On 18.06.2018 14:21, Kirill Tkhai wrote: > On 17.06.2018 21:58, David Ahern wrote: >> On 6/15/18 3:44 AM, Kirill Tkhai wrote: >>> Hm, but is this a likely case, when real device is moved to net ns, so it >>> requires moving to init_net back? It seems the most devices moved to >>> !init_net >>> are virtual and they just destroyed in default_device_exit_batch(). Or we >>> have >>> more devices to care here? >>> >>> I don't much want to insert here something like below: >>> >>> if (__dev_get_by_name(&init_net, dev->name)) >>> snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); >>> err = dev_change_net_namespace(dev, &init_net, "dev%d"); >>> >>> because dev_change_net_namespace() is generic interface and it's used not >>> only here, >>> and this will crumble the code in corner cases. >>> >>> Maybe you have better ideas about this? >> >> There are a lot of use cases these days (e.g., switch NOS) with 1000's >> (10's of 1000's) of netdevices. On top of that support for port netdevs >> in a namespace to create virtual switches needs to happen (and I suspect >> will happen in the next few years). That becomes one example where >> netdevices representing physical ports can be pushed back to init_net. > > Oh, then we really need to do something with rtnl_mutex. Otherwise > this will stop working at all. > >> That said, not many easy options at the moment for the bug you are fixing. >> >> Further, panic'ing a node because the move back to init_net fails is >> just wrong. > > So, let's fix it for now like in the patch to avoid the panic. Then we > can rework this in generic way to make the generic fallback name for moved > devices. Maybe, something like to give all moved device a fallback name > like "__moved-<hash of jiffies>-<generated id>".
Just to clarify, I mean that this small fix will be easy to backport to stable. Kirill
