[PATCH] really close all extra fds in bb_daemonize_or_rexec

[Rob Landley]

On Tuesday 01 July 2008 03:26:31 Andreas Wetzel wrote:
> hi,
>
> the current implementation seems not to reliably close all open file
> descriptors if the flag DAEMON_CLOSE_EXTRA_FDS is set. it is assumed,
> that the fd returned by opening a file (bb_dev_null) is the top-most fd,
> but this is not true. open returns the first free fd. so, potentially,
> some fds remain open.
>
> the attached patch fixes this by calling close() on every possible file
> descriptor (diffed against the busybox_1_9_stable branch).

I wonder how that interacts with the non-sequential file descriptor stuff?  (I 
never did follow to see if it was merged, but http://lwn.net/Articles/237722/ 
is a starting point.)

Personally, I'd probably just iterate through /proc/$PID/fd and use the actual 
list of the process's open file descriptors to know what to close.  But I can 
see the potential downsides of that probably being larger and depending 
on /proc.  (Not that getdtablesize is the most portable call out there 
either, considering that even when it exists it can return INT_MAX...)

Rob
-- 
"One of my most productive days was throwing away 1000 lines of code."
  - Ken Thompson.