[FFmpeg-devel] [PATCH] avcodec/av1dec: Always set ret before goto end

[James Almer]

On 5/2/2024 6:05 AM, Andreas Rheinhardt wrote:
> Hendrik Leppkes:
>> On Thu, May 2, 2024 at 10:22 AM Andreas Rheinhardt
>> <andreas.rheinhardt at outlook.com> wrote:
>>>
>>> Before 0f8763fbea4e8816cd54c2a481d4c048fec58394, av1_frame_ref()
>>> and update_reference_list() could fail and therefore needed to
>>> be checked, which incidentally set ret. This is no longer happening,
>>> leading to a potential use of an uninitialized value which is
>>> also the subject of Coverity ticket #1596605.
>>>
>>> Fix this by always setting ret before goto end; do not return
>>> some random ancient value.
>>>
>>> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at outlook.com>
>>> ---
>>>   libavcodec/av1dec.c | 12 ++++++++++--
>>>   1 file changed, 10 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/libavcodec/av1dec.c b/libavcodec/av1dec.c
>>> index 79a30a114d..c3f255a29a 100644
>>> --- a/libavcodec/av1dec.c
>>> +++ b/libavcodec/av1dec.c
>>> @@ -1335,6 +1335,12 @@ static int av1_receive_frame_internal(AVCodecContext *avctx, AVFrame *frame)
>>>                       ret = set_output_frame(avctx, frame);
>>>                       if (ret < 0)
>>>                           av_log(avctx, AV_LOG_ERROR, "Set output frame error.\n");
>>> +                } else {
>>> +                    // CBS checks for us that the frame to be shown actually existed
>>> +                    // in the bitstream; if it doesn't it could be e.g. due to
>>> +                    // skip_frame setting. Return EAGAIN to indicate that we are
>>> +                    // currently unable to produce output.
>>> +                    ret = AVERROR(EAGAIN);
>>>                   }
>>>
>>
>> In the vein of this comment, set_output_frame will also return 0
>> without returning a frame in some cases - eg. with multiple layers.
>> Should this equally return EAGAIN rather than claiming success without
>> a frame?
>>
> 
> Thanks for pointing this out. There is a translation 0->AVERROR(EAGAIN)
> at the end of this function if the output frame is unset. Maybe this
> commit should rather set ret to 0 and rely on this (as it was before
> 0f8763fbea)? Let's hear what James says about this.

Yes, i prefer that.