xref: /freebsd/sys/netinet/tcp_timer.c (revision 85d94372502647815bd6bd3fc62f8867f663a71e)
1c398230bSWarner Losh /*-
2e79adb8eSGarrett Wollman  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3df8bae1dSRodney W. Grimes  *	The Regents of the University of California.  All rights reserved.
4df8bae1dSRodney W. Grimes  *
5df8bae1dSRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
6df8bae1dSRodney W. Grimes  * modification, are permitted provided that the following conditions
7df8bae1dSRodney W. Grimes  * are met:
8df8bae1dSRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
9df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
10df8bae1dSRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
11df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
12df8bae1dSRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
13df8bae1dSRodney W. Grimes  * 4. Neither the name of the University nor the names of its contributors
14df8bae1dSRodney W. Grimes  *    may be used to endorse or promote products derived from this software
15df8bae1dSRodney W. Grimes  *    without specific prior written permission.
16df8bae1dSRodney W. Grimes  *
17df8bae1dSRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18df8bae1dSRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19df8bae1dSRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20df8bae1dSRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21df8bae1dSRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22df8bae1dSRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23df8bae1dSRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24df8bae1dSRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25df8bae1dSRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26df8bae1dSRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27df8bae1dSRodney W. Grimes  * SUCH DAMAGE.
28df8bae1dSRodney W. Grimes  *
29e79adb8eSGarrett Wollman  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
30c3aac50fSPeter Wemm  * $FreeBSD$
31df8bae1dSRodney W. Grimes  */
32df8bae1dSRodney W. Grimes 
33fb59c426SYoshinobu Inoue #include "opt_inet6.h"
340cc12cc5SJoerg Wunsch #include "opt_tcpdebug.h"
350cc12cc5SJoerg Wunsch 
36df8bae1dSRodney W. Grimes #include <sys/param.h>
3798163b98SPoul-Henning Kamp #include <sys/kernel.h>
38c74af4faSBruce Evans #include <sys/lock.h>
3908517d53SMike Silbersack #include <sys/mbuf.h>
40c74af4faSBruce Evans #include <sys/mutex.h>
41c74af4faSBruce Evans #include <sys/protosw.h>
42df8bae1dSRodney W. Grimes #include <sys/socket.h>
43df8bae1dSRodney W. Grimes #include <sys/socketvar.h>
44c74af4faSBruce Evans #include <sys/sysctl.h>
45c74af4faSBruce Evans #include <sys/systm.h>
46e79adb8eSGarrett Wollman 
47df8bae1dSRodney W. Grimes #include <net/route.h>
48df8bae1dSRodney W. Grimes 
49df8bae1dSRodney W. Grimes #include <netinet/in.h>
50df8bae1dSRodney W. Grimes #include <netinet/in_pcb.h>
51c74af4faSBruce Evans #include <netinet/in_systm.h>
52fb59c426SYoshinobu Inoue #ifdef INET6
53fb59c426SYoshinobu Inoue #include <netinet6/in6_pcb.h>
54fb59c426SYoshinobu Inoue #endif
55df8bae1dSRodney W. Grimes #include <netinet/ip_var.h>
56df8bae1dSRodney W. Grimes #include <netinet/tcp.h>
57df8bae1dSRodney W. Grimes #include <netinet/tcp_fsm.h>
58df8bae1dSRodney W. Grimes #include <netinet/tcp_timer.h>
59df8bae1dSRodney W. Grimes #include <netinet/tcp_var.h>
60df8bae1dSRodney W. Grimes #include <netinet/tcpip.h>
61af7a2999SDavid Greenman #ifdef TCPDEBUG
62af7a2999SDavid Greenman #include <netinet/tcp_debug.h>
63af7a2999SDavid Greenman #endif
64df8bae1dSRodney W. Grimes 
659b8b58e0SJonathan Lemon int	tcp_keepinit;
66ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
67ccb4d0c6SJonathan Lemon     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "");
687b40aa32SPaul Traina 
699b8b58e0SJonathan Lemon int	tcp_keepidle;
70ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
71ccb4d0c6SJonathan Lemon     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "");
7298163b98SPoul-Henning Kamp 
739b8b58e0SJonathan Lemon int	tcp_keepintvl;
74ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
75ccb4d0c6SJonathan Lemon     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "");
7698163b98SPoul-Henning Kamp 
779b8b58e0SJonathan Lemon int	tcp_delacktime;
786489fe65SAndre Oppermann SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
796489fe65SAndre Oppermann     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
80ccb4d0c6SJonathan Lemon     "Time before a delayed ACK is sent");
819b8b58e0SJonathan Lemon 
829b8b58e0SJonathan Lemon int	tcp_msl;
83ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
84ccb4d0c6SJonathan Lemon     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
859b8b58e0SJonathan Lemon 
86701bec5aSMatthew Dillon int	tcp_rexmit_min;
87701bec5aSMatthew Dillon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
886489fe65SAndre Oppermann     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
896489fe65SAndre Oppermann     "Minimum Retransmission Timeout");
90701bec5aSMatthew Dillon 
91701bec5aSMatthew Dillon int	tcp_rexmit_slop;
92701bec5aSMatthew Dillon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
936489fe65SAndre Oppermann     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
946489fe65SAndre Oppermann     "Retransmission Timer Slop");
95701bec5aSMatthew Dillon 
96c39a614eSRobert Watson static int	always_keepalive = 1;
973d177f46SBill Fumerola SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
983d177f46SBill Fumerola     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
9934be9bf3SPoul-Henning Kamp 
1007c72af87SMohan Srinivasan int    tcp_fast_finwait2_recycle = 0;
1017c72af87SMohan Srinivasan SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
1026489fe65SAndre Oppermann     &tcp_fast_finwait2_recycle, 0,
1036489fe65SAndre Oppermann     "Recycle closed FIN_WAIT_2 connections faster");
1047c72af87SMohan Srinivasan 
1057c72af87SMohan Srinivasan int    tcp_finwait2_timeout;
1067c72af87SMohan Srinivasan SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
1076489fe65SAndre Oppermann     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
1087c72af87SMohan Srinivasan 
1097c72af87SMohan Srinivasan 
1100312fbe9SPoul-Henning Kamp static int	tcp_keepcnt = TCPTV_KEEPCNT;
1110312fbe9SPoul-Henning Kamp 	/* max idle probes */
1129b8b58e0SJonathan Lemon int	tcp_maxpersistidle;
1130312fbe9SPoul-Henning Kamp 	/* max idle time in persist */
114df8bae1dSRodney W. Grimes int	tcp_maxidle;
115e79adb8eSGarrett Wollman 
116df8bae1dSRodney W. Grimes /*
117df8bae1dSRodney W. Grimes  * Tcp protocol timeout routine called every 500 ms.
1189b8b58e0SJonathan Lemon  * Updates timestamps used for TCP
119df8bae1dSRodney W. Grimes  * causes finite state machine actions if timers expire.
120df8bae1dSRodney W. Grimes  */
121df8bae1dSRodney W. Grimes void
12285d94372SRobert Watson tcp_slowtimo()
123df8bae1dSRodney W. Grimes {
12415bd2b43SDavid Greenman 
125e79adb8eSGarrett Wollman 	tcp_maxidle = tcp_keepcnt * tcp_keepintvl;
126607b0b0cSJonathan Lemon 	INP_INFO_WLOCK(&tcbinfo);
1272104448fSAndre Oppermann 	(void) tcp_tw_2msl_scan(0);
128607b0b0cSJonathan Lemon 	INP_INFO_WUNLOCK(&tcbinfo);
129df8bae1dSRodney W. Grimes }
130df8bae1dSRodney W. Grimes 
1317d42e30cSJonathan Lemon int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
1327d42e30cSJonathan Lemon     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
1337d42e30cSJonathan Lemon 
134df8bae1dSRodney W. Grimes int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
135f058535dSJeffrey Hsu     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
136df8bae1dSRodney W. Grimes 
137f058535dSJeffrey Hsu static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
138e79adb8eSGarrett Wollman 
139623dce13SRobert Watson static int tcp_timer_race;
140623dce13SRobert Watson SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
141623dce13SRobert Watson     0, "Count of t_inpcb races on tcp_discardcb");
142623dce13SRobert Watson 
143df8bae1dSRodney W. Grimes /*
144df8bae1dSRodney W. Grimes  * TCP timer processing.
145df8bae1dSRodney W. Grimes  */
14685d94372SRobert Watson 
14785d94372SRobert Watson void
14885d94372SRobert Watson tcp_timer_delack(void *xtp)
149df8bae1dSRodney W. Grimes {
15085d94372SRobert Watson 	struct tcpcb *tp = xtp;
15185d94372SRobert Watson 	struct inpcb *inp;
15285d94372SRobert Watson 
15385d94372SRobert Watson 	INP_INFO_RLOCK(&tcbinfo);
15485d94372SRobert Watson 	inp = tp->t_inpcb;
15585d94372SRobert Watson 	/*
15685d94372SRobert Watson 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
15785d94372SRobert Watson 	 * tear-down mean we need it as a work-around for races between
15885d94372SRobert Watson 	 * timers and tcp_discardcb().
