manual_setup

Manual Topology Setup

In this example we show how to create a virtual topology without using any helper (mininet, p4utils, ...). For that we will manually create hosts using unix namespaces and interconnect them using virtual interfaces and bmv2 switches.

The goal of this example is to show what does mininet do when you define a topology. Note that mininet provides you with much richer set of options and does many more things. Here, we just show the minimum needed to try a very basic p4 example.

Topology:

+--+     +--+     ++-+     +--+
|h1+-----+s1+-----+s2+-----+h2|
+--+     +-++     +--+     +--+  

Create Linux Network Namespaces

We first create a pair of linux network namespaces (see more: article, man page).

Linux namespaces provide a network stack for all the processes running within the namespace. Including network interfaces, routing tables, iptables rules, etc. Thus, network virtualization tools (like mininet) use linux network namespaces to instanciate isolated nodes in a topology (hosts, switches, routers, etc)

To create network namespaces we can use the ip tool, specifically ip-netns command.

$ sudo ip netns add h1
$ sudo ip netns add h2

Once we have the namespaces we can run shell commands using:

$ sudo ip netns exec <namespace-name> <cmd>

Alternatively we can get a shell:

$ sudo ip netns exec <namespace-name> bash

Adding Virtual Interfaces

To interconnect namespaces between them we need to create virtual ethernet interfaces and assign them. Virtual interfaces come in connected pairs, thus if you send a packet to one side of the pair you get the packet in the other side.

AS before, we use ip to create three veth pairs:

$ sudo ip link add h1-eth0 type veth peer name s1-eth0
$ sudo ip link add h2-eth0 type veth peer name s2-eth0
$ sudo ip link add s1-eth1 type veth peer name s2-eth1

If you list the number of interfaces, you will see that six new interfaces have been added to the default/global namespace. Next we will move two of those interfaces to h1 and h2 network namespaces:

sudo ip link set h1-eth0 netns h1
sudo ip link set h2-eth0 netns h2

We leave the rest of interfaces (the ones we will connect to the switches) in the global namespace due to the fact that bmv2 switches can not be run inside network namespaces (or at least they do not seem to work). However, that is fine since the switches do not need to run the TCP/IP stack they just need to receive packets from one interface, process them, and send them to another interface.

Optional:

Optionally we can change the interfaces MTU to 9500 so we can play with bigger packets. Furthermore, we can disable ipv6, to avoid seeing some automatically generated packets by the ipv6 stack.

declare -a arr=("h1-eth0" "h2-eth0" "s1-eth0" "s1-eth1" "s2-eth0" "s2-eth1")
for intf in "${arr[@]}"
do
    ip link set "$intf" mtu 9500
    sysctl net.ipv6.conf.${intf}.disable_ipv6=1
done

In the next steps we need to bring the interfaces up, assign IP, configure gateways and fill arp tables.

Setup Interfaces

First we bring the loopback interfaces at the namespaces:

sudo ip netns exec h1 ifconfig lo up
sudo ip netns exec h2 ifconfig lo up

Then we bring the virtual interfaces up, and configure the MAC and IP addresses at the hosts.

sudo ip netns exec h1 ifconfig h1-eth0 hw ether 00:00:00:00:01:01 10.0.1.1/24 up
sudo ip netns exec h2 ifconfig h2-eth0 hw ether 00:00:00:00:02:02 10.0.2.2/24 up
sudo ip link set dev s1-eth0 up
sudo ip link set dev s1-eth1 up
sudo ip link set dev s2-eth0 up
sudo ip link set dev s2-eth1 up

Since hosts belong to different subnetworks we need to configure a gateway that will be used for packets going to 0.0.0.0. For that we assign as a gateway a fake IP and which interface has to be used as a gateway.

sudo ip netns exec h1 route add default gw 10.0.1.254 h1-eth0
sudo ip netns exec h2 route add default gw 10.0.2.254 h2-eth0

Setting the gateway address and interface is not enough, when the host tries to communicate with the gateway it will use the link layer (L2), for that to happen the ARP (address resolution protocol) will be used in order to discover the MAC address of 10.0.X.254, however the switches in this example are not programmed to respond to ARP. To bypass that, we just populate the ARP tables at each host with fake MAC addresses:

sudo ip netns exec h1 arp -i h1-eth0 -s 10.0.1.254 00:00:00:01:01:00
sudo ip netns exec h2 arp -i h2-eth0 -s 10.0.2.254 00:00:00:02:02:00

Compile P4 Program

To run the p4 program in the bmv2 switch we first have to compile it using the p4c compiler:

p4c --target bmv2 --arch v1model --std p4-16 forwarding.p4 -o .

Run Bmv2 Switches

Next we run two simple_switches instances, attach the interfaces we want to have, and provide the compile p4 program.

sudo simple_switch -i 1@s1-eth0 -i 2@s1-eth1  --thrift-port 9090 --nanolog ipc:///tmp/bm-0-log.ipc --device-id 0 forwarding.json &
sudo simple_switch -i 1@s2-eth0 -i 2@s2-eth1  --thrift-port 9091 --nanolog ipc:///tmp/bm-1-log.ipc --device-id 1 forwarding.json &

Populate Tables with Simple Switch CLI

As a last step we populate the ipv4_lpm table using the simple_switch_CLI and a file with the commands we want to type in.

simple_switch_CLI --thrift-port 9090 < s1-commands.txt
simple_switch_CLI --thrift-port 9091 < s2-commands.txt

Ping Test

To test our virtual network, we access h1, spawn a shell, and ping h2:

ip netns exec h1 bash

$ ping 10.0.2.2
PING 10.0.2.2 (10.0.2.2) 56(84) bytes of data.
64 bytes from 10.0.2.2: icmp_seq=1 ttl=62 time=4.89 ms
64 bytes from 10.0.2.2: icmp_seq=2 ttl=62 time=4.19 ms