Re: [patch net-next v4 5/9] genetlink: introduce per-sock family private pointer storage

[Jiri Pirko <jiri@resnulli.us>]

Mon, Nov 27, 2023 at 12:13:09PM CET, pabeni@redhat.com wrote:
>On Thu, 2023-11-23 at 19:15 +0100, Jiri Pirko wrote:
>[...]
>> +/**
>> + * genl_sk_priv_store - Store per-socket private pointer for family
>> + *
>> + * @sk: socket
>> + * @family: family
>> + * @priv: private pointer
>> + *
>> + * Store a private pointer per-socket for a specified
>> + * Generic netlink family.
>> + *
>> + * Caller has to make sure this is not called in parallel multiple times
>> + * for the same sock and also in parallel to genl_release() for the same sock.
>> + *
>> + * Returns: previously stored private pointer for the family (could be NULL)
>> + * on success, otherwise negative error value encoded by ERR_PTR().
>> + */
>> +void *genl_sk_priv_store(struct sock *sk, struct genl_family *family,
>> +			 void *priv)
>> +{
>> +	struct genl_sk_ctx *ctx;
>> +	void *old_priv;
>> +
>> +	ctx = rcu_dereference_raw(nlk_sk(sk)->priv);
>
>Minor nit: Looking at the following patch, this will be called under
>the rtnl lock. Since a look is needed to ensure the priv ptr

Well, depends on the user (as the comment says). Devlink does not call
it with rtnl lock. Relies on the socket skb processing and cleanup
flows.


>consistency, what about adding the relevant lockdep annotation here?
>
>No need to repost for the above.
>
>Cheers,
>
>Paolo
>