2 * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
4 * @APPLE_LICENSE_HEADER_START@
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
20 * @APPLE_LICENSE_HEADER_END@
23 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
24 * The Regents of the University of California. All rights reserved.
26 * Redistribution and use in source and binary forms, with or without
27 * modification, are permitted provided that the following conditions
29 * 1. Redistributions of source code must retain the above copyright
30 * notice, this list of conditions and the following disclaimer.
31 * 2. Redistributions in binary form must reproduce the above copyright
32 * notice, this list of conditions and the following disclaimer in the
33 * documentation and/or other materials provided with the distribution.
34 * 3. All advertising materials mentioning features or use of this software
35 * must display the following acknowledgement:
36 * This product includes software developed by the University of
37 * California, Berkeley and its contributors.
38 * 4. Neither the name of the University nor the names of its contributors
39 * may be used to endorse or promote products derived from this software
40 * without specific prior written permission.
42 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
54 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
55 * $FreeBSD: src/sys/netinet/tcp_timer.c,v 1.34.2.11 2001/08/22 00:59:12 silby Exp $
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
63 #include <sys/sysctl.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
68 #include <kern/cpu_number.h> /* before tcp_seq.h, for tcp_random18() */
70 #include <net/route.h>
72 #include <netinet/in.h>
73 #include <netinet/in_systm.h>
74 #include <netinet/in_pcb.h>
76 #include <netinet6/in6_pcb.h>
78 #include <netinet/ip_var.h>
79 #include <netinet/tcp.h>
80 #include <netinet/tcp_fsm.h>
81 #include <netinet/tcp_seq.h>
82 #include <netinet/tcp_timer.h>
83 #include <netinet/tcp_var.h>
84 #include <netinet/tcpip.h>
86 #include <netinet/tcp_debug.h>
88 #include <sys/kdebug.h>
90 #define DBG_FNC_TCP_FAST NETDBG_CODE(DBG_NETTCP, (5 << 8))
91 #define DBG_FNC_TCP_SLOW NETDBG_CODE(DBG_NETTCP, (5 << 8) | 1)
101 sysctl_msec_to_ticks SYSCTL_HANDLER_ARGS
105 tt
= *(int *)oidp
->oid_arg1
;
108 error
= sysctl_handle_int(oidp
, &s
, 0, req
);
109 if (error
|| !req
->newptr
)
116 *(int *)oidp
->oid_arg1
= tt
;
121 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPINIT
, keepinit
, CTLTYPE_INT
|CTLFLAG_RW
,
122 &tcp_keepinit
, 0, sysctl_msec_to_ticks
, "I", "");
125 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPIDLE
, keepidle
, CTLTYPE_INT
|CTLFLAG_RW
,
126 &tcp_keepidle
, 0, sysctl_msec_to_ticks
, "I", "");
129 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_KEEPINTVL
, keepintvl
, CTLTYPE_INT
|CTLFLAG_RW
,
130 &tcp_keepintvl
, 0, sysctl_msec_to_ticks
, "I", "");
133 SYSCTL_PROC(_net_inet_tcp
, TCPCTL_DELACKTIME
, delacktime
,
134 CTLTYPE_INT
|CTLFLAG_RW
, &tcp_delacktime
, 0, sysctl_msec_to_ticks
, "I",
135 "Time before a delayed ACK is sent");
138 SYSCTL_PROC(_net_inet_tcp
, OID_AUTO
, msl
, CTLTYPE_INT
|CTLFLAG_RW
,
139 &tcp_msl
, 0, sysctl_msec_to_ticks
, "I", "Maximum segment lifetime");
141 static int always_keepalive
= 0;
142 SYSCTL_INT(_net_inet_tcp
, OID_AUTO
, always_keepalive
, CTLFLAG_RW
,
143 &always_keepalive
, 0, "Assume SO_KEEPALIVE on all TCP connections");
