xref: /netbsd-src/sys/netinet/tcp_input.c (revision dc306354b0b29af51801a7632f1e95265a68cd81)
1 /*	$NetBSD: tcp_input.c,v 1.72 1998/12/18 21:38:02 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
9  * Facility, NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
42  *	The Regents of the University of California.  All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  * 1. Redistributions of source code must retain the above copyright
48  *    notice, this list of conditions and the following disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  * 3. All advertising materials mentioning features or use of this software
53  *    must display the following acknowledgement:
54  *	This product includes software developed by the University of
55  *	California, Berkeley and its contributors.
56  * 4. Neither the name of the University nor the names of its contributors
57  *    may be used to endorse or promote products derived from this software
58  *    without specific prior written permission.
59  *
60  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
61  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
62  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
63  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
64  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
65  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
66  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
67  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
68  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
69  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70  * SUCH DAMAGE.
71  *
72  *	@(#)tcp_input.c	8.12 (Berkeley) 5/24/95
73  */
74 
75 /*
76  *	TODO list for SYN cache stuff:
77  *
78  *	Find room for a "state" field, which is needed to keep a
79  *	compressed state for TIME_WAIT TCBs.  It's been noted already
80  *	that this is fairly important for very high-volume web and
81  *	mail servers, which use a large number of short-lived
82  *	connections.
83  */
84 
85 #include <sys/param.h>
86 #include <sys/systm.h>
87 #include <sys/malloc.h>
88 #include <sys/mbuf.h>
89 #include <sys/protosw.h>
90 #include <sys/socket.h>
91 #include <sys/socketvar.h>
92 #include <sys/errno.h>
93 #include <sys/syslog.h>
94 #include <sys/pool.h>
95 
96 #include <net/if.h>
97 #include <net/route.h>
98 
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/ip.h>
102 #include <netinet/in_pcb.h>
103 #include <netinet/ip_var.h>
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_fsm.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet/tcpip.h>
110 #include <netinet/tcp_debug.h>
111 
112 #include <machine/stdarg.h>
113 
114 int	tcprexmtthresh = 3;
115 struct	tcpiphdr tcp_saveti;
116 
117 extern u_long sb_max;
118 
119 #define TCP_PAWS_IDLE	(24 * 24 * 60 * 60 * PR_SLOWHZ)
120 
121 /* for modulo comparisons of timestamps */
122 #define TSTMP_LT(a,b)	((int)((a)-(b)) < 0)
123 #define TSTMP_GEQ(a,b)	((int)((a)-(b)) >= 0)
124 
125 /*
126  * Macro to compute ACK transmission behavior.  Delay the ACK unless
127  * we have already delayed an ACK (must send an ACK every two segments).
128  * We also ACK immediately if we received a PUSH and the ACK-on-PUSH
129  * option is enabled.
130  */
131 #define	TCP_SETUP_ACK(tp, ti) \
132 do { \
133 	if ((tp)->t_flags & TF_DELACK || \
134 	    (tcp_ack_on_push && (ti)->ti_flags & TH_PUSH)) \
135 		tp->t_flags |= TF_ACKNOW; \
136 	else \
137 		TCP_SET_DELACK(tp); \
138 } while (0)
139 
140 /*
141  * Insert segment ti into reassembly queue of tcp with
142  * control block tp.  Return TH_FIN if reassembly now includes
143  * a segment with FIN.  The macro form does the common case inline
144  * (segment is the next to be received on an established connection,
145  * and the queue is empty), avoiding linkage into and removal
146  * from the queue and repetition of various conversions.
147  * Set DELACK for segments received in order, but ack immediately
148  * when segments are out of order (so fast retransmit can work).
149  */
150 #define	TCP_REASS(tp, ti, m, so, flags) { \
151 	TCP_REASS_LOCK((tp)); \
152 	if ((ti)->ti_seq == (tp)->rcv_nxt && \
153 	    (tp)->segq.lh_first == NULL && \
154 	    (tp)->t_state == TCPS_ESTABLISHED) { \
155 		TCP_SETUP_ACK(tp, ti); \
156 		(tp)->rcv_nxt += (ti)->ti_len; \
157 		flags = (ti)->ti_flags & TH_FIN; \
158 		tcpstat.tcps_rcvpack++;\
159 		tcpstat.tcps_rcvbyte += (ti)->ti_len;\
160 		sbappend(&(so)->so_rcv, (m)); \
161 		sorwakeup(so); \
162 	} else { \
163 		(flags) = tcp_reass((tp), (ti), (m)); \
164 		tp->t_flags |= TF_ACKNOW; \
165 	} \
166 	TCP_REASS_UNLOCK((tp)); \
167 }
168 
169 int
170 tcp_reass(tp, ti, m)
171 	register struct tcpcb *tp;
172 	register struct tcpiphdr *ti;
173 	struct mbuf *m;
174 {
175 	register struct ipqent *p, *q, *nq, *tiqe = NULL;
176 	struct socket *so = tp->t_inpcb->inp_socket;
177 	int pkt_flags;
178 	tcp_seq pkt_seq;
179 	unsigned pkt_len;
180 	u_long rcvpartdupbyte = 0;
181 	u_long rcvoobyte;
182 
183 	TCP_REASS_LOCK_CHECK(tp);
184 
185 	/*
186 	 * Call with ti==0 after become established to
187 	 * force pre-ESTABLISHED data up to user socket.
188 	 */
189 	if (ti == 0)
190 		goto present;
191 
192 	rcvoobyte = ti->ti_len;
193 	/*
194 	 * Copy these to local variables because the tcpiphdr
195 	 * gets munged while we are collapsing mbufs.
196 	 */
197 	pkt_seq = ti->ti_seq;
198 	pkt_len = ti->ti_len;
199 	pkt_flags = ti->ti_flags;
200 	/*
201 	 * Find a segment which begins after this one does.
202 	 */
203 	for (p = NULL, q = tp->segq.lh_first; q != NULL; q = nq) {
204 		nq = q->ipqe_q.le_next;
205 		/*
206 		 * If the received segment is just right after this
207 		 * fragment, merge the two together and then check
208 		 * for further overlaps.
209 		 */
210 		if (q->ipqe_seq + q->ipqe_len == pkt_seq) {
211 #ifdef TCPREASS_DEBUG
212 			printf("tcp_reass[%p]: concat %u:%u(%u) to %u:%u(%u)\n",
213 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
214 			       q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len);
215 #endif
216 			pkt_len += q->ipqe_len;
217 			pkt_flags |= q->ipqe_flags;
218 			pkt_seq = q->ipqe_seq;
219 			m_cat(q->ipqe_m, m);
220 			m = q->ipqe_m;
221 			goto free_ipqe;
222 		}
223 		/*
224 		 * If the received segment is completely past this
225 		 * fragment, we need to go the next fragment.
226 		 */
227 		if (SEQ_LT(q->ipqe_seq + q->ipqe_len, pkt_seq)) {
228 			p = q;
229 			continue;
230 		}
231 		/*
232 		 * If the fragment is past the received segment,
233 		 * it (or any following) can't be concatenated.
234 		 */
235 		if (SEQ_GT(q->ipqe_seq, pkt_seq + pkt_len))
236 			break;
237 		/*
238 		 * We've received all the data in this segment before.
239 		 * mark it as a duplicate and return.
240 		 */
241 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq) &&
242 		    SEQ_GEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
243 			tcpstat.tcps_rcvduppack++;
244 			tcpstat.tcps_rcvdupbyte += pkt_len;
245 			m_freem(m);
246 			if (tiqe != NULL)
247 				pool_put(&ipqent_pool, tiqe);
248 			return (0);
249 		}
250 		/*
251 		 * Received segment completely overlaps this fragment
252 		 * so we drop the fragment (this keeps the temporal
253 		 * ordering of segments correct).
254 		 */
255 		if (SEQ_GEQ(q->ipqe_seq, pkt_seq) &&
256 		    SEQ_LEQ(q->ipqe_seq + q->ipqe_len, pkt_seq + pkt_len)) {
257 			rcvpartdupbyte += q->ipqe_len;
258 			m_freem(q->ipqe_m);
259 			goto free_ipqe;
260 		}
261 		/*
262 		 * RX'ed segment extends past the end of the
263 		 * fragment.  Drop the overlapping bytes.  Then
264 		 * merge the fragment and segment then treat as
265 		 * a longer received packet.
266 		 */
267 		if (SEQ_LT(q->ipqe_seq, pkt_seq)
268 		    && SEQ_GT(q->ipqe_seq + q->ipqe_len, pkt_seq))  {
269 			int overlap = q->ipqe_seq + q->ipqe_len - pkt_seq;
270 #ifdef TCPREASS_DEBUG
271 			printf("tcp_reass[%p]: trim starting %d bytes of %u:%u(%u)\n",
272 			       tp, overlap,
273 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
274 #endif
275 			m_adj(m, overlap);
276 			rcvpartdupbyte += overlap;
277 			m_cat(q->ipqe_m, m);
278 			m = q->ipqe_m;
279 			pkt_seq = q->ipqe_seq;
280 			pkt_len += q->ipqe_len - overlap;
281 			rcvoobyte -= overlap;
282 			goto free_ipqe;
283 		}
284 		/*
285 		 * RX'ed segment extends past the front of the
286 		 * fragment.  Drop the overlapping bytes on the
287 		 * received packet.  The packet will then be
288 		 * contatentated with this fragment a bit later.
289 		 */
290 		if (SEQ_GT(q->ipqe_seq, pkt_seq)
291 		    && SEQ_LT(q->ipqe_seq, pkt_seq + pkt_len))  {
292 			int overlap = pkt_seq + pkt_len - q->ipqe_seq;
293 #ifdef TCPREASS_DEBUG
294 			printf("tcp_reass[%p]: trim trailing %d bytes of %u:%u(%u)\n",
295 			       tp, overlap,
296 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
297 #endif
298 			m_adj(m, -overlap);
299 			pkt_len -= overlap;
300 			rcvpartdupbyte += overlap;
301 			rcvoobyte -= overlap;
302 		}
303 		/*
304 		 * If the received segment immediates precedes this
305 		 * fragment then tack the fragment onto this segment
306 		 * and reinsert the data.
307 		 */
308 		if (q->ipqe_seq == pkt_seq + pkt_len) {
309 #ifdef TCPREASS_DEBUG
310 			printf("tcp_reass[%p]: append %u:%u(%u) to %u:%u(%u)\n",
311 			       tp, q->ipqe_seq, q->ipqe_seq + q->ipqe_len, q->ipqe_len,
312 			       pkt_seq, pkt_seq + pkt_len, pkt_len);
313 #endif
314 			pkt_len += q->ipqe_len;
315 			pkt_flags |= q->ipqe_flags;
316 			m_cat(m, q->ipqe_m);
317 			LIST_REMOVE(q, ipqe_q);
318 			LIST_REMOVE(q, ipqe_timeq);
319 			if (tiqe == NULL) {
320 			    tiqe = q;
321 			} else {
322 			    pool_put(&ipqent_pool, q);
323 			}
324 			break;
325 		}
326 		/*
327 		 * If the fragment is before the segment, remember it.
328 		 * When this loop is terminated, p will contain the
329 		 * pointer to fragment that is right before the received
330 		 * segment.
331 		 */
332 		if (SEQ_LEQ(q->ipqe_seq, pkt_seq))
333 			p = q;
334 
335 		continue;
336 
337 		/*
338 		 * This is a common operation.  It also will allow
339 		 * to save doing a malloc/free in most instances.
340 		 */
341 	  free_ipqe:
342 		LIST_REMOVE(q, ipqe_q);
343 		LIST_REMOVE(q, ipqe_timeq);
344 		if (tiqe == NULL) {
345 		    tiqe = q;
346 		} else {
347 		    pool_put(&ipqent_pool, q);
348 		}
349 	}
350 
351 	/*
352 	 * Allocate a new queue entry since the received segment did not
353 	 * collapse onto any other out-of-order block; thus we are allocating
354 	 * a new block.  If it had collapsed, tiqe would not be NULL and
355 	 * we would be reusing it.
