tcp_timer.c revision 21673 
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 *	$FreeBSD: head/sys/netinet/tcp_timer.c 21673 1997-01-14 07:20:47Z jkh $
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#ifdef TCPDEBUG
67#include <netinet/tcp_debug.h>
68#endif
69
70int	tcp_keepinit = TCPTV_KEEP_INIT;
71SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
72	CTLFLAG_RW, &tcp_keepinit , 0, "");
73
74int	tcp_keepidle = TCPTV_KEEP_IDLE;
75SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
76	CTLFLAG_RW, &tcp_keepidle , 0, "");
77
78static int	tcp_keepintvl = TCPTV_KEEPINTVL;
79SYSCTL_INT(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
80	CTLFLAG_RW, &tcp_keepintvl , 0, "");
81
82static int	always_keepalive = 0;
83SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive,
84	CTLFLAG_RW, &always_keepalive , 0, "");
85
86static int	tcp_keepcnt = TCPTV_KEEPCNT;
87	/* max idle probes */
88static int	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
89	/* max idle time in persist */
90int	tcp_maxidle;
91#else /* TUBA_INCLUDE */
92
93static	int tcp_maxpersistidle;
94#endif /* TUBA_INCLUDE */
95
96/*
97 * Fast timeout routine for processing delayed acks
98 */
99void
100tcp_fasttimo()
101{
102	register struct inpcb *inp;
103	register struct tcpcb *tp;
104	int s;
105
106	s = splnet();
107
108	for (inp = tcb.lh_first; inp != NULL; inp = inp->inp_list.le_next) {
109		if ((tp = (struct tcpcb *)inp->inp_ppcb) &&
110		    (tp->t_flags & TF_DELACK)) {
111			tp->t_flags &= ~TF_DELACK;
112			tp->t_flags |= TF_ACKNOW;
113			tcpstat.tcps_delack++;
114			(void) tcp_output(tp);
115		}
116	}
117	splx(s);
118}
119
120/*
121 * Tcp protocol timeout routine called every 500 ms.
122 * Updates the timers in all active tcb's and
123 * causes finite state machine actions if timers expire.
124 */
125void
126tcp_slowtimo()
127{
128	register struct inpcb *ip, *ipnxt;
129	register struct tcpcb *tp;
130	register int i;
131	int s;
132#ifdef TCPDEBUG
133	int ostate;
134#endif
135
136	s = splnet();
137
138	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
139
140	ip = tcb.lh_first;
141	if (ip == NULL) {
142		splx(s);
143		return;
144	}
145	/*
146	 * Search through tcb's and update active timers.
147	 */
148	for (; ip != NULL; ip = ipnxt) {
149		ipnxt = ip->inp_list.le_next;
150		tp = intotcpcb(ip);
151		if (tp == 0 || tp->t_state == TCPS_LISTEN)
152			continue;
153		for (i = 0; i < TCPT_NTIMERS; i++) {
154			if (tp->t_timer[i] && --tp->t_timer[i] == 0) {
155#ifdef TCPDEBUG
156				ostate = tp->t_state;
157#endif
158				tp = tcp_timers(tp, i);
159				if (tp == NULL)
160					goto tpgone;
161#ifdef TCPDEBUG
162				if (tp->t_inpcb->inp_socket->so_options
163				    & SO_DEBUG)
164					tcp_trace(TA_USER, ostate, tp,
165						  (struct tcpiphdr *)0,
166						  PRU_SLOWTIMO);
167#endif
168			}
169		}
170		tp->t_idle++;
171		tp->t_duration++;
172		if (tp->t_rtt)
173			tp->t_rtt++;
174tpgone:
175		;
176	}
177	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
178#ifdef TCP_COMPAT_42
179	if ((int)tcp_iss < 0)
180		tcp_iss = TCP_ISSINCR;			/* XXX */
181#endif
182	tcp_now++;					/* for timestamps */
183	splx(s);
184}
185#ifndef TUBA_INCLUDE
186
187/*
188 * Cancel all timers for TCP tp.
189 */
190void
191tcp_canceltimers(tp)
192	struct tcpcb *tp;
193{
194	register int i;
195
196	for (i = 0; i < TCPT_NTIMERS; i++)
197		tp->t_timer[i] = 0;
198}
199
200int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
201    { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
202
203static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
204
205/*
206 * TCP timer processing.
207 */
208struct tcpcb *
209tcp_timers(tp, timer)
210	register struct tcpcb *tp;
211	int timer;
212{
213	register int rexmt;
214
215	switch (timer) {
216
217	/*
218	 * 2 MSL timeout in shutdown went off.  If we're closed but
219	 * still waiting for peer to close and connection has been idle
220	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
221	 * control block.  Otherwise, check again in a bit.
222	 */
223	case TCPT_2MSL:
224		if (tp->t_state != TCPS_TIME_WAIT &&
225		    tp->t_idle <= tcp_maxidle)
226			tp->t_timer[TCPT_2MSL] = tcp_keepintvl;
227		else
228			tp = tcp_close(tp);
229		break;
230
231	/*
232	 * Retransmission timer went off.  Message has not
233	 * been acked within retransmit interval.  Back off
234	 * to a longer retransmit interval and retransmit one segment.
235	 */
236	case TCPT_REXMT:
237		if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
238			tp->t_rxtshift = TCP_MAXRXTSHIFT;
239			tcpstat.tcps_timeoutdrop++;
240			tp = tcp_drop(tp, tp->t_softerror ?
