xref: /freebsd/sys/netinet/tcp_timer.c (revision 1a2cdef4962b47be5057809ce730a733b7f3c27c)
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_syn_backoff[TCP_MAXRXTSHIFT + 1] =
157     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
158 
159 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
160     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
161 
162 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
163 
164 /*
165  * TCP timer processing.
166  */
167 void
168 tcp_timer_delack(xtp)
169 	void *xtp;
170 {
171 	struct tcpcb *tp = xtp;
172 	int s;
173 
174 	s = splnet();
175 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
176 		splx(s);
177 		return;
178 	}
179 	callout_deactivate(tp->tt_delack);
180 
181 	tp->t_flags |= TF_ACKNOW;
182 	tcpstat.tcps_delack++;
183 	(void) tcp_output(tp);
184 	splx(s);
185 }
186 
187 void
188 tcp_timer_2msl(xtp)
189 	void *xtp;
190 {
191 	struct tcpcb *tp = xtp;
192 	int s;
193 #ifdef TCPDEBUG
194 	int ostate;
195 
196 	ostate = tp->t_state;
197 #endif
198 	s = splnet();
199 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
200 		splx(s);
201 		return;
202 	}
203 	callout_deactivate(tp->tt_2msl);
204 	/*
205 	 * 2 MSL timeout in shutdown went off.  If we're closed but
206 	 * still waiting for peer to close and connection has been idle
207 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
208 	 * control block.  Otherwise, check again in a bit.
209 	 */
210 	if (tp->t_state != TCPS_TIME_WAIT &&
211 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
212 		callout_reset(tp->tt_2msl, tcp_keepintvl,
213 			      tcp_timer_2msl, tp);
214 	else
215 		tp = tcp_close(tp);
216 
217 #ifdef TCPDEBUG
218 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
219 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
220 			  PRU_SLOWTIMO);
221 #endif
222 	splx(s);
223 }
224 
225 void
226 tcp_timer_keep(xtp)
227 	void *xtp;
228 {
229 	struct tcpcb *tp = xtp;
230 	int s;
231 #ifdef TCPDEBUG
232 	int ostate;
233 
234 	ostate = tp->t_state;
235 #endif
236 	s = splnet();
237 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
238 		splx(s);
239 		return;
240 	}
241 	callout_deactivate(tp->tt_keep);
242 	/*
243 	 * Keep-alive timer went off; send something
244 	 * or drop connection if idle for too long.
245 	 */
246 	tcpstat.tcps_keeptimeo++;
247 	if (tp->t_state < TCPS_ESTABLISHED)
248 		goto dropit;
249 	if ((always_keepalive ||
250 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
251 	    tp->t_state <= TCPS_CLOSING) {
252 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
253 			goto dropit;
254 		/*
255 		 * Send a packet designed to force a response
256 		 * if the peer is up and reachable:
257 		 * either an ACK if the connection is still alive,
258 		 * or an RST if the peer has closed the connection
259 		 * due to timeout or reboot.
260 		 * Using sequence number tp->snd_una-1
261 		 * causes the transmitted zero-length segment
262 		 * to lie outside the receive window;
263 		 * by the protocol spec, this requires the
264 		 * correspondent TCP to respond.
265 		 */
266 		tcpstat.tcps_keepprobe++;
267 #ifdef TCP_COMPAT_42
268 		/*
269 		 * The keepalive packet must have nonzero length
270 		 * to get a 4.2 host to respond.
271 		 */
272 		tcp_respond(tp, tp->t_template->tt_ipgen,
273 			    &tp->t_template->tt_t, (struct mbuf *)NULL,
274 			    tp->rcv_nxt - 1, tp->snd_una - 1, 0);
275 #else
276 		tcp_respond(tp, tp->t_template->tt_ipgen,
277 			    &tp->t_template->tt_t, (struct mbuf *)NULL,
278 			    tp->rcv_nxt, tp->snd_una - 1, 0);
279 #endif
280 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
281 	} else
282 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
283 
284 #ifdef TCPDEBUG
285 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
286 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
287 			  PRU_SLOWTIMO);
288 #endif
289 	splx(s);
290 	return;
291 
292 dropit:
293 	tcpstat.tcps_keepdrops++;
294 	tp = tcp_drop(tp, ETIMEDOUT);
295 
296 #ifdef TCPDEBUG
297 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
298 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
299 			  PRU_SLOWTIMO);
300 #endif
301 	splx(s);
302 }
303 
304 void
305 tcp_timer_persist(xtp)
306 	void *xtp;
307 {
308 	struct tcpcb *tp = xtp;
309 	int s;
310 #ifdef TCPDEBUG
311 	int ostate;
312 
313 	ostate = tp->t_state;
314 #endif
315 	s = splnet();
316 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
317 		splx(s);
318 		return;
319 	}
320 	callout_deactivate(tp->tt_persist);
321 	/*
322 	 * Persistance timer into zero window.
323 	 * Force a byte to be output, if possible.