145 static int tcp_keepcnt
= TCPTV_KEEPCNT
;
146 /* max idle probes */
147 int tcp_maxpersistidle
;
148 /* max idle time in persist */
151 struct inpcbhead time_wait_slots
[N_TIME_WAIT_SLOTS
];
154 u_long
*delack_bitmask
;
157 void add_to_time_wait(tp
)
162 LIST_REMOVE(tp
->t_inpcb
, inp_list
);
164 if (tp
->t_timer
[TCPT_2MSL
] == 0)
165 tp
->t_timer
[TCPT_2MSL
] = 1;
167 tp
->t_rcvtime
+= tp
->t_timer
[TCPT_2MSL
] & (N_TIME_WAIT_SLOTS
- 1);
168 tw_slot
= (tp
->t_timer
[TCPT_2MSL
] & (N_TIME_WAIT_SLOTS
- 1)) + cur_tw_slot
;
169 if (tw_slot
>= N_TIME_WAIT_SLOTS
)
170 tw_slot
-= N_TIME_WAIT_SLOTS
;
172 LIST_INSERT_HEAD(&time_wait_slots
[tw_slot
], tp
->t_inpcb
, inp_list
);
180 * Fast timeout routine for processing delayed acks
185 register struct inpcb
*inp
;
186 register struct tcpcb
*tp
;
190 register u_long temp_mask
;
191 register u_long elem_base
= 0;
192 struct inpcbhead
*head
;
196 int delack_checked
= 0;
198 KERNEL_DEBUG(DBG_FNC_TCP_FAST
| DBG_FUNC_START
, 0,0,0,0,0);
200 if (!tcp_delack_enabled
)
203 for (i
=0; i
< (tcbinfo
.hashsize
/ 32); i
++) {
204 if (delack_bitmask
[i
]) {
206 for (j
=0; j
< 32; j
++) {
207 if (temp_mask
& delack_bitmask
[i
]) {
208 head
= &tcbinfo
.hashbase
[elem_base
+ j
];
209 for (inp
=head
->lh_first
; inp
!= 0; inp
= inp
->inp_hash
.le_next
) {
211 if ((tp
= (struct tcpcb
*)inp
->inp_ppcb
) && (tp
->t_flags
& TF_DELACK
)) {
212 tp
->t_flags
&= ~TF_DELACK
;
213 tp
->t_flags
|= TF_ACKNOW
;
214 tcpstat
.tcps_delack
++;
215 (void) tcp_output(tp
);
221 delack_bitmask
[i
] = 0;
225 KERNEL_DEBUG(DBG_FNC_TCP_FAST
| DBG_FUNC_END
, delack_checked
,tcpstat
.tcps_delack
,0,0,0);
231 * Tcp protocol timeout routine called every 500 ms.
232 * Updates the timers in all active tcb's and
233 * causes finite state machine actions if timers expire.
238 register struct inpcb
*ip
, *ipnxt
;
239 register struct tcpcb
*tp
;
246 static int tws_checked
;
249 KERNEL_DEBUG(DBG_FNC_TCP_SLOW
| DBG_FUNC_START
, 0,0,0,0,0);
252 tcp_maxidle
= tcp_keepcnt
* tcp_keepintvl
;
260 * Search through tcb's and update active timers.
262 for (; ip
!= NULL
; ip
= ipnxt
) {
263 ipnxt
= ip
->inp_list
.le_next
;
265 if (tp
== 0 || tp
->t_state
== TCPS_LISTEN
)
268 * Bogus state when port owned by SharedIP with loopback as the
269 * only configured interface: BlueBox does not filters loopback
271 if (tp
->t_state
== TCP_NSTATES
)
274 for (i
= 0; i
< TCPT_NTIMERS
; i
++) {
275 if (tp
->t_timer
[i
] && --tp
->t_timer
[i
] == 0) {
277 ostate
= tp
->t_state
;
279 tp
= tcp_timers(tp
, i
);
283 if (tp
->t_inpcb
->inp_socket
->so_options
285 tcp_trace(TA_USER
, ostate
, tp
,
303 KERNEL_DEBUG(DBG_FNC_TCP_SLOW
| DBG_FUNC_NONE
, tws_checked
,0,0,0,0);
306 * Process the items in the current time-wait slot
309 for (ip
= time_wait_slots
[cur_tw_slot
].lh_first
; ip
; ip
= ipnxt
)
314 ipnxt
= ip
->inp_list
.le_next
;
316 if (tp
== NULL
) { /* tp already closed, remove from list */
317 LIST_REMOVE(ip
, inp_list
);
320 if (tp
->t_timer
[TCPT_2MSL
] >= N_TIME_WAIT_SLOTS
) {
321 tp
->t_timer
[TCPT_2MSL
] -= N_TIME_WAIT_SLOTS
;
322 tp
->t_rcvtime
+= N_TIME_WAIT_SLOTS
;
325 tp
->t_timer
[TCPT_2MSL
] = 0;
327 if (tp
->t_timer
[TCPT_2MSL
] == 0)
328 tp
= tcp_timers(tp
, TCPT_2MSL
);
331 if (++cur_tw_slot
>= N_TIME_WAIT_SLOTS
)
333 tcp_now
++; /* for timestamps */
335 KERNEL_DEBUG(DBG_FNC_TCP_SLOW
| DBG_FUNC_END
, tws_checked
, cur_tw_slot
,0,0,0);
339 * Cancel all timers for TCP tp.
