Bug#774227: busybox-static: execs applets when chrooting
Control: tags -1 confirmed
Control: severity -1 normal
On 03/07/17 11:49, Alexander Kurtz wrote:
> As you can see, chroot will run the applet rather than the binary
> unless the full path is given. While this *may* be useful in some
> situations, it can also lead to *really* subtle failures in others, so
> I'm raising the severity of this bug. Please consider applying
> Bernhard's patch if possible or at least documenting this behaviour!
Hi Alexander, Bernhard,
The busybox package in Debian is under new management, and we (the new
maintainers) are going over the existing bugs to see what can and should
be fixed or closed.
I've spent some time investigating this bug, and it seems that this
behaviour is actually deliberate in busybox-static. The -static package
is specifically configured in this way:
> bootc@muta busybox $ grep -r CONFIG_FEATURE_PREFER_APPLETS debian/config/pkg/
> debian/config/pkg/udeb:# CONFIG_FEATURE_PREFER_APPLETS is not set
> debian/config/pkg/static:CONFIG_FEATURE_PREFER_APPLETS=y
> debian/config/pkg/deb:# CONFIG_FEATURE_PREFER_APPLETS is not set
I can understand that you might find this behaviour surprising, but I
believe that the static package would have been configured this way due
to its intended usage: as a useful tool to rescue a broken system. If
your system is broken in such a way that you require a statically-linked
version of busybox to dig yourself out of that hole, then it's likely
that your system-provided tools will not function. There is no point
trying to run them, so the built-in applets should be preferred.
The standard flavour of busybox in Debian is the dynamically linked
'busybox' package, which doesn't have this behaviour because it's mainly
intended for use in the initramfs.
I guess this means that the behaviour is probably correct and simply
needs documenting. Aside from the package description, is there
somewhere in particular that makes sense to you (for example, where you
would have looked) to find such a difference documented?
Because I suspect this is a documentation bug rather than actual broken
behaviour, I am reducing the severity of this bug.
Cheers,
Chris
--
Chris Boot
bootc@debian.org
Reply to: