LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
234 	} else {
235 		LIST_INSERT_AFTER(p, tiqe, ipqe_q);
236 	}
237 
238 present:
239 	/*
240 	 * Present data to user, advancing rcv_nxt through
241 	 * completed sequence space.
242 	 */
243 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
244 		return (0);
245 	q = tp->segq.lh_first;
246 	if (q == NULL || q->ipqe_tcp->ti_seq != tp->rcv_nxt)
247 		return (0);
248 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->ti_len)
249 		return (0);
250 	do {
251 		tp->rcv_nxt += q->ipqe_tcp->ti_len;
252 		flags = q->ipqe_tcp->ti_flags & TH_FIN;
253 
254 		nq = q->ipqe_q.le_next;
255 		LIST_REMOVE(q, ipqe_q);
256 		if (so->so_state & SS_CANTRCVMORE)
257 			m_freem(q->ipqe_m);
258 		else
259 			sbappend(&so->so_rcv, q->ipqe_m);
260 		FREE(q, M_IPQ);
261 		q = nq;
262 	} while (q != NULL && q->ipqe_tcp->ti_seq == tp->rcv_nxt);
263 	sorwakeup(so);
264 	return (flags);
265 }
266 
267 /*
268  * TCP input routine, follows pages 65-76 of the
269  * protocol specification dated September, 1981 very closely.
270  */
271 void
272 #if __STDC__
273 tcp_input(struct mbuf *m, ...)
274 #else
275 tcp_input(m, va_alist)
276 	register struct mbuf *m;
277 #endif
278 {
279 	register struct tcpiphdr *ti;
280 	register struct inpcb *inp;
281 	caddr_t optp = NULL;
282 	int optlen = 0;
283 	int len, tlen, off, hdroptlen;
284 	register struct tcpcb *tp = 0;
285 	register int tiflags;
286 	struct socket *so = NULL;
287 	int todrop, acked, ourfinisacked, needoutput = 0;
288 	short ostate = 0;
289 	int iss = 0;
290 	u_long tiwin;
291 	struct tcp_opt_info opti;
292 	int iphlen;
293 	va_list ap;
294 
295 	va_start(ap, m);
296 	iphlen = va_arg(ap, int);
297 	va_end(ap);
298 
299 	tcpstat.tcps_rcvtotal++;
300 
301 	opti.ts_present = 0;
302 	opti.maxseg = 0;
303 
304 	/*
305 	 * Get IP and TCP header together in first mbuf.
306 	 * Note: IP leaves IP header in first mbuf.
307 	 */
308 	ti = mtod(m, struct tcpiphdr *);
309 	if (iphlen > sizeof (struct ip))
310 		ip_stripoptions(m, (struct mbuf *)0);
311 	if (m->m_len < sizeof (struct tcpiphdr)) {
312 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
313 			tcpstat.tcps_rcvshort++;
314 			return;
315 		}
316 		ti = mtod(m, struct tcpiphdr *);
317 	}
318 
319 	/*
320 	 * Checksum extended TCP header and data.
321 	 */
322 	tlen = ((struct ip *)ti)->ip_len;
323 	len = sizeof (struct ip) + tlen;
324 	bzero(ti->ti_x1, sizeof ti->ti_x1);
325 	ti->ti_len = (u_int16_t)tlen;
326 	HTONS(ti->ti_len);
327 	if ((ti->ti_sum = in_cksum(m, len)) != 0) {
328 		tcpstat.tcps_rcvbadsum++;
329 		goto drop;
330 	}
331 #endif /* TUBA_INCLUDE */
332 
333 	/*
334 	 * Check that TCP offset makes sense,
335 	 * pull out TCP options and adjust length.		XXX
336 	 */
337 	off = ti->ti_off << 2;
338 	if (off < sizeof (struct tcphdr) || off > tlen) {
339 		tcpstat.tcps_rcvbadoff++;
340 		goto drop;
341 	}
342 	tlen -= off;
343 	ti->ti_len = tlen;
344 	if (off > sizeof (struct tcphdr)) {
345 		if (m->m_len < sizeof(struct ip) + off) {
346 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
347 				tcpstat.tcps_rcvshort++;
348 				return;
349 			}
350 			ti = mtod(m, struct tcpiphdr *);
351 		}
352 		optlen = off - sizeof (struct tcphdr);
353 		optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
354 		/*
355 		 * Do quick retrieval of timestamp options ("options
356 		 * prediction?").  If timestamp is the only option and it's
357 		 * formatted as recommended in RFC 1323 appendix A, we
358 		 * quickly get the values now and not bother calling
359 		 * tcp_dooptions(), etc.
360 		 */
361 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
362 		     (optlen > TCPOLEN_TSTAMP_APPA &&
363 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
364 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
365 		     (ti->ti_flags & TH_SYN) == 0) {
366 			opti.ts_present = 1;
367 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
368 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
369 			optp = NULL;	/* we've parsed the options */
370 		}
371 	}
372 	tiflags = ti->ti_flags;
373 
374 	/*
375 	 * Convert TCP protocol specific fields to host format.
376 	 */
377 	NTOHL(ti->ti_seq);
378 	NTOHL(ti->ti_ack);
379 	NTOHS(ti->ti_win);
380 	NTOHS(ti->ti_urp);
381 
382 	/*
383 	 * Locate pcb for segment.
384 	 */
385 findpcb:
386 	inp = in_pcblookup_connect(&tcbtable, ti->ti_src, ti->ti_sport,
387 	    ti->ti_dst, ti->ti_dport);
388 	if (inp == 0) {
389 		++tcpstat.tcps_pcbhashmiss;
390 		inp = in_pcblookup_bind(&tcbtable, ti->ti_dst, ti->ti_dport);
391 		if (inp == 0) {
392 			++tcpstat.tcps_noport;
393 			goto dropwithreset;
394 		}
395 	}
396 
397 	/*
398 	 * If the state is CLOSED (i.e., TCB does not exist) then
399 	 * all data in the incoming segment is discarded.
400 	 * If the TCB exists but is in CLOSED state, it is embryonic,
401 	 * but should either do a listen or a connect soon.
402 	 */
403 	tp = intotcpcb(inp);
404 	if (tp == 0)
405 		goto dropwithreset;
406 	if (tp->t_state == TCPS_CLOSED)
407 		goto drop;
408 
409 	/* Unscale the window into a 32-bit value. */
410 	if ((tiflags & TH_SYN) == 0)
411 		tiwin = ti->ti_win << tp->snd_scale;
412 	else
413 		tiwin = ti->ti_win;
414 
415 	so = inp->inp_socket;
416 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
417 		if (so->so_options & SO_DEBUG) {
418 			ostate = tp->t_state;
419 			tcp_saveti = *ti;
420 		}
421 		if (so->so_options & SO_ACCEPTCONN) {
422   			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
423 				if (tiflags & TH_RST)
424 					syn_cache_reset(ti);
425 				else if (tiflags & TH_ACK) {
426 					so = syn_cache_get(so, m);
427 					if (so == NULL) {
428 						/*
429 						 * We don't have a SYN for
430 						 * this ACK; send an RST.
431 						 */
432 						tcpstat.tcps_badsyn++;
433 						tp = NULL;
434 						goto dropwithreset;
435 					} else if (so ==
436 					    (struct socket *)(-1)) {
437 						/*
438 						 * We were unable to create
439 						 * the connection.  If the
440 						 * 3-way handshake was
441 						 * completeed, and RST has
442 						 * been sent to the peer.
443 						 * Since the mbuf might be
444 						 * in use for the reply,
445 						 * do not free it.
446 						 */
447 						m = NULL;
448 					} else {
449 						/*
450 						 * We have created a
451 						 * full-blown connection.
