2 * Copyright (c) 2003-2004 Jeffrey Hsu. All rights reserved.
3 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
4 * The Regents of the University of California. All rights reserved.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. All advertising materials mentioning features or use of this software
15 * must display the following acknowledgement:
16 * This product includes software developed by the University of
17 * California, Berkeley and its contributors.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
35 * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
36 * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.7 2004/04/07 17:01:25 dillon Exp $
39 #include "opt_compat.h"
40 #include "opt_inet6.h"
41 #include "opt_tcpdebug.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/thread.h>
52 #include <sys/globaldata.h>
54 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
56 #include <net/route.h>
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/in_pcb.h>
62 #include <netinet6/in6_pcb.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcpip.h>
72 #include <netinet/tcp_debug.h>
76 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
80 tt = *(int *)oidp->oid_arg1;
81 s = (int)((int64_t)tt * 1000 / hz);
83 error = sysctl_handle_int(oidp, &s, 0, req);
84 if (error || !req->newptr)
87 tt = (int)((int64_t)s * hz / 1000);
91 *(int *)oidp->oid_arg1 = tt;
96 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
97 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
100 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
101 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
104 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
105 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
108 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
109 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
110 "Time before a delayed ACK is sent");
113 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
114 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
117 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
118 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
121 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
122 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
124 static int always_keepalive = 0;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
126 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
128 static int tcp_keepcnt = TCPTV_KEEPCNT;
129 /* max idle probes */
130 int tcp_maxpersistidle;
131 /* max idle time in persist */
135 * Tcp protocol timeout routine called every 500 ms.
136 * Updates timestamps used for TCP
137 * causes finite state machine actions if timers expire.
146 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
152 * Cancel all timers for TCP tp.
158 callout_stop(tp->tt_2msl);
159 callout_stop(tp->tt_persist);
160 callout_stop(tp->tt_keep);
161 callout_stop(tp->tt_rexmt);
164 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
165 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
167 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
168 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
170 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
173 * TCP timer processing.
176 tcp_timer_delack(xtp)
179 struct tcpcb *tp = xtp;
183 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
187 callout_deactivate(tp->tt_delack);
189 tp->t_flags |= TF_ACKNOW;
190 tcpstat.tcps_delack++;
191 (void) tcp_output(tp);
199 struct tcpcb *tp = xtp;
204 ostate = tp->t_state;
207 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
211 callout_deactivate(tp->tt_2msl);
213 * 2 MSL timeout in shutdown went off. If we're closed but
214 * still waiting for peer to close and connection has been idle
215 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
216 * control block. Otherwise, check again in a bit.
218 if (tp->t_state != TCPS_TIME_WAIT &&
219 (ticks - tp->t_rcvtime) <= tcp_maxidle)
220 callout_reset(tp->tt_2msl, tcp_keepintvl,
226 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
227 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
237 struct tcpcb *tp = xtp;
238 struct tcptemp *t_template;
243 ostate = tp->t_state;
246 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
250 callout_deactivate(tp->tt_keep);
252 * Keep-alive timer went off; send something
253 * or drop connection if idle for too long.
255 tcpstat.tcps_keeptimeo++;
256 if (tp->t_state < TCPS_ESTABLISHED)
258 if ((always_keepalive ||
259 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
260 tp->t_state <= TCPS_CLOSING) {
261 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
264 * Send a packet designed to force a response
265 * if the peer is up and reachable:
266 * either an ACK if the connection is still alive,
267 * or an RST if the peer has closed the connection
268 * due to timeout or reboot.
269 * Using sequence number tp->snd_una-1
270 * causes the transmitted zero-length segment
271 * to lie outside the receive window;
272 * by the protocol spec, this requires the
273 * correspondent TCP to respond.
275 tcpstat.tcps_keepprobe++;
276 t_template = tcp_maketemplate(tp);
278 tcp_respond(tp, t_template->tt_ipgen,
279 &t_template->tt_t, (struct mbuf *)NULL,
280 tp->rcv_nxt, tp->snd_una - 1, 0);
281 (void) m_free(dtom(t_template));
283 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
285 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
288 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
289 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
296 tcpstat.tcps_keepdrops++;
297 tp = tcp_drop(tp, ETIMEDOUT);
300 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
301 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
308 tcp_timer_persist(xtp)
311 struct tcpcb *tp = xtp;
316 ostate = tp->t_state;
319 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
323 callout_deactivate(tp->tt_persist);
325 * Persistance timer into zero window.
326 * Force a byte to be output, if possible.
328 tcpstat.tcps_persisttimeo++;
330 * Hack: if the peer is dead/unreachable, we do not
331 * time out if the window is closed. After a full
332 * backoff, drop the connection if the idle time
333 * (no responses to probes) reaches the maximum
334 * backoff that we would use if retransmitting.
