Networking services sockets is based on the Berkeley Software Distribution version 4.4 sockets implementation.

The IBM OS/2 socket implementation differs from the Berkeley socket implementation as follows:

• Sockets are not OS/2 files or devices. Socket numbers have no relationship to OS/2 file handles. Therefore, the read(), write(), and close() calls do not work for sockets: using them gives incorrect results. Use the recv(), send(), and soclose() calls instead.

• Error codes set by the OS/2 TCP/IP sockets implementation are not made available via the global errno variable. Instead, error codes are accessed by using the sock_errno() call (see sock_errno()). Use the psock_errno() call, instead of the perror() call, to write a short error message to the standard error device describing the last error encountered during a call to a socket library function. To access system return values, use the errno.h include statement supplied with the compiler.

  This is intended to obtain per-thread error codes in a multithreaded application environment and to avoid conflict with standard ANSI C error constants.

  BSD-style error checking is as follows:

  • rt = recv(s, buf, sizeof(buf), 0);
    if (rt == -1 && errno == EWOULDBLOCK)
    {...}
    

  • if (recv(s, buf, sizeof(buf), 0) < 0) {
    perror("Recv()");
    exit(1);
    }
    

  The preferred OS/2-style error checking is as follows:

  • rt = recv(s, buf, sizeof(buf), 0);
    if (rt == -1 && sock_errno() == SOCEWOULDBLOCK)
    {...}
    

  • if (recv(s, buf, sizeof(buf), 0) < 0)
    {
    psock_errno("Recv()");
    exit(1);
    }
    

  Error constants consistent with BSD sockets are provided for compatibility purposes; your application can use the error constant EWOULDBLOCK, instead of SOCEWOULDBLOCK. See Socket Error Constants, or the <NERRNO.H> file for definitions of error constants.

Unlike the Berkeley select() call, you cannot use the OS/2 select() call to wait for activity on devices other than sockets. See select() for more information.

The ioctl(), getsockopt(), setsockopt(), and sysctl() calls don't support all of the options supported by BSD and add some options not supported by BSD. See ioctl(), getsockopt(), setsockopt(), and sysctl() for more information.