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) 1982, 1986, 1988, 1990, 1993, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.14 2003/02/03 02:33:41 hsu Exp $
68 * $DragonFly: src/sys/netinet/tcp_timer.c,v 1.16 2007/03/04 18:51:59 swildner Exp $
71 #include "opt_compat.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/protosw.h>
83 #include <sys/thread.h>
84 #include <sys/globaldata.h>
85 #include <sys/thread2.h>
87 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
89 #include <net/route.h>
91 #include <netinet/in.h>
92 #include <netinet/in_systm.h>
93 #include <netinet/in_pcb.h>
95 #include <netinet6/in6_pcb.h>
97 #include <netinet/ip_var.h>
98 #include <netinet/tcp.h>
99 #include <netinet/tcp_fsm.h>
100 #include <netinet/tcp_seq.h>
101 #include <netinet/tcp_timer.h>
102 #include <netinet/tcp_var.h>
103 #include <netinet/tcpip.h>
105 #include <netinet/tcp_debug.h>
109 sysctl_msec_to_ticks(SYSCTL_HANDLER_ARGS
)
113 tt
= *(int *)oidp
->oid_arg1
;
114 s
= (int)((int64_t)tt
* 1000 / hz
);
116 error
= sysctl_handle_int(oidp
, &s
, 0, req
);
117 if (error
|| !req
->newptr
)
120 tt
= (int)((int64_t)s
* hz
/ 1000);
124 *(int *)oidp
->oid_arg1
= tt
;
129 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPINIT
, keepinit
, CTLTYPE_INT
|CTLFLAG_RW
,
130 &tcp_keepinit
, 0, sysctl_msec_to_ticks
, "I", "");
133 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPIDLE
, keepidle
, CTLTYPE_INT
|CTLFLAG_RW
,
134 &tcp_keepidle
, 0, sysctl_msec_to_ticks
, "I", "");
137 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPINTVL
, keepintvl
, CTLTYPE_INT
|CTLFLAG_RW
,
138 &tcp_keepintvl
, 0, sysctl_msec_to_ticks
, "I", "");
141 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_DELACKTIME
, delacktime
,
142 CTLTYPE_INT
|CTLFLAG_RW
, &tcp_delacktime
, 0, sysctl_msec_to_ticks
, "I",
143 "Time before a delayed ACK is sent");
146 SYSCTL_PROC(_net_inet_tcp
, OID_AUTO
, msl
, CTLTYPE_INT
|CTLFLAG_RW
,
147 &tcp_msl
, 0, sysctl_msec_to_ticks
, "I", "Maximum segment lifetime");
150 SYSCTL_PROC(_net_inet_tcp
, OID_AUTO
, rexmit_min
, CTLTYPE_INT
|CTLFLAG_RW
,
151 &tcp_rexmit_min
, 0, sysctl_msec_to_ticks
, "I", "Minimum Retransmission Timeout");
154 SYSCTL_PROC(_net_inet_tcp
, OID_AUTO
, rexmit_slop
, CTLTYPE_INT
|CTLFLAG_RW
,
155 &tcp_rexmit_slop
, 0, sysctl_msec_to_ticks
, "I",
156 "Retransmission Timer Slop");
158 static int always_keepalive
= 0;
159 SYSCTL_INT(_net_inet_tcp
, OID_AUTO
, always_keepalive
, CTLFLAG_RW
,
160 &always_keepalive
, 0, "Assume SO_KEEPALIVE on all TCP connections");
162 static int tcp_keepcnt
= TCPTV_KEEPCNT
;
163 /* max idle probes */
164 int tcp_maxpersistidle
;
165 /* max idle time in persist */
169 * Tcp protocol timeout routine called every 500 ms.
170 * Updates timestamps used for TCP
171 * causes finite state machine actions if timers expire.
177 tcp_maxidle
= tcp_keepcnt
* tcp_keepintvl
;
182 * Cancel all timers for TCP tp.
