[FFmpeg-devel] [PATCH 3/3] avformat/mxfdec: Reduce overflows in essence_length computation

[Tomas Härdin]

fre 2022-12-23 klockan 00:25 +0100 skrev Michael Niedermayer:
> On Wed, Nov 16, 2022 at 12:49:10PM +0100, Tomas Härdin wrote:
> > sön 2022-11-13 klockan 00:44 +0100 skrev Michael Niedermayer:
> > > Fixes: signed integer overflow: -3741319169 - 9223372036823449370
> > > cannot be represented in type 'long'
> > > Fixes: 51896/clusterfuzz-testcase-minimized-
> > > ffmpeg_dem_MXF_fuzzer-
> > > 513039428681728
> > > 
> > > 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/mxfdec.c | 10 +++++++---
> > >  1 file changed, 7 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/libavformat/mxfdec.c b/libavformat/mxfdec.c
> > > index e6118e141d..42109cb43a 100644
> > > --- a/libavformat/mxfdec.c
> > > +++ b/libavformat/mxfdec.c
> > > @@ -100,7 +100,7 @@ typedef struct MXFPartition {
> > >      uint64_t previous_partition;
> > >      int index_sid;
> > >      int body_sid;
> > > -    int64_t this_partition;
> > > +    uint64_t this_partition;
> > >      int64_t essence_offset;         ///< absolute offset of
> > > essence
> > >      int64_t essence_length;
> > >      int32_t kag_size;
> > > @@ -3519,8 +3519,12 @@ static void
> > > mxf_compute_essence_containers(AVFormatContext *s)
> > >              p->essence_offset = p->first_essence_klv.offset;
> > >  
> > >              /* essence container spans to the next partition */
> > > -            if (x < mxf->partitions_count - 1)
> > > -                p->essence_length = mxf-
> > > > partitions[x+1].this_partition - p->essence_offset;
> > > +            if (x < mxf->partitions_count - 1) {
> > > +                if (mxf->partitions[x+1].this_partition < p-
> > > > essence_offset) {
> > > +                    p->essence_length = -1;
> > > +                } else
> > > +                    p->essence_length = mxf-
> > > > partitions[x+1].this_partition - p->essence_offset;
> > > +            }
> > 
> > A better solution might be to record the actual offset of the
> > partitions rather than relying on ThisPartition. Then we can
> > guarantee
> > that they are strictly increasing.
> 
> do you mean that pack_ofs could be used here ?

this_partition should be == pack_ofs - run_in. I think we can just nuke
this_partition in the struct and use pack_ofs - run_in everywhere. Emit
a warning or bail out if any partition does not obey this contraint,
since such files have likely been written by bad muxers.


> but as pack_ofs is signed that either means dealing with potentially
> negative
> or changing it to uint64_t or adding a new field and avio_tell()
> somewhere?

pack_ofs can only be negative in case of error, in which case we should
bail out anyway. Something must be very wrong if lavf can't avio_tell()

For every partition, pack_ofs >= run_in. this_partition >= 0, and in
particular this_partition == 0 for the very first partition.

This is reminding me that I had some finer points to fix related to
run_in and deleting the superfluous sync stuff that's currently in the
code. Baptiste is fine with that, as it's a relic from when mxfdec
didn't have proper seek support.

/Tomas