452 						 */
453 						inp = sotoinpcb(so);
454 						tp = intotcpcb(inp);
455 						tiwin <<= tp->snd_scale;
456 						goto after_listen;
457 					}
458   				}
459   			} else {
460 				/*
461 				 * Received a SYN; create compressed
462 				 * TCP state for it.
463 				 */
464 				if (so->so_qlen <= so->so_qlimit &&
465 				    syn_cache_add(so, m, optp, optlen, &opti))
466 					m = NULL;
467 			}
468 			goto drop;
469 		}
470 	}
471 
472 after_listen:
473 #ifdef DIAGNOSTIC
474 	/*
475 	 * Should not happen now that all embryonic connections
476 	 * are handled with compressed state.
477 	 */
478 	if (tp->t_state == TCPS_LISTEN)
479 		panic("tcp_input: TCPS_LISTEN");
480 #endif
481 
482 	/*
483 	 * Segment received on connection.
484 	 * Reset idle time and keep-alive timer.
485 	 */
486 	tp->t_idle = 0;
487 	if (TCPS_HAVEESTABLISHED(tp->t_state))
488 		tp->t_timer[TCPT_KEEP] = tcp_keepidle;
489 
490 	/*
491 	 * Process options.
492 	 */
493 	if (optp)
494 		tcp_dooptions(tp, optp, optlen, ti, &opti);
495 
496 	/*
497 	 * Header prediction: check for the two common cases
498 	 * of a uni-directional data xfer.  If the packet has
499 	 * no control flags, is in-sequence, the window didn't
500 	 * change and we're not retransmitting, it's a
501 	 * candidate.  If the length is zero and the ack moved
502 	 * forward, we're the sender side of the xfer.  Just
503 	 * free the data acked & wake any higher level process
504 	 * that was blocked waiting for space.  If the length
505 	 * is non-zero and the ack didn't move, we're the
506 	 * receiver side.  If we're getting packets in-order
507 	 * (the reassembly queue is empty), add the data to
508 	 * the socket buffer and note that we need a delayed ack.
509 	 */
510 	if (tp->t_state == TCPS_ESTABLISHED &&
511 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
512 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
513 	    ti->ti_seq == tp->rcv_nxt &&
514 	    tiwin && tiwin == tp->snd_wnd &&
515 	    tp->snd_nxt == tp->snd_max) {
516 
517 		/*
518 		 * If last ACK falls within this segment's sequence numbers,
519 		 *  record the timestamp.
520 		 */
521 		if (opti.ts_present &&
522 		    SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
523 		    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
524 			tp->ts_recent_age = tcp_now;
525 			tp->ts_recent = opti.ts_val;
526 		}
527 
528 		if (ti->ti_len == 0) {
529 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
530 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
531 			    tp->snd_cwnd >= tp->snd_wnd &&
532 			    tp->t_dupacks < tcprexmtthresh) {
533 				/*
534 				 * this is a pure ack for outstanding data.
535 				 */
536 				++tcpstat.tcps_predack;
537 				if (opti.ts_present)
538 					tcp_xmit_timer(tp,
539 					    tcp_now-opti.ts_ecr+1);
540 				else if (tp->t_rtt &&
541 				    SEQ_GT(ti->ti_ack, tp->t_rtseq))
542 					tcp_xmit_timer(tp, tp->t_rtt);
543 				acked = ti->ti_ack - tp->snd_una;
544 				tcpstat.tcps_rcvackpack++;
545 				tcpstat.tcps_rcvackbyte += acked;
546 				sbdrop(&so->so_snd, acked);
547 				tp->snd_una = ti->ti_ack;
548 				m_freem(m);
549 
550 				/*
551 				 * If all outstanding data are acked, stop
552 				 * retransmit timer, otherwise restart timer
553 				 * using current (possibly backed-off) value.
554 				 * If process is waiting for space,
555 				 * wakeup/selwakeup/signal.  If data
556 				 * are ready to send, let tcp_output
557 				 * decide between more output or persist.
558 				 */
559 				if (tp->snd_una == tp->snd_max)
560 					tp->t_timer[TCPT_REXMT] = 0;
561 				else if (tp->t_timer[TCPT_PERSIST] == 0)
562 					tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
563 
564 				if (sb_notify(&so->so_snd))
565 					sowwakeup(so);
566 				if (so->so_snd.sb_cc)
567 					(void) tcp_output(tp);
568 				return;
569 			}
570 		} else if (ti->ti_ack == tp->snd_una &&
571 		    tp->segq.lh_first == NULL &&
572 		    ti->ti_len <= sbspace(&so->so_rcv)) {
573 			/*
574 			 * this is a pure, in-sequence data packet
575 			 * with nothing on the reassembly queue and
576 			 * we have enough buffer space to take it.
577 			 */
578 			++tcpstat.tcps_preddat;
579 			tp->rcv_nxt += ti->ti_len;
580 			tcpstat.tcps_rcvpack++;
581 			tcpstat.tcps_rcvbyte += ti->ti_len;
582 			/*
583 			 * Drop TCP, IP headers and TCP options then add data
584 			 * to socket buffer.
585 			 */
586 			m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
587 			m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
588 			sbappend(&so->so_rcv, m);
589 			sorwakeup(so);
590 			if (ti->ti_flags & TH_PUSH)
591 				tp->t_flags |= TF_ACKNOW;
592 			else
593 				tp->t_flags |= TF_DELACK;
594 			return;
595 		}
596 	}
597 
598 	/*
599 	 * Drop TCP, IP headers and TCP options.
600 	 */
601 	hdroptlen  = sizeof(struct tcpiphdr) + off - sizeof(struct tcphdr);
602 	m->m_data += hdroptlen;
603 	m->m_len  -= hdroptlen;
604 
605 	/*
606 	 * Calculate amount of space in receive window,
607 	 * and then do TCP input processing.
608 	 * Receive window is amount of space in rcv queue,
609 	 * but not less than advertised window.
610 	 */
611 	{ int win;
612 
613 	win = sbspace(&so->so_rcv);
614 	if (win < 0)
615 		win = 0;
616 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
617 	}
618 
619 	switch (tp->t_state) {
620 
621 	/*
622 	 * If the state is SYN_SENT:
623 	 *	if seg contains an ACK, but not for our SYN, drop the input.
624 	 *	if seg contains a RST, then drop the connection.
625 	 *	if seg does not contain SYN, then drop it.
626 	 * Otherwise this is an acceptable SYN segment
627 	 *	initialize tp->rcv_nxt and tp->irs
628 	 *	if seg contains ack then advance tp->snd_una
629 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
630 	 *	arrange for segment to be acked (eventually)
631 	 *	continue processing rest of data/controls, beginning with URG
632 	 */
633 	case TCPS_SYN_SENT:
634 		if ((tiflags & TH_ACK) &&
635 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
636 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
637 			goto dropwithreset;
638 		if (tiflags & TH_RST) {
639 			if (tiflags & TH_ACK)
640 				tp = tcp_drop(tp, ECONNREFUSED);
641 			goto drop;
642 		}
643 		if ((tiflags & TH_SYN) == 0)
644 			goto drop;
645 		if (tiflags & TH_ACK) {
646 			tp->snd_una = ti->ti_ack;
647 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
648 				tp->snd_nxt = tp->snd_una;
649 		}
650 		tp->t_timer[TCPT_REXMT] = 0;
651 		tp->irs = ti->ti_seq;
652 		tcp_rcvseqinit(tp);
653 		tp->t_flags |= TF_ACKNOW;
654 		tcp_mss_from_peer(tp, opti.maxseg);
655 		tcp_rmx_rtt(tp);
656 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
657 			tcpstat.tcps_connects++;
658 			soisconnected(so);
659 			tcp_established(tp);
660 			/* Do window scaling on this connection? */
661 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
662 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
663 				tp->snd_scale = tp->requested_s_scale;
664 				tp->rcv_scale = tp->request_r_scale;
665 			}
666 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
667 				(struct mbuf *)0);
668 			/*
669 			 * if we didn't have to retransmit the SYN,
670 			 * use its rtt as our initial srtt & rtt var.