185 tcp_canceltimers(struct tcpcb
*tp
)
187 callout_stop(tp
->tt_2msl
);
188 callout_stop(tp
->tt_persist
);
189 callout_stop(tp
->tt_keep
);
190 callout_stop(tp
->tt_rexmt
);
193 int tcp_syn_backoff
[TCP_MAXRXTSHIFT
+ 1] =
194 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
196 int tcp_backoff
[TCP_MAXRXTSHIFT
+ 1] =
197 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
199 static int tcp_totbackoff
= 511; /* sum of tcp_backoff[] */
202 * TCP timer processing.
205 tcp_timer_delack(void *xtp
)
207 struct tcpcb
*tp
= xtp
;
210 if (callout_pending(tp
->tt_delack
) || !callout_active(tp
->tt_delack
)) {
214 callout_deactivate(tp
->tt_delack
);
216 tp
->t_flags
|= TF_ACKNOW
;
217 tcpstat
.tcps_delack
++;
223 tcp_timer_2msl(void *xtp
)
225 struct tcpcb
*tp
= xtp
;
229 ostate
= tp
->t_state
;
232 if (callout_pending(tp
->tt_2msl
) || !callout_active(tp
->tt_2msl
)) {
236 callout_deactivate(tp
->tt_2msl
);
238 * 2 MSL timeout in shutdown went off. If we're closed but
239 * still waiting for peer to close and connection has been idle
240 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
241 * control block. Otherwise, check again in a bit.
243 if (tp
->t_state
!= TCPS_TIME_WAIT
&&
244 (ticks
- tp
->t_rcvtime
) <= tcp_maxidle
)
245 callout_reset(tp
->tt_2msl
, tcp_keepintvl
,
251 if (tp
&& (tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
))
252 tcp_trace(TA_USER
, ostate
, tp
, NULL
, NULL
, PRU_SLOWTIMO
);
258 tcp_timer_keep(void *xtp
)
260 struct tcpcb
*tp
= xtp
;
261 struct tcptemp
*t_template
;
265 ostate
= tp
->t_state
;
268 if (callout_pending(tp
->tt_keep
) || !callout_active(tp
->tt_keep
)) {
272 callout_deactivate(tp
->tt_keep
);
274 * Keep-alive timer went off; send something
275 * or drop connection if idle for too long.
277 tcpstat
.tcps_keeptimeo
++;
278 if (tp
->t_state
< TCPS_ESTABLISHED
)
280 if ((always_keepalive
||
281 tp
->t_inpcb
->inp_socket
->so_options
& SO_KEEPALIVE
) &&
282 tp
->t_state
<= TCPS_CLOSING
) {
283 if ((ticks
- tp
->t_rcvtime
) >= tcp_keepidle
+ tcp_maxidle
)
286 * Send a packet designed to force a response
287 * if the peer is up and reachable:
288 * either an ACK if the connection is still alive,
289 * or an RST if the peer has closed the connection
290 * due to timeout or reboot.
291 * Using sequence number tp->snd_una-1
292 * causes the transmitted zero-length segment
293 * to lie outside the receive window;
294 * by the protocol spec, this requires the
295 * correspondent TCP to respond.
297 tcpstat
.tcps_keepprobe
++;
298 t_template
= tcp_maketemplate(tp
);
300 tcp_respond(tp
, t_template
->tt_ipgen
,
301 &t_template
->tt_t
, (struct mbuf
*)NULL
,
302 tp
->rcv_nxt
, tp
->snd_una
- 1, 0);
303 tcp_freetemplate(t_template
);
305 callout_reset(tp
->tt_keep
, tcp_keepintvl
, tcp_timer_keep
, tp
);
307 callout_reset(tp
->tt_keep
, tcp_keepidle
, tcp_timer_keep
, tp
);
310 if (tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
)
311 tcp_trace(TA_USER
, ostate
, tp
, NULL
, NULL
, PRU_SLOWTIMO
);
317 tcpstat
.tcps_keepdrops
++;
318 tp
= tcp_drop(tp
, ETIMEDOUT
);
321 if (tp
&& (tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
))
322 tcp_trace(TA_USER
, ostate
, tp
, NULL
, NULL
, PRU_SLOWTIMO
);
328 tcp_timer_persist(void *xtp
)
330 struct tcpcb
*tp
= xtp
;
334 ostate
= tp
->t_state
;
337 if (callout_pending(tp
->tt_persist
) || !callout_active(tp
->tt_persist
)){
341 callout_deactivate(tp
->tt_persist
);
343 * Persistance timer into zero window.
