Re: [v3 3/3] d3dx9: Remove redundant parameter size check in set_constants().

[Matteo Bruni]

2017-06-15 0:00 GMT+02:00 Paul Gofman :
> Signed-off-by: Paul Gofman
>
> v3:
>     no changes.
>
> dlls/d3dx9_36/preshader.c | 5 -----
>  1 file changed, 5 deletions(-)
>
> diff --git a/dlls/d3dx9_36/preshader.c b/dlls/d3dx9_36/preshader.c
> index 2785ca3..0bc31cb 100644
> --- a/dlls/d3dx9_36/preshader.c
> +++ b/dlls/d3dx9_36/preshader.c
> @@ -1178,11 +1178,6 @@ static void set_constants(struct d3dx_regstore *rs, struct d3dx_const_tab *const
>                          param_offset = i + j * info.major;
>                      else
>                          param_offset = i * info.minor + j;
> -
>                if (param_offset * sizeof(unsigned int) >= param->bytes)
> -
>                {
> -
>                    WARN("Parameter data is too short, name %s, component %u.\n", debugstr_a(param->name), i);
> -
>                    break;
> -
>                }
>                out[offset] = data[param_offset];

Is there anything else covering this case though? If I'm not mistaken
such a parameter / constant would be accepted currently so just
dropping the check would potentially allow broken preshaders to read
outside of the parameter data.

Now I certainly agree that the check shouldn't be there, it should
probably be in init_set_constants_param() instead. If this is already
checked somewhere in the code then I'm okay with the patch as is.

[v3 3/3] d3dx9: Remove redundant parameter size check in set_constants().

[Paul Gofman]

On 06/16/2017 06:29 PM, Matteo Bruni wrote:
> 2017-06-15 0:00 GMT+02:00 Paul Gofman :
> > Signed-off-by: Paul Gofman
> >
> > v3:
> >      no changes.
> >
> > dlls/d3dx9_36/preshader.c | 5 -----
> >   1 file changed, 5 deletions(-)
> >
> > diff --git a/dlls/d3dx9_36/preshader.c b/dlls/d3dx9_36/preshader.c
> > index 2785ca3..0bc31cb 100644
> > --- a/dlls/d3dx9_36/preshader.c
> > +++ b/dlls/d3dx9_36/preshader.c
> > @@ -1178,11 +1178,6 @@ static void set_constants(struct d3dx_regstore *rs, struct d3dx_const_tab *const
> >                           param_offset = i + j * info.major;
> >                       else
> >                           param_offset = i * info.minor + j;
> > -
> >                if (param_offset * sizeof(unsigned int) >= param->bytes)
> > -
> >                {
> > -
> >                    WARN("Parameter data is too short, name %s, component %u.\n", debugstr_a(param->name), i);
> > -
> >                    break;
> > -
> >                }
> >                 out[offset] = data[param_offset];
>
> Is there anything else covering this case though? If I'm not mistaken
> such a parameter / constant would be accepted currently so just
> dropping the check would potentially allow broken preshaders to read
> outside of the parameter data.
>
> Now I certainly agree that the check shouldn't be there, it should
> probably be in init_set_constants_param() instead. If this is already
> checked somewhere in the code then I'm okay with the patch as is.

I've checked that:

- there is no independent size for matrices and vectors during 
effect parse, the size is set as: 'param->bytes = 4 * param->rows * 
param->columns;' in parse_effect_typedef();

- parameter length in bytes is not necessarily checked in matrix 
and vector set functions in effect.c.

So the check currently present in set_constants() looks really 
redundant to me. Even if I am missing some case when 'bytes' can 
actually be not sufficient to hold 'rows * columns' value it seems 
reasonable to me to handle it during effect creation rather than in 
'init_set_constants_param()'.
     I suggest to add an assert instead of error & FIXME to 
'init_set_constants_param()' the same way we just did for value sizes.