# utility functions for ctdb event scripts
-PATH=/bin:/usr/bin:/usr/sbin:/sbin:$PATH
+if [ -z "$CTDB_BASE" ] ; then
+ echo 'CTDB_BASE unset in CTDB functions file'
+ exit 1
+fi
-[ -z "$CTDB_VARDIR" ] && {
- if [ -d "/var/lib/ctdb" ] ; then
- export CTDB_VARDIR="/var/lib/ctdb"
- else
- export CTDB_VARDIR="/var/ctdb"
- fi
-}
-[ -z "$CTDB_ETCDIR" ] && {
- export CTDB_ETCDIR="/etc"
-}
+CTDB_VARDIR="/usr/local/var/lib/ctdb"
+ctdb_rundir="/usr/local/var/run/ctdb"
+
+# Only (and always) override these variables in test code
+
+if [ -z "$CTDB_SCRIPT_VARDIR" ] ; then
+ CTDB_SCRIPT_VARDIR="/usr/local/var/lib/ctdb/state"
+fi
+
+if [ -z "$CTDB_SYS_ETCDIR" ] ; then
+ CTDB_SYS_ETCDIR="/etc"
+fi
#######################################
# pull in a system config file, if any
+
+rewrite_ctdb_options ()
+{
+ case "$CTDB_DBDIR" in
+ tmpfs|tmpfs:*)
+ _opts_defaults="mode=700"
+ # Get any extra options specified after colon
+ if [ "$CTDB_DBDIR" = "tmpfs" ] ; then
+ _opts=""
+ else
+ _opts="${CTDB_DBDIR#tmpfs:}"
+ fi
+ # This is an internal variable, only used by ctdbd_wrapper.
+ # It is OK to repeat mount options - last value wins
+ CTDB_DBDIR_TMPFS_OPTIONS="${_opts_defaults}${_opts:+,}${_opts}"
+
+ CTDB_DBDIR="${ctdb_rundir}/CTDB_DBDIR"
+ ;;
+ *)
+ CTDB_DBDIR_TMPFS_OPTIONS=""
+ esac
+}
+
_loadconfig() {
if [ -z "$1" ] ; then
foo="${service_config:-${service_name}}"
if [ -n "$foo" ] ; then
loadconfig "$foo"
+ return
fi
- elif [ "$1" != "ctdb" ] ; then
+ fi
+
+ if [ "$1" != "ctdb" ] ; then
loadconfig "ctdb"
fi
- if [ -f $CTDB_ETCDIR/sysconfig/$1 ]; then
- . $CTDB_ETCDIR/sysconfig/$1
- elif [ -f $CTDB_ETCDIR/default/$1 ]; then
- . $CTDB_ETCDIR/default/$1
+ if [ -z "$1" ] ; then
+ return
+ fi
+
+ if [ -f $CTDB_SYS_ETCDIR/sysconfig/$1 ]; then
+ . $CTDB_SYS_ETCDIR/sysconfig/$1
+ elif [ -f $CTDB_SYS_ETCDIR/default/$1 ]; then
+ . $CTDB_SYS_ETCDIR/default/$1
elif [ -f $CTDB_BASE/sysconfig/$1 ]; then
. $CTDB_BASE/sysconfig/$1
fi
+
+ if [ "$1" = "ctdb" ] ; then
+ _config="${CTDB_BASE}/ctdbd.conf"
+ if [ -r "$_config" ] ; then
+ . "$_config"
+ fi
+ rewrite_ctdb_options
+ fi
}
loadconfig () {
# here document.
if [ -n "$1" ] ; then
echo "DEBUG: $*"
- elif ! tty -s ; then
+ else
sed -e 's@^@DEBUG: @'
fi
+ else
+ if [ -z "$1" ] ; then
+ cat >/dev/null
+ fi
fi
}
{
_tag="$1" ; shift
- _using_syslog=false
- if [ "$CTDB_SYSLOG" = "yes" -o -z "$CTDB_LOGFILE" ] ; then
- _using_syslog=true
- fi
- case "$CTDB_OPTIONS" in
- *--syslog*) _using_syslog=true ;;
- esac
-
- if $_using_syslog ; then
- logger -t "ctdbd: ${_tag}" $*
- else
- {
- if [ -n "$*" ] ; then
- echo "$*"
+ case "$CTDB_LOGGING" in
+ file:*|"")
+ if [ -n "$CTDB_LOGGING" ] ; then
+ _file="${CTDB_LOGGING#file:}"
else
- cat
+ _file="/usr/local/var/log/log.ctdb"
fi
- } >>"${CTDB_LOGFILE:-/var/log/log.ctdb}"
- fi
+ {
+ if [ -n "$*" ] ; then
+ echo "$*"
+ else
+ cat
+ fi
+ } >>"$_file"
+ ;;
+ *)
+ # Handle all syslog:* variants here too. There's no tool to do
+ # the lossy things, so just use logger.
