Re: [PATCH] p54: add null pointer check before releasing socket buffer

Re: [PATCH] p54: add null pointer check before releasing socket buffer

[Christian Lamparter]

(Added linux-wireless, since this is a wireless driver)

On Sunday, April 9, 2017 10:23:20 PM CEST Myungho Jung wrote:
> Kernel panic is caused by trying to dereference null pointer. Check if
> the pointer is null before freeing space.
Do you have the kernel panic somewhere?
I think you have an even bigger problem: You see, in order to get EEPROM
readback and rx_stats feedback you need to sent a request to the firmware
and if the response's req_id cookies don't match, you end up filling up 
the very limited device address space.

As for adding if (!skb) checks. I think kfree, kfree_skb, dev_kfree_skb 
(aka consume_skb) all check for null pointers already. So the logical
thing to do would be to make dev_kfree_skb_irq (which would also fix
dev_kfree_skb_any) consistent with kfree, kfree_skb, dev_kfree_skb and
add the check there.

> Signed-off-by: Myungho Jung <mhjungk@gmail.com>
> ---
>  drivers/net/wireless/intersil/p54/txrx.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/wireless/intersil/p54/txrx.c b/drivers/net/wireless/intersil/p54/txrx.c
> index 1af7da0..8956061 100644
> --- a/drivers/net/wireless/intersil/p54/txrx.c
> +++ b/drivers/net/wireless/intersil/p54/txrx.c
> @@ -503,7 +503,9 @@ static void p54_rx_eeprom_readback(struct p54_common *priv,
>  
>  	priv->eeprom = NULL;
>  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> -	dev_kfree_skb_any(tmp);
> +	if (unlikely(!tmp))
> +		dev_kfree_skb_any(tmp);
> +
>  	complete(&priv->eeprom_comp);
>  }
>  
> @@ -597,7 +599,9 @@ static void p54_rx_stats(struct p54_common *priv, struct sk_buff *skb)
>  	}
>  
>  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> -	dev_kfree_skb_any(tmp);
> +	if (unlikely(!tmp))
> +		dev_kfree_skb_any(tmp);
> +
>  	complete(&priv->stat_comp);
>  }
>  
> 

Re: [PATCH] p54: add null pointer check before releasing socket buffer

[Myungho Jung]

On Mon, Apr 10, 2017 at 02:12:54PM +0200, Christian Lamparter wrote:
> (Added linux-wireless, since this is a wireless driver)
> 
> On Sunday, April 9, 2017 10:23:20 PM CEST Myungho Jung wrote:
> > Kernel panic is caused by trying to dereference null pointer. Check if
> > the pointer is null before freeing space.
> Do you have the kernel panic somewhere?
> I think you have an even bigger problem: You see, in order to get EEPROM
> readback and rx_stats feedback you need to sent a request to the firmware
> and if the response's req_id cookies don't match, you end up filling up 
> the very limited device address space.
> 
> As for adding if (!skb) checks. I think kfree, kfree_skb, dev_kfree_skb 
> (aka consume_skb) all check for null pointers already. So the logical
> thing to do would be to make dev_kfree_skb_irq (which would also fix
> dev_kfree_skb_any) consistent with kfree, kfree_skb, dev_kfree_skb and
> add the check there.
> 
> > Signed-off-by: Myungho Jung <mhjungk@gmail.com>
> > ---
> >  drivers/net/wireless/intersil/p54/txrx.c | 8 ++++++--
> >  1 file changed, 6 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/net/wireless/intersil/p54/txrx.c b/drivers/net/wireless/intersil/p54/txrx.c
> > index 1af7da0..8956061 100644
> > --- a/drivers/net/wireless/intersil/p54/txrx.c
> > +++ b/drivers/net/wireless/intersil/p54/txrx.c
> > @@ -503,7 +503,9 @@ static void p54_rx_eeprom_readback(struct p54_common *priv,
> >  
> >  	priv->eeprom = NULL;
> >  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> > -	dev_kfree_skb_any(tmp);
> > +	if (unlikely(!tmp))
> > +		dev_kfree_skb_any(tmp);
> > +
> >  	complete(&priv->eeprom_comp);
> >  }
> >  
> > @@ -597,7 +599,9 @@ static void p54_rx_stats(struct p54_common *priv, struct sk_buff *skb)
> >  	}
> >  
> >  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> > -	dev_kfree_skb_any(tmp);
> > +	if (unlikely(!tmp))
> > +		dev_kfree_skb_any(tmp);
> > +
> >  	complete(&priv->stat_comp);
> >  }
> >  
> > 
> 
> 

I found that the fix was totally opposite to my thought. Sorry about
confusion. I'm not sure it actually caused kernel panic but guessed from
a bug report [https://bugzilla.kernel.org/show_bug.cgi?id=195289]. And
correct fix will be like this:
	if (likely(tmp))
		dev_kfree_skb_any(tmp);

But, like you said, I think null pointer should be checked in
dev_kfree_skb_irq although already checking before calling it in many
other places. I'll try another patch. Thank you for your advice.