324 	 */
325 	tcpstat.tcps_persisttimeo++;
326 	/*
327 	 * Hack: if the peer is dead/unreachable, we do not
328 	 * time out if the window is closed.  After a full
329 	 * backoff, drop the connection if the idle time
330 	 * (no responses to probes) reaches the maximum
331 	 * backoff that we would use if retransmitting.
332 	 */
333 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
334 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
335 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
336 		tcpstat.tcps_persistdrop++;
337 		tp = tcp_drop(tp, ETIMEDOUT);
338 		goto out;
339 	}
340 	tcp_setpersist(tp);
341 	tp->t_force = 1;
342 	(void) tcp_output(tp);
343 	tp->t_force = 0;
344 
345 out:
346 #ifdef TCPDEBUG
347 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
348 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
349 			  PRU_SLOWTIMO);
350 #endif
351 	splx(s);
352 }
353 
354 void
355 tcp_timer_rexmt(xtp)
356 	void *xtp;
357 {
358 	struct tcpcb *tp = xtp;
359 	int s;
360 	int rexmt;
361 #ifdef TCPDEBUG
362 	int ostate;
363 
364 	ostate = tp->t_state;
365 #endif
366 	s = splnet();
367 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
368 		splx(s);
369 		return;
370 	}
371 	callout_deactivate(tp->tt_rexmt);
372 	/*
373 	 * Retransmission timer went off.  Message has not
374 	 * been acked within retransmit interval.  Back off
375 	 * to a longer retransmit interval and retransmit one segment.
376 	 */
377 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
378 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
379 		tcpstat.tcps_timeoutdrop++;
380 		tp = tcp_drop(tp, tp->t_softerror ?
381 			      tp->t_softerror : ETIMEDOUT);
382 		goto out;
383 	}
384 	if (tp->t_rxtshift == 1) {
385 		/*
386 		 * first retransmit; record ssthresh and cwnd so they can
387 	 	 * be recovered if this turns out to be a "bad" retransmit.
388 		 * A retransmit is considered "bad" if an ACK for this
389 		 * segment is received within RTT/2 interval; the assumption
390 		 * here is that the ACK was already in flight.  See
391 		 * "On Estimating End-to-End Network Path Properties" by
392 		 * Allman and Paxson for more details.
393 		 */
394 		tp->snd_cwnd_prev = tp->snd_cwnd;
395 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
396 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
397 	}
398 	tcpstat.tcps_rexmttimeo++;
399 	if (tp->t_state == TCPS_SYN_SENT)
400 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
401 	else
402 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
403 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
404 		      tp->t_rttmin, TCPTV_REXMTMAX);
405 	/*
406 	 * If losing, let the lower level know and try for
407 	 * a better route.  Also, if we backed off this far,
408 	 * our srtt estimate is probably bogus.  Clobber it
409 	 * so we'll take the next rtt measurement as our srtt;
410 	 * move the current srtt into rttvar to keep the current
411 	 * retransmit times until then.
412 	 */
413 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
414 #ifdef INET6
415 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
416 			in6_losing(tp->t_inpcb);
417 		else
418 #endif
419 		in_losing(tp->t_inpcb);
420 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
421 		tp->t_srtt = 0;
422 	}
423 	tp->snd_nxt = tp->snd_una;
424 	/*
425 	 * Note:  We overload snd_recover to function also as the
426 	 * snd_last variable described in RFC 2582
427 	 */
428 	tp->snd_recover = tp->snd_max;
429 	/*
430 	 * Force a segment to be sent.
431 	 */
432 	tp->t_flags |= TF_ACKNOW;
433 	/*
434 	 * If timing a segment in this window, stop the timer.
435 	 */
436 	tp->t_rtttime = 0;
437 	/*
438 	 * Close the congestion window down to one segment
439 	 * (we'll open it by one segment for each ack we get).
440 	 * Since we probably have a window's worth of unacked
441 	 * data accumulated, this "slow start" keeps us from
442 	 * dumping all that data as back-to-back packets (which
443 	 * might overwhelm an intermediate gateway).
444 	 *
445 	 * There are two phases to the opening: Initially we
446 	 * open by one mss on each ack.  This makes the window
447 	 * size increase exponentially with time.  If the
448 	 * window is larger than the path can handle, this
449 	 * exponential growth results in dropped packet(s)
450 	 * almost immediately.  To get more time between
451 	 * drops but still "push" the network to take advantage
452 	 * of improving conditions, we switch from exponential
453 	 * to linear window opening at some threshhold size.
454 	 * For a threshhold, we use half the current window
455 	 * size, truncated to a multiple of the mss.
456 	 *
457 	 * (the minimum cwnd that will give us exponential
458 	 * growth is 2 mss.  We don't allow the threshhold
459 	 * to go below this.)
460 	 */
461 	{
462 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
463 		if (win < 2)
464 			win = 2;
465 		tp->snd_cwnd = tp->t_maxseg;
466 		tp->snd_ssthresh = win * tp->t_maxseg;
467 		tp->t_dupacks = 0;
468 	}
469 	(void) tcp_output(tp);
470 
471 out:
472 #ifdef TCPDEBUG
473 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
474 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
475 			  PRU_SLOWTIMO);
476 #endif
477 	splx(s);
478 }
479