The minimalistic x86 x64 api hooking library for windows

Posted in Programming by Michael (Ramen Software) on September 30th, 2013.
Tags: minhook

• Download Source (v1.3.3) — 49.1 KB
• Download Precompiled DLLs (v1.3.3) — 19.0 KB
• Download Precompiled Static Libraries (v1.3.3) — 1.8 MB
• Download Samples (v1.3.2) — 120.6 KB

• Release archive of my GitHub repository
• NuGet Package for the latest version (NuGet Gallery)

Background

As you who are interested in Windows API hooking know, there is an excellent library for it by Microsoft Research named Detours. It’s really useful, but its free edition (called ‘Express’) doesn’t support the x64 environment. Though its commercial edition (called ‘Professional’) supports x64, it’s too expensive for me to afford. It costs around US$10,000!

So I decided to write my own library or «poorman’s Detours» from scratch. But I haven’t designed my library as the perfect clone of Detours. It has just the API hooking functionality because that’s all I want.

As of April 2016, this library is used in some projects: 7+ Taskbar Tweaker, Better Explorer, ConEmu, DxWnd, Mumble, NonVisual Desktop Access, Open Broadcaster Software, QTTabBar, x360ce, mods for some games and more. I am happy finding that this project is helpful to people.

Using the Library

Look at the sample code below. That’s all. It hooks the MessageBoxW() function and modifies its text. This example is included in the source and binary archive. Please try it in both x64 and x86 modes.

The NuGet Package is now available for those who are familiar with NuGet.

#include <Windows.h>
#include "MinHook.h"

#if defined _M_X64
#pragma comment(lib, "libMinHook.x64.lib")
#elif defined _M_IX86
#pragma comment(lib, "libMinHook.x86.lib")
#endif

typedef int (WINAPI *MESSAGEBOXW)(HWND, LPCWSTR, LPCWSTR, UINT);

// Pointer for calling original MessageBoxW.
MESSAGEBOXW fpMessageBoxW = NULL;

// Detour function which overrides MessageBoxW.
int WINAPI DetourMessageBoxW(HWND hWnd, LPCWSTR lpText, LPCWSTR lpCaption, UINT uType)
{
    return fpMessageBoxW(hWnd, L"Hooked!", lpCaption, uType);
}

int main()
{
    // Initialize MinHook.
    if (MH_Initialize() != MH_OK)
    {
        return 1;
    }

    // Create a hook for MessageBoxW, in disabled state.
    if (MH_CreateHook(&MessageBoxW, &DetourMessageBoxW, 
        reinterpret_cast<LPVOID*>(&fpMessageBoxW)) != MH_OK)
    {
        return 1;
    }

    // or you can use the new helper function like this.
    //if (MH_CreateHookApiEx(
    //    L"user32", "MessageBoxW", &DetourMessageBoxW, &fpMessageBoxW) != MH_OK)
    //{
    //    return 1;
    //}

    // Enable the hook for MessageBoxW.
    if (MH_EnableHook(&MessageBoxW) != MH_OK)
    {
        return 1;
    }

    // Expected to tell "Hooked!".
    MessageBoxW(NULL, L"Not hooked...", L"MinHook Sample", MB_OK);

    // Disable the hook for MessageBoxW.
    if (MH_DisableHook(&MessageBoxW) != MH_OK)
    {
        return 1;
    }

    // Expected to tell "Not hooked...".
    MessageBoxW(NULL, L"Not hooked...", L"MinHook Sample", MB_OK);

    // Uninitialize MinHook.
    if (MH_Uninitialize() != MH_OK)
    {
        return 1;
    }

    return 0;
}

If you are a C++ user, you can write a small wrapper for MH_CreateHook() and MH_CreateHookApi(). It allows you to remove annoying reinterpret_casts like this:

template <typename T>
inline MH_STATUS MH_CreateHookEx(LPVOID pTarget, LPVOID pDetour, T** ppOriginal)
{
    return MH_CreateHook(pTarget, pDetour, reinterpret_cast<LPVOID*>(ppOriginal));
}

template <typename T>
inline MH_STATUS MH_CreateHookApiEx(
    LPCWSTR pszModule, LPCSTR pszProcName, LPVOID pDetour, T** ppOriginal)
{
    return MH_CreateHookApi(
        pszModule, pszProcName, pDetour, reinterpret_cast<LPVOID*>(ppOriginal));
}

...

// Create a hook for MessageBoxW, in disabled state.
if (MH_CreateHookApiEx(L"user32", "MessageBoxW", &DetourMessageBoxW, &fpMessageBoxW) != MH_OK)
{
    return 1;
}

How It Works

The basic concept of this software is the same as Microsoft Detours and Daniel Pistelli’s Hook-Engine. It replaces the prologue of the target function with x86’s JMP (unconditional jump) instruction to the detour function. It’s safe, stable, and a proven method.

