Re: [Intel-gfx] [PATCH] fbmem: Reject FB_ACTIVATE_KD_TEXT from userspace

[Javier Martinez Canillas <javierm@redhat.com>]

Daniel Vetter <daniel.vetter@ffwll.ch> writes:

> This is an oversight from dc5bdb68b5b3 ("drm/fb-helper: Fix vt
> restore") - I failed to realize that nasty userspace could set this.
>
> It's not pretty to mix up kernel-internal and userspace uapi flags
> like this, but since the entire fb_var_screeninfo structure is uapi
> we'd need to either add a new parameter to the ->fb_set_par callback
> and fb_set_par() function, which has a _lot_ of users. Or some other
> fairly ugly side-channel int fb_info. Neither is a pretty prospect.
>
> Instead just correct the issue at hand by filtering out this
> kernel-internal flag in the ioctl handling code.
>
> Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
> Fixes: dc5bdb68b5b3 ("drm/fb-helper: Fix vt restore")

[..]

> diff --git a/drivers/video/fbdev/core/fbmem.c b/drivers/video/fbdev/core/fbmem.c
> index 875541ff185b..3fd95a79e4c3 100644
> --- a/drivers/video/fbdev/core/fbmem.c
> +++ b/drivers/video/fbdev/core/fbmem.c
> @@ -1116,6 +1116,8 @@ static long do_fb_ioctl(struct fb_info *info, unsigned int cmd,
>  	case FBIOPUT_VSCREENINFO:
>  		if (copy_from_user(&var, argp, sizeof(var)))
>  			return -EFAULT;
> +		/* only for kernel-internal use */
> +		var.activate &= ~FB_ACTIVATE_KD_TEXT;
>  		console_lock();

I don't have a better idea on how to fix this and as you said the whole
struct fb_var_screeninfo is an uAPI anyways...

Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>

-- 
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat