mctpd: remove_peer if get endpoint id fails

[jinliangw]

method_net_learn_endpoint shall only keep the peer which is alive (responds to the query_get_peer_msgtypes).

Tested:

busctl call au.com.codeconstruct.MCTP1 /au/com/codeconstruct/mctp1/networks/1 au.com.codeconstruct.MCTP.Network1 LearnEndpoint y 8
sb "/au/com/codeconstruct/mctp1/networks/1/endpoints/8" false

busctl call au.com.codeconstruct.MCTP1 /au/com/codeconstruct/mctp1/networks/1 au.com.codeconstruct.MCTP.Network1 LearnEndpoint y 9
Call failed: MCTP Endpoint did not respond

busctl call au.com.codeconstruct.MCTP1 /au/com/codeconstruct/mctp1/networks/1 au.com.codeconstruct.MCTP.Network1 LearnEndpoint y 10
sb "/au/com/codeconstruct/mctp1/networks/1/endpoints/10" true

busctl call au.com.codeconstruct.MCTP1 /au/com/codeconstruct/mctp1/networks/1 au.com.codeconstruct.MCTP.Network1 LearnEndpoint y 15
Call failed: Request failed

[jk-ozlabs]

While the endpoint should respond to the Get MCTP Message Type Support command (the spec lists it as mandatory), we don't consider this a fatal error for enumeration in general.

The spec does call out Get Endpoint ID as the preferred mechanism for liveness detection - we use this in various parts of the recovery process, for example. If we want a liveness check included in the Network1/LearnEndpoint call, I think we should introduce a Get Endpoint ID command here.

[jinliangw]

If the bridged endpoint doesn't respond to "Get MCTP Message Type Support command", application layer daemons have no "official" way to know what messages(nvme-mi, pldm or spdm...) this endpoints supports. From this point of view, it doesn't have much value to expose a endpoint which we don't know which messages it supports.

Please let me know if you still prefer "Get Endpoint ID" as indication of live scan, I can update the pull request.

[jinliangw]

BTW, the original code just uses query_peer_properties(peer) without Get Endpoint ID anyway.

[jk-ozlabs]

If the bridged endpoint doesn't respond to "Get MCTP Message Type Support command", application layer daemons have no "official" way to know what messages(nvme-mi, pldm or spdm...) this endpoints supports

That's true, but there are some use-cases that do not consume the type data for endpoint discovery (ie, they use platform-specific mechanisms for endpoint enumeration). That's not really a primary concern in general, but we do still need to support it.

The presence of the endpoint on dbus should reflect whether or not the endpoint exists; the presence of types data in the SupportedMessageTypes property should reflect whether we have type data for that endpoint. I'd prefer to keep those concepts decoupled if possible.

(the disadvantage is that we have to issue an extra command in this process, but I think that's justified here)

[jinliangw]

Updated to use query_get_endpoint_id as liveness detection.

[jk-ozlabs]

I have just pushed some fixes for CI, which is likely causing the failures seen above. Are you able to rebase and push?