xref: /freebsd/sys/netinet/tcp_timer.c (revision afe61c15161c324a7af299a9b8457aba5afc92db)
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  */
35 
36 #ifndef TUBA_INCLUDE
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/protosw.h>
44 #include <sys/errno.h>
45 
46 #include <net/if.h>
47 #include <net/route.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/in_systm.h>
51 #include <netinet/ip.h>
52 #include <netinet/in_pcb.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/tcp.h>
55 #include <netinet/tcp_fsm.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/tcp_timer.h>
58 #include <netinet/tcp_var.h>
59 #include <netinet/tcpip.h>
60 
61 int	tcp_keepidle = TCPTV_KEEP_IDLE;
62 int	tcp_keepintvl = TCPTV_KEEPINTVL;
63 int	tcp_maxidle;
64 #endif /* TUBA_INCLUDE */
65 /*
66  * Fast timeout routine for processing delayed acks
67  */
68 void
69 tcp_fasttimo()
70 {
71 	register struct inpcb *inp;
72 	register struct tcpcb *tp;
73 	int s = splnet();
74 
75 	inp = tcb.inp_next;
76 	if (inp)
77 	for (; inp != &tcb; inp = inp->inp_next)
78 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
79 		    (tp->t_flags & TF_DELACK)) {
80 			tp->t_flags &= ~TF_DELACK;
81 			tp->t_flags |= TF_ACKNOW;
82 			tcpstat.tcps_delack++;
83 			(void) tcp_output(tp);
84 		}
85 	splx(s);
86 }
87 
88 /*
89  * Tcp protocol timeout routine called every 500 ms.
90  * Updates the timers in all active tcb's and
91  * causes finite state machine actions if timers expire.
92  */
93 void
94 tcp_slowtimo()
95 {
96 	register struct inpcb *ip, *ipnxt;
97 	register struct tcpcb *tp;
98 	int s = splnet();
99 	register int i;
100 
101 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
102 	/*
103 	 * Search through tcb's and update active timers.
104 	 */
105 	ip = tcb.inp_next;
106 	if (ip == 0) {
107 		splx(s);
108 		return;
109 	}
110 	for (; ip != &tcb; ip = ipnxt) {
111 		ipnxt = ip->inp_next;
112 		tp = intotcpcb(ip);
113 		if (tp == 0)
114 			continue;
115 		for (i = 0; i < TCPT_NTIMERS; i++) {
116 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
117 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
118 				    PRU_SLOWTIMO, (struct mbuf *)0,
119 				    (struct mbuf *)i, (struct mbuf *)0);
120 				if (ipnxt->inp_prev != ip)
121 					goto tpgone;
122 			}
123 		}
124 		tp->t_idle++;
125 		if (tp->t_rtt)
126 			tp->t_rtt++;
127 tpgone:
128 		;
129 	}
130 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
131 #ifdef TCP_COMPAT_42
132 	if ((int)tcp_iss < 0)
133 		tcp_iss = 0;				/* XXX */
134 #endif
135 	tcp_now++;					/* for timestamps */
136 	splx(s);
137 }
138 #ifndef TUBA_INCLUDE
139 
140 /*
141  * Cancel all timers for TCP tp.
142  */
143 void
144 tcp_canceltimers(tp)
145 	struct tcpcb *tp;
146 {
147 	register int i;
148 
149 	for (i = 0; i < TCPT_NTIMERS; i++)
150 		tp->t_timer[i] = 0;
151 }
152 
153 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
154     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
155 
156 /*
157  * TCP timer processing.
158  */
159 struct tcpcb *
160 tcp_timers(tp, timer)
161 	register struct tcpcb *tp;
162 	int timer;
163 {
164 	register int rexmt;
165 
166 	switch (timer) {
167 
168 	/*
169 	 * 2 MSL timeout in shutdown went off.  If we're closed but
170 	 * still waiting for peer to close and connection has been idle
171 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
172 	 * control block.  Otherwise, check again in a bit.
173 	 */
174 	case TCPT_2MSL:
175 		if (tp->t_state != TCPS_TIME_WAIT &&
176 		    tp->t_idle <= tcp_maxidle)
177 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
178 		else
179 			tp = tcp_close(tp);
180 		break;
181 
182 	/*
183 	 * Retransmission timer went off.  Message has not
184 	 * been acked within retransmit interval.  Back off
185 	 * to a longer retransmit interval and retransmit one segment.
