xref: /netbsd-src/sys/netinet/tcp_timer.c (revision 481fca6e59249d8ffcf24fef7cfbe7b131bfb080)
1 /*	$NetBSD: tcp_timer.c,v 1.46 2000/03/30 13:25:10 augustss Exp $	*/
2 
3 /*
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*-
33  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
34  * All rights reserved.
35  *
36  * This code is derived from software contributed to The NetBSD Foundation
37  * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38  * Facility, NASA Ames Research Center.
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  * 1. Redistributions of source code must retain the above copyright
44  *    notice, this list of conditions and the following disclaimer.
45  * 2. Redistributions in binary form must reproduce the above copyright
46  *    notice, this list of conditions and the following disclaimer in the
47  *    documentation and/or other materials provided with the distribution.
48  * 3. All advertising materials mentioning features or use of this software
49  *    must display the following acknowledgement:
50  *	This product includes software developed by the NetBSD
51  *	Foundation, Inc. and its contributors.
52  * 4. Neither the name of The NetBSD Foundation nor the names of its
53  *    contributors may be used to endorse or promote products derived
54  *    from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66  * POSSIBILITY OF SUCH DAMAGE.
67  */
68 
69 /*
70  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71  *	The Regents of the University of California.  All rights reserved.
72  *
73  * Redistribution and use in source and binary forms, with or without
74  * modification, are permitted provided that the following conditions
75  * are met:
76  * 1. Redistributions of source code must retain the above copyright
77  *    notice, this list of conditions and the following disclaimer.
78  * 2. Redistributions in binary form must reproduce the above copyright
79  *    notice, this list of conditions and the following disclaimer in the
80  *    documentation and/or other materials provided with the distribution.
81  * 3. All advertising materials mentioning features or use of this software
82  *    must display the following acknowledgement:
83  *	This product includes software developed by the University of
84  *	California, Berkeley and its contributors.
85  * 4. Neither the name of the University nor the names of its contributors
86  *    may be used to endorse or promote products derived from this software
87  *    without specific prior written permission.
88  *
89  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
90  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
91  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
92  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
93  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
94  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
95  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
96  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
97  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
98  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
99  * SUCH DAMAGE.
100  *
101  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
102  */
103 
104 #include "opt_inet.h"
105 
106 #include <sys/param.h>
107 #include <sys/systm.h>
108 #include <sys/malloc.h>
109 #include <sys/mbuf.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/protosw.h>
113 #include <sys/errno.h>
114 
115 #include <net/if.h>
116 #include <net/route.h>
117 
118 #include <netinet/in.h>
119 #include <netinet/in_systm.h>
120 #include <netinet/ip.h>
121 #include <netinet/in_pcb.h>
122 #include <netinet/ip_var.h>
123 
124 #ifdef INET6
125 #ifndef INET
126 #include <netinet/in.h>
127 #endif
128 #include <netinet/ip6.h>
129 #include <netinet6/in6_pcb.h>
130 #endif
131 
132 #include <netinet/tcp.h>
133 #include <netinet/tcp_fsm.h>
134 #include <netinet/tcp_seq.h>
135 #include <netinet/tcp_timer.h>
136 #include <netinet/tcp_var.h>
137 #include <netinet/tcpip.h>
138 
139 int	tcp_keepidle = TCPTV_KEEP_IDLE;
140 int	tcp_keepintvl = TCPTV_KEEPINTVL;
141 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
142 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
143 int	tcp_maxidle;
144 
145 struct tcp_delack_head tcp_delacks;
146 
147 /*
148  * Fast timeout routine for processing delayed acks
149  */
150 void
151 tcp_fasttimo()
152 {
153 	struct tcpcb *tp, *ntp;
154 	int s;
155 
156 	s = splsoftnet();
157 	for (tp = tcp_delacks.lh_first; tp != NULL; tp = ntp) {
158 		/*
159 		 * If tcp_output() can't transmit the ACK for whatever
160 		 * reason, it will remain on the queue for the next
161 		 * time the heartbeat ticks.
