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>
48 #include <sys/vimage.h>
51 #include <net/route.h>
53 #include <netinet/in.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/in_systm.h>
57 #include <netinet6/in6_pcb.h>
59 #include <netinet/ip_var.h>
60 #include <netinet/tcp.h>
61 #include <netinet/tcp_fsm.h>
62 #include <netinet/tcp_timer.h>
63 #include <netinet/tcp_var.h>
64 #include <netinet/tcpip.h>
66 #include <netinet/tcp_debug.h>
68 #include <netinet/vinet.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 */
122 * Tcp protocol timeout routine called every 500 ms.
123 * Updates timestamps used for TCP
124 * causes finite state machine actions if timers expire.
129 VNET_ITERATOR_DECL(vnet_iter);
132 VNET_FOREACH(vnet_iter) {
133 CURVNET_SET(vnet_iter);
134 INIT_VNET_INET(vnet_iter);
135 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
136 INP_INFO_WLOCK(&V_tcbinfo);
137 (void) tcp_tw_2msl_scan(0);
138 INP_INFO_WUNLOCK(&V_tcbinfo);
144 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
145 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
147 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
148 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
150 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
152 static int tcp_timer_race;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
154 0, "Count of t_inpcb races on tcp_discardcb");
157 * TCP timer processing.
161 tcp_timer_delack(void *xtp)
163 struct tcpcb *tp = xtp;
165 CURVNET_SET(tp->t_vnet);
166 INIT_VNET_INET(tp->t_vnet);
168 INP_INFO_RLOCK(&V_tcbinfo);
171 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
172 * tear-down mean we need it as a work-around for races between
173 * timers and tcp_discardcb().
175 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
179 INP_INFO_RUNLOCK(&V_tcbinfo);
184 INP_INFO_RUNLOCK(&V_tcbinfo);
185 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack)
186 || !callout_active(&tp->t_timers->tt_delack)) {
191 callout_deactivate(&tp->t_timers->tt_delack);
193 tp->t_flags |= TF_ACKNOW;
194 V_tcpstat.tcps_delack++;
195 (void) tcp_output(tp);
201 tcp_timer_2msl(void *xtp)
203 struct tcpcb *tp = xtp;
205 CURVNET_SET(tp->t_vnet);
206 INIT_VNET_INET(tp->t_vnet);
210 ostate = tp->t_state;
213 * XXXRW: Does this actually happen?
215 INP_INFO_WLOCK(&V_tcbinfo);
218 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
219 * tear-down mean we need it as a work-around for races between
220 * timers and tcp_discardcb().
222 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
226 INP_INFO_WUNLOCK(&V_tcbinfo);
231 tcp_free_sackholes(tp);
232 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) ||
233 !callout_active(&tp->t_timers->tt_2msl)) {
234 INP_WUNLOCK(tp->t_inpcb);
235 INP_INFO_WUNLOCK(&V_tcbinfo);
239 callout_deactivate(&tp->t_timers->tt_2msl);
241 * 2 MSL timeout in shutdown went off. If we're closed but
242 * still waiting for peer to close and connection has been idle
243 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
244 * control block. Otherwise, check again in a bit.
246 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
247 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
248 * Ignore fact that there were recent incoming segments.
250 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
251 tp->t_inpcb && tp->t_inpcb->inp_socket &&
252 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
253 V_tcpstat.tcps_finwait2_drops++;
256 if (tp->t_state != TCPS_TIME_WAIT &&
257 (ticks - tp->t_rcvtime) <= tcp_maxidle)
258 callout_reset(&tp->t_timers->tt_2msl, tcp_keepintvl,
265 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
266 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
271 INP_INFO_WUNLOCK(&V_tcbinfo);
276 tcp_timer_keep(void *xtp)
278 struct tcpcb *tp = xtp;
279 struct tcptemp *t_template;
281 CURVNET_SET(tp->t_vnet);
282 INIT_VNET_INET(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 V_tcpstat.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 V_tcpstat.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(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
345 callout_reset(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
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 V_tcpstat.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);
378 INIT_VNET_INET(tp->t_vnet);
382 ostate = tp->t_state;
384 INP_INFO_WLOCK(&V_tcbinfo);
387 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
388 * tear-down mean we need it as a work-around for races between
389 * timers and tcp_discardcb().
391 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
395 INP_INFO_WUNLOCK(&V_tcbinfo);
400 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist)
401 || !callout_active(&tp->t_timers->tt_persist)) {
403 INP_INFO_WUNLOCK(&V_tcbinfo);
407 callout_deactivate(&tp->t_timers->tt_persist);
409 * Persistance timer into zero window.
410 * Force a byte to be output, if possible.
412 V_tcpstat.tcps_persisttimeo++;
414 * Hack: if the peer is dead/unreachable, we do not
415 * time out if the window is closed. After a full
416 * backoff, drop the connection if the idle time
417 * (no responses to probes) reaches the maximum
418 * backoff that we would use if retransmitting.
420 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
421 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
422 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
423 V_tcpstat.tcps_persistdrop++;
424 tp = tcp_drop(tp, ETIMEDOUT);
428 tp->t_flags |= TF_FORCEDATA;
429 (void) tcp_output(tp);
430 tp->t_flags &= ~TF_FORCEDATA;
434 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
435 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
439 INP_INFO_WUNLOCK(&V_tcbinfo);
444 tcp_timer_rexmt(void * xtp)
446 struct tcpcb *tp = xtp;
447 CURVNET_SET(tp->t_vnet);
448 INIT_VNET_INET(tp->t_vnet);
455 ostate = tp->t_state;
457 INP_INFO_WLOCK(&V_tcbinfo);
461 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
462 * tear-down mean we need it as a work-around for races between
463 * timers and tcp_discardcb().
