get rid of mptcp hooking in tcp_check_req()

[pabeni]

We currently have the following code in tcp_check_req():

      if (own_req && sk_is_mptcp(child) && mptcp_sk_is_subflow(child)) {
                reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
                inet_csk_reqsk_queue_drop_and_put(sk, req);
                return child;
        }

Can we instead, for mp_join child, in subflow_syn_recv_sock():

• complete the hashdance
• adjust sk refcnt,
• return *own_req == false
  ?
  If not, document properly why (here and/or directly in the code)

Note: no need to call subflow_syn_recv_sock() / sock_rps_save_rxhash() in subflow_syn_recv_sock(), tcp_check_req() will do that, since we will still return child != NULL

[pabeni]

The above schema cant' work:

• if syn_recv_sock() returns a != NULL child socket ptr, the request socket pointer must still be valid (will be de-referenced by tcp_check_req() before inet_csk_complete_hashdance());

• if syn_recv_sock() return own_req == false, tcp_check_req() will NOT release the req sock

Overall we lack a way to dispose the req socket 'stolen' by the MP_JOIN without our custom chunk

[pabeni]

see upstream commit:

commit 90bf451
Author: Paolo Abeni [email protected]
Date: Fri May 15 19:22:15 2020 +0200

mptcp: add new sock flag to deal with join subflows

I'm unable to do any better than that