[FFmpeg-devel] [PATCH] avcodec/hevcdec: do not let updated extradata corrupt state
Michael Niedermayer
michael at niedermayer.cc
Thu Jul 6 17:29:21 EEST 2017
On Thu, Jul 06, 2017 at 12:35:28PM +0200, wm4 wrote:
> On Thu, 6 Jul 2017 11:58:47 +0200
> Michael Niedermayer <michael at niedermayer.cc> wrote:
>
> > On Wed, Jul 05, 2017 at 04:39:05PM +0200, wm4 wrote:
> > > On Wed, 5 Jul 2017 16:08:38 +0200
> > > Michael Niedermayer <michael at niedermayer.cc> wrote:
> > >
> > > > On Wed, Jul 05, 2017 at 09:56:10AM +0200, wm4 wrote:
> > > > > On Tue, 4 Jul 2017 22:33:52 +0200
> > > > > Michael Niedermayer <michael at niedermayer.cc> wrote:
> > > > >
> > > > > > Fixes: out of array access
> > > > > > Fixes: 2451/clusterfuzz-testcase-minimized-4781613957251072
> > > > > >
> > > > > > 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/hevcdec.c | 8 ++++----
> > > > > > 1 file changed, 4 insertions(+), 4 deletions(-)
> > > > > >
> > > > > > diff --git a/libavcodec/hevcdec.c b/libavcodec/hevcdec.c
> > > > > > index cc8ac82164..55f51211c3 100644
> > > > > > --- a/libavcodec/hevcdec.c
> > > > > > +++ b/libavcodec/hevcdec.c
> > > > > > @@ -3057,7 +3057,7 @@ static int verify_md5(HEVCContext *s, AVFrame *frame)
> > > > > > return 0;
> > > > > > }
> > > > > >
> > > > > > -static int hevc_decode_extradata(HEVCContext *s, uint8_t *buf, int length)
> > > > > > +static int hevc_decode_extradata(HEVCContext *s, uint8_t *buf, int length, int first)
> > > > > > {
> > > > > > int ret, i;
> > > > > >
> > > > > > @@ -3069,7 +3069,7 @@ static int hevc_decode_extradata(HEVCContext *s, uint8_t *buf, int length)
> > > > > >
> > > > > > /* export stream parameters from the first SPS */
> > > > > > for (i = 0; i < FF_ARRAY_ELEMS(s->ps.sps_list); i++) {
> > > > > > - if (s->ps.sps_list[i]) {
> > > > > > + if (first && s->ps.sps_list[i]) {
> > > > > > const HEVCSPS *sps = (const HEVCSPS*)s->ps.sps_list[i]->data;
> > > > > > export_stream_params(s->avctx, &s->ps, sps);
> > > > > > break;
> > > > > > @@ -3099,7 +3099,7 @@ static int hevc_decode_frame(AVCodecContext *avctx, void *data, int *got_output,
> > > > > > new_extradata = av_packet_get_side_data(avpkt, AV_PKT_DATA_NEW_EXTRADATA,
> > > > > > &new_extradata_size);
> > > > > > if (new_extradata && new_extradata_size > 0) {
> > > > > > - ret = hevc_decode_extradata(s, new_extradata, new_extradata_size);
> > > > > > + ret = hevc_decode_extradata(s, new_extradata, new_extradata_size, 0);
> > > > > > if (ret < 0)
> > > > > > return ret;
> > > > > > }
> > > > > > @@ -3387,7 +3387,7 @@ static av_cold int hevc_decode_init(AVCodecContext *avctx)
> > > > > > s->threads_number = 1;
> > > > > >
> > > > > > if (avctx->extradata_size > 0 && avctx->extradata) {
> > > > > > - ret = hevc_decode_extradata(s, avctx->extradata, avctx->extradata_size);
> > > > > > + ret = hevc_decode_extradata(s, avctx->extradata, avctx->extradata_size, 1);
> > > > > > if (ret < 0) {
> > > > > > hevc_decode_free(avctx);
> > > > > > return ret;
> > > > >
> > > > > Couldn't that have done in a less confusing way? What the heck does
> > > > > "first" even mean? (Also you have to look up what that means on the
> > > > > caller site.)
> > > > >
> > > > > Can you explain this?
> > > >
> > > > first means "Preceding all others of a series or kind;" https://en.wiktionary.org/wiki/first
> > >
> > > Thanks for the English lesson. By the way, you're the only person on
> > > the internet who uses "iam".
> > >
> >
> > > > in the context of hevc_decode_extradata(), a function decoding extradata
> > > > that signifies the first of a series of extradata.
> > > >
> > > > No question it can be done differently, everything can be done
> > > > differently.
> > >
> > > So can you explain why "first" is checked every loop, instead of
> > > putting the loop into an if, or
> >
> > Every piece of code can be implemented im many ways.
> > Theres many equivalent and near equivalent ways to do it.
> > add a if(), add a condition to a existing if(), ...
> > some of these are a bit simpler, some are a bit faster, speedwise
> > theres nothing one can gain here as this executes too rarely
>
> Well this one has literally no advantage, and it's like it was written
> in this way just to be tricky.
>
> >
> >
> > > (what should have been done) moving the
> > > entire loop out of the function and putting it inline into the only
> > > place where it's effectively used?
> >
> > For a security fix i like to make sure it can be robustly backported.
> > If i add a "int first" the compiler makes sure that every call has
> > the parameter set.
> > If i move code around the compiler wouldnt notice if there was another
> > call.
> >
> > For git master, i agree moving the code is slightly nicer, ill post a
> > patch to make that change
> >
> > thx
> >
> > [...]
>
> I don't think the backport argument works either.
> If it's a subtle
> security issue, you should probably do something more careful than
> cherry-pick + see if it compiles.
yes absolutely but
Is it a subtle issue, noone has claimed it is nor that it isnt. And
where did i say that one would just do "cherry-pick + see if it compiles."
I do like to use all tools available that help produce better code
with less effort. If i can use the compiler to check against some
issues, thats a good thing
The fewer things the compiler can check the more things will be buggy
on average when the human time spend on it stays constant. No matter
how much or little time the human spends on it
That said, if you want to help backport fixes, that surely is welcome.
With whatever code style you prefer.
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
I do not agree with what you have to say, but I'll defend to the death your
right to say it. -- Voltaire
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20170706/4f940401/attachment.sig>
More information about the ffmpeg-devel
mailing list