Re: [PATCH] nfc: st-nci: array index overflow in st_nci_se_get_bwi()

[Krzysztof Kozlowski]

On 19/12/2022 16:41, Alexander Duyck wrote:
> On Mon, Dec 19, 2022 at 1:06 AM Krzysztof Kozlowski
> <krzysztof.kozlowski@xxxxxxxxxx> wrote:
>>
>> On 14/12/2022 19:35, Alexander H Duyck wrote:
>>> On Tue, 2022-12-13 at 09:12 -0500, Aleksandr Burakov wrote:
>>>> Index of info->se_info.atr can be overflow due to unchecked increment
>>>> in the loop "for". The patch checks the value of current array index
>>>> and doesn't permit increment in case of the index is equal to
>>>> ST_NCI_ESE_MAX_LENGTH - 1.
>>>>
>>>> Found by Linux Verification Center (linuxtesting.org) with SVACE.
>>>>
>>>> Fixes: ed06aeefdac3 ("nfc: st-nci: Rename st21nfcb to st-nci")
>>>> Signed-off-by: Aleksandr Burakov <a.burakov@xxxxxxxxxxxx>
>>>> ---
>>>>  drivers/nfc/st-nci/se.c | 5 +++--
>>>>  1 file changed, 3 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/nfc/st-nci/se.c b/drivers/nfc/st-nci/se.c
>>>> index ec87dd21e054..ff8ac1784880 100644
>>>> --- a/drivers/nfc/st-nci/se.c
>>>> +++ b/drivers/nfc/st-nci/se.c
>>>> @@ -119,10 +119,11 @@ static u8 st_nci_se_get_bwi(struct nci_dev *ndev)
>>>>      /* Bits 8 to 5 of the first TB for T=1 encode BWI from zero to nine */
>>>>      for (i = 1; i < ST_NCI_ESE_MAX_LENGTH; i++) {
>>>>              td = ST_NCI_ATR_GET_Y_FROM_TD(info->se_info.atr[i]);
>>>> -            if (ST_NCI_ATR_TA_PRESENT(td))
>>>> +            if (ST_NCI_ATR_TA_PRESENT(td) && i < ST_NCI_ESE_MAX_LENGTH - 1)
>>>>                      i++;
>>>>              if (ST_NCI_ATR_TB_PRESENT(td)) {
>>>> -                    i++;
>>>> +                    if (i < ST_NCI_ESE_MAX_LENGTH - 1)
>>>> +                            i++;
>>>>                      return info->se_info.atr[i] >> 4;
>>>>              }
>>>>      }
>>>
>>> Rather than adding 2 checks you could do this all with one check.
>>> Basically you would just need to replace:
>>>   if (ST_NCI_ATR_TB_PRESENT(td)) {
>>>       i++;
>>>
>>> with:
>>>   if (ST_NCI_ATR_TB_PRESENT(td) && ++i < ST_NCI_ESE_MAX_LENGTH)
>>>
>>> Basically it is fine to increment "i" as long as it isn't being used as
>>> an index so just restricting the last access so that we don't
>>> dereference using it as an index should be enough.
>>
>> These are different checks - TA and TB. By skipping TA, your code is not
>> equivalent. Was it intended?
> 
> Sorry, I wasn't talking about combining the TA and TB checks. I was
> talking about combining the TB check and the bounds check so that you
> didn't return and se_info_atr for a value that may not have actually
> aligned due to the fact you had overflowed. Specifically, is skipping
> the i++ the correct response to going out of bounds? I'm wondering if
> you should be returning the default instead in the case of overflow?
> 
> The TA check could be modified so that it checks for "++i =
> ST_NCI_ESE_MAX_LENGTH" and if that is true break rather than continue
> in the loop.

Ah, right. From that point of view, the first check (TA) also does not
look correct or equivalent. If we reached end of
ST_NCI_ESE_MAX_LENGTH(), we should not check TB on that entry. I would
propose to end the loop at that stage.

Best regards,
Krzysztof