Re: [PATCH v2 2/2] livepatch: Delete the associated module of disabled livepatch

[Petr Mladek <pmladek@suse.com>]

On Thu 2024-05-09 13:53:17, Yafang Shao wrote:
> On Thu, May 9, 2024 at 1:20 PM Josh Poimboeuf <jpoimboe@kernel.org> wrote:
> >
> > On Thu, May 09, 2024 at 10:17:43AM +0800, Yafang Shao wrote:
> > > On Wed, May 8, 2024 at 3:03 PM Josh Poimboeuf <jpoimboe@kernel.org> wrote:
> > > >
> > > > On Wed, May 08, 2024 at 02:01:29PM +0800, Yafang Shao wrote:
> > > > > On Wed, May 8, 2024 at 1:16 PM Josh Poimboeuf <jpoimboe@kernel.org> wrote:
> > > > > > If klp_patch.replace is set on the new patch then it will replace all
> > > > > > previous patches.
> > > > >
> > > > > A scenario exists wherein a user could simultaneously disable a loaded
> > > > > livepatch, potentially resulting in it not being replaced by the new
> > > > > patch. While theoretical, this possibility is not entirely
> > > > > implausible.
> > > >
> > > > Why does it matter whether it was replaced, or was disabled beforehand?
> > > > Either way the end result is the same.
> > >
> > > When users disable the livepatch, the corresponding kernel module may
> > > sometimes be removed, while other times it remains intact. This
> > > inconsistency has the potential to confuse users.
> >
> > I'm afraid I don't understand.  Can you give an example scenario?
> >
> 
> As previously mentioned, this scenario may occur if user-space tools
> remove all pertinent kernel modules from /sys/livepatch/* while a user
> attempts to load a new atomic-replace livepatch.
> 
> For instance:
> 
> User-A                                                       User-B
> 
> echo 0 > /sys/livepatch/A/enable              insmod atomic-replace-patch.ko
> 
> >From User-A's viewpoint, the A.ko module might sometimes be removed,
> while at other times it remains intact. The reason is that User-B
> removed a module that he shouldn't remove.

Why would User-A want to keep the module, please? The livepatches
could not longer be re-enabled since the commit 958ef1e39d24d6
("livepatch: Simplify API by removing registration step") which
was added in v5.1-rc1.

The only problem might be that User-A can't remove the module
because it has already been removed by User-B or vice versa.
Is this really a problem?

Have you seen the problem in practice or is it just theoretical?

Is anyone really combining livepatches with and without atomic
replace?

Best Regards,
Petr