xref: /dflybsd-src/sys/netinet/tcp_timer.c (revision 984263bcb83ad82313113c6ac840d99124d8f90c)
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: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
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 = (int)((int64_t)tt * 1000 / hz);
78 
79 	error = sysctl_handle_int(oidp, &s, 0, req);
80 	if (error || !req->newptr)
81 		return (error);
82 
83 	tt = (int)((int64_t)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 int	tcp_rexmit_min;
113 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
114     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
115 
116 int	tcp_rexmit_slop;
117 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
118     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
119 
120 static int	always_keepalive = 0;
121 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
122     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
123 
124 static int	tcp_keepcnt = TCPTV_KEEPCNT;
125 	/* max idle probes */
126 int	tcp_maxpersistidle;
127 	/* max idle time in persist */
128 int	tcp_maxidle;
129 
130 /*
131  * Tcp protocol timeout routine called every 500 ms.
132  * Updates timestamps used for TCP
133  * causes finite state machine actions if timers expire.
134  */
135 void
136 tcp_slowtimo()
137 {
138 	int s;
139 
140 	s = splnet();
141 
142 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
143 
144 	splx(s);
145 }
146 
147 /*
148  * Cancel all timers for TCP tp.
149  */
150 void
151 tcp_canceltimers(tp)
152 	struct tcpcb *tp;
153 {
154 	callout_stop(tp->tt_2msl);
155 	callout_stop(tp->tt_persist);
156 	callout_stop(tp->tt_keep);
157 	callout_stop(tp->tt_rexmt);
158 }
159 
160 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
161     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
162 
163 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
164     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
165 
166 static int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
167 
168 /*
169  * TCP timer processing.
170  */
171 void
172 tcp_timer_delack(xtp)
173 	void *xtp;
174 {
175 	struct tcpcb *tp = xtp;
176 	int s;
177 
178 	s = splnet();
179 	if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
180 		splx(s);
181 		return;
182 	}
183 	callout_deactivate(tp->tt_delack);
184 
185 	tp->t_flags |= TF_ACKNOW;
186 	tcpstat.tcps_delack++;
187 	(void) tcp_output(tp);
188 	splx(s);
189 }
190 
191 void
192 tcp_timer_2msl(xtp)
193 	void *xtp;
194 {
195 	struct tcpcb *tp = xtp;
196 	int s;
197 #ifdef TCPDEBUG
198 	int ostate;
199 
200 	ostate = tp->t_state;
201 #endif
202 	s = splnet();
203 	if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
204 		splx(s);
205 		return;
206 	}
207 	callout_deactivate(tp->tt_2msl);
208 	/*
209 	 * 2 MSL timeout in shutdown went off.  If we're closed but
210 	 * still waiting for peer to close and connection has been idle
211 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
212 	 * control block.  Otherwise, check again in a bit.
213 	 */
214 	if (tp->t_state != TCPS_TIME_WAIT &&
215 	    (ticks - tp->t_rcvtime) <= tcp_maxidle)
216 		callout_reset(tp->tt_2msl, tcp_keepintvl,
217 			      tcp_timer_2msl, tp);
218 	else
219 		tp = tcp_close(tp);
220 
221 #ifdef TCPDEBUG
222 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
223 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
224 			  PRU_SLOWTIMO);
225 #endif
226 	splx(s);
227 }
228 
229 void
230 tcp_timer_keep(xtp)
231 	void *xtp;
232 {
233 	struct tcpcb *tp = xtp;
234 	struct tcptemp *t_template;
235 	int s;
236 #ifdef TCPDEBUG
237 	int ostate;
238 
239 	ostate = tp->t_state;
240 #endif
241 	s = splnet();
242 	if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
243 		splx(s);
244 		return;
245 	}
246 	callout_deactivate(tp->tt_keep);
247 	/*
248 	 * Keep-alive timer went off; send something
249 	 * or drop connection if idle for too long.
250 	 */
251 	tcpstat.tcps_keeptimeo++;
252 	if (tp->t_state < TCPS_ESTABLISHED)
253 		goto dropit;
254 	if ((always_keepalive ||
255 	     tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
256 	    tp->t_state <= TCPS_CLOSING) {
257 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
258 			goto dropit;
259 		/*
260 		 * Send a packet designed to force a response
261 		 * if the peer is up and reachable:
262 		 * either an ACK if the connection is still alive,
263 		 * or an RST if the peer has closed the connection
264 		 * due to timeout or reboot.
265 		 * Using sequence number tp->snd_una-1
266 		 * causes the transmitted zero-length segment
267 		 * to lie outside the receive window;
268 		 * by the protocol spec, this requires the
269 		 * correspondent TCP to respond.
270 		 */
271 		tcpstat.tcps_keepprobe++;
272 		t_template = tcp_maketemplate(tp);
273 		if (t_template) {
274 			tcp_respond(tp, t_template->tt_ipgen,
275 				    &t_template->tt_t, (struct mbuf *)NULL,
276 				    tp->rcv_nxt, tp->snd_una - 1, 0);
277 			(void) m_free(dtom(t_template));
278 		}
279 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
280 	} else
281 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
282 
283 #ifdef TCPDEBUG
284 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
285 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
286 			  PRU_SLOWTIMO);
287 #endif
288 	splx(s);
289 	return;
290 
291 dropit:
292 	tcpstat.tcps_keepdrops++;
293 	tp = tcp_drop(tp, ETIMEDOUT);
294 
295 #ifdef TCPDEBUG
296 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
297 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
298 			  PRU_SLOWTIMO);
299 #endif
300 	splx(s);
301 }
302 
303 void
304 tcp_timer_persist(xtp)
305 	void *xtp;
306 {
307 	struct tcpcb *tp = xtp;
308 	int s;
309 #ifdef TCPDEBUG
310 	int ostate;
311 
312 	ostate = tp->t_state;
313 #endif
314 	s = splnet();
315 	if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
316 		splx(s);
317 		return;
318 	}
319 	callout_deactivate(tp->tt_persist);
320 	/*
321 	 * Persistance timer into zero window.