15985d94372SRobert Watson 	 *
16085d94372SRobert Watson 	 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
16185d94372SRobert Watson 	 */
16285d94372SRobert Watson 	if (inp == NULL) {
16385d94372SRobert Watson 		tcp_timer_race++;
16485d94372SRobert Watson 		INP_INFO_RUNLOCK(&tcbinfo);
16585d94372SRobert Watson 		return;
16685d94372SRobert Watson 	}
16785d94372SRobert Watson 	INP_LOCK(inp);
16885d94372SRobert Watson 	INP_INFO_RUNLOCK(&tcbinfo);
16985d94372SRobert Watson 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_delack)
17085d94372SRobert Watson 	    || !callout_active(tp->tt_delack)) {
17185d94372SRobert Watson 		INP_UNLOCK(inp);
17285d94372SRobert Watson 		return;
17385d94372SRobert Watson 	}
17485d94372SRobert Watson 	callout_deactivate(tp->tt_delack);
175df8bae1dSRodney W. Grimes 
1769b8b58e0SJonathan Lemon 	tp->t_flags |= TF_ACKNOW;
1779b8b58e0SJonathan Lemon 	tcpstat.tcps_delack++;
1789b8b58e0SJonathan Lemon 	(void) tcp_output(tp);
17985d94372SRobert Watson 	INP_UNLOCK(inp);
1809b8b58e0SJonathan Lemon }
1819b8b58e0SJonathan Lemon 
18285d94372SRobert Watson void
18385d94372SRobert Watson tcp_timer_2msl(void *xtp)
1849b8b58e0SJonathan Lemon {
18585d94372SRobert Watson 	struct tcpcb *tp = xtp;
18685d94372SRobert Watson 	struct inpcb *inp;
1879b8b58e0SJonathan Lemon #ifdef TCPDEBUG
1889b8b58e0SJonathan Lemon 	int ostate;
1899b8b58e0SJonathan Lemon 
1909b8b58e0SJonathan Lemon 	ostate = tp->t_state;
1919b8b58e0SJonathan Lemon #endif
192623dce13SRobert Watson 	/*
19385d94372SRobert Watson 	 * XXXRW: Does this actually happen?
19485d94372SRobert Watson 	 */
19585d94372SRobert Watson 	INP_INFO_WLOCK(&tcbinfo);
19685d94372SRobert Watson 	inp = tp->t_inpcb;
19785d94372SRobert Watson 	/*
19885d94372SRobert Watson 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
19985d94372SRobert Watson 	 * tear-down mean we need it as a work-around for races between
20085d94372SRobert Watson 	 * timers and tcp_discardcb().
20185d94372SRobert Watson 	 *
20285d94372SRobert Watson 	 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
20385d94372SRobert Watson 	 */
20485d94372SRobert Watson 	if (inp == NULL) {
20585d94372SRobert Watson 		tcp_timer_race++;
20685d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
20785d94372SRobert Watson 		return;
20885d94372SRobert Watson 	}
20985d94372SRobert Watson 	INP_LOCK(inp);
21085d94372SRobert Watson 	tcp_free_sackholes(tp);
21185d94372SRobert Watson 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_2msl) ||
21285d94372SRobert Watson 	    !callout_active(tp->tt_2msl)) {
21385d94372SRobert Watson 		INP_UNLOCK(tp->t_inpcb);
21485d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
21585d94372SRobert Watson 		return;
21685d94372SRobert Watson 	}
21785d94372SRobert Watson 	callout_deactivate(tp->tt_2msl);
21885d94372SRobert Watson 	/*
219df8bae1dSRodney W. Grimes 	 * 2 MSL timeout in shutdown went off.  If we're closed but
220df8bae1dSRodney W. Grimes 	 * still waiting for peer to close and connection has been idle
221df8bae1dSRodney W. Grimes 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
222df8bae1dSRodney W. Grimes 	 * control block.  Otherwise, check again in a bit.
2237c72af87SMohan Srinivasan 	 *
2247c72af87SMohan Srinivasan 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
2257c72af87SMohan Srinivasan 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
2267c72af87SMohan Srinivasan 	 * Ignore fact that there were recent incoming segments.
227df8bae1dSRodney W. Grimes 	 */
2287c72af87SMohan Srinivasan 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
22985d94372SRobert Watson 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
2307c72af87SMohan Srinivasan 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
2317c72af87SMohan Srinivasan 		tcpstat.tcps_finwait2_drops++;
23285d94372SRobert Watson 		tp = tcp_close(tp);
2337c72af87SMohan Srinivasan 	} else {
234df8bae1dSRodney W. Grimes 		if (tp->t_state != TCPS_TIME_WAIT &&
2359b8b58e0SJonathan Lemon 		   (ticks - tp->t_rcvtime) <= tcp_maxidle)
23685d94372SRobert Watson 		       callout_reset(tp->tt_2msl, tcp_keepintvl,
23785d94372SRobert Watson 				     tcp_timer_2msl, tp);
238df8bae1dSRodney W. Grimes 	       else
23985d94372SRobert Watson 		       tp = tcp_close(tp);
2407c72af87SMohan Srinivasan        }
241df8bae1dSRodney W. Grimes 
2429b8b58e0SJonathan Lemon #ifdef TCPDEBUG
243c214db75SRobert Watson 	if (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
244fb59c426SYoshinobu Inoue 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
2459b8b58e0SJonathan Lemon 			  PRU_SLOWTIMO);
2469b8b58e0SJonathan Lemon #endif
24785d94372SRobert Watson 	if (tp != NULL)
24885d94372SRobert Watson 		INP_UNLOCK(inp);
24985d94372SRobert Watson 	INP_INFO_WUNLOCK(&tcbinfo);
2509b8b58e0SJonathan Lemon }
2519b8b58e0SJonathan Lemon 
25285d94372SRobert Watson void
25385d94372SRobert Watson tcp_timer_keep(void *xtp)
2549b8b58e0SJonathan Lemon {
25585d94372SRobert Watson 	struct tcpcb *tp = xtp;
25608517d53SMike Silbersack 	struct tcptemp *t_template;
25785d94372SRobert Watson 	struct inpcb *inp;
2589b8b58e0SJonathan Lemon #ifdef TCPDEBUG
2599b8b58e0SJonathan Lemon 	int ostate;
2609b8b58e0SJonathan Lemon 
2619b8b58e0SJonathan Lemon 	ostate = tp->t_state;
2629b8b58e0SJonathan Lemon #endif
26385d94372SRobert Watson 	INP_INFO_WLOCK(&tcbinfo);
26485d94372SRobert Watson 	inp = tp->t_inpcb;
26585d94372SRobert Watson 	/*
26685d94372SRobert Watson 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
26785d94372SRobert Watson 	 * tear-down mean we need it as a work-around for races between
26885d94372SRobert Watson 	 * timers and tcp_discardcb().
