[FFmpeg-devel] [PATCH 1/2] avcodec/cbs_h265_syntax_template: Better check for num_long_term_sps

[Mark Thompson]

On 21/11/2020 17:37, Michael Niedermayer wrote:
> On Sun, Nov 15, 2020 at 09:37:45PM +0000, Mark Thompson wrote:
>> On 14/11/2020 10:18, Michael Niedermayer wrote:
>>> Fixes: index 26 out of bounds for type 'uint8_t [16]'
>>> Fixes: 24913/clusterfuzz-testcase-minimized-ffmpeg_BSF_HEVC_METADATA_fuzzer-6261760693370880
>>>
>>> Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
>>> Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
>>> ---
>>>    libavcodec/cbs_h265_syntax_template.c | 2 ++
>>>    1 file changed, 2 insertions(+)
>>>
>>> diff --git a/libavcodec/cbs_h265_syntax_template.c b/libavcodec/cbs_h265_syntax_template.c
>>> index 48fae82d04..8eb6e159f4 100644
>>> --- a/libavcodec/cbs_h265_syntax_template.c
>>> +++ b/libavcodec/cbs_h265_syntax_template.c
>>> @@ -1405,6 +1405,8 @@ static int FUNC(slice_segment_header)(CodedBitstreamContext *ctx, RWContext *rw,
>>>                        infer(num_long_term_sps, 0);
>>>                        idx_size = 0;
>>>                    }
>>> +                if (HEVC_MAX_REFS < current->num_long_term_sps)
>>> +                    return AVERROR_INVALIDDATA;
>>
>> Please don't put isolated tests in the middle of the template.  If it constrains a value then add it to the constraints on that value.
>>
>>>                    ue(num_long_term_pics, 0, HEVC_MAX_REFS - current->num_long_term_sps);
>>>                    for (i = 0; i < current->num_long_term_sps +
>>>
>>
>> It would be good if the commit message could include an explanation derived from the standard, too.
>>
>> As far as I can tell the constraint doesn't appear explicitly, but the SPS is allowed to define more possible long term frames than are actually allowed to be present at any given moment so we need the tighter bound.
> 
> Is the change below what you had in mind ?
> 
> commit 72c6c46bb2b31b2822331aff461acccd0a4f9159 (HEAD -> master)
> Author: Michael Niedermayer <michael at niedermayer.cc>
> Date:   Fri Nov 13 23:15:52 2020 +0100
> 
>      avcodec/cbs_h265_syntax_template: Better check for num_long_term_sps
>      
>      As far as we can tell the constraint doesn't appear explicitly, but the SPS is allowed to
>      define more possible long term frames than are actually allowed to be present at any given moment so we need the tighter bound.

I meant write a commit message which explains where in the standard the constraint is coming from.  I wrote that because I didn't see any extra constraint written in the standard for num_long_term_sps but num_long_term_ref_pics_sps is indeed bigger, so presumably it must be implied by something else.

>      
>      Fixes: index 26 out of bounds for type 'uint8_t [16]'
>      Fixes: 24913/clusterfuzz-testcase-minimized-ffmpeg_BSF_HEVC_METADATA_fuzzer-6261760693370880
>      
>      Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
>      Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> 
> diff --git a/libavcodec/cbs_h265_syntax_template.c b/libavcodec/cbs_h265_syntax_template.c
> index 48fae82d04..30f2bd328c 100644
> --- a/libavcodec/cbs_h265_syntax_template.c
> +++ b/libavcodec/cbs_h265_syntax_template.c
> @@ -1399,7 +1399,7 @@ static int FUNC(slice_segment_header)(CodedBitstreamContext *ctx, RWContext *rw,
>                   unsigned int idx_size;
>   
>                   if (sps->num_long_term_ref_pics_sps > 0) {
> -                    ue(num_long_term_sps, 0, sps->num_long_term_ref_pics_sps);
> +                    ue(num_long_term_sps, 0, FFMIN(sps->num_long_term_ref_pics_sps, HEVC_MAX_REFS));

Yes to this part.

>                       idx_size = av_log2(sps->num_long_term_ref_pics_sps - 1) + 1;
>                   } else {
>                       infer(num_long_term_sps, 0);
> 
> [...]

- Mark