322 	 * Force a byte to be output, if possible.
323 	 */
324 	tcpstat.tcps_persisttimeo++;
325 	/*
326 	 * Hack: if the peer is dead/unreachable, we do not
327 	 * time out if the window is closed.  After a full
328 	 * backoff, drop the connection if the idle time
329 	 * (no responses to probes) reaches the maximum
330 	 * backoff that we would use if retransmitting.
331 	 */
332 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
333 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
334 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
335 		tcpstat.tcps_persistdrop++;
336 		tp = tcp_drop(tp, ETIMEDOUT);
337 		goto out;
338 	}
339 	tcp_setpersist(tp);
340 	tp->t_force = 1;
341 	(void) tcp_output(tp);
342 	tp->t_force = 0;
343 
344 out:
345 #ifdef TCPDEBUG
346 	if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
347 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
348 			  PRU_SLOWTIMO);
349 #endif
350 	splx(s);
351 }
352 
353 void
354 tcp_timer_rexmt(xtp)
355 	void *xtp;
356 {
357 	struct tcpcb *tp = xtp;
358 	int s;
359 	int rexmt;
360 #ifdef TCPDEBUG
361 	int ostate;
362 
363 	ostate = tp->t_state;
364 #endif
365 	s = splnet();
366 	if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
367 		splx(s);
368 		return;
369 	}
370 	callout_deactivate(tp->tt_rexmt);
371 	/*
372 	 * Retransmission timer went off.  Message has not
373 	 * been acked within retransmit interval.  Back off
374 	 * to a longer retransmit interval and retransmit one segment.
375 	 */
376 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
377 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
378 		tcpstat.tcps_timeoutdrop++;
379 		tp = tcp_drop(tp, tp->t_softerror ?
380 			      tp->t_softerror : ETIMEDOUT);
381 		goto out;
382 	}
383 	if (tp->t_rxtshift == 1) {
384 		/*
385 		 * first retransmit; record ssthresh and cwnd so they can
386 	 	 * be recovered if this turns out to be a "bad" retransmit.
387 		 * A retransmit is considered "bad" if an ACK for this
388 		 * segment is received within RTT/2 interval; the assumption
389 		 * here is that the ACK was already in flight.  See
390 		 * "On Estimating End-to-End Network Path Properties" by
391 		 * Allman and Paxson for more details.
392 		 */
393 		tp->snd_cwnd_prev = tp->snd_cwnd;
394 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
395 		tp->snd_high_prev = tp->snd_high;
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 	 * Disable rfc1323 and rfc1644 if we havn't got any response to
407 	 * our third SYN to work-around some broken terminal servers
408 	 * (most of which have hopefully been retired) that have bad VJ
409 	 * header compression code which trashes TCP segments containing
410 	 * unknown-to-them TCP options.
411 	 */
412 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
413 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
414 	/*
415 	 * If losing, let the lower level know and try for
416 	 * a better route.  Also, if we backed off this far,
417 	 * our srtt estimate is probably bogus.  Clobber it
418 	 * so we'll take the next rtt measurement as our srtt;
419 	 * move the current srtt into rttvar to keep the current
420 	 * retransmit times until then.
421 	 */
422 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
423 #ifdef INET6
424 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
425 			in6_losing(tp->t_inpcb);
426 		else
427 #endif
428 		in_losing(tp->t_inpcb);
429 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
430 		tp->t_srtt = 0;
431 	}
432 	tp->snd_nxt = tp->snd_una;
433 	tp->snd_high = tp->snd_max;
434 	/*
435 	 * Force a segment to be sent.
436 	 */
437 	tp->t_flags |= TF_ACKNOW;
438 	/*
439 	 * If timing a segment in this window, stop the timer.
440 	 */
441 	tp->t_rtttime = 0;
442 	/*
443 	 * Close the congestion window down to one segment
444 	 * (we'll open it by one segment for each ack we get).
445 	 * Since we probably have a window's worth of unacked
446 	 * data accumulated, this "slow start" keeps us from
447 	 * dumping all that data as back-to-back packets (which
448 	 * might overwhelm an intermediate gateway).
449 	 *
450 	 * There are two phases to the opening: Initially we
451 	 * open by one mss on each ack.  This makes the window
452 	 * size increase exponentially with time.  If the
453 	 * window is larger than the path can handle, this
454 	 * exponential growth results in dropped packet(s)
455 	 * almost immediately.  To get more time between
456 	 * drops but still "push" the network to take advantage
457 	 * of improving conditions, we switch from exponential
458 	 * to linear window opening at some threshhold size.
459 	 * For a threshhold, we use half the current window
460 	 * size, truncated to a multiple of the mss.
461 	 *
462 	 * (the minimum cwnd that will give us exponential
463 	 * growth is 2 mss.  We don't allow the threshhold
464 	 * to go below this.)
465 	 */
466 	{
467 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
468 		if (win < 2)
469 			win = 2;
470 		tp->snd_cwnd = tp->t_maxseg;
471 		tp->snd_ssthresh = win * tp->t_maxseg;
472 		tp->t_dupacks = 0;
473 	}
474 	(void) tcp_output(tp);
475 
476 out:
477 #ifdef TCPDEBUG
478 	if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
479 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
480 			  PRU_SLOWTIMO);
481 #endif
482 	splx(s);
483 }
484