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
33 #include "opt_inet6.h"
34 #include "opt_tcpdebug.h"
35 #include "opt_tcp_sack.h"
37 #include <sys/param.h>
38 #include <sys/kernel.h>
41 #include <sys/mutex.h>
42 #include <sys/protosw.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/sysctl.h>
46 #include <sys/systm.h>
48 #include <net/route.h>
50 #include <netinet/in.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet/in_systm.h>
54 #include <netinet6/in6_pcb.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/tcp.h>
58 #include <netinet/tcp_fsm.h>
59 #include <netinet/tcp_timer.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/tcpip.h>
63 #include <netinet/tcp_debug.h>
67 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
68 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
71 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
72 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
75 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
76 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
80 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
81 "Time before a delayed ACK is sent");
84 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
85 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
88 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
89 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
92 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
93 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
95 static int always_keepalive = 1;
96 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
97 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
99 int tcp_fast_finwait2_recycle = 0;
100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
101 &tcp_fast_finwait2_recycle, 0, "Recycle closed FIN_WAIT_2 connections faster");
103 int tcp_finwait2_timeout;
104 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
105 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "");
108 static int tcp_keepcnt = TCPTV_KEEPCNT;
109 /* max idle probes */
110 int tcp_maxpersistidle;
111 /* max idle time in persist */
115 * Tcp protocol timeout routine called every 500 ms.
116 * Updates timestamps used for TCP
117 * causes finite state machine actions if timers expire.
123 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
124 INP_INFO_WLOCK(&tcbinfo);
125 (void) tcp_timer_2msl_tw(0);
126 INP_INFO_WUNLOCK(&tcbinfo);
129 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
130 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
132 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
133 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
135 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
137 static int tcp_timer_race;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
139 0, "Count of t_inpcb races on tcp_discardcb");
142 * TCP timer processing.
146 tcp_timer_delack(xtp)
149 struct tcpcb *tp = xtp;
152 INP_INFO_RLOCK(&tcbinfo);
155 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
156 * tear-down mean we need it as a work-around for races between
157 * timers and tcp_discardcb().
159 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
163 INP_INFO_RUNLOCK(&tcbinfo);
167 INP_INFO_RUNLOCK(&tcbinfo);
168 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_delack)
169 || !callout_active(tp->tt_delack)) {
173 callout_deactivate(tp->tt_delack);
175 tp->t_flags |= TF_ACKNOW;
176 tcpstat.tcps_delack++;
177 (void) tcp_output(tp);
185 struct tcpcb *tp = xtp;
190 ostate = tp->t_state;
193 * XXXRW: Does this actually happen?
195 INP_INFO_WLOCK(&tcbinfo);
198 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
199 * tear-down mean we need it as a work-around for races between
200 * timers and tcp_discardcb().
202 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
206 INP_INFO_WUNLOCK(&tcbinfo);
210 tcp_free_sackholes(tp);
211 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_2msl) ||
212 !callout_active(tp->tt_2msl)) {
213 INP_UNLOCK(tp->t_inpcb);
214 INP_INFO_WUNLOCK(&tcbinfo);
217 callout_deactivate(tp->tt_2msl);
219 * 2 MSL timeout in shutdown went off. If we're closed but
220 * still waiting for peer to close and connection has been idle
221 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
222 * control block. Otherwise, check again in a bit.
224 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
225 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
226 * Ignore fact that there were recent incoming segments.
228 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
229 tp->t_inpcb && tp->t_inpcb->inp_socket &&
230 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
231 tcpstat.tcps_finwait2_drops++;
234 if (tp->t_state != TCPS_TIME_WAIT &&
235 (ticks - tp->t_rcvtime) <= tcp_maxidle)
236 callout_reset(tp->tt_2msl, tcp_keepintvl,
243 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
244 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
249 INP_INFO_WUNLOCK(&tcbinfo);
253 * The timed wait queue contains references to each of the TCP sessions
254 * currently in the TIME_WAIT state. The queue pointers, including the
255 * queue pointers in each tcptw structure, are protected using the global
256 * tcbinfo lock, which must be held over queue iteration and modification.
