Re: [PATCH] RDMA/mlx4: Fix potential memory leak

[Leon Romanovsky]

On Thu, Nov 04, 2021 at 10:14:12AM +0000, Jiasheng Jiang wrote:
> In the error path, the dev->dev isn't released.
> Therefore, it might be better to fix it to avoid
> potential memory leak.
> 
> Fixes: 9376932 ("IB/mlx4_ib: Add support for user MR re-registration")
> Signed-off-by: Jiasheng Jiang <jiasheng@xxxxxxxxxxx>
> ---
>  drivers/infiniband/hw/mlx4/mr.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

I don't understand about which release you are talking, but this patch
can't be right. You are supposed to call to "release_mpt_entry" only in
the error flows which after successful mlx4_mr_hw_get_mpt().

Thanks

> 
> diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
> index 50becc0..d8ae92e 100644
> --- a/drivers/infiniband/hw/mlx4/mr.c
> +++ b/drivers/infiniband/hw/mlx4/mr.c
> @@ -473,7 +473,7 @@ struct ib_mr *mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags, u64 start,
>  	 */
>  	err =  mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry);
>  	if (err)
> -		return ERR_PTR(err);
> +		goto release_mpt_entry;
>  
>  	if (flags & IB_MR_REREG_PD) {
>  		err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry,
> -- 
> 2.7.4
>