xref: /csrg-svn/sys/netinet/tcp_timer.c (revision 33746)
1 /*
2  * Copyright (c) 1982, 1986 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are permitted
6  * provided that this notice is preserved and that due credit is given
7  * to the University of California at Berkeley. The name of the University
8  * may not be used to endorse or promote products derived from this
9  * software without specific prior written permission. This software
10  * is provided ``as is'' without express or implied warranty.
11  *
12  *	@(#)tcp_timer.c	7.12 (Berkeley) 03/16/88
13  */
14 
15 #include "param.h"
16 #include "systm.h"
17 #include "mbuf.h"
18 #include "socket.h"
19 #include "socketvar.h"
20 #include "protosw.h"
21 #include "errno.h"
22 
23 #include "../net/if.h"
24 #include "../net/route.h"
25 
26 #include "in.h"
27 #include "in_pcb.h"
28 #include "in_systm.h"
29 #include "ip.h"
30 #include "ip_var.h"
31 #include "tcp.h"
32 #include "tcp_fsm.h"
33 #include "tcp_seq.h"
34 #include "tcp_timer.h"
35 #include "tcp_var.h"
36 #include "tcpip.h"
37 
38 int	tcpnodelack = 0;
39 int	tcp_keepidle = TCPTV_KEEP_IDLE;
40 int	tcp_keepintvl = TCPTV_KEEPINTVL;
41 int	tcp_maxidle;
42 /*
43  * Fast timeout routine for processing delayed acks
44  */
45 tcp_fasttimo()
46 {
47 	register struct inpcb *inp;
48 	register struct tcpcb *tp;
49 	int s = splnet();
50 
51 	inp = tcb.inp_next;
52 	if (inp)
53 	for (; inp != &tcb; inp = inp->inp_next)
54 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
55 		    (tp->t_flags & TF_DELACK)) {
56 			tp->t_flags &= ~TF_DELACK;
57 			tp->t_flags |= TF_ACKNOW;
58 			tcpstat.tcps_delack++;
59 			(void) tcp_output(tp);
60 		}
61 	splx(s);
62 }
63 
64 /*
65  * Tcp protocol timeout routine called every 500 ms.
66  * Updates the timers in all active tcb's and
67  * causes finite state machine actions if timers expire.
68  */
69 tcp_slowtimo()
70 {
71 	register struct inpcb *ip, *ipnxt;
72 	register struct tcpcb *tp;
73 	int s = splnet();
74 	register int i;
75 
76 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
77 	/*
78 	 * Search through tcb's and update active timers.
79 	 */
80 	ip = tcb.inp_next;
81 	if (ip == 0) {
82 		splx(s);
83 		return;
84 	}
85 	for (; ip != &tcb; ip = ipnxt) {
86 		ipnxt = ip->inp_next;
87 		tp = intotcpcb(ip);
88 		if (tp == 0)
89 			continue;
90 		for (i = 0; i < TCPT_NTIMERS; i++) {
91 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
92 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
93 				    PRU_SLOWTIMO, (struct mbuf *)0,
94 				    (struct mbuf *)i, (struct mbuf *)0);
95 				if (ipnxt->inp_prev != ip)
96 					goto tpgone;
97 			}
98 		}
99 		tp->t_idle++;
100 		if (tp->t_rtt)
101 			tp->t_rtt++;
102 tpgone:
103 		;
104 	}
105 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
106 #ifdef TCP_COMPAT_42
107 	if ((int)tcp_iss < 0)
108 		tcp_iss = 0;				/* XXX */
109 #endif
110 	splx(s);
111 }
112 
113 /*
114  * Cancel all timers for TCP tp.
115  */
116 tcp_canceltimers(tp)
117 	struct tcpcb *tp;
118 {
119 	register int i;
120 
121 	for (i = 0; i < TCPT_NTIMERS; i++)
122 		tp->t_timer[i] = 0;
123 }
124 
125 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
126     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
127 
128 /*
129  * TCP timer processing.
130  */
131 struct tcpcb *
132 tcp_timers(tp, timer)
133 	register struct tcpcb *tp;
134 	int timer;
135 {
136 	register int rexmt;
137 
138 	switch (timer) {
139 
140 	/*
141 	 * 2 MSL timeout in shutdown went off.  If we're closed but
142 	 * still waiting for peer to close and connection has been idle
143 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
144 	 * control block.  Otherwise, check again in a bit.
145 	 */
146 	case TCPT_2MSL:
147 		if (tp->t_state != TCPS_TIME_WAIT &&
148 		    tp->t_idle <= tcp_maxidle)
149 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
150 		else
151 			tp = tcp_close(tp);
152 		break;
153 
154 	/*
155 	 * Retransmission timer went off.  Message has not
156 	 * been acked within retransmit interval.  Back off
157 	 * to a longer retransmit interval and retransmit one segment.
