2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
84 * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
85 * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.12 2004/11/14 00:49:08 hsu Exp $
88 #include "opt_compat.h"
89 #include "opt_inet6.h"
90 #include "opt_tcpdebug.h"
92 #include <sys/param.h>
93 #include <sys/systm.h>
94 #include <sys/kernel.h>
96 #include <sys/sysctl.h>
97 #include <sys/socket.h>
98 #include <sys/socketvar.h>
99 #include <sys/protosw.h>
100 #include <sys/thread.h>
101 #include <sys/globaldata.h>
103 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
105 #include <net/route.h>
107 #include <netinet/in.h>
108 #include <netinet/in_systm.h>
109 #include <netinet/in_pcb.h>
111 #include <netinet6/in6_pcb.h>
113 #include <netinet/ip_var.h>
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_var.h>
119 #include <netinet/tcpip.h>
121 #include <netinet/tcp_debug.h>
125 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS)
129 tt = *(int *)oidp->oid_arg1;
130 s = (int)((int64_t)tt * 1000 / hz);
132 error = sysctl_handle_int(oidp, &s, 0, req);
133 if (error || !req->newptr)
136 tt = (int)((int64_t)s * hz / 1000);
140 *(int *)oidp->oid_arg1 = tt;
145 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
146 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
149 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
150 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
153 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
154 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
157 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
158 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
159 "Time before a delayed ACK is sent");
162 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
163 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
166 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
167 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
170 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
171 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
173 static int always_keepalive = 0;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
175 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
177 static int tcp_keepcnt = TCPTV_KEEPCNT;
178 /* max idle probes */
179 int tcp_maxpersistidle;
180 /* max idle time in persist */
184 * Tcp protocol timeout routine called every 500 ms.
185 * Updates timestamps used for TCP
186 * causes finite state machine actions if timers expire.
195 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
201 * Cancel all timers for TCP tp.
204 tcp_canceltimers(struct tcpcb *tp)
206 callout_stop(tp->tt_2msl);
207 callout_stop(tp->tt_persist);
208 callout_stop(tp->tt_keep);
209 callout_stop(tp->tt_rexmt);
212 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
213 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
215 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
216 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
218 static int tcp_totbackoff = 511; /* sum of tcp_backoff[] */
221 * TCP timer processing.
224 tcp_timer_delack(void *xtp)
226 struct tcpcb *tp = xtp;
230 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
234 callout_deactivate(tp->tt_delack);
236 tp->t_flags |= TF_ACKNOW;
237 tcpstat.tcps_delack++;
238 (void) tcp_output(tp);
243 tcp_timer_2msl(void *xtp)
245 struct tcpcb *tp = xtp;
250 ostate = tp->t_state;
253 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
257 callout_deactivate(tp->tt_2msl);
259 * 2 MSL timeout in shutdown went off. If we're closed but
260 * still waiting for peer to close and connection has been idle
261 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
262 * control block. Otherwise, check again in a bit.
264 if (tp->t_state != TCPS_TIME_WAIT &&
265 (ticks - tp->t_rcvtime) <= tcp_maxidle)
266 callout_reset(tp->tt_2msl, tcp_keepintvl,
272 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
273 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
280 tcp_timer_keep(void *xtp)
282 struct tcpcb *tp = xtp;
283 struct tcptemp *t_template;
288 ostate = tp->t_state;
291 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
295 callout_deactivate(tp->tt_keep);
297 * Keep-alive timer went off; send something
298 * or drop connection if idle for too long.
300 tcpstat.tcps_keeptimeo++;
301 if (tp->t_state < TCPS_ESTABLISHED)
303 if ((always_keepalive ||
304 tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) &&
305 tp->t_state <= TCPS_CLOSING) {
306 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
309 * Send a packet designed to force a response
310 * if the peer is up and reachable:
311 * either an ACK if the connection is still alive,
312 * or an RST if the peer has closed the connection
313 * due to timeout or reboot.
314 * Using sequence number tp->snd_una-1
315 * causes the transmitted zero-length segment
316 * to lie outside the receive window;
317 * by the protocol spec, this requires the
318 * correspondent TCP to respond.
320 tcpstat.tcps_keepprobe++;
321 t_template = tcp_maketemplate(tp);
323 tcp_respond(tp, t_template->tt_ipgen,
324 &t_template->tt_t, (struct mbuf *)NULL,
325 tp->rcv_nxt, tp->snd_una - 1, 0);
326 tcp_freetemplate(t_template);
328 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
330 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
333 if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
334 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
341 tcpstat.tcps_keepdrops++;
342 tp = tcp_drop(tp, ETIMEDOUT);
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_timer_persist(void *xtp)
355 struct tcpcb *tp = xtp;
360 ostate = tp->t_state;
363 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
367 callout_deactivate(tp->tt_persist);
369 * Persistance timer into zero window.
370 * Force a byte to be output, if possible.
372 tcpstat.tcps_persisttimeo++;
374 * Hack: if the peer is dead/unreachable, we do not
375 * time out if the window is closed. After a full
376 * backoff, drop the connection if the idle time
377 * (no responses to probes) reaches the maximum
378 * backoff that we would use if retransmitting.
