<!doctype linuxdoc system>

<!-- This is the Linux NAT HOWTO.
 -->

<!-- $Id: NAT-HOWTO.sgml,v 1.18 2002/01/14 09:35:13 laforge Exp $ -->

<article>

<!-- Title information -->

<title>Linux 2.4 NAT HOWTO
<author>Rusty Russell, mailing list <tt>netfilter@lists.samba.org</tt>
<date>$Revision: 1.18 $ $Date: 2002/01/14 09:35:13 $
<abstract>
This document describes how to do masquerading, transparent proxying,
port forwarding, and other forms of Network Address Translations with
the 2.4 Linux Kernels.
</abstract>

<!-- Table of contents -->
<toc>

<!-- Begin the document -->

<sect>Introduction<label id="intro">

<p>
Welcome, gentle reader.

<p>
You are about to delve into the fascinating (and sometimes horrid)
world of NAT: Network Address Translation, and this HOWTO is going to
be your somewhat accurate guide to the 2.4 Linux Kernel and beyond.

<p>In Linux 2.4, an infrastructure for mangling packets was
introduced, called `netfilter'.  A layer on top of this provides NAT,
completely reimplemented from previous kernels.

<p>(C) 2000 Paul `Rusty' Russell.  Licensed under the GNU GPL.

<sect>Where is the official Web Site and List?

<p>There are three official sites:
<itemize>
<item>Thanks to <url url="http://netfilter.filewatcher.org/" name="Filewatcher">.
<item>Thanks to <url url="http://netfilter.samba.org/" name="The Samba Team and SGI">.
<item>Thanks to <url url="http://netfilter.gnumonks.org/" name="Harald Welte">.
</itemize>

<p>You can reach all of them using round-robin DNS via
<url url="http://www.netfilter.org/"> and <url url="http://www.iptables.org/">

<p>For the official netfilter mailing list, see
<url url="http://www.netfilter.org/contact.html#list" name="netfilter List">.

<sect1>What is Network Address Translation?

<p>
Normally, packets on a network travel from their source (such as your
home computer) to their destination (such as www.gnumonks.org)
through many different links: about 19 from where I am in Australia.
None of these links really alter your packet: they just send it
onward.

<p>
If one of these links were to do NAT, then they would alter the source
or destinations of the packet as it passes through.  As you can
imagine, this is not how the system was designed to work, and hence
NAT is always something of a crock.  Usually the link doing NAT will
remember how it mangled a packet, and when a reply packet passes
through the other way, it will do the reverse mangling on that reply
packet, so everything works.

<sect1>Why Would I Want To Do NAT?

<p>In a perfect world, you wouldn't.  Meanwhile, the main reasons are:

<descrip>
<tag/Modem Connections To The Internet/ Most ISPs give you a single IP
address when you dial up to them.  You can send out packets with any
source address you want, but only replies to packets with this source
IP address will return to you.  If you want to use multiple different
machines (such as a home network) to connect to the Internet through
this one link, you'll need NAT.

<p>This is by far the most common use of NAT today, commonly known as
`masquerading' in the Linux world.  I call this SNAT, because you
change the <bf>source</bf> address of the first packet.

<tag/Multiple Servers/ Sometimes you want to change where packets
heading into your network will go.  Frequently this is because (as
above), you have only one IP address, but you want people to be able
to get into the boxes behind the one with the `real' IP address.  If
you rewrite the destination of incoming packets, you can manage this.
This type of NAT was called port-forwarding under previous versions of
Linux.

<p>A common variation of this is load-sharing, where the mapping
ranges over a set of machines, fanning packets out to them.  If you're
doing this on a serious scale, you may want to look at

<url url="http://linuxvirtualserver.org/" name="Linux Virtual Server">.

<tag/Transparent Proxying/ Sometimes you want to pretend that each
packet which passes through your Linux box is destined for a program
on the Linux box itself.  This is used to make transparent proxies: a
proxy is a program which stands between your network and the outside
world, shuffling communication between the two.  The transparent part
is because your network won't even know it's talking to a proxy,
unless of course, the proxy doesn't work.

<p>Squid can be configured to work this way, and it is called
redirection or transparent proxying under previous Linux versions.
</descrip>

<sect>The Two Types of NAT

<p>I divide NAT into two different types: <bf>Source NAT</bf> (SNAT)
and <bf>Destination NAT</bf> (DNAT).

<p>Source NAT is when you alter the source address of the first
packet: i.e. you are changing where the connection is coming from.
Source NAT is always done post-routing, just before the packet goes
out onto the wire.  Masquerading is a specialized form of SNAT.

