xref: /netbsd-src/sys/netinet/tcp_timer.c (revision 5aefcfdc06931dd97e76246d2fe0302f7b3fe094)
1 /*	$NetBSD: tcp_timer.c,v 1.48 2000/10/19 20:23:00 itojun 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 #ifdef INET6
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 #ifdef INET6
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 #ifdef DIAGNOSTIC
303 	if (tp->t_inpcb && tp->t_in6pcb)
304 		panic("tcp_timers: both t_inpcb and t_in6pcb are set");
305 #endif
306 
307 	switch (timer) {
308 
309 	/*
310 	 * 2 MSL timeout in shutdown went off.  If we're closed but
311 	 * still waiting for peer to close and connection has been idle
312 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
313 	 * control block.  Otherwise, check again in a bit.
314 	 */
315 	case TCPT_2MSL:
316 		if (tp->t_state != TCPS_TIME_WAIT &&
317 		    ((tcp_maxidle == 0) || (tp->t_idle <= tcp_maxidle)))
318 			TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
319 		else
320 			tp = tcp_close(tp);
321 		break;
322 
323 	/*
324 	 * Retransmission timer went off.  Message has not
325 	 * been acked within retransmit interval.  Back off
326 	 * to a longer retransmit interval and retransmit one segment.
327 	 */
328 	case TCPT_REXMT:
329 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
330 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
331 			tcpstat.tcps_timeoutdrop++;
332 			tp = tcp_drop(tp, tp->t_softerror ?
333 			    tp->t_softerror : ETIMEDOUT);
334 			break;
335 		}
336 		tcpstat.tcps_rexmttimeo++;
337 		rto = TCP_REXMTVAL(tp);
338 		if (rto < tp->t_rttmin)
339 			rto = tp->t_rttmin;
340 		TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
341 		    tp->t_rttmin, TCPTV_REXMTMAX);
342 		TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
343 #if 0
344 		/*
345 		 * If we are losing and we are trying path MTU discovery,
346 		 * try turning it off.  This will avoid black holes in
347 		 * the network which suppress or fail to send "packet
348 		 * too big" ICMP messages.  We should ideally do
349 		 * lots more sophisticated searching to find the right
350 		 * value here...
351 		 */
352 		if (ip_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
353 			struct rtentry *rt = NULL;
354 
355 #ifdef INET
356 			if (tp->t_inpcb)
357 				rt = in_pcbrtentry(tp->t_inpcb);
358 #endif
359 #ifdef INET6
360 			if (tp->t_in6pcb)
361 				rt = in6_pcbrtentry(tp->t_in6pcb);
362 #endif
363 
364 			/* XXX:  Black hole recovery code goes here */
365 		}
366 #endif
367 		/*
368 		 * If losing, let the lower level know and try for
369 		 * a better route.  Also, if we backed off this far,
370 		 * our srtt estimate is probably bogus.  Clobber it
371 		 * so we'll take the next rtt measurement as our srtt;
372 		 * move the current srtt into rttvar to keep the current
373 		 * retransmit times until then.
374 		 */
375 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
376 #ifdef INET
377 			if (tp->t_inpcb)
378 				in_losing(tp->t_inpcb);
379 #endif
380 #ifdef INET6
381 			if (tp->t_in6pcb)
382 				in6_losing(tp->t_in6pcb);
383 #endif
384 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
385 			tp->t_srtt = 0;
386 		}
387 		tp->snd_nxt = tp->snd_una;
388 		/*
389 		 * If timing a segment in this window, stop the timer.
390 		 */
391 		tp->t_rtt = 0;
392 		/*
393 		 * Remember if we are retransmitting a SYN, because if
394 		 * we do, set the initial congestion window must be set
395 		 * to 1 segment.
396 		 */
397 		if (tp->t_state == TCPS_SYN_SENT)
398 			tp->t_flags |= TF_SYN_REXMT;
399 		/*
400 		 * Close the congestion window down to one segment
401 		 * (we'll open it by one segment for each ack we get).
402 		 * Since we probably have a window's worth of unacked
403 		 * data accumulated, this "slow start" keeps us from
404 		 * dumping all that data as back-to-back packets (which
405 		 * might overwhelm an intermediate gateway).