162 		 */
163 		ntp = tp->t_delack.le_next;
164 		tp->t_flags |= TF_ACKNOW;
165 		(void) tcp_output(tp);
166 	}
167 	splx(s);
168 }
169 
170 /*
171  * Tcp protocol timeout routine called every 500 ms.
172  * Updates the timers in all active tcb's and
173  * causes finite state machine actions if timers expire.
174  */
175 void
176 tcp_slowtimo()
177 {
178 	struct inpcb *inp, *ninp;
179 	struct tcpcb *tp;
180 #if defined(INET6) && !defined(TCP6)
181 	struct in6pcb *in6p, *nin6p;
182 #endif
183 	int s;
184 	long i;
185 	static int syn_cache_last = 0;
186 	int skip, mask;
187 
188 	skip = mask = 0;
189 
190 	s = splsoftnet();
191 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
192 	/*
193 	 * Search through tcb's and update active timers.
194 	 */
195 	mask |= 1;
196 	inp = tcbtable.inpt_queue.cqh_first;
197 	if (inp == (struct inpcb *)0) {				/* XXX */
198 		skip |= 1;
199 		goto dotcb6;
200 	}
201 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) {
202 		ninp = inp->inp_queue.cqe_next;
203 		tp = intotcpcb(inp);
204 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
205 			continue;
206 		for (i = 0; i < TCPT_NTIMERS; i++) {
207 			if (TCP_TIMER_ISEXPIRED(tp, i)) {
208 				TCP_TIMER_DISARM(tp, i);
209 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
210 				    PRU_SLOWTIMO, (struct mbuf *)0,
211 				    (struct mbuf *)i, (struct mbuf *)0,
212 				    (struct proc *)0);
213 				/* XXX NOT MP SAFE */
214 				if ((ninp == (void *)&tcbtable.inpt_queue &&
215 				    tcbtable.inpt_queue.cqh_last != inp) ||
216 				    ninp->inp_queue.cqe_prev != inp)
217 					goto tpgone;
218 			}
219 		}
220 		tp->t_idle++;
221 		if (tp->t_rtt)
222 			tp->t_rtt++;
223 tpgone:
224 		;
225 	}
226 dotcb6:
227 #if defined(INET6) && !defined(TCP6)
228 	mask |= 2;
229 	in6p = tcb6.in6p_next;
230 	if (in6p == (struct in6pcb *)0) {			/* XXX */
231 		skip |= 2;
232 		goto doiss;
233 	}
234 	for (; in6p != (struct in6pcb *)&tcb6; in6p = nin6p) {
235 		nin6p = in6p->in6p_next;
236 		tp = in6totcpcb(in6p);
237 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
238 			continue;
239 		for (i = 0; i < TCPT_NTIMERS; i++) {
240 			if (TCP_TIMER_ISEXPIRED(tp, i)) {
241 				TCP_TIMER_DISARM(tp, i);
242 				(void) tcp_usrreq(tp->t_in6pcb->in6p_socket,
243 				    PRU_SLOWTIMO, (struct mbuf *)0,
244 				    (struct mbuf *)i, (struct mbuf *)0,
245 				    (struct proc *)0);
246 				/* XXX NOT MP SAFE */
247 				if ((nin6p == (void *)&tcb6 &&
248 				    tcb6.in6p_prev != in6p) ||
249 				    nin6p->in6p_prev != in6p)
250 					goto tp6gone;
251 			}
252 		}
253 		tp->t_idle++;
254 		if (tp->t_rtt)
255 			tp->t_rtt++;
256 tp6gone:
257 		;
258 	}
259 
260 doiss:
261 #endif
262 	if (mask == skip)
263 		goto done;
264 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
265 	tcp_now++;					/* for timestamps */
266 	if (++syn_cache_last >= tcp_syn_cache_interval) {
267 		syn_cache_timer();
268 		syn_cache_last = 0;
269 	}
270 done:
271 	splx(s);
272 }
273 
274 /*
275  * Cancel all timers for TCP tp.
276  */
277 void
278 tcp_canceltimers(tp)
279 	struct tcpcb *tp;
280 {
281 	int i;
282 
283 	for (i = 0; i < TCPT_NTIMERS; i++)
284 		TCP_TIMER_DISARM(tp, i);
285 }
286 
287 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
288     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
289 
290 int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
291 
292 /*
293  * TCP timer processing.
