path: root/heartbeat/portblock

#!/bin/sh
#
# portblock: iptables temporary portblocking control 
#
# Author:	Sun Jiang Dong (initial version)
#               Philipp Reisner (per-IP filtering)
#
# License:	GNU General Public License (GPL)
#
# Copyright:	(C) 2005 International Business Machines
#
#	  OCF parameters are as below:
#		OCF_RESKEY_protocol
#		OCF_RESKEY_portno
#		OCF_RESKEY_action
#		OCF_RESKEY_ip
#		OCF_RESKEY_tickle_dir
#		OCF_RESKEY_sync_script
#######################################################################
# Initialization:

: ${OCF_FUNCTIONS_DIR=${OCF_ROOT}/lib/heartbeat}
. ${OCF_FUNCTIONS_DIR}/ocf-shellfuncs

# Defaults
OCF_RESKEY_protocol_default=""
OCF_RESKEY_portno_default=""
OCF_RESKEY_direction_default="in"
OCF_RESKEY_action_default=""
OCF_RESKEY_ip_default="0.0.0.0/0"
OCF_RESKEY_reset_local_on_unblock_stop_default="false"
OCF_RESKEY_tickle_dir_default=""
OCF_RESKEY_sync_script_default=""

: ${OCF_RESKEY_protocol=${OCF_RESKEY_protocol_default}}
: ${OCF_RESKEY_portno=${OCF_RESKEY_portno_default}}
: ${OCF_RESKEY_direction=${OCF_RESKEY_direction_default}}
: ${OCF_RESKEY_action=${OCF_RESKEY_action_default}}
: ${OCF_RESKEY_ip=${OCF_RESKEY_ip_default}}
: ${OCF_RESKEY_reset_local_on_unblock_stop=${OCF_RESKEY_reset_local_on_unblock_stop_default}}
: ${OCF_RESKEY_tickle_dir=${OCF_RESKEY_tickle_dir_default}}
: ${OCF_RESKEY_sync_script=${OCF_RESKEY_sync_script_default}}
#######################################################################
CMD=`basename $0`
TICKLETCP=$HA_BIN/tickle_tcp

usage()
{
	cat <<END >&2
	usage: $CMD {start|stop|status|monitor|meta-data|validate-all}

	$CMD is used to temporarily block ports using iptables.

	It can be used to blackhole a port before bringing
	up an IP address, and enable it after a service is started.
	To do that for samba, the following can be used:

	crm configure <<EOF
	primitive portblock-samba ocf:heartbeat:portblock \\
	  params protocol=tcp portno=137,138 action=block
	primitive portunblock-samba ocf:heartbeat:portblock \\
	  params protocol=tcp portno=137,138 action=unblock
	primitive samba-vip ocf:heartbeat:IPaddr2 \\
	  params ip=10.10.10.20
	group g-samba \\
	  portblock-samba samba-vip nmbd smbd portunblock-samba
	EOF

	This will do the following things:

	  - DROP all incoming packets for TCP ports 137 and 138
	  - Bring up the IP alias 10.10.10.20
	  - start the nmbd and smbd services
	  - Re-enable TCP ports 137 and 138
	        (enable normal firewall rules on those ports)

	This prevents clients from getting TCP RST if they try to reconnect
	to the service after the alias is enabled but before nmbd and smbd
	are running.  These packets will cause some clients to give up
	attempting to reconnect to the server.

	Attempts to connect to UDP and other non-TCP ports which have nothing
	listening can result in ICMP port unreachable responses, which can
	have the same undesirable affect on some clients.

	NOTE: iptables is Linux-specific.

	An additional feature in the portblock RA is the tickle ACK function
	enabled by specifying the tickle_dir parameter. The tickle ACK 
	triggers the clients to faster reconnect their TCP connections to the 
	fail-overed server.

	Please note that this feature is often used for the floating IP fail-
	over scenario where the long-lived TCP connections need to be tickled.
	It doesn't support the cluster alias IP scenario.

