[PATCH][net-next] gianfar: Fix alloc_skb_resources on -ENOMEM cleanup path

[PATCH][net-next] gianfar: Fix alloc_skb_resources on -ENOMEM cleanup path

[Claudiu Manoil]

Should gfar_init_bds() return with -ENOMEM inside gfar_alloc_skb_resources(),
free_skb_resources() will be called twice in a row on the "cleanup" path,
leading to duplicate kfree() calls for rx_|tx_queue->rx_|tx_skbuff resulting
in segmentation fault.
This patch prevents the segmentation fault to happen in the future
(rx_|tx_sbkbuff set to NULL), and corrects the error path handling
for gfar_init_bds().

Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: "David S. Miller" <davem@davemloft.net>

Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
---
 drivers/net/ethernet/freescale/gianfar.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
index 1d03dcd..c5c82ad 100644
--- a/drivers/net/ethernet/freescale/gianfar.c
+++ b/drivers/net/ethernet/freescale/gianfar.c
@@ -311,7 +311,7 @@ static int gfar_alloc_skb_resources(struct net_device *ndev)
 	}
 
 	if (gfar_init_bds(ndev))
-		goto cleanup;
+		return -ENOMEM;
 
 	return 0;
 
@@ -1356,7 +1356,9 @@ static int gfar_restore(struct device *dev)
 	if (!netif_running(ndev))
 		return 0;
 
-	gfar_init_bds(ndev);
+	if (gfar_init_bds(ndev))
+		return -ENOMEM;
+
 	init_registers(ndev);
 	gfar_set_mac_address(ndev);
 	gfar_init_mac(ndev);
@@ -1709,6 +1711,7 @@ static void free_skb_tx_queue(struct gfar_priv_tx_q *tx_queue)
 		tx_queue->tx_skbuff[i] = NULL;
 	}
 	kfree(tx_queue->tx_skbuff);
+	tx_queue->tx_skbuff = NULL;
 }
 
 static void free_skb_rx_queue(struct gfar_priv_rx_q *rx_queue)
@@ -1732,6 +1735,7 @@ static void free_skb_rx_queue(struct gfar_priv_rx_q *rx_queue)
 		rxbdp++;
 	}
 	kfree(rx_queue->rx_skbuff);
+	rx_queue->rx_skbuff = NULL;
 }
 
 /* If there are any tx skbs or rx skbs still around, free them.
-- 
1.6.6

Re: [PATCH][net-next] gianfar: Fix alloc_skb_resources on -ENOMEM cleanup path

[Paul Gortmaker]

On 12-11-08 08:40 AM, Claudiu Manoil wrote:
> Should gfar_init_bds() return with -ENOMEM inside gfar_alloc_skb_resources(),
> free_skb_resources() will be called twice in a row on the "cleanup" path,
> leading to duplicate kfree() calls for rx_|tx_queue->rx_|tx_skbuff resulting
> in segmentation fault.
> This patch prevents the segmentation fault to happen in the future
> (rx_|tx_sbkbuff set to NULL), and corrects the error path handling
> for gfar_init_bds().

Since gfar_init_bds is more like a slave routine to gfar_alloc_skb_resources,
I think the dup free_skb_resources should remain in the parent, and be removed
from gfar_init_bds.  Otherwise the gfar_alloc_skb_resources will appear
confusing -- one will think it it allocates some resources, hits a failure
and then returns without bothering to do any cleanup of the parts it
did manage to allocate. (Then gfar_restore will have to call the free
itself _if_ gfar_init_bds fails too.)

Paul.
--

> 
> Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
> Cc: "David S. Miller" <davem@davemloft.net>
> 
> Signed-off-by: Claudiu Manoil <claudiu.manoil@freescale.com>
> ---
>  drivers/net/ethernet/freescale/gianfar.c |    8 ++++++--
>  1 files changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/ethernet/freescale/gianfar.c b/drivers/net/ethernet/freescale/gianfar.c
> index 1d03dcd..c5c82ad 100644
> --- a/drivers/net/ethernet/freescale/gianfar.c
> +++ b/drivers/net/ethernet/freescale/gianfar.c
> @@ -311,7 +311,7 @@ static int gfar_alloc_skb_resources(struct net_device *ndev)
>  	}
>  
>  	if (gfar_init_bds(ndev))
> -		goto cleanup;
> +		return -ENOMEM;
>  
>  	return 0;
>  
> @@ -1356,7 +1356,9 @@ static int gfar_restore(struct device *dev)
>  	if (!netif_running(ndev))
>  		return 0;
>  
> -	gfar_init_bds(ndev);
> +	if (gfar_init_bds(ndev))
> +		return -ENOMEM;
> +
>  	init_registers(ndev);
>  	gfar_set_mac_address(ndev);
>  	gfar_init_mac(ndev);
> @@ -1709,6 +1711,7 @@ static void free_skb_tx_queue(struct gfar_priv_tx_q *tx_queue)
>  		tx_queue->tx_skbuff[i] = NULL;
>  	}
>  	kfree(tx_queue->tx_skbuff);
> +	tx_queue->tx_skbuff = NULL;
>  }
>  
>  static void free_skb_rx_queue(struct gfar_priv_rx_q *rx_queue)
> @@ -1732,6 +1735,7 @@ static void free_skb_rx_queue(struct gfar_priv_rx_q *rx_queue)
>  		rxbdp++;
>  	}
>  	kfree(rx_queue->rx_skbuff);
> +	rx_queue->rx_skbuff = NULL;
>  }
>  
>  /* If there are any tx skbs or rx skbs still around, free them.
> 

Re: [PATCH][net-next] gianfar: Fix alloc_skb_resources on -ENOMEM cleanup path

[Claudiu Manoil]

On 11/8/2012 4:21 PM, Paul Gortmaker wrote:
> On 12-11-08 08:40 AM, Claudiu Manoil wrote:
>> Should gfar_init_bds() return with -ENOMEM inside gfar_alloc_skb_resources(),
>> free_skb_resources() will be called twice in a row on the "cleanup" path,
>> leading to duplicate kfree() calls for rx_|tx_queue->rx_|tx_skbuff resulting
>> in segmentation fault.
>> This patch prevents the segmentation fault to happen in the future
>> (rx_|tx_sbkbuff set to NULL), and corrects the error path handling
>> for gfar_init_bds().
>
> Since gfar_init_bds is more like a slave routine to gfar_alloc_skb_resources,
> I think the dup free_skb_resources should remain in the parent, and be removed
> from gfar_init_bds.  Otherwise the gfar_alloc_skb_resources will appear
> confusing -- one will think it it allocates some resources, hits a failure
> and then returns without bothering to do any cleanup of the parts it
> did manage to allocate. (Then gfar_restore will have to call the free
> itself _if_ gfar_init_bds fails too.)
>
> Paul.

You're right. I'll send the v1 patch shortly.
Thanks.

Claudiu