The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_timer.c

Version: -  FREEBSD  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    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.
   16  *
   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
   27  * SUCH DAMAGE.
   28  *
   29  *      @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
   30  */
   31 
   32 #include <sys/cdefs.h>
   33 __FBSDID("$FreeBSD: head/sys/netinet/tcp_timer.c 286227 2015-08-03 12:13:54Z jch $");
   34 
   35 #include "opt_inet.h"
   36 #include "opt_inet6.h"
   37 #include "opt_tcpdebug.h"
   38 #include "opt_rss.h"
   39 
   40 #include <sys/param.h>
   41 #include <sys/kernel.h>
   42 #include <sys/lock.h>
   43 #include <sys/mbuf.h>
   44 #include <sys/mutex.h>
   45 #include <sys/protosw.h>
   46 #include <sys/smp.h>
   47 #include <sys/socket.h>
   48 #include <sys/socketvar.h>
   49 #include <sys/sysctl.h>
   50 #include <sys/systm.h>
   51 
   52 #include <net/if.h>
   53 #include <net/route.h>
   54 #include <net/rss_config.h>
   55 #include <net/vnet.h>
   56 #include <net/netisr.h>
   57 
   58 #include <netinet/cc.h>
   59 #include <netinet/in.h>
   60 #include <netinet/in_pcb.h>
   61 #include <netinet/in_rss.h>
   62 #include <netinet/in_systm.h>
   63 #ifdef INET6
   64 #include <netinet6/in6_pcb.h>
   65 #endif
   66 #include <netinet/ip_var.h>
   67 #include <netinet/tcp_fsm.h>
   68 #include <netinet/tcp_timer.h>
   69 #include <netinet/tcp_var.h>
   70 #ifdef INET6
   71 #include <netinet6/tcp6_var.h>
   72 #endif
   73 #include <netinet/tcpip.h>
   74 #ifdef TCPDEBUG
   75 #include <netinet/tcp_debug.h>
   76 #endif
   77 
   78 int     tcp_keepinit;
   79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
   80     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
   81 
   82 int     tcp_keepidle;
   83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
   84     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
   85 
   86 int     tcp_keepintvl;
   87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
   88     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
   89 
   90 int     tcp_delacktime;
   91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
   92     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
   93     "Time before a delayed ACK is sent");
   94 
   95 int     tcp_msl;
   96 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
   97     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
   98 
   99 int     tcp_rexmit_min;
  100 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
  101     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
  102     "Minimum Retransmission Timeout");
  103 
  104 int     tcp_rexmit_slop;
  105 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
  106     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
  107     "Retransmission Timer Slop");
  108 
  109 static int      always_keepalive = 1;
  110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
  111     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
  112 
  113 int    tcp_fast_finwait2_recycle = 0;
  114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW, 
  115     &tcp_fast_finwait2_recycle, 0,
  116     "Recycle closed FIN_WAIT_2 connections faster");
  117 
  118 int    tcp_finwait2_timeout;
  119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
  120     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
  121 
  122 int     tcp_keepcnt = TCPTV_KEEPCNT;
  123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
  124     "Number of keepalive probes to send");
  125 
  126         /* max idle probes */
  127 int     tcp_maxpersistidle;
  128 
  129 static int      tcp_rexmit_drop_options = 0;
  130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
  131     &tcp_rexmit_drop_options, 0,
  132     "Drop TCP options from 3rd and later retransmitted SYN");
  133 
  134 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
  135 #define V_tcp_pmtud_blackhole_detect    VNET(tcp_pmtud_blackhole_detect)
  136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
  137     CTLFLAG_RW|CTLFLAG_VNET,
  138     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
  139     "Path MTU Discovery Black Hole Detection Enabled");
  140 
  141 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
  142 #define V_tcp_pmtud_blackhole_activated \
  143     VNET(tcp_pmtud_blackhole_activated)
  144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
  145     CTLFLAG_RD|CTLFLAG_VNET,
  146     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
  147     "Path MTU Discovery Black Hole Detection, Activation Count");
  148 
  149 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
  150 #define V_tcp_pmtud_blackhole_activated_min_mss \
  151     VNET(tcp_pmtud_blackhole_activated_min_mss)
  152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
  153     CTLFLAG_RD|CTLFLAG_VNET,
  154     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
  155     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
  156 
  157 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
  158 #define V_tcp_pmtud_blackhole_failed    VNET(tcp_pmtud_blackhole_failed)
  159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
  160     CTLFLAG_RD|CTLFLAG_VNET,
  161     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
  162     "Path MTU Discovery Black Hole Detection, Failure Count");
  163 
  164 #ifdef INET
  165 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
  166 #define V_tcp_pmtud_blackhole_mss       VNET(tcp_pmtud_blackhole_mss)
  167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
  168     CTLFLAG_RW|CTLFLAG_VNET,
  169     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
  170     "Path MTU Discovery Black Hole Detection lowered MSS");
  171 #endif
  172 
  173 #ifdef INET6
  174 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
  175 #define V_tcp_v6pmtud_blackhole_mss     VNET(tcp_v6pmtud_blackhole_mss)
  176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
  177     CTLFLAG_RW|CTLFLAG_VNET,
  178     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
  179     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
  180 #endif
  181 
  182 #ifdef  RSS
  183 static int      per_cpu_timers = 1;
  184 #else
  185 static int      per_cpu_timers = 0;
  186 #endif
  187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
  188     &per_cpu_timers , 0, "run tcp timers on all cpus");
  189 
  190 #if 0
  191 #define INP_CPU(inp)    (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
  192                 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
  193 #endif
  194 
  195 /*
  196  * Map the given inp to a CPU id.
