Threading and Sockets

[vipergt023]

Hi,

I have a program that reads information from an IP device (camera). It establishes the Telnet connection and then via a thread, initializes the camera (runs a bunch of commands to get the camera's basic configuration). I keep getting the same two errors: "User called breakpoint at 0xXXXXXX" or Access violation. It seems that it always quits whenever I do a CString concatenation on the receive side. I'm using a CString to act as a FIFO buffer and process whatever data is in the buffer. If I step through the program in debug, it often works fine, but with no breakpoints set, it crashes intermittently. I use CCriticalSection to try and protect the operations that modify the buffer and it reduces it, but doesn't eliminate it. Any ideas? Thanks.

Vinoj

[mac11]

I'll try to offer some general ideas, but mostly plead for more details...

I would bet that if you're getting intermittent crashes it's likely some buffer overrun, or unbound array access where you're clobbering memory. Though I have never seen this particular error and it could end up being something completely different.

What OS and compiler are you using?

Are you doing memory allocation and memory copy stuff alot? Maybe it would help if you can use c++ stl containers (string, streams, streambuf, and such) so you won't have to worry about memory management so much (I assume you are).

It's possible something in your thread handling is causing the issue, maybe you can try to post some code example?

[RRick]

The access violation means you are trying to access some restricted area. This is usually caused by bad pointers or bad data. Unfortunately, there are a lot of things that can cause this type of memory corruption.

A memory checker would be helpful in this situation, or logs of the pointer values/data before you try to access the string.

You mentioned you are using threads. You might have a race condition where one thread is adding data and another is extracting or accessing the data before the data is available. Its typical for a race condition to not show up in the debugger, because breakpoints, etc. tend to slow down the program and the race condition doesn't happen.

[vipergt023]

mac11, I'm using XP and VC++ 6. I'm using CStrings so I think that is a self-managed entity. I add received data to a CString buffer, and delete it when I've processed it. I wasn't doing any explicit memory management.

RRick, I run the heap check function to check the status of the heap and it always returns ok. Is there a good/free memory checker tool you'd recommend? I'll try and post the errors and some code.

[vipergt023]

Ok,

Here's the code causing problems. I've cut out some of the repetitive parts to make it more concise. The code calls ReceiveTCP to append rx'd data to a CString buffer, then calls OnTcpip() to process the data. The error occurs during a CString destructor - FreeData call.

int CIrCamera::SendTCP()
{
    _critSect.Lock();
    int bytesSent = 0, nRet = 0;
    char toSend = 0;
    CString tempSendBuff = m_strSendBuff;
 
    //insert delay on sending, 5 ms per character, 50 ms per line
    while(tempSendBuff.GetLength() > 0){
        if(tempSendBuff[0] == '\n')// || tempSendBuff[0] == '\r')
            Sleep(50);
        else
            Sleep(5);
        toSend = tempSendBuff[0];
        bytesSent += m_pSocket->Send(&toSend, 1, 0);
        if(bytesSent < 0){
            bytesSent = GetLastError();
            break;            
        }
        tempSendBuff.Delete(0);
    }
    m_strSendBuff.Empty();
    tempSendBuff.Empty();
 
    if(bytesSent == WSAECONNABORTED || bytesSent == WSAECONNRESET){
        Disconnect();
        Connect();
    }
    m_iPending++;
 
    //giving time for responses to 
    Sleep(50);
    _critSect.Unlock();
 
    return bytesSent;
}
 
int CIrCamera::ReceiveTCP() 
{
    _critSect.Lock();
    int nRet = 0;
    memset(m_cBuff, 0, sizeof(m_cBuff));
    nRet = m_pSocket->Receive(m_cBuff, BUFFER_SIZE, 0);
    temp = m_strRecvBuff + m_cBuff;
    m_strRecvBuff += m_cBuff;
    _critSect.Unlock();
    return nRet;
}
 
void CIrCamera::OnTcpip()
{
 
    int bFailedOpCtr = 0;
    _critSect.Lock();
 
    while(m_strRecvBuff.GetLength() > 0 && !m_bIncompleteRx){
        CheckForTelnetText();
        switch(m_iCurrentState){
            case 0:                
                //initialize telnet
                TelnetInit();
                m_iPending=0;
                break;
            case 1:
                m_iPending--;
                //telnet command processing state
                switch(GetCurrentOperation()){
                    case 1:
                        m_bOpCompleted = ReadFOVLimits();
                        break;
                    case 2:
                        m_bOpCompleted = ReadFocusLimits();
                        break;
                    case 3:
                        m_bOpCompleted = ReadFOV();
                        break;
                    case 4:
                        m_bOpCompleted = ReadFocus();
                        break;
//..cut out case statements that call parsing functions
          case -1:
                        m_bIncompleteRx = TRUE;
                        break;
                }
                if(m_strRecvBuff.GetLength() <=0)
                    m_pExtDoc->StoreIrCamera();
 
                break;                
        }
        //if a failed op, reset buffer in case of garbage
        if(!m_bOpCompleted){
            bFailedOpCtr++;
            if(bFailedOpCtr > 25)
                m_strRecvBuff.Empty();
        }
 
    }
    //when rx completes, check current op
    if(m_bOpCompleted){
        m_cCurrentOperation = 0;
        m_bIncompleteRx = FALSE;
    }
    _critSect.Unlock();
 
}
 
char CIrCamera::GetCurrentOperation()
{
 
    //store the global in case an incomplete command comes through
    _critSect.Lock();
    char currentOp = m_cCurrentOperation;
    CString tempRecvBuff = m_strRecvBuff;
 
    //if there is intermittent telnet commands in the stream
    StripSpecialChars();
 
    if(m_bOpCompleted){
        CString temp;
        temp = tempRecvBuff.Left(tempRecvBuff.Find('\n'));
        //if no '\n' in buffer
        if(temp.GetLength() == 0){
            temp = tempRecvBuff;
        }
 
        //Set Commands - do not require parsing
        if(temp.Find("overlay on") > -1){
            tempRecvBuff.Delete(0, temp.GetLength()+1);
            currentOp = 0;
        }
        else if(temp.Find("overlay off") > -1){
            tempRecvBuff.Delete(0, temp.GetLength()+1);
            currentOp = 0;
        }
        //... cut out a bunch of find() commands...//
        }
        //Telnet Echo
        else if(temp.Find(m_strSend) > -1){
            tempRecvBuff.Delete(0, temp.GetLength()+1);
            currentOp=10;
        }
        else{
            currentOp = -1;
        }
 
        //set the global
        m_cCurrentOperation = currentOp;
    }
    m_strRecvBuff.Empty();
    m_strRecvBuff = tempRecvBuff;
    _critSect.Unlock();
    return m_cCurrentOperation;
}

[vipergt023]

The main error right now are Access Violations in the ReceiveTCP() function during the concatenation of the string (the +=). I think I may have a race condition as previously mentioned. How do I check for that? I thought CriticalSections prevented more than one thread from accessing a section...

[gadams]

I would try looking into a .net hotfix. I have found a similar issue with a telnet device and a race condition, that I believe we are having with our intermec scan guns. Please see hotfix: http://support.microsoft.com/kb/2759112 for asp.net. This is for a later issue of windows.... but, I believe from what you describe a very similar issue.