Re: [OE-core] [PATCH] busybox: drop net-tools from defconfig

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Sun, 2025-01-26 at 21:30 +0100, Enrico Jörns wrote:
> Am Sonntag, dem 26.01.2025 um 12:58 +0000 schrieb Richard Purdie:
> > On Sun, 2025-01-26 at 12:51 +0100, Enrico Jörns via
> > lists.openembedded.org wrote:
> > > The 'net-tools' have been deprecated 15 years ago! [1]
> > > Let's remove their busybox pendants from the defconfig to prevent
> > > people
> > > from accidentally starting projects with ancient technology.
> > > 
> > > [1] https://lists.debian.org/debian-devel/2009/03/msg00780.html
> > > 
> > > Signed-off-by: Enrico Jörns <ejo@pengutronix.de>
> > > ---
> > >  meta/recipes-core/busybox/busybox/defconfig | 10 +++++-----
> > >  1 file changed, 5 insertions(+), 5 deletions(-)
> > 
> > I did a quick grep of OE-Core and we still have a handful of sites
> > using this on target, particularly but not limited to our QA, the ones
> > that caught my eye:
> 
> thank you for looking into this!
> 
> Well, not that I hadn't done a grep before. 😉
> I looked through the occurrences of 'ifconfig' but concluded that starting with disabling it for
> busybox could be an initial step that should not break most setups. But not sure if I was too
> optimistic..

The udev/initramfs and some of the QA references do worry me a bit.

> I started working on a few patches to address occurrences of ifconfig, but I ran into challenges
> with testing my changes. I decided to send this patch on its own, hoping it doesn’t break the
> autobuilder, and also to check if there are any objections to removing net-tools in general.

I don't see a problem with switching the references over and
personally, I think we'll need to remove net-tools but one step at a
time I guess.

I think what I'd suggest is sending a series of what you have removing
the references you can. We can then test that on the autobuilder and
see what it shows. Please just make it clear you're asking for help
with testing. Copying Mathieu as he may be able to help.

> I've now had a second look to see what could be affected by the busybox config change.
> 
> > meta/recipes-core/udev/udev-extraconf/network.sh:       ifconfig | grep -q "^$INTERFACE" || ifup $INTERFACE
> 
> Wasn't sure about this one but it seems to be used by the initramfs framework and thus probably with
> busybox.

Right, that was my worry.

> -> Probably affected. Might make sense to replace it with ip anyway.
> 
> > meta/recipes-core/images/build-appliance-image/README_VirtualBox_Toaster.txt:    $ ifconfig
> > meta/recipes-core/images/build-appliance-image/README_VirtualBox_Toaster.txt:    $ ifconfig
> 
> Documentation that hasn't been touched since 2016.
> Also, the commands mentioned there are for the host machine commands.

Agreed, those ones aren't a blocker.

> 
> -> Unrelated. ifconfig could be easily removed here, though.
> 
> > meta/recipes-core/busybox/files/simple.script:                                /SBIN_DIR/ifconfig $interface 0.0.0.0
> > meta/recipes-core/busybox/files/simple.script:                        /SBIN_DIR/ifconfig $interface $ip $BROADCAST $NETMASK
> 
> Both are guarded by 'if [ $have_bin_ip -eq 1 ]; then' and should also work with iproute2 equivalents
> though.

Good, I didn't check that, I was just trying to see what references we
had left.

> -> Related but no need to fix. Fallback could be removed maybe.
> 
> > meta/lib/oeqa/selftest/cases/devtool.py:            result = runCmd('PATH="$PATH:/sbin:/usr/sbin" ifconfig -a', ignore_status=True)
> > meta/lib/oeqa/selftest/cases/devtool.py:                self.skipTest('Failed to determine if tap devices exist with ifconfig or ip: %s' % result.output)
> 
> This is fallback handling for missing 'ip' on the host.

Ok.

> -> Unrelated. Could be removed anyway maybe (already had a patch for this).
> 
> > meta/lib/oeqa/utils/qemurunner.py:                    cmd = "ifconfig eth0 %s netmask %s up\n" % (self.ip, self.netmask)
> 
> This I had overseen and it actually seems to be called on the target.

Right, that one does seem potentially problematic. I didn't look into
which set of runner options trigger it though.

> 
> -> Probably affected and should be changed.
> 
> > meta/lib/oeqa/runtime/cases/ethernet_ip_connman.py:        (status, output) = self.target.run("ifconfig eth0 | grep 'inet ' | awk '{print $2}'")
> > meta/lib/oeqa/runtime/cases/ethernet_ip_connman.py:        (status, output) = self.target.run("ifconfig eth0:1 %s netmask 255.255.255.0 && sleep 2 && ping -c 5 %s && ifconfig eth0:1 down" % (virtual_ip,virtual_ip))
> 
> I had made a patch for this but did not figure out how to run this test or if it is actually used by
> CI.

I'd guess a "bitbake XXX-image-XXX -c testimage" where the image
contains connman?

> -> Probably affected and needs more research.
> 
> > but there are others too. We probably need to finish resolving these to
> > other commands before we can turn this off.
> > 
> > If you aren't able to help with that, we should at least have an open
> > bug to sort it out.
> 
> I'd try to address the potential issues first.
> 
> Another challenge would be to remove the actual 'net-tools' package because ltp 
> still seems to rely on it (with explicit RDEPENDS).

I'd guess there are tests in ltp which call ifconfig and friends?

> But we could start with removing it from packagegroup-core-{base-utils,full-cmdline}.bb IMHO.

Yes, definitely agreed.

> Do you have an opinion about the host tooling? Could we drop net-tools here, too?

If we don't need it anywhere now, I'd be fine with that.

Cheers,

Richard