tcp_timer.c revision 35056 
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 1.26 1998/02/26 05:25:33 dg Exp $
35 */
36
37#include "opt_compat.h"
38#include "opt_tcpdebug.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/kernel.h>
43#include <sys/sysctl.h>
44#include <sys/socket.h>
45#include <sys/socketvar.h>
46#include <sys/protosw.h>
47
48#include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
49
50#include <net/route.h>
51
52#include <netinet/in.h>
53#include <netinet/in_systm.h>
54#include <netinet/in_pcb.h>
55#include <netinet/ip_var.h>
56#include <netinet/tcp.h>
57#include <netinet/tcp_fsm.h>
58#include <netinet/tcp_seq.h>
59#include <netinet/tcp_timer.h>
60#include <netinet/tcp_var.h>
61#include <netinet/tcpip.h>
62#ifdef TCPDEBUG
63#include <netinet/tcp_debug.h>
64#endif
65
66int	tcp_keepinit = TCPTV_KEEP_INIT;
67SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
68	CTLFLAG_RW, &tcp_keepinit , 0, "");
69
70int	tcp_keepidle = TCPTV_KEEP_IDLE;
71SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
72	CTLFLAG_RW, &tcp_keepidle , 0, "");
73
74static int	tcp_keepintvl = TCPTV_KEEPINTVL;
75SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
76	CTLFLAG_RW, &tcp_keepintvl , 0, "");
77
78static int	always_keepalive = 0;
79SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive,
80	CTLFLAG_RW, &always_keepalive , 0, "");
81
82static int	tcp_keepcnt = TCPTV_KEEPCNT;
83	/* max idle probes */
84static int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
85	/* max idle time in persist */
86int	tcp_maxidle;
87
88/*
89 * Fast timeout routine for processing delayed acks
90 */
91void
92tcp_fasttimo()
93{
94	register struct inpcb *inp;
95	register struct tcpcb *tp;
96	int s;
97
98	if (tcp_delack_enabled) {
99		s = splnet();
100		for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
101			if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
102			    (tp->t_flags & TF_DELACK)) {
103				tp->t_flags &= ~TF_DELACK;
104				tp->t_flags |= TF_ACKNOW;
105				tcpstat.tcps_delack++;
106				(void) tcp_output(tp);
107			}
108		}
109		splx(s);
110	}
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#ifdef TCPDEBUG
126	int ostate;
127#endif
128
129	s = splnet();
130
131	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
132
133	ip = tcb.lh_first;
134	if (ip == NULL) {
135		splx(s);
136		return;
137	}
138	/*
139	 * Search through tcb's and update active timers.
140	 */
141	for (; ip != NULL; ip = ipnxt) {
142		ipnxt = ip->inp_list.le_next;
143		tp = intotcpcb(ip);
144		if (tp == 0 || tp->t_state == TCPS_LISTEN)
145			continue;
146		for (i = 0; i < TCPT_NTIMERS; i++) {
147			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
148#ifdef TCPDEBUG
149				ostate = tp->t_state;
150#endif
151				tp = tcp_timers(tp, i);
152				if (tp == NULL)
153					goto tpgone;
154#ifdef TCPDEBUG
155				if (tp->t_inpcb->inp_socket->so_options
156				    & SO_DEBUG)
157					tcp_trace(TA_USER, ostate, tp,
158						  (struct tcpiphdr *)0,
159						  PRU_SLOWTIMO);
160#endif
161			}
162		}
163		tp->t_idle++;
164		tp->t_duration++;
165		if (tp->t_rtt)
166			tp->t_rtt++;
167tpgone:
168		;
169	}
170	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
171#ifdef TCP_COMPAT_42
172	if ((int)tcp_iss < 0)
173		tcp_iss = TCP_ISSINCR;			/* XXX */
174#endif
175	tcp_now++;					/* for timestamps */
176	splx(s);
177}
178
179/*
180 * Cancel all timers for TCP tp.
181 */
182void
183tcp_canceltimers(tp)
184	struct tcpcb *tp;
185{
186	register int i;
187
188	for (i = 0; i < TCPT_NTIMERS; i++)
189		tp->t_timer[i] = 0;
190}
191
192int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
193    { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
194
195static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
196
197/*
198 * TCP timer processing.
199 */
200struct tcpcb *
201tcp_timers(tp, timer)
202	register struct tcpcb *tp;
203	int timer;
204{
205	register int rexmt;
206
207	switch (timer) {
208
209	/*
210	 * 2 MSL timeout in shutdown went off.  If we're closed but
211	 * still waiting for peer to close and connection has been idle
212	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
213	 * control block.  Otherwise, check again in a bit.
214	 */
215	case TCPT_2MSL:
216		if (tp->t_state != TCPS_TIME_WAIT &&
217		    tp->t_idle <= tcp_maxidle)
218			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
219		else
220			tp = tcp_close(tp);
221		break;
222
223	/*
224	 * Retransmission timer went off.  Message has not
225	 * been acked within retransmit interval.  Back off
226	 * to a longer retransmit interval and retransmit one segment.