  197  *
  198  * This queries RSS if it's compiled in, else it defaults to the current
  199  * CPU ID.
  200  */
  201 static inline int
  202 inp_to_cpuid(struct inpcb *inp)
  203 {
  204         u_int cpuid;
  205 
  206 #ifdef  RSS
  207         if (per_cpu_timers) {
  208                 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
  209                 if (cpuid == NETISR_CPUID_NONE)
  210                         return (curcpu);        /* XXX */
  211                 else
  212                         return (cpuid);
  213         }
  214 #else
  215         /* Legacy, pre-RSS behaviour */
  216         if (per_cpu_timers) {
  217                 /*
  218                  * We don't have a flowid -> cpuid mapping, so cheat and
  219                  * just map unknown cpuids to curcpu.  Not the best, but
  220                  * apparently better than defaulting to swi 0.
  221                  */
  222                 cpuid = inp->inp_flowid % (mp_maxid + 1);
  223                 if (! CPU_ABSENT(cpuid))
  224                         return (cpuid);
  225                 return (curcpu);
  226         }
  227 #endif
  228         /* Default for RSS and non-RSS - cpuid 0 */
  229         else {
  230                 return (0);
  231         }
  232 }
  233 
  234 /*
  235  * Tcp protocol timeout routine called every 500 ms.
  236  * Updates timestamps used for TCP
  237  * causes finite state machine actions if timers expire.
  238  */
  239 void
  240 tcp_slowtimo(void)
  241 {
  242         VNET_ITERATOR_DECL(vnet_iter);
  243 
  244         VNET_LIST_RLOCK_NOSLEEP();
  245         VNET_FOREACH(vnet_iter) {
  246                 CURVNET_SET(vnet_iter);
  247                 (void) tcp_tw_2msl_scan(0);
  248                 CURVNET_RESTORE();
  249         }
  250         VNET_LIST_RUNLOCK_NOSLEEP();
  251 }
  252 
  253 int     tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
  254     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
  255 
  256 int     tcp_backoff[TCP_MAXRXTSHIFT + 1] =
  257     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
  258 
  259 static int tcp_totbackoff = 2559;       /* sum of tcp_backoff[] */
  260 
  261 /*
  262  * TCP timer processing.
