Re: [PATCH v3] mm/slab_common: Deleting kobject in kmem_cache_destroy() without holding slab_mutex/cpu_hotplug_lock

[Waiman Long <longman@xxxxxxxxxx>]

On 8/13/22 11:00, Hyeonggon Yoo wrote:
> On Fri, Aug 12, 2022 at 02:30:33PM -0400, Waiman Long wrote:
> > A circular locking problem is reported by lockdep due to the following
> > circular locking dependency.
> >
> >    +--> cpu_hotplug_lock --> slab_mutex --> kn->active --+
> >    |                                                     |
> >    +-----------------------------------------------------+
> >
> > The forward cpu_hotplug_lock ==> slab_mutex ==> kn->active dependency
> > happens in
> >
> >    kmem_cache_destroy():	cpus_read_lock(); mutex_lock(&slab_mutex);
> >    ==> sysfs_slab_unlink()
> >        ==> kobject_del()
> >            ==> kernfs_remove()
> > 	      ==> __kernfs_remove()
> > 	          ==> kernfs_drain(): rwsem_acquire(&kn->dep_map, ...);
> Maybe you mean this?
>
>          /* but everyone should wait for draining */
>          wait_event(root->deactivate_waitq,
>                     atomic_read(&kn->active) == KN_DEACTIVATED_BIAS);
This is part of the kernfs_drain() operation. However, I am focusing on 
the behavior of the pseudo lock that causes the lockdep splat in the 
first place. That is why I am showcasing the rwsem_acquire() call here.


> > The backward kn->active ==> cpu_hotplug_lock dependency happens in
> >
> >    kernfs_fop_write_iter(): kernfs_get_active();
> >    ==> slab_attr_store()
> >        ==> cpu_partial_store()
> >            ==> flush_all(): cpus_read_lock()
> >
> > One way to break this circular locking chain is to avoid holding
> > cpu_hotplug_lock and slab_mutex while deleting the kobject in
> > sysfs_slab_unlink() which should be equivalent to doing a write_lock
> > and write_unlock pair of the kn->active virtual lock.
> >
> > Since the kobject structures are not protected by slab_mutex or the
> > cpu_hotplug_lock, we can certainly release those locks before doing
> > the delete operation.
> >
> > Move sysfs_slab_unlink() and sysfs_slab_release() to the newly
> > created kmem_cache_release() and call it outside the slab_mutex &
> > cpu_hotplug_lock critical sections. There will be a slight delay
> > in the deletion of sysfs files if kmem_cache_release() is called
> > indirectly from a work function.
> >
> > Signed-off-by: Waiman Long <longman@xxxxxxxxxx>
> > ---
> >
> >   [v3] Move sysfs_slab_unlink() out to kmem_cache_release() and move
> >        schedule_work() back right after list_add_tail().
> >
> >   mm/slab_common.c | 45 +++++++++++++++++++++++++++++----------------
> >   1 file changed, 29 insertions(+), 16 deletions(-)
> >
> > diff --git a/mm/slab_common.c b/mm/slab_common.c
> > index 17996649cfe3..07b948288f84 100644
> > --- a/mm/slab_common.c
> > +++ b/mm/slab_common.c
> > @@ -392,6 +392,28 @@ kmem_cache_create(const char *name, unsigned int size, unsigned int align,
> >   }
> >   EXPORT_SYMBOL(kmem_cache_create);
> >   
> > +#ifdef SLAB_SUPPORTS_SYSFS
> > +/*
> > + * For a given kmem_cache, kmem_cache_destroy() should only be called
> > + * once or there will be a use-after-free problem. The actual deletion
> > + * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
> > + * protection. So they are now done without holding those locks.
> > + *
> > + * Note that there will be a slight delay in the deletion of sysfs files
> > + * if kmem_cache_release() is called indrectly from a work function.
> > + */
> > +static void kmem_cache_release(struct kmem_cache *s)
> > +{
> > +	sysfs_slab_unlink(s);
> > +	sysfs_slab_release(s);
> > +}
> > +#else
> > +static void kmem_cache_release(struct kmem_cache *s)
> > +{
> > +	slab_kmem_cache_release(s);
> > +}
> > +#endif
> > +
> >   static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
> >   {
> >   	LIST_HEAD(to_destroy);
> > @@ -418,11 +440,7 @@ static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
> >   	list_for_each_entry_safe(s, s2, &to_destroy, list) {
> >   		debugfs_slab_release(s);
> >   		kfence_shutdown_cache(s);
> > -#ifdef SLAB_SUPPORTS_SYSFS
> > -		sysfs_slab_release(s);
> > -#else
> > -		slab_kmem_cache_release(s);
> > -#endif
> > +		kmem_cache_release(s);
> >   	}
> >   }
> >   
> > @@ -437,20 +455,11 @@ static int shutdown_cache(struct kmem_cache *s)
> >   	list_del(&s->list);
> >   
> >   	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
> > -#ifdef SLAB_SUPPORTS_SYSFS
> > -		sysfs_slab_unlink(s);
> > -#endif
> >   		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
> >   		schedule_work(&slab_caches_to_rcu_destroy_work);
> >   	} else {
> >   		kfence_shutdown_cache(s);
> >   		debugfs_slab_release(s);
> > -#ifdef SLAB_SUPPORTS_SYSFS
> > -		sysfs_slab_unlink(s);
> > -		sysfs_slab_release(s);
> > -#else
> > -		slab_kmem_cache_release(s);
> > -#endif
> >   	}
> >   
> >   	return 0;
> > @@ -465,14 +474,16 @@ void slab_kmem_cache_release(struct kmem_cache *s)
> >   
> >   void kmem_cache_destroy(struct kmem_cache *s)
> >   {
> > +	int refcnt;
> > +
> >   	if (unlikely(!s) || !kasan_check_byte(s))
> >   		return;
> >   
> >   	cpus_read_lock();
> >   	mutex_lock(&slab_mutex);
> >   
> > -	s->refcount--;
> > -	if (s->refcount)
> > +	refcnt = --s->refcount;
> > +	if (refcnt)
> >   		goto out_unlock;
> >   
> >   	WARN(shutdown_cache(s),
> > @@ -481,6 +492,8 @@ void kmem_cache_destroy(struct kmem_cache *s)
> >   out_unlock:
> >   	mutex_unlock(&slab_mutex);
> >   	cpus_read_unlock();
> > +	if (!refcnt && !(s->flags & SLAB_TYPESAFE_BY_RCU))
> > +		kmem_cache_release(s);
> >   }
> >   EXPORT_SYMBOL(kmem_cache_destroy);
> >   
> > --
> > 2.31.1
> little bit complicated but looks good to me.
> Thank you for fixing this.
>
> Reviewed-by: Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx>
Thanks for the review.

Cheers,
Longman