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");
115 static int tcp_keepcnt = TCPTV_KEEPCNT;
116 /* max idle probes */
117 int tcp_maxpersistidle;
118 /* max idle time in persist */
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 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
142 INP_INFO_WLOCK(&V_tcbinfo);
143 (void) tcp_tw_2msl_scan(0);
144 INP_INFO_WUNLOCK(&V_tcbinfo);
147 VNET_LIST_RUNLOCK_NOSLEEP();
150 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
151 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
153 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
154 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
156 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
158 static int tcp_timer_race;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
160 0, "Count of t_inpcb races on tcp_discardcb");
163 * TCP timer processing.
167 tcp_timer_delack(void *xtp)
169 struct tcpcb *tp = xtp;
171 CURVNET_SET(tp->t_vnet);
175 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
176 * tear-down mean we need it as a work-around for races between
177 * timers and tcp_discardcb().
179 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
187 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack)
188 || !callout_active(&tp->t_timers->tt_delack)) {
193 callout_deactivate(&tp->t_timers->tt_delack);
195 tp->t_flags |= TF_ACKNOW;
196 TCPSTAT_INC(tcps_delack);
197 (void) tcp_output(tp);
203 tcp_timer_2msl(void *xtp)
205 struct tcpcb *tp = xtp;
207 CURVNET_SET(tp->t_vnet);
211 ostate = tp->t_state;
214 * XXXRW: Does this actually happen?
216 INP_INFO_WLOCK(&V_tcbinfo);
219 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
220 * tear-down mean we need it as a work-around for races between
221 * timers and tcp_discardcb().
223 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
227 INP_INFO_WUNLOCK(&V_tcbinfo);
232 tcp_free_sackholes(tp);
233 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) ||
234 !callout_active(&tp->t_timers->tt_2msl)) {
235 INP_WUNLOCK(tp->t_inpcb);
236 INP_INFO_WUNLOCK(&V_tcbinfo);
240 callout_deactivate(&tp->t_timers->tt_2msl);
242 * 2 MSL timeout in shutdown went off. If we're closed but
243 * still waiting for peer to close and connection has been idle
244 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
245 * control block. Otherwise, check again in a bit.
247 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
248 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
249 * Ignore fact that there were recent incoming segments.
251 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
252 tp->t_inpcb && tp->t_inpcb->inp_socket &&
253 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
254 TCPSTAT_INC(tcps_finwait2_drops);
257 if (tp->t_state != TCPS_TIME_WAIT &&
258 ticks - tp->t_rcvtime <= tcp_maxidle)
259 callout_reset_on(&tp->t_timers->tt_2msl, tcp_keepintvl,
260 tcp_timer_2msl, tp, INP_CPU(inp));
266 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
267 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
272 INP_INFO_WUNLOCK(&V_tcbinfo);
277 tcp_timer_keep(void *xtp)
279 struct tcpcb *tp = xtp;
280 struct tcptemp *t_template;
282 CURVNET_SET(tp->t_vnet);
286 ostate = tp->t_state;
288 INP_INFO_WLOCK(&V_tcbinfo);
291 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
292 * tear-down mean we need it as a work-around for races between
293 * timers and tcp_discardcb().
295 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
299 INP_INFO_WUNLOCK(&V_tcbinfo);
304 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep)
305 || !callout_active(&tp->t_timers->tt_keep)) {
307 INP_INFO_WUNLOCK(&V_tcbinfo);
311 callout_deactivate(&tp->t_timers->tt_keep);
313 * Keep-alive timer went off; send something
314 * or drop connection if idle for too long.
316 TCPSTAT_INC(tcps_keeptimeo);
317 if (tp->t_state < TCPS_ESTABLISHED)
319 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
320 tp->t_state <= TCPS_CLOSING) {
321 if (ticks - tp->t_rcvtime >= tcp_keepidle + tcp_maxidle)
324 * Send a packet designed to force a response
325 * if the peer is up and reachable:
326 * either an ACK if the connection is still alive,
327 * or an RST if the peer has closed the connection
328 * due to timeout or reboot.
329 * Using sequence number tp->snd_una-1
330 * causes the transmitted zero-length segment
331 * to lie outside the receive window;
332 * by the protocol spec, this requires the
333 * correspondent TCP to respond.
