xref: /openbsd-src/sys/netinet/tcp_timer.c (revision ae3cb403620ab940fbaabb3055fac045a63d56b7)
1 /*	$OpenBSD: tcp_timer.c,v 1.60 2017/10/29 14:56:36 florian 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 
109 	/*
110 	 * If tcp_output() wasn't able to transmit the ACK
111 	 * for whatever reason, it will restart the delayed
112 	 * ACK callout.
113 	 */
114 	NET_LOCK();
115 	if (tp->t_flags & TF_DEAD)
116 		goto out;
117 	tp->t_flags |= TF_ACKNOW;
118 	(void) tcp_output(tp);
119  out:
120 	NET_UNLOCK();
121 }
122 
123 /*
124  * Tcp protocol timeout routine called every 500 ms.
125  * Updates the timers in all active tcb's and
126  * causes finite state machine actions if timers expire.
127  */
128 void
129 tcp_slowtimo(void)
130 {
131 	NET_LOCK();
132 
133 	tcp_maxidle = TCPTV_KEEPCNT * tcp_keepintvl;
134 	tcp_iss += TCP_ISSINCR2/PR_SLOWHZ;		/* increment iss */
135 	tcp_now++;					/* for timestamps */
136 
137 	NET_UNLOCK();
138 }
139 
140 /*
141  * Cancel all timers for TCP tp.
142  */
143 void
144 tcp_canceltimers(struct tcpcb *tp)
145 {
146 	int i;
147 
148 	for (i = 0; i < TCPT_NTIMERS; i++)
149 		TCP_TIMER_DISARM(tp, i);
150 }
151 
152 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
153     { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
154 
155 int tcp_totbackoff = 511;	/* sum of tcp_backoff[] */
156 
157 /*
158  * TCP timer processing.
159  */
160 
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 }
178 
179 void
180 tcp_timer_rexmt(void *arg)
181 {
182 	struct tcpcb *tp = arg;
183 	uint32_t rto;
184 
185 	NET_LOCK();
186 	if (tp->t_flags & TF_DEAD)
187 		goto out;
188 
189 	if ((tp->t_flags & TF_PMTUD_PEND) && tp->t_inpcb &&
190 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
191 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) {
192 		struct sockaddr_in sin;
193 		struct icmp icmp;
194 
195 		tp->t_flags &= ~TF_PMTUD_PEND;
196 
197 		/* XXX create fake icmp message with relevant entries */
198 		icmp.icmp_nextmtu = tp->t_pmtud_nextmtu;
199 		icmp.icmp_ip.ip_len = tp->t_pmtud_ip_len;
200 		icmp.icmp_ip.ip_hl = tp->t_pmtud_ip_hl;
201 		icmp.icmp_ip.ip_dst = tp->t_inpcb->inp_faddr;
202 		icmp_mtudisc(&icmp, tp->t_inpcb->inp_rtableid);
203 
204 		/*
205 		 * Notify all connections to the same peer about
206 		 * new mss and trigger retransmit.
207 		 */
208 		bzero(&sin, sizeof(sin));
209 		sin.sin_len = sizeof(sin);
210 		sin.sin_family = AF_INET;
211 		sin.sin_addr = tp->t_inpcb->inp_faddr;
212 		in_pcbnotifyall(&tcbtable, sintosa(&sin),
213 		    tp->t_inpcb->inp_rtableid, EMSGSIZE, tcp_mtudisc);
214 		goto out;
215 	}
216 
217 	tcp_timer_freesack(tp);
218 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
219 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
220 		tcpstat_inc(tcps_timeoutdrop);
221 		(void)tcp_drop(tp, tp->t_softerror ?
222 		    tp->t_softerror : ETIMEDOUT);
223 		goto out;
224 	}
225 	tcpstat_inc(tcps_rexmttimeo);
226 	rto = TCP_REXMTVAL(tp);
227 	if (rto < tp->t_rttmin)
228 		rto = tp->t_rttmin;
229 	TCPT_RANGESET(tp->t_rxtcur,
230 	    rto * tcp_backoff[tp->t_rxtshift],
231 	    tp->t_rttmin, TCPTV_REXMTMAX);
232 	TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
233 
234 	/*
235 	 * If we are losing and we are trying path MTU discovery,
236 	 * try turning it off.  This will avoid black holes in
237 	 * the network which suppress or fail to send "packet
238 	 * too big" ICMP messages.  We should ideally do
239 	 * lots more sophisticated searching to find the right
240 	 * value here...
