Re: [PATCH 1/4] exec: Change uselib(2) IS_SREG() failure to EACCES

On Mon, May 18, 2020 at 06:57:15PM -0500, Eric W. Biederman wrote:

Christian Brauner [off-list ref] writes:

quoted

On Mon, May 18, 2020 at 04:43:20PM +0200, Jann Horn wrote:

quoted

On Mon, May 18, 2020 at 3:03 PM Christian Brauner
[off-list ref] wrote:

quoted

Also - gulp (puts on flame proof suit) - may I suggest we check if there
are any distros out there that still set CONFIG_USELIB=y

Debian seems to have it enabled on x86...

https://salsa.debian.org/kernel-team/linux/-/blob/master/debian/config/kernelarch-x86/config#L1896

A random Ubuntu 19.10 VM I have here has it enabled, too.

I wonder if there's any program - apart from _ancient_ glibc out there
that actually use it...
I looked at uselib in codsearch but the results were quite unspecific
but I didn't look too close.

So the thing to do is to have a polite word with people who build Ubuntu
and Debian kernels and get them to disable the kernel .config.

Yeah, I think that's a sane thing to do.
I filed a bug for Ubuntu to start a discussion. I can't see an obvious
reason why not.

A quick look suggets it is already disabled in RHEL8.  It cannot be
disabled in RHEL7.

Then in a few years we can come back and discuss removing the uselib
system call, base on no distributions having it enabled.

If it was only libc4 and libc5 that used the uselib system call then it
can probably be removed after enough time.

We can probably reorganize the code before the point it is clearly safe
to drop support for USELIB to keep it off to the side so USELIB does not
have any ongoing mainteance costs.

For this patchset I think we need to assume uselib will need to be
maintained for a bit longer.

Yeah, agreed. It doesn't matter as long as we have a plan for the future
to remove it. I don't think keeping this cruft around forever should be
the only outlook.

Christian