	When using the tickle ACK function, in addition to the normal usage
	of portblock RA, the parameter tickle_dir must be specified in the 
	action=unblock instance of the portblock resources.
	For example, you may stack resources like below:
		portblock action=block
		services
		portblock action=unblock tickle_dir=/tickle/state/dir

	If you want to tickle all the TCP connections which connected to _one_
	floating IP but different ports, no matter how many portblock resources 
	you have defined, you should enable tickles for _one_ portblock 
	resource(action=unblock) only.
	
	The tickle_dir is a location which stores the established TCP 
	connections. It can be a shared directory(which is cluster-visible to 
	all nodes) or a local directory.
	If you use the shared directory, you needn't do any other things.
	If you use the local directory, you must also specify the sync_script
	paramater. We recommend you to use csync2 as the sync_script.
	For example, if you use the local directory /tmp/tickle as tickle_dir, 
	you could setup the csync2 as the csync2 documentation says and 
	configure your /etc/csync2/csync2.cfg like:
		group ticklegroup {
		  host node1;
		  host node2;
		  key  /etc/csync2/ticklegroup.key;
		  include /etc/csync2/csync2.cfg;
		  include /tmp/tickle;
		  auto younger;
		}
	Then specify the parameter sync_script as "csync2 -xv".

END
}

meta_data() {
	cat <<END
<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="portblock" version="1.0">
<version>1.0</version>

<longdesc lang="en">
Resource script for portblock. It is used to temporarily block ports 
using iptables. In addition, it may allow for faster TCP reconnects
for clients on failover. Use that if there are long lived TCP
connections to an HA service. This feature is enabled by setting the
tickle_dir parameter and only in concert with action set to unblock.
Note that the tickle ACK function is new as of version 3.0.2 and
hasn't yet seen widespread use.
</longdesc>
<shortdesc lang="en">Block and unblocks access to TCP and UDP ports</shortdesc>

<parameters>
<parameter name="protocol" unique="0" required="1">
<longdesc lang="en">
The protocol used to be blocked/unblocked.
</longdesc>
<shortdesc lang="en">protocol</shortdesc>
<content type="string" default="${OCF_RESKEY_protocol_default}" />
</parameter>

<parameter name="portno" unique="0" required="1">
<longdesc lang="en">
The port number used to be blocked/unblocked.
</longdesc>
<shortdesc lang="en">portno</shortdesc>
<content type="string" default="${OCF_RESKEY_portno_default}" />
</parameter>

<parameter name="action" unique="0" required="1">
<longdesc lang="en">
The action (block/unblock) to be done on the protocol::portno.
</longdesc>
<shortdesc lang="en">action</shortdesc>
<content type="string" default="${OCF_RESKEY_action_default}" />
</parameter>

<parameter name="reset_local_on_unblock_stop" unique="0" required="0">
<longdesc lang="en">
If for some reason the long lived server side TCP sessions won't be cleaned up
by a reconfiguration/flush/stop of whatever services this portblock protects,
they would linger in the connection table, even after the IP is gone
and services have been switched over to another node.

An example would be the default NFS kernel server.

These "known" connections may seriously confuse and delay a later switchback.

Enabling this option will cause this agent to try to get rid of these connections
by injecting a temporary iptables rule to TCP-reset outgoing packets from the
blocked ports, and additionally tickle them locally,
just before it starts to DROP incoming packets on "unblock stop".
</longdesc>
<shortdesc lang="en">(try to) reset server TCP sessions when unblock stops</shortdesc>
<content type="boolean" default="${OCF_RESKEY_reset_local_on_unblock_stop_default}" />
</parameter>

<parameter name="ip" unique="0" required="0">
<longdesc lang="en">
The IP address used to be blocked/unblocked.
</longdesc>
<shortdesc lang="en">ip</shortdesc>
<content type="string" default="${OCF_RESKEY_ip_default}" />
</parameter>

