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 30df8bae1dSRodney W. Grimes */ 31df8bae1dSRodney W. Grimes 324b421e2dSMike Silbersack #include <sys/cdefs.h> 334b421e2dSMike Silbersack __FBSDID("$FreeBSD$"); 344b421e2dSMike Silbersack 35fb59c426SYoshinobu Inoue #include "opt_inet6.h" 360cc12cc5SJoerg Wunsch #include "opt_tcpdebug.h" 370cc12cc5SJoerg Wunsch 38df8bae1dSRodney W. Grimes #include <sys/param.h> 3998163b98SPoul-Henning Kamp #include <sys/kernel.h> 40c74af4faSBruce Evans #include <sys/lock.h> 4108517d53SMike Silbersack #include <sys/mbuf.h> 42c74af4faSBruce Evans #include <sys/mutex.h> 43c74af4faSBruce Evans #include <sys/protosw.h> 44df8bae1dSRodney W. Grimes #include <sys/socket.h> 45df8bae1dSRodney W. Grimes #include <sys/socketvar.h> 46c74af4faSBruce Evans #include <sys/sysctl.h> 47c74af4faSBruce Evans #include <sys/systm.h> 48e79adb8eSGarrett Wollman 49df8bae1dSRodney W. Grimes #include <net/route.h> 50df8bae1dSRodney W. Grimes 51df8bae1dSRodney W. Grimes #include <netinet/in.h> 52df8bae1dSRodney W. Grimes #include <netinet/in_pcb.h> 53c74af4faSBruce Evans #include <netinet/in_systm.h> 54fb59c426SYoshinobu Inoue #ifdef INET6 55fb59c426SYoshinobu Inoue #include <netinet6/in6_pcb.h> 56fb59c426SYoshinobu Inoue #endif 57df8bae1dSRodney W. Grimes #include <netinet/ip_var.h> 58df8bae1dSRodney W. Grimes #include <netinet/tcp.h> 59df8bae1dSRodney W. Grimes #include <netinet/tcp_fsm.h> 60df8bae1dSRodney W. Grimes #include <netinet/tcp_timer.h> 61df8bae1dSRodney W. Grimes #include <netinet/tcp_var.h> 62df8bae1dSRodney W. Grimes #include <netinet/tcpip.h> 63af7a2999SDavid Greenman #ifdef TCPDEBUG 64af7a2999SDavid Greenman #include <netinet/tcp_debug.h> 65af7a2999SDavid Greenman #endif 66df8bae1dSRodney W. Grimes 679b8b58e0SJonathan Lemon int tcp_keepinit; 68ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW, 69ccb4d0c6SJonathan Lemon &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", ""); 707b40aa32SPaul Traina 719b8b58e0SJonathan Lemon int tcp_keepidle; 72ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW, 73ccb4d0c6SJonathan Lemon &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", ""); 7498163b98SPoul-Henning Kamp 759b8b58e0SJonathan Lemon int tcp_keepintvl; 76ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW, 77ccb4d0c6SJonathan Lemon &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", ""); 7898163b98SPoul-Henning Kamp 799b8b58e0SJonathan Lemon int tcp_delacktime; 806489fe65SAndre Oppermann SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW, 816489fe65SAndre Oppermann &tcp_delacktime, 0, sysctl_msec_to_ticks, "I", 82ccb4d0c6SJonathan Lemon "Time before a delayed ACK is sent"); 839b8b58e0SJonathan Lemon 849b8b58e0SJonathan Lemon int tcp_msl; 85ccb4d0c6SJonathan Lemon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW, 86ccb4d0c6SJonathan Lemon &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime"); 879b8b58e0SJonathan Lemon 88701bec5aSMatthew Dillon int tcp_rexmit_min; 89701bec5aSMatthew Dillon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW, 906489fe65SAndre Oppermann &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I", 916489fe65SAndre Oppermann "Minimum Retransmission Timeout"); 92701bec5aSMatthew Dillon 93701bec5aSMatthew Dillon int tcp_rexmit_slop; 94701bec5aSMatthew Dillon SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW, 956489fe65SAndre Oppermann &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I", 966489fe65SAndre Oppermann "Retransmission Timer Slop"); 97701bec5aSMatthew Dillon 98c39a614eSRobert Watson static int always_keepalive = 1; 993d177f46SBill Fumerola SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW, 1003d177f46SBill Fumerola &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections"); 10134be9bf3SPoul-Henning Kamp 1027c72af87SMohan Srinivasan int tcp_fast_finwait2_recycle = 0; 1037c72af87SMohan Srinivasan SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW, 1046489fe65SAndre