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 1.10 2004/06/04 04:32:23 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.
155 tcp_canceltimers(struct tcpcb *tp)
157 callout_stop(tp->tt_2msl);
158 callout_stop(tp->tt_persist);
159 callout_stop(tp->tt_keep);
160 callout_stop(tp->tt_rexmt);
163 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
164 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
166 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
167 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
169 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
172 * TCP timer processing.
175 tcp_timer_delack(void *xtp)
177 struct tcpcb *tp = xtp;
181 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
185 callout_deactivate(tp->tt_delack);
187 tp->t_flags |= TF_ACKNOW;
188 tcpstat.tcps_delack++;
189 (void) tcp_output(tp);
194 tcp_timer_2msl(void *xtp)
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,
231 tcp_timer_keep(void *xtp)
233 struct tcpcb *tp = xtp;
234 struct tcptemp *t_template;
239 ostate = tp->t_state;
242 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
246 callout_deactivate(tp->tt_keep);
248 * Keep-alive timer went off; send something
249 * or drop connection if idle for too long.
251 tcpstat.tcps_keeptimeo++;
252 if (tp->t_state < TCPS_ESTABLISHED)
254 if ((always_keepalive ||
255 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
256 tp->t_state <= TCPS_CLOSING) {
257 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
260 * Send a packet designed to force a response
261 * if the peer is up and reachable:
262 * either an ACK if the connection is still alive,
263 * or an RST if the peer has closed the connection
264 * due to timeout or reboot.
265 * Using sequence number tp->snd_una-1
266 * causes the transmitted zero-length segment
267 * to lie outside the receive window;
268 * by the protocol spec, this requires the
269 * correspondent TCP to respond.
271 tcpstat.tcps_keepprobe++;
272 t_template = tcp_maketemplate(tp);
274 tcp_respond(tp, t_template->tt_ipgen,
275 &t_template->tt_t, (struct mbuf *)NULL,
276 tp->rcv_nxt, tp->snd_una - 1, 0);
277 tcp_freetemplate(t_template);
279 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
281 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
284 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
285 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
292 tcpstat.tcps_keepdrops++;
293 tp = tcp_drop(tp, ETIMEDOUT);
296 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
297 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
304 tcp_timer_persist(void *xtp)
306 struct tcpcb *tp = xtp;
311 ostate = tp->t_state;
314 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
318 callout_deactivate(tp->tt_persist);
320 * Persistance timer into zero window.
321 * Force a byte to be output, if possible.
323 tcpstat.tcps_persisttimeo++;
325 * Hack: if the peer is dead/unreachable, we do not
326 * time out if the window is closed. After a full
327 * backoff, drop the connection if the idle time
328 * (no responses to probes) reaches the maximum
329 * backoff that we would use if retransmitting.
331 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
332 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
333 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
334 tcpstat.tcps_persistdrop++;
335 tp = tcp_drop(tp, ETIMEDOUT);
339 tp->t_flags |= TF_FORCE;
340 (void) tcp_output(tp);
341 tp->t_flags &= ~TF_FORCE;
345 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
346 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
353 tcp_save_congestion_state(struct tcpcb *tp)
355 tp->snd_cwnd_prev = tp->snd_cwnd;
356 tp->snd_ssthresh_prev = tp->snd_ssthresh;
357 tp->snd_recover_prev = tp->snd_recover;
358 if (IN_FASTRECOVERY(tp))
359 tp->t_flags |= TF_WASFRECOVERY;
361 tp->t_flags &= ~TF_WASFRECOVERY;
362 if (tp->t_flags & TF_RCVD_TSTMP) {
363 tp->t_rexmtTS = ticks;
364 tp->t_flags |= TF_FIRSTACCACK;
369 tcp_revert_congestion_state(struct tcpcb *tp)
371 tp->snd_cwnd = tp->snd_cwnd_prev;
372 tp->snd_ssthresh = tp->snd_ssthresh_prev;
373 tp->snd_recover = tp->snd_recover_prev;
374 if (tp->t_flags & TF_WASFRECOVERY)
375 ENTER_FASTRECOVERY(tp);
376 if (tp->t_flags & TF_FASTREXMT) {
377 ++tcpstat.tcps_sndfastrexmitbad;
378 if (tp->t_flags & TF_EARLYREXMT)
379 ++tcpstat.tcps_sndearlyrexmitbad;
381 ++tcpstat.tcps_sndrtobad;
384 tp->snd_nxt = tp->snd_max;
388 tcp_timer_rexmt(void *xtp)
390 struct tcpcb *tp = xtp;
396 ostate = tp->t_state;
399 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
403 callout_deactivate(tp->tt_rexmt);
405 * Retransmission timer went off. Message has not
406 * been acked within retransmit interval. Back off
407 * to a longer retransmit interval and retransmit one segment.
