Re: rtnl_newlink() cleanup on namespace change

[Cong Wang <xiyou.wangcong@gmail.com>]

On Tue, Mar 10, 2015 at 7:00 PM, David Miller <davem@davemloft.net> wrote:
> From: Cong Wang <xiyou.wangcong@gmail.com>
> Date: Tue, 10 Mar 2015 16:44:06 -0700
>
>> On Tue, Mar 10, 2015 at 3:43 PM, David Miller <davem@davemloft.net> wrote:
>>>
>>> Cong, I was reviewing commit 7afb8886a05be68e376655539a064ec672de8a8e
>>> ("rtnetlink: call ->dellink on failure when ->newlink exists") during
>>> a stable backport and I noticed that the function, after the change
>>> you made, subsequently goes:
>>>
>>>                 if (link_net) {
>>>                         err = dev_change_net_namespace(dev, dest_net, ifname);
>>>                         if (err < 0)
>>>                                 unregister_netdevice(dev);
>>>                 }
>>>
>>> Isn't the potential ->dellink() unwinding necessary in this path too?
>>>
>>
>> Right, I wasn't aware of the link_net change.
>> Do you want me to send a fix?
>
> I took care of it, as follows:

Thanks, it looks good.

>
> ====================
> [PATCH] net: Handle unregister properly when netdev namespace change fails.
>
> If rtnl_newlink() fails on it's call to dev_change_net_namespace(), we
> have to make use of the ->dellink() method, if present, just like we
> do when rtnl_configure_link() fails.
>
> Fixes: 317f4810e45e ("rtnl: allow to create device with IFLA_LINK_NETNSID set")
> Signed-off-by: David S. Miller <davem@davemloft.net>

Acked-by: Cong Wang <xiyou.wangcong@gmail.com>