Oppermann &tcp_fast_finwait2_recycle, 0, 1056489fe65SAndre Oppermann "Recycle closed FIN_WAIT_2 connections faster"); 1067c72af87SMohan Srinivasan 1077c72af87SMohan Srinivasan int tcp_finwait2_timeout; 1087c72af87SMohan Srinivasan SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW, 1096489fe65SAndre Oppermann &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout"); 1107c72af87SMohan Srinivasan 1117c72af87SMohan Srinivasan 1120312fbe9SPoul-Henning Kamp static int tcp_keepcnt = TCPTV_KEEPCNT; 1130312fbe9SPoul-Henning Kamp /* max idle probes */ 1149b8b58e0SJonathan Lemon int tcp_maxpersistidle; 1150312fbe9SPoul-Henning Kamp /* max idle time in persist */ 116df8bae1dSRodney W. Grimes int tcp_maxidle; 117e79adb8eSGarrett Wollman 118df8bae1dSRodney W. Grimes /* 119df8bae1dSRodney W. Grimes * Tcp protocol timeout routine called every 500 ms. 1209b8b58e0SJonathan Lemon * Updates timestamps used for TCP 121df8bae1dSRodney W. Grimes * causes finite state machine actions if timers expire. 122df8bae1dSRodney W. Grimes */ 123df8bae1dSRodney W. Grimes void 124e2f2059fSMike Silbersack tcp_slowtimo(void) 125df8bae1dSRodney W. Grimes { 12615bd2b43SDavid Greenman 127e79adb8eSGarrett Wollman tcp_maxidle = tcp_keepcnt * tcp_keepintvl; 128607b0b0cSJonathan Lemon INP_INFO_WLOCK(&tcbinfo); 1292104448fSAndre Oppermann (void) tcp_tw_2msl_scan(0); 130607b0b0cSJonathan Lemon INP_INFO_WUNLOCK(&tcbinfo); 131df8bae1dSRodney W. Grimes } 132df8bae1dSRodney W. Grimes 1337d42e30cSJonathan Lemon int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] = 1347d42e30cSJonathan Lemon { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 }; 1357d42e30cSJonathan Lemon 136df8bae1dSRodney W. Grimes int tcp_backoff[TCP_MAXRXTSHIFT + 1] = 137f058535dSJeffrey Hsu { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 }; 138df8bae1dSRodney W. Grimes 139f058535dSJeffrey Hsu static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */ 140e79adb8eSGarrett Wollman 141623dce13SRobert Watson static int tcp_timer_race; 142623dce13SRobert Watson SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race, 143623dce13SRobert Watson 0, "Count of t_inpcb races on tcp_discardcb"); 144623dce13SRobert Watson 145df8bae1dSRodney W. Grimes /* 146df8bae1dSRodney W. Grimes * TCP timer processing. 147df8bae1dSRodney W. Grimes */ 14885d94372SRobert Watson 14985d94372SRobert Watson void 15085d94372SRobert Watson tcp_timer_delack(void *xtp) 151df8bae1dSRodney W. Grimes { 15285d94372SRobert Watson struct tcpcb *tp = xtp; 15385d94372SRobert Watson struct inpcb *inp; 15485d94372SRobert Watson 15585d94372SRobert Watson INP_INFO_RLOCK(&tcbinfo); 15685d94372SRobert Watson inp = tp->t_inpcb; 15785d94372SRobert Watson /* 15885d94372SRobert Watson * XXXRW: While this assert is in fact correct, bugs in the tcpcb 15985d94372SRobert Watson * tear-down mean we need it as a work-around for races between 16085d94372SRobert Watson * timers and tcp_discardcb(). 16185d94372SRobert Watson * 16285d94372SRobert Watson * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL")); 16385d94372SRobert Watson */ 16485d94372SRobert Watson if (inp == NULL) { 16585d94372SRobert Watson tcp_timer_race++; 16685d94372SRobert Watson INP_INFO_RUNLOCK(&tcbinfo); 16785d94372SRobert Watson return; 16885d94372SRobert Watson } 16985d94372SRobert Watson INP_LOCK(inp); 17085d94372SRobert Watson INP_INFO_RUNLOCK(&tcbinfo); 171e2f2059fSMike Silbersack if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_delack) 172e2f2059fSMike Silbersack || !callout_active(&tp->t_timers->tt_delack)) { 17385d94372SRobert Watson INP_UNLOCK(inp); 17485d94372SRobert Watson return; 17585d94372SRobert Watson } 176e2f2059fSMike Silbersack callout_deactivate(&tp->t_timers->tt_delack); 177df8bae1dSRodney W. Grimes 1789b8b58e0SJonathan Lemon tp->t_flags |= TF_ACKNOW; 1799b8b58e0SJonathan Lemon tcpstat.tcps_delack++; 1809b8b58e0SJonathan Lemon (void) tcp_output(tp); 18185d94372SRobert Watson INP_UNLOCK(inp); 1829b8b58e0SJonathan Lemon } 1839b8b58e0SJonathan Lemon 18485d94372SRobert Watson void 18585d94372SRobert Watson tcp_timer_2msl(void *xtp) 1869b8b58e0SJonathan Lemon { 18785d94372SRobert Watson struct tcpcb *tp = xtp; 18885d94372SRobert Watson struct inpcb *inp; 1899b8b58e0SJonathan Lemon #ifdef TCPDEBUG 1909b8b58e0SJonathan Lemon int ostate; 1919b8b58e0SJonathan Lemon 1929b8b58e0SJonathan Lemon ostate = tp->t_state; 1939b8b58e0SJonathan Lemon #endif 194623dce13SRobert Watson /* 19585d94372SRobert Watson * XXXRW: Does this actually happen? 