xref: /netbsd-src/sys/netinet/tcp_subr.c (revision dc306354b0b29af51801a7632f1e95265a68cd81)
1 /*	$NetBSD: tcp_subr.c,v 1.63 1998/12/18 21:38:03 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, 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_subr.c	8.2 (Berkeley) 5/24/95
73  */
74 
75 #include "opt_tcp_compat_42.h"
76 #include "rnd.h"
77 
78 #include <sys/param.h>
79 #include <sys/proc.h>
80 #include <sys/systm.h>
81 #include <sys/malloc.h>
82 #include <sys/mbuf.h>
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/protosw.h>
86 #include <sys/errno.h>
87 #include <sys/kernel.h>
88 #include <sys/pool.h>
89 #if NRND > 0
90 #include <sys/rnd.h>
91 #endif
92 
93 #include <net/route.h>
94 #include <net/if.h>
95 
96 #include <netinet/in.h>
97 #include <netinet/in_systm.h>
98 #include <netinet/ip.h>
99 #include <netinet/in_pcb.h>
100 #include <netinet/ip_var.h>
101 #include <netinet/ip_icmp.h>
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcpip.h>
108 
109 /* patchable/settable parameters for tcp */
110 int 	tcp_mssdflt = TCP_MSS;
111 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
112 int	tcp_do_rfc1323 = 1;	/* window scaling / timestamps (obsolete) */
113 int	tcp_do_sack = 1;	/* selective acknowledgement */
114 int	tcp_do_win_scale = 1;	/* RFC1323 window scaling */
115 int	tcp_do_timestamps = 1;	/* RFC1323 timestamps */
116 int	tcp_do_newreno = 0;	/* Use the New Reno algorithms */
117 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
118 int	tcp_init_win = 1;
119 int	tcp_mss_ifmtu = 0;
120 #ifdef TCP_COMPAT_42
121 int	tcp_compat_42 = 1;
122 #else
123 int	tcp_compat_42 = 0;
124 #endif
125 
126 #ifndef TCBHASHSIZE
127 #define	TCBHASHSIZE	128
128 #endif
129 int	tcbhashsize = TCBHASHSIZE;
130 
131 int	tcp_freeq __P((struct tcpcb *));
132 
133 struct pool tcpcb_pool;
134 struct pool tcp_template_pool;
135 
136 /*
137  * Tcp initialization
138  */
139 void
140 tcp_init()
141 {
142 
143 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
144 	    0, NULL, NULL, M_PCB);
145 	pool_init(&tcp_template_pool, sizeof(struct tcpiphdr), 0, 0, 0,
146 	    "tcptmpl", 0, NULL, NULL, M_MBUF);
147 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
148 	LIST_INIT(&tcp_delacks);
149 	if (max_protohdr < sizeof(struct tcpiphdr))
150 		max_protohdr = sizeof(struct tcpiphdr);
151 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
152 		panic("tcp_init");
153 
154 	/* Initialize the compressed state engine. */
155 	syn_cache_init();
156 }
157 
158 /*
159  * Create template to be used to send tcp packets on a connection.
160  * Call after host entry created, allocates an mbuf and fills
161  * in a skeletal tcp/ip header, minimizing the amount of work
162  * necessary when the connection is used.
163  */
164 struct tcpiphdr *
165 tcp_template(tp)
166 	struct tcpcb *tp;
167 {
168 	register struct inpcb *inp = tp->t_inpcb;
169 	register struct tcpiphdr *n;
170 
171 	if ((n = tp->t_template) == 0) {
172 		n = pool_get(&tcp_template_pool, PR_NOWAIT);
173 		if (n == NULL)
174 			return (NULL);
175 	}
176 	bzero(n->ti_x1, sizeof n->ti_x1);
177 	n->ti_pr = IPPROTO_TCP;
178 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
179 	n->ti_src = inp->inp_laddr;
180 	n->ti_dst = inp->inp_faddr;
181 	n->ti_sport = inp->inp_lport;
182 	n->ti_dport = inp->inp_fport;
183 	n->ti_seq = 0;
184 	n->ti_ack = 0;
185 	n->ti_x2 = 0;
186 	n->ti_off = 5;
187 	n->ti_flags = 0;
188 	n->ti_win = 0;
189 	n->ti_sum = 0;
190 	n->ti_urp = 0;
191 	return (n);
192 }
193 
194 /*
195  * Send a single message to the TCP at address specified by
196  * the given TCP/IP header.  If m == 0, then we make a copy
197  * of the tcpiphdr at ti and send directly to the addressed host.