<p>Destination NAT is when you alter the destination address of the
first packet: i.e. you are changing where the connection is going to.
Destination NAT is always done before routing, when the packet first
comes off the wire.  Port forwarding, load sharing, and transparent
proxying are all forms of DNAT.

<sect>Quick Translation From 2.0 and 2.2 Kernels

<p>Sorry to those of you still shell-shocked from the 2.0 (ipfwadm) to
2.2 (ipchains) transition.  There's good and bad news.

<p>Firstly, you can simply use ipchains and ipfwadm as before.  To do
this, you need to insmod the `ipchains.o' or `ipfwadm.o' kernel
modules found in the latest netfilter distribution.  These are
mutually exclusive (you have been warned), and should not be combined
with any other netfilter modules.

<p>Once one of these modules is installed, you can use ipchains and
ipfwadm as normal, with the following differences:

<itemize>
<item> Setting the masquerading timeouts with ipchains -M -S, or
	ipfwadm -M -s does nothing.  Since the timeouts are longer for
	the new NAT infrastructure, this should not matter.

<item> The init_seq, delta and previous_delta fields in the verbose
	masquerade listing are always zero.

<item> Zeroing and listing the counters at the same time `-Z -L' does
        not work any more: the counters will not be zeroed.

<item> The backward compatibility layer doesn't scale very well for
	large numbers of connections: don't use it for your corporate
	gateway!
</itemize>

Hackers may also notice:

<itemize>
<item> You can now bind to ports 61000-65095 even if you're
	masquerading.  The masquerading code used to assume anything
	in this range was fair game, so programs couldn't use it.

<item> The (undocumented) `getsockname' hack, which transparent proxy
	programs could use to find out the real destinations of
	connections no longer works.

<item> The (undocumented) bind-to-foreign-address hack is also not
	implemented; this was used to complete the illusion of
	transparent proxying.

</itemize>

<sect1> I just want masquerading!  Help!

<p>This is what most people want.  If you have a dynamically allocated
IP PPP dialup (if you don't know, this is you), you simply want to
tell your box that all packets coming from your internal network
should be made to look like they are coming from the PPP dialup box.

<tscreen><verb>
# Load the NAT module (this pulls in all the others).
modprobe iptable_nat

# In the NAT table (-t nat), Append a rule (-A) after routing
# (POSTROUTING) for all packets going out ppp0 (-o ppp0) which says to
# MASQUERADE the connection (-j MASQUERADE).
iptables -t nat -A POSTROUTING -o ppp0 -j MASQUERADE

# Turn on IP forwarding
echo 1 > /proc/sys/net/ipv4/ip_forward
</verb></tscreen>

Note that you are not doing any packet filtering here: for that, see
the Packet Filtering HOWTO: `Mixing NAT and Packet Filtering'.

<sect1> What about ipmasqadm?

<p>This is a much more niche user base, so I didn't worry about
backward compatibility as much.  You can simply use `iptables -t nat'
to do port forwarding.  So for example, in Linux 2.2 you might have
done:

<tscreen><verb>
# Linux 2.2
# Forward TCP packets going to port 8080 on 1.2.3.4 to 192.168.1.1's port 80
ipmasqadm portfw -a -P tcp -L 1.2.3.4 8080 -R 192.168.1.1 80
</verb></tscreen>

Now you would do:

<tscreen><verb>
# Linux 2.4
# Append a rule before routing (-A PREROUTING) to the NAT table (-t nat) that
# TCP packets (-p tcp) going to 1.2.3.4 (-d 1.2.3.4) port 8080 (--dport 8080)
# have their destination mapped (-j DNAT) to 192.168.1.1, port 80
# (--to 192.168.1.1:80).
iptables -A PREROUTING -t nat -p tcp -d 1.2.3.4 --dport 8080 \
        -j DNAT --to 192.168.1.1:80
</verb></tscreen>

<sect>Controlling What To NAT

<p>You need to create NAT rules which tell the kernel what connections
to change, and how to change them.  To do this, we use the very
versatile <tt>iptables</tt> tool, and tell it to alter the NAT table by
specifying the `-t nat' option.

<p>The table of NAT rules contains three lists called `chains': each
rule is examined in order until one matches.  The two chains are
called PREROUTING (for Destination NAT, as packets first come in), and
POSTROUTING (for Source NAT, as packets leave).  The third (OUTPUT)
will be ignored here.

