Add MIB counters for fallback to TCP

[matttbe]

Because there is nothing about that, hard to track what's wrong.

[Dwyane-Yan]

Hi Matthieu:
I noticed there exists a patch named "mptcp: handle consistently DSS corruption." which adds two MIB counters, but it seems not the same as this issue, right?
Furthermore, upon reviewing the code, I believe that simply adding the code snippet below should work. Do you have any additional suggestions?

--- a/net/mptcp/subflow.c
+++ b/net/mptcp/subflow.c
@@ -1394,7 +1394,7 @@ static bool subflow_check_data_avail(struct sock *ssk)
                        WRITE_ONCE(subflow->data_avail, false);
                        return false;
                }
-
+               MPTCP_INC_STATS(sock_net(ssk), MPTCP_MIB_MIDDLEFALLBACK);
                mptcp_do_fallback(ssk);
        }

[matttbe]

Hi Gang Yan,

I noticed there exists a patch named "mptcp: handle consistently DSS corruption." which adds two MIB counters, but it seems not the same as this issue, right?

Correct: this patch adds new MIB counters, but for new code triggering a fallback.

The goal of this ticket here is to add counters where we do a fallback / reset subflow, due to an issue with the MPTCP option. Something likely triggered by a middlebox missing up with MPTCP options.

Furthermore, upon reviewing the code, I believe that simply adding the code snippet below should work. Do you have any additional suggestions?

Is there no other places where a fallback to TCP / subflows reset are done, and there are no counters? For example, I think that there are already counters when such fallback / reset are done at the beginning of the connection, but better to double-check that as well (e.g. in case of simultaneous connects, no MPTCP options after packet establishment, if MD5 (or TCP_AO?) is used, etc.).

Also, it would be good to cover that with tests, probably easier with packetdrill tests.

[Dwyane-Yan]

Is there no other places where a fallback to TCP / subflows reset are done, and there are no counters? For example, I think that there are already counters when such fallback / reset are done at the beginning of the connection, but better to double-check that as well (e.g. in case of simultaneous connects, no MPTCP options after packet establishment, if MD5 (or TCP_AO?) is used, etc.).

OK, I'll check it carefully.

Also, it would be good to cover that with tests, probably easier with packetdrill tests.

Thanks for your advice, I'll look into how to use it.

[geliangtang]

Hi Matt, I assigned this issue to myself since my colleague Gang Yan (Dwyane-Yan) in my group is looking at it.