198  * This is used to force keep alive messages out using the TCP
199  * template for a connection tp->t_template.  If flags are given
200  * then we send a message back to the TCP which originated the
201  * segment ti, and discard the mbuf containing it and any other
202  * attached mbufs.
203  *
204  * In any case the ack and sequence number of the transmitted
205  * segment are as specified by the parameters.
206  */
207 int
208 tcp_respond(tp, ti, m, ack, seq, flags)
209 	struct tcpcb *tp;
210 	register struct tcpiphdr *ti;
211 	register struct mbuf *m;
212 	tcp_seq ack, seq;
213 	int flags;
214 {
215 	register int tlen;
216 	int win = 0;
217 	struct route *ro = 0;
218 
219 	if (tp) {
220 		if ((flags & TH_RST) == 0)
221 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
222 		ro = &tp->t_inpcb->inp_route;
223 	}
224 	if (m == 0) {
225 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
226 		if (m == NULL)
227 			return (ENOBUFS);
228 
229 		if (tcp_compat_42)
230 			tlen = 1;
231 		else
232 			tlen = 0;
233 
234 		m->m_data += max_linkhdr;
235 		*mtod(m, struct tcpiphdr *) = *ti;
236 		ti = mtod(m, struct tcpiphdr *);
237 		flags = TH_ACK;
238 	} else {
239 		m_freem(m->m_next);
240 		m->m_next = 0;
241 		m->m_data = (caddr_t)ti;
242 		m->m_len = sizeof (struct tcpiphdr);
243 		tlen = 0;
244 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
245 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
246 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
247 #undef xchg
248 	}
249 	bzero(ti->ti_x1, sizeof ti->ti_x1);
250 	ti->ti_seq = htonl(seq);
251 	ti->ti_ack = htonl(ack);
252 	ti->ti_x2 = 0;
253 	if ((flags & TH_SYN) == 0) {
254 		if (tp)
255 			ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
256 		else
257 			ti->ti_win = htons((u_int16_t)win);
258 		ti->ti_off = sizeof (struct tcphdr) >> 2;
259 		tlen += sizeof (struct tcphdr);
260 	} else
261 		tlen += ti->ti_off << 2;
262 	ti->ti_len = htons((u_int16_t)tlen);
263 	tlen += sizeof (struct ip);
264 	m->m_len = tlen;
265 	m->m_pkthdr.len = tlen;
266 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
267 	ti->ti_flags = flags;
268 	ti->ti_urp = 0;
269 	ti->ti_sum = 0;
270 	ti->ti_sum = in_cksum(m, tlen);
271 	((struct ip *)ti)->ip_len = tlen;
272 	((struct ip *)ti)->ip_ttl = ip_defttl;
273 	return ip_output(m, NULL, ro, 0, NULL);
274 }
275 
276 /*
277  * Create a new TCP control block, making an
278  * empty reassembly queue and hooking it to the argument
279  * protocol control block.
280  */
281 struct tcpcb *
282 tcp_newtcpcb(inp)
283 	struct inpcb *inp;
284 {
285 	register struct tcpcb *tp;
286 
287 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
288 	if (tp == NULL)
289 		return (NULL);
290 	bzero((caddr_t)tp, sizeof(struct tcpcb));
291 	LIST_INIT(&tp->segq);
292 	LIST_INIT(&tp->timeq);
293 	tp->t_peermss = tcp_mssdflt;
294 	tp->t_ourmss = tcp_mssdflt;
295 	tp->t_segsz = tcp_mssdflt;
296 
297 	tp->t_flags = 0;
298 	if (tcp_do_rfc1323 && tcp_do_win_scale)
299 		tp->t_flags |= TF_REQ_SCALE;
300 	if (tcp_do_rfc1323 && tcp_do_timestamps)
301 		tp->t_flags |= TF_REQ_TSTMP;
302 	if (tcp_do_sack == 2)
303 		tp->t_flags |= TF_WILL_SACK;
304 	else if (tcp_do_sack == 1)
305 		tp->t_flags |= TF_WILL_SACK|TF_IGNR_RXSACK;
306 	tp->t_flags |= TF_CANT_TXSACK;
307 	tp->t_inpcb = inp;
308 	/*
309 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
310 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
311 	 * reasonable initial retransmit time.
