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 static int tcp_keepcnt = TCPTV_KEEPCNT;
100 /* max idle probes */
101 int tcp_maxpersistidle;
102 /* max idle time in persist */
106 * Tcp protocol timeout routine called every 500 ms.
107 * Updates timestamps used for TCP
108 * causes finite state machine actions if timers expire.
114 tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
115 INP_INFO_WLOCK(&tcbinfo);
116 (void) tcp_timer_2msl_tw(0);
117 INP_INFO_WUNLOCK(&tcbinfo);
120 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
121 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
123 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
124 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
126 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
129 * TCP timer processing.
133 tcp_timer_delack(xtp)
136 struct tcpcb *tp = xtp;
139 INP_INFO_RLOCK(&tcbinfo);
142 INP_INFO_RUNLOCK(&tcbinfo);
146 INP_INFO_RUNLOCK(&tcbinfo);
147 if (callout_pending(tp->tt_delack) || !callout_active(tp->tt_delack)) {
151 callout_deactivate(tp->tt_delack);
153 tp->t_flags |= TF_ACKNOW;
154 tcpstat.tcps_delack++;
155 (void) tcp_output(tp);
163 struct tcpcb *tp = xtp;
168 ostate = tp->t_state;
170 INP_INFO_WLOCK(&tcbinfo);
173 INP_INFO_WUNLOCK(&tcbinfo);
177 tcp_free_sackholes(tp);
178 if (callout_pending(tp->tt_2msl) || !callout_active(tp->tt_2msl)) {
179 INP_UNLOCK(tp->t_inpcb);
180 INP_INFO_WUNLOCK(&tcbinfo);
183 callout_deactivate(tp->tt_2msl);
185 * 2 MSL timeout in shutdown went off. If we're closed but
186 * still waiting for peer to close and connection has been idle
187 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
188 * control block. Otherwise, check again in a bit.
190 if (tp->t_state != TCPS_TIME_WAIT &&
191 (ticks - tp->t_rcvtime) <= tcp_maxidle)
192 callout_reset(tp->tt_2msl, tcp_keepintvl,
198 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
199 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
204 INP_INFO_WUNLOCK(&tcbinfo);
208 * The timed wait lists contain references to each of the TCP sessions
209 * currently TIME_WAIT state. The list pointers, including the list pointers
210 * in each tcptw structure, are protected using the global tcbinfo lock,
211 * which must be held over list iteration and modification.
214 LIST_HEAD(, tcptw) tw_list;
215 struct tcptw tw_tail;
217 #define TWLIST_NLISTS 2
218 static struct twlist twl_2msl[TWLIST_NLISTS];
219 static struct twlist *tw_2msl_list[] = { &twl_2msl[0], &twl_2msl[1], NULL };
227 for (i = 0; i < TWLIST_NLISTS; i++) {
229 LIST_INIT(&twl->tw_list);
230 LIST_INSERT_HEAD(&twl->tw_list, &twl->tw_tail, tw_2msl);
235 tcp_timer_2msl_reset(struct tcptw *tw, int timeo)
238 struct tcptw *tw_tail;
240 INP_INFO_WLOCK_ASSERT(&tcbinfo);
241 INP_LOCK_ASSERT(tw->tw_inpcb);
242 if (tw->tw_time != 0)
243 LIST_REMOVE(tw, tw_2msl);
244 tw->tw_time = timeo + ticks;
245 i = timeo > tcp_msl ? 1 : 0;
246 tw_tail = &twl_2msl[i].tw_tail;
247 LIST_INSERT_BEFORE(tw_tail, tw, tw_2msl);
251 tcp_timer_2msl_stop(struct tcptw *tw)
254 INP_INFO_WLOCK_ASSERT(&tcbinfo);
255 if (tw->tw_time != 0)
256 LIST_REMOVE(tw, tw_2msl);
260 tcp_timer_2msl_tw(int reuse)
262 struct tcptw *tw, *tw_tail;
266 INP_INFO_WLOCK_ASSERT(&tcbinfo);
267 for (i = 0; i < 2; i++) {
268 twl = tw_2msl_list[i];
269 tw_tail = &twl->tw_tail;
271 tw = LIST_FIRST(&twl->tw_list);
272 if (tw == tw_tail || (!reuse && tw->tw_time > ticks))
274 INP_LOCK(tw->tw_inpcb);
275 if (tcp_twclose(tw, reuse) != NULL)
286 struct tcpcb *tp = xtp;
287 struct tcptemp *t_template;
292 ostate = tp->t_state;
294 INP_INFO_WLOCK(&tcbinfo);
297 INP_INFO_WUNLOCK(&tcbinfo);
301 if (callout_pending(tp->tt_keep) || !callout_active(tp->tt_keep)) {
303 INP_INFO_WUNLOCK(&tcbinfo);
306 callout_deactivate(tp->tt_keep);
308 * Keep-alive timer went off; send something
309 * or drop connection if idle for too long.
311 tcpstat.tcps_keeptimeo++;
312 if (tp->t_state < TCPS_ESTABLISHED)
314 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
315 tp->t_state <= TCPS_CLOSING) {
316 if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
319 * Send a packet designed to force a response
320 * if the peer is up and reachable:
321 * either an ACK if the connection is still alive,
322 * or an RST if the peer has closed the connection
323 * due to timeout or reboot.
324 * Using sequence number tp->snd_una-1
325 * causes the transmitted zero-length segment
326 * to lie outside the receive window;
327 * by the protocol spec, this requires the
328 * correspondent TCP to respond.
330 tcpstat.tcps_keepprobe++;
331 t_template = tcpip_maketemplate(inp);
333 tcp_respond(tp, t_template->tt_ipgen,
334 &t_template->tt_t, (struct mbuf *)NULL,
335 tp->rcv_nxt, tp->snd_una - 1, 0);
336 (void) m_free(dtom(t_template));
338 callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
340 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
343 if (inp->inp_socket->so_options & SO_DEBUG)
344 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
348 INP_INFO_WUNLOCK(&tcbinfo);
352 tcpstat.tcps_keepdrops++;
353 tp = tcp_drop(tp, ETIMEDOUT);
356 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
357 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
361 INP_UNLOCK(tp->t_inpcb);
362 INP_INFO_WUNLOCK(&tcbinfo);
366 tcp_timer_persist(xtp)
369 struct tcpcb *tp = xtp;
374 ostate = tp->t_state;
376 INP_INFO_WLOCK(&tcbinfo);
379 INP_INFO_WUNLOCK(&tcbinfo);
383 if (callout_pending(tp->tt_persist) || !callout_active(tp->tt_persist)){
385 INP_INFO_WUNLOCK(&tcbinfo);
388 callout_deactivate(tp->tt_persist);
390 * Persistance timer into zero window.
391 * Force a byte to be output, if possible.
393 tcpstat.tcps_persisttimeo++;
395 * Hack: if the peer is dead/unreachable, we do not
396 * time out if the window is closed. After a full
397 * backoff, drop the connection if the idle time
398 * (no responses to probes) reaches the maximum
399 * backoff that we would use if retransmitting.
401 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
402 ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
403 (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
404 tcpstat.tcps_persistdrop++;
405 tp = tcp_drop(tp, ETIMEDOUT);
409 tp->t_flags |= TF_FORCEDATA;
410 (void) tcp_output(tp);
411 tp->t_flags &= ~TF_FORCEDATA;
415 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
416 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
421 INP_INFO_WUNLOCK(&tcbinfo);
428 struct tcpcb *tp = xtp;
435 ostate = tp->t_state;
437 INP_INFO_WLOCK(&tcbinfo);
441 INP_INFO_WUNLOCK(&tcbinfo);
445 if (callout_pending(tp->tt_rexmt) || !callout_active(tp->tt_rexmt)) {
447 INP_INFO_WUNLOCK(&tcbinfo);
450 callout_deactivate(tp->tt_rexmt);
451 tcp_free_sackholes(tp);
453 * Retransmission timer went off. Message has not
454 * been acked within retransmit interval. Back off
455 * to a longer retransmit interval and retransmit one segment.
