xref: /freebsd/sys/netinet/tcp_timer.c (revision daf1cffce2e07931f27c6c6998652e90df6ba87e)
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  * $FreeBSD$
35  */
36 
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/protosw.h>
48 
49 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
50 
51 #include <net/route.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_systm.h>
55 #include <netinet/in_pcb.h>
56 #ifdef INET6
57 #include <netinet6/in6_pcb.h>
58 #endif
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 
70 static int
71 sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS
72 {
73 	int error, s, tt;
74 
75 	tt = *(int *)oidp->oid_arg1;
76 	s = tt * 1000 / hz;
77 
78 	error = sysctl_handle_int(oidp, &s, 0, req);
79 	if (error || !req->newptr)
80 		return (error);
81 
82 	tt = s * hz / 1000;
83 	if (tt < 1)
84 		return (EINVAL);
85 
86 	*(int *)oidp->oid_arg1 = tt;
87         return (0);
88 }
89 
90 int	tcp_keepinit;
91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
92     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
93 
94 int	tcp_keepidle;
95 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
96     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
97 
98 int	tcp_keepintvl;
99 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
100     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
101 
102 int	tcp_delacktime;
103 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
104     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
105     "Time before a delayed ACK is sent");
106 
107 int	tcp_msl;
108 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
109     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
110 
111 static int	always_keepalive = 0;
112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
113     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
114 
115 static int	tcp_keepcnt = TCPTV_KEEPCNT;
116 	/* max idle probes */
117 int	tcp_maxpersistidle;
118 	/* max idle time in persist */
119 int	tcp_maxidle;
120 
121 /*
122  * Tcp protocol timeout routine called every 500 ms.
123  * Updates timestamps used for TCP
124  * causes finite state machine actions if timers expire.
125  */
126 void
127 tcp_slowtimo()
128 {
129 	int s;
130 
131 	s = splnet();
132 
133 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
134 
135 	tcp_iss += TCP_ISSINCR/PR_SLOWHZ;		/* increment iss */
136 #ifdef TCP_COMPAT_42
137 	if ((int)tcp_iss < 0)
138 		tcp_iss = TCP_ISSINCR;			/* XXX */
139 #endif
140 	splx(s);
141 }
142 
143 /*
144  * Cancel all timers for TCP tp.
145  */
146 void
147 tcp_canceltimers(tp)
148 	struct tcpcb *tp;
149 {
150 	callout_stop(tp->tt_2msl);
151 	callout_stop(tp->tt_persist);
152 	callout_stop(tp->tt_keep);
153 	callout_stop(tp->tt_rexmt);
154 }
155 
156 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
157     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
158 
159 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
160 
161 /*
162  * TCP timer processing.
163  */
164 void
165 tcp_timer_delack(xtp)
166 	void *xtp;
167 {
168 	struct tcpcb *tp = xtp;
169 	int s;
170 
171 	s = splnet();
172 	if (callout_pending(tp->tt_delack)) {
173 		splx(s);
174 		return;
175 	}
176 	callout_deactivate(tp->tt_delack);
177 
178 	tp->t_flags |= TF_ACKNOW;
179 	tcpstat.tcps_delack++;
180 	(void) tcp_output(tp);
181 	splx(s);
182 }
183 
184 void
185 tcp_timer_2msl(xtp)
186 	void *xtp;
187 {
188 	struct tcpcb *tp = xtp;
189 	int s;
190 #ifdef TCPDEBUG
191 	int ostate;
192 
193 	ostate = tp->t_state;
194 #endif
195 	s = splnet();
196 	if (callout_pending(tp->tt_2msl)) {
197 		splx(s);
198 		return;
199 	}
200 	callout_deactivate(tp->tt_2msl);
201 	/*
202 	 * 2 MSL timeout in shutdown went off.  If we're closed but
203 	 * still waiting for peer to close and connection has been idle
204 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
205 	 * control block.  Otherwise, check again in a bit.
