Re: [PATCH mptcp-next 1/2] mptcp: give MPTCP sockets a distinct lockdep class

[Matthieu Baerts <matttbe@kernel.org>]

Hi Geliang,

Thank you for the patches!

On 15/10/2026 12:42, Geliang Tang wrote:
> From: Geliang Tang <tanggeliang@kylinos.cn>
> 
> MPTCP sockets and their subflows share the same k-sk_lock-AF_INET
> lockdep class. When both are kernel sockets, lockdep flags every path
> that holds the MPTCP socket lock and then locks a subflow (setsockopt,
> connect, bind, worker send-ack) as "possible recursive locking".
> 
> Reclassify the MPTCP socket lock in mptcp_init_sock() to use a distinct
> lock class (sk_lock-AF_INET-MPTCP for IPv4, sk_lock-AF_INET6-MPTCP for
> IPv6). This approach is similar to nvme-tcp and avoids the need to
> reclassify each subflow socket individually.
> 
> The implementation is based on nvme_tcp_reclassify_socket() from
> drivers/nvme/host/tcp.c.

Note: we also do that in pm_kernel.c for the extra listening socket.

> Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
> ---
>  net/mptcp/protocol.c | 37 +++++++++++++++++++++++++++++++++++++
>  1 file changed, 37 insertions(+)
> 
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index 60cbd91be632..e63896a60595 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -56,6 +56,41 @@ DEFINE_PER_CPU(struct mptcp_delegated_action, mptcp_delegated_actions) = {
>  };
>  static struct net_device *mptcp_napi_dev;
>  
> +#ifdef CONFIG_DEBUG_LOCK_ALLOC

In pm_kernel.c we don't have that. Do we really need it? If all the
functions are nops, maybe not?

> +/* distinct lockdep class for MPTCP sockets */
> +static struct lock_class_key mptcp_sk_key[2];
> +static struct lock_class_key mptcp_slock_key[2];
> +
> +static void mptcp_reclassify_socket(struct socket *sock)
> +{
> +	struct sock *sk = sock->sk;
> +
> +	if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
> +		return;
> +
> +	switch (sk->sk_family) {
> +	case AF_INET:
> +		sock_lock_init_class_and_name(sk, "slock-AF_INET-MPTCP",
> +					      &mptcp_slock_key[0],
> +					      "sk_lock-AF_INET-MPTCP",
> +					      &mptcp_sk_key[0]);
> +		break;
> +#if IS_ENABLED(CONFIG_MPTCP_IPV6)
> +	case AF_INET6:
> +		sock_lock_init_class_and_name(sk, "slock-AF_INET6-MPTCP",
> +					      &mptcp_slock_key[1],
> +					      "sk_lock-AF_INET6-MPTCP",
> +					      &mptcp_sk_key[1]);

In pm_kernel.c we only call sock_lock_init_class_and_name() once. Maybe
we should do the same here, then no need to have a special helper?

> +		break;
> +#endif
> +	default:
> +		WARN_ON_ONCE(1);
> +	}
> +}
> +#else
> +static void mptcp_reclassify_socket(struct socket *sock) { }
> +#endif
> +
>  /* Returns end sequence number of the receiver's advertised window */
>  u64 mptcp_wnd_end(const struct mptcp_sock *msk)
>  {
> @@ -3292,6 +3327,8 @@ static int mptcp_init_sock(struct sock *sk)
>  	sk->sk_sndbuf = READ_ONCE(net->ipv4.sysctl_tcp_wmem[1]);
>  	sk->sk_write_space = sk_stream_write_space;
>  
> +	mptcp_reclassify_socket(sk->sk_socket);

Indeed, better in __mptcp_init_sock().

> +
>  	return 0;
>  }
>  

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.