Re: [PATCH v2] rapidio: fix possible UAF when kfifo_alloc() fails

[wangweiyang]

Hi Andrew,

on 2022/11/22 6:04, Andrew Morton wrote:
> On Sat, 19 Nov 2022 12:03:35 +0800 Wang Weiyang <wangweiyang2@xxxxxxxxxx> wrote:
> 
>> If kfifo_alloc() fails in mport_cdev_open(), goto err_fifo and just free
>> priv. But priv is still in the chdev->file_list, then list traversal
>> may cause UAF. This fixes the following smatch warning:
>>
>> drivers/rapidio/devices/rio_mport_cdev.c:1930 mport_cdev_open() warn: '&priv->list' not removed from list
>>
>> ...
>>
>> Changes in v2:
>> - Avoid adding the new instance onto the list until the new instance
>>   has been fully initialized.
> 
> But it still isn't fully initialized.

Yes, I ignored that the initialization is still going on.

> 
>>  drivers/rapidio/devices/rio_mport_cdev.c | 7 +++----
>>  1 file changed, 3 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/rapidio/devices/rio_mport_cdev.c b/drivers/rapidio/devices/rio_mport_cdev.c
>> index 3cc83997a1f8..86b28c4cd906 100644
>> --- a/drivers/rapidio/devices/rio_mport_cdev.c
>> +++ b/drivers/rapidio/devices/rio_mport_cdev.c
>> @@ -1904,10 +1904,6 @@ static int mport_cdev_open(struct inode *inode, struct file *filp)
>>  
>>  	priv->md = chdev;
>>  
>> -	mutex_lock(&chdev->file_mutex);
>> -	list_add_tail(&priv->list, &chdev->file_list);
>> -	mutex_unlock(&chdev->file_mutex);
>> -
>>  	INIT_LIST_HEAD(&priv->db_filters);
>>  	INIT_LIST_HEAD(&priv->pw_filters);
>>  	spin_lock_init(&priv->fifo_lock);
>> @@ -1920,6 +1916,9 @@ static int mport_cdev_open(struct inode *inode, struct file *filp)
>>  		ret = -ENOMEM;
>>  		goto err_fifo;
>>  	}
>> +	mutex_lock(&chdev->file_mutex);
>> +	list_add_tail(&priv->list, &chdev->file_list);
>> +	mutex_unlock(&chdev->file_mutex);
>>  
>>  #ifdef CONFIG_RAPIDIO_DMA_ENGINE
>>  	INIT_LIST_HEAD(&priv->async_list);
> 
> Here we're still setting up state at *priv.  I'm thinking the list
> addition shouldn't occur until after the 
> 
> 	filp->private_data = priv;
> 
> Or just prior to it.  I'm not sure what the atomicity rules are for
> file.private versus mport_dev.file_list.

I prefer to put the list addition just prior to the

	filp->private_data = priv;

to keep mport_cdev_release() in reverse order of init. If it's ok with
you, I'll send a new patch later.

> .
>