336 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
337 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
338 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
339 tcpstat.tcps_persistdrop++;
340 tp = tcp_drop(tp, ETIMEDOUT);
345 (void) tcp_output(tp);
350 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
351 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
358 tcp_save_congestion_state(struct tcpcb *tp)
360 tp->snd_cwnd_prev = tp->snd_cwnd;
361 tp->snd_ssthresh_prev = tp->snd_ssthresh;
362 tp->snd_recover_prev = tp->snd_recover;
363 if (IN_FASTRECOVERY(tp))
364 tp->t_flags |= TF_WASFRECOVERY;
366 tp->t_flags &= ~TF_WASFRECOVERY;
367 if (tp->t_flags & TF_RCVD_TSTMP) {
368 tp->t_rexmtTS = ticks;
369 tp->t_flags |= TF_FIRSTACCACK;
374 tcp_revert_congestion_state(struct tcpcb *tp)
376 tp->snd_cwnd = tp->snd_cwnd_prev;
377 tp->snd_ssthresh = tp->snd_ssthresh_prev;
378 tp->snd_recover = tp->snd_recover_prev;
379 if (tp->t_flags & TF_WASFRECOVERY)
380 ENTER_FASTRECOVERY(tp);
381 if (tp->t_flags & TF_FASTREXMT) {
382 ++tcpstat.tcps_sndfastrexmitbad;
383 if (tp->t_flags & TF_EARLYREXMT)
384 ++tcpstat.tcps_sndearlyrexmitbad;
386 ++tcpstat.tcps_sndrtobad;
389 tp->snd_nxt = tp->snd_max;
396 struct tcpcb *tp = xtp;
402 ostate = tp->t_state;
405 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
409 callout_deactivate(tp->tt_rexmt);
411 * Retransmission timer went off. Message has not
412 * been acked within retransmit interval. Back off
413 * to a longer retransmit interval and retransmit one segment.
415 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
416 tp->t_rxtshift = TCP_MAXRXTSHIFT;
417 tcpstat.tcps_timeoutdrop++;
418 tp = tcp_drop(tp, tp->t_softerror ?
419 tp->t_softerror : ETIMEDOUT);
422 if (tp->t_rxtshift == 1) {
424 * first retransmit; record ssthresh and cwnd so they can
425 * be recovered if this turns out to be a "bad" retransmit.
426 * A retransmit is considered "bad" if an ACK for this
427 * segment is received within RTT/2 interval; the assumption
428 * here is that the ACK was already in flight. See
429 * "On Estimating End-to-End Network Path Properties" by
430 * Allman and Paxson for more details.
432 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
433 tcp_save_congestion_state(tp);
434 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
436 tcpstat.tcps_rexmttimeo++;
437 if (tp->t_state == TCPS_SYN_SENT)
438 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
440 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
441 TCPT_RANGESET(tp->t_rxtcur, rexmt,
442 tp->t_rttmin, TCPTV_REXMTMAX);
444 * Disable rfc1323 and rfc1644 if we havn't got any response to
445 * our third SYN to work-around some broken terminal servers
446 * (most of which have hopefully been retired) that have bad VJ
447 * header compression code which trashes TCP segments containing
448 * unknown-to-them TCP options.
450 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
451 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
453 * If losing, let the lower level know and try for
454 * a better route. Also, if we backed off this far,
455 * our srtt estimate is probably bogus. Clobber it
456 * so we'll take the next rtt measurement as our srtt;
457 * move the current srtt into rttvar to keep the current
458 * retransmit times until then.
460 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
462 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
463 in6_losing(tp->t_inpcb);
466 in_losing(tp->t_inpcb);
467 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
470 tp->snd_nxt = tp->snd_una;
471 tp->snd_recover = tp->snd_max;
473 * Force a segment to be sent.
475 tp->t_flags |= TF_ACKNOW;
477 * If timing a segment in this window, stop the timer.
481 * Close the congestion window down to one segment
482 * (we'll open it by one segment for each ack we get).
483 * Since we probably have a window's worth of unacked
484 * data accumulated, this "slow start" keeps us from
485 * dumping all that data as back-to-back packets (which
486 * might overwhelm an intermediate gateway).
488 * There are two phases to the opening: Initially we
489 * open by one mss on each ack. This makes the window
490 * size increase exponentially with time. If the
491 * window is larger than the path can handle, this
492 * exponential growth results in dropped packet(s)
493 * almost immediately. To get more time between
494 * drops but still "push" the network to take advantage
495 * of improving conditions, we switch from exponential
496 * to linear window opening at some threshhold size.
497 * For a threshhold, we use half the current window
498 * size, truncated to a multiple of the mss.
500 * (the minimum cwnd that will give us exponential
501 * growth is 2 mss. We don't allow the threshhold
505 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
508 tp->snd_cwnd = tp->t_maxseg;
509 tp->snd_ssthresh = win * tp->t_maxseg;
512 EXIT_FASTRECOVERY(tp);
513 (void) tcp_output(tp);
517 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
518 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,