#!/bin/sh

[ -n "$CTDB_BASE" ] || \
    export CTDB_BASE=$(cd -P $(dirname "$0") ; dirname "$PWD")

. $CTDB_BASE/functions
loadconfig

service_name=per_ip_routing

# Do nothing if unconfigured 
[ -n "$CTDB_PER_IP_ROUTING_CONF" ] || exit 0

table_id_prefix="ctdb."

[ -n "$CTDB_PER_IP_ROUTING_RULE_PREF" ] || \
    die "error: CTDB_PER_IP_ROUTING_RULE_PREF not configured"

[ "$CTDB_PER_IP_ROUTING_TABLE_ID_LOW" -lt "$CTDB_PER_IP_ROUTING_TABLE_ID_HIGH" ] 2>/dev/null || \
    die "error: CTDB_PER_IP_ROUTING_TABLE_ID_LOW[$CTDB_PER_IP_ROUTING_TABLE_ID_LOW] and/or CTDB_PER_IP_ROUTING_TABLE_ID_HIGH[$CTDB_PER_IP_ROUTING_TABLE_ID_HIGH] improperly configured"

if [ "$CTDB_PER_IP_ROUTING_TABLE_ID_LOW" -le 253 -a \
    255 -le "$CTDB_PER_IP_ROUTING_TABLE_ID_HIGH" ] ; then
    die "error: range CTDB_PER_IP_ROUTING_TABLE_ID_LOW[$CTDB_PER_IP_ROUTING_TABLE_ID_LOW]..CTDB_PER_IP_ROUTING_TABLE_ID_HIGH[$CTDB_PER_IP_ROUTING_TABLE_ID_HIGH] must not include 253-255"
fi

have_link_local_config ()
{
    [ "$CTDB_PER_IP_ROUTING_CONF" = "__auto_link_local__" ]
}

if ! have_link_local_config && [ ! -r "$CTDB_PER_IP_ROUTING_CONF" ] ; then
    die "error: CTDB_PER_IP_ROUTING_CONF=$CTDB_PER_IP_ROUTING_CONF file not found"
fi

######################################################################

ipv4_is_valid_addr()
{
    _ip="$1"

    _count=0
    # Get the shell to break up the address into 1 word per octet 
    for _o in $(export IFS="." ; echo $_ip) ; do
	# The 2>/dev/null stops output from failures where an "octet"
	# is not numeric.  The test will still fail.
	if ! [ 0 -le $_o -a $_o -le 255 ] 2>/dev/null ; then
	    return 1
	fi
	_count=$(($_count + 1))
    done

    # A valid IPv4 address has 4 octets
    [ $_count -eq 4 ]
}

ensure_ipv4_is_valid_addr ()
{
    _event="$1"
    _ip="$2"

    ipv4_is_valid_addr "$_ip" || {
	echo "$0: $_event not an ipv4 address skipping IP:$_ip"
	exit 0
    }
}

ipv4_host_addr_to_net ()
{
    _host="$1"
    _maskbits="$2"

    # Convert the host address to an unsigned long by splitting out
    # the octets and doing the math.
    _host_ul=0
    for _o in $(export IFS="." ; echo $_host) ; do
	_host_ul=$(( ($_host_ul << 8) + $_o)) # work around Emacs color bug
    done

    # Calculate the mask and apply it.
    _mask_ul=$(( 0xffffffff << (32 - $_maskbits) ))
    _net_ul=$(( $_host_ul & $_mask_ul ))
 
    # Now convert to a network address one byte at a time.
    _net=""
    for _o in $(seq 1 4) ; do
	_net="$(($_net_ul & 255))${_net:+.}${_net}"
	_net_ul=$(($_net_ul >> 8))
    done

    echo "${_net}/${_maskbits}"
}

######################################################################

# Setup a table id to use for the given IP.  We don't need to know it,
# it just needs to exist in /etc/iproute2/rt_tables.  Fail if no free
# table id could be found in the configured range.
ensure_table_id_for_ip ()
{
    _ip=$1

    _f="$CTDB_ETCDIR/iproute2/rt_tables"
    # This file should always exist, but...
    if [ ! -f "$_f" ] ; then
	mkdir -p $(dirname "$_f")
	touch "$_f"
    fi

    # Maintain a table id for each IP address we've ever seen in
    # rt_tables.  We use a "ctdb." prefix on the label.
    _label="${table_id_prefix}${_ip}"

    # This finds either the table id corresponding to the label or a
    # new unused one (that is greater than all the used ones in the
    # range).
    (
	# Note that die() just gets us out of the subshell...
	flock --timeout 30 0 || \
	    die "ensure_table_id_for_ip: failed to lock file $_f"