+ logger -t "ctdbd: ${_tag}" $*
+ ;;
+ esac
}
# When things are run in the background in an eventscript then logging
##############################################################
# determine on what type of system (init style) we are running
-detect_init_style() {
+detect_init_style()
+{
# only do detection if not already set:
- test "x$CTDB_INIT_STYLE" != "x" && return
+ [ -z "$CTDB_INIT_STYLE" ] || return
if [ -x /sbin/startproc ]; then
CTDB_INIT_STYLE="suse"
if [ -x /sbin/service ]; then
$_nice /sbin/service "$_service_name" "$_op"
- elif [ -x $CTDB_ETCDIR/init.d/$_service_name ]; then
- $_nice $CTDB_ETCDIR/init.d/$_service_name "$_op"
- elif [ -x $CTDB_ETCDIR/rc.d/init.d/$_service_name ]; then
- $_nice $CTDB_ETCDIR/rc.d/init.d/$_service_name "$_op"
+ elif [ -x /usr/sbin/service ]; then
+ $_nice /usr/sbin/service "$_service_name" "$_op"
+ elif [ -x $CTDB_SYS_ETCDIR/init.d/$_service_name ]; then
+ $_nice $CTDB_SYS_ETCDIR/init.d/$_service_name "$_op"
+ elif [ -x $CTDB_SYS_ETCDIR/rc.d/init.d/$_service_name ]; then
+ $_nice $CTDB_SYS_ETCDIR/rc.d/init.d/$_service_name "$_op"
fi
}
_service "$@"
}
+######################################################
+# Cached retrieval of PNN from local node. This never changes so why
+# open a client connection to the server each time this is needed?
+# This sets $pnn - this avoid an unnecessary subprocess.
+ctdb_get_pnn ()
+{
+ _pnn_file="${CTDB_SCRIPT_VARDIR}/my-pnn"
+ if [ ! -f "$_pnn_file" ] ; then
+ ctdb pnn | sed -e 's@.*:@@' >"$_pnn_file"
+ fi
+
+ read pnn <"$_pnn_file"
+}
+
######################################################
# wrapper around /proc/ settings to allow them to be hooked
# for testing
echo "$2" >"/proc/$1"
}
+set_proc_maybe ()
+{
+ if [ -w "/proc/$1" ] ; then
+ set_proc "$1" "$2"
+ fi
+}
+
######################################################
# wrapper around getting file contents from /proc/ to allow
# this to be hooked for testing
}
######################################################
-# Check that an RPC service is healthy -
-# this includes allowing a certain number of failures
-# before marking the NFS service unhealthy.
-#
-# usage: nfs_check_rpc_service SERVICE_NAME [ triple ...]
-#
-# each triple is a set of 3 arguments: an operator, a
-# fail count limit and an action string.
-#
-# For example:
-#
-# nfs_check_rpc_service "lockd" \
-# -ge 15 "verbose restart unhealthy" \
-# -eq 10 "restart:bs"
-#
-# says that if lockd is down for 15 iterations then do
-# a verbose restart of lockd and mark the node unhealthy.
-# Before this, after 10 iterations of failure, the
-# service is restarted silently in the background.
-# Order is important: the number of failures need to be
-# specified in reverse order because processing stops
-# after the first condition that is true.
-######################################################
-nfs_check_rpc_service ()
+# Print up to $_max kernel stack traces for processes named $_program
+program_stack_traces ()
{
- _prog_name="$1" ; shift
-
- _v=""
- case "$1" in
- -*) : ;;
- *) _v="$1" ; shift ;;
- esac
-
- _version=${_v:-1}
- _rpc_prog="$_prog_name"
- _restart=""
- _opts=""
- case "$_prog_name" in
- knfsd)
- _rpc_prog=nfs
- _version=${_v:-3}
- _restart="echo 'Trying to restart NFS service'"
- _restart="${_restart}; startstop_nfs restart"
- ;;
- ganesha)
- _rpc_prog=nfs
- _version=${_v:-3}
- _restart="echo 'Trying to restart Ganesha NFS service'"
- _restart="${_restart}; startstop_ganesha restart"
- ;;
- mountd)
- _opts="${MOUNTD_PORT:+ -p }${MOUNTD_PORT}"
- ;;
- rquotad)
- _opts="${RQUOTAD_PORT:+ -p }${RQUOTAD_PORT}"
- ;;
- lockd)
- _rpc_prog=nlockmgr
- _version=${_v:-4}
- _restart="echo 'Trying to restart lock manager service'"
- _restart="${_restart}; startstop_nfslock restart"
- ;;
- statd)
- _rpc_prog=status
- _opts="${STATD_HOSTNAME:+ -n }${STATD_HOSTNAME}"
- _opts="${_opts}${STATD_PORT:+ -p }${STATD_PORT}"
- _opts="${_opts}${STATD_OUTGOING_PORT:+ -o }${STATD_OUTGOING_PORT}"
- ;;
- *)
- echo "Internal error: unknown RPC program \"$_prog_name\"."
- exit 1
- esac
-
- _service_name="nfs_${_prog_name}"
-
- if ctdb_check_rpc "$_rpc_prog" $_version >/dev/null ; then
- ctdb_counter_init "$_service_name"
- return 0
- fi
-
- ctdb_counter_incr "$_service_name"
-
- while [ -n "$3" ] ; do
- ctdb_check_counter "quiet" "$1" "$2" "$_service_name" || {
- for _action in $3 ; do
- case "$_action" in
- verbose)
- echo "$ctdb_check_rpc_out"
- ;;
- restart|restart:*)
- # No explicit command specified, construct rpc command.
- if [ -z "$_restart" ] ; then
- _p="rpc.${_prog_name}"
- _restart="echo 'Trying to restart $_prog_name [${_p}${_opts}]'"
- _restart="${_restart}; killall -q -9 $_p"
- _restart="${_restart}; $_p $_opts"
- fi
-
- # Process restart flags...
- _flags="${_action#restart:}"
- # There may not have been a colon...