  263  */
  264 
  265 void
  266 tcp_timer_delack(void *xtp)
  267 {
  268         struct tcpcb *tp = xtp;
  269         struct inpcb *inp;
  270         CURVNET_SET(tp->t_vnet);
  271 
  272         inp = tp->t_inpcb;
  273         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
  274         INP_WLOCK(inp);
  275         if (callout_pending(&tp->t_timers->tt_delack) ||
  276             !callout_active(&tp->t_timers->tt_delack)) {
  277                 INP_WUNLOCK(inp);
  278                 CURVNET_RESTORE();
  279                 return;
  280         }
  281         callout_deactivate(&tp->t_timers->tt_delack);
  282         if ((inp->inp_flags & INP_DROPPED) != 0) {
  283                 INP_WUNLOCK(inp);
  284                 CURVNET_RESTORE();
  285                 return;
  286         }
  287         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
  288                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
  289         KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
  290                 ("%s: tp %p delack callout should be running", __func__, tp));
  291 
  292         tp->t_flags |= TF_ACKNOW;
  293         TCPSTAT_INC(tcps_delack);
  294         (void) tcp_output(tp);
  295         INP_WUNLOCK(inp);
  296         CURVNET_RESTORE();
  297 }
  298 
  299 void
  300 tcp_timer_2msl(void *xtp)
  301 {
  302         struct tcpcb *tp = xtp;
  303         struct inpcb *inp;
  304         CURVNET_SET(tp->t_vnet);
  305 #ifdef TCPDEBUG
  306         int ostate;
  307 
  308         ostate = tp->t_state;
  309 #endif
  310         INP_INFO_RLOCK(&V_tcbinfo);
  311         inp = tp->t_inpcb;
  312         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
  313         INP_WLOCK(inp);
  314         tcp_free_sackholes(tp);
  315         if (callout_pending(&tp->t_timers->tt_2msl) ||
  316             !callout_active(&tp->t_timers->tt_2msl)) {
  317                 INP_WUNLOCK(tp->t_inpcb);
  318                 INP_INFO_RUNLOCK(&V_tcbinfo);
  319                 CURVNET_RESTORE();
  320                 return;
  321         }
  322         callout_deactivate(&tp->t_timers->tt_2msl);
  323         if ((inp->inp_flags & INP_DROPPED) != 0) {
  324                 INP_WUNLOCK(inp);
  325                 INP_INFO_RUNLOCK(&V_tcbinfo);
  326                 CURVNET_RESTORE();
  327                 return;
  328         }
  329         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
  330                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
  331         KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
  332                 ("%s: tp %p 2msl callout should be running", __func__, tp));
  333         /*
  334          * 2 MSL timeout in shutdown went off.  If we're closed but
  335          * still waiting for peer to close and connection has been idle
  336          * too long, or if 2MSL time is up from TIME_WAIT, delete connection
  337          * control block.  Otherwise, check again in a bit.
  338          *
  339          * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed, 
  340          * there's no point in hanging onto FIN_WAIT_2 socket. Just close it. 
  341          * Ignore fact that there were recent incoming segments.
  342          */
  343         if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
  344             tp->t_inpcb && tp->t_inpcb->inp_socket && 
  345             (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
  346                 TCPSTAT_INC(tcps_finwait2_drops);
  347                 tp = tcp_close(tp);             
  348         } else {
  349                 if (tp->t_state != TCPS_TIME_WAIT &&
  350                    ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
  351                         if (!callout_reset(&tp->t_timers->tt_2msl,
  352                            TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
  353                                 tp->t_timers->tt_flags &= ~TT_2MSL_RST;
  354                         }
  355                 } else
  356                        tp = tcp_close(tp);
  357        }
  358 
  359 #ifdef TCPDEBUG
  360         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
  361                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
  362                           PRU_SLOWTIMO);
  363 #endif
  364         if (tp != NULL)
  365                 INP_WUNLOCK(inp);
  366         INP_INFO_RUNLOCK(&V_tcbinfo);
  367         CURVNET_RESTORE();
  368 }
  369 
  370 void
  371 tcp_timer_keep(void *xtp)
  372 {
  373         struct tcpcb *tp = xtp;
  374         struct tcptemp *t_template;
  375         struct inpcb *inp;
  376         CURVNET_SET(tp->t_vnet);
  377 #ifdef TCPDEBUG
  378         int ostate;
  379 
  380         ostate = tp->t_state;
  381 #endif
  382         INP_INFO_RLOCK(&V_tcbinfo);
  383         inp = tp->t_inpcb;
  384         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
  385         INP_WLOCK(inp);
  386         if (callout_pending(&tp->t_timers->tt_keep) ||
  387             !callout_active(&tp->t_timers->tt_keep)) {
  388                 INP_WUNLOCK(inp);
  389                 INP_INFO_RUNLOCK(&V_tcbinfo);
  390                 CURVNET_RESTORE();
  391                 return;
  392         }
  393         callout_deactivate(&tp->t_timers->tt_keep);
  394         if ((inp->inp_flags & INP_DROPPED) != 0) {
  395                 INP_WUNLOCK(inp);
  396                 INP_INFO_RUNLOCK(&V_tcbinfo);
  397                 CURVNET_RESTORE();
  398                 return;
  399         }
  400         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
  401                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
  402         KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
  403                 ("%s: tp %p keep callout should be running", __func__, tp));
  404         /*
  405          * Keep-alive timer went off; send something
  406          * or drop connection if idle for too long.
