xref: /freebsd/sys/netinet/tcp_timer.c (revision 609e0c94f2ea3e5e75ddf58a45ec23613265f2a6)
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 1.24 1997/09/16 18:36:06 joerg Exp $
35  */
36 
37 #include "opt_compat.h"
38 #include "opt_tcpdebug.h"
39 
40 #ifndef TUBA_INCLUDE
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/protosw.h>
48 
49 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
50 
51 #include <net/route.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/in_pcb.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcp_fsm.h>
59 #include <netinet/tcp_seq.h>
60 #include <netinet/tcp_timer.h>
61 #include <netinet/tcp_var.h>
62 #include <netinet/tcpip.h>
63 #ifdef TCPDEBUG
64 #include <netinet/tcp_debug.h>
65 #endif
66 
67 int	tcp_keepinit = TCPTV_KEEP_INIT;
68 SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
69 	CTLFLAG_RW, &tcp_keepinit , 0, "");
70 
71 int	tcp_keepidle = TCPTV_KEEP_IDLE;
72 SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
73 	CTLFLAG_RW, &tcp_keepidle , 0, "");
74 
75 static int	tcp_keepintvl = TCPTV_KEEPINTVL;
76 SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
77 	CTLFLAG_RW, &tcp_keepintvl , 0, "");
78 
79 static int	always_keepalive = 0;
80 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive,
81 	CTLFLAG_RW, &always_keepalive , 0, "");
82 
83 static int	tcp_keepcnt = TCPTV_KEEPCNT;
84 	/* max idle probes */
85 static int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
86 	/* max idle time in persist */
87 int	tcp_maxidle;
88 #else /* TUBA_INCLUDE */
89 
90 static	int tcp_maxpersistidle;
91 #endif /* TUBA_INCLUDE */
92 
93 /*
94  * Fast timeout routine for processing delayed acks
95  */
96 void
97 tcp_fasttimo()
98 {
99 	register struct inpcb *inp;
100 	register struct tcpcb *tp;
101 	int s;
102 
103 	s = splnet();
104 
105 	for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
106 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
107 		    (tp->t_flags & TF_DELACK)) {
108 			tp->t_flags &= ~TF_DELACK;
109 			tp->t_flags |= TF_ACKNOW;
110 			tcpstat.tcps_delack++;
111 			(void) tcp_output(tp);
112 		}
113 	}
114 	splx(s);
115 }
116 
117 /*
118  * Tcp protocol timeout routine called every 500 ms.
119  * Updates the timers in all active tcb's and
120  * causes finite state machine actions if timers expire.
121  */
122 void
123 tcp_slowtimo()
124 {
125 	register struct inpcb *ip, *ipnxt;
126 	register struct tcpcb *tp;
127 	register int i;
128 	int s;
129 #ifdef TCPDEBUG
130 	int ostate;
131 #endif
132 
133 	s = splnet();
134 
135 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
136 
137 	ip = tcb.lh_first;
138 	if (ip == NULL) {
139 		splx(s);
140 		return;
141 	}
142 	/*
143 	 * Search through tcb's and update active timers.
144 	 */
145 	for (; ip != NULL; ip = ipnxt) {
146 		ipnxt = ip->inp_list.le_next;
147 		tp = intotcpcb(ip);
148 		if (tp == 0 || tp->t_state == TCPS_LISTEN)
149 			continue;
150 		for (i = 0; i < TCPT_NTIMERS; i++) {
151 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
152 #ifdef TCPDEBUG
153 				ostate = tp->t_state;
154 #endif
155 				tp = tcp_timers(tp, i);
156 				if (tp == NULL)
157 					goto tpgone;
158 #ifdef TCPDEBUG
159 				if (tp->t_inpcb->inp_socket->so_options
160 				    & SO_DEBUG)
161 					tcp_trace(TA_USER, ostate, tp,
162 						  (struct tcpiphdr *)0,
163 						  PRU_SLOWTIMO);
164 #endif
165 			}
166 		}
167 		tp->t_idle++;
168 		tp->t_duration++;
169 		if (tp->t_rtt)
170 			tp->t_rtt++;
171 tpgone:
172 		;
173 	}
174 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
175 #ifdef TCP_COMPAT_42
176 	if ((int)tcp_iss < 0)
177 		tcp_iss = TCP_ISSINCR;			/* XXX */
178 #endif
179 	tcp_now++;					/* for timestamps */
180 	splx(s);
181 }
182 #ifndef TUBA_INCLUDE
183 
184 /*
185  * Cancel all timers for TCP tp.
186  */
187 void
188 tcp_canceltimers(tp)
189 	struct tcpcb *tp;
190 {
191 	register int i;
192 
193 	for (i = 0; i < TCPT_NTIMERS; i++)
194 		tp->t_timer[i] = 0;
195 }
196 
197 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
198     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
199 
200 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
201 
202 /*
203  * TCP timer processing.
204  */
205 struct tcpcb *
206 tcp_timers(tp, timer)
207 	register struct tcpcb *tp;
208 	int timer;
209 {
210 	register int rexmt;
211 
212 	switch (timer) {
213 
214 	/*
215 	 * 2 MSL timeout in shutdown went off.  If we're closed but
216 	 * still waiting for peer to close and connection has been idle
217 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
218 	 * control block.  Otherwise, check again in a bit.
219 	 */
220 	case TCPT_2MSL:
221 		if (tp->t_state != TCPS_TIME_WAIT &&
222 		    tp->t_idle <= tcp_maxidle)
223 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
224 		else
225 			tp = tcp_close(tp);
226 		break;
227 
228 	/*
229 	 * Retransmission timer went off.  Message has not
230 	 * been acked within retransmit interval.  Back off
231 	 * to a longer retransmit interval and retransmit one segment.
