mirror of
https://github.com/RIPE-NCC/ripe-atlas-software-probe.git
synced 2024-05-11 05:55:02 +00:00
Michel Stam
350 lines
9.1 KiB
Bash
Executable File
350 lines
9.1 KiB
Bash
Executable File
#!/bin/sh
|
|
# The ATLAS Main Loop
|
|
# 2010 atlas@ripe.net
|
|
# Written by Daniel Karrenberg, Antony Antony
|
|
# RIPE NCC
|
|
|
|
#exec >/tmp/ATLAS.out 2>/tmp/ATLAS.err
|
|
#set -x
|
|
|
|
if [ -f bin/config.sh ]
|
|
then
|
|
. bin/config.sh
|
|
export DEVICE_NAME SUB_ARCH ATLAS_STATIC
|
|
else
|
|
echo no 'bin/config.sh' >&2
|
|
exit 1
|
|
fi
|
|
|
|
. $ATLAS_STATIC/bin/arch/$DEVICE_NAME/$DEVICE_NAME-ATLAS.sh
|
|
|
|
. $BIN_DIR/common.sh
|
|
|
|
source $ATLAS_STATIC/bin/support.lib.sh
|
|
source $ATLAS_STATIC/bin/class.lib.sh
|
|
source $ATLAS_STATIC/bin/array.lib.sh
|
|
source $ATLAS_STATIC/bin/json.lib.sh
|
|
source $ATLAS_STATIC/bin/atlas_log.lib.sh
|
|
|
|
# config variables
|
|
if [ $(config_lookup RXTXRPT no) = yes ]
|
|
then
|
|
do_rxtxrpt=yes
|
|
fi
|
|
httppost_port="$(config_lookup HTTP_POST_PORT '')"
|
|
if [ -n "$httppost_port" ]
|
|
then
|
|
export HTTPPOST_PORT="$httppost_port"
|
|
fi
|
|
export httppost_port
|
|
export TELNETD_PORT=$(config_lookup TELNETD_PORT 2023)
|
|
|
|
$MOUNT_FS_CMD
|
|
|
|
mkdir -p $STATUS_DIR $STATE_DIR $RUN_DIR
|
|
mkdir -p $DATA_DIR $DATA_NEW_DIR $DATA_OUT_DIR
|
|
mkdir -p $DATA_DIR/out/ooq
|
|
mkdir -p $DATA_DIR/out/ooq10 # For wifi, can't hurt to just create this
|
|
|
|
$CHOWN_DATA_DIRS
|
|
|
|
if dfrm -A 9018 $DATA_DIR 600 $DATA_NEW_DIR $DATA_OUT_DIR $DATA_DIR/home/atlas/data/new $DATA_DIR/home/atlas/data/out >> $DATA_NEW_DIR/simpleping
|
|
then
|
|
D=`epoch`
|
|
echo "RESULT 9007 done $D $ETHER_SCANNED data full available space < $LOW_DISK_LIMIT KB"
|
|
echo "RESULT 9007 done $D $ETHER_SCANNED data full available space < $LOW_DISK_LIMIT KB" >> $DATA_NEW_DIR/simpleping
|
|
fi
|
|
|
|
$BUDDYINFO $LOW_MEM_T $DATA_NEW_DIR/simpleping
|
|
|
|
/bin/mkdir -p $BASE_DIR/crons/main $BASE_DIR/crons/2
|
|
rm -f $DATA_NEW_DIR/v6addr.vol
|
|
rm -f $DATA_OUT_DIR/v6addr.txt
|
|
rm -f $SSH_ERR
|
|
|
|
# Try to get the date from a previous incarnation
|
|
$LOAD_STORAGE_CURRENT_TIME
|
|
$SET_DATE_FROM_CURRENTTIME_TXT
|
|
|
|
# Source system specific initializing files
|
|
if [ -f $BIN_DIR/passwdset ] ; then
|
|
. $BIN_DIR/passwdset
|
|
rm -f $BIN_DIR/passwdset
|
|
$AFTER_PASSWDSET
|
|
fi
|
|
|
|
if [ -f $BIN_DIR/rc.local ] ; then
|
|
. $BIN_DIR/rc.local
|
|
fi
|
|
|
|
$TRY_UPGRADE_CMD
|
|
|
|
echo "P_TO_C_REPORT" > $STATUS_DIR/p_to_c_report_header
|
|
|
|
mkdir -p $SSH_DIR
|
|
cp $KNOWN_HOSTS_REG $SSH_DIR/known_hosts
|
|
if [ -f $STATUS_DIR/known_hosts_controllers ]