<parameter name="tickle_dir" unique="0" required="0">
<longdesc lang="en">
The shared or local directory (_must_ be absolute path) which 
stores the established TCP connections.
</longdesc>
<shortdesc lang="en">Tickle directory</shortdesc>
<content type="string" default="${OCF_RESKEY_tickle_dir_default}" />
</parameter>

<parameter name="sync_script" unique="0" required="0">
<longdesc lang="en">
If the tickle_dir is a local directory, then the TCP connection state
file has to be replicated to other nodes in the cluster. It can be
csync2 (default), some wrapper of rsync, or whatever. It takes the
file name as a single argument. For csync2, set it to "csync2 -xv".
</longdesc>
<shortdesc lang="en">Connection state file synchronization script</shortdesc>
<content type="string" default="${OCF_RESKEY_sync_script_default}" />
</parameter>

<parameter name="direction" unique="0" required="0">
<longdesc lang="en">
Whether to block incoming or outgoing traffic. Can be either "in",
"out", or "both".
If "in" is used, the incoming ports are blocked on the INPUT chain.
If "out" is used, the outgoing ports are blocked on the OUTPUT chain.
If "both" is used, both the incoming and outgoing ports are blocked.
</longdesc>
<shortdesc lang="en">Whether to block incoming or outgoing traffic, or both</shortdesc>
<content type="string" default="${OCF_RESKEY_direction_default}" />
</parameter>
</parameters>

<actions>
<action name="start" timeout="20s" />
<action name="stop" timeout="20s" />
<action name="status" depth="0" timeout="10s" interval="10s" />
<action name="monitor" depth="0" timeout="10s" interval="10s" />
<action name="meta-data" timeout="5s" />
<action name="validate-all" timeout="5s" />
</actions>
</resource-agent>
END
}


#
#	Because this is the normal usage, we consider "block"
#	resources to be pseudo-resources -- that is, their status can't
#	be reliably determined through external means.
#	This is because we expect an "unblock" resource to come along
#	and disable us -- but we're still in some sense active...
#

#active_grep_pat {udp|tcp} portno,portno ip {d|s}
# d = look for destination ports
# s = look for source ports
active_grep_pat()
{
  w="[ 	][ 	]*"
  any="0\\.0\\.0\\.0/0"
  src=$any dst=$3
  if [ "$4" = "s" ]; then
    local src=$3
    local dst=$any
  fi
  echo "^DROP${w}${1}${w}--${w}${src}${w}${dst}${w}multiport${w}${4}ports${w}${2}$"
}

#chain_isactive  {udp|tcp} portno,portno ip chain
chain_isactive()
{
  [ "$4" = "OUTPUT" ] && ds="s" || ds="d"
  PAT=$(active_grep_pat "$1" "$2" "$3" "$ds")
  $IPTABLES $wait -n -L "$4" | grep "$PAT" >/dev/null
}

# netstat -tn and ss -Htn, split on whitespace and colon,
# look very similar:
#  tcp    0  0  10.43.55.1  675  10.43.9.8  2049  ESTABLISHED
#  ESTAB  0  0  10.43.55.1  675  10.43.9.8  2049
# so we can write one awk script for both
get_established_tcp_connections()
{
	local columns
	if [ -z "$1" ] ; then
		columns='$4,$5, $6,$7'
	else
		# swap local and remote for "tickle_local"
		columns='$6,$7, $4,$5'
	fi
	$ss_or_netstat | awk -F '[:[:space:]]+' '
		( $8 == "ESTABLISHED" || $1 == "ESTAB" ) && $4 == "'$OCF_RESKEY_ip'" \
		{printf "%s:%s\t%s:%s\n", '"$columns"'}'
}

