xref: /freebsd/sys/netinet/tcp_timer.c (revision 48991a368427cadb9cdac39581d1676c29619c52)
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 1.11 1995/11/09 20:23:07 phk Exp $
35  */
36 
37 #ifndef TUBA_INCLUDE
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/protosw.h>
47 #include <sys/errno.h>
48 #include <sys/queue.h>
49 
50 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
51 
52 #include <net/if.h>
53 #include <net/route.h>
54 
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/ip.h>
58 #include <netinet/in_pcb.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/tcp.h>
61 #include <netinet/tcp_fsm.h>
62 #include <netinet/tcp_seq.h>
63 #include <netinet/tcp_timer.h>
64 #include <netinet/tcp_var.h>
65 #include <netinet/tcpip.h>
66 
67 int	tcp_keepidle = TCPTV_KEEP_IDLE;
68 SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
69 	CTLFLAG_RW, &tcp_keepidle , 0, "");
70 
71 static int	tcp_keepintvl = TCPTV_KEEPINTVL;
72 SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
73 	CTLFLAG_RW, &tcp_keepintvl , 0, "");
74 
75 static int	tcp_keepcnt = TCPTV_KEEPCNT;
76 	/* max idle probes */
77 static int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
78 	/* max idle time in persist */
79 int	tcp_maxidle;
80 #else /* TUBA_INCLUDE */
81 
82 static	int tcp_maxpersistidle;
83 #endif /* TUBA_INCLUDE */
84 
85 /*
86  * Fast timeout routine for processing delayed acks
87  */
88 void
89 tcp_fasttimo()
90 {
91 	register struct inpcb *inp;
92 	register struct tcpcb *tp;
93 	int s;
94 
95 	s = splnet();
96 
97 	for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
98 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
99 		    (tp->t_flags & TF_DELACK)) {
100 			tp->t_flags &= ~TF_DELACK;
101 			tp->t_flags |= TF_ACKNOW;
102 			tcpstat.tcps_delack++;
103 			(void) tcp_output(tp);
104 		}
105 	}
106 	splx(s);
107 }
108 
109 /*
110  * Tcp protocol timeout routine called every 500 ms.
111  * Updates the timers in all active tcb's and
112  * causes finite state machine actions if timers expire.
113  */
114 void
115 tcp_slowtimo()
116 {
117 	register struct inpcb *ip, *ipnxt;
118 	register struct tcpcb *tp;
119 	register int i;
120 	int s;
121 
122 	s = splnet();
123 
124 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
125 
126 	ip = tcb.lh_first;
127 	if (ip == NULL) {
128 		splx(s);
129 		return;
130 	}
131 	/*
132 	 * Search through tcb's and update active timers.
133 	 */
134 	for (; ip != NULL; ip = ipnxt) {
135 		ipnxt = ip->inp_list.le_next;
136 		tp = intotcpcb(ip);
137 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
138 			continue;
139 		for (i = 0; i < TCPT_NTIMERS; i++) {
140 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
141 				if (tcp_usrreq(tp->t_inpcb->inp_socket,
142 				    PRU_SLOWTIMO, (struct mbuf *)0,
143 				    (struct mbuf *)i, (struct mbuf *)0) == NULL)
144 					goto tpgone;
145 			}
146 		}
147 		tp->t_idle++;
148 		tp->t_duration++;
149 		if (tp->t_rtt)
150 			tp->t_rtt++;
151 tpgone:
152 		;
153 	}
154 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
155 #ifdef TCP_COMPAT_42
156 	if ((int)tcp_iss < 0)
157 		tcp_iss = TCP_ISSINCR;			/* XXX */
158 #endif
159 	tcp_now++;					/* for timestamps */
160 	splx(s);
161 }
162 #ifndef TUBA_INCLUDE
163 
164 /*
165  * Cancel all timers for TCP tp.
166  */
167 void
168 tcp_canceltimers(tp)
169 	struct tcpcb *tp;
170 {
171 	register int i;
172 
173 	for (i = 0; i < TCPT_NTIMERS; i++)
174 		tp->t_timer[i] = 0;
175 }
176 
177 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
178     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
179 
180 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
181 
182 /*
183  * TCP timer processing.
184  */
185 struct tcpcb *
186 tcp_timers(tp, timer)
187 	register struct tcpcb *tp;
188 	int timer;
189 {
190 	register int rexmt;
191 
192 	switch (timer) {
193 
194 	/*
195 	 * 2 MSL timeout in shutdown went off.  If we're closed but
196 	 * still waiting for peer to close and connection has been idle
197 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
198 	 * control block.  Otherwise, check again in a bit.
199 	 */
200 	case TCPT_2MSL:
201 		if (tp->t_state != TCPS_TIME_WAIT &&
202 		    tp->t_idle <= tcp_maxidle)
203 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
204 		else
205 			tp = tcp_close(tp);
206 		break;
207 
208 	/*
209 	 * Retransmission timer went off.  Message has not
210 	 * been acked within retransmit interval.  Back off
211 	 * to a longer retransmit interval and retransmit one segment.