26985d94372SRobert Watson 	 *
27085d94372SRobert Watson 	 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
27185d94372SRobert Watson 	 */
27285d94372SRobert Watson 	if (inp == NULL) {
27385d94372SRobert Watson 		tcp_timer_race++;
27485d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
27585d94372SRobert Watson 		return;
27685d94372SRobert Watson 	}
27785d94372SRobert Watson 	INP_LOCK(inp);
27885d94372SRobert Watson 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_keep)
27985d94372SRobert Watson 	    || !callout_active(tp->tt_keep)) {
28085d94372SRobert Watson 		INP_UNLOCK(inp);
28185d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
28285d94372SRobert Watson 		return;
28385d94372SRobert Watson 	}
28485d94372SRobert Watson 	callout_deactivate(tp->tt_keep);
2859b8b58e0SJonathan Lemon 	/*
2869b8b58e0SJonathan Lemon 	 * Keep-alive timer went off; send something
2879b8b58e0SJonathan Lemon 	 * or drop connection if idle for too long.
2889b8b58e0SJonathan Lemon 	 */
2899b8b58e0SJonathan Lemon 	tcpstat.tcps_keeptimeo++;
2909b8b58e0SJonathan Lemon 	if (tp->t_state < TCPS_ESTABLISHED)
2919b8b58e0SJonathan Lemon 		goto dropit;
2922a074620SSam Leffler 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
2939b8b58e0SJonathan Lemon 	    tp->t_state <= TCPS_CLOSING) {
2949b8b58e0SJonathan Lemon 		if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle)
2959b8b58e0SJonathan Lemon 			goto dropit;
2969b8b58e0SJonathan Lemon 		/*
2979b8b58e0SJonathan Lemon 		 * Send a packet designed to force a response
2989b8b58e0SJonathan Lemon 		 * if the peer is up and reachable:
2999b8b58e0SJonathan Lemon 		 * either an ACK if the connection is still alive,
3009b8b58e0SJonathan Lemon 		 * or an RST if the peer has closed the connection
3019b8b58e0SJonathan Lemon 		 * due to timeout or reboot.
3029b8b58e0SJonathan Lemon 		 * Using sequence number tp->snd_una-1
3039b8b58e0SJonathan Lemon 		 * causes the transmitted zero-length segment
3049b8b58e0SJonathan Lemon 		 * to lie outside the receive window;
3059b8b58e0SJonathan Lemon 		 * by the protocol spec, this requires the
3069b8b58e0SJonathan Lemon 		 * correspondent TCP to respond.