344 * Force a byte to be output, if possible.
346 tcpstat
.tcps_persisttimeo
++;
348 * Hack: if the peer is dead/unreachable, we do not
349 * time out if the window is closed. After a full
350 * backoff, drop the connection if the idle time
351 * (no responses to probes) reaches the maximum
352 * backoff that we would use if retransmitting.
354 if (tp
->t_rxtshift
== TCP_MAXRXTSHIFT
&&
355 ((ticks
- tp
->t_rcvtime
) >= tcp_maxpersistidle
||
356 (ticks
- tp
->t_rcvtime
) >= TCP_REXMTVAL(tp
) * tcp_totbackoff
)) {
357 tcpstat
.tcps_persistdrop
++;
358 tp
= tcp_drop(tp
, ETIMEDOUT
);
362 tp
->t_flags
|= TF_FORCE
;
364 tp
->t_flags
&= ~TF_FORCE
;
368 if (tp
&& tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
)
369 tcp_trace(TA_USER
, ostate
, tp
, NULL
, NULL
, PRU_SLOWTIMO
);
375 tcp_save_congestion_state(struct tcpcb
*tp
)
377 tp
->snd_cwnd_prev
= tp
->snd_cwnd
;
378 tp
->snd_wacked_prev
= tp
->snd_wacked
;
379 tp
->snd_ssthresh_prev
= tp
->snd_ssthresh
;
380 tp
->snd_recover_prev
= tp
->snd_recover
;
381 if (IN_FASTRECOVERY(tp
))
382 tp
->t_flags
|= TF_WASFRECOVERY
;
384 tp
->t_flags
&= ~TF_WASFRECOVERY
;
385 if (tp
->t_flags
& TF_RCVD_TSTMP
) {
386 tp
->t_rexmtTS
= ticks
;
387 tp
->t_flags
|= TF_FIRSTACCACK
;
390 tcp_sack_save_scoreboard(&tp
->scb
);
395 tcp_revert_congestion_state(struct tcpcb
*tp
)
397 tp
->snd_cwnd
= tp
->snd_cwnd_prev
;
398 tp
->snd_wacked
= tp
->snd_wacked_prev
;
399 tp
->snd_ssthresh
= tp
->snd_ssthresh_prev
;
400 tp
->snd_recover
= tp
->snd_recover_prev
;
401 if (tp
->t_flags
& TF_WASFRECOVERY
)
402 ENTER_FASTRECOVERY(tp
);
403 if (tp
->t_flags
& TF_FASTREXMT
) {
404 ++tcpstat
.tcps_sndfastrexmitbad
;
405 if (tp
->t_flags
& TF_EARLYREXMT
)
406 ++tcpstat
.tcps_sndearlyrexmitbad
;
408 ++tcpstat
.tcps_sndrtobad
;
411 tp
->snd_nxt
= tp
->snd_max
;
413 tcp_sack_revert_scoreboard(&tp
->scb
, tp
->snd_una
);
418 tcp_timer_rexmt(void *xtp
)
420 struct tcpcb
*tp
= xtp
;
425 ostate
= tp
->t_state
;
428 if (callout_pending(tp
->tt_rexmt
) || !callout_active(tp
->tt_rexmt
)) {
432 callout_deactivate(tp
->tt_rexmt
);
434 * Retransmission timer went off. Message has not
435 * been acked within retransmit interval. Back off
436 * to a longer retransmit interval and retransmit one segment.
438 if (++tp
->t_rxtshift
> TCP_MAXRXTSHIFT
) {
439 tp
->t_rxtshift
= TCP_MAXRXTSHIFT
;
440 tcpstat
.tcps_timeoutdrop
++;
441 tp
= tcp_drop(tp
, tp
->t_softerror
?
442 tp
->t_softerror
: ETIMEDOUT
);
445 if (tp
->t_rxtshift
== 1) {
447 * first retransmit; record ssthresh and cwnd so they can
448 * be recovered if this turns out to be a "bad" retransmit.
