Wed, Nov 02, 2022 at 04:52:49PM CET, kuba@xxxxxxxxxx wrote: >On Wed, 2 Nov 2022 16:37:00 +0100 Jiri Pirko wrote: >> >> Maybe it's time to plumb policies thru to classic netlink, instead of >> >> creating weird attribute constructs? >> > >> >Not a blocker, FWIW, just pointing out a better alternative. >> >> Or, even better, move RTnetlink to generic netlink. Really, there is no >> point to have it as non-generic netlink forever. We moved ethtool there, >> why not RTnetlink? > >As a rewrite? We could plug in the same callbacks into a genl family >but the replies / notifications would have different headers depending >on the socket type which gets hairy, no? I mean like ethtool, completely side iface, independent, new attrs etc. We can start with NetdevNetlink for example. Just cover netdev part of RTNetlink. That is probably most interesting anyway.
