Re: [PATCH 1/1] Address an issue with hardened user copy

[Leon Romanovsky <leon@kernel.org>]

On Tue, Aug 11, 2020 at 02:14:57PM -0500, Bob Pearson wrote:
> by copying to user space from the stack instead of slab cache.
> This affects the rdma_rxe driver causing a warning once per boot.
> The alternative is to ifigure out how to whitelist the xxx_qp struct

ifigure -> figure

> but this seems simple and clean.


We have multiple cases like this in the code, what is the error exactly?
And what is "hardened user copy"?

>
> ---

Signed-off-by is missing.

>  drivers/infiniband/core/uverbs_std_types_qp.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/infiniband/core/uverbs_std_types_qp.c b/drivers/infiniband/core/uverbs_std_types_qp.c
> index 3bf8dcdfe7eb..2f8b14003b95 100644
> --- a/drivers/infiniband/core/uverbs_std_types_qp.c
> +++ b/drivers/infiniband/core/uverbs_std_types_qp.c
> @@ -98,6 +98,7 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
>  	struct ib_device *device;
>  	u64 user_handle;
>  	int ret;
> +	int qp_num;
>
>  	ret = uverbs_copy_from_or_zero(&cap, attrs,
>  			       UVERBS_ATTR_CREATE_QP_CAP);
> @@ -293,9 +294,10 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
>  	if (ret)
>  		return ret;
>
> +	/* copy from stack to avoid whitelisting issues */
> +	qp_num = qp->qp_num;
>  	ret = uverbs_copy_to(attrs, UVERBS_ATTR_CREATE_QP_RESP_QP_NUM,
> -			     &qp->qp_num,
> -			     sizeof(qp->qp_num));
> +			     &qp_num, sizeof(qp_num));
>
>  	return ret;
>  err_put:
> --
> 2.25.1
>