241 	 */
242 	if (ip_mtudisc && tp->t_inpcb &&
243 	    TCPS_HAVEESTABLISHED(tp->t_state) &&
244 	    tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
245 		struct inpcb *inp = tp->t_inpcb;
246 		struct rtentry *rt = NULL;
247 
248 		/* No data to send means path mtu is not a problem */
249 		if (!inp->inp_socket->so_snd.sb_cc)
250 			goto leave;
251 
252 		rt = in_pcbrtentry(inp);
253 		/* Check if path MTU discovery is disabled already */
254 		if (rt && (rt->rt_flags & RTF_HOST) &&
255 		    (rt->rt_locks & RTV_MTU))
256 			goto leave;
257 
258 		rt = NULL;
259 		switch(tp->pf) {
260 #ifdef INET6
261 		case PF_INET6:
262 			/*
263 			 * We can not turn off path MTU for IPv6.
264 			 * Do nothing for now, maybe lower to
265 			 * minimum MTU.
266 			 */
267 			break;
268 #endif
269 		case PF_INET:
270 			rt = icmp_mtudisc_clone(inp->inp_faddr,
271 			    inp->inp_rtableid);
272 			break;
273 		}
274 		if (rt != NULL) {
275 			/* Disable path MTU discovery */
276 			if ((rt->rt_locks & RTV_MTU) == 0) {
277 				rt->rt_locks |= RTV_MTU;
278 				in_rtchange(inp, 0);
279 			}
280 
281 			rtfree(rt);
282 		}
283 	leave:
284 		;
285 	}
286 
287 	/*
288 	 * If losing, let the lower level know and try for
289 	 * a better route.  Also, if we backed off this far,
290 	 * our srtt estimate is probably bogus.  Clobber it
291 	 * so we'll take the next rtt measurement as our srtt;
292 	 * move the current srtt into rttvar to keep the current
293 	 * retransmit times until then.
294 	 */
295 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
296 		in_losing(tp->t_inpcb);
297 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
298 		tp->t_srtt = 0;
299 	}
300 	tp->snd_nxt = tp->snd_una;
301 	/*
302 	 * Note:  We overload snd_last to function also as the
303 	 * snd_last variable described in RFC 2582
304 	 */
305 	tp->snd_last = tp->snd_max;
306 	/*
307 	 * If timing a segment in this window, stop the timer.
308 	 */
309 	tp->t_rtttime = 0;
310 #ifdef TCP_ECN
311 	/*
312 	 * if ECN is enabled, there might be a broken firewall which
313 	 * blocks ecn packets.  fall back to non-ecn.
314 	 */
315 	if ((tp->t_state == TCPS_SYN_SENT || tp->t_state == TCPS_SYN_RECEIVED)
316 	    && tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN))
317 		tp->t_flags |= TF_DISABLE_ECN;
318 #endif
319 	/*
320 	 * Close the congestion window down to one segment
321 	 * (we'll open it by one segment for each ack we get).
322 	 * Since we probably have a window's worth of unacked
323 	 * data accumulated, this "slow start" keeps us from
324 	 * dumping all that data as back-to-back packets (which
325 	 * might overwhelm an intermediate gateway).
326 	 *
327 	 * There are two phases to the opening: Initially we
328 	 * open by one mss on each ack.  This makes the window
329 	 * size increase exponentially with time.  If the
330 	 * window is larger than the path can handle, this
331 	 * exponential growth results in dropped packet(s)
332 	 * almost immediately.  To get more time between
333 	 * drops but still "push" the network to take advantage
334 	 * of improving conditions, we switch from exponential
335 	 * to linear window opening at some threshold size.
336 	 * For a threshold, we use half the current window
337 	 * size, truncated to a multiple of the mss.
338 	 *
339 	 * (the minimum cwnd that will give us exponential
340 	 * growth is 2 mss.  We don't allow the threshold
341 	 * to go below this.)