232 	 */
233 	case TCPT_REXMT:
234 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
235 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
236 			tcpstat.tcps_timeoutdrop++;
237 			tp = tcp_drop(tp, tp->t_softerror ?
238 			    tp->t_softerror : ETIMEDOUT);
239 			break;
240 		}
241 		tcpstat.tcps_rexmttimeo++;
242 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
243 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
244 		    tp->t_rttmin, TCPTV_REXMTMAX);
245 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
246 		/*
247 		 * If losing, let the lower level know and try for
248 		 * a better route.  Also, if we backed off this far,
249 		 * our srtt estimate is probably bogus.  Clobber it
250 		 * so we'll take the next rtt measurement as our srtt;
251 		 * move the current srtt into rttvar to keep the current
252 		 * retransmit times until then.
253 		 */
254 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
255 			in_losing(tp->t_inpcb);
256 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
257 			tp->t_srtt = 0;
258 		}
259 		tp->snd_nxt = tp->snd_una;
260 		/*
261 		 * Force a segment to be sent.
262 		 */
263 		tp->t_flags |= TF_ACKNOW;
264 		/*
265 		 * If timing a segment in this window, stop the timer.
266 		 */
267 		tp->t_rtt = 0;
268 		/*
269 		 * Close the congestion window down to one segment
270 		 * (we'll open it by one segment for each ack we get).
271 		 * Since we probably have a window's worth of unacked
272 		 * data accumulated, this "slow start" keeps us from
273 		 * dumping all that data as back-to-back packets (which
274 		 * might overwhelm an intermediate gateway).
275 		 *
276 		 * There are two phases to the opening: Initially we
277 		 * open by one mss on each ack.  This makes the window
278 		 * size increase exponentially with time.  If the
279 		 * window is larger than the path can handle, this
280 		 * exponential growth results in dropped packet(s)
281 		 * almost immediately.  To get more time between
282 		 * drops but still "push" the network to take advantage
283 		 * of improving conditions, we switch from exponential
284 		 * to linear window opening at some threshhold size.
285 		 * For a threshhold, we use half the current window
286 		 * size, truncated to a multiple of the mss.
287 		 *
288 		 * (the minimum cwnd that will give us exponential
289 		 * growth is 2 mss.  We don't allow the threshhold
290 		 * to go below this.)
291 		 */
292 		{
293 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
294 		if (win < 2)
295 			win = 2;
296 		tp->snd_cwnd = tp->t_maxseg;
297 		tp->snd_ssthresh = win * tp->t_maxseg;
298 		tp->t_dupacks = 0;
299 		}
300 		(void) tcp_output(tp);
301 		break;
302 
303 	/*
304 	 * Persistance timer into zero window.
305 	 * Force a byte to be output, if possible.
306 	 */
307 	case TCPT_PERSIST:
308 		tcpstat.tcps_persisttimeo++;
309 		/*
310 		 * Hack: if the peer is dead/unreachable, we do not
311 		 * time out if the window is closed.  After a full
312 		 * backoff, drop the connection if the idle time
313 		 * (no responses to probes) reaches the maximum
314 		 * backoff that we would use if retransmitting.
315 		 */
316 		if (tp->t_rxtshift == TCP_MAXRXTSHIFT) {
317 			u_long maxidle = TCP_REXMTVAL(tp);
318 			if (maxidle < tp->t_rttmin)
319 				maxidle = tp->t_rttmin;
320 			maxidle *= tcp_totbackoff;
321 			if (tp->t_idle >= tcp_maxpersistidle ||
322 			    tp->t_idle >= maxidle) {
323 				tcpstat.tcps_persistdrop++;
324 				tp = tcp_drop(tp, ETIMEDOUT);
325 				break;
326 			}
327 		}
328 		tcp_setpersist(tp);
329 		tp->t_force = 1;
330 		(void) tcp_output(tp);
331 		tp->t_force = 0;
332 		break;
333 
334 	/*
335 	 * Keep-alive timer went off; send something
336 	 * or drop connection if idle for too long.
337 	 */
338 	case TCPT_KEEP:
339 		tcpstat.tcps_keeptimeo++;
340 		if (tp->t_state < TCPS_ESTABLISHED)
341 			goto dropit;
342 		if ((always_keepalive ||
343 		    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
344 		    tp->t_state <= TCPS_CLOSING) {
345 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
346 				goto dropit;
347 			/*
348 			 * Send a packet designed to force a response
349 			 * if the peer is up and reachable:
350 			 * either an ACK if the connection is still alive,
351 			 * or an RST if the peer has closed the connection
352 			 * due to timeout or reboot.
353 			 * Using sequence number tp->snd_una-1
354 			 * causes the transmitted zero-length segment
355 			 * to lie outside the receive window;
356 			 * by the protocol spec, this requires the
357 			 * correspondent TCP to respond.
358 			 */
359 			tcpstat.tcps_keepprobe++;
360 #ifdef TCP_COMPAT_42
361 			/*
362 			 * The keepalive packet must have nonzero length
363 			 * to get a 4.2 host to respond.
364 			 */
365 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
366 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
367 #else
368 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
369 			    tp->rcv_nxt, tp->snd_una - 1, 0);
370 #endif
371 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
372 		} else
373 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
374 		break;
375 	dropit:
376 		tcpstat.tcps_keepdrops++;
377 		tp = tcp_drop(tp, ETIMEDOUT);
378 		break;
379 	}
380 	return (tp);
381 }
382 #endif /* TUBA_INCLUDE */
383