3079b8b58e0SJonathan Lemon 		 */
3089b8b58e0SJonathan Lemon 		tcpstat.tcps_keepprobe++;
30979909384SJonathan Lemon 		t_template = tcpip_maketemplate(inp);
31008517d53SMike Silbersack 		if (t_template) {
31108517d53SMike Silbersack 			tcp_respond(tp, t_template->tt_ipgen,
31208517d53SMike Silbersack 				    &t_template->tt_t, (struct mbuf *)NULL,
3139b8b58e0SJonathan Lemon 				    tp->rcv_nxt, tp->snd_una - 1, 0);
31408517d53SMike Silbersack 			(void) m_free(dtom(t_template));
31508517d53SMike Silbersack 		}
31685d94372SRobert Watson 		callout_reset(tp->tt_keep, tcp_keepintvl, tcp_timer_keep, tp);
3174cc20ab1SSeigo Tanimura 	} else
31885d94372SRobert Watson 		callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
3199b8b58e0SJonathan Lemon 
3209b8b58e0SJonathan Lemon #ifdef TCPDEBUG
3212a074620SSam Leffler 	if (inp->inp_socket->so_options & SO_DEBUG)
322fb59c426SYoshinobu Inoue 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
3239b8b58e0SJonathan Lemon 			  PRU_SLOWTIMO);
3249b8b58e0SJonathan Lemon #endif
32585d94372SRobert Watson 	INP_UNLOCK(inp);
32685d94372SRobert Watson 	INP_INFO_WUNLOCK(&tcbinfo);
32785d94372SRobert Watson 	return;
3289b8b58e0SJonathan Lemon 
3299b8b58e0SJonathan Lemon dropit:
3309b8b58e0SJonathan Lemon 	tcpstat.tcps_keepdrops++;
33185d94372SRobert Watson 	tp = tcp_drop(tp, ETIMEDOUT);
33285d94372SRobert Watson 
33385d94372SRobert Watson #ifdef TCPDEBUG
33485d94372SRobert Watson 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
33585d94372SRobert Watson 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
33685d94372SRobert Watson 			  PRU_SLOWTIMO);
33785d94372SRobert Watson #endif
33885d94372SRobert Watson 	if (tp != NULL)
33985d94372SRobert Watson 		INP_UNLOCK(tp->t_inpcb);
34085d94372SRobert Watson 	INP_INFO_WUNLOCK(&tcbinfo);
3419b8b58e0SJonathan Lemon }
3429b8b58e0SJonathan Lemon 
34385d94372SRobert Watson void
34485d94372SRobert Watson tcp_timer_persist(void *xtp)
3459b8b58e0SJonathan Lemon {
34685d94372SRobert Watson 	struct tcpcb *tp = xtp;
34785d94372SRobert Watson 	struct inpcb *inp;
3489b8b58e0SJonathan Lemon #ifdef TCPDEBUG
3499b8b58e0SJonathan Lemon 	int ostate;
3509b8b58e0SJonathan Lemon 
3519b8b58e0SJonathan Lemon 	ostate = tp->t_state;
3529b8b58e0SJonathan Lemon #endif
35385d94372SRobert Watson 	INP_INFO_WLOCK(&tcbinfo);
35485d94372SRobert Watson 	inp = tp->t_inpcb;
35585d94372SRobert Watson 	/*
35685d94372SRobert Watson 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
35785d94372SRobert Watson 	 * tear-down mean we need it as a work-around for races between
35885d94372SRobert Watson 	 * timers and tcp_discardcb().
35985d94372SRobert Watson 	 *
36085d94372SRobert Watson 	 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
36185d94372SRobert Watson 	 */
36285d94372SRobert Watson 	if (inp == NULL) {
36385d94372SRobert Watson 		tcp_timer_race++;
36485d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
36585d94372SRobert Watson 		return;
36685d94372SRobert Watson 	}
36785d94372SRobert Watson 	INP_LOCK(inp);
36885d94372SRobert Watson 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_persist)
36985d94372SRobert Watson 	    || !callout_active(tp->tt_persist)) {
37085d94372SRobert Watson 		INP_UNLOCK(inp);
37185d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
37285d94372SRobert Watson 		return;
37385d94372SRobert Watson 	}
37485d94372SRobert Watson 	callout_deactivate(tp->tt_persist);
3759b8b58e0SJonathan Lemon 	/*
3769b8b58e0SJonathan Lemon 	 * Persistance timer into zero window.
3779b8b58e0SJonathan Lemon 	 * Force a byte to be output, if possible.
3789b8b58e0SJonathan Lemon 	 */
3799b8b58e0SJonathan Lemon 	tcpstat.tcps_persisttimeo++;
3809b8b58e0SJonathan Lemon 	/*
3819b8b58e0SJonathan Lemon 	 * Hack: if the peer is dead/unreachable, we do not
3829b8b58e0SJonathan Lemon 	 * time out if the window is closed.  After a full
3839b8b58e0SJonathan Lemon 	 * backoff, drop the connection if the idle time
3849b8b58e0SJonathan Lemon 	 * (no responses to probes) reaches the maximum
3859b8b58e0SJonathan Lemon 	 * backoff that we would use if retransmitting.