671 			 */
672 			if (tp->t_rtt)
673 				tcp_xmit_timer(tp, tp->t_rtt);
674 		} else
675 			tp->t_state = TCPS_SYN_RECEIVED;
676 
677 		/*
678 		 * Advance ti->ti_seq to correspond to first data byte.
679 		 * If data, trim to stay within window,
680 		 * dropping FIN if necessary.
681 		 */
682 		ti->ti_seq++;
683 		if (ti->ti_len > tp->rcv_wnd) {
684 			todrop = ti->ti_len - tp->rcv_wnd;
685 			m_adj(m, -todrop);
686 			ti->ti_len = tp->rcv_wnd;
687 			tiflags &= ~TH_FIN;
688 			tcpstat.tcps_rcvpackafterwin++;
689 			tcpstat.tcps_rcvbyteafterwin += todrop;
690 		}
691 		tp->snd_wl1 = ti->ti_seq - 1;
692 		tp->rcv_up = ti->ti_seq;
693 		goto step6;
694 
695 	/*
696 	 * If the state is SYN_RECEIVED:
697 	 *	If seg contains an ACK, but not for our SYN, drop the input
698 	 *	and generate an RST.  See page 36, rfc793
699 	 */
700 	case TCPS_SYN_RECEIVED:
701 		if ((tiflags & TH_ACK) &&
702 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
703 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
704 			goto dropwithreset;
705 		break;
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 (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
719 	    TSTMP_LT(opti.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 = tcp_new_iss(tp, sizeof(struct tcpcb),
815 						  tp->rcv_nxt);
816 				tp = tcp_close(tp);
817 				/*
818 				 * We have already advanced the mbuf
819 				 * pointers past the IP+TCP headers and
820 				 * options.  Restore those pointers before
821 				 * attempting to use the TCP header again.
822 				 */
823 				m->m_data -= hdroptlen;
824 				m->m_len  += hdroptlen;
825 				goto findpcb;
826 			}
827 			/*
828 			 * If window is closed can only take segments at
829 			 * window edge, and have to drop data and PUSH from
830 			 * incoming segments.  Continue processing, but
831 			 * remember to ack.  Otherwise, drop segment
832 			 * and ack.
833 			 */
834 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
835 				tp->t_flags |= TF_ACKNOW;
836 				tcpstat.tcps_rcvwinprobe++;
837 			} else
838 				goto dropafterack;
839 		} else
840 			tcpstat.tcps_rcvbyteafterwin += todrop;
841 		m_adj(m, -todrop);
842 		ti->ti_len -= todrop;
843 		tiflags &= ~(TH_PUSH|TH_FIN);
844 	}
845 
846 	/*
847 	 * If last ACK falls within this segment's sequence numbers,
848 	 * record its timestamp.
849 	 */
850 	if (opti.ts_present && SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
851 	    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
852 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
853 		tp->ts_recent_age = tcp_now;
854 		tp->ts_recent = opti.ts_val;
855 	}
856 
857 	/*
858 	 * If the RST bit is set examine the state:
859 	 *    SYN_RECEIVED STATE:
860 	 *	If passive open, return to LISTEN state.
861 	 *	If active open, inform user that connection was refused.
862 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
863 	 *	Inform user that connection was reset, and close tcb.
864 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
865 	 *	Close the tcb.
866 	 */
867 	if (tiflags&TH_RST) switch (tp->t_state) {
868 
869 	case TCPS_SYN_RECEIVED:
870 		so->so_error = ECONNREFUSED;
871 		goto close;
872 
873 	case TCPS_ESTABLISHED:
874 	case TCPS_FIN_WAIT_1:
875 	case TCPS_FIN_WAIT_2:
876 	case TCPS_CLOSE_WAIT:
877 		so->so_error = ECONNRESET;
878 	close:
879 		tp->t_state = TCPS_CLOSED;
880 		tcpstat.tcps_drops++;
881 		tp = tcp_close(tp);
882 		goto drop;
883 
884 	case TCPS_CLOSING:
885 	case TCPS_LAST_ACK:
886 	case TCPS_TIME_WAIT:
887 		tp = tcp_close(tp);
888 		goto drop;
889 	}
890 
891 	/*
892 	 * If a SYN is in the window, then this is an
893 	 * error and we send an RST and drop the connection.
894 	 */
895 	if (tiflags & TH_SYN) {
896 		tp = tcp_drop(tp, ECONNRESET);
897 		goto dropwithreset;
898 	}
899 
900 	/*
901 	 * If the ACK bit is off we drop the segment and return.
902 	 */
903 	if ((tiflags & TH_ACK) == 0)
904 		goto drop;
905 
906 	/*
907 	 * Ack processing.
908 	 */
909 	switch (tp->t_state) {
910 
911 	/*
912 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
913 	 * ESTABLISHED state and continue processing, otherwise
914 	 * send an RST.
915 	 */
916 	case TCPS_SYN_RECEIVED:
917 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
918 		    SEQ_GT(ti->ti_ack, tp->snd_max))
919 			goto dropwithreset;
920 		tcpstat.tcps_connects++;
921 		soisconnected(so);
922 		tcp_established(tp);
923 		/* Do window scaling? */
924 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
925 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
926 			tp->snd_scale = tp->requested_s_scale;
927 			tp->rcv_scale = tp->request_r_scale;
928 		}
929 		(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
930 		tp->snd_wl1 = ti->ti_seq - 1;
931 		/* fall into ... */
932 
933 	/*
934 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
935 	 * ACKs.  If the ack is in the range
936 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
937 	 * then advance tp->snd_una to ti->ti_ack and drop
938 	 * data from the retransmission queue.  If this ACK reflects
939 	 * more up to date window information we update our window information.
940 	 */
941 	case TCPS_ESTABLISHED:
942 	case TCPS_FIN_WAIT_1:
943 	case TCPS_FIN_WAIT_2:
944 	case TCPS_CLOSE_WAIT:
945 	case TCPS_CLOSING:
946 	case TCPS_LAST_ACK:
947 	case TCPS_TIME_WAIT:
948 
949 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
950 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
951 				tcpstat.tcps_rcvdupack++;
952 				/*
953 				 * If we have outstanding data (other than
954 				 * a window probe), this is a completely
955 				 * duplicate ack (ie, window info didn't
956 				 * change), the ack is the biggest we've
957 				 * seen and we've seen exactly our rexmt
958 				 * threshhold of them, assume a packet
959 				 * has been dropped and retransmit it.
960 				 * Kludge snd_nxt & the congestion
961 				 * window so we send only this one
962 				 * packet.
963 				 *
964 				 * We know we're losing at the current
965 				 * window size so do congestion avoidance
966 				 * (set ssthresh to half the current window
967 				 * and pull our congestion window back to
968 				 * the new ssthresh).
969 				 *
970 				 * Dup acks mean that packets have left the
971 				 * network (they're now cached at the receiver)
972 				 * so bump cwnd by the amount in the receiver
973 				 * to keep a constant cwnd packets in the
974 				 * network.
