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_msec_to_ticks(SYSCTL_HANDLER_ARGS)
71 tt = *(int *)oidp->oid_arg1;
72 s = (int)((int64_t)tt * 1000 / hz);
74 error = sysctl_handle_int(oidp, &s, 0, req);
75 if (error || !req->newptr)
78 tt = (int)((int64_t)s * hz / 1000);
82 *(int *)oidp->oid_arg1 = tt;
87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
88 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
92 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
95 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
96 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
99 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
100 CTLTYPE_INT|CTLFLAG_RW, &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
101 "Time before a delayed ACK is sent");
104 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
105 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
108 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
109 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", "Minimum Retransmission Timeout");
112 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
113 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", "Retransmission Timer Slop");
115 static int always_keepalive = 1;
116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
117 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
119 static int tcp_keepcnt = TCPTV_KEEPCNT;
120 /* max idle probes */
121 int tcp_maxpersistidle;
122 /* max idle time in persist */
126 * Tcp protocol timeout routine called every 500 ms.
127 * Updates timestamps used for TCP
128 * causes finite state machine actions if timers expire.
134 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
135 INP_INFO_WLOCK(&tcbinfo);
136 (void) tcp_timer_2msl_tw(0);
137 INP_INFO_WUNLOCK(&tcbinfo);
140 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
141 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
143 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
144 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
146 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
149 * TCP timer processing.
153 tcp_timer_delack(xtp)
156 struct tcpcb *tp = xtp;
159 INP_INFO_RLOCK(&tcbinfo);
162 INP_INFO_RUNLOCK(&tcbinfo);
166 INP_INFO_RUNLOCK(&tcbinfo);
167 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
171 callout_deactivate(tp->tt_delack);
173 tp->t_flags |= TF_ACKNOW;
174 tcpstat.tcps_delack++;
175 (void) tcp_output(tp);
183 struct tcpcb *tp = xtp;
188 ostate = tp->t_state;
190 INP_INFO_WLOCK(&tcbinfo);
193 INP_INFO_WUNLOCK(&tcbinfo);
197 tcp_free_sackholes(tp);
198 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
199 INP_UNLOCK(tp->t_inpcb);
200 INP_INFO_WUNLOCK(&tcbinfo);
203 callout_deactivate(tp->tt_2msl);
205 * 2 MSL timeout in shutdown went off. If we're closed but
206 * still waiting for peer to close and connection has been idle
207 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
208 * control block. Otherwise, check again in a bit.
210 if (tp->t_state != TCPS_TIME_WAIT &&
211 (ticks - tp->t_rcvtime) <= tcp_maxidle)
212 callout_reset(tp->tt_2msl, tcp_keepintvl,
218 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
219 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
224 INP_INFO_WUNLOCK(&tcbinfo);
228 * The timed wait lists contain references to each of the TCP sessions
229 * currently TIME_WAIT state. The list pointers, including the list pointers
230 * in each tcptw structure, are protected using the global tcbinfo lock,
231 * which must be held over list iteration and modification.
