Skip to content

Latest commit

 

History

History
264 lines (218 loc) · 12.4 KB

README.md

File metadata and controls

264 lines (218 loc) · 12.4 KB
Raw

Campus Network Project

This repository contains:

  • the set of scripts to provision VMs, build a virtual exchange point and run a virtual campus networks on a VM for the course LINGI2142 at UCLouvain in order to emulate campus networks and interconnect them.
  • Some of the student's implementation, showcasing various aspects of the configuration and management of an IPv6-only multihomed network using provider-assigned prefixes, in the student_projects folder. An overview of these project, as well as a feature highlight is available in the README file.
  • host defines the script to hosts a remote set of VMs for each group.

Virtual machine

A sample virtual machine definition to run the script is provided and managed using Vagrant, using a VirtualBox provider.

You need to install both of these softwares on your machine in order to run an emulated network.

Commands summary

  • ./build_vm.sh will create and provision the virtual machine.
  • vagrant up will boot the VM (once it has been built).
  • vagrant ssh (from this directory) will create an ssh connection to the VM.
  • vagrant halt will stop the VM (i.e. shutdown properly the guest OS).

Virtual network

The main directory of this repository contains the set of scripts to start a virtual network as well as loads and apply its configuration files. You should only run such a network within the VM.

Description

  • ./create_network.sh will create a virtual network and execute startup scripts in every node. It takes one mandatory argument: a bash script defining a mk_topo function to describe the topology (i.e., example_topo or ucl_topo). This scripts can also redefine the following environment variables:
    • CONFIGDIR to specify which directory contains the configuration files for the topology,
    • BOOT to specify the prefix of the scripts to run when a network node has booted,
    • STARTUP to specify the prefix of the scripts to run when a node has an established connectivity. The script exposes various functions that can be used to create and connect a virtual network, such as
    • add_link $1 $2 to create and link nodes $1 and $2,
    • mk_LAN $router $@ to create a LAN (a network bridged at layer 2) connecting a $router to one or more hosts (separated by space),
    • bridge_node $node $itf $name to bridge the physical interface $itf of the VM (i.e., one connected to the host machine) with the virtual interface named $name of the network node $node.

When executed, the script creates the network according to the mk_topo function. For each network node, the directory named $CONFIGDIR/<node name> (i.e. cfg/<node name>) will be mounted on /etc (i.e., providing an overlay only visible to that node). The $CONFIGDIR/templates will be mounted on /templates and $CONFIGDIR/puppetmodules will be mounted on /puppetmodules. Two scripts are then executed on each node, if available and executable: * $CONFIGDIR/<node name>_$BOOT is executed right after the creation of the node, and before the creation of its links. * $CONFIGDIR/<node name>_$STARTUP is executed on every node once the whole network has been created (i.e. links, LANs, ...), thus once the mk_topo function returns (i.e. do not make blocking calls in it ...).

  • ./connect_to.sh takes two arguments, a configuration folder, and a node's name as defined in the mk_topo function (i.e., BXL for the example_topo), and opens a shell in its environment. This is useful to access a node through an out of band channel (i.e., it will work even if the network itself is not working), thus for debug purposes.
  • ./cleanup.sh will shutdown the network and attempt to clean all associated resources such as links, temp files, net NS, processes. You should extend this script to account for you own tempfiles.

Two topologies are pre-defined to be started with the create_topo script.

  1. example_topo is a sample topology showing how to start a network, configure basic routing, as well as execute some tests in it. See the next section for a detailed description.
  2. ucl_topo defines the base UCL topology with OSPF configuration. See the "UCL Minimal Topology" section below for a detailed description.

By default, the create_network.sh script imports a few files from /etc in the different nodes order to help you. If you need additional files, edit the base array and append their names in ETC_IMPORT at the top of that script.

Example topology

example_topo defines a small network topology, running OSPF as IGP using two different software stacks.

Description

BELNET --- BXL --- LLN
|                  |
+- B1          L1 -+- L3
|                  |
+- B2          L2 -+- L4

The example_topo script defines a small topology with routers: BXL, BELNET and LLN.

It then attaches 2 LANs in the network: one on BELNET and one in LLN, with a variable number of hosts.

Scripts

The routers all define a boot script, which reloads the sysctl configuration in every net NS (in this case: Enable IPv6 and IPv6 forwarding). Their startup script then assign IPv6 addresses to the interfaces and/or start a routing daemon.

More specifically:

  • BXL is the simplest node. It statically defines the IP addresses/prefixes on every interface, and then start the BIRD routing daemon with a minimal OSPFv3 configuration.
  • LLN leverages the features of its routing daemon, Quagga. As a result, its startup script only starts the zebra daemon, and then later ospf6d. The IPv6 addresses, prefixes, ... of the interfaces are all managed by the zebra daemon, i.e. defined in $CONFIGDIR/LLN/quagga/zebra.conf.
  • BELNET is a (crude) example of a more generic configuration process. Based on network-wide settings defined in example_topo, it derives the IP addresses for its interfaces automatically. It then derives its BIRD configuration file from a template. (A much better way would to use a real templating engine and set of script, e.g. using python/mako, ...)

