[ovs-dev] [PATCH] nsh: rework NSH netlink keys and actions

[Jan Scheurich]

> > As discussed please adjust the netlink uAPI for NSH as follows:
> >
> > OVS_KEY_ATTR_NSH
> > -- OVS_NSH_KEY_ATTR_BASE		mandatory
> > -- OVS_NSH_KEY_ATTR_MD1 		conditional: if mdtype=MD1
> >
> > OVS_ACTION_ATTR_PUSH_NSH
> > -- OVS_NSH_KEY_ATTR_BASE		mandatory
> > -- OVS_NSH_PUSH_ATTR_CONTEXT	conditional: currently if MD1 or
> > 					MD2 with TLV encap properties
> >
> > OVS_ACTION_ATTR_POP_NSH		no data
> >
> > Remove OVS_NSH_KEY_ATTR_MD2 because the MD2 context headers
> will not
> > be modelled as dedicated key fields but mapped to some generic TLV
> fields,
> > similar to but independent from the current tunnel metadata TLV
> registers.
> >
> > Using the name OVS_NSH_KEY_ATTR_MD2 for the opaque context data
> in
> > PUSH_NSH would be misleading as the variable length byte array can
> carry
> > any type of metadata format: MD1, MD2, or any future MD type. For
> that
> > reason I prefer the generic name OVS_NSH_PUSH_ATTR_CONTEXT.
> 
> Then why even have ATTR_MD1? Sounds a single attribute is enough to
> support all cases.

OVS_NSH_KEY_ATTR_MD1 represents a struct containing the four fixed length context headers defined in MD1. These are match fields in OVS and must be represented explicitly in OVS_KEY_ATTR_NSH. But their presence is conditional on MD type 1, that's why Jiri rightly insisted on having it split from OVS_NSH_KEY_ATTR_BASE.

In OVS_ACTION_ATTR_PUSH_NSH we again have the same mandatory header part represented by OVS_NSH_KEY_ATTR_BASE, this time followed by an optional variable length opaque container for up to 248 octets of context data (the format is of no concern for the push_nsh action). The content of the ATTR_CONTEXT, if any, is compiled by ofproto when the action is sent.

BR, Jan