reproducible bug in 2.2.19 & 2.4.x

[Rainer Keller]

Hi Balazs,

Well, reading & running (on UP, no SIGSEGVs) Your code, I get the
impression, that there might be a small bug in it.

Your while-loop looks like this:
 while (1)
    {
      int newfd;
      int tmplen;
      pthread_t t;
      tmplen = sizeof(sin);
      newfd = accept(fd, (struct sockaddr *) &sin, &tmplen);
      ...
      pthread_create(&t, NULL, thread_func, &newfd);
   }

Now, if the thread_func is not started immediately on Your SMP-machine,
the next iteration of the loop might already have started the accept
with a new file-descriptor, leaving two threads reading on the same fd.

This should be per-thread data.

Just my 2 cents.

Greetings,
raY