xref: /netbsd-src/sys/netinet/tcp_timer.c (revision 2a399c6883d870daece976daec6ffa7bb7f934ce)
1 /*	$NetBSD: tcp_timer.c,v 1.28 1998/01/05 10:32:11 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
36  */
37 
38 #ifndef TUBA_INCLUDE
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 
48 #include <net/if.h>
49 #include <net/route.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/ip_var.h>
56 #include <netinet/tcp.h>
57 #include <netinet/tcp_fsm.h>
58 #include <netinet/tcp_seq.h>
59 #include <netinet/tcp_timer.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/tcpip.h>
62 
63 int	tcp_keepidle = TCPTV_KEEP_IDLE;
64 int	tcp_keepintvl = TCPTV_KEEPINTVL;
65 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
66 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
67 int	tcp_maxidle;
68 #else /* TUBA_INCLUDE */
69 
70 extern	int tcp_keepcnt;
71 extern	int tcp_maxpersistidle;
72 #endif /* TUBA_INCLUDE */
73 
74 struct tcp_delack_head tcp_delacks;
75 
76 /*
77  * Fast timeout routine for processing delayed acks
78  */
79 void
80 tcp_fasttimo()
81 {
82 	register struct tcpcb *tp, *ntp;
83 	int s;
84 
85 	s = splsoftnet();
86 	for (tp = tcp_delacks.lh_first; tp != NULL; tp = ntp) {
87 		/*
88 		 * If tcp_output() can't transmit the ACK for whatever
89 		 * reason, it will remain on the queue for the next
90 		 * time the heartbeat ticks.
91 		 */
92 		ntp = tp->t_delack.le_next;
93 		tp->t_flags |= TF_ACKNOW;
94 		(void) tcp_output(tp);
95 	}
96 	splx(s);
97 }
98 
99 /*
100  * Tcp protocol timeout routine called every 500 ms.
101  * Updates the timers in all active tcb's and
102  * causes finite state machine actions if timers expire.
103  */
104 void
105 tcp_slowtimo()
106 {
107 	register struct inpcb *inp, *ninp;
108 	register struct tcpcb *tp;
109 	int s;
110 	register long i;
111 	static int syn_cache_last = 0;
112 
113 	s = splsoftnet();
114 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
115 	/*
116 	 * Search through tcb's and update active timers.
117 	 */
118 	inp = tcbtable.inpt_queue.cqh_first;
119 	if (inp == (struct inpcb *)0) {				/* XXX */
120 		splx(s);
121 		return;
122 	}
123 	for (; inp != (struct inpcb *)&tcbtable.inpt_queue; inp = ninp) {
124 		ninp = inp->inp_queue.cqe_next;
125 		tp = intotcpcb(inp);
126 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
127 			continue;
128 		for (i = 0; i < TCPT_NTIMERS; i++) {
129 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
130 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
131 				    PRU_SLOWTIMO, (struct mbuf *)0,
132 				    (struct mbuf *)i, (struct mbuf *)0,
133 				    (struct proc *)0);
134 				/* XXX NOT MP SAFE */
135 				if ((ninp == (void *)&tcbtable.inpt_queue &&
136 				    tcbtable.inpt_queue.cqh_last != inp) ||
137 				    ninp->inp_queue.cqe_prev != inp)
138 					goto tpgone;
139 			}
140 		}
141 		tp->t_idle++;
142 		if (tp->t_rtt)
143 			tp->t_rtt++;
144 tpgone:
145 		;
146 	}
147 #if NRND == 0 /* Do we need to do this when using random() ? */
148 	tcp_iss_seq += TCP_ISSINCR;			/* increment iss */
149 #ifdef TCP_COMPAT_42
150 	if ((int)tcp_iss_seq < 0)
151 		tcp_iss_seq = 0;			/* XXX */
152 #endif
153 #endif
154 	tcp_now++;					/* for timestamps */
155 	if (++syn_cache_last >= tcp_syn_cache_interval) {
156 		syn_cache_timer(syn_cache_last);
157 		syn_cache_last = 0;
158 	}
159 	splx(s);
160 }
161 #ifndef TUBA_INCLUDE
162 
163 /*
164  * Cancel all timers for TCP tp.
165  */
166 void
167 tcp_canceltimers(tp)
168 	struct tcpcb *tp;
169 {
170 	register int i;
171 
172 	for (i = 0; i < TCPT_NTIMERS; i++)
173 		tp->t_timer[i] = 0;
174 }
175 
176 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
177     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
178 
179 int	tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
180 
181 /*
182  * TCP timer processing.
183  */
184 struct tcpcb *
185 tcp_timers(tp, timer)
186 	register struct tcpcb *tp;
187 	int timer;
188 {
189 
190 	switch (timer) {
191 
192 	/*
193 	 * 2 MSL timeout in shutdown went off.  If we're closed but
194 	 * still waiting for peer to close and connection has been idle
195 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
196 	 * control block.  Otherwise, check again in a bit.
197 	 */
198 	case TCPT_2MSL:
199 		if (tp->t_state != TCPS_TIME_WAIT &&
200 		    tp->t_idle <= tcp_maxidle)
201 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
202 		else
203 			tp = tcp_close(tp);
204 		break;
205 
206 	/*
207 	 * Retransmission timer went off.  Message has not
208 	 * been acked within retransmit interval.  Back off
209 	 * to a longer retransmit interval and retransmit one segment.
