[FFmpeg-devel] [PATCH] avformat/mxfdec: Check index_duration

[Marton Balint]

On Tue, 27 Dec 2022, Marton Balint wrote:

>
>
> On Tue, 27 Dec 2022, Michael Niedermayer wrote:
>
>>  On Tue, Dec 27, 2022 at 07:05:44PM +0100, Marton Balint wrote:
>>> 
>>>
>>>  On Mon, 26 Dec 2022, Michael Niedermayer wrote:
>>>
>>>>  On Mon, Dec 26, 2022 at 11:32:48AM +0100, Tomas Härdin wrote:
>>>>>  lör 2022-12-24 klockan 23:50 +0100 skrev Michael Niedermayer:
>>>>>>
>>>>>>           index_table->nb_ptses += s->index_duration;
>>>>>>  +        // If index_duration is substantially larger than
>>>>>>  nb_index_entries then this algorithm which
>>>>>>  +        // allocates index_duration elements is a bad idea. All
>>>>>>  files i tried have it equal
>>>>>>  +        if (s->index_duration > 10LL * s->nb_index_entries)
>>>>>>  +            return AVERROR_PATCHWELCOME;
>>>>>
>>>>>  I was going to say this can overflow but the 10LL ensures it can't. So
>>>>>  looks OK.
>>>>
>>>>  will apply
>>>
>>>  Please don't, as far as I see this disallows the usage of partial index
>>>  tables, so practically rejecting valid files, which is not OK.
>>
>>  can you share a file that would break ?
>
> I don't have such file. But the MXF specs (SMPTE 377-1-2009) explictly 
> defines the concept of partial index tables:
>
> "Where all Index Table segments are contiguous, or there is only one segment, 
> but not all Edit Units in the Essence Container are indexed, these tables are 
> called Partial Index Tables."
>
> As far as I see here nb_index_entries is corresponding to the number of 
> indexed edit units, and the number is allowed to be smaller than the index 
> duration, because not all edit units have to be indexed.

I read the specs again, and it seems that I misread it the first time, 
because partial index tables mean that the index segments have no gaps 
between them, but the index still not cover the whole essence. So it is 
not referring to the index entries in the segment.

So, in principal your patch *might* be OK. However, existing code simply 
ignores a corrupt index table, does not reject it. I kind of prefer we 
make the check more strict, but gracefully allow corrupted index by 
ignoring it fully.

I will post a follow up patch series.

Regards,
Marton