tcp_timer.c revision 1.67 1 /* $NetBSD: tcp_timer.c,v 1.67 2005/01/26 21:49:27 mycroft Exp $ */
2
3 /*
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 /*-
33 * Copyright (c) 1997, 1998, 2001 The NetBSD Foundation, Inc.
34 * All rights reserved.
35 *
36 * This code is derived from software contributed to The NetBSD Foundation
37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38 * Facility, NASA Ames Research Center.
39 *
40 * Redistribution and use in source and binary forms, with or without
41 * modification, are permitted provided that the following conditions
42 * are met:
43 * 1. Redistributions of source code must retain the above copyright
44 * notice, this list of conditions and the following disclaimer.
45 * 2. Redistributions in binary form must reproduce the above copyright
46 * notice, this list of conditions and the following disclaimer in the
47 * documentation and/or other materials provided with the distribution.
48 * 3. All advertising materials mentioning features or use of this software
49 * must display the following acknowledgement:
50 * This product includes software developed by the NetBSD
51 * Foundation, Inc. and its contributors.
52 * 4. Neither the name of The NetBSD Foundation nor the names of its
53 * contributors may be used to endorse or promote products derived
54 * from this software without specific prior written permission.
55 *
56 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
57 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
58 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
59 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
60 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
61 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
62 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
63 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
64 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
65 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
66 * POSSIBILITY OF SUCH DAMAGE.
67 */
68
69 /*
70 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
71 * The Regents of the University of California. All rights reserved.
72 *
73 * Redistribution and use in source and binary forms, with or without
74 * modification, are permitted provided that the following conditions
75 * are met:
76 * 1. Redistributions of source code must retain the above copyright
77 * notice, this list of conditions and the following disclaimer.
78 * 2. Redistributions in binary form must reproduce the above copyright
79 * notice, this list of conditions and the following disclaimer in the
80 * documentation and/or other materials provided with the distribution.
81 * 3. Neither the name of the University nor the names of its contributors
82 * may be used to endorse or promote products derived from this software
83 * without specific prior written permission.
84 *
85 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
86 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
87 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
88 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
89 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
90 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
91 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
92 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
93 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
94 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
95 * SUCH DAMAGE.
96 *
97 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
98 */
99
100 #include <sys/cdefs.h>
101 __KERNEL_RCSID(0, "$NetBSD: tcp_timer.c,v 1.67 2005/01/26 21:49:27 mycroft Exp $");
102
103 #include "opt_inet.h"
104 #include "opt_tcp_debug.h"
105
106 #include <sys/param.h>
107 #include <sys/systm.h>
108 #include <sys/malloc.h>
109 #include <sys/mbuf.h>
110 #include <sys/socket.h>
111 #include <sys/socketvar.h>
112 #include <sys/protosw.h>
113 #include <sys/errno.h>
114 #include <sys/kernel.h>
115
116 #include <net/if.h>
117 #include <net/route.h>
118
119 #include <netinet/in.h>
120 #include <netinet/in_systm.h>
121 #include <netinet/ip.h>
122 #include <netinet/in_pcb.h>
123 #include <netinet/ip_var.h>
124
125 #ifdef INET6
126 #ifndef INET
127 #include <netinet/in.h>
128 #endif
129 #include <netinet/ip6.h>
130 #include <netinet6/in6_pcb.h>
131 #endif
132
133 #include <netinet/tcp.h>
134 #include <netinet/tcp_fsm.h>
135 #include <netinet/tcp_seq.h>
136 #include <netinet/tcp_timer.h>
137 #include <netinet/tcp_var.h>
138 #include <netinet/tcpip.h>
139 #ifdef TCP_DEBUG
140 #include <netinet/tcp_debug.h>
141 #endif
142
143 /*
144 * Various tunable timer parameters. These are initialized in tcp_init(),
145 * unless they are patched.
