#!/bin/sh # # Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved. # # init.cssd - Control script for the Oracle CSS daemon. # # In a full RAC install, this should not be in an rcX.d # directory. It should only be invoked from init.crs. # Never by hand. # # No manual invocation of init.cssd is supported on a cluster. # # In a local, non-cluster, installation without RAC, it should # be placed in an rcX.d directory. It may be invoked by hand # if necessary, however there are a series of risks and complications # that should be kept in mind when doing so. # # Actions usable in local-only configuration are: # start, stop, disable and enable. # # ========================================================== # Porting Definitions and requirements: # # FAST_REBOOT - take out the machine now. We are concerned about # data integrity since the other node has evicted us. # SLOW_REBOOT - We can rely on vendor clusterware to delay sending # the reconfig to the other node due to its IO fencing # guarantees. So trigger its diagnostic reboot. # VC_UP - Determine whether Vendor clusterware processes are active. # If they are, then the remote node will hear that CSS/CLSVMON # have died, and we will be forced to use FAST_REBOOT. # This is also used at startup time for dependency checking. # Returns 0 if they are up, 1 otherwise. This should be # an extremely fast check. # CLINFO - Determine whether we are booted in non-cluster mode. # Returns 0 for cluster mode, 1 for non-cluster mode # This call is allowed to take a long time to decide. # # GETBOOTID - Returns a string that uniquely identifies this boot. # This string must never change while the machine is up, # and must change the next time the machine boots. ORA_CRS_HOME=%ORA_CRS_HOME% ORACLE_USER=%OwnerId% ORACLE_HOME=$ORA_CRS_HOME export ORACLE_HOME export ORA_CRS_HOME export ORACLE_USER # Set DISABLE_OPROCD to false. Platforms that do not ship an oprocd # binary should override this below. DISABLE_OPROCD=false # Default OPROCD timeout values defined here, so that it can be # over-ridden as needed by a platform. # default Timout of 1000 ms and a margin of 500ms OPROCD_DEFAULT_TIMEOUT=1000 OPROCD_DEFAULT_MARGIN=500 # default Timeout for other actions OPROCD_CHECK_TIMEOUT=2000 OPROCD_STOP_TIMEOUT=2000 OPROCD_DEFAULT_HISTORGRAM= # Incase /bin/hostname is not present in a particular platform, we # may have to do something different. HOSTN=/bin/hostname EXPRN=/usr/bin/expr CUT=/usr/bin/cut AWK='/bin/awk' ECHO='echo' TR=/bin/tr #solaris on amd and SPARC has issue with /bin/tr [ 'SunOS' = `/bin/uname` ] && TR=/usr/xpg4/bin/tr #on Linux tr is at /usr/bin/tr [ 'Linux' = `/bin/uname` ] && TR=/usr/bin/tr #If the hostname is an IP address, let hostname #remain as IP address HOST=`$HOSTN` len1=`$EXPRN "$HOST" : '.*'` len2=`$EXPRN "$HOST" : '[0-9]*\.[0-9]*\.[0-9]*\.[0-9]*'` # Strip off domain name in case /bin/hostname returns # FQDN hostname if [ $len1 != $len2 ]; then HOST=`$ECHO $HOST | $CUT -d'.' -f1 ` fi HOST=`$ECHO $HOST | $TR '[:upper:]' '[:lower:]'` # Default Location for commands on most platforms PS='/bin/ps' # ps -e is expected to search for all processes on the box and provide # terse binary name output so that column count does not truncate binary # names and confuse grep. PSE='/bin/ps -e' PSEF='/bin/ps -ef' HEAD='/bin/head' GREP='/bin/grep' KILL='/bin/kill' KILLTERM='/bin/kill -TERM' KILLDIE='/bin/kill -9' KILLCHECK="/bin/kill -0 $$" SLEEP='/bin/sleep' NULL='/dev/null' UNAME='/bin/uname' CAT='/bin/cat' RMF='/bin/rm -f' TEST='/bin/test' WCL='/bin/wc -l' TOUCH='/bin/touch' SU='/bin/su' SED='/bin/sed' RENICE='/bin/renice' RTGPID='/bin/priocntl -s -c RT -i pgid' XARGS='/bin/xargs' PWD_COMMAND=/usr/bin/pwd MVF="/bin/mv -f" EVAL=eval LS='/bin/ls' BASENAME=/bin/basename DATE="/bin/date" MKDIRP="/bin/mkdir -p" CHOWN="/bin/chown" SYNC='/bin/sync' USING_VC=0 # not using vendor clusterware by default USING_DISKMON='false' # not using diskmon by default # This should be a noop on non-Linux platforms. OMONSLEEP="" # Default value for sleep when doing a sync SYNCSLEEP='0' UNIQUEDATE=`/bin/date '+%Y-%m-%d-%H:%M:%S'` UNIQUECORE=core.$UNIQUEDATE FINDCLSVMON="$PS -e -o comm,pid | $GREP '^oclsvmon'" FINDCLSOMON="$PS -e -o comm,pid | $GREP '^oclsomon'" FINDCSSD="$PS -e -o comm,pid | $GREP '^ocssd'" FINDPROCD="$PS -e -o comm,pid | $GREP '^oprocd'" FINDDISKMON="$PS -e -o comm,pid | $GREP '^diskmon'" # Error reporting commands. These need to be kept in sync with # the value of $MSGFILE in root.sh. LOGERR="/bin/logger -puser.alert" LOGMSG="/bin/logger -puser.err" # See discussion in LocalFence below. CLEANREBOOTLOCK=/bin/true OPROCD=$ORA_CRS_HOME/bin/oprocd CRSCTL=$ORA_CRS_HOME/bin/crsctl DISKMON=$ORA_CRS_HOME/bin/diskmon # default diagnostic collection script OPROCD_DIAGSCRIPT=$ORA_CRS_HOME/bin/oprocd_diag # Temp file for crsctl output CRSCTLOUT=/tmp/crsctl.$$ # indicates that the key doesnt exist in OCR CRSCTL_KEY_UNDEFINED=12 # Default path for SCR control files. SCRDIR=/var/opt/oracle/scls_scr/$HOST # Directory for OPROCD logfiles OPROCDIR=/var/opt/oracle/oprocd # Command to make a process run in realtime REALTIME_PID= # How long to sleep in startcheck to cover asynchronous startups of things like # networks, disks, and filesystems not readily checked by crsctl. This time should # be 0 except in extenuating circumstances. STARTCHECKSLEEP=0 # Rate at which we recheck the flag files for a manual startup, after # deciding not to start. RUNRECHECKTIME=30 # Defines for the priority boost mechanism PRIORITY_BOOST_DISABLED=0 PRIORITY_BOOST_LOW=1 PRIORITY_BOOST_MID=2 PRIORITY_BOOST_HIGH=3 PRIORITY_BOOST_REALTIME=4 # If a platform wants to override usage of priority boost functionality # that platform should override PRIORITY_BOOST_DEFAULT in the platform # specific section below. PRIORITY_BOOST_DEFAULT=$PRIORITY_BOOST_HIGH # How long to wait (in seconds) before rechecking a dependency, # and printing out messages about it. DEP_CHECK_WAIT=60 #ulimit set to unlimited ULIMIT_CORE="ulimit -c unlimited" #Check if SUBSYSFILE is created CHECK_SUBSYSFILE="" PLATFORM=`$UNAME` MAXFILE=65536 case $PLATFORM in Linux) # Presence of this file indicates that vendor