xref: /freebsd/sys/netinet/tcp_timer.c (revision 17ee9d00bc1ae1e598c38f25826f861e4bc6c3ce)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
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	8.1 (Berkeley) 6/10/93
34  * $Id: tcp_timer.c,v 1.3 1995/02/09 23:13:26 wollman 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 
47 #include <net/if.h>
48 #include <net/route.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip_var.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_fsm.h>
57 #include <netinet/tcp_seq.h>
58 #include <netinet/tcp_timer.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/tcpip.h>
61 
62 int	tcp_keepidle = TCPTV_KEEP_IDLE;
63 int	tcp_keepintvl = TCPTV_KEEPINTVL;
64 int	tcp_maxidle;
65 #endif /* TUBA_INCLUDE */
66 /*
67  * Fast timeout routine for processing delayed acks
68  */
69 void
70 tcp_fasttimo()
71 {
72 	register struct inpcb *inp;
73 	register struct tcpcb *tp;
74 	int s = splnet();
75 
76 	inp = tcb.inp_next;
77 	if (inp)
78 	for (; inp != &tcb; inp = inp->inp_next)
79 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
80 		    (tp->t_flags & TF_DELACK)) {
81 			tp->t_flags &= ~TF_DELACK;
82 			tp->t_flags |= TF_ACKNOW;
83 			tcpstat.tcps_delack++;
84 			(void) tcp_output(tp);
85 		}
86 	splx(s);
87 }
88 
89 /*
90  * Tcp protocol timeout routine called every 500 ms.
91  * Updates the timers in all active tcb's and
92  * causes finite state machine actions if timers expire.
93  */
94 void
95 tcp_slowtimo()
96 {
97 	register struct inpcb *ip, *ipnxt;
98 	register struct tcpcb *tp;
99 	int s = splnet();
100 	register int i;
101 
102 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
103 	/*
104 	 * Search through tcb's and update active timers.
105 	 */
106 	ip = tcb.inp_next;
107 	if (ip == 0) {
108 		splx(s);
109 		return;
110 	}
111 	for (; ip != &tcb; ip = ipnxt) {
112 		ipnxt = ip->inp_next;
113 		tp = intotcpcb(ip);
114 		if (tp == 0)
115 			continue;
116 		for (i = 0; i < TCPT_NTIMERS; i++) {
117 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
118 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
119 				    PRU_SLOWTIMO, (struct mbuf *)0,
120 				    (struct mbuf *)i, (struct mbuf *)0);
121 				if (ipnxt->inp_prev != ip)
122 					goto tpgone;
123 			}
124 		}
125 		tp->t_idle++;
126 		tp->t_duration++;
127 		if (tp->t_rtt)
128 			tp->t_rtt++;
129 tpgone:
130 		;
131 	}
132 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
133 #ifdef TCP_COMPAT_42
134 	if ((int)tcp_iss < 0)
135 		tcp_iss = 0;				/* XXX */
136 #endif
137 	tcp_now++;					/* for timestamps */
138 	splx(s);
139 }
140 #ifndef TUBA_INCLUDE
141 
142 /*
143  * Cancel all timers for TCP tp.
144  */
145 void
146 tcp_canceltimers(tp)
147 	struct tcpcb *tp;
148 {
149 	register int i;
150 
151 	for (i = 0; i < TCPT_NTIMERS; i++)
152 		tp->t_timer[i] = 0;
153 }
154 
155 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
156     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
157 
158 /*
159  * TCP timer processing.
160  */
161 struct tcpcb *
162 tcp_timers(tp, timer)
163 	register struct tcpcb *tp;
164 	int timer;
165 {
166 	register int rexmt;
167 
168 	switch (timer) {
169 
170 	/*
171 	 * 2 MSL timeout in shutdown went off.  If we're closed but
172 	 * still waiting for peer to close and connection has been idle
173 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
174 	 * control block.  Otherwise, check again in a bit.
175 	 */
176 	case TCPT_2MSL:
177 		if (tp->t_state != TCPS_TIME_WAIT &&
178 		    tp->t_idle <= tcp_maxidle)
179 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
180 		else
181 			tp = tcp_close(tp);
182 		break;
183 
184 	/*
185 	 * Retransmission timer went off.  Message has not
186 	 * been acked within retransmit interval.  Back off
187 	 * to a longer retransmit interval and retransmit one segment.
188 	 */
189 	case TCPT_REXMT:
190 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
191 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
192 			tcpstat.tcps_timeoutdrop++;
193 			tp = tcp_drop(tp, tp->t_softerror ?
