xref: /openbsd-src/sys/netinet/tcp_timer.c (revision d4741794dd2f512d997014f8bd85fbb24d935059)
1 /*	$OpenBSD: tcp_timer.c,v 1.53 2016/12/19 08:36:49 mpi Exp $	*/
2 /*	$NetBSD: tcp_timer.c,v 1.14 1996/02/13 23:44:09 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)tcp_timer.c	8.1 (Berkeley) 6/10/93
33  */
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/mbuf.h>
38 #include <sys/socket.h>
39 #include <sys/socketvar.h>
40 #include <sys/protosw.h>
41 #include <sys/kernel.h>
42 #include <sys/pool.h>
43 
44 #include <net/route.h>
45 
46 #include <netinet/in.h>
47 #include <netinet/ip.h>
48 #include <netinet/in_pcb.h>
49 #include <netinet/ip_var.h>
50 #include <netinet/tcp.h>
51 #include <netinet/tcp_fsm.h>
52 #include <netinet/tcp_timer.h>
53 #include <netinet/tcp_var.h>
54 #include <netinet/ip_icmp.h>
55 #include <netinet/tcp_seq.h>
56 
57 int	tcp_always_keepalive;
58 int	tcp_keepidle;
59 int	tcp_keepintvl;
60 int	tcp_maxpersistidle;	/* max idle time in persist */
61 int	tcp_maxidle;
62 
63 /*
64  * Time to delay the ACK.  This is initialized in tcp_init(), unless
65  * its patched.
66  */
67 int	tcp_delack_ticks;
68 
69 void	tcp_timer_rexmt(void *);
70 void	tcp_timer_persist(void *);
71 void	tcp_timer_keep(void *);
72 void	tcp_timer_2msl(void *);
73 
74 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
75 	tcp_timer_rexmt,
76 	tcp_timer_persist,
77 	tcp_timer_keep,
78 	tcp_timer_2msl,
79 };
80 
81 /*
82  * Timer state initialization, called from tcp_init().
83  */
84 void
85 tcp_timer_init(void)
86 {
87 
88 	if (tcp_keepidle == 0)
89 		tcp_keepidle = TCPTV_KEEP_IDLE;
90 
91 	if (tcp_keepintvl == 0)
92 		tcp_keepintvl = TCPTV_KEEPINTVL;
93 
94 	if (tcp_maxpersistidle == 0)
95 		tcp_maxpersistidle = TCPTV_KEEP_IDLE;
96 
97 	if (tcp_delack_ticks == 0)
98 		tcp_delack_ticks = TCP_DELACK_TICKS;
99 }
100 
101 /*
102  * Callout to process delayed ACKs for a TCPCB.
103  */
104 void
105 tcp_delack(void *arg)
106 {
107 	struct tcpcb *tp = arg;
108 	int s;
109 
110 	/*
111 	 * If tcp_output() wasn't able to transmit the ACK
112 	 * for whatever reason, it will restart the delayed
113 	 * ACK callout.
114 	 */
115 	NET_LOCK(s);
116 	if (tp->t_flags & TF_DEAD)
117 		goto out;
118 	tp->t_flags |= TF_ACKNOW;
119 	(void) tcp_output(tp);
120  out:
121 	NET_UNLOCK(s);
122 }
123 
124 /*
125  * Tcp protocol timeout routine called every 500 ms.
126  * Updates the timers in all active tcb's and
127  * causes finite state machine actions if timers expire.
128  */
129 void
130 tcp_slowtimo(void)
131 {
132 	splsoftassert(IPL_SOFTNET);
133 
134 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
135 	tcp_iss += TCP_ISSINCR2/PR_SLOWHZ;		/* increment iss */
136 	tcp_now++;					/* for timestamps */
137 }
138 
139 /*
140  * Cancel all timers for TCP tp.
141  */
142 void
143 tcp_canceltimers(struct tcpcb *tp)
144 {
145 	int i;
146 
147 	for (i = 0; i < TCPT_NTIMERS; i++)
148 		TCP_TIMER_DISARM(tp, i);
149 }
150 
151 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
152     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
153 
154 int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
155 
156 /*
157  * TCP timer processing.
158  */
159 
160 #ifdef TCP_SACK
161 void	tcp_timer_freesack(struct tcpcb *);
162 
163 void
164 tcp_timer_freesack(struct tcpcb *tp)
165 {
166 	struct sackhole *p, *q;
167 	/*
168 	 * Free SACK holes for 2MSL and REXMT timers.
