On 10/1/24 12:56, Ilya Maximets via discuss wrote: > On 10/1/24 10:44, Adrián Moreno via discuss wrote: >> On Mon, Sep 30, 2024 at 01:52:13PM GMT, Harald Albrecht via discuss wrote: >>> My "quest" is to discover the topology of OVS switches and their netdev >>> ports >>> to provide some (albeit limited) OVS-related information to users of the >>> Ghostwire Linux kernel networking discovery engine >>> (https://github.com/siemens/ >>> ghostwire). This engine is used, for instance, by containerlab (https:// >>> containerlab.dev/) users to quickly pick and "shoot" their capture targets >>> in >>> containers acting as routers. >>> >>> Before asking here I've searched for information in the OVS documentation >>> itself, trying to find other sources, and looked at the ovs_vport NETLINK >>> generic family description (https://docs.kernel.org/networking/netlink_spec/ >>> ovs_vport.html). >>> >>> What I would like to achieve: discover which netdev-based network interfaces >>> are related to which OVS bridge instances. What I have to avoid is bringing >>> in >>> user-space CLI binaries that produce stdout output and then having to parse >>> this output, as it is at least slow and has turned out to be a nightmare in >>> other similar situations (podman is an infamous example here). >>> >>> The current discovery sees a "master"-"slave" relationship between the >>> "system" >>> OVS kernel bridge and all netdev-based ports of all the other OVS kernel >>> bridges. This differs from Linux stock kernel bridges where this >>> relationship >>> uniquely represents the switch-port topology. Reading >>> https://docs.kernel.org/ >>> networking/netlink_spec/ovs_vport.html I'm at a loss at how to figure out >>> which >>> OVS kernel bridge a netdev port has been attached to. Can you please be so >>> kind >>> as to point me to information that I've missed or maybe give me here more >>> information on how I can discover this topology without the need for >>> user-space >>> CLI tools?
FWIW, you don't need CLI tools to communicate with usersapce OVS daemons. At least, you may discover the topology by talking to ovsdb-server via the well-defined JSON-RPC protocol over the unix socket: https://datatracker.ietf.org/doc/html/rfc7047 This communication method and the database schema supposed to be backward compatible. Best regards, Ilya Maximets. >>> >>> With best regards, >>> TheDiveO >>> >> >> Hi Harald, >> >> A similar discussion occurred on netdev recently. >> The OVS kernel datapath does not contain any "logical" topology >> information (i.e: bridges, ports, etc). Furthermore, internal interfaces >> called after OpenFlow bridges are not stacked under "ovs-system". > > Note also, that some datapath ports, like tunnel backing interfaces, can > represent multiple OpenFlow ports from multiple bridges. So, they cannot > be associated with a single OVS bridge at all. > >> >> From netlink, the best you can get is the list of OVS-managed netdevs. >> To extract the OpenFlow topology, you need to use the OpenFlow API, >> in userspace or look at the OVSDB. You should be able to access both >> programatically without parsing the output of CLI tools. >> >> Thanks. >> Adrián >> >>> _______________________________________________ >>> discuss mailing list >>> disc...@openvswitch.org >>> https://mail.openvswitch.org/mailman/listinfo/ovs-discuss >> >> _______________________________________________ >> discuss mailing list >> disc...@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-discuss > > _______________________________________________ > discuss mailing list > disc...@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-discuss _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
