The code is released under the MIT license. Bugs will be addressed as soon I've got time available if any are found, all the code in the repo was written in about a morning and an afternoon, so some oversights might be present. PRs are welcome.
Generally, I didn't go over kdmapper's code too much and I only commented parts of code that I edited for this PoC's needs. I did, however, comment almost everything in kdriver and the namespaces and functions I added to kdmapper's code to achieve communication with the kernel payload, please feel free to DM me on twitter, open issues or discussions here if there's anything you'd like to ask or if there's anything you'd want to see fixed.
This repository contains a proof of concept demonstrating communication via mapped shared memory structures between a user-mode process and a kernel-mode payload on Windows 10 20H2. The usermode component of this repository, kdmapper, was originally created by z195, kdmapper is a driver mapper that exploits a vulnerability in a signed intel driver to map a kernel payload and execute it. This project, however, started from a slightly modified version of kdmapper from TheCruZ. This project's kdmapper has also been largely refactored to follow the snake_case convention as closely as possible without, however, editing windows' structures.
Mapped drivers have some limitations, and work differently in some occasions:
- They cannot be unloaded by simply stopping their service, since they don't have one (this is solved by simply jumping to ExFreePool in unmap.asm, hence freeing the memory the driver is mapped in and never returning to it)
- PatchGuard (windows' kernel patch protection) won't let the payload register callbacks regularly and will cause the system to crash with a blue screen of death (bsod) in case the payload tries doing so. To get around the callbacks issue that may be needed to get other processes' modules, the utils.hpp file in the kdriver project comes in handy, implementing PEB walking to iterate the linked list of modules in memory so that we can grab the one we need.
__try {} __except() {}blocks will not work in a mapped driver to catch exceptions. Or in non module-backed memory at all, for the matter.
Just using a manually mapped driver, however, will not cause a blue screen of death; manually mapping your driver is one of the many ways to overcome DSE (driver signature enforcement) on Windows 10, that requires all drivers to be digitally signed to be loaded regularly as system services. The terms payload and mapped driver are used interchangeably throughout this README file. To manually map a driver, it is generally needed to find a vulnerable driver that allows arbitrary kernel memory reading and writing to gain code execution in the kernel.
Now on to the actual proof of concept. This solution contains two projects: kdriver and kdmapper. Access to the shared memory structures is well synchronized between the kernel payload and the usermode process. This project can easily be modified to make kdriver handle multiple maps at once in multiple threads, effectively making communication multithreaded.
This proof of concept's kdmapper is not like the ordinary one. It contains some extra features to aid in remaining stealthy and overall avoiding detection via traditional memory scans after the kernel payload has been mapped, namely:
- Kernel payload PE header randomization
- The vulnerable intel driver's removal from memory
- Extra parameters are passed to the entry point to make communication possible
- An entire comms namespace to setup and maintain communication with the kernel payload in the kdmapper namespace
kdriver is a driver that was specifically designed to work with this project's kdmapper. It takes the memory structure's virtual address in the kdmapper process' address space, takes its physical address, maps it and instaurates communication with the usermode process, making virtual memory read and write operations without a handle from/to the process you want to read and write frmo/to trivial. kdriver is designed to work as "driverless" without a DRIVER_OBJECT assigned to it, therefore it is more of a payload than a proper driver. kdriver's project features a few nice things:
- A DEBUG_MODE constant to enable/disable printing
- A pattern scanner to scan memory
- Somewhat safe MmCopyVirtualMemory calls
- PEB structures
- PEB walking to get usermode process modules bases without needing to set a load notify routine with
PsSetLoadImageNotifyRoutine, as it would trigger PatchGuard and make it bsod the system - A centralized project structure centered around header-only namespaces and a single header for the necessary includes
This project can be used if, for any reason, it is necessary to conduct research on processes that may have had their memory protected by a driver in kernel-mode via a callback that strips handles to the process registered via ObRegisterCallbacks. It can also be easily re-worked to work as a .dll and, with a hooking library, ReadProcessMemory and WriteProcessMemory could be hooked, bringing back to life analysis tools that rely on usermode APIs for their functioning by emulating the behavior of said APIs in the kernel payload. For instance, one could hook ReadProcessMemory with something like MinHook so that kdmapper::comms::read_vm<T> is called instead.
This project, as stealthy as it tries to be via very simple and straightforward methods, can easily be detected by anti-virus and anti-cheating solutions, it takes a few tricks to be able to map a kernel payload and go undetected which, unfortunately, I can't share (as of now) on GitHub for ethical reasons. Some of the methods via which detection could be achieved are:
- System thread enumeration (We're starting a system thread with a start address that is not backed by any real module).
- Stackwalking (Even if the start address is spoofed either via DKOM or via code-caves, APCs can be issued to the thread, and the stack can be walked, revealing code execution in addresses that are not backed by any real module. The thread can also be "asynchronously stackwalked" by just copying the stack asynchronously from the system thread's data structures.).
- MmUnloadedDrivers & PiDDBCacheTable (Unloading the intel driver is not stealthy at all, it will leave traces in the MmUnloadedDrivers array and in the undocumented PiDDBCacheTable linked list).
- The windows system evt logs (Unloading the intel driver will leave logs in there).
Some ways of getting around some of these detection vectors will be added in the future for educational purposes.
- Install VS2019
- Install the WDK (Windows driver kit) + the WDK extension for VS2019
- Build as-is, release mode in x64 already has what needs set-up to build a working kdmapper payload (A custom entry point).
- Open a CMD prompt as admin.
- cd into the directory containing the files that were just built.
- Run
kdmapper.exe kdriver.sys