2023-05-22 20:10:09

by Thomas Bogendoerfer

Subject: [PATCH net] net: mellanox: mlxbf_gige: Fix skb_panic splat under memory pressure

Do skp_put() after a new skb has been successfully allocated otherwise
the reused skb leads to skp_panics or incorrect packet sizes.

Signed-off-by: Thomas Bogendoerfer <[email protected]>
---
.../ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
index afa3b92a6905..2c132849a76d 100644
--- a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
+++ b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
@@ -245,18 +245,19 @@ static bool mlxbf_gige_rx_packet(struct mlxbf_gige *priv, int *rx_pkts)

skb = priv->rx_skb[rx_pi_rem];

- skb_put(skb, datalen);
-
- skb->ip_summed = CHECKSUM_NONE; /* device did not checksum packet */
-
- skb->protocol = eth_type_trans(skb, netdev);
-
/* Alloc another RX SKB for this same index */
rx_skb = mlxbf_gige_alloc_skb(priv, MLXBF_GIGE_DEFAULT_BUF_SZ,
&rx_buf_dma, DMA_FROM_DEVICE);
if (!rx_skb)
return false;
priv->rx_skb[rx_pi_rem] = rx_skb;
+
+ skb_put(skb, datalen);
+
+ skb->ip_summed = CHECKSUM_NONE; /* device did not checksum packet */
+
+ skb->protocol = eth_type_trans(skb, netdev);
+
dma_unmap_single(priv->dev, *rx_wqe_addr,
MLXBF_GIGE_DEFAULT_BUF_SZ, DMA_FROM_DEVICE);
*rx_wqe_addr = rx_buf_dma;
--
2.35.3

2023-05-24 04:00:32

by Jakub Kicinski

Subject: Re: [PATCH net] net: mellanox: mlxbf_gige: Fix skb_panic splat under memory pressure

On Mon, 22 May 2023 21:07:13 +0200 Thomas Bogendoerfer wrote:
> Do skp_put() after a new skb has been successfully allocated otherwise
> the reused skb leads to skp_panics or incorrect packet sizes.

s/skp/skb/

Please add a Fixes tag pointing at the patch which introduced
the problem (could be the first commit adding this driver).

> Signed-off-by: Thomas Bogendoerfer <[email protected]>
> ---
> .../ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c | 13 +++++++------
> 1 file changed, 7 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
> index afa3b92a6905..2c132849a76d 100644
> --- a/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
> +++ b/drivers/net/ethernet/mellanox/mlxbf_gige/mlxbf_gige_rx.c
> @@ -245,18 +245,19 @@ static bool mlxbf_gige_rx_packet(struct mlxbf_gige *priv, int *rx_pkts)
>
> skb = priv->rx_skb[rx_pi_rem];
>
> - skb_put(skb, datalen);
> -
> - skb->ip_summed = CHECKSUM_NONE; /* device did not checksum packet */
> -
> - skb->protocol = eth_type_trans(skb, netdev);
> -
> /* Alloc another RX SKB for this same index */
> rx_skb = mlxbf_gige_alloc_skb(priv, MLXBF_GIGE_DEFAULT_BUF_SZ,
> &rx_buf_dma, DMA_FROM_DEVICE);
> if (!rx_skb)
> return false;
> priv->rx_skb[rx_pi_rem] = rx_skb;
> +
> + skb_put(skb, datalen);
> +
> + skb->ip_summed = CHECKSUM_NONE; /* device did not checksum packet */
> +
> + skb->protocol = eth_type_trans(skb, netdev);
>
> dma_unmap_single(priv->dev, *rx_wqe_addr,
> MLXBF_GIGE_DEFAULT_BUF_SZ, DMA_FROM_DEVICE);

You should move the code here, after the dma_unmap().
eth_type_trans() will access the payload and until dma_unmap
the data should not be accessed by the CPU.

> *rx_wqe_addr = rx_buf_dma;

--
pw-bot: cr