xref: /freebsd/sys/netinet/tcp_timer.c (revision a3e8fd0b7f663db7eafff527d5c3ca3bcfa8a537)
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 	INP_LOCK(inp);
182 	INP_INFO_RUNLOCK(&tcbinfo);
183 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
184 		INP_UNLOCK(inp);
185 		splx(s);
186 		return;
187 	}
188 	callout_deactivate(tp->tt_delack);
189 
190 	tp->t_flags |= TF_ACKNOW;
191 	tcpstat.tcps_delack++;
192 	(void) tcp_output(tp);
193 	INP_UNLOCK(inp);
194 	splx(s);
195 }
196 
197 void
198 tcp_timer_2msl(xtp)
199 	void *xtp;
200 {
201 	struct tcpcb *tp = xtp;
202 	int s;
203 	struct inpcb *inp;
204 #ifdef TCPDEBUG
205 	int ostate;
206 
207 	ostate = tp->t_state;
208 #endif
209 	s = splnet();
210 	INP_INFO_WLOCK(&tcbinfo);
211 	inp = tp->t_inpcb;
212 	INP_LOCK(inp);
213 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
214 		INP_UNLOCK(tp->t_inpcb);
215 		INP_INFO_WUNLOCK(&tcbinfo);
216 		splx(s);
217 		return;
218 	}
219 	callout_deactivate(tp->tt_2msl);
220 	/*
221 	 * 2 MSL timeout in shutdown went off.  If we're closed but
222 	 * still waiting for peer to close and connection has been idle
223 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
224 	 * control block.  Otherwise, check again in a bit.
225 	 */
226 	if (tp->t_state != TCPS_TIME_WAIT &&
227 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
228 		callout_reset(tp->tt_2msl, tcp_keepintvl,
229 			      tcp_timer_2msl, tp);
230 	else
231 		tp = tcp_close(tp);
232 
233 #ifdef TCPDEBUG
234 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
235 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
236 			  PRU_SLOWTIMO);
237 #endif
238 	if (tp)
239 		INP_UNLOCK(inp);
240 	INP_INFO_WUNLOCK(&tcbinfo);
241 	splx(s);
242 }
243 
244 void
245 tcp_timer_keep(xtp)
246 	void *xtp;
247 {
248 	struct tcpcb *tp = xtp;
249 	struct tcptemp *t_template;
250 	int s;
251 	struct inpcb *inp;
252 #ifdef TCPDEBUG
253 	int ostate;
254 
255 	ostate = tp->t_state;
256 #endif
257 	s = splnet();
258 	INP_INFO_WLOCK(&tcbinfo);
259 	inp = tp->t_inpcb;
260 	INP_LOCK(inp);
261 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
262 		INP_UNLOCK(inp);
263 		INP_INFO_WUNLOCK(&tcbinfo);
264 		splx(s);
265 		return;
266 	}
267 	callout_deactivate(tp->tt_keep);
268 	/*
269 	 * Keep-alive timer went off; send something
270 	 * or drop connection if idle for too long.
271 	 */
272 	tcpstat.tcps_keeptimeo++;
273 	if (tp->t_state < TCPS_ESTABLISHED)
274 		goto dropit;
275 	if ((always_keepalive ||
276 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
277 	    tp->t_state <= TCPS_CLOSING) {
278 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
279 			goto dropit;
280 		/*
281 		 * Send a packet designed to force a response
282 		 * if the peer is up and reachable:
283 		 * either an ACK if the connection is still alive,
284 		 * or an RST if the peer has closed the connection
285 		 * due to timeout or reboot.
286 		 * Using sequence number tp->snd_una-1
287 		 * causes the transmitted zero-length segment
288 		 * to lie outside the receive window;
289 		 * by the protocol spec, this requires the
290 		 * correspondent TCP to respond.