356 	 * XXX If we can't, just drop the packet.  XXX
357 	 */
358 	if (tiqe == NULL) {
359 		tiqe = pool_get(&ipqent_pool, PR_NOWAIT);
360 		if (tiqe == NULL) {
361 			tcpstat.tcps_rcvmemdrop++;
362 			m_freem(m);
363 			return (0);
364 		}
365 	}
366 
367 	/*
368 	 * Update the counters.
369 	 */
370 	tcpstat.tcps_rcvoopack++;
371 	tcpstat.tcps_rcvoobyte += rcvoobyte;
372 	if (rcvpartdupbyte) {
373 	    tcpstat.tcps_rcvpartduppack++;
374 	    tcpstat.tcps_rcvpartdupbyte += rcvpartdupbyte;
375 	}
376 
377 	/*
378 	 * Insert the new fragment queue entry into both queues.
379 	 */
380 	tiqe->ipqe_m = m;
381 	tiqe->ipqe_seq = pkt_seq;
382 	tiqe->ipqe_len = pkt_len;
383 	tiqe->ipqe_flags = pkt_flags;
384 	if (p == NULL) {
385 		LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q);
386 #ifdef TCPREASS_DEBUG
387 		if (tiqe->ipqe_seq != tp->rcv_nxt)
388 			printf("tcp_reass[%p]: insert %u:%u(%u) at front\n",
389 			       tp, pkt_seq, pkt_seq + pkt_len, pkt_len);
390 #endif
391 	} else {
392 		LIST_INSERT_AFTER(p, tiqe, ipqe_q);
393 #ifdef TCPREASS_DEBUG
394 		printf("tcp_reass[%p]: insert %u:%u(%u) after %u:%u(%u)\n",
395 		       tp, pkt_seq, pkt_seq + pkt_len, pkt_len,
396 		       p->ipqe_seq, p->ipqe_seq + p->ipqe_len, p->ipqe_len);
397 #endif
398 	}
399 
400 	LIST_INSERT_HEAD(&tp->timeq, tiqe, ipqe_timeq);
401 
402 present:
403 	/*
404 	 * Present data to user, advancing rcv_nxt through
405 	 * completed sequence space.
406 	 */
407 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
408 		return (0);
409 	q = tp->segq.lh_first;
410 	if (q == NULL || q->ipqe_seq != tp->rcv_nxt)
411 		return (0);
412 	if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_len)
413 		return (0);
414 
415 	tp->rcv_nxt += q->ipqe_len;
416 	pkt_flags = q->ipqe_flags & TH_FIN;
417 
418 	LIST_REMOVE(q, ipqe_q);
419 	LIST_REMOVE(q, ipqe_timeq);
420 	if (so->so_state & SS_CANTRCVMORE)
421 		m_freem(q->ipqe_m);
422 	else
423 		sbappend(&so->so_rcv, q->ipqe_m);
424 	pool_put(&ipqent_pool, q);
425 	sorwakeup(so);
426 	return (pkt_flags);
427 }
428 
429 /*
430  * TCP input routine, follows pages 65-76 of the
431  * protocol specification dated September, 1981 very closely.
432  */
433 void
434 #if __STDC__
435 tcp_input(struct mbuf *m, ...)
436 #else
437 tcp_input(m, va_alist)
438 	register struct mbuf *m;
439 #endif
440 {
441 	register struct tcpiphdr *ti;
442 	register struct inpcb *inp;
443 	caddr_t optp = NULL;
444 	int optlen = 0;
445 	int len, tlen, off, hdroptlen;
446 	register struct tcpcb *tp = 0;
447 	register int tiflags;
448 	struct socket *so = NULL;
449 	int todrop, acked, ourfinisacked, needoutput = 0;
450 	short ostate = 0;
451 	int iss = 0;
452 	u_long tiwin;
453 	struct tcp_opt_info opti;
454 	int iphlen;
455 	va_list ap;
456 
457 	va_start(ap, m);
458 	iphlen = va_arg(ap, int);
459 	va_end(ap);
460 
461 	tcpstat.tcps_rcvtotal++;
462 
463 	opti.ts_present = 0;
464 	opti.maxseg = 0;
465 
466 	/*
467 	 * Get IP and TCP header together in first mbuf.
468 	 * Note: IP leaves IP header in first mbuf.
469 	 */
470 	ti = mtod(m, struct tcpiphdr *);
471 	if (iphlen > sizeof (struct ip))
472 		ip_stripoptions(m, (struct mbuf *)0);
473 	if (m->m_len < sizeof (struct tcpiphdr)) {
474 		if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
475 			tcpstat.tcps_rcvshort++;
476 			return;
477 		}
478 		ti = mtod(m, struct tcpiphdr *);
479 	}
480 
481 	/*
482 	 * Checksum extended TCP header and data.
483 	 */
484 	tlen = ((struct ip *)ti)->ip_len;
485 	len = sizeof (struct ip) + tlen;
486 	bzero(ti->ti_x1, sizeof ti->ti_x1);
487 	ti->ti_len = (u_int16_t)tlen;
488 	HTONS(ti->ti_len);
489 	if ((ti->ti_sum = in_cksum(m, len)) != 0) {
490 		tcpstat.tcps_rcvbadsum++;
491 		goto drop;
492 	}
493 
494 	/*
495 	 * Check that TCP offset makes sense,
496 	 * pull out TCP options and adjust length.		XXX
497 	 */
498 	off = ti->ti_off << 2;
499 	if (off < sizeof (struct tcphdr) || off > tlen) {
500 		tcpstat.tcps_rcvbadoff++;
501 		goto drop;
502 	}
503 	tlen -= off;
504 	ti->ti_len = tlen;
505 	if (off > sizeof (struct tcphdr)) {
506 		if (m->m_len < sizeof(struct ip) + off) {
507 			if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
508 				tcpstat.tcps_rcvshort++;
509 				return;
510 			}
511 			ti = mtod(m, struct tcpiphdr *);
512 		}
513 		optlen = off - sizeof (struct tcphdr);
514 		optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
515 		/*
516 		 * Do quick retrieval of timestamp options ("options
517 		 * prediction?").  If timestamp is the only option and it's
518 		 * formatted as recommended in RFC 1323 appendix A, we
519 		 * quickly get the values now and not bother calling
520 		 * tcp_dooptions(), etc.
521 		 */
522 		if ((optlen == TCPOLEN_TSTAMP_APPA ||
523 		     (optlen > TCPOLEN_TSTAMP_APPA &&
524 			optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
525 		     *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
526 		     (ti->ti_flags & TH_SYN) == 0) {
527 			opti.ts_present = 1;
528 			opti.ts_val = ntohl(*(u_int32_t *)(optp + 4));
529 			opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8));
530 			optp = NULL;	/* we've parsed the options */
531 		}
532 	}
533 	tiflags = ti->ti_flags;
534 
535 	/*
536 	 * Convert TCP protocol specific fields to host format.
537 	 */
538 	NTOHL(ti->ti_seq);
539 	NTOHL(ti->ti_ack);
540 	NTOHS(ti->ti_win);
541 	NTOHS(ti->ti_urp);
542 
543 	/*
544 	 * Locate pcb for segment.
545 	 */
546 findpcb:
547 	inp = in_pcblookup_connect(&tcbtable, ti->ti_src, ti->ti_sport,
548 	    ti->ti_dst, ti->ti_dport);
549 	if (inp == 0) {
550 		++tcpstat.tcps_pcbhashmiss;
551 		inp = in_pcblookup_bind(&tcbtable, ti->ti_dst, ti->ti_dport);
552 		if (inp == 0) {
553 			++tcpstat.tcps_noport;
554 			goto dropwithreset;
555 		}
556 	}
557 
558 	/*
559 	 * If the state is CLOSED (i.e., TCB does not exist) then
560 	 * all data in the incoming segment is discarded.
561 	 * If the TCB exists but is in CLOSED state, it is embryonic,
562 	 * but should either do a listen or a connect soon.
563 	 */
564 	tp = intotcpcb(inp);
565 	if (tp == 0)
566 		goto dropwithreset;
567 	if (tp->t_state == TCPS_CLOSED)
568 		goto drop;
569 
570 	/* Unscale the window into a 32-bit value. */
571 	if ((tiflags & TH_SYN) == 0)
572 		tiwin = ti->ti_win << tp->snd_scale;
573 	else
574 		tiwin = ti->ti_win;
575 
576 	so = inp->inp_socket;
577 	if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
578 		if (so->so_options & SO_DEBUG) {
579 			ostate = tp->t_state;
580 			tcp_saveti = *ti;
581 		}
582 		if (so->so_options & SO_ACCEPTCONN) {
583   			if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
584 				if (tiflags & TH_RST) {
585 					syn_cache_reset(ti);
586 				} else if ((tiflags & (TH_ACK|TH_SYN)) ==
587 				    (TH_ACK|TH_SYN)) {
588 					/*
589 					 * Received a SYN,ACK.  This should
590 					 * never happen while we are in
591 					 * LISTEN.  Send an RST.
592 					 */
593 					goto badsyn;
594 				} else if (tiflags & TH_ACK) {
595 					so = syn_cache_get(so, m);
596 					if (so == NULL) {
597 						/*
598 						 * We don't have a SYN for
599 						 * this ACK; send an RST.
600 						 */
601 						goto badsyn;
602 					} else if (so ==
603 					    (struct socket *)(-1)) {
604 						/*
605 						 * We were unable to create
606 						 * the connection.  If the
607 						 * 3-way handshake was
608 						 * completed, and RST has
609 						 * been sent to the peer.
610 						 * Since the mbuf might be
611 						 * in use for the reply,
612 						 * do not free it.
613 						 */
614 						m = NULL;
615 					} else {
616 						/*
617 						 * We have created a
618 						 * full-blown connection.
619 						 */
620 						inp = sotoinpcb(so);
621 						tp = intotcpcb(inp);
622 						tiwin <<= tp->snd_scale;
623 						goto after_listen;
624 					}
625   				} else {
626 					/*
627 					 * None of RST, SYN or ACK was set.
628 					 * This is an invalid packet for a
629 					 * TCB in LISTEN state.  Send a RST.
630 					 */
631 					goto badsyn;
632 				}
633   			} else {
634 				/*
635 				 * Received a SYN.
636 				 */
637 				if (in_hosteq(ti->ti_src, ti->ti_dst) &&
638 				    ti->ti_sport == ti->ti_dport) {
639 					/*
640 					 * LISTEN socket received a SYN
641 					 * from itself?  This can't possibly
642 					 * be valid; drop the packet.
643 					 */
644 					tcpstat.tcps_badsyn++;
645 					goto drop;
646 				}
647 				/*
648 				 * SYN looks ok; create compressed TCP
649 				 * state for it.
650 				 */
651 				if (so->so_qlen <= so->so_qlimit &&
652 				    syn_cache_add(so, m, optp, optlen, &opti))
653 					m = NULL;
654 			}
655 			goto drop;
656 		}
657 	}
658 
659 after_listen:
660 #ifdef DIAGNOSTIC
661 	/*
662 	 * Should not happen now that all embryonic connections
663 	 * are handled with compressed state.
664 	 */
665 	if (tp->t_state == TCPS_LISTEN)
666 		panic("tcp_input: TCPS_LISTEN");
667 #endif
668 
669 	/*
670 	 * Segment received on connection.
671 	 * Reset idle time and keep-alive timer.
672 	 */
673 	tp->t_idle = 0;
674 	if (TCPS_HAVEESTABLISHED(tp->t_state))
675 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
676 
677 	/*
678 	 * Process options.