clusterware is installed # And on Linux, we don't support any vendor clusterware. SKGXNLIB=$ORA_CRS_HOME/lib/INVALID_DIRECTORY # touch /var/lock/subsys/init.crs so that the shutdown # scripts get called during shutdown time (refer bug 4997126) SUBSYSFILE="/var/lock/subsys/init.crs" if [ ! -f $TEST ]; then TEST=/usr/bin/test fi if [ -f $SKGXNLIB ]; then USING_VC=1 fi LD_LIBRARY_PATH=$ORA_CRS_HOME/lib export LD_LIBRARY_PATH HEAD='/usr/bin/head' # Use runuser on systems which have SElinux, to avoid # interactive scripts when su is used. if [ -f "/sbin/runuser" ]; then SU="/sbin/runuser -l" else SU="/bin/su -l" fi # Over-ride SCRDIR SCRDIR=/etc/oracle/scls_scr/$HOST # Command to make a process run in realtime REALTIME_CMD=/usr/bin/chrt if [ ! -f $REALTIME_CMD ]; then REALTIME_CMD=/bin/chrt fi if [ -x $REALTIME_CMD ]; then REALTIME_PID="$REALTIME_CMD -r -p 99" fi USING_DISKMON='true' # using diskmon on this platform CLINFO=/bin/true if [ $USING_VC -eq 1 ]; then # We are using PrimeCluster as vendor clusterware SLOW_REBOOT="/usr/bin/pkill -HUP syslogd ; /bin/sync & /bin/sleep 2; /sbin/reboot -n -f" FAST_REBOOT="$ECHO b > /proc/sysrq-trigger" VC_UP=/bin/true # No known way to detect the clusterware via PS VC_NAME="PrimeCluster" else # No valid vendor clusterware FAST_REBOOT="$ECHO b > /proc/sysrq-trigger" VC_UP=/bin/true VC_NAME="" fi ID=/etc/init.d LOGGER="/usr/bin/logger" if [ ! -f "$LOGGER" ];then LOGGER="/bin/logger" fi LOGMSG="$LOGGER -puser.err" LOGERR="$LOGGER -puser.alert" RENICE="/usr/bin/renice" OPROCDIR=/etc/oracle/oprocd # Enable kernel sysrqs so that FAST_REBOOT can run. Occasionally # a user will run init.cssd as non-root. Avoid the error message # by confirming the userid first. MYID=`/usr/bin/id -u` if [ "$MYID" = "0" ]; then $ECHO 1 > /proc/sys/kernel/sysrq # Don't get timezone when the stack is not up RUNFILE=$SCRDIR/root/cssrun RUNID= if [ -r $RUNFILE ]; then RUNID=`$CAT $RUNFILE` fi # if RUNFILE not readable, RUNID will be null if [ "norun" != "$RUNID" ]; then $SU $ORACLE_USER -c "$ECHO \$TZ > /tmp/oratz.$$ " > /dev/null 2>&1 if [ -f "/tmp/oratz.$$" ]; then NEWTZ=`$SED 's/^[ \t]*//;s/[ \t]*$//' < /tmp/oratz.$$` $RMF /tmp/oratz.$$ fi if [ ! -z "$NEWTZ" ]; then TZ=$NEWTZ; export TZ; fi fi fi # Make the CSSD run realtime by default PRIORITY_BOOST_DEFAULT=$PRIORITY_BOOST_REALTIME # See discussion in LocalFence REBOOTLOCKFILE=/var/tmp/.orarblock CLEANREBOOTLOCK="$RMF $REBOOTLOCKFILE" # We are trying to get a timestamp of when the machine booted. It # appears that access time on /proc will do this, even if it looks odd. GETBOOTID="eval $LS -ld --full-time /proc | $AWK '{print \$6\$7\$8\$9\$10;}'" DIRNAME=/usr/bin/dirname XARGS=/usr/bin/xargs $SU $ORACLE_USER -c "$ECHO \$TZ > /tmp/oratz.$$ " > /dev/null 2>&1 NEWTZ=`$SED 's/^[ \t]*//;s/[ \t]*$//' < /tmp/oratz.$$` $RMF /tmp/oratz.$$ if [ ! -z "$NEWTZ" ]; then TZ=$NEWTZ; export TZ; fi # Delay node fence invoked by oclsomon by 1 second on Linux. OMONSLEEP="$SLEEP 1" ;; HP-UX) MACH_HARDWARE=`/bin/uname -m` CLUSTERDIR=/opt/nmapi/nmapi2 if [ "$MACH_HARDWARE" = "ia64" ]; then SO_EXT=so NMAPIDIR_64=$CLUSTERDIR/lib/hpux64 else SO_EXT=sl NMAPIDIR_64=$CLUSTERDIR/lib/pa20_64 fi LD_LIBRARY_PATH=$ORA_CRS_HOME/lib:$NMAPIDIR_64:/usr/lib:$LD_LIBRARY_PATH export LD_LIBRARY_PATH # Presence of this file indicates that vendor clusterware is installed SKGXNLIB=${NMAPIDIR_64}/libnmapi2.${SO_EXT} if [ -f $SKGXNLIB ]; then USING_VC=1 fi VC_UP="$PSEF | $GREP '/usr/lbin/cm[g]msd' 1>$NULL 2>$NULL" # NOTE: Not supporting Hardware partitioning machines. REBOOT_TOC="/var/opt/oracle/bin/reboot_toc" FAST_REBOOT="$REBOOT_TOC || /usr/sbin/reboot -r -n -q" SLOW_REBOOT="/bin/kill -HUP `$CAT /var/run/syslog.pid` ; /bin/sync & $SLEEP 2 ; $REBOOT_TOC || /usr/sbin/reboot -r -n -q" DFL_CLSINFO=$CLUSTERDIR/bin/clsinfo if [ $USING_VC -eq 0 ]; then # We are not using vendor clusterware. VC_NAME="" VC_UP=/bin/true CLINFO=/bin/true FAST_REBOOT="$REBOOT_TOC || /usr/sbin/reboot -r -n -q" SLOW_REBOOT="$REBOOT_TOC || /usr/sbin/reboot -r -n -q" elif [ -f $DFL_CLSINFO ]; then # We are using Generic HP Vendor Clusterware VC_NAME="Generic HP-UX Vendor Clusterware" VC_UP=/bin/true CLINFO=$DFL_CLSINFO else # We are using HPUX ServiceGuard Clusterware VC_NAME="HP-UX Service Guard" CLINFO=/bin/true fi ID=/sbin/init.d LVINFO=/usr/sbin/lvdisplay VGINFO=/usr/sbin/vgdisplay VG_ACTIVATE="/usr/sbin/vgchange -a s" VG_DEACTIVATE="/usr/sbin/vgchange -a n" GETBOOTID="/bin/who -b" RENICE='/usr/bin/renice -n ' DIRNAME=/bin/dirname UNIX95=true export UNIX95 # Make the CSSD run realtime by default PRIORITY_BOOST_DEFAULT=$PRIORITY_BOOST_REALTIME USING_DISKMON='true' # using diskmon on this platform # Command to make a process run in realtime REALTIME_CMD=/usr/bin/rtprio if [ ! -f $REALTIME_CMD ]; then REALTIME_CMD=/bin/rtprio fi if [ -x $REALTIME_CMD ]; then REALTIME_PID="$REALTIME_CMD 127" fi ;; SunOS) MACH_HARDWARE=`/bin/uname -i` ARCH=`/usr/bin/isainfo -b` CLUSTERDIR=/opt/ORCLcluster LD_LIBRARY_PATH=$ORA_CRS_HOME/lib:$CLUSTERDIR/lib:/usr/lib:/usr/ucblib:$LD_LIBRARY_PATH LD_LIBRARY_PATH_64=$ORA_CRS_HOME/lib:$CLUSTERDIR/lib:/usr/lib:/usr/ucblib:$LD_LIBRARY_PATH_64 if [ "${MACH_HARDWARE}${ARCH}" = "i86pc64" ]; then LD_LIBRARY_PATH=$ORA_CRS_HOME/lib:$CLUSTERDIR/lib:/usr/lib/amd64:/usr/ucblib/amd64:$LD_LIBRARY_PATH LD_LIBRARY_PATH_64=$ORA_CRS_HOME/lib:$CLUSTERDIR/lib:/usr/lib/amd64:/usr/ucblib/amd64:$LD_LIBRARY_PATH_64 fi export LD_LIBRARY_PATH export LD_LIBRARY_PATH_64 # Presence of this file indicates that vendor clusterware is installed SKGXNLIB=$CLUSTERDIR/lib/libskgxn2.so if [ -f $SKGXNLIB ]; then USING_VC=1 fi # Define some variables based on the presence of vendor clusterware. DFL_CLSINFO=$CLUSTERDIR/bin/clsinfo if [ $USING_VC -eq 0 ]; then # We are not using vendor clusterware. SLOW_REBOOT="/sbin/uadmin 1 1" FAST_REBOOT="/sbin/uadmin 1 1" VC_UP=/bin/true CLINFO=/bin/true VC_NAME="" elif [ -f $DFL_CLSINFO ]; then # We are using vendor clusterware other than Sparc64 SunCluster SLOW_REBOOT="/bin/pkill -HUP syslogd ; /bin/sync & /bin/sleep 2; /sbin/uadmin 5 1" FAST_REBOOT="/sbin/uadmin 1 1" CLINFO=$DFL_CLSINFO VC_UP=/bin/true # No known way to detect the clusterware via PS VC_NAME="Generic