|
|
then
|
|
cat $STATUS_DIR/known_hosts_controllers >> $SSH_DIR/known_hosts
|
|
fi
|
|
|
|
$SETUP_NETWORK_CMD
|
|
$SET_HOSTNAME $ETHER_ADDR
|
|
|
|
# Kill deamons that may be still running before deleting the pid files
|
|
$KILL_TELNETD_CMD
|
|
$KILL_PERDS_CMD
|
|
|
|
# the system part is done. now set env for ATLAS.
|
|
rm -f $STATUS_DIR/*.vol
|
|
rm -f $RUN_DIR/*.vol
|
|
|
|
# Crontabs may be bad enough that we can't actually do anything. Delete them
|
|
# after too many failed boots
|
|
if [ -f $DATA_NEW_DIR/reboot-count.txt ]
|
|
then
|
|
a=`cat $DATA_NEW_DIR/reboot-count.txt`
|
|
b=`expr $a + 1`
|
|
echo $b > $DATA_NEW_DIR/reboot-count.txt
|
|
echo Reboot count is now $b
|
|
if [ $b -gt 10 ]
|
|
then
|
|
echo Removing crontabs
|
|
rm -f $BASE_DIR/crons/main/root
|
|
rm -f $BASE_DIR/crons/*/root
|
|
fi
|
|
else
|
|
echo 0 > $DATA_NEW_DIR/reboot-count.txt
|
|
fi
|
|
|
|
D=`epoch`
|
|
MSG="RESULT 9000 done $D $ETHER_SCANNED STARTING ATLAS system initialized (reboot count `cat $DATA_NEW_DIR/reboot-count.txt`)"
|
|
echo "$MSG" >> $DATA_NEW_DIR/simpleping
|
|
echo "$MSG"
|
|
|
|
echo "RESULT 9000 done $D $ETHER_SCANNED STARTING TELNETD LOCALLY"
|
|
echo "RESULT 9000 done $D $ETHER_SCANNED STARTING TELNETD LOCALLY" >> $DATA_NEW_DIR/simpleping
|
|
telnetd -b 127.0.0.1 -p $TELNETD_PORT -P $RUN_DIR/telnetd-port$TELNETD_PORT-pid.vol &
|
|
|
|
$BUDDYINFO $LOW_MEM_T $DATA_NEW_DIR/simpleping
|
|
|
|
ifconfig
|
|
|
|
mkdir -p $BASE_DIR/crons/main
|
|
mkdir -p $BASE_DIR/crons/2 $BASE_DIR/crons/3 $BASE_DIR/crons/4
|
|
mkdir -p $BASE_DIR/crons/5 $BASE_DIR/crons/6 $BASE_DIR/crons/7
|
|
mkdir -p $BASE_DIR/crons/8 $BASE_DIR/crons/9 $BASE_DIR/crons/10
|
|
mkdir -p $BASE_DIR/crons/11 $BASE_DIR/crons/12 $BASE_DIR/crons/13
|
|
mkdir -p $BASE_DIR/crons/14 $BASE_DIR/crons/15 $BASE_DIR/crons/16
|
|
mkdir -p $BASE_DIR/crons/17 $BASE_DIR/crons/18 $BASE_DIR/crons/19
|
|
mkdir -p $BASE_DIR/crons/20
|
|
mkdir -p $DATA_DIR/oneoff
|
|
|
|
$CHOWN_FOR_MSM
|
|
$CHMOD_FOR_MSM
|
|
|
|
start_perd()
|
|
{
|
|
perd -c $BASE_DIR/crons/main -A 9801 -P $RUN_DIR/perd-main.pid.vol
|
|
}
|
|
start_eperd()
|
|
{
|
|
eperd -c $BASE_DIR/crons/7 -A 9807 -P $RUN_DIR/perd-7.pid.vol -O /home/atlas/data/new/7 -i 7
|
|
}
|
|
start_eooqd()
|
|
{
|
|
eooqd $BASE_DIR/crons/oneoff -A 9809 -P $RUN_DIR/eooqd.pid.vol -i 9 &
|
|
}
|
|
|
|
start_perd
|
|
start_eperd
|
|
start_eooqd
|
|
|
|
|
|
$NTPCLIENT_CMD &
|
|
|
|
if [ -n "$RESOLVCONF_CMD" ]
|
|
then
|
|
if [ -n "$IPV6_INF" ]
|
|
then
|
|
rptra6_inf="-I $IPV6_INF"
|
|
else
|
|
rptra6_inf=''
|
|
fi
|
|
rptra6 -P $RUN_DIR/rptra6.pid.vol -r $DATA_NEW_DIR/resolv.conf.ra6 $rptra6_inf &
