xref: /freebsd/sys/netinet/tcp_timer.c (revision 99e8005137088aafb1350e23b113d69b01b0820f)
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 	splx(s);
136 }
137 
138 /*
139  * Cancel all timers for TCP tp.
140  */
141 void
142 tcp_canceltimers(tp)
143 	struct tcpcb *tp;
144 {
145 	callout_stop(tp->tt_2msl);
146 	callout_stop(tp->tt_persist);
147 	callout_stop(tp->tt_keep);
148 	callout_stop(tp->tt_rexmt);
149 }
150 
151 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
152     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
153 
154 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
155     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
156 
157 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
158 
159 /*
160  * TCP timer processing.
161  */
162 void
163 tcp_timer_delack(xtp)
164 	void *xtp;
165 {
166 	struct tcpcb *tp = xtp;
167 	int s;
168 
169 	s = splnet();
170 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
171 		splx(s);
172 		return;
173 	}
174 	callout_deactivate(tp->tt_delack);
175 
176 	tp->t_flags |= TF_ACKNOW;
177 	tcpstat.tcps_delack++;
178 	(void) tcp_output(tp);
179 	splx(s);
180 }
181 
182 void
183 tcp_timer_2msl(xtp)
184 	void *xtp;
185 {
186 	struct tcpcb *tp = xtp;
187 	int s;
188 #ifdef TCPDEBUG
189 	int ostate;
190 
191 	ostate = tp->t_state;
192 #endif
193 	s = splnet();
194 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
195 		splx(s);
196 		return;
197 	}
198 	callout_deactivate(tp->tt_2msl);
199 	/*
200 	 * 2 MSL timeout in shutdown went off.  If we're closed but
201 	 * still waiting for peer to close and connection has been idle
202 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
203 	 * control block.  Otherwise, check again in a bit.
204 	 */
205 	if (tp->t_state != TCPS_TIME_WAIT &&
206 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
207 		callout_reset(tp->tt_2msl, tcp_keepintvl,
208 			      tcp_timer_2msl, tp);
209 	else
210 		tp = tcp_close(tp);
211 
212 #ifdef TCPDEBUG
213 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
214 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
215 			  PRU_SLOWTIMO);
216 #endif
217 	splx(s);
218 }
219 
220 void
221 tcp_timer_keep(xtp)
222 	void *xtp;
223 {
224 	struct tcpcb *tp = xtp;
225 	int s;
226 #ifdef TCPDEBUG
227 	int ostate;
228 
229 	ostate = tp->t_state;
230 #endif
231 	s = splnet();
232 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
233 		splx(s);
234 		return;
235 	}
236 	callout_deactivate(tp->tt_keep);
237 	/*
238 	 * Keep-alive timer went off; send something
239 	 * or drop connection if idle for too long.
240 	 */
241 	tcpstat.tcps_keeptimeo++;
242 	if (tp->t_state < TCPS_ESTABLISHED)
243 		goto dropit;
244 	if ((always_keepalive ||
245 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
246 	    tp->t_state <= TCPS_CLOSING) {
247 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
248 			goto dropit;
249 		/*
250 		 * Send a packet designed to force a response
251 		 * if the peer is up and reachable:
252 		 * either an ACK if the connection is still alive,
253 		 * or an RST if the peer has closed the connection
254 		 * due to timeout or reboot.
255 		 * Using sequence number tp->snd_una-1
256 		 * causes the transmitted zero-length segment
257 		 * to lie outside the receive window;
258 		 * by the protocol spec, this requires the
259 		 * correspondent TCP to respond.
260 		 */
261 		tcpstat.tcps_keepprobe++;
262 		tcp_respond(tp, tp->t_template->tt_ipgen,
263 			    &tp->t_template->tt_t, (struct mbuf *)NULL,
264 			    tp->rcv_nxt, tp->snd_una - 1, 0);
265 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
266 	} else
267 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
268 
269 #ifdef TCPDEBUG
270 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
271 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
272 			  PRU_SLOWTIMO);
273 #endif
274 	splx(s);
275 	return;
276 
277 dropit:
278 	tcpstat.tcps_keepdrops++;
279 	tp = tcp_drop(tp, ETIMEDOUT);
280 
281 #ifdef TCPDEBUG
282 	if (tp && (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 }
288 
289 void
290 tcp_timer_persist(xtp)
291 	void *xtp;
292 {
293 	struct tcpcb *tp = xtp;
294 	int s;
295 #ifdef TCPDEBUG
296 	int ostate;
297 
298 	ostate = tp->t_state;
299 #endif
300 	s = splnet();
301 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
302 		splx(s);
303 		return;
304 	}
305 	callout_deactivate(tp->tt_persist);
306 	/*
307 	 * Persistance timer into zero window.
308 	 * Force a byte to be output, if possible.
309 	 */
310 	tcpstat.tcps_persisttimeo++;
311 	/*
312 	 * Hack: if the peer is dead/unreachable, we do not
313 	 * time out if the window is closed.  After a full
314 	 * backoff, drop the connection if the idle time
315 	 * (no responses to probes) reaches the maximum
316 	 * backoff that we would use if retransmitting.