3869b8b58e0SJonathan Lemon 	 */
3879b8b58e0SJonathan Lemon 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
3889b8b58e0SJonathan Lemon 	    ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle ||
3899b8b58e0SJonathan Lemon 	     (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
3909b8b58e0SJonathan Lemon 		tcpstat.tcps_persistdrop++;
39185d94372SRobert Watson 		tp = tcp_drop(tp, ETIMEDOUT);
39285d94372SRobert Watson 		goto out;
3939b8b58e0SJonathan Lemon 	}
3949b8b58e0SJonathan Lemon 	tcp_setpersist(tp);
3952cdbfa66SPaul Saab 	tp->t_flags |= TF_FORCEDATA;
3969b8b58e0SJonathan Lemon 	(void) tcp_output(tp);
3972cdbfa66SPaul Saab 	tp->t_flags &= ~TF_FORCEDATA;
3989b8b58e0SJonathan Lemon 
39985d94372SRobert Watson out:
4009b8b58e0SJonathan Lemon #ifdef TCPDEBUG
401ffb761f6SGleb Smirnoff 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
402ffb761f6SGleb Smirnoff 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
4039b8b58e0SJonathan Lemon #endif
40485d94372SRobert Watson 	if (tp != NULL)
40585d94372SRobert Watson 		INP_UNLOCK(inp);
40685d94372SRobert Watson 	INP_INFO_WUNLOCK(&tcbinfo);
4079b8b58e0SJonathan Lemon }
4089b8b58e0SJonathan Lemon 
40985d94372SRobert Watson void
41085d94372SRobert Watson tcp_timer_rexmt(void * xtp)
4119b8b58e0SJonathan Lemon {
41285d94372SRobert Watson 	struct tcpcb *tp = xtp;
4139b8b58e0SJonathan Lemon 	int rexmt;
41485d94372SRobert Watson 	int headlocked;
41585d94372SRobert Watson 	struct inpcb *inp;
4169b8b58e0SJonathan Lemon #ifdef TCPDEBUG
4179b8b58e0SJonathan Lemon 	int ostate;
4189b8b58e0SJonathan Lemon 
4199b8b58e0SJonathan Lemon 	ostate = tp->t_state;
4209b8b58e0SJonathan Lemon #endif
42185d94372SRobert Watson 	INP_INFO_WLOCK(&tcbinfo);
42285d94372SRobert Watson 	headlocked = 1;
42385d94372SRobert Watson 	inp = tp->t_inpcb;
42485d94372SRobert Watson 	/*
42585d94372SRobert Watson 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
42685d94372SRobert Watson 	 * tear-down mean we need it as a work-around for races between
42785d94372SRobert Watson 	 * timers and tcp_discardcb().
42885d94372SRobert Watson 	 *
42985d94372SRobert Watson 	 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
43085d94372SRobert Watson 	 */
43185d94372SRobert Watson 	if (inp == NULL) {
43285d94372SRobert Watson 		tcp_timer_race++;
43385d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
43485d94372SRobert Watson 		return;
43585d94372SRobert Watson 	}
43685d94372SRobert Watson 	INP_LOCK(inp);
43785d94372SRobert Watson 	if ((inp->inp_vflag & INP_DROPPED) || callout_pending(tp->tt_rexmt)
43885d94372SRobert Watson 	    || !callout_active(tp->tt_rexmt)) {
43985d94372SRobert Watson 		INP_UNLOCK(inp);
44085d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
44185d94372SRobert Watson 		return;
44285d94372SRobert Watson 	}
44385d94372SRobert Watson 	callout_deactivate(tp->tt_rexmt);
4446d90faf3SPaul Saab 	tcp_free_sackholes(tp);
445df8bae1dSRodney W. Grimes 	/*
446df8bae1dSRodney W. Grimes 	 * Retransmission timer went off.  Message has not
447df8bae1dSRodney W. Grimes 	 * been acked within retransmit interval.  Back off
448df8bae1dSRodney W. Grimes 	 * to a longer retransmit interval and retransmit one segment.
449df8bae1dSRodney W. Grimes 	 */
450df8bae1dSRodney W. Grimes 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
451df8bae1dSRodney W. Grimes 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
452df8bae1dSRodney W. Grimes 		tcpstat.tcps_timeoutdrop++;
45385d94372SRobert Watson 		tp = tcp_drop(tp, tp->t_softerror ?
45485d94372SRobert Watson 			      tp->t_softerror : ETIMEDOUT);
45585d94372SRobert Watson 		goto out;
4569b8b58e0SJonathan Lemon 	}
45785d94372SRobert Watson 	INP_INFO_WUNLOCK(&tcbinfo);
45885d94372SRobert Watson 	headlocked = 0;
4599b8b58e0SJonathan Lemon 	if (tp->t_rxtshift == 1) {
4609b8b58e0SJonathan Lemon 		/*
4619b8b58e0SJonathan Lemon 		 * first retransmit; record ssthresh and cwnd so they can
4629b8b58e0SJonathan Lemon 		 * be recovered if this turns out to be a "bad" retransmit.
4639b8b58e0SJonathan Lemon 		 * A retransmit is considered "bad" if an ACK for this
4649b8b58e0SJonathan Lemon 		 * segment is received within RTT/2 interval; the assumption
4659b8b58e0SJonathan Lemon 		 * here is that the ACK was already in flight.  See
4669b8b58e0SJonathan Lemon 		 * "On Estimating End-to-End Network Path Properties" by
4679b8b58e0SJonathan Lemon 		 * Allman and Paxson for more details.