Overwriting the Target Function

In the x64/x86 instruction set, there are some forms of the JMP instruction. I decided to always use a 32 bit relative JMP of 5 bytes. It’s the shortest form that can be used in reality. In this case, shorter is better.

In x86 mode, 32bit relative JMP covers the whole address space. Because the overflowed bits are just ignored in the relative address arithmetic, in x86 mode, the function addresses don’t matter.

; x86 mode (assumed that the target function is at 0x40000000)

; 32bit relative JMPs of 5 bytes cover whole address space
0x40000000:  E9 FBFFFFBF      JMP 0x0        (EIP+0xBFFFFFFB)
0x40000000:  E9 FAFFFFBF      JMP 0xFFFFFFFF (EIP+0xBFFFFFFA)

; Shorter forms are useless in this case
; 8bit JMPs of 2 bytes cover -126 ~ +129 bytes
0x40000000:  EB 80            JMP 0x3FFFFF82 (EIP-0x80)
0x40000000:  EB 7F            JMP 0x40000081 (EIP+0x7F)
; 16bit JMPs of 4 bytes cover -32764 ~ +32771 bytes
0x40000000:  66E9 0080        JMP 0x3FFF8004 (EIP-0x8000)
0x40000000:  66E9 FF7F        JMP 0x40008003 (EIP+0x7FFF)

But, in x64 mode, it’s a problem. It only covers the very narrow range in comparison with the whole address space. So I introduced a new function called ‘Relay Function’ which just has a 64 bit jump to the detour function and is placed near the target function. Fortunately, the VirtualAlloc() API function can accept the address to allocate, and it’s an easy job to look for unallocated regions near the target function.

; x64 mode (assumed that the target function is at 0x140000000)

; 32bit relative JMPs of 5 bytes cover about -2GB ~ +2GB
0x140000000: E9 00000080      JMP 0xC0000005  (RIP-0x80000000)
0x140000000: E9 FFFFFF7F      JMP 0x1C0000004 (RIP+0x7FFFFFFF)

; Target function (Jump to the Relay Function)
0x140000000: E9 FBFF0700      JMP 0x140080000 (RIP+0x7FFFB)

; Relay function (Jump to the Detour Function)
0x140080000: FF25 FAFF0000    JMP [0x140090000 (RIP+0xFFFA)]
0x140090000: xxxxxxxxxxxxxxxx ; 64bit address of the Detour Function

Building the Trampoline Function

The target function is overwritten to detour. And, how do we call the original target function? In many cases, we have to call the original function from within the detour function. MinHook has a function called «Trampoline Function» for the purpose of calling the original function (and Daniel Pistelli call it «Bridge Function»). This is a clone of the prologue of the original function with the trailing unconditional jump for resuming into the original function. The real world examples are here. They are what MinHook actually creates.

We should disassemble the original function to know the instructions boundary and the instructions to be copied. I adopted Vyacheslav Patkov‘s Hacker Disassembler Engine (HDE) as a disassembler. It’s small, light-weight and suitable for my purpose. I disassembled thousands of API functions on Windows XP, Vista, and 7 for examination purposes, and built the trampoline function for them.

; Original "USER32.dll!MessageBoxW" in x64 mode
0x770E11E4: 4883EC 38         SUB RSP, 0x38
0x770E11E8: 4533DB            XOR R11D, R11D
; Trampoline
0x77064BD0: 4883EC 38         SUB RSP, 0x38
0x77064BD4: 4533DB            XOR R11D, R11D
0x77064BD7: FF25 5BE8FEFF     JMP QWORD NEAR [0x77053438 (RIP-0x117A5)]
; Address Table
0x77053438: EB110E7700000000  ; Address of the Target Function +7 (for resuming)

; Original "USER32.dll!MessageBoxW" in x86 mode
0x7687FECF: 8BFF              MOV EDI, EDI
0x7687FED1: 55                PUSH EBP
0x7687FED2: 8BEC              MOV EBP, ESP
; Trampoline
0x0014BE10: 8BFF              MOV EDI, EDI
0x0014BE12: 55                PUSH EBP
0x0014BE13: 8BEC              MOV EBP, ESP
0x0014BE15: E9 BA407376       JMP 0x7687FED4

What if the original function contains the branch instructions? Of course, they should be modified to point to the same address as the original.

