In this assignment you will implement a fully functional Internet router that routes real network traffic. The goal is to give you hands-on experience as to how a router really works. Your router will run as a user process locally, and when finished will route real packets that are flowing across the Internet to application servers located at Stanford. We'll be giving you a skeleton, incomplete router (the "sr" or simple router) that you have to complete, and then demonstrate that it works by performing traceroutes, pings and downloading some files from a web server via your router.
The VNS was designed at Stanford, for their introductory networking course and they're nice enough to let us use it too. It gives you hands-on experience working on projects requiring low level network access, such as routers. The VNS is comprised of two components: (1) The VNS Server which runs in a lab at Stanford, and (2) A number of VNS Clients which connect to the server. Your router is an example of a VNS Client. The server intercepts packets on the network, forwards the packets to the clients, receives packets from the client and injects them back into the network. The physical setup of the VNS is shown in the figure.
The server is a user level process running at Stanford. The machine hosting the server is connected to a hub which is connected to two Linux servers running a few internet services (http, ftp, and a streaming music server on port 8888), referred to as application servers. The VN Server simulates a network topology which consists of multiple links and VN Clients. The application servers sit on the other side of the network topology. For example, the simplest topology is one with a single VNS Client and one application server, as shown below in the figure.
A client wanting access to traffic in the network connects to the server via a normal TCP socket and requests the traffic seen on links in the topology, in this case, topology 0. Assuming the traffic is not already being sent to some other user, the server accepts the request and sends the traffic on the link to the client over the TCP socket. The client would then inspect the packet, determine where the next hop in the network (which would be fairly easy in the case of topology 0) and send the packet back to the server to be injected back into the network.
The VNS Server can handle multiple (2^16) topologies simultaneously. This means that each student can have his or her own topology to connect to and route over. The VNS Server ensures that clients are only sent traffic belonging to their topology.
A VNS client is any program that speaks the VNS protocol and connects to the VNS server. In the case of this assignment we provide you with the code for a basic VNS client (called sr or Simple Router) that can connect to the VNS server. The clients are run locally by the students as regular user processes and connect to the server via normal TCP sockets. Clients, once connected to the server, are forwarded all packets that they are supposed to see in the topology. The clients can manipulate the packets in any way they wish, generate responses based on the packets, or make routing decisions for those packets and send the replies back to the server to place back onto the network. For example, on the above topology (topology 0), the VNS Server might receive a TCP SYN packet destined for vns-app-1.stanford.edu. The VNS Server sends the packet to the VNS Client which would receive the packet on interface zero, decrement the TTL, recalculate the header checksum, consult the routing table and send the packet back to the server with directions to inject it back onto the network out of interface one. What will the destination hardware address be for the packet sent back by the client? What if the client doesn't know the hardware address for www-server-1?
In this assignment you will implement a fully functional router by extending the sr code given to you.
The following scenario is a step by step explanation of how a client routes traffic on a simple topology.
Nick has just finished developing his router for programming assignment #3. He was assigned topology 42 for testing which is shown in the figure below.
To test, Nick runs his router from mycomputer.home.edu and connects to the VNS server at vns-1.stanford.edu, topology 42. The VNS server sends Nick's router the list of interfaces and their IP addresses.
To generate traffic for routing, Nick fires up a standard web browser from his local computer pointed at the IP of the application server on topology 42. Nick's router will now get the opportunity to route all packets between his web browser and the web server.
We'll now walk through the first few significant steps that take place when packets flow between Nick's web browser and the web server.
Before beginning development you should first get familiar with the sr stub code and some of the functionality it provides. Download the Stub Code Tarball and save it locally. As described before, it handles all of the dirty-work required for connecting and communicating with the server. To run the code, untar the package (tar -zxvf sr_stub.tar.gz) and compile it via make. Once compiled, you can connect to the VNS server as follows:
./sr -s vns-1.stanford.edu -t <topo-id>
for example, connecting to the server on topology 0 would look like:
./sr -s vns-1.stanford.edu -t 0
(you can use ./sr -h to print a list of the accepted command line options)
After you connect successfully, the server will send you a description of the host including all the interfaces and their IP addresses. The stub code uses this to build the interface list in the router (the head of the list is member if_list for struct sr_instance). The routing table is constructed from the file rtable and by default consists of only the default route which is the firewall. The routing table format is as follows:
ip gateway mask interface
a valid rtable file might look like this:
172.24.74.213 172.24.74.213 255.255.255.255 eth1
172.24.74.228 172.24.74.228 255.255.255.255 eth2
0.0.0.0 172.24.74.17 0.0.0.0 eth0
The VNS Server, on connection should return the IP addresses associated with each one of the interfaces. The output for each interface should look something like:
INTERFACE: eth0
Speed: 10
Hardware Address: 70:00:00:00:00:01
Ethernet IP: 172.24.74.41
Subnet: 0.0.0.0
Mask: 0.0.0.0
To test if the router is actually receiving packets try pinging or running traceroute to the IP address of eth0 (which is connected to the firewall in the assignment topology). The sr should print out that it received a packet. What type of packet do you think this is?
