vmware-nsx/neutron/agent/linux/iptables_firewall.py
Peter Feiner e31c9fd3fe defer firewall updates to iptables data structures
One of two patches that fixes bug #1194438.

In the iptables firewall driver, each port method (update_port_filter,
prepare_port_filter, remove_port_filter) makes O(N) calls, where
N=len(firewall.ports), to IptablesManager methods that update dozens
of data structures. When the firewall methods are called in sequence,
e.g., by SecurityGroupAgentRpcMixin, the calls to IptablesManager's
methods start to add up.

This patch changes IptablesFirewallDriver to defer and coalesce calls
to IptablesManager. Now a sequence of M port method calls results in
O(N) calls to IptablesManager methods instead of O(N*M) as before.

Change-Id: If17eeaec197beae8b8aecffca1f19d4535a7226e
2013-07-29 14:23:19 +00:00

343 lines
14 KiB
Python

# vim: tabstop=4 shiftwidth=4 softtabstop=4
#
# Copyright 2012, Nachi Ueno, NTT MCL, Inc.
# All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
import netaddr
from oslo.config import cfg
from neutron.agent import firewall
from neutron.agent.linux import iptables_manager
from neutron.common import constants
from neutron.openstack.common import log as logging
LOG = logging.getLogger(__name__)
SG_CHAIN = 'sg-chain'
INGRESS_DIRECTION = 'ingress'
EGRESS_DIRECTION = 'egress'
IP_SPOOF_FILTER = 'ip-spoof-filter'
CHAIN_NAME_PREFIX = {INGRESS_DIRECTION: 'i',
EGRESS_DIRECTION: 'o',
IP_SPOOF_FILTER: 's'}
LINUX_DEV_LEN = 14
class IptablesFirewallDriver(firewall.FirewallDriver):
"""Driver which enforces security groups through iptables rules."""
IPTABLES_DIRECTION = {INGRESS_DIRECTION: 'physdev-out',
EGRESS_DIRECTION: 'physdev-in'}
def __init__(self):
self.iptables = iptables_manager.IptablesManager(
root_helper=cfg.CONF.AGENT.root_helper,
use_ipv6=True)
# list of port which has security group
self.filtered_ports = {}
self._add_fallback_chain_v4v6()
self._defer_apply = False
self._pre_defer_filtered_ports = None
@property
def ports(self):
return self.filtered_ports
def prepare_port_filter(self, port):
LOG.debug(_("Preparing device (%s) filter"), port['device'])
self._remove_chains()
self.filtered_ports[port['device']] = port
# each security group has it own chains
self._setup_chains()
self.iptables.apply()
def update_port_filter(self, port):
LOG.debug(_("Updating device (%s) filter"), port['device'])
if port['device'] not in self.filtered_ports:
LOG.info(_('Attempted to update port filter which is not '
'filtered %s'), port['device'])
return
self._remove_chains()
self.filtered_ports[port['device']] = port
self._setup_chains()
self.iptables.apply()
def remove_port_filter(self, port):
LOG.debug(_("Removing device (%s) filter"), port['device'])
if not self.filtered_ports.get(port['device']):
LOG.info(_('Attempted to remove port filter which is not '
'filtered %r'), port)
return
self._remove_chains()
self.filtered_ports.pop(port['device'], None)
self._setup_chains()
self.iptables.apply()
def _setup_chains(self):
"""Setup ingress and egress chain for a port."""
if not self._defer_apply:
self._setup_chains_apply(self.filtered_ports)
def _setup_chains_apply(self, ports):
self._add_chain_by_name_v4v6(SG_CHAIN)
for port in ports.values():
self._setup_chain(port, INGRESS_DIRECTION)
self._setup_chain(port, EGRESS_DIRECTION)
self.iptables.ipv4['filter'].add_rule(SG_CHAIN, '-j ACCEPT')
self.iptables.ipv6['filter'].add_rule(SG_CHAIN, '-j ACCEPT')
def _remove_chains(self):
"""Remove ingress and egress chain for a port."""
if not self._defer_apply:
self._remove_chains_apply(self.filtered_ports)
def _remove_chains_apply(self, ports):
for port in ports.values():
self._remove_chain(port, INGRESS_DIRECTION)
self._remove_chain(port, EGRESS_DIRECTION)
self._remove_chain(port, IP_SPOOF_FILTER)
self._remove_chain_by_name_v4v6(SG_CHAIN)
