xref: /freebsd/sys/netinet/tcp_timer.c (revision eacee0ff7ec955b32e09515246bd97b6edcd2b0f)
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/mbuf.h>
45 #include <sys/sysctl.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 
50 #include <machine/cpu.h>	/* before tcp_seq.h, for tcp_random18() */
51 
52 #include <net/route.h>
53 
54 #include <netinet/in.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/in_pcb.h>
57 #ifdef INET6
58 #include <netinet6/in6_pcb.h>
59 #endif
60 #include <netinet/ip_var.h>
61 #include <netinet/tcp.h>
62 #include <netinet/tcp_fsm.h>
63 #include <netinet/tcp_seq.h>
64 #include <netinet/tcp_timer.h>
65 #include <netinet/tcp_var.h>
66 #include <netinet/tcpip.h>
67 #ifdef TCPDEBUG
68 #include <netinet/tcp_debug.h>
69 #endif
70 
71 static int
72 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
73 {
74 	int error, s, tt;
75 
76 	tt = *(int *)oidp->oid_arg1;
77 	s = tt * 1000 / hz;
78 
79 	error = sysctl_handle_int(oidp, &s, 0, req);
80 	if (error || !req->newptr)
81 		return (error);
82 
83 	tt = s * hz / 1000;
84 	if (tt < 1)
85 		return (EINVAL);
86 
87 	*(int *)oidp->oid_arg1 = tt;
88         return (0);
89 }
90 
91 int	tcp_keepinit;
92 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
93     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
94 
95 int	tcp_keepidle;
96 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
97     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
98 
99 int	tcp_keepintvl;
100 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
101     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
102 
103 int	tcp_delacktime;
104 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
105     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
106     "Time before a delayed ACK is sent");
107 
108 int	tcp_msl;
109 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
110     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
111 
112 static int	always_keepalive = 1;
113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
114     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
115 
116 static int	tcp_keepcnt = TCPTV_KEEPCNT;
117 	/* max idle probes */
118 int	tcp_maxpersistidle;
119 	/* max idle time in persist */
120 int	tcp_maxidle;
121 
122 /*
123  * Tcp protocol timeout routine called every 500 ms.
124  * Updates timestamps used for TCP
125  * causes finite state machine actions if timers expire.
126  */
127 void
128 tcp_slowtimo()
129 {
130 	int s;
131 
132 	s = splnet();
133 
134 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
135 
136 	splx(s);
137 }
138 
139 /*
140  * Cancel all timers for TCP tp.
141  */
142 void
143 tcp_canceltimers(tp)
144 	struct tcpcb *tp;
145 {
146 	callout_stop(tp->tt_2msl);
147 	callout_stop(tp->tt_persist);
148 	callout_stop(tp->tt_keep);
149 	callout_stop(tp->tt_rexmt);
150 }
151 
152 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
153     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
154 
155 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
156     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
157 
158 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
159 
160 /*
161  * TCP timer processing.
162  */
163 void
164 tcp_timer_delack(xtp)
165 	void *xtp;
166 {
167 	struct tcpcb *tp = xtp;
168 	int s;
169 
170 	s = splnet();
171 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
172 		splx(s);
173 		return;
174 	}
175 	callout_deactivate(tp->tt_delack);
176 
177 	tp->t_flags |= TF_ACKNOW;
178 	tcpstat.tcps_delack++;
179 	(void) tcp_output(tp);
180 	splx(s);
181 }
182 
183 void
184 tcp_timer_2msl(xtp)
185 	void *xtp;
186 {
187 	struct tcpcb *tp = xtp;
188 	int s;
189 #ifdef TCPDEBUG
190 	int ostate;
191 
192 	ostate = tp->t_state;
193 #endif
194 	s = splnet();
195 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
196 		splx(s);
197 		return;
198 	}
199 	callout_deactivate(tp->tt_2msl);
200 	/*
201 	 * 2 MSL timeout in shutdown went off.  If we're closed but
202 	 * still waiting for peer to close and connection has been idle
203 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
204 	 * control block.  Otherwise, check again in a bit.
