RE: [net-next-2.6 PATCH 02/10] ethtool: add ntuple flow specifier to network flow classifier

[Ben Hutchings <bhutchings@solarflare.com>]

On Wed, 2011-03-02 at 11:11 -0800, Dimitrios Michailidis wrote:
> Ben Hutchings wrote:
> > /**
> >  * struct ethtool_flow_ext - flow spec common extension fields
> >  * @vlan_etype: EtherType for vlan tagged packet to match
> >  * @vlan_tci: VLAN tag to match
> >  * @data: Driver-dependent data to match
> >  *
> >  * Note: Additional fields may be inserted before @vlan_etype in future,
> >  * but the offset of the existing fields within the containing structure
> >  * (&struct ethtool_rx_flow_spec) will be stable.
> >  */
> > struct ethtool_flow_ext {
> > 	__be16	vlan_etype;
> > 	__be16	vlan_tci;
> > 	__be32	data[2];
> > };
> 
> I am wondering about the semantics of these vlan_* fields.  Is vlan_etype the
> Ethertype in the VLAN header or the type after it?

It would be the the type in the VLAN tag.  The nested ethertype is
normally implied by flow_type to be ETH_P_IP.

This does leave the question of what this would mean:

struct ethtool_rx_flow_spec fs = {
	.flow_type = ... | FLOW_EXT,
	...
	.h_ext.vlan_tci = htons(0x1234),
	.m_ext.vlan_etype = 0xffff,
};

This says the TCI must be == 0x1234 but the type can be anything.  But
the type surely has to be be one assigned for use in VLAN tags.  Should
we leave it to the driver/hardware to determine what those valid types
are, or should we reject this as valid?

> Specifically, I am wondering
> whether these fields are limited to VLANs or can work with general Ethertypes and the
> 2 bytes after them.

If you set flow_type = ETHER_FLOW you can then specify any combination
of Ethernet header fields to match in {h_u,m_u}.ether_spec.  However, so
far only sfc implements it and it only allows matching on the
destination MAC address.

Ben.

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.