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>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/sysctl.h>
48 #include <sys/systm.h>
51 #include <net/route.h>
54 #include <netinet/cc.h>
55 #include <netinet/in.h>
56 #include <netinet/in_pcb.h>
57 #include <netinet/in_systm.h>
59 #include <netinet6/in6_pcb.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/tcp_fsm.h>
63 #include <netinet/tcp_timer.h>
64 #include <netinet/tcp_var.h>
65 #include <netinet/tcpip.h>
67 #include <netinet/tcp_debug.h>
71 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
72 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
75 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
76 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
80 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
84 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
85 "Time before a delayed ACK is sent");
88 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
89 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
92 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
93 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
94 "Minimum Retransmission Timeout");
97 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
98 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
99 "Retransmission Timer Slop");
101 static int always_keepalive = 1;
102 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
103 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
105 int tcp_fast_finwait2_recycle = 0;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
107 &tcp_fast_finwait2_recycle, 0,
108 "Recycle closed FIN_WAIT_2 connections faster");
110 int tcp_finwait2_timeout;
111 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
112 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
114 int tcp_keepcnt = TCPTV_KEEPCNT;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
116 "Number of keepalive probes to send");
118 /* max idle probes */
119 int tcp_maxpersistidle;
121 static int per_cpu_timers = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
123 &per_cpu_timers , 0, "run tcp timers on all cpus");
125 #define INP_CPU(inp) (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
126 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
129 * Tcp protocol timeout routine called every 500 ms.
130 * Updates timestamps used for TCP
131 * causes finite state machine actions if timers expire.
136 VNET_ITERATOR_DECL(vnet_iter);
138 VNET_LIST_RLOCK_NOSLEEP();
139 VNET_FOREACH(vnet_iter) {
140 CURVNET_SET(vnet_iter);
141 INP_INFO_WLOCK(&V_tcbinfo);
142 (void) tcp_tw_2msl_scan(0);
143 INP_INFO_WUNLOCK(&V_tcbinfo);
146 VNET_LIST_RUNLOCK_NOSLEEP();
149 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
150 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
152 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
153 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
155 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
157 static int tcp_timer_race;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
159 0, "Count of t_inpcb races on tcp_discardcb");
162 * TCP timer processing.
166 tcp_timer_delack(void *xtp)
168 struct tcpcb *tp = xtp;
170 CURVNET_SET(tp->t_vnet);
174 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
175 * tear-down mean we need it as a work-around for races between
176 * timers and tcp_discardcb().
178 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
186 if (callout_pending(&tp->t_timers->tt_delack) ||
187 !callout_active(&tp->t_timers->tt_delack)) {
192 callout_deactivate(&tp->t_timers->tt_delack);
193 if ((inp->inp_flags & INP_DROPPED) != 0) {
199 tp->t_flags |= TF_ACKNOW;
200 TCPSTAT_INC(tcps_delack);
201 (void) tcp_output(tp);
207 tcp_timer_2msl(void *xtp)
209 struct tcpcb *tp = xtp;
211 CURVNET_SET(tp->t_vnet);
215 ostate = tp->t_state;
218 * XXXRW: Does this actually happen?
220 INP_INFO_WLOCK(&V_tcbinfo);
223 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
224 * tear-down mean we need it as a work-around for races between
225 * timers and tcp_discardcb().
227 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
231 INP_INFO_WUNLOCK(&V_tcbinfo);
236 tcp_free_sackholes(tp);
237 if (callout_pending(&tp->t_timers->tt_2msl) ||
238 !callout_active(&tp->t_timers->tt_2msl)) {
239 INP_WUNLOCK(tp->t_inpcb);
240 INP_INFO_WUNLOCK(&V_tcbinfo);
244 callout_deactivate(&tp->t_timers->tt_2msl);
245 if ((inp->inp_flags & INP_DROPPED) != 0) {
247 INP_INFO_WUNLOCK(&V_tcbinfo);
252 * 2 MSL timeout in shutdown went off. If we're closed but
253 * still waiting for peer to close and connection has been idle
254 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
255 * control block. Otherwise, check again in a bit.
