search for: target_gp

On Tue, Nov 28, 2017 at 3:22 PM, James Zern <jzern at google.com> wrote: > On Mon, Nov 20, 2017 at 1:07 PM, James Zern <jzern at google.com> wrote: >> Just an attempt to avoid overflows with an explicit check, I don't know if >> there's a better way to identify corrupt input here. >> >> James Zern (2): >> op_pcm_seek: fix int64 overflow

...frames (the default), we'd discard 90 ms on >> average.*/ >> if(discard_count>=0&&OP_UNLIKELY(discard_count<90*48)){ > > > I think that better than adding a custom overflow check here, we should use > if(OP_LIKELY(!op_granpos_diff(&discard_count,target_gp,gp))) directly > (because _pcm_offset == (target_gp - pcm_start) and diff == (gp - > pcm_start). > This works. > [...] > >> @@ -2078,7 +2078,10 @@ static int op_fetch_and_process_page(OggOpusFile >> *_of, >> &&OP_LIKELY(diff<total_duration)){...

...> Assuming 20 ms frames (the default), we'd discard 90 ms on average.*/ > if(discard_count>=0&&OP_UNLIKELY(discard_count<90*48)){ I think that better than adding a custom overflow check here, we should use if(OP_LIKELY(!op_granpos_diff(&discard_count,target_gp,gp))) directly (because _pcm_offset == (target_gp - pcm_start) and diff == (gp - pcm_start). The way you have this now, if the gp of whatever current page we happen to be decoding comes before the start of the stream, you'll set discard_count to 0 and then we will pass the following if() a...

Just an attempt to avoid overflows with an explicit check, I don't know if there's a better way to identify corrupt input here. James Zern (2): op_pcm_seek: fix int64 overflow op_fetch_and_process_page: fix int64 overflow src/opusfile.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) -- 2.15.0.448.gf294e3d99a-goog