Problem with ash redirection changing behaviour

[Richard Purdie]

On Wed, 2010-06-02 at 14:08 +0200, Cristian Ionescu-Idbohrn wrote:
> On Wed, 2 Jun 2010, Richard Purdie wrote:
> > I have an issue having recently upgraded busybox. The change that is
> > causing me problems is:
> >
> > http://git.busybox.net/busybox/commit/?id=6a0ad2506116f4ddc3f9f617a90ba04a57eeef88
> >
> > which is sanity checking file descriptors before allowing redirection.
> >
> > A script that triggers the problem is:
> >
> > """
> > exec 9>&0 </etc/fstab
> > while read fs mnt type opts dump pass junk
> > do
> > 	echo $fs
> > done
> > exec 0>&9 9>&-
> > """
> 
> That should work.
> 
> > The idea being to save stdin, read from a file and then restore stdin
> > without forking.
> 
> But so should this (no forks here either, and simplier):
> 
> while read fs mnt type opts dump pass junk; do
> 	case $fs in
> 		\#*|'')
> 			continue
> 			;;
> 		*)
> 			echo $fs
> 			;;
> 	esac
> done < /etc/fstab
> 
> Does the above work for you?

Yes, that works fine and is simpler. Its as soon as you touch the stdin
fd that things go wrong.

I already changed my script to do:

exec 9< /etc/fstab
while read fs mnt type opts dump pass junk <&9

which is still more complicated than your version but also works

> > This used to work but now when run as ". script" from ash
> 
> Any particular reason to 'source'?  You end up with environment
> clutter:
> 
> dump=''
> fs=''
> junk=''
> mnt=''
> opts=''
> pass=''
> type=''
> 
> Is that intentional?

Its sourced to avoid forking. As you've probably guessed, this is a
fragment of an initscript and when you stop forking for many
initscripts, it does significantly improve boot time. The environment
clutter is not intentional and something I should look into though, good
point.

Cheers,

Richard