[ovs-discuss] QoS: linux- QoS configuration reset when port enslaved to ovs bridge

[Reshma Sreekumar]

Hello All,

I see a peculiar behavior where the following steps resets the QoS
configuration on the netdevice:

*sudo ip link add dummy0 type dummy*

*sudo tc qdisc add dev dummy0 root handle 1: htb default 1*
*sudo ovs-vsctl add-port global_fwbr dummy0*

i.e. when a port with linux htb config having qdisc "1:" is added to ovs
bridge, the qdisc is reset to default type ('noqueue').

*However, this is only true for qdisc "1:" and not for any other qdiscs!*

On further checking the code, it looks like *netdev_linux_set_qos *eventually
calls *tc_del_qdisc* for this port due to a mismatch in  *new_ops *and
*netdev->tc->ops!*

And eventhough the description of *tc_del_qdisc *says that it's equivalent
to  *`tc qdisc del dev <name> root`, *it's not so! *tc_del_qdisc  *only
deletes the qdisc with handle "1:0" and nothing else!

So, what exactly is the expected behaviour? Is it a valid configuration to
have a port with its own linux htb configuration to be enslaved into an ovs
bridge while the configs remain intact?

Thanks in advance!

Thanks,
Reshma
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20211014/a9a26e79/attachment.html>