Should there be a separate 'host loopback' path? about pna

[stolsma]

If the 'host loopback' arrow is eliminated, would it be possible for a packet to skip message processing by using recirculation?

@rst0git As I recall correctly (but I could be wrong here) the current spec doesn't allow recirculation to enter the other pipeline direction (net-to-host -> recirculate -> host-to-net) currently it stays in the same direction (net-to-host -> recirculate -> net-to-host and host-to-net -> recirculate -> host-to-net). My opinion is that changing this recirculate requirement would not be the right way to 'solve' the host loopback path question...

[rst0git]

If the 'host loopback' arrow is eliminated, would it be possible for a packet to skip message processing by using recirculation?

[stolsma]

After thinking about this question a long time I came to the following opinion, hopefully it will help...

If it means that this decision will also remove the packet-from-host-loopback packetpath type indication its a definitive no go for me because it then will be impossible to separate host originating packets from host side recirculated packets. We can add packet headers field indicating host recirculated packets but that would be not so nice as it will use precious parsing and control code space.

Also, if the recirculated packets go through the message processing block, the message processing block (and maybe future host side loaded MP block programs) must obey the host-loopback indication given by the from-net-to-host pipeline. In the case of separate control of the pipeline and message processing module (as described at the end of paragraph 1.2 'Message processing') that could be not the case and and my gut feeling says this may result in a serious security risk (again, gut feeling...).

Very curious what other opinions are!