380 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
381 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
382 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
383 tcpstat.tcps_persistdrop++;
384 tp = tcp_drop(tp, ETIMEDOUT);
388 tp->t_flags |= TF_FORCE;
389 (void) tcp_output(tp);
390 tp->t_flags &= ~TF_FORCE;
394 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
395 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
402 tcp_save_congestion_state(struct tcpcb *tp)
404 tp->snd_cwnd_prev = tp->snd_cwnd;
405 tp->snd_ssthresh_prev = tp->snd_ssthresh;
406 tp->snd_recover_prev = tp->snd_recover;
407 if (IN_FASTRECOVERY(tp))
408 tp->t_flags |= TF_WASFRECOVERY;
410 tp->t_flags &= ~TF_WASFRECOVERY;
411 if (tp->t_flags & TF_RCVD_TSTMP) {
412 tp->t_rexmtTS = ticks;
413 tp->t_flags |= TF_FIRSTACCACK;
416 tcp_sack_save_scoreboard(&tp->scb);
421 tcp_revert_congestion_state(struct tcpcb *tp)
423 tp->snd_cwnd = tp->snd_cwnd_prev;
424 tp->snd_ssthresh = tp->snd_ssthresh_prev;
425 tp->snd_recover = tp->snd_recover_prev;
426 if (tp->t_flags & TF_WASFRECOVERY)
427 ENTER_FASTRECOVERY(tp);
428 if (tp->t_flags & TF_FASTREXMT) {
429 ++tcpstat.tcps_sndfastrexmitbad;
430 if (tp->t_flags & TF_EARLYREXMT)
431 ++tcpstat.tcps_sndearlyrexmitbad;
433 ++tcpstat.tcps_sndrtobad;
436 tp->snd_nxt = tp->snd_max;
438 tcp_sack_revert_scoreboard(&tp->scb, tp->snd_una);
443 tcp_timer_rexmt(void *xtp)
445 struct tcpcb *tp = xtp;
451 ostate = tp->t_state;
454 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
458 callout_deactivate(tp->tt_rexmt);
460 * Retransmission timer went off. Message has not
461 * been acked within retransmit interval. Back off
462 * to a longer retransmit interval and retransmit one segment.
464 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
465 tp->t_rxtshift = TCP_MAXRXTSHIFT;
466 tcpstat.tcps_timeoutdrop++;
467 tp = tcp_drop(tp, tp->t_softerror ?
468 tp->t_softerror : ETIMEDOUT);
471 if (tp->t_rxtshift == 1) {
473 * first retransmit; record ssthresh and cwnd so they can
474 * be recovered if this turns out to be a "bad" retransmit.
475 * A retransmit is considered "bad" if an ACK for this
476 * segment is received within RTT/2 interval; the assumption
477 * here is that the ACK was already in flight. See
478 * "On Estimating End-to-End Network Path Properties" by
479 * Allman and Paxson for more details.
481 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
482 tcp_save_congestion_state(tp);
483 tp->t_flags &= ~(TF_FASTREXMT | TF_EARLYREXMT);
485 /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
486 tcp_sack_cleanup(&tp->scb);
487 tcpstat.tcps_rexmttimeo++;
488 if (tp->t_state == TCPS_SYN_SENT)
489 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
491 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
492 TCPT_RANGESET(tp->t_rxtcur, rexmt,
493 tp->t_rttmin, TCPTV_REXMTMAX);
495 * Disable rfc1323 and rfc1644 if we havn't got any response to
496 * our third SYN to work-around some broken terminal servers
497 * (most of which have hopefully been retired) that have bad VJ
498 * header compression code which trashes TCP segments containing
499 * unknown-to-them TCP options.
501 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
502 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_REQ_CC);
504 * If losing, let the lower level know and try for
505 * a better route. Also, if we backed off this far,
506 * our srtt estimate is probably bogus. Clobber it
507 * so we'll take the next rtt measurement as our srtt;
508 * move the current srtt into rttvar to keep the current
509 * retransmit times until then.
511 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
513 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
514 in6_losing(tp->t_inpcb);
517 in_losing(tp->t_inpcb);
518 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
521 tp->snd_nxt = tp->snd_una;
522 tp->rexmt_high = tp->snd_una;
523 tp->snd_recover = tp->snd_max;
525 * Force a segment to be sent.
527 tp->t_flags |= TF_ACKNOW;
529 * If timing a segment in this window, stop the timer.
533 * Close the congestion window down to one segment
534 * (we'll open it by one segment for each ack we get).
535 * Since we probably have a window's worth of unacked
536 * data accumulated, this "slow start" keeps us from
537 * dumping all that data as back-to-back packets (which
538 * might overwhelm an intermediate gateway).
540 * There are two phases to the opening: Initially we
541 * open by one mss on each ack. This makes the window
542 * size increase exponentially with time. If the
543 * window is larger than the path can handle, this
544 * exponential growth results in dropped packet(s)
545 * almost immediately. To get more time between
546 * drops but still "push" the network to take advantage
547 * of improving conditions, we switch from exponential
548 * to linear window opening at some threshhold size.
549 * For a threshhold, we use half the current window
550 * size, truncated to a multiple of the mss.
552 * (the minimum cwnd that will give us exponential
553 * growth is 2 mss. We don't allow the threshhold
557 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
560 tp->snd_cwnd = tp->t_maxseg;
561 tp->snd_ssthresh = win * tp->t_maxseg;
564 EXIT_FASTRECOVERY(tp);
565 (void) tcp_output(tp);
569 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
570 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,