Routing and IP Gateways
I mentioned previously that IP
is an unreliable, connection-less protocol.
That is, it contains no provision to ensure that the packet
arrives correctly at the destination, nor
is there anything that guarantees that when packets do arrive they arrive in the
correct order. Although IP
is responsible to ensure that the packets get to the right machine, it has
essentially no understanding of the physical connection between the two
machines. IP will happily run on machines that are
connected with something as simple as a telephone wire, to something as complex
as satellites. IP depends on some other means to
"physically" carry it across the network.
What this means is that the system administrator
is responsible for laying the "map" that is used to define which
network address go with what sets of machine and what
IP addresses are assigned to individual
One important job that IP
does is routing. That is, getting the packet
to the right machine. If the source and destination machines are directly
connected, that is on the same network, then routing is easy. Essentially there isn't any. IP
sees this fact and simply hands the packets off to the data link
layer. Otherwise, IP
has to figure out how and where to send it.
Usually the 'how' is over a router. A router
is some piece of hardware that
acts like an air traffic controller send one packet
off one way and others off a different
way. Often routers are separate pieces of equipment that can be configured in very
detailed ways. The disadvantage to this is that with power comes price. The ability to
configure a router
in many different ways usually means a high price tag. Fortunately,
many operating systems, including Linux allow IP
to serve as router-software. Thereby
avoiding the cost of router
In comparison to the router
is the concept of a gateway. Like a router, a
gateway has knowledge of other networks and how to reach them. In general, we can think of
as a special piece of hardware that does the work for us. In fact, there are
companies that sell equipment called routers. A gateway
is more of a concept, in that is
the means by which you go from one network
to another. Today, the distinction between a
router and a gateway
is blurred. Originally, a gateway was a machine that converted from
to another. However, in common usage today, routers can server as gateways,
gateways can serve as routers.
The path the packet
takes from one machine to the next is called a route.
Although each machine can maintain static routes for specific destinations, the default
gateway is usually used to find remote destinations. (The default gateway
is needed only for computers that are part of an internetwork.) If you have a gateway connected to
several other networks, there will (probably) be route
definitions for each of those other
Let's look at this process as if we were sending a letter, as we did a little while
ago. Each letter we send has an envelope which contains a message. On the envelope we
write the source and destination addresses. When we mail the letter it gets to the post
office and the person sorting the mail checks the destination zip code. If it's the same
as the local zip code, the envelope is sent to one of the carriers for delivery. If the zip
code is different, then it is sent to some other location. Perhaps all non-local mail is
sent to the same place.
If you live across the country from me, the letter probably doesn't go directly from
my local post office to yours. (assuming I don't live in San Francisco and you don't live
in New York) The same applies to IP
packets. My letter first goes to my local post office,
if it is destined for a local address
it is processed there. If not, it is sent along to a
larger post office. If I sent a letter from Santa Cruz, California destined for Annsville,
Pennsylvania, it will probably go first to San Francisco and then to New York (or
Philadelphia) before it gets sent to Annsville.
Again, the same applies to IP
packets. If I were communicating with a network
other side of the country, my machine needs to know how to get to the other one. This is
the concept of a "gateway". A gateway
is the first step in the path, or "route" to the
remote machine. Just as there are a couple of post offices between Santa Cruz and
Annsville, there can be multiple gateways between computers.
Since San Francisco is the closest "major" city to Santa Cruz, it is possible that all
mail bound for points beyond must first go through there. What if I lived in Fresno, which
is about halfway between San Francisco and Los Angeles? If I sent a letter to Annsville,
it could go through Los Angeles or it could go through San Francisco. To make things easy,
it might always get sent through San Francisco it not destined for a local address.
What if the letter is bound for Los Angeles? It seems silly to go through San Francisco first
when it is bound for LA. At the post office in Fresno, they might have a special procedure
that says all remote mail goes through San Francisco, except for those with a zip code in
a special range.
Here, too, the same applies to IP
addresses. One machine may be defined as the "default" gateway,
but if an IP packet
was bound for a particular network
it could be told to use a completely different gateway. Which gateway to use to
get to a particular machine or network is the concept of "routes." If I want
all remotely-bound packets to use a particular route, I
add that route as a default to my machine. If packets bound for a particular
network are to go via a different route, I can add that route as well.
Image - Network gateway (interactive)
prepares to send a "message", it inserts the local (source) and destination
IP addresses in the IP header.
