Re: [PATCH -repost 05/21] kgr: update Kconfig documentation

[Jiri Kosina <jkosina@suse.cz>]

On Thu, 26 Jun 2014, Jiri Slaby wrote:

> >> ---
> >>  kernel/Kconfig.kgraft | 3 +++
> >>  samples/Kconfig       | 4 ++++
> >>  2 files changed, 7 insertions(+)
> >>
> >> diff --git a/kernel/Kconfig.kgraft b/kernel/Kconfig.kgraft
> >> index f38d82c06580..bead93646071 100644
> >> --- a/kernel/Kconfig.kgraft
> >> +++ b/kernel/Kconfig.kgraft
> >> @@ -5,3 +5,6 @@ config KGRAFT
> >>  	bool "kGraft infrastructure"
> >>  	depends on DYNAMIC_FTRACE_WITH_REGS
> >>  	depends on HAVE_KGRAFT
> >> +	help
> >> +	  Select this to enable kGraft online kernel patching. The
> >> +	  runtime price is zero, so it is safe to say Y here.
> >> diff --git a/samples/Kconfig b/samples/Kconfi
> > 
> > The runtime impact is that you've just introduced a virus and trojan
> > writers delight into your kernel. There's a balance between convenience
> > and security but given most users will never use kgraft this advice seems
> > incorrect.
> 
> This now writes:
> +       help
> +         Select this to enable kGraft online kernel patching. The
> +         runtime price is nearly zero, so it is safe to say Y here
> +         provided you are aware of all the consequences (e.g. in
> +         security).
> 
> Is it OK with you?

This might cause a false impression that we are actually opening a 
security hole into a system, which is not true at all.

Yes, backdoor writeres might (or might not) make use of kGraft API, but 
they have gazillion of other comparable options (*probes, ftrace, 
text_poke(), ...).

I'd perhaps propose something like

"Select this to enable kGraft live kernel patching. The runtime penalty is 
nearly zero, so it is safe to say Y here if you want the kernel to expose 
API for live patching to modules".

-- 
Jiri Kosina
SUSE Labs