  407          */
  408         TCPSTAT_INC(tcps_keeptimeo);
  409         if (tp->t_state < TCPS_ESTABLISHED)
  410                 goto dropit;
  411         if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
  412             tp->t_state <= TCPS_CLOSING) {
  413                 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
  414                         goto dropit;
  415                 /*
  416                  * Send a packet designed to force a response
  417                  * if the peer is up and reachable:
  418                  * either an ACK if the connection is still alive,
  419                  * or an RST if the peer has closed the connection
  420                  * due to timeout or reboot.
  421                  * Using sequence number tp->snd_una-1
  422                  * causes the transmitted zero-length segment
  423                  * to lie outside the receive window;
  424                  * by the protocol spec, this requires the
  425                  * correspondent TCP to respond.
  426                  */
  427                 TCPSTAT_INC(tcps_keepprobe);
  428                 t_template = tcpip_maketemplate(inp);
  429                 if (t_template) {
  430                         tcp_respond(tp, t_template->tt_ipgen,
  431                                     &t_template->tt_t, (struct mbuf *)NULL,
  432                                     tp->rcv_nxt, tp->snd_una - 1, 0);
  433                         free(t_template, M_TEMP);
  434                 }
  435                 if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
  436                     tcp_timer_keep, tp)) {
  437                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
  438                 }
  439         } else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
  440                     tcp_timer_keep, tp)) {
  441                         tp->t_timers->tt_flags &= ~TT_KEEP_RST;
  442                 }
  443 
  444 #ifdef TCPDEBUG
  445         if (inp->inp_socket->so_options & SO_DEBUG)
  446                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
  447                           PRU_SLOWTIMO);
  448 #endif
  449         INP_WUNLOCK(inp);
  450         INP_INFO_RUNLOCK(&V_tcbinfo);
  451         CURVNET_RESTORE();
  452         return;
  453 
  454 dropit:
  455         TCPSTAT_INC(tcps_keepdrops);
  456         tp = tcp_drop(tp, ETIMEDOUT);
  457 
  458 #ifdef TCPDEBUG
  459         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
  460                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
  461                           PRU_SLOWTIMO);
  462 #endif
  463         if (tp != NULL)
  464                 INP_WUNLOCK(tp->t_inpcb);
  465         INP_INFO_RUNLOCK(&V_tcbinfo);
  466         CURVNET_RESTORE();
  467 }
  468 
  469 void
  470 tcp_timer_persist(void *xtp)
  471 {
  472         struct tcpcb *tp = xtp;
  473         struct inpcb *inp;
  474         CURVNET_SET(tp->t_vnet);
  475 #ifdef TCPDEBUG
  476         int ostate;
  477 
  478         ostate = tp->t_state;
  479 #endif
  480         INP_INFO_RLOCK(&V_tcbinfo);
  481         inp = tp->t_inpcb;
  482         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
  483         INP_WLOCK(inp);
  484         if (callout_pending(&tp->t_timers->tt_persist) ||
  485             !callout_active(&tp->t_timers->tt_persist)) {
  486                 INP_WUNLOCK(inp);
  487                 INP_INFO_RUNLOCK(&V_tcbinfo);
  488                 CURVNET_RESTORE();
  489                 return;
  490         }
  491         callout_deactivate(&tp->t_timers->tt_persist);
  492         if ((inp->inp_flags & INP_DROPPED) != 0) {
  493                 INP_WUNLOCK(inp);
  494                 INP_INFO_RUNLOCK(&V_tcbinfo);
  495                 CURVNET_RESTORE();
  496                 return;
  497         }
  498         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
  499                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
  500         KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
  501                 ("%s: tp %p persist callout should be running", __func__, tp));
  502         /*
  503          * Persistance timer into zero window.
  504          * Force a byte to be output, if possible.
  505          */
  506         TCPSTAT_INC(tcps_persisttimeo);
  507         /*
  508          * Hack: if the peer is dead/unreachable, we do not
  509          * time out if the window is closed.  After a full
  510          * backoff, drop the connection if the idle time
  511          * (no responses to probes) reaches the maximum
  512          * backoff that we would use if retransmitting.