212 	 */
213 	case TCPT_REXMT:
214 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
215 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
216 			tcpstat.tcps_timeoutdrop++;
217 			tp = tcp_drop(tp, tp->t_softerror ?
218 			    tp->t_softerror : ETIMEDOUT);
219 			break;
220 		}
221 		tcpstat.tcps_rexmttimeo++;
222 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
223 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
224 		    tp->t_rttmin, TCPTV_REXMTMAX);
225 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
226 		/*
227 		 * If losing, let the lower level know and try for
228 		 * a better route.  Also, if we backed off this far,
229 		 * our srtt estimate is probably bogus.  Clobber it
230 		 * so we'll take the next rtt measurement as our srtt;
231 		 * move the current srtt into rttvar to keep the current
232 		 * retransmit times until then.
233 		 */
234 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
235 			in_losing(tp->t_inpcb);
236 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
237 			tp->t_srtt = 0;
238 		}
239 		tp->snd_nxt = tp->snd_una;
240 		/*
241 		 * If timing a segment in this window, stop the timer.
242 		 */
243 		tp->t_rtt = 0;
244 		/*
245 		 * Close the congestion window down to one segment
246 		 * (we'll open it by one segment for each ack we get).
247 		 * Since we probably have a window's worth of unacked
248 		 * data accumulated, this "slow start" keeps us from
249 		 * dumping all that data as back-to-back packets (which
250 		 * might overwhelm an intermediate gateway).
251 		 *
252 		 * There are two phases to the opening: Initially we
253 		 * open by one mss on each ack.  This makes the window
254 		 * size increase exponentially with time.  If the
255 		 * window is larger than the path can handle, this
256 		 * exponential growth results in dropped packet(s)
257 		 * almost immediately.  To get more time between
258 		 * drops but still "push" the network to take advantage
259 		 * of improving conditions, we switch from exponential
260 		 * to linear window opening at some threshhold size.
261 		 * For a threshhold, we use half the current window
262 		 * size, truncated to a multiple of the mss.
263 		 *
264 		 * (the minimum cwnd that will give us exponential
265 		 * growth is 2 mss.  We don't allow the threshhold
266 		 * to go below this.)
267 		 */
268 		{
269 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
270 		if (win < 2)
271 			win = 2;
272 		tp->snd_cwnd = tp->t_maxseg;
273 		tp->snd_ssthresh = win * tp->t_maxseg;
274 		tp->t_dupacks = 0;
275 		}
276 		(void) tcp_output(tp);
277 		break;
278 
279 	/*
280 	 * Persistance timer into zero window.
281 	 * Force a byte to be output, if possible.
282 	 */
283 	case TCPT_PERSIST:
284 		tcpstat.tcps_persisttimeo++;
285 		/*
286 		 * Hack: if the peer is dead/unreachable, we do not
287 		 * time out if the window is closed.  After a full
288 		 * backoff, drop the connection if the idle time
289 		 * (no responses to probes) reaches the maximum
290 		 * backoff that we would use if retransmitting.
291 		 */
292 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
293 		    (tp->t_idle >= tcp_maxpersistidle ||
294 		    tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
295 		        tcpstat.tcps_persistdrop++;
296 		        tp = tcp_drop(tp, ETIMEDOUT);
297 		        break;
298 		}
299 		tcp_setpersist(tp);
300 		tp->t_force = 1;
301 		(void) tcp_output(tp);
302 		tp->t_force = 0;
303 		break;
304 
305 	/*
306 	 * Keep-alive timer went off; send something
307 	 * or drop connection if idle for too long.
308 	 */
309 	case TCPT_KEEP:
310 		tcpstat.tcps_keeptimeo++;
311 		if (tp->t_state < TCPS_ESTABLISHED)
312 			goto dropit;
313 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
314 		    tp->t_state <= TCPS_CLOSE_WAIT) {
315 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
316 				goto dropit;
317 			/*
318 			 * Send a packet designed to force a response
319 			 * if the peer is up and reachable:
320 			 * either an ACK if the connection is still alive,
321 			 * or an RST if the peer has closed the connection
322 			 * due to timeout or reboot.
323 			 * Using sequence number tp->snd_una-1
324 			 * causes the transmitted zero-length segment
325 			 * to lie outside the receive window;
326 			 * by the protocol spec, this requires the
327 			 * correspondent TCP to respond.
328 			 */
329 			tcpstat.tcps_keepprobe++;
330 #ifdef TCP_COMPAT_42
331 			/*
332 			 * The keepalive packet must have nonzero length
333 			 * to get a 4.2 host to respond.
334 			 */
335 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
336 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
337 #else
338 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
339 			    tp->rcv_nxt, 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