Re: [PATCH 1/2] kobject: don't delay to cleanup module kobject

[Ming Lei]

On Fri, Nov 26, 2021 at 05:33:45PM +0100, Greg Kroah-Hartman wrote:
> On Sat, Nov 27, 2021 at 12:28:48AM +0800, Ming Lei wrote:
> > On Fri, Nov 26, 2021 at 05:08:16PM +0100, Greg Kroah-Hartman wrote:
> > > On Fri, Nov 05, 2021 at 02:37:09PM +0800, Ming Lei wrote:
> > > > CONFIG_DEBUG_KOBJECT_RELEASE is used for debugging kobject release/cleanup
> > > > issue. The module kobject is released after module_exit() returns. If
> > > > this kobject is delayed too much, and may cause other kobject's
> > > > cleaned up a bit earlier before freeing module, then real issue is
> > > > hidden.
> > > > 
> > > > So don't delay module kobject's cleanup, meantime module kobject is
> > > > always cleaned up synchronously, and we needn't module kobject's
> > > > cleanup.
> > > > 
> > > > Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx>
> > > > ---
> > > >  lib/kobject.c | 5 +++++
> > > >  1 file changed, 5 insertions(+)
> > > > 
> > > > diff --git a/lib/kobject.c b/lib/kobject.c
> > > > index ea53b30cf483..4c0dbe11be3d 100644
> > > > --- a/lib/kobject.c
> > > > +++ b/lib/kobject.c
> > > > @@ -16,6 +16,7 @@
> > > >  #include <linux/stat.h>
> > > >  #include <linux/slab.h>
> > > >  #include <linux/random.h>
> > > > +#include <linux/module.h>
> > > >  
> > > >  /**
> > > >   * kobject_namespace() - Return @kobj's namespace tag.
> > > > @@ -727,6 +728,10 @@ static void kobject_release(struct kref *kref)
> > > >  	struct kobject *kobj = container_of(kref, struct kobject, kref);
> > > >  #ifdef CONFIG_DEBUG_KOBJECT_RELEASE
> > > >  	unsigned long delay = HZ + HZ * (get_random_int() & 0x3);
> > > > +
> > > > +	if (kobj->ktype == &module_ktype)
> > > > +		delay = 0;
> > > 
> > > No, there should not be anything "special" about module kobjects to get
> > > this kind of treatment.  They should work like any other kobject and
> > > clean up properly when needed.
> > 
> > Here setting 0 delay for module kobject is just for making DEBUG_KOBJECT_RELEASE
> > reliable to detect/report issues. Otherwise if the random delay for module
> > kobject is bigger than other kobjects, potential use-after-after won't
> > be exposed.
> 
> So you now can not debug the module kobject code?

So far, module kobject code is always released in sync way, see
mod_kobject_put(), and CONFIG_DEBUG_KOBJECT_RELEASE is basically useless
for module kobject.

> 
> This needs to be documented really really really well why this kobject
> type is somehow "special" in the code.  We should not special-case these
> things unless you have a great reason, and I am not yet convinced.

OK, I will add comment on this special usage in V2 so that you can
review it further.


Thanks,
Ming