234 LIST_HEAD(, tcptw) tw_list;
235 struct tcptw tw_tail;
237 #define TWLIST_NLISTS 2
238 static struct twlist twl_2msl[TWLIST_NLISTS];
239 static struct twlist *tw_2msl_list[] = { &twl_2msl[0], &twl_2msl[1], NULL };
247 for (i = 0; i < TWLIST_NLISTS; i++) {
249 LIST_INIT(&twl->tw_list);
250 LIST_INSERT_HEAD(&twl->tw_list, &twl->tw_tail, tw_2msl);
255 tcp_timer_2msl_reset(struct tcptw *tw, int timeo)
258 struct tcptw *tw_tail;
260 INP_INFO_WLOCK_ASSERT(&tcbinfo);
261 INP_LOCK_ASSERT(tw->tw_inpcb);
262 if (tw->tw_time != 0)
263 LIST_REMOVE(tw, tw_2msl);
264 tw->tw_time = timeo + ticks;
265 i = timeo > tcp_msl ? 1 : 0;
266 tw_tail = &twl_2msl[i].tw_tail;
267 LIST_INSERT_BEFORE(tw_tail, tw, tw_2msl);
271 tcp_timer_2msl_stop(struct tcptw *tw)
274 INP_INFO_WLOCK_ASSERT(&tcbinfo);
275 if (tw->tw_time != 0)
276 LIST_REMOVE(tw, tw_2msl);
280 tcp_timer_2msl_tw(int reuse)
282 struct tcptw *tw, *tw_tail;
286 INP_INFO_WLOCK_ASSERT(&tcbinfo);
287 for (i = 0; i < 2; i++) {
288 twl = tw_2msl_list[i];
289 tw_tail = &twl->tw_tail;
291 tw = LIST_FIRST(&twl->tw_list);
292 if (tw == tw_tail || (!reuse && tw->tw_time > ticks))
294 INP_LOCK(tw->tw_inpcb);
295 if (tcp_twclose(tw, reuse) != NULL)
306 struct tcpcb *tp = xtp;
307 struct tcptemp *t_template;
312 ostate = tp->t_state;
314 INP_INFO_WLOCK(&tcbinfo);
317 INP_INFO_WUNLOCK(&tcbinfo);
321 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
323 INP_INFO_WUNLOCK(&tcbinfo);
326 callout_deactivate(tp->tt_keep);
328 * Keep-alive timer went off; send something
329 * or drop connection if idle for too long.
331 tcpstat.tcps_keeptimeo++;
332 if (tp->t_state < TCPS_ESTABLISHED)
334 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
335 tp->t_state <= TCPS_CLOSING) {
336 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
339 * Send a packet designed to force a response
340 * if the peer is up and reachable:
341 * either an ACK if the connection is still alive,
342 * or an RST if the peer has closed the connection
343 * due to timeout or reboot.
344 * Using sequence number tp->snd_una-1
345 * causes the transmitted zero-length segment
346 * to lie outside the receive window;
347 * by the protocol spec, this requires the
348 * correspondent TCP to respond.
350 tcpstat.tcps_keepprobe++;
351 t_template = tcpip_maketemplate(inp);
353 tcp_respond(tp, t_template->tt_ipgen,
354 &t_template->tt_t, (struct mbuf *)NULL,
355 tp->rcv_nxt, tp->snd_una - 1, 0);
356 (void) m_free(dtom(t_template));
358 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
360 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
363 if (inp->inp_socket->so_options & SO_DEBUG)
364 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
368 INP_INFO_WUNLOCK(&tcbinfo);
372 tcpstat.tcps_keepdrops++;
373 tp = tcp_drop(tp, ETIMEDOUT);
376 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
377 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
381 INP_UNLOCK(tp->t_inpcb);
382 INP_INFO_WUNLOCK(&tcbinfo);
386 tcp_timer_persist(xtp)
389 struct tcpcb *tp = xtp;
394 ostate = tp->t_state;
396 INP_INFO_WLOCK(&tcbinfo);
399 INP_INFO_WUNLOCK(&tcbinfo);
403 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
405 INP_INFO_WUNLOCK(&tcbinfo);
408 callout_deactivate(tp->tt_persist);
410 * Persistance timer into zero window.
411 * Force a byte to be output, if possible.
413 tcpstat.tcps_persisttimeo++;
415 * Hack: if the peer is dead/unreachable, we do not
416 * time out if the window is closed. After a full
417 * backoff, drop the connection if the idle time
418 * (no responses to probes) reaches the maximum
419 * backoff that we would use if retransmitting.