  513          */
  514         if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
  515             (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
  516              ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
  517                 TCPSTAT_INC(tcps_persistdrop);
  518                 tp = tcp_drop(tp, ETIMEDOUT);
  519                 goto out;
  520         }
  521         /*
  522          * If the user has closed the socket then drop a persisting
  523          * connection after a much reduced timeout.
  524          */
  525         if (tp->t_state > TCPS_CLOSE_WAIT &&
  526             (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
  527                 TCPSTAT_INC(tcps_persistdrop);
  528                 tp = tcp_drop(tp, ETIMEDOUT);
  529                 goto out;
  530         }
  531         tcp_setpersist(tp);
  532         tp->t_flags |= TF_FORCEDATA;
  533         (void) tcp_output(tp);
  534         tp->t_flags &= ~TF_FORCEDATA;
  535 
  536 out:
  537 #ifdef TCPDEBUG
  538         if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
  539                 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
  540 #endif
  541         if (tp != NULL)
  542                 INP_WUNLOCK(inp);
  543         INP_INFO_RUNLOCK(&V_tcbinfo);
  544         CURVNET_RESTORE();
  545 }
  546 
  547 void
  548 tcp_timer_rexmt(void * xtp)
  549 {
  550         struct tcpcb *tp = xtp;
  551         CURVNET_SET(tp->t_vnet);
  552         int rexmt;
  553         int headlocked;
  554         struct inpcb *inp;
  555 #ifdef TCPDEBUG
  556         int ostate;
  557 
  558         ostate = tp->t_state;
  559 #endif
  560 
  561         INP_INFO_RLOCK(&V_tcbinfo);
  562         inp = tp->t_inpcb;
  563         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
  564         INP_WLOCK(inp);
  565         if (callout_pending(&tp->t_timers->tt_rexmt) ||
  566             !callout_active(&tp->t_timers->tt_rexmt)) {
  567                 INP_WUNLOCK(inp);
  568                 INP_INFO_RUNLOCK(&V_tcbinfo);
  569                 CURVNET_RESTORE();
  570                 return;
  571         }
  572         callout_deactivate(&tp->t_timers->tt_rexmt);
  573         if ((inp->inp_flags & INP_DROPPED) != 0) {
  574                 INP_WUNLOCK(inp);
  575                 INP_INFO_RUNLOCK(&V_tcbinfo);
  576                 CURVNET_RESTORE();
  577                 return;
  578         }
  579         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
  580                 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
  581         KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
  582                 ("%s: tp %p rexmt callout should be running", __func__, tp));
  583         tcp_free_sackholes(tp);
  584         /*
  585          * Retransmission timer went off.  Message has not
  586          * been acked within retransmit interval.  Back off
  587          * to a longer retransmit interval and retransmit one segment.
  588          */
  589         if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
  590                 tp->t_rxtshift = TCP_MAXRXTSHIFT;
  591                 TCPSTAT_INC(tcps_timeoutdrop);
  592 
  593                 tp = tcp_drop(tp, tp->t_softerror ?
  594                               tp->t_softerror : ETIMEDOUT);
  595                 headlocked = 1;
  596                 goto out;
  597         }
  598         INP_INFO_RUNLOCK(&V_tcbinfo);
  599         headlocked = 0;
  600         if (tp->t_state == TCPS_SYN_SENT) {
  601                 /*
  602                  * If the SYN was retransmitted, indicate CWND to be
  603                  * limited to 1 segment in cc_conn_init().
  604                  */
  605                 tp->snd_cwnd = 1;
  606         } else if (tp->t_rxtshift == 1) {
  607                 /*
  608                  * first retransmit; record ssthresh and cwnd so they can
  609                  * be recovered if this turns out to be a "bad" retransmit.
  610                  * A retransmit is considered "bad" if an ACK for this
  611                  * segment is received within RTT/2 interval; the assumption
  612                  * here is that the ACK was already in flight.  See
  613                  * "On Estimating End-to-End Network Path Properties" by
  614                  * Allman and Paxson for more details.
