[Question] MCTP Packets destined to NULL Address unless sees any RTN_LOCAL routes, will not propagate to AF_NET binded sockets

[faizana-nvidia]

Hello Jeremy,

I came across a peculiar situation while handling MCTP Packet [MCTP header - ver=0x01, dest=0x00, src=0x00, flags_seq_tag=0xc8] referring to DSP0283 USB MCTP Endpoint Discovery Discovery Notify command.

Observation:

1. Packets are being dropped as they fail to find any RTN_LOCAL routes (if local routes/address where not assigned to interface) mctp_route_lookup_null
2. Such control commands are not reaching to any AF_MCTP control type binded sockets.

Question:

Is this a mandated thing to pass as a pre-conditional check before any such MCTP packet to send to any bounded socket as well because for me it makes sense when it's destined for any specific EIDs and not NULL_EID?

[jk-ozlabs]

Is this a mandated thing to pass as a pre-conditional check before any such MCTP packet to send to any bounded socket as well because for me it makes sense when it's destined for any specific EIDs and not NULL_EID?

I am having trouble parsing that question.

In general though: yes, we currently require a device-specific route for packet ingress. This is typically the local-address route created during address assignment. That's not based on any spec requirement though, it's an implementation detail. We may be able to relax that restriction following the dst-routing changes that went in a while ago.

[faizana-nvidia]

May be I could give you a better picture of my use-case and issue which I see with the current routing of incoming packet.

In our setup, as soon as end device show up on USB bus and it would immediately send out Discovery notify packet with NULL src and NULL dest header. By the time this happens our rules are not fast enough to assign address and create RTN_LOCAL routes for that mctpusb0 net device (but my question is irrespective of this). Thus when route lookup is being seen, even at route_lookup_null() this check for routes would fail and we'll be completely dropping the packet.

My question on this was, for packets which are destined to NULL_EID (dest=NULL), shouldn't these reach to socket rx queue to later propagate to any user space AF_MCTP sockets which are binded to NET_ANY, MCTP_ADD_ANY irrespective of the fact that local routes were there for net device or not.

[jk-ozlabs]

My question on this was, for packets which are destined to NULL_EID (dest=NULL), shouldn't these reach to socket rx queue to later propagate to any user space AF_MCTP sockets which are binded to NET_ANY, MCTP_ADD_ANY irrespective of the fact that local routes were there for net device or not.

Yes, that would be good.