449 * A retransmit is considered "bad" if an ACK for this
450 * segment is received within RTT/2 interval; the assumption
451 * here is that the ACK was already in flight. See
452 * "On Estimating End-to-End Network Path Properties" by
453 * Allman and Paxson for more details.
455 tp
->t_badrxtwin
= ticks
+ (tp
->t_srtt
>> (TCP_RTT_SHIFT
+ 1));
456 tcp_save_congestion_state(tp
);
457 tp
->t_flags
&= ~(TF_FASTREXMT
| TF_EARLYREXMT
);
459 /* Throw away SACK blocks on a RTO, as specified by RFC2018. */
460 tcp_sack_cleanup(&tp
->scb
);
461 tcpstat
.tcps_rexmttimeo
++;
462 if (tp
->t_state
== TCPS_SYN_SENT
)
463 rexmt
= TCP_REXMTVAL(tp
) * tcp_syn_backoff
[tp
->t_rxtshift
];
465 rexmt
= TCP_REXMTVAL(tp
) * tcp_backoff
[tp
->t_rxtshift
];
466 TCPT_RANGESET(tp
->t_rxtcur
, rexmt
,
467 tp
->t_rttmin
, TCPTV_REXMTMAX
);
469 * Disable rfc1323 and rfc1644 if we havn't got any response to
470 * our third SYN to work-around some broken terminal servers
471 * (most of which have hopefully been retired) that have bad VJ
472 * header compression code which trashes TCP segments containing
473 * unknown-to-them TCP options.
475 if ((tp
->t_state
== TCPS_SYN_SENT
) && (tp
->t_rxtshift
== 3))
476 tp
->t_flags
&= ~(TF_REQ_SCALE
|TF_REQ_TSTMP
|TF_REQ_CC
);
478 * If losing, let the lower level know and try for
479 * a better route. Also, if we backed off this far,
480 * our srtt estimate is probably bogus. Clobber it
481 * so we'll take the next rtt measurement as our srtt;
482 * move the current srtt into rttvar to keep the current
483 * retransmit times until then.
485 if (tp
->t_rxtshift
> TCP_MAXRXTSHIFT
/ 4) {
487 if ((tp
->t_inpcb
->inp_vflag
& INP_IPV6
) != 0)
488 in6_losing(tp
->t_inpcb
);
491 in_losing(tp
->t_inpcb
);
492 tp
->t_rttvar
+= (tp
->t_srtt
>> TCP_RTT_SHIFT
);
495 tp
->snd_nxt
= tp
->snd_una
;
496 tp
->rexmt_high
= tp
->snd_una
;
497 tp
->snd_recover
= tp
->snd_max
;
499 * Force a segment to be sent.
501 tp
->t_flags
|= TF_ACKNOW
;
503 * If timing a segment in this window, stop the timer.
507 * Close the congestion window down to one segment
508 * (we'll open it by one segment for each ack we get).
509 * Since we probably have a window's worth of unacked
510 * data accumulated, this "slow start" keeps us from
511 * dumping all that data as back-to-back packets (which
512 * might overwhelm an intermediate gateway).
514 * There are two phases to the opening: Initially we
515 * open by one mss on each ack. This makes the window
516 * size increase exponentially with time. If the
517 * window is larger than the path can handle, this
518 * exponential growth results in dropped packet(s)
519 * almost immediately. To get more time between
520 * drops but still "push" the network to take advantage
521 * of improving conditions, we switch from exponential
522 * to linear window opening at some threshhold size.
523 * For a threshhold, we use half the current window
524 * size, truncated to a multiple of the mss.
526 * (the minimum cwnd that will give us exponential
527 * growth is 2 mss. We don't allow the threshhold
531 u_int win
= min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 / tp
->t_maxseg
;
535 tp
->snd_cwnd
= tp
->t_maxseg
;
537 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
540 EXIT_FASTRECOVERY(tp
);
545 if (tp
&& (tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
))
546 tcp_trace(TA_USER
, ostate
, tp
, NULL
, NULL
, PRU_SLOWTIMO
);