457 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
458 tp->t_rxtshift = TCP_MAXRXTSHIFT;
459 tcpstat.tcps_timeoutdrop++;
460 tp = tcp_drop(tp, tp->t_softerror ?
461 tp->t_softerror : ETIMEDOUT);
464 INP_INFO_WUNLOCK(&tcbinfo);
466 if (tp->t_rxtshift == 1) {
468 * first retransmit; record ssthresh and cwnd so they can
469 * be recovered if this turns out to be a "bad" retransmit.
470 * A retransmit is considered "bad" if an ACK for this
471 * segment is received within RTT/2 interval; the assumption
472 * here is that the ACK was already in flight. See
473 * "On Estimating End-to-End Network Path Properties" by
474 * Allman and Paxson for more details.
476 tp->snd_cwnd_prev = tp->snd_cwnd;
477 tp->snd_ssthresh_prev = tp->snd_ssthresh;
478 tp->snd_recover_prev = tp->snd_recover;
479 if (IN_FASTRECOVERY(tp))
480 tp->t_flags |= TF_WASFRECOVERY;
482 tp->t_flags &= ~TF_WASFRECOVERY;
483 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
485 tcpstat.tcps_rexmttimeo++;
486 if (tp->t_state == TCPS_SYN_SENT)
487 rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
489 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
490 TCPT_RANGESET(tp->t_rxtcur, rexmt,
491 tp->t_rttmin, TCPTV_REXMTMAX);
493 * Disable rfc1323 if we havn't got any response to
494 * our third SYN to work-around some broken terminal servers
495 * (most of which have hopefully been retired) that have bad VJ
496 * header compression code which trashes TCP segments containing
497 * unknown-to-them TCP options.
499 if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
500 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
502 * If we backed off this far, our srtt estimate is probably bogus.
503 * Clobber it so we'll take the next rtt measurement as our srtt;
504 * move the current srtt into rttvar to keep the current
505 * retransmit times until then.
507 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
509 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
510 in6_losing(tp->t_inpcb);
513 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
516 tp->snd_nxt = tp->snd_una;
517 tp->snd_recover = tp->snd_max;
519 * Force a segment to be sent.
521 tp->t_flags |= TF_ACKNOW;
523 * If timing a segment in this window, stop the timer.
527 * Close the congestion window down to one segment
528 * (we'll open it by one segment for each ack we get).
529 * Since we probably have a window's worth of unacked
530 * data accumulated, this "slow start" keeps us from
531 * dumping all that data as back-to-back packets (which
532 * might overwhelm an intermediate gateway).
534 * There are two phases to the opening: Initially we
535 * open by one mss on each ack. This makes the window
536 * size increase exponentially with time. If the
537 * window is larger than the path can handle, this
538 * exponential growth results in dropped packet(s)
539 * almost immediately. To get more time between
540 * drops but still "push" the network to take advantage
541 * of improving conditions, we switch from exponential
542 * to linear window opening at some threshhold size.
543 * For a threshhold, we use half the current window
544 * size, truncated to a multiple of the mss.
546 * (the minimum cwnd that will give us exponential
547 * growth is 2 mss. We don't allow the threshhold
551 u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
554 tp->snd_cwnd = tp->t_maxseg;
555 tp->snd_ssthresh = win * tp->t_maxseg;
558 EXIT_FASTRECOVERY(tp);
559 (void) tcp_output(tp);
563 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
564 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
570 INP_INFO_WUNLOCK(&tcbinfo);