465 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
469 INP_INFO_WUNLOCK(&V_tcbinfo);
474 if ((inp->inp_flags & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt)
475 || !callout_active(&tp->t_timers->tt_rexmt)) {
477 INP_INFO_WUNLOCK(&V_tcbinfo);
481 callout_deactivate(&tp->t_timers->tt_rexmt);
482 tcp_free_sackholes(tp);
484 * Retransmission timer went off. Message has not
485 * been acked within retransmit interval. Back off
486 * to a longer retransmit interval and retransmit one segment.
488 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
489 tp->t_rxtshift = TCP_MAXRXTSHIFT;
490 V_tcpstat.tcps_timeoutdrop++;
491 tp = tcp_drop(tp, tp->t_softerror ?
492 tp->t_softerror : ETIMEDOUT);
495 INP_INFO_WUNLOCK(&V_tcbinfo);
497 if (tp->t_rxtshift == 1) {
499 * first retransmit; record ssthresh and cwnd so they can
500 * be recovered if this turns out to be a "bad" retransmit.
501 * A retransmit is considered "bad" if an ACK for this
502 * segment is received within RTT/2 interval; the assumption
503 * here is that the ACK was already in flight. See
504 * "On Estimating End-to-End Network Path Properties" by
505 * Allman and Paxson for more details.
507 tp->snd_cwnd_prev = tp->snd_cwnd;
508 tp->snd_ssthresh_prev = tp->snd_ssthresh;
509 tp->snd_recover_prev = tp->snd_recover;
510 if (IN_FASTRECOVERY(tp))
511 tp->t_flags |= TF_WASFRECOVERY;
513 tp->t_flags &= ~TF_WASFRECOVERY;
514 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
516 V_tcpstat.tcps_rexmttimeo++;
517 if (tp->t_state == TCPS_SYN_SENT)
518 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
520 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
521 TCPT_RANGESET(tp->t_rxtcur, rexmt,
522 tp->t_rttmin, TCPTV_REXMTMAX);
524 * Disable rfc1323 if we havn't got any response to
525 * our third SYN to work-around some broken terminal servers
526 * (most of which have hopefully been retired) that have bad VJ
527 * header compression code which trashes TCP segments containing
528 * unknown-to-them TCP options.
530 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
531 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
533 * If we backed off this far, our srtt estimate is probably bogus.
534 * Clobber it so we'll take the next rtt measurement as our srtt;
535 * move the current srtt into rttvar to keep the current
536 * retransmit times until then.
538 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
540 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
541 in6_losing(tp->t_inpcb);
544 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
547 tp->snd_nxt = tp->snd_una;
548 tp->snd_recover = tp->snd_max;
550 * Force a segment to be sent.
552 tp->t_flags |= TF_ACKNOW;
554 * If timing a segment in this window, stop the timer.
558 * Close the congestion window down to one segment
559 * (we'll open it by one segment for each ack we get).
560 * Since we probably have a window's worth of unacked
561 * data accumulated, this "slow start" keeps us from
562 * dumping all that data as back-to-back packets (which
563 * might overwhelm an intermediate gateway).
565 * There are two phases to the opening: Initially we
566 * open by one mss on each ack. This makes the window
567 * size increase exponentially with time. If the
568 * window is larger than the path can handle, this
569 * exponential growth results in dropped packet(s)
570 * almost immediately. To get more time between
571 * drops but still "push" the network to take advantage
572 * of improving conditions, we switch from exponential
573 * to linear window opening at some threshhold size.
574 * For a threshhold, we use half the current window
575 * size, truncated to a multiple of the mss.
577 * (the minimum cwnd that will give us exponential
578 * growth is 2 mss. We don't allow the threshhold
582 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
585 tp->snd_cwnd = tp->t_maxseg;
586 tp->snd_ssthresh = win * tp->t_maxseg;
589 EXIT_FASTRECOVERY(tp);
590 tp->t_bytes_acked = 0;
591 (void) tcp_output(tp);
595 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
596 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
602 INP_INFO_WUNLOCK(&V_tcbinfo);
607 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
609 struct callout *t_callout;
612 switch (timer_type) {
614 t_callout = &tp->t_timers->tt_delack;
615 f_callout = tcp_timer_delack;
618 t_callout = &tp->t_timers->tt_rexmt;
619 f_callout = tcp_timer_rexmt;
622 t_callout = &tp->t_timers->tt_persist;
623 f_callout = tcp_timer_persist;
626 t_callout = &tp->t_timers->tt_keep;
627 f_callout = tcp_timer_keep;
630 t_callout = &tp->t_timers->tt_2msl;
631 f_callout = tcp_timer_2msl;
634 panic("bad timer_type");
637 callout_stop(t_callout);
639 callout_reset(t_callout, delta, f_callout, tp);
644 tcp_timer_active(struct tcpcb *tp, int timer_type)
646 struct callout *t_callout;
648 switch (timer_type) {
650 t_callout = &tp->t_timers->tt_delack;
653 t_callout = &tp->t_timers->tt_rexmt;
656 t_callout = &tp->t_timers->tt_persist;
659 t_callout = &tp->t_timers->tt_keep;
662 t_callout = &tp->t_timers->tt_2msl;
665 panic("bad timer_type");
667 return callout_active(t_callout);