xref: /freebsd/sys/netinet/tcp_timer.c (revision 8e6b01171e30297084bb0b4457c4183c2746aacc)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
34  *	$Id: tcp_timer.c,v 1.8 1995/07/29 18:48:43 davidg Exp $
35  */
36 
37 #ifndef TUBA_INCLUDE
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/protosw.h>
45 #include <sys/errno.h>
46 #include <sys/queue.h>
47 
48 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
49 
50 #include <net/if.h>
51 #include <net/route.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/ip.h>
56 #include <netinet/in_pcb.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcp_fsm.h>
60 #include <netinet/tcp_seq.h>
61 #include <netinet/tcp_timer.h>
62 #include <netinet/tcp_var.h>
63 #include <netinet/tcpip.h>
64 
65 int	tcp_keepidle = TCPTV_KEEP_IDLE;
66 int	tcp_keepintvl = TCPTV_KEEPINTVL;
67 int	tcp_keepcnt = TCPTV_KEEPCNT;		/* max idle probes */
68 int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;	/* max idle time in persist */
69 int	tcp_maxidle;
70 #else /* TUBA_INCLUDE */
71 
72 extern	int tcp_maxpersistidle;
73 #endif /* TUBA_INCLUDE */
74 
75 /*
76  * Fast timeout routine for processing delayed acks
77  */
78 void
79 tcp_fasttimo()
80 {
81 	register struct inpcb *inp;
82 	register struct tcpcb *tp;
83 	int s;
84 
85 	s = splnet();
86 
87 	for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
88 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
89 		    (tp->t_flags & TF_DELACK)) {
90 			tp->t_flags &= ~TF_DELACK;
91 			tp->t_flags |= TF_ACKNOW;
92 			tcpstat.tcps_delack++;
93 			(void) tcp_output(tp);
94 		}
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 *ip, *ipnxt;
108 	register struct tcpcb *tp;
109 	register int i;
110 	int s;
111 
112 	s = splnet();
113 
114 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
115 
116 	ip = tcb.lh_first;
117 	if (ip == NULL) {
118 		splx(s);
119 		return;
120 	}
121 	/*
122 	 * Search through tcb's and update active timers.
123 	 */
124 	for (; ip != NULL; ip = ipnxt) {
125 		ipnxt = ip->inp_list.le_next;
126 		tp = intotcpcb(ip);
127 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
128 			continue;
129 		for (i = 0; i < TCPT_NTIMERS; i++) {
130 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
131 				if (tcp_usrreq(tp->t_inpcb->inp_socket,
132 				    PRU_SLOWTIMO, (struct mbuf *)0,
133 				    (struct mbuf *)i, (struct mbuf *)0) == NULL)
134 					goto tpgone;
135 			}
136 		}
137 		tp->t_idle++;
138 		tp->t_duration++;
139 		if (tp->t_rtt)
140 			tp->t_rtt++;
141 tpgone:
142 		;
143 	}
144 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
145 #ifdef TCP_COMPAT_42
146 	if ((int)tcp_iss < 0)
147 		tcp_iss = TCP_ISSINCR;			/* XXX */
148 #endif
149 	tcp_now++;					/* for timestamps */
150 	splx(s);
151 }
152 #ifndef TUBA_INCLUDE
153 
154 /*
155  * Cancel all timers for TCP tp.
156  */
157 void
158 tcp_canceltimers(tp)
159 	struct tcpcb *tp;
160 {
161 	register int i;
162 
163 	for (i = 0; i < TCPT_NTIMERS; i++)
164 		tp->t_timer[i] = 0;
165 }
166 
167 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
168     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
169 
170 int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
171 
172 /*
173  * TCP timer processing.
174  */
175 struct tcpcb *
176 tcp_timers(tp, timer)
177 	register struct tcpcb *tp;
178 	int timer;
179 {
180 	register int rexmt;
181 
182 	switch (timer) {
183 
184 	/*
185 	 * 2 MSL timeout in shutdown went off.  If we're closed but
186 	 * still waiting for peer to close and connection has been idle
187 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
188 	 * control block.  Otherwise, check again in a bit.
189 	 */
190 	case TCPT_2MSL:
191 		if (tp->t_state != TCPS_TIME_WAIT &&
192 		    tp->t_idle <= tcp_maxidle)
193 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
194 		else
195 			tp = tcp_close(tp);
196 		break;
197 
198 	/*
199 	 * Retransmission timer went off.  Message has not
200 	 * been acked within retransmit interval.  Back off
201 	 * to a longer retransmit interval and retransmit one segment.
202 	 */
203 	case TCPT_REXMT:
204 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
205 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
206 			tcpstat.tcps_timeoutdrop++;
207 			tp = tcp_drop(tp, tp->t_softerror ?