342 	 */
343 	{
344 		u_long win = ulmin(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
345 		if (win < 2)
346 			win = 2;
347 		tp->snd_cwnd = tp->t_maxseg;
348 		tp->snd_ssthresh = win * tp->t_maxseg;
349 		tp->t_dupacks = 0;
350 #ifdef TCP_ECN
351 		tp->snd_last = tp->snd_max;
352 		tp->t_flags |= TF_SEND_CWR;
353 #endif
354 #if 1 /* TCP_ECN */
355 		tcpstat_inc(tcps_cwr_timeout);
356 #endif
357 	}
358 	(void) tcp_output(tp);
359 
360  out:
361 	NET_UNLOCK();
362 }
363 
364 void
365 tcp_timer_persist(void *arg)
366 {
367 	struct tcpcb *tp = arg;
368 	uint32_t rto;
369 
370 	NET_LOCK();
371 	if ((tp->t_flags & TF_DEAD) ||
372             TCP_TIMER_ISARMED(tp, TCPT_REXMT)) {
373 		goto out;
374 	}
375 	tcpstat_inc(tcps_persisttimeo);
376 	/*
377 	 * Hack: if the peer is dead/unreachable, we do not
378 	 * time out if the window is closed.  After a full
379 	 * backoff, drop the connection if the idle time
380 	 * (no responses to probes) reaches the maximum
381 	 * backoff that we would use if retransmitting.
382 	 */
383 	rto = TCP_REXMTVAL(tp);
384 	if (rto < tp->t_rttmin)
385 		rto = tp->t_rttmin;
386 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
387 	    ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
388 	    (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
389 		tcpstat_inc(tcps_persistdrop);
390 		tp = tcp_drop(tp, ETIMEDOUT);
391 		goto out;
392 	}
393 	tcp_setpersist(tp);
394 	tp->t_force = 1;
395 	(void) tcp_output(tp);
396 	tp->t_force = 0;
397  out:
398 	NET_UNLOCK();
399 }
400 
401 void
402 tcp_timer_keep(void *arg)
403 {
404 	struct tcpcb *tp = arg;
405 
406 	NET_LOCK();
407 	if (tp->t_flags & TF_DEAD)
408 		goto out;
409 
410 	tcpstat_inc(tcps_keeptimeo);
411 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
412 		goto dropit;
413 	if ((tcp_always_keepalive ||
414 	    tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
415 	    tp->t_state <= TCPS_CLOSING) {
416 		if ((tcp_maxidle > 0) &&
417 		    ((tcp_now - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle))
418 			goto dropit;
419 		/*
420 		 * Send a packet designed to force a response
421 		 * if the peer is up and reachable:
422 		 * either an ACK if the connection is still alive,
423 		 * or an RST if the peer has closed the connection
424 		 * due to timeout or reboot.
425 		 * Using sequence number tp->snd_una-1
426 		 * causes the transmitted zero-length segment
427 		 * to lie outside the receive window;
428 		 * by the protocol spec, this requires the
429 		 * correspondent TCP to respond.
430 		 */
431 		tcpstat_inc(tcps_keepprobe);
432 		tcp_respond(tp, mtod(tp->t_template, caddr_t),
433 		    NULL, tp->rcv_nxt, tp->snd_una - 1, 0, 0);
434 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
435 	} else
436 		TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
437  out:
438 	NET_UNLOCK();
439 	return;
440 
441  dropit:
442 	tcpstat_inc(tcps_keepdrops);
443 	tp = tcp_drop(tp, ETIMEDOUT);
444 	NET_UNLOCK();
445 }
446 
447 void
448 tcp_timer_2msl(void *arg)
449 {
450 	struct tcpcb *tp = arg;
451 
452 	NET_LOCK();
453 	if (tp->t_flags & TF_DEAD)
454 		goto out;
455 
456 	tcp_timer_freesack(tp);
457 
458 	if (tp->t_state != TCPS_TIME_WAIT &&
459 	    ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle)))
460 		TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
461 	else
462 		tp = tcp_close(tp);
463 
464  out:
465 	NET_UNLOCK();
466 }
467