	_new=$CTDB_PER_IP_ROUTING_TABLE_ID_LOW
	while read _t _l ; do
	    # Skip comments
	    case "$_t" in
		\#*) continue ;;
	    esac
	    # Found existing: done
	    if [ "$_l" = "$_label" ] ; then
		return 0
	    fi
	    # Potentially update the new table id to be used.  The
	    # redirect stops error spam for a non-numeric value.
	    if [ $_new -le $_t -a \
		$_t -le $CTDB_PER_IP_ROUTING_TABLE_ID_HIGH ] 2>/dev/null ; then
		_new=$(($_t + 1))
	    fi
	done

	# If the new table id is legal then add it to the file and
	# print it.
	if [ $_new -le $CTDB_PER_IP_ROUTING_TABLE_ID_HIGH ] ; then
	    printf "%d\t%s\n" "$_new" "$_label" >>"$_f"
	    return 0
	else
	    return 1
	fi
    ) <"$_f"
}

# Clean up all the table ids that we might own.
clean_up_table_ids ()
{
    _f="$CTDB_ETCDIR/iproute2/rt_tables"
    # Even if this didn't exist on the system, adding a route will
    # have created it.  What if we startup and immediately shutdown?
    if [ ! -f "$_f" ] ; then
	mkdir -p $(dirname "$_f")
	touch "$_f"
    fi

    (
	# Note that die() just gets us out of the subshell...
	flock --timeout 30 0 || \
	    die "clean_up_table_ids: failed to lock file $_f"

	# Delete any items from the file that have a table id in our
	# range or a label matching our label.  Preserve comments.
	_tmp="${_f}.$$.ctdb"
	awk -v min="$CTDB_PER_IP_ROUTING_TABLE_ID_LOW" \
	    -v max="$CTDB_PER_IP_ROUTING_TABLE_ID_HIGH" \
	    -v pre="$table_id_prefix" \
	    '/^#/ || \
	     !(min <= $1 && $1 <= max) && \
	     !(index($2, pre) == 1) \
	     { print $0 }' "$_f" >"$_tmp"

	mv "$_tmp" "$_f"
	# The lock is gone - don't do anything else here
    ) <"$_f"
}

######################################################################

# This prints the config for an IP, which is either relevant entries
# from the config file or, if set to the magic link local value, some
# link local routing config for the IP.
get_config_for_ip ()
{
    _ip="$1"

    if have_link_local_config ; then
	# When parsing public_addresses also split on '/'.  This means
	# that we get the maskbits as item #2 without further parsing.
	while IFS="/$IFS" read _i _maskbits _x ; do
	    if [ "$_ip" = "$_i" ] ; then
		echo -n "$_ip "; ipv4_host_addr_to_net "$_ip" "$_maskbits"
	    fi
	done <"${CTDB_PUBLIC_ADDRESSES:-/dev/null}"
    else
	while read _i _rest ; do
	    if [ "$_ip" = "$_i" ] ; then
		printf "%s\t%s\n" "$_ip" "$_rest"
	    fi
	done <"$CTDB_PER_IP_ROUTING_CONF"
    fi
}

ip_has_configuration ()
{
    _ip="$1"

    [ -n "$(get_config_for_ip $_ip)" ]
}

add_routing_for_ip ()
{
    _iface="$1"
    _ip="$2"

    # Do nothing if no config for this IP.
    ip_has_configuration "$_ip" || return 0

    ensure_table_id_for_ip "$_ip" || \
	die "add_routing_for_ip: out of table ids in range $CTDB_PER_IP_ROUTING_TABLE_ID_LOW - $CTDB_PER_IP_ROUTING_TABLE_ID_HIGH"

    _pref="$CTDB_PER_IP_ROUTING_RULE_PREF"
    _table_id="${table_id_prefix}${_ip}"

    del_routing_for_ip "$_ip" >/dev/null 2>&1

    ip rule add from "$_ip" pref "$_pref" table "$_table_id" || \
	die "add_routing_for_ip: failed to add rule for $_ip"

    # Add routes to table for any lines matching the IP.
    get_config_for_ip "$_ip" |
    while read _i _dest _gw ; do
	_r="$_dest ${_gw:+via} $_gw dev $_iface table $_table_id"
	ip route add $_r || \
	    die "add_routing_for_ip: failed to add route: $_r"
    done
}

del_routing_for_ip ()
{
    _ip="$1"

    _pref="$CTDB_PER_IP_ROUTING_RULE_PREF"
    _table_id="${table_id_prefix}${_ip}"

    # Do this unconditionally since we own any matching table ids.
    # However, print a meaningful message if something goes wrong.
    _cmd="ip rule del from $_ip pref $_pref table $_table_id"
    _out=$($_cmd 2>&1) || \
	cat <<EOF
WARNING: Failed to delete policy routing rule
  Command "$_cmd" failed:
  $_out
EOF
    # This should never usually fail, so don't redirect output.
    # However, it can fail when deleting a rogue IP, since there will
    # be no routes for that IP.  In this case it should only fail when
    # the rule deletion above has already failed because the table id
    # is invalid.  Therefore, go to a little bit of trouble to indent
    # the failure message so that it is associated with the above
    # warning message and doesn't look too nasty.
    ip route flush table $_table_id 2>&1 | sed -e 's@^.@  &@'
}