save_tcp_connections()
{
	[ -z "$OCF_RESKEY_tickle_dir" ] && return
	statefile=$OCF_RESKEY_tickle_dir/$OCF_RESKEY_ip
	# If we have _no_ sync script, we probably have a shared
	# (or replicated) directory, and need to fsync, or we might
	# end up with the just truncated file after failover, exactly
	# when we need it.
	#
	# If we _do_ have a sync script, it is not that important whether
	# the local state file is fsync'ed or not, the sync script is
	# responsible to "atomically" communicate the state to the peer(s).
	if [ -z "$OCF_RESKEY_sync_script" ]; then
		get_established_tcp_connections |
		dd of="$statefile".new conv=fsync status=none &&
		mv "$statefile".new "$statefile"
	else
		get_established_tcp_connections > $statefile
		$OCF_RESKEY_sync_script $statefile > /dev/null 2>&1 &
	fi
}

tickle_remote()
{
	[ -z "$OCF_RESKEY_tickle_dir" ] && return
	echo 1 > /proc/sys/net/ipv4/tcp_tw_recycle
	f=$OCF_RESKEY_tickle_dir/$OCF_RESKEY_ip
	[ -r $f ] || return
	$TICKLETCP -n 3 < $f
}

tickle_local()
{
	[ -z "$OCF_RESKEY_tickle_dir" ] && return
	f=$OCF_RESKEY_tickle_dir/$OCF_RESKEY_ip
	[ -r $f ] || return

	# swap "local" and "remote" address,
	# so we tickle ourselves.
	# We set up a REJECT with tcp-reset before we do so, so we get rid of
	# the no longer wanted potentially long lived "ESTABLISHED" connection
	# entries on the IP we are going to delet in a sec.  These would get in
	# the way if we switch-over and then switch-back in quick succession.
	local i
	awk '{ print $2, $1; }' $f | $TICKLETCP
	$ss_or_netstat | grep -Fw $OCF_RESKEY_ip || return
	for i in 0.1 0.5 1 2 4 ; do
		sleep $i
		# now kill what is currently in the list,
		# not what was recorded during last monitor
		get_established_tcp_connections swap | $TICKLETCP
		$ss_or_netstat | grep -Fw $OCF_RESKEY_ip || break
	done
}

SayActive()
{
  echo "$CMD DROP rule [$*]  is running (OK)"
}

SayConsideredActive()
{
  echo "$CMD DROP rule [$*] considered to be running (OK)"
}

SayInactive()
{
  echo "$CMD DROP rule [$*] is inactive"
}

#IptablesStatus  {udp|tcp} portno,portno ip {in|out|both} {block|unblock}
IptablesStatus() {
  local rc
  rc=$OCF_ERR_GENERIC
  is_active=0
  if [ "$4" = "in" ] || [ "$4" = "both" ]; then
    chain_isactive "$1" "$2" "$3" INPUT
    is_active=$?
  fi
  if [ "$4" = "out" ] || [ "$4" = "both" ]; then
    chain_isactive "$1" "$2" "$3" OUTPUT
    r=$?
    [ $r -gt $is_active ] && is_active=$r
  fi
  if [ $is_active -eq 0 ]; then
    case $5 in
      block)
        SayActive $*
        rc=$OCF_SUCCESS
        ;;
      *)
        SayInactive $*
        rc=$OCF_NOT_RUNNING
        ;;
    esac
  else
    case $5 in
      block)
        if ha_pseudo_resource "${OCF_RESOURCE_INSTANCE}" status; then
          SayConsideredActive $*
          rc=$OCF_SUCCESS
        else
          SayInactive $*
          rc=$OCF_NOT_RUNNING
        fi
        ;;
      *)
        if ha_pseudo_resource "${OCF_RESOURCE_INSTANCE}" status; then
          SayActive $*
          #This is only run on real monitor events.
          save_tcp_connections
          rc=$OCF_SUCCESS
        else
          SayInactive $*
          rc=$OCF_NOT_RUNNING
        fi
        ;;
    esac
  fi
  return $rc
}

