[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <736d1f6d-1b07-4671-857d-a0841828fce1@openvpn.net>
Date: Thu, 29 May 2025 12:07:48 +0200
From: Antonio Quartulli <antonio@...nvpn.net>
To: Paolo Abeni <pabeni@...hat.com>, netdev@...r.kernel.org
Cc: Sabrina Dubroca <sd@...asysnail.net>,
"David S . Miller" <davem@...emloft.net>, Eric Dumazet
<edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>,
Oleksandr Natalenko <oleksandr@...alenko.name>,
Qingfang Deng <dqfext@...il.com>, Gert Doering <gert@...enie.muc.de>
Subject: Re: [PATCH net-next 2/4] ovpn: ensure sk is still valid during
cleanup
On 29/05/2025 11:44, Paolo Abeni wrote:
> On 5/27/25 3:46 PM, Antonio Quartulli wrote:
>> diff --git a/drivers/net/ovpn/netlink.c b/drivers/net/ovpn/netlink.c
>> index bea03913bfb1..1dd7e763c168 100644
>> --- a/drivers/net/ovpn/netlink.c
>> +++ b/drivers/net/ovpn/netlink.c
>> @@ -423,9 +423,14 @@ int ovpn_nl_peer_new_doit(struct sk_buff *skb, struct genl_info *info)
>> ovpn_sock = ovpn_socket_new(sock, peer);
>> /* at this point we unconditionally drop the reference to the socket:
>> * - in case of error, the socket has to be dropped
>> - * - if case of success, the socket is configured and let
>> + * - if case of success, the socket is configured and we let
>> * userspace own the reference, so that the latter can
>> - * trigger the final close()
>> + * trigger the final close().
>> + *
>> + * NOTE: at this point ovpn_socket_new() has acquired a reference
>> + * to sock->sk. That's needed especially to avoid race conditions
>> + * during cleanup, where sock may still be alive, but sock->sk may be
>> + * getting released concurrently.
>
> This comment duplicate some wording from commit message contents and
> don't add much value IMHO. It could be dropped.
ACK
>
>> @@ -192,19 +189,30 @@ struct ovpn_socket *ovpn_socket_new(struct socket *sock, struct ovpn_peer *peer)
>> rcu_read_unlock();
>> }
>>
>> + /* increase sk refcounter as we'll store a reference in
>> + * ovpn_socket.
>> + * ovpn_socket_release() will decrement the refcounter.
>> + */
>> + if (!refcount_inc_not_zero(&sk->sk_refcnt)) {
>
> How could sk_refcnt be zero here? likely just sock_hold() is sufficient.
> Also I think the reference could be acquired a little later, avoiding at
> least the following chunk.
I just wanted to be safe.
But as you point out I don't see any reason why sk_refcnt would be 0 at
this point in time.
If we don't have to check the return value, I can definitely postpone
the sock_hold() a bit more.
>
> Also IMHO the comment is not very clear. I think it should state
> explicitly which entity is retaining the reference (AFAICS the peer hash
> table).
Ok, I will elaborate a bit more.
Essentially: sk -> ovpn_socket -> ovpn_peer -> peer hash
Thanks a lot!
Regards,
>
> /P
>
--
Antonio Quartulli
OpenVPN Inc.
Powered by blists - more mailing lists