- [ "$_flags" != "$_action" ] || _flags=""
- # q=quiet - everything to /dev/null
- if [ "${_flags#*q}" != "$_flags" ] ; then
- _restart="{ ${_restart} ; } >/dev/null 2>&1"
- fi
- # s=stealthy - last command to /dev/null
- if [ "${_flags#*s}" != "$_flags" ] ; then
- _restart="${_restart} >/dev/null 2>&1"
- fi
- # b=background - the whole thing, easy and reliable
- if [ "${_flags#*b}" != "$_flags" ] ; then
- _restart="{ ${_restart} ; } &"
- fi
-
- # Do it!
- eval "${_restart}"
- ;;
- unhealthy)
- exit 1
- ;;
- *)
- echo "Internal error: unknown action \"$_action\"."
- exit 1
- esac
- done
-
- # Only process the first action group.
- break
- }
- shift 3
+ _prog="$1"
+ _max="${2:-1}"
+
+ _count=1
+ for _pid in $(pidof "$_prog") ; do
+ [ $_count -le $_max ] || break
+
+ # Do this first to avoid racing with process exit
+ _stack=$(get_proc "${_pid}/stack" 2>/dev/null)
+ if [ -n "$_stack" ] ; then
+ echo "Stack trace for ${_prog}[${_pid}]:"
+ echo "$_stack"
+ _count=$(($_count + 1))
+ fi
done
}
-######################################################
-# check that a rpc server is registered with portmap
-# and responding to requests
-# usage: ctdb_check_rpc SERVICE_NAME VERSION
-######################################################
-ctdb_check_rpc ()
-{
- progname="$1"
- version="$2"
-
- if ! ctdb_check_rpc_out=$(rpcinfo -u localhost $progname $version 2>&1) ; then
- ctdb_check_rpc_out="ERROR: $progname failed RPC check:
-$ctdb_check_rpc_out"
- echo "$ctdb_check_rpc_out"
- return 1
- fi
-}
-
######################################################
# Ensure $service_name is set
assert_service_name ()
_ctdb_check_tcp_common ()
{
assert_service_name
- _ctdb_service_started_file="$ctdb_fail_dir/$service_name.started"
+ _d="${CTDB_SCRIPT_VARDIR}/failcount"
+ _ctdb_service_started_file="${_d}/${service_name}.started"
}
ctdb_check_tcp_init ()
touch "$_ctdb_service_started_file"
}
+# Check whether something is listening on all of the given TCP ports
+# using the "ctdb checktcpport" command.
ctdb_check_tcp_ports()
{
if [ -z "$1" ] ; then
exit 1
fi
- # Set default value for CTDB_TCP_PORT_CHECKS if unset.
- # If any of these defaults are unsupported then this variable can
- # be overridden in /etc/sysconfig/ctdb or via a file in
- # /etc/ctdb/rc.local.d/.
- : ${CTDB_TCP_PORT_CHECKERS:=ctdb nmap netstat}
-
- for _c in $CTDB_TCP_PORT_CHECKERS ; do
- ctdb_check_tcp_ports_$_c "$@"
- case "$?" in
+ for _p ; do # process each function argument (port)
+ _cmd="ctdb checktcpport $_p"
+ _out=$($_cmd 2>&1)
+ _ret=$?
+ case "$_ret" in
0)
- _ctdb_check_tcp_common
- rm -f "$_ctdb_service_started_file"
- return 0
- ;;
- 1)
_ctdb_check_tcp_common
if [ ! -f "$_ctdb_service_started_file" ] ; then
echo "ERROR: $service_name tcp port $_p is not responding"
- debug <<EOF
-$ctdb_check_tcp_ports_debug
-EOF
+ debug "\"ctdb checktcpport $_p\" was able to bind to port"
else
echo "INFO: $service_name tcp port $_p is not responding"
fi
- return 1
- ;;
- 127)
- debug <<EOF
-ctdb_check_ports - checker $_c not implemented
-output from checker was:
-$ctdb_check_tcp_ports_debug
-EOF
- ;;
- *)
-
- esac
- done
-
- echo "INTERNAL ERROR: ctdb_check_ports - no working checkers in CTDB_TCP_PORT_CHECKERS=\"$CTDB_TCP_PORT_CHECKERS\""
-
- return 127
-}
-
-ctdb_check_tcp_ports_netstat ()
-{
- _cmd='netstat -l -t -n'
- _ns=$($_cmd 2>&1)
- if [ $? -eq 127 ] ; then
- # netstat probably not installed - unlikely?
- ctdb_check_tcp_ports_debug="$_ns"
- return 127
- fi
-
- for _p ; do # process each function argument (port)
- for _a in '0\.0\.0\.0' '::' ; do
- _pat="[[:space:]]${_a}:${_p}[[:space:]]+[^[:space:]]+[[:space:]]+LISTEN"
- if echo "$_ns" | grep -E -q "$_pat" ; then
- # We matched the port, so process next port
- continue 2
- fi
- done
-
- # We didn't match the port, so flag an error.
