xref: /csrg-svn/sys/netinet/tcp_timer.c (revision 57433)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)tcp_timer.c	7.20 (Berkeley) 01/08/93
8  */
9 
10 #include <sys/param.h>
11 #include <sys/systm.h>
12 #include <sys/malloc.h>
13 #include <sys/mbuf.h>
14 #include <sys/socket.h>
15 #include <sys/socketvar.h>
16 #include <sys/protosw.h>
17 #include <sys/errno.h>
18 
19 #include <net/if.h>
20 #include <net/route.h>
21 
22 #include <netinet/in.h>
23 #include <netinet/in_systm.h>
24 #include <netinet/ip.h>
25 #include <netinet/in_pcb.h>
26 #include <netinet/ip_var.h>
27 #include <netinet/tcp.h>
28 #include <netinet/tcp_fsm.h>
29 #include <netinet/tcp_seq.h>
30 #include <netinet/tcp_timer.h>
31 #include <netinet/tcp_var.h>
32 #include <netinet/tcpip.h>
33 
34 int	tcp_keepidle = TCPTV_KEEP_IDLE;
35 int	tcp_keepintvl = TCPTV_KEEPINTVL;
36 int	tcp_maxidle;
37 /*
38  * Fast timeout routine for processing delayed acks
39  */
40 tcp_fasttimo()
41 {
42 	register struct inpcb *inp;
43 	register struct tcpcb *tp;
44 	int s = splnet();
45 
46 	inp = tcb.inp_next;
47 	if (inp)
48 	for (; inp != &tcb; inp = inp->inp_next)
49 		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
50 		    (tp->t_flags & TF_DELACK)) {
51 			tp->t_flags &= ~TF_DELACK;
52 			tp->t_flags |= TF_ACKNOW;
53 			tcpstat.tcps_delack++;
54 			(void) tcp_output(tp);
55 		}
56 	splx(s);
57 }
58 
59 /*
60  * Tcp protocol timeout routine called every 500 ms.
61  * Updates the timers in all active tcb's and
62  * causes finite state machine actions if timers expire.
63  */
64 tcp_slowtimo()
65 {
66 	register struct inpcb *ip, *ipnxt;
67 	register struct tcpcb *tp;
68 	int s = splnet();
69 	register int i;
70 
71 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
72 	/*
73 	 * Search through tcb's and update active timers.
74 	 */
75 	ip = tcb.inp_next;
76 	if (ip == 0) {
77 		splx(s);
78 		return;
79 	}
80 	for (; ip != &tcb; ip = ipnxt) {
81 		ipnxt = ip->inp_next;
82 		tp = intotcpcb(ip);
83 		if (tp == 0)
84 			continue;
85 		for (i = 0; i < TCPT_NTIMERS; i++) {
86 			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
87 				(void) tcp_usrreq(tp->t_inpcb->inp_socket,
88 				    PRU_SLOWTIMO, (struct mbuf *)0,
89 				    (struct mbuf *)i, (struct mbuf *)0);
90 				if (ipnxt->inp_prev != ip)
91 					goto tpgone;
92 			}
93 		}
94 		tp->t_idle++;
95 		if (tp->t_rtt)
96 			tp->t_rtt++;
97 tpgone:
98 		;
99 	}
100 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
101 #ifdef TCP_COMPAT_42
102 	if ((int)tcp_iss < 0)
103 		tcp_iss = 0;				/* XXX */
104 #endif
105 	tcp_now++;					/* for timestamps */
106 	splx(s);
107 }
108 
109 /*
110  * Cancel all timers for TCP tp.
111  */
112 tcp_canceltimers(tp)
113 	struct tcpcb *tp;
114 {
115 	register int i;
116 
117 	for (i = 0; i < TCPT_NTIMERS; i++)
118 		tp->t_timer[i] = 0;
119 }
120 
121 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
122     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
123 
124 /*
125  * TCP timer processing.
126  */
127 struct tcpcb *
128 tcp_timers(tp, timer)
129 	register struct tcpcb *tp;
130 	int timer;
131 {
132 	register int rexmt;
133 
134 	switch (timer) {
135 
136 	/*
137 	 * 2 MSL timeout in shutdown went off.  If we're closed but
138 	 * still waiting for peer to close and connection has been idle
139 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
140 	 * control block.  Otherwise, check again in a bit.
141 	 */
142 	case TCPT_2MSL:
143 		if (tp->t_state != TCPS_TIME_WAIT &&
144 		    tp->t_idle <= tcp_maxidle)
145 			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
146 		else
147 			tp = tcp_close(tp);
148 		break;
149 
150 	/*
151 	 * Retransmission timer went off.  Message has not
152 	 * been acked within retransmit interval.  Back off
153 	 * to a longer retransmit interval and retransmit one segment.
