Re: [PATCH net-next v20 09/25] ovpn: implement basic RX path (UDP)

[Antonio Quartulli <antonio@openvpn.net>]

On 28/02/2025 16:25, Sabrina Dubroca wrote:
> (I'm still reviewing how everything fits together, but one small thing
> here:)
> 
> 2025-02-27, 02:21:34 +0100, Antonio Quartulli wrote:
>> +static void ovpn_udp_close(struct sock *sk, long timeout)
>> +{
>> +	struct ovpn_socket *sock;
>> +	struct ovpn_priv *ovpn;
>> +
>> +	rcu_read_lock();
>> +	sock = rcu_dereference_sk_user_data(sk);
>> +	if (!sock || !sock->ovpn) {
>> +		rcu_read_unlock();
>> +		return;
>> +	}
>> +	ovpn = sock->ovpn;
>> +	rcu_read_unlock();
>> +
>> +	if (ovpn->mode == OVPN_MODE_P2P)
>> +		ovpn_peer_release_p2p(ovpn, sk,
>> +				      OVPN_DEL_PEER_REASON_TRANSPORT_DISCONNECT);
>> +	sock->udp_prot->close(sk, timeout);
>> +}
>> +
>> +static void ovpn_udp_build_protos(struct proto *new_prot,
>> +				  const struct proto *orig_prot)
>> +{
>> +	memcpy(new_prot, orig_prot, sizeof(*new_prot));
>> +	new_prot->close = ovpn_udp_close;
>> +}
> 
> Could you set ->encap_destroy in udp_tunnel_sock_cfg, instead of
> making up your own struct proto? It should allow to do basically the
> same thing, but with less code in ovpn:
> 
> ->close = udp_lib_close -> sk_common_release -> sk_prot->destroy = udp_destroy_sock -> encap_destroy

Oh, this is very nice.
I'll give it a go and stick to encap_destroy if no issue arise.

Cheers,

> 
> 

-- 
Antonio Quartulli
OpenVPN Inc.