shorewall_code/docs/ProxyARP.xml

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.4//EN"
"http://www.oasis-open.org/docbook/xml/4.4/docbookx.dtd">
<article id="ProxyARP">
  <!--$Id$-->

  <articleinfo>
    <title>Proxy ARP</title>

    <authorgroup>
      <author>
        <firstname>Tom</firstname>

        <surname>Eastep</surname>
      </author>
    </authorgroup>

    <pubdate><?dbtimestamp format="Y/m/d"?></pubdate>

    <copyright>
      <year>2001-2006</year>

      <holder>Thomas M. Eastep</holder>
    </copyright>

    <legalnotice>
      <para>Permission is granted to copy, distribute and/or modify this
      document under the terms of the GNU Free Documentation License, Version
      1.2 or any later version published by the Free Software Foundation; with
      no Invariant Sections, with no Front-Cover, and with no Back-Cover
      Texts. A copy of the license is included in the section entitled
      <quote><ulink url="GnuCopyright.htm">GNU Free Documentation
      License</ulink></quote>.</para>
    </legalnotice>
  </articleinfo>

  <section>
    <title>Overview</title>

    <para>Proxy ARP (RFC 1027) is a way to make a machine physically located
    on one network appear to be logically part of a different physical network
    connected to the same router/firewall. Typically it allows us to hide a
    machine with a public IP address on a private network behind a router, and
    still have the machine appear to be on the public network "in front of"
    the router. The router "proxys" ARP requests and all network traffic to
    and from the hidden machine to make this fiction possible.</para>

    <para>Consider a router with two interface cards, one connected to a
    public network PUBNET and one connected to a private network PRIVNET. We
    want to hide a server machine on the PRIVNET network but have it
    accessible from the PUBNET network. The IP address of the server machine
    lies in the PUBNET network, even though we are placing the machine on the
    PRIVNET network behind the router.</para>

    <para>By enabling proxy ARP on the router, any machine on the PUBNET
    network that issues an ARP "who has" request for the server's MAC address
    will get a proxy ARP reply from the router containing the router's MAC
    address. This tells machines on the PUBNET network that they should be
    sending packets destined for the server via the router. The router
    forwards the packets from the machines on the PUBNET network to the server
    on the PRIVNET network.</para>

    <para>Similarly, when the server on the PRIVNET network issues a "who has"
    request for any machines on the PUBNET network, the router provides its
    own MAC address via proxy ARP. This tells the server to send packets for
    machines on the PUBNET network via the router. The router forwards the
    packets from the server on the PRIVNET network to the machines on the
    PUBNET network.</para>

    <para>The proxy ARP provided by the router allows the server on the
    PRIVNETnetwork to appear to be on the PUBNET network. It lets the router
    pass ARP requests and other network packets in both directions between the
    server machine and the PUBNET network, making the server machine appear to
    be connected to the PUBNET network even though it is on the PRIVNET
    network hidden behind the router.</para>

    <para>Before you try to use this technique, I strongly recommend that you
    read the <ulink url="shorewall_setup_guide.htm">Shorewall Setup
    Guide</ulink>.</para>
  </section>

  <section id="Example">
    <title>Example</title>

    <para>The following figure represents a Proxy ARP environment.</para>

    <graphic align="center" fileref="images/proxyarp.png" />

    <para>Proxy ARP can be used to make the systems with addresses
    130.252.100.18 and 130.252.100.19 appear to be on the upper
    (130.252.100.*) subnet. Assuming that the upper firewall interface is eth0
    and the lower interface is eth1, this is accomplished using the following
    entries in <filename>/etc/shorewall/proxyarp</filename>:</para>

    <programlisting>#ADDRESS          INTERFACE   EXTERNAL  HAVEROUTE  PERSISTENT
130.252.100.18    eth1        eth0      no         yes
130.252.100.19    eth1        eth0      no         yes  </programlisting>

    <para><emphasis role="bold">Be sure that the internal systems
    (130.242.100.18 and 130.252.100.19 in the above example) are not included
    in any specification in <filename>/etc/shorewall/masq</filename> or
    <filename>/etc/shorewall/nat</filename>.</emphasis></para>

    <note>
      <para>I've used an RFC1918 IP address for eth1 - that IP address is
      largely irrelevant (see below).</para>
    </note>

    <para>The lower systems (130.252.100.18 and 130.252.100.19) <emphasis
    role="bold">should have their subnet mask and default gateway configured
    exactly the same way that the Firewall system's eth0 is configured. In
    other words, they should be configured just like they would be if they
    were parallel to the firewall rather than behind it.</emphasis></para>

    <warning>
      <para>Do not add the Proxy ARP'ed address(es) (130.252.100.18 and
      130.252.100.19 in the above example) to the external interface (eth0 in
      this example) of the firewall.</para>
    </warning>

