Re: [PATCH 1/1] drivers/usb/chipidea/debuc.c: avoid out of bound read

[Heinrich Schuchardt <xypron.glpk@gmx.de>]

On 22.04.2015 03:26, Peter Chen wrote:
> On Tue, Apr 21, 2015 at 09:25:41PM +0200, Heinrich Schuchardt wrote:
>> Hello Peter,
>>
>> thanks for reviewing.
>>
>> On 21.04.2015 03:32, Peter Chen wrote:
>>> On Fri, Apr 17, 2015 at 08:04:13AM +0200, Heinrich Schuchardt wrote:
>>>> A string written by the user may not be zero terminated.
>>>>
>>>> sscanf may read memory beyond the buffer if no zero byte
>>>> is found.
>>>>
>>>> Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
>>>> ---
>>>>  drivers/usb/chipidea/debug.c | 6 +++++-
>>>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/usb/chipidea/debug.c b/drivers/usb/chipidea/debug.c
>>>> index dfb05ed..ef08af3 100644
>>>> --- a/drivers/usb/chipidea/debug.c
>>>> +++ b/drivers/usb/chipidea/debug.c
>>>> @@ -88,9 +88,13 @@ static ssize_t ci_port_test_write(struct file *file, const char __user *ubuf,
>>>>  	char buf[32];
>>>>  	int ret;
>>>>  
>>>> -	if (copy_from_user(buf, ubuf, min_t(size_t, sizeof(buf) - 1, count)))
>>>> +	count = min_t(size_t, sizeof(buf) - 1, count);
>>>> +	if (copy_from_user(buf, ubuf, count))
>>>>  		return -EFAULT;
>>>
>>> Any reasons to change above?
>>
>> Otherwise we would have two lines with the term
>> min_t(size_t, sizeof(buf) - 1, count).
> 
> Sorry, two lines of min_t(..)? Why I can't find it?

Hello Peter,

in my patch I write:

count = min_t(size_t, sizeof(buf) - 1, count);
if (copy_from_user(buf, ubuf, count))
	return -EFAULT;

/* sscanf requires a zero terminated string */
buf[count] = 0;

Without the first part of the change I would have had to write:

if (copy_from_user(buf, ubuf, min_t(size_t, sizeof(buf) - 1, count)))
	return -EFAULT;

/* sscanf requires a zero terminated string */
buf[min_t(size_t, sizeof(buf) - 1, count)] = 0;

We should do the same calculation
"min_t(size_t, sizeof(buf) - 1, count)"
only once in the coding.

Best regards

Heinrich


> 
> 
>>
>> This would make the code harder to read.
>>
>>>>  
>>>> +	/* sscanf requires a zero terminated string */
>>>> +	buf[count] = 0;
>>>> +
>>>
>>> I prefer using '\0'
>>
>> If you confirm the rest of the patch is ok, I will send an updated patch.
>>
>> Best regards
>>
>> Heinrich
>>
>>>
>>>>  	if (sscanf(buf, "%u", &mode) != 1)
>>>>  		return -EINVAL;
>>>>  
>>
>