186 	 */
187 	case TCPT_REXMT:
188 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
189 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
190 			tcpstat.tcps_timeoutdrop++;
191 			tp = tcp_drop(tp, tp->t_softerror ?
192 			    tp->t_softerror : ETIMEDOUT);
193 			break;
194 		}
195 		tcpstat.tcps_rexmttimeo++;
196 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
197 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
198 		    tp->t_rttmin, TCPTV_REXMTMAX);
199 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
200 		/*
201 		 * If losing, let the lower level know and try for
202 		 * a better route.  Also, if we backed off this far,
203 		 * our srtt estimate is probably bogus.  Clobber it
204 		 * so we'll take the next rtt measurement as our srtt;
205 		 * move the current srtt into rttvar to keep the current
206 		 * retransmit times until then.
207 		 */
208 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
209 			in_losing(tp->t_inpcb);
210 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
211 			tp->t_srtt = 0;
212 		}
213 		tp->snd_nxt = tp->snd_una;
214 		/*
215 		 * If timing a segment in this window, stop the timer.
216 		 */
217 		tp->t_rtt = 0;
218 		/*
219 		 * Close the congestion window down to one segment
220 		 * (we'll open it by one segment for each ack we get).
221 		 * Since we probably have a window's worth of unacked
222 		 * data accumulated, this "slow start" keeps us from
223 		 * dumping all that data as back-to-back packets (which
224 		 * might overwhelm an intermediate gateway).
225 		 *
226 		 * There are two phases to the opening: Initially we
227 		 * open by one mss on each ack.  This makes the window
228 		 * size increase exponentially with time.  If the
229 		 * window is larger than the path can handle, this
230 		 * exponential growth results in dropped packet(s)
231 		 * almost immediately.  To get more time between
232 		 * drops but still "push" the network to take advantage
233 		 * of improving conditions, we switch from exponential
234 		 * to linear window opening at some threshhold size.
235 		 * For a threshhold, we use half the current window
236 		 * size, truncated to a multiple of the mss.
237 		 *
238 		 * (the minimum cwnd that will give us exponential
239 		 * growth is 2 mss.  We don't allow the threshhold
240 		 * to go below this.)
241 		 */
242 		{
243 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
244 		if (win < 2)
245 			win = 2;
246 		tp->snd_cwnd = tp->t_maxseg;
247 		tp->snd_ssthresh = win * tp->t_maxseg;
248 		tp->t_dupacks = 0;
249 		}
250 		(void) tcp_output(tp);
251 		break;
252 
253 	/*
254 	 * Persistance timer into zero window.
255 	 * Force a byte to be output, if possible.
256 	 */
257 	case TCPT_PERSIST:
258 		tcpstat.tcps_persisttimeo++;
259 		tcp_setpersist(tp);
260 		tp->t_force = 1;
261 		(void) tcp_output(tp);
262 		tp->t_force = 0;
263 		break;
264 
265 	/*
266 	 * Keep-alive timer went off; send something
267 	 * or drop connection if idle for too long.
268 	 */
269 	case TCPT_KEEP:
270 		tcpstat.tcps_keeptimeo++;
271 		if (tp->t_state < TCPS_ESTABLISHED)
272 			goto dropit;
273 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
274 		    tp->t_state <= TCPS_CLOSE_WAIT) {
275 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
276 				goto dropit;
277 			/*
278 			 * Send a packet designed to force a response
279 			 * if the peer is up and reachable:
280 			 * either an ACK if the connection is still alive,
281 			 * or an RST if the peer has closed the connection
282 			 * due to timeout or reboot.
283 			 * Using sequence number tp->snd_una-1
284 			 * causes the transmitted zero-length segment
285 			 * to lie outside the receive window;
286 			 * by the protocol spec, this requires the
287 			 * correspondent TCP to respond.
288 			 */
289 			tcpstat.tcps_keepprobe++;
290 #ifdef TCP_COMPAT_42
291 			/*
292 			 * The keepalive packet must have nonzero length
293 			 * to get a 4.2 host to respond.
294 			 */
295 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
296 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
297 #else
298 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
299 			    tp->rcv_nxt, tp->snd_una - 1, 0);
300 #endif
301 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
302 		} else
303 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
304 		break;
305 	dropit:
306 		tcpstat.tcps_keepdrops++;
307 		tp = tcp_drop(tp, ETIMEDOUT);
308 		break;
309 	}
310 	return (tp);
311 }
312 #endif /* TUBA_INCLUDE */
313