257 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
258 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
259 * Ignore fact that there were recent incoming segments.
261 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
262 tp->t_inpcb && tp->t_inpcb->inp_socket &&
263 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
264 TCPSTAT_INC(tcps_finwait2_drops);
267 if (tp->t_state != TCPS_TIME_WAIT &&
268 ticks - tp->t_rcvtime <= TP_MAXIDLE(tp))
269 callout_reset_on(&tp->t_timers->tt_2msl,
270 TP_KEEPINTVL(tp), tcp_timer_2msl, tp, INP_CPU(inp));
276 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
277 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
282 INP_INFO_WUNLOCK(&V_tcbinfo);
287 tcp_timer_keep(void *xtp)
289 struct tcpcb *tp = xtp;
290 struct tcptemp *t_template;
292 CURVNET_SET(tp->t_vnet);
296 ostate = tp->t_state;
298 INP_INFO_WLOCK(&V_tcbinfo);
301 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
302 * tear-down mean we need it as a work-around for races between
303 * timers and tcp_discardcb().
305 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
309 INP_INFO_WUNLOCK(&V_tcbinfo);
314 if (callout_pending(&tp->t_timers->tt_keep) ||
315 !callout_active(&tp->t_timers->tt_keep)) {
317 INP_INFO_WUNLOCK(&V_tcbinfo);
321 callout_deactivate(&tp->t_timers->tt_keep);
322 if ((inp->inp_flags & INP_DROPPED) != 0) {
324 INP_INFO_WUNLOCK(&V_tcbinfo);
329 * Keep-alive timer went off; send something
330 * or drop connection if idle for too long.
332 TCPSTAT_INC(tcps_keeptimeo);
333 if (tp->t_state < TCPS_ESTABLISHED)
335 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
336 tp->t_state <= TCPS_CLOSING) {
337 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
340 * Send a packet designed to force a response
341 * if the peer is up and reachable:
342 * either an ACK if the connection is still alive,
343 * or an RST if the peer has closed the connection
344 * due to timeout or reboot.
345 * Using sequence number tp->snd_una-1
346 * causes the transmitted zero-length segment
347 * to lie outside the receive window;
348 * by the protocol spec, this requires the
349 * correspondent TCP to respond.
351 TCPSTAT_INC(tcps_keepprobe);
352 t_template = tcpip_maketemplate(inp);
354 tcp_respond(tp, t_template->tt_ipgen,
355 &t_template->tt_t, (struct mbuf *)NULL,
356 tp->rcv_nxt, tp->snd_una - 1, 0);
357 free(t_template, M_TEMP);
359 callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
360 tcp_timer_keep, tp, INP_CPU(inp));
362 callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
363 tcp_timer_keep, tp, INP_CPU(inp));
366 if (inp->inp_socket->so_options & SO_DEBUG)
367 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
371 INP_INFO_WUNLOCK(&V_tcbinfo);
376 TCPSTAT_INC(tcps_keepdrops);
377 tp = tcp_drop(tp, ETIMEDOUT);
380 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
381 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
385 INP_WUNLOCK(tp->t_inpcb);
386 INP_INFO_WUNLOCK(&V_tcbinfo);
391 tcp_timer_persist(void *xtp)
393 struct tcpcb *tp = xtp;
395 CURVNET_SET(tp->t_vnet);
399 ostate = tp->t_state;
401 INP_INFO_WLOCK(&V_tcbinfo);
404 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
405 * tear-down mean we need it as a work-around for races between
406 * timers and tcp_discardcb().
