xref: /freebsd/sys/netinet/tcp_timer.c (revision f9218d3d4fd34f082473b3a021c6d4d109fb47cf)
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_inet6.h"
38 #include "opt_tcpdebug.h"
39 
40 #include <sys/param.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
50 
51 #include <net/route.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/in_systm.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_timer.h>
63 #include <netinet/tcp_var.h>
64 #include <netinet/tcpip.h>
65 #ifdef TCPDEBUG
66 #include <netinet/tcp_debug.h>
67 #endif
68 
69 static int
70 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
71 {
72 	int error, s, tt;
73 
74 	tt = *(int *)oidp->oid_arg1;
75 	s = (int)((int64_t)tt * 1000 / hz);
76 
77 	error = sysctl_handle_int(oidp, &s, 0, req);
78 	if (error || !req->newptr)
79 		return (error);
80 
81 	tt = (int)((int64_t)s * hz / 1000);
82 	if (tt < 1)
83 		return (EINVAL);
84 
85 	*(int *)oidp->oid_arg1 = tt;
86         return (0);
87 }
88 
89 int	tcp_keepinit;
90 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
91     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
92 
93 int	tcp_keepidle;
94 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
95     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
96 
97 int	tcp_keepintvl;
98 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
99     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
100 
101 int	tcp_delacktime;
102 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
103     CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
104     "Time before a delayed ACK is sent");
105 
106 int	tcp_msl;
107 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
108     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
109 
110 int	tcp_rexmit_min;
111 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
112     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
113 
114 int	tcp_rexmit_slop;
115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
116     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
117 
118 static int	always_keepalive = 1;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
120     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
121 
122 static int	tcp_keepcnt = TCPTV_KEEPCNT;
123 	/* max idle probes */
124 int	tcp_maxpersistidle;
125 	/* max idle time in persist */
126 int	tcp_maxidle;
127 
128 /*
129  * Tcp protocol timeout routine called every 500 ms.
130  * Updates timestamps used for TCP
131  * causes finite state machine actions if timers expire.
132  */
133 void
134 tcp_slowtimo()
135 {
136 	int s;
137 
138 	s = splnet();
139 
140 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
141 
142 	splx(s);
143 }
144 
145 /*
146  * Cancel all timers for TCP tp.
147  */
148 void
149 tcp_canceltimers(tp)
150 	struct tcpcb *tp;
151 {
152 	callout_stop(tp->tt_2msl);
153 	callout_stop(tp->tt_persist);
154 	callout_stop(tp->tt_keep);
155 	callout_stop(tp->tt_rexmt);
156 }
157 
158 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
159     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
160 
161 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
162     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
163 
164 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
165 
166 /*
167  * TCP timer processing.
168  */
169 
170 void
171 tcp_timer_delack(xtp)
172 	void *xtp;
173 {
174 	struct tcpcb *tp = xtp;
175 	int s;
176 	struct inpcb *inp;
177 
178 	s = splnet();
179 	INP_INFO_RLOCK(&tcbinfo);
180 	inp = tp->t_inpcb;
181 	if (!inp) {
182 		INP_INFO_RUNLOCK(&tcbinfo);
183 		splx(s);
184 		return;
185 	}
186 	INP_LOCK(inp);
187 	INP_INFO_RUNLOCK(&tcbinfo);
188 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
189 		INP_UNLOCK(inp);
190 		splx(s);
191 		return;
192 	}
193 	callout_deactivate(tp->tt_delack);
194 
195 	tp->t_flags |= TF_ACKNOW;
196 	tcpstat.tcps_delack++;
197 	(void) tcp_output(tp);
198 	INP_UNLOCK(inp);
199 	splx(s);
200 }
201 
202 void
203 tcp_timer_2msl(xtp)
204 	void *xtp;
205 {
206 	struct tcpcb *tp = xtp;
207 	int s;
208 	struct inpcb *inp;
209 #ifdef TCPDEBUG
210 	int ostate;
211 
212 	ostate = tp->t_state;
213 #endif
214 	s = splnet();
215 	INP_INFO_WLOCK(&tcbinfo);
216 	inp = tp->t_inpcb;
217 	if (!inp) {
218 		INP_INFO_WUNLOCK(&tcbinfo);
219 		splx(s);
220 		return;
221 	}
222 	INP_LOCK(inp);
223 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
224 		INP_UNLOCK(tp->t_inpcb);
225 		INP_INFO_WUNLOCK(&tcbinfo);
226 		splx(s);
227 		return;
228 	}
229 	callout_deactivate(tp->tt_2msl);
230 	/*
231 	 * 2 MSL timeout in shutdown went off.  If we're closed but
232 	 * still waiting for peer to close and connection has been idle
233 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
234 	 * control block.  Otherwise, check again in a bit.
