start.sh脚本解读

We set up the topology in start.sh. First, we start four docker containers in the topology. Take the command for switch1 as an example.

sudo docker run --net=none --privileged --entrypoint /bin/bash --name switch1 -it -d -v $PWD/switch1:/sonic docker-sonic-p4:latest

We specify --net=none to prevent Docker engine from adding its docker0 interface, which may interfere with the topology being tested. --privileged is to enable each container to configure their own interfaces. -v $PWD/switch1:/sonic mounts the configuration folder into the switch containers.

Then we create three links. Take the link between switch1 and switch2 as an example. The following commands connect switch1’s interface eth1 with switch2’s interface eth1.

sudo ovs-vsctl add-br switch1_switch2
sudo ovs-docker add-port switch1_switch2 eth1 switch1
sudo ovs-docker add-port switch1_switch2 eth1 switch2

We also configure the interface IP and default routes on the host1 and host2. Take host1 as an example.

sudo docker exec -d host1 ifconfig eth1 192.168.1.2/24 mtu 1400
sudo docker exec -d host1 ip route replace default via 192.168.1.1

Finally, we invoke the startup script for switch1 and switch2.

sudo docker exec -d switch1 sh /sonic/scripts/startup.sh
sudo docker exec -d switch2 sh /sonic/scripts/startup.sh

SONiC-P4 配置细节

In start.sh, we have mounted the configuration folder into the switch container, at /sonic. The most important configurations are at /sonic/scripts/startup.sh, /sonic/etc/config_db/vlan_config.json, and /sonic/etc/quagga/bgpd.conf.

In /sonic/scripts/startup.sh, we start all SONiC services and a P4 software switch. The P4 software switch is started by this line (see supervisord.conf)

simple_switch --log-console -i 1@eth1 -i 2@eth2 ...

It binds interface eth1 to P4 software switch’s port 1, eth2 to port 2, and so on. These ethX interfaces are usually referred as front-panel interfaces, and directly used by the P4 switches for carrying data plane packets. However, SONiC operates on another type of interfaces, called host interfaces. Host interfaces are for SONiC control plane, and are NOT carrying data plane packets. The host interfaces are named as EthernetX. We configure peer-to-peer IP and MTU on host interfaces. SONiC reads the configurations, like IP and MTU, from host interfaces, then configures these values on the P4 software switch using SAI. The mapping between host interfaces and switch ports is specified in /port_config.ini:

# alias         lanes
Ethernet0       1
Ethernet1       2
...

Together with the simple_switch command in /sonic/scripts/startup.sh, we have configured this mapping Ethernet0 –> lane 1 –> eth1. It is essentially a mapping between host interfaces and front-panel interfaces.

/sonic/etc/config_db/vlan_config.json configures the switch vlan interfaces used in this test, using ConfigDB interface, look here for details:

{
    "VLAN": {
        "Vlan15": {
            "members": [
                "Ethernet0"
            ], 
            "vlanid": "15"
        }, 
        "Vlan10": {
            "members": [
                "Ethernet1"
            ], 
            "vlanid": "10"
        }
    },
    "VLAN_MEMBER": {
        "Vlan15|Ethernet0": {
            "tagging_mode": "untagged"
        },
        "Vlan10|Ethernet1": {
            "tagging_mode": "untagged"
        }
    },
    "VLAN_INTERFACE": {
        "Vlan15|10.0.0.0/31": {},
        "Vlan10|192.168.1.1/24": {}
    }
}

/sonic/etc/quagga/bgpd.conf configures the BGP session on the switch. Here is the BGP configuration for switch1, which peers with switch2 using peer-to-peer IP 10.0.0.0/31, and announces 192.168.1.0/24.

router bgp 10001                        
  bgp router-id 192.168.1.1             
  network 192.168.1.0 mask 255.255.255.0
  neighbor 10.0.0.1 remote-as 10002     
  neighbor 10.0.0.1 timers 1 3          
  neighbor 10.0.0.1 send-community      
  neighbor 10.0.0.1 allowas-in          
  maximum-paths 64                      
!                                       
access-list all permit any