Possible regression in exec_path handling

Denys Vlasenko vda.linux at googlemail.com
Fri Aug 26 22:56:08 UTC 2011


On Thursday 25 August 2011 20:49, Tom Gundersen wrote:
> Hi guys,
> 
> Since commit 4a2a86d5e7e7bf284a31af604a738dfa1f1a2240 (improve
> --install operation in chroot jail) I have had some problems. I see
> that the patch is meant to help when exec_path is /proc/self/exec, but
> it is causing problems for me when I use exec_path=/bin/busybox.
> 
> My usecase (somewhat simplified) is the following:
> 
> busybox is installed in /lib/busybox/busybox and not in /bin/busybox
> 
> cp /lib/busybox/busybox /mnt/bin/busybox
> /lib/busybox/busybox -s --install /mnt/bin
> chroot /mnt /bin/sh
> 
> Before the aforementioned commit this worked fine, but now /mnt/bin/sh
> points to /lib/busybox/busybox.
> 
> Any chance the commit could be reverted, or changed to only apply to
> the case when exec_path is /proc/self/exec?

I am leaning towards ripping out --install altogether,
and telling users that they should use something like

for applet in `busybox --list`; do
	ln -s busybox /mnt/bin/$applet
done

instead.

-- 
vda


More information about the busybox mailing list