One solution to this problem is "tunneling" -- that is, encapsulating the raw TCP/IP traffic into another format. This is often used to create VPN's, or "Virtual Private Networks." A tunnel does not necessarily have to be encrypted, although it frequently is.
One protocol that is used on Cisco devices to create tunnels is GRE, which will be the focus of today's post.
Consider the following network:
In this example, I have created two LAN segments, one in Anchorage (R3 and R5) and one in Bethel(R4 and R6), connected by a service provider's network (R1 and R2). The configuration on R1 and R2 are very straightforward: turn up the necessary network interfaces, set up OSPF to share routes between R1 and R2, then configure BGP to peer with the Anchorage and Bethel routers (R3 and R4), all of which we've done before. R5 and R6 are also very straightforward: turn up FA1/0 on both routers, assign the appropriate IP addresses to each router, then set up a default route through R3 for R5 and through R4 for R6. As usual, these two routers are simply simulating PCs on the internal LAN.
R3 and R4 are the routers that where the magic happens in this lab. Essentially, a tunnel interface is just a logical interface on the router that is configured much like a physical interface. For GRE, the interface name is "Tunnel1" and the IP address of the tunnel interface is configured exactly like the IP address of any other interface on the router. On R3...:
interface Tunnel1
ip address 10.255.255.1 255.255.255.252
R4, of course, is configured with a tunnel IP address of 10.255.255.2/30.
Next, you must configure a tunnel source and a tunnel destination. That is, you must tell the router what interface you want the tunnel to terminate at on the local router ("tunnel source") and what router the far end of the interface should exist upon ("tunnel destination"). On R3, also under the "interface Tunnel1" configuration...:
tunnel source 172.16.0.2
tunnel destination 172.16.1.2
There is one additional caveat to consider, and this is one that is easy to forget. On TCP/IP networks, there is a parameter known as MTU size. Essentially, this is the largest size of a packet that a network device can forward without having to break the packet into smaller pieces. Consider it like this: if you have a shipment that is delivered to you in a semi truck, but you need to carry that shipment in your pickup truck, you will have to make several trips to deliver the load, right? Same thing in networking. If one network device can handle a 1500 byte packet, but the next device can only handle a 1400 byte packet, then the second network device will have to break the packet into a 1400 byte chunk, then send the remaining 100 bytes in a second chunk. This is called "fragmentation." MTU size becomes an issue with GRE tunnels, because GRE adds several bytes to the packet size when it encapsulates a TCP packet. Consequently, if your network can only support 1500 byte MTUs, and you are receiving 1500 byte packets from, say, a LAN interface, then you will have a GRE-encapsulated packet that is larger than 1500 bytes exiting through your WAN interface. You can avoid this by setting the MTU size and adjusting the maximum packet size that the tunnel will encapsulate. This is done with two commands in the tunnel interface setting:
ip mtu 1460
ip tcp adjust-mss 1420
The complete tunnel interface configuration looks like this:
interface Tunnel1
ip address 10.255.255.1 255.255.255.252
ip mtu 1460
ip tcp adjust-mss 1420
tunnel source 172.16.0.2
tunnel destination 172.16.1.2
There is one last factor to consider. Since tunnels are used to isolate certain types of traffic, for example, LAN traffic, you probably don't want to advertise all of your routes when tunneling between two networks. In this lab, I modeled a very common scenario: two LAN networks with a tunnel to allow them to communicate over a public, untrusted network. This is, in fact, exactly how a network I manage in my day job is built: two LAN networks with a VPN tunnel across the Internet to bridge them together. Obviously, I don't want to advertise routes to my internal, RFC-1918 networks to the Internet, nor do I want the entire Internet routing tables stored on my LAN routers. To address this, I set up BGP peering between R3 and R1, and between R4 and R2, to provide routes to public (Internet) hosts (note: yes, I'm using private IP space on all of these hosts. Just pretend that 192.168.0.0/16 is private IP space, but 10.0.0.0/8 and 172.16.0.0/12 are public for this lab). Then, I set up OSPF to advertise routes to my LAN networks only across the GRE tunnel. R3 and R4 are NOT sharing their OSPF routes with R1 or R2.
To do this, I first created the access list and route map to match the LAN network on each router. The config is identical on R3 and R4:
access-list 10 permit 192.168.0.0 0.0.255.255
!
route-map lan permit 10
match ip address 10
Then, I applied the route maps to the OSPF configuration on each router. On R3:
router ospf 2112
router-id 10.255.255.1
log-adjacency-changes
redistribute connected subnets route-map lan
network 10.255.255.0 0.0.0.3 area 0
...and on R4:
router ospf 2112
router-id 10.255.255.2
log-adjacency-changes
redistribute connected subnets route-map lan
network 10.255.255.0 0.0.0.3 area 0
For BGP, R3 and R4 are essentially set up to receive routes only. On R3:
router bgp 64513
bgp log-neighbor-changes
neighbor 172.16.0.1 remote-as 64512
!
address-family ipv4
neighbor 172.16.0.1 activate
no auto-summary
no synchronization
exit-address-family
...and on R4:
router bgp 64515
bgp log-neighbor-changes
neighbor 172.16.1.1 remote-as 64514
!
address-family ipv4
neighbor 172.16.1.1 activate
no auto-summary
no synchronization
exit-address-family
With that, I have a virtual connection from R3 to R4, as shown by the traceroute from R5 to R6:
R5#traceroute 192.168.1.2
Type escape sequence to abort.
Tracing the route to 192.168.1.2
1 192.168.0.1 0 msec 0 msec 8 msec
2 10.255.255.2 28 msec 24 msec 40 msec
3 192.168.1.2 32 msec 24 msec *
R5#
Notice that the traceroute does not show that it crossed 172.16.0.0/30, 10.0.0.0/30 or 172.16.1.0/30? It appears to go directly from R3 to R4. That is due to the tunnel that bridges R3 and R4 together. Even though, in reality, the traffic is indeed passing through the "ISP" network, to the LAN hosts, it is as if R3 has a direct physical connection to R4.
