Re: port leak in select

[Samuel Thibault]

Richard Braun, on Fri 02 Dec 2016 10:56:40 +0100, wrote:
> On Fri, Dec 02, 2016 at 10:43:04AM +0100, Samuel Thibault wrote:
> > Or simply
> > 
> >   (d[i].reply_port != MACH_PORT_NULL))
> 
> Aren't there cases where an entry could remain uninitialized ?

Here, no, we have the same structure:


  if (firstfd == -1)
    /* But not if there were no ports to deal with at all.
       We are just a pure timeout.  */
    portset = __mach_reply_port ();
  else
    {
      portset = MACH_PORT_NULL;

      for (i = firstfd; i <= lastfd; ++i)
+       if (d[i].type & SELECT_ERROR)
+           d[i].reply_port = MACH_PORT_NULL
+       else
-       if (d[i].type & ~SELECT_ERROR)
          {
            int type = d[i].type;
            d[i].reply_port = __mach_reply_port ();
            if (timeout == NULL)
              err = __io_select_request (d[i].io_port, d[i].reply_port, type);
            else



and

  if (firstfd != -1)
    for (i = firstfd; i <= lastfd; ++i)
-     if (d[i].type & ~(SELECT_ERROR | SELECT_RETURNED))
+     if (d[i].reply_port != MACH_PORT_NULL)
        __mach_port_destroy (__mach_task_self (), d[i].reply_port);

Samuel