2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
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 * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.5 2003/08/14 23:09:33 hsu Exp $
38 #include "opt_compat.h"
39 #include "opt_inet6.h"
40 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
51 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
53 #include <net/route.h>
55 #include <netinet/in.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/in_pcb.h>
59 #include <netinet6/in6_pcb.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/tcp.h>
63 #include <netinet/tcp_fsm.h>
64 #include <netinet/tcp_seq.h>
65 #include <netinet/tcp_timer.h>
66 #include <netinet/tcp_var.h>
67 #include <netinet/tcpip.h>
69 #include <netinet/tcp_debug.h>
73 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
77 tt = *(int *)oidp->oid_arg1;
78 s = (int)((int64_t)tt * 1000 / hz);
80 error = sysctl_handle_int(oidp, &s, 0, req);
81 if (error || !req->newptr)
84 tt = (int)((int64_t)s * hz / 1000);
88 *(int *)oidp->oid_arg1 = tt;
93 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
94 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
97 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
98 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
101 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
102 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
105 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
106 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
107 "Time before a delayed ACK is sent");
110 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
111 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
114 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
115 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
118 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
119 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
121 static int always_keepalive = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
123 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
125 static int tcp_keepcnt = TCPTV_KEEPCNT;
126 /* max idle probes */
127 int tcp_maxpersistidle;
128 /* max idle time in persist */
132 * Tcp protocol timeout routine called every 500 ms.
133 * Updates timestamps used for TCP
134 * causes finite state machine actions if timers expire.
143 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
149 * Cancel all timers for TCP tp.
155 callout_stop(tp->tt_2msl);
156 callout_stop(tp->tt_persist);
157 callout_stop(tp->tt_keep);
158 callout_stop(tp->tt_rexmt);
161 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
162 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
164 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
165 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
167 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
170 * TCP timer processing.
173 tcp_timer_delack(xtp)
176 struct tcpcb *tp = xtp;
180 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
184 callout_deactivate(tp->tt_delack);
186 tp->t_flags |= TF_ACKNOW;
187 tcpstat.tcps_delack++;
188 (void) tcp_output(tp);
196 struct tcpcb *tp = xtp;
201 ostate = tp->t_state;
204 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
208 callout_deactivate(tp->tt_2msl);
210 * 2 MSL timeout in shutdown went off. If we're closed but
211 * still waiting for peer to close and connection has been idle
212 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
213 * control block. Otherwise, check again in a bit.
215 if (tp->t_state != TCPS_TIME_WAIT &&
216 (ticks - tp->t_rcvtime) <= tcp_maxidle)
217 callout_reset(tp->tt_2msl, tcp_keepintvl,
223 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
224 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
234 struct tcpcb *tp = xtp;
235 struct tcptemp *t_template;
240 ostate = tp->t_state;
243 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
247 callout_deactivate(tp->tt_keep);
249 * Keep-alive timer went off; send something
250 * or drop connection if idle for too long.
252 tcpstat.tcps_keeptimeo++;
253 if (tp->t_state < TCPS_ESTABLISHED)
255 if ((always_keepalive ||
256 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
257 tp->t_state <= TCPS_CLOSING) {
258 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
261 * Send a packet designed to force a response
262 * if the peer is up and reachable:
263 * either an ACK if the connection is still alive,
264 * or an RST if the peer has closed the connection
265 * due to timeout or reboot.
266 * Using sequence number tp->snd_una-1
267 * causes the transmitted zero-length segment
268 * to lie outside the receive window;
269 * by the protocol spec, this requires the
270 * correspondent TCP to respond.
272 tcpstat.tcps_keepprobe++;
273 t_template = tcp_maketemplate(tp);
275 tcp_respond(tp, t_template->tt_ipgen,
276 &t_template->tt_t, (struct mbuf *)NULL,
277 tp->rcv_nxt, tp->snd_una - 1, 0);
278 (void) m_free(dtom(t_template));
280 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
282 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
285 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
286 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
293 tcpstat.tcps_keepdrops++;
294 tp = tcp_drop(tp, ETIMEDOUT);
297 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
298 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
305 tcp_timer_persist(xtp)
308 struct tcpcb *tp = xtp;
313 ostate = tp->t_state;
316 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
320 callout_deactivate(tp->tt_persist);
322 * Persistance timer into zero window.
323 * Force a byte to be output, if possible.
325 tcpstat.tcps_persisttimeo++;
327 * Hack: if the peer is dead/unreachable, we do not
328 * time out if the window is closed. After a full
329 * backoff, drop the connection if the idle time
330 * (no responses to probes) reaches the maximum
331 * backoff that we would use if retransmitting.