235 	 */
236 	if (tp->t_state != TCPS_TIME_WAIT &&
237 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
238 		callout_reset(tp->tt_2msl, tcp_keepintvl,
239 			      tcp_timer_2msl, tp);
240 	else
241 		tp = tcp_close(tp);
242 
243 #ifdef TCPDEBUG
244 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
245 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
246 			  PRU_SLOWTIMO);
247 #endif
248 	if (tp)
249 		INP_UNLOCK(inp);
250 	INP_INFO_WUNLOCK(&tcbinfo);
251 	splx(s);
252 }
253 
254 void
255 tcp_timer_2msl_tw(xtw)
256 	void *xtw;
257 {
258 	struct tcptw *tw = xtw;
259 	int s;
260 
261 	s = splnet();
262 	INP_INFO_WLOCK(&tcbinfo);
263 	if (tw->tw_inpcb == NULL) {
264 		INP_INFO_WUNLOCK(&tcbinfo);
265 		splx(s);
266 		return;
267 	}
268 	INP_LOCK(tw->tw_inpcb);
269 	if (callout_pending(tw->tt_2msl) || !callout_active(tw->tt_2msl)) {
270 		INP_UNLOCK(tw->tw_inpcb);
271 		INP_INFO_WUNLOCK(&tcbinfo);
272 		splx(s);
273 		return;
274 	}
275 	callout_deactivate(tw->tt_2msl);
276 	tcp_twclose(tw);
277 	INP_INFO_WUNLOCK(&tcbinfo);
278 	splx(s);
279 }
280 
281 void
282 tcp_timer_keep(xtp)
283 	void *xtp;
284 {
285 	struct tcpcb *tp = xtp;
286 	struct tcptemp *t_template;
287 	int s;
288 	struct inpcb *inp;
289 #ifdef TCPDEBUG
290 	int ostate;
291 
292 	ostate = tp->t_state;
293 #endif
294 	s = splnet();
295 	INP_INFO_WLOCK(&tcbinfo);
296 	inp = tp->t_inpcb;
297 	if (!inp) {
298 		INP_INFO_WUNLOCK(&tcbinfo);
299 		splx(s);
300 		return;
301 	}
302 	INP_LOCK(inp);
303 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
304 		INP_UNLOCK(inp);
305 		INP_INFO_WUNLOCK(&tcbinfo);
306 		splx(s);
307 		return;
308 	}
309 	callout_deactivate(tp->tt_keep);
310 	/*
311 	 * Keep-alive timer went off; send something
312 	 * or drop connection if idle for too long.
313 	 */
314 	tcpstat.tcps_keeptimeo++;
315 	if (tp->t_state < TCPS_ESTABLISHED)
316 		goto dropit;
317 	if ((always_keepalive ||
318 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
319 	    tp->t_state <= TCPS_CLOSING) {
320 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
321 			goto dropit;
322 		/*
323 		 * Send a packet designed to force a response
324 		 * if the peer is up and reachable:
325 		 * either an ACK if the connection is still alive,
326 		 * or an RST if the peer has closed the connection
327 		 * due to timeout or reboot.
328 		 * Using sequence number tp->snd_una-1
329 		 * causes the transmitted zero-length segment
330 		 * to lie outside the receive window;
331 		 * by the protocol spec, this requires the
332 		 * correspondent TCP to respond.