409 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
410 tp->t_rxtshift = TCP_MAXRXTSHIFT;
411 tcpstat.tcps_timeoutdrop++;
412 tp = tcp_drop(tp, tp->t_softerror ?
413 tp->t_softerror : ETIMEDOUT);
416 if (tp->t_rxtshift == 1) {
418 * first retransmit; record ssthresh and cwnd so they can
419 * be recovered if this turns out to be a "bad" retransmit.
420 * A retransmit is considered "bad" if an ACK for this
421 * segment is received within RTT/2 interval; the assumption
422 * here is that the ACK was already in flight. See
423 * "On Estimating End-to-End Network Path Properties" by
424 * Allman and Paxson for more details.
426 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
427 tcp_save_congestion_state(tp);
428 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
430 tcpstat.tcps_rexmttimeo++;
431 if (tp->t_state == TCPS_SYN_SENT)
432 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
434 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
435 TCPT_RANGESET(tp->t_rxtcur, rexmt,
436 tp->t_rttmin, TCPTV_REXMTMAX);
438 * Disable rfc1323 and rfc1644 if we havn't got any response to
439 * our third SYN to work-around some broken terminal servers
440 * (most of which have hopefully been retired) that have bad VJ
441 * header compression code which trashes TCP segments containing
442 * unknown-to-them TCP options.
444 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
445 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
447 * If losing, let the lower level know and try for
448 * a better route. Also, if we backed off this far,
449 * our srtt estimate is probably bogus. Clobber it
450 * so we'll take the next rtt measurement as our srtt;
451 * move the current srtt into rttvar to keep the current
452 * retransmit times until then.
454 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
456 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
457 in6_losing(tp->t_inpcb);
460 in_losing(tp->t_inpcb);
461 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
464 tp->snd_nxt = tp->snd_una;
465 tp->snd_recover = tp->snd_max;
467 * Force a segment to be sent.
469 tp->t_flags |= TF_ACKNOW;
471 * If timing a segment in this window, stop the timer.
475 * Close the congestion window down to one segment
476 * (we'll open it by one segment for each ack we get).
477 * Since we probably have a window's worth of unacked
478 * data accumulated, this "slow start" keeps us from
479 * dumping all that data as back-to-back packets (which
480 * might overwhelm an intermediate gateway).
482 * There are two phases to the opening: Initially we
483 * open by one mss on each ack. This makes the window
484 * size increase exponentially with time. If the
485 * window is larger than the path can handle, this
486 * exponential growth results in dropped packet(s)
487 * almost immediately. To get more time between
488 * drops but still "push" the network to take advantage
489 * of improving conditions, we switch from exponential
490 * to linear window opening at some threshhold size.
491 * For a threshhold, we use half the current window
492 * size, truncated to a multiple of the mss.
494 * (the minimum cwnd that will give us exponential
495 * growth is 2 mss. We don't allow the threshhold
499 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
502 tp->snd_cwnd = tp->t_maxseg;
503 tp->snd_ssthresh = win * tp->t_maxseg;
506 EXIT_FASTRECOVERY(tp);
507 (void) tcp_output(tp);
511 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
512 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,