258 static TAILQ_HEAD(, tcptw) twq_2msl;
264 TAILQ_INIT(&twq_2msl);
268 tcp_timer_2msl_reset(struct tcptw *tw, int rearm)
271 INP_INFO_WLOCK_ASSERT(&tcbinfo);
272 INP_LOCK_ASSERT(tw->tw_inpcb);
274 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
275 tw->tw_time = ticks + 2 * tcp_msl;
276 TAILQ_INSERT_TAIL(&twq_2msl, tw, tw_2msl);
280 tcp_timer_2msl_stop(struct tcptw *tw)
283 INP_INFO_WLOCK_ASSERT(&tcbinfo);
284 TAILQ_REMOVE(&twq_2msl, tw, tw_2msl);
288 tcp_timer_2msl_tw(int reuse)
292 INP_INFO_WLOCK_ASSERT(&tcbinfo);
294 tw = TAILQ_FIRST(&twq_2msl);
295 if (tw == NULL || (!reuse && tw->tw_time > ticks))
297 INP_LOCK(tw->tw_inpcb);
298 tcp_twclose(tw, reuse);
309 struct tcpcb *tp = xtp;
310 struct tcptemp *t_template;
315 ostate = tp->t_state;
317 INP_INFO_WLOCK(&tcbinfo);
320 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
321 * tear-down mean we need it as a work-around for races between
322 * timers and tcp_discardcb().
324 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
328 INP_INFO_WUNLOCK(&tcbinfo);
332 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_keep)
333 || !callout_active(tp->tt_keep)) {
335 INP_INFO_WUNLOCK(&tcbinfo);
338 callout_deactivate(tp->tt_keep);
340 * Keep-alive timer went off; send something
341 * or drop connection if idle for too long.
343 tcpstat.tcps_keeptimeo++;
344 if (tp->t_state < TCPS_ESTABLISHED)
346 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
347 tp->t_state <= TCPS_CLOSING) {
348 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
351 * Send a packet designed to force a response
352 * if the peer is up and reachable:
353 * either an ACK if the connection is still alive,
354 * or an RST if the peer has closed the connection
355 * due to timeout or reboot.
356 * Using sequence number tp->snd_una-1
357 * causes the transmitted zero-length segment
358 * to lie outside the receive window;
359 * by the protocol spec, this requires the
360 * correspondent TCP to respond.
362 tcpstat.tcps_keepprobe++;
363 t_template = tcpip_maketemplate(inp);
365 tcp_respond(tp, t_template->tt_ipgen,
366 &t_template->tt_t, (struct mbuf *)NULL,
367 tp->rcv_nxt, tp->snd_una - 1, 0);
368 (void) m_free(dtom(t_template));
370 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
372 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
375 if (inp->inp_socket->so_options & SO_DEBUG)
376 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
380 INP_INFO_WUNLOCK(&tcbinfo);
384 tcpstat.tcps_keepdrops++;
385 tp = tcp_drop(tp, ETIMEDOUT);
388 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
389 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
393 INP_UNLOCK(tp->t_inpcb);
394 INP_INFO_WUNLOCK(&tcbinfo);
398 tcp_timer_persist(xtp)
401 struct tcpcb *tp = xtp;
406 ostate = tp->t_state;
408 INP_INFO_WLOCK(&tcbinfo);
411 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
412 * tear-down mean we need it as a work-around for races between
413 * timers and tcp_discardcb().
415 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
419 INP_INFO_WUNLOCK(&tcbinfo);
423 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_persist)
424 || !callout_active(tp->tt_persist)) {
426 INP_INFO_WUNLOCK(&tcbinfo);
429 callout_deactivate(tp->tt_persist);
431 * Persistance timer into zero window.
432 * Force a byte to be output, if possible.
434 tcpstat.tcps_persisttimeo++;
436 * Hack: if the peer is dead/unreachable, we do not
437 * time out if the window is closed. After a full
438 * backoff, drop the connection if the idle time
439 * (no responses to probes) reaches the maximum
440 * backoff that we would use if retransmitting.
442 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
443 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
444 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
445 tcpstat.tcps_persistdrop++;
446 tp = tcp_drop(tp, ETIMEDOUT);
450 tp->t_flags |= TF_FORCEDATA;
451 (void) tcp_output(tp);
452 tp->t_flags &= ~TF_FORCEDATA;
456 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
457 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
461 INP_INFO_WUNLOCK(&tcbinfo);
468 struct tcpcb *tp = xtp;
475 ostate = tp->t_state;
477 INP_INFO_WLOCK(&tcbinfo);
481 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
482 * tear-down mean we need it as a work-around for races between
483 * timers and tcp_discardcb().
