[FFmpeg-devel] [PATCH 1/5] avcodec/hevc_parse: check for parameter set decoding failure

[Hendrik Leppkes]

On Wed, Apr 5, 2017 at 11:40 PM, James Almer <jamrial at gmail.com> wrote:
> On 4/3/2017 10:46 AM, James Almer wrote:
>> On 4/3/2017 7:00 AM, Michael Niedermayer wrote:
>>> On Sun, Apr 02, 2017 at 10:45:41PM -0300, James Almer wrote:
>>>> Signed-off-by: James Almer <jamrial at gmail.com>
>>>> ---
>>>>  libavcodec/hevc_parse.c | 32 +++++++++++++++++++++++++-------
>>>>  1 file changed, 25 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/libavcodec/hevc_parse.c b/libavcodec/hevc_parse.c
>>>> index 6c1138e015..028ca5afe5 100644
>>>> --- a/libavcodec/hevc_parse.c
>>>> +++ b/libavcodec/hevc_parse.c
>>>> @@ -22,7 +22,8 @@
>>>>  #include "hevc_parse.h"
>>>>
>>>>  static int hevc_decode_nal_units(const uint8_t *buf, int buf_size, HEVCParamSets *ps,
>>>> -                                 int is_nalff, int nal_length_size, void *logctx)
>>>> +                                 int is_nalff, int nal_length_size, int err_recognition,
>>>> +                                 void *logctx)
>>>>  {
>>>>      int i;
>>>>      int ret = 0;
>>>> @@ -38,9 +39,21 @@ static int hevc_decode_nal_units(const uint8_t *buf, int buf_size, HEVCParamSets
>>>>
>>>>          /* ignore everything except parameter sets and VCL NALUs */
>>>>          switch (nal->type) {
>>>> -        case HEVC_NAL_VPS: ff_hevc_decode_nal_vps(&nal->gb, logctx, ps);    break;
>>>> -        case HEVC_NAL_SPS: ff_hevc_decode_nal_sps(&nal->gb, logctx, ps, 1); break;
>>>> -        case HEVC_NAL_PPS: ff_hevc_decode_nal_pps(&nal->gb, logctx, ps);    break;
>>>> +        case HEVC_NAL_VPS:
>>>> +            ret = ff_hevc_decode_nal_vps(&nal->gb, logctx, ps);
>>>> +            if (ret < 0)
>>>> +                goto done;
>>>> +            break;
>>>> +        case HEVC_NAL_SPS:
>>>> +            ret = ff_hevc_decode_nal_sps(&nal->gb, logctx, ps, 1);
>>>> +            if (ret < 0)
>>>> +                goto done;
>>>> +            break;
>>>> +        case HEVC_NAL_PPS:
>>>> +            ret = ff_hevc_decode_nal_pps(&nal->gb, logctx, ps);
>>>> +            if (ret < 0)
>>>> +                goto done;
>>>> +            break;
>>>>          case HEVC_NAL_TRAIL_R:
>>>>          case HEVC_NAL_TRAIL_N:
>>>>          case HEVC_NAL_TSA_N:
>>>
>>> I didnt investigate how exactly this is used but from just the patch
>>> this seems not optimal
>>> For example, if you have 3 PPS NALs and the first fails to decode
>>> you might still be able to fully decode the other 2
>>
>> I'm mimicking the behavior of decode_nal_unit() in hevcdec.c, which is
>> currently used during frame decoding and extradata decoding.
>> This patchset aims at removing duplicate code while keeping the hevc
>> decoder behaving the same as it was before. I could skip this change
>> if that's preferred, but if this behavior is not optimal then someone
>> who better understands the codec should look at it.
>
> To add some context, the functions in hevc_parse.c are currently used
> only by the mediacodec based hevc decoder to decode extradata, and it's
> a duplicate of existing functionality in hevcdec.c used by the internal
> hevc decoder.
>
> This set aims at putting the hevc_parse.c version on par with the
> hevcdec.c one (in the scope of extradata parsing, not frame) to share it
> between the two decoders and any other that may need it in the future.
> This patch checks for PS failures and aborts if err_recog is set to
> explode. The second makes apply_defdispwin user defined instead of
> having it hardcoded to 1. The third avoids aborting on NALs not needed
> or expected in extradata, and the last two patches make hevcdec.c use
> ff_hevc_decode_extradata().
>
> Not aborting on PS NAL failures here would technically mean a change in
> behavior on the internal hevc decoder once patch five is applied, and
> I'd rather no do that as part of this set.
>

Keeping the current behavior for the HEVC software decoder seems
ideal. If a change in behavior is wanted afterwards, it should be
dealt with in a separate change, and not this refactoring.

- Hendrik