975 				 */
976 				if (tp->t_timer[TCPT_REXMT] == 0 ||
977 				    ti->ti_ack != tp->snd_una)
978 					tp->t_dupacks = 0;
979 				else if (++tp->t_dupacks == tcprexmtthresh) {
980 					tcp_seq onxt = tp->snd_nxt;
981 					u_int win =
982 					    min(tp->snd_wnd, tp->snd_cwnd) / 2 /
983 						tp->t_maxseg;
984 
985 					if (win < 2)
986 						win = 2;
987 					tp->snd_ssthresh = win * tp->t_maxseg;
988 					tp->t_timer[TCPT_REXMT] = 0;
989 					tp->t_rtt = 0;
990 					tp->snd_nxt = ti->ti_ack;
991 					tp->snd_cwnd = tp->t_maxseg;
992 					(void) tcp_output(tp);
993 					tp->snd_cwnd = tp->snd_ssthresh +
994 					       tp->t_maxseg * tp->t_dupacks;
995 					if (SEQ_GT(onxt, tp->snd_nxt))
996 						tp->snd_nxt = onxt;
997 					goto drop;
998 				} else if (tp->t_dupacks > tcprexmtthresh) {
999 					tp->snd_cwnd += tp->t_maxseg;
1000 					(void) tcp_output(tp);
1001 					goto drop;
1002 				}
1003 			} else
1004 				tp->t_dupacks = 0;
1005 			break;
1006 		}
1007 		/*
1008 		 * If the congestion window was inflated to account
1009 		 * for the other side's cached packets, retract it.
1010 		 */
1011 		if (tp->t_dupacks >= tcprexmtthresh &&
1012 		    tp->snd_cwnd > tp->snd_ssthresh)
1013 			tp->snd_cwnd = tp->snd_ssthresh;
1014 		tp->t_dupacks = 0;
1015 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1016 			tcpstat.tcps_rcvacktoomuch++;
1017 			goto dropafterack;
1018 		}
1019 		acked = ti->ti_ack - tp->snd_una;
1020 		tcpstat.tcps_rcvackpack++;
1021 		tcpstat.tcps_rcvackbyte += acked;
1022 
1023 		/*
1024 		 * If we have a timestamp reply, update smoothed
1025 		 * round trip time.  If no timestamp is present but
1026 		 * transmit timer is running and timed sequence
1027 		 * number was acked, update smoothed round trip time.
1028 		 * Since we now have an rtt measurement, cancel the
1029 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1030 		 * Recompute the initial retransmit timer.
1031 		 */
1032 		if (opti.ts_present)
1033 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1);
1034 		else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1035 			tcp_xmit_timer(tp,tp->t_rtt);
1036 
1037 		/*
1038 		 * If all outstanding data is acked, stop retransmit
1039 		 * timer and remember to restart (more output or persist).
1040 		 * If there is more data to be acked, restart retransmit
1041 		 * timer, using current (possibly backed-off) value.
1042 		 */
1043 		if (ti->ti_ack == tp->snd_max) {
1044 			tp->t_timer[TCPT_REXMT] = 0;
1045 			needoutput = 1;
1046 		} else if (tp->t_timer[TCPT_PERSIST] == 0)
1047 			tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1048 		/*
1049 		 * When new data is acked, open the congestion window.
1050 		 * If the window gives us less than ssthresh packets
1051 		 * in flight, open exponentially (maxseg per packet).
1052 		 * Otherwise open linearly: maxseg per window
1053 		 * (maxseg^2 / cwnd per packet), plus a constant
1054 		 * fraction of a packet (maxseg/8) to help larger windows
1055 		 * open quickly enough.
1056 		 */
1057 		{
1058 		register u_int cw = tp->snd_cwnd;
1059 		register u_int incr = tp->t_maxseg;
1060 
1061 		if (cw > tp->snd_ssthresh)
1062 			incr = incr * incr / cw;
1063 		tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1064 		}
1065 		if (acked > so->so_snd.sb_cc) {
1066 			tp->snd_wnd -= so->so_snd.sb_cc;
1067 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1068 			ourfinisacked = 1;
1069 		} else {
1070 			sbdrop(&so->so_snd, acked);
1071 			tp->snd_wnd -= acked;
1072 			ourfinisacked = 0;
1073 		}
1074 		if (sb_notify(&so->so_snd))
1075 			sowwakeup(so);
1076 		tp->snd_una = ti->ti_ack;
1077 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1078 			tp->snd_nxt = tp->snd_una;
1079 
1080 		switch (tp->t_state) {
1081 
1082 		/*
1083 		 * In FIN_WAIT_1 STATE in addition to the processing
1084 		 * for the ESTABLISHED state if our FIN is now acknowledged
1085 		 * then enter FIN_WAIT_2.
1086 		 */
1087 		case TCPS_FIN_WAIT_1:
1088 			if (ourfinisacked) {
1089 				/*
1090 				 * If we can't receive any more
1091 				 * data, then closing user can proceed.
1092 				 * Starting the timer is contrary to the
1093 				 * specification, but if we don't get a FIN
1094 				 * we'll hang forever.
1095 				 */
1096 				if (so->so_state & SS_CANTRCVMORE) {
1097 					soisdisconnected(so);
1098 					tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1099 				}
1100 				tp->t_state = TCPS_FIN_WAIT_2;
1101 			}
1102 			break;
1103 
1104 	 	/*
1105 		 * In CLOSING STATE in addition to the processing for
1106 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1107 		 * then enter the TIME-WAIT state, otherwise ignore
1108 		 * the segment.
1109 		 */
1110 		case TCPS_CLOSING:
1111 			if (ourfinisacked) {
1112 				tp->t_state = TCPS_TIME_WAIT;
1113 				tcp_canceltimers(tp);
1114 				tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1115 				soisdisconnected(so);
1116 			}
1117 			break;
1118 
1119 		/*
1120 		 * In LAST_ACK, we may still be waiting for data to drain
1121 		 * and/or to be acked, as well as for the ack of our FIN.
1122 		 * If our FIN is now acknowledged, delete the TCB,
1123 		 * enter the closed state and return.
1124 		 */
1125 		case TCPS_LAST_ACK:
1126 			if (ourfinisacked) {
1127 				tp = tcp_close(tp);
1128 				goto drop;
1129 			}
1130 			break;
1131 
1132 		/*
1133 		 * In TIME_WAIT state the only thing that should arrive
1134 		 * is a retransmission of the remote FIN.  Acknowledge
1135 		 * it and restart the finack timer.
1136 		 */
1137 		case TCPS_TIME_WAIT:
1138 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1139 			goto dropafterack;
1140 		}
1141 	}
1142 
1143 step6:
1144 	/*
1145 	 * Update window information.
1146 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1147 	 */
1148 	if (((tiflags & TH_ACK) && SEQ_LT(tp->snd_wl1, ti->ti_seq)) ||
1149 	    (tp->snd_wl1 == ti->ti_seq && SEQ_LT(tp->snd_wl2, ti->ti_ack)) ||
1150 	    (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)) {
1151 		/* keep track of pure window updates */
1152 		if (ti->ti_len == 0 &&
1153 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1154 			tcpstat.tcps_rcvwinupd++;
1155 		tp->snd_wnd = tiwin;
1156 		tp->snd_wl1 = ti->ti_seq;
1157 		tp->snd_wl2 = ti->ti_ack;
1158 		if (tp->snd_wnd > tp->max_sndwnd)
1159 			tp->max_sndwnd = tp->snd_wnd;
1160 		needoutput = 1;
1161 	}
1162 
1163 	/*
1164 	 * Process segments with URG.
1165 	 */
1166 	if ((tiflags & TH_URG) && ti->ti_urp &&
1167 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1168 		/*
1169 		 * This is a kludge, but if we receive and accept
1170 		 * random urgent pointers, we'll crash in
1171 		 * soreceive.  It's hard to imagine someone
1172 		 * actually wanting to send this much urgent data.
1173 		 */
1174 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1175 			ti->ti_urp = 0;			/* XXX */
1176 			tiflags &= ~TH_URG;		/* XXX */
1177 			goto dodata;			/* XXX */
1178 		}
1179 		/*
1180 		 * If this segment advances the known urgent pointer,
1181 		 * then mark the data stream.  This should not happen
1182 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1183 		 * a FIN has been received from the remote side.
1184 		 * In these states we ignore the URG.
1185 		 *
1186 		 * According to RFC961 (Assigned Protocols),
1187 		 * the urgent pointer points to the last octet
1188 		 * of urgent data.  We continue, however,
1189 		 * to consider it to indicate the first octet
1190 		 * of data past the urgent section as the original
1191 		 * spec states (in one of two places).