The BELNET node also configures two BGP sessions that are available on the remote VM.

Most hosts (B2 L2 L3 L4) are not configured: without startup scripts and no router advertisement daemons, no IPv6 addresses are assigned. As a result, they only have IPv6 link-local addresses -- thus only have connectivity in their LAN.

L1 statically assigns its IPv6 address, and then starts an iperf server. B1 also assign itself an IPv6 address. Both hosts auto-assign a default route to the closest router.

Example test

  1. The following commands will instantiate the example_topo topology, then enter the node B1 and execute an iperf throughput test with the IPv6 address of the node L1. 
    sudo ./create_network.sh example_topo
sudo ./connect_to.sh example_cfg B1
iperf -V -c fd00:255:11::1
  2. Use the connect_to.sh script to get a shell in different nodes. There, you can check the IPv6 addresses assigned to the various interfaces (ip -6 address), view the kernel routing table (ip -6 route), connect to the BIRD routing daemon if in use (e.g. bird6c -s /tmp/LLN.ctl, then type ? to see the list of available commands), or Quagga (telnet localhost zebra or telnet localhost ospf6d ), ...

UCL Minimal Topology

ucl_topo defines a minimal UCL network, with only routers running OSPFv3 using bird.

Description

    +---- SH1C ----+
    |              |
Michotte        Halles ------ belnetb
    |              |
 Carnoy ----- Pythagore ----- belneta
    |              |
    +--- Stevin ---+

The ucl_topo script defines a small topology with 6 routers and 2 links to two providers.

Scripts

The routers all define a boot script, which reloads the sysctl configuration in every net NS (in this case: Enable IPv6 and IPv6 forwarding). Their startup script run a puppet scripts that fills a template and run bird6 on the produced configuration file.

More specifically, the puppet ucl_minimal_cfg/<node name>/puppet/site.pp (mounted to /etc/puppet/site.pp) is executed. This script imports (i.e., execute) a puppet module called bird6.

Modules are common to all the nodes and stored in ucl_minimal_cfg/puppetmodules/ (mounted to /puppetmodules/). The puppet code is in ucl_minimal_cfg/puppetmodules/bird6/manifests/init.pp. This module defines a class with two parameters. Parameters are automatically filled by hiera that will look at ucl_minimal_cfg/<node name>/puppet/data/node.yaml for a value with key <class-name>::<parameter-name>. A hiera lookup for a given key can be explicitely requested (there is an example in the module). The node.yaml (a YAML file) is a database of any information specific to your node.

The module is filling a template of a bird6 config stored in ucl_minimal_cfg/templates/. This template is written in ERB and it has access to any local variable or parameters of the module. After filling this template, the puppet module launches bird6.

Concepts

We emulate networks on a single machine by leveraging Linux network namespace, to which we interface using the ip netns command family (from iproute2, see man ip-netns).

Conceptually, a net NS is a new instance of the kernel networking stack. This enables to have multiple routing tables, set of interfaces, isolated across different namespaces. We can thus emulate a new network node by representing it as a single network namespace, with its local set of interfaces.

Connecting network namespaces is achieved using pairs of virtual ethernet interfaces (see main ip-link and look for the veth link type). Once a pair of interfaces is created, we can then move each interface in the net NS of the corresponding network node, such that:

  • the node 'sees' a new interface (its own end of the link)
  • the node has no visibility of the interfaces of other nodes.

The filesystem of the host machine is shared across all network namespace, i.e. if two nodes try to write to a file named /tmp/temp, they will conflict with eachother. Make sure to read to the documentation of the programs you want to run in each namespace in order to figure out which file they create (e.g. bird6 creates its control socket in /var/run/bird6.ctl, as a result, we override this default value in the node BXL/BELNET's startup script). The content of $CONFIGDIR/$node_name is 'overlaid' on top of /etc for each node.

LAN are represented by creating a virtual switch (using the default linux bridge implementation), connecting it to a router, then connecting multiple hosts to the switch.

MISC

  • The DNS server from the project setup authorize zone transfers originating from the project network (i.e., dig -t AXFR ingi. should work inside the emulated network if the setup is correct)
  • You can use your local web browser to access the emulated network web servers if you host any: 1. Connect to the emulated network using ssh (possibly using ProxyJump if you need to first connect to the remote server) and use the -D <port> option to create a SOCKS proxy on a local port. 2. Configure your web browser to use the new SOCKS proxy to access content from nodes in your virtual network (i.e. on Firefox you would go to Preferences / Advanced / Connection / Settings / Manual Proxy / SOCKS host: localhost, port / SOCKS 5 / Tick the Proxy DNS option at the bottom of the pane).