335 TCPSTAT_INC(tcps_keepprobe);
336 t_template = tcpip_maketemplate(inp);
338 tcp_respond(tp, t_template->tt_ipgen,
339 &t_template->tt_t, (struct mbuf *)NULL,
340 tp->rcv_nxt, tp->snd_una - 1, 0);
341 free(t_template, M_TEMP);
343 callout_reset_on(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp, INP_CPU(inp));
345 callout_reset_on(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp, INP_CPU(inp));
348 if (inp->inp_socket->so_options & SO_DEBUG)
349 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
353 INP_INFO_WUNLOCK(&V_tcbinfo);
358 TCPSTAT_INC(tcps_keepdrops);
359 tp = tcp_drop(tp, ETIMEDOUT);
362 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
363 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
367 INP_WUNLOCK(tp->t_inpcb);
368 INP_INFO_WUNLOCK(&V_tcbinfo);
373 tcp_timer_persist(void *xtp)
375 struct tcpcb *tp = xtp;
377 CURVNET_SET(tp->t_vnet);
381 ostate = tp->t_state;
383 INP_INFO_WLOCK(&V_tcbinfo);
386 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
387 * tear-down mean we need it as a work-around for races between
388 * timers and tcp_discardcb().
390 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
394 INP_INFO_WUNLOCK(&V_tcbinfo);
399 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
400 || !callout_active(&tp->t_timers->tt_persist)) {
402 INP_INFO_WUNLOCK(&V_tcbinfo);
406 callout_deactivate(&tp->t_timers->tt_persist);
408 * Persistance timer into zero window.
409 * Force a byte to be output, if possible.
411 TCPSTAT_INC(tcps_persisttimeo);
413 * Hack: if the peer is dead/unreachable, we do not
414 * time out if the window is closed. After a full
415 * backoff, drop the connection if the idle time
416 * (no responses to probes) reaches the maximum
417 * backoff that we would use if retransmitting.
419 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
420 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
421 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
422 TCPSTAT_INC(tcps_persistdrop);
423 tp = tcp_drop(tp, ETIMEDOUT);
427 tp->t_flags |= TF_FORCEDATA;
428 (void) tcp_output(tp);
429 tp->t_flags &= ~TF_FORCEDATA;
433 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
434 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
438 INP_INFO_WUNLOCK(&V_tcbinfo);
443 tcp_timer_rexmt(void * xtp)
445 struct tcpcb *tp = xtp;
446 CURVNET_SET(tp->t_vnet);
453 ostate = tp->t_state;
455 INP_INFO_RLOCK(&V_tcbinfo);
458 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
459 * tear-down mean we need it as a work-around for races between
460 * timers and tcp_discardcb().
462 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
466 INP_INFO_RUNLOCK(&V_tcbinfo);
471 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt)
472 || !callout_active(&tp->t_timers->tt_rexmt)) {
474 INP_INFO_RUNLOCK(&V_tcbinfo);
478 callout_deactivate(&tp->t_timers->tt_rexmt);
479 tcp_free_sackholes(tp);
481 * Retransmission timer went off. Message has not
482 * been acked within retransmit interval. Back off
483 * to a longer retransmit interval and retransmit one segment.
485 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
486 tp->t_rxtshift = TCP_MAXRXTSHIFT;
487 TCPSTAT_INC(tcps_timeoutdrop);
489 INP_INFO_RUNLOCK(&V_tcbinfo);
491 INP_INFO_WLOCK(&V_tcbinfo);
493 if (in_pcbrele(inp)) {
494 INP_INFO_WUNLOCK(&V_tcbinfo);
498 tp = tcp_drop(tp, tp->t_softerror ?
499 tp->t_softerror : ETIMEDOUT);
503 INP_INFO_RUNLOCK(&V_tcbinfo);
505 if (tp->t_rxtshift == 1) {
507 * first retransmit; record ssthresh and cwnd so they can
508 * be recovered if this turns out to be a "bad" retransmit.
509 * A retransmit is considered "bad" if an ACK for this
510 * segment is received within RTT/2 interval; the assumption
511 * here is that the ACK was already in flight. See
512 * "On Estimating End-to-End Network Path Properties" by
513 * Allman and Paxson for more details.
