Bug 2045848: Avoid avoid re-initting usrsctp a=pascalc

    DC_DEBUG(("Calling usrsctp_init %p", this));

    MOZ_DIAGNOSTIC_ASSERT(!sInitted);

    // usrsctp is a process-global singleton. If a previous DeinitUsrSctp()

    // could not tear the stack down (see the comment there), usrsctp is still

    // initialized and its timer thread is still running; re-initializing would

    // reset global locks and the callout queue out from under it and start a

    // second timer thread, corrupting memory. Reuse the live stack; the

    // callback and sysctl configuration below are still in effect.

    if (sInitted) {

      return;

    }

    usrsctp_init(0, DataChannelRegistry::SctpDtlsOutput, debug_printf);

    sInitted = true;

    MOZ_DIAGNOSTIC_ASSERT(NS_IsMainThread());

    MOZ_DIAGNOSTIC_ASSERT(sInitted);

    DC_DEBUG(("Calling usrsctp_finish %p", this));

    usrsctp_finish();

    // usrsctp_finish() is fallible: it returns -1 and tears nothing down (the

    // timer thread keeps running) while an endpoint whose destruction was

    // deferred (e.g. by an in-flight timer handler) is still on the global

    // list. Only clear sInitted on success, so InitUsrSctp() never

    // re-initializes usrsctp over a still-live stack.

    if (usrsctp_finish() == 0) {

      sInitted = false;

    }

  }

  uintptr_t mNextId = 1;

  std::map<uintptr_t, RefPtr<DataChannelConnection>> mConnections;