347 for (i
= 0; i
< TCPT_NTIMERS
; i
++)
351 int tcp_syn_backoff
[TCP_MAXRXTSHIFT
+ 1] =
352 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
354 int tcp_backoff
[TCP_MAXRXTSHIFT
+ 1] =
355 { 1, 2, 4, 8, 16, 32, 64, 64, 64, 64, 64, 64, 64 };
357 static int tcp_totbackoff
= 511; /* sum of tcp_backoff[] */
360 * TCP timer processing.
363 tcp_timers(tp
, timer
)
364 register struct tcpcb
*tp
;
368 struct socket
*so_tmp
;
369 struct tcptemp
*t_template
;
376 int isipv6
= (tp
->t_inpcb
->inp_vflag
& INP_IPV4
) == 0;
383 * 2 MSL timeout in shutdown went off. If we're closed but
384 * still waiting for peer to close and connection has been idle
385 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
386 * control block. Otherwise, check again in a bit.
389 if (tp
->t_state
!= TCPS_TIME_WAIT
&&
390 tp
->t_rcvtime
<= tcp_maxidle
) {
391 tp
->t_timer
[TCPT_2MSL
] = tcp_keepintvl
;
392 add_to_time_wait(tp
);
399 * Retransmission timer went off. Message has not
400 * been acked within retransmit interval. Back off
401 * to a longer retransmit interval and retransmit one segment.
404 if (++tp
->t_rxtshift
> TCP_MAXRXTSHIFT
) {
405 tp
->t_rxtshift
= TCP_MAXRXTSHIFT
;
406 tcpstat
.tcps_timeoutdrop
++;
407 so_tmp
= tp
->t_inpcb
->inp_socket
;
408 tp
= tcp_drop(tp
, tp
->t_softerror
?
409 tp
->t_softerror
: ETIMEDOUT
);
410 postevent(so_tmp
, 0, EV_TIMEOUT
);
414 if (tp
->t_rxtshift
== 1) {
416 * first retransmit; record ssthresh and cwnd so they can
417 * be recovered if this turns out to be a "bad" retransmit.
418 * A retransmit is considered "bad" if an ACK for this
419 * segment is received within RTT/2 interval; the assumption
420 * here is that the ACK was already in flight. See
421 * "On Estimating End-to-End Network Path Properties" by
422 * Allman and Paxson for more details.
424 tp
->snd_cwnd_prev
= tp
->snd_cwnd
;
425 tp
->snd_ssthresh_prev
= tp
->snd_ssthresh
;
426 tp
->t_badrxtwin
= tcp_now
+ (tp
->t_srtt
>> (TCP_RTT_SHIFT
+ 1));
428 tcpstat
.tcps_rexmttimeo
++;
429 if (tp
->t_state
== TCPS_SYN_SENT
)
430 rexmt
= TCP_REXMTVAL(tp
) * tcp_syn_backoff
[tp
->t_rxtshift
];
432 rexmt
= TCP_REXMTVAL(tp
) * tcp_backoff
[tp
->t_rxtshift
];
433 TCPT_RANGESET(tp
->t_rxtcur
, rexmt
,
434 tp
->t_rttmin
, TCPTV_REXMTMAX
);
435 tp
->t_timer
[TCPT_REXMT
] = tp
->t_rxtcur
;
438 * Disable rfc1323 and rfc1644 if we havn't got any response to
439 * our third SYN to work-around some broken terminal servers
440 * (most of which have hopefully been retired) that have bad VJ
441 * header compression code which trashes TCP segments containing
442 * unknown-to-them TCP options.
444 if ((tp
->t_state
== TCPS_SYN_SENT
) && (tp
->t_rxtshift
== 3))
445 tp
->t_flags
&= ~(TF_REQ_SCALE
|TF_REQ_TSTMP
|TF_REQ_CC
);
447 * If losing, let the lower level know and try for
448 * a better route. Also, if we backed off this far,
449 * our srtt estimate is probably bogus. Clobber it
450 * so we'll take the next rtt measurement as our srtt;
451 * move the current srtt into rttvar to keep the current
452 * retransmit times until then.
454 if (tp
->t_rxtshift
> TCP_MAXRXTSHIFT
/ 4) {
457 in6_losing(tp
->t_inpcb
);
460 in_losing(tp
->t_inpcb
);
461 tp
->t_rttvar
+= (tp
->t_srtt
>> TCP_RTT_SHIFT
);
464 tp
->snd_nxt
= tp
->snd_una
;
466 * Note: We overload snd_recover to function also as the
467 * snd_last variable described in RFC 2582
469 tp
->snd_recover
= tp
->snd_max
;
471 * Force a segment to be sent.