154 	 */
155 	case TCPT_REXMT:
156 		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
157 			tp->t_rxtshift = TCP_MAXRXTSHIFT;
158 			tcpstat.tcps_timeoutdrop++;
159 			tp = tcp_drop(tp, tp->t_softerror ?
160 			    tp->t_softerror : ETIMEDOUT);
161 			break;
162 		}
163 		tcpstat.tcps_rexmttimeo++;
164 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
165 		TCPT_RANGESET(tp->t_rxtcur, rexmt,
166 		    tp->t_rttmin, TCPTV_REXMTMAX);
167 		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
168 		/*
169 		 * If losing, let the lower level know and try for
170 		 * a better route.  Also, if we backed off this far,
171 		 * our srtt estimate is probably bogus.  Clobber it
172 		 * so we'll take the next rtt measurement as our srtt;
173 		 * move the current srtt into rttvar to keep the current
174 		 * retransmit times until then.
175 		 */
176 		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
177 			in_losing(tp->t_inpcb);
178 			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
179 			tp->t_srtt = 0;
180 		}
181 		tp->snd_nxt = tp->snd_una;
182 		/*
183 		 * If timing a segment in this window, stop the timer.
184 		 */
185 		tp->t_rtt = 0;
186 		/*
187 		 * Close the congestion window down to one segment
188 		 * (we'll open it by one segment for each ack we get).
189 		 * Since we probably have a window's worth of unacked
190 		 * data accumulated, this "slow start" keeps us from
191 		 * dumping all that data as back-to-back packets (which
192 		 * might overwhelm an intermediate gateway).
193 		 *
194 		 * There are two phases to the opening: Initially we
195 		 * open by one mss on each ack.  This makes the window
196 		 * size increase exponentially with time.  If the
197 		 * window is larger than the path can handle, this
198 		 * exponential growth results in dropped packet(s)
199 		 * almost immediately.  To get more time between
200 		 * drops but still "push" the network to take advantage
201 		 * of improving conditions, we switch from exponential
202 		 * to linear window opening at some threshhold size.
203 		 * For a threshhold, we use half the current window
204 		 * size, truncated to a multiple of the mss.
205 		 *
206 		 * (the minimum cwnd that will give us exponential
207 		 * growth is 2 mss.  We don't allow the threshhold
208 		 * to go below this.)
209 		 */
210 		{
211 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
212 		if (win < 2)
213 			win = 2;
214 		tp->snd_cwnd = tp->t_maxseg;
215 		tp->snd_ssthresh = win * tp->t_maxseg;
216 		tp->t_dupacks = 0;
217 		}
218 		(void) tcp_output(tp);
219 		break;
220 
221 	/*
222 	 * Persistance timer into zero window.
223 	 * Force a byte to be output, if possible.
224 	 */
225 	case TCPT_PERSIST:
226 		tcpstat.tcps_persisttimeo++;
227 		tcp_setpersist(tp);
228 		tp->t_force = 1;
229 		(void) tcp_output(tp);
230 		tp->t_force = 0;
231 		break;
232 
233 	/*
234 	 * Keep-alive timer went off; send something
235 	 * or drop connection if idle for too long.
236 	 */
237 	case TCPT_KEEP:
238 		tcpstat.tcps_keeptimeo++;
239 		if (tp->t_state < TCPS_ESTABLISHED)
240 			goto dropit;
241 		if (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE &&
242 		    tp->t_state <= TCPS_CLOSE_WAIT) {
243 		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
244 				goto dropit;
245 			/*
246 			 * Send a packet designed to force a response
247 			 * if the peer is up and reachable:
248 			 * either an ACK if the connection is still alive,
249 			 * or an RST if the peer has closed the connection
250 			 * due to timeout or reboot.
251 			 * Using sequence number tp->snd_una-1
252 			 * causes the transmitted zero-length segment
253 			 * to lie outside the receive window;
254 			 * by the protocol spec, this requires the
255 			 * correspondent TCP to respond.
256 			 */
257 			tcpstat.tcps_keepprobe++;
258 #ifdef TCP_COMPAT_42
259 			/*
260 			 * The keepalive packet must have nonzero length
261 			 * to get a 4.2 host to respond.
262 			 */
263 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
264 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
265 #else
266 			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
267 			    tp->rcv_nxt, tp->snd_una - 1, 0);
268 #endif
269 			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
270 		} else
271 			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
272 		break;
273 	dropit:
274 		tcpstat.tcps_keepdrops++;
275 		tp = tcp_drop(tp, ETIMEDOUT);
276 		break;
277 	}
278 	return (tp);
279 }
280