Re: [MPTCP][PATCH mptcp-next 0/9] fullmesh path manager support

[Paolo Abeni <pabeni@redhat.com>]

On Fri, 2021-07-09 at 19:04 +0800, Geliang Tang wrote:
> Implement the in-kernel fullmesh path manager like on the mptcp.org
> kernel.
> 
> Closes: https://github.com/multipath-tcp/mptcp_net-next/issues/193

Following-up the yday disucssion in the public mtg, I skimmed over this
series.

I think this approach leads to quite a bit of duplicate code and
avoidable complexity.

I also think we could obtain a full-mash topology with some not to-
complex extension to the current NL PM:

- add and manage a new per endpoint flag, something alike 'fullmesh'
- in mptcp_pm_create_subflow_or_signal_addr(), if such flag is set,
  instead of:

	remote_address((struct sock_common *)sk, &remote);

  fill a temporary allocated array of all known remote address.
  After releaseing the pm lock loop on such array and create a subflow 
  for each remote address from the given local.
  Note that the we could still use an array even for non 'fullmash' 
  endpoint: with a single entry corresponding to the primary MPC
  subflow remote address.
- mptcp_pm_nl_add_addr_received(), fill a temporary allocate array of 
  all local address corresponding to fullmash endpoint. If such array 
  is empty, keep the current behavior.
  Elsewhere loop on such array and create a subflow for each local 
  address towards the given remote address

I hope that overall the above would require a limited amount of
changes. If so, I believe this way to be preferrable:
- 1 in kernel path manager
- a simple one
- a single configuration interface
- should cover the full-mesh use case and possibly more
- the idea is - some far day - use ebpf for more funcy stuff, if
needed.

WDYT?

Thanks!

Paolo