#!/bin/sh # # Copyright (c) 2008, 2012, Oracle and/or its affiliates. All rights reserved. # # SNDR start script # # Description: This is the SNDR start script. It must be located # in /etc/init.d with links to the appropriate rc2.d and # rc0.d files. # It can also be used to start or stop a specified cluster # resource group when invoked from the data service cluster # failover script. # # # PATH=/etc:/bin RDCBOOT="/usr/sbin/sndrboot" USAGE="Usage: $0 {start|stop} [cluster_resource]" SVCS=/usr/bin/svcs DSCFG_DEPEND_NOCHK="/tmp/.dscfgadm_pid" OS_MINOR=`/usr/bin/uname -r | /usr/bin/cut -d '.' -f2` . /lib/svc/share/smf_include.sh # Make sure prior SMF dependents are not 'online' # $1 = name of SMF service to validate dependents # do_smf_depends () { times=0 count=1 if [ $OS_MINOR -ge 11 ] then return 0 elif [ -f $DSCFG_DEPEND_NOCHK ] then for pid in `pgrep dscfgadm` do if [ `grep -c $pid $DSCFG_DEPEND_NOCHK` -gt 0 ] then return 0 fi done elif [ `ps -ef | grep preremove | grep -c SUNWrdcu` -gt 0 ] then return 0 fi while [ $count -ne 0 ] do count=`$SVCS -o STATE -D $1 2>>/dev/null | grep "^online" | wc -l` if [ $count -ne 0 ] then # Output banner after waiting first 5 seconds # if [ $times -eq 1 ] then echo "Waiting for $1 dependents to be 'offline'" $SVCS -D $1 2>>/dev/null | grep "^online" fi # Has it been longer then 5 minutes? (60 * 5 secs.) # if [ $times -eq 60 ] then echo "Error: Failed waiting for $1 dependents to be 'offline'" $SVCS -D $1 2>>/dev/null | grep "^online" exit $SMF_EXIT_ERR_FATAL fi # Now sleep, giving other services time to stop # sleep 5 times=`expr $times + 1` fi done return 0 } CLINFO=/usr/sbin/clinfo killproc() { # kill the named process(es) pid=`/usr/bin/ps -e | /usr/bin/grep -w $1 | /usr/bin/sed -e 's/^ *//' -e 's/ .*//'` [ "$pid" != "" ] && kill $pid } # There is a cluster and default instance of this service. # We only want to run this script once when the cluster mode # and service ($SMF_FMRI) match. # Check if in cluster (cl_mode = 0) or not (cl_mode=1) # and do the following on start: # cl_mode=0 and nws_rdc:cluster, continue # cl_mode=0 and nws_rdc:default, exit # cl_mode=1 and nws_rdc:default, continue # cl_mode=1 and nws_rdc:cluster, disable nws_rdc:cluster to prevent # dependency problems and exit # # on stop, just exit if cluster mode and service do not match # if [ -x "$CLINFO" ] then ${CLINFO} cl_mode=$? else cl_mode=1 fi case "$1" in 'start') if [ $cl_mode -eq 1 -a "$SMF_FMRI" = svc:/system/nws_rdc:cluster ] then smf_method_exit $SMF_EXIT_TEMP_DISABLE cluster_mode \ "$SMF_FMRI is not supported in cluster mode" fi if [ $cl_mode -eq 0 -a "$SMF_FMRI" = svc:/system/nws_rdc:default ] then exit $SMF_EXIT_OK fi COPT= if [ -x ${RDCBOOT} ] then if ${CLINFO} then if [ "$2" != "" ] then ${RDCBOOT} -r -C $2 else # SNDR 3.2 SetIDs fixup ${RDCBOOT} -C post-patch-setids -r -s COPT="-C -" ${RDCBOOT} ${COPT} -r fi else # non-clustered start ${RDCBOOT} -r fi fi ;; 'stop') if [ $cl_mode -eq 0 -a "$SMF_FMRI" = svc:/system/nws_rdc:cluster ] || [ $cl_mode -eq 1 -a "$SMF_FMRI" != svc:/system/nws_rdc:default ] then exit $SMF_EXIT_OK fi COPT= if [ ! -r /dev/rdc ] then RDCBOOT=/usr/bin/true fi do_smf_depends "$SMF_FMRI" if [ -x ${RDCBOOT} ] then if ${CLINFO} then if [ "$2" != "" ] then ${RDCBOOT} -s -C $2 else COPT="-C -" ${RDCBOOT} ${COPT} -s echo "killing SNDR daemons" killproc sndrd killproc sndrsync fi else # non-clustered stop ${RDCBOOT} -s echo "killing SNDR daemons" killproc sndrd killproc sndrsync fi else # no sndr boot command, kill daemon anyway echo "killing SNDR daemons" killproc sndrd killproc sndrsync fi ;; *) echo $USAGE exit 1 ;; esac exit $SMF_EXIT_OK