Hi Souptick,

On Thu, Feb 18, 2016 at 12:49 AM, Souptick Joarder <[email protected]> wrote:
> Hi Julian,
>
> On Wed, Feb 17, 2016 at 5:38 AM, Julian Calaby <[email protected]> 
> wrote:
>> Hi Souptick,
>>
>> On Tue, Feb 16, 2016 at 5:36 PM, Souptick Joarder <[email protected]> 
>> wrote:
>>> On Tue, Feb 16, 2016 at 12:34 AM, Alan <[email protected]> wrote:
>>>> The buffer needs to be zero terminated in case the user data is not.
>>>> Otherwise we run off the end of the buffer.
>>>>
>>>> Signed-off-by: Alan Cox <[email protected]>
>>>> ---
>>>>  drivers/net/wireless/ralink/rt2x00/rt2x00debug.c |    5 +++--
>>>>  1 file changed, 3 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00debug.c 
>>>> b/drivers/net/wireless/ralink/rt2x00/rt2x00debug.c
>>>> index 25ee3cb..72ae530 100644
>>>> --- a/drivers/net/wireless/ralink/rt2x00/rt2x00debug.c
>>>> +++ b/drivers/net/wireless/ralink/rt2x00/rt2x00debug.c
>>>> @@ -478,7 +478,7 @@ static ssize_t rt2x00debug_write_##__name(struct file 
>>>> *file,        \
>>>>  {                                                              \
>>>>         struct rt2x00debug_intf *intf = file->private_data;     \
>>>>         const struct rt2x00debug *debug = intf->debug;          \
>>>> -       char line[16];                                          \
>>>> +       char line[17];                                          \
>>>>         size_t size;                                            \
>>>>         unsigned int index = intf->offset_##__name;             \
>>>>         __type value;                                           \
>>>> @@ -494,7 +494,8 @@ static ssize_t rt2x00debug_write_##__name(struct file 
>>>> *file,        \
>>>>                                                                 \
>>>>         if (copy_from_user(line, buf, length))                  \
>>>>                 return -EFAULT;                                 \
>>>> -                                                               \
>>>> +       line[16] = 0;
>>>
>>>             line[length] = '\0';
>>>             correct me if I am wrong.                          \
>>
>> I believe that in this case the data in buf will already be null
>> terminated, so ensuring that line is null terminated is only needed if
>> there are exactly 16 bytes in buf. IMHO
>>
>> line[16] = 0;
>
> I think, if there are 16 bytes in buf, we end up loosing 1 byte data.
> If there are 15 bytes in buf, buf is already be null terminated.
> So is this really required?
> Correct me if I am wrong.

Counting in C starts at 0.

We started out with line being a 16 character string, so it initially
looked like this:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 <- index
_ _ _ _ _ _ _ _ _ _ __ __ __ __ __ __ <- data, i.e. line[index]

If we are given a 4 byte null terminated string, we end up with:

0 1 2 3  4 5 6 7 8 9 10 11 12 13 14 15
a b c d \0 _ _ _ _ _ __ __ __ __ __ __

If we are given a 15 byte null terminated string, we end up with:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
a b c d e f g h i j  k  l  m  n  o \0

If we are given a 16 byte null terminated string, we end up with:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
a b c d e f g h i j  k  l  m  n  o  p

Note that it is not null terminated. The call to strlen() will now run
off into memory that doesn't belong to line and potentially cause a
crash. This is a bug.

Alan fixed this by adding another character to the string, so it's now
17 characters long and line[16] will always be set to 0 after
copy_from_user() is called. So now line initially looks like:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
_ _ _ _ _ _ _ _ _ _ __ __ __ __ __ __ __

If we are given a 4 byte null terminated string, we end up with:

0 1 2 3  4 5 6 7 8 9 10 11 12 13 14 15 16
a b c d \0 _ _ _ _ _ __ __ __ __ __ __ \0

If we are given a 15 byte null terminated string, we end up with:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
a b c d e f g h i j  k  l  m  n  o \0 \0

If we are given a 16 byte null terminated string, we end up with:

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
a b c d e f g h i j  k  l  m  n  o  p \0

Note that the final null terminator is set by the line[16] = 0; line,
not copy_from_user().

As line is now null terminated, strlen() will behave correctly and
there is no bug.

There are a few different ways to fix this bug. All of them will
involve increasing the size of line to 17, and the exact details of
how we ensure the null terminator is set can vary. Alan has chosen
what is likely to be the smallest option in compiled code size, the
fastest, and one of the most obvious options.

Understanding how arrays and strings work in C is essential knowledge
if you're going to submit or review patches for the Linux kernel.
Could you please refresh your knowledge on those topics before
submitting or reviewing any more patches?

Thanks,

-- 
Julian Calaby

Email: [email protected]
Profile: http://www.google.com/profiles/julian.calaby/
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to