What should your router do on receipt of an ARP request packet?
As you work with the sr router, you will want to take a look at the packets that the router is sending and receiving. The easiest way to do this is by logging packets to a file and then displaying them using a program called tcpdump.
First, tell your router to log packets to a file in a format that tcpdump can read by passing it the -l option and a filename:
./sr -t <topo-id> -s vns-1.stanford.edu -l <logfile>
As the router runs, it will log the packets that it receives and sends (including headers) to the indicated file. After the router has run for a bit, use tcpdump to display the packets in a readable form:
tcpdump -r <logfile> -e -vvv -x
The -r switch tells tcpdump where to look for the logfile. -e tells tcpdump to print the headers of the packets, not just their payload. -vvv makes the output very verbose, and -x puts the packets in a hex format that is usually easier to read than ASCII. You may want to specify the -xx option instead of -x to print the link-level (Ethernet) header in hex as well.
The two most important methods for developers to get familiar with are:
void sr_handlepacket(struct sr_instance* sr, uint8_t * packet/* lent */, unsigned int len, char* interface/* lent */)
This method, located in sr_router.c, is called by the router each time a packet is received. The "packet" argument points to the packet buffer which contains the full packet including the ethernet header. The name of the receiving interface is passed into the method as well.
int sr_send_packet(struct sr_instance* sr /* borrowed */, uint8_t* buf /* borrowed */, unsigned int len, const char* iface /* borrowed */)
This method, located in sr_vns_comm.c, will send an arbitrary packet of length, len, to the network out of the interface specified by iface.
Within the sr framework you will be dealing directly with raw Ethernet packets. There are a number of resources which describe the protocol headers in detail, including Stevens UNP, www.networksorcery.com and the Internet RFC's for ARP (RFC826), IP (RFC791), and ICMP (RFC792). The stub code itself provides some data structures in sr_protocols.h which you may use to manipulate headers. There is no requirement that you use the provided data structures, you may prefer to write your own or use standard system includes.
In addition to routing packets between the local networks and the Internet, a virtual router can act as a simple firewall, controlling which packets can reach the application servers connected to the router.
Specifically, a VNS firewall should support the following functionality.
1. The ability to declare an interface as internal or external. For simplicity only a single external interface is supported -- this is the first interface in your rtable. Internal interfaces are connected to local (protected) networks while the external interface is connected to the Internet. 2. By default packets arriving to the external interface and destined to an internal interface are silently dropped. That is, no response (e.g., ICMP) packet is generated as a result of dropping the original packet. This dropped packet is logged, for the correct format see 'Log'. 3. By default, packets arriving from an internal interface that need to be forwarded through the external interface are allowed to pass through the firewall. Doing so, allows end-hosts within the protected networks to access services on the public Internet. See 'Generating Outbound Traffic' below for more information. 4. The ability to create exceptions that override the default behaviors. See 'Adding Exceptions' below.
Note that simply allowing packets from the internal hosts to go through the firewall is not enough to establish a working connection to an external service, because most (if not all) TCP/IP services entail two-way communications. Therefore, packets that belong to a flow initiated by an internal end-host that arrive to the external interface must be allowed through the firewall. To support this feature the firewall maintains a "flow table" that contains all the active (and allowed) flows that traverse the firewall. In this context, a flow is defined as a 5-tuple <srcIP, dstIP, IPprotocol, src-port, dst-port>.
When the first "internal" packet arrives at the firewall, two entries are added to the flow table, one for each direction of communication. The entry for the external-to-internal flow can be generated by inverting the order of source and destination IP addresses and ports. When a packet arrives to the external interface, the firewall checks if it matches one of the entries in the flow table.If it does then the packet is not dropped and it is forwarded to the internal interface.
