gSwitch

README

gSwitch

This is a user space switch (network pipeline processing) written in go for learning purposes (still work in progress).

Configuration:

[Redis]
Address = "127.0.0.1"
Port = 6379
Password = ""
DB = 0
Prefix = ""

[SwitchPorts]
    [SwitchPorts.sw1]
    Trunk = false
    AllowedVLANs = [1]
    Up = true

    [SwitchPorts.sw2]
    Trunk = true
    AllowedVLANs = [1, 10]
    Up = true


[[ControlProcess]]
Layer = 2
Name = "L2Switch"

[[ControlProcess]]
Layer = 2
Name = "ARP"
ConfigFile = "etc/l2/ARPConfig.toml"

[[ControlProcess]]
Layer = 2
Name = "L2Adapter"
ConfigFile = "etc/l2/L2Adapter.toml"

[[ControlProcess]]
Layer = 3
Name = "IPv4"

[[ControlProcess]]
Layer = 3
Name = "Routing"
ConfigFile = "etc/l3/RoutingTable.toml"

[[ControlProcess]]
Layer = 3
Name = "ICMP"
ConfigFile = "etc/l3/ICMPConfig.toml"

1- Redis (not used yet):

Redis is the backend datastore for this switch during runtime.

2- SwitchPorts:

This represents the ports that will be added to the switch.

• Trunk: whether the port is trunk or access port (not implemented yet)

• AllowedVLANs: in case Trunk is false, specify only one vlan number, otherwise it includes the allowed vlans on the trunk (eg. [10, 11, 12])

• Up: represents the initial status of the port whether it should be brought up on startup or not

3- ControlProcess:

Control processes are what defines how the traffic is handled by the switch. The switch is built heavily around the distributed asynchronous pipeline github.com/m-motawea/pipeline.

With the recent shift to the distributed pipeline branch, Goroutine channel bottlenecks were alleviated with the localqueue messaging drivers over 8-byte IDs (MessageStore centralized registries) rather than pointer traversal.

Currently L2Switch, L2Adapter, ARP, IPv4, Routing, and ICMP processes are beautifully orchestrated and natively interact with one another bidirectionally. Processes are executed iteratively depending on Layer and Name order during Ingress and Reverse order during Egress.

• Layer: represents the layer this process handles

• Name: name of the process

• ConfigFile: path to process configuration file (if needed)

Try It:

1- Get the Package

go get github.com/m-motawea/gSwitch

2- Go the package dir and build it

cd ~/go/src/github.com/m-motawea/gSwitch # change the location if you installed go in a custom location
go build

3- Initialize a test environment by network namespaces:

sudo ./scripts/env_setup.sh

• this will create 5 namespaces as hosts (h1,..h4) and a one as switch sw
• h1 & h2 IP address are 10.1.1.10 and 10.1.1.20
• h2 & h3 IP address are 10.10.1.30 and 10.10.1.40

4- Start the switch in the sw namepace with the default config in the package:

sudo ip netns exec sw ./gSwitch

• h1 and h2 are connected to sw as access ports on vlan 1
• h3and h4are connected to sw as access ports on vlan 10
• Control processes include the L2Switch, ARP, IPv4, ICMP and Routing as well as each layer adapter process.

5- Test connectivity example:

sudo ip netns exec h1 ping 10.1.1.20 # connection to h2
sudo ip netns exec h1 ping 10.10.1.40 # connection to h4 (routed)

6- Clean the test environment:

sudo ./scripts/env_destroy.sh

Running unit and integration tests locally:

Before executing full integration tests verification, you must be successfully running the distributed_pipeline branch of the github.com/m-motawea/pipeline framework. This architecture replaces blocking concurrent Go channels with the localqueue driver and ID-based message map StoreMessage() pointer bypassing logic.

go mod edit -dropreplace github.com/m-motawea/pipeline
go get github.com/m-motawea/pipeline@distributed_pipeline

Run the robust E2E full-stack verification test natively:

go test -v ./controlplane -run TestSwitchEndToEnd_ICMP

This comprehensive unit test mimics a complete end-to-end framework, sending ICMP Echo Requests resolving dynamically back and forth using physical L2/L3 encapsulation layers and intercepting active dynamic hardware ARP negotiation requests automatically whenever a mapped address table drops.

TODO:

1- Try to Fix Trunk Ports (due to stripped vlan tags)

• currently trunk link is not working but to get around this you can use subinterfaces.
• create a sub interface for each vlan and use the sub interface in configuration instaed of the Master.

ip link add link <master> name <sub name> type vlan id <id>

• make sure to set the trunk option for the subinterface as false other wise there will be two layers of 802.1Q.

[SwitchPorts."sw5.10"]
Trunk = false
AllowedVLANs = [10]
Up = true

[SwitchPorts."sw5.1"]
Trunk = false
AllowedVLANs = [1]
Up = true

2- Handle L2 Loops Using STP (needs support for 802.3)

3- Document Current Processes