  615                  */
  616                 tp->snd_cwnd_prev = tp->snd_cwnd;
  617                 tp->snd_ssthresh_prev = tp->snd_ssthresh;
  618                 tp->snd_recover_prev = tp->snd_recover;
  619                 if (IN_FASTRECOVERY(tp->t_flags))
  620                         tp->t_flags |= TF_WASFRECOVERY;
  621                 else
  622                         tp->t_flags &= ~TF_WASFRECOVERY;
  623                 if (IN_CONGRECOVERY(tp->t_flags))
  624                         tp->t_flags |= TF_WASCRECOVERY;
  625                 else
  626                         tp->t_flags &= ~TF_WASCRECOVERY;
  627                 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
  628                 tp->t_flags |= TF_PREVVALID;
  629         } else
  630                 tp->t_flags &= ~TF_PREVVALID;
  631         TCPSTAT_INC(tcps_rexmttimeo);
  632         if (tp->t_state == TCPS_SYN_SENT)
  633                 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
  634         else
  635                 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
  636         TCPT_RANGESET(tp->t_rxtcur, rexmt,
  637                       tp->t_rttmin, TCPTV_REXMTMAX);
  638 
  639         /*
  640          * We enter the path for PLMTUD if connection is established or, if
  641          * connection is FIN_WAIT_1 status, reason for the last is that if
  642          * amount of data we send is very small, we could send it in couple of
  643          * packets and process straight to FIN. In that case we won't catch
  644          * ESTABLISHED state.
  645          */
  646         if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
  647             || (tp->t_state == TCPS_FIN_WAIT_1))) {
  648                 int optlen;
  649 #ifdef INET6
  650                 int isipv6;
  651 #endif
  652 
  653                 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
  654                     (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
  655                     (tp->t_rxtshift <= 2)) {
  656                         /*
  657                          * Enter Path MTU Black-hole Detection mechanism:
  658                          * - Disable Path MTU Discovery (IP "DF" bit).
  659                          * - Reduce MTU to lower value than what we
  660                          *   negotiated with peer.
  661                          */
  662                         /* Record that we may have found a black hole. */
  663                         tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
  664 
  665                         /* Keep track of previous MSS. */
  666                         optlen = tp->t_maxopd - tp->t_maxseg;
  667                         tp->t_pmtud_saved_maxopd = tp->t_maxopd;
  668 
  669                         /* 
  670                          * Reduce the MSS to blackhole value or to the default
  671                          * in an attempt to retransmit.
  672                          */
  673 #ifdef INET6
  674                         isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
  675                         if (isipv6 &&
  676                             tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
  677                                 /* Use the sysctl tuneable blackhole MSS. */
  678                                 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
  679                                 V_tcp_pmtud_blackhole_activated++;
  680                         } else if (isipv6) {
  681                                 /* Use the default MSS. */
  682                                 tp->t_maxopd = V_tcp_v6mssdflt;
  683                                 /*
  684                                  * Disable Path MTU Discovery when we switch to
  685                                  * minmss.
  686                                  */
  687                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
  688                                 V_tcp_pmtud_blackhole_activated_min_mss++;
  689                         }
  690 #endif
  691 #if defined(INET6) && defined(INET)
  692                         else
  693 #endif
  694 #ifdef INET
  695                         if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
  696                                 /* Use the sysctl tuneable blackhole MSS. */
  697                                 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
  698                                 V_tcp_pmtud_blackhole_activated++;
  699                         } else {
  700                                 /* Use the default MSS. */
  701                                 tp->t_maxopd = V_tcp_mssdflt;
  702                                 /*
  703                                  * Disable Path MTU Discovery when we switch to
  704                                  * minmss.
  705                                  */
  706                                 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
  707                                 V_tcp_pmtud_blackhole_activated_min_mss++;
  708                         }
  709 #endif
  710                         tp->t_maxseg = tp->t_maxopd - optlen;
  711                         /*
  712                          * Reset the slow-start flight size
  713                          * as it may depend on the new MSS.
  714                          */
  715                         if (CC_ALGO(tp)->conn_init != NULL)
  716                                 CC_ALGO(tp)->conn_init(tp->ccv);
  717                 } else {
  718                         /*
  719                          * If further retransmissions are still unsuccessful
  720                          * with a lowered MTU, maybe this isn't a blackhole and
  721                          * we restore the previous MSS and blackhole detection
  722                          * flags.
  723                          */
  724                         if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
  725                             (tp->t_rxtshift > 4)) {
  726                                 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
  727                                 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
  728                                 optlen = tp->t_maxopd - tp->t_maxseg;
  729                                 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
  730                                 tp->t_maxseg = tp->t_maxopd - optlen;
  731                                 V_tcp_pmtud_blackhole_failed++;
  732                                 /*
  733                                  * Reset the slow-start flight size as it
  734                                  * may depend on the new MSS.
