Re: [PATCH v2] mm/zsmalloc.c: close race window between zs_pool_dec_isolated() and zs_unregister_migration()

[Miaohe Lin]

Sorry for the disturbs! Please ignore this duplicated one...

On 2021/7/8 19:51, Miaohe Lin wrote:
> There has one possible race window between zs_pool_dec_isolated() and
> zs_unregister_migration() because wait_for_isolated_drain() checks the
> isolated count without holding class->lock and there is no order inside
> zs_pool_dec_isolated(). Thus the below race window could be possible:
> 
> zs_pool_dec_isolated		zs_unregister_migration
>   check pool->destroying != 0
> 				  pool->destroying = true;
> 				  smp_mb();
> 				  wait_for_isolated_drain()
> 				    wait for pool->isolated_pages == 0
>   atomic_long_dec(&pool->isolated_pages);
>   atomic_long_read(&pool->isolated_pages) == 0
> 
> Since we observe the pool->destroying (false) before atomic_long_dec()
> for pool->isolated_pages, waking pool->migration_wait up is missed.
> 
> Fix this by ensure checking pool->destroying is happened after the
> atomic_long_dec(&pool->isolated_pages).
> 
> Fixes: 701d678599d0 ("mm/zsmalloc.c: fix race condition in zs_destroy_pool")
> Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> ---
> v1->v2:
>   Fix potential race window rather than simply combine atomic_long_dec
> and atomic_long_read.
> 
> Hi Andrew,
>   This patch is the version 2 of
> mm-zsmallocc-combine-two-atomic-ops-in-zs_pool_dec_isolated.patch.
> Many thanks.
> ---
>  mm/zsmalloc.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
> index 5f3df680f0a2..0fc388a0202d 100644
> --- a/mm/zsmalloc.c
> +++ b/mm/zsmalloc.c
> @@ -1830,10 +1830,11 @@ static inline void zs_pool_dec_isolated(struct zs_pool *pool)
>  	VM_BUG_ON(atomic_long_read(&pool->isolated_pages) <= 0);
>  	atomic_long_dec(&pool->isolated_pages);
>  	/*
> -	 * There's no possibility of racing, since wait_for_isolated_drain()
> -	 * checks the isolated count under &class->lock after enqueuing
> -	 * on migration_wait.
> +	 * Checking pool->destroying must happen after atomic_long_dec()
> +	 * for pool->isolated_pages above. Paired with the smp_mb() in
> +	 * zs_unregister_migration().
>  	 */
> +	smp_mb__after_atomic();
>  	if (atomic_long_read(&pool->isolated_pages) == 0 && pool->destroying)
>  		wake_up_all(&pool->migration_wait);
>  }
>