4689b8b58e0SJonathan Lemon 		 */
4699b8b58e0SJonathan Lemon 		tp->snd_cwnd_prev = tp->snd_cwnd;
4709b8b58e0SJonathan Lemon 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
4719d11646dSJeffrey Hsu 		tp->snd_recover_prev = tp->snd_recover;
4729d11646dSJeffrey Hsu 		if (IN_FASTRECOVERY(tp))
4739d11646dSJeffrey Hsu 		  tp->t_flags |= TF_WASFRECOVERY;
4749d11646dSJeffrey Hsu 		else
4759d11646dSJeffrey Hsu 		  tp->t_flags &= ~TF_WASFRECOVERY;
4769b8b58e0SJonathan Lemon 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
477df8bae1dSRodney W. Grimes 	}
478df8bae1dSRodney W. Grimes 	tcpstat.tcps_rexmttimeo++;
4797d42e30cSJonathan Lemon 	if (tp->t_state == TCPS_SYN_SENT)
4807d42e30cSJonathan Lemon 		rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift];
4817d42e30cSJonathan Lemon 	else
482df8bae1dSRodney W. Grimes 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
483df8bae1dSRodney W. Grimes 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
484df8bae1dSRodney W. Grimes 		      tp->t_rttmin, TCPTV_REXMTMAX);
485df8bae1dSRodney W. Grimes 	/*
486c94c54e4SAndre Oppermann 	 * Disable rfc1323 if we havn't got any response to
4877ceb7783SJesper Skriver 	 * our third SYN to work-around some broken terminal servers
4887ceb7783SJesper Skriver 	 * (most of which have hopefully been retired) that have bad VJ
4897ceb7783SJesper Skriver 	 * header compression code which trashes TCP segments containing
4907ceb7783SJesper Skriver 	 * unknown-to-them TCP options.
4917ceb7783SJesper Skriver 	 */
4927ceb7783SJesper Skriver 	if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3))
493c94c54e4SAndre Oppermann 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP);
4947ceb7783SJesper Skriver 	/*
49597d8d152SAndre Oppermann 	 * If we backed off this far, our srtt estimate is probably bogus.
49697d8d152SAndre Oppermann 	 * Clobber it so we'll take the next rtt measurement as our srtt;
497df8bae1dSRodney W. Grimes 	 * move the current srtt into rttvar to keep the current
498df8bae1dSRodney W. Grimes 	 * retransmit times until then.
499df8bae1dSRodney W. Grimes 	 */
500df8bae1dSRodney W. Grimes 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
501fb59c426SYoshinobu Inoue #ifdef INET6
502fb59c426SYoshinobu Inoue 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
503fb59c426SYoshinobu Inoue 			in6_losing(tp->t_inpcb);
504fb59c426SYoshinobu Inoue 		else
505fb59c426SYoshinobu Inoue #endif
506df8bae1dSRodney W. Grimes 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
507df8bae1dSRodney W. Grimes 		tp->t_srtt = 0;
508df8bae1dSRodney W. Grimes 	}
509df8bae1dSRodney W. Grimes 	tp->snd_nxt = tp->snd_una;
5109d11646dSJeffrey Hsu 	tp->snd_recover = tp->snd_max;
51146f58482SJonathan Lemon 	/*
51274b48c1dSAndras Olah 	 * Force a segment to be sent.
51374b48c1dSAndras Olah 	 */
51474b48c1dSAndras Olah 	tp->t_flags |= TF_ACKNOW;
51574b48c1dSAndras Olah 	/*
516df8bae1dSRodney W. Grimes 	 * If timing a segment in this window, stop the timer.
517df8bae1dSRodney W. Grimes 	 */
5189b8b58e0SJonathan Lemon 	tp->t_rtttime = 0;
519df8bae1dSRodney W. Grimes 	/*
520df8bae1dSRodney W. Grimes 	 * Close the congestion window down to one segment
521df8bae1dSRodney W. Grimes 	 * (we'll open it by one segment for each ack we get).
522df8bae1dSRodney W. Grimes 	 * Since we probably have a window's worth of unacked
523df8bae1dSRodney W. Grimes 	 * data accumulated, this "slow start" keeps us from
524df8bae1dSRodney W. Grimes 	 * dumping all that data as back-to-back packets (which
525df8bae1dSRodney W. Grimes 	 * might overwhelm an intermediate gateway).
526df8bae1dSRodney W. Grimes 	 *
527df8bae1dSRodney W. Grimes 	 * There are two phases to the opening: Initially we
528df8bae1dSRodney W. Grimes 	 * open by one mss on each ack.  This makes the window
529df8bae1dSRodney W. Grimes 	 * size increase exponentially with time.  If the
530df8bae1dSRodney W. Grimes 	 * window is larger than the path can handle, this
531df8bae1dSRodney W. Grimes 	 * exponential growth results in dropped packet(s)
532df8bae1dSRodney W. Grimes 	 * almost immediately.  To get more time between
533df8bae1dSRodney W. Grimes 	 * drops but still "push" the network to take advantage
534df8bae1dSRodney W. Grimes 	 * of improving conditions, we switch from exponential
535df8bae1dSRodney W. Grimes 	 * to linear window opening at some threshhold size.
536df8bae1dSRodney W. Grimes 	 * For a threshhold, we use half the current window
537df8bae1dSRodney W. Grimes 	 * size, truncated to a multiple of the mss.
538df8bae1dSRodney W. Grimes 	 *
539df8bae1dSRodney W. Grimes 	 * (the minimum cwnd that will give us exponential
540df8bae1dSRodney W. Grimes 	 * growth is 2 mss.  We don't allow the threshhold
541df8bae1dSRodney W. Grimes 	 * to go below this.)
