Re: [PATCH 4/5] ramfb: add sanity checks to ramfb_create_display_surface

[Laszlo Ersek <lersek@redhat.com>]

On 04/22/20 12:02, Gerd Hoffmann wrote:
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> ---
>  hw/display/ramfb.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/hw/display/ramfb.c b/hw/display/ramfb.c
> index fbe959147dc9..d1b1cb9bb294 100644
> --- a/hw/display/ramfb.c
> +++ b/hw/display/ramfb.c
> @@ -15,6 +15,7 @@
>  #include "qapi/error.h"
>  #include "hw/loader.h"
>  #include "hw/display/ramfb.h"
> +#include "hw/display/bochs-vbe.h" /* for limits */
>  #include "ui/console.h"
>  #include "sysemu/reset.h"
>  
> @@ -49,6 +50,11 @@ static DisplaySurface *ramfb_create_display_surface(int width, int height,
>      hwaddr size;
>      void *data;
>  
> +    if (width < 16 || width > VBE_DISPI_MAX_XRES ||
> +        height < 16 || height > VBE_DISPI_MAX_YRES ||

Seems to make sense (I've checked the constants).

> +        format == 0 /* unknown format */)

OK, this is from qemu_drm_format_to_pixman().

> +        return NULL;
> +
>      if (linesize == 0) {
>          linesize = width * PIXMAN_FORMAT_BPP(format) / 8;
>      }
> 

I would suggest four more sanity checks:

- if "linesize" is nonzero, make sure it is a whole multiple of the
required word size (?)

- if "linesize" is nonzero, make sure it is not bogus with relation to
"width". I'm thinking something like:

    if (linesize > 0) {
        min_linesize = width * PIXMAN_FORMAT_BPP(format) / 8;
        if (linesize < min_linesize) {
            return NULL;
        }
    }

May not be the best way to put it, but you get the idea.

- We might want to put an upper bound on "linesize" too. I realize
"(hwaddr)linesize * height" should be safe in this function, as the
multiplication is done in uint64_t. But we also pass "linesize" to
qemu_create_displaysurface_from(), and who knows how it is used for
multiplication there.

- in the ramfb_create_display_surface() function, we should change the
type of the parameters "width", "height", and "linesize", from "int" to
"uint32_t". In ramfb_fw_cfg_write(), we do take them as uint32_t from
the guest, but then pass them as "int"s. And so the current state can
produce negative values for any of "width", "height", and "linesize" --
and I'd rather not investigate where those lead. (The new checks catch a
negative "width" and "height" already, but not "linesize".) Casting a
negative "linesize" to (hwaddr) produces a big, ugly value, FWIW.

Thanks
Laszlo