146 */
147 int tcp_keepidle = 0;
148 int tcp_keepintvl = 0;
149 int tcp_keepcnt = 0; /* max idle probes */
150 int tcp_maxpersistidle = 0; /* max idle time in persist */
151 int tcp_maxidle; /* computed in tcp_slowtimo() */
152
153 /*
154 * Time to delay the ACK. This is initialized in tcp_init(), unless
155 * its patched.
156 */
157 int tcp_delack_ticks = 0;
158
159 void tcp_timer_rexmt(void *);
160 void tcp_timer_persist(void *);
161 void tcp_timer_keep(void *);
162 void tcp_timer_2msl(void *);
163
164 const tcp_timer_func_t tcp_timer_funcs[TCPT_NTIMERS] = {
165 tcp_timer_rexmt,
166 tcp_timer_persist,
167 tcp_timer_keep,
168 tcp_timer_2msl,
169 };
170
171 /*
172 * Timer state initialization, called from tcp_init().
173 */
174 void
175 tcp_timer_init(void)
176 {
177
178 if (tcp_keepidle == 0)
179 tcp_keepidle = TCPTV_KEEP_IDLE;
180
181 if (tcp_keepintvl == 0)
182 tcp_keepintvl = TCPTV_KEEPINTVL;
183
184 if (tcp_keepcnt == 0)
185 tcp_keepcnt = TCPTV_KEEPCNT;
186
187 if (tcp_maxpersistidle == 0)
188 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
189
190 if (tcp_delack_ticks == 0)
191 tcp_delack_ticks = TCP_DELACK_TICKS;
192 }
193
194 /*
195 * Return how many timers are currently being invoked.
196 */
197 int
198 tcp_timers_invoking(struct tcpcb *tp)
199 {
200 int i;
201 int count = 0;
202
203 for (i = 0; i < TCPT_NTIMERS; i++)
204 if (callout_invoking(&tp->t_timer[i]))
205 count++;
206 if (callout_invoking(&tp->t_delack_ch))
207 count++;
208
209 return count;
210 }
211
212 /*
213 * Callout to process delayed ACKs for a TCPCB.
214 */
215 void
216 tcp_delack(void *arg)
217 {
218 struct tcpcb *tp = arg;
219 int s;
220
221 /*
222 * If tcp_output() wasn't able to transmit the ACK
223 * for whatever reason, it will restart the delayed
224 * ACK callout.
225 */
226
227 s = splsoftnet();
228 callout_ack(&tp->t_delack_ch);
229 if (tcp_isdead(tp)) {
230 splx(s);
231 return;
232 }
233
234 tp->t_flags |= TF_ACKNOW;
235 (void) tcp_output(tp);
236 splx(s);
237 }
238
239 /*
240 * Tcp protocol timeout routine called every 500 ms.
241 * Updates the timers in all active tcb's and
242 * causes finite state machine actions if timers expire.
243 */
244 void
245 tcp_slowtimo()
246 {
247 int s;
248
249 s = splsoftnet();
250 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
251 tcp_iss_seq += TCP_ISSINCR; /* increment iss */
252 tcp_now++; /* for timestamps */
253 splx(s);
254 }
255
256 /*
257 * Cancel all timers for TCP tp.
258 */
259 void
260 tcp_canceltimers(tp)
261 struct tcpcb *tp;
262 {
263 int i;
264
265 for (i = 0; i < TCPT_NTIMERS; i++)
266 TCP_TIMER_DISARM(tp, i);
267 }
268
269 const int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
270 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
271
272 const int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
273
274 /*
275 * TCP timer processing.
