Re: [PATCH v1 2/5] connector/cn_proc: Add filtering to fix some bugs

[Jakub Kicinski <kuba@kernel.org>]

On Tue, 14 Mar 2023 02:32:13 +0000 Anjali Kulkarni wrote:
> This is clearly a layering violation, right?
> Please don't add "if (family_x)" to the core netlink code.
>
> ANJALI> Yes, it is, but there does not seem a very clean way to do it
> ANJALI> otherwise and I saw a check for protocol NETLINK_GENERIC just
> ANJALI> below it, so used it for connector as well. There is no
> ANJALI> release or free callback in the netlink_sock. Is it ok to add
> ANJALI> it? There was another bug (for which I have not yet sent a
> ANJALI> patch) in which, we need to decrement
> ANJALI> proc_event_num_listeners, when client exits without calling
> ANJALI> IGNORE, else that count again gets out of status of actual no
> ANJALI> of listeners.   
> The other option is to add a flag in netlink_sock, something like
> NETLINK_F_SK_USER_DATA_FREE, which will free the sk_user_data, if
> this flag is set. But it does not solve the above scenario.

Please fix your email setup, it's really hard to read your replies.

There is an unbind callback, and a notifier. Can neither of those 
be made to work? ->sk_user_data is not a great choice of a field,
either, does any other netlink family use it this way?
Adding a new field for family use to struct netlink_sock may be better.