679 	 */
680 	if (optp)
681 		tcp_dooptions(tp, optp, optlen, ti, &opti);
682 
683 	/*
684 	 * Header prediction: check for the two common cases
685 	 * of a uni-directional data xfer.  If the packet has
686 	 * no control flags, is in-sequence, the window didn't
687 	 * change and we're not retransmitting, it's a
688 	 * candidate.  If the length is zero and the ack moved
689 	 * forward, we're the sender side of the xfer.  Just
690 	 * free the data acked & wake any higher level process
691 	 * that was blocked waiting for space.  If the length
692 	 * is non-zero and the ack didn't move, we're the
693 	 * receiver side.  If we're getting packets in-order
694 	 * (the reassembly queue is empty), add the data to
695 	 * the socket buffer and note that we need a delayed ack.
696 	 */
697 	if (tp->t_state == TCPS_ESTABLISHED &&
698 	    (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
699 	    (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) &&
700 	    ti->ti_seq == tp->rcv_nxt &&
701 	    tiwin && tiwin == tp->snd_wnd &&
702 	    tp->snd_nxt == tp->snd_max) {
703 
704 		/*
705 		 * If last ACK falls within this segment's sequence numbers,
706 		 *  record the timestamp.
707 		 */
708 		if (opti.ts_present &&
709 		    SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
710 		    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len)) {
711 			tp->ts_recent_age = tcp_now;
712 			tp->ts_recent = opti.ts_val;
713 		}
714 
715 		if (ti->ti_len == 0) {
716 			if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
717 			    SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
718 			    tp->snd_cwnd >= tp->snd_wnd &&
719 			    tp->t_dupacks < tcprexmtthresh) {
720 				/*
721 				 * this is a pure ack for outstanding data.
722 				 */
723 				++tcpstat.tcps_predack;
724 				if (opti.ts_present)
725 					tcp_xmit_timer(tp,
726 					    tcp_now-opti.ts_ecr+1);
727 				else if (tp->t_rtt &&
728 				    SEQ_GT(ti->ti_ack, tp->t_rtseq))
729 					tcp_xmit_timer(tp, tp->t_rtt);
730 				acked = ti->ti_ack - tp->snd_una;
731 				tcpstat.tcps_rcvackpack++;
732 				tcpstat.tcps_rcvackbyte += acked;
733 				sbdrop(&so->so_snd, acked);
734 				tp->snd_una = ti->ti_ack;
735 				m_freem(m);
736 
737 				/*
738 				 * If all outstanding data are acked, stop
739 				 * retransmit timer, otherwise restart timer
740 				 * using current (possibly backed-off) value.
741 				 * If process is waiting for space,
742 				 * wakeup/selwakeup/signal.  If data
743 				 * are ready to send, let tcp_output
744 				 * decide between more output or persist.
745 				 */
746 				if (tp->snd_una == tp->snd_max)
747 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
748 				else if (TCP_TIMER_ISARMED(tp,
749 				    TCPT_PERSIST) == 0)
750 					TCP_TIMER_ARM(tp, TCPT_REXMT,
751 					    tp->t_rxtcur);
752 
753 				sowwakeup(so);
754 				if (so->so_snd.sb_cc)
755 					(void) tcp_output(tp);
756 				return;
757 			}
758 		} else if (ti->ti_ack == tp->snd_una &&
759 		    tp->segq.lh_first == NULL &&
760 		    ti->ti_len <= sbspace(&so->so_rcv)) {
761 			/*
762 			 * this is a pure, in-sequence data packet
763 			 * with nothing on the reassembly queue and
764 			 * we have enough buffer space to take it.
765 			 */
766 			++tcpstat.tcps_preddat;
767 			tp->rcv_nxt += ti->ti_len;
768 			tcpstat.tcps_rcvpack++;
769 			tcpstat.tcps_rcvbyte += ti->ti_len;
770 			/*
771 			 * Drop TCP, IP headers and TCP options then add data
772 			 * to socket buffer.
773 			 */
774 			m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
775 			m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
776 			sbappend(&so->so_rcv, m);
777 			sorwakeup(so);
778 			TCP_SETUP_ACK(tp, ti);
779 			if (tp->t_flags & TF_ACKNOW)
780 				(void) tcp_output(tp);
781 			return;
782 		}
783 	}
784 
785 	/*
786 	 * Drop TCP, IP headers and TCP options.
787 	 */
788 	hdroptlen  = sizeof(struct tcpiphdr) + off - sizeof(struct tcphdr);
789 	m->m_data += hdroptlen;
790 	m->m_len  -= hdroptlen;
791 
792 	/*
793 	 * Calculate amount of space in receive window,
794 	 * and then do TCP input processing.
795 	 * Receive window is amount of space in rcv queue,
796 	 * but not less than advertised window.
797 	 */
798 	{ int win;
799 
800 	win = sbspace(&so->so_rcv);
801 	if (win < 0)
802 		win = 0;
803 	tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
804 	}
805 
806 	switch (tp->t_state) {
807 
808 	/*
809 	 * If the state is SYN_SENT:
810 	 *	if seg contains an ACK, but not for our SYN, drop the input.
811 	 *	if seg contains a RST, then drop the connection.
812 	 *	if seg does not contain SYN, then drop it.
813 	 * Otherwise this is an acceptable SYN segment
814 	 *	initialize tp->rcv_nxt and tp->irs
815 	 *	if seg contains ack then advance tp->snd_una
816 	 *	if SYN has been acked change to ESTABLISHED else SYN_RCVD state
817 	 *	arrange for segment to be acked (eventually)
818 	 *	continue processing rest of data/controls, beginning with URG
819 	 */
820 	case TCPS_SYN_SENT:
821 		if ((tiflags & TH_ACK) &&
822 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
823 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
824 			goto dropwithreset;
825 		if (tiflags & TH_RST) {
826 			if (tiflags & TH_ACK)
827 				tp = tcp_drop(tp, ECONNREFUSED);
828 			goto drop;
829 		}
830 		if ((tiflags & TH_SYN) == 0)
831 			goto drop;
832 		if (tiflags & TH_ACK) {
833 			tp->snd_una = ti->ti_ack;
834 			if (SEQ_LT(tp->snd_nxt, tp->snd_una))
835 				tp->snd_nxt = tp->snd_una;
836 		}
837 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
838 		tp->irs = ti->ti_seq;
839 		tcp_rcvseqinit(tp);
840 		tp->t_flags |= TF_ACKNOW;
841 		tcp_mss_from_peer(tp, opti.maxseg);
842 
843 		/*
844 		 * Initialize the initial congestion window.  If we
845 		 * had to retransmit the SYN, we must initialize cwnd
846 		 * to 1 segment (i.e. the Loss Window).
847 		 */
848 		if (tp->t_flags & TF_SYN_REXMT)
849 			tp->snd_cwnd = tp->t_peermss;
850 		else
851 			tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
852 			    tp->t_peermss);
853 
854 		tcp_rmx_rtt(tp);
855 		if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
856 			tcpstat.tcps_connects++;
857 			soisconnected(so);
858 			tcp_established(tp);
859 			/* Do window scaling on this connection? */
860 			if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
861 				(TF_RCVD_SCALE|TF_REQ_SCALE)) {
862 				tp->snd_scale = tp->requested_s_scale;
863 				tp->rcv_scale = tp->request_r_scale;
864 			}
865 			TCP_REASS_LOCK(tp);
866 			(void) tcp_reass(tp, (struct tcpiphdr *)0,
867 				(struct mbuf *)0);
868 			TCP_REASS_UNLOCK(tp);
869 			/*
870 			 * if we didn't have to retransmit the SYN,
871 			 * use its rtt as our initial srtt & rtt var.
872 			 */
873 			if (tp->t_rtt)
874 				tcp_xmit_timer(tp, tp->t_rtt);
875 		} else
876 			tp->t_state = TCPS_SYN_RECEIVED;
877 
878 		/*
879 		 * Advance ti->ti_seq to correspond to first data byte.
880 		 * If data, trim to stay within window,
881 		 * dropping FIN if necessary.
882 		 */
883 		ti->ti_seq++;
884 		if (ti->ti_len > tp->rcv_wnd) {
885 			todrop = ti->ti_len - tp->rcv_wnd;
886 			m_adj(m, -todrop);
887 			ti->ti_len = tp->rcv_wnd;
888 			tiflags &= ~TH_FIN;
889 			tcpstat.tcps_rcvpackafterwin++;
890 			tcpstat.tcps_rcvbyteafterwin += todrop;
891 		}
892 		tp->snd_wl1 = ti->ti_seq - 1;
893 		tp->rcv_up = ti->ti_seq;
894 		goto step6;
895 
896 	/*
897 	 * If the state is SYN_RECEIVED:
898 	 *	If seg contains an ACK, but not for our SYN, drop the input
899 	 *	and generate an RST.  See page 36, rfc793
900 	 */
901 	case TCPS_SYN_RECEIVED:
902 		if ((tiflags & TH_ACK) &&
903 		    (SEQ_LEQ(ti->ti_ack, tp->iss) ||
904 		     SEQ_GT(ti->ti_ack, tp->snd_max)))
905 			goto dropwithreset;
906 		break;
907 	}
908 
909 	/*
910 	 * States other than LISTEN or SYN_SENT.
911 	 * First check timestamp, if present.
912 	 * Then check that at least some bytes of segment are within
913 	 * receive window.  If segment begins before rcv_nxt,
914 	 * drop leading data (and SYN); if nothing left, just ack.
915 	 *
916 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
917 	 * and it's less than ts_recent, drop it.
918 	 */
919 	if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
920 	    TSTMP_LT(opti.ts_val, tp->ts_recent)) {
921 
922 		/* Check to see if ts_recent is over 24 days old.  */
923 		if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
924 			/*
925 			 * Invalidate ts_recent.  If this segment updates
926 			 * ts_recent, the age will be reset later and ts_recent
927 			 * will get a valid value.  If it does not, setting
928 			 * ts_recent to zero will at least satisfy the
929 			 * requirement that zero be placed in the timestamp
930 			 * echo reply when ts_recent isn't valid.  The
931 			 * age isn't reset until we get a valid ts_recent
932 			 * because we don't want out-of-order segments to be
933 			 * dropped when ts_recent is old.
934 			 */
935 			tp->ts_recent = 0;
936 		} else {
937 			tcpstat.tcps_rcvduppack++;
938 			tcpstat.tcps_rcvdupbyte += ti->ti_len;
939 			tcpstat.tcps_pawsdrop++;
940 			goto dropafterack;
941 		}
942 	}
943 
944 	todrop = tp->rcv_nxt - ti->ti_seq;
945 	if (todrop > 0) {
946 		if (tiflags & TH_SYN) {
947 			tiflags &= ~TH_SYN;
948 			ti->ti_seq++;
949 			if (ti->ti_urp > 1)
950 				ti->ti_urp--;
951 			else {
952 				tiflags &= ~TH_URG;
953 				ti->ti_urp = 0;
954 			}
955 			todrop--;
956 		}
957 		if (todrop > ti->ti_len ||
958 		    (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
959 			/*
960 			 * Any valid FIN must be to the left of the window.
961 			 * At this point the FIN must be a duplicate or
962 			 * out of sequence; drop it.
963 			 */
964 			tiflags &= ~TH_FIN;
965 			/*
966 			 * Send an ACK to resynchronize and drop any data.
967 			 * But keep on processing for RST or ACK.
968 			 */
969 			tp->t_flags |= TF_ACKNOW;
970 			todrop = ti->ti_len;
971 			tcpstat.tcps_rcvdupbyte += todrop;
972 			tcpstat.tcps_rcvduppack++;
973 		} else {
974 			tcpstat.tcps_rcvpartduppack++;
975 			tcpstat.tcps_rcvpartdupbyte += todrop;
976 		}
977 		m_adj(m, todrop);
978 		ti->ti_seq += todrop;
979 		ti->ti_len -= todrop;
980 		if (ti->ti_urp > todrop)
981 			ti->ti_urp -= todrop;
982 		else {
983 			tiflags &= ~TH_URG;
984 			ti->ti_urp = 0;
985 		}
986 	}
987 
988 	/*
989 	 * If new data are received on a connection after the
990 	 * user processes are gone, then RST the other end.