<p>The following diagram would illustrate it quite well if I had any
artistic talent:

<tscreen><verb>
      _____                                     _____
     /     \                                   /     \
   PREROUTING -->[Routing ]----------------->POSTROUTING----->
     \D-NAT/     [Decision]                    \S-NAT/
                     |                            ^
                     |                            |  
                     |                            |
                     |                            |
                     |                            |
                     |                            |
                     |                            |
                     --------> Local Process ------
</verb></tscreen>

At each of the points above, when a packet passes we look up what
connection it is associated with.  If it's a new connection, we look
up the corresponding chain in the NAT table to see what to do with it.
The answer it gives will apply to all future packets on that
connection.

<sect1>Simple Selection using iptables

<p><tt>iptables</tt> takes a number of standard options as listed
below.  All the double-dash options can be abbreviated, as long as
<tt>iptables</tt> can still tell them apart from the other possible
options.  If your kernel has iptables support as a module, you'll need
to load the ip_tables.o module first: `insmod ip_tables'.

<p>The most important option here is the table selection option, `-t'.
For all NAT operations, you will want to use `-t nat' for the NAT
table.  The second most important option to use is `-A' to append a
new rule at the end of the chain (e.g. `-A POSTROUTING'), or `-I' to
insert one at the beginning (e.g. `-I PREROUTING').

<p>You can specify the source (`-s' or `--source') and destination
(`-d' or `--destination') of the packets you want to NAT.  These
options can be followed by a single IP address (e.g. 192.168.1.1), a
name (e.g. www.gnumonks.org), or a network address
(e.g. 192.168.1.0/24 or 192.168.1.0/255.255.255.0).

<p>You can specify the incoming (`-i' or `--in-interface') or outgoing
(`-o' or `--out-interface') interface to match, but which you can
specify depends on which chain you are putting the rule into: at
PREROUTING you can only select incoming interface, and at POSTROUTING
you can only select outgoing interface.  If you use the
wrong one, <tt>iptables</tt> will give an error.

<sect1>Finer Points Of Selecting What Packets To Mangle

<p>I said above that you can specify a source and destination address.
If you omit the source address option, then any source address will
do.  If you omit the destination address option, then any destination
address will do.

<p>You can also indicate a specific protocol (`-p' or `--protocol'),
such as TCP or UDP; only packets of this protocol will match the rule.
The main reason for doing this is that specifying a protocol of tcp or
udp then allows extra options: specifically the `--source-port' and
`--destination-port' options (abbreviated as `--sport' and `--dport').

<p>These options allow you to specify that only packets with a certain
source and destination port will match the rule.  This is useful for
redirecting web requests (TCP port 80 or 8080) and leaving other
packets alone.

<p>These options must follow the `-p' option (which has a side-effect
of loading the shared library extension for that protocol).  You can
use port numbers, or a name from the /etc/services file.

<p>All the different qualities you can select a packet by are detailed
in painful detail in the manual page (<tt>man iptables</tt>).

<sect>Saying How To Mangle The Packets

<p>So now we know how to select the packets we want to mangle.  To
complete our rule, we need to tell the kernel exactly what we want it
to do to the packets.

<sect1>Source NAT

<p>You want to do Source NAT; change the source address of connections
to something different.  This is done in the POSTROUTING chain, just
before it is finally sent out; this is an important detail, since it
means that anything else on the Linux box itself (routing, packet
filtering) will see the packet unchanged.  It also means that the `-o'
(outgoing interface) option can be used.

<p>Source NAT is specified using `-j SNAT', and the `--to-source'
option specifies an IP address, a range of IP addresses, and an
optional port or range of ports (for UDP and TCP protocols only).

<tscreen><verb>
## Change source addresses to 1.2.3.4.
# iptables -t nat -A POSTROUTING -o eth0 -j SNAT --to 1.2.3.4

## Change source addresses to 1.2.3.4, 1.2.3.5 or 1.2.3.6
# iptables -t nat -A POSTROUTING -o eth0 -j SNAT --to 1.2.3.4-1.2.3.6

## Change source addresses to 1.2.3.4, ports 1-1023
# iptables -t nat -A POSTROUTING -p tcp -o eth0 -j SNAT --to 1.2.3.4:1-1023
</verb></tscreen>

<sect2>Masquerading

<p>There is a specialized case of Source NAT called masquerading: it
should only be used for dynamically-assigned IP addresses, such as
standard dialups (for static IP addresses, use SNAT above).

<p>You don't need to put in the source address explicitly with
masquerading: it will use the source address of the interface the
packet is going out from.  But more importantly, if the link goes
down, the connections (which are now lost anyway) are forgotten,
meaning fewer glitches when connection comes back up with a new IP
address.