- ctdb_check_tcp_ports_debug="$_cmd shows this output:
-$_ns"
- return 1
- done
-
- return 0
-}
-
-ctdb_check_tcp_ports_nmap ()
-{
- # nmap wants a comma-separated list of ports
- _ports=""
- for _p ; do
- _ports="${_ports}${_ports:+,}${_p}"
- done
-
- _cmd="nmap -n -oG - -PS 127.0.0.1 -p $_ports"
-
- _nmap_out=$($_cmd 2>&1)
- if [ $? -eq 127 ] ; then
- # nmap probably not installed
- ctdb_check_tcp_ports_debug="$_nmap_out"
- return 127
- fi
-
- # get the port-related output
- _port_info=$(echo "$_nmap_out" | sed -n -r -e 's@^.*Ports:[[:space:]]@@p')
-
- for _p ; do
- # looking for something like this:
- # 445/open/tcp//microsoft-ds///
- # possibly followed by a comma
- _t="$_p/open/tcp//"
- case "$_port_info" in
- # The info we're after must be either at the beginning of
- # the string or it must follow a space.
- $_t*|*\ $_t*) : ;;
- *)
- # Nope, flag an error...
- ctdb_check_tcp_ports_debug="$_cmd shows this output:
-$_nmap_out"
- return 1
- esac
- done
-
- return 0
-}
-
-# Use the new "ctdb checktcpport" command to check the port.
-# This is very cheap.
-ctdb_check_tcp_ports_ctdb ()
-{
- for _p ; do # process each function argument (port)
- _cmd="ctdb checktcpport $_p"
- _out=$($_cmd 2>&1)
- _ret=$?
- case "$_ret" in
- 0)
- ctdb_check_tcp_ports_debug="\"$_cmd\" was able to bind to port"
return 1
;;
98)
continue
;;
*)
- ctdb_check_tcp_ports_debug="$_cmd (exited with $_ret) with output:
+ echo "ERROR: unexpected error running \"ctdb checktcpport\""
+ debug <<EOF
+ctdb checktcpport (exited with $_ret) with output:
$_out"
- # assume not implemented
- return 127
+EOF
+ return $_ret
esac
done
+ # All ports listening
+ _ctdb_check_tcp_common
+ rm -f "$_ctdb_service_started_file"
return 0
}
################################################
# kill off any TCP connections with the given IP
################################################
-kill_tcp_connections() {
- _IP="$1"
+kill_tcp_connections ()
+{
+ _ip="$1"
_oneway=false
if [ "$2" = "oneway" ] ; then
_oneway=true
fi
- _killcount=0
-
- get_tcp_connections_for_ip "$_IP" | {
- _failed=false
-
- while read dest src; do
- echo "Killing TCP connection $src $dest"
- ctdb killtcp $src $dest >/dev/null 2>&1 || _failed=true
- _destport="${dest##*:}"
+ get_tcp_connections_for_ip "$_ip" | {
+ _killcount=0
+ _connections=""
+ _nl="
+"
+ while read _dst _src; do
+ _destport="${_dst##*:}"
__oneway=$_oneway
case $_destport in
# we only do one-way killtcp for CIFS
139|445) __oneway=true ;;
esac
+
+ echo "Killing TCP connection $_src $_dst"
+ _connections="${_connections}${_nl}${_src} ${_dst}"
if ! $__oneway ; then
- ctdb killtcp $dest $src >/dev/null 2>&1 || _failed=true
+ _connections="${_connections}${_nl}${_dst} ${_src}"
fi
- _killcount=`expr $_killcount + 1`
+ _killcount=$(($_killcount + 1))
done
- if $_failed ; then
- echo "Failed to send killtcp control"
+ if [ $_killcount -eq 0 ] ; then
return
fi
- [ $_killcount -gt 0 ] || {
- return;
+
+ echo "$_connections" | ctdb killtcp || {
+ echo "Failed to send killtcp control"
+ return
}
_count=0
while : ; do
- if [ -z "$(get_tcp_connections_for_ip $_IP)" ] ; then
- echo "killed $_killcount TCP connections to released IP $_IP"
+ _remaining=$(get_tcp_connections_for_ip $_ip | wc -l)
+
+ if [ $_remaining -eq 0 ] ; then
+ echo "Killed $_killcount TCP connections to released IP $_ip"
return
fi
_count=$(($_count + 1))
if [ $_count -gt 3 ] ; then
- echo "Timed out killing tcp connections for IP $_IP"
+ echo "Timed out killing tcp connections for IP $_ip ($_remaining remaining)"
return
fi
+ echo "Waiting for $_remaining connections to be killed for IP $_ip"
sleep 1
done
}
##################################################################
# kill off the local end for any TCP connections with the given IP
##################################################################
-kill_tcp_connections_local_only()
+kill_tcp_connections_local_only ()
{
kill_tcp_connections "$1" "oneway"
}
##################################################################
# tickle any TCP connections with the given IP
##################################################################
-tickle_tcp_connections() {
- _IP="$1"
+tickle_tcp_connections ()
+{
+ _ip="$1"
- get_tcp_connections_for_ip "$_IP" |
+ get_tcp_connections_for_ip "$_ip" |
{
_failed=false
}
########################################################
-# start/stop the Ganesha nfs service
-########################################################
-startstop_ganesha()
-{
- _service_name="nfs-ganesha-$CTDB_CLUSTER_FILESYSTEM_TYPE"
- case "$1" in
- start)
- service "$_service_name" start
- ;;
- stop)
