[FFmpeg-devel] [PATCH] apng: use correct size for output buffer

[Paul B Mahol]

On 11/6/15, Andreas Cadhalpun <andreas.cadhalpun at googlemail.com> wrote:
> On 06.11.2015 22:29, wm4 wrote:
>> On Fri, 6 Nov 2015 22:18:04 +0100
>> Andreas Cadhalpun <andreas.cadhalpun at googlemail.com> wrote:
>>
>>> This fixes a stack buffer overflow.
>>>
>>> Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun at googlemail.com>
>>> ---
>>>  libavcodec/pngdec.c | 3 ++-
>>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/libavcodec/pngdec.c b/libavcodec/pngdec.c
>>> index 689aa2b..c974654 100644
>>> --- a/libavcodec/pngdec.c
>>> +++ b/libavcodec/pngdec.c
>>> @@ -1010,13 +1010,13 @@ static int handle_p_frame_apng(AVCodecContext
>>> *avctx, PNGDecContext *s,
>>>              memcpy(buffer + row_start, p->data[0] + row_start, s->bpp *
>>> s->cur_w);
>>>          }
>>>      } else { // APNG_BLEND_OP_OVER
>>> +        uint8_t *output = av_malloc(s->bpp);
>>>          for (y = s->y_offset; y < s->y_offset + s->cur_h; ++y) {
>>>              uint8_t *foreground = p->data[0] + s->image_linesize * y +
>>> s->bpp * s->x_offset;
>>>              uint8_t *background = buffer + s->image_linesize * y +
>>> s->bpp * s->x_offset;
>>>              for (x = s->x_offset; x < s->x_offset + s->cur_w; ++x,
>>> foreground += s->bpp, background += s->bpp) {
>>>                  size_t b;
>>>                  uint8_t foreground_alpha, background_alpha,
>>> output_alpha;
>>> -                uint8_t output[4];
>>>
>>>                  // Since we might be blending alpha onto alpha, we use
>>> the following equations:
>>>                  // output_alpha = foreground_alpha + (1 -
>>> foreground_alpha) * background_alpha
>>> @@ -1069,6 +1069,7 @@ static int handle_p_frame_apng(AVCodecContext
>>> *avctx, PNGDecContext *s,
>>>                  memcpy(background, output, s->bpp);
>>>              }
>>>          }
>>> +        av_freep(&output);
>>>      }
>>>
>>>      // Copy blended buffer into the frame and free
>>
>> This seems wasteful, can't it just be output[8]?
>
> I think s->bpp can be up to 10:
>     size_t byte_depth = s->bit_depth > 8 ? 2 : 1;                  //
> maximal 2
> ...
>         s->channels       = ff_png_get_nb_channels(s->color_type); //
> maximal 4
>         s->bits_per_pixel = s->bit_depth * s->channels;            //
> bit_depth is maximal 16
>         s->bpp            = (s->bits_per_pixel + 7) >> 3;          //
> maximal 8
> ...
>         if (s->has_trns && s->color_type != PNG_COLOR_TYPE_PALETTE) {
> ...
>             s->bpp += byte_depth;                                  //
> maximal 10
>
>> It also adds a bug (unchecked malloc).
>
> Right, sorry for that.
> Attached is a patch increasing the buffer size to 10 and
> adding an assert that s->bpp is not larger.

Should be fine.

>
> Best regards,
> Andreas
>
>