exit 0
}
-case $cmd in
+monitor_interfaces()
+{
+ local INTERFACES=`cat $CTDB_PUBLIC_ADDRESSES |
+ sed -e "s/^[^\t ]*[\t ]*//" -e "s/,/ /g" -e "s/[\t ]*$//"`
+
+ [ "$CTDB_PUBLIC_INTERFACE" ] && INTERFACES="$CTDB_PUBLIC_INTERFACE $INTERFACES"
+ [ "$CTDB_NATGW_PUBLIC_IFACE" ] && INTERFACES="$CTDB_NATGW_PUBLIC_IFACE $INTERFACES"
+
+ local IFACES=`ctdb ifaces -Y | grep -v '^:Name:LinkStatus:References:'`
+
+ local I
+ local IFACE
+
+ for I in $IFACES; do
+ IFACE=`echo -n "$I" | cut -d ':' -f2`
+ INTERFACES="$IFACE $INTERFACES"
+ done
+
+ INTERFACES=`for IFACE in $INTERFACES ; do echo $IFACE ; done | sort | uniq`
+
+ local fail=0
+ local ok=0
+ for IFACE in $INTERFACES ; do
+
+ ip addr show $IFACE 2>/dev/null >/dev/null || {
+ echo Interface $IFACE does not exist but it is used by public addresses.
+ exit 1
+ }
+
+ # These interfaces are sometimes bond devices
+ # When we use VLANs for bond interfaces, there will only
+ # be an entry in /proc for the underlying real interface
+ local REALIFACE=`echo $IFACE |sed -e 's/\..*$//'`
+ [ -f /proc/net/bonding/$REALIFACE ] && {
+ grep -q 'Currently Active Slave: None' /proc/net/bonding/$REALIFACE && {
+ echo "ERROR: No active slaves for bond device $REALIFACE"
+ fail=1
+ ctdb setifacelink $IFACE down
+ continue;
+ }
+ grep -q '^MII Status: up' /proc/net/bonding/$REALIFACE || {
+ echo "ERROR: public network interface $REALIFACE is down"
+ fail=1
+ ctdb setifacelink $IFACE down
+ continue;
+ }
+ grep -q '^Bonding Mode: IEEE 802.3ad Dynamic link aggregation' /proc/net/bonding/$REALIFACE && {
+ grep 'MII Status:' /proc/net/bonding/$REALIFACE | tail -n +2 | grep -q '^MII Status: up' || {
+ echo No active slaves for 802.ad bond device $REALIFACE
+ ctdb setifacelink $IFACE down
+ fail=1
+ continue
+ }
+ }
+ ok=1 # we only set ok for interfaces known to ctdbd
+ ctdb setifacelink $IFACE up
+ continue;
+ }
+
+ case $IFACE in
+ lo*)
+ # loopback is always working
+ ok=1 # we only set ok for interfaces known to ctdbd
+ ctdb setifacelink $IFACE up
+ ;;
+ ib*)
+ # we dont know how to test ib links
+ ok=1 # we only set ok for interfaces known to ctdbd
+ ctdb setifacelink $IFACE up
+ ;;
+ *)
+ [ -z "$IFACE" ] || {
+ [ "$(basename $(readlink /sys/class/net/$IFACE/device/driver) 2>/dev/null)" = virtio_net ] ||
+ ethtool $IFACE | grep -q 'Link detected: yes' || {
+ # On some systems, this is not successful when a
+ # cable is plugged but the interface has not been
+ # brought up previously. Bring the interface up and
+ # try again...
+ /sbin/ip link set $IFACE up
+ ethtool $IFACE | grep -q 'Link detected: yes' || {
+ echo "ERROR: No link on the public network interface $IFACE"
+ fail=1
+ ctdb setifacelink $IFACE down
+ continue
+ }
+ }
+ ok=1 # we only set ok for interfaces known to ctdbd
+ ctdb setifacelink $IFACE up
+ }
+ ;;
+ esac
+
+ done
+
+ test x"$fail" = x"0" && {
+ return 0;
+ }
+
+ test x"$ok" = x"1" && {
+ return 2;
+ }
+
+ return 1;
+}
+
+case "$1" in
#############################
# called when ctdbd starts up
- startup)
+ init)
# make sure that we only respond to ARP messages from the NIC where
# a particular ip address is associated.
[ -f /proc/sys/net/ipv4/conf/all/arp_filter ] && {
echo 1 > /proc/sys/net/ipv4/conf/all/arp_filter
}
- cat "$CTDB_PUBLIC_ADDRESSES" | cut -d/ -f1 | while read _IP; do
- _IP_HELD=`/sbin/ip addr show | grep "inet $_IP/"`
- [ -z "$_IP_HELD" ] || {
- _IFACE=`echo $_IP_HELD | sed -e "s/.*\s//"`
- _NM=`echo $_IP_HELD | sed -e "s/.*$_IP\///" -e "s/\s.*//"`
- echo "Removing public address $_IP/$_NM from device $_IFACE"
- /sbin/ip addr del $_IP/$_NM dev $_IFACE
- }
+ ;;
+
+ #############################
+ # called after ctdbd has done its initial recovery
+ # and we start the services to become healthy
+ startup)
+ # Assume all links are good initially
+ INTERFACES=`for IFACE in $INTERFACES ; do echo $IFACE ; done | sort | uniq`
+
+ for IFACE in $INTERFACES ; do
+ ctdb setifacelink $IFACE down
done
+
+ monitor_interfaces
+
;;
################################################
# called when ctdbd wants to claim an IP address
takeip)
- if [ $# != 3 ]; then
+ if [ $# != 4 ]; then
echo "must supply interface, IP and maskbits"
exit 1
fi
- iface=$1
- ip=$2
- maskbits=$3
-
- # we make sure the interface is up first
- /sbin/ip link set $iface up || {
- echo "Failed to bringup interface $iface"
- exit 1
- }
- /sbin/ip addr add $ip/$maskbits brd + dev $iface || {
- echo "Failed to add $ip/$maskbits on dev $iface"
+ iface=$2
+ ip=$3
+ maskbits=$4
+
+ add_ip_to_iface $iface $ip $maskbits || {
+ exit 1;
}
+
# cope with the script being killed while we have the interface blocked
iptables -D INPUT -i $iface -d $ip -j DROP 2> /dev/null
##################################################
# called when ctdbd wants to release an IP address
releaseip)
- if [ $# != 3 ]; then
+ if [ $# != 4 ]; then
echo "must supply interface, IP and maskbits"
exit 1
fi
# 2) use netstat -tn to find existing connections, and kill them
# 3) remove the IP from the interface
# 4) remove the firewall rule
- iface=$1
- ip=$2
- maskbits=$3
+ iface=$2
+ ip=$3
+ maskbits=$4
failed=0
# we do an extra delete to cope with the script being killed
iptables -I INPUT -i $iface -d $ip -j DROP
kill_tcp_connections $ip
- # the ip tool will delete all secondary IPs if this is the primary. To work around
- # this _very_ annoying behaviour we have to keep a record of the secondaries and re-add
- # them afterwards. yuck
- secondaries=""
- if /sbin/ip addr list dev $iface primary | grep -q "inet $ip/$maskbits " ; then
- secondaries=`/sbin/ip addr list dev $iface secondary | grep " inet " | awk '{print $2}'`
- fi
- /sbin/ip addr del $ip/$maskbits dev $iface || failed=1
- [ -z "$secondaries" ] || {
- for i in $secondaries; do
- if /sbin/ip addr list dev $iface | grep -q "inet $i" ; then
- echo "kept secondary $i on dev $iface"
- else
- echo "re-adding secondary address $i to dev $iface"
- /sbin/ip addr add $i dev $iface || failed=1
- fi
- done
+ delete_ip_from_iface $iface $ip $maskbits || {
+ iptables -D INPUT -i $iface -d $ip -j DROP 2> /dev/null
+ exit 1;
}
+
iptables -D INPUT -i $iface -d $ip -j DROP 2> /dev/null
- [ $failed = 0 ] || {
- echo "Failed to del $ip on dev $iface"
- exit 1
+
+ # flush our route cache
+ echo 1 > /proc/sys/net/ipv4/route/flush
+ ;;
+
+ ##################################################
+ # called when ctdbd wants to update an IP address
+ updateip)
+ if [ $# != 5 ]; then
+ echo "must supply old interface, new interface, IP and maskbits"
+ exit 1
+ fi
+
+ # moving an IP is a bit more complex than it seems.