SunOS Vendor Clusterware" else # We are using SunCluster as vendor clusterware SLOW_REBOOT="/bin/pkill -HUP syslogd ; /bin/sync & /bin/sleep 2; /bin/pkill -TERM ucmmd ; /bin/sleep 35 ; /sbin/uadmin 5 1" FAST_REBOOT="/sbin/uadmin 1 1" CLINFO=/usr/sbin/clinfo VC_UP="$PSE -u 0 | $GREP 'ucm[m]d' 1>$NULL 2>$NULL" if [ "$MACH_HARDWARE" = "i86pc" ]; then SLOW_REBOOT="/bin/pkill -HUP syslogd ; /bin/sync & /bin/sleep 2; /bin/pkill -TERM ucmmd ; /bin/sleep 45 ; /sbin/uadmin 5 1" VC_NAME="SunCluster on x86" else VC_NAME="SunCluster on Sparc" fi fi ID=/etc/init.d GETBOOTID="/bin/who -b" DIRNAME=/bin/dirname # Changes for bug 7551160 ZONENAME="/bin/zonename" MAJOROS=`$UNAME -r | $AWK -F. '{print $1}'` MINOROS=`$UNAME -r | $AWK -F. '{print $2}'` if [ $MAJOROS -ge 5 -a $MINOROS -ge 10 -a -f $ZONENAME ]; then FINDCLSVMON="$PS -z `$ZONENAME` -o fname,pid | $GREP '^oclsvmon'" FINDCLSOMON="$PS -z `$ZONENAME` -o fname,pid | $GREP '^oclsomon'" FINDDISKMON="$PS -z `$ZONENAME` -o fname,pid | $GREP '^diskmon'" FINDCSSD="$PS -z `$ZONENAME` -o fname,pid | $GREP '^ocssd'" else FINDCLSVMON="$PS -e -o fname,pid | $GREP '^oclsvmon'" FINDCLSOMON="$PS -e -o fname,pid | $GREP '^oclsomon'" FINDDISKMON="$PS -e -o fname,pid | $GREP '^diskmon'" FINDCSSD="$PS -e -o fname,pid | $GREP '^ocssd'" fi # Make the CSSD run realtime by default PRIORITY_BOOST_DEFAULT=$PRIORITY_BOOST_REALTIME # Command to make a process run in realtime REALTIME_CMD=/usr/bin/priocntl if [ ! -f $REALTIME_CMD ]; then REALTIME_CMD=/bin/priocntl fi if [ -x $REALTIME_CMD ]; then # RT priority at 110 for Solaris REALTIME_PID="$REALTIME_CMD -s -c RT -p 10 -i pid" fi ;; AIX) CLUSTERDIR=/opt/ORCLcluster LIBPATH=$ORA_CRS_HOME/lib:$CLUSTERDIR/lib:/usr/lib:$LIBPATH export LIBPATH RENICE="/usr/sbin/renice" FAST_REBOOT="/usr/bin/sysdumpstart -p" SLOW_REBOOT="/bin/kill -HUP `$CAT /etc/syslog.pid`; /bin/sync & $SLEEP 2; /usr/sbin/fastboot -n -q" # Presence of this file indicates that vendor clusterware is installed SKGXNLIB=/usr/sbin/cluster/utilities/cldomain if [ -f $SKGXNLIB ]; then USING_VC=1 fi # Over-ride SCRDIR OPROCDIR=/etc/oracle/oprocd SCRDIR=/etc/oracle/scls_scr/$HOST VC_UP="$PSEF | $GREP 'ha[g]sd' | $GREP '[g]rpsvcs' 1>$NULL 2>$NULL" # NOTE: If RPD happens to get supported later, modify clustername # accordingly DFL_CLSINFO=$CLUSTERDIR/bin/clsinfo if [ $USING_VC -eq 0 ]; then # We are not using vendor clusterware. VC_NAME="" VC_UP=/bin/true CLINFO=/bin/true FAST_REBOOT="/usr/bin/sysdumpstart -p" SLOW_REBOOT="/usr/bin/sysdumpstart -p" elif [ -f $DFL_CLSINFO ]; then # We are running Generic Clusterware VC_NAME="Generic AIX Vendor Cluster" CLINFO=$DFL_CLSINFO VC_UP=/bin/true else # We are running AIX HACMP VC_NAME="AIX HACMP Vendor Clusterware" VC_UP=/bin/true CLINFO=/bin/false USING_VC=0 fi ID=/etc LVINFO=/usr/sbin/lslv VGINFO=/usr/sbin/lsvg VG_ACTIVATE="/etc/varyonvg -c" VG_DEACTIVATE="/etc/varyoffvg" GETBOOTID="/bin/who -b" DIRNAME=/bin/dirname # Make the CSSD run realtime by default PRIORITY_BOOST_DEFAULT=$PRIORITY_BOOST_REALTIME # Command to make a process run in higher priority OPRIPROC=opriproc OPRIPROC_DIR="/etc/oracle/bin" if [ -f "$OPRIPROC_DIR/$OPRIPROC" ]; then REALTIME_CMD="$OPRIPROC_DIR/$OPRIPROC" else REALTIME_CMD=$RENICE fi if [ -x "$OPRIPROC_DIR/$OPRIPROC" ]; then REALTIME_PID="$OPRIPROC_DIR/$OPRIPROC -p " elif [ -x $REALTIME_CMD ]; then REALTIME_PID="$REALTIME_CMD -20" fi ;; OSF1) LD_LIBRARY_PATH=$ORA_CRS_HOME/lib:/usr/lib export LD_LIBRARY_PATH FAST_REBOOT="/usr/sbin/reboot -n -q" SLOW_REBOOT="/bin/kill -HUP `$CAT /var/run/syslog.pid`; /bin/sync & $SLEEP 2 ; /usr/sbin/reboot -n -q" # No need for OPROCD on Tru64. We don't support a configuration # without Tru64 clusterware, AND, Tru64 clusterware is responsible # for I/O Fencing. DISABLE_OPROCD=true # Presence of this file indicates that vendor clusterware is installed SKGXNLIB=/usr/ccs/lib/libdlm.a if [ -f $SKGXNLIB ]; then USING_VC=1 fi VC_UP=/bin/true VC_NAME="TruCluster" CLINFO=/bin/true ID=/sbin/init.d GETBOOTID="/bin/who -b" DIRNAME=/usr/bin/dirname MAXFILE=unlimited ;; *) /bin/echo "ERROR: Unknown Operating System" exit -1 ;; esac OPROCDLOG=$OPROCDIR/$HOST.oprocd.log OPROCDLGL=$OPROCDIR/$HOST.oprocd.lgl # If no immediate reboot specified, use FAST_REBOOT if [ -z "$SLOW_REBOOT" ] then SLOW_REBOOT=$FAST_REBOOT fi # These six files contain system settings used for controlling CSS, and CRS. AUTOSTARTFILE=$SCRDIR/root/crsstart FATALFILE=$SCRDIR/$ORACLE_USER/cssfatal NOCLSVMON=$SCRDIR/root/noclsvmon NOCLSOMON=$SCRDIR/root/noclsomon NOOPROCD=$SCRDIR/root/nooprocd RUNFILE=$SCRDIR/root/cssrun CLSVMONPIDFILE=$SCRDIR/root/clsvmonpid CLSOMONPIDFILE=$SCRDIR/root/clsomonpid DAEMONPIDFILE=$SCRDIR/root/daemonpid OPROCDPIDFILE=$SCRDIR/root/oprocdpid DISKMONPIDFILE=$SCRDIR/root/oprocdpid CSSFBOOT=$SCRDIR/root/cssfboot NODISKMON=$SCRDIR/root/nodiskmon # start diskmon or not DISKMONUSED=$SCRDIR/$ORACLE_USER/diskmonused # use diskmon or not # This function is applicable only on Linux flavors createSUBSYSFILE() { if [ "$PLATFORM" = "Linux" ] ; then # touch /var/lock/subsys/init.crs so that the shutdown # scripts get called during shutdown time (refer bug 4997126) if [ "$CHECK_SUBSYSFILE" == "" ]; then $TOUCH $SUBSYSFILE CHECK_SUBSYSFILE=1 fi fi } validateORACLE_USER() { ret=0; case $PLATFORM in Linux) ;; HP-UX) ;; SunOS) ;; AIX) LSUSER=/usr/sbin/lsuser CHUSER=/usr/bin/chuser SED=/usr/bin/sed USER=$ORACLE_USER CURRENT_CAPABILITIES=`$LSUSER -a capabilities $USER | $SED -e "s-$USER--g" -e "s-.*=--" -e "s-,- -g"` REQUIRED_CAPABILITIES="CAP_NUMA_ATTACH CAP_BYPASS_RAC_VMM CAP_PROPAGATE" NEEDED_CAPABILITIES="" for capabilities in $REQUIRED_CAPABILITIES do if [ `$ECHO $CURRENT_CAPABILITIES | $GREP -w -q $capabilities; $ECHO $?