Libsai

Libsai is a small SAI implementation for testing SAI applications.

The goal of the project is to help both hardware vendors and SAI application developers by

• Becoming a reference SAI implementation
• Providing an easy way to test SAI application on top of it

The codes include not only some glue codes but also actual data-plane codes using a software switch, thus you can try playing with it on a single PC!

The functionalities it supports as for now are:

• L2 tenant isolation using VLAN (saiport, saivlan)
• L3 inter-VLAN routing using virtual router (sainexthop, sairoute, sairouter)

We are working hard to support other functionalities as well.

How to build

Please refer to doc/Building.md

How to run

Before starting, make sure you finished everything in the "How to build" section.

1. Run the switch VM

   host-OS$ sudo qemu-system-x86_64 [image of switch VM] -enable-kvm -device rocker,name=sw1,len-ports=4,ports[0]=dev0,ports[1]=dev1,ports[2]=dev2 -netdev bridge,br=br0,id=dev0 -netdev bridge,br=br1,id=dev1 -netdev bridge,br=br2,id=dev2
   

2. Run the three other VMs, which you created by copying the switch VM in doc/Building.md.

   host-OS$ sudo qemu-system-x86_64 [image of VM1] -enable-kvm -net nic,macaddr=52:54:00:00:00:11 -net bridge,br=br0
   host-OS$ sudo qemu-system-x86_64 [image of VM2] -enable-kvm -net nic,macaddr=52:54:00:00:00:12 -net bridge,br=br1
   host-OS$ sudo qemu-system-x86_64 [image of VM3] -enable-kvm -net nic,macaddr=52:54:00:00:00:13 -net bridge,br=br2
   

The three VMs are attached to the switch VM, each to one port. Note that the VMs are still not yet "connected" each other, thus no pakcets can be exchanged even if you try to set addresses to the VMs.

3. Assing addresses to the VMs, and execute libsai in the switch VM.

   VM1$ sudo ip addr add 192.168.1.2/24 dev eth0
   VM1$ sudo ip link set up eth0
   VM1$ sudo ip route add 192.168.2.0/24 via 192.168.1.1 dev eth0
   
   VM2$ sudo ip addr add 192.168.1.3/24 dev eth0
   VM2$ sudo ip link set up eth0
   VM2$ sudo ip route add 192.168.2.0/24 via 192.168.1.1 dev eth0
   
   VM3$ sudo ip addr add 192.168.2.2/24 dev eth0   # be careful it's 2.2, not 1.2 nor 1.4
   VM3$ sudo ip link set up eth0
   VM3$ sudo ip route add 192.168.1.0/24 via 192.168.2.1 dev eth0  # be careful again, VM3 is under 192.168.2.0/24
   
   switch-VM$ cd libsai
   switch-VM$ sudo ./a.out
   switch-VM$ sudo su -
   switch-VM# echo "1" > /proc/sys/net/ipv4/ip_forward
   

At this point, the three VMs are connected and routed under the topology in the figure below. VM1 and VM2 are connected to the same VLAN (saivlan), and VM3 is connected to another VLAN. The two VLANs are connected to virtual router (sairouter) and routed each other so that VMs in different VLANs can communicate each other.

4. Try ping between the VMs and see what is happening.

• Ping from VM1 to VM2 while executing tcpdump -vvv on VM2: You will first see an ARP request arriving to VM2 from VM1, and then ICMP packets start arriving. This is because VM1 and VM2 are connected to the same VLAN. Note that you cannot see the VLAN tags on the VMs because they are automatically pushed/dropped inside the switch VM.
• Ping from VM1 to VM3 while executing tcpdump -vvv on VM3: You will again see an ARP request arriving to VM3, but this time not from VM1 but from 192.168.2.1. This is because VM1 and VM3 are connected to different VLANs and the two VLANs are routed on L3.

How to hack

In main.c the topology used above is intutively coded.

// initialize the switch
sw = switch_init();

......

Vlan* vlan1 = create_vlan(2, &sw->ports[0], &sw->ports[1]);
Vlan* vlan2 = create_vlan(2, &sw->ports[2], &sw->ports[3]);

// create a virtual router
Router* router = create_router(sw);

// assign 192.168.1.0/24 to vlan1
add_vlan_to_router(router, vlan1, make_ip4("192.168.1.1"));
create_new_route(router, make_ip4("192.168.1.0"), make_ip4("255.255.255.0"), vlan1);

// assign 192.168.2.0/24 to vlan2
add_vlan_to_router(router, vlan2, make_ip4("192.168.2.1"));
create_new_route(router, make_ip4("192.168.2.0"), make_ip4("255.255.255.0"), vlan2);

You can create as many VLANs as you want and create more complex routes by using these abstractions.

Detailed descriptions of the directories and files in the repository can be found ar doc/Files.md

Copyright

Copyright (C) 2015 Nippon Telegraph and Telephone Corporation. Released under Apache License 2.0.