991 	 */
992 	if ((so->so_state & SS_NOFDREF) &&
993 	    tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
994 		tp = tcp_close(tp);
995 		tcpstat.tcps_rcvafterclose++;
996 		goto dropwithreset;
997 	}
998 
999 	/*
1000 	 * If segment ends after window, drop trailing data
1001 	 * (and PUSH and FIN); if nothing left, just ACK.
1002 	 */
1003 	todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1004 	if (todrop > 0) {
1005 		tcpstat.tcps_rcvpackafterwin++;
1006 		if (todrop >= ti->ti_len) {
1007 			tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1008 			/*
1009 			 * If a new connection request is received
1010 			 * while in TIME_WAIT, drop the old connection
1011 			 * and start over if the sequence numbers
1012 			 * are above the previous ones.
1013 			 */
1014 			if (tiflags & TH_SYN &&
1015 			    tp->t_state == TCPS_TIME_WAIT &&
1016 			    SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1017 				iss = tcp_new_iss(tp, sizeof(struct tcpcb),
1018 						  tp->rcv_nxt);
1019 				tp = tcp_close(tp);
1020 				/*
1021 				 * We have already advanced the mbuf
1022 				 * pointers past the IP+TCP headers and
1023 				 * options.  Restore those pointers before
1024 				 * attempting to use the TCP header again.
1025 				 */
1026 				m->m_data -= hdroptlen;
1027 				m->m_len  += hdroptlen;
1028 				goto findpcb;
1029 			}
1030 			/*
1031 			 * If window is closed can only take segments at
1032 			 * window edge, and have to drop data and PUSH from
1033 			 * incoming segments.  Continue processing, but
1034 			 * remember to ack.  Otherwise, drop segment
1035 			 * and ack.
1036 			 */
1037 			if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1038 				tp->t_flags |= TF_ACKNOW;
1039 				tcpstat.tcps_rcvwinprobe++;
1040 			} else
1041 				goto dropafterack;
1042 		} else
1043 			tcpstat.tcps_rcvbyteafterwin += todrop;
1044 		m_adj(m, -todrop);
1045 		ti->ti_len -= todrop;
1046 		tiflags &= ~(TH_PUSH|TH_FIN);
1047 	}
1048 
1049 	/*
1050 	 * If last ACK falls within this segment's sequence numbers,
1051 	 * and the timestamp is newer, record it.
1052 	 */
1053 	if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) &&
1054 	    SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) &&
1055 	    SEQ_LT(tp->last_ack_sent, ti->ti_seq + ti->ti_len +
1056 		   ((tiflags & (TH_SYN|TH_FIN)) != 0))) {
1057 		tp->ts_recent_age = tcp_now;
1058 		tp->ts_recent = opti.ts_val;
1059 	}
1060 
1061 	/*
1062 	 * If the RST bit is set examine the state:
1063 	 *    SYN_RECEIVED STATE:
1064 	 *	If passive open, return to LISTEN state.
1065 	 *	If active open, inform user that connection was refused.
1066 	 *    ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1067 	 *	Inform user that connection was reset, and close tcb.
1068 	 *    CLOSING, LAST_ACK, TIME_WAIT STATES
1069 	 *	Close the tcb.
1070 	 */
1071 	if (tiflags&TH_RST) switch (tp->t_state) {
1072 
1073 	case TCPS_SYN_RECEIVED:
1074 		so->so_error = ECONNREFUSED;
1075 		goto close;
1076 
1077 	case TCPS_ESTABLISHED:
1078 	case TCPS_FIN_WAIT_1:
1079 	case TCPS_FIN_WAIT_2:
1080 	case TCPS_CLOSE_WAIT:
1081 		so->so_error = ECONNRESET;
1082 	close:
1083 		tp->t_state = TCPS_CLOSED;
1084 		tcpstat.tcps_drops++;
1085 		tp = tcp_close(tp);
1086 		goto drop;
1087 
1088 	case TCPS_CLOSING:
1089 	case TCPS_LAST_ACK:
1090 	case TCPS_TIME_WAIT:
1091 		tp = tcp_close(tp);
1092 		goto drop;
1093 	}
1094 
1095 	/*
1096 	 * If a SYN is in the window, then this is an
1097 	 * error and we send an RST and drop the connection.
1098 	 */
1099 	if (tiflags & TH_SYN) {
1100 		tp = tcp_drop(tp, ECONNRESET);
1101 		goto dropwithreset;
1102 	}
1103 
1104 	/*
1105 	 * If the ACK bit is off we drop the segment and return.
1106 	 */
1107 	if ((tiflags & TH_ACK) == 0)
1108 		goto drop;
1109 
1110 	/*
1111 	 * Ack processing.
1112 	 */
1113 	switch (tp->t_state) {
1114 
1115 	/*
1116 	 * In SYN_RECEIVED state if the ack ACKs our SYN then enter
1117 	 * ESTABLISHED state and continue processing, otherwise
1118 	 * send an RST.
1119 	 */
1120 	case TCPS_SYN_RECEIVED:
1121 		if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
1122 		    SEQ_GT(ti->ti_ack, tp->snd_max))
1123 			goto dropwithreset;
1124 		tcpstat.tcps_connects++;
1125 		soisconnected(so);
1126 		tcp_established(tp);
1127 		/* Do window scaling? */
1128 		if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1129 			(TF_RCVD_SCALE|TF_REQ_SCALE)) {
1130 			tp->snd_scale = tp->requested_s_scale;
1131 			tp->rcv_scale = tp->request_r_scale;
1132 		}
1133 		TCP_REASS_LOCK(tp);
1134 		(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
1135 		TCP_REASS_UNLOCK(tp);
1136 		tp->snd_wl1 = ti->ti_seq - 1;
1137 		/* fall into ... */
1138 
1139 	/*
1140 	 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1141 	 * ACKs.  If the ack is in the range
1142 	 *	tp->snd_una < ti->ti_ack <= tp->snd_max
1143 	 * then advance tp->snd_una to ti->ti_ack and drop
1144 	 * data from the retransmission queue.  If this ACK reflects
1145 	 * more up to date window information we update our window information.
1146 	 */
1147 	case TCPS_ESTABLISHED:
1148 	case TCPS_FIN_WAIT_1:
1149 	case TCPS_FIN_WAIT_2:
1150 	case TCPS_CLOSE_WAIT:
1151 	case TCPS_CLOSING:
1152 	case TCPS_LAST_ACK:
1153 	case TCPS_TIME_WAIT:
1154 
1155 		if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1156 			if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1157 				tcpstat.tcps_rcvdupack++;
1158 				/*
1159 				 * If we have outstanding data (other than
1160 				 * a window probe), this is a completely
1161 				 * duplicate ack (ie, window info didn't
1162 				 * change), the ack is the biggest we've
1163 				 * seen and we've seen exactly our rexmt
1164 				 * threshhold of them, assume a packet
1165 				 * has been dropped and retransmit it.
1166 				 * Kludge snd_nxt & the congestion
1167 				 * window so we send only this one
1168 				 * packet.
1169 				 *
1170 				 * We know we're losing at the current
1171 				 * window size so do congestion avoidance
1172 				 * (set ssthresh to half the current window
1173 				 * and pull our congestion window back to
1174 				 * the new ssthresh).
1175 				 *
1176 				 * Dup acks mean that packets have left the
1177 				 * network (they're now cached at the receiver)
1178 				 * so bump cwnd by the amount in the receiver
1179 				 * to keep a constant cwnd packets in the
1180 				 * network.
1181 				 */
1182 				if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0 ||
1183 				    ti->ti_ack != tp->snd_una)
1184 					tp->t_dupacks = 0;
1185 				else if (++tp->t_dupacks == tcprexmtthresh) {
1186 					tcp_seq onxt = tp->snd_nxt;
1187 					u_int win =
1188 					    min(tp->snd_wnd, tp->snd_cwnd) /
1189 					    2 /	tp->t_segsz;
1190 					if (SEQ_LT(ti->ti_ack, tp->snd_recover)) {
1191 						/*
1192 						 * False fast retransmit after
1193 						 * timeout.  Do not cut window.
1194 						 */
1195 						tp->snd_cwnd += tp->t_segsz;
1196 						tp->t_dupacks = 0;
1197 						(void) tcp_output(tp);
1198 						goto drop;
1199 					}
1200 
1201 					if (win < 2)
1202 						win = 2;
1203 					tp->snd_ssthresh = win * tp->t_segsz;
1204 					tp->snd_recover = tp->snd_max;
1205 					TCP_TIMER_DISARM(tp, TCPT_REXMT);
1206 					tp->t_rtt = 0;
1207 					tp->snd_nxt = ti->ti_ack;
1208 					tp->snd_cwnd = tp->t_segsz;
1209 					(void) tcp_output(tp);
1210 					tp->snd_cwnd = tp->snd_ssthresh +
1211 					       tp->t_segsz * tp->t_dupacks;
1212 					if (SEQ_GT(onxt, tp->snd_nxt))
1213 						tp->snd_nxt = onxt;
1214 					goto drop;
1215 				} else if (tp->t_dupacks > tcprexmtthresh) {
1216 					tp->snd_cwnd += tp->t_segsz;
1217 					(void) tcp_output(tp);
1218 					goto drop;
1219 				}
1220 			} else
1221 				tp->t_dupacks = 0;
1222 			break;
1223 		}
1224 		/*
1225 		 * If the congestion window was inflated to account
1226 		 * for the other side's cached packets, retract it.
1227 		 */
1228 		if (!tcp_do_newreno) {
1229 			if (tp->t_dupacks >= tcprexmtthresh &&
1230 			    tp->snd_cwnd > tp->snd_ssthresh)
1231 				tp->snd_cwnd = tp->snd_ssthresh;
1232 			tp->t_dupacks = 0;
1233 		} else if (tp->t_dupacks >= tcprexmtthresh
1234 		    && !tcp_newreno(tp, ti)) {
1235 			tp->snd_cwnd = tp->snd_ssthresh;
1236 			/*
1237 			 * Window inflation should have left us with approx.
1238 			 * snd_ssthresh outstanding data.  But in case we
1239 			 * would be inclined to send a burst, better to do
1240 			 * it via the slow start mechanism.
1241 			 */
1242 			if (SEQ_SUB(tp->snd_max, ti->ti_ack) < tp->snd_ssthresh)
1243 				tp->snd_cwnd = SEQ_SUB(tp->snd_max, ti->ti_ack)
1244 				     + tp->t_segsz;
1245 			tp->t_dupacks = 0;
1246 		}
1247 		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1248 			tcpstat.tcps_rcvacktoomuch++;
1249 			goto dropafterack;
1250 		}
1251 		acked = ti->ti_ack - tp->snd_una;
1252 		tcpstat.tcps_rcvackpack++;
1253 		tcpstat.tcps_rcvackbyte += acked;
1254 
1255 		/*
1256 		 * If we have a timestamp reply, update smoothed
1257 		 * round trip time.  If no timestamp is present but
1258 		 * transmit timer is running and timed sequence
1259 		 * number was acked, update smoothed round trip time.
1260 		 * Since we now have an rtt measurement, cancel the
1261 		 * timer backoff (cf., Phil Karn's retransmit alg.).
1262 		 * Recompute the initial retransmit timer.
1263 		 */
1264 		if (opti.ts_present)
1265 			tcp_xmit_timer(tp, tcp_now - opti.ts_ecr + 1);
1266 		else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1267 			tcp_xmit_timer(tp,tp->t_rtt);
1268 
1269 		/*
1270 		 * If all outstanding data is acked, stop retransmit
1271 		 * timer and remember to restart (more output or persist).
1272 		 * If there is more data to be acked, restart retransmit
1273 		 * timer, using current (possibly backed-off) value.
1274 		 */
1275 		if (ti->ti_ack == tp->snd_max) {
1276 			TCP_TIMER_DISARM(tp, TCPT_REXMT);
1277 			needoutput = 1;
1278 		} else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0)
1279 			TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
1280 		/*
1281 		 * When new data is acked, open the congestion window.
1282 		 * If the window gives us less than ssthresh packets
1283 		 * in flight, open exponentially (segsz per packet).
