Re: [PATCH net] NFC: hci: fix sleep in atomic context bugs in nfc_hci_hcp_message_tx

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 16/05/2022 12:18, duoming@zju.edu.cn wrote:
> Hello,
> 
> On Mon, 16 May 2022 08:44:39 +0200 Krzysztof wrote:
> 
>>> There are sleep in atomic context bugs when the request to secure
>>> element of st21nfca is timeout. The root cause is that kzalloc and
>>> alloc_skb with GFP_KERNEL parameter is called in st21nfca_se_wt_timeout
>>> which is a timer handler. The call tree shows the execution paths that
>>> could lead to bugs:
>>>
>>>    (Interrupt context)
>>> st21nfca_se_wt_timeout
>>>   nfc_hci_send_event
>>>     nfc_hci_hcp_message_tx
>>>       kzalloc(..., GFP_KERNEL) //may sleep
>>>       alloc_skb(..., GFP_KERNEL) //may sleep
>>>
>>> This patch changes allocation mode of kzalloc and alloc_skb from
>>> GFP_KERNEL to GFP_ATOMIC in order to prevent atomic context from
>>> sleeping. The GFP_ATOMIC flag makes memory allocation operation
>>> could be used in atomic context.
>>>
>>> Fixes: 8b8d2e08bf0d ("NFC: HCI support")
>>> Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
>>> ---
>>>  net/nfc/hci/hcp.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/net/nfc/hci/hcp.c b/net/nfc/hci/hcp.c
>>> index 05c60988f59..1caf9c2086f 100644
>>> --- a/net/nfc/hci/hcp.c
>>> +++ b/net/nfc/hci/hcp.c
>>> @@ -30,7 +30,7 @@ int nfc_hci_hcp_message_tx(struct nfc_hci_dev *hdev, u8 pipe,
>>>  	int hci_len, err;
>>>  	bool firstfrag = true;
>>>  
>>> -	cmd = kzalloc(sizeof(struct hci_msg), GFP_KERNEL);
>>> +	cmd = kzalloc(sizeof(*cmd), GFP_ATOMIC);
>>
>> No, this does not look correct. This function can sleep, so it can use
>> GFP_KERNEL. Please just look at the function before replacing any flags...
> 
> I am sorry, I don`t understand why nfc_hci_hcp_message_tx() can sleep.

Why? because in line 93 it uses a mutex (which is a sleeping primitve).

> 
> I think st21nfca_se_wt_timeout() is a timer handler, which is in interrupt context.
> The call tree shows the execution paths that could lead to bugs:
> 
> st21nfca_hci_se_io()
>   mod_timer(&info->se_info.bwi_timer,...)
>     st21nfca_se_wt_timeout()  //timer handler, interrupt context
>       nfc_hci_send_event()
>         nfc_hci_hcp_message_tx()
>           kzalloc(..., GFP_KERNEL) //may sleep
>           alloc_skb(..., GFP_KERNEL) //may sleep
> 
> What`s more, I think the "mutex_lock(&hdev->msg_tx_mutex)" called by nfc_hci_hcp_message_tx()
> is also improper.
> 
> Please correct me, If you think I am wrong. Thanks for your time.

Your patch is not correct in current semantics of this function. This
function can sleep (because it uses a mutex), so the mistake is rather
calling it from interrupt context.


Best regards,
Krzysztof