408 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
412 INP_INFO_WUNLOCK(&V_tcbinfo);
417 if (callout_pending(&tp->t_timers->tt_persist) ||
418 !callout_active(&tp->t_timers->tt_persist)) {
420 INP_INFO_WUNLOCK(&V_tcbinfo);
424 callout_deactivate(&tp->t_timers->tt_persist);
425 if ((inp->inp_flags & INP_DROPPED) != 0) {
427 INP_INFO_WUNLOCK(&V_tcbinfo);
432 * Persistance timer into zero window.
433 * Force a byte to be output, if possible.
435 TCPSTAT_INC(tcps_persisttimeo);
437 * Hack: if the peer is dead/unreachable, we do not
438 * time out if the window is closed. After a full
439 * backoff, drop the connection if the idle time
440 * (no responses to probes) reaches the maximum
441 * backoff that we would use if retransmitting.
443 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
444 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
445 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
446 TCPSTAT_INC(tcps_persistdrop);
447 tp = tcp_drop(tp, ETIMEDOUT);
451 tp->t_flags |= TF_FORCEDATA;
452 (void) tcp_output(tp);
453 tp->t_flags &= ~TF_FORCEDATA;
457 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
458 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
462 INP_INFO_WUNLOCK(&V_tcbinfo);
467 tcp_timer_rexmt(void * xtp)
469 struct tcpcb *tp = xtp;
470 CURVNET_SET(tp->t_vnet);
477 ostate = tp->t_state;
479 INP_INFO_RLOCK(&V_tcbinfo);
482 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
483 * tear-down mean we need it as a work-around for races between
484 * timers and tcp_discardcb().
486 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
490 INP_INFO_RUNLOCK(&V_tcbinfo);
495 if (callout_pending(&tp->t_timers->tt_rexmt) ||
496 !callout_active(&tp->t_timers->tt_rexmt)) {
498 INP_INFO_RUNLOCK(&V_tcbinfo);
502 callout_deactivate(&tp->t_timers->tt_rexmt);
503 if ((inp->inp_flags & INP_DROPPED) != 0) {
505 INP_INFO_RUNLOCK(&V_tcbinfo);
509 tcp_free_sackholes(tp);
511 * Retransmission timer went off. Message has not
512 * been acked within retransmit interval. Back off
513 * to a longer retransmit interval and retransmit one segment.
515 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
516 tp->t_rxtshift = TCP_MAXRXTSHIFT;
517 TCPSTAT_INC(tcps_timeoutdrop);
519 INP_INFO_RUNLOCK(&V_tcbinfo);
521 INP_INFO_WLOCK(&V_tcbinfo);
523 if (in_pcbrele_wlocked(inp)) {
524 INP_INFO_WUNLOCK(&V_tcbinfo);
528 if (inp->inp_flags & INP_DROPPED) {
530 INP_INFO_WUNLOCK(&V_tcbinfo);
535 tp = tcp_drop(tp, tp->t_softerror ?
536 tp->t_softerror : ETIMEDOUT);
540 INP_INFO_RUNLOCK(&V_tcbinfo);
542 if (tp->t_rxtshift == 1) {
544 * first retransmit; record ssthresh and cwnd so they can
545 * be recovered if this turns out to be a "bad" retransmit.
546 * A retransmit is considered "bad" if an ACK for this
547 * segment is received within RTT/2 interval; the assumption
548 * here is that the ACK was already in flight. See
549 * "On Estimating End-to-End Network Path Properties" by
550 * Allman and Paxson for more details.