1284 		 * Otherwise open linearly: segsz per window
1285 		 * (segsz^2 / cwnd per packet), plus a constant
1286 		 * fraction of a packet (segsz/8) to help larger windows
1287 		 * open quickly enough.
1288 		 */
1289 		{
1290 		register u_int cw = tp->snd_cwnd;
1291 		register u_int incr = tp->t_segsz;
1292 
1293 		if (cw > tp->snd_ssthresh)
1294 			incr = incr * incr / cw;
1295 		if (!tcp_do_newreno || SEQ_GEQ(ti->ti_ack, tp->snd_recover))
1296 			tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale);
1297 		}
1298 		if (acked > so->so_snd.sb_cc) {
1299 			tp->snd_wnd -= so->so_snd.sb_cc;
1300 			sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1301 			ourfinisacked = 1;
1302 		} else {
1303 			sbdrop(&so->so_snd, acked);
1304 			tp->snd_wnd -= acked;
1305 			ourfinisacked = 0;
1306 		}
1307 		sowwakeup(so);
1308 		tp->snd_una = ti->ti_ack;
1309 		if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1310 			tp->snd_nxt = tp->snd_una;
1311 
1312 		switch (tp->t_state) {
1313 
1314 		/*
1315 		 * In FIN_WAIT_1 STATE in addition to the processing
1316 		 * for the ESTABLISHED state if our FIN is now acknowledged
1317 		 * then enter FIN_WAIT_2.
1318 		 */
1319 		case TCPS_FIN_WAIT_1:
1320 			if (ourfinisacked) {
1321 				/*
1322 				 * If we can't receive any more
1323 				 * data, then closing user can proceed.
1324 				 * Starting the timer is contrary to the
1325 				 * specification, but if we don't get a FIN
1326 				 * we'll hang forever.
1327 				 */
1328 				if (so->so_state & SS_CANTRCVMORE) {
1329 					soisdisconnected(so);
1330 					if (tcp_maxidle > 0)
1331 						TCP_TIMER_ARM(tp, TCPT_2MSL,
1332 						    tcp_maxidle);
1333 				}
1334 				tp->t_state = TCPS_FIN_WAIT_2;
1335 			}
1336 			break;
1337 
1338 	 	/*
1339 		 * In CLOSING STATE in addition to the processing for
1340 		 * the ESTABLISHED state if the ACK acknowledges our FIN
1341 		 * then enter the TIME-WAIT state, otherwise ignore
1342 		 * the segment.
1343 		 */
1344 		case TCPS_CLOSING:
1345 			if (ourfinisacked) {
1346 				tp->t_state = TCPS_TIME_WAIT;
1347 				tcp_canceltimers(tp);
1348 				TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1349 				soisdisconnected(so);
1350 			}
1351 			break;
1352 
1353 		/*
1354 		 * In LAST_ACK, we may still be waiting for data to drain
1355 		 * and/or to be acked, as well as for the ack of our FIN.
1356 		 * If our FIN is now acknowledged, delete the TCB,
1357 		 * enter the closed state and return.
1358 		 */
1359 		case TCPS_LAST_ACK:
1360 			if (ourfinisacked) {
1361 				tp = tcp_close(tp);
1362 				goto drop;
1363 			}
1364 			break;
1365 
1366 		/*
1367 		 * In TIME_WAIT state the only thing that should arrive
1368 		 * is a retransmission of the remote FIN.  Acknowledge
1369 		 * it and restart the finack timer.
1370 		 */
1371 		case TCPS_TIME_WAIT:
1372 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1373 			goto dropafterack;
1374 		}
1375 	}
1376 
1377 step6:
1378 	/*
1379 	 * Update window information.
1380 	 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1381 	 */
1382 	if (((tiflags & TH_ACK) && SEQ_LT(tp->snd_wl1, ti->ti_seq)) ||
1383 	    (tp->snd_wl1 == ti->ti_seq && SEQ_LT(tp->snd_wl2, ti->ti_ack)) ||
1384 	    (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)) {
1385 		/* keep track of pure window updates */
1386 		if (ti->ti_len == 0 &&
1387 		    tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1388 			tcpstat.tcps_rcvwinupd++;
1389 		tp->snd_wnd = tiwin;
1390 		tp->snd_wl1 = ti->ti_seq;
1391 		tp->snd_wl2 = ti->ti_ack;
1392 		if (tp->snd_wnd > tp->max_sndwnd)
1393 			tp->max_sndwnd = tp->snd_wnd;
1394 		needoutput = 1;
1395 	}
1396 
1397 	/*
1398 	 * Process segments with URG.
1399 	 */
1400 	if ((tiflags & TH_URG) && ti->ti_urp &&
1401 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1402 		/*
1403 		 * This is a kludge, but if we receive and accept
1404 		 * random urgent pointers, we'll crash in
1405 		 * soreceive.  It's hard to imagine someone
1406 		 * actually wanting to send this much urgent data.
1407 		 */
1408 		if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1409 			ti->ti_urp = 0;			/* XXX */
1410 			tiflags &= ~TH_URG;		/* XXX */
1411 			goto dodata;			/* XXX */
1412 		}
1413 		/*
1414 		 * If this segment advances the known urgent pointer,
1415 		 * then mark the data stream.  This should not happen
1416 		 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1417 		 * a FIN has been received from the remote side.
1418 		 * In these states we ignore the URG.
1419 		 *
1420 		 * According to RFC961 (Assigned Protocols),
1421 		 * the urgent pointer points to the last octet
1422 		 * of urgent data.  We continue, however,
1423 		 * to consider it to indicate the first octet
1424 		 * of data past the urgent section as the original
1425 		 * spec states (in one of two places).
1426 		 */
1427 		if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1428 			tp->rcv_up = ti->ti_seq + ti->ti_urp;
1429 			so->so_oobmark = so->so_rcv.sb_cc +
1430 			    (tp->rcv_up - tp->rcv_nxt) - 1;
1431 			if (so->so_oobmark == 0)
1432 				so->so_state |= SS_RCVATMARK;
1433 			sohasoutofband(so);
1434 			tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1435 		}
1436 		/*
1437 		 * Remove out of band data so doesn't get presented to user.
1438 		 * This can happen independent of advancing the URG pointer,
1439 		 * but if two URG's are pending at once, some out-of-band
1440 		 * data may creep in... ick.
1441 		 */
1442 		if (ti->ti_urp <= (u_int16_t) ti->ti_len
1443 #ifdef SO_OOBINLINE
1444 		     && (so->so_options & SO_OOBINLINE) == 0
1445 #endif
1446 		     )
1447 			tcp_pulloutofband(so, ti, m);
1448 	} else
1449 		/*
1450 		 * If no out of band data is expected,
1451 		 * pull receive urgent pointer along
1452 		 * with the receive window.
1453 		 */
1454 		if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1455 			tp->rcv_up = tp->rcv_nxt;
1456 dodata:							/* XXX */
1457 
1458 	/*
1459 	 * Process the segment text, merging it into the TCP sequencing queue,
1460 	 * and arranging for acknowledgment of receipt if necessary.
1461 	 * This process logically involves adjusting tp->rcv_wnd as data
1462 	 * is presented to the user (this happens in tcp_usrreq.c,
1463 	 * case PRU_RCVD).  If a FIN has already been received on this
1464 	 * connection then we just ignore the text.
1465 	 */
1466 	if ((ti->ti_len || (tiflags & TH_FIN)) &&
1467 	    TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1468 		TCP_REASS(tp, ti, m, so, tiflags);
1469 		/*
1470 		 * Note the amount of data that peer has sent into
1471 		 * our window, in order to estimate the sender's
1472 		 * buffer size.
1473 		 */
1474 		len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1475 	} else {
1476 		m_freem(m);
1477 		tiflags &= ~TH_FIN;
1478 	}
1479 
1480 	/*
1481 	 * If FIN is received ACK the FIN and let the user know
1482 	 * that the connection is closing.  Ignore a FIN received before
1483 	 * the connection is fully established.
1484 	 */
1485 	if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) {
1486 		if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1487 			socantrcvmore(so);
1488 			tp->t_flags |= TF_ACKNOW;
1489 			tp->rcv_nxt++;
1490 		}
1491 		switch (tp->t_state) {
1492 
1493 	 	/*
1494 		 * In ESTABLISHED STATE enter the CLOSE_WAIT state.
1495 		 */
1496 		case TCPS_ESTABLISHED:
1497 			tp->t_state = TCPS_CLOSE_WAIT;
1498 			break;
1499 
1500 	 	/*
1501 		 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1502 		 * enter the CLOSING state.
1503 		 */
1504 		case TCPS_FIN_WAIT_1:
1505 			tp->t_state = TCPS_CLOSING;
1506 			break;
1507 
1508 	 	/*
1509 		 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1510 		 * starting the time-wait timer, turning off the other
1511 		 * standard timers.
1512 		 */
1513 		case TCPS_FIN_WAIT_2:
1514 			tp->t_state = TCPS_TIME_WAIT;
1515 			tcp_canceltimers(tp);
1516 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1517 			soisdisconnected(so);
1518 			break;
1519 
1520 		/*
1521 		 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1522 		 */
1523 		case TCPS_TIME_WAIT:
1524 			TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL);
1525 			break;
1526 		}
1527 	}
1528 	if (so->so_options & SO_DEBUG)
1529 		tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1530 
1531 	/*
1532 	 * Return any desired output.
1533 	 */
1534 	if (needoutput || (tp->t_flags & TF_ACKNOW))
1535 		(void) tcp_output(tp);
1536 	return;
1537 
1538 badsyn:
1539 	/*
1540 	 * Received a bad SYN.  Increment counters and dropwithreset.
1541 	 */
1542 	tcpstat.tcps_badsyn++;
1543 	tp = NULL;
1544 	goto dropwithreset;
1545 
1546 dropafterack:
1547 	/*
1548 	 * Generate an ACK dropping incoming segment if it occupies
1549 	 * sequence space, where the ACK reflects our state.
1550 	 */
1551 	if (tiflags & TH_RST)
1552 		goto drop;
1553 	m_freem(m);
1554 	tp->t_flags |= TF_ACKNOW;
1555 	(void) tcp_output(tp);
1556 	return;
1557 
1558 dropwithreset:
1559 	/*
1560 	 * Generate a RST, dropping incoming segment.
1561 	 * Make ACK acceptable to originator of segment.
1562 	 * Don't bother to respond if destination was broadcast/multicast.
1563 	 */
1564 	if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST) ||
1565 	    IN_MULTICAST(ti->ti_dst.s_addr))
1566 		goto drop;
1567 	if (tiflags & TH_ACK)
1568 		(void)tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1569 	else {
1570 		if (tiflags & TH_SYN)
1571 			ti->ti_len++;
1572 		(void)tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1573 		    TH_RST|TH_ACK);
1574 	}
1575 	return;
1576 
1577 drop:
1578 	/*
1579 	 * Drop space held by incoming segment and return.