1192 		 */
1193 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1194 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1195 			so->so_oobmark = so->so_rcv.sb_cc +
1196 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1197 			if (so->so_oobmark == 0)
1198 				so->so_state |= SS_RCVATMARK;
1199 			sohasoutofband(so);
1200 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1201 		}
1202 		/*
1203 		 * Remove out of band data so doesn't get presented to user.
1204 		 * This can happen independent of advancing the URG pointer,
1205 		 * but if two URG's are pending at once, some out-of-band
1206 		 * data may creep in... ick.
1207 		 */
1208 		if (ti->ti_urp <= (u_int16_t) ti->ti_len
1209 #ifdef SO_OOBINLINE
1210 		     && (so->so_options & SO_OOBINLINE) == 0
1211 #endif
1212 		     )
1213 			tcp_pulloutofband(so, ti, m);
1214 	} else
1215 		/*
1216 		 * If no out of band data is expected,
1217 		 * pull receive urgent pointer along
1218 		 * with the receive window.
1219 		 */
1220 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1221 			tp->rcv_up = tp->rcv_nxt;
1222 dodata:							/* XXX */
1223 
1224 	/*
1225 	 * Process the segment text, merging it into the TCP sequencing queue,
1226 	 * and arranging for acknowledgment of receipt if necessary.
1227 	 * This process logically involves adjusting tp->rcv_wnd as data
1228 	 * is presented to the user (this happens in tcp_usrreq.c,
1229 	 * case PRU_RCVD).  If a FIN has already been received on this
1230 	 * connection then we just ignore the text.
1231 	 */
1232 	if ((ti->ti_len || (tiflags & TH_FIN)) &&
1233 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1234 		TCP_REASS(tp, ti, m, so, tiflags);
1235 		/*
1236 		 * Note the amount of data that peer has sent into
1237 		 * our window, in order to estimate the sender's
1238 		 * buffer size.
1239 		 */
1240 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1241 	} else {
1242 		m_freem(m);
1243 		tiflags &= ~TH_FIN;
1244 	}
1245 
1246 	/*
1247 	 * If FIN is received ACK the FIN and let the user know
1248 	 * that the connection is closing.  Ignore a FIN received before
1249 	 * the connection is fully established.
1250 	 */
1251 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
1252 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1253 			socantrcvmore(so);
1254 			tp->t_flags |= TF_ACKNOW;
1255 			tp->rcv_nxt++;
1256 		}
1257 		switch (tp->t_state) {
1258 
1259 	 	/*
1260 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
1261 		 */
1262 		case TCPS_ESTABLISHED:
1263 			tp->t_state = TCPS_CLOSE_WAIT;
1264 			break;
1265 
1266 	 	/*
1267 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1268 		 * enter the CLOSING state.
1269 		 */
1270 		case TCPS_FIN_WAIT_1:
1271 			tp->t_state = TCPS_CLOSING;
1272 			break;
1273 
1274 	 	/*
1275 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1276 		 * starting the time-wait timer, turning off the other
1277 		 * standard timers.
1278 		 */
1279 		case TCPS_FIN_WAIT_2:
1280 			tp->t_state = TCPS_TIME_WAIT;
1281 			tcp_canceltimers(tp);
1282 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1283 			soisdisconnected(so);
1284 			break;
1285 
1286 		/*
1287 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1288 		 */
1289 		case TCPS_TIME_WAIT:
1290 			tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1291 			break;
1292 		}
1293 	}
1294 	if (so->so_options & SO_DEBUG)
1295 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1296 
1297 	/*
1298 	 * Return any desired output.
1299 	 */
1300 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1301 		(void) tcp_output(tp);
1302 	return;
1303 
1304 dropafterack:
1305 	/*
1306 	 * Generate an ACK dropping incoming segment if it occupies
1307 	 * sequence space, where the ACK reflects our state.
1308 	 */
1309 	if (tiflags & TH_RST)
1310 		goto drop;
1311 	m_freem(m);
1312 	tp->t_flags |= TF_ACKNOW;
1313 	(void) tcp_output(tp);
1314 	return;
1315 
1316 dropwithreset:
1317 	/*
1318 	 * Generate a RST, dropping incoming segment.
1319 	 * Make ACK acceptable to originator of segment.
1320 	 * Don't bother to respond if destination was broadcast/multicast.
1321 	 */
1322 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1323 	    IN_MULTICAST(ti->ti_dst.s_addr))
1324 		goto drop;
1325 	if (tiflags & TH_ACK)
1326 		(void)tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1327 	else {
1328 		if (tiflags & TH_SYN)
1329 			ti->ti_len++;
1330 		(void)tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1331 		    TH_RST|TH_ACK);
1332 	}
1333 	return;
1334 
1335 drop:
1336 	/*
1337 	 * Drop space held by incoming segment and return.
1338 	 */
1339 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1340 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1341 	m_freem(m);
1342 	return;
1343 #ifndef TUBA_INCLUDE
1344 }
1345 
1346 void
1347 tcp_dooptions(tp, cp, cnt, ti, oi)
1348 	struct tcpcb *tp;
1349 	u_char *cp;
1350 	int cnt;
1351 	struct tcpiphdr *ti;
1352 	struct tcp_opt_info *oi;
1353 {
1354 	u_int16_t mss;
1355 	int opt, optlen;
1356 
1357 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1358 		opt = cp[0];
1359 		if (opt == TCPOPT_EOL)
1360 			break;
1361 		if (opt == TCPOPT_NOP)
1362 			optlen = 1;
1363 		else {
1364 			optlen = cp[1];
1365 			if (optlen <= 0)
1366 				break;
1367 		}
1368 		switch (opt) {
1369 
1370 		default:
1371 			continue;
1372 
1373 		case TCPOPT_MAXSEG:
1374 			if (optlen != TCPOLEN_MAXSEG)
1375 				continue;
1376 			if (!(ti->ti_flags & TH_SYN))
1377 				continue;
1378 			bcopy(cp + 2, &mss, sizeof(mss));
1379 			oi->maxseg = ntohs(mss);
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 			oi->ts_present = 1;
1395 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
1396 			NTOHL(oi->ts_val);
1397 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
1398 			NTOHL(oi->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 = oi->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 << 2;
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 << 2;
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  * TCP compressed state engine.  Currently used to hold compressed
1524  * state for SYN_RECEIVED.
1525  */
1526 
1527 u_long	syn_cache_count;
1528 u_int32_t syn_hash1, syn_hash2;
1529 
1530 #define SYN_HASH(sa, sp, dp) \
1531 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
1532 				     ((u_int32_t)(sp)))^syn_hash2)) \
1533 	 & 0x7fffffff)
1534 
1535 #define	eptosp(ep, e, s)	((struct s *)((char *)(ep) - \
1536 			    ((char *)(&((struct s *)0)->e) - (char *)0)))
1537 
1538 #define	SYN_CACHE_RM(sc, p, scp) {					\
1539 	*(p) = (sc)->sc_next;						\
1540 	if ((sc)->sc_next)						\
1541 		(sc)->sc_next->sc_timer += (sc)->sc_timer;		\
1542 	else {								\
1543 		(scp)->sch_timer_sum -= (sc)->sc_timer;			\
1544 		if ((scp)->sch_timer_sum <= 0)				\
1545 			(scp)->sch_timer_sum = -1;			\
1546 		/* If need be, fix up the last pointer */		\
1547 		if ((scp)->sch_first)					\
1548 			(scp)->sch_last = eptosp(p, sc_next, syn_cache); \
1549 	}								\
1550 	(scp)->sch_length--;						\
1551 	syn_cache_count--;						\
1552 }
1553 
1554 void
1555 syn_cache_insert(sc, prevp, headp)
1556 	struct syn_cache *sc;
1557 	struct syn_cache ***prevp;
1558 	struct syn_cache_head **headp;
1559 {
1560 	struct syn_cache_head *scp, *scp2, *sce;
1561 	struct syn_cache *sc2;
1562 	static u_int timeo_val;
1563 	int s;
1564 
1565 	/* Initialize the hash secrets when adding the first entry */
1566 	if (syn_cache_count == 0) {
1567 		struct timeval tv;
1568 		microtime(&tv);
1569 		syn_hash1 = random() ^ (u_long)&sc;
1570 		syn_hash2 = random() ^ tv.tv_usec;
1571 	}
1572 
1573 	sc->sc_hash = SYN_HASH(&sc->sc_src, sc->sc_sport, sc->sc_dport);
1574 	sc->sc_next = NULL;
1575 	scp = &tcp_syn_cache[sc->sc_hash % tcp_syn_cache_size];
1576 	*headp = scp;
1577 
1578 	/*
1579 	 * Make sure that we don't overflow the per-bucket
1580 	 * limit or the total cache size limit.