241			    tp->t_softerror : ETIMEDOUT);
242			break;
243		}
244		tcpstat.tcps_rexmttimeo++;
245		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
246		TCPT_RANGESET(tp->t_rxtcur, rexmt,
247		    tp->t_rttmin, TCPTV_REXMTMAX);
248		tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
249		/*
250		 * If losing, let the lower level know and try for
251		 * a better route.  Also, if we backed off this far,
252		 * our srtt estimate is probably bogus.  Clobber it
253		 * so we'll take the next rtt measurement as our srtt;
254		 * move the current srtt into rttvar to keep the current
255		 * retransmit times until then.
256		 */
257		if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
258			in_losing(tp->t_inpcb);
259			tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
260			tp->t_srtt = 0;
261		}
262		tp->snd_nxt = tp->snd_una;
263		/*
264		 * Force a segment to be sent.
265		 */
266		tp->t_flags |= TF_ACKNOW;
267		/*
268		 * If timing a segment in this window, stop the timer.
269		 */
270		tp->t_rtt = 0;
271		/*
272		 * Close the congestion window down to one segment
273		 * (we'll open it by one segment for each ack we get).
274		 * Since we probably have a window's worth of unacked
275		 * data accumulated, this "slow start" keeps us from
276		 * dumping all that data as back-to-back packets (which
277		 * might overwhelm an intermediate gateway).
278		 *
279		 * There are two phases to the opening: Initially we
280		 * open by one mss on each ack.  This makes the window
281		 * size increase exponentially with time.  If the
282		 * window is larger than the path can handle, this
283		 * exponential growth results in dropped packet(s)
284		 * almost immediately.  To get more time between
285		 * drops but still "push" the network to take advantage
286		 * of improving conditions, we switch from exponential
287		 * to linear window opening at some threshhold size.
288		 * For a threshhold, we use half the current window
289		 * size, truncated to a multiple of the mss.
290		 *
291		 * (the minimum cwnd that will give us exponential
292		 * growth is 2 mss.  We don't allow the threshhold
293		 * to go below this.)
294		 */
295		{
296		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
297		if (win < 2)
298			win = 2;
299		tp->snd_cwnd = tp->t_maxseg;
300		tp->snd_ssthresh = win * tp->t_maxseg;
301		tp->t_dupacks = 0;
302		}
303		(void) tcp_output(tp);
304		break;
305
306	/*
307	 * Persistance timer into zero window.
308	 * Force a byte to be output, if possible.
309	 */
310	case TCPT_PERSIST:
311		tcpstat.tcps_persisttimeo++;
312		/*
313		 * Hack: if the peer is dead/unreachable, we do not
314		 * time out if the window is closed.  After a full
315		 * backoff, drop the connection if the idle time
316		 * (no responses to probes) reaches the maximum
317		 * backoff that we would use if retransmitting.
318		 */
319		if (tp->t_rxtshift == TCP_MAXRXTSHIFT) {
320			u_long maxidle = TCP_REXMTVAL(tp);
321			if (maxidle < tp->t_rttmin)
322				maxidle = tp->t_rttmin;
323			maxidle *= tcp_totbackoff;
324			if (tp->t_idle >= tcp_maxpersistidle ||
325			    tp->t_idle >= maxidle) {
326				tcpstat.tcps_persistdrop++;
327				tp = tcp_drop(tp, ETIMEDOUT);
328				break;
329			}
330		}
331		tcp_setpersist(tp);
332		tp->t_force = 1;
333		(void) tcp_output(tp);
334		tp->t_force = 0;
335		break;
336
337	/*
338	 * Keep-alive timer went off; send something
339	 * or drop connection if idle for too long.
340	 */
341	case TCPT_KEEP:
342		tcpstat.tcps_keeptimeo++;
343		if (tp->t_state < TCPS_ESTABLISHED)
344			goto dropit;
345		if ((always_keepalive ||
346		    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
347		    tp->t_state <= TCPS_CLOSING) {
348		    	if (tp->t_idle >= tcp_keepidle + tcp_maxidle)
349				goto dropit;
350			/*
351			 * Send a packet designed to force a response
352			 * if the peer is up and reachable:
353			 * either an ACK if the connection is still alive,
354			 * or an RST if the peer has closed the connection
355			 * due to timeout or reboot.
356			 * Using sequence number tp->snd_una-1
357			 * causes the transmitted zero-length segment
358			 * to lie outside the receive window;
359			 * by the protocol spec, this requires the
360			 * correspondent TCP to respond.
361			 */
362			tcpstat.tcps_keepprobe++;
363#ifdef TCP_COMPAT_42
364			/*
365			 * The keepalive packet must have nonzero length
366			 * to get a 4.2 host to respond.
367			 */
368			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
369			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
370#else
371			tcp_respond(tp, tp->t_template, (struct mbuf *)NULL,
372			    tp->rcv_nxt, tp->snd_una - 1, 0);
373#endif
374			tp->t_timer[TCPT_KEEP] = tcp_keepintvl;
375		} else
376			tp->t_timer[TCPT_KEEP] = tcp_keepidle;
377		break;
378	dropit:
379		tcpstat.tcps_keepdrops++;
380		tp = tcp_drop(tp, ETIMEDOUT);
381		break;
382	}
383	return (tp);
384}
385#endif /* TUBA_INCLUDE */
386