19685d94372SRobert Watson */ 19785d94372SRobert Watson INP_INFO_WLOCK(&tcbinfo); 19885d94372SRobert Watson inp = tp->t_inpcb; 19985d94372SRobert Watson /* 20085d94372SRobert Watson * XXXRW: While this assert is in fact correct, bugs in the tcpcb 20185d94372SRobert Watson * tear-down mean we need it as a work-around for races between 20285d94372SRobert Watson * timers and tcp_discardcb(). 20385d94372SRobert Watson * 20485d94372SRobert Watson * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL")); 20585d94372SRobert Watson */ 20685d94372SRobert Watson if (inp == NULL) { 20785d94372SRobert Watson tcp_timer_race++; 20885d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 20985d94372SRobert Watson return; 21085d94372SRobert Watson } 21185d94372SRobert Watson INP_LOCK(inp); 21285d94372SRobert Watson tcp_free_sackholes(tp); 213e2f2059fSMike Silbersack if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_2msl) || 214e2f2059fSMike Silbersack !callout_active(&tp->t_timers->tt_2msl)) { 21585d94372SRobert Watson INP_UNLOCK(tp->t_inpcb); 21685d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 21785d94372SRobert Watson return; 21885d94372SRobert Watson } 219e2f2059fSMike Silbersack callout_deactivate(&tp->t_timers->tt_2msl); 22085d94372SRobert Watson /* 221df8bae1dSRodney W. Grimes * 2 MSL timeout in shutdown went off. If we're closed but 222df8bae1dSRodney W. Grimes * still waiting for peer to close and connection has been idle 223df8bae1dSRodney W. Grimes * too long, or if 2MSL time is up from TIME_WAIT, delete connection 224df8bae1dSRodney W. Grimes * control block. Otherwise, check again in a bit. 2257c72af87SMohan Srinivasan * 2267c72af87SMohan Srinivasan * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed, 2277c72af87SMohan Srinivasan * there's no point in hanging onto FIN_WAIT_2 socket. Just close it. 2287c72af87SMohan Srinivasan * Ignore fact that there were recent incoming segments. 229df8bae1dSRodney W. Grimes */ 2307c72af87SMohan Srinivasan if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 && 23185d94372SRobert Watson tp->t_inpcb && tp->t_inpcb->inp_socket && 2327c72af87SMohan Srinivasan (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) { 2337c72af87SMohan Srinivasan tcpstat.tcps_finwait2_drops++; 23485d94372SRobert Watson tp = tcp_close(tp); 2357c72af87SMohan Srinivasan } else { 236df8bae1dSRodney W. Grimes if (tp->t_state != TCPS_TIME_WAIT && 2379b8b58e0SJonathan Lemon (ticks - tp->t_rcvtime) <= tcp_maxidle) 238e2f2059fSMike Silbersack callout_reset(&tp->t_timers->tt_2msl, tcp_keepintvl, 23985d94372SRobert Watson tcp_timer_2msl, tp); 240df8bae1dSRodney W. Grimes else 24185d94372SRobert Watson tp = tcp_close(tp); 2427c72af87SMohan Srinivasan } 243df8bae1dSRodney W. Grimes 2449b8b58e0SJonathan Lemon #ifdef TCPDEBUG 245586b4a0eSKonstantin Belousov if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 246fb59c426SYoshinobu Inoue tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 2479b8b58e0SJonathan Lemon PRU_SLOWTIMO); 2489b8b58e0SJonathan Lemon #endif 24985d94372SRobert Watson if (tp != NULL) 25085d94372SRobert Watson INP_UNLOCK(inp); 25185d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 2529b8b58e0SJonathan Lemon } 2539b8b58e0SJonathan Lemon 25485d94372SRobert Watson void 25585d94372SRobert Watson tcp_timer_keep(void *xtp) 2569b8b58e0SJonathan Lemon { 25785d94372SRobert Watson struct tcpcb *tp = xtp; 25808517d53SMike