tcp data modified after socket transmission - c++

Question

I am writing a program to transmit data between two machines with tcp/ip socket in C/C++. The programs checks the size of each read and write to make sure the correct size of data being transferred. I also implement a checksum verification to make sure the correct data are transferred. Most of time there is no problem. Once a while, like, once in 50,000 times, I notice the data received is the same length as the sending side but some of its content were altered. The altered portion is always 64 bytes long. Below is my code of send and read. Please help me figure out how this can happen and how to prevent it. Thanks!
bool WriteSocket(int pSocket, unsigned char* pData, uint32_t pNumBytes, uint32_t &pNumBytesWritten, uint32_t &pNumWrites)
{
int rc;
unsigned char* currPos = pData;
uint32_t numFailures = 0;
pNumBytesWritten = 0;
pNumWrites = 0;
while ( true )
{
rc = write(pSocket, currPos, pNumBytes);
pNumWrites++;
if ( rc < 0 )
{
if ( errno == EINTR )
continue;
numFailures++;
if ( numFailures >= 5 )
break;
else
continue;
}
pNumBytesWritten += rc;
if ( rc == pNumBytes )
return true;
pNumBytes -= rc;
currPos += rc;
numFailures = 0;
};
return false;
};
bool ReadSocket(int pSocket, unsigned char* pData, uint32_t pNumBytes)
{
int rc;
unsigned char* currPos = pData;
uint32_t numFailures = 0;
while ( true )
{
rc = read(pSocket, currPos, pNumBytes);
if ( rc < 0 )
{
if ( errno == EINTR )
continue;
numFailures++;
if ( numFailures >= 5 )
break;
else
continue;
}
if ( rc == pNumBytes )
return true;
pNumBytes -= rc;
currPos += rc;
numFailures = 0;
};
return false;
};

Answer 1

If a socket read/write fails for any reason other than EINTR or EAGAIN/EWOULDBLOCK, you should consider that as a fatal error. Do not retry the operation, just stop and close the socket. But your code doesn't do that, it keeps trying to repeat the failed operation.
Also, when reading, you are not handling the case where the read returns 0 when the peer disconnects gracefully. A socket error is not reported in that case.
Try something more like this instead:
bool WriteSocket(int pSocket, unsigned char* pData, uint32_t pNumBytes, uint32_t &pNumBytesWritten, uint32_t &pNumWrites)
{
int rc;
uint32_t numRetries = 0;
pNumBytesWritten = 0;
pNumWrites = 0;
while ( pNumBytes > 0 )
{
rc = write(pSocket, pData, pNumBytes);
if ( rc < 0 )
{
if ( errno == EINTR )
continue;
if ( (errno == EAGAIN) || (errno == EWOULDBLOCK) )
{
++numRetries;
if ( numRetries < 5 )
{
// TODO: use select() or epoll() w/ timeout to
// detect when then socket is writable again...
continue;
}
}
return false;
}
++pNumWrites;
pNumBytesWritten += rc;
pNumBytes -= rc;
pData += rc;
numRetries = 0;
}
return true;
}
bool ReadSocket(int pSocket, unsigned char* pData, uint32_t pNumBytes)
{
int rc;
uint32_t numRetries = 0;
while ( pNumBytes > 0 )
{
rc = read(pSocket, pData, pNumBytes);
if ( rc < 0 )
{
if ( errno == EINTR )
continue;
if ( (errno == EAGAIN) || (errno == EWOULDBLOCK) )
{
++numRetries;
if ( numRetries < 5 )
{
// TODO: use select() or epoll() w/ timeout to
// detect when then socket is readable again...
continue;
}
}
return false;
}
else if ( rc == 0 )
{
return false;
}
else
{
pNumBytes -= rc;
pData += rc;
numRetries = 0;
}
}
return true;
}