tcp_timer.c revision 15262 
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 1.15 1996/04/04 11:17:04 phk Exp $
35 */
36
37#ifndef TUBA_INCLUDE
38#include <sys/param.h>
39#include <sys/queue.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/sysctl.h>
43#include <sys/malloc.h>
44#include <sys/mbuf.h>
45#include <sys/socket.h>
46#include <sys/socketvar.h>
47#include <sys/protosw.h>
48#include <sys/errno.h>
49
50#include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
51
52#include <net/if.h>
53#include <net/route.h>
54
55#include <netinet/in.h>
56#include <netinet/in_systm.h>
57#include <netinet/ip.h>
58#include <netinet/in_pcb.h>
59#include <netinet/ip_var.h>
60#include <netinet/tcp.h>
61#include <netinet/tcp_fsm.h>
62#include <netinet/tcp_seq.h>
63#include <netinet/tcp_timer.h>
64#include <netinet/tcp_var.h>
65#include <netinet/tcpip.h>
66
67int	tcp_keepidle = TCPTV_KEEP_IDLE;
68SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
69	CTLFLAG_RW, &tcp_keepidle , 0, "");
70
71static int	tcp_keepintvl = TCPTV_KEEPINTVL;
72SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
73	CTLFLAG_RW, &tcp_keepintvl , 0, "");
74
75static int	always_keepalive = 0;
76SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive,
77	CTLFLAG_RW, &always_keepalive , 0, "");
78
79static int	tcp_keepcnt = TCPTV_KEEPCNT;
80	/* max idle probes */
81static int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
82	/* max idle time in persist */
83int	tcp_maxidle;
84#else /* TUBA_INCLUDE */
85
86static	int tcp_maxpersistidle;
87#endif /* TUBA_INCLUDE */
88
89/*
90 * Fast timeout routine for processing delayed acks
91 */
92void
93tcp_fasttimo()
94{
95	register struct inpcb *inp;
96	register struct tcpcb *tp;
97	int s;
98
99	s = splnet();
100
101	for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
102		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
103		    (tp->t_flags & TF_DELACK)) {
104			tp->t_flags &= ~TF_DELACK;
105			tp->t_flags |= TF_ACKNOW;
106			tcpstat.tcps_delack++;
107			(void) tcp_output(tp);
108		}
109	}
110	splx(s);
111}
112
113/*
114 * Tcp protocol timeout routine called every 500 ms.
115 * Updates the timers in all active tcb's and
116 * causes finite state machine actions if timers expire.
117 */
118void
119tcp_slowtimo()
120{
121	register struct inpcb *ip, *ipnxt;
122	register struct tcpcb *tp;
123	register int i;
124	int s;
125
126	s = splnet();
127
128	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
129
130	ip = tcb.lh_first;
131	if (ip == NULL) {
132		splx(s);
133		return;
134	}
135	/*
136	 * Search through tcb's and update active timers.
137	 */
138	for (; ip != NULL; ip = ipnxt) {
139		ipnxt = ip->inp_list.le_next;
140		tp = intotcpcb(ip);
141		if (tp == 0 || tp->t_state == TCPS_LISTEN)
142			continue;
143		for (i = 0; i < TCPT_NTIMERS; i++) {
144			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
145				if (tcp_usrreq(tp->t_inpcb->inp_socket,
146				    PRU_SLOWTIMO, (struct mbuf *)0,
147				    (struct mbuf *)i, (struct mbuf *)0) == NULL)
148					goto tpgone;
149			}
150		}
151		tp->t_idle++;
152		tp->t_duration++;
153		if (tp->t_rtt)
154			tp->t_rtt++;
155tpgone:
156		;
157	}
158	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
159#ifdef TCP_COMPAT_42
160	if ((int)tcp_iss < 0)
161		tcp_iss = TCP_ISSINCR;			/* XXX */
162#endif
163	tcp_now++;					/* for timestamps */
164	splx(s);
165}
166#ifndef TUBA_INCLUDE
167
168/*
169 * Cancel all timers for TCP tp.
170 */
171void
172tcp_canceltimers(tp)
173	struct tcpcb *tp;
174{
175	register int i;
176
177	for (i = 0; i < TCPT_NTIMERS; i++)
178		tp->t_timer[i] = 0;
179}
180
181int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
182    { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
183
184static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
185
186/*
187 * TCP timer processing.
188 */
189struct tcpcb *
190tcp_timers(tp, timer)
191	register struct tcpcb *tp;
192	int timer;
193{
194	register int rexmt;
195
196	switch (timer) {
197
198	/*
199	 * 2 MSL timeout in shutdown went off.  If we're closed but
200	 * still waiting for peer to close and connection has been idle
201	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
202	 * control block.  Otherwise, check again in a bit.
203	 */
204	case TCPT_2MSL:
205		if (tp->t_state != TCPS_TIME_WAIT &&
206		    tp->t_idle <= tcp_maxidle)
207			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
208		else
209			tp = tcp_close(tp);
210		break;
211
212	/*
213	 * Retransmission timer went off.  Message has not
214	 * been acked within retransmit interval.  Back off
215	 * to a longer retransmit interval and retransmit one segment.