312 	 */
313 	tp->t_srtt = TCPTV_SRTTBASE;
314 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
315 	tp->t_rttmin = TCPTV_MIN;
316 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
317 	    TCPTV_MIN, TCPTV_REXMTMAX);
318 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
319 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
320 	inp->inp_ip.ip_ttl = ip_defttl;
321 	inp->inp_ppcb = (caddr_t)tp;
322 	return (tp);
323 }
324 
325 /*
326  * Drop a TCP connection, reporting
327  * the specified error.  If connection is synchronized,
328  * then send a RST to peer.
329  */
330 struct tcpcb *
331 tcp_drop(tp, errno)
332 	register struct tcpcb *tp;
333 	int errno;
334 {
335 	struct socket *so = tp->t_inpcb->inp_socket;
336 
337 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
338 		tp->t_state = TCPS_CLOSED;
339 		(void) tcp_output(tp);
340 		tcpstat.tcps_drops++;
341 	} else
342 		tcpstat.tcps_conndrops++;
343 	if (errno == ETIMEDOUT && tp->t_softerror)
344 		errno = tp->t_softerror;
345 	so->so_error = errno;
346 	return (tcp_close(tp));
347 }
348 
349 /*
350  * Close a TCP control block:
351  *	discard all space held by the tcp
352  *	discard internet protocol block
353  *	wake up any sleepers
354  */
355 struct tcpcb *
356 tcp_close(tp)
357 	register struct tcpcb *tp;
358 {
359 	struct inpcb *inp = tp->t_inpcb;
360 	struct socket *so = inp->inp_socket;
361 #ifdef RTV_RTT
362 	register struct rtentry *rt;
363 
364 	/*
365 	 * If we sent enough data to get some meaningful characteristics,
366 	 * save them in the routing entry.  'Enough' is arbitrarily
367 	 * defined as the sendpipesize (default 4K) * 16.  This would
368 	 * give us 16 rtt samples assuming we only get one sample per
369 	 * window (the usual case on a long haul net).  16 samples is
370 	 * enough for the srtt filter to converge to within 5% of the correct
371 	 * value; fewer samples and we could save a very bogus rtt.
372 	 *
373 	 * Don't update the default route's characteristics and don't
374 	 * update anything that the user "locked".
375 	 */
376 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
377 	    (rt = inp->inp_route.ro_rt) &&
378 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
379 		register u_long i = 0;
380 
381 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
382 			i = tp->t_srtt *
383 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
384 			if (rt->rt_rmx.rmx_rtt && i)
385 				/*
386 				 * filter this update to half the old & half
387 				 * the new values, converting scale.
388 				 * See route.h and tcp_var.h for a
389 				 * description of the scaling constants.
390 				 */
391 				rt->rt_rmx.rmx_rtt =
392 				    (rt->rt_rmx.rmx_rtt + i) / 2;
393 			else
394 				rt->rt_rmx.rmx_rtt = i;
395 		}
396 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
397 			i = tp->t_rttvar *
398 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
399 			if (rt->rt_rmx.rmx_rttvar && i)
400 				rt->rt_rmx.rmx_rttvar =
401 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
402 			else
403 				rt->rt_rmx.rmx_rttvar = i;
404 		}
405 		/*
406 		 * update the pipelimit (ssthresh) if it has been updated
407 		 * already or if a pipesize was specified & the threshhold
408 		 * got below half the pipesize.  I.e., wait for bad news
409 		 * before we start updating, then update on both good
410 		 * and bad news.
411 		 */
412 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
413 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
414 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
415 			/*
416 			 * convert the limit from user data bytes to
417 			 * packets then to packet data bytes.