1580 	 */
1581 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1582 		tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1583 	m_freem(m);
1584 	return;
1585 }
1586 
1587 void
1588 tcp_dooptions(tp, cp, cnt, ti, oi)
1589 	struct tcpcb *tp;
1590 	u_char *cp;
1591 	int cnt;
1592 	struct tcpiphdr *ti;
1593 	struct tcp_opt_info *oi;
1594 {
1595 	u_int16_t mss;
1596 	int opt, optlen;
1597 
1598 	for (; cnt > 0; cnt -= optlen, cp += optlen) {
1599 		opt = cp[0];
1600 		if (opt == TCPOPT_EOL)
1601 			break;
1602 		if (opt == TCPOPT_NOP)
1603 			optlen = 1;
1604 		else {
1605 			optlen = cp[1];
1606 			if (optlen <= 0)
1607 				break;
1608 		}
1609 		switch (opt) {
1610 
1611 		default:
1612 			continue;
1613 
1614 		case TCPOPT_MAXSEG:
1615 			if (optlen != TCPOLEN_MAXSEG)
1616 				continue;
1617 			if (!(ti->ti_flags & TH_SYN))
1618 				continue;
1619 			bcopy(cp + 2, &mss, sizeof(mss));
1620 			oi->maxseg = ntohs(mss);
1621 			break;
1622 
1623 		case TCPOPT_WINDOW:
1624 			if (optlen != TCPOLEN_WINDOW)
1625 				continue;
1626 			if (!(ti->ti_flags & TH_SYN))
1627 				continue;
1628 			tp->t_flags |= TF_RCVD_SCALE;
1629 			tp->requested_s_scale = cp[2];
1630 			if (tp->requested_s_scale > TCP_MAX_WINSHIFT) {
1631 				log(LOG_ERR, "TCP: invalid wscale %d from "
1632 				    "0x%08x, assuming %d\n",
1633 				    tp->requested_s_scale,
1634 				    ntohl(ti->ti_src.s_addr),
1635 				    TCP_MAX_WINSHIFT);
1636 				tp->requested_s_scale = TCP_MAX_WINSHIFT;
1637 			}
1638 			break;
1639 
1640 		case TCPOPT_TIMESTAMP:
1641 			if (optlen != TCPOLEN_TIMESTAMP)
1642 				continue;
1643 			oi->ts_present = 1;
1644 			bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val));
1645 			NTOHL(oi->ts_val);
1646 			bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr));
1647 			NTOHL(oi->ts_ecr);
1648 
1649 			/*
1650 			 * A timestamp received in a SYN makes
1651 			 * it ok to send timestamp requests and replies.
1652 			 */
1653 			if (ti->ti_flags & TH_SYN) {
1654 				tp->t_flags |= TF_RCVD_TSTMP;
1655 				tp->ts_recent = oi->ts_val;
1656 				tp->ts_recent_age = tcp_now;
1657 			}
1658 			break;
1659 		case TCPOPT_SACK_PERMITTED:
1660 			if (optlen != TCPOLEN_SACK_PERMITTED)
1661 				continue;
1662 			if (!(ti->ti_flags & TH_SYN))
1663 				continue;
1664 			tp->t_flags &= ~TF_CANT_TXSACK;
1665 			break;
1666 
1667 		case TCPOPT_SACK:
1668 			if (tp->t_flags & TF_IGNR_RXSACK)
1669 				continue;
1670 			if (optlen % 8 != 2 || optlen < 10)
1671 				continue;
1672 			cp += 2;
1673 			optlen -= 2;
1674 			for (; optlen > 0; cp -= 8, optlen -= 8) {
1675 				tcp_seq lwe, rwe;
1676 				bcopy((char *)cp, (char *) &lwe, sizeof(lwe));
1677 				NTOHL(lwe);
1678 				bcopy((char *)cp, (char *) &rwe, sizeof(rwe));
1679 				NTOHL(rwe);
1680 				/* tcp_mark_sacked(tp, lwe, rwe); */
1681 			}
1682 			break;
1683 		}
1684 	}
1685 }
1686 
1687 /*
1688  * Pull out of band byte out of a segment so
1689  * it doesn't appear in the user's data queue.
1690  * It is still reflected in the segment length for
1691  * sequencing purposes.
1692  */
1693 void
1694 tcp_pulloutofband(so, ti, m)
1695 	struct socket *so;
1696 	struct tcpiphdr *ti;
1697 	register struct mbuf *m;
1698 {
1699 	int cnt = ti->ti_urp - 1;
1700 
1701 	while (cnt >= 0) {
1702 		if (m->m_len > cnt) {
1703 			char *cp = mtod(m, caddr_t) + cnt;
1704 			struct tcpcb *tp = sototcpcb(so);
1705 
1706 			tp->t_iobc = *cp;
1707 			tp->t_oobflags |= TCPOOB_HAVEDATA;
1708 			bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1709 			m->m_len--;
1710 			return;
1711 		}
1712 		cnt -= m->m_len;
1713 		m = m->m_next;
1714 		if (m == 0)
1715 			break;
1716 	}
1717 	panic("tcp_pulloutofband");
1718 }
1719 
1720 /*
1721  * Collect new round-trip time estimate
1722  * and update averages and current timeout.
1723  */
1724 void
1725 tcp_xmit_timer(tp, rtt)
1726 	register struct tcpcb *tp;
1727 	short rtt;
1728 {
1729 	register short delta;
1730 	short rttmin;
1731 
1732 	tcpstat.tcps_rttupdated++;
1733 	--rtt;
1734 	if (tp->t_srtt != 0) {
1735 		/*
1736 		 * srtt is stored as fixed point with 3 bits after the
1737 		 * binary point (i.e., scaled by 8).  The following magic
1738 		 * is equivalent to the smoothing algorithm in rfc793 with
1739 		 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
1740 		 * point).  Adjust rtt to origin 0.
1741 		 */
1742 		delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
1743 		if ((tp->t_srtt += delta) <= 0)
1744 			tp->t_srtt = 1 << 2;
1745 		/*
1746 		 * We accumulate a smoothed rtt variance (actually, a
1747 		 * smoothed mean difference), then set the retransmit
1748 		 * timer to smoothed rtt + 4 times the smoothed variance.
1749 		 * rttvar is stored as fixed point with 2 bits after the
1750 		 * binary point (scaled by 4).  The following is
1751 		 * equivalent to rfc793 smoothing with an alpha of .75
1752 		 * (rttvar = rttvar*3/4 + |delta| / 4).  This replaces
1753 		 * rfc793's wired-in beta.
1754 		 */
1755 		if (delta < 0)
1756 			delta = -delta;
1757 		delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT);
1758 		if ((tp->t_rttvar += delta) <= 0)
1759 			tp->t_rttvar = 1 << 2;
1760 	} else {
1761 		/*
1762 		 * No rtt measurement yet - use the unsmoothed rtt.
1763 		 * Set the variance to half the rtt (so our first
1764 		 * retransmit happens at 3*rtt).
1765 		 */
1766 		tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2);
1767 		tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1);
1768 	}
1769 	tp->t_rtt = 0;
1770 	tp->t_rxtshift = 0;
1771 
1772 	/*
1773 	 * the retransmit should happen at rtt + 4 * rttvar.
1774 	 * Because of the way we do the smoothing, srtt and rttvar
1775 	 * will each average +1/2 tick of bias.  When we compute
1776 	 * the retransmit timer, we want 1/2 tick of rounding and
1777 	 * 1 extra tick because of +-1/2 tick uncertainty in the
1778 	 * firing of the timer.  The bias will give us exactly the
1779 	 * 1.5 tick we need.  But, because the bias is
1780 	 * statistical, we have to test that we don't drop below
1781 	 * the minimum feasible timer (which is 2 ticks).
1782 	 */
1783 	if (tp->t_rttmin > rtt + 2)
1784 		rttmin = tp->t_rttmin;
1785 	else
1786 		rttmin = rtt + 2;
1787 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX);
1788 
1789 	/*
1790 	 * We received an ack for a packet that wasn't retransmitted;
1791 	 * it is probably safe to discard any error indications we've
1792 	 * received recently.  This isn't quite right, but close enough
1793 	 * for now (a route might have failed after we sent a segment,
1794 	 * and the return path might not be symmetrical).
1795 	 */
1796 	tp->t_softerror = 0;
1797 }
1798 
1799 /*
1800  * Checks for partial ack.  If partial ack arrives, force the retransmission
1801  * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
1802  * 1.  By setting snd_nxt to ti_ack, this forces retransmission timer to
1803  * be started again.  If the ack advances at least to tp->snd_recover, return 0.
1804  */
1805 int
1806 tcp_newreno(tp, ti)
1807 	struct tcpcb *tp;
1808 	struct tcpiphdr *ti;
1809 {
1810 	if (SEQ_LT(ti->ti_ack, tp->snd_recover)) {
1811 	        tcp_seq onxt = tp->snd_nxt;
1812 	        tcp_seq ouna = tp->snd_una;  /* Haven't updated snd_una yet*/
1813 	        u_long  ocwnd = tp->snd_cwnd;
1814 		TCP_TIMER_DISARM(tp, TCPT_REXMT);
1815 	        tp->t_rtt = 0;
1816 	        tp->snd_nxt = ti->ti_ack;
1817 	        tp->snd_cwnd = tp->t_segsz;
1818 	        tp->snd_una = ti->ti_ack;
1819 	        (void) tcp_output(tp);
1820 	        tp->snd_cwnd = ocwnd;
1821 	        tp->snd_una = ouna;
1822 	        if (SEQ_GT(onxt, tp->snd_nxt))
1823 	                tp->snd_nxt = onxt;
1824 	        /*
1825 	         * Partial window deflation.  Relies on fact that tp->snd_una
1826 	         * not updated yet.
1827 	         */
1828 	        tp->snd_cwnd -= (ti->ti_ack - tp->snd_una - tp->t_segsz);
1829 	        return 1;
1830 	}
1831 	return 0;
1832 }
1833 
1834 
1835 /*
1836  * TCP compressed state engine.  Currently used to hold compressed
1837  * state for SYN_RECEIVED.
1838  */
1839 
1840 u_long	syn_cache_count;
1841 u_int32_t syn_hash1, syn_hash2;
1842 
1843 #define SYN_HASH(sa, sp, dp) \
1844 	((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \
1845 				     ((u_int32_t)(sp)))^syn_hash2)))
1846 
1847 LIST_HEAD(, syn_cache_head) tcp_syn_cache_queue;
1848 
1849 #define	SYN_CACHE_RM(sc, scp)						\
1850 do {									\
1851 	TAILQ_REMOVE(&(scp)->sch_queue, (sc), sc_queue);		\
1852 	if (--(scp)->sch_length	== 0)					\
1853 		LIST_REMOVE((scp), sch_headq);				\
1854 	syn_cache_count--;						\
1855 } while (0)
1856 
1857 struct pool syn_cache_pool;
1858 
1859 void
1860 syn_cache_init()
1861 {
1862 	int i;
1863 
1864 	/* Initialize the hash bucket queues. */
1865 	for (i = 0; i < tcp_syn_cache_size; i++)
1866 		TAILQ_INIT(&tcp_syn_cache[i].sch_queue);
1867 
1868 	/* Initialize the active hash bucket cache. */
1869 	LIST_INIT(&tcp_syn_cache_queue);
1870 
1871 	/* Initialize the syn cache pool. */
1872 	pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0,
1873 	    "synpl", 0, NULL, NULL, M_PCB);
1874 }
1875 
1876 void
1877 syn_cache_insert(sc)
1878 	struct syn_cache *sc;
1879 {
1880 	struct syn_cache_head *scp, *scp2, *sce;
1881 	struct syn_cache *sc2;
1882 	int s;
1883 
1884 	/*
1885 	 * If there are no entries in the hash table, reinitialize
1886 	 * the hash secrets.
1887 	 */
1888 	if (syn_cache_count == 0) {
1889 		struct timeval tv;
1890 		microtime(&tv);
1891 		syn_hash1 = random() ^ (u_long)&sc;
1892 		syn_hash2 = random() ^ tv.tv_usec;
1893 	}
1894 
1895 	sc->sc_hash = SYN_HASH(&sc->sc_src, sc->sc_sport, sc->sc_dport);
1896 	scp = &tcp_syn_cache[sc->sc_hash % tcp_syn_cache_size];
1897 
1898 	/*
1899 	 * Make sure that we don't overflow the per-bucket
1900 	 * limit or the total cache size limit.
1901 	 */
1902 	s = splsoftnet();
1903 	if (scp->sch_length >= tcp_syn_bucket_limit) {
1904 		tcpstat.tcps_sc_bucketoverflow++;
1905 		/*
1906 		 * The bucket is full.  Toss the first (i.e. oldest)
1907 		 * element in this bucket.
1908 		 */
1909 		sc2 = TAILQ_FIRST(&scp->sch_queue);
1910 		SYN_CACHE_RM(sc2, scp);
1911 		if (sc2->sc_ipopts)
1912 			(void) m_free(sc2->sc_ipopts);
1913 		pool_put(&syn_cache_pool, sc2);
1914 	} else if (syn_cache_count >= tcp_syn_cache_limit) {
1915 		tcpstat.tcps_sc_overflowed++;
1916 		/*
1917 		 * The cache is full.  Toss the first (i.e. oldest)
1918 		 * element in the first non-empty bucket we can find.
