Re: [PATCH 4.19] IB/umem: fix reference count leak in ib_umem_odp_get()

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Jul 14, 2020 at 10:57:48AM +0000, Yang Yingliang wrote:
> Add missing mmput() on error path to avoid ref-count leak.
> 
> This problem has already been resolved in mainline by
> f27a0d50a4bc ("RDMA/umem: Use umem->owning_mm inside ODP").
> 
> Fixes: 79bb5b7ee177 ("RDMA/umem: Fix missing mmap_sem in get umem ODP call")
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
> ---
>  drivers/infiniband/core/umem_odp.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c
> index eeafdc0beec7..08ef654ea9b8 100644
> --- a/drivers/infiniband/core/umem_odp.c
> +++ b/drivers/infiniband/core/umem_odp.c
> @@ -347,7 +347,8 @@ int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem,
>  		vma = find_vma(mm, ib_umem_start(umem));
>  		if (!vma || !is_vm_hugetlb_page(vma)) {
>  			up_read(&mm->mmap_sem);
> -			return -EINVAL;
> +			ret_val = -EINVAL;
> +			goto out_mm;
>  		}
>  		h = hstate_vma(vma);
>  		umem->page_shift = huge_page_shift(h);

This patch does look correct, please address Greg's remarks.

Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>

Thanks,
Jason