418 			 */
419 			i = (i + tp->t_segsz / 2) / tp->t_segsz;
420 			if (i < 2)
421 				i = 2;
422 			i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
423 			if (rt->rt_rmx.rmx_ssthresh)
424 				rt->rt_rmx.rmx_ssthresh =
425 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
426 			else
427 				rt->rt_rmx.rmx_ssthresh = i;
428 		}
429 	}
430 #endif /* RTV_RTT */
431 	/* free the reassembly queue, if any */
432 	TCP_REASS_LOCK(tp);
433 	(void) tcp_freeq(tp);
434 	TCP_REASS_UNLOCK(tp);
435 
436 	TCP_CLEAR_DELACK(tp);
437 
438 	if (tp->t_template)
439 		pool_put(&tcp_template_pool, tp->t_template);
440 	pool_put(&tcpcb_pool, tp);
441 	inp->inp_ppcb = 0;
442 	soisdisconnected(so);
443 	in_pcbdetach(inp);
444 	tcpstat.tcps_closed++;
445 	return ((struct tcpcb *)0);
446 }
447 
448 int
449 tcp_freeq(tp)
450 	struct tcpcb *tp;
451 {
452 	register struct ipqent *qe;
453 	int rv = 0;
454 #ifdef TCPREASS_DEBUG
455 	int i = 0;
456 #endif
457 
458 	TCP_REASS_LOCK_CHECK(tp);
459 
460 	while ((qe = tp->segq.lh_first) != NULL) {
461 #ifdef TCPREASS_DEBUG
462 		printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
463 			tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
464 			qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
465 #endif
466 		LIST_REMOVE(qe, ipqe_q);
467 		LIST_REMOVE(qe, ipqe_timeq);
468 		m_freem(qe->ipqe_m);
469 		pool_put(&ipqent_pool, qe);
470 		rv = 1;
471 	}
472 	return (rv);
473 }
474 
475 /*
476  * Protocol drain routine.  Called when memory is in short supply.
477  */
478 void
479 tcp_drain()
480 {
481 	register struct inpcb *inp;
482 	register struct tcpcb *tp;
483 
484 	/*
485 	 * Free the sequence queue of all TCP connections.
486 	 */
487 	inp = tcbtable.inpt_queue.cqh_first;
488 	if (inp)						/* XXX */
489 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue;
490 	    inp = inp->inp_queue.cqe_next) {
491 		if ((tp = intotcpcb(inp)) != NULL) {
492 			/*
493 			 * We may be called from a device's interrupt
494 			 * context.  If the tcpcb is already busy,
495 			 * just bail out now.
496 			 */
497 			if (tcp_reass_lock_try(tp) == 0)
498 				continue;
499 			if (tcp_freeq(tp))
500 				tcpstat.tcps_connsdrained++;
501 			TCP_REASS_UNLOCK(tp);
502 		}
503 	}
504 }
505 
506 /*
507  * Notify a tcp user of an asynchronous error;
508  * store error as soft error, but wake up user
509  * (for now, won't do anything until can select for soft error).
510  */
511 void
512 tcp_notify(inp, error)
513 	struct inpcb *inp;
514 	int error;
515 {
516 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
517 	register struct socket *so = inp->inp_socket;
518 
519 	/*
520 	 * Ignore some errors if we are hooked up.
521 	 * If connection hasn't completed, has retransmitted several times,
522 	 * and receives a second error, give up now.  This is better
523 	 * than waiting a long time to establish a connection that
524 	 * can never complete.
525 	 */
526 	if (tp->t_state == TCPS_ESTABLISHED &&
527 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
528 	      error == EHOSTDOWN)) {
529 		return;
530 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
531 	    tp->t_rxtshift > 3 && tp->t_softerror)
532 		so->so_error = error;
533 	else
534 		tp->t_softerror = error;
535 	wakeup((caddr_t) &so->so_timeo);
536 	sorwakeup(so);
537 	sowwakeup(so);
538 }
539 
540 void *
541 tcp_ctlinput(cmd, sa, v)
542 	int cmd;
543 	struct sockaddr *sa;
544 	register void *v;
545 {
546 	register struct ip *ip = v;
547 	register struct tcphdr *th;
548 	extern int inetctlerrmap[];
549 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
550 	int errno;
551 	int nmatch;
552 
553 	if ((unsigned)cmd >= PRC_NCMDS)
554 		return NULL;
555 	errno = inetctlerrmap[cmd];
556 	if (cmd == PRC_QUENCH)
557 		notify = tcp_quench;
558 	else if (PRC_IS_REDIRECT(cmd))
559 		notify = in_rtchange, ip = 0;
560 	else if (cmd == PRC_MSGSIZE && ip_mtudisc)
561 		notify = tcp_mtudisc, ip = 0;
562 	else if (cmd == PRC_HOSTDEAD)
563 		ip = 0;
564 	else if (errno == 0)
565 		return NULL;
566 	if (ip) {
567 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
568 		nmatch = in_pcbnotify(&tcbtable, satosin(sa)->sin_addr,
569 		    th->th_dport, ip->ip_src, th->th_sport, errno, notify);
570 		if (nmatch == 0 && syn_cache_count &&
571 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
572 		    inetctlerrmap[cmd] == ENETUNREACH ||
573 		    inetctlerrmap[cmd] == EHOSTDOWN))
574 			syn_cache_unreach(ip, th);
575 	} else
576 		(void)in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
577 		    notify);
578 	return NULL;
579 }
580 
581 /*
582  * When a source quence is received, we are being notifed of congestion.
583  * Close the congestion window down to the Loss Window (one segment).
584  * We will gradually open it again as we proceed.
585  */
586 void
587 tcp_quench(inp, errno)
588 	struct inpcb *inp;
589 	int errno;
590 {
591 	struct tcpcb *tp = intotcpcb(inp);
592 
593 	if (tp)
594 		tp->snd_cwnd = tp->t_segsz;
595 }
596 
597 /*
598  * On receipt of path MTU corrections, flush old route and replace it
599  * with the new one.  Retransmit all unacknowledged packets, to ensure
600  * that all packets will be received.
601  */
602 void
603 tcp_mtudisc(inp, errno)
604 	struct inpcb *inp;
605 	int errno;
606 {
607 	struct tcpcb *tp = intotcpcb(inp);
608 	struct rtentry *rt = in_pcbrtentry(inp);
609 
610 	if (tp != 0) {
611 		if (rt != 0) {
612 			/*
613 			 * If this was not a host route, remove and realloc.
614 			 */
615 			if ((rt->rt_flags & RTF_HOST) == 0) {
616 				in_rtchange(inp, errno);
617 				if ((rt = in_pcbrtentry(inp)) == 0)
618 					return;
619 			}
620 
621 			/*
622 			 * Slow start out of the error condition.  We
623 			 * use the MTU because we know it's smaller
624 			 * than the previously transmitted segment.
625 			 *
626 			 * Note: This is more conservative than the
627 			 * suggestion in draft-floyd-incr-init-win-03.
628 			 */
629 			if (rt->rt_rmx.rmx_mtu != 0)
630 				tp->snd_cwnd =
631 				    TCP_INITIAL_WINDOW(tcp_init_win,
632 				    rt->rt_rmx.rmx_mtu);
633 		}
634 
635 		/*
636 		 * Resend unacknowledged packets.
637 		 */
638 		tp->snd_nxt = tp->snd_una;
639 		tcp_output(tp);
640 	}
641 }
642 
643 
644 /*
645  * Compute the MSS to advertise to the peer.  Called only during
646  * the 3-way handshake.  If we are the server (peer initiated
647  * connection), we are called with a pointer to the interface
648  * on which the SYN packet arrived.  If we are the client (we
649  * initiated connection), we are called with a pointer to the
650  * interface out which this connection should go.
651  */
652 u_long
653 tcp_mss_to_advertise(ifp)
654 	const struct ifnet *ifp;
655 {
656 	extern u_long in_maxmtu;
657 	u_long mss = 0;
658 
659 	/*
660 	 * In order to avoid defeating path MTU discovery on the peer,
661 	 * we advertise the max MTU of all attached networks as our MSS,
662 	 * per RFC 1191, section 3.1.
663 	 *
664 	 * We provide the option to advertise just the MTU of
665 	 * the interface on which we hope this connection will
666 	 * be receiving.  If we are responding to a SYN, we
667 	 * will have a pretty good idea about this, but when
668 	 * initiating a connection there is a bit more doubt.
669 	 *
670 	 * We also need to ensure that loopback has a large enough
671 	 * MSS, as the loopback MTU is never included in in_maxmtu.