Entries remain on the flow table as long as packets that match these entries go through the firewall. To support this feature each entry has a time-to-live (TTL). Each time a packet matching the flow entry is received, the entry's TTL is set to X seconds. We should be able to set X with the -T option (e.g., ./sr -T 120). The firewall periodically scans the flow table and removes all entries whose TTL has expired. (Note: You should update the entries associated with both directions of a flow when a packet is received.)
The flow table can hold up to Y entries at each time. The parameter Y is configurable and we should be able to set it with the -F option (e.g., ./sr -F 100). If a new entry needs to be added when the flow table is full, first a scan is initiated to determine if one or more stale entries exist in the flow table. If all entries are valid, then an ICMP response is returned (Destination Unreachable - Port Unreachable) to the originator and a log entry is generated.
The firewall supports adding explicit rules to allow/disallow flows to traverse the firewall. For example, the firewall's administrator might decide to allow TCP packets with destination port 80 and destination address X.Y.Z.W to go through the firewall. Such rules are entered in the firewall's "rule table". A rule table entry has the following format <srcIP, dstIP, proto, srcPort, dstPort, 0 or 1> where each of the entry's first five components can be a wildcard and the last specifies if it allows (last field is 1) or disallows (last field is 0) those flows. A wildcard entry matches all values of the corresponding field in an actual packet. Your firewall should read the rule table from the file specified with the -R option. There is a sample rule table here, you will need to change the IP addresses in it to refer to your topology. The precedence of rules is determined by their order in the rule table (i.e., use the first matching rule you find).
When a packet arrives at the firewall it first checks to see if the packet matches one of the entries in the flow table. If not, then the firewall determines if the packet matches one of the entries in the rule table. If this is true, then the firewall allows or disallows it based on the rule. If a rule disallows a packet it is logged and no response is generated. If the packet matches a rule table entry but the flow table is full then an ICMP response is returned and an entry is added to the firewall's log. Your firewall should write it's log to the file specified with the -L option.
In order to fully test the firewall's functionality, you will need to generate outbound traffic from within your topology. You can do this by connecting to the application server IP via SSH and then using ping and wget to generate outbound requests that will establish TCP/IP flows between the (internal) application server and an external server. Note: Your program does not actually need to handle ping traffic for this assignment because it is not TCP or UDP traffic!
In order to support connecting to your application server IP via SSH, your firewall will need to have at least one inbound allowed flow defined in its configuration: you must allow inbound connections on port 22 (the standard SSH port) to the IP of your applications servers.
Full instructions for loggin in via SSH were emailed to you with the announcement of this assignment.
To generate traffic using ping, you must specifiy the IP address to use as the source address. You can do this with the following command:
ping -I <IP Address> host
For example, if I were logged into the IP 171.67.71.22 and wanted to ping the server at www.cs.princeton.com, I would use the following command:
ping -I 171.67.71.22 www.cs.princeton.com
You can also generate outbound HTTP requests using the wget utility.
wget --bind-address <IP> URL
For example, to retrieve the Princeton CS homepage while logged into 171.67.71.22, I would use the following command:
wget --bind-address 171.67.71.22 http://www.cs.princeton.edu
Your log of dropped packets should have the following format:
<srcIP, dstIP, protocol, src-port, dst-port, drop-code>
The drop-code for 'flow not allowed' is 2. The drop-code for 'flow table full' is 3.
Here's an example:
<1.2.3.4, 5.6.7.8, UDP, 54321, 23, 2> <4.3.2.1, 9.8.6.5, TCP, 12345, 80, 3>
There is no support for fragment reassembly for firewall purposes. That is if a packet fragment arrives with the transport level header missing then the fragment (as well all other subsequent fragments) are dropped.
Your firewall does not need to respond properly to pings or traceroutes.
We will declare that your firewall is functioning correctly if and only if:
Not Required but Smiled Upon:
Currently the stub code is event based. That is, code is executed each time a packet is received. This makes it hard to correctly enforce timeouts. For example, if the router is waiting for an ARP request that doesn't come, it will have to wait for another packet to arrive before it can handle the timeout. Of course, if a packet never arrives, the timeout will never be serviced. Though not required, an implementer may choose to enforce stronger guarantees on timeouts.
Last updated: Mon Apr 27 01:46:15 -0400 2009