Re: Raw can socket close and disabling filters

[Oliver Hartkopp <socketcan@hartkopp.net>]

Hi Laurent,

On 04/22/2016 01:39 PM, laurent vaudoit wrote:

> i'm using a 3.10 kernel with rtpatch applied, with raw can features.
> Running some debug in the release function, i've seen something weird (for
> me).
>
> In the function, we get the socket informations
> and then we unregister the notifier.
>
> After this, we test if the socket is bound and which is the ifindex.
>
> But when the unregister function is called, there is a call to the notifier
> with the event NETDEV_UNREGISTER, who leads to clear the socket inforamtions
> like bound and ifindex.
>
> So in the release function we can never enter in the code specific to a
> device.

The relevant removal is protected by lock_sock().

So when the notifier runs with NETDEV_UNREGISTER it will remove the 
filters & bind status.

In this case the block in raw_release() won't have anything to do to 
release the filters and the bind state, right?

Looks double work - but I wonder why we can rely on a notifier call when 
we call unregister_netdevice_notifier() ??

Regards,
Oliver