def _setup_chain(self, port, DIRECTION):
self._add_chain(port, DIRECTION)
self._add_rule_by_security_group(port, DIRECTION)
def _remove_chain(self, port, DIRECTION):
chain_name = self._port_chain_name(port, DIRECTION)
self._remove_chain_by_name_v4v6(chain_name)
def _add_fallback_chain_v4v6(self):
self.iptables.ipv4['filter'].add_chain('sg-fallback')
self.iptables.ipv4['filter'].add_rule('sg-fallback', '-j DROP')
self.iptables.ipv6['filter'].add_chain('sg-fallback')
self.iptables.ipv6['filter'].add_rule('sg-fallback', '-j DROP')
def _add_chain_by_name_v4v6(self, chain_name):
self.iptables.ipv6['filter'].add_chain(chain_name)
self.iptables.ipv4['filter'].add_chain(chain_name)
def _remove_chain_by_name_v4v6(self, chain_name):
self.iptables.ipv4['filter'].ensure_remove_chain(chain_name)
self.iptables.ipv6['filter'].ensure_remove_chain(chain_name)
def _add_rule_to_chain_v4v6(self, chain_name, ipv4_rules, ipv6_rules):
for rule in ipv4_rules:
self.iptables.ipv4['filter'].add_rule(chain_name, rule)
for rule in ipv6_rules:
self.iptables.ipv6['filter'].add_rule(chain_name, rule)
def _get_device_name(self, port):
return port['device']
def _add_chain(self, port, direction):
chain_name = self._port_chain_name(port, direction)
self._add_chain_by_name_v4v6(chain_name)
# Note(nati) jump to the security group chain (SG_CHAIN)
# This is needed because the packet may much two rule in port
# if the two port is in the same host
# We accept the packet at the end of SG_CHAIN.
# jump to the security group chain
device = self._get_device_name(port)
jump_rule = ['-m physdev --%s %s --physdev-is-bridged '
'-j $%s' % (self.IPTABLES_DIRECTION[direction],
device,
SG_CHAIN)]
self._add_rule_to_chain_v4v6('FORWARD', jump_rule, jump_rule)
# jump to the chain based on the device
jump_rule = ['-m physdev --%s %s --physdev-is-bridged '
'-j $%s' % (self.IPTABLES_DIRECTION[direction],
device,
chain_name)]
self._add_rule_to_chain_v4v6(SG_CHAIN, jump_rule, jump_rule)
if direction == EGRESS_DIRECTION:
self._add_rule_to_chain_v4v6('INPUT', jump_rule, jump_rule)
def _split_sgr_by_ethertype(self, security_group_rules):
ipv4_sg_rules = []
ipv6_sg_rules = []
for rule in security_group_rules:
if rule.get('ethertype') == constants.IPv4:
ipv4_sg_rules.append(rule)
elif rule.get('ethertype') == constants.IPv6:
if rule.get('protocol') == 'icmp':
rule['protocol'] = 'icmpv6'
ipv6_sg_rules.append(rule)
return ipv4_sg_rules, ipv6_sg_rules
def _select_sgr_by_direction(self, port, direction):
return [rule
for rule in port.get('security_group_rules', [])
if rule['direction'] == direction]
def _arp_spoofing_rule(self, port):
return ['-m mac ! --mac-source %s -j DROP' % port['mac_address']]
def _setup_ip_spoof_filter_chain(self, port, table, addresses, rules):
if len(addresses) == 1:
rules.append('! -s %s -j DROP' % addresses[0])
elif addresses:
chain_name = self._port_chain_name(port, IP_SPOOF_FILTER)
table.add_chain(chain_name)
for ip in addresses:
table.add_rule(chain_name, '-s %s -j RETURN' % ip)
table.add_rule(chain_name, '-j DROP')
rules.append('-j $%s' % chain_name)
def _ip_spoofing_rule(self, port, ipv4_rules, ipv6_rules):
#Note(nati) allow dhcp or RA packet
ipv4_rules += ['-p udp -m udp --sport 68 --dport 67 -j RETURN']
ipv6_rules += ['-p icmpv6 -j RETURN']
ipv4_addresses = []
ipv6_addresses = []
for ip in port['fixed_ips']:
if netaddr.IPAddress(ip).version == 4:
ipv4_addresses.append(ip)
else:
ipv6_addresses.append(ip)
self._setup_ip_spoof_filter_chain(port, self.iptables.ipv4['filter'],
ipv4_addresses, ipv4_rules)
self._setup_ip_spoof_filter_chain(port, self.iptables.ipv6['filter'],
ipv6_addresses, ipv6_rules)
def _drop_dhcp_rule(self):
#Note(nati) Drop dhcp packet from VM
return ['-p udp -m udp --sport 67 --dport 68 -j DROP']
def _add_rule_by_security_group(self, port, direction):