  735                                  */
  736                                 if (CC_ALGO(tp)->conn_init != NULL)
  737                                         CC_ALGO(tp)->conn_init(tp->ccv);
  738                         }
  739                 }
  740         }
  741 
  742         /*
  743          * Disable RFC1323 and SACK if we haven't got any response to
  744          * our third SYN to work-around some broken terminal servers
  745          * (most of which have hopefully been retired) that have bad VJ
  746          * header compression code which trashes TCP segments containing
  747          * unknown-to-them TCP options.
  748          */
  749         if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
  750             (tp->t_rxtshift == 3))
  751                 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
  752         /*
  753          * If we backed off this far, our srtt estimate is probably bogus.
  754          * Clobber it so we'll take the next rtt measurement as our srtt;
  755          * move the current srtt into rttvar to keep the current
  756          * retransmit times until then.
  757          */
  758         if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
  759 #ifdef INET6
  760                 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
  761                         in6_losing(tp->t_inpcb);
  762 #endif
  763                 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
  764                 tp->t_srtt = 0;
  765         }
  766         tp->snd_nxt = tp->snd_una;
  767         tp->snd_recover = tp->snd_max;
  768         /*
  769          * Force a segment to be sent.
  770          */
  771         tp->t_flags |= TF_ACKNOW;
  772         /*
  773          * If timing a segment in this window, stop the timer.
  774          */
  775         tp->t_rtttime = 0;
  776 
  777         cc_cong_signal(tp, NULL, CC_RTO);
  778 
  779         (void) tcp_output(tp);
  780 
  781 out:
  782 #ifdef TCPDEBUG
  783         if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
  784                 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
  785                           PRU_SLOWTIMO);
  786 #endif
  787         if (tp != NULL)
  788                 INP_WUNLOCK(inp);
  789         if (headlocked)
  790                 INP_INFO_RUNLOCK(&V_tcbinfo);
  791         CURVNET_RESTORE();
  792 }
  793 
  794 void
  795 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
  796 {
  797         struct callout *t_callout;
  798         timeout_t *f_callout;
  799         struct inpcb *inp = tp->t_inpcb;
  800         int cpu = inp_to_cpuid(inp);
  801         uint32_t f_reset;
  802 
  803 #ifdef TCP_OFFLOAD
  804         if (tp->t_flags & TF_TOE)
  805                 return;
  806 #endif
  807 
  808         if (tp->t_timers->tt_flags & TT_STOPPED)
  809                 return;
  810 
  811         switch (timer_type) {
  812                 case TT_DELACK:
  813                         t_callout = &tp->t_timers->tt_delack;
  814                         f_callout = tcp_timer_delack;
  815                         f_reset = TT_DELACK_RST;
  816                         break;
  817                 case TT_REXMT:
  818                         t_callout = &tp->t_timers->tt_rexmt;
  819                         f_callout = tcp_timer_rexmt;
  820                         f_reset = TT_REXMT_RST;
  821                         break;
  822                 case TT_PERSIST:
  823                         t_callout = &tp->t_timers->tt_persist;
  824                         f_callout = tcp_timer_persist;
  825                         f_reset = TT_PERSIST_RST;
  826                         break;
  827                 case TT_KEEP:
  828                         t_callout = &tp->t_timers->tt_keep;
  829                         f_callout = tcp_timer_keep;
  830                         f_reset = TT_KEEP_RST;
  831                         break;
  832                 case TT_2MSL:
  833                         t_callout = &tp->t_timers->tt_2msl;
  834                         f_callout = tcp_timer_2msl;
  835                         f_reset = TT_2MSL_RST;
  836                         break;
  837                 default:
  838                         panic("tp %p bad timer_type %#x", tp, timer_type);
  839                 }
  840         if (delta == 0) {
  841                 if ((tp->t_timers->tt_flags & timer_type) &&
  842                     callout_stop(t_callout) &&
  843                     (tp->t_timers->tt_flags & f_reset)) {
  844                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
  845                 }
  846         } else {
  847                 if ((tp->t_timers->tt_flags & timer_type) == 0) {
  848                         tp->t_timers->tt_flags |= (timer_type | f_reset);
  849                         callout_reset_on(t_callout, delta, f_callout, tp, cpu);
  850                 } else {
  851                         /* Reset already running callout on the same CPU. */
  852                         if (!callout_reset(t_callout, delta, f_callout, tp)) {
  853                                 /*
  854                                  * Callout not cancelled, consider it as not
  855                                  * properly restarted. */