    <note>
      <para>It should be stressed that entries in the proxyarp file do not
      automatically enable traffic between the external network and the
      internal host(s) — such traffic is still subject to your policies and
      rules.</para>
    </note>

    <para>While the address given to the firewall interface is largely
    irrelevant, one approach you can take is to make that address the same as
    the address of your external interface!</para>

    <graphic align="center" fileref="images/proxyarp1.png" />

    <para>In the diagram above, <filename class="devicefile">eth1</filename>
    has been given the address 130.252.100.17, the same as
    <filename>eth0</filename>. Note though that the VLSM is 32 so there is no
    network associated with this address. This is the approach <ulink
    url="XenMyWay.html">that I take with my DMZ</ulink>.</para>

    <para>To permit Internet hosts to connect to the local systems, you use
    ACCEPT rules. For example, if you run a web server on 130.252.100.19 which
    you have configured to be in the <emphasis role="bold">loc</emphasis> zone
    then you would need this entry in /etc/shorewall/rules:</para>

    <programlisting>#ACTION SOURCE          DEST                    PROTO   DEST
#                                                       PORT
ACCEPT  net             loc:130.252.100.19      tcp     80</programlisting>

    <warning>
      <para>Your distribution's network configuration GUI may not be capable
      of configuring a device in this way. It may complain about the duplicate
      address or it may configure the address incorrectly. Here is what the
      above configuration should look like when viewed using
      <command>ip</command> (the line "inet 130.252.100.17/32 scope global
      eth1" is the most important):</para>

      <programlisting>gateway:~# <command>ip addr ls eth1</command>
3: eth1: &lt;BROADCAST,MULTICAST,UP&gt; mtu 1500 qdisc pfifo_fast qlen 1000
    link/ether 00:a0:cc:d1:db:12 brd ff:ff:ff:ff:ff:ff
    <emphasis role="bold">inet 130.252.100.17/32 scope global eth1</emphasis>
gateway:~#</programlisting>

      <para>Note in particular that there is no broadcast address. Here is an
      <filename>ifcfg-eth-id-00:a0:cc:d1:db:12</filename> file from SUSE that
      produces this result (Note: SUSE ties the configuration file to the card
      by embedding the card's MAC address in the file name):</para>

      <programlisting>BOOTPROTO='static'
BROADCAST='130.252.100.17'
IPADDR='130.252.100.17'
MTU=''
NETMASK='255.255.255.255'
NETWORK='130.252.100.17'
REMOTE_IPADDR=''
STARTMODE='onboot'
UNIQUE='8otl.IPwRm6bNMRD'
_nm_name='bus-pci-0000:00:04.0'</programlisting>

      <para>Here is an excerpt from a Debian /etc/network/interfaces file that
      does the same thing:</para>

      <programlisting>...
auto eth1
iface eth1 inet static
        address 130.252.100.17
        netmask 255.255.255.255
        broadcast 0.0.0.0
...</programlisting>
    </warning>
  </section>

  <section id="ARP">
    <title>ARP cache</title>

    <para>A word of warning is in order here. ISPs typically configure their
    routers with a long ARP cache timeout. If you move a system from parallel
    to your firewall to behind your firewall with Proxy ARP, it will probably
    be <emphasis role="bold">HOURS</emphasis> before that system can
    communicate with the Internet.</para>

    <para>If you sniff traffic on the firewall's external interface, you can
    see incoming traffic for the internal system(s) but the traffic is never
    sent out the internal interface.</para>

    <para>You can determine if your ISP's gateway ARP cache is stale using
    ping and tcpdump. Suppose that we suspect that the gateway router has a
    stale ARP cache entry for 130.252.100.19. On the firewall, run tcpdump as
    follows:</para>

    <programlisting>tcpdump -nei eth0 icmp</programlisting>

    <para>Now from 130.252.100.19, ping the ISP's gateway (which we will
    assume is 130.252.100.254):</para>

    <programlisting>ping 130.252.100.254</programlisting>

    <para>We can now observe the tcpdump output:</para>

    <programlisting>13:35:12.159321 <emphasis role="bold">0:4:e2:20:20:33</emphasis> 0:0:77:95:dd:19 ip 98: 130.252.100.19 &gt; 130.252.100.254: icmp: echo request (DF)
13:35:12.207615 0:0:77:95:dd:19 <emphasis role="bold">0:c0:a8:50:b2:57 </emphasis>ip 98: 130.252.100.254 &gt; 130.252.100.19 : icmp: echo reply</programlisting>

