2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include "opt_inet6.h"
36 #include "opt_tcpdebug.h"
38 #include <sys/param.h>
39 #include <sys/kernel.h>
42 #include <sys/mutex.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
49 #include <net/route.h>
51 #include <netinet/in.h>
52 #include <netinet/in_pcb.h>
53 #include <netinet/in_systm.h>
55 #include <netinet6/in6_pcb.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcp_fsm.h>
60 #include <netinet/tcp_timer.h>
61 #include <netinet/tcp_var.h>
62 #include <netinet/tcpip.h>
64 #include <netinet/tcp_debug.h>
68 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
69 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
72 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
73 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
76 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
77 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
80 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
81 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
82 "Time before a delayed ACK is sent");
85 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
86 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
89 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
90 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
91 "Minimum Retransmission Timeout");
94 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
95 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
96 "Retransmission Timer Slop");
98 static int always_keepalive = 1;
99 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
100 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
102 int tcp_fast_finwait2_recycle = 0;
103 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
104 &tcp_fast_finwait2_recycle, 0,
105 "Recycle closed FIN_WAIT_2 connections faster");
107 int tcp_finwait2_timeout;
108 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
109 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
112 static int tcp_keepcnt = TCPTV_KEEPCNT;
113 /* max idle probes */
114 int tcp_maxpersistidle;
115 /* max idle time in persist */
119 * Tcp protocol timeout routine called every 500 ms.
120 * Updates timestamps used for TCP
121 * causes finite state machine actions if timers expire.
127 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
128 INP_INFO_WLOCK(&tcbinfo);
129 (void) tcp_tw_2msl_scan(0);
130 INP_INFO_WUNLOCK(&tcbinfo);
133 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
134 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
136 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
137 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
139 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
141 static int tcp_timer_race;
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
143 0, "Count of t_inpcb races on tcp_discardcb");
146 * TCP timer processing.
150 tcp_timer_delack(void *xtp)
152 struct tcpcb *tp = xtp;
155 INP_INFO_RLOCK(&tcbinfo);
158 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
159 * tear-down mean we need it as a work-around for races between
160 * timers and tcp_discardcb().
162 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
166 INP_INFO_RUNLOCK(&tcbinfo);
170 INP_INFO_RUNLOCK(&tcbinfo);
171 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack)
172 || !callout_active(&tp->t_timers->tt_delack)) {
176 callout_deactivate(&tp->t_timers->tt_delack);
178 tp->t_flags |= TF_ACKNOW;
179 tcpstat.tcps_delack++;
180 (void) tcp_output(tp);
185 tcp_timer_2msl(void *xtp)
187 struct tcpcb *tp = xtp;
192 ostate = tp->t_state;
195 * XXXRW: Does this actually happen?
197 INP_INFO_WLOCK(&tcbinfo);
200 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
201 * tear-down mean we need it as a work-around for races between
202 * timers and tcp_discardcb().
204 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
208 INP_INFO_WUNLOCK(&tcbinfo);
212 tcp_free_sackholes(tp);
213 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) ||
214 !callout_active(&tp->t_timers->tt_2msl)) {
215 INP_WUNLOCK(tp->t_inpcb);
216 INP_INFO_WUNLOCK(&tcbinfo);
219 callout_deactivate(&tp->t_timers->tt_2msl);
221 * 2 MSL timeout in shutdown went off. If we're closed but
222 * still waiting for peer to close and connection has been idle
223 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
224 * control block. Otherwise, check again in a bit.
226 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
227 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
228 * Ignore fact that there were recent incoming segments.
230 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
231 tp->t_inpcb && tp->t_inpcb->inp_socket &&
232 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
233 tcpstat.tcps_finwait2_drops++;
236 if (tp->t_state != TCPS_TIME_WAIT &&
237 (ticks - tp->t_rcvtime) <= tcp_maxidle)
238 callout_reset(&tp->t_timers->tt_2msl, tcp_keepintvl,
245 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
246 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
251 INP_INFO_WUNLOCK(&tcbinfo);
255 tcp_timer_keep(void *xtp)
257 struct tcpcb *tp = xtp;
258 struct tcptemp *t_template;
263 ostate = tp->t_state;
265 INP_INFO_WLOCK(&tcbinfo);
268 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
269 * tear-down mean we need it as a work-around for races between
270 * timers and tcp_discardcb().
272 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
276 INP_INFO_WUNLOCK(&tcbinfo);
280 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep)
281 || !callout_active(&tp->t_timers->tt_keep)) {
283 INP_INFO_WUNLOCK(&tcbinfo);
286 callout_deactivate(&tp->t_timers->tt_keep);
288 * Keep-alive timer went off; send something
289 * or drop connection if idle for too long.
291 tcpstat.tcps_keeptimeo++;
292 if (tp->t_state < TCPS_ESTABLISHED)
294 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
295 tp->t_state <= TCPS_CLOSING) {
296 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
299 * Send a packet designed to force a response
300 * if the peer is up and reachable:
301 * either an ACK if the connection is still alive,
302 * or an RST if the peer has closed the connection
303 * due to timeout or reboot.
304 * Using sequence number tp->snd_una-1
305 * causes the transmitted zero-length segment
306 * to lie outside the receive window;
307 * by the protocol spec, this requires the
308 * correspondent TCP to respond.
310 tcpstat.tcps_keepprobe++;
311 t_template = tcpip_maketemplate(inp);
313 tcp_respond(tp, t_template->tt_ipgen,
314 &t_template->tt_t, (struct mbuf *)NULL,
315 tp->rcv_nxt, tp->snd_una - 1, 0);
316 (void) m_free(dtom(t_template));
318 callout_reset(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
320 callout_reset(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
323 if (inp->inp_socket->so_options & SO_DEBUG)
324 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
328 INP_INFO_WUNLOCK(&tcbinfo);
332 tcpstat.tcps_keepdrops++;
333 tp = tcp_drop(tp, ETIMEDOUT);
336 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
337 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
341 INP_WUNLOCK(tp->t_inpcb);
342 INP_INFO_WUNLOCK(&tcbinfo);
346 tcp_timer_persist(void *xtp)
348 struct tcpcb *tp = xtp;
353 ostate = tp->t_state;
355 INP_INFO_WLOCK(&tcbinfo);
358 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
359 * tear-down mean we need it as a work-around for races between
360 * timers and tcp_discardcb().
362 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
366 INP_INFO_WUNLOCK(&tcbinfo);
370 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
371 || !callout_active(&tp->t_timers->tt_persist)) {
373 INP_INFO_WUNLOCK(&tcbinfo);
376 callout_deactivate(&tp->t_timers->tt_persist);
378 * Persistance timer into zero window.
379 * Force a byte to be output, if possible.
381 tcpstat.tcps_persisttimeo++;
383 * Hack: if the peer is dead/unreachable, we do not
384 * time out if the window is closed. After a full
385 * backoff, drop the connection if the idle time
386 * (no responses to probes) reaches the maximum
387 * backoff that we would use if retransmitting.
389 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
390 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
391 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
392 tcpstat.tcps_persistdrop++;
393 tp = tcp_drop(tp, ETIMEDOUT);
397 tp->t_flags |= TF_FORCEDATA;
398 (void) tcp_output(tp);
399 tp->t_flags &= ~TF_FORCEDATA;
403 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
404 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
408 INP_INFO_WUNLOCK(&tcbinfo);
412 tcp_timer_rexmt(void * xtp)
414 struct tcpcb *tp = xtp;
421 ostate = tp->t_state;
423 INP_INFO_WLOCK(&tcbinfo);
427 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
428 * tear-down mean we need it as a work-around for races between
429 * timers and tcp_discardcb().
431 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
435 INP_INFO_WUNLOCK(&tcbinfo);
439 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt)
440 || !callout_active(&tp->t_timers->tt_rexmt)) {
442 INP_INFO_WUNLOCK(&tcbinfo);
445 callout_deactivate(&tp->t_timers->tt_rexmt);
446 tcp_free_sackholes(tp);
448 * Retransmission timer went off. Message has not
449 * been acked within retransmit interval. Back off
450 * to a longer retransmit interval and retransmit one segment.
452 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
453 tp->t_rxtshift = TCP_MAXRXTSHIFT;
454 tcpstat.tcps_timeoutdrop++;
455 tp = tcp_drop(tp, tp->t_softerror ?
456 tp->t_softerror : ETIMEDOUT);
459 INP_INFO_WUNLOCK(&tcbinfo);
461 if (tp->t_rxtshift == 1) {
463 * first retransmit; record ssthresh and cwnd so they can
464 * be recovered if this turns out to be a "bad" retransmit.
