[vlc-devel] [PATCH] gme: fixed out-of-memory/out-of-range tests

[Rémi Denis-Courmont]

Le vendredi 23 mars 2012 19:00:04 Piotr Fusik, vous avez écrit :
> Hello everyone, I'm new here. :)
> 
> Dnia 2012-03-23 18:49 Rémi Denis-Courmont napisał(a):
> >Le vendredi 23 mars 2012 18:15:33 Piotr Fusik, vous avez écrit :
> >> ---
> >> 
> >>  modules/demux/gme.c |    6 +++---
> >>  1 files changed, 3 insertions(+), 3 deletions(-)
> >> 
> >> diff --git a/modules/demux/gme.c b/modules/demux/gme.c
> >> index 464ba9e..8f56887 100644
> >> --- a/modules/demux/gme.c
> >> +++ b/modules/demux/gme.c
> >> @@ -261,8 +261,8 @@ static int Control (demux_t *demux, int query,
> 
> va_list
> 
> >> args)
> >> 
> >>               || (sys->titlev[sys->track_id]->i_length == 0))
> >>               || 
> >>                  break;
> >> 
> >> -            int seek = (sys->titlev[sys->track_id]->i_length / 1000) *
> >> pos; -            if (seek > INT_MAX || gme_seek (sys->emu, seek))
> >> +            double seek = (sys->titlev[sys->track_id]->i_length / 1000)
> 
> *
> 
> >> pos;
> >> +            if (seek > INT_MAX || gme_seek (sys->emu, (int) seek))
> >
> >So signed overflow is undefined. Unsigned would be enough and avoid going
> 
> to
> 
> >FPU...
> 
> Correct me if I'm wrong, but I thought double->unsigned conversion is only
> defined within the range of unsigned?

Yeah? So?

I mean, we do care about not crashing or exhibiting undefined behaviour. We 
don't care about not playing corrupt or overly long files correctly.

Unsigned should be enough to protect against undefined behaviour, even if it 
might cause integer overflow, no?

-- 
Rémi Denis-Courmont
http://www.remlab.net/
http://fi.linkedin.com/in/remidenis