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 1.6 2004/03/08 00:39:00 hsu 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>
52 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
54 #include <net/route.h>
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/in_pcb.h>
60 #include <netinet6/in6_pcb.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_fsm.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/tcp_timer.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet/tcpip.h>
70 #include <netinet/tcp_debug.h>
74 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
78 tt = *(int *)oidp->oid_arg1;
79 s = (int)((int64_t)tt * 1000 / hz);
81 error = sysctl_handle_int(oidp, &s, 0, req);
82 if (error || !req->newptr)
85 tt = (int)((int64_t)s * hz / 1000);
89 *(int *)oidp->oid_arg1 = tt;
94 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
95 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
98 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
99 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
102 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
103 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
106 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
107 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
108 "Time before a delayed ACK is sent");
111 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
112 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
116 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
120 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
122 static int always_keepalive = 0;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
124 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
126 static int tcp_keepcnt = TCPTV_KEEPCNT;
127 /* max idle probes */
128 int tcp_maxpersistidle;
129 /* max idle time in persist */
133 * Tcp protocol timeout routine called every 500 ms.
134 * Updates timestamps used for TCP
135 * causes finite state machine actions if timers expire.
144 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
150 * Cancel all timers for TCP tp.
156 callout_stop(tp->tt_2msl);
157 callout_stop(tp->tt_persist);
158 callout_stop(tp->tt_keep);
159 callout_stop(tp->tt_rexmt);
162 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
163 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
165 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
166 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
168 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
171 * TCP timer processing.
174 tcp_timer_delack(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);
197 struct tcpcb *tp = xtp;
202 ostate = tp->t_state;
205 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
209 callout_deactivate(tp->tt_2msl);
211 * 2 MSL timeout in shutdown went off. If we're closed but
212 * still waiting for peer to close and connection has been idle
213 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
214 * control block. Otherwise, check again in a bit.
216 if (tp->t_state != TCPS_TIME_WAIT &&
217 (ticks - tp->t_rcvtime) <= tcp_maxidle)
218 callout_reset(tp->tt_2msl, tcp_keepintvl,
224 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
225 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
235 struct tcpcb *tp = xtp;
236 struct tcptemp *t_template;
241 ostate = tp->t_state;
244 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
248 callout_deactivate(tp->tt_keep);
250 * Keep-alive timer went off; send something
251 * or drop connection if idle for too long.
253 tcpstat.tcps_keeptimeo++;
254 if (tp->t_state < TCPS_ESTABLISHED)
256 if ((always_keepalive ||
257 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
258 tp->t_state <= TCPS_CLOSING) {
259 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
262 * Send a packet designed to force a response
263 * if the peer is up and reachable:
264 * either an ACK if the connection is still alive,
265 * or an RST if the peer has closed the connection
266 * due to timeout or reboot.
267 * Using sequence number tp->snd_una-1
268 * causes the transmitted zero-length segment
269 * to lie outside the receive window;
270 * by the protocol spec, this requires the
271 * correspondent TCP to respond.
273 tcpstat.tcps_keepprobe++;
274 t_template = tcp_maketemplate(tp);
276 tcp_respond(tp, t_template->tt_ipgen,
277 &t_template->tt_t, (struct mbuf *)NULL,
278 tp->rcv_nxt, tp->snd_una - 1, 0);
279 (void) m_free(dtom(t_template));
281 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
283 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
286 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
287 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
294 tcpstat.tcps_keepdrops++;
295 tp = tcp_drop(tp, ETIMEDOUT);
298 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
299 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
306 tcp_timer_persist(xtp)
309 struct tcpcb *tp = xtp;
314 ostate = tp->t_state;
317 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
321 callout_deactivate(tp->tt_persist);
323 * Persistance timer into zero window.
324 * Force a byte to be output, if possible.
326 tcpstat.tcps_persisttimeo++;
328 * Hack: if the peer is dead/unreachable, we do not
329 * time out if the window is closed. After a full
330 * backoff, drop the connection if the idle time
331 * (no responses to probes) reaches the maximum
332 * backoff that we would use if retransmitting.