194 			    tp->t_softerror : ETIMEDOUT);
195 			break;
196 		}
197 		tcpstat.tcps_rexmttimeo++;
198 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
199 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
200 		    tp->t_rttmin, TCPTV_REXMTMAX);
201 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
202 		/*
203 		 * If losing, let the lower level know and try for
204 		 * a better route.  Also, if we backed off this far,
205 		 * our srtt estimate is probably bogus.  Clobber it
206 		 * so we'll take the next rtt measurement as our srtt;
207 		 * move the current srtt into rttvar to keep the current
208 		 * retransmit times until then.
209 		 */
210 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
211 			in_losing(tp->t_inpcb);
212 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
213 			tp->t_srtt = 0;
214 		}
215 		tp->snd_nxt = tp->snd_una;
216 		/*
217 		 * Force a segment to be sent.
218 		 */
219 		tp->t_flags |= TF_ACKNOW;
220 		/*
221 		 * If timing a segment in this window, stop the timer.
222 		 */
223 		tp->t_rtt = 0;
224 		/*
225 		 * Close the congestion window down to one segment
226 		 * (we'll open it by one segment for each ack we get).
227 		 * Since we probably have a window's worth of unacked
228 		 * data accumulated, this "slow start" keeps us from
229 		 * dumping all that data as back-to-back packets (which
230 		 * might overwhelm an intermediate gateway).
231 		 *
232 		 * There are two phases to the opening: Initially we
233 		 * open by one mss on each ack.  This makes the window
234 		 * size increase exponentially with time.  If the
235 		 * window is larger than the path can handle, this
236 		 * exponential growth results in dropped packet(s)
237 		 * almost immediately.  To get more time between
238 		 * drops but still "push" the network to take advantage
239 		 * of improving conditions, we switch from exponential
240 		 * to linear window opening at some threshhold size.
241 		 * For a threshhold, we use half the current window
242 		 * size, truncated to a multiple of the mss.
243 		 *
244 		 * (the minimum cwnd that will give us exponential
245 		 * growth is 2 mss.  We don't allow the threshhold
246 		 * to go below this.)
247 		 */
248 		{
249 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
250 		if (win < 2)
251 			win = 2;
252 		tp->snd_cwnd = tp->t_maxseg;
253 		tp->snd_ssthresh = win * tp->t_maxseg;
254 		tp->t_dupacks = 0;
255 		}
256 		(void) tcp_output(tp);
257 		break;
258 
259 	/*
260 	 * Persistance timer into zero window.
261 	 * Force a byte to be output, if possible.
262 	 */
263 	case TCPT_PERSIST:
264 		tcpstat.tcps_persisttimeo++;
265 		tcp_setpersist(tp);
266 		tp->t_force = 1;
267 		(void) tcp_output(tp);
268 		tp->t_force = 0;
269 		break;
270 
271 	/*
272 	 * Keep-alive timer went off; send something
273 	 * or drop connection if idle for too long.
274 	 */
275 	case TCPT_KEEP:
276 		tcpstat.tcps_keeptimeo++;
277 		if (tp->t_state < TCPS_ESTABLISHED)
278 			goto dropit;
279 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
280 		    tp->t_state <= TCPS_CLOSE_WAIT) {
281 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
282 				goto dropit;
283 			/*
284 			 * Send a packet designed to force a response
285 			 * if the peer is up and reachable:
286 			 * either an ACK if the connection is still alive,
287 			 * or an RST if the peer has closed the connection
288 			 * due to timeout or reboot.
289 			 * Using sequence number tp->snd_una-1
290 			 * causes the transmitted zero-length segment
291 			 * to lie outside the receive window;
292 			 * by the protocol spec, this requires the
293 			 * correspondent TCP to respond.
294 			 */
295 			tcpstat.tcps_keepprobe++;
296 #ifdef TCP_COMPAT_42
297 			/*
298 			 * The keepalive packet must have nonzero length
299 			 * to get a 4.2 host to respond.
300 			 */
301 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
302 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
303 #else
304 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
305 			    tp->rcv_nxt, tp->snd_una - 1, 0);
306 #endif
307 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
308 		} else
309 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
310 		break;
311 	dropit:
312 		tcpstat.tcps_keepdrops++;
313 		tp = tcp_drop(tp, ETIMEDOUT);
314 		break;
315 	}
316 	return (tp);
317 }
318 #endif /* TUBA_INCLUDE */
319