333 		 */
334 		tcpstat.tcps_keepprobe++;
335 		t_template = tcpip_maketemplate(inp);
336 		if (t_template) {
337 			tcp_respond(tp, t_template->tt_ipgen,
338 				    &t_template->tt_t, (struct mbuf *)NULL,
339 				    tp->rcv_nxt, tp->snd_una - 1, 0);
340 			(void) m_free(dtom(t_template));
341 		}
342 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
343 	} else
344 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
345 
346 #ifdef TCPDEBUG
347 	if (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 	INP_UNLOCK(inp);
352 	INP_INFO_WUNLOCK(&tcbinfo);
353 	splx(s);
354 	return;
355 
356 dropit:
357 	tcpstat.tcps_keepdrops++;
358 	tp = tcp_drop(tp, ETIMEDOUT);
359 
360 #ifdef TCPDEBUG
361 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
362 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
363 			  PRU_SLOWTIMO);
364 #endif
365 	if (tp)
366 		INP_UNLOCK(tp->t_inpcb);
367 	INP_INFO_WUNLOCK(&tcbinfo);
368 	splx(s);
369 }
370 
371 void
372 tcp_timer_persist(xtp)
373 	void *xtp;
374 {
375 	struct tcpcb *tp = xtp;
376 	int s;
377 	struct inpcb *inp;
378 #ifdef TCPDEBUG
379 	int ostate;
380 
381 	ostate = tp->t_state;
382 #endif
383 	s = splnet();
384 	INP_INFO_WLOCK(&tcbinfo);
385 	inp = tp->t_inpcb;
386 	if (!inp) {
387 		INP_INFO_WUNLOCK(&tcbinfo);
388 		splx(s);
389 		return;
390 	}
391 	INP_LOCK(inp);
392 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
393 		INP_UNLOCK(inp);
394 		INP_INFO_WUNLOCK(&tcbinfo);
395 		splx(s);
396 		return;
397 	}
398 	callout_deactivate(tp->tt_persist);
399 	/*
400 	 * Persistance timer into zero window.
401 	 * Force a byte to be output, if possible.
402 	 */
403 	tcpstat.tcps_persisttimeo++;
404 	/*
405 	 * Hack: if the peer is dead/unreachable, we do not
406 	 * time out if the window is closed.  After a full
407 	 * backoff, drop the connection if the idle time
408 	 * (no responses to probes) reaches the maximum
409 	 * backoff that we would use if retransmitting.
410 	 */
411 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
412 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
413 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
414 		tcpstat.tcps_persistdrop++;
415 		tp = tcp_drop(tp, ETIMEDOUT);
416 		goto out;
417 	}
418 	tcp_setpersist(tp);
419 	tp->t_force = 1;
420 	(void) tcp_output(tp);
421 	tp->t_force = 0;
422 
423 out:
424 #ifdef TCPDEBUG
425 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
426 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
427 			  PRU_SLOWTIMO);
428 #endif
429 	if (tp)
430 		INP_UNLOCK(inp);
431 	INP_INFO_WUNLOCK(&tcbinfo);
432 	splx(s);
433 }
434 
435 void
436 tcp_timer_rexmt(xtp)
437 	void *xtp;
438 {
439 	struct tcpcb *tp = xtp;
440 	int s;
441 	int rexmt;
442 	int headlocked;
443 	struct inpcb *inp;
444 #ifdef TCPDEBUG
445 	int ostate;
446 
447 	ostate = tp->t_state;
448 #endif
449 	s = splnet();
450 	INP_INFO_WLOCK(&tcbinfo);
451 	headlocked = 1;
452 	inp = tp->t_inpcb;
453 	if (!inp) {
454 		INP_INFO_WUNLOCK(&tcbinfo);
455 		splx(s);
456 		return;
457 	}
458 	INP_LOCK(inp);
459 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
460 		INP_UNLOCK(inp);
461 		INP_INFO_WUNLOCK(&tcbinfo);
462 		splx(s);
463 		return;
464 	}
465 	callout_deactivate(tp->tt_rexmt);
466 	/*
467 	 * Retransmission timer went off.  Message has not
468 	 * been acked within retransmit interval.  Back off
469 	 * to a longer retransmit interval and retransmit one segment.
470 	 */
471 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
472 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
473 		tcpstat.tcps_timeoutdrop++;
474 		tp = tcp_drop(tp, tp->t_softerror ?
