h1d3r/gpu_shellcodeHides reflective payloads in GPU memory via CUDA APIs, wiping host memory during sleep and using a VEH handler to restore executable memory, evading memory forensics.
1- first, we need to setup and initialize some structs / and run our ve handler.
2- second, we run the stageless payload, we must use stageless, to know where
the reflection will land (im using cobalt strike and it uses dll reflective loader), we only care about the 2nd stage cz thats the actual
payload. i didnt add any tricks to this step, just a virtualalloc and a rwx section
3- now when the payload goes to sleep, we copy the payload to the gpu memory, and clean the payload in the real memory.
4- when the sleep is done, the veh will handle the exception (EXCEPTION_ACCESS_VIOLATION) by re-setting the memory permissions to PAGE_EXECUTE_READWRITE, and placing the payload back in place from the gpu.