208 			    tp->t_softerror : ETIMEDOUT);
209 			break;
210 		}
211 		tcpstat.tcps_rexmttimeo++;
212 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
213 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
214 		    tp->t_rttmin, TCPTV_REXMTMAX);
215 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
216 		/*
217 		 * If losing, let the lower level know and try for
218 		 * a better route.  Also, if we backed off this far,
219 		 * our srtt estimate is probably bogus.  Clobber it
220 		 * so we'll take the next rtt measurement as our srtt;
221 		 * move the current srtt into rttvar to keep the current
222 		 * retransmit times until then.
223 		 */
224 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
225 			in_losing(tp->t_inpcb);
226 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
227 			tp->t_srtt = 0;
228 		}
229 		tp->snd_nxt = tp->snd_una;
230 		/*
231 		 * Force a segment to be sent.
232 		 */
233 		tp->t_flags |= TF_ACKNOW;
234 		/*
235 		 * If timing a segment in this window, stop the timer.
236 		 */
237 		tp->t_rtt = 0;
238 		/*
239 		 * Close the congestion window down to one segment
240 		 * (we'll open it by one segment for each ack we get).
241 		 * Since we probably have a window's worth of unacked
242 		 * data accumulated, this "slow start" keeps us from
243 		 * dumping all that data as back-to-back packets (which
244 		 * might overwhelm an intermediate gateway).
245 		 *
246 		 * There are two phases to the opening: Initially we
247 		 * open by one mss on each ack.  This makes the window
248 		 * size increase exponentially with time.  If the
249 		 * window is larger than the path can handle, this
250 		 * exponential growth results in dropped packet(s)
251 		 * almost immediately.  To get more time between
252 		 * drops but still "push" the network to take advantage
253 		 * of improving conditions, we switch from exponential
254 		 * to linear window opening at some threshhold size.
255 		 * For a threshhold, we use half the current window
256 		 * size, truncated to a multiple of the mss.
257 		 *
258 		 * (the minimum cwnd that will give us exponential
259 		 * growth is 2 mss.  We don't allow the threshhold
260 		 * to go below this.)
261 		 */
262 		{
263 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
264 		if (win < 2)
265 			win = 2;
266 		tp->snd_cwnd = tp->t_maxseg;
267 		tp->snd_ssthresh = win * tp->t_maxseg;
268 		tp->t_dupacks = 0;
269 		}
270 		(void) tcp_output(tp);
271 		break;
272 
273 	/*
274 	 * Persistance timer into zero window.
275 	 * Force a byte to be output, if possible.
276 	 */
277 	case TCPT_PERSIST:
278 		tcpstat.tcps_persisttimeo++;
279 		/*
280 		 * Hack: if the peer is dead/unreachable, we do not
281 		 * time out if the window is closed.  After a full
282 		 * backoff, drop the connection if the idle time
283 		 * (no responses to probes) reaches the maximum
284 		 * backoff that we would use if retransmitting.
285 		 */
286 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
287 		    (tp->t_idle >= tcp_maxpersistidle ||
288 		    tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
289 		        tcpstat.tcps_persistdrop++;
290 		        tp = tcp_drop(tp, ETIMEDOUT);
291 		        break;
292 		}
293 		tcp_setpersist(tp);
294 		tp->t_force = 1;
295 		(void) tcp_output(tp);
296 		tp->t_force = 0;
297 		break;
298 
299 	/*
300 	 * Keep-alive timer went off; send something
301 	 * or drop connection if idle for too long.
302 	 */
303 	case TCPT_KEEP:
304 		tcpstat.tcps_keeptimeo++;
305 		if (tp->t_state < TCPS_ESTABLISHED)
306 			goto dropit;
307 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
308 		    tp->t_state <= TCPS_CLOSE_WAIT) {
309 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
310 				goto dropit;
311 			/*
312 			 * Send a packet designed to force a response
313 			 * if the peer is up and reachable:
314 			 * either an ACK if the connection is still alive,
315 			 * or an RST if the peer has closed the connection
316 			 * due to timeout or reboot.
317 			 * Using sequence number tp->snd_una-1
318 			 * causes the transmitted zero-length segment
319 			 * to lie outside the receive window;
320 			 * by the protocol spec, this requires the
321 			 * correspondent TCP to respond.
322 			 */
323 			tcpstat.tcps_keepprobe++;
324 #ifdef TCP_COMPAT_42
325 			/*
326 			 * The keepalive packet must have nonzero length
327 			 * to get a 4.2 host to respond.
328 			 */
329 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
330 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
331 #else
332 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
333 			    tp->rcv_nxt, tp->snd_una - 1, 0);
334 #endif
335 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
336 		} else
337 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
338 		break;
339 	dropit:
340 		tcpstat.tcps_keepdrops++;
341 		tp = tcp_drop(tp, ETIMEDOUT);
342 		break;
343 	}
344 	return (tp);
345 }
346 #endif /* TUBA_INCLUDE */
347