334 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
335 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
336 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
337 tcpstat.tcps_persistdrop++;
338 tp = tcp_drop(tp, ETIMEDOUT);
343 (void) tcp_output(tp);
348 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
349 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
356 tcp_save_congestion_state(struct tcpcb *tp)
358 tp->snd_cwnd_prev = tp->snd_cwnd;
359 tp->snd_ssthresh_prev = tp->snd_ssthresh;
360 tp->snd_recover_prev = tp->snd_recover;
361 if (IN_FASTRECOVERY(tp))
362 tp->t_flags |= TF_WASFRECOVERY;
364 tp->t_flags &= ~TF_WASFRECOVERY;
365 if (tp->t_flags & TF_RCVD_TSTMP) {
366 tp->t_rexmtTS = ticks;
367 tp->t_flags |= TF_FIRSTACCACK;
372 tcp_revert_congestion_state(struct tcpcb *tp)
374 tp->snd_cwnd = tp->snd_cwnd_prev;
375 tp->snd_ssthresh = tp->snd_ssthresh_prev;
376 tp->snd_recover = tp->snd_recover_prev;
377 if (tp->t_flags & TF_WASFRECOVERY)
378 ENTER_FASTRECOVERY(tp);
379 if (tp->t_flags & TF_FASTREXMT) {
380 ++tcpstat.tcps_sndfastrexmitbad;
381 if (tp->t_flags & TF_EARLYREXMT)
382 ++tcpstat.tcps_sndearlyrexmitbad;
384 ++tcpstat.tcps_sndrtobad;
387 tp->snd_nxt = tp->snd_max;
394 struct tcpcb *tp = xtp;
400 ostate = tp->t_state;
403 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
407 callout_deactivate(tp->tt_rexmt);
409 * Retransmission timer went off. Message has not
410 * been acked within retransmit interval. Back off
411 * to a longer retransmit interval and retransmit one segment.
413 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
414 tp->t_rxtshift = TCP_MAXRXTSHIFT;
415 tcpstat.tcps_timeoutdrop++;
416 tp = tcp_drop(tp, tp->t_softerror ?
417 tp->t_softerror : ETIMEDOUT);
420 if (tp->t_rxtshift == 1) {
422 * first retransmit; record ssthresh and cwnd so they can
423 * be recovered if this turns out to be a "bad" retransmit.
424 * A retransmit is considered "bad" if an ACK for this
425 * segment is received within RTT/2 interval; the assumption
426 * here is that the ACK was already in flight. See
427 * "On Estimating End-to-End Network Path Properties" by
428 * Allman and Paxson for more details.
430 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
431 tcp_save_congestion_state(tp);
432 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
434 tcpstat.tcps_rexmttimeo++;
435 if (tp->t_state == TCPS_SYN_SENT)
436 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
438 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
439 TCPT_RANGESET(tp->t_rxtcur, rexmt,
440 tp->t_rttmin, TCPTV_REXMTMAX);
442 * Disable rfc1323 and rfc1644 if we havn't got any response to
443 * our third SYN to work-around some broken terminal servers
444 * (most of which have hopefully been retired) that have bad VJ
445 * header compression code which trashes TCP segments containing
446 * unknown-to-them TCP options.
448 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
449 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
451 * If losing, let the lower level know and try for
452 * a better route. Also, if we backed off this far,
453 * our srtt estimate is probably bogus. Clobber it
454 * so we'll take the next rtt measurement as our srtt;
455 * move the current srtt into rttvar to keep the current
456 * retransmit times until then.
458 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
460 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
461 in6_losing(tp->t_inpcb);
464 in_losing(tp->t_inpcb);
465 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
468 tp->snd_nxt = tp->snd_una;
469 tp->snd_recover = tp->snd_max;
471 * Force a segment to be sent.
473 tp->t_flags |= TF_ACKNOW;
475 * If timing a segment in this window, stop the timer.
479 * Close the congestion window down to one segment
480 * (we'll open it by one segment for each ack we get).
481 * Since we probably have a window's worth of unacked
482 * data accumulated, this "slow start" keeps us from
483 * dumping all that data as back-to-back packets (which
484 * might overwhelm an intermediate gateway).
486 * There are two phases to the opening: Initially we
487 * open by one mss on each ack. This makes the window
488 * size increase exponentially with time. If the
489 * window is larger than the path can handle, this
490 * exponential growth results in dropped packet(s)
491 * almost immediately. To get more time between
492 * drops but still "push" the network to take advantage
493 * of improving conditions, we switch from exponential
494 * to linear window opening at some threshhold size.
495 * For a threshhold, we use half the current window
496 * size, truncated to a multiple of the mss.
498 * (the minimum cwnd that will give us exponential
499 * growth is 2 mss. We don't allow the threshhold
503 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
506 tp->snd_cwnd = tp->t_maxseg;
507 tp->snd_ssthresh = win * tp->t_maxseg;
510 EXIT_FASTRECOVERY(tp);
511 (void) tcp_output(tp);
515 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
516 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,