672 	 */
673 
674 	if (ifp != NULL)
675 		mss = ifp->if_mtu;
676 
677 	if (tcp_mss_ifmtu == 0)
678 		mss = max(in_maxmtu, mss);
679 
680 	if (mss > sizeof(struct tcpiphdr))
681 		mss -= sizeof(struct tcpiphdr);
682 
683 	mss = max(tcp_mssdflt, mss);
684 	return (mss);
685 }
686 
687 /*
688  * Set connection variables based on the peer's advertised MSS.
689  * We are passed the TCPCB for the actual connection.  If we
690  * are the server, we are called by the compressed state engine
691  * when the 3-way handshake is complete.  If we are the client,
692  * we are called when we recieve the SYN,ACK from the server.
693  *
694  * NOTE: Our advertised MSS value must be initialized in the TCPCB
695  * before this routine is called!
696  */
697 void
698 tcp_mss_from_peer(tp, offer)
699 	struct tcpcb *tp;
700 	int offer;
701 {
702 	struct inpcb *inp = tp->t_inpcb;
703 	struct socket *so = inp->inp_socket;
704 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
705 	struct rtentry *rt = in_pcbrtentry(inp);
706 #endif
707 	u_long bufsize;
708 	int mss;
709 
710 	/*
711 	 * As per RFC1122, use the default MSS value, unless they
712 	 * sent us an offer.  Do not accept offers less than 32 bytes.
713 	 */
714 	mss = tcp_mssdflt;
715 	if (offer)
716 		mss = offer;
717 	mss = max(mss, 32);		/* sanity */
718 	tp->t_peermss = mss;
719 	mss -= (tcp_optlen(tp) + ip_optlen(tp->t_inpcb));
720 
721 	/*
722 	 * If there's a pipesize, change the socket buffer to that size.
723 	 * Make the socket buffer an integral number of MSS units.  If
724 	 * the MSS is larger than the socket buffer, artificially decrease
725 	 * the MSS.
726 	 */
727 #ifdef RTV_SPIPE
728 	if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
729 		bufsize = rt->rt_rmx.rmx_sendpipe;
730 	else
731 #endif
732 		bufsize = so->so_snd.sb_hiwat;
733 	if (bufsize < mss)
734 		mss = bufsize;
735 	else {
736 		bufsize = roundup(bufsize, mss);
737 		if (bufsize > sb_max)
738 			bufsize = sb_max;
739 		(void) sbreserve(&so->so_snd, bufsize);
740 	}
741 	tp->t_segsz = mss;
742 
743 #ifdef RTV_SSTHRESH
744 	if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
745 		/*
746 		 * There's some sort of gateway or interface buffer
747 		 * limit on the path.  Use this to set the slow
748 		 * start threshold, but set the threshold to no less
749 		 * than 2 * MSS.
750 		 */
751 		tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
752 	}
753 #endif
754 }
755 
756 /*
757  * Processing necessary when a TCP connection is established.
758  */
759 void
760 tcp_established(tp)
761 	struct tcpcb *tp;
762 {
763 	struct inpcb *inp = tp->t_inpcb;
764 	struct socket *so = inp->inp_socket;
765 #ifdef RTV_RPIPE
766 	struct rtentry *rt = in_pcbrtentry(inp);
767 #endif
768 	u_long bufsize;
769 
770 	tp->t_state = TCPS_ESTABLISHED;
771 	TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
772 
773 #ifdef RTV_RPIPE
774 	if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
775 		bufsize = rt->rt_rmx.rmx_recvpipe;
776 	else
777 #endif
778 		bufsize = so->so_rcv.sb_hiwat;
779 	if (bufsize > tp->t_ourmss) {
780 		bufsize = roundup(bufsize, tp->t_ourmss);
781 		if (bufsize > sb_max)
782 			bufsize = sb_max;
783 		(void) sbreserve(&so->so_rcv, bufsize);
784 	}
785 }
786 
787 /*
788  * Check if there's an initial rtt or rttvar.  Convert from the
789  * route-table units to scaled multiples of the slow timeout timer.
790  * Called only during the 3-way handshake.
