Re: sh_eth.c::sh_eth_rx(): mdp->rx_skbuff[entry] can be NULL

[Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>]

Hello.

On 10/19/2015 06:01 PM, Yasushi SHOJI wrote:

> In a low memory situation with netdev_alloc_skb() failure,
> mdp->rx_skbuff[entry] can be left NULL, however, sh_eth_rx() seems to
> access it without checking NULL or not in the following code:
>
> 			skb = mdp->rx_skbuff[entry];
> 			mdp->rx_skbuff[entry] = NULL;
> 			if (mdp->cd->rpadir)
> 				skb_reserve(skb, NET_IP_ALIGN);
> 			dma_unmap_single(&ndev->dev, rxdesc->addr,
> 					 ALIGN(mdp->rx_buf_sz, 16),
> 					 DMA_FROM_DEVICE);
>
> I've put BUG_ON() to test skb and got the following backtrace.  I can
> also enable slub poisoning to see some bad access.
>
> I'm not that familiar with this code base so I'm note including any
> patch yet.  I appreciate if someone with insight in this code give a
> quick look and tell me that it's a real one or not.  if this is a real
> case, I can take a deep look.

    If you got the oops, it's real. Thanks for the reporting. I guess I should 
check the new ravb driver as well...
    Do you want to try fixing the bug yourself?

> BTW, the backtrace is from old 3.4.74+ kernel but the current tip is
> very close.

     Yeah, this part didn't change in a long time...

> Thanks,

MBR, Sergei