#DoIptables  {-I|-D} {udp|tcp} portno,portno ip chain
DoIptables()
{
  op=$1 proto=$2 ports=$3 ip=$4 chain=$5
  active=0; chain_isactive "$proto" "$ports" "$ip" "$chain" && active=1
  want_active=0; [ "$op" = "-I" ] && want_active=1
  echo "active: $active want_active: $want_active"
  if [ $active -eq $want_active ] ; then
    : Chain already in desired state
  else
    [ "$chain" = "OUTPUT" ] && ds="s" || ds="d"
    $IPTABLES $wait "$op" "$chain" -p "$proto" -${ds} "$ip" -m multiport --${ds}ports "$ports" -j DROP
  fi
}

#IptablesBLOCK  {udp|tcp} portno,portno ip {in|out|both} {block|unblock}
IptablesBLOCK()
{
  local rc_in=0
  local rc_out=0
  if [ "$4" = "in" ] || [ "$4" = "both" ]; then
    local try_reset=false
    if	[ "$1/$5/$__OCF_ACTION" = tcp/unblock/stop ] &&
    ocf_is_true $reset_local_on_unblock_stop
    then
      try_reset=true
    fi
    if
      chain_isactive "$1" "$2" "$3" INPUT
    then
      : OK -- chain already active
    else
      if $try_reset ; then
        $IPTABLES $wait -I OUTPUT -p "$1" -s "$3" -m multiport --sports "$2" -j REJECT --reject-with tcp-reset
        tickle_local
      fi
      $IPTABLES $wait -I INPUT -p "$1" -d "$3" -m multiport --dports "$2" -j DROP
      rc_in=$?
      if $try_reset ; then
        $IPTABLES $wait -D OUTPUT -p "$1" -s "$3" -m multiport --sports "$2" -j REJECT --reject-with tcp-reset
      fi
    fi
  fi
  if [ "$4" = "out" ] || [ "$4" = "both" ]; then
    DoIptables -I "$1" "$2" "$3" OUTPUT
    rc_out=$?
  fi

  [ $rc_in -gt $rc_out ] && return $rc_in || return $rc_out
}

#IptablesUNBLOCK  {udp|tcp} portno,portno ip {in|out|both}
IptablesUNBLOCK()
{
  if [ "$4" = "in" ] || [ "$4" = "both" ]; then
    DoIptables -D "$1" "$2" "$3" INPUT
  fi
  if [ "$4" = "out" ] || [ "$4" = "both" ]; then
    DoIptables -D "$1" "$2" "$3" OUTPUT
  fi

  return $?
}

#IptablesStart  {udp|tcp} portno,portno ip {in|out|both} {block|unblock}
IptablesStart()
{
  ha_pseudo_resource "${OCF_RESOURCE_INSTANCE}" start
  case $5 in
    block)	IptablesBLOCK "$@";;
    unblock)
		IptablesUNBLOCK "$@"
		rc=$?
		tickle_remote
		#ignore run_tickle_tcp exit code!
		return $rc
		;;
    *)		usage; return 1;
  esac

  return $?
}

#IptablesStop  {udp|tcp} portno,portno ip {in|out|both} {block|unblock}
IptablesStop()
{
  ha_pseudo_resource "${OCF_RESOURCE_INSTANCE}" stop
  case $5 in
    block)	IptablesUNBLOCK "$@";;
    unblock)
		save_tcp_connections
		IptablesBLOCK "$@"
		;;
    *)		usage; return 1;;
  esac

  return $?
}

#
#	Check if the port is valid, this function code is not decent, but works
#
CheckPort() {
#	Examples of valid port: "1080", "1", "0080"
#	Examples of invalid port: "1080bad", "0", "0000", ""
  echo $1 |egrep -qx '[0-9]+(:[0-9]+)?(,[0-9]+(:[0-9]+)?)*'
}

