[FFmpeg-devel] [PATCH 1/6] avformat/avidec: support huge durations

[Tomas Härdin]

mån 2023-10-02 klockan 21:03 +0200 skrev Michael Niedermayer:
> On Mon, Oct 02, 2023 at 11:07:47AM +0200, Anton Khirnov wrote:
> > Quoting Michael Niedermayer (2023-10-01 00:28:56)
> > > On Sat, Sep 30, 2023 at 10:18:38PM +0200, Anton Khirnov wrote:
> > > > Quoting Michael Niedermayer (2023-09-30 16:31:43)
> > > > > On Sat, Sep 30, 2023 at 04:04:03PM +0200, Michael Niedermayer
> > > > > wrote:
> > > > > > On Sat, Sep 30, 2023 at 11:35:19AM +0200, Marton Balint
> > > > > > wrote:
> > > > > > > 
> > > > > > > 
> > > > > > > On Sat, 30 Sep 2023, Michael Niedermayer wrote:
> > > > > > > 
> > > > > > > > Fixes: signed integer overflow: 109817402400 *
> > > > > > > > 301990077 cannot be represented in type 'long long'
> > > > > > > > Fixes: 51896/clusterfuzz-testcase-minimized-
> > > > > > > > ffmpeg_dem_AVI_fuzzer-6706191715139584
> > > > > > > > 
> > > > > > > > Found-by: continuous fuzzing process
> > > > > > > > https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
> > > > > > > > Signed-off-by: Michael Niedermayer
> > > > > > > > <michael at niedermayer.cc>
> > > > > > > > ---
> > > > > > > > libavformat/avidec.c | 12 ++++++++++--
> > > > > > > > 1 file changed, 10 insertions(+), 2 deletions(-)
> > > > > > > > 
> > > > > > > > diff --git a/libavformat/avidec.c
> > > > > > > > b/libavformat/avidec.c
> > > > > > > > index 00bd7a98a9d..cfc693842b7 100644
> > > > > > > > --- a/libavformat/avidec.c
> > > > > > > > +++ b/libavformat/avidec.c
> > > > > > > > @@ -27,6 +27,7 @@
> > > > > > > > #include "libavutil/avstring.h"
> > > > > > > > #include "libavutil/opt.h"
> > > > > > > > #include "libavutil/dict.h"
> > > > > > > > +#include "libavutil/integer.h"
> > > > > > > > #include "libavutil/internal.h"
> > > > > > > > #include "libavutil/intreadwrite.h"
> > > > > > > > #include "libavutil/mathematics.h"
> > > > > > > > @@ -476,7 +477,7 @@ static int
> > > > > > > > calculate_bitrate(AVFormatContext *s)
> > > > > > > >         AVStream *st = s->streams[i];
> > > > > > > >         FFStream *const sti = ffstream(st);
> > > > > > > >         int64_t duration;
> > > > > > > > -        int64_t bitrate;
> > > > > > > > +        int64_t bitrate = 0;
> > > > > > > > 
> > > > > > > >         for (j = 0; j < sti->nb_index_entries; j++)
> > > > > > > >             len += sti->index_entries[j].size;
> > > > > > > > @@ -484,7 +485,14 @@ static int
> > > > > > > > calculate_bitrate(AVFormatContext *s)
> > > > > > > >         if (sti->nb_index_entries < 2 || st->codecpar-
> > > > > > > > >bit_rate > 0)
> > > > > > > >             continue;
> > > > > > > >         duration = sti->index_entries[j-1].timestamp -
> > > > > > > > sti->index_entries[0].timestamp;
> > > > > > > > -        bitrate = av_rescale(8*len, st->time_base.den,
> > > > > > > > duration * st->time_base.num);
> > > > > > > > +        if (INT64_MAX / duration >= st->time_base.num)
> > > > > > > > {
> > > > > > > > +            bitrate = av_rescale(8*len, st-
> > > > > > > > >time_base.den, duration * st->time_base.num);
> > > > > > > 
> > > > > > > Why not always use the AVInteger version? This is not
> > > > > > > performance sensitive
> > > > > > > as far as I see.
> > > > > > 
> > > > > > We can, i will have to fix the rounding though so it
> > > > > > matches av_rescale()
> > > > > 
> > > > > will apply this with just AVInteger and fixed rounding with
> > > > > my next push probably
> > > > 
> > > > This seems MUCH less readable to me.
> > > 
> > > we can add a av_rescale_2den()
> > 
> > Won't av_rescale_q(len, (AVRational){8, duration}, st->time_base)
> > achieve the same effect?
> 
> duration is 64bit AVRational is 32/32 bit, so i would expect that to
> not
> work.
> If duration was fitting in 32bit then duration * st->time_base.num
> would not
> have overflowed

Maybe we need a 64/64-bit version of AVRational..

/Tomas