158 	 */
159 	case TCPT_REXMT:
160 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
161 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
162 			tcpstat.tcps_timeoutdrop++;
163 			tp = tcp_drop(tp, ETIMEDOUT);
164 			break;
165 		}
166 		tcpstat.tcps_rexmttimeo++;
167 		rexmt = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1;
168 		rexmt *= tcp_backoff[tp->t_rxtshift];
169 		TCPT_RANGESET(tp->t_rxtcur, rexmt, TCPTV_MIN, TCPTV_REXMTMAX);
170 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
171 		/*
172 		 * If losing, let the lower level know and try for
173 		 * a better route.  Also, if we backed off this far,
174 		 * our srtt estimate is probably bogus.  Clobber it
175 		 * so we'll take the next rtt measurement as our srtt;
176 		 * move the current srtt into rttvar to keep the current
177 		 * retransmit times until then.
178 		 */
179 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
180 			in_losing(tp->t_inpcb);
181 			tp->t_rttvar += (tp->t_srtt >> 2);
182 			tp->t_srtt = 0;
183 		}
184 		tp->snd_nxt = tp->snd_una;
185 		/*
186 		 * If timing a segment in this window, stop the timer.
187 		 */
188 		tp->t_rtt = 0;
189 		/*
190 		 * Close the congestion window down to one segment
191 		 * (we'll open it by one segment for each ack we get).
192 		 * Since we probably have a window's worth of unacked
193 		 * data accumulated, this "slow start" keeps us from
194 		 * dumping all that data as back-to-back packets (which
195 		 * might overwhelm an intermediate gateway).
196 		 *
197 		 * There are two phases to the opening: Initially we
198 		 * open by one mss on each ack.  This makes the window
199 		 * size increase exponentially with time.  If the
200 		 * window is larger than the path can handle, this
201 		 * exponential growth results in dropped packet(s)
202 		 * almost immediately.  To get more time between
203 		 * drops but still "push" the network to take advantage
204 		 * of improving conditions, we switch from exponential
205 		 * to linear window opening at some threshhold size.
206 		 * For a threshhold, we use half the current window
207 		 * size, truncated to a multiple of the mss.
208 		 *
209 		 * (the minimum cwnd that will give us exponential
210 		 * growth is 2 mss.  We don't allow the threshhold
211 		 * to go below this.)
212 		 */
213 		{
214 		u_int win = MIN(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
215 		if (win < 2)
216 			win = 2;
217 		tp->snd_cwnd = tp->t_maxseg;
218 		tp->snd_ssthresh = win * tp->t_maxseg;
219 		}
220 		(void) tcp_output(tp);
221 		break;
222 
223 	/*
224 	 * Persistance timer into zero window.
225 	 * Force a byte to be output, if possible.
226 	 */
227 	case TCPT_PERSIST:
228 		tcpstat.tcps_persisttimeo++;
229 		tcp_setpersist(tp);
230 		tp->t_force = 1;
231 		(void) tcp_output(tp);
232 		tp->t_force = 0;
233 		break;
234 
235 	/*
236 	 * Keep-alive timer went off; send something
237 	 * or drop connection if idle for too long.
238 	 */
239 	case TCPT_KEEP:
240 		tcpstat.tcps_keeptimeo++;
241 		if (tp->t_state < TCPS_ESTABLISHED)
242 			goto dropit;
243 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
244 		    tp->t_state <= TCPS_CLOSE_WAIT) {
245 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
246 				goto dropit;
247 			/*
248 			 * Send a packet designed to force a response
249 			 * if the peer is up and reachable:
250 			 * either an ACK if the connection is still alive,
251 			 * or an RST if the peer has closed the connection
252 			 * due to timeout or reboot.
253 			 * Using sequence number tp->snd_una-1
254 			 * causes the transmitted zero-length segment
255 			 * to lie outside the receive window;
256 			 * by the protocol spec, this requires the
257 			 * correspondent TCP to respond.
258 			 */
259 			tcpstat.tcps_keepprobe++;
260 #ifdef TCP_COMPAT_42
261 			/*
262 			 * The keepalive packet must have nonzero length
263 			 * to get a 4.2 host to respond.
264 			 */
265 			tcp_respond(tp, tp->t_template,
266 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
267 #else
268 			tcp_respond(tp, tp->t_template,
269 			    tp->rcv_nxt, tp->snd_una - 1, 0);
270 #endif
271 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
272 		} else
273 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
274 		break;
275 	dropit:
276 		tcpstat.tcps_keepdrops++;
277 		tp = tcp_drop(tp, ETIMEDOUT);
278 		break;
279 	}
280 	return (tp);
281 }
282