rtlwifi: Fix kernel oops introduced with commit e49656147359

Commit Message

Larry Finger Dec. 20, 2016, 2:38 a.m. UTC

With commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of
kfree_skb"), the method used to free an skb was changed because the
kfree_skb() was inside a spinlock. What was forgotten is that kfree_skb()
guards against a NULL value for the argument. Routine dev_kfree_skb_irq()
does not, and a test is needed to prevent kernel panics.

Fixes: commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")
Signed-off-by: Larry Finger <[e-mail]>
Cc: Stable <[e-mail]> (4.9+)
Cc: Wei Yongjun <[e-mail]>
---
Kalle,

This change should be sent to mainline during the 4.10 merge period,
or as soon as possible.

Thanks,

Larry
---
 drivers/net/wireless/realtek/rtlwifi/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Kalle Valo Dec. 20, 2016, 11:21 a.m. UTC | #1

Larry Finger <[e-mail]> writes:

> With commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of
> kfree_skb"), the method used to free an skb was changed because the
> kfree_skb() was inside a spinlock. What was forgotten is that kfree_skb()
> guards against a NULL value for the argument. Routine dev_kfree_skb_irq()
> does not, and a test is needed to prevent kernel panics.
>
> Fixes: commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")

This should be:

Fixes: e49656147359 ("rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")

> Signed-off-by: Larry Finger <[e-mail]>
> Cc: Stable <[e-mail]> (4.9+)

And this:

Cc: Stable <[e-mail]> # 4.9+

I can fix both of them.

> Cc: Wei Yongjun <[e-mail]>
> ---
> Kalle,
>
> This change should be sent to mainline during the 4.10 merge period,
> or as soon as possible.

Ok, I'll queue this to 4.10. Most likely I'll send a pull request to
Dave later this week or so.

Larry Finger Dec. 20, 2016, 3:25 p.m. UTC | #2

On 12/20/2016 05:21 AM, Kalle Valo wrote:
> Larry Finger <[e-mail]> writes:
>
>> With commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of
>> kfree_skb"), the method used to free an skb was changed because the
>> kfree_skb() was inside a spinlock. What was forgotten is that kfree_skb()
>> guards against a NULL value for the argument. Routine dev_kfree_skb_irq()
>> does not, and a test is needed to prevent kernel panics.
>>
>> Fixes: commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")
>
> This should be:
>
> Fixes: e49656147359 ("rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")
>
>> Signed-off-by: Larry Finger <[e-mail]>
>> Cc: Stable <[e-mail]> (4.9+)
>
> And this:
>
> Cc: Stable <[e-mail]> # 4.9+
>
> I can fix both of them.
>
>> Cc: Wei Yongjun <[e-mail]>
>> ---
>> Kalle,
>>
>> This change should be sent to mainline during the 4.10 merge period,
>> or as soon as possible.
>
> Ok, I'll queue this to 4.10. Most likely I'll send a pull request to
> Dave later this week or so.

Thanks for the suggested changes, and for the quick action.

Larry

Kalle Valo Dec. 21, 2016, 2:35 p.m. UTC | #3

Larry Finger <[e-mail]> wrote:
> With commit e49656147359 {"rtlwifi: Use dev_kfree_skb_irq instead of
> kfree_skb"), the method used to free an skb was changed because the
> kfree_skb() was inside a spinlock. What was forgotten is that kfree_skb()
> guards against a NULL value for the argument. Routine dev_kfree_skb_irq()
> does not, and a test is needed to prevent kernel panics.
> 
> Fixes: e49656147359 ("rtlwifi: Use dev_kfree_skb_irq instead of kfree_skb")
> Signed-off-by: Larry Finger <[e-mail]>
> Cc: Stable <[e-mail]> # 4.9+
> Cc: Wei Yongjun <[e-mail]>

Patch applied to wireless-drivers.git, thanks.

22b68b93ae25 rtlwifi: Fix kernel oops introduced with commit e49656147359

Patch

diff --git a/drivers/net/wireless/realtek/rtlwifi/core.c b/drivers/net/wireless/realtek/rtlwifi/core.c
index 955055b..df8b977 100644
--- a/drivers/net/wireless/realtek/rtlwifi/core.c
+++ b/drivers/net/wireless/realtek/rtlwifi/core.c
@@ -1823,7 +1823,8 @@  bool rtl_cmd_send_packet(struct ieee80211_hw *hw, struct sk_buff *skb)
 
 	spin_lock_irqsave(&rtlpriv->locks.irq_th_lock, flags);
 	pskb = __skb_dequeue(&ring->queue);
-	dev_kfree_skb_irq(pskb);
+	if (pskb)
+		dev_kfree_skb_irq(pskb);
 
 	/*this is wrong, fill_tx_cmddesc needs update*/
 	pdesc = &ring->desc[0];