<tscreen><verb>
## Masquerade everything out ppp0.
# iptables -t nat -A POSTROUTING -o ppp0 -j MASQUERADE
</verb></tscreen>

<sect1>Destination NAT

<p>This is done in the PREROUTING chain, just as the packet comes in;
this means that anything else on the Linux box itself (routing, packet
filtering) will see the packet going to its `real' destination.  It
also means that the `-i' (incoming interface) option can be used.

<p>Destination NAT is specified using `-j DNAT', and the
`--to-destination' option specifies an IP address, a range of IP
addresses, and an optional port or range of ports (for UDP and TCP
protocols only).

<tscreen><verb>
## Change destination addresses to 5.6.7.8
# iptables -t nat -A PREROUTING -i eth0 -j DNAT --to 5.6.7.8

## Change destination addresses to 5.6.7.8, 5.6.7.9 or 5.6.7.10.
# iptables -t nat -A PREROUTING -i eth0 -j DNAT --to 5.6.7.8-5.6.7.10

## Change destination addresses of web traffic to 5.6.7.8, port 8080.
# iptables -t nat -A PREROUTING -p tcp --dport 80 -i eth0 \
	-j DNAT --to 5.6.7.8:8080
</verb></tscreen>

<sect2>Redirection

<p>There is a specialized case of Destination NAT called redirection:
it is a simple convenience which is exactly equivalent to doing DNAT
to the address of the incoming interface.

<tscreen><verb>
## Send incoming port-80 web traffic to our squid (transparent) proxy
# iptables -t nat -A PREROUTING -i eth1 -p tcp --dport 80 \
	-j REDIRECT --to-port 3128
</verb></tscreen>

Note that squid needs to be configured to know it's a transparent proxy!

<sect1>Mappings In Depth

<p>There are some subtleties to NAT which most people will never have
to deal with.  They are documented here for the curious.

<sect2>Selection Of Multiple Addresses in a Range

<p>If a range of IP addresses is given, the IP address to use is
chosen based on the least currently used IP for connections the
machine knows about.  This gives primitive load-balancing.

<sect2>Creating Null NAT Mappings

<p>You can use the `-j ACCEPT' target to let a connection through
without any NAT taking place.

<sect2>Standard NAT Behavior

<p>The default behavior is to alter the connection as little as
possible, within the constraints of the rule given by the user.  This
means we won't remap ports unless we have to.

<sect2>Implicit Source Port Mapping

<p>Even when no NAT is requested for a connection, source port
translation may occur implicitly, if another connection has been
mapped over the new one.  Consider the case of masquerading, which
is rather common:

<enum>
<item> A web connection is established by a box 192.1.1.1 from port
   1024 to www.netscape.com port 80.

<item> This is masqueraded by the masquerading box to use its source
   IP address (1.2.3.4).

<item> The masquerading box tries to make a web connection to
   www.netscape.com port 80 from 1.2.3.4 (its external interface
   address) port 1024.

<item> The NAT code will alter the source port of the second
   connection to 1025, so that the two don't clash.
</enum>

<p>When this implicit source mapping occurs, ports are divided into
three classes:
<itemize>
<item> Ports below 512
<item> Ports between 512 and 1023
<item> Ports 1024 and above.
</itemize>

A port will never be implicitly mapped into a different class.

<sect2>What Happens When NAT Fails

<p>If there is no way to uniquely map a connection as the user
requests, it will be dropped.  This also applies to packets which
could not be classified as part of any connection, because they are
malformed, or the box is out of memory, etc.

<sect2>Multiple Mappings, Overlap and Clashes

<p>You can have NAT rules which map packets onto the same range; the
NAT code is clever enough to avoid clashes.  Hence having two rules
which map the source address 192.168.1.1 and 192.168.1.2 respectively
onto 1.2.3.4 is fine.

<p>Furthermore, you can map over real, used IP addresses, as long as
those addresses pass through the mapping box as well.  So if you have
an assigned network (1.2.3.0/24), but have one internal network using
those addresses and one using the Private Internet Addresses
192.168.1.0/24, you can simply NAT the 192.168.1.0/24 source addresses
onto the 1.2.3.0 network, without fear of clashing:

<tscreen><verb>
# iptables -t nat -A POSTROUTING -s 192.168.1.0/24 -o eth1 \
	-j SNAT --to 1.2.3.0/24
</verb></tscreen>

<p>The same logic applies to addresses used by the NAT box itself:
this is how masquerading works (by sharing the interface address
between masqueraded packets and `real' packets coming from the box
itself).

