[PATCH] can: mcba_usb: fix possible double dev_kfree_skb in mcba_usb_start_xmit

[PATCH] can: mcba_usb: fix possible double dev_kfree_skb in mcba_usb_start_xmit

[Hangyu Hua]

There is no need to call dev_kfree_skb when usb_submit_urb fails beacause
can_put_echo_skb deletes original skb and can_free_echo_skb deletes the cloned
skb.

Fixes: 51f3baad7de9 ("can: mcba_usb: Add support for Microchip CAN BUS Analyzer")
Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
---
 drivers/net/can/usb/mcba_usb.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/net/can/usb/mcba_usb.c b/drivers/net/can/usb/mcba_usb.c
index 77bddff86252..7c198eb5bc9c 100644
--- a/drivers/net/can/usb/mcba_usb.c
+++ b/drivers/net/can/usb/mcba_usb.c
@@ -364,7 +364,6 @@ static netdev_tx_t mcba_usb_start_xmit(struct sk_buff *skb,
 xmit_failed:
 	can_free_echo_skb(priv->netdev, ctx->ndx, NULL);
 	mcba_usb_free_ctx(ctx);
-	dev_kfree_skb(skb);
 	stats->tx_dropped++;
 
 	return NETDEV_TX_OK;
-- 
2.25.1

Re: [PATCH] can: mcba_usb: fix possible double dev_kfree_skb in mcba_usb_start_xmit

[Hangyu Hua]

Gentle ping.

On 2022/3/11 16:02, Hangyu Hua wrote:
> There is no need to call dev_kfree_skb when usb_submit_urb fails beacause
> can_put_echo_skb deletes original skb and can_free_echo_skb deletes the cloned
> skb.
> 
> Fixes: 51f3baad7de9 ("can: mcba_usb: Add support for Microchip CAN BUS Analyzer")
> Signed-off-by: Hangyu Hua <hbh25y@gmail.com>
> ---
>   drivers/net/can/usb/mcba_usb.c | 1 -
>   1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/net/can/usb/mcba_usb.c b/drivers/net/can/usb/mcba_usb.c
> index 77bddff86252..7c198eb5bc9c 100644
> --- a/drivers/net/can/usb/mcba_usb.c
> +++ b/drivers/net/can/usb/mcba_usb.c
> @@ -364,7 +364,6 @@ static netdev_tx_t mcba_usb_start_xmit(struct sk_buff *skb,
>   xmit_failed:
>   	can_free_echo_skb(priv->netdev, ctx->ndx, NULL);
>   	mcba_usb_free_ctx(ctx);
> -	dev_kfree_skb(skb);
>   	stats->tx_dropped++;
>   
>   	return NETDEV_TX_OK;

Re: [PATCH] can: mcba_usb: fix possible double dev_kfree_skb in mcba_usb_start_xmit

[Yasushi SHOJI]

Hi Hangyu,

On Fri, Mar 11, 2022 at 5:02 PM Hangyu Hua <hbh25y@gmail.com> wrote:
>
> There is no need to call dev_kfree_skb when usb_submit_urb fails beacause
> can_put_echo_skb deletes original skb and can_free_echo_skb deletes the cloned
> skb.

So, it's more like, "we don't need to call dev_kfree_skb() after
can_put_echo_skb()
because can_put_echo_skb() consumes the given skb.".  It seems it doesn't depend
on the condition of usb_submit_urb().  Plus, we don't see the "cloned
skb" at the
call site.

Would you mind adding a comment on can_put_echo_skb(), in a separate patch,
saying the fact that it consumes the skb?

ems_usb.c, gs_usb.c and possibly some others seem to call
dev_kfree_skb() as well.
Are they affected?

Best,
-- 
           yashi

Re: [PATCH] can: mcba_usb: fix possible double dev_kfree_skb in mcba_usb_start_xmit

[Hangyu Hua]

Hi yashi,

You are right. There are a series of the same problems about 
can_put_echo_skb.

This issue was discovered while I was discussing a incorrect patch with 
Marc. You can check this in

https://lore.kernel.org/all/20220225060019.21220-1-hbh25y@gmail.com/

So i submitted a new patch. This was the first place where the problem 
occurs. You can check this in

[1] https://lore.kernel.org/all/20220228083639.38183-1-hbh25y@gmail.com/

After a week, I realized it could be a series of problems. So
i submitted the following patches

[2] 
https://lore.kernel.org/all/0d2f9980-fb1d-4068-7868-effc77892a97@gmail.com/
[3] 
https://lore.kernel.org/all/de416319-c027-837d-4b8c-b8c3c37ed88e@gmail.com/
[4] https://lore.kernel.org/all/20220317081305.739554-1-mkl@pengutronix.de/

I think this is all affected. None of the four have been merged 
upstream. Do i need to remake all these four patches?

Thanks,
Hangyu






On 2022/3/23 07:08, Yasushi SHOJI wrote:
> Hi Hangyu,
> 
> On Fri, Mar 11, 2022 at 5:02 PM Hangyu Hua <hbh25y@gmail.com> wrote:
>>
>> There is no need to call dev_kfree_skb when usb_submit_urb fails beacause
>> can_put_echo_skb deletes original skb and can_free_echo_skb deletes the cloned
>> skb.
> 
> So, it's more like, "we don't need to call dev_kfree_skb() after
> can_put_echo_skb()
> because can_put_echo_skb() consumes the given skb.".  It seems it doesn't depend
> on the condition of usb_submit_urb().  Plus, we don't see the "cloned
> skb" at the
> call site.
> 
> Would you mind adding a comment on can_put_echo_skb(), in a separate patch,
> saying the fact that it consumes the skb?
> 
> ems_usb.c, gs_usb.c and possibly some others seem to call
> dev_kfree_skb() as well.
> Are they affected?
> 
> Best,