552 tp->snd_cwnd_prev = tp->snd_cwnd;
553 tp->snd_ssthresh_prev = tp->snd_ssthresh;
554 tp->snd_recover_prev = tp->snd_recover;
555 if (IN_FASTRECOVERY(tp->t_flags))
556 tp->t_flags |= TF_WASFRECOVERY;
558 tp->t_flags &= ~TF_WASFRECOVERY;
559 if (IN_CONGRECOVERY(tp->t_flags))
560 tp->t_flags |= TF_WASCRECOVERY;
562 tp->t_flags &= ~TF_WASCRECOVERY;
563 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
564 tp->t_flags |= TF_PREVVALID;
566 tp->t_flags &= ~TF_PREVVALID;
567 TCPSTAT_INC(tcps_rexmttimeo);
568 if (tp->t_state == TCPS_SYN_SENT)
569 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
571 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
572 TCPT_RANGESET(tp->t_rxtcur, rexmt,
573 tp->t_rttmin, TCPTV_REXMTMAX);
575 * Disable rfc1323 if we haven't got any response to
576 * our third SYN to work-around some broken terminal servers
577 * (most of which have hopefully been retired) that have bad VJ
578 * header compression code which trashes TCP segments containing
579 * unknown-to-them TCP options.
581 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
582 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
584 * If we backed off this far, our srtt estimate is probably bogus.
585 * Clobber it so we'll take the next rtt measurement as our srtt;
586 * move the current srtt into rttvar to keep the current
587 * retransmit times until then.
589 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
591 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
592 in6_losing(tp->t_inpcb);
595 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
598 tp->snd_nxt = tp->snd_una;
599 tp->snd_recover = tp->snd_max;
601 * Force a segment to be sent.
603 tp->t_flags |= TF_ACKNOW;
605 * If timing a segment in this window, stop the timer.
609 cc_cong_signal(tp, NULL, CC_RTO);
611 (void) tcp_output(tp);
615 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
616 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
622 INP_INFO_WUNLOCK(&V_tcbinfo);
627 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
629 struct callout *t_callout;
631 struct inpcb *inp = tp->t_inpcb;
632 int cpu = INP_CPU(inp);
635 if (tp->t_flags & TF_TOE)
639 switch (timer_type) {
641 t_callout = &tp->t_timers->tt_delack;
642 f_callout = tcp_timer_delack;
645 t_callout = &tp->t_timers->tt_rexmt;
646 f_callout = tcp_timer_rexmt;
649 t_callout = &tp->t_timers->tt_persist;
650 f_callout = tcp_timer_persist;
653 t_callout = &tp->t_timers->tt_keep;
654 f_callout = tcp_timer_keep;
657 t_callout = &tp->t_timers->tt_2msl;
658 f_callout = tcp_timer_2msl;
661 panic("bad timer_type");
664 callout_stop(t_callout);
666 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
671 tcp_timer_active(struct tcpcb *tp, int timer_type)
673 struct callout *t_callout;
675 switch (timer_type) {
677 t_callout = &tp->t_timers->tt_delack;
680 t_callout = &tp->t_timers->tt_rexmt;
683 t_callout = &tp->t_timers->tt_persist;
686 t_callout = &tp->t_timers->tt_keep;
689 t_callout = &tp->t_timers->tt_2msl;
692 panic("bad timer_type");
694 return callout_active(t_callout);
697 #define ticks_to_msecs(t) (1000*(t) / hz)
700 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer, struct xtcp_timer *xtimer)
702 bzero(xtimer, sizeof(struct xtcp_timer));
705 if (callout_active(&timer->tt_delack))
706 xtimer->tt_delack = ticks_to_msecs(timer->tt_delack.c_time - ticks);
707 if (callout_active(&timer->tt_rexmt))
708 xtimer->tt_rexmt = ticks_to_msecs(timer->tt_rexmt.c_time - ticks);
709 if (callout_active(&timer->tt_persist))
710 xtimer->tt_persist = ticks_to_msecs(timer->tt_persist.c_time - ticks);
711 if (callout_active(&timer->tt_keep))
712 xtimer->tt_keep = ticks_to_msecs(timer->tt_keep.c_time - ticks);
713 if (callout_active(&timer->tt_2msl))
714 xtimer->tt_2msl = ticks_to_msecs(timer->tt_2msl.c_time - ticks);
715 xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);