205 	 */
206 	if (tp->t_state != TCPS_TIME_WAIT &&
207 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
208 		callout_reset(tp->tt_2msl, tcp_keepintvl,
209 			      tcp_timer_2msl, tp);
210 	else
211 		tp = tcp_close(tp);
212 
213 #ifdef TCPDEBUG
214 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
215 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
216 			  PRU_SLOWTIMO);
217 #endif
218 	splx(s);
219 }
220 
221 void
222 tcp_timer_keep(xtp)
223 	void *xtp;
224 {
225 	struct tcpcb *tp = xtp;
226 	struct tcptemp *t_template;
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) || !callout_active(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 		t_template = tcp_maketemplate(tp);
265 		if (t_template) {
266 			tcp_respond(tp, t_template->tt_ipgen,
267 				    &t_template->tt_t, (struct mbuf *)NULL,
268 				    tp->rcv_nxt, tp->snd_una - 1, 0);
269 			(void) m_free(dtom(t_template));
270 		}
271 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
272 	} else
273 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
274 
275 #ifdef TCPDEBUG
276 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
277 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
278 			  PRU_SLOWTIMO);
279 #endif
280 	splx(s);
281 	return;
282 
283 dropit:
284 	tcpstat.tcps_keepdrops++;
285 	tp = tcp_drop(tp, ETIMEDOUT);
286 
287 #ifdef TCPDEBUG
288 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
289 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
290 			  PRU_SLOWTIMO);
291 #endif
292 	splx(s);
293 }
294 
295 void
296 tcp_timer_persist(xtp)
297 	void *xtp;
298 {
299 	struct tcpcb *tp = xtp;
300 	int s;
301 #ifdef TCPDEBUG
302 	int ostate;
303 
304 	ostate = tp->t_state;
305 #endif
306 	s = splnet();
307 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
308 		splx(s);
309 		return;
310 	}
311 	callout_deactivate(tp->tt_persist);
312 	/*
313 	 * Persistance timer into zero window.
314 	 * Force a byte to be output, if possible.
315 	 */
316 	tcpstat.tcps_persisttimeo++;
317 	/*
318 	 * Hack: if the peer is dead/unreachable, we do not
319 	 * time out if the window is closed.  After a full
320 	 * backoff, drop the connection if the idle time
321 	 * (no responses to probes) reaches the maximum
322 	 * backoff that we would use if retransmitting.
323 	 */
324 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
325 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
326 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
327 		tcpstat.tcps_persistdrop++;
328 		tp = tcp_drop(tp, ETIMEDOUT);
329 		goto out;
330 	}
331 	tcp_setpersist(tp);
332 	tp->t_force = 1;
333 	(void) tcp_output(tp);
334 	tp->t_force = 0;
335 
336 out:
337 #ifdef TCPDEBUG
338 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
339 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
340 			  PRU_SLOWTIMO);
341 #endif
342 	splx(s);
343 }
344 
345 void
346 tcp_timer_rexmt(xtp)
347 	void *xtp;
348 {
349 	struct tcpcb *tp = xtp;
350 	int s;
351 	int rexmt;
352 #ifdef TCPDEBUG
353 	int ostate;
354 
355 	ostate = tp->t_state;
356 #endif
357 	s = splnet();
358 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
359 		splx(s);
360 		return;
361 	}
362 	callout_deactivate(tp->tt_rexmt);
363 	/*
364 	 * Retransmission timer went off.  Message has not
365 	 * been acked within retransmit interval.  Back off
366 	 * to a longer retransmit interval and retransmit one segment.
367 	 */
368 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
369 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
370 		tcpstat.tcps_timeoutdrop++;
371 		tp = tcp_drop(tp, tp->t_softerror ?
372 			      tp->t_softerror : ETIMEDOUT);
373 		goto out;
374 	}
375 	if (tp->t_rxtshift == 1) {
376 		/*
377 		 * first retransmit; record ssthresh and cwnd so they can
378 	 	 * be recovered if this turns out to be a "bad" retransmit.
379 		 * A retransmit is considered "bad" if an ACK for this
380 		 * segment is received within RTT/2 interval; the assumption
381 		 * here is that the ACK was already in flight.  See
382 		 * "On Estimating End-to-End Network Path Properties" by
383 		 * Allman and Paxson for more details.
384 		 */
385 		tp->snd_cwnd_prev = tp->snd_cwnd;
386 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
387 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
388 	}
389 	tcpstat.tcps_rexmttimeo++;
390 	if (tp->t_state == TCPS_SYN_SENT)
391 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
392 	else
393 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
394 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
395 		      tp->t_rttmin, TCPTV_REXMTMAX);
396 	/*
397 	 * Disable rfc1323 and rfc1644 if we havn't got any response to
398 	 * our third SYN to work-around some broken terminal servers
399 	 * (most of which have hopefully been retired) that have bad VJ
400 	 * header compression code which trashes TCP segments containing
401 	 * unknown-to-them TCP options.
402 	 */
403 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
404 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
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