Re: [patch net-next v6 5/9] genetlink: introduce per-sock family private storage

[Jiri Pirko <jiri@resnulli.us>]

Wed, Dec 13, 2023 at 12:49:38AM CET, kuba@kernel.org wrote:
>On Tue, 12 Dec 2023 11:17:32 +0100 Jiri Pirko wrote:
>> +static void genl_sk_priv_free(struct genl_sk_priv *priv)
>> +{
>> +	spin_lock(&priv->family->sock_priv_list->lock);
>> +	list_del(&priv->list);
>> +	spin_unlock(&priv->family->sock_priv_list->lock);
>> +	if (priv->destructor)
>> +		priv->destructor(priv->priv);
>> +	kfree(priv);
>
>> +static void genl_sk_priv_list_free(const struct genl_family *family)
>> +{
>> +	struct genl_sk_priv *priv, *tmp;
>> +
>> +	if (!family->sock_priv_size)
>> +		return;
>> +
>> +	list_for_each_entry_safe(priv, tmp, &family->sock_priv_list->list, list)
>> +		genl_sk_priv_free(priv);
>> +	kfree(family->sock_priv_list);
>
>Is this not racy for socket close vs family unregister?
>Once family starts to unregister no new privs can be installed
>(on the basis that such family's callbacks can no longer be called).
>But a socket may get closed in the meantime, and we'll end up entering
>genl_sk_priv_free() both from genl_release() and genl_sk_priv_list_free().
>
>Also I'm afraid there is a race still between removing the entry from
>the list and calling destroy.

Moreover, priv is not erased from gsk->family_privs xarray. Sigh, this
is pain.