294  */
295 struct tcpcb *
296 tcp_timers(tp, timer)
297 	struct tcpcb *tp;
298 	int timer;
299 {
300 	short	rto;
301 
302 	switch (timer) {
303 
304 	/*
305 	 * 2 MSL timeout in shutdown went off.  If we're closed but
306 	 * still waiting for peer to close and connection has been idle
307 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
308 	 * control block.  Otherwise, check again in a bit.
309 	 */
310 	case TCPT_2MSL:
311 		if (tp->t_state != TCPS_TIME_WAIT &&
312 		    ((tcp_maxidle == 0) || (tp->t_idle <= tcp_maxidle)))
313 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
314 		else
315 			tp = tcp_close(tp);
316 		break;
317 
318 	/*
319 	 * Retransmission timer went off.  Message has not
320 	 * been acked within retransmit interval.  Back off
321 	 * to a longer retransmit interval and retransmit one segment.
322 	 */
323 	case TCPT_REXMT:
324 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
325 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
326 			tcpstat.tcps_timeoutdrop++;
327 			tp = tcp_drop(tp, tp->t_softerror ?
328 			    tp->t_softerror : ETIMEDOUT);
329 			break;
330 		}
331 		tcpstat.tcps_rexmttimeo++;
332 		rto = TCP_REXMTVAL(tp);
333 		if (rto < tp->t_rttmin)
334 			rto = tp->t_rttmin;
335 		TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
336 		    tp->t_rttmin, TCPTV_REXMTMAX);
337 		TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
338 #if 0
339 		/*
340 		 * If we are losing and we are trying path MTU discovery,
341 		 * try turning it off.  This will avoid black holes in
342 		 * the network which suppress or fail to send "packet
343 		 * too big" ICMP messages.  We should ideally do
344 		 * lots more sophisticated searching to find the right
345 		 * value here...
346 		 */
347 		if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
348 			struct rtentry *rt = NULL;
349 
350 			if (tp->t_inpcb)
351 				rt = in_pcbrtentry(tp->t_inpcb);
352 #ifdef INET6
353 			else if (tp->t_in6pcb)
354 				rt = in6_pcbrtentry(tp->t_in6pcb);
355 #endif
356 
357 			/* XXX:  Black hole recovery code goes here */
358 		}
359 #endif
360 		/*
361 		 * If losing, let the lower level know and try for
362 		 * a better route.  Also, if we backed off this far,
363 		 * our srtt estimate is probably bogus.  Clobber it
364 		 * so we'll take the next rtt measurement as our srtt;
365 		 * move the current srtt into rttvar to keep the current
366 		 * retransmit times until then.
367 		 */
368 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
369 			if (tp->t_inpcb)
370 				in_losing(tp->t_inpcb);
371 #ifdef INET6
372 			else if (tp->t_in6pcb)
373 				in6_losing(tp->t_in6pcb);
374 #endif
375 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
376 			tp->t_srtt = 0;
377 		}
378 		tp->snd_nxt = tp->snd_una;
379 		/*
380 		 * If timing a segment in this window, stop the timer.
381 		 */
382 		tp->t_rtt = 0;
383 		/*
384 		 * Remember if we are retransmitting a SYN, because if
385 		 * we do, set the initial congestion window must be set
386 		 * to 1 segment.
387 		 */
388 		if (tp->t_state == TCPS_SYN_SENT)
389 			tp->t_flags |= TF_SYN_REXMT;
390 		/*
391 		 * Close the congestion window down to one segment
392 		 * (we'll open it by one segment for each ack we get).
393 		 * Since we probably have a window's worth of unacked
394 		 * data accumulated, this "slow start" keeps us from
395 		 * dumping all that data as back-to-back packets (which
396 		 * might overwhelm an intermediate gateway).
397 		 *
398 		 * There are two phases to the opening: Initially we
399 		 * open by one mss on each ack.  This makes the window
400 		 * size increase exponentially with time.  If the
401 		 * window is larger than the path can handle, this
402 		 * exponential growth results in dropped packet(s)
403 		 * almost immediately.  To get more time between
404 		 * drops but still "push" the network to take advantage
405 		 * of improving conditions, we switch from exponential
406 		 * to linear window opening at some threshhold size.
