[PATCH] staging: most: don't access hdm_ch before checking it valid

[PATCH] staging: most: don't access hdm_ch before checking it valid

[Jing Xiangfeng]

In try_start_dim_transfer(), pointer hdm_ch is accessed before checking.
This may lead to a potential null pointer dereference. Fix this by
dereferencing hdm_ch after calling BUG_ON().

Signed-off-by: Jing Xiangfeng <jingxiangfeng@huawei.com>
---
 drivers/staging/most/dim2/dim2.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/staging/most/dim2/dim2.c b/drivers/staging/most/dim2/dim2.c
index 509c8012d20b..ccd7cc7545e4 100644
--- a/drivers/staging/most/dim2/dim2.c
+++ b/drivers/staging/most/dim2/dim2.c
@@ -148,7 +148,7 @@ void dimcb_on_error(u8 error_id, const char *error_message)
 static int try_start_dim_transfer(struct hdm_channel *hdm_ch)
 {
 	u16 buf_size;
-	struct list_head *head = &hdm_ch->pending_list;
+	struct list_head *head;
 	struct mbo *mbo;
 	unsigned long flags;
 	struct dim_ch_state_t st;
@@ -156,6 +156,7 @@ static int try_start_dim_transfer(struct hdm_channel *hdm_ch)
 	BUG_ON(!hdm_ch);
 	BUG_ON(!hdm_ch->is_initialized);
 
+	head = &hdm_ch->pending_list;
 	spin_lock_irqsave(&dim_lock, flags);
 	if (list_empty(head)) {
 		spin_unlock_irqrestore(&dim_lock, flags);
-- 
2.17.1

Re: [PATCH] staging: most: don't access hdm_ch before checking it valid

[Dan Carpenter]

On Mon, Sep 28, 2020 at 06:48:38PM +0800, Jing Xiangfeng wrote:
> In try_start_dim_transfer(), pointer hdm_ch is accessed before checking.
> This may lead to a potential null pointer dereference. Fix this by
> dereferencing hdm_ch after calling BUG_ON().
> 
> Signed-off-by: Jing Xiangfeng <jingxiangfeng@huawei.com>
> ---
>  drivers/staging/most/dim2/dim2.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/staging/most/dim2/dim2.c b/drivers/staging/most/dim2/dim2.c
> index 509c8012d20b..ccd7cc7545e4 100644
> --- a/drivers/staging/most/dim2/dim2.c
> +++ b/drivers/staging/most/dim2/dim2.c
> @@ -148,7 +148,7 @@ void dimcb_on_error(u8 error_id, const char *error_message)
>  static int try_start_dim_transfer(struct hdm_channel *hdm_ch)
>  {
>  	u16 buf_size;
> -	struct list_head *head = &hdm_ch->pending_list;

This is not a dereference, it's just pointer math.  In other words:

	struct list_head *head = hdm_ch + offsetof(struct hdm_channel, pending_list);

So the commit message is wrong because this cannot lead to a NULL
dereference.  It's better to just delete the BUG_ON().  We don't really
like BUG_ON().  Checkpatch will complain about them.  An Oops gives
basically the same information as a BUG_ON() without completely killing
the kernel so just dereferencing a NULL is preferable.  Finally, we can
see from the callers that "hdm_ch" is never NULL.

regards,
dan carpenter

Re: [PATCH] staging: most: don't access hdm_ch before checking it valid

[Jing Xiangfeng]

On 2020/9/28 19:48, Dan Carpenter wrote:
> On Mon, Sep 28, 2020 at 06:48:38PM +0800, Jing Xiangfeng wrote:
>> In try_start_dim_transfer(), pointer hdm_ch is accessed before checking.
>> This may lead to a potential null pointer dereference. Fix this by
>> dereferencing hdm_ch after calling BUG_ON().
>>
>> Signed-off-by: Jing Xiangfeng <jingxiangfeng@huawei.com>
>> ---
>>   drivers/staging/most/dim2/dim2.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/staging/most/dim2/dim2.c b/drivers/staging/most/dim2/dim2.c
>> index 509c8012d20b..ccd7cc7545e4 100644
>> --- a/drivers/staging/most/dim2/dim2.c
>> +++ b/drivers/staging/most/dim2/dim2.c
>> @@ -148,7 +148,7 @@ void dimcb_on_error(u8 error_id, const char *error_message)
>>   static int try_start_dim_transfer(struct hdm_channel *hdm_ch)
>>   {
>>   	u16 buf_size;
>> -	struct list_head *head = &hdm_ch->pending_list;
> This is not a dereference, it's just pointer math.  In other words:
>
> 	struct list_head *head = hdm_ch + offsetof(struct hdm_channel, pending_list);
Thanks for correcting!

>
> So the commit message is wrong because this cannot lead to a NULL
> dereference.  It's better to just delete the BUG_ON().  We don't really
> like BUG_ON().  Checkpatch will complain about them.  An Oops gives
> basically the same information as a BUG_ON() without completely killing
> the kernel so just dereferencing a NULL is preferable.  Finally, we can
> see from the callers that "hdm_ch" is never NULL.
>
> regards,
> dan carpenter
>
> .
>