Silbersack struct tcptemp *t_template; 25985d94372SRobert Watson struct inpcb *inp; 2609b8b58e0SJonathan Lemon #ifdef TCPDEBUG 2619b8b58e0SJonathan Lemon int ostate; 2629b8b58e0SJonathan Lemon 2639b8b58e0SJonathan Lemon ostate = tp->t_state; 2649b8b58e0SJonathan Lemon #endif 26585d94372SRobert Watson INP_INFO_WLOCK(&tcbinfo); 26685d94372SRobert Watson inp = tp->t_inpcb; 26785d94372SRobert Watson /* 26885d94372SRobert Watson * XXXRW: While this assert is in fact correct, bugs in the tcpcb 26985d94372SRobert Watson * tear-down mean we need it as a work-around for races between 27085d94372SRobert Watson * timers and tcp_discardcb(). 27185d94372SRobert Watson * 27285d94372SRobert Watson * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL")); 27385d94372SRobert Watson */ 27485d94372SRobert Watson if (inp == NULL) { 27585d94372SRobert Watson tcp_timer_race++; 27685d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 27785d94372SRobert Watson return; 27885d94372SRobert Watson } 27985d94372SRobert Watson INP_LOCK(inp); 280e2f2059fSMike Silbersack if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_keep) 281e2f2059fSMike Silbersack || !callout_active(&tp->t_timers->tt_keep)) { 28285d94372SRobert Watson INP_UNLOCK(inp); 28385d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 28485d94372SRobert Watson return; 28585d94372SRobert Watson } 286e2f2059fSMike Silbersack callout_deactivate(&tp->t_timers->tt_keep); 2879b8b58e0SJonathan Lemon /* 2889b8b58e0SJonathan Lemon * Keep-alive timer went off; send something 2899b8b58e0SJonathan Lemon * or drop connection if idle for too long. 2909b8b58e0SJonathan Lemon */ 2919b8b58e0SJonathan Lemon tcpstat.tcps_keeptimeo++; 2929b8b58e0SJonathan Lemon if (tp->t_state < TCPS_ESTABLISHED) 2939b8b58e0SJonathan Lemon goto dropit; 2942a074620SSam Leffler if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) && 2959b8b58e0SJonathan Lemon tp->t_state <= TCPS_CLOSING) { 2969b8b58e0SJonathan Lemon if ((ticks - tp->t_rcvtime) >= tcp_keepidle + tcp_maxidle) 2979b8b58e0SJonathan Lemon goto dropit; 2989b8b58e0SJonathan Lemon /* 2999b8b58e0SJonathan Lemon * Send a packet designed to force a response 3009b8b58e0SJonathan Lemon * if the peer is up and reachable: 3019b8b58e0SJonathan Lemon * either an ACK if the connection is still alive, 3029b8b58e0SJonathan Lemon * or an RST if the peer has closed the connection 3039b8b58e0SJonathan Lemon * due to timeout or reboot. 3049b8b58e0SJonathan Lemon * Using sequence number tp->snd_una-1 3059b8b58e0SJonathan Lemon * causes the transmitted zero-length segment 3069b8b58e0SJonathan Lemon * to lie outside the receive window; 3079b8b58e0SJonathan Lemon * by the protocol spec, this requires the 3089b8b58e0SJonathan Lemon * correspondent TCP to respond. 3099b8b58e0SJonathan Lemon */ 3109b8b58e0SJonathan Lemon tcpstat.tcps_keepprobe++; 31179909384SJonathan Lemon t_template = tcpip_maketemplate(inp); 31208517d53SMike Silbersack if (t_template) { 31308517d53SMike Silbersack tcp_respond(tp, t_template->tt_ipgen, 31408517d53SMike Silbersack &t_template->tt_t, (struct mbuf *)NULL, 3159b8b58e0SJonathan Lemon tp->rcv_nxt, tp->snd_una - 1, 0); 31608517d53SMike Silbersack (void) m_free(dtom(t_template)); 31708517d53SMike Silbersack } 318e2f2059fSMike Silbersack callout_reset(&tp->t_timers->tt_keep, tcp_keepintvl, tcp_timer_keep, tp); 3194cc20ab1SSeigo Tanimura } else 320e2f2059fSMike Silbersack callout_reset(&tp->t_timers->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 3219b8b58e0SJonathan Lemon 3229b8b58e0SJonathan Lemon #ifdef TCPDEBUG 3232a074620SSam Leffler if (inp->inp_socket->so_options & SO_DEBUG) 324fb59c426SYoshinobu Inoue tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 3259b8b58e0SJonathan Lemon PRU_SLOWTIMO); 3269b8b58e0SJonathan Lemon #endif 32785d94372SRobert Watson INP_UNLOCK(inp); 32885d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 32985d94372SRobert Watson return; 3309b8b58e0SJonathan Lemon 