` -ne 0 ];then NEEDED_CAPABILITIES="$capabilities $NEEDED_CAPABILITIES" fi done if [ ! -z "$NEEDED_CAPABILITIES" ]; then $LOGERR "INIT.CSSD Failure. Change <$USER> userid to have the following OS capabilities :"\ "<$CURRENT_CAPABILITIES $NEEDED_CAPABILITIES> by running <$CHUSER> command." ret=1; fi ;; *) ;; esac return $ret; } # Initialize diagnositic collection system if any exists on the system StartDiagCollect() { # we only need to start diagnostic collection if the diagcollect script # exists on the users system, otherwise we skip it if [ -f $OPROCD_DIAGSCRIPT ]; then $OPROCD_DIAGSCRIPT start > $NULL 2>&1 & OPROCD_DIAGSCRIPT_PID=$! fi } # Termination of the diagnostic collection system StopDiagCollect() { if [ ! -z "$OPROCD_DIAGSCRIPT_PID" ]; then $OPROCD_DIAGSCRIPT stop $OPROCD_DIAGSCRIPT_PID > $NULL 2>&1 fi } # Define the reboot action which will be used to prevent IO from the local # node from corrupting the database. LocalFence() { # We want to reboot here as fast as possible. It is imperative # that we do not flush any IO to the shared disks. Choosing not # to flush local disks or kill off processes gracefully shuts # us down quickly. But there are a couple cases that allow us to go # down more gracefully as outlined below. case `$CAT $FATALFILE` in disable*) # No reboot action is necessary at this time. We have not finished # the install and allowed database activity to start. REBOOT_CMD="$SLEEP 1" ;; *) $LOGERR "Oracle CRS failure. Rebooting for cluster integrity." # If there is no vendor clusterware, then we do not have a slow reboot # mechanism. if [ -r "$SKGXNLIB" ]; then READABLE_SKGXN=1 else READABLE_SKGXN=0 fi if [ $USING_VC -eq 0 -o $READABLE_SKGXN -eq 0 ]; then REBOOT_CMD="$FAST_REBOOT" else # is clsvmon up? VMONUP=`$EVAL $FINDCLSVMON` # is vendor clusterware alive? $EVAL $VC_UP RC=$? # If vmon is down and vendor clusterware is up, then the other node # is about to evict us, and we should kill the node quickly. if [ -z "$VMONUP" ] && [ "$RC" = "0" ]; then REBOOT_CMD="$FAST_REBOOT" else REBOOT_CMD="$SLOW_REBOOT" fi fi ;; esac # If diagwait is set, $SYNCSLEEP should be greater than 0, so # we can sync and sleep a small amount to get better diagnostics. if [ $SYNCSLEEP -gt 0 ]; then $SYNC& $SLEEP $SYNCSLEEP fi # Workaround to Redhat 3 issue with multiple invocations of reboot. # Here if oclsomon and ocssd are attempting a reboot at the same time # then the kernel could lock up. Here we have a crude lock which # doesn't eliminate but drastically reduces the likelihood of getting # two reboots at once. if [ "$PLATFORM" = "Linux" ]; then CEDETO= if [ -e "$REBOOTLOCKFILE" ]; then CEDETO=`$CAT $REBOOTLOCKFILE` fi $ECHO $$ > $REBOOTLOCKFILE if [ ! -z "$CEDETO" ]; then REBOOT_CMD="$SLEEP 0" $LOGMSG "Oracle init script ceding reboot to sibling $CEDETO." fi fi $EVAL $REBOOT_CMD $SLEEP 120 } # Priority boost is only applied to the CSS daemon, invoked by one of these two actions. if [ $1 = 'daemon' ] || [ $1 = 'run' ] || [ $1 = 'diag' ]; then # since we got the node name, now try to get the actual # boost value for the node PRIORITY_BOOST_VALUE=`$CRSCTL get css priority` RC=$? # be sure to check if we hit any errors case $RC in 0) # no errors, all things are good ;; $CRSCTL_KEY_UNDEFINED) # if we get back an error saying the value wasn't there, # then we can just assume the default PRIORITY_BOOST_VALUE=$PRIORITY_BOOST_DEFAULT ;; *) # for any other errors we still assume the defaults PRIORITY_BOOST_VALUE=$PRIORITY_BOOST_DEFAULT ;; esac fi # Get appropriate values for diagwait and reboottime if [ $1 = 'daemon' ] || [ $1 = 'oprocd' ] || [ $1 = 'oclsomon' ]; then # Get diagwait value DIAGWAIT=`$CRSCTL get css diagwait` RC=$? if [ "$RC" != "0" ]; then # No diagwait set DIAGWAIT=0 fi # Get reboottime value. A value is always returned REBOOTTIME=`$CRSCTL get css reboottime` if [ $DIAGWAIT -gt $REBOOTTIME ]; then # Max time we can sleep waiting for a sync to complete SYNCSLEEP=`$EXPRN $DIAGWAIT - $REBOOTTIME` # Relax oprocd margin to allow for poor scheduling OPROCD_DEFAULT_MARGIN=`$EXPRN $SYNCSLEEP \* 1000` # Calculate 500 ms and 1000 ms offsets based on OPROCD_DEFAULT_MARGIN OPROCD_HISTOGRAM_BASE_500=`$EXPRN 500 \* 100 / $OPROCD_DEFAULT_MARGIN` OPROCD_HISTOGRAM_BASE_1000=`$EXPRN 1000 \* 100 / $OPROCD_DEFAULT_MARGIN` OPROCD_DEFAULT_HISTOGRAM="-hsi $OPROCD_HISTOGRAM_BASE_500:$OPROCD_HISTOGRAM_BASE_1000:50:75:90" else # do not print error msg if not set if [ $DIAGWAIT -gt 0 ]; then $LOGERR "INIT.CSSD ERROR: Diagwait set to $DIAGWAIT, must be"\ "greater than reboottime of $REBOOTTIME" $LOGERR "INIT.CSSD ERROR: diagwait option disabled" $LOGERR "INIT.CSSD ERROR: Use '$CRSCTL set css diagwait N' to correct" fi fi fi # This will be consumed by the CSS daemon binary later export PRIORITY_BOOST_VALUE # Validate ORACLE_USER validateORACLE_USER || { $ECHO $?; exit 1; } case $1 in 'home') $ECHO $ORA_CRS_HOME exit 0; ;; 'user') $ECHO $ORACLE_USER exit 0; ;; 'diag') $ECHO CRS init script diagnostics. $ECHO TZ=$TZ $ECHO LANG=$LANG $ECHO USER=$ORACLE_USER $ECHO HOME=$ORA_CRS_HOME $ECHO AUTO=`$CAT $AUTOSTARTFILE` $ECHO RUNID=`$CAT $RUNFILE` $ECHO BOOTID=`$ID/init.cssd bootid` $ECHO FATAL=`$CAT $FATALFILE` $ECHO SCS=$STARTCHECKSLEEP $ECHO RCT=$RUNCHECKTIME $ECHO PRIORITY_BOOST_VALUE=$PRIORITY_BOOST_VALUE if [ -f "$SKGXNLIB" ]; then $ECHO VC=Yes else $ECHO VC=No fi if [ -f "$NOCLSVMON" ]; then $ECHO CLSV=No else $ECHO CLSV=Yes fi if [ -f "$NOCLSOMON" ]; then $ECHO CLSO=No else $ECHO CLSO=Yes fi if [ -f "$NOOPROCD" ]; then $ECHO OPROC=No else $ECHO OPROC=Yes fi if [ -f "$NODISKMON" ]; then $ECHO DISKMON=No else $ECHO DISKMON=Yes fi exit 0; ;; 'bootid') # Set a null timezone. Some implementations of GETBOOTID are sensitive # to the timezone in the multiple environments this can get invoked from. TZ= export TZ # Set a null language id. GETBOOTID can also be sensitive to the language # codes, when init.cssd is run in a user shell compared to being run # under init. Setting a blank language code should guarantee consistency. LANG= export LANG $GETBOOTID exit 0; ;; 'disable') # Disable the CSS daemon from starting at boot time. $ECHO "Automatic startup disabled for system boot." $ECHO disable > $AUTOSTARTFILE exit 0; ;; 'enable') # Enable automatic CSS startup at boot time. $ECHO "Automatic startup enabled for system boot." $ECHO enable > $AUTOSTARTFILE exit 0; ;; 'norun') # Mark the norun flag to disable any daemons from restarting. $ECHO "norun" > $RUNFILE exit 0; ;; 'start') CMD=`$BASENAME $0` # If we are being invoked by the user, perform manual startup. # If we are being invoked as an RC script, check for autostart. if [ "$CMD" = "init.cssd" ]; then $LOGMSG "Oracle Cluster Synchronization Service starting by user request." $ID/init.cssd manualstart else $ID/init.cssd autostart fi ;; 'manualstart') # Create lock file SUBSYSFILE createSUBSYSFILE; # Queue a startup to the CRS scripts invoked by inittab. BOOTID=`$ID/init.cssd bootid` $ECHO "$BOOTID" > $RUNFILE STARTTIME=`$EXPRN $RUNRECHECKTIME + $STARTCHECKSLEEP` $ECHO "Startup will be queued to init within $STARTTIME seconds." ;; 'autostart') # Create lock file SUBSYSFILE createSUBSYSFILE; # Check to see whether the user has enabled automatic startup. # Is the flag file still there? if [ -r $AUTOSTARTFILE ] then case `$CAT $AUTOSTARTFILE` in disable*) $LOGMSG "Oracle Cluster Ready Services disabled by administrator." $ID/init.cssd norun ;; enable*) # Continue on to the next check $LOGMSG "Oracle Cluster Ready Services starting up automatically." $ID/init.cssd manualstart ;; *) $LOGERR "Oracle Cluster Ready Services disabled by damaged install." $LOGERR " Unexpected settings found in $AUTOSTARTFILE." $ID/init.cssd norun ;; esac else # If the SCR directory does not exist, then either the user changed # their hostname, or they wiped sections of the disk. Give them a # slightly less frightening message. if [ ! -d "$SCRDIR" ] then $LOGERR Oracle Cluster Ready Services startup disabled. $LOGERR Did the hostname change? Could not access $AUTOSTARTFILE. else $LOGERR Oracle Cluster Ready Services is disabled by damaged install. $LOGERR Could not access $AUTOSTARTFILE. fi $ID/init.cssd norun fi exit 0; ;; # Runcheck is intended to check the run status for any of the # run or fatal scripts (evm/crs/css) as initialized by the automatic startup # or manual startup routines. # Runcheck is supposed to be a very fast check, and be silent since it # will get invoked regularly while the system is up. # Exit codes should match the requirements of startcheck. 'runcheck') if [ ! -r "$RUNFILE" ]; then # If this file does not exist, then the automatic startup checks # in the "start" target will warn the user of problems. exit 3; fi BOOTID=`$ID/init.cssd bootid` RUNID=`$CAT $RUNFILE` if [ "$BOOTID" != "$RUNID" ]; then if [ "$RUNID" = "startstack" ]; then $ID/init.cssd manualstart else # We do not have clearance to start up. So respect that authority. # Nothing needs to be logged in this case. exit 3; fi fi exit 0; ;; # Startcheck is intended to check the run status for any of the # run or fatal scripts (evm/crs/css) (see runcheck), and additionally # check on vendor clusterware and filesystem dependencies. # Startcheck is supposed to block for a long time waiting on dependencies. 'startcheck') # Returns 0 if we should start # Returns 1 on a non-cluster boot # Returns 2 if disabled by admin # Returns 3 on error # Check our startup run status, initialized by the automatic startup # or manual startup routes. $ID/init.cssd runcheck STATUS=$? if [ "$STATUS" != "0" ]; then exit $STATUS; fi # If we have vendor clusterware, and CLINFO indicates a non-clustered boot # then prevent CRS startup. if [ $USING_VC -eq 1 ] then # Run CLINFO to get cluster status. It returns 0 if the machine is # booting in clustered mode. $EVAL $CLINFO if [ "$?" != "0" ]; then $LOGMSG "Oracle Cluster Ready Services disabled by non-clustered boot." $ID/init.cssd norun exit 1; fi # Wait for the Vendor Clusterware to start $EVAL $VC_UP RC=$? while [ $RC -ne 0 ]; do $LOGMSG "Oracle Cluster Ready Services waiting for $VC_NAME to start." $SLEEP $DEP_CHECK_WAIT $EVAL $VC_UP RC=$? done fi # Wait until it is safe to start CRS daemons while [ ! -r $CRSCTL ] do $LOGMSG "Waiting for filesystem containing $CRSCTL." $SLEEP $DEP_CHECK_WAIT done # Now bring up the OCR, Voting Disk Volume Group(s), but only if we are # called by the CSS startup if [ "$2" = "CSS" ] then $SLEEP 0 # do nothing for now. fi # Wait for additional filesystems and objects to become available # crsctl should print out a message indicating cause of failure. $SU $ORACLE_USER -c "$CRSCTL check boot > $CRSCTLOUT" RC=$? while [ "$RC" != "0" ] do $LOGMSG Cluster Ready Services waiting on dependencies. Diagnostics in $CRSCTLOUT. $SLEEP $DEP_CHECK_WAIT $SU $ORACLE_USER -c "$CRSCTL check boot > $CRSCTLOUT" RC=$? done # Finished waiting on dependencies. Diagnostics are no longer necessary. $LOGMSG Cluster Ready Services completed waiting on dependencies. $RMF $CRSCTLOUT # Some platforms may need additional sleeps to be done here in # rare circumstances. $SLEEP $STARTCHECKSLEEP # all good exit 0; ;; 'oclsomon') # This action is expected to start and monitor the clsomon daemon. # Its purpose is to assist the CSS daemon by monitoring it for hangs. # In the event of a CSS daemon hang, the remote nodes may evict the # current node and so it is necessary to terminate the local node. # On rare occasion this will get re-invoked while a shutdown is # in progress. If so, this script should exit. $ID/init.cssd runcheck if [ "$?" != "0" ]; then exit 0; fi # Arguments. By default none. CLSOMON_ARGS= # Run clsomon out of its own directory in case it drops a core CLSOMONDIR=$ORA_CRS_HOME/log/$HOST/cssd/oclsomon # Backup the oclsomon logfiles $MVF $CLSOMONDIR/oclsomon.ba4 $CLSOMONDIR/oclsomon.ba5 $MVF $CLSOMONDIR/oclsomon.ba3 $CLSOMONDIR/oclsomon.ba4 $MVF $CLSOMONDIR/oclsomon.ba2 $CLSOMONDIR/oclsomon.ba3 $MVF $CLSOMONDIR/oclsomon.ba1 $CLSOMONDIR/oclsomon.ba2 $MVF $CLSOMONDIR/oclsomon.log $CLSOMONDIR/oclsomon.ba1 if [ ! -d "$CLSOMONDIR" ]; then $MKDIRP $CLSOMONDIR $CHOWN $ORACLE_USER $CLSOMONDIR fi # If we can set ourselves into realtime, do so, otherwise renice # ourselves, and the child scripts. This is useful to ensure # that we can perform the fence action if CSS kills itself due to # unreasonable load. This script and oclsomon will remain asleep # most of the time so a high priority shouldn't bother anything. if [ "$REALTIME_PID" != "" ]; then case $PLATFORM in HP-UX) $REALTIME_PID -$$ ;; *) $REALTIME_PID $$ ;; esac else $RENICE -5 $$ fi $SU $ORACLE_USER -c \ "/bin/sh -c 'cd $CLSOMONDIR; $ULIMIT_CORE; $ORA_CRS_HOME/bin/oclsomon $CLSOMON_ARGS || exit \$?'" RC=$? # What kind of return code have we received? case "$RC" in # Graceful shutdown 0) $LOGMSG "Oracle clsomon shutdown