169 	 */
170 	q = tp->snd_holes;
171 	while (q != NULL) {
172 		p = q;
173 		q = q->next;
174 		pool_put(&sackhl_pool, p);
175 	}
176 	tp->snd_holes = 0;
177 #ifdef TCP_FACK
178 	tp->snd_fack = tp->snd_una;
179 	tp->retran_data = 0;
180 	tp->snd_awnd = 0;
181 #endif /* TCP_FACK */
182 }
183 #endif /* TCP_SACK */
184 
185 void
186 tcp_timer_rexmt(void *arg)
187 {
188 	struct tcpcb *tp = arg;
189 	uint32_t rto;
190 	int s;
191 
192 	NET_LOCK(s);
193 	if (tp->t_flags & TF_DEAD)
194 		goto out;
195 
196 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
197 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
198 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) {
199 		struct sockaddr_in sin;
200 		struct icmp icmp;
201 
202 		tp->t_flags &= ~TF_PMTUD_PEND;
203 
204 		/* XXX create fake icmp message with relevant entries */
205 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
206 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
207 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
208 		icmp.icmp_ip.ip_dst = tp->t_inpcb->inp_faddr;
209 		icmp_mtudisc(&icmp, tp->t_inpcb->inp_rtableid);
210 
211 		/*
212 		 * Notify all connections to the same peer about
213 		 * new mss and trigger retransmit.
214 		 */
215 		bzero(&sin, sizeof(sin));
216 		sin.sin_len = sizeof(sin);
217 		sin.sin_family = AF_INET;
218 		sin.sin_addr = tp->t_inpcb->inp_faddr;
219 		in_pcbnotifyall(&tcbtable, sintosa(&sin),
220 		    tp->t_inpcb->inp_rtableid, EMSGSIZE, tcp_mtudisc);
221 		goto out;
222 	}
223 
224 #ifdef TCP_SACK
225 	tcp_timer_freesack(tp);
226 #endif
227 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
228 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
229 		tcpstat.tcps_timeoutdrop++;
230 		(void)tcp_drop(tp, tp->t_softerror ?
231 		    tp->t_softerror : ETIMEDOUT);
232 		goto out;
233 	}
234 	tcpstat.tcps_rexmttimeo++;
235 	rto = TCP_REXMTVAL(tp);
236 	if (rto < tp->t_rttmin)
237 		rto = tp->t_rttmin;
238 	TCPT_RANGESET(tp->t_rxtcur,
239 	    rto * tcp_backoff[tp->t_rxtshift],
240 	    tp->t_rttmin, TCPTV_REXMTMAX);
241 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
242 
243 	/*
244 	 * If we are losing and we are trying path MTU discovery,
245 	 * try turning it off.  This will avoid black holes in
246 	 * the network which suppress or fail to send "packet
247 	 * too big" ICMP messages.  We should ideally do
248 	 * lots more sophisticated searching to find the right
249 	 * value here...
250 	 */
251 	if (ip_mtudisc && tp->t_inpcb &&
252 	    TCPS_HAVEESTABLISHED(tp->t_state) &&
253 	    tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
254 		struct inpcb *inp = tp->t_inpcb;
255 		struct rtentry *rt = NULL;
256 
257 		/* No data to send means path mtu is not a problem */
258 		if (!inp->inp_socket->so_snd.sb_cc)
259 			goto leave;
260 
261 		rt = in_pcbrtentry(inp);
262 		/* Check if path MTU discovery is disabled already */
263 		if (rt && (rt->rt_flags & RTF_HOST) &&
264 		    (rt->rt_rmx.rmx_locks & RTV_MTU))
265 			goto leave;
266 
267 		rt = NULL;
268 		switch(tp->pf) {
269 #ifdef INET6
270 		case PF_INET6:
271 			/*
272 			 * We can not turn off path MTU for IPv6.
273 			 * Do nothing for now, maybe lower to
274 			 * minimum MTU.
275 			 */
276 			break;
277 #endif
278 		case PF_INET:
279 			rt = icmp_mtudisc_clone(inp->inp_faddr,
280 			    inp->inp_rtableid);
281 			break;
282 		}
283 		if (rt != NULL) {
284 			/* Disable path MTU discovery */
285 			if ((rt->rt_rmx.rmx_locks & RTV_MTU) == 0) {
286 				rt->rt_rmx.rmx_locks |= RTV_MTU;
287 				in_rtchange(inp, 0);
288 			}
289 
290 			rtfree(rt);
291 		}
292 	leave:
293 		;
294 	}
295 
296 	/*
297 	 * If losing, let the lower level know and try for
298 	 * a better route.  Also, if we backed off this far,
299 	 * our srtt estimate is probably bogus.  Clobber it
300 	 * so we'll take the next rtt measurement as our srtt;
301 	 * move the current srtt into rttvar to keep the current
302 	 * retransmit times until then.
303 	 */
304 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
305 		in_losing(tp->t_inpcb);
306 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
307 		tp->t_srtt = 0;
308 	}
309 	tp->snd_nxt = tp->snd_una;
310 #if defined(TCP_SACK)
311 	/*
312 	 * Note:  We overload snd_last to function also as the
313 	 * snd_last variable described in RFC 2582
314 	 */
315 	tp->snd_last = tp->snd_max;
316 #endif /* TCP_SACK */
317 	/*
318 	 * If timing a segment in this window, stop the timer.
319 	 */
320 	tp->t_rtttime = 0;
321 #ifdef TCP_ECN
322 	/*
323 	 * if ECN is enabled, there might be a broken firewall which
324 	 * blocks ecn packets.  fall back to non-ecn.
325 	 */
326 	if ((tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED)
327 	    && tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
328 		tp->t_flags |= TF_DISABLE_ECN;
329 #endif
330 	/*
331 	 * Close the congestion window down to one segment
332 	 * (we'll open it by one segment for each ack we get).