3319b8b58e0SJonathan Lemon dropit: 3329b8b58e0SJonathan Lemon tcpstat.tcps_keepdrops++; 33385d94372SRobert Watson tp = tcp_drop(tp, ETIMEDOUT); 33485d94372SRobert Watson 33585d94372SRobert Watson #ifdef TCPDEBUG 33685d94372SRobert Watson if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 33785d94372SRobert Watson tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 33885d94372SRobert Watson PRU_SLOWTIMO); 33985d94372SRobert Watson #endif 34085d94372SRobert Watson if (tp != NULL) 34185d94372SRobert Watson INP_UNLOCK(tp->t_inpcb); 34285d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 3439b8b58e0SJonathan Lemon } 3449b8b58e0SJonathan Lemon 34585d94372SRobert Watson void 34685d94372SRobert Watson tcp_timer_persist(void *xtp) 3479b8b58e0SJonathan Lemon { 34885d94372SRobert Watson struct tcpcb *tp = xtp; 34985d94372SRobert Watson struct inpcb *inp; 3509b8b58e0SJonathan Lemon #ifdef TCPDEBUG 3519b8b58e0SJonathan Lemon int ostate; 3529b8b58e0SJonathan Lemon 3539b8b58e0SJonathan Lemon ostate = tp->t_state; 3549b8b58e0SJonathan Lemon #endif 35585d94372SRobert Watson INP_INFO_WLOCK(&tcbinfo); 35685d94372SRobert Watson inp = tp->t_inpcb; 35785d94372SRobert Watson /* 35885d94372SRobert Watson * XXXRW: While this assert is in fact correct, bugs in the tcpcb 35985d94372SRobert Watson * tear-down mean we need it as a work-around for races between 36085d94372SRobert Watson * timers and tcp_discardcb(). 36185d94372SRobert Watson * 36285d94372SRobert Watson * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL")); 36385d94372SRobert Watson */ 36485d94372SRobert Watson if (inp == NULL) { 36585d94372SRobert Watson tcp_timer_race++; 36685d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 36785d94372SRobert Watson return; 36885d94372SRobert Watson } 36985d94372SRobert Watson INP_LOCK(inp); 370e2f2059fSMike Silbersack if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_persist) 371e2f2059fSMike Silbersack || !callout_active(&tp->t_timers->tt_persist)) { 37285d94372SRobert Watson INP_UNLOCK(inp); 37385d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 37485d94372SRobert Watson return; 37585d94372SRobert Watson } 376e2f2059fSMike Silbersack callout_deactivate(&tp->t_timers->tt_persist); 3779b8b58e0SJonathan Lemon /* 3789b8b58e0SJonathan Lemon * Persistance timer into zero window. 3799b8b58e0SJonathan Lemon * Force a byte to be output, if possible. 3809b8b58e0SJonathan Lemon */ 3819b8b58e0SJonathan Lemon tcpstat.tcps_persisttimeo++; 3829b8b58e0SJonathan Lemon /* 3839b8b58e0SJonathan Lemon * Hack: if the peer is dead/unreachable, we do not 3849b8b58e0SJonathan Lemon * time out if the window is closed. After a full 3859b8b58e0SJonathan Lemon * backoff, drop the connection if the idle time 3869b8b58e0SJonathan Lemon * (no responses to probes) reaches the maximum 3879b8b58e0SJonathan Lemon * backoff that we would use if retransmitting. 3889b8b58e0SJonathan Lemon */ 3899b8b58e0SJonathan Lemon if (tp->t_rxtshift == TCP_MAXRXTSHIFT && 3909b8b58e0SJonathan Lemon ((ticks - tp->t_rcvtime) >= tcp_maxpersistidle || 3919b8b58e0SJonathan Lemon (ticks - tp->t_rcvtime) >= TCP_REXMTVAL(tp) * tcp_totbackoff)) { 3929b8b58e0SJonathan Lemon tcpstat.tcps_persistdrop++; 39385d94372SRobert Watson tp = tcp_drop(tp, ETIMEDOUT); 39485d94372SRobert Watson goto out; 3959b8b58e0SJonathan Lemon } 3969b8b58e0SJonathan Lemon tcp_setpersist(tp); 3972cdbfa66SPaul Saab tp->t_flags |= TF_FORCEDATA; 3989b8b58e0SJonathan Lemon (void) tcp_output(tp); 3992cdbfa66SPaul Saab tp->t_flags &= ~TF_FORCEDATA; 4009b8b58e0SJonathan Lemon 40185d94372SRobert Watson out: 4029b8b58e0SJonathan Lemon #ifdef TCPDEBUG 403ffb761f6SGleb Smirnoff if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG) 404ffb761f6SGleb Smirnoff tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO); 4059b8b58e0SJonathan Lemon #endif 40685d94372SRobert Watson if (tp != NULL) 40785d94372SRobert Watson INP_UNLOCK(inp); 40885d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 