successful." $TOUCH $NOCLSOMON exit 0; ;; # oclsomon is already up. Restart it. 56) $EVAL $FINDCLSOMON | $AWK '{ print $2 ; }' | $XARGS $KILLTERM > $NULL 2>&1 $LOGMSG "Duplicate clsomon found. Restarting." exit 0; ;; # Initialization errors 4*|5*|6*) $LOGMSG "Oracle clsomon startup failed with status $RC. Restarting." $SLEEP 5 exit 0; ;; # Steady state errors. # note: segfault and signals may have status 130+, which falls under 1* 1*|2*|3*) $LOGMSG "Oracle clsomon failed with fatal status $RC." $EVAL $OMONSLEEP LocalFence; $ID/init.cssd norun exit 0; ;; # Unknown error *) $LOGMSG "Oracle clsomon failed with unexpected status $RC." $EVAL $OMONSLEEP LocalFence; $ID/init.cssd norun exit 0; esac # all good. exit 0; ;; 'oclsvmon') # This action is expected to start and monitor the clsvmon daemon. # Its purpose is to assist the CSS daemon in monitoring vendor clusterware # and allow additional diagnostics to be obtained in the case of system # failures. # On rare occasion this will get re-invoked while a shutdown is # in progress. If so, this script should exit. $ID/init.cssd runcheck if [ "$?" != "0" ]; then exit 0; fi # Arguments. By default none. CLSVMON_ARGS= # Run clsvmon out of its own directory in case it drops a core CLSVMONDIR=$ORA_CRS_HOME/log/$HOST/cssd/oclsvmon if [ ! -d "$CLSVMONDIR" ]; then $MKDIRP $CLSVMONDIR $CHOWN $ORACLE_USER $CLSVMONDIR fi $SU $ORACLE_USER -c \ "/bin/sh -c 'cd $CLSVMONDIR; $ULIMIT_CORE; $ORA_CRS_HOME/bin/oclsvmon $CLSVMON_ARGS || exit \$?'" RC=$? # what kind of return code have we received? case "$RC" in # CLSVMON will return a status code of 15 if it has seen a failure in # a stable CSS group. This can happen for several reasons, but the most # interesting of these is when an IO freeze causes the daemon to lock up. # In this case, this codepath is the only one available for killing the # machine before the rest of the cluster evicts us. 15) $LOGMSG "Oracle CSS daemon detected IO freeze to its voting disks." $EVAL $FAST_REBOOT $TOUCH $NOCLSVMON exit 0; ;; # CLSVMON will return a status code of 9 if it has received a TERM signal. # TERM signal will either be caused by init when we are shutting down, # or be caused by the graceful shutdown script. In either case, this # monitor script has nothing left to do, and no log message is warranted. 9) $TOUCH $NOCLSVMON exit 0; ;; # CLSVMON will return a status code of 11 if it has a failure in # a vendor clusterware API. Retry. 11) $LOGMSG "Oracle CLSVMON was unable to interact with vendor clusterware. Retrying" $RMF $NOCLSVMON exit 1; ;; # CLSVMON will return a status code of 13 if it is unable to find # vendor clusterware. Do not retry. # The clsvmon part of init.cssd is not supposed to be run without vendor # clusterware, though, so this should not happen. 13) $LOGMSG "Oracle CLSVMON did not find any vendor clusterware." $TOUCH $NOCLSVMON exit 0; ;; # CLSVMON will return a status code of 14 if it thinks that another # copy of clsvmon is alive. This will generally only happen if the # user has come along and killed off a wrapper script and then duplicates # got spawned. Here we shut down the other clsvmon, exit requesting a respawn, # and start up again. If there is another copy of init.cssd clsvmon hanging # around then it will also exit, but not get respawned. Our respawn will # create a new copy of the daemon, and then be able to properly wait on it. 14) $LOGMSG "Duplicate Oracle CLSVMON found. Killing and restarting it." $EVAL $FINDCLSVMON | $AWK '{ print $2 ; }' | $XARGS $KILLDIE > $NULL 2>&1 $RMF $NOCLSVMON exit 1; ;; *) $LOGMSG "Oracle CLSVMON terminated with unexpected status $RC." $RMF $NOCLSVMON exit 1; ;; esac ;; 'oprocd') # Runs the oprocd daemon. This script is supposed to exit 0 # if it decides that it should not be respawned, and exit 1 if it # wants to be respawned. The purpose of oprocd is to detect system # hangs, which often occur due to faulty drivers or hardware. # On rare occasion this will get re-invoked while a shutdown is # in progress. If so, this script should exit. $ID/init.cssd runcheck if [ "$?" != "0" ]; then exit 0; fi # If we are in fatal mode, then let oprocd know right away. if [ -f $FATALFILE ] && [ `$CAT $FATALFILE` = "enable" ] ; then FATALARG="-f" else FATALARG= fi # Backup the oprocd logfiles if [ -f $OPROCDLOG ] ; then FILENAME=$OPROCDLOG.$UNIQUEDATE $MVF $OPROCDLOG "$FILENAME" fi # Backup the oprocd last gasp files if [ -f $OPROCDLGL ] ; then FILENAME=$OPROCDLGL.$UNIQUEDATE $MVF $OPROCDLGL "$FILENAME" fi # Run oprocd synchronously and look for its status code cd $OPROCDIR # startup the some diagnostic collection scripts if any StartDiagCollect; $OPROCD run -t $OPROCD_DEFAULT_TIMEOUT -m $OPROCD_DEFAULT_MARGIN \ $OPROCD_DEFAULT_HISTOGRAM $FATALARG RC=$? # shutdown diagnostic collection StopDiagCollect; # Oprocd will return a status code of 0 if a graceful shutdown has been # requested. In this case, we do not want to be respawned. if [ "$RC" = "0" ]; then $LOGMSG "Oprocd received graceful shutdown request. Shutting down." $TOUCH $NOOPROCD exit 0; fi # Oprocd will return a status code of 1 if it notices that it is a # duplicate copy of the daemon. This will generally only happen if the # user has come along and killed off a wrapper script and then duplicates # got spawned. Here we shut down the other oprocd, exit requesting a respawn, # and start up again. If there is another copy of init.cssd oprocd hanging # around then it will also exit, but not get respawned. Our respawn will # create a new copy of the daemon, and then be able to properly wait on it. if [ "$RC" = "1" ]; then PROCDDUP=`$EVAL $FINDPROCD` if [ ! -z "$PROCDDUP" ]; then $OPROCD stop -t $OPROCD_STOP_TIMEOUT 2>$NULL $LOGMSG "Duplicate spawn of Oprocd. See init.cssd for description." else $LOGMSG "WARNING: Oprocd exited unexpectedly. Check oprocd binary/log and restart the stack" fi $RMF $NOOPROCD exit 1; fi # Oprocd will return a status code of 2 if it has noticed a system hang # while in nonfatal mode. In this case we need to respawn it, so that if # CSS starts allowing database IO that we have fencing guarantees. if [ "$RC" = "2" ]; then $LOGMSG "Oprocd detected system hang while in nonfatal