It then checks whether the network
ID of the destination and source match (the zip codes). If so, the packet
is sent directly to the destination, since it is on the local network. If the
network IDs don't match, the routing table is examined for static routes. If
none are found, the packet is forwarded to the default
The default gateway
is a computer connected to the local subnet
and other networks that has knowledge of the
IDs for other networks and how to reach them. Because the default gateway knows
how to reach the other networks, it can forward the
packet, either to other gateways or directly to that
machine if the gateway is on the same network as the destination. This process
is known as routing.
If you only have a single network,
there is no reason to have a gateway,
as each machine is directly connected to every other. It's possible that you
only want certain machines within your network to go beyond the local net to
the outside. In this case, these machine can have a default (or static)
route default, while the others have none. However, users
can add routes themselves, using the route command.
As we talked about earlier, TCP
connections are not the only ones that are managed by inetd. Basically all
network connections are. This can be understood if we go
back to the telephone operator analogy. If the operator (inetd) is also the
receptionist, we can then think of TCP connections as incoming telephone calls
and UDP packets as incoming letters. Like incoming phones
calls, the receptionist is responsible to route the
incoming mail to the right person. (This is a valid analogy, because like regular
mail, there is nothing to guarantee the delivery of the message in the letter,
although with TCP connections you can ask your partner to resend the message.)
Like TCP connections, UDP daemons are "listening" on specific ports. Also like
TCP connections, these well-known ports are listed in /etc/services.
One common UDP
connection is the routing daemon: routed. Routed supplies (as you might have
guessed) routing information in the form of routing packets. If your system is
serving as a router, then routed periodically sends copies
of its routing tables to other machines.
One key difference is that routed is not actually started by inetd. Instead,
it is normally started through one of the scripts under /etc/rc.d as the system
is booting. This actually calls the script
When it starts, routed makes the assumption that it will forward packets
between all interfaces on the system. This only includes those that are "up"
and does not include the loopback driver. (The loopback driver is a special
TCP/IP interface that simply loops the packets back to the local machine. Hence
the name.) Routed then transmits a REQUEST packet on each
of these interfaces and waits for a RESPONSE packet for any other hosts.
Potentially there are other machines on the network that
are also sending REQUESTS packets, so routed can also respond to them.
The response routed gives is based on information it has in it's routing
tables. This contains information about known routes, including how far
away the destination machine is in turns of hops or intermediary
machines. When routed receives a RESPONSE packet, it uses
the information contained in that packet to update it's own routing tables. Look
at the routed man-page for more information.
Routes are added to and removed from the system using the route
command. The general syntax is:
route <option> command destination gateway
The two commands used are add and delete. The destination is the IP
of the machine or network
you want to reach. You can also use tokens for the network name by including
entries in the /etc/networks file. This is an ASCII file
containing two columns. The first is the name of the network and the second
column is the network address. You can then use that name in the
is the IP
of the interface to which the packets need to be addressed. Keep in mind that
the system must already know how to get to the gateway for this to work.
The metric is a value that normally indicates the number of intermediate
machines (hops). The system uses this value in determining the shortest
route to a particular machine.
For example, let's assume we have an entry in /etc/networks like this:
Let's also assume that the machine I need to use to access this route
has an IP
of 22.214.171.124. I could then run the route command like this:
This says that any packet
destined for the siemau network
(as defined in /etc/networks) should go to the IP
126.96.36.199 with a metric of 0. Normally, 0 is used when the IP address
you specify is directly connected to your machine.
If you have a single machine that serves as your gateway
to the rest of the world, you can specify default instead of a specific
address or network as your
destination. In the example above, if we wanted to use the same machine for all
networks instead of just siemau, the command would look like this:
As you move about the network,
dynamic entries are created by the routing protocol
that you use. (Most commonly routed) The routing protocol communicates with it's
counterpart on other machines and adds entries to the routing tables
When it starts, routed looks for the file /etc/gateways, which contains a
list of gateways. (What else?) The general format for this file is:
<net|host> name gateway
The first field specifies whether the gateway
is to a specific machine or network.
The name field is the name of the destination host
or network. This can either be an IP
or a token.
If using a token, then the hostname
must be located in /etc/hosts or can be determined through DNS.
If through DNS, routed must be started after named. If a network, the name
must be in /etc/networks.
field is the name or address
of the gateway that is to be used. The metric is the same as for routes and
indicates the number of hops. The type can be either
passive, active or external. A passive gateway is one that is not expected to
exchange routing information. Active gateways will exchange information and
usually have routed running on them. External gateways are ones that are managed
by another system, but alternate routes should not be installed.