On Fri, 7 Feb 2025 10:10:49 +0100 Nicolas Dichtel wrote: > Le 07/02/2025 à 00:39, Jakub Kicinski a écrit : > > On Thu, 6 Feb 2025 17:50:26 +0100 Nicolas Dichtel wrote: > >> Since the below commit, there is no way to see if the netns_local property > >> is set on a device. Let's add a netlink attribute to advertise it. > > > > I think the motivation for the change may be worth elaborating on. > > It's a bit unclear to me what user space would care about this > > information, a bit of a "story" on how you hit the issue could > > be useful perhaps? The uAPI is new but the stable tag indicates > > regression.. > To make it short: we were trying a new NIC with a custom distro provided by a > vendor (with out of tree drivers). We were unable to move the interface in > another netns. Thanks to ethtool we were able to confirm that the 'netns-local' > flag was set. Having this information helps debugging. Thanks, makes sense. Still a bit unsure if this is a stable candidate, if you don't mind net-next that'd be my preference. If you do mind, I'll live with it :) > >> @@ -2041,6 +2042,7 @@ static int rtnl_fill_ifinfo(struct sk_buff *skb, > >> netif_running(dev) ? READ_ONCE(dev->operstate) : > >> IF_OPER_DOWN) || > >> nla_put_u8(skb, IFLA_LINKMODE, READ_ONCE(dev->link_mode)) || > >> + nla_put_u8(skb, IFLA_NETNS_LOCAL, dev->netns_local) || > > > > Maybe nla_put_flag() ? Or do you really care about false being there? > It depends if the commit is backported or not. If it won't be backported, having > the false value helps to know that the kernel support this attribute (and so > that the property is not set). Wish we had a good solution for this, it's always the argument against flags :(
