[FFmpeg-devel] [PATCH 2/2] avcodec/g2meet: Check for end of input in jpg_decode_block()

[Tomas Härdin]

tor 2019-09-12 klockan 00:21 +0200 skrev Michael Niedermayer:
> On Wed, Sep 11, 2019 at 11:18:47PM +0200, Tomas Härdin wrote:
> > tis 2019-09-10 klockan 16:16 +0200 skrev Michael Niedermayer:
> > > On Mon, Sep 09, 2019 at 11:04:32PM +0200, Tomas Härdin wrote:
> > > > mån 2019-09-09 klockan 22:16 +0200 skrev Michael Niedermayer:
> > > > > Fixes: Timeout (100sec -> 0.7sec)
> > > > > Fixes: 8668/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_G2M_fuzzer-5174143888130048
> > > > > 
> > > > > 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/g2meet.c | 3 +++
> > > > >  1 file changed, 3 insertions(+)
> > > > > 
> > > > > diff --git a/libavcodec/g2meet.c b/libavcodec/g2meet.c
> > > > > index 19e1c130ce..731d29a5d4 100644
> > > > > --- a/libavcodec/g2meet.c
> > > > > +++ b/libavcodec/g2meet.c
> > > > > @@ -244,6 +244,9 @@ static int jpg_decode_block(JPGContext *c, GetBitContext *gb,
> > > > >      const int is_chroma = !!plane;
> > > > >      const uint8_t *qmat = is_chroma ? chroma_quant : luma_quant;
> > > > >  
> > > > > +    if (get_bits_left(gb) < 1)
> > > > > +        return AVERROR_INVALIDDATA;
> > > > > +
> > > > >      c->bdsp.clear_block(block);
> > > > >      dc = get_vlc2(gb, c->dc_vlc[is_chroma].table, 9, 3);
> > > > 
> > > > Why doesn't the VLC stuff have EOF handling? 
> > > 
> > > Because it doesnt need it in most of the cases and the get_vlc code
> > > is quite speed critical in some codecs.
> > > Also it would add a error return to cases that previously never
> > > could receive an error. That would require callers to be changed
> > > and check for this error in some cases
> > > 
> > > 
> > > > There's bound to be a
> > > > metric bajillion of cases like this strewn across the codebase..
> > > 
> > > if that was the case, the fuzzer would have likely found more cases.
> > That's only a matter of time. 
> 
> There is something wrong with this argument.
> Because one could say this about anything that is not static
> no matter if true or false. And theres no  practical way to 
> disproof it even if one waited, because maybe one just didnt wait
> long enough ...

You can actually prove this in many cases, which is something I've been
on about for a while now. The exception is formats that approach Turing
completeness, for obvious reasons. Even context-sensitive formats can
be tricky. The common case of nested length fields can take O(N²) to
parse and validate, if I understand the literature correctly. Pointer
based formats certainly do, like ZIP.

That is of course not really something we can do anything about,
formats are a given.

> [...]
> > I've said multiple times that worrying about these timeout things is
> > mostly a waste of time since any serious user will know to put time
> > limits on jobs. 
> 
> Everyone probably has timelimits on jobs but these timeouts are still
> a big problem. And i think this was discussed before ...
> 
> I think if you just think about what timeout to use for each case
> A. a web browser loading image, video and audio files

Presumably browser devs know how to use TBB and friends. They also
don't use g2meet, or cinepak, or anything else that isn't H.26* or VP*
etc. Closest thing is GIF

> > Resources would be better spent gearing the fuzzing
> > toward finding memory corruption issues, since the harm from them is
> > far more serious.
> 
> Then fixing the timeouts would be a priority as they hold the fuzzer
> up from finding other issues.
> Time spend waiting for a timeout is time the fuzzer cannot search for
> other issues

I see this more as a fuzzer tuning thing. When I last did fuzzing with
afl I certainly made sure to give it tiny samples and not a lot of time
per round

Question: is the fuzzer really allowed to spend 120 seconds on a test
case like this one? Or is that timing just an after-the-fact thing?

/Tomas