Re: WARNING: kobject bug in br_add_if

[Nikolay Aleksandrov <nikolay@cumulusnetworks.com>]

On 26/04/18 13:37, Hangbin Liu wrote:
> On Thu, Apr 26, 2018 at 10:04:16AM +0200, Dmitry Vyukov wrote:
>> On Thu, Apr 26, 2018 at 8:13 AM, Hangbin Liu <liuhangbin@gmail.com> wrote:
>>> On Wed, Apr 11, 2018 at 05:18:23PM +0200, Dmitry Vyukov wrote:
>>>> On Wed, Apr 11, 2018 at 5:15 PM, syzbot
>>>> <syzbot+de73361ee4971b6e6f75@syzkaller.appspotmail.com> wrote:
>>>>> kobject_add_internal failed for brport (error: -12 parent: bond0)
[snip]
> 
> Re-checked the error. This is a -ENOMEM. So normally we could ignore it.
> 
> But on the other hand, although we could find out the slave iface's
> master in netdev_master_upper_dev_link(). It already go much further
> and allocate some resource and change iface state. e.g.
> 
> [54273.968516] br0: port 1(em1) entered blocking state
> [54273.973979] br0: port 1(em1) entered disabled state
> 
> So I think we'd better return as early as possible. I will post a fix
> for this.
> 
> Thanks
> Hangbin

If I'm not mistaken the bridge allocated resources for the port are
cleaned on kobject_init_and_add() error return. Or are you talking
about some other resources ?