1919 		 */
1920 		scp2 = scp;
1921 		if (TAILQ_FIRST(&scp2->sch_queue) == NULL) {
1922 			sce = &tcp_syn_cache[tcp_syn_cache_size];
1923 			for (++scp2; scp2 != scp; scp2++) {
1924 				if (scp2 >= sce)
1925 					scp2 = &tcp_syn_cache[0];
1926 				if (TAILQ_FIRST(&scp2->sch_queue) != NULL)
1927 					break;
1928 			}
1929 		}
1930 		sc2 = TAILQ_FIRST(&scp2->sch_queue);
1931 		if (sc2 == NULL) {
1932 			if (sc->sc_ipopts)
1933 				(void) m_free(sc->sc_ipopts);
1934 			pool_put(&syn_cache_pool, sc);
1935 			return;
1936 		}
1937 		SYN_CACHE_RM(sc2, scp2);
1938 		if (sc2->sc_ipopts)
1939 			(void) m_free(sc2->sc_ipopts);
1940 		pool_put(&syn_cache_pool, sc2);
1941 	}
1942 
1943 	/* Set entry's timer. */
1944 	PRT_SLOW_ARM(sc->sc_timer, tcp_syn_cache_timeo);
1945 
1946 	/* Put it into the bucket. */
1947 	TAILQ_INSERT_TAIL(&scp->sch_queue, sc, sc_queue);
1948 	if (++scp->sch_length == 1)
1949 		LIST_INSERT_HEAD(&tcp_syn_cache_queue, scp, sch_headq);
1950 	syn_cache_count++;
1951 
1952 	tcpstat.tcps_sc_added++;
1953 	splx(s);
1954 }
1955 
1956 /*
1957  * Walk down the cache list, looking for expired entries in each bucket.
1958  */
1959 void
1960 syn_cache_timer()
1961 {
1962 	struct syn_cache_head *scp, *nscp;
1963 	struct syn_cache *sc, *nsc;
1964 	int s;
1965 
1966 	s = splsoftnet();
1967 	for (scp = LIST_FIRST(&tcp_syn_cache_queue); scp != NULL; scp = nscp) {
1968 #ifdef DIAGNOSTIC
1969 		if (TAILQ_FIRST(&scp->sch_queue) == NULL)
1970 			panic("syn_cache_timer: queue inconsistency");
1971 #endif
1972 		nscp = LIST_NEXT(scp, sch_headq);
1973 		for (sc = TAILQ_FIRST(&scp->sch_queue);
1974 		     sc != NULL && PRT_SLOW_ISEXPIRED(sc->sc_timer);
1975 		     sc = nsc) {
1976 			nsc = TAILQ_NEXT(sc, sc_queue);
1977 			tcpstat.tcps_sc_timed_out++;
1978 			SYN_CACHE_RM(sc, scp);
1979 			if (sc->sc_ipopts)
1980 				(void) m_free(sc->sc_ipopts);
1981 			pool_put(&syn_cache_pool, sc);
1982 		}
1983 	}
1984 	splx(s);
1985 }
1986 
1987 /*
1988  * Find an entry in the syn cache.
1989  */
1990 struct syn_cache *
1991 syn_cache_lookup(ti, headp)
1992 	struct tcpiphdr *ti;
1993 	struct syn_cache_head **headp;
1994 {
1995 	struct syn_cache *sc;
1996 	struct syn_cache_head *scp;
1997 	u_int32_t hash;
1998 	int s;
1999 
2000 	hash = SYN_HASH(&ti->ti_src, ti->ti_sport, ti->ti_dport);
2001 
2002 	scp = &tcp_syn_cache[hash % tcp_syn_cache_size];
2003 	*headp = scp;
2004 	s = splsoftnet();
2005 	for (sc = TAILQ_FIRST(&scp->sch_queue); sc != NULL;
2006 	     sc = TAILQ_NEXT(sc, sc_queue)) {
2007 		if (sc->sc_hash != hash)
2008 			continue;
2009 		if (sc->sc_src.s_addr == ti->ti_src.s_addr &&
2010 		    sc->sc_sport == ti->ti_sport &&
2011 		    sc->sc_dport == ti->ti_dport &&
2012 		    sc->sc_dst.s_addr == ti->ti_dst.s_addr) {
2013 			splx(s);
2014 			return (sc);
2015 		}
2016 	}
2017 	splx(s);
2018 	return (NULL);
2019 }
2020 
2021 /*
2022  * This function gets called when we receive an ACK for a
2023  * socket in the LISTEN state.  We look up the connection
2024  * in the syn cache, and if its there, we pull it out of
2025  * the cache and turn it into a full-blown connection in
2026  * the SYN-RECEIVED state.
2027  *
2028  * The return values may not be immediately obvious, and their effects
2029  * can be subtle, so here they are:
2030  *
2031  *	NULL	SYN was not found in cache; caller should drop the
2032  *		packet and send an RST.
2033  *
2034  *	-1	We were unable to create the new connection, and are
2035  *		aborting it.  An ACK,RST is being sent to the peer
2036  *		(unless we got screwey sequence numbners; see below),
2037  *		because the 3-way handshake has been completed.  Caller
2038  *		should not free the mbuf, since we may be using it.  If
2039  *		we are not, we will free it.
2040  *
2041  *	Otherwise, the return value is a pointer to the new socket
2042  *	associated with the connection.
2043  */
2044 struct socket *
2045 syn_cache_get(so, m)
2046 	struct socket *so;
2047 	struct mbuf *m;
2048 {
2049 	struct syn_cache *sc;
2050 	struct syn_cache_head *scp;
2051 	register struct inpcb *inp;
2052 	register struct tcpcb *tp = 0;
2053 	register struct tcpiphdr *ti;
2054 	struct sockaddr_in *sin;
2055 	struct mbuf *am;
2056 	long win;
2057 	int s;
2058 
2059 	ti = mtod(m, struct tcpiphdr *);
2060 	s = splsoftnet();
2061 	if ((sc = syn_cache_lookup(ti, &scp)) == NULL) {
2062 		splx(s);
2063 		return (NULL);
2064 	}
2065 
2066 	win = sbspace(&so->so_rcv);
2067 	if (win > TCP_MAXWIN)
2068 		win = TCP_MAXWIN;
2069 
2070 	/*
2071 	 * Verify the sequence and ack numbers.
2072 	 */
2073 	if ((ti->ti_ack != sc->sc_iss + 1) ||
2074 	    SEQ_LEQ(ti->ti_seq, sc->sc_irs) ||
2075 	    SEQ_GT(ti->ti_seq, sc->sc_irs + 1 + win)) {
2076 		(void) syn_cache_respond(sc, m, ti, win, 0);
2077 		splx(s);
2078 		return ((struct socket *)(-1));
2079 	}
2080 
2081 	/* Remove this cache entry */
2082 	SYN_CACHE_RM(sc, scp);
2083 	splx(s);
2084 
2085 	/*
2086 	 * Ok, create the full blown connection, and set things up
2087 	 * as they would have been set up if we had created the
2088 	 * connection when the SYN arrived.  If we can't create
2089 	 * the connection, abort it.
2090 	 */
2091 	so = sonewconn(so, SS_ISCONNECTED);
2092 	if (so == NULL)
2093 		goto resetandabort;
2094 
2095 	inp = sotoinpcb(so);
2096 	inp->inp_laddr = sc->sc_dst;
2097 	inp->inp_lport = sc->sc_dport;
2098 	in_pcbstate(inp, INP_BOUND);
2099 	inp->inp_options = ip_srcroute();
2100 	if (inp->inp_options == NULL) {
2101 		inp->inp_options = sc->sc_ipopts;
2102 		sc->sc_ipopts = NULL;
2103 	}
2104 
2105 	am = m_get(M_DONTWAIT, MT_SONAME);	/* XXX */
2106 	if (am == NULL)
2107 		goto resetandabort;
2108 	am->m_len = sizeof(struct sockaddr_in);
2109 	sin = mtod(am, struct sockaddr_in *);
2110 	sin->sin_family = AF_INET;
2111 	sin->sin_len = sizeof(*sin);
2112 	sin->sin_addr = sc->sc_src;
2113 	sin->sin_port = sc->sc_sport;
2114 	bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
2115 	if (in_pcbconnect(inp, am)) {
2116 		(void) m_free(am);
2117 		goto resetandabort;
2118 	}
2119 	(void) m_free(am);
2120 
2121 	tp = intotcpcb(inp);
2122 	if (sc->sc_request_r_scale != 15) {
2123 		tp->requested_s_scale = sc->sc_requested_s_scale;
2124 		tp->request_r_scale = sc->sc_request_r_scale;
2125 		tp->snd_scale = sc->sc_requested_s_scale;
2126 		tp->rcv_scale = sc->sc_request_r_scale;
2127 		tp->t_flags |= TF_RCVD_SCALE;
2128 	}
2129 	if (sc->sc_flags & SCF_TIMESTAMP)
2130 		tp->t_flags |= TF_RCVD_TSTMP;
2131 
2132 	tp->t_template = tcp_template(tp);
2133 	if (tp->t_template == 0) {
2134 		tp = tcp_drop(tp, ENOBUFS);	/* destroys socket */
2135 		so = NULL;
2136 		m_freem(m);
2137 		goto abort;
2138 	}
2139 
2140 	tp->iss = sc->sc_iss;
2141 	tp->irs = sc->sc_irs;
2142 	tcp_sendseqinit(tp);
2143 	tcp_rcvseqinit(tp);
2144 	tp->t_state = TCPS_SYN_RECEIVED;
2145 	TCP_TIMER_ARM(tp, TCPT_KEEP, TCPTV_KEEP_INIT);
2146 	tcpstat.tcps_accepts++;
2147 
2148 	/* Initialize tp->t_ourmss before we deal with the peer's! */
2149 	tp->t_ourmss = sc->sc_ourmaxseg;
2150 	tcp_mss_from_peer(tp, sc->sc_peermaxseg);
2151 
2152 	/*
2153 	 * Initialize the initial congestion window.  If we
2154 	 * had to retransmit the SYN,ACK, we must initialize cwnd
2155 	 * to 1 segment (i.e. the Loss Window).
2156 	 */
2157 	if (sc->sc_rexmt_count)
2158 		tp->snd_cwnd = tp->t_peermss;
2159 	else
2160 		tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win, tp->t_peermss);
2161 
2162 	tcp_rmx_rtt(tp);
2163 	tp->snd_wl1 = sc->sc_irs;
2164 	tp->rcv_up = sc->sc_irs + 1;
2165 
2166 	/*
2167 	 * This is what whould have happened in tcp_ouput() when
2168 	 * the SYN,ACK was sent.
2169 	 */
2170 	tp->snd_up = tp->snd_una;
2171 	tp->snd_max = tp->snd_nxt = tp->iss+1;
2172 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
2173 	if (win > 0 && SEQ_GT(tp->rcv_nxt+win, tp->rcv_adv))
2174 		tp->rcv_adv = tp->rcv_nxt + win;
2175 	tp->last_ack_sent = tp->rcv_nxt;
2176 
2177 	tcpstat.tcps_sc_completed++;
2178 	if (sc->sc_ipopts)
2179 		(void) m_free(sc->sc_ipopts);
2180 	pool_put(&syn_cache_pool, sc);
2181 	return (so);
2182 
2183 resetandabort:
2184 	(void) tcp_respond(NULL, ti, m, ti->ti_seq+ti->ti_len,
2185 	    (tcp_seq)0, TH_RST|TH_ACK);
2186 abort:
2187 	if (so != NULL)
2188 		(void) soabort(so);
2189 	if (sc->sc_ipopts)
2190 		(void) m_free(sc->sc_ipopts);
2191 	pool_put(&syn_cache_pool, sc);
2192 	tcpstat.tcps_sc_aborted++;
2193 	return ((struct socket *)(-1));
2194 }
2195 
2196 /*
2197  * This function is called when we get a RST for a
2198  * non-existant connection, so that we can see if the
2199  * connection is in the syn cache.  If it is, zap it.