475 			      tp->t_softerror : ETIMEDOUT);
476 		goto out;
477 	}
478 	INP_INFO_WUNLOCK(&tcbinfo);
479 	headlocked = 0;
480 	if (tp->t_rxtshift == 1) {
481 		/*
482 		 * first retransmit; record ssthresh and cwnd so they can
483 	 	 * be recovered if this turns out to be a "bad" retransmit.
484 		 * A retransmit is considered "bad" if an ACK for this
485 		 * segment is received within RTT/2 interval; the assumption
486 		 * here is that the ACK was already in flight.  See
487 		 * "On Estimating End-to-End Network Path Properties" by
488 		 * Allman and Paxson for more details.
489 		 */
490 		tp->snd_cwnd_prev = tp->snd_cwnd;
491 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
492 		tp->snd_high_prev = tp->snd_high;
493 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
494 	}
495 	tcpstat.tcps_rexmttimeo++;
496 	if (tp->t_state == TCPS_SYN_SENT)
497 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
498 	else
499 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
500 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
501 		      tp->t_rttmin, TCPTV_REXMTMAX);
502 	/*
503 	 * Disable rfc1323 and rfc1644 if we havn't got any response to
504 	 * our third SYN to work-around some broken terminal servers
505 	 * (most of which have hopefully been retired) that have bad VJ
506 	 * header compression code which trashes TCP segments containing
507 	 * unknown-to-them TCP options.
508 	 */
509 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
510 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
511 	/*
512 	 * If losing, let the lower level know and try for
513 	 * a better route.  Also, if we backed off this far,
514 	 * our srtt estimate is probably bogus.  Clobber it
515 	 * so we'll take the next rtt measurement as our srtt;
516 	 * move the current srtt into rttvar to keep the current
517 	 * retransmit times until then.
518 	 */
519 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
520 #ifdef INET6
521 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
522 			in6_losing(tp->t_inpcb);
523 		else
524 #endif
525 		in_losing(tp->t_inpcb);
526 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
527 		tp->t_srtt = 0;
528 	}
529 	tp->snd_nxt = tp->snd_una;
530 	tp->snd_high = tp->snd_max;
531 	/*
532 	 * Force a segment to be sent.
533 	 */
534 	tp->t_flags |= TF_ACKNOW;
535 	/*
536 	 * If timing a segment in this window, stop the timer.
537 	 */
538 	tp->t_rtttime = 0;
539 	/*
540 	 * Close the congestion window down to one segment
541 	 * (we'll open it by one segment for each ack we get).
542 	 * Since we probably have a window's worth of unacked
543 	 * data accumulated, this "slow start" keeps us from
544 	 * dumping all that data as back-to-back packets (which
545 	 * might overwhelm an intermediate gateway).
546 	 *
547 	 * There are two phases to the opening: Initially we
548 	 * open by one mss on each ack.  This makes the window
549 	 * size increase exponentially with time.  If the
550 	 * window is larger than the path can handle, this
551 	 * exponential growth results in dropped packet(s)
552 	 * almost immediately.  To get more time between
553 	 * drops but still "push" the network to take advantage
554 	 * of improving conditions, we switch from exponential
555 	 * to linear window opening at some threshhold size.
556 	 * For a threshhold, we use half the current window
557 	 * size, truncated to a multiple of the mss.
558 	 *
559 	 * (the minimum cwnd that will give us exponential
560 	 * growth is 2 mss.  We don't allow the threshhold
561 	 * to go below this.)
562 	 */
563 	{
564 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
565 		if (win < 2)
566 			win = 2;
567 		tp->snd_cwnd = tp->t_maxseg;
568 		tp->snd_ssthresh = win * tp->t_maxseg;
569 		tp->t_dupacks = 0;
570 	}
571 	(void) tcp_output(tp);
572 
573 out:
574 #ifdef TCPDEBUG
575 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
576 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
577 			  PRU_SLOWTIMO);
578 #endif
579 	if (tp)
580 		INP_UNLOCK(inp);
581 	if (headlocked)
582 		INP_INFO_WUNLOCK(&tcbinfo);
583 	splx(s);
584 }
585