; Original "kernel32.dll!IsProcessorFeaturePresent" in x64 mode
0x771BD130: 83F9 03           CMP ECX, 0x3
0x771BD133: 7414              JE 0x771BD149
; Trampoline
; (Became a little complex, because 64 bit version of JE doesn't exist)
0x77069860: 83F9 03           CMP ECX, 0x3
0x77069863: 74 02             JE 0x77069867
0x77069865: EB 06             JMP 0x7706986D
0x77069867: FF25 1BE1FEFF     JMP QWORD NEAR [0x77057988 (RIP-0x11EE5)]
0x7706986D: FF25 1DE1FEFF     JMP QWORD NEAR [0x77057990 (RIP-0x11EE3)]
; Address Table
0x77057988: 49D11B7700000000  ; Where the original JE points.
0x77057990: 35D11B7700000000  ; Address of the Target Function +5 (for resuming)

; Original "gdi32.DLL!GdiFlush" in x86 mode
0x76479FF4: E8 DDFFFFFF       CALL 0x76479FD6
; Trampoline
0x00147D64: E8 6D223376       CALL 0x76479FD6
0x00147D69: E9 8B223376       JMP 0x76479FF9

; Original "kernel32.dll!CloseProfileUserMapping" in x86 mode
0x763B7918: 33C0              XOR EAX, EAX
0x763B791A: 40                INC EAX
0x763B791B: C3                RET
0x763B791C: 90                NOP
; Trampoline (Additional jump is not required, because this is a perfect function)
0x0014585C: 33C0              XOR EAX, EAX
0x0014585E: 40                INC EAX
0x0014585F: C3                RET 

The RIP relative addressing mode is also a problem in the x64 mode. Their relative addresses should be modified to point to the same addresses.

; Original "kernel32.dll!GetConsoleInputWaitHandle" in x64 mode
0x771B27F0: 488B05 11790C00   MOV RAX, [0x7727A108 (RIP+0xC7911)]
; Trampoline
0x77067EB8: 488B05 49222100   MOV RAX, [0x7727A108 (RIP+0x212249)]
0x77067EBF: FF25 4BE3FEFF     JMP QWORD NEAR [0x77056210 (RIP-0x11CB5)]
; Address Table
0x77056210: F7271B7700000000  ; Address of the Target Function +7 (for resuming)

; Original "user32.dll!TileWindows" in x64 mode
0x770E023C: 4883EC 38         SUB RSP, 0x38
0x770E0240: 488D05 71FCFFFF   LEA RAX, [0x770DFEB8 (RIP-0x38F)]
; Trampoline
0x77064A80: 4883EC 38         SUB RSP, 0x38
0x77064A84: 488D05 2DB40700   LEA RAX, [0x770DFEB8 (RIP+0x7B42D)]
0x77064A8B: FF25 CFE8FEFF     JMP QWORD NEAR [0x77053360 (RIP-0x11731)]
; Address Table
0x77053360: 47020E7700000000 ; Address of the Target Function +11 (for resuming)

Conclusion

Though this library is small and simple, I think it’s practical enough. Please enjoy!

History

• 8 Jan, 2017: Updated the source to v1.3.3.
  1. Added a helper function MH_CreateHookApiEx. (Thanks to asm256)
  2. Support Visual Studio 2017 RC.
• 1 Nov, 2015: Updated the source to v1.3.2.
  1. Support Visual Studio 2015.
  2. Support MinGW.
• 18 May, 2015: Updated the source to v1.3.2-beta2.
  1. Fixed some subtle bugs. (Thanks to RaMMicHaeL)
  2. Added a helper function MH_StatusToString. (Thanks to Jan Klass)
• 12 May, 2015: Updated the source to v1.3.2-beta.
  1. Fixed a possible thread deadlock in x64 mode. (Thanks to Aleh Kazakevich)
  2. Reduced the footprint a little more.
  3. Support Visual Studio 2015 RC. (Experimental)
• 19 Mar, 2015: Updated the source to v1.3.1.
  1. No major changes from v1.3.1-beta.
• 11 Mar, 2015: Updated the source to v1.3.1-beta.
  1. Added the MH_CreateHookApi function.
  2. Fixed a false memory leak reported by some tools.
  3. Fixed a degradated compatibility issue.
• 13 Sep, 2014: Updated the source to v1.3.
  1. No major changes from v1.3-beta3.
• 31 Jul, 2014: Updated the source to v1.3-beta3.
  1. Fixed some small bugs.
  2. Improved the memory management.
• 21 Jul, 2014: Updated the source to v1.3-beta2.
  1. Changed the parameters to Windows-friendly types. (void* to LPVOID)
  2. Fixed some small bugs.
  3. Reorganized the source files.
  4. Reduced the footprint a little more.
• 17 Jul, 2014: Updated the source to v1.3-beta.
  1. Rewrote in plain C to reduce the footprint and memory usage. (suggested by Andrey Unis)
  2. Simplified the overall code base to make it more readable and maintainable