monitoring/OLD/check_dsl.sh
2017-02-21 01:19:22 +01:00

325 lines
13 KiB
Bash
Executable File

#!/usr/bin/env bash
## --------------------------
## --- Make your changes here
## --------------------------
LOGGING_CONSOLE=false
DEBUG=false
log_file=/var/log/check_dsl.log
## - Put in your DSL devices (refers to your network configuration)
## - youe wish be congigured by that script
## -
DSL_DEVICES="ppp-surf2 ppp-surf1 ppp-st"
admin_email=root
from_address="check-dsl@`hostname -f`"
company="Gemeinschaft Altenschlirf"
content_type='Content-Type: text/plain;\n charset="utf-8"'
##########################################
## --- Dont't make changes after this line
##########################################
## ------------------
## --- Some functions
## ------------------
## - Check if a given array (parameter 2) contains a given string (parameter 1)
## -
containsElement () {
local e
for e in "${@:2}"; do [[ "$e" == "$1" ]] && return 0; done
return 1
}
## -------------------------------------------------
## --- If script is already running, stop execution
## -------------------------------------------------
LOCK_DIR=/tmp/`basename $0`.LOCK
if mkdir "$LOCK_DIR" 2> /dev/null ; then
## - Remove lockdir when the script finishes, or when it receives a signal
trap 'rm -rf "$LOCK_DIR"' 0 2 15
else
datum=`date +"%d.%m.%Y"`
msg="[ Error ]: A previos instance of script \"`basename $0`\" seems already be running.\n\n Exiting now.."
echo -e "To:${admin_email}\n${content_type}\nSubject:DSL Script Error $company -- $datum\n\n${msg}\n" | /usr/sbin/sendmail -F "DSL Monitoring $company" -f $from_address $admin_email
if $LOGGING_CONSOLE ; then
echo ""
echo "[ Error ]: A previos instance script \"`basename $0`\" seems already be running."
echo ""
echo " Exiting now.."
echo ""
fi
exit 1
fi
## - Define (non associative) array
## -
declare -a dsl_devices_arr
for _dsl_device in $DSL_DEVICES ; do
dsl_devices_arr+=("$_dsl_device")
done
declare -a online_devices_arr
for _online_device in "$@" ; do
online_devices_arr+=("$_online_device")
done
## - Define associative array
## -
declare -A remote_gw_arr
declare -A filetime_PID_arr
for dsl_device in "${online_devices_arr[@]}" ; do
remote_gw_address=`ifconfig $dsl_device 2> /dev/null | grep "inet Adresse" | cut -d":" -f3 | cut -d" " -f1`
remote_gw_arr[$dsl_device]=$remote_gw_address
_pid_file=/var/run/${dsl_device}.pid
if [ -f $_pid_file ]; then
filetime_PID_arr[$dsl_device]=`stat -c %Y /var/run/${dsl_device}.pid`
else
filetime_PID_arr[$dsl_device]="NOT FOUND"
fi
done
ping_test_hosts="oopen.de google.com heise.de debian.org ubuntu.com"
declare -a ping_ip_arr;
#for _key in "${!filetime_PID_arr[@]}" ; do
# echo "filetime_PID_arr[$_key]: ${filetime_PID_arr[$_key]}"
#done
#for _key in "${!remote_gw_arr[@]}" ; do
# echo "remote_gw_arr[$_key]: ${remote_gw_arr[$_key]}"
#done
#for _device in ${online_devices_arr[@]} ; do
# echo "$_device"
#done
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] ## ---" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] ## --- Starting script `basename $0`" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] ## --- Devices Online...: ${online_devices_arr[@]}" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] ## --- Remote Gateways..: ${remote_gw_arr[@]}" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] ## ---" >> $log_file
while true ; do
changed=false
for dsl_device in "${dsl_devices_arr[@]}" ; do
if $DEBUG ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] ## --- Device $dsl_device " >> $log_file
fi
remote_gw_address=`ifconfig $dsl_device 2> /dev/null | grep "inet Adresse" | cut -d":" -f3 | cut -d" " -f1`
iface_name="dsl-`echo $dsl_device | cut -d '-' -f2`"
rt_name="dsl_`echo $dsl_device | cut -d '-' -f2`"
if [ -z "$remote_gw_address" ]; then
if containsElement "$dsl_device" ${online_devices_arr[@]} ; then
## - <before> <now>
## -
## - online offline
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Status Devices \"$dsl_device\" changed" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Devices \"$dsl_device\" is now OFFLINE" >> $log_file
if ps -x | grep -E "/usr/sbin/pppd\s+call\s+$iface_name" > /dev/null 2>&1 ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] pppd for \"$iface_name\" is running: Waiting another period" >> $log_file
else
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] Trying to start pppd for \"$dsl_device\".." >> $log_file
/usr/sbin/pppd call $iface_name > /dev/null 2>&1
fi
## - Remove device from array online_devices_arr
## -
for _index in ${!online_devices_arr[@]} ; do
if [ "${online_devices_arr[$_index]}" = "$dsl_device" ]; then
unset online_devices_arr[$_index]
break
fi
done
## - Also remove device from remote_gw_arr
## -
unset remote_gw_arr[$dsl_device]
changed=true
else
## - <before> <now>
## -
## - offline offline
if $DEBUG ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] $dsl_device is still offline" >> $log_file
fi
if ps -x | grep -E "/usr/sbin/pppd\s+call\s+$iface_name" > /dev/null 2>&1 ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] pppd for \"$iface_name\" is running: Waiting another period" >> $log_file
else
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] Trying to start pppd for \"$dsl_device\".." >> $log_file