406 		 *
407 		 * There are two phases to the opening: Initially we
408 		 * open by one mss on each ack.  This makes the window
409 		 * size increase exponentially with time.  If the
410 		 * window is larger than the path can handle, this
411 		 * exponential growth results in dropped packet(s)
412 		 * almost immediately.  To get more time between
413 		 * drops but still "push" the network to take advantage
414 		 * of improving conditions, we switch from exponential
415 		 * to linear window opening at some threshhold size.
416 		 * For a threshhold, we use half the current window
417 		 * size, truncated to a multiple of the mss.
418 		 *
419 		 * (the minimum cwnd that will give us exponential
420 		 * growth is 2 mss.  We don't allow the threshhold
421 		 * to go below this.)
422 		 */
423 		{
424 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
425 		if (win < 2)
426 			win = 2;
427 		/* Loss Window MUST be one segment. */
428 		tp->snd_cwnd = tp->t_segsz;
429 		tp->snd_ssthresh = win * tp->t_segsz;
430 		tp->t_dupacks = 0;
431 		}
432 		(void) tcp_output(tp);
433 		break;
434 
435 	/*
436 	 * Persistance timer into zero window.
437 	 * Force a byte to be output, if possible.
438 	 */
439 	case TCPT_PERSIST:
440 		/*
441 		 * Hack: if the peer is dead/unreachable, we do not
442 		 * time out if the window is closed.  After a full
443 		 * backoff, drop the connection if the idle time
444 		 * (no responses to probes) reaches the maximum
445 		 * backoff that we would use if retransmitting.
446 		 */
447 		rto = TCP_REXMTVAL(tp);
448 		if (rto < tp->t_rttmin)
449 			rto = tp->t_rttmin;
450 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
451 		    (tp->t_idle >= tcp_maxpersistidle ||
452 		    tp->t_idle >= rto * tcp_totbackoff)) {
453 			tcpstat.tcps_persistdrops++;
454 			tp = tcp_drop(tp, ETIMEDOUT);
455 			break;
456 		}
457 		tcpstat.tcps_persisttimeo++;
458 		tcp_setpersist(tp);
459 		tp->t_force = 1;
460 		(void) tcp_output(tp);
461 		tp->t_force = 0;
462 		break;
463 
464 	/*
465 	 * Keep-alive timer went off; send something
466 	 * or drop connection if idle for too long.
467 	 */
468 	case TCPT_KEEP:
469 	    {
470 		struct socket *so = NULL;
471 
472 		tcpstat.tcps_keeptimeo++;
473 		if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
474 			goto dropit;
475 #ifdef INET
476 		if (tp->t_inpcb)
477 			so = tp->t_inpcb->inp_socket;
478 #endif
479 #ifdef INET6
480 		if (tp->t_in6pcb)
481 			so = tp->t_in6pcb->in6p_socket;
482 #endif
483 		if (so->so_options & SO_KEEPALIVE &&
484 		    tp->t_state <= TCPS_CLOSE_WAIT) {
485 		    	if ((tcp_maxidle > 0) &&
486 			    (tp->t_idle >= tcp_keepidle + tcp_maxidle))
487 				goto dropit;
488 			/*
489 			 * Send a packet designed to force a response
490 			 * if the peer is up and reachable:
491 			 * either an ACK if the connection is still alive,
492 			 * or an RST if the peer has closed the connection
493 			 * due to timeout or reboot.
494 			 * Using sequence number tp->snd_una-1
495 			 * causes the transmitted zero-length segment
496 			 * to lie outside the receive window;
497 			 * by the protocol spec, this requires the
498 			 * correspondent TCP to respond.
499 			 */
500 			tcpstat.tcps_keepprobe++;
501 			if (tcp_compat_42) {
502 				/*
503 				 * The keepalive packet must have nonzero
504 				 * length to get a 4.2 host to respond.
505 				 */
506 				(void)tcp_respond(tp, tp->t_template,
507 				    (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
508 				    tp->snd_una - 1, 0);
509 			} else {
510 				(void)tcp_respond(tp, tp->t_template,
511 				    (struct mbuf *)NULL, NULL, tp->rcv_nxt,
512 				    tp->snd_una - 1, 0);
513 			}
514 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
515 		} else
516 			TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
517 		break;
518 	    }
519 	dropit:
520 		tcpstat.tcps_keepdrops++;
521 		tp = tcp_drop(tp, ETIMEDOUT);
522 		break;
523 	}
524 	return (tp);
525 }
526