mode. Respawning." $RMF $NOOPROCD exit 1; fi # If it is not one of the above, then request a respawn. Delay a bit # in case this is a recurring failure, to avoid spinning. $LOGMSG "Oprocd terminated with unexpected status $RC. Respawning." $RMF $NOOPROCD exit 1; ;; 'diskmon') # Runs the diskmon daemon. This script is supposed to exit 0 # if it decides that it should not be respawned, and exit 1 if it # wants to be respawned. The purpose of diskmon is to issue SAGE # fence requests # On rare occasion this will get re-invoked while a shutdown is # in progress. If so, this script should exit. $ID/init.cssd runcheck if [ "$?" != "0" ]; then exit 0; fi # If a diskmon is up, we want to kill it so that the new execution will # be the only one up $EVAL $FINDDISKMON RC=$? if [ $RC -eq 0 ]; then $LOGMSG "Killing diskmon from previous startup" $EVAL $FINDDISKMON | $AWK '{ print $2 ; }' | $XARGS $KILLDIE > $NULL 2>&1 fi # Tracing and force takeover of local endpt args by default DISKMON_ARGS='-d -f' # Run diskmon synchronously and look for its status code $DISKMON $DISKMON_ARGS RC=$? # Diskmon will return a status code of 0 if a graceful shutdown has been # requested. In this case, we do not want to be respawned. if [ "$RC" = "0" ]; then $LOGMSG "Diskmon shutdown gracefully." $TOUCH $NODISKMON else $LOGMSG "Diskmon terminated with unexpected status $RC. Respawning." $RMF $NODISKMON fi exit $RC; ;; 'stop') $LOGMSG "Oracle CSSD being stopped" # disable CSS startup until the next boot $ID/init.cssd norun # shutdown the OPROCD process if it is running if [ ! -f $NOOPROCD ]; then $OPROCD stop -t $OPROCD_STOP_TIMEOUT 2>$NULL fi # No steps are necessary for shutting down clsomon. It will go down # automatically when CSS is shutdown. # Shut down oclsvmon if it is up. if [ ! -f $NOCLSVMON ]; then $EVAL $FINDCLSVMON | $AWK '{ print $2 ; }' | $XARGS $KILLTERM > $NULL 2>&1 fi # shutdown diskmon if [ ! -f $NODISKMON ]; then DISKMON_STOPARGS='-s' $DISKMON $DISKMON_STOPARGS fi # Invalidate init.cssd fatal pidfiles. $ECHO "stopped" > $CSSFBOOT $TOUCH $NOOPROCD $TOUCH $NOCLSVMON $TOUCH $NOCLSOMON $TOUCH $NODISKMON # Remove SUBSYSFILE on Linux if [ "$PLATFORM" = "Linux" ]; then $RMF $SUBSYSFILE fi # Now tell it to shut down. if [ -x "$CRSCTL" ]; then $CRSCTL stop crs fi $ECHO "Shutdown has begun. The daemons should exit soon." ;; 'run') # Foreground run, for single instance or single-node installs only. # If this is used in a cluster install, RDBMS datafile corruption is # likely. # Run the startcheck to see whether we should continue $ID/init.cssd startcheck while [ "$?" != "0" ]; do $SLEEP $RUNRECHECKTIME $ID/init.cssd startcheck done cd $ORA_CRS_HOME/log/$HOST/cssd # If there is an old corefile by such a collision prone name, then # rename it to something safe. if [ -f ./core ]; then $MVF ./core "$UNIQUECORE" fi $ECHO $USING_DISKMON > $DISKMONUSED $CHOWN $ORACLE_USER $DISKMONUSED STARTUP_OK=1 if [ "$USING_DISKMON" = "false" ]; then $TOUCH $NODISKMON else # Use diskmon $RMF $NODISKMON STARTUP_OK=0 # Set cluster GUID if [ -x "$CRSCTL" ]; then STARTUP_OK=1 $CRSCTL get css clusterguid > /dev/null 2>&1 RC=$? if [ $RC -ne 0 ]; then GUID=`$CRSCTL set css clusterguid -force` RC=$? # The cluster GUID is required for SAGE, so do not start if # unable to set cluster GUID if [ $RC -ne 0 ]; then STARTUP_OK=0 $LOGERR "Unable to set cluster GUID, cannot start clusterware" fi fi fi fi # If we are unable to start, exit here if [ $STARTUP_OK -eq 0 ]; then exit 1 fi $TOUCH $NOOPROCD # Never use OPROCD in local-only $TOUCH $NOCLSVMON $ECHO "disable" > $FATALFILE # disable reboot # Start the CSSD $ID/init.cssd daemon & # We need to loop here to restart diskmon if it fails while $TRUE; do # Poll every second to see which of our children has gone away. $SLEEP 1 # Check to see whether shutdown has been invoked. if [ ! -r "$RUNFILE" ]; then $LOGMSG "Filesystem containing $RUNFILE vanished." $LOGMSG "Unpredictable behavior from Oracle CRS may ensue." $SLEEP 30 continue; fi RUNID=`$CAT $RUNFILE` if [ "norun" = "$RUNID" ]; then # Shutdown was clearly invoked. Go away cleanly. $LOGMSG "Oracle CSS family monitor shutting down gracefully." # Invalidate pidfiles. $ECHO "stopped" > $CSSFBOOT exit 0; fi # Did the diskmon script go away? if [ ! -f $NODISKMON ]; then $KILL -0 $DISKMONSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $DISKMONSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $ID/init.cssd diskmon & DISKMONSCRIPTPID=$! $ECHO $DISKMONSCRIPTPID > $DISKMONPIDFILE fi fi fi done ;; 'fatal') # This action is invoked to start the CSS daemon in cluster mode, # and one or more of its accompanying daemons oprocd or clsvmon or clsomon # This respawn wrapper is done in lieu of adding new entries to inittab. # Check to see if we are supposed to run this boot. $ID/init.cssd startcheck while [ "$?" != "0" ]; do $SLEEP $RUNRECHECKTIME $ID/init.cssd startcheck done # See discussion in LocalFence $EVAL $CLEANREBOOTLOCK # Determine which co-daemons we should be using on this system. $TOUCH $NOCLSVMON $TOUCH $NOCLSOMON $TOUCH $NOOPROCD $TOUCH $NODISKMON # exit if we failed to create the nooprocd flag file if [ "$?" != "0" ]; then $LOGMSG "Failed to create $NOOPROCD. Exiting..." exit 1; fi # if init.cssd fatal is respawning, then read in the children # to monitor. CSSFID=`$CAT $CSSFBOOT` BOOTID=`$ID/init.cssd bootid` if [ "$CSSFID" = "$BOOTID" ]; then OPROCDSCRIPTPID=`$CAT $OPROCDPIDFILE` CLSOMONSCRIPTPID=`$CAT $CLSOMONPIDFILE` CLSVMONSCRIPTPID=`$CAT $CLSVMONPIDFILE` DAEMONSCRIPTPID=`$CAT $DAEMONPIDFILE` $LOGMSG "Oracle CSS Family monitor restarting." else $LOGMSG "Oracle CSS Family monitor starting." fi $ECHO "$BOOTID" > $CSSFBOOT # Set some default pids so that the main loop will # spawn the daemons on the first pass through. if [ -z "$OPROCDSCRIPTPID" ]; then OPROCDSCRIPTPID=nopid fi if [ -z "$DAEMONSCRIPTPID" ]; then DAEMONSCRIPTPID=nopid fi if [ -z "$CLSVMONSCRIPTPID" ]; then CLSVMONSCRIPTPID=nopid fi if [ -z "$CLSOMONSCRIPTPID" ]; then CLSOMONSCRIPTPID=nopid fi # Only setup environment for OPROCD if # 1) it is Not DISABLED for this platform # 2) the daemon is installed # 3) and vendor clusterware is not provided if [ "$DISABLE_OPROCD" = "false" ] && [ -x "$OPROCD" ] && [ ! -f "$SKGXNLIB" ]; then $RMF $NOOPROCD fi # Only use clsvmon if vendor clusterware is present. if [ $USING_VC -eq 1 ]; then $RMF $NOCLSVMON fi # Always use clsomon. $RMF $NOCLSOMON # Use diskmon if configured $ECHO $USING_DISKMON > $DISKMONUSED $CHOWN $ORACLE_USER $DISKMONUSED STARTUP_OK=1 if [ "$USING_DISKMON" = "false" ]; then $TOUCH $NODISKMON else # Use diskmon $RMF $NODISKMON fi while $TRUE; do # Poll every second to see which of our children has gone away. $SLEEP 1 # Check to see whether shutdown has been invoked. if [ ! -r "$RUNFILE" ]; then $LOGMSG "Filesystem containing $RUNFILE vanished." $LOGMSG "Unpredictable behavior from Oracle CRS may ensue." $SLEEP 30 continue; fi RUNID=`$CAT $RUNFILE` if [ "norun" = "$RUNID" ]; then # Shutdown was clearly invoked. Go away cleanly. $LOGMSG "Oracle CSS family monitor shutting down gracefully." # Invalidate pidfiles. $ECHO "stopped" > $CSSFBOOT exit 0; fi # Did the oprocd script go away? if [ ! -f $NOOPROCD ]; then $KILL -0 $OPROCDSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $OPROCDSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $ID/init.cssd oprocd & OPROCDSCRIPTPID=$! $ECHO $OPROCDSCRIPTPID > $OPROCDPIDFILE fi fi fi # Did the clsvmon script go away? if [ ! -f $NOCLSVMON ]; then $KILL -0 $CLSVMONSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $CLSVMONSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $ID/init.cssd oclsvmon & CLSVMONSCRIPTPID=$! $ECHO $CLSVMONSCRIPTPID > $CLSVMONPIDFILE fi fi fi # Did the clsomon script go away? if [ ! -f $NOCLSOMON ]; then $KILL -0 $CLSOMONSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $CLSOMONSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $ID/init.cssd oclsomon & CLSOMONSCRIPTPID=$! $ECHO $CLSOMONSCRIPTPID > $CLSOMONPIDFILE fi fi fi # Did the diskmon script go away? if [ ! -f $NODISKMON ]; then $KILL -0 $DISKMONSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $DISKMONSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $ID/init.cssd diskmon & DISKMONSCRIPTPID=$! $ECHO $DISKMONSCRIPTPID > $DISKMONPIDFILE fi fi fi # Did the daemon script go away? Respawning here is not # an intention to restart CSS. It allows the system of scripts # to distinguish between several failure and shutdown scenarios. $KILL -0 $DAEMONSCRIPTPID > $NULL 2>&1 if [ "$?" != "0" ]; then $EVAL $KILLCHECK > $NULL 2>&1 KC=$? $KILL -0 $DAEMONSCRIPTPID > $NULL 2>&1 RK=$? if [ "$KC" = "0" -a "$RK" != "0" ]; then $LOGMSG "Oracle CSS restart. $KC, $RK" $ID/init.cssd daemon & DAEMONSCRIPTPID=$! $ECHO $DAEMONSCRIPTPID > $DAEMONPIDFILE fi fi done # We will never reach this exit 0; ;; 'daemon') # Run the CSS daemon in cluster mode. This will enable the fencing # functionality not required in single-node mode. The CSS daemon # monitors node health, and provides endpoint discovery for RDBMS. # On rare occasion this will get re-invoked while a shutdown is # in progress. If so, this script should exit. $ID/init.cssd runcheck if [ "$?" != "0" ]; then exit 0; fi # Check to see if CSS is already up. If so, this indicates that # someone has killed the "init.cssd daemon" script that was parenting # that CSS daemon. This is a bad thing since we can't reattach a # new parent to the CSS daemon. CSSDUP=`$EVAL $FINDCSSD` if [ ! -z "$CSSDUP" ]; then $LOGMSG "Oracle CSSD shell script failure. Duplicate CSSD." $EVAL $SLOW_REBOOT $SLEEP 300 exit 0; fi # If we can set ourselves into realtime, do so, otherwise renice # ourselves, and the child scripts. This is useful to ensure # that we can perform the fence action if CSS kills itself due to # unreasonable load. This script will remain asleep most of the time # so a high priority shouldn't bother anything. if [ "$REALTIME_PID" != "" ]; then case $PLATFORM in HP-UX) $REALTIME_PID -$$ ;; *) $REALTIME_PID $$ ;; esac else $RENICE -5 $$ fi # Wait for oprocd to come up. if [ ! -f $NOOPROCD ]; then $OPROCD check -t $OPROCD_CHECK_TIMEOUT > $NULL 2>&1 RC=$? while [ "$RC" != "0" ] && [ "$RC" != "1" ]; do $LOGMSG "Oracle CSSD waiting for OPROCD to start" $SLEEP 5 $OPROCD check -t $OPROCD_CHECK_TIMEOUT > $NULL 2>&1 RC=$? done fi cd $ORA_CRS_HOME/log/$HOST/cssd # If there is an old corefile by such a collision prone name, then # rename it to something safe. if [ -f ./core ]; then $MVF ./core "$UNIQUECORE" fi # Arguments. By default none. OCSSD_ARGS= # Invoke the CSS daemon. Since SU can lose a few important settings, it # is necessary to perform them inside the subshell. case $PLATFORM in AIX) OCSSDDIR=$ORA_CRS_HOME/log/$HOST/cssd $SU $ORACLE_USER -c \ "/bin/sh -c '$ULIMIT_CORE; ulimit -n $MAXFILE; cd $OCSSDDIR; $ORA_CRS_HOME/bin/ocssd $OCSSD_ARGS || exit \$?'" RC=$? ;; *) $ULIMIT_CORE ulimit -n $MAXFILE $ORA_CRS_HOME/bin/ocssd $OCSSD_ARGS RC=$? ;; esac # Three ways of going down. Graceful, Failing to Initialize, and # Steady-state Failure. case "$RC" in 0) $LOGMSG "Oracle CSSD graceful shutdown" ;; 100) $LOGMSG "Oracle CSS daemon failed to start up. Check CRS logs for diagnostics." ;; 110) $LOGMSG "Oracle CSS daemon failed in non-fatal mode" ;; *) # Catch all remaining un-graceful return codes and perform the reboot # action here. $LOGERR "Oracle CSSD failure $RC." LocalFence; esac # Shut down OPROCD and OCLSVMON. This gets shut down earlier if the # user ran init.crs stop, but if they are using crsctl this was skipped. if [ ! -f $NOOPROCD ]; then $OPROCD stop 2>$NULL fi if [ ! -f $NOCLSVMON ]; then $EVAL $FINDCLSVMON | $AWK '{ print $2 ; }' | $XARGS $KILLTERM > $NULL 2>&1 fi # shutdown diskmon if [ ! -f $NODISKMON ]; then DISKMON_STOPARGS='-s' $DISKMON $DISKMON_STOPARGS fi # Disable the CSS daemon from getting restarted until the next # user intiated startup, or the next reboot cycle. $ID/init.cssd norun exit 0; ;; 'activatevg') # Do nothing for now exit 0; ;; 'start_msg') # start message if we are being invoked as an RC script (only on HP-UX) $ECHO "Start Oracle Cluster Synchronization Service" ;; 'stop_msg') # stop message if we are being invoked as an RC script (only on HP-UX) $ECHO "Stop Oracle Cluster Synchronization Service" ;; *) $ECHO "See documentation at the top of $0 about supported commands." exit 1; ;; esac