227	 */
228	case TCPT_REXMT:
229		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
230			tp->t_rxtshift = TCP_MAXRXTSHIFT;
231			tcpstat.tcps_timeoutdrop++;
232			tp = tcp_drop(tp, tp->t_softerror ?
233			    tp->t_softerror : ETIMEDOUT);
234			break;
235		}
236		tcpstat.tcps_rexmttimeo++;
237		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
238		TCPT_RANGESET(tp->t_rxtcur, rexmt,
239		    tp->t_rttmin, TCPTV_REXMTMAX);
240		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
241		/*
242		 * If losing, let the lower level know and try for
243		 * a better route.  Also, if we backed off this far,
244		 * our srtt estimate is probably bogus.  Clobber it
245		 * so we'll take the next rtt measurement as our srtt;
246		 * move the current srtt into rttvar to keep the current
247		 * retransmit times until then.
248		 */
249		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
250			in_losing(tp->t_inpcb);
251			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
252			tp->t_srtt = 0;
253		}
254		tp->snd_nxt = tp->snd_una;
255		/*
256		 * Force a segment to be sent.
257		 */
258		tp->t_flags |= TF_ACKNOW;
259		/*
260		 * If timing a segment in this window, stop the timer.
261		 */
262		tp->t_rtt = 0;
263		/*
264		 * Close the congestion window down to one segment
265		 * (we'll open it by one segment for each ack we get).
266		 * Since we probably have a window's worth of unacked
267		 * data accumulated, this "slow start" keeps us from
268		 * dumping all that data as back-to-back packets (which
269		 * might overwhelm an intermediate gateway).
270		 *
271		 * There are two phases to the opening: Initially we
272		 * open by one mss on each ack.  This makes the window
273		 * size increase exponentially with time.  If the
274		 * window is larger than the path can handle, this
275		 * exponential growth results in dropped packet(s)
276		 * almost immediately.  To get more time between
277		 * drops but still "push" the network to take advantage
278		 * of improving conditions, we switch from exponential
279		 * to linear window opening at some threshhold size.
280		 * For a threshhold, we use half the current window
281		 * size, truncated to a multiple of the mss.
282		 *
283		 * (the minimum cwnd that will give us exponential
284		 * growth is 2 mss.  We don't allow the threshhold
285		 * to go below this.)
286		 */
287		{
288		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
289		if (win < 2)
290			win = 2;
291		tp->snd_cwnd = tp->t_maxseg;
292		tp->snd_ssthresh = win * tp->t_maxseg;
293		tp->t_dupacks = 0;
294		}
295		(void) tcp_output(tp);
296		break;
297
298	/*
299	 * Persistance timer into zero window.
300	 * Force a byte to be output, if possible.
301	 */
302	case TCPT_PERSIST:
303		tcpstat.tcps_persisttimeo++;
304		/*
305		 * Hack: if the peer is dead/unreachable, we do not
306		 * time out if the window is closed.  After a full
307		 * backoff, drop the connection if the idle time
308		 * (no responses to probes) reaches the maximum
309		 * backoff that we would use if retransmitting.
310		 */
311		if (tp->t_rxtshift == TCP_MAXRXTSHIFT) {
312			u_long maxidle = TCP_REXMTVAL(tp);
313			if (maxidle < tp->t_rttmin)
314				maxidle = tp->t_rttmin;
315			maxidle *= tcp_totbackoff;
316			if (tp->t_idle >= tcp_maxpersistidle ||
317			    tp->t_idle >= maxidle) {
318				tcpstat.tcps_persistdrop++;
319				tp = tcp_drop(tp, ETIMEDOUT);
320				break;
321			}
322		}
323		tcp_setpersist(tp);
324		tp->t_force = 1;
325		(void) tcp_output(tp);
326		tp->t_force = 0;
327		break;
328
329	/*
330	 * Keep-alive timer went off; send something
331	 * or drop connection if idle for too long.
332	 */
333	case TCPT_KEEP:
334		tcpstat.tcps_keeptimeo++;
335		if (tp->t_state < TCPS_ESTABLISHED)
336			goto dropit;
337		if ((always_keepalive ||
338		    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
339		    tp->t_state <= TCPS_CLOSING) {
340		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
341				goto dropit;
342			/*
343			 * Send a packet designed to force a response
344			 * if the peer is up and reachable:
345			 * either an ACK if the connection is still alive,
346			 * or an RST if the peer has closed the connection
347			 * due to timeout or reboot.
348			 * Using sequence number tp->snd_una-1
349			 * causes the transmitted zero-length segment
350			 * to lie outside the receive window;
351			 * by the protocol spec, this requires the
352			 * correspondent TCP to respond.
353			 */
354			tcpstat.tcps_keepprobe++;
355#ifdef TCP_COMPAT_42
356			/*
357			 * The keepalive packet must have nonzero length
358			 * to get a 4.2 host to respond.
359			 */
360			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
361			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
362#else
363			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
364			    tp->rcv_nxt, tp->snd_una - 1, 0);
365#endif
366			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
367		} else
368			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
369		break;
370	dropit:
371		tcpstat.tcps_keepdrops++;
372		tp = tcp_drop(tp, ETIMEDOUT);
373		break;
374	}
375	return (tp);
376}
377