485 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
489 INP_INFO_WUNLOCK(&tcbinfo);
493 if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_rexmt)
494 || !callout_active(tp->tt_rexmt)) {
496 INP_INFO_WUNLOCK(&tcbinfo);
499 callout_deactivate(tp->tt_rexmt);
500 tcp_free_sackholes(tp);
502 * Retransmission timer went off. Message has not
503 * been acked within retransmit interval. Back off
504 * to a longer retransmit interval and retransmit one segment.
506 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
507 tp->t_rxtshift = TCP_MAXRXTSHIFT;
508 tcpstat.tcps_timeoutdrop++;
509 tp = tcp_drop(tp, tp->t_softerror ?
510 tp->t_softerror : ETIMEDOUT);
513 INP_INFO_WUNLOCK(&tcbinfo);
515 if (tp->t_rxtshift == 1) {
517 * first retransmit; record ssthresh and cwnd so they can
518 * be recovered if this turns out to be a "bad" retransmit.
519 * A retransmit is considered "bad" if an ACK for this
520 * segment is received within RTT/2 interval; the assumption
521 * here is that the ACK was already in flight. See
522 * "On Estimating End-to-End Network Path Properties" by
523 * Allman and Paxson for more details.
525 tp->snd_cwnd_prev = tp->snd_cwnd;
526 tp->snd_ssthresh_prev = tp->snd_ssthresh;
527 tp->snd_recover_prev = tp->snd_recover;
528 if (IN_FASTRECOVERY(tp))
529 tp->t_flags |= TF_WASFRECOVERY;
531 tp->t_flags &= ~TF_WASFRECOVERY;
532 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
534 tcpstat.tcps_rexmttimeo++;
535 if (tp->t_state == TCPS_SYN_SENT)
536 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
538 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
539 TCPT_RANGESET(tp->t_rxtcur, rexmt,
540 tp->t_rttmin, TCPTV_REXMTMAX);
542 * Disable rfc1323 if we havn't got any response to
543 * our third SYN to work-around some broken terminal servers
544 * (most of which have hopefully been retired) that have bad VJ
545 * header compression code which trashes TCP segments containing
546 * unknown-to-them TCP options.
548 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
549 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
551 * If we backed off this far, our srtt estimate is probably bogus.
552 * Clobber it so we'll take the next rtt measurement as our srtt;
553 * move the current srtt into rttvar to keep the current
554 * retransmit times until then.
556 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
558 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
559 in6_losing(tp->t_inpcb);
562 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
565 tp->snd_nxt = tp->snd_una;
566 tp->snd_recover = tp->snd_max;
568 * Force a segment to be sent.
570 tp->t_flags |= TF_ACKNOW;
572 * If timing a segment in this window, stop the timer.
576 * Close the congestion window down to one segment
577 * (we'll open it by one segment for each ack we get).
578 * Since we probably have a window's worth of unacked
579 * data accumulated, this "slow start" keeps us from
580 * dumping all that data as back-to-back packets (which
581 * might overwhelm an intermediate gateway).
583 * There are two phases to the opening: Initially we
584 * open by one mss on each ack. This makes the window
585 * size increase exponentially with time. If the
586 * window is larger than the path can handle, this
587 * exponential growth results in dropped packet(s)
588 * almost immediately. To get more time between
589 * drops but still "push" the network to take advantage
590 * of improving conditions, we switch from exponential
591 * to linear window opening at some threshhold size.
592 * For a threshhold, we use half the current window
593 * size, truncated to a multiple of the mss.
595 * (the minimum cwnd that will give us exponential
596 * growth is 2 mss. We don't allow the threshhold
600 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
603 tp->snd_cwnd = tp->t_maxseg;
604 tp->snd_ssthresh = win * tp->t_maxseg;
607 EXIT_FASTRECOVERY(tp);
608 (void) tcp_output(tp);
612 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
613 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
619 INP_INFO_WUNLOCK(&tcbinfo);