Re: [PATCH mptcp-next v16 5/8] mptcp: add get_subflow wrapper

[Paolo Abeni <pabeni@redhat.com>]

On Wed, 2022-04-27 at 10:27 +0200, Paolo Abeni wrote:
> > +
> > +	msk->sched ? INDIRECT_CALL_INET_1(msk->sched->get_subflow,
> > +					  mptcp_get_subflow_default,
> > +					  msk, reinject, &data) :
> > +		     mptcp_get_subflow_default(msk, reinject, &data);
> 
> With this we have quite a lot of conditionals for the default
> scheduler. I think we can drop the INDIRECT_CALL_INET_1() wrapper, and
> rework the code so that msk->sched is always NULL with the default
> scheduler.

Even better, we could drop the mptcp_get_subflow_default() wrapper and
call directly the current scheduler when msk->sched is NULL, even
before filling the 'data' struct. That should fit nicely expecially if
we use 2 separate ops for plain xmit and retrans.

Overall that would lead to something alike:

---
static inline struct sock *mptcp_sched_get_send(struct mptcp_sock msk)
{
	struct mptcp_sched_data data;

        sock_owned_by_me(sk);

	/* the following check is moved out of mptcp_subflow_get_send */
        if (__mptcp_check_fallback(msk)) {
                if (!msk->first)
                        return NULL;
                return sk_stream_memory_free(msk->first) ? msk->first : NULL;
        }

	if (!msk->sched)
		return mptcp_subflow_get_send(msk);

	data.sock = msk->first; /* if needed */
	data.call_again = 0;	/* if needed */
	msk->sched->get_send(msk, &data);
	msk->last_snd = data.sock;
	return data.sock;
}
---
plus something similar for retrans.

Thanks!

Paolo