216	 */
217	case TCPT_REXMT:
218		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
219			tp->t_rxtshift = TCP_MAXRXTSHIFT;
220			tcpstat.tcps_timeoutdrop++;
221			tp = tcp_drop(tp, tp->t_softerror ?
222			    tp->t_softerror : ETIMEDOUT);
223			break;
224		}
225		tcpstat.tcps_rexmttimeo++;
226		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
227		TCPT_RANGESET(tp->t_rxtcur, rexmt,
228		    tp->t_rttmin, TCPTV_REXMTMAX);
229		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
230		/*
231		 * If losing, let the lower level know and try for
232		 * a better route.  Also, if we backed off this far,
233		 * our srtt estimate is probably bogus.  Clobber it
234		 * so we'll take the next rtt measurement as our srtt;
235		 * move the current srtt into rttvar to keep the current
236		 * retransmit times until then.
237		 */
238		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
239			in_losing(tp->t_inpcb);
240			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
241			tp->t_srtt = 0;
242		}
243		tp->snd_nxt = tp->snd_una;
244		/*
245		 * Force a segment to be sent.
246		 */
247		tp->t_flags |= TF_ACKNOW;
248		/*
249		 * If timing a segment in this window, stop the timer.
250		 */
251		tp->t_rtt = 0;
252		/*
253		 * Close the congestion window down to one segment
254		 * (we'll open it by one segment for each ack we get).
255		 * Since we probably have a window's worth of unacked
256		 * data accumulated, this "slow start" keeps us from
257		 * dumping all that data as back-to-back packets (which
258		 * might overwhelm an intermediate gateway).
259		 *
260		 * There are two phases to the opening: Initially we
261		 * open by one mss on each ack.  This makes the window
262		 * size increase exponentially with time.  If the
263		 * window is larger than the path can handle, this
264		 * exponential growth results in dropped packet(s)
265		 * almost immediately.  To get more time between
266		 * drops but still "push" the network to take advantage
267		 * of improving conditions, we switch from exponential
268		 * to linear window opening at some threshhold size.
269		 * For a threshhold, we use half the current window
270		 * size, truncated to a multiple of the mss.
271		 *
272		 * (the minimum cwnd that will give us exponential
273		 * growth is 2 mss.  We don't allow the threshhold
274		 * to go below this.)
275		 */
276		{
277		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
278		if (win < 2)
279			win = 2;
280		tp->snd_cwnd = tp->t_maxseg;
281		tp->snd_ssthresh = win * tp->t_maxseg;
282		tp->t_dupacks = 0;
283		}
284		(void) tcp_output(tp);
285		break;
286
287	/*
288	 * Persistance timer into zero window.
289	 * Force a byte to be output, if possible.
290	 */
291	case TCPT_PERSIST:
292		tcpstat.tcps_persisttimeo++;
293		/*
294		 * Hack: if the peer is dead/unreachable, we do not
295		 * time out if the window is closed.  After a full
296		 * backoff, drop the connection if the idle time
297		 * (no responses to probes) reaches the maximum
298		 * backoff that we would use if retransmitting.
299		 */
300		if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
301		    (tp->t_idle >= tcp_maxpersistidle ||
302		    tp->t_idle >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
303		        tcpstat.tcps_persistdrop++;
304		        tp = tcp_drop(tp, ETIMEDOUT);
305		        break;
306		}
307		tcp_setpersist(tp);
308		tp->t_force = 1;
309		(void) tcp_output(tp);
310		tp->t_force = 0;
311		break;
312
313	/*
314	 * Keep-alive timer went off; send something
315	 * or drop connection if idle for too long.
316	 */
317	case TCPT_KEEP:
318		tcpstat.tcps_keeptimeo++;
319		if (tp->t_state < TCPS_ESTABLISHED)
320			goto dropit;
321		if ((always_keepalive ||
322		    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
323		    tp->t_state <= TCPS_CLOSING) {
324		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
325				goto dropit;
326			/*
327			 * Send a packet designed to force a response
328			 * if the peer is up and reachable:
329			 * either an ACK if the connection is still alive,
330			 * or an RST if the peer has closed the connection
331			 * due to timeout or reboot.
332			 * Using sequence number tp->snd_una-1
333			 * causes the transmitted zero-length segment
334			 * to lie outside the receive window;
335			 * by the protocol spec, this requires the
336			 * correspondent TCP to respond.
337			 */
338			tcpstat.tcps_keepprobe++;
339#ifdef TCP_COMPAT_42
340			/*
341			 * The keepalive packet must have nonzero length
342			 * to get a 4.2 host to respond.
343			 */
344			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
345			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
346#else
347			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
348			    tp->rcv_nxt, tp->snd_una - 1, 0);
349#endif
350			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
351		} else
352			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
353		break;
354	dropit:
355		tcpstat.tcps_keepdrops++;
356		tp = tcp_drop(tp, ETIMEDOUT);
357		break;
358	}
359	return (tp);
360}
361#endif /* TUBA_INCLUDE */
362