Question

This is my first question on this site, having used it many times to help me in my c++ coding.

I am only asking because i’ve hit a wall, and just don’t get whats happening, and think its a nice little mystery for someone to solve.

I have a client / server architecture, with 1 server and multiple clients.

The code snippet below works fine, apart from the fact that the only socket that blocks is the last one in mList, so if 2 clients are connected only the second one to be connected receives data.
This code is not my own, i am attempting to debug a segmentation fault that occurs when moving an if statement.

int SocketManager::block(int secs, int usecs)
{
int ret = 0;
int result = 0;
struct timeval tv;
fd_set set;
int max_sd = 0;

if (mList.empty())
{
  return -1;
}

tv.tv_sec = secs;
tv.tv_usec = usecs;

FD_ZERO(&set);

SocketList::const_iterator iter;
int n = 0;

for (iter = mList.begin(); iter != mList.end(); iter++)
{
  if ((*iter)->socket() > max_sd)
  {
    max_sd = (*iter)->socket();
  }

  FD_SET((*iter)->socket(), &set);
  n++;
}

errno = 0;
if ((result = select(max_sd + 1, &set, NULL, NULL, &tv)) > 0)
{
  ret = 1;
  Socket *s = NULL;
  for (iter = mList.begin(); iter != mList.end(); iter++)
  {
    if (FD_ISSET((*iter)->socket(), &set))
    {
      s = (*iter);
    }
  }

  if (s)
  {
    mLastSocketPtr = s;
    s->checkForData();
    mLastSocketPtr = NULL;
  }
}
else if (result == 0)
{
  // Timeout
  ret = 0;
}
else
{
  // Error
  ret = -1;
}

return ret;
}

The problem occurs with the line “if (FD_ISSET((*iter)->socket(), &set))”, when the if statement

if (s)
{
  mLastSocketPtr = s;
  s->checkForData();
  mLastSocketPtr = NULL;
}

is moved inside the FD_ISSET function like so

for (iter = mList.begin(); iter != mList.end(); iter++)
      {
        if (FD_ISSET((*iter)->socket(), &set))
        {
          s = (*iter);
          if (s)
          {
            mLastSocketPtr = s;
            s->checkForData();
            mLastSocketPtr = NULL;
          }
        }
      }

Any help?

Answer 1

Most likely, the problem is that you are modifying mList while you are traversing it. By moving checkForData inside the for loop, any modification to mList inside checkForData is not allowed as it invalidates the iterator you are about to increment and dereference.

There are a lot of ways to fix this code. The best way would be not to write code like this in the first place, but I guess that ship has sailed. Here are a few suggestions:

1) Go through the FD_SET instead of going through mList. For each hit you find in the FD_SET, find the corresponding entry in mList. That way, you’ll be going through something that won’t be changing.

2) When you get a hit in the FD_SET, clear the flag in the FD_SET. Start the for loop over with a fresh call to begin.

3) Go through the list once and make a vector of all the sockets that need processing. Then iterate over that vector to call checkForData.

Here’s an ugly fix showing method 2:

for (iter = mList.begin(); iter != mList.end(); iter++)
  {
    if (FD_ISSET((*iter)->socket(), &set))
    {
      s = (*iter);
      FD_CLR(s->socket(), &set); // don't find this again
      mLastSocketPtr = s;
      s->checkForData();
      mLastSocketPtr = NULL;
      iter = mList.begin(); // make the iterator valid
      continue; // don't increment the iterator
    }
  }