Port Forwarding
This is my first attempt at documenting the code I wrote to expand on
the functionality of the IP Masquerading code included in Linux 2.0.*
kernels.
New
Using Port Forwarding with Linux 2.2.
Port Forwarding - Why?
Before I describe what port forwarding is, let me describe the
situation that persuaded me to write this code. A local computer group had
3 computers and 1 IP address. They had implemented an IP masquerading
solution to allow people to use all 3 computers with the configuration:
INTERNET
| (194.160.1.1)
linux box 1
| (10.0.0.1)
+--------------------------+ internal ethernet
| (10.0.0.2) | (10.0.0.3)
linux box 2 linux box 3
The first linux box was a 386SX20 with only 4Mb and couldn't have coped
with any real users. We then redirected telnet, mail, web, etc. to one of
the two internal machines with plug-gw or nc (netcat) from inetd.
This had a number of problems:
- Some services such as rlogin in to the cluster didn't work properly
- Logs on internal machines were difficult to interpret as all
connections originated at the gateway host.
- Making internal services which use UDP visible externally was difficult.
- Redirecting inward connections was much slower than masquerading
outward ones.
- It confused the users.
Port Forwarding - What is it?
The easiest way to imagine Port Forwarding is a combination of
routing by port combined with packet rewriting. A convention router
examines the packet header and dispatches the packet on one of it's other
interfaces, depending on the packet's destination address. Port
Forwarding examines the packet header and forwards it on to another
host (after a little header rewriting) depending on the destination
port.
In more detail: Port forwarding forwards all packets intended
for one forwarding port on the gateway from the external networks to
routed on a specified port on one of the internal machines (after a little
rewriting of headers). This is (in some ways) a reverse of masquerading and
uses many of the maquerading functions - particularly the packet header
rewriting code.
Confused? Here's an example:
On the gateway, we setup the rule that all connections to port 80/tcp
should be redirected to port 80 on 10.0.0.2 (an internal machine).
The incoming web connection would be labelled:
Source: 163.158.1.2/7890 Dest: 194.160.1.1/80
This would be forwarded on to the internal host as:
Source: 163.158.1.2/7890 Dest: 10.0.0.2/80
Replies would be labelled:
Source: 10.0.0.2/80 Dest: 163.158.1.2/7890
and would be rewritten by the gateway to:
Source 194.160.1.1 Dest: 163.158.1.2/7890
This has a number of advantages over using tools like nc and
plug-gw to do the forwarding:
- As there is less copying of packets in memory, it's much faster.
- Internal hosts see the original connection source so so logs are
meaningful.
- It's easy to load split between the internal hosts. The
choice of host can also depend on live feedback such as which
machine has the lower load average. This decision making is done
entirely in userspace so is easy to implement.
- As redirection is done at a packet level, you don't need
different gatewaying tools for different services.
- It's very stable. Our gateway machine has been crashed many times (it's
not difficult to crash a machine with only 4Mb of memory) but it's carried
on forwarding/masquerading although someone notices that they can't log in
and reboots it.
Port Forwarding - How does it work?
[This section is optional reading and is not necessary to use Port
Forwarding]
Port forwarding uses the existing masquerading scheme to do all the
rewriting of packets. The masquerading table (what you see when you
type netstat -M or ipfwadm -M -l) is setup as if the connection started
internally. When the existing masquerading code receives a packet from the
external interface, it checks whether the destination port is in the range
61000-64999 and, if so, checks for any current entries in the masquerading
table. If there is a matching entry, it rewrites the packet header and
forwards it onto its new desintation. Port forwarding performs an
additional check on the destination port if it isn't in the masquerading
range to see whether it's a forwarding port. If it is, we let the
existing code check for an entry in the masquerading table. If a
corresponding entry exists in the masquerading table, the existing
masquerading code rewrites the header and sends the packet out. If the
destination port is a forwarding port but doesn't have an entry in
the masquerading table, we create a suitable entry in the table
before rewriting the packet and sending it out.
Port Forwarding - How do I use it?
The code was original written for Linux 2.0.27-29 and worked with many
other 2.0.* kernel. Changes were made to the kerne's masquerading code in
2.0.30 and this provoked me to improve and upgrade my patches while
converting them for 2.0.30. The new patches use 17k less memory when in
use and are smaller, neater and better integrated into the kernel
distribution and with a Makefile option.
Using the Linux 2.0.27-29 patches.
Using the Linux 2.0.30-38 patches.
Using Port Forwarding with Linux 2.2.
Port Forwarding - Anything else?
When I wrote this code, I didn't imagine anyone else ever using this code
other than the computer group I mentioned. I would appreciate all comments
on the idea of port forwarding and its implementation, whether it's
"didn't you know there's a much easier way of doing it" or "I liked
the idea but I extended it by doing ...". If there are any mistakes or
areas that aren't clear in the above documentation, please also let me
know. My email address is:
steven@monmouth.demon.co.uk
First FAQ: I now know what port forwarding is but what's IP
Substitution or IPSubs?
Answer: IP Substitution was the original name I gave to the
code I wrote. I subsequently decided I didn't like the name and changed it
to Port Forwarding instead.
Lastly...
Last updated on 22nd October, 2000
SC