|
|
$RESOLVCONF_CMD
|
|
fi
|
|
|
|
while :
|
|
do
|
|
$CHECK_RO_USB
|
|
|
|
if [ $DHCP = "Temp" ] ; then
|
|
#static config failed started DHCP tempeorly
|
|
evping_no_check -4 -c 2 $IPV4_GW
|
|
ARP=`arp -n $IPV4_GW`
|
|
set $ARP
|
|
MAC=$4
|
|
|
|
# Use dynamic resolvers until we switch to static
|
|
# addresses
|
|
if [ -n "$RESOLVCONF_CMD" ]
|
|
then
|
|
$RESOLVCONF_CMD "" dynamic
|
|
fi
|
|
|
|
if [ -n $MAC ] ; then
|
|
if [ $MAC != '<incomplete>' ] ; then
|
|
$KILL_DHCPC_CMD
|
|
/sbin/route del default
|
|
echo "RESULT 9100 done $D DEFAULT GW complete stop DHCP" > $DATA_NEW_DIR/simpleping
|
|
echo "RESULT 9100 done $D DEFAULT GW complete stop DHCP"
|
|
. $NETCONFIG_V4_DEST
|
|
/sbin/ifconfig $LANINF 0.0.0.0
|
|
DHCP=False
|
|
if [ -n "$RESOLVCONF_CMD" ]
|
|
then
|
|
# Regenerate resolv.conf
|
|
$RESOLVCONF_CMD
|
|
else
|
|
cp $RESOLV_CONF_STATIC $RESOLV_CONF
|
|
fi
|
|
fi
|
|
fi
|
|
fi
|
|
|
|
# Check if ssh is still running (when we expect it to be running)
|
|
if [ -f $STATUS_DIR/reginit.vol ]; then
|
|
$FINDPID_SSH_CMD
|
|
FOUNDPID=$?
|
|
if [ $FOUNDPID -eq 1 ] ; then
|
|
echo "no ssh client matching $KEEP_PID. cleanup state files. for next restart"
|
|
rm -f $NETCONFIG_V4_VOL
|
|
rm -f $NETCONFIG_V6_VOL
|
|
rm -f $STATUS_DIR/reginit.vol
|
|
rm -f $STATUS_DIR/con_hello_sent.vol
|
|
rm -f $STATUS_DIR/network_info.vol
|
|
rm -f $STATUS_DIR/network_info_sent.vol
|
|
D=`epoch`
|
|
echo "RESULT 9010 done $D $ETHER_SCANNED no ssh client running restart registration" >> $DATA_NEW_DIR/simpleping
|
|
echo "RESULT 9011 done $D $ETHER_SCANNED stderr" `cat $SSH_ERR` >> $DATA_NEW_DIR/simpleping
|
|
else
|
|
$SET_LEDS_CMD net-ok
|
|
$SET_LEDS_CMD keep-found
|
|
echo "found KEEP ssh client running"
|
|
fi
|
|
fi
|
|
|
|
if [ -s $STATUS_DIR/con_keep_reply.vol ] ; then
|
|
echo "Controller kicked us out. KEEP result is nonzero"
|
|
echo "kill the session and forced reregister"
|
|
$KILL_SSH_CMD
|
|
echo "REASON=CON_KEEP_BROKEN" > $FORCE_REG
|
|
rm -f $STATUS_DIR/con_keep_reply.txt
|
|
rm -f $STATUS_DIR/reginit.vol
|
|
D=`epoch`
|
|
echo "RESULT 9003 done $D $ETHER_SCANNED REREGISTER Controller kicked us out from KEEP" >> $DATA_NEW_DIR/simpleping
|
|
echo "RESULT 9003 done $D $ETHER_SCANNED REREGISTER Controller kicked us out from KEEP"
|
|
fi
|
|
|
|
$TRIGGER_MANUAL_UPGRADE_CMD
|
|
|
|
if [ ! -f $STATUS_DIR/reginit.vol ] ; then
|
|
D=`epoch`
|
|
echo "RESULT 9006 done $D $ETHER_SCANNED no reginit.vol start registration"
|
|
echo "$STATUS_DIR/reginit.vol does not exist try new reg"
|
|
UPTIME=`onlyuptime`
|
|
$SET_LEDS_CMD net-try
|
|