276 */
277
278 void
279 tcp_timer_rexmt(void *arg)
280 {
281 struct tcpcb *tp = arg;
282 uint32_t rto;
283 int s;
284 #ifdef TCP_DEBUG
285 struct socket *so = NULL;
286 short ostate;
287 #endif
288
289 s = splsoftnet();
290 callout_ack(&tp->t_timer[TCPT_REXMT]);
291 if (tcp_isdead(tp)) {
292 splx(s);
293 return;
294 }
295
296 #ifdef TCP_DEBUG
297 #ifdef INET
298 if (tp->t_inpcb)
299 so = tp->t_inpcb->inp_socket;
300 #endif
301 #ifdef INET6
302 if (tp->t_in6pcb)
303 so = tp->t_in6pcb->in6p_socket;
304 #endif
305 ostate = tp->t_state;
306 #endif /* TCP_DEBUG */
307
308 /*
309 * Retransmission timer went off. Message has not
310 * been acked within retransmit interval. Back off
311 * to a longer retransmit interval and retransmit one segment.
312 */
313
314 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
315 tp->t_rxtshift = TCP_MAXRXTSHIFT;
316 tcpstat.tcps_timeoutdrop++;
317 tp = tcp_drop(tp, tp->t_softerror ?
318 tp->t_softerror : ETIMEDOUT);
319 goto out;
320 }
321 tcpstat.tcps_rexmttimeo++;
322 rto = TCP_REXMTVAL(tp);
323 if (rto < tp->t_rttmin)
324 rto = tp->t_rttmin;
325 TCPT_RANGESET(tp->t_rxtcur, rto * tcp_backoff[tp->t_rxtshift],
326 tp->t_rttmin, TCPTV_REXMTMAX);
327 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur);
328
329 /*
330 * If we are losing and we are trying path MTU discovery,
331 * try turning it off. This will avoid black holes in
332 * the network which suppress or fail to send "packet
333 * too big" ICMP messages. We should ideally do
334 * lots more sophisticated searching to find the right
335 * value here...
336 */
337 if (tp->t_mtudisc && tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
338 tcpstat.tcps_pmtublackhole++;
339
340 #ifdef INET
341 /* try turning PMTUD off */
342 if (tp->t_inpcb)
343 tp->t_mtudisc = 0;
344 #endif
345 #ifdef INET6
346 /* try using IPv6 minimum MTU */
347 if (tp->t_in6pcb)
348 tp->t_mtudisc = 0;
349 #endif
350
351 /* XXX: more sophisticated Black hole recovery code? */
352 }
353
354 /*
355 * If losing, let the lower level know and try for
356 * a better route. Also, if we backed off this far,
357 * our srtt estimate is probably bogus. Clobber it
358 * so we'll take the next rtt measurement as our srtt;
359 * move the current srtt into rttvar to keep the current
360 * retransmit times until then.
361 */
362 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
363 #ifdef INET
364 if (tp->t_inpcb)
365 in_losing(tp->t_inpcb);
366 #endif
367 #ifdef INET6
368 if (tp->t_in6pcb)
369 in6_losing(tp->t_in6pcb);
370 #endif
371 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
372 tp->t_srtt = 0;
373 }
374 tp->snd_nxt = tp->snd_una;
375 tp->snd_high = tp->snd_max;
376 /*
377 * If timing a segment in this window, stop the timer.
378 */
379 tp->t_rtttime = 0;
380 /*
381 * Remember if we are retransmitting a SYN, because if
382 * we do, set the initial congestion window must be set
383 * to 1 segment.
384 */
385 if (tp->t_state == TCPS_SYN_SENT)
386 tp->t_flags |= TF_SYN_REXMT;
387 /*
388 * Close the congestion window down to one segment
389 * (we'll open it by one segment for each ack we get).
390 * Since we probably have a window's worth of unacked
391 * data accumulated, this "slow start" keeps us from
392 * dumping all that data as back-to-back packets (which
393 * might overwhelm an intermediate gateway).
394 *
395 * There are two phases to the opening: Initially we
396 * open by one mss on each ack. This makes the window
397 * size increase exponentially with time. If the
398 * window is larger than the path can handle, this
399 * exponential growth results in dropped packet(s)
400 * almost immediately. To get more time between
401 * drops but still "push" the network to take advantage
402 * of improving conditions, we switch from exponential
403 * to linear window opening at some threshhold size.
