Re: [kernel-hardening] Re: [PATCH v5 next 5/5] net: modules: use request_module_cap() to load 'netdev-%s' modules

[Theodore Ts'o]

On Tue, Nov 28, 2017 at 04:18:59PM -0800, Kees Cook wrote:
> There's also a difference between immutable CONFIG options that cannot
> be disabled at runtime, those that can, global sysctls, per-namespace
> controls, etc etc. The kernel is all about providing admins with knobs
> to tweak their performance and security. Suddenly being told that we
> can't create optional improvements is very odd.

I just think that tweakable knobs are mostly pointless.  From my
experience the number of sysadmins that adjust knobs is ***tiny***[1].
Put another way, the effort to determine whether tweaking a knob will
result in breakages or will be safe is as much work as creating a
white list of modules that are allowed to be loaded.

[1] And I say that having providing a lot of knobs for ext4.  :-)

This is why some on the kernel-hardening list have argued for making
the default to be opt-out, which means some users will be breaken (and
their answer to that seems to be, "oh well --- gotta break some eggs
to make an omlette".  Sucks if you're one of the eggs, though.)

And I don't see how systemd magically means no one will be broken.  If
you have a non-root process trying to invoke a line discpline which
has to be loaded as a module, if you flip the switch, that process
will be broken.  How does using systemd make the problem go away?

					- Ted