xref: /freebsd/sys/netinet/tcp_timer.c (revision 4b2eaea43fec8e8792be611dea204071a10b655a)
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_keep(xtp)
256 	void *xtp;
257 {
258 	struct tcpcb *tp = xtp;
259 	struct tcptemp *t_template;
260 	int s;
261 	struct inpcb *inp;
262 #ifdef TCPDEBUG
263 	int ostate;
264 
265 	ostate = tp->t_state;
266 #endif
267 	s = splnet();
268 	INP_INFO_WLOCK(&tcbinfo);
269 	inp = tp->t_inpcb;
270 	if (!inp) {
271 		INP_INFO_WUNLOCK(&tcbinfo);
272 		splx(s);
273 		return;
274 	}
275 	INP_LOCK(inp);
276 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
277 		INP_UNLOCK(inp);
278 		INP_INFO_WUNLOCK(&tcbinfo);
279 		splx(s);
280 		return;
281 	}
282 	callout_deactivate(tp->tt_keep);
283 	/*
284 	 * Keep-alive timer went off; send something
285 	 * or drop connection if idle for too long.
286 	 */
287 	tcpstat.tcps_keeptimeo++;
288 	if (tp->t_state < TCPS_ESTABLISHED)
289 		goto dropit;
290 	if ((always_keepalive ||
291 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
292 	    tp->t_state <= TCPS_CLOSING) {
293 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
294 			goto dropit;
295 		/*
296 		 * Send a packet designed to force a response
297 		 * if the peer is up and reachable:
298 		 * either an ACK if the connection is still alive,
299 		 * or an RST if the peer has closed the connection
300 		 * due to timeout or reboot.
301 		 * Using sequence number tp->snd_una-1
302 		 * causes the transmitted zero-length segment
303 		 * to lie outside the receive window;
304 		 * by the protocol spec, this requires the
305 		 * correspondent TCP to respond.
306 		 */
307 		tcpstat.tcps_keepprobe++;
308 		t_template = tcp_maketemplate(tp);
309 		if (t_template) {
310 			tcp_respond(tp, t_template->tt_ipgen,
311 				    &t_template->tt_t, (struct mbuf *)NULL,
312 				    tp->rcv_nxt, tp->snd_una - 1, 0);
313 			(void) m_free(dtom(t_template));
314 		}
315 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
316 	} else
317 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
318 
319 #ifdef TCPDEBUG
320 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
321 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
322 			  PRU_SLOWTIMO);
323 #endif
324 	INP_UNLOCK(inp);
325 	INP_INFO_WUNLOCK(&tcbinfo);
326 	splx(s);
327 	return;
328 
329 dropit:
330 	tcpstat.tcps_keepdrops++;
331 	tp = tcp_drop(tp, ETIMEDOUT);
332 
333 #ifdef TCPDEBUG
334 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
335 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
336 			  PRU_SLOWTIMO);
337 #endif
338 	if (tp)
339 		INP_UNLOCK(tp->t_inpcb);
340 	INP_INFO_WUNLOCK(&tcbinfo);
341 	splx(s);
342 }
343 
344 void
345 tcp_timer_persist(xtp)
346 	void *xtp;
347 {
348 	struct tcpcb *tp = xtp;
349 	int s;
350 	struct inpcb *inp;
351 #ifdef TCPDEBUG
352 	int ostate;
353 
354 	ostate = tp->t_state;
355 #endif
356 	s = splnet();
357 	INP_INFO_WLOCK(&tcbinfo);
358 	inp = tp->t_inpcb;
359 	if (!inp) {
360 		INP_INFO_WUNLOCK(&tcbinfo);
361 		splx(s);
362 		return;
363 	}
364 	INP_LOCK(inp);
365 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
366 		INP_UNLOCK(inp);
367 		INP_INFO_WUNLOCK(&tcbinfo);
368 		splx(s);
369 		return;
370 	}
371 	callout_deactivate(tp->tt_persist);
372 	/*
373 	 * Persistance timer into zero window.
374 	 * Force a byte to be output, if possible.
375 	 */
376 	tcpstat.tcps_persisttimeo++;
377 	/*
378 	 * Hack: if the peer is dead/unreachable, we do not
379 	 * time out if the window is closed.  After a full
380 	 * backoff, drop the connection if the idle time
381 	 * (no responses to probes) reaches the maximum
382 	 * backoff that we would use if retransmitting.