333 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
334 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
335 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
336 tcpstat.tcps_persistdrop++;
337 tp = tcp_drop(tp, ETIMEDOUT);
342 (void) tcp_output(tp);
347 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
348 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
355 tcp_save_congestion_state(struct tcpcb *tp)
357 tp->snd_cwnd_prev = tp->snd_cwnd;
358 tp->snd_ssthresh_prev = tp->snd_ssthresh;
359 tp->snd_recover_prev = tp->snd_recover;
360 if (IN_FASTRECOVERY(tp))
361 tp->t_flags |= TF_WASFRECOVERY;
363 tp->t_flags &= ~TF_WASFRECOVERY;
364 if (tp->t_flags & TF_RCVD_TSTMP) {
365 tp->t_rexmtTS = ticks;
366 tp->t_flags |= TF_FIRSTACCACK;
371 tcp_revert_congestion_state(struct tcpcb *tp)
373 tp->snd_cwnd = tp->snd_cwnd_prev;
374 tp->snd_ssthresh = tp->snd_ssthresh_prev;
375 tp->snd_recover = tp->snd_recover_prev;
376 if (tp->t_flags & TF_WASFRECOVERY)
377 ENTER_FASTRECOVERY(tp);
378 if (tp->t_flags & TF_FASTREXMT)
379 ++tcpstat.tcps_sndfastrexmitbad;
381 ++tcpstat.tcps_sndrtobad;
384 tp->snd_nxt = tp->snd_max;
391 struct tcpcb *tp = xtp;
397 ostate = tp->t_state;
400 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
404 callout_deactivate(tp->tt_rexmt);
406 * Retransmission timer went off. Message has not
407 * been acked within retransmit interval. Back off
408 * to a longer retransmit interval and retransmit one segment.
410 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
411 tp->t_rxtshift = TCP_MAXRXTSHIFT;
412 tcpstat.tcps_timeoutdrop++;
413 tp = tcp_drop(tp, tp->t_softerror ?
414 tp->t_softerror : ETIMEDOUT);
417 if (tp->t_rxtshift == 1) {
419 * first retransmit; record ssthresh and cwnd so they can
420 * be recovered if this turns out to be a "bad" retransmit.
421 * A retransmit is considered "bad" if an ACK for this
422 * segment is received within RTT/2 interval; the assumption
423 * here is that the ACK was already in flight. See
424 * "On Estimating End-to-End Network Path Properties" by
425 * Allman and Paxson for more details.
427 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
428 tcp_save_congestion_state(tp);
429 tp->t_flags &= ~TF_FASTREXMT;
431 tcpstat.tcps_rexmttimeo++;
432 if (tp->t_state == TCPS_SYN_SENT)
433 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
435 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
436 TCPT_RANGESET(tp->t_rxtcur, rexmt,
437 tp->t_rttmin, TCPTV_REXMTMAX);
439 * Disable rfc1323 and rfc1644 if we havn't got any response to
440 * our third SYN to work-around some broken terminal servers
441 * (most of which have hopefully been retired) that have bad VJ
442 * header compression code which trashes TCP segments containing
443 * unknown-to-them TCP options.
445 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
446 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
448 * If losing, let the lower level know and try for
449 * a better route. Also, if we backed off this far,
450 * our srtt estimate is probably bogus. Clobber it
451 * so we'll take the next rtt measurement as our srtt;
452 * move the current srtt into rttvar to keep the current
453 * retransmit times until then.
455 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
457 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
458 in6_losing(tp->t_inpcb);
461 in_losing(tp->t_inpcb);
462 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
465 tp->snd_nxt = tp->snd_una;
466 tp->snd_recover = tp->snd_max;
468 * Force a segment to be sent.
470 tp->t_flags |= TF_ACKNOW;
472 * If timing a segment in this window, stop the timer.
476 * Close the congestion window down to one segment
477 * (we'll open it by one segment for each ack we get).
478 * Since we probably have a window's worth of unacked
479 * data accumulated, this "slow start" keeps us from
480 * dumping all that data as back-to-back packets (which
481 * might overwhelm an intermediate gateway).
483 * There are two phases to the opening: Initially we
484 * open by one mss on each ack. This makes the window
485 * size increase exponentially with time. If the
486 * window is larger than the path can handle, this
487 * exponential growth results in dropped packet(s)
488 * almost immediately. To get more time between
489 * drops but still "push" the network to take advantage
490 * of improving conditions, we switch from exponential
491 * to linear window opening at some threshhold size.
492 * For a threshhold, we use half the current window
493 * size, truncated to a multiple of the mss.
495 * (the minimum cwnd that will give us exponential
496 * growth is 2 mss. We don't allow the threshhold
500 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
503 tp->snd_cwnd = tp->t_maxseg;
504 tp->snd_ssthresh = win * tp->t_maxseg;
507 EXIT_FASTRECOVERY(tp);
508 (void) tcp_output(tp);
512 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
513 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,