On 01.02.24 11:24, Stefan Kerkmann wrote:
> Hello Ahmad,
>
> On 31.01.24 18:21, Ahmad Fatoum wrote:
>> On 31.01.24 17:57, Stefan Kerkmann wrote:
>>> OOB writes can be triggered when fdt->dt_size * 2 is still smaller than
>>> the property for which memory should be allocated. This can happen under
>>> rare circumstances when editing a fdt with the of_property command and a
>>> property is larger than 128k in size.
>>>
>>> This happend when editing a FIT image (which is a ftd) with the
>>> of_property command and the Kernel image was around 8M in size.
>>>
>>> The simplified call chain is the following:
>>>
>>> of_property -> of_flatten_dtb -> create new fdt with 64k in size (this is
>>> fixed) -> __of_flatten_dtb -> attempt to copy kernel image (8M) ->
>>> fdt_ensure_space -> allocate only 128k for fdt->dt -> memcopy 8M into fdt->dt
>>> buffer -> crash
>>>
>>> The fix is to grow fdt->dt to hold at least the new property. The power
>>> of 2 increment is untouched to keep the same behaviour otherwise.
>>>
>>> Signed-off-by: Stefan Kerkmann <s.kerkmann@xxxxxxxxxxxxxx>
>>> ---
>>> drivers/of/fdt.c | 4 ++++
>>> 1 file changed, 4 insertions(+)
>>>
>>> diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
>>> index 4f79a6120f..1f24ed0bbc 100644
>>> --- a/drivers/of/fdt.c
>>> +++ b/drivers/of/fdt.c
>>> @@ -399,6 +399,10 @@ static int fdt_ensure_space(struct fdt *fdt, int dtsize)
>>> previous = fdt->dt;
>>> new_size = fdt->dt_size * 2;
>>> + while (new_size <= dtsize) {
>>> + new_size *= 2;
>>> + }
>>
>> A nitpick that I solely note because I already had feedback on the first patch:
>> Kernel coding style is to omit { braces } for single statement blocks.
>>
>> In your case you could just do:
>>
>> if (new_size <= dtsize)
>> new_size = roundup_pow_of_two(new_size + dtsize);
>>
>> I think to skip the loop.
>>
>
> Thanks! That is the better solution.
>
> To not over provision memory I changed the new size to be `roundup_pow_of_two(fdt->dt_size + dtsize)` as we know for sure that `dtsize` is already larger than `fdt->dt_size * 2`.
>
> In (made up) case that we already have 8k space for the fdt and got a 17k property we would allocate 65k (8k + 8k + 17k = 33k ⇾ rounded ⇾ 65k) and only 32k (8k + 17k = 25k → rounded → 32k) with `fdt->dt_size + dtsize`.
Sounds good.
>
>> Cheers,
>> Ahmad
>>
>>> +
>>> if ((fdt->dt = memalign_realloc(previous, fdt->dt_size,
>>> new_size)) == NULL) {
>>> free(previous);
>>>
>>
>
> Cheers
> Stefan
>
--
Pengutronix e.K. | |
Steuerwalder Str. 21 | http://www.pengutronix.de/ |
31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
