C++-socket接包切数据时遇到的诡异问题

Question

最近在写一个socket方面的小程序，遇见一个我个人觉得很诡异问题。
客户端发包我用php写的，代码很简单并且我也测试过，没有任何问题，我就不贴出来献丑了，调用接口为：
function send($socket, $data)
$data参数就是php客户端要发送给服务器的数据，这个send函数会把数据切割成一个一个的片段，并给每个片段增加协议头，然后调用socket_write发送给服务器。

服务器程序用c++写的，一个简单的测试代码，功能是：连接好socket后，先尝试接收固定长度的数据，这个数据就是协议头，从协议头中取出给定好的数据长度和数字签名，然后按照数据长度获取数据，拿到数据后再和数字签名比对看看是否合法，若合法则表明本次收包没问题，接续接受下一个包。。。。

整个流程很简单，下面我贴一下我写的c++代码：

/**
 * receive the data from the socket
 *
 * protocol format: [flag(1) | size(4) | md5(32) | data(size) ]
 *
 * @param socket
 * @param data
 */
int Receive(int socket, std::string *data)
{
    while(1)
    {
        //first, get the protocol header
        int total = HEADERSIZE;
        char header[HEADERSIZE] = {'0'};
        int howMany = 0;
        while(total > 0)
        {
            howMany = read(socket, header + howMany, total);
            if(howMany < 0)
                return -1;

            total -= howMany;
        }

        //verify the flah
        if('M' != header[0])
            return -2;

        //get the data size
        char tmpLength[4];
        strncpy(tmpLength, header+1, 4);
        int length = atoi(tmpLength);
        if(length <= 0)
            return length;  //if the length < 0, means something wrong happen, or if the length = 0, means the data finish.

        //get the data
        char buffer[length];
        //memset(buffer, '0', length);
        howMany = 0;
        while(length > 0)
        {
            howMany = read(socket, buffer + howMany, length);
            if(howMany < 0)
                return -1;

            length -= howMany;
        }

        //verify the data
        std::string onePart(buffer);
        MD5 md5(onePart);
        std::string verification(md5.md5());
        if(verification.compare(0, 32, header + 5, 32) != 0)
            return -3;

        //merge the part of data
        data->append(onePart);
    }

    std::cout << std::endl;   //i really fucking do not know why must do this code .
    return 0;
}

Answer 1

/**
* receive the data from the socket
*
* protocol format: [flag(1) | size(4) | md5(32) | data(size) ]
*
* @param socket
* @param data
*/
int Receive(int socket, std::string *data)
{
while(1)
{
//first, get the protocol header
int total = HEADERSIZE;
char header[HEADERSIZE] = {'0'};
int howMany = 0;
while(total > 0)
{
howMany = read(socket, header + howMany, total);
if(howMany < 0)
return -1;

total -= howMany;
}

//verify the flah
if('M' != header[0])
return -2;

//get the data size
char tmpLength[4];
strncpy(tmpLength, header+1, 4);
int length = atoi(tmpLength);
if(length <= 0)
return length; //if the length < 0, means something wrong happen, or if the length = 0, means the data finish.

//get the data
char buffer[length];
//memset(buffer, '0', length);
howMany = 0;
total = length;
while(total > 0)
{
howMany = read(socket, buffer + howMany, total);
if(howMany < 0)
return -1;

total -= howMany;
}

//verify the data
std::string onePart(buffer);
onePart = onePart.substr(0, length); //***only get the valid size of data, this is so important
MD5 md5(onePart);
std::string verification(md5.md5());

if(verification.compare(0, 32, header + 5, 32) != 0)
return -3;

//merge the part of data
data->append(onePart);
}

return 0;
}

这是我更正了问题中的相关代码后的一个版本，请看我问题下面的留言