Re: [PATCH mptcp-next v1 5/5] mptcp: poll allow write call before actual connect

[Paolo Abeni <pabeni@redhat.com>]

On Wed, 2022-09-21 at 17:25 +0200, Benjamin Hesmans wrote:
> If fastopen is used, poll must allow a first write that will trigger
> the SYN+data
> 
> Similar to what is done in tcp_poll().
> 
> Signed-off-by: Benjamin Hesmans <benjamin.hesmans@tessares.net>
> ---
>  net/mptcp/protocol.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index aa6e13949b23..308a09b3c0b6 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -3733,10 +3733,12 @@ static __poll_t mptcp_poll(struct file *file, struct socket *sock,
>  {
>  	struct sock *sk = sock->sk;
>  	struct mptcp_sock *msk;
> +	struct socket *ssock;
>  	__poll_t mask = 0;
>  	int state;
>  
>  	msk = mptcp_sk(sk);
> +	ssock = __mptcp_nmpc_socket(msk);

The above is not safe. We need to own the msk socket lock to invoke
__mptcp_nmpc_socket(), and we really want to avoid acquiring such lock
here.

Instead you can copy the ssk 'defer_connect' into the msk at
mptcp_stream_connect() time...

> 
>  	sock_poll_wait(file, sock, wait);
>  
>  	state = inet_sk_state_load(sk);
> @@ -3751,6 +3753,9 @@ static __poll_t mptcp_poll(struct file *file, struct socket *sock,
>  	if (state != TCP_SYN_SENT && state != TCP_SYN_RECV) {
>  		mask |= mptcp_check_readable(msk);
>  		mask |= mptcp_check_writeable(msk);
> +	} else if (ssock && state == TCP_SYN_SENT && inet_sk(ssock->sk)->defer_connect) {

... and check it here.

Cheers,

Paolo