Thread safety using WSAOVERLAPPED

[maxsharabayko]

PR #2834 has a problem, The srt::CChannel::sendto will be call in CSndQueue(thread1) and CRcvQueue(thread2), and that not thread safety.

Them used same WSAOVERLAPPED, so WSASendTo maybe not completed when WSAWaitForMultipleEvents return.

From https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-wsasendto

The lpOverlapped parameter must be valid for the duration of the overlapped operation. If multiple I/O operations are simultaneously outstanding, each must reference a separate WSAOVERLAPPED structure.

Originally posted by @mGaosi in #2834 (comment)

[mGaosi]

We found this issue during our app testing. When sending 400mbps of data in our program, some exceptions quickly occurred and the stack was damaged. The probability of this problem occurring is related to your actual traffic and the performance of the network subsystem.

I am currently using a temporary solution, using TLS WSAEvent and stack WSAOVERLAPPED.

    WSAOVERLAPPED overlapped;
    ::SecureZeroMemory(&overlapped, sizeof(overlapped));
    class WSAEventRef
    {
    public:
        WSAEventRef()
            : e(::WSACreateEvent())
        {
        }
        ~WSAEventRef()
        {
            ::WSACloseEvent(e);
            e = NULL;
        }
        void reset()
        {
            ::WSAResetEvent(e);
        }
        WSAEVENT handle()
        {
            return e;
        }

    private:
        WSAEVENT e;
    };
    thread_local WSAEventRef lEvent;
    overlapped.hEvent = lEvent.handle();
    int res = ::WSASendTo(m_iSocket, (LPWSABUF)packet.m_PacketVector, 2, &size, 0, addr.get(), addrsize, &overlapped, nullptr);

[maxsharabayko]

The Windows build of SRT enables C++11 by default. If disabled, the pthread port is being used. But it is waiting to be deprecated.

@mGaosi
I believe your solution works well and is worth to make a PR out of it. I would just do something like this:

#if HAVE_CXX11
thread_local WSAEventRef lEvent;
#else
WSAEventRef lEvent;
#endif

P.S. Look like HAVE_CXX11 and ENABLE_CXX11 are not working together. 🤔

[ethouris]

If we state that Windows can only be built with C++11, then it should suffice to declare thread_local WSAEventRef lEvent under a condition for WIN32 only.

[maxsharabayko]

Resolved by #2929.