Re: Shader security model discussions: breaking it down?

On 24 Aug 2017, at 3:13 pm, Kai Ninomiya <kainino@google.com> wrote:
> Yes. These security constraints apply to the entire API usage, both inside and outside shaders.

I read through the Vulkan spec today, thinking about how I could try and break things. Most of my ideas were outside of the scope of the Vulkan specification, so I suspect there is a need to have some examples that do break things.

Is there any Vulkan, DirectX and Metal code that demonstrates how a malicious shader causes a system to crash? If we can build up a library, then ideas and suggestions can be tested on some real examples being used as a test suite.

I had a thought about hashing resource identifiers, then only needing to monitor resource binding function calls... but I would need to know more real world detail to really have an idea of how effective this approach would be.

--
  steve@adam.com.au

Received on Saturday, 26 August 2017 12:53:32 UTC