IptablesValidateAll()
{
  check_binary $IPTABLES
  case $protocol in
    tcp|udp)
	;;
    *)
	ocf_log err "Invalid protocol $protocol!"
	exit $OCF_ERR_CONFIGURED
	;;
  esac

  if CheckPort "$portno"; then
	:
  else
	ocf_log err "Invalid port number $portno!"
	exit $OCF_ERR_CONFIGURED
  fi

  if [ -n "$OCF_RESKEY_tickle_dir" ]; then
	if [ x"$action" != x"unblock" ]; then
		ocf_log err "Tickles are only useful with action=unblock!"
		exit $OCF_ERR_CONFIGURED
	fi
	if [ ! -d "$OCF_RESKEY_tickle_dir" ]; then
		ocf_log err "The tickle dir doesn't exist!"
		exit $OCF_ERR_INSTALLED	  	
	fi
  fi

  case $action in
    block|unblock)	
	;;
    *)		
	ocf_log err "Invalid action $action!"
	exit $OCF_ERR_CONFIGURED
	;; 
  esac

  if ocf_is_true $reset_local_on_unblock_stop; then
	if [ $action != unblock ] ; then
		ocf_log err "reset_local_on_unblock_stop is only relevant with action=unblock"
		exit $OCF_ERR_CONFIGURED
	fi
	if [ -z $OCF_RESKEY_tickle_dir ] ; then
		ocf_log warn "reset_local_on_unblock_stop works best with tickle_dir enabled as well"
	fi
  fi

  return $OCF_SUCCESS
}

if
  ( [ $# -ne 1 ] )
then
  usage
  exit $OCF_ERR_ARGS
fi

case $1 in
  meta-data)		meta_data
			exit $OCF_SUCCESS
			;;

  usage)		usage
			exit $OCF_SUCCESS
			;;
  *)			;;
esac

if [ -z "$OCF_RESKEY_protocol" ]; then
  ocf_log err "Please set OCF_RESKEY_protocol"
  exit $OCF_ERR_CONFIGURED
fi 

if [ -z "$OCF_RESKEY_portno" ]; then
  ocf_log err "Please set OCF_RESKEY_portno"
  exit $OCF_ERR_CONFIGURED
fi 

if [ -z "$OCF_RESKEY_action" ]; then
  ocf_log err "Please set OCF_RESKEY_action"
  exit $OCF_ERR_CONFIGURED
fi 

# iptables v1.4.20+ is required to use -w (wait)
version=$(iptables -V | awk -F ' v' '{print $NF}')
ocf_version_cmp "$version" "1.4.19.1"
if [ "$?" -eq "2" ]; then
    wait="-w"
else
    wait=""
fi

protocol=$OCF_RESKEY_protocol
portno=$OCF_RESKEY_portno
direction=$OCF_RESKEY_direction
action=$OCF_RESKEY_action
ip=$OCF_RESKEY_ip
reset_local_on_unblock_stop=$OCF_RESKEY_reset_local_on_unblock_stop


# If "tickle" is enabled, we need to record the list of currently established
# connections during monitor.  Use ss where available, and netstat otherwise.
if [ -n "$OCF_RESKEY_tickle_dir" ] ; then
	if have_binary ss ; then
		ss_or_netstat="ss -Htn"
	elif have_binary netstat ; then
		ss_or_netstat="netstat -tn"
	else
		ocf_log err "Neither ss nor netstat found, but needed to record estblished connections."
		exit $OCF_ERR_INSTALLED
	fi
fi

case $1 in
  start)	
			IptablesStart $protocol $portno $ip $direction $action
			;;

  stop)		
			IptablesStop $protocol $portno $ip $direction $action
			;;

  status|monitor)	
			IptablesStatus $protocol $portno $ip $direction $action
			;;

  validate-all)
			IptablesValidateAll
			;;

  *)			usage
			exit $OCF_ERR_UNIMPLEMENTED
			;;
esac

exit $?