317 	 */
318 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
319 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
320 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
321 		tcpstat.tcps_persistdrop++;
322 		tp = tcp_drop(tp, ETIMEDOUT);
323 		goto out;
324 	}
325 	tcp_setpersist(tp);
326 	tp->t_force = 1;
327 	(void) tcp_output(tp);
328 	tp->t_force = 0;
329 
330 out:
331 #ifdef TCPDEBUG
332 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
333 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
334 			  PRU_SLOWTIMO);
335 #endif
336 	splx(s);
337 }
338 
339 void
340 tcp_timer_rexmt(xtp)
341 	void *xtp;
342 {
343 	struct tcpcb *tp = xtp;
344 	int s;
345 	int rexmt;
346 #ifdef TCPDEBUG
347 	int ostate;
348 
349 	ostate = tp->t_state;
350 #endif
351 	s = splnet();
352 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
353 		splx(s);
354 		return;
355 	}
356 	callout_deactivate(tp->tt_rexmt);
357 	/*
358 	 * Retransmission timer went off.  Message has not
359 	 * been acked within retransmit interval.  Back off
360 	 * to a longer retransmit interval and retransmit one segment.
361 	 */
362 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
363 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
364 		tcpstat.tcps_timeoutdrop++;
365 		tp = tcp_drop(tp, tp->t_softerror ?
366 			      tp->t_softerror : ETIMEDOUT);
367 		goto out;
368 	}
369 	if (tp->t_rxtshift == 1) {
370 		/*
371 		 * first retransmit; record ssthresh and cwnd so they can
372 	 	 * be recovered if this turns out to be a "bad" retransmit.
373 		 * A retransmit is considered "bad" if an ACK for this
374 		 * segment is received within RTT/2 interval; the assumption
375 		 * here is that the ACK was already in flight.  See
376 		 * "On Estimating End-to-End Network Path Properties" by
377 		 * Allman and Paxson for more details.
378 		 */
379 		tp->snd_cwnd_prev = tp->snd_cwnd;
380 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
381 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
382 	}
383 	tcpstat.tcps_rexmttimeo++;
384 	if (tp->t_state == TCPS_SYN_SENT)
385 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
386 	else
387 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
388 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
389 		      tp->t_rttmin, TCPTV_REXMTMAX);
390 	/*
391 	 * If losing, let the lower level know and try for
392 	 * a better route.  Also, if we backed off this far,
393 	 * our srtt estimate is probably bogus.  Clobber it
394 	 * so we'll take the next rtt measurement as our srtt;
395 	 * move the current srtt into rttvar to keep the current
396 	 * retransmit times until then.
397 	 */
398 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
399 #ifdef INET6
400 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
401 			in6_losing(tp->t_inpcb);
402 		else
403 #endif
404 		in_losing(tp->t_inpcb);
405 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
406 		tp->t_srtt = 0;
407 	}
408 	tp->snd_nxt = tp->snd_una;
409 	/*
410 	 * Note:  We overload snd_recover to function also as the
411 	 * snd_last variable described in RFC 2582
412 	 */
413 	tp->snd_recover = tp->snd_max;
414 	/*
415 	 * Force a segment to be sent.
416 	 */
417 	tp->t_flags |= TF_ACKNOW;
418 	/*
419 	 * If timing a segment in this window, stop the timer.
420 	 */
421 	tp->t_rtttime = 0;
422 	/*
423 	 * Close the congestion window down to one segment
424 	 * (we'll open it by one segment for each ack we get).
425 	 * Since we probably have a window's worth of unacked
426 	 * data accumulated, this "slow start" keeps us from
427 	 * dumping all that data as back-to-back packets (which
428 	 * might overwhelm an intermediate gateway).
429 	 *
430 	 * There are two phases to the opening: Initially we
431 	 * open by one mss on each ack.  This makes the window
432 	 * size increase exponentially with time.  If the
433 	 * window is larger than the path can handle, this
434 	 * exponential growth results in dropped packet(s)
435 	 * almost immediately.  To get more time between
436 	 * drops but still "push" the network to take advantage
437 	 * of improving conditions, we switch from exponential
438 	 * to linear window opening at some threshhold size.
439 	 * For a threshhold, we use half the current window
440 	 * size, truncated to a multiple of the mss.
441 	 *
442 	 * (the minimum cwnd that will give us exponential
443 	 * growth is 2 mss.  We don't allow the threshhold
444 	 * to go below this.)
445 	 */
446 	{
447 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
448 		if (win < 2)
449 			win = 2;
450 		tp->snd_cwnd = tp->t_maxseg;
451 		tp->snd_ssthresh = win * tp->t_maxseg;
452 		tp->t_dupacks = 0;
453 	}
454 	(void) tcp_output(tp);
455 
456 out:
457 #ifdef TCPDEBUG
458 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
459 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
460 			  PRU_SLOWTIMO);
461 #endif
462 	splx(s);
463 }
464