Re: [PATCH v2] mm: thp: Fix refcount leak in thpsize_create() error path

["David Hildenbrand (Arm)" <david@kernel.org>]

On 4/12/26 19:54, Guangshuo Li wrote:
> After kobject_init_and_add(), the lifetime of the embedded struct
> kobject is expected to be managed through the kobject core reference
> counting.
> 
> In thpsize_create(), if kobject_init_and_add() fails, thpsize is freed
> directly with kfree() rather than releasing the kobject reference with
> kobject_put(). This may leave the reference count of the embedded struct
> kobject unbalanced, resulting in a refcount leak.
> 
> Fix this by using kobject_put(&thpsize->kobj) in the failure path and
> letting thpsize_release() handle the final cleanup.

Fortunately, that's only called during init, and the stars would have to
align for this to ever trigger. I don't think this would ever trigger on
any reasonable system, really.

And if it triggers, other stuff would already be severely messed up.

> 
> Fixes: 3485b88390b0 ("mm: thp: introduce multi-size THP sysfs interface")
> Cc: stable@vger.kernel.org

Which also makes me wonder whether we even care about stable here.

> Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
> ---
> v2:
>   - Drop the incorrect UAF mention from the commit message
>   - Clarify that the bug is an unbalanced kobject reference in the
>   - kobject_init_and_add() failure path
> 
>  mm/huge_memory.c | 4 +---
>  1 file changed, 1 insertion(+), 3 deletions(-)
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 40cf59301c21..c8ffa188a198 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -726,10 +726,8 @@ static struct thpsize *thpsize_create(int order, struct kobject *parent)
>  
>  	ret = kobject_init_and_add(&thpsize->kobj, &thpsize_ktype, parent,
>  				   "hugepages-%lukB", size);
> -	if (ret) {
> -		kfree(thpsize);
> +	if (ret)
>  		goto err;
> -	}

kobject_init_and_add() indeed documents "If this function returns an
error, kobject_put() must be called".

As Andrew says, that's not what the "goto err" does.

-- 
Cheers,

David