Re: Re: [PATCH] nfc: nci: Fix uninit-value in nci_rx_work()

[zhanghao]

On 04/08/2024 12:57, Simon Horman wrote:
> > On Sat, Aug 03, 2024 at 08:18:17PM +0800, zhanghao wrote:
> >> Commit e624e6c3e777 ("nfc: Add a virtual nci device driver")
> >> calls alloc_skb() with GFP_KERNEL as the argument flags.The
> >> allocated heap memory was not initialized.This causes KMSAN
> >> to detect an uninitialized value.
> >>
> >> Reported-by: syzbot+3da70a0abd7f5765b6ea@xxxxxxxxxxxxxxxxxxxxxxxxx
> >> Closes: https://syzkaller.appspot.com/bug?extid=3da70a0abd7f5765b6ea
> > 
> > Hi,
> > 
> > I wonder if the problem reported above is caused by accessing packet
> > data which is past the end of what is copied in virtual_ncidev_write().
> > I.e. count is unusually short and this is not being detected.
> > 
Yes, you're right.I debug the kernel and find that skb->data is " ".The 
length is less than 2.Using nci_plen to access skb->data in nci_rx_work()
triggers kmsan to detect access to uninitialized memory.
 
I wonder if I can use skb->len to determine length in the
nci_rx_work().I tested it and there was no problem.
 
> >> Fixes: e624e6c3e777 ("nfc: Add a virtual nci device driver")
> >> Link: https://lore.kernel.org/all/000000000000747dd6061a974686@xxxxxxxxxx/T/
> >> Signed-off-by: zhanghao <zhanghao1@xxxxxxxxxx>
> >> ---
> >>  drivers/nfc/virtual_ncidev.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/nfc/virtual_ncidev.c b/drivers/nfc/virtual_ncidev.c
> >> index 6b89d596ba9a..ae1592db131e 100644
> >> --- a/drivers/nfc/virtual_ncidev.c
> >> +++ b/drivers/nfc/virtual_ncidev.c
> >> @@ -117,7 +117,7 @@ static ssize_t virtual_ncidev_write(struct file *file,
> >>  	struct virtual_nci_dev *vdev = file->private_data;
> >>  	struct sk_buff *skb;
> >>  
> >> -	skb = alloc_skb(count, GFP_KERNEL);
> >> +	skb = alloc_skb(count, GFP_KERNEL|__GFP_ZERO);
> >>  	if (!skb)
> >>  		return -ENOMEM;
> > 
> > I'm not sure this helps wrt initialising the memory as immediately below there
> > is;
> > 
> > 	if (copy_from_user(skb_put(skb, count), buf, count)) {
> > 		...
> > 
> > Which I assume will initialise count bytes of skb data.
> 
> Yeah, this looks like hiding the real issue.
> 
> Best regards,
> Krzysztof