542df8bae1dSRodney W. Grimes 	 */
543df8bae1dSRodney W. Grimes 	{
544df8bae1dSRodney W. Grimes 		u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
545df8bae1dSRodney W. Grimes 		if (win < 2)
546df8bae1dSRodney W. Grimes 			win = 2;
547df8bae1dSRodney W. Grimes 		tp->snd_cwnd = tp->t_maxseg;
548df8bae1dSRodney W. Grimes 		tp->snd_ssthresh = win * tp->t_maxseg;
549df8bae1dSRodney W. Grimes 		tp->t_dupacks = 0;
550df8bae1dSRodney W. Grimes 	}
5519d11646dSJeffrey Hsu 	EXIT_FASTRECOVERY(tp);
552df8bae1dSRodney W. Grimes 	(void) tcp_output(tp);
553df8bae1dSRodney W. Grimes 
55485d94372SRobert Watson out:
5559b8b58e0SJonathan Lemon #ifdef TCPDEBUG
5561c53f806SRobert Watson 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
557fb59c426SYoshinobu Inoue 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
5589b8b58e0SJonathan Lemon 			  PRU_SLOWTIMO);
559df8bae1dSRodney W. Grimes #endif
56085d94372SRobert Watson 	if (tp != NULL)
56185d94372SRobert Watson 		INP_UNLOCK(inp);
56285d94372SRobert Watson 	if (headlocked)
56385d94372SRobert Watson 		INP_INFO_WUNLOCK(&tcbinfo);
56485d94372SRobert Watson }
56585d94372SRobert Watson 
56685d94372SRobert Watson void
56785d94372SRobert Watson tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
56885d94372SRobert Watson {
56985d94372SRobert Watson 	struct callout *t_callout;
57085d94372SRobert Watson 	void *f_callout;
57185d94372SRobert Watson 
57285d94372SRobert Watson 	switch (timer_type) {
57385d94372SRobert Watson 		case TT_DELACK:
57485d94372SRobert Watson 			t_callout = tp->tt_delack;
57585d94372SRobert Watson 			f_callout = tcp_timer_delack;
57685d94372SRobert Watson 			break;
57785d94372SRobert Watson 		case TT_REXMT:
57885d94372SRobert Watson 			t_callout = tp->tt_rexmt;
57985d94372SRobert Watson 			f_callout = tcp_timer_rexmt;
58085d94372SRobert Watson 			break;
58185d94372SRobert Watson 		case TT_PERSIST:
58285d94372SRobert Watson 			t_callout = tp->tt_persist;
58385d94372SRobert Watson 			f_callout = tcp_timer_persist;
58485d94372SRobert Watson 			break;
58585d94372SRobert Watson 		case TT_KEEP:
58685d94372SRobert Watson 			t_callout = tp->tt_keep;
58785d94372SRobert Watson 			f_callout = tcp_timer_keep;
58885d94372SRobert Watson 			break;
58985d94372SRobert Watson 		case TT_2MSL:
59085d94372SRobert Watson 			t_callout = tp->tt_2msl;
59185d94372SRobert Watson 			f_callout = tcp_timer_2msl;
59285d94372SRobert Watson 			break;
59385d94372SRobert Watson 		default:
59485d94372SRobert Watson 			panic("bad timer_type");
59585d94372SRobert Watson 		}
59685d94372SRobert Watson 	if (delta == 0) {
59785d94372SRobert Watson 		callout_stop(t_callout);
59885d94372SRobert Watson 	} else {
59985d94372SRobert Watson 		callout_reset(t_callout, delta, f_callout, tp);
60085d94372SRobert Watson 	}
60185d94372SRobert Watson }
60285d94372SRobert Watson 
60385d94372SRobert Watson int
60485d94372SRobert Watson tcp_timer_active(struct tcpcb *tp, int timer_type)
60585d94372SRobert Watson {
60685d94372SRobert Watson 	struct callout *t_callout;
60785d94372SRobert Watson 
60885d94372SRobert Watson 	switch (timer_type) {
60985d94372SRobert Watson 		case TT_DELACK:
61085d94372SRobert Watson 			t_callout = tp->tt_delack;
61185d94372SRobert Watson 			break;
61285d94372SRobert Watson 		case TT_REXMT:
61385d94372SRobert Watson 			t_callout = tp->tt_rexmt;
61485d94372SRobert Watson 			break;
61585d94372SRobert Watson 		case TT_PERSIST:
61685d94372SRobert Watson 			t_callout = tp->tt_persist;
61785d94372SRobert Watson 			break;
61885d94372SRobert Watson 		case TT_KEEP:
61985d94372SRobert Watson 			t_callout = tp->tt_keep;
62085d94372SRobert Watson 			break;
62185d94372SRobert Watson 		case TT_2MSL:
62285d94372SRobert Watson 			t_callout = tp->tt_2msl;
62385d94372SRobert Watson 			break;
62485d94372SRobert Watson 		default:
62585d94372SRobert Watson 			panic("bad timer_type");
62685d94372SRobert Watson 		}
62785d94372SRobert Watson 	return callout_active(t_callout);
628df8bae1dSRodney W. Grimes }
629