404 * For a threshhold, we use half the current window
405 * size, truncated to a multiple of the mss.
406 *
407 * (the minimum cwnd that will give us exponential
408 * growth is 2 mss. We don't allow the threshhold
409 * to go below this.)
410 */
411 {
412 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_segsz;
413 if (win < 2)
414 win = 2;
415 /* Loss Window MUST be one segment. */
416 tp->snd_cwnd = tp->t_segsz;
417 tp->snd_ssthresh = win * tp->t_segsz;
418 tp->t_dupacks = 0;
419 }
420 (void) tcp_output(tp);
421
422 out:
423 #ifdef TCP_DEBUG
424 if (tp && so->so_options & SO_DEBUG)
425 tcp_trace(TA_USER, ostate, tp, NULL,
426 PRU_SLOWTIMO | (TCPT_REXMT << 8));
427 #endif
428 splx(s);
429 }
430
431 void
432 tcp_timer_persist(void *arg)
433 {
434 struct tcpcb *tp = arg;
435 uint32_t rto;
436 int s;
437 #ifdef TCP_DEBUG
438 struct socket *so = NULL;
439 short ostate;
440 #endif
441
442 s = splsoftnet();
443 callout_ack(&tp->t_timer[TCPT_PERSIST]);
444 if (tcp_isdead(tp)) {
445 splx(s);
446 return;
447 }
448
449 #ifdef TCP_DEBUG
450 #ifdef INET
451 if (tp->t_inpcb)
452 so = tp->t_inpcb->inp_socket;
453 #endif
454 #ifdef INET6
455 if (tp->t_in6pcb)
456 so = tp->t_in6pcb->in6p_socket;
457 #endif
458
459 ostate = tp->t_state;
460 #endif /* TCP_DEBUG */
461
462 /*
463 * Persistance timer into zero window.
464 * Force a byte to be output, if possible.
465 */
466
467 /*
468 * Hack: if the peer is dead/unreachable, we do not
469 * time out if the window is closed. After a full
470 * backoff, drop the connection if the idle time
471 * (no responses to probes) reaches the maximum
472 * backoff that we would use if retransmitting.
473 */
474 rto = TCP_REXMTVAL(tp);
475 if (rto < tp->t_rttmin)
476 rto = tp->t_rttmin;
477 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
478 ((tcp_now - tp->t_rcvtime) >= tcp_maxpersistidle ||
479 (tcp_now - tp->t_rcvtime) >= rto * tcp_totbackoff)) {
480 tcpstat.tcps_persistdrops++;
481 tp = tcp_drop(tp, ETIMEDOUT);
482 goto out;
483 }
484 tcpstat.tcps_persisttimeo++;
485 tcp_setpersist(tp);
486 tp->t_force = 1;
487 (void) tcp_output(tp);
488 tp->t_force = 0;
489
490 out:
491 #ifdef TCP_DEBUG
492 if (tp && so->so_options & SO_DEBUG)
493 tcp_trace(TA_USER, ostate, tp, NULL,
494 PRU_SLOWTIMO | (TCPT_PERSIST << 8));
495 #endif
496 splx(s);
497 }
498
499 void
500 tcp_timer_keep(void *arg)
501 {
502 struct tcpcb *tp = arg;
503 struct socket *so = NULL; /* Quell compiler warning */
504 int s;
505 #ifdef TCP_DEBUG
506 short ostate;
507 #endif
508
509 s = splsoftnet();
510 callout_ack(&tp->t_timer[TCPT_KEEP]);
511 if (tcp_isdead(tp)) {
512 splx(s);
513 return;
514 }
515
516 #ifdef TCP_DEBUG
517 ostate = tp->t_state;
518 #endif /* TCP_DEBUG */
519
520 /*
521 * Keep-alive timer went off; send something
522 * or drop connection if idle for too long.