515 tp->snd_cwnd_prev = tp->snd_cwnd;
516 tp->snd_ssthresh_prev = tp->snd_ssthresh;
517 tp->snd_recover_prev = tp->snd_recover;
518 if (IN_FASTRECOVERY(tp->t_flags))
519 tp->t_flags |= TF_WASFRECOVERY;
521 tp->t_flags &= ~TF_WASFRECOVERY;
522 if (IN_CONGRECOVERY(tp->t_flags))
523 tp->t_flags |= TF_WASCRECOVERY;
525 tp->t_flags &= ~TF_WASCRECOVERY;
526 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
527 tp->t_flags |= TF_PREVVALID;
529 tp->t_flags &= ~TF_PREVVALID;
530 TCPSTAT_INC(tcps_rexmttimeo);
531 if (tp->t_state == TCPS_SYN_SENT)
532 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
534 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
535 TCPT_RANGESET(tp->t_rxtcur, rexmt,
536 tp->t_rttmin, TCPTV_REXMTMAX);
538 * Disable rfc1323 if we haven't got any response to
539 * our third SYN to work-around some broken terminal servers
540 * (most of which have hopefully been retired) that have bad VJ
541 * header compression code which trashes TCP segments containing
542 * unknown-to-them TCP options.
544 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
545 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
547 * If we backed off this far, our srtt estimate is probably bogus.
548 * Clobber it so we'll take the next rtt measurement as our srtt;
549 * move the current srtt into rttvar to keep the current
550 * retransmit times until then.
552 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
554 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
555 in6_losing(tp->t_inpcb);
558 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
561 tp->snd_nxt = tp->snd_una;
562 tp->snd_recover = tp->snd_max;
564 * Force a segment to be sent.
566 tp->t_flags |= TF_ACKNOW;
568 * If timing a segment in this window, stop the timer.
572 cc_cong_signal(tp, NULL, CC_RTO);
574 (void) tcp_output(tp);
578 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
579 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
585 INP_INFO_WUNLOCK(&V_tcbinfo);
590 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
592 struct callout *t_callout;
594 struct inpcb *inp = tp->t_inpcb;
595 int cpu = INP_CPU(inp);
597 switch (timer_type) {
599 t_callout = &tp->t_timers->tt_delack;
600 f_callout = tcp_timer_delack;
603 t_callout = &tp->t_timers->tt_rexmt;
604 f_callout = tcp_timer_rexmt;
607 t_callout = &tp->t_timers->tt_persist;
608 f_callout = tcp_timer_persist;
611 t_callout = &tp->t_timers->tt_keep;
612 f_callout = tcp_timer_keep;
615 t_callout = &tp->t_timers->tt_2msl;
616 f_callout = tcp_timer_2msl;
619 panic("bad timer_type");
622 callout_stop(t_callout);
624 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
629 tcp_timer_active(struct tcpcb *tp, int timer_type)
631 struct callout *t_callout;
633 switch (timer_type) {
635 t_callout = &tp->t_timers->tt_delack;
638 t_callout = &tp->t_timers->tt_rexmt;
641 t_callout = &tp->t_timers->tt_persist;
644 t_callout = &tp->t_timers->tt_keep;
647 t_callout = &tp->t_timers->tt_2msl;
650 panic("bad timer_type");
652 return callout_active(t_callout);
655 #define ticks_to_msecs(t) (1000*(t) / hz)
658 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer, struct xtcp_timer *xtimer)
660 bzero(xtimer, sizeof(struct xtcp_timer));
663 if (callout_active(&timer->tt_delack))
664 xtimer->tt_delack = ticks_to_msecs(timer->tt_delack.c_time - ticks);
665 if (callout_active(&timer->tt_rexmt))
666 xtimer->tt_rexmt = ticks_to_msecs(timer->tt_rexmt.c_time - ticks);
667 if (callout_active(&timer->tt_persist))
668 xtimer->tt_persist = ticks_to_msecs(timer->tt_persist.c_time - ticks);
669 if (callout_active(&timer->tt_keep))
670 xtimer->tt_keep = ticks_to_msecs(timer->tt_keep.c_time - ticks);
671 if (callout_active(&timer->tt_2msl))
672 xtimer->tt_2msl = ticks_to_msecs(timer->tt_2msl.c_time - ticks);
673 xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);