291 		 */
292 		tcpstat.tcps_keepprobe++;
293 		t_template = tcp_maketemplate(tp);
294 		if (t_template) {
295 			tcp_respond(tp, t_template->tt_ipgen,
296 				    &t_template->tt_t, (struct mbuf *)NULL,
297 				    tp->rcv_nxt, tp->snd_una - 1, 0);
298 			(void) m_free(dtom(t_template));
299 		}
300 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
301 	} else
302 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
303 
304 #ifdef TCPDEBUG
305 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
306 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
307 			  PRU_SLOWTIMO);
308 #endif
309 	INP_UNLOCK(inp);
310 	INP_INFO_WUNLOCK(&tcbinfo);
311 	splx(s);
312 	return;
313 
314 dropit:
315 	tcpstat.tcps_keepdrops++;
316 	tp = tcp_drop(tp, ETIMEDOUT);
317 
318 #ifdef TCPDEBUG
319 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
320 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
321 			  PRU_SLOWTIMO);
322 #endif
323 	if (tp)
324 		INP_UNLOCK(tp->t_inpcb);
325 	INP_INFO_WUNLOCK(&tcbinfo);
326 	splx(s);
327 }
328 
329 void
330 tcp_timer_persist(xtp)
331 	void *xtp;
332 {
333 	struct tcpcb *tp = xtp;
334 	int s;
335 	struct inpcb *inp;
336 #ifdef TCPDEBUG
337 	int ostate;
338 
339 	ostate = tp->t_state;
340 #endif
341 	s = splnet();
342 	INP_INFO_WLOCK(&tcbinfo);
343 	inp = tp->t_inpcb;
344 	INP_LOCK(inp);
345 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
346 		INP_UNLOCK(inp);
347 		INP_INFO_WUNLOCK(&tcbinfo);
348 		splx(s);
349 		return;
350 	}
351 	callout_deactivate(tp->tt_persist);
352 	/*
353 	 * Persistance timer into zero window.
354 	 * Force a byte to be output, if possible.
355 	 */
356 	tcpstat.tcps_persisttimeo++;
357 	/*
358 	 * Hack: if the peer is dead/unreachable, we do not
359 	 * time out if the window is closed.  After a full
360 	 * backoff, drop the connection if the idle time
361 	 * (no responses to probes) reaches the maximum
362 	 * backoff that we would use if retransmitting.
363 	 */
364 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
365 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
366 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
367 		tcpstat.tcps_persistdrop++;
368 		tp = tcp_drop(tp, ETIMEDOUT);
369 		goto out;
370 	}
371 	tcp_setpersist(tp);
372 	tp->t_force = 1;
373 	(void) tcp_output(tp);
374 	tp->t_force = 0;
375 
376 out:
377 #ifdef TCPDEBUG
378 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
379 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
380 			  PRU_SLOWTIMO);
381 #endif
382 	if (tp)
383 		INP_UNLOCK(inp);
384 	INP_INFO_WUNLOCK(&tcbinfo);
385 	splx(s);
386 }
387 
388 void
389 tcp_timer_rexmt(xtp)
390 	void *xtp;
391 {
392 	struct tcpcb *tp = xtp;
393 	int s;
394 	int rexmt;
395 	int headlocked;
396 	struct inpcb *inp;
397 #ifdef TCPDEBUG
398 	int ostate;
399 
400 	ostate = tp->t_state;
401 #endif
402 	s = splnet();
403 	INP_INFO_WLOCK(&tcbinfo);
404 	headlocked = 1;
405 	inp = tp->t_inpcb;
406 	INP_LOCK(inp);
407 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
408 		INP_UNLOCK(inp);
409 		INP_INFO_WUNLOCK(&tcbinfo);
410 		splx(s);
411 		return;
412 	}
413 	callout_deactivate(tp->tt_rexmt);
414 	/*
415 	 * Retransmission timer went off.  Message has not
416 	 * been acked within retransmit interval.  Back off
417 	 * to a longer retransmit interval and retransmit one segment.
418 	 */
419 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
420 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
421 		tcpstat.tcps_timeoutdrop++;
422 		tp = tcp_drop(tp, tp->t_softerror ?