    <para>Notice that the source MAC address in the echo request is different
    from the destination MAC address in the echo reply!! In this case
    0:4:e2:20:20:33 was the MAC of the firewall's eth0 NIC while
    0:c0:a8:50:b2:57 was the MAC address of the system on the lower left. In
    other words, the gateway's ARP cache still associates 130.252.100.19 with
    the NIC in that system rather than with the firewall's eth0.</para>

    <para>If you have this problem, there are a couple of things that you can
    try:</para>

    <orderedlist>
      <listitem>
        <para>A reading of <citetitle>TCP/IP Illustrated, Vol 1</citetitle> by
        Stevens reveals<footnote>
            <para>Courtesy of Bradey Honsinger</para>
          </footnote> that a <quote>gratuitous</quote> ARP packet should cause
        the ISP's router to refresh their ARP cache (section 4.7). A
        gratuitous ARP is simply a host requesting the MAC address for its own
        IP; in addition to ensuring that the IP address isn't a
        duplicate...</para>

        <blockquote>
          <para>if the host sending the gratuitous ARP has just changed its
          hardware address..., this packet causes any other host...that has an
          entry in its cache for the old hardware address to update its ARP
          cache entry accordingly.</para>
        </blockquote>

        <para>Which is, of course, exactly what you want to do when you switch
        a host from being exposed to the Internet to behind Shorewall using
        proxy ARP (or one-to-one NAT for that matter). Happily enough, recent
        versions of Redhat's iputils package include <quote>arping</quote>,
        whose <quote>-U</quote> flag does just that:</para>

        <programlisting>arping -U -I &lt;<emphasis>net if</emphasis>&gt; &lt;<emphasis>newly proxied IP</emphasis>&gt;
arping -U -I eth0 66.58.99.83             # for example</programlisting>

        <para>Stevens goes on to mention that not all systems respond
        correctly to gratuitous ARPs, but googling for <quote>arping
        -U</quote> seems to support the idea that it works most of the
        time.</para>

        <para>To use arping with Proxy ARP in the above example, you would
        have to:</para>

        <programlisting>shorewall clear
ip addr add 130.252.100.18 dev eth0
ip addr add 130.252.100.19 dev eth0
arping -U -c 10 -I eth0 130.252.100.18
arping -U -c 10 -I eth0 130.252.100.19
ip addr del 130.252.100.18 dev eth0
ip addr del 130.252.100.19 dev eth0
shorewall start</programlisting>
      </listitem>

      <listitem>
        <para>You can call your ISP and ask them to purge the stale ARP cache
        entry but many either can't or won't purge individual entries.</para>
      </listitem>
    </orderedlist>

    <warning>
      <para>There are two distinct versions of <command>arping</command>
      available:</para>

      <orderedlist>
        <listitem>
          <para><command>arping</command> by Thomas Habets (Debian package
          <emphasis>arping</emphasis>).</para>
        </listitem>

        <listitem>
          <para><command>arping</command> as part of the iputils package by
          Alexey Kuznetsov (Debian package
          <emphasis>iputils-arping</emphasis>).</para>
        </listitem>
      </orderedlist>

      <para>You want the second one by Alexey Kuznetsov.</para>
    </warning>
  </section>

  <section>
    <title>IPv6 - Proxy NDP</title>

    <para>The IPv6 analog of Proxy ARP is Proxy NDP (Neighbor Discovery
    Protocol). Begiinning with Shorewall 4.4.16, Shorewall6 supports Proxy NDP
    in a manner similar to Proxy ARP support in Shorewall:</para>

    <itemizedlist>
      <listitem>
        <para>The configuration file is /etc/shorewall6/proxyndp (see <ulink
        url="manpages6/shorewall6-proxyndp.html">shorewall6-proxyndp
        </ulink>(5)).</para>
      </listitem>

      <listitem>
        <para>The ADDRESS column of that file contains an IPv6 address.</para>
      </listitem>
    </itemizedlist>

    <para>It should be noted that IPv6 implements a "strong host model"
    whereas Linux IPv4 implements a "weak host model". In the strong model, IP
    addresses are associated with interfaces; in the weak model, they are
    associated with the host. This is relevant with respect to Proxy NDP in
    that a multi-homed Linux IPv6 host will only respond to neighbor
    discoverey requests for IPv6 addresses configured on the interface
    receiving the request. So if eth0 has address 2001:470:b:227::44/128 and
    eth1 has address 2001:470:b:227::1/64 then in order for eth1 to respond to
    neighbor discovery requests for 2001:470:b:227::44, the following entry in
    /etc/shorewall6/proxyndp is required:</para>

    <programlisting>#ADDRESS	          INTERFACE    EXTERNAL    HAVEROUTE    PERSISTENT
2001:470:b:227::44 -            eth1        Yes</programlisting>

    <para>A practical application is shown in the Linux <ulink
    url="Vserver.html#NDP">Vserver article</ulink>.</para>
  </section>
</article>