######################################################################

flush_rules_and_routes ()
{
	ip rule show |
	while read _p _x _i _x _t ; do
	    # Remove trailing colon after priority/preference.
	    _p="${_p%:}"
	    # Only remove rules that match our priority/preference.
	    [ "$CTDB_PER_IP_ROUTING_RULE_PREF" = "$_p" ] || continue

	    echo "Removing ip rule for public address $_i for routing table $_t"
	    ip rule del from "$_i" table "$_t" pref "$_p"
	    ip route flush table "$_t" 2>/dev/null
	done
}

# Add any missing routes.  Some might have gone missing if, for
# example, all IPs on the network were removed (possibly if the
# primary was removed).  If $1 is "force" then (re-)add all the
# routes.
add_missing_routes ()
{
    ctdb ip -v -X | {
	read _x # skip header line

	# Read the rest of the lines.  We're only interested in the
	# "IP" and "ActiveInterface" columns.  The latter is only set
	# for addresses local to this node, making it easy to skip
	# non-local addresses.  For each IP local address we check if
	# the relevant routing table is populated and populate it if
	# not.
	while IFS="|" read _x _ip _x _iface _x ; do
	    [ -n "$_iface" ] || continue
	    
	    _table_id="${table_id_prefix}${_ip}"
	    if [ -z "$(ip route show table $_table_id 2>/dev/null)" -o \
		"$1" = "force" ]  ; then
		add_routing_for_ip "$_iface" "$_ip"
	    fi
	done
    } || exit $?
}

# Remove rules/routes for addresses that we're not hosting.  If a
# releaseip event failed in an earlier script then we might not have
# had a chance to remove the corresponding rules/routes.
remove_bogus_routes ()
{
    # Get a IPs current hosted by this node, each anchored with '@'.
    _ips=$(ctdb ip -v -X | awk -F'|' 'NR > 1 && $4 != "" {printf "@%s@\n", $2}')

    ip rule show |
    while read _p _x _i _x _t ; do
	# Remove trailing colon after priority/preference.
	_p="${_p%:}"
	# Only remove rules that match our priority/preference.
	[ "$CTDB_PER_IP_ROUTING_RULE_PREF" = "$_p" ] || continue
	# Only remove rules for which we don't have an IP.  This could
	# be done with grep, but let's do it with shell prefix removal
	# to avoid unnecessary processes.  This falls through if
	# "@${_i}@" isn't present in $_ips.
	[ "$_ips" = "${_ips#*@${_i}@}" ] || continue

	echo "Removing ip rule/routes for unhosted public address $_i"
	del_routing_for_ip "$_i"
    done
}

######################################################################

service_reconfigure ()
{
    add_missing_routes "force"
    remove_bogus_routes

    # flush our route cache
    set_proc sys/net/ipv4/route/flush 1
}

######################################################################

ctdb_check_args "$@"

ctdb_service_check_reconfigure

case "$1" in
    startup)
	flush_rules_and_routes

	# make sure that we only respond to ARP messages from the NIC
	# where a particular ip address is associated.
	get_proc sys/net/ipv4/conf/all/arp_filter >/dev/null 2>&1 && {
	    set_proc sys/net/ipv4/conf/all/arp_filter 1
	}
	;;

    shutdown)
	flush_rules_and_routes
	clean_up_table_ids
	;;

    takeip)
	iface=$2
	ip=$3
	maskbits=$4

	ensure_ipv4_is_valid_addr "$1" "$ip"
	add_routing_for_ip "$iface" "$ip"

	# flush our route cache
	set_proc sys/net/ipv4/route/flush 1

	ctdb gratiousarp "$ip" "$iface"
	;;

    updateip)
	oiface=$2
	niface=$3
	ip=$4
	maskbits=$5

	ensure_ipv4_is_valid_addr "$1" "$ip"
	add_routing_for_ip "$niface" "$ip"

	# flush our route cache
	set_proc sys/net/ipv4/route/flush 1

	ctdb gratiousarp "$ip" "$niface"
	tickle_tcp_connections "$ip"
	;;

    releaseip)
	iface=$2
	ip=$3
	maskbits=$4

	ensure_ipv4_is_valid_addr "$1" "$ip"
	del_routing_for_ip "$ip"
	;;

    ipreallocated)
	add_missing_routes
	remove_bogus_routes
	;;

    *)
	ctdb_standard_event_handler "$@"
	;;
esac

exit 0