473 tp
->t_flags
|= TF_ACKNOW
;
475 * If timing a segment in this window, stop the timer.
479 * Close the congestion window down to one segment
480 * (we'll open it by one segment for each ack we get).
481 * Since we probably have a window's worth of unacked
482 * data accumulated, this "slow start" keeps us from
483 * dumping all that data as back-to-back packets (which
484 * might overwhelm an intermediate gateway).
486 * There are two phases to the opening: Initially we
487 * open by one mss on each ack. This makes the window
488 * size increase exponentially with time. If the
489 * window is larger than the path can handle, this
490 * exponential growth results in dropped packet(s)
491 * almost immediately. To get more time between
492 * drops but still "push" the network to take advantage
493 * of improving conditions, we switch from exponential
494 * to linear window opening at some threshhold size.
495 * For a threshhold, we use half the current window
496 * size, truncated to a multiple of the mss.
498 * (the minimum cwnd that will give us exponential
499 * growth is 2 mss. We don't allow the threshhold
503 u_int win
= min(tp
->snd_wnd
, tp
->snd_cwnd
) / 2 / tp
->t_maxseg
;
506 tp
->snd_cwnd
= tp
->t_maxseg
;
507 tp
->snd_ssthresh
= win
* tp
->t_maxseg
;
510 (void) tcp_output(tp
);
514 * Persistance timer into zero window.
515 * Force a byte to be output, if possible.
518 tcpstat
.tcps_persisttimeo
++;
520 * Hack: if the peer is dead/unreachable, we do not
521 * time out if the window is closed. After a full
522 * backoff, drop the connection if the idle time
523 * (no responses to probes) reaches the maximum
524 * backoff that we would use if retransmitting.
526 if (tp
->t_rxtshift
== TCP_MAXRXTSHIFT
&&
527 (tp
->t_rcvtime
>= tcp_maxpersistidle
||
528 tp
->t_rcvtime
>= TCP_REXMTVAL(tp
) * tcp_totbackoff
)) {
529 tcpstat
.tcps_persistdrop
++;
530 so_tmp
= tp
->t_inpcb
->inp_socket
;
531 tp
= tcp_drop(tp
, ETIMEDOUT
);
532 postevent(so_tmp
, 0, EV_TIMEOUT
);
537 (void) tcp_output(tp
);
542 * Keep-alive timer went off; send something
543 * or drop connection if idle for too long.
546 tcpstat
.tcps_keeptimeo
++;
547 if (tp
->t_state
< TCPS_ESTABLISHED
)
549 if ((always_keepalive
||
550 tp
->t_inpcb
->inp_socket
->so_options
& SO_KEEPALIVE
) &&
551 tp
->t_state
<= TCPS_CLOSING
) {
552 if (tp
->t_rcvtime
>= TCP_KEEPIDLE(tp
) + tcp_maxidle
)
555 * Send a packet designed to force a response
556 * if the peer is up and reachable:
557 * either an ACK if the connection is still alive,
558 * or an RST if the peer has closed the connection
559 * due to timeout or reboot.
560 * Using sequence number tp->snd_una-1
561 * causes the transmitted zero-length segment
562 * to lie outside the receive window;
563 * by the protocol spec, this requires the
564 * correspondent TCP to respond.
566 tcpstat
.tcps_keepprobe
++;
567 t_template
= tcp_maketemplate(tp
);
569 tcp_respond(tp
, t_template
->tt_ipgen
,
570 &t_template
->tt_t
, (struct mbuf
*)NULL
,
571 tp
->rcv_nxt
, tp
->snd_una
- 1, 0);
572 (void) m_free(dtom(t_template
));
574 tp
->t_timer
[TCPT_KEEP
] = tcp_keepintvl
;
576 tp
->t_timer
[TCPT_KEEP
] = TCP_KEEPIDLE(tp
);
580 if (tp
->t_inpcb
->inp_socket
->so_options
& SO_DEBUG
)
581 tcp_trace(TA_USER
, ostate
, tp
, (void *)0, (struct tcphdr
*)0,
585 tcpstat
.tcps_keepdrops
++;
586 so_tmp
= tp
->t_inpcb
->inp_socket
;
587 tp
= tcp_drop(tp
, ETIMEDOUT
);
588 postevent(so_tmp
, 0, EV_TIMEOUT
);