if evping -A "9017" -e -O $DATA_NEW_DIR/simpleping "U$UPTIME.$SOS_ID.sos.atlas.ripe.net"
|
|
then
|
|
echo "Ping works"
|
|
$SET_LEDS_CMD net-ok
|
|
else
|
|
echo "Ping failed"
|
|
$SET_LEDS_CMD net-fail
|
|
fi
|
|
$KILL_SSH_CMD
|
|
echo "start reg"
|
|
$BIN_DIR/reginit.sh $STATUS_DIR/reginit.vol
|
|
fi
|
|
|
|
$BUDDYINFO $LOW_MEM_T $DATA_NEW_DIR/simpleping
|
|
if [ $? -eq 1 ] ; then
|
|
D=`epoch`
|
|
echo "RESULT 9005 done $D $ETHER_SCANNED REBOOT low memeory. $LOW_MEM_T K blocks "
|
|
echo "RESULT 9005 done $D $ETHER_SCANNED REBOOT low memeory. $LOW_MEM_T K blocks " >> $DATA_NEW_DIR/simpleping
|
|
reboot_probe
|
|
fi
|
|
|
|
if [ -n "$do_rxtxrpt" ]
|
|
then
|
|
rxtxrpt -A "9002" >> $DATA_NEW_DIR/simpleping
|
|
rxtxrpt
|
|
fi
|
|
rptaddrs -A 9104 -c $DATA_NEW_DIR/v6addr.vol -O $DATA_NEW_DIR/v6addr.txt
|
|
rptuptime >> $DATA_NEW_DIR/simpleping
|
|
condmv $DATA_NEW_DIR/v6addr.txt $DATA_OUT_DIR/v6addr.txt
|
|
condmv $DATA_NEW_DIR/simpleping $DATA_OUT_DIR/simpleping
|
|
|
|
epoch >$STATUS_DIR/currenttime.txt
|
|
$HANDLE_STORAGE_CURRENT_TIME
|
|
|
|
sleepkick 180
|
|
|
|
if [ -f $DATA_NEW_DIR/reboot-count.txt ]
|
|
then
|
|
UPTIME=`onlyuptime`
|
|
if [ $UPTIME -gt 600 ]
|
|
then
|
|
# Use condmv because it is build-in
|
|
echo Moving reboot-count.txt
|
|
condmv -f $DATA_NEW_DIR/reboot-count.txt $DATA_NEW_DIR/reboot-count.old
|
|
fi
|
|
else
|
|
if dfrm -A 9018 $DATA_DIR $LOW_DISK_LIMIT $DATA_NEW_DIR $DATA_OUT_DIR >> $DATA_NEW_DIR/simpleping
|
|
then
|
|
D=`epoch`
|
|
echo "RESULT 9007 done $D $ETHER_SCANNED REBOOT data full available space < $LOW_DISK_LIMIT KB"
|
|
echo "RESULT 9007 done $D $ETHER_SCANNED REBOOT data full available space < $LOW_DISK_LIMIT KB" >> $DATA_NEW_DIR/simpleping
|
|
reboot_probe
|
|
fi
|
|
fi
|
|
|
|
if [ -n "$RESOLVCONF_CMD" -a -f $DATA_NEW_DIR/resolv.conf.ra6 ]
|
|
then
|
|
mv $DATA_NEW_DIR/resolv.conf.ra6 /tmp/resolv.conf.ra6
|
|
$RESOLVCONF_CMD
|
|
fi
|
|
|
|
# Check if perd is still running
|
|
if check_pid $(cat $RUN_DIR/perd-main.pid.vol)
|
|
then
|
|
: # perd is running
|
|
else
|
|
echo restarting perd
|
|
start_perd
|
|
atlas_log 9811 'perd' 'perd died (restarted)'
|
|
fi
|
|
# Check if eperd is still running
|
|
if check_pid $(cat $RUN_DIR/perd-7.pid.vol)
|
|
then
|
|
: # eperd is running
|
|
else
|
|
echo restarting eperd
|
|
start_eperd
|
|
atlas_log 9817 'eperd' 'eperd died (restarted)'
|
|
fi
|
|
# Check if eooqd is still running
|
|
if check_pid $(cat $RUN_DIR/eooqd.pid.vol)
|
|
then
|
|
: # eooqd is running
|
|
else
|
|
echo restarting eooqd
|
|
start_eooqd
|
|
atlas_log 9819 'eooqd' 'eooqd died (restarted)'
|
|
fi
|
|
|
|
sync
|
|
$DROP_CACHES
|
|
done
|