421 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
422 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
423 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
424 tcpstat.tcps_persistdrop++;
425 tp = tcp_drop(tp, ETIMEDOUT);
429 tp->t_flags |= TF_FORCEDATA;
430 (void) tcp_output(tp);
431 tp->t_flags &= ~TF_FORCEDATA;
435 if (tp && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
436 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
441 INP_INFO_WUNLOCK(&tcbinfo);
448 struct tcpcb *tp = xtp;
455 ostate = tp->t_state;
457 INP_INFO_WLOCK(&tcbinfo);
461 INP_INFO_WUNLOCK(&tcbinfo);
465 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
467 INP_INFO_WUNLOCK(&tcbinfo);
470 callout_deactivate(tp->tt_rexmt);
471 tcp_free_sackholes(tp);
473 * Retransmission timer went off. Message has not
474 * been acked within retransmit interval. Back off
475 * to a longer retransmit interval and retransmit one segment.
477 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
478 tp->t_rxtshift = TCP_MAXRXTSHIFT;
479 tcpstat.tcps_timeoutdrop++;
480 tp = tcp_drop(tp, tp->t_softerror ?
481 tp->t_softerror : ETIMEDOUT);
484 INP_INFO_WUNLOCK(&tcbinfo);
486 if (tp->t_rxtshift == 1) {
488 * first retransmit; record ssthresh and cwnd so they can
489 * be recovered if this turns out to be a "bad" retransmit.
490 * A retransmit is considered "bad" if an ACK for this
491 * segment is received within RTT/2 interval; the assumption
492 * here is that the ACK was already in flight. See
493 * "On Estimating End-to-End Network Path Properties" by
494 * Allman and Paxson for more details.
496 tp->snd_cwnd_prev = tp->snd_cwnd;
497 tp->snd_ssthresh_prev = tp->snd_ssthresh;
498 tp->snd_recover_prev = tp->snd_recover;
499 if (IN_FASTRECOVERY(tp))
500 tp->t_flags |= TF_WASFRECOVERY;
502 tp->t_flags &= ~TF_WASFRECOVERY;
503 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
505 tcpstat.tcps_rexmttimeo++;
506 if (tp->t_state == TCPS_SYN_SENT)
507 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
509 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
510 TCPT_RANGESET(tp->t_rxtcur, rexmt,
511 tp->t_rttmin, TCPTV_REXMTMAX);
513 * Disable rfc1323 if we havn't got any response to
514 * our third SYN to work-around some broken terminal servers
515 * (most of which have hopefully been retired) that have bad VJ
516 * header compression code which trashes TCP segments containing
517 * unknown-to-them TCP options.
519 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
520 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
522 * If we backed off this far, our srtt estimate is probably bogus.
523 * Clobber it so we'll take the next rtt measurement as our srtt;
524 * move the current srtt into rttvar to keep the current
525 * retransmit times until then.
527 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
529 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
530 in6_losing(tp->t_inpcb);
533 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
536 tp->snd_nxt = tp->snd_una;
537 tp->snd_recover = tp->snd_max;
539 * Force a segment to be sent.
541 tp->t_flags |= TF_ACKNOW;
543 * If timing a segment in this window, stop the timer.
547 * Close the congestion window down to one segment
548 * (we'll open it by one segment for each ack we get).
549 * Since we probably have a window's worth of unacked
550 * data accumulated, this "slow start" keeps us from
551 * dumping all that data as back-to-back packets (which
552 * might overwhelm an intermediate gateway).
554 * There are two phases to the opening: Initially we
555 * open by one mss on each ack. This makes the window
556 * size increase exponentially with time. If the
557 * window is larger than the path can handle, this
558 * exponential growth results in dropped packet(s)
559 * almost immediately. To get more time between
560 * drops but still "push" the network to take advantage
561 * of improving conditions, we switch from exponential
562 * to linear window opening at some threshhold size.
563 * For a threshhold, we use half the current window
564 * size, truncated to a multiple of the mss.
566 * (the minimum cwnd that will give us exponential
567 * growth is 2 mss. We don't allow the threshhold
571 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
574 tp->snd_cwnd = tp->t_maxseg;
575 tp->snd_ssthresh = win * tp->t_maxseg;
578 EXIT_FASTRECOVERY(tp);
579 (void) tcp_output(tp);
583 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
584 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
590 INP_INFO_WUNLOCK(&tcbinfo);