Re: [PATCH] livepatch: Enhance livepatch to support remove patch module dynamically

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Wed, Apr 01, 2015 at 10:40:19PM +0800, Minfei Huang wrote:
> 2015-04-01 22:13 GMT+08:00 Jiri Kosina <jkosina@suse.cz>:
> > On Wed, 1 Apr 2015, Minfei Huang wrote:
> >
> >> > > diff --git a/kernel/livepatch/core.c b/kernel/livepatch/core.c
> >> > > index 3f9f1d6..0266950 100644
> >> > > --- a/kernel/livepatch/core.c
> >> > > +++ b/kernel/livepatch/core.c
> >> > > @@ -502,6 +502,17 @@ static int __klp_disable_patch(struct klp_patch *patch)
> >> > >       return 0;
> >> > >  }
> >> > >
> >> > > +static int __klp_disable_patch_nolock(struct klp_patch *patch)
> >> > > +{
> >> > > +     int ret = 0;
> >> > > +
> >> > > +     ret = __klp_disable_patch(patch);
> >> > > +     if (ret)
> >> > > +             return ret;
> >> > > +     module_put(patch->mod);
> >> > > +     return ret;
> >> > > +}
> >> > > +
> >> >
> >> > Your patch doesn't solve the problem at all.
> >> >
> >> > There is no guarantee that once __klp_disable_patch() returns noone is
> >> > using the old code any more.
> >> >
> >>
> >> Yes, thanks.
> >>
> >> The __klp_disable_patch only guarantees that we will never call the
> >> function in patch module. For now, patch module can never be removed
> >> from the kernel once it was loaded. It may be inconvenience if we want
> >> re-load a new patch module replace the old patch module.
> >
> > Patch stacking is possible though.
> >
> 
> How about check the stack when start to remove the patch module? We
> can use the interface stop_machine to hang other CPU, then check and
> remove if no function in patch module is used.

Yeah, we'll eventually have something like that as part of the
consistency model.  The exact implementation is still undecided.  Until
then I think we want to disallow removing the patch module.

-- 
Josh