1581 	 */
1582 	s = splsoftnet();
1583 	if (scp->sch_length >= tcp_syn_bucket_limit) {
1584 		tcpstat.tcps_sc_bucketoverflow++;
1585 		sc2 = scp->sch_first;
1586 		scp->sch_first = sc2->sc_next;
1587 		FREE(sc2, M_PCB);
1588 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
1589 		tcpstat.tcps_sc_overflowed++;
1590 		/*
1591 		 * The cache is full.  Toss the first (i.e, oldest)
1592 		 * element in this bucket.
1593 		 */
1594 		scp2 = scp;
1595 		if (scp2->sch_first == NULL) {
1596 			sce = &tcp_syn_cache[tcp_syn_cache_size];
1597 			for (++scp2; scp2 != scp; scp2++) {
1598 				if (scp2 >= sce)
1599 					scp2 = &tcp_syn_cache[0];
1600 				if (scp2->sch_first)
1601 					break;
1602 			}
1603 		}
1604 		sc2 = scp2->sch_first;
1605 		if (sc2 == NULL) {
1606 			FREE(sc, M_PCB);
1607 			return;
1608 		}
1609 		if ((scp2->sch_first = sc2->sc_next) == NULL)
1610 			scp2->sch_last = NULL;
1611 		else
1612 			sc2->sc_next->sc_timer += sc2->sc_timer;
1613 		FREE(sc2, M_PCB);
1614 	} else {
1615 		scp->sch_length++;
1616 		syn_cache_count++;
1617 	}
1618 	tcpstat.tcps_sc_added++;
1619 
1620 	/*
1621 	 * Put it into the bucket.
1622 	 */
1623 	if (scp->sch_first == NULL)
1624 		*prevp = &scp->sch_first;
1625 	else {
1626 		*prevp = &scp->sch_last->sc_next;
1627 		tcpstat.tcps_sc_collisions++;
1628 	}
1629 	**prevp = sc;
1630 	scp->sch_last = sc;
1631 
1632 	/*
1633 	 * If the timeout value has changed
1634 	 *   1) force it to fit in a u_char
1635 	 *   2) Run the timer routine to truncate all
1636 	 *	existing entries to the new timeout value.
1637 	 */
1638 	if (timeo_val != tcp_syn_cache_timeo) {
1639 		tcp_syn_cache_timeo = min(tcp_syn_cache_timeo, UCHAR_MAX);
1640 		if (timeo_val > tcp_syn_cache_timeo)
1641 			syn_cache_timer(timeo_val - tcp_syn_cache_timeo);
1642 		timeo_val = tcp_syn_cache_timeo;
1643 	}
1644 	if (scp->sch_timer_sum > 0)
1645 		sc->sc_timer = tcp_syn_cache_timeo - scp->sch_timer_sum;
1646 	else if (scp->sch_timer_sum == 0) {
1647 		/* When the bucket timer is 0, it is not in the cache queue.  */
1648 		scp->sch_headq = tcp_syn_cache_first;
1649 		tcp_syn_cache_first = scp;
1650 		sc->sc_timer = tcp_syn_cache_timeo;
1651 	}
1652 	scp->sch_timer_sum = tcp_syn_cache_timeo;
1653 	splx(s);
1654 }
1655 
1656 /*
1657  * Walk down the cache list, decrementing the timer of
1658  * the first element on each entry.  If the timer goes
1659  * to zero, remove it and all successive entries with
1660  * a zero timer.
1661  */
1662 void
1663 syn_cache_timer(interval)
1664 	int interval;
1665 {
1666 	struct syn_cache_head *scp, **pscp;
1667 	struct syn_cache *sc, *scn;
1668 	int n, s;
1669 
1670 	pscp = &tcp_syn_cache_first;
1671 	scp = tcp_syn_cache_first;
1672 	s = splsoftnet();
1673 	while (scp) {
1674 		/*
1675 		 * Remove any empty hash buckets
1676 		 * from the cache queue.
1677 		 */
1678 		if ((sc = scp->sch_first) == NULL) {
1679 			*pscp = scp->sch_headq;
1680 			scp->sch_headq = NULL;
1681 			scp->sch_timer_sum = 0;
1682 			scp->sch_first = scp->sch_last = NULL;
1683 			scp->sch_length = 0;
1684 			scp = *pscp;
1685 			continue;
1686 		}
1687 
1688 		scp->sch_timer_sum -= interval;
1689 		if (scp->sch_timer_sum <= 0)
1690 			scp->sch_timer_sum = -1;
1691 		n = interval;
1692 		while (sc->sc_timer <= n) {
1693 			n -= sc->sc_timer;
1694 			scn = sc->sc_next;
1695 			tcpstat.tcps_sc_timed_out++;
1696 			syn_cache_count--;
1697 			FREE(sc, M_PCB);
1698 			scp->sch_length--;
1699 			if ((sc = scn) == NULL)
1700 				break;
1701 		}
1702 		if ((scp->sch_first = sc) != NULL) {
1703 			sc->sc_timer -= n;
1704 			pscp = &scp->sch_headq;
1705 			scp = scp->sch_headq;
1706 		}
1707 	}
1708 	splx(s);
1709 }
1710 
1711 /*
1712  * Find an entry in the syn cache.
1713  */
1714 struct syn_cache *
1715 syn_cache_lookup(ti, prevp, headp)
1716 	struct tcpiphdr *ti;
1717 	struct syn_cache ***prevp;
1718 	struct syn_cache_head **headp;
1719 {
1720 	struct syn_cache *sc, **prev;
1721 	struct syn_cache_head *head;
1722 	u_int32_t hash;
1723 	int s;
1724 
1725 	hash = SYN_HASH(&ti->ti_src, ti->ti_sport, ti->ti_dport);
1726 
1727 	head = &tcp_syn_cache[hash % tcp_syn_cache_size];
1728 	*headp = head;
1729 	prev = &head->sch_first;
1730 	s = splsoftnet();
1731 	for (sc = head->sch_first; sc; prev = &sc->sc_next, sc = sc->sc_next) {
1732 		if (sc->sc_hash != hash)
1733 			continue;
1734 		if (sc->sc_src.s_addr == ti->ti_src.s_addr &&
1735 		    sc->sc_sport == ti->ti_sport &&
1736 		    sc->sc_dport == ti->ti_dport &&
1737 		    sc->sc_dst.s_addr == ti->ti_dst.s_addr) {
1738 			*prevp = prev;
1739 			splx(s);
1740 			return (sc);
1741 		}
1742 	}
1743 	splx(s);
1744 	return (NULL);
1745 }
1746 
1747 /*
1748  * This function gets called when we receive an ACK for a
1749  * socket in the LISTEN state.  We look up the connection
1750  * in the syn cache, and if its there, we pull it out of
1751  * the cache and turn it into a full-blown connection in
1752  * the SYN-RECEIVED state.
