[FFmpeg-devel] [PATCH v2 1/3] lavc/hevc_mp4toannexb: Fix integer overflow

[Andriy Gelman]

On Tue, 03. Dec 21:42, Andriy Gelman wrote:
> On Wed, 04. Dec 01:26, Andreas Rheinhardt wrote:
> > On Wed, Dec 4, 2019 at 1:04 AM Andriy Gelman <andriy.gelman at gmail.com>
> > wrote:
> > 
> > > From: Andriy Gelman <andriy.gelman at gmail.com>
> > >
> > > Check packet grow size against INT_MAX instead of SIZE_MAX.
> > >
> > > Found with libFuzzer:
> > > 4294967044 cannot be represented as int.
> > >
> > > Signed-off-by: Andriy Gelman <andriy.gelman at gmail.com>
> > > ---
> > >  libavcodec/hevc_mp4toannexb_bsf.c | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/libavcodec/hevc_mp4toannexb_bsf.c
> > > b/libavcodec/hevc_mp4toannexb_bsf.c
> > > index 09bce5b34c2..bf4c25b3b7b 100644
> > > --- a/libavcodec/hevc_mp4toannexb_bsf.c
> > > +++ b/libavcodec/hevc_mp4toannexb_bsf.c
> > > @@ -66,7 +66,7 @@ static int hevc_extradata_to_annexb(AVBSFContext *ctx)
> > >          for (j = 0; j < cnt; j++) {
> > >              int nalu_len = bytestream2_get_be16(&gb);
> > >
> > > -            if (4 + AV_INPUT_BUFFER_PADDING_SIZE + nalu_len > SIZE_MAX -
> > > new_extradata_size) {
> > > +            if (4 + AV_INPUT_BUFFER_PADDING_SIZE + nalu_len > INT_MAX -
> > > new_extradata_size) {
> > >
> 
> > 
> > There is actually FF_MAX_EXTRADATA_SIZE constant to limit the size of
> > extradata (it is chosen so small that one can parse the extradata via a
> > GetBitContext). (But honestly, it is not used consistently.)
> 
> If we decide to use it should be a separate commit IMO.
> 
> > 
> > 
> > >                  ret = AVERROR_INVALIDDATA;
> > >                  goto fail;
> > >              }
> > > @@ -152,8 +152,8 @@ static int hevc_mp4toannexb_filter(AVBSFContext *ctx,
> > > AVPacket *out)
> > >          extra_size    = add_extradata * ctx->par_out->extradata_size;
> > >          got_irap     |= is_irap;
> > >
> > > -        if (SIZE_MAX - nalu_size < 4 ||
> > > -            SIZE_MAX - 4 - nalu_size < extra_size) {
> > > +        if (INT_MAX < 4 + (uint64_t)nalu_size ||
> > >
> > 
> > The cast is unnecessary: INT_MAX - 4 < nalu_size does the same. And INT_MAX
> > - 4 can even be computed at compile-time.
> > 
> > 
> 
> > > +            INT_MAX - 4 < extra_size + (uint64_t)nalu_size) {
> > >
> > 
> > If you check that the extradata is not too big (namely <= INT_MAX - 4),
> > then you can simplify this to INT_MAX - 4 - extra_size < nalu_size. The
> > left side is positive, so it does not wrap around when converting it to
> > uint32_t. This makes the first check superfluous.
> 
> Yes, that's a good idea. And I think that extra_size <= INT_MAX - 4 always holds
> because we pad extradata with AV_INPUT_PADDING_SIZE (and the checks are done in
> the initialization). But I'll double check.

The above check also needs to take into account the current packet size, i.e.: 

INT_MAX - 4 - extra_size < (uint64_t)nalu_size + out->size

I don't think it's possible to get rid of the cast on the rhs then.

Thanks,
-- 
Andriy