Thanks,
Myungho

Re: [PATCH] p54: add null pointer check before releasing socket buffer

[Christian Lamparter]

On Monday, April 10, 2017 1:54:14 PM CEST Myungho Jung wrote:
> On Mon, Apr 10, 2017 at 02:12:54PM +0200, Christian Lamparter wrote:
> > On Sunday, April 9, 2017 10:23:20 PM CEST Myungho Jung wrote:
> > > Kernel panic is caused by trying to dereference null pointer. Check if
> > > the pointer is null before freeing space.
> >  [...]
> > As for adding if (!skb) checks. I think kfree, kfree_skb, dev_kfree_skb 
> > (aka consume_skb) all check for null pointers already. So the logical
> > thing to do would be to make dev_kfree_skb_irq (which would also fix
> > dev_kfree_skb_any) consistent with kfree, kfree_skb, dev_kfree_skb and
> > add the check there.
> > > ---
> > >  drivers/net/wireless/intersil/p54/txrx.c | 8 ++++++--
> > >  1 file changed, 6 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/net/wireless/intersil/p54/txrx.c b/drivers/net/wireless/intersil/p54/txrx.c
> > > index 1af7da0..8956061 100644
> > > --- a/drivers/net/wireless/intersil/p54/txrx.c
> > > +++ b/drivers/net/wireless/intersil/p54/txrx.c
> > > @@ -503,7 +503,9 @@ static void p54_rx_eeprom_readback(struct p54_common *priv,
> > >  
> > >  	priv->eeprom = NULL;
> > >  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> > > -	dev_kfree_skb_any(tmp);
> > > +	if (unlikely(!tmp))
> > > +		dev_kfree_skb_any(tmp);
> > > +
> > >  	complete(&priv->eeprom_comp);
> > >  }
> > >  
> > > @@ -597,7 +599,9 @@ static void p54_rx_stats(struct p54_common *priv, struct sk_buff *skb)
> > >  	}
> > >  
> > >  	tmp = p54_find_and_unlink_skb(priv, hdr->req_id);
> > > -	dev_kfree_skb_any(tmp);
> > > +	if (unlikely(!tmp))
> > > +		dev_kfree_skb_any(tmp);
> > > +
> > >  	complete(&priv->stat_comp);
> > >  }
> > >  
> > > 
> > 
> > 
> [...] I'm not sure it actually caused kernel panic but guessed from
> a bug report [https://bugzilla.kernel.org/show_bug.cgi?id=195289].
Thank you for bringing this to my attention. Reading the bugreport, it
does sound like there's a bigger issue with the USB Ports. I'll see if
this can be fixed. But it does sound like a hardware issue at this 
point.

> And correct fix will be like this:
> 	if (likely(tmp))
> 		dev_kfree_skb_any(tmp);
> 
> But, like you said, I think null pointer should be checked in
> dev_kfree_skb_irq although already checking before calling it in many
> other places. I'll try another patch. Thank you for your advice.

Well, the patch could be as simple as this:
---
diff --git a/net/core/dev.c b/net/core/dev.c
index 7869ae3837ca..44f7d5a1c67c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2450,6 +2450,9 @@ void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
 {
 	unsigned long flags;
 
+	if (!skb)
+		return;
+
 	if (likely(atomic_read(&skb->users) == 1)) {
 		smp_rmb();
 		atomic_set(&skb->users, 0);
---

The question is: would David or Eric support the change. Any comments,
what's the prefered solution? Just patch __dev_kfree_skb_irq to make
it consistent with *kfree*, or patch the driver? I'm fine either way,
but I would prefere patching __dev_kfree_skb_irq.

Regards,
Christian

Re: [PATCH] p54: add null pointer check before releasing socket buffer

[Eric Dumazet]

On Mon, Apr 10, 2017 at 2:22 PM, Christian Lamparter
<chunkeey@googlemail.com> wrote:

> Well, the patch could be as simple as this:
> ---
> diff --git a/net/core/dev.c b/net/core/dev.c
> index 7869ae3837ca..44f7d5a1c67c 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -2450,6 +2450,9 @@ void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
>  {
>         unsigned long flags;
>
> +       if (!skb)
> +               return;
> +
>         if (likely(atomic_read(&skb->users) == 1)) {
>                 smp_rmb();
>                 atomic_set(&skb->users, 0);
> ---
>
> The question is: would David or Eric support the change. Any comments,
> what's the prefered solution? Just patch __dev_kfree_skb_irq to make
> it consistent with *kfree*, or patch the driver? I'm fine either way,
> but I would prefere patching __dev_kfree_skb_irq.

This is fine, same check happens in consume_skb()