1753  *
1754  * The return values may not be immediately obvious, and their effects
1755  * can be subtle, so here they are:
1756  *
1757  *	NULL	SYN was not found in cache; caller should drop the
1758  *		packet and send an RST.
1759  *
1760  *	-1	We were unable to create the new connection, and are
1761  *		aborting it.  An ACK,RST is being sent to the peer
1762  *		(unless we got screwey sequence numbners; see below),
1763  *		because the 3-way handshake has been completed.  Caller
1764  *		should not free the mbuf, since we may be using it.  If
1765  *		we are not, we will free it.
1766  *
1767  *	Otherwise, the return value is a pointer to the new socket
1768  *	associated with the connection.
1769  */
1770 struct socket *
1771 syn_cache_get(so, m)
1772 	struct socket *so;
1773 	struct mbuf *m;
1774 {
1775 	struct syn_cache *sc, **sc_prev;
1776 	struct syn_cache_head *head;
1777 	register struct inpcb *inp;
1778 	register struct tcpcb *tp = 0;
1779 	register struct tcpiphdr *ti;
1780 	struct sockaddr_in *sin;
1781 	struct mbuf *am;
1782 	long win;
1783 	int s;
1784 
1785 	ti = mtod(m, struct tcpiphdr *);
1786 	s = splsoftnet();
1787 	if ((sc = syn_cache_lookup(ti, &sc_prev, &head)) == NULL) {
1788 		splx(s);
1789 		return (NULL);
1790 	}
1791 
1792 	win = sbspace(&so->so_rcv);
1793 	if (win > TCP_MAXWIN)
1794 		win = TCP_MAXWIN;
1795 
1796 	/*
1797 	 * Verify the sequence and ack numbers.
1798 	 */
1799 	if ((ti->ti_ack != sc->sc_iss + 1) ||
1800 	    SEQ_LEQ(ti->ti_seq, sc->sc_irs) ||
1801 	    SEQ_GT(ti->ti_seq, sc->sc_irs + 1 + win)) {
1802 		(void) syn_cache_respond(sc, m, ti, win, 0);
1803 		splx(s);
1804 		return ((struct socket *)(-1));
1805 	}
1806 
1807 	/* Remove this cache entry */
1808 	SYN_CACHE_RM(sc, sc_prev, head);
1809 	splx(s);
1810 
1811 	/*
1812 	 * Ok, create the full blown connection, and set things up
1813 	 * as they would have been set up if we had created the
1814 	 * connection when the SYN arrived.  If we can't create
1815 	 * the connection, abort it.
1816 	 */
1817 	so = sonewconn(so, SS_ISCONNECTED);
1818 	if (so == NULL)
1819 		goto resetandabort;
1820 
1821 	inp = sotoinpcb(so);
1822 	inp->inp_laddr = sc->sc_dst;
1823 	inp->inp_lport = sc->sc_dport;
1824 	in_pcbstate(inp, INP_BOUND);
1825 #if BSD>=43
1826 	inp->inp_options = ip_srcroute();
1827 #endif
1828 
1829 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
1830 	if (am == NULL) {
1831 		m_freem(m);
1832 		goto resetandabort;
1833 	}
1834 	am->m_len = sizeof(struct sockaddr_in);
1835 	sin = mtod(am, struct sockaddr_in *);
1836 	sin->sin_family = AF_INET;
1837 	sin->sin_len = sizeof(*sin);
1838 	sin->sin_addr = sc->sc_src;
1839 	sin->sin_port = sc->sc_sport;
1840 	bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
1841 	if (in_pcbconnect(inp, am)) {
1842 		(void) m_free(am);
1843 		m_freem(m);
1844 		goto resetandabort;
1845 	}
1846 	(void) m_free(am);
1847 
1848 	tp = intotcpcb(inp);
1849 	if (sc->sc_request_r_scale != 15) {
1850 		tp->requested_s_scale = sc->sc_requested_s_scale;
1851 		tp->request_r_scale = sc->sc_request_r_scale;
1852 		tp->snd_scale = sc->sc_requested_s_scale;
1853 		tp->rcv_scale = sc->sc_request_r_scale;
1854 		tp->t_flags |= TF_RCVD_SCALE;
1855 	}
1856 	if (sc->sc_tstmp)
1857 		tp->t_flags |= TF_RCVD_TSTMP;
1858 
1859 	tp->t_template = tcp_template(tp);
1860 	if (tp->t_template == 0) {
1861 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
1862 		so = NULL;
1863 		m_freem(m);
1864 		goto abort;
1865 	}
1866 
1867 	tp->iss = sc->sc_iss;
1868 	tp->irs = sc->sc_irs;
1869 	tcp_sendseqinit(tp);
1870 	tcp_rcvseqinit(tp);
1871 	tp->t_state = TCPS_SYN_RECEIVED;
1872 	tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
1873 	tcpstat.tcps_accepts++;
1874 
1875 	/* Initialize tp->t_ourmss before we deal with the peer's! */
1876 	tp->t_ourmss = sc->sc_ourmaxseg;
1877 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
1878 	tcp_rmx_rtt(tp);
1879 	tp->snd_wl1 = sc->sc_irs;
1880 	tp->rcv_up = sc->sc_irs + 1;
1881 
1882 	/*
1883 	 * This is what whould have happened in tcp_ouput() when
1884 	 * the SYN,ACK was sent.
1885 	 */
1886 	tp->snd_up = tp->snd_una;
1887 	tp->snd_max = tp->snd_nxt = tp->iss+1;
1888 	tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1889 	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
1890 		tp->rcv_adv = tp->rcv_nxt + win;
1891 	tp->last_ack_sent = tp->rcv_nxt;
1892 
1893 	tcpstat.tcps_sc_completed++;
1894 	FREE(sc, M_PCB);
1895 	return (so);
1896 
1897 resetandabort:
1898 	(void) tcp_respond(NULL, ti, m, ti->ti_seq+ti->ti_len,
1899 	    (tcp_seq)0, TH_RST|TH_ACK);
1900 abort:
1901 	if (so != NULL)
1902 		(void) soabort(so);
1903 	FREE(sc, M_PCB);
1904 	tcpstat.tcps_sc_aborted++;
1905 	return ((struct socket *)(-1));
1906 }
1907 
1908 /*
1909  * This function is called when we get a RST for a
1910  * non-existant connection, so that we can see if the
1911  * connection is in the syn cache.  If it is, zap it.
1912  */
1913 
1914 void
1915 syn_cache_reset(ti)
1916 	register struct tcpiphdr *ti;
1917 {
1918 	struct syn_cache *sc, **sc_prev;
1919 	struct syn_cache_head *head;
1920 	int s = splsoftnet();
1921 
1922 	if ((sc = syn_cache_lookup(ti, &sc_prev, &head)) == NULL) {
1923 		splx(s);
1924 		return;
1925 	}
1926 	if (SEQ_LT(ti->ti_seq,sc->sc_irs) ||
1927 	    SEQ_GT(ti->ti_seq, sc->sc_irs+1)) {
1928 		splx(s);
1929 		return;
1930 	}
1931 	SYN_CACHE_RM(sc, sc_prev, head);
1932 	splx(s);
1933 	tcpstat.tcps_sc_reset++;
1934 	FREE(sc, M_PCB);
1935 }
1936 
1937 void
1938 syn_cache_unreach(ip, th)
1939 	struct ip *ip;
1940 	struct tcphdr *th;
1941 {
1942 	struct syn_cache *sc, **sc_prev;
1943 	struct syn_cache_head *head;
1944 	struct tcpiphdr ti2;
1945 	int s;
1946 
1947 	ti2.ti_src.s_addr = ip->ip_dst.s_addr;
1948 	ti2.ti_dst.s_addr = ip->ip_src.s_addr;
1949 	ti2.ti_sport = th->th_dport;
1950 	ti2.ti_dport = th->th_sport;
1951 
1952 	s = splsoftnet();
1953 	if ((sc = syn_cache_lookup(&ti2, &sc_prev, &head)) == NULL) {
1954 		splx(s);
1955 		return;
1956 	}
1957 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
1958 	if (ntohl (th->th_seq) != sc->sc_iss) {
1959 		splx(s);
1960 		return;
1961 	}
1962 	SYN_CACHE_RM(sc, sc_prev, head);
1963 	splx(s);
1964 	tcpstat.tcps_sc_unreach++;
1965 	FREE(sc, M_PCB);
1966 }
1967 
1968 /*
1969  * Given a LISTEN socket and an inbound SYN request, add
1970  * this to the syn cache, and send back a segment:
1971  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1972  * to the source.