chain_name = self._port_chain_name(port, direction)
# select rules for current direction
security_group_rules = self._select_sgr_by_direction(port, direction)
# split groups by ip version
# for ipv4, iptables command is used
# for ipv6, iptables6 command is used
ipv4_sg_rules, ipv6_sg_rules = self._split_sgr_by_ethertype(
security_group_rules)
ipv4_iptables_rule = []
ipv6_iptables_rule = []
if direction == EGRESS_DIRECTION:
ipv4_iptables_rule += self._arp_spoofing_rule(port)
ipv6_iptables_rule += self._arp_spoofing_rule(port)
self._ip_spoofing_rule(port,
ipv4_iptables_rule,
ipv6_iptables_rule)
ipv4_iptables_rule += self._drop_dhcp_rule()
ipv4_iptables_rule += self._convert_sgr_to_iptables_rules(
ipv4_sg_rules)
ipv6_iptables_rule += self._convert_sgr_to_iptables_rules(
ipv6_sg_rules)
self._add_rule_to_chain_v4v6(chain_name,
ipv4_iptables_rule,
ipv6_iptables_rule)
def _convert_sgr_to_iptables_rules(self, security_group_rules):
iptables_rules = []
self._drop_invalid_packets(iptables_rules)
self._allow_established(iptables_rules)
for rule in security_group_rules:
# These arguments MUST be in the format iptables-save will
# display them: source/dest, protocol, sport, dport, target
# Otherwise the iptables_manager code won't be able to find
# them to preserve their [packet:byte] counts.
args = self._ip_prefix_arg('s',
rule.get('source_ip_prefix'))
args += self._ip_prefix_arg('d',
rule.get('dest_ip_prefix'))
args += self._protocol_arg(rule.get('protocol'))
args += self._port_arg('sport',
rule.get('protocol'),
rule.get('source_port_range_min'),
rule.get('source_port_range_max'))
args += self._port_arg('dport',
rule.get('protocol'),
rule.get('port_range_min'),
rule.get('port_range_max'))
args += ['-j RETURN']
iptables_rules += [' '.join(args)]
iptables_rules += ['-j $sg-fallback']
return iptables_rules
def _drop_invalid_packets(self, iptables_rules):
# Always drop invalid packets
iptables_rules += ['-m state --state ' 'INVALID -j DROP']
return iptables_rules
def _allow_established(self, iptables_rules):
# Allow established connections
iptables_rules += ['-m state --state RELATED,ESTABLISHED -j RETURN']
return iptables_rules
def _protocol_arg(self, protocol):
if not protocol:
return []
iptables_rule = ['-p', protocol]
# iptables always adds '-m protocol' for udp and tcp
if protocol in ['udp', 'tcp']:
iptables_rule += ['-m', protocol]
return iptables_rule
def _port_arg(self, direction, protocol, port_range_min, port_range_max):
if not (protocol in ['udp', 'tcp'] and port_range_min):
return []
if port_range_min == port_range_max:
return ['--%s' % direction, '%s' % (port_range_min,)]
else:
return ['-m', 'multiport',
'--%ss' % direction,
'%s:%s' % (port_range_min, port_range_max)]
def _ip_prefix_arg(self, direction, ip_prefix):
#NOTE (nati) : source_group_id is converted to list of source_
# ip_prefix in server side
if ip_prefix:
return ['-%s' % direction, ip_prefix]
return []
def _port_chain_name(self, port, direction):
return iptables_manager.get_chain_name(
'%s%s' % (CHAIN_NAME_PREFIX[direction], port['device'][3:]))
def filter_defer_apply_on(self):
if not self._defer_apply:
self.iptables.defer_apply_on()
self._pre_defer_filtered_ports = dict(self.filtered_ports)
self._defer_apply = True
def filter_defer_apply_off(self):
if self._defer_apply:
self._defer_apply = False
self._remove_chains_apply(self._pre_defer_filtered_ports)
self._pre_defer_filtered_ports = None
self._setup_chains_apply(self.filtered_ports)
self.iptables.defer_apply_off()
class OVSHybridIptablesFirewallDriver(IptablesFirewallDriver):
OVS_HYBRID_TAP_PREFIX = 'tap'
def _port_chain_name(self, port, direction):
return iptables_manager.get_chain_name(
'%s%s' % (CHAIN_NAME_PREFIX[direction], port['device']))
def _get_device_name(self, port):
return (self.OVS_HYBRID_TAP_PREFIX + port['device'])[:LINUX_DEV_LEN]