4099b8b58e0SJonathan Lemon } 4109b8b58e0SJonathan Lemon 41185d94372SRobert Watson void 41285d94372SRobert Watson tcp_timer_rexmt(void * xtp) 4139b8b58e0SJonathan Lemon { 41485d94372SRobert Watson struct tcpcb *tp = xtp; 4159b8b58e0SJonathan Lemon int rexmt; 41685d94372SRobert Watson int headlocked; 41785d94372SRobert Watson struct inpcb *inp; 4189b8b58e0SJonathan Lemon #ifdef TCPDEBUG 4199b8b58e0SJonathan Lemon int ostate; 4209b8b58e0SJonathan Lemon 4219b8b58e0SJonathan Lemon ostate = tp->t_state; 4229b8b58e0SJonathan Lemon #endif 42385d94372SRobert Watson INP_INFO_WLOCK(&tcbinfo); 42485d94372SRobert Watson headlocked = 1; 42585d94372SRobert Watson inp = tp->t_inpcb; 42685d94372SRobert Watson /* 42785d94372SRobert Watson * XXXRW: While this assert is in fact correct, bugs in the tcpcb 42885d94372SRobert Watson * tear-down mean we need it as a work-around for races between 42985d94372SRobert Watson * timers and tcp_discardcb(). 43085d94372SRobert Watson * 43185d94372SRobert Watson * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL")); 43285d94372SRobert Watson */ 43385d94372SRobert Watson if (inp == NULL) { 43485d94372SRobert Watson tcp_timer_race++; 43585d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 43685d94372SRobert Watson return; 43785d94372SRobert Watson } 43885d94372SRobert Watson INP_LOCK(inp); 439e2f2059fSMike Silbersack if ((inp->inp_vflag & INP_DROPPED) || callout_pending(&tp->t_timers->tt_rexmt) 440e2f2059fSMike Silbersack || !callout_active(&tp->t_timers->tt_rexmt)) { 44185d94372SRobert Watson INP_UNLOCK(inp); 44285d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 44385d94372SRobert Watson return; 44485d94372SRobert Watson } 445e2f2059fSMike Silbersack callout_deactivate(&tp->t_timers->tt_rexmt); 4466d90faf3SPaul Saab tcp_free_sackholes(tp); 447df8bae1dSRodney W. Grimes /* 448df8bae1dSRodney W. Grimes * Retransmission timer went off. Message has not 449df8bae1dSRodney W. Grimes * been acked within retransmit interval. Back off 450df8bae1dSRodney W. Grimes * to a longer retransmit interval and retransmit one segment. 451df8bae1dSRodney W. Grimes */ 452df8bae1dSRodney W. Grimes if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) { 453df8bae1dSRodney W. Grimes tp->t_rxtshift = TCP_MAXRXTSHIFT; 454df8bae1dSRodney W. Grimes tcpstat.tcps_timeoutdrop++; 45585d94372SRobert Watson tp = tcp_drop(tp, tp->t_softerror ? 45685d94372SRobert Watson tp->t_softerror : ETIMEDOUT); 45785d94372SRobert Watson goto out; 4589b8b58e0SJonathan Lemon } 45985d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 46085d94372SRobert Watson headlocked = 0; 4619b8b58e0SJonathan Lemon if (tp->t_rxtshift == 1) { 4629b8b58e0SJonathan Lemon /* 4639b8b58e0SJonathan Lemon * first retransmit; record ssthresh and cwnd so they can 4649b8b58e0SJonathan Lemon * be recovered if this turns out to be a "bad" retransmit. 4659b8b58e0SJonathan Lemon * A retransmit is considered "bad" if an ACK for this 4669b8b58e0SJonathan Lemon * segment is received within RTT/2 interval; the assumption 4679b8b58e0SJonathan Lemon * here is that the ACK was already in flight. See 4689b8b58e0SJonathan Lemon * "On Estimating End-to-End Network Path Properties" by 4699b8b58e0SJonathan Lemon * Allman and Paxson for more details. 4709b8b58e0SJonathan Lemon */ 4719b8b58e0SJonathan Lemon tp->snd_cwnd_prev = tp->snd_cwnd; 4729b8b58e0SJonathan Lemon tp->snd_ssthresh_prev = tp->snd_ssthresh; 4739d11646dSJeffrey Hsu tp->snd_recover_prev = tp->snd_recover; 4749d11646dSJeffrey Hsu if (IN_FASTRECOVERY(tp)) 4759d11646dSJeffrey Hsu tp->t_flags |= TF_WASFRECOVERY; 4769d11646dSJeffrey Hsu else 4779d11646dSJeffrey Hsu tp->t_flags &= ~TF_WASFRECOVERY; 4789b8b58e0SJonathan Lemon tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1)); 479df8bae1dSRodney W. Grimes } 480df8bae1dSRodney W. Grimes tcpstat.tcps_rexmttimeo++; 4817d42e30cSJonathan Lemon if (tp->t_state == TCPS_SYN_SENT) 4827d42e30cSJonathan Lemon rexmt = TCP_REXMTVAL(tp) * tcp_syn_backoff[tp->t_rxtshift]; 4837d42e30cSJonathan Lemon else 484df8bae1dSRodney W. Grimes rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 485df8bae1dSRodney W. Grimes TCPT_RANGESET(tp->t_rxtcur, rexmt, 486df8bae1dSRodney W. Grimes tp->t_rttmin, TCPTV_REXMTMAX); 487df8bae1dSRodney W. Grimes /* 488c94c54e4SAndre Oppermann * Disable rfc1323 if we havn't got any response to 4897ceb7783SJesper Skriver * our third SYN to work-around some broken terminal servers 4907ceb7783SJesper Skriver * (most of which have hopefully been retired) that have bad VJ 4917ceb7783SJesper Skriver * header compression code which trashes TCP segments containing 4927ceb7783SJesper Skriver * unknown-to-them TCP options. 4937ceb7783SJesper Skriver */ 4947ceb7783SJesper Skriver if ((tp->t_state == TCPS_SYN_SENT) && (tp->t_rxtshift == 3)) 495c94c54e4SAndre Oppermann tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP); 4967ceb7783SJesper Skriver /* 49797d8d152SAndre Oppermann * If we backed off this far, our srtt estimate is probably bogus. 49897d8d152SAndre Oppermann * Clobber it so we'll take the next rtt measurement as our srtt; 499df8bae1dSRodney W. Grimes * move the current srtt into rttvar to keep the current 500df8bae1dSRodney W. Grimes * retransmit times until then. 501df8bae1dSRodney W. Grimes */ 502df8bae1dSRodney W. Grimes if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) { 503fb59c426SYoshinobu Inoue #ifdef INET6 504fb59c426SYoshinobu Inoue if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 505fb59c426SYoshinobu Inoue in6_losing(tp->t_inpcb); 506fb59c426SYoshinobu Inoue else 507fb59c426SYoshinobu Inoue #endif 508df8bae1dSRodney W. Grimes tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT); 509df8bae1dSRodney W. Grimes tp->t_srtt = 0; 510df8bae1dSRodney W. Grimes } 511df8bae1dSRodney W. Grimes tp->snd_nxt = tp->snd_una; 5129d11646dSJeffrey Hsu tp->snd_recover = tp->snd_max; 51346f58482SJonathan Lemon /* 51474b48c1dSAndras Olah * Force a segment to be sent. 51574b48c1dSAndras Olah */ 51674b48c1dSAndras Olah tp->t_flags |= TF_ACKNOW; 51774b48c1dSAndras Olah /* 518df8bae1dSRodney W. Grimes * If timing a segment in this window, stop the timer. 519df8bae1dSRodney W. Grimes */ 5209b8b58e0SJonathan Lemon tp->t_rtttime = 0; 521df8bae1dSRodney W. Grimes /* 522df8bae1dSRodney W. Grimes * Close the congestion window down to one segment 523df8bae1dSRodney W. Grimes * (we'll open it by one segment for each ack we get). 524df8bae1dSRodney W. Grimes * Since we probably have a window's worth of unacked 525df8bae1dSRodney W. Grimes * data accumulated, this "slow start" keeps us from 526df8bae1dSRodney W. Grimes * dumping all that data as back-to-back packets (which 527df8bae1dSRodney W. Grimes * might overwhelm an intermediate gateway). 528df8bae1dSRodney W. Grimes * 529df8bae1dSRodney W. Grimes * There are two phases to the opening: Initially we 530df8bae1dSRodney W. Grimes * open by one mss on each ack. This makes the window 531df8bae1dSRodney W. Grimes * size increase exponentially with time. If the 532df8bae1dSRodney W. Grimes * window is larger than the path can handle, this 533df8bae1dSRodney W. Grimes * exponential growth results in dropped packet(s) 534df8bae1dSRodney W. Grimes * almost immediately. To get more time between 535df8bae1dSRodney W. Grimes * drops but still "push" the network to take advantage 536df8bae1dSRodney W. Grimes * of improving conditions, we switch from exponential 537df8bae1dSRodney W. Grimes * to linear window opening at some threshhold size. 538df8bae1dSRodney W. Grimes * For a threshhold, we use half the current window 539df8bae1dSRodney W. Grimes * size, truncated to a multiple of the mss. 540df8bae1dSRodney W. Grimes * 541df8bae1dSRodney W. Grimes * (the minimum cwnd that will give us exponential 542df8bae1dSRodney W. Grimes * growth is 2 mss. We don't allow the threshhold 543df8bae1dSRodney W. Grimes * to go below this.) 544df8bae1dSRodney