Project

General

Profile

Doc-overview » Protocol information »

Bridge loop avoidance II

Further pages on this topic:

Situation

This is the example we will use to discuss the concept: multiple mesh nodes (called backbone gateways throughout the text) are connected to a big, single LAN (the backbone) as well as mesh nodes which are not connected to the backbone. backbone connections are drawn with solid lines, mesh connections with dotted lines. You can see an internet gateway connected through the LAN (this could be any other network device as well) at the top and a client (e.g. your laptop) connected/bridged through a mesh node in the bottom.

Definitions:

  • backbone gateway: A mesh node which is connected to both - a mesh network and a backbone (e.g. LAN).
  • client: A non-mesh network participant which is sending data via the mesh. The client is always identified by the source MAC address of the payload Ethernet header.
  • Originator: An originator is a mesh participant in batman. If we talk about the originator address in this document, we mean the hardware address of the primary interface.

Goals:

  • The bridge loop avoidance should be able to scale to hundreds of gateways connected to the same backbone.
  • Communication between the mesh nodes and the backbone should be via the best backbone gateway.
  • Minimize broadcast traffic in the backbone.
  • Roaming should still be possible.

Key Concepts:

Claiming clients:

Only one backbone gateway (out of possibly many gateways) should forward traffic from a non-mesh client (coming via the mesh) to the backbone. Every backbone gateway announces the mac addresses of the non-mesh clients it feels responsible for in the form of "claim frames" on the soft-interface bat0. Each backbone gateway will save a claim list of other backbone gateways. By doing this, it can:

  • see which clients are already tracked (claimed)
  • see which other backbone gateways exist in the backbone

Note that backbone gateways may overwrite a claim of another backbone gateway by simply claiming the same client. In this case, the newest claim wins, and local databases are updated accordingly.

Backbone gateways can also "unclaim" their own clients. After doing this, the client is not claimed anymore for this backbone.

Details and Cases:

The concept will be explained by discussing the different cases: unicast and broadcast frames, both for the backbone->mesh and mesh->backbone case. Loop cases are considered too: backbone->mesh->backbone and mesh->backbone->mesh. Furthermore, roaming (when a client changes from the backbone to another mesh node and back) is discussed.

Broadcast, mesh->backbone:


If the originator of the broadcast is not known (not in the originator table), don't forward it into the backbone.

If the originator of the broadcast is another backbone gateway from the same backbone, don't forward.

If a backbone gateway has already claimed the client, only the responsible backbone gateway will deliver the frame to the backbone/LAN. Other backbone gateways will discard the packet.

If no backbone gateway has claimed the client, the first gateway receiving the broadcast will claim it in the backbone by sending a claim frame, and send the actual broadcast frame.

Unicast, mesh->backbone:

If the client is not claimed by the backbone gateway receiving the unicast, a claim packet is sent out first. This backbone gateway is now the (new) responsible backbone gateway for this client.

Then, the unicast packet is delivered to the soft interface.

Broadcast, backbone->mesh:

If a claim for the mesh-client exists:

  • all not responsible backbone gateways should discard the frame - this might be a loop.
  • if the responsible backbone gateway (which claimed this client) also receives the packet, "unclaim" this client by sending an according un-claim packet, and forward the packet into the mesh. This should only happen in the roaming case, but not in normal situations.

If the client is not claimed, all backbone gateways will send the broadcast into the mesh. The mesh nodes will avoid duplicates by using the duplicate lists (see section below)

Unicast, backbone->mesh:

Nothing fancy should happen here, as a switch should do the right thing. If the sender (source of the Ethernet frame) is not claimed, the frame is simply sent into the mesh.

However, when using hubs (or broken switches), also unicast packets might loop. Therefore, if the sender is claimed by any backbone gateway, don't forward.

mesh->backbone->mesh loop:

A broadcast frame is only sent to backbone by the (yet to be) responsible backbone gateway. Back from the backbone to the mesh, all not responsible backbone gateways will not forward the frame. The responsible backbone gateway should not receive it on normal switched networks.

To avoid loops for hubs, unicast packets are not forwarded from backbone to mesh if the client is claimed.

backbone->mesh->backbone loop:

Although a broadcast frame is sent into the mesh, it won't loop back into the backbone as a backbone gateway will not forward it if was sent from another backbone gateway (from the same backbone).

backbone->mesh Roaming:

This is the easy case: Sooner or later, a claim will be added for the roamed client.

mesh->backbone Roaming:

We assume that the client was already claimed by a backbone gateway before, otherwise there is nothing to consider.

When a broadcast frame is sent, the responsible backbone gateway can detect the roam and will unclaim the client. This first broadcast will only be broadcasted by the (previously) responsible backbone gateway into the mesh, later broadcasts will be sent by all backbone gateways again.

Gateway selection:

All gateways announce the clients within their backbone and ignore roaming messages from other gateways on the same backbone.

Other non-gateway mesh nodes may select their best gateway to the backbone based on the TQ value.

Broadcasting by all backbone gateways

All backbone gateways send the broadcasts from backbone clients into the mesh. We have to make sure that the duplicates are recognized and ignored to deliver the broadcast packets only once within the mesh.

To do this, we use a "mesh duplicate list": This list is kept for every backbone gateway mesh node which is connected to a backbone with multiple backbone gateways. It checks broadcast packets which are sent from the backbone into the mesh through multiple backbone gateways (and thus multiple times).

If the client is not claimed by any backbone gateway, every backbone gateway shall forward the packet into the mesh. This will lead to duplicates of the broadcast with different meta information (different originators, different sequence numbers). To avoid duplicates within the mesh, every mesh node will use the "mesh duplicate list":

  • after the (old) seqno window check: match the frame to the "mesh duplicate list" of other backbone gateways to see if another backbone gateway from the same backbone has sent a broadcast with the same payload checksum.
  • if true, drop the packet. Otherwise, add the checksum to the entries and re-forward it.

Discussion:

Features:

  • no single "super" gateway => should scale better
  • the only additional BATMAN backbone packets are claim packets, which are only sent for new claims and regular announcements
  • no BATMAN packets on the backbone
  • broadcasts are sent from all the gateways into the mesh
  • nodes can select gateways, and change among them (gateways will automatically re-claiming)
  • should not loop ;)

Limitations:

  • loops in higher-level structures may not be avoided. For example, if there are two meshes and two backbones are interconnected as in the illustration below, a loop is formed which can't be detected, since the claim frames of one mesh won't travel along the mesh network of the other.