407 		 * For a threshhold, we use half the current window
408 		 * size, truncated to a multiple of the mss.
409 		 *
410 		 * (the minimum cwnd that will give us exponential
411 		 * growth is 2 mss.  We don't allow the threshhold
412 		 * to go below this.)
413 		 */
414 		{
415 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
416 		if (win < 2)
417 			win = 2;
418 		/* Loss Window MUST be one segment. */
419 		tp->snd_cwnd = tp->t_segsz;
420 		tp->snd_ssthresh = win * tp->t_segsz;
421 		tp->t_dupacks = 0;
422 		}
423 		(void) tcp_output(tp);
424 		break;
425 
426 	/*
427 	 * Persistance timer into zero window.
428 	 * Force a byte to be output, if possible.
429 	 */
430 	case TCPT_PERSIST:
431 		/*
432 		 * Hack: if the peer is dead/unreachable, we do not
433 		 * time out if the window is closed.  After a full
434 		 * backoff, drop the connection if the idle time
435 		 * (no responses to probes) reaches the maximum
436 		 * backoff that we would use if retransmitting.
437 		 */
438 		rto = TCP_REXMTVAL(tp);
439 		if (rto < tp->t_rttmin)
440 			rto = tp->t_rttmin;
441 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
442 		    (tp->t_idle >= tcp_maxpersistidle ||
443 		    tp->t_idle >= rto * tcp_totbackoff)) {
444 			tcpstat.tcps_persistdrops++;
445 			tp = tcp_drop(tp, ETIMEDOUT);
446 			break;
447 		}
448 		tcpstat.tcps_persisttimeo++;
449 		tcp_setpersist(tp);
450 		tp->t_force = 1;
451 		(void) tcp_output(tp);
452 		tp->t_force = 0;
453 		break;
454 
455 	/*
456 	 * Keep-alive timer went off; send something
457 	 * or drop connection if idle for too long.
458 	 */
459 	case TCPT_KEEP:
460 	    {
461 		struct socket *so = NULL;
462 
463 		tcpstat.tcps_keeptimeo++;
464 		if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
465 			goto dropit;
466 		if (tp->t_inpcb)
467 			so = tp->t_inpcb->inp_socket;
468 #ifdef INET6
469 		else if (tp->t_in6pcb)
470 			so = tp->t_in6pcb->in6p_socket;
471 #endif
472 		if (so->so_options & SO_KEEPALIVE &&
473 		    tp->t_state <= TCPS_CLOSE_WAIT) {
474 		    	if ((tcp_maxidle > 0) &&
475 			    (tp->t_idle >= tcp_keepidle + tcp_maxidle))
476 				goto dropit;
477 			/*
478 			 * Send a packet designed to force a response
479 			 * if the peer is up and reachable:
480 			 * either an ACK if the connection is still alive,
481 			 * or an RST if the peer has closed the connection
482 			 * due to timeout or reboot.
483 			 * Using sequence number tp->snd_una-1
484 			 * causes the transmitted zero-length segment
485 			 * to lie outside the receive window;
486 			 * by the protocol spec, this requires the
487 			 * correspondent TCP to respond.
488 			 */
489 			tcpstat.tcps_keepprobe++;
490 			if (tcp_compat_42) {
491 				/*
492 				 * The keepalive packet must have nonzero
493 				 * length to get a 4.2 host to respond.
494 				 */
495 				(void)tcp_respond(tp, tp->t_template,
496 				    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
497 				    tp->snd_una - 1, 0);
498 			} else {
499 				(void)tcp_respond(tp, tp->t_template,
500 				    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
501 				    tp->snd_una - 1, 0);
502 			}
503 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
504 		} else
505 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
506 		break;
507 	    }
508 	dropit:
509 		tcpstat.tcps_keepdrops++;
510 		tp = tcp_drop(tp, ETIMEDOUT);
511 		break;
512 	}
513 	return (tp);
514 }
515