791  */
792 void
793 tcp_rmx_rtt(tp)
794 	struct tcpcb *tp;
795 {
796 #ifdef RTV_RTT
797 	struct rtentry *rt;
798 	int rtt;
799 
800 	if ((rt = in_pcbrtentry(tp->t_inpcb)) == NULL)
801 		return;
802 
803 	if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
804 		/*
805 		 * XXX The lock bit for MTU indicates that the value
806 		 * is also a minimum value; this is subject to time.
807 		 */
808 		if (rt->rt_rmx.rmx_locks & RTV_RTT)
809 			TCPT_RANGESET(tp->t_rttmin,
810 			    rtt / (RTM_RTTUNIT / PR_SLOWHZ),
811 			    TCPTV_MIN, TCPTV_REXMTMAX);
812 		tp->t_srtt = rtt /
813 		    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
814 		if (rt->rt_rmx.rmx_rttvar) {
815 			tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
816 			    ((RTM_RTTUNIT / PR_SLOWHZ) >>
817 				(TCP_RTTVAR_SHIFT + 2));
818 		} else {
819 			/* Default variation is +- 1 rtt */
820 			tp->t_rttvar =
821 			    tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
822 		}
823 		TCPT_RANGESET(tp->t_rxtcur,
824 		    ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
825 		    tp->t_rttmin, TCPTV_REXMTMAX);
826 	}
827 #endif
828 }
829 
830 tcp_seq	 tcp_iss_seq = 0;	/* tcp initial seq # */
831 
832 /*
833  * Get a new sequence value given a tcp control block
834  */
835 tcp_seq
836 tcp_new_iss(tp, len, addin)
837 	void            *tp;
838 	u_long           len;
839 	tcp_seq		 addin;
840 {
841 	tcp_seq          tcp_iss;
842 
843 	/*
844 	 * add randomness about this connection, but do not estimate
845 	 * entropy from the timing, since the physical device driver would
846 	 * have done that for us.
847 	 */
848 #if NRND > 0
849 	if (tp != NULL)
850 		rnd_add_data(NULL, tp, len, 0);
851 #endif
852 
853 	/*
854 	 * randomize.
855 	 */
856 #if NRND > 0
857 	rnd_extract_data(&tcp_iss, sizeof(tcp_iss), RND_EXTRACT_ANY);
858 #else
859 	tcp_iss = random();
860 #endif
861 
862 	/*
863 	 * If we were asked to add some amount to a known value,
864 	 * we will take a random value obtained above, mask off the upper
865 	 * bits, and add in the known value.  We also add in a constant to
866 	 * ensure that we are at least a certain distance from the original
867 	 * value.
868 	 *
869 	 * This is used when an old connection is in timed wait
870 	 * and we have a new one coming in, for instance.
871 	 */
872 	if (addin != 0) {
873 #ifdef TCPISS_DEBUG
874 		printf("Random %08x, ", tcp_iss);
875 #endif
876 		tcp_iss &= TCP_ISS_RANDOM_MASK;
877 		tcp_iss += addin + TCP_ISSINCR;
878 #ifdef TCPISS_DEBUG
879 		printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
880 #endif
881 	} else {
882 		tcp_iss &= TCP_ISS_RANDOM_MASK;
883 		tcp_iss += tcp_iss_seq;
884 		tcp_iss_seq += TCP_ISSINCR;
885 #ifdef TCPISS_DEBUG
886 		printf("ISS %08x\n", tcp_iss);
887 #endif
888 	}
889 
890 	if (tcp_compat_42) {
891 		/*
892 		 * Limit it to the positive range for really old TCP
893 		 * implementations.
894 		 */
895 		if (tcp_iss >= 0x80000000)
896 			tcp_iss &= 0x7fffffff;		/* XXX */
897 	}
898 
899 	return tcp_iss;
900 }
901 
902 
903 /*
904  * Determine the length of the TCP options for this connection.
905  *
906  * XXX:  What do we do for SACK, when we add that?  Just reserve
907  *       all of the space?  Otherwise we can't exactly be incrementing
908  *       cwnd by an amount that varies depending on the amount we last
909  *       had to SACK!
910  */
911 
912 u_int
913 tcp_optlen(tp)
914 	struct tcpcb *tp;
915 {
916 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
917 	    (TF_REQ_TSTMP | TF_RCVD_TSTMP))
918 		return TCPOLEN_TSTAMP_APPA;
919 	else
920 		return 0;
921 }
922