206 	 */
207 	if (tp->t_state != TCPS_TIME_WAIT &&
208 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
209 		callout_reset(tp->tt_2msl, tcp_keepintvl,
210 			      tcp_timer_2msl, tp);
211 	else
212 		tp = tcp_close(tp);
213 
214 #ifdef TCPDEBUG
215 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
216 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
217 			  PRU_SLOWTIMO);
218 #endif
219 	splx(s);
220 }
221 
222 void
223 tcp_timer_keep(xtp)
224 	void *xtp;
225 {
226 	struct tcpcb *tp = xtp;
227 	int s;
228 #ifdef TCPDEBUG
229 	int ostate;
230 
231 	ostate = tp->t_state;
232 #endif
233 	s = splnet();
234 	if (callout_pending(tp->tt_keep)) {
235 		splx(s);
236 		return;
237 	}
238 	callout_deactivate(tp->tt_keep);
239 	/*
240 	 * Keep-alive timer went off; send something
241 	 * or drop connection if idle for too long.
242 	 */
243 	tcpstat.tcps_keeptimeo++;
244 	if (tp->t_state < TCPS_ESTABLISHED)
245 		goto dropit;
246 	if ((always_keepalive ||
247 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
248 	    tp->t_state <= TCPS_CLOSING) {
249 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
250 			goto dropit;
251 		/*
252 		 * Send a packet designed to force a response
253 		 * if the peer is up and reachable:
254 		 * either an ACK if the connection is still alive,
255 		 * or an RST if the peer has closed the connection
256 		 * due to timeout or reboot.
257 		 * Using sequence number tp->snd_una-1
258 		 * causes the transmitted zero-length segment
259 		 * to lie outside the receive window;
260 		 * by the protocol spec, this requires the
261 		 * correspondent TCP to respond.
262 		 */
263 		tcpstat.tcps_keepprobe++;
264 #ifdef TCP_COMPAT_42
265 		/*
266 		 * The keepalive packet must have nonzero length
267 		 * to get a 4.2 host to respond.
268 		 */
269 		tcp_respond(tp, tp->t_template->tt_ipgen,
270 			    &tp->t_template->tt_t, (struct mbuf *)NULL,
271 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
272 #else
273 		tcp_respond(tp, tp->t_template->tt_ipgen,
274 			    &tp->t_template->tt_t, (struct mbuf *)NULL,
275 			    tp->rcv_nxt, tp->snd_una - 1, 0);
276 #endif
277 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
278 	} else
279 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
280 
281 #ifdef TCPDEBUG
282 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
283 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
284 			  PRU_SLOWTIMO);
285 #endif
286 	splx(s);
287 	return;
288 
289 dropit:
290 	tcpstat.tcps_keepdrops++;
291 	tp = tcp_drop(tp, ETIMEDOUT);
292 
293 #ifdef TCPDEBUG
294 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
295 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
296 			  PRU_SLOWTIMO);
297 #endif
298 	splx(s);
299 }
300 
301 void
302 tcp_timer_persist(xtp)
303 	void *xtp;
304 {
305 	struct tcpcb *tp = xtp;
306 	int s;
307 #ifdef TCPDEBUG
308 	int ostate;
309 
310 	ostate = tp->t_state;
311 #endif
312 	s = splnet();
313 	if (callout_pending(tp->tt_persist)) {
314 		splx(s);
315 		return;
316 	}
317 	callout_deactivate(tp->tt_persist);
318 	/*
319 	 * Persistance timer into zero window.
320 	 * Force a byte to be output, if possible.
321 	 */
322 	tcpstat.tcps_persisttimeo++;
323 	/*
324 	 * Hack: if the peer is dead/unreachable, we do not
325 	 * time out if the window is closed.  After a full
326 	 * backoff, drop the connection if the idle time
327 	 * (no responses to probes) reaches the maximum
328 	 * backoff that we would use if retransmitting.