523 */
524
525 tcpstat.tcps_keeptimeo++;
526 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
527 goto dropit;
528 #ifdef INET
529 if (tp->t_inpcb)
530 so = tp->t_inpcb->inp_socket;
531 #endif
532 #ifdef INET6
533 if (tp->t_in6pcb)
534 so = tp->t_in6pcb->in6p_socket;
535 #endif
536 if (so->so_options & SO_KEEPALIVE &&
537 tp->t_state <= TCPS_CLOSE_WAIT) {
538 if ((tcp_maxidle > 0) &&
539 ((tcp_now - tp->t_rcvtime) >=
540 tcp_keepidle + tcp_maxidle))
541 goto dropit;
542 /*
543 * Send a packet designed to force a response
544 * if the peer is up and reachable:
545 * either an ACK if the connection is still alive,
546 * or an RST if the peer has closed the connection
547 * due to timeout or reboot.
548 * Using sequence number tp->snd_una-1
549 * causes the transmitted zero-length segment
550 * to lie outside the receive window;
551 * by the protocol spec, this requires the
552 * correspondent TCP to respond.
553 */
554 tcpstat.tcps_keepprobe++;
555 if (tcp_compat_42) {
556 /*
557 * The keepalive packet must have nonzero
558 * length to get a 4.2 host to respond.
559 */
560 (void)tcp_respond(tp, tp->t_template,
561 (struct mbuf *)NULL, NULL, tp->rcv_nxt - 1,
562 tp->snd_una - 1, 0);
563 } else {
564 (void)tcp_respond(tp, tp->t_template,
565 (struct mbuf *)NULL, NULL, tp->rcv_nxt,
566 tp->snd_una - 1, 0);
567 }
568 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepintvl);
569 } else
570 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle);
571
572 #ifdef TCP_DEBUG
573 if (tp && so->so_options & SO_DEBUG)
574 tcp_trace(TA_USER, ostate, tp, NULL,
575 PRU_SLOWTIMO | (TCPT_KEEP << 8));
576 #endif
577 splx(s);
578 return;
579
580 dropit:
581 tcpstat.tcps_keepdrops++;
582 (void) tcp_drop(tp, ETIMEDOUT);
583 splx(s);
584 }
585
586 void
587 tcp_timer_2msl(void *arg)
588 {
589 struct tcpcb *tp = arg;
590 int s;
591 #ifdef TCP_DEBUG
592 struct socket *so = NULL;
593 short ostate;
594 #endif
595
596 s = splsoftnet();
597 callout_ack(&tp->t_timer[TCPT_2MSL]);
598 if (tcp_isdead(tp)) {
599 splx(s);
600 return;
601 }
602
603 #ifdef TCP_DEBUG
604 #ifdef INET
605 if (tp->t_inpcb)
606 so = tp->t_inpcb->inp_socket;
607 #endif
608 #ifdef INET6
609 if (tp->t_in6pcb)
610 so = tp->t_in6pcb->in6p_socket;
611 #endif
612
613 ostate = tp->t_state;
614 #endif /* TCP_DEBUG */
615
616 /*
617 * 2 MSL timeout in shutdown went off. If we're closed but
618 * still waiting for peer to close and connection has been idle
619 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
620 * control block. Otherwise, check again in a bit.
621 */
622 if (tp->t_state != TCPS_TIME_WAIT &&
623 ((tcp_maxidle == 0) || ((tcp_now - tp->t_rcvtime) <= tcp_maxidle)))
624 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_keepintvl);
625 else
626 tp = tcp_close(tp);
627
628 #ifdef TCP_DEBUG
629 if (tp && so->so_options & SO_DEBUG)
630 tcp_trace(TA_USER, ostate, tp, NULL,
631 PRU_SLOWTIMO | (TCPT_2MSL << 8));
632 #endif
633 splx(s);
634 }
635