465 * A retransmit is considered "bad" if an ACK for this
466 * segment is received within RTT/2 interval; the assumption
467 * here is that the ACK was already in flight. See
468 * "On Estimating End-to-End Network Path Properties" by
469 * Allman and Paxson for more details.
471 tp->snd_cwnd_prev = tp->snd_cwnd;
472 tp->snd_ssthresh_prev = tp->snd_ssthresh;
473 tp->snd_recover_prev = tp->snd_recover;
474 if (IN_FASTRECOVERY(tp))
475 tp->t_flags |= TF_WASFRECOVERY;
477 tp->t_flags &= ~TF_WASFRECOVERY;
478 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
480 tcpstat.tcps_rexmttimeo++;
481 if (tp->t_state == TCPS_SYN_SENT)
482 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
484 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
485 TCPT_RANGESET(tp->t_rxtcur, rexmt,
486 tp->t_rttmin, TCPTV_REXMTMAX);
488 * Disable rfc1323 if we havn't got any response to
489 * our third SYN to work-around some broken terminal servers
490 * (most of which have hopefully been retired) that have bad VJ
491 * header compression code which trashes TCP segments containing
492 * unknown-to-them TCP options.
494 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
495 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
497 * If we backed off this far, our srtt estimate is probably bogus.
498 * Clobber it so we'll take the next rtt measurement as our srtt;
499 * move the current srtt into rttvar to keep the current
500 * retransmit times until then.
502 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
504 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
505 in6_losing(tp->t_inpcb);
508 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
511 tp->snd_nxt = tp->snd_una;
512 tp->snd_recover = tp->snd_max;
514 * Force a segment to be sent.
516 tp->t_flags |= TF_ACKNOW;
518 * If timing a segment in this window, stop the timer.
522 * Close the congestion window down to one segment
523 * (we'll open it by one segment for each ack we get).
524 * Since we probably have a window's worth of unacked
525 * data accumulated, this "slow start" keeps us from
526 * dumping all that data as back-to-back packets (which
527 * might overwhelm an intermediate gateway).
529 * There are two phases to the opening: Initially we
530 * open by one mss on each ack. This makes the window
531 * size increase exponentially with time. If the
532 * window is larger than the path can handle, this
533 * exponential growth results in dropped packet(s)
534 * almost immediately. To get more time between
535 * drops but still "push" the network to take advantage
536 * of improving conditions, we switch from exponential
537 * to linear window opening at some threshhold size.
538 * For a threshhold, we use half the current window
539 * size, truncated to a multiple of the mss.
541 * (the minimum cwnd that will give us exponential
542 * growth is 2 mss. We don't allow the threshhold
546 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
549 tp->snd_cwnd = tp->t_maxseg;
550 tp->snd_ssthresh = win * tp->t_maxseg;
553 EXIT_FASTRECOVERY(tp);
554 (void) tcp_output(tp);
558 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
559 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
565 INP_INFO_WUNLOCK(&tcbinfo);
569 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
571 struct callout *t_callout;
574 switch (timer_type) {
576 t_callout = &tp->t_timers->tt_delack;
577 f_callout = tcp_timer_delack;
580 t_callout = &tp->t_timers->tt_rexmt;
581 f_callout = tcp_timer_rexmt;
584 t_callout = &tp->t_timers->tt_persist;
585 f_callout = tcp_timer_persist;
588 t_callout = &tp->t_timers->tt_keep;
589 f_callout = tcp_timer_keep;
592 t_callout = &tp->t_timers->tt_2msl;
593 f_callout = tcp_timer_2msl;
596 panic("bad timer_type");
599 callout_stop(t_callout);
601 callout_reset(t_callout, delta, f_callout, tp);
606 tcp_timer_active(struct tcpcb *tp, int timer_type)
608 struct callout *t_callout;
610 switch (timer_type) {
612 t_callout = &tp->t_timers->tt_delack;
615 t_callout = &tp->t_timers->tt_rexmt;
618 t_callout = &tp->t_timers->tt_persist;
621 t_callout = &tp->t_timers->tt_keep;
624 t_callout = &tp->t_timers->tt_2msl;
627 panic("bad timer_type");
629 return callout_active(t_callout);