329 	 */
330 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
331 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
332 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
333 		tcpstat.tcps_persistdrop++;
334 		tp = tcp_drop(tp, ETIMEDOUT);
335 		goto out;
336 	}
337 	tcp_setpersist(tp);
338 	tp->t_force = 1;
339 	(void) tcp_output(tp);
340 	tp->t_force = 0;
341 
342 out:
343 #ifdef TCPDEBUG
344 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
345 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
346 			  PRU_SLOWTIMO);
347 #endif
348 	splx(s);
349 }
350 
351 void
352 tcp_timer_rexmt(xtp)
353 	void *xtp;
354 {
355 	struct tcpcb *tp = xtp;
356 	int s;
357 	int rexmt;
358 #ifdef TCPDEBUG
359 	int ostate;
360 
361 	ostate = tp->t_state;
362 #endif
363 	s = splnet();
364 	if (callout_pending(tp->tt_rexmt)) {
365 		splx(s);
366 		return;
367 	}
368 	callout_deactivate(tp->tt_rexmt);
369 	/*
370 	 * Retransmission timer went off.  Message has not
371 	 * been acked within retransmit interval.  Back off
372 	 * to a longer retransmit interval and retransmit one segment.
373 	 */
374 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
375 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
376 		tcpstat.tcps_timeoutdrop++;
377 		tp = tcp_drop(tp, tp->t_softerror ?
378 			      tp->t_softerror : ETIMEDOUT);
379 		goto out;
380 	}
381 	if (tp->t_rxtshift == 1) {
382 		/*
383 		 * first retransmit; record ssthresh and cwnd so they can
384 	 	 * be recovered if this turns out to be a "bad" retransmit.
385 		 * A retransmit is considered "bad" if an ACK for this
386 		 * segment is received within RTT/2 interval; the assumption
387 		 * here is that the ACK was already in flight.  See
388 		 * "On Estimating End-to-End Network Path Properties" by
389 		 * Allman and Paxson for more details.
390 		 */
391 		tp->snd_cwnd_prev = tp->snd_cwnd;
392 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
393 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
394 	}
395 	tcpstat.tcps_rexmttimeo++;
396 	rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
397 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
398 		      tp->t_rttmin, TCPTV_REXMTMAX);
399 	/*
400 	 * If losing, let the lower level know and try for
401 	 * a better route.  Also, if we backed off this far,
402 	 * our srtt estimate is probably bogus.  Clobber it
403 	 * so we'll take the next rtt measurement as our srtt;
404 	 * move the current srtt into rttvar to keep the current
405 	 * retransmit times until then.
406 	 */
407 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
408 #ifdef INET6
409 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
410 			in6_losing(tp->t_inpcb);
411 		else
412 #endif
413 		in_losing(tp->t_inpcb);
414 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
415 		tp->t_srtt = 0;
416 	}
417 	tp->snd_nxt = tp->snd_una;
418 	/*
419 	 * Force a segment to be sent.
420 	 */
421 	tp->t_flags |= TF_ACKNOW;
422 	/*
423 	 * If timing a segment in this window, stop the timer.
424 	 */
425 	tp->t_rtttime = 0;
426 	/*
427 	 * Close the congestion window down to one segment
428 	 * (we'll open it by one segment for each ack we get).
429 	 * Since we probably have a window's worth of unacked
430 	 * data accumulated, this "slow start" keeps us from
431 	 * dumping all that data as back-to-back packets (which
432 	 * might overwhelm an intermediate gateway).
433 	 *
434 	 * There are two phases to the opening: Initially we
435 	 * open by one mss on each ack.  This makes the window
436 	 * size increase exponentially with time.  If the
437 	 * window is larger than the path can handle, this
438 	 * exponential growth results in dropped packet(s)
439 	 * almost immediately.  To get more time between
440 	 * drops but still "push" the network to take advantage
441 	 * of improving conditions, we switch from exponential
442 	 * to linear window opening at some threshhold size.
443 	 * For a threshhold, we use half the current window
444 	 * size, truncated to a multiple of the mss.
445 	 *
446 	 * (the minimum cwnd that will give us exponential
447 	 * growth is 2 mss.  We don't allow the threshhold
448 	 * to go below this.)
449 	 */
450 	{
451 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
452 		if (win < 2)
453 			win = 2;
454 		tp->snd_cwnd = tp->t_maxseg;
455 		tp->snd_ssthresh = win * tp->t_maxseg;
456 		tp->t_dupacks = 0;
457 	}
458 	(void) tcp_output(tp);
459 
460 out:
461 #ifdef TCPDEBUG
462 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
463 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
464 			  PRU_SLOWTIMO);
465 #endif
466 	splx(s);
467 }
468