- service "$_service_name" stop
- ;;
- restart)
- service "$_service_name" restart
- ;;
- esac
-}
-
-########################################################
-# start/stop the nfs service on different platforms
-########################################################
-startstop_nfs() {
- PLATFORM="unknown"
- [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
- PLATFORM="sles"
- }
- [ -x $CTDB_ETCDIR/init.d/nfslock ] && {
- PLATFORM="rhel"
- }
-
- case $PLATFORM in
- sles)
- case $1 in
- start)
- service nfsserver start
- ;;
- stop)
- service nfsserver stop > /dev/null 2>&1
- ;;
- restart)
- set_proc "fs/nfsd/threads" 0
- service nfsserver stop > /dev/null 2>&1
- pkill -9 nfsd
- service nfsserver start
- ;;
- esac
- ;;
- rhel)
- case $1 in
- start)
- service nfslock start
- service nfs start
- ;;
- stop)
- service nfs stop
- service nfslock stop
- ;;
- restart)
- set_proc "fs/nfsd/threads" 0
- service nfs stop > /dev/null 2>&1
- service nfslock stop > /dev/null 2>&1
- pkill -9 nfsd
- service nfslock start
- service nfs start
- ;;
- esac
- ;;
- *)
- echo "Unknown platform. NFS is not supported with ctdb"
- exit 1
- ;;
- esac
-}
-
-########################################################
-# start/stop the nfs lockmanager service on different platforms
-########################################################
-startstop_nfslock() {
- PLATFORM="unknown"
- [ -x $CTDB_ETCDIR/init.d/nfsserver ] && {
- PLATFORM="sles"
- }
- [ -x $CTDB_ETCDIR/init.d/nfslock ] && {
- PLATFORM="rhel"
- }
- case $PLATFORM in
- sles)
- # for sles there is no service for lockmanager
- # so we instead just shutdown/restart nfs
- case $1 in
- start)
- service nfsserver start
- ;;
- stop)
- service nfsserver stop > /dev/null 2>&1
- ;;
- restart)
- service nfsserver stop
- service nfsserver start
- ;;
- esac
- ;;
- rhel)
- case $1 in
- start)
- service nfslock start
- ;;
- stop)
- service nfslock stop > /dev/null 2>&1
- ;;
- restart)
- service nfslock stop
- service nfslock start
- ;;
- esac
- ;;
- *)
- echo "Unknown platform. NFS locking is not supported with ctdb"
- exit 1
- ;;
- esac
-}
-
-add_ip_to_iface()
+add_ip_to_iface ()
{
_iface=$1
_ip=$2
_maskbits=$3
- _lockfile="${CTDB_VARDIR}/state/interface_modify_${_iface}.flock"
- mkdir -p "${_lockfile%/*}" # dirname
- [ -f "$_lockfile" ] || touch "$_lockfile"
-
- (
- # Note: use of return/exit/die() below only gets us out of the
- # sub-shell, which is actually what we want. That is, the
- # function should just return non-zero.
+ # Ensure interface is up
+ ip link set "$_iface" up || \
+ die "Failed to bringup interface $_iface"
- flock --timeout 30 0 || \
- die "add_ip_to_iface: unable to get lock for ${_iface}"
+ # Only need to define broadcast for IPv4
+ case "$ip" in
+ *:*) _bcast="" ;;
+ *) _bcast="brd +" ;;
+ esac
- # Ensure interface is up
- ip link set "$_iface" up || \
- die "Failed to bringup interface $_iface"
+ ip addr add "$_ip/$_maskbits" $_bcast dev "$_iface" || {
+ echo "Failed to add $_ip/$_maskbits on dev $_iface"
+ return 1
+ }
- ip addr add "$_ip/$_maskbits" brd + dev "$_iface" || \
- die "Failed to add $_ip/$_maskbits on dev $_iface"
- ) <"$_lockfile"
+ # Wait 5 seconds for IPv6 addresses to stop being tentative...
+ if [ -z "$_bcast" ] ; then
+ for _x in $(seq 1 10) ; do
+ ip addr show to "${_ip}/128" | grep -q "tentative" || break
+ sleep 0.5
+ done
- # Do nothing here - return above only gets us out of the subshell
- # and doing anything here will affect the return code.
+ # If the address was a duplicate then it won't be on the
+ # interface so flag an error.
+ _t=$(ip addr show to "${_ip}/128")
+ case "$_t" in
+ "")
+ echo "Failed to add $_ip/$_maskbits on dev $_iface"
+ return 1
+ ;;
+ *tentative*|*dadfailed*)
+ echo "Failed to add $_ip/$_maskbits on dev $_iface"
+ ip addr del "$_ip/$_maskbits" dev "$_iface"
+ return 1
+ ;;
+ esac
+ fi
}
delete_ip_from_iface()
_ip=$2
_maskbits=$3
- _lockfile="${CTDB_VARDIR}/state/interface_modify_${_iface}.flock"
- mkdir -p "${_lockfile%/*}" # dirname
- [ -f "$_lockfile" ] || touch "$_lockfile"
-
- (
- # Note: use of return/exit/die() below only gets us out of the
- # sub-shell, which is actually what we want. That is, the
- # function should just return non-zero.
-
- flock --timeout 30 0 || \
- die "delete_ip_from_iface: unable to get lock for ${_iface}"
-
- _im="$_ip/$_maskbits" # shorthand for readability
-
- # "ip addr del" will delete all secondary IPs if this is the
- # primary. To work around this _very_ annoying behaviour we
- # have to keep a record of the secondaries and re-add them
- # afterwards. Yuck!
-
- _secondaries=""
- if ip addr list dev "$_iface" primary | grep -Fq "inet $_im " ; then
- _secondaries=$(ip addr list dev "$_iface" secondary | \
- awk '$1 == "inet" { print $2 }')
- fi
-
- local _rc=0
- ip addr del "$_im" dev "$_iface" || {
- echo "Failed to del $_ip on dev $_iface"
- _rc=1
- }
+ # This could be set globally for all interfaces but it is probably
+ # better to avoid surprises, so limit it the interfaces where CTDB
+ # has public IP addresses. There isn't anywhere else convenient
+ # to do this so just set it each time. This is much cheaper than
+ # remembering and re-adding secondaries.
+ set_proc "sys/net/ipv4/conf/${_iface}/promote_secondaries" 1
- if [ -n "$_secondaries" ] ; then
- for _i in $_secondaries; do
- if ip addr list dev "$_iface" | grep -Fq "inet $_i" ; then
- echo "Kept secondary $_i on dev $_iface"
- else
- echo "Re-adding secondary address $_i to dev $_iface"
- ip addr add $_i brd + dev $_iface || {
- echo "Failed to re-add address $_i to dev $_iface"
- _rc=1
- }
- fi
- done
- fi
-
- return $_rc
- ) <"$_lockfile"
-
- # Do nothing here - return above only gets us out of the subshell
- # and doing anything here will affect the return code.
+ ip addr del "$_ip/$_maskbits" dev "$_iface" || {
+ echo "Failed to del $_ip on dev $_iface"
+ return 1
+ }
}
-# If the given IP is hosted then print 2 items: maskbits and iface
+# If the given IP is hosted then print 2 items: maskbits and iface
ip_maskbits_iface ()
{
_addr="$1"
- ip addr show to "${_addr}/32" 2>/dev/null | \
- awk '$1 == "inet" { print gensub(".*/", "", 1, $2), $NF }'
+ case "$_addr" in
+ *:*) _family="inet6" ; _bits=128 ;;
+ *) _family="inet" ; _bits=32 ;;
+ esac
+
+ ip addr show to "${_addr}/${_bits}" 2>/dev/null | \
+ awk -v family="${_family}" \
+ 'NR == 1 { iface = $2; sub(":$", "", iface) ; \
+ sub("@.*", "", iface) } \
+ $1 ~ /inet/ { mask = $2; sub(".*/", "", mask); \
+ print mask, iface, family }'
}
drop_ip ()
{
_addr="${1%/*}" # Remove optional maskbits
- _log_tag="$2"
set -- $(ip_maskbits_iface $_addr)
if [ -n "$1" ] ; then
_maskbits="$1"
_iface="$2"
- if [ -n "$_log_tag" ] ; then
- script_log "$_log_tag" \
- "Removing public address $_addr/$_maskbits from device $_iface"
- fi
- ip addr del $_addr/$_maskbits dev $_iface >/dev/null 2>&1
+ echo "Removing public address $_addr/$_maskbits from device $_iface"
+ delete_ip_from_iface $_iface $_addr $_maskbits >/dev/null 2>&1
fi
}
drop_all_public_ips ()
{
- _log_tag="$1"
-
while read _ip _x ; do
- drop_ip "$_ip" "$_log_tag"
+ drop_ip "$_ip"
done <"${CTDB_PUBLIC_ADDRESSES:-/dev/null}"
}
+flush_route_cache ()
+{
+ set_proc_maybe sys/net/ipv4/route/flush 1
+ set_proc_maybe sys/net/ipv6/route/flush 1
+}
+
########################################################
-# some simple logic for counting events - per eventscript
-# usage: ctdb_counter_init
-# ctdb_counter_incr
-# ctdb_check_counter_limit <limit>
-# ctdb_check_counter_limit fails when count >= <limit>
-########################################################
+# Simple counters
_ctdb_counter_common () {
_service_name="${1:-${service_name:-${script_name}}}"
- _counter_file="$ctdb_fail_dir/$_service_name"
+ _counter_file="${CTDB_SCRIPT_VARDIR}/failcount/${_service_name}"
mkdir -p "${_counter_file%/*}" # dirname
}
ctdb_counter_init () {
# unary counting!
echo -n 1 >> "$_counter_file"
}
-ctdb_check_counter_limit () {
- _ctdb_counter_common
-
- _limit="${1:-${service_fail_limit}}"
- _quiet="$2"
-
+ctdb_counter_get () {
+ _ctdb_counter_common "$1"
# unary counting!
- _size=$(stat -c "%s" "$_counter_file" 2>/dev/null || echo 0)
- if [ $_size -ge $_limit ] ; then
- echo "ERROR: more than $_limit consecutive failures for $service_name, marking cluster unhealthy"
- exit 1
- elif [ $_size -gt 0 -a -z "$_quiet" ] ; then
- echo "WARNING: less than $_limit consecutive failures ($_size) for $service_name, not unhealthy yet"
- fi
+ stat -c "%s" "$_counter_file" 2>/dev/null || echo 0
}
ctdb_check_counter () {
_msg="${1:-error}" # "error" - anything else is silent on fail
_op="${2:--ge}" # an integer operator supported by test
_limit="${3:-${service_fail_limit}}"
shift 3
- _ctdb_counter_common "$1"
- # unary counting!
- _size=$(stat -c "%s" "$_counter_file" 2>/dev/null || echo 0)
- if [ $_size $_op $_limit ] ; then
+ _size=$(ctdb_counter_get "$1")
+
+ _hit=false
+ if [ "$_op" != "%" ] ; then
+ if [ $_size $_op $_limit ] ; then
+ _hit=true
+ fi
+ else
+ if [ $(($_size $_op $_limit)) -eq 0 ] ; then
+ _hit=true
+ fi
+ fi
+ if $_hit ; then
if [ "$_msg" = "error" ] ; then
- echo "ERROR: $_limit consecutive failures for $_service_name, marking node unhealthy"
+ echo "ERROR: $_size consecutive failures for $_service_name, marking node unhealthy"
exit 1
else
return 1
########################################################
-ctdb_status_dir="$CTDB_VARDIR/status"
-ctdb_fail_dir="$CTDB_VARDIR/failcount"
-
ctdb_setup_service_state_dir ()
{
- service_state_dir="$CTDB_VARDIR/state/${1:-${service_name}}"
+ service_state_dir="${CTDB_SCRIPT_VARDIR}/service_state/${1:-${service_name}}"
mkdir -p "$service_state_dir" || {
echo "Error creating state dir \"$service_state_dir\""
exit 1
########################################################
# Managed status history, for auto-start/stop
-ctdb_managed_dir="$CTDB_VARDIR/managed_history"
-
_ctdb_managed_common ()
{
- _ctdb_managed_file="$ctdb_managed_dir/$service_name"
+ _ctdb_managed_file="${CTDB_SCRIPT_VARDIR}/managed_history/${service_name}"
}
ctdb_service_managed ()
{
_ctdb_managed_common
- mkdir -p "$ctdb_managed_dir"
+ mkdir -p "${_ctdb_managed_file%/*}" # dirname
touch "$_ctdb_managed_file"
}
[ -f "$_ctdb_managed_file" ]
}
-########################################################
-# Check and set status
-
-log_status_cat ()
-{
- echo "node is \"$1\", \"${script_name}\" reports problem: $(cat $2)"
-}
-
-ctdb_checkstatus ()
-{
- if [ -r "$ctdb_status_dir/$script_name/unhealthy" ] ; then
- log_status_cat "unhealthy" "$ctdb_status_dir/$script_name/unhealthy"
- return 1
- elif [ -r "$ctdb_status_dir/$script_name/banned" ] ; then
- log_status_cat "banned" "$ctdb_status_dir/$script_name/banned"
- return 2
- else
- return 0
- fi
-}
-
-ctdb_setstatus ()
-{
- d="$ctdb_status_dir/$script_name"
- case "$1" in
- unhealthy|banned)
- mkdir -p "$d"
- cat "$2" >"$d/$1"
- ;;
- *)
- for i in "banned" "unhealthy" ; do
- rm -f "$d/$i"
- done
- ;;
- esac
-}
-
##################################################################
# Reconfigure a service on demand
_ctdb_service_reconfigure_common ()
{
- _d="$ctdb_status_dir/${service_name}"
+ _d="${CTDB_SCRIPT_VARDIR}/service_status/${service_name}"
mkdir -p "$_d"
_ctdb_service_reconfigure_flag="$_d/reconfigure"
}
:
}
-ctdb_reconfigure_try_lock ()
+ctdb_reconfigure_take_lock ()
{
_ctdb_service_reconfigure_common
_lock="${_d}/reconfigure_lock"
) <"$_lock"
}
+ctdb_reconfigure_release_lock ()
+{
+ _ctdb_service_reconfigure_common
+ _lock="${_d}/reconfigure_lock"
+
+ rm -f "$_lock"
+}
+
ctdb_replay_monitor_status ()
{
echo "Replaying previous status for this script due to reconfigure..."
- # Leading colon (':') is missing in some versions...
- _out=$(ctdb scriptstatus -Y | grep -E "^:?monitor:${script_name}:")
+ # Leading separator ('|') is missing in some versions...
+ _out=$(ctdb scriptstatus -X | grep -E "^\|?monitor\|${script_name}\|")
# Output looks like this:
- # :monitor:60.nfs:1:ERROR:1314764004.030861:1314764004.035514:foo bar:
+ # |monitor|60.nfs|1|ERROR|1314764004.030861|1314764004.035514|foo bar|
# This is the cheapest way of getting fields in the middle.
- set -- $(IFS=":" ; echo $_out)
+ set -- $(IFS="|" ; echo $_out)
_code="$3"
_status="$4"
# The error output field can include colons so we'll try to
# preserve them. The weak checking at the beginning tries to make
- # this work for both broken (no leading ':') and fixed output.
- _out="${_out%:}"
- _err_out="${_out#*monitor:${script_name}:*:*:*:*:}"
+ # this work for both broken (no leading '|') and fixed output.
+ _out="${_out%|}"
+ _err_out="${_out#*monitor|${script_name}|*|*|*|*|}"
case "$_status" in
OK) : ;; # Do nothing special.
TIMEDOUT)
;;
*) : ;; # Must be ERROR, do nothing special.
esac
- echo "$_err_out"
+ if [ -n "$_err_out" ] ; then
+ echo "$_err_out"
+ fi
exit $_code
}
*) return 0 ;;
esac
- if ctdb_reconfigure_try_lock ; then
+ if ctdb_reconfigure_take_lock ; then
# No events covered by this function are running, so proceed
# with gay abandon.
case "$event_name" in
ctdb_service_reconfigure
fi
;;
- monitor)
- if ctdb_service_needs_reconfigure ; then
- ctdb_service_reconfigure
- # Given that the reconfigure might not have
- # resulted in the service being stable yet, we
- # replay the previous status since that's the best
- # information we have.