1973  *
1974  * XXX We don't properly handle SYN-with-data!
1975  */
1976 
1977 int
1978 syn_cache_add(so, m, optp, optlen, oi)
1979 	struct socket *so;
1980 	struct mbuf *m;
1981 	u_char *optp;
1982 	int optlen;
1983 	struct tcp_opt_info *oi;
1984 {
1985 	register struct tcpiphdr *ti;
1986 	struct tcpcb tb, *tp;
1987 	long win;
1988 	struct syn_cache *sc, **sc_prev;
1989 	struct syn_cache_head *scp;
1990 	extern int tcp_do_rfc1323;
1991 
1992 	tp = sototcpcb(so);
1993 	ti = mtod(m, struct tcpiphdr *);
1994 
1995 	/*
1996 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1997 	 * in_broadcast() should never return true on a received
1998 	 * packet with M_BCAST not set.
1999 	 */
2000 	if (m->m_flags & (M_BCAST|M_MCAST) ||
2001 	    IN_MULTICAST(ti->ti_src.s_addr) ||
2002 	    IN_MULTICAST(ti->ti_dst.s_addr))
2003 		return (0);
2004 
2005 	/*
2006 	 * Initialize some local state.
2007 	 */
2008 	win = sbspace(&so->so_rcv);
2009 	if (win > TCP_MAXWIN)
2010 		win = TCP_MAXWIN;
2011 
2012 	if (optp) {
2013 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
2014 		tcp_dooptions(&tb, optp, optlen, ti, oi);
2015 	} else
2016 		tb.t_flags = 0;
2017 
2018 	/*
2019 	 * See if we already have an entry for this connection.
2020 	 */
2021 	if ((sc = syn_cache_lookup(ti, &sc_prev, &scp)) != NULL) {
2022 		tcpstat.tcps_sc_dupesyn++;
2023 		if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) {
2024 			tcpstat.tcps_sndacks++;
2025 			tcpstat.tcps_sndtotal++;
2026 		}
2027 		return (1);
2028 	}
2029 
2030 	MALLOC(sc, struct syn_cache *, sizeof(*sc), M_PCB, M_NOWAIT);
2031 	if (sc == NULL)
2032 		return (0);
2033 	/*
2034 	 * Fill in the cache, and put the necessary TCP
2035 	 * options into the reply.
2036 	 */
2037 	sc->sc_src.s_addr = ti->ti_src.s_addr;
2038 	sc->sc_dst.s_addr = ti->ti_dst.s_addr;
2039 	sc->sc_sport = ti->ti_sport;
2040 	sc->sc_dport = ti->ti_dport;
2041 	sc->sc_irs = ti->ti_seq;
2042 	sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0);
2043 	sc->sc_peermaxseg = oi->maxseg;
2044 	sc->sc_ourmaxseg = tcp_mss_to_advertise(tp);
2045 	sc->sc_tstmp = (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP)) ? 1 : 0;
2046 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2047 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2048 		sc->sc_requested_s_scale = tb.requested_s_scale;
2049 		sc->sc_request_r_scale = 0;
2050 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
2051 		    TCP_MAXWIN << sc->sc_request_r_scale <
2052 		    so->so_rcv.sb_hiwat)
2053 			sc->sc_request_r_scale++;
2054 	} else {
2055 		sc->sc_requested_s_scale = 15;
2056 		sc->sc_request_r_scale = 15;
2057 	}
2058 	if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) {
2059 		syn_cache_insert(sc, &sc_prev, &scp);
2060 		tcpstat.tcps_sndacks++;
2061 		tcpstat.tcps_sndtotal++;
2062 	} else {
2063 		FREE(sc, M_PCB);
2064 		tcpstat.tcps_sc_dropped++;
2065 	}
2066 	return (1);
2067 }
2068 
2069 int
2070 syn_cache_respond(sc, m, ti, win, ts)
2071 	struct syn_cache *sc;
2072 	struct mbuf *m;
2073 	register struct tcpiphdr *ti;
2074 	long win;
2075 	u_long ts;
2076 {
2077 	u_int8_t *optp;
2078 	int optlen;
2079 
2080 	/*
2081 	 * Tack on the TCP options.  If there isn't enough trailing
2082 	 * space for them, move up the fixed header to make space.
2083 	 */
2084 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
2085 	    (sc->sc_tstmp ? TCPOLEN_TSTAMP_APPA : 0);
2086 	if (optlen > M_TRAILINGSPACE(m)) {
2087 		if (M_LEADINGSPACE(m) >= optlen) {
2088 			m->m_data -= optlen;
2089 			m->m_len += optlen;
2090 		} else {
2091 			struct mbuf *m0 = m;
2092 			if ((m = m_gethdr(M_DONTWAIT, MT_HEADER)) == NULL) {
2093 				m_freem(m0);
2094 				return (ENOBUFS);
2095 			}
2096 			MH_ALIGN(m, sizeof(*ti) + optlen);
2097 			m->m_next = m0; /* this gets freed below */
2098 		}
2099 		ovbcopy((caddr_t)ti, mtod(m, caddr_t), sizeof(*ti));
2100 		ti = mtod(m, struct tcpiphdr *);
2101 	}
2102 
2103 	optp = (u_int8_t *)(ti + 1);
2104 	optp[0] = TCPOPT_MAXSEG;
2105 	optp[1] = 4;
2106 	optp[2] = (sc->sc_ourmaxseg >> 8) & 0xff;
2107 	optp[3] = sc->sc_ourmaxseg & 0xff;
2108 	optlen = 4;
2109 
2110 	if (sc->sc_request_r_scale != 15) {
2111 		*((u_int32_t *)(optp + optlen)) = htonl(TCPOPT_NOP << 24 |
2112 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
2113 		    sc->sc_request_r_scale);
2114 		optlen += 4;
2115 	}
2116 
2117 	if (sc->sc_tstmp) {
2118 		u_int32_t *lp = (u_int32_t *)(optp + optlen);
2119 		/* Form timestamp option as shown in appendix A of RFC 1323. */
2120 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
2121 		*lp++ = htonl(tcp_now);
2122 		*lp   = htonl(ts);
2123 		optlen += TCPOLEN_TSTAMP_APPA;
2124 	}
2125 
2126 	/*
2127 	 * Toss any trailing mbufs.  No need to worry about
2128 	 * m_len and m_pkthdr.len, since tcp_respond() will
2129 	 * unconditionally set them.
2130 	 */
2131 	if (m->m_next) {
2132 		m_freem(m->m_next);
2133 		m->m_next = NULL;
2134   	}
2135 
2136 	/*
2137 	 * Fill in the fields that tcp_respond() will not touch, and
2138 	 * then send the response.
2139 	 */
2140 	ti->ti_off = (sizeof(struct tcphdr) + optlen) >> 2;
2141 	ti->ti_win = htons(win);
2142 	return (tcp_respond(NULL, ti, m, sc->sc_irs + 1, sc->sc_iss,
2143 	    TH_SYN|TH_ACK));
2144 }
2145 #endif /* TUBA_INCLUDE */
2146