423 			      tp->t_softerror : ETIMEDOUT);
424 		goto out;
425 	}
426 	INP_INFO_WUNLOCK(&tcbinfo);
427 	headlocked = 0;
428 	if (tp->t_rxtshift == 1) {
429 		/*
430 		 * first retransmit; record ssthresh and cwnd so they can
431 	 	 * be recovered if this turns out to be a "bad" retransmit.
432 		 * A retransmit is considered "bad" if an ACK for this
433 		 * segment is received within RTT/2 interval; the assumption
434 		 * here is that the ACK was already in flight.  See
435 		 * "On Estimating End-to-End Network Path Properties" by
436 		 * Allman and Paxson for more details.
437 		 */
438 		tp->snd_cwnd_prev = tp->snd_cwnd;
439 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
440 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
441 	}
442 	tcpstat.tcps_rexmttimeo++;
443 	if (tp->t_state == TCPS_SYN_SENT)
444 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
445 	else
446 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
447 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
448 		      tp->t_rttmin, TCPTV_REXMTMAX);
449 	/*
450 	 * Disable rfc1323 and rfc1644 if we havn't got any response to
451 	 * our third SYN to work-around some broken terminal servers
452 	 * (most of which have hopefully been retired) that have bad VJ
453 	 * header compression code which trashes TCP segments containing
454 	 * unknown-to-them TCP options.
455 	 */
456 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
457 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
458 	/*
459 	 * If losing, let the lower level know and try for
460 	 * a better route.  Also, if we backed off this far,
461 	 * our srtt estimate is probably bogus.  Clobber it
462 	 * so we'll take the next rtt measurement as our srtt;
463 	 * move the current srtt into rttvar to keep the current
464 	 * retransmit times until then.
465 	 */
466 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
467 #ifdef INET6
468 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
469 			in6_losing(tp->t_inpcb);
470 		else
471 #endif
472 		in_losing(tp->t_inpcb);
473 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
474 		tp->t_srtt = 0;
475 	}
476 	tp->snd_nxt = tp->snd_una;
477 	/*
478 	 * Note:  We overload snd_recover to function also as the
479 	 * snd_last variable described in RFC 2582
480 	 */
481 	tp->snd_recover = tp->snd_max;
482 	/*
483 	 * Force a segment to be sent.
484 	 */
485 	tp->t_flags |= TF_ACKNOW;
486 	/*
487 	 * If timing a segment in this window, stop the timer.
488 	 */
489 	tp->t_rtttime = 0;
490 	/*
491 	 * Close the congestion window down to one segment
492 	 * (we'll open it by one segment for each ack we get).
493 	 * Since we probably have a window's worth of unacked
494 	 * data accumulated, this "slow start" keeps us from
495 	 * dumping all that data as back-to-back packets (which
496 	 * might overwhelm an intermediate gateway).
497 	 *
498 	 * There are two phases to the opening: Initially we
499 	 * open by one mss on each ack.  This makes the window
500 	 * size increase exponentially with time.  If the
501 	 * window is larger than the path can handle, this
502 	 * exponential growth results in dropped packet(s)
503 	 * almost immediately.  To get more time between
504 	 * drops but still "push" the network to take advantage
505 	 * of improving conditions, we switch from exponential
506 	 * to linear window opening at some threshhold size.
507 	 * For a threshhold, we use half the current window
508 	 * size, truncated to a multiple of the mss.
509 	 *
510 	 * (the minimum cwnd that will give us exponential
511 	 * growth is 2 mss.  We don't allow the threshhold
512 	 * to go below this.)
513 	 */
514 	{
515 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
516 		if (win < 2)
517 			win = 2;
518 		tp->snd_cwnd = tp->t_maxseg;
519 		tp->snd_ssthresh = win * tp->t_maxseg;
520 		tp->t_dupacks = 0;
521 	}
522 	(void) tcp_output(tp);
523 
524 out:
525 #ifdef TCPDEBUG
526 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
527 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
528 			  PRU_SLOWTIMO);
529 #endif
530 	if (tp)
531 		INP_UNLOCK(inp);
532 	if (headlocked)
533 		INP_INFO_WUNLOCK(&tcbinfo);
534 	splx(s);
535 }
536