- ctdb_replay_monitor_status
- fi
- ;;
esac
+
+ ctdb_reconfigure_release_lock
else
# Somebody else is running an event we don't want to collide
# with. We proceed with caution.
# backward compatibility and try again.
ctdb_compat_managed_service "$CTDB_MANAGES_VSFTPD" "vsftpd"
ctdb_compat_managed_service "$CTDB_MANAGES_SAMBA" "samba"
- ctdb_compat_managed_service "$CTDB_MANAGES_SCP" "scp"
ctdb_compat_managed_service "$CTDB_MANAGES_WINBIND" "winbind"
ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD" "apache2"
ctdb_compat_managed_service "$CTDB_MANAGES_HTTPD" "httpd"
ctdb_compat_managed_service "$CTDB_MANAGES_ISCSI" "iscsi"
ctdb_compat_managed_service "$CTDB_MANAGES_CLAMD" "clamd"
ctdb_compat_managed_service "$CTDB_MANAGES_NFS" "nfs"
- ctdb_compat_managed_service "$CTDB_MANAGES_NFS" "nfs-ganesha-gpfs"
t=" $CTDB_MANAGED_SERVICES "
# Default service_start() and service_stop() functions.
-# These may be overridden in an eventscript. When overriding, the
-# following convention must be followed. If these functions are
-# called with no arguments then they may use internal logic to
-# determine whether the service is managed and, therefore, whether
-# they should take any action. However, if the service name is
-# specified as an argument then an attempt must be made to start or
-# stop the service. This is because the auto-start/stop code calls
-# them with the service name as an argument.
+# These may be overridden in an eventscript.
service_start ()
{
service "$service_name" start
ctdb_standard_event_handler ()
{
- case "$1" in
- status)
- ctdb_checkstatus
- exit
- ;;
- setstatus)
- shift
- ctdb_setstatus "$@"
- exit
- ;;
- esac
+ :
}
-# iptables doesn't like being re-entered, so flock-wrap it.
-iptables()
+iptables_wrapper ()
{
- flock -w 30 $CTDB_VARDIR/iptables-ctdb.flock /sbin/iptables "$@"
+ _family="$1" ; shift
+ if [ "$_family" = "inet6" ] ; then
+ _iptables_cmd="ip6tables"
+ else
+ _iptables_cmd="iptables"
+ fi
+
+ # iptables doesn't like being re-entered, so flock-wrap it.
+ flock -w 30 "${CTDB_SCRIPT_VARDIR}/iptables.flock" "$_iptables_cmd" "$@"
}
+# AIX (and perhaps others?) doesn't have mktemp
+if ! type mktemp >/dev/null 2>&1 ; then
+ mktemp ()
+ {
+ _dir=false
+ if [ "$1" = "-d" ] ; then
+ _dir=true
+ shift
+ fi
+ _d="${TMPDIR:-/tmp}"
+ _hex10=$(dd if=/dev/urandom count=20 2>/dev/null | \
+ md5sum | \
+ sed -e 's@\(..........\).*@\1@')
+ _t="${_d}/tmp.${_hex10}"
+ (
+ umask 077
+ if $_dir ; then
+ mkdir "$_t"
+ else
+ >"$_t"
+ fi
+ )
+ echo "$_t"
+ }
+fi
+
########################################################
# tickle handling
########################################################
{
_port="$1"
- tickledir="$CTDB_VARDIR/state/tickles"
+ tickledir="${CTDB_SCRIPT_VARDIR}/tickles"
mkdir -p "$tickledir"
- # Who am I?
- _pnn=$(ctdb pnn) ; _pnn=${_pnn#PNN:}
+ ctdb_get_pnn
# What public IPs do I hold?
- _ips=$(ctdb -Y ip | awk -F: -v pnn=$_pnn '$3 == pnn {print $2}')
+ _ips=$(ctdb -X ip | awk -F'|' -v pnn=$pnn '$3 == pnn {print $2}')
# IPs as a regexp choice
_ipschoice="($(echo $_ips | sed -e 's/ /|/g' -e 's/\./\\\\./g'))"
- # Record connections to our public IPs in a temporary file
- _my_connections="${tickledir}/${_port}.connections"
- rm -f "$_my_connections"
+ # Record connections to our public IPs in a temporary file.
+ # This temporary file is in CTDB's private state directory and
+ # $$ is used to avoid a very rare race involving CTDB's script
+ # debugging. No security issue, nothing to see here...
+ _my_connections="${tickledir}/${_port}.connections.$$"
netstat -tn |
awk -v destpat="^${_ipschoice}:${_port}\$" \
'$1 == "tcp" && $6 == "ESTABLISHED" && $4 ~ destpat {print $5, $4}' |
sort >"$_my_connections"
# Record our current tickles in a temporary file
- _my_tickles="${tickledir}/${_port}.tickles"
- rm -f "$_my_tickles"
+ _my_tickles="${tickledir}/${_port}.tickles.$$"
for _i in $_ips ; do
- ctdb -Y gettickles $_i $_port |
- awk -F: 'NR > 1 { printf "%s:%s %s:%s\n", $2, $3, $4, $5 }'
+ ctdb -X gettickles $_i $_port |
+ awk -F'|' 'NR > 1 { printf "%s:%s %s:%s\n", $2, $3, $4, $5 }'
done |
sort >"$_my_tickles"
ctdb deltickle $_src $_dst
done
- rm -f "$_my_connections" "$_my_tickles"
+ rm -f "$_my_connections" "$_my_tickles"
+
+ # Remove stale files from killed scripts
+ find "$tickledir" -type f -mmin +10 | xargs -r rm
}
########################################################