2200  */
2201 
2202 void
2203 syn_cache_reset(ti)
2204 	register struct tcpiphdr *ti;
2205 {
2206 	struct syn_cache *sc;
2207 	struct syn_cache_head *scp;
2208 	int s = splsoftnet();
2209 
2210 	if ((sc = syn_cache_lookup(ti, &scp)) == NULL) {
2211 		splx(s);
2212 		return;
2213 	}
2214 	if (SEQ_LT(ti->ti_seq,sc->sc_irs) ||
2215 	    SEQ_GT(ti->ti_seq, sc->sc_irs+1)) {
2216 		splx(s);
2217 		return;
2218 	}
2219 	SYN_CACHE_RM(sc, scp);
2220 	splx(s);
2221 	tcpstat.tcps_sc_reset++;
2222 	if (sc->sc_ipopts)
2223 		(void) m_free(sc->sc_ipopts);
2224 	pool_put(&syn_cache_pool, sc);
2225 }
2226 
2227 void
2228 syn_cache_unreach(ip, th)
2229 	struct ip *ip;
2230 	struct tcphdr *th;
2231 {
2232 	struct syn_cache *sc;
2233 	struct syn_cache_head *scp;
2234 	struct tcpiphdr ti2;
2235 	int s;
2236 
2237 	ti2.ti_src.s_addr = ip->ip_dst.s_addr;
2238 	ti2.ti_dst.s_addr = ip->ip_src.s_addr;
2239 	ti2.ti_sport = th->th_dport;
2240 	ti2.ti_dport = th->th_sport;
2241 
2242 	s = splsoftnet();
2243 	if ((sc = syn_cache_lookup(&ti2, &scp)) == NULL) {
2244 		splx(s);
2245 		return;
2246 	}
2247 	/* If the sequence number != sc_iss, then it's a bogus ICMP msg */
2248 	if (ntohl (th->th_seq) != sc->sc_iss) {
2249 		splx(s);
2250 		return;
2251 	}
2252 
2253 	/*
2254 	 * If we've rertransmitted 3 times and this is our second error,
2255 	 * we remove the entry.  Otherwise, we allow it to continue on.
2256 	 * This prevents us from incorrectly nuking an entry during a
2257 	 * spurious network outage.
2258 	 *
2259 	 * See tcp_notify().
2260 	 */
2261 	if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rexmt_count < 3) {
2262 		sc->sc_flags |= SCF_UNREACH;
2263 		splx(s);
2264 		return;
2265 	}
2266 
2267 	SYN_CACHE_RM(sc, scp);
2268 	splx(s);
2269 	tcpstat.tcps_sc_unreach++;
2270 	if (sc->sc_ipopts)
2271 		(void) m_free(sc->sc_ipopts);
2272 	pool_put(&syn_cache_pool, sc);
2273 }
2274 
2275 /*
2276  * Given a LISTEN socket and an inbound SYN request, add
2277  * this to the syn cache, and send back a segment:
2278  *	<SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
2279  * to the source.
2280  *
2281  * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN.
2282  * Doing so would require that we hold onto the data and deliver it
2283  * to the application.  However, if we are the target of a SYN-flood
2284  * DoS attack, an attacker could send data which would eventually
2285  * consume all available buffer space if it were ACKed.  By not ACKing
2286  * the data, we avoid this DoS scenario.
2287  */
2288 
2289 int
2290 syn_cache_add(so, m, optp, optlen, oi)
2291 	struct socket *so;
2292 	struct mbuf *m;
2293 	u_char *optp;
2294 	int optlen;
2295 	struct tcp_opt_info *oi;
2296 {
2297 	register struct tcpiphdr *ti;
2298 	struct tcpcb tb, *tp;
2299 	long win;
2300 	struct syn_cache *sc;
2301 	struct syn_cache_head *scp;
2302 	struct mbuf *ipopts;
2303 
2304 	tp = sototcpcb(so);
2305 	ti = mtod(m, struct tcpiphdr *);
2306 
2307 	/*
2308 	 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
2309 	 * in_broadcast() should never return true on a received
2310 	 * packet with M_BCAST not set.
2311 	 */
2312 	if (m->m_flags & (M_BCAST|M_MCAST) ||
2313 	    IN_MULTICAST(ti->ti_src.s_addr) ||
2314 	    IN_MULTICAST(ti->ti_dst.s_addr))
2315 		return (0);
2316 
2317 	/*
2318 	 * Initialize some local state.
2319 	 */
2320 	win = sbspace(&so->so_rcv);
2321 	if (win > TCP_MAXWIN)
2322 		win = TCP_MAXWIN;
2323 
2324 	/*
2325 	 * Remember the IP options, if any.
2326 	 */
2327 	ipopts = ip_srcroute();
2328 
2329 	if (optp) {
2330 		tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
2331 		tcp_dooptions(&tb, optp, optlen, ti, oi);
2332 	} else
2333 		tb.t_flags = 0;
2334 
2335 	/*
2336 	 * See if we already have an entry for this connection.
2337 	 * If we do, resend the SYN,ACK, and remember since the
2338 	 * initial congestion window must be initialized to 1
2339 	 * segment when the connection completes.
2340 	 */
2341 	if ((sc = syn_cache_lookup(ti, &scp)) != NULL) {
2342 		tcpstat.tcps_sc_dupesyn++;
2343 		sc->sc_rexmt_count++;
2344 		if (sc->sc_rexmt_count == 0) {
2345 			/*
2346 			 * Eeek!  We rolled the counter.  Just set it
2347 			 * to the max value.  This shouldn't ever happen,
2348 			 * but there's no real reason to panic here, since
2349 			 * the count doesn't have to be very precise.
2350 			 */
2351 			sc->sc_rexmt_count = USHRT_MAX;
2352 		}
2353 
2354 		if (ipopts) {
2355 			/*
2356 			 * If we were remembering a previous source route,
2357 			 * forget it and use the new one we've been given.
2358 			 */
2359 			if (sc->sc_ipopts)
2360 				(void) m_free(sc->sc_ipopts);
2361 			sc->sc_ipopts = ipopts;
2362 		}
2363 
2364 		if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) {
2365 			tcpstat.tcps_sndacks++;
2366 			tcpstat.tcps_sndtotal++;
2367 		}
2368 		return (1);
2369 	}
2370 
2371 	sc = pool_get(&syn_cache_pool, PR_NOWAIT);
2372 	if (sc == NULL) {
2373 		if (ipopts)
2374 			(void) m_free(ipopts);
2375 		return (0);
2376 	}
2377 
2378 	/*
2379 	 * Fill in the cache, and put the necessary IP and TCP
2380 	 * options into the reply.
2381 	 */
2382 	sc->sc_src.s_addr = ti->ti_src.s_addr;
2383 	sc->sc_dst.s_addr = ti->ti_dst.s_addr;
2384 	sc->sc_sport = ti->ti_sport;
2385 	sc->sc_dport = ti->ti_dport;
2386 	sc->sc_flags = 0;
2387 	sc->sc_ipopts = ipopts;
2388 	sc->sc_irs = ti->ti_seq;
2389 	sc->sc_iss = tcp_new_iss(sc, sizeof(struct syn_cache), 0);
2390 	sc->sc_peermaxseg = oi->maxseg;
2391 	sc->sc_ourmaxseg = tcp_mss_to_advertise(m->m_flags & M_PKTHDR ?
2392 						m->m_pkthdr.rcvif : NULL);
2393 	if (tcp_do_rfc1323 && (tb.t_flags & TF_RCVD_TSTMP))
2394 		sc->sc_flags |= SCF_TIMESTAMP;
2395 	if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2396 	    (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2397 		sc->sc_requested_s_scale = tb.requested_s_scale;
2398 		sc->sc_request_r_scale = 0;
2399 		while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT &&
2400 		    TCP_MAXWIN << sc->sc_request_r_scale <
2401 		    so->so_rcv.sb_hiwat)
2402 			sc->sc_request_r_scale++;
2403 	} else {
2404 		sc->sc_requested_s_scale = 15;
2405 		sc->sc_request_r_scale = 15;
2406 	}
2407 	if (syn_cache_respond(sc, m, ti, win, tb.ts_recent) == 0) {
2408 		syn_cache_insert(sc);
2409 		tcpstat.tcps_sndacks++;
2410 		tcpstat.tcps_sndtotal++;
2411 	} else {
2412 		if (sc->sc_ipopts)
2413 			(void) m_free(sc->sc_ipopts);
2414 		pool_put(&syn_cache_pool, sc);
2415 		tcpstat.tcps_sc_dropped++;
2416 	}
2417 	return (1);
2418 }
2419 
2420 int
2421 syn_cache_respond(sc, m, ti, win, ts)
2422 	struct syn_cache *sc;
2423 	struct mbuf *m;
2424 	register struct tcpiphdr *ti;
2425 	long win;
2426 	u_long ts;
2427 {
2428 	u_int8_t *optp;
2429 	int optlen;
2430 
2431 	/*
2432 	 * Tack on the TCP options.  If there isn't enough trailing
2433 	 * space for them, move up the fixed header to make space.
2434 	 */
2435 	optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) +
2436 	    ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0);
2437 	if (optlen > M_TRAILINGSPACE(m)) {
2438 		if (M_LEADINGSPACE(m) >= optlen) {
2439 			m->m_data -= optlen;
2440 			m->m_len += optlen;
2441 		} else {
2442 			struct mbuf *m0 = m;
2443 			if ((m = m_gethdr(M_DONTWAIT, MT_HEADER)) == NULL) {
2444 				m_freem(m0);
2445 				return (ENOBUFS);
2446 			}
2447 			MH_ALIGN(m, sizeof(*ti) + optlen);
2448 			m->m_next = m0; /* this gets freed below */
2449 		}
2450 		bcopy((caddr_t)ti, mtod(m, caddr_t), sizeof(*ti));
2451 		ti = mtod(m, struct tcpiphdr *);
2452 	}
2453 
2454 	optp = (u_int8_t *)(ti + 1);
2455 	optp[0] = TCPOPT_MAXSEG;
2456 	optp[1] = 4;
2457 	optp[2] = (sc->sc_ourmaxseg >> 8) & 0xff;
2458 	optp[3] = sc->sc_ourmaxseg & 0xff;
2459 	optlen = 4;
2460 
2461 	if (sc->sc_request_r_scale != 15) {
2462 		*((u_int32_t *)(optp + optlen)) = htonl(TCPOPT_NOP << 24 |
2463 		    TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 |
2464 		    sc->sc_request_r_scale);
2465 		optlen += 4;
2466 	}
2467 
2468 	if (sc->sc_flags & SCF_TIMESTAMP) {
2469 		u_int32_t *lp = (u_int32_t *)(optp + optlen);
2470 		/* Form timestamp option as shown in appendix A of RFC 1323. */
2471 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
2472 		*lp++ = htonl(tcp_now);
2473 		*lp   = htonl(ts);
2474 		optlen += TCPOLEN_TSTAMP_APPA;
2475 	}
2476 
2477 	/*
2478 	 * Toss any trailing mbufs.  No need to worry about
2479 	 * m_len and m_pkthdr.len, since tcp_respond() will
2480 	 * unconditionally set them.
2481 	 */
2482 	if (m->m_next) {
2483 		m_freem(m->m_next);
2484 		m->m_next = NULL;
2485   	}
2486 
2487 	/*
2488 	 * Fill in the fields that tcp_respond() will not touch, and
2489 	 * then send the response.
2490 	 */
2491 	ti->ti_off = (sizeof(struct tcphdr) + optlen) >> 2;
2492 	ti->ti_win = htons(win);
2493 	return (tcp_respond(NULL, ti, m, sc->sc_irs + 1, sc->sc_iss,
2494 	    TH_SYN|TH_ACK));
2495 }
2496