383 	 */
384 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
385 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
386 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
387 		tcpstat.tcps_persistdrop++;
388 		tp = tcp_drop(tp, ETIMEDOUT);
389 		goto out;
390 	}
391 	tcp_setpersist(tp);
392 	tp->t_force = 1;
393 	(void) tcp_output(tp);
394 	tp->t_force = 0;
395 
396 out:
397 #ifdef TCPDEBUG
398 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
399 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
400 			  PRU_SLOWTIMO);
401 #endif
402 	if (tp)
403 		INP_UNLOCK(inp);
404 	INP_INFO_WUNLOCK(&tcbinfo);
405 	splx(s);
406 }
407 
408 void
409 tcp_timer_rexmt(xtp)
410 	void *xtp;
411 {
412 	struct tcpcb *tp = xtp;
413 	int s;
414 	int rexmt;
415 	int headlocked;
416 	struct inpcb *inp;
417 #ifdef TCPDEBUG
418 	int ostate;
419 
420 	ostate = tp->t_state;
421 #endif
422 	s = splnet();
423 	INP_INFO_WLOCK(&tcbinfo);
424 	headlocked = 1;
425 	inp = tp->t_inpcb;
426 	if (!inp) {
427 		INP_INFO_WUNLOCK(&tcbinfo);
428 		splx(s);
429 		return;
430 	}
431 	INP_LOCK(inp);
432 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
433 		INP_UNLOCK(inp);
434 		INP_INFO_WUNLOCK(&tcbinfo);
435 		splx(s);
436 		return;
437 	}
438 	callout_deactivate(tp->tt_rexmt);
439 	/*
440 	 * Retransmission timer went off.  Message has not
441 	 * been acked within retransmit interval.  Back off
442 	 * to a longer retransmit interval and retransmit one segment.
443 	 */
444 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
445 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
446 		tcpstat.tcps_timeoutdrop++;
447 		tp = tcp_drop(tp, tp->t_softerror ?
448 			      tp->t_softerror : ETIMEDOUT);
449 		goto out;
450 	}
451 	INP_INFO_WUNLOCK(&tcbinfo);
452 	headlocked = 0;
453 	if (tp->t_rxtshift == 1) {
454 		/*
455 		 * first retransmit; record ssthresh and cwnd so they can
456 	 	 * be recovered if this turns out to be a "bad" retransmit.
457 		 * A retransmit is considered "bad" if an ACK for this
458 		 * segment is received within RTT/2 interval; the assumption
459 		 * here is that the ACK was already in flight.  See
460 		 * "On Estimating End-to-End Network Path Properties" by
461 		 * Allman and Paxson for more details.
462 		 */
463 		tp->snd_cwnd_prev = tp->snd_cwnd;
464 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
465 		tp->snd_high_prev = tp->snd_high;
466 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
467 	}
468 	tcpstat.tcps_rexmttimeo++;
469 	if (tp->t_state == TCPS_SYN_SENT)
470 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
471 	else
472 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
473 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
474 		      tp->t_rttmin, TCPTV_REXMTMAX);
475 	/*
476 	 * Disable rfc1323 and rfc1644 if we havn't got any response to
477 	 * our third SYN to work-around some broken terminal servers
478 	 * (most of which have hopefully been retired) that have bad VJ
479 	 * header compression code which trashes TCP segments containing
480 	 * unknown-to-them TCP options.
481 	 */
482 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
483 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
484 	/*
485 	 * If losing, let the lower level know and try for
486 	 * a better route.  Also, if we backed off this far,
487 	 * our srtt estimate is probably bogus.  Clobber it
488 	 * so we'll take the next rtt measurement as our srtt;
489 	 * move the current srtt into rttvar to keep the current
490 	 * retransmit times until then.
491 	 */
492 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
493 #ifdef INET6
494 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
495 			in6_losing(tp->t_inpcb);
496 		else
497 #endif
498 		in_losing(tp->t_inpcb);
499 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
500 		tp->t_srtt = 0;
501 	}
502 	tp->snd_nxt = tp->snd_una;
503 	tp->snd_high = tp->snd_max;
504 	/*
505 	 * Force a segment to be sent.
506 	 */
507 	tp->t_flags |= TF_ACKNOW;
508 	/*
509 	 * If timing a segment in this window, stop the timer.
510 	 */
511 	tp->t_rtttime = 0;
512 	/*
513 	 * Close the congestion window down to one segment
514 	 * (we'll open it by one segment for each ack we get).
515 	 * Since we probably have a window's worth of unacked
516 	 * data accumulated, this "slow start" keeps us from
517 	 * dumping all that data as back-to-back packets (which
518 	 * might overwhelm an intermediate gateway).
519 	 *
520 	 * There are two phases to the opening: Initially we
521 	 * open by one mss on each ack.  This makes the window
522 	 * size increase exponentially with time.  If the
523 	 * window is larger than the path can handle, this
524 	 * exponential growth results in dropped packet(s)
525 	 * almost immediately.  To get more time between
526 	 * drops but still "push" the network to take advantage
527 	 * of improving conditions, we switch from exponential
528 	 * to linear window opening at some threshhold size.
529 	 * For a threshhold, we use half the current window
530 	 * size, truncated to a multiple of the mss.
531 	 *
532 	 * (the minimum cwnd that will give us exponential
533 	 * growth is 2 mss.  We don't allow the threshhold
534 	 * to go below this.)
535 	 */
536 	{
537 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
538 		if (win < 2)
539 			win = 2;
540 		tp->snd_cwnd = tp->t_maxseg;
541 		tp->snd_ssthresh = win * tp->t_maxseg;
542 		tp->t_dupacks = 0;
543 	}
544 	(void) tcp_output(tp);
545 
546 out:
547 #ifdef TCPDEBUG
548 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
549 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
550 			  PRU_SLOWTIMO);
551 #endif
552 	if (tp)
553 		INP_UNLOCK(inp);
554 	if (headlocked)
555 		INP_INFO_WUNLOCK(&tcbinfo);
556 	splx(s);
557 }
558