333 	 * Since we probably have a window's worth of unacked
334 	 * data accumulated, this "slow start" keeps us from
335 	 * dumping all that data as back-to-back packets (which
336 	 * might overwhelm an intermediate gateway).
337 	 *
338 	 * There are two phases to the opening: Initially we
339 	 * open by one mss on each ack.  This makes the window
340 	 * size increase exponentially with time.  If the
341 	 * window is larger than the path can handle, this
342 	 * exponential growth results in dropped packet(s)
343 	 * almost immediately.  To get more time between
344 	 * drops but still "push" the network to take advantage
345 	 * of improving conditions, we switch from exponential
346 	 * to linear window opening at some threshold size.
347 	 * For a threshold, we use half the current window
348 	 * size, truncated to a multiple of the mss.
349 	 *
350 	 * (the minimum cwnd that will give us exponential
351 	 * growth is 2 mss.  We don't allow the threshold
352 	 * to go below this.)
353 	 */
354 	{
355 		u_long win = ulmin(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
356 		if (win < 2)
357 			win = 2;
358 		tp->snd_cwnd = tp->t_maxseg;
359 		tp->snd_ssthresh = win * tp->t_maxseg;
360 		tp->t_dupacks = 0;
361 #ifdef TCP_ECN
362 		tp->snd_last = tp->snd_max;
363 		tp->t_flags |= TF_SEND_CWR;
364 #endif
365 #if 1 /* TCP_ECN */
366 		tcpstat.tcps_cwr_timeout++;
367 #endif
368 	}
369 	(void) tcp_output(tp);
370 
371  out:
372 	NET_UNLOCK(s);
373 }
374 
375 void
376 tcp_timer_persist(void *arg)
377 {
378 	struct tcpcb *tp = arg;
379 	uint32_t rto;
380 	int s;
381 
382 	NET_LOCK(s);
383 	if ((tp->t_flags & TF_DEAD) ||
384             TCP_TIMER_ISARMED(tp, TCPT_REXMT)) {
385 		goto out;
386 	}
387 	tcpstat.tcps_persisttimeo++;
388 	/*
389 	 * Hack: if the peer is dead/unreachable, we do not
390 	 * time out if the window is closed.  After a full
391 	 * backoff, drop the connection if the idle time
392 	 * (no responses to probes) reaches the maximum
393 	 * backoff that we would use if retransmitting.
394 	 */
395 	rto = TCP_REXMTVAL(tp);
396 	if (rto < tp->t_rttmin)
397 		rto = tp->t_rttmin;
398 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
399 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
400 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
401 		tcpstat.tcps_persistdrop++;
402 		tp = tcp_drop(tp, ETIMEDOUT);
403 		goto out;
404 	}
405 	tcp_setpersist(tp);
406 	tp->t_force = 1;
407 	(void) tcp_output(tp);
408 	tp->t_force = 0;
409  out:
410 	NET_UNLOCK(s);
411 }
412 
413 void
414 tcp_timer_keep(void *arg)
415 {
416 	struct tcpcb *tp = arg;
417 	int s;
418 
419 	NET_LOCK(s);
420 	if (tp->t_flags & TF_DEAD)
421 		goto out;
422 
423 	tcpstat.tcps_keeptimeo++;
424 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
425 		goto dropit;
426 	if ((tcp_always_keepalive ||
427 	    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
428 	    tp->t_state <= TCPS_CLOSING) {
429 		if ((tcp_maxidle > 0) &&
430 		    ((tcp_now - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle))
431 			goto dropit;
432 		/*
433 		 * Send a packet designed to force a response
434 		 * if the peer is up and reachable:
435 		 * either an ACK if the connection is still alive,
436 		 * or an RST if the peer has closed the connection
437 		 * due to timeout or reboot.
438 		 * Using sequence number tp->snd_una-1
439 		 * causes the transmitted zero-length segment
440 		 * to lie outside the receive window;
441 		 * by the protocol spec, this requires the
442 		 * correspondent TCP to respond.
443 		 */
444 		tcpstat.tcps_keepprobe++;
445 		tcp_respond(tp, mtod(tp->t_template, caddr_t),
446 		    NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0);
447 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
448 	} else
449 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
450  out:
451 	NET_UNLOCK(s);
452 	return;
453 
454  dropit:
455 	tcpstat.tcps_keepdrops++;
456 	tp = tcp_drop(tp, ETIMEDOUT);
457 	NET_UNLOCK(s);
458 }
459 
460 void
461 tcp_timer_2msl(void *arg)
462 {
463 	struct tcpcb *tp = arg;
464 	int s;
465 
466 	NET_LOCK(s);
467 	if (tp->t_flags & TF_DEAD)
468 		goto out;
469 
470 #ifdef TCP_SACK
471 	tcp_timer_freesack(tp);
472 #endif
473 
474 	if (tp->t_state != TCPS_TIME_WAIT &&
475 	    ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle)))
476 		TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
477 	else
478 		tp = tcp_close(tp);
479 
480  out:
481 	NET_UNLOCK(s);
482 }
483