W. Grimes */ 545df8bae1dSRodney W. Grimes { 546df8bae1dSRodney W. Grimes u_int win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg; 547df8bae1dSRodney W. Grimes if (win < 2) 548df8bae1dSRodney W. Grimes win = 2; 549df8bae1dSRodney W. Grimes tp->snd_cwnd = tp->t_maxseg; 550df8bae1dSRodney W. Grimes tp->snd_ssthresh = win * tp->t_maxseg; 551df8bae1dSRodney W. Grimes tp->t_dupacks = 0; 552df8bae1dSRodney W. Grimes } 5539d11646dSJeffrey Hsu EXIT_FASTRECOVERY(tp); 554df8bae1dSRodney W. Grimes (void) tcp_output(tp); 555df8bae1dSRodney W. Grimes 55685d94372SRobert Watson out: 5579b8b58e0SJonathan Lemon #ifdef TCPDEBUG 5581c53f806SRobert Watson if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 559fb59c426SYoshinobu Inoue tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0, 5609b8b58e0SJonathan Lemon PRU_SLOWTIMO); 561df8bae1dSRodney W. Grimes #endif 56285d94372SRobert Watson if (tp != NULL) 56385d94372SRobert Watson INP_UNLOCK(inp); 56485d94372SRobert Watson if (headlocked) 56585d94372SRobert Watson INP_INFO_WUNLOCK(&tcbinfo); 56685d94372SRobert Watson } 56785d94372SRobert Watson 56885d94372SRobert Watson void 56985d94372SRobert Watson tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta) 57085d94372SRobert Watson { 57185d94372SRobert Watson struct callout *t_callout; 57285d94372SRobert Watson void *f_callout; 57385d94372SRobert Watson 57485d94372SRobert Watson switch (timer_type) { 57585d94372SRobert Watson case TT_DELACK: 576e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_delack; 57785d94372SRobert Watson f_callout = tcp_timer_delack; 57885d94372SRobert Watson break; 57985d94372SRobert Watson case TT_REXMT: 580e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_rexmt; 58185d94372SRobert Watson f_callout = tcp_timer_rexmt; 58285d94372SRobert Watson break; 58385d94372SRobert Watson case TT_PERSIST: 584e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_persist; 58585d94372SRobert Watson f_callout = tcp_timer_persist; 58685d94372SRobert Watson break; 58785d94372SRobert Watson case TT_KEEP: 588e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_keep; 58985d94372SRobert Watson f_callout = tcp_timer_keep; 59085d94372SRobert Watson break; 59185d94372SRobert Watson case TT_2MSL: 592e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_2msl; 59385d94372SRobert Watson f_callout = tcp_timer_2msl; 59485d94372SRobert Watson break; 59585d94372SRobert Watson default: 59685d94372SRobert Watson panic("bad timer_type"); 59785d94372SRobert Watson } 59885d94372SRobert Watson if (delta == 0) { 59985d94372SRobert Watson callout_stop(t_callout); 60085d94372SRobert Watson } else { 60185d94372SRobert Watson callout_reset(t_callout, delta, f_callout, tp); 60285d94372SRobert Watson } 60385d94372SRobert Watson } 60485d94372SRobert Watson 60585d94372SRobert Watson int 60685d94372SRobert Watson tcp_timer_active(struct tcpcb *tp, int timer_type) 60785d94372SRobert Watson { 60885d94372SRobert Watson struct callout *t_callout; 60985d94372SRobert Watson 61085d94372SRobert Watson switch (timer_type) { 61185d94372SRobert Watson case TT_DELACK: 612e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_delack; 61385d94372SRobert Watson break; 61485d94372SRobert Watson case TT_REXMT: 615e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_rexmt; 61685d94372SRobert Watson break; 61785d94372SRobert Watson case TT_PERSIST: 618e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_persist; 61985d94372SRobert Watson break; 62085d94372SRobert Watson case TT_KEEP: 621e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_keep; 62285d94372SRobert Watson break; 62385d94372SRobert Watson case TT_2MSL: 624e2f2059fSMike Silbersack t_callout = &tp->t_timers->tt_2msl; 62585d94372SRobert Watson break; 62685d94372SRobert Watson default: 62785d94372SRobert Watson panic("bad timer_type"); 62885d94372SRobert Watson } 62985d94372SRobert Watson return callout_active(t_callout); 630df8bae1dSRodney W. Grimes } 631