  856                                 tp->t_timers->tt_flags &= ~f_reset;
  857                         }
  858                 }
  859         }
  860 }
  861 
  862 int
  863 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
  864 {
  865         struct callout *t_callout;
  866 
  867         switch (timer_type) {
  868                 case TT_DELACK:
  869                         t_callout = &tp->t_timers->tt_delack;
  870                         break;
  871                 case TT_REXMT:
  872                         t_callout = &tp->t_timers->tt_rexmt;
  873                         break;
  874                 case TT_PERSIST:
  875                         t_callout = &tp->t_timers->tt_persist;
  876                         break;
  877                 case TT_KEEP:
  878                         t_callout = &tp->t_timers->tt_keep;
  879                         break;
  880                 case TT_2MSL:
  881                         t_callout = &tp->t_timers->tt_2msl;
  882                         break;
  883                 default:
  884                         panic("tp %p bad timer_type %#x", tp, timer_type);
  885                 }
  886         return callout_active(t_callout);
  887 }
  888 
  889 void
  890 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
  891 {
  892         struct callout *t_callout;
  893         timeout_t *f_callout;
  894         uint32_t f_reset;
  895 
  896         tp->t_timers->tt_flags |= TT_STOPPED;
  897 
  898         switch (timer_type) {
  899                 case TT_DELACK:
  900                         t_callout = &tp->t_timers->tt_delack;
  901                         f_callout = tcp_timer_delack_discard;
  902                         f_reset = TT_DELACK_RST;
  903                         break;
  904                 case TT_REXMT:
  905                         t_callout = &tp->t_timers->tt_rexmt;
  906                         f_callout = tcp_timer_rexmt_discard;
  907                         f_reset = TT_REXMT_RST;
  908                         break;
  909                 case TT_PERSIST:
  910                         t_callout = &tp->t_timers->tt_persist;
  911                         f_callout = tcp_timer_persist_discard;
  912                         f_reset = TT_PERSIST_RST;
  913                         break;
  914                 case TT_KEEP:
  915                         t_callout = &tp->t_timers->tt_keep;
  916                         f_callout = tcp_timer_keep_discard;
  917                         f_reset = TT_KEEP_RST;
  918                         break;
  919                 case TT_2MSL:
  920                         t_callout = &tp->t_timers->tt_2msl;
  921                         f_callout = tcp_timer_2msl_discard;
  922                         f_reset = TT_2MSL_RST;
  923                         break;
  924                 default:
  925                         panic("tp %p bad timer_type %#x", tp, timer_type);
  926                 }
  927 
  928         if (tp->t_timers->tt_flags & timer_type) {
  929                 if (callout_stop(t_callout) &&
  930                     (tp->t_timers->tt_flags & f_reset)) {
  931                         tp->t_timers->tt_flags &= ~(timer_type | f_reset);
  932                 } else {
  933                         /*
  934                          * Can't stop the callout, defer tcpcb actual deletion
  935                          * to the last tcp timer discard callout.
  936                          * The TT_STOPPED flag will ensure that no tcp timer
  937                          * callouts can be restarted on our behalf, and
  938                          * past this point currently running callouts waiting
  939                          * on inp lock will return right away after the
  940                          * classical check for callout reset/stop events:
  941                          * callout_pending() || !callout_active()
  942                          */
  943                         callout_reset(t_callout, 1, f_callout, tp);
  944                 }
  945         }
  946 }
  947 
  948 #define ticks_to_msecs(t)       (1000*(t) / hz)
  949 
  950 void
  951 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
  952     struct xtcp_timer *xtimer)
  953 {
  954         sbintime_t now;
  955 
  956         bzero(xtimer, sizeof(*xtimer));
  957         if (timer == NULL)
  958                 return;
  959         now = getsbinuptime();
  960         if (callout_active(&timer->tt_delack))
  961                 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
  962         if (callout_active(&timer->tt_rexmt))
  963                 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
  964         if (callout_active(&timer->tt_persist))
  965                 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
  966         if (callout_active(&timer->tt_keep))
  967                 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
  968         if (callout_active(&timer->tt_2msl))
  969                 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
  970         xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
  971 }

Cache object: e4d014ddac5215416e54c3e00c1e37b9


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.