<p>Moreover, you can map the same packets onto many different targets,
and they will be shared.  For example, if you don't want to map
anything over 1.2.3.5, you could do:

<tscreen><verb>
# iptables -t nat -A POSTROUTING -s 192.168.1.0/24 -o eth1 \
	-j SNAT --to 1.2.3.0-1.2.3.4 --to 1.2.3.6-1.2.3.254
</verb></tscreen>

<sect2>Altering the Destination of Locally-Generated Connections

<p>The NAT code allows you to insert DNAT rules in the OUTPUT chain,
but this is not fully supported in 2.4 (it can be, but it requires a
new configuration option, some testing, and a fair bit of coding, so
unless someone contracts Rusty to write it, I wouldn't expect it
soon).

<p>The current limitation is that you can only change the destination
to the local machine (e.g. `j DNAT --to 127.0.0.1'), not to any other
machine, otherwise the replies won't be translated correctly.

<sect>Special Protocols

<p>Some protocols do not like being NAT'ed.  For each of these
protocols, two extensions must be written; one for the connection
tracking of the protocol, and one for the actual NAT.

<p>Inside the netfilter distribution, there are currently modules for
ftp: ip_conntrack_ftp.o and ip_nat_ftp.o.  If you insmod these into
your kernel (or you compile them in permanently), then doing any kind
of NAT on ftp connections should work.  If you don't, then you can
only use passive ftp, and even that might not work reliably if you're
doing more than simple Source NAT.

<sect>Caveats on NAT

<p>If you are doing NAT on a connection, all packets passing
<bf>both</bf> ways (in and out of the network) must pass through the
NAT'ed box, otherwise it won't work reliably.  In particular, the
connection tracking code reassembles fragments, which means that not
only will connection tracking not be reliable, but your packets may
not get through at all, as fragments will be withheld.

<sect>Source NAT and Routing

<p>If you are doing SNAT, you will want to make sure that every
machine the SNAT'ed packets goes to will send replies back to the NAT
box.  For example, if you are mapping some outgoing packets onto the
source address 1.2.3.4, then the outside router must know that it is
to send reply packets (which will have <bf>destination</bf> 1.2.3.4)
back to this box.  This can be done in the following ways:

<enum>
<item> If you are doing SNAT onto the box's own address (for which
       routing and everything already works), you don't need to do
       anything.

<item> If you are doing SNAT onto an unused address on the local LAN
       (for example, you're mapping onto 1.2.3.99, a free IP on your
       1.2.3.0/24 network), your NAT box will need to respond to ARP
       requests for that address as well as its own: the easiest way
       to do this is create an IP alias, e.g.:
<tscreen><verb>
# ip address add 1.2.3.99 dev eth0
</verb></tscreen>

<item> If you are doing SNAT onto a completely different address, you
       will have to ensure that the machines the SNAT packets will hit
       will route this address back to the NAT box.  This is already
       achieved if the NAT box is their default gateway, otherwise you
       will need to advertise a route (if running a routing protocol)
       or manually add routes to each machine involved.
</enum>

<sect>Destination NAT Onto the Same Network

<p>If you are doing port forwarding back onto the same network, you
need to make sure that both future packets and reply packets pass
through the NAT box (so they can be altered).  The NAT code will now
(since 2.4.0-test6), block the outgoing ICMP redirect which is
produced when the NAT'ed packet heads out the same interface it came
in on, but the receiving server will still try to reply directly to
the client (which won't recognize the reply).

<p>The classic case is that internal staff try to access your `public'
web server, which is actually DNAT'ed from the public address
(1.2.3.4) to an internal machine (192.168.1.1), like so:

<tscreen><verb>
# iptables -t nat -A PREROUTING -d 1.2.3.4 \
	-p tcp --dport 80 -j DNAT --to 192.168.1.1
</verb></tscreen>

<p>One way is to run an internal DNS server which knows the real
(internal) IP address of your public web site, and forward all other
requests to an external DNS server.  This means that the logging on
your web server will show the internal IP addresses correctly.

<p>The other way is to have the NAT box also map the source IP address
to its own for these connections, fooling the server into replying
through it.  In this example, we would do the following (assuming the
internal IP address of the NAT box is 192.168.1.250):

<tscreen><verb>
# iptables -t nat -A POSTROUTING -d 192.168.1.1 -s 192.168.1.0/24 \
	-p tcp --dport 80 -j SNAT --to 192.168.1.250
</verb></tscreen>

Because the <bf>PREROUTING</bf> rule gets run first, the packets will
already be destined for the internal web server: we can tell which
ones are internally sourced by the source IP addresses.

<sect>Thanks

<p>Thanks first to WatchGuard, and David Bonn, who believed in the
netfilter idea enough to support me while I worked on it.

<p>And to everyone else who put up with my ranting as I learnt about
the ugliness of NAT, especially those who read my diary.

<p>Rusty.
</article>