210 	 */
211 	case TCPT_REXMT:
212 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
213 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
214 			tcpstat.tcps_timeoutdrop++;
215 			tp = tcp_drop(tp, tp->t_softerror ?
216 			    tp->t_softerror : ETIMEDOUT);
217 			break;
218 		}
219 		tcpstat.tcps_rexmttimeo++;
220 		TCPT_RANGESET(tp->t_rxtcur,
221 		    TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift],
222 		    tp->t_rttmin, TCPTV_REXMTMAX);
223 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
224 		/*
225 		 * If losing, let the lower level know and try for
226 		 * a better route.  Also, if we backed off this far,
227 		 * our srtt estimate is probably bogus.  Clobber it
228 		 * so we'll take the next rtt measurement as our srtt;
229 		 * move the current srtt into rttvar to keep the current
230 		 * retransmit times until then.
231 		 */
232 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
233 			in_losing(tp->t_inpcb);
234 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
235 			tp->t_srtt = 0;
236 		}
237 		tp->snd_nxt = tp->snd_una;
238 		/*
239 		 * If timing a segment in this window, stop the timer.
240 		 */
241 		tp->t_rtt = 0;
242 		/*
243 		 * Close the congestion window down to the initial window
244 		 * (we'll open it by one segment for each ack we get).
245 		 * Since we probably have a window's worth of unacked
246 		 * data accumulated, this "slow start" keeps us from
247 		 * dumping all that data as back-to-back packets (which
248 		 * might overwhelm an intermediate gateway).
249 		 *
250 		 * There are two phases to the opening: Initially we
251 		 * open by one mss on each ack.  This makes the window
252 		 * size increase exponentially with time.  If the
253 		 * window is larger than the path can handle, this
254 		 * exponential growth results in dropped packet(s)
255 		 * almost immediately.  To get more time between
256 		 * drops but still "push" the network to take advantage
257 		 * of improving conditions, we switch from exponential
258 		 * to linear window opening at some threshhold size.
259 		 * For a threshhold, we use half the current window
260 		 * size, truncated to a multiple of the mss.
261 		 *
262 		 * (the minimum cwnd that will give us exponential
263 		 * growth is 2 mss.  We don't allow the threshhold
264 		 * to go below this.)
265 		 */
266 		{
267 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
268 		if (win < 2)
269 			win = 2;
270 		tp->snd_cwnd = TCP_INITIAL_WINDOW(tp->t_segsz);
271 		tp->snd_ssthresh = win * tp->t_segsz;
272 		tp->t_dupacks = 0;
273 		}
274 		(void) tcp_output(tp);
275 		break;
276 
277 	/*
278 	 * Persistance timer into zero window.
279 	 * Force a byte to be output, if possible.
280 	 */
281 	case TCPT_PERSIST:
282 		/*
283 		 * Hack: if the peer is dead/unreachable, we do not
284 		 * time out if the window is closed.  After a full
285 		 * backoff, drop the connection if the idle time
286 		 * (no responses to probes) reaches the maximum
287 		 * backoff that we would use if retransmitting.
288 		 */
289 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
290 		    (tp->t_idle >= tcp_maxpersistidle ||
291 		    tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
292 			tcpstat.tcps_persistdrops++;
293 			tp = tcp_drop(tp, ETIMEDOUT);
294 			break;
295 		}
296 		tcpstat.tcps_persisttimeo++;
297 		tcp_setpersist(tp);
298 		tp->t_force = 1;
299 		(void) tcp_output(tp);
300 		tp->t_force = 0;
301 		break;
302 
303 	/*
304 	 * Keep-alive timer went off; send something
305 	 * or drop connection if idle for too long.
306 	 */
307 	case TCPT_KEEP:
308 		tcpstat.tcps_keeptimeo++;
309 		if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
310 			goto dropit;
311 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
312 		    tp->t_state <= TCPS_CLOSE_WAIT) {
313 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
314 				goto dropit;
315 			/*
316 			 * Send a packet designed to force a response
317 			 * if the peer is up and reachable:
318 			 * either an ACK if the connection is still alive,
319 			 * or an RST if the peer has closed the connection
320 			 * due to timeout or reboot.
321 			 * Using sequence number tp->snd_una-1
322 			 * causes the transmitted zero-length segment
323 			 * to lie outside the receive window;
324 			 * by the protocol spec, this requires the
325 			 * correspondent TCP to respond.
326 			 */
327 			tcpstat.tcps_keepprobe++;
328 #ifdef TCP_COMPAT_42
329 			/*
330 			 * The keepalive packet must have nonzero length
331 			 * to get a 4.2 host to respond.
332 			 */
333 			(void)tcp_respond(tp, tp->t_template,
334 			    (struct mbuf *)NULL, tp->rcv_nxt - 1,
335 			    tp->snd_una - 1, 0);
336 #else
337 			(void)tcp_respond(tp, tp->t_template,
338 			    (struct mbuf *)NULL, tp->rcv_nxt,
339 			    tp->snd_una - 1, 0);
340 #endif
341 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
342 		} else
343 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
344 		break;
345 	dropit:
346 		tcpstat.tcps_keepdrops++;
347 		tp = tcp_drop(tp, ETIMEDOUT);
348 		break;
349 	}
350 	return (tp);
351 }
352 #endif /* TUBA_INCLUDE */
353