2019-03-11 06:34:38

by Kangjie Lu

[permalink] [raw]
Subject: [PATCH] net: mlx5: fix a potential NULL pointer dereference

In case skb_header_pointer fails, the fix issues a warning.

A better fix requires modifying the signature of mlx5e_get_fcs to
pass an error code upstream.

Signed-off-by: Kangjie Lu <[email protected]>
---
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
index be396e5e4e39..6ec1c110e4be 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
@@ -719,6 +719,8 @@ static u32 mlx5e_get_fcs(const struct sk_buff *skb)

fcs_bytes = skb_header_pointer(skb, skb->len - ETH_FCS_LEN,
ETH_FCS_LEN, &_fcs_bytes);
+ if (unlikely(!fcs_bytes))
+ pr_warn_once("skb_header_pointer returns NULL\n");

return __get_unaligned_cpu32(fcs_bytes);
}
--
2.17.1



2019-03-11 14:07:26

by Jason Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH] net: mlx5: fix a potential NULL pointer dereference

On Mon, Mar 11, 2019 at 01:33:43AM -0500, Kangjie Lu wrote:
> In case skb_header_pointer fails, the fix issues a warning.
>
> A better fix requires modifying the signature of mlx5e_get_fcs to
> pass an error code upstream.
>
> Signed-off-by: Kangjie Lu <[email protected]>
> drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> index be396e5e4e39..6ec1c110e4be 100644
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> @@ -719,6 +719,8 @@ static u32 mlx5e_get_fcs(const struct sk_buff *skb)
>
> fcs_bytes = skb_header_pointer(skb, skb->len - ETH_FCS_LEN,
> ETH_FCS_LEN, &_fcs_bytes);
> + if (unlikely(!fcs_bytes))
> + pr_warn_once("skb_header_pointer returns NULL\n");

This is really pointless. Fix stuff like this so it doesn't oops or
don't bother.

Jason

> return __get_unaligned_cpu32(fcs_bytes);
> }

2019-03-11 18:47:33

by Saeed Mahameed

[permalink] [raw]
Subject: Re: [PATCH] net: mlx5: fix a potential NULL pointer dereference

On Mon, 2019-03-11 at 01:33 -0500, Kangjie Lu wrote:
> In case skb_header_pointer fails, the fix issues a warning.
>

This case is impossible: mlx5 driver builds this skb itself so we are
sure skb_header_pointer never fails.

What is the motivation behind this fix? static checker or actual issue
?

> A better fix requires modifying the signature of mlx5e_get_fcs to
> pass an error code upstream.
>
> Signed-off-by: Kangjie Lu <[email protected]>
> ---
> drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> index be396e5e4e39..6ec1c110e4be 100644
> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c
> @@ -719,6 +719,8 @@ static u32 mlx5e_get_fcs(const struct sk_buff
> *skb)
>
> fcs_bytes = skb_header_pointer(skb, skb->len - ETH_FCS_LEN,
> ETH_FCS_LEN, &_fcs_bytes);
> + if (unlikely(!fcs_bytes))
> + pr_warn_once("skb_header_pointer returns NULL\n");
>
> return __get_unaligned_cpu32(fcs_bytes);
> }

2019-03-11 20:39:55

by David Miller

[permalink] [raw]
Subject: Re: [PATCH] net: mlx5: fix a potential NULL pointer dereference

From: Kangjie Lu <[email protected]>
Date: Mon, 11 Mar 2019 01:33:43 -0500

> In case skb_header_pointer fails, the fix issues a warning.
>
> A better fix requires modifying the signature of mlx5e_get_fcs to
> pass an error code upstream.
>
> Signed-off-by: Kangjie Lu <[email protected]>

Saeed, I'll let you take this.

Thanks.