+ # First we drop all traffic on the old interface.
+ # Then we try to add the ip to the new interface and before
+ # we finally remove it from the old interface.
+ #
+ # 1) firewall this IP, so no new external packets arrive for it
+ # 2) add the IP to the new interface
+ # 3) remove the IP from the old interface
+ # 4) remove the firewall rule
+ # 5) use ctdb gratiousarp to propagate the new mac address
+ # 6) use netstat -tn to find existing connections, and tickle them
+ oiface=$2
+ niface=$3
+ ip=$4
+ maskbits=$5
+
+ failed=0
+ # we do an extra delete to cope with the script being killed
+ iptables -D INPUT -i $oiface -d $ip -j DROP 2> /dev/null
+ iptables -I INPUT -i $oiface -d $ip -j DROP
+
+ delete_ip_from_iface $oiface $ip $maskbits 2>/dev/null
+ delete_ip_from_iface $niface $ip $maskbits 2>/dev/null
+
+ add_ip_to_iface $niface $ip $maskbits || {
+ iptables -D INPUT -i $oiface -d $ip -j DROP 2> /dev/null
+ exit 1;
}
+ # cope with the script being killed while we have the interface blocked
+ iptables -D INPUT -i $oiface -d $ip -j DROP 2> /dev/null
+
# flush our route cache
echo 1 > /proc/sys/net/ipv4/route/flush
+
+ # propagate the new mac address
+ ctdb gratiousarp $ip $niface
+
+ # tickle all existing connections, so that dropped packets
+ # are retransmited and the tcp streams work
+
+ tickle_tcp_connections $ip
+
;;
;;
monitor)
- INTERFACES=`cat $CTDB_PUBLIC_ADDRESSES |
- sed -e "s/^[^\t ]*[\t ]*//" -e "s/[\t ]*$//"`
+ monitor_interfaces
+ ret=$?
- [ "$CTDB_PUBLIC_INTERFACE" ] && INTERFACES="$CTDB_PUBLIC_INTERFACE $INTERFACES"
-
- INTERFACES=`for IFACE in $INTERFACES ; do echo $IFACE ; done | sort | uniq`
-
- for IFACE in $INTERFACES ; do
- case $IFACE in
- ethX*|bond*)
- IFACE=`echo $IFACE |sed -e 's/\....$//'`
- grep -q 'Currently Active Slave: None' /proc/net/bonding/$IFACE && {
- echo "ERROR: No active slaves for bond device $IFACE"
- exit 1
- }
- grep -q '^MII Status: up' /proc/net/bonding/$IFACE || {
- echo "ERROR: public network interface $IFACE is down"
- exit 1
- }
- ;;
- ib*)
- # we dont know how to test ib links
- ;;
- *)
- [ -z "$IFACE" ] || {
- /usr/sbin/ethtool $IFACE | grep -q 'Link detected: yes' || {
- # On some systems, this is not successful when a
- # cable is plugged but the interface has not been
- # brought up previously. Bring the interface up and
- # try again...
- /sbin/ip link set $IFACE up
- /usr/sbin/ethtool $IFACE | grep -q 'Link detected: yes' || {
- echo "ERROR: No link on the public network interface $IFACE"
- exit 1
- }
- }
+ test x"$ret" = x"2" && {
+ test x"$CTDB_PARTIALLY_ONLINE_INTERFACES" != x"yes" && {
+ exit 1;
}
- ;;
- esac
- done
+ # as long as we have one interface available don't become
+ # unhealthy
+ ret=0
+ }
+
+ test x"$ret" != x"0" && {
+ exit 1;
+ }
;;
- status)
- ctdb_checkstatus || exit $?
+ *)
+ ctdb_standard_event_handler "$@"
;;
esac