/usr/sbin/pppd call $iface_name > /dev/null 2>&1
fi
fi
else
if containsElement "$dsl_device" ${online_devices_arr[@]} ; then
## - <before> <now>
## -
## - online online
if $DEBUG ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] $dsl_device is still online" >> $log_file
fi
## - Check if remote gateway has changed
## -
if [ "$remote_gw_address" != "${remote_gw_arr[$dsl_device]}" ]; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Remote Gateway on device \"$dsl_device\" has changed: ${remote_gw_arr[$dsl_device]} --> $remote_gw_address" >> $log_file
remote_gw_arr[$dsl_device]=$remote_gw_address
filetime_PID_arr[$dsl_device]=`stat -c %Y /var/run/${dsl_device}.pid`
changed=true
else
if $DEBUG ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] Remote Gateway on device \"$dsl_device\": still ${remote_gw_arr[$dsl_device]}" >> $log_file
fi
## - Test if pid-file's modify time hs changed
## -
## - Notice: that happens if your provider forces a reconnect (mostly one time a day
## - or in other words after 1440 minutes)
## -
_pid_file=/var/run/${dsl_device}.pid
if [ -f $_pid_file ]; then
if [ "`stat -c %Y $_pid_file`" != "${filetime_PID_arr[$dsl_device]}" ]; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Modify time for pid-file \"${dsl_device}.pid\" has changed" >> $log_file
filetime_PID_arr[$dsl_device]=`stat -c %Y $_pid_file`
changed=true
else
## - Check if routing through this dsl connection realy works
## -
failed=true
if [ ${#ping_ip_arr[@]} -lt 1 ]; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] No ip-address for ping-test is set. Skipping test.." >> $log_file
else
for _key in ${!ping_ip_arr[@]} ; do
/sbin/ip rule add to ${ping_ip_arr[$_key]} table $rt_name
if ping -q -c2 ${ping_ip_arr[$_key]} >/dev/null 2>&1 ; then
if $DEBUG ; then
_local_gw=`curl -4 https://meine-ip.oopen.de 2> /dev/null`
if [ -n "$_local_gw" ]; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] Using local gateway \"$_local_gw\" for ping test" >> $log_file
fi
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Debug ] Ping test (to ${ping_ip_arr[$_key]}) for device \"${dsl_device}\" was successful." >> $log_file
fi
/sbin/ip rule del to ${ping_ip_arr[$_key]} table $rt_name
failed=false
break
fi
/sbin/ip rule del to ${ping_ip_arr[$_key]} table $rt_name
done
if $failed ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Ping test for device \"${dsl_device}\" failed" >> $log_file
i echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Status Devices \"$dsl_device\" changed" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Devices \"$dsl_device\" is now OFFLINE" >> $log_file
_pid=`ps -ax | grep -e "pppd call $iface_name" | grep -v grep | awk '{print$1}'`
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Info ] Kill ppp-daemon for $iface_name (pid $_pid)" >> $log_file
## - Remove device from array online_devices_arr
## -
for _index in ${!online_devices_arr[@]} ; do
if [ "${online_devices_arr[$_index]}" = "$dsl_device" ]; then
unset online_devices_arr[$_index]
break
fi
done
## - Also remove device from remote_gw_arr
## -
unset remote_gw_arr[$dsl_device]
kill -9 $_pid
changed=true
fi
fi
fi
else
filetime_PID_arr[$dsl_device]="NOT FOUND"
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] No pid-file for connection \"${dsl_device}\"found" >> $log_file
changed=true
fi
fi
else
## - <before> <now>
## -
## - offline online
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Status Devices \"$dsl_device\" changed" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Devices \"$dsl_device\" is now online" >> $log_file
## - Add device to array online_devices_arr
## -
online_devices_arr+=("$dsl_device")
remote_gw_arr[$dsl_device]=$remote_gw_address
filetime_PID_arr[$dsl_device]=`stat -c %Y /var/run/${dsl_device}.pid`
changed=true
fi
fi
done
if $changed ; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Status Online Devices changed" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Online Devices: ${online_devices_arr[@]}" >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Warn ] Reconfigure Routing: invoking script \"netconfig_dsl.sh\".." >> $log_file
/root/bin/netconfig_dsl.sh > /dev/null 2>&1
#/root/bin/netconfig_dsl_round-robin.sh > /dev/null 2>&1
datum=`date +"%d.%m.%Y"`
msg="[ `date +\"%H:%M:%S\"` ]: Status Online Devices changed..\n Online Devices: ${online_devices_arr[@]}\n\n Script \"netconfig_dsl.sh\" was invoked to reconfigure routing."
echo -e "To:${admin_email}\n${content_type}\nSubject:DSL Status changed $company -- $datum\n\n${msg}\n" | /usr/sbin/sendmail -F "DSL Monitoring $company" -f $from_address $admin_email
fi
## - Set IP-adresses for Ping-Test at next run
## -
if [ ${#online_devices_arr[@]} -gt 0 ]; then
ping_test_ip=""
unset ping_ip_arr
declare -i i=0
for _host in $ping_test_hosts ; do
while [ $i -lt 2 ]; do
ping_test_ip=`dig +short $_host | head -1`
if ping -q -c2 $ping_test_ip >/dev/null 2>&1 ; then
ping_ip_arr+=("$ping_test_ip")
let i++
break
fi
done
done
if [ ${#ping_ip_arr[@]} -lt 1 ]; then
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] No ip-address for ping test found " >> $log_file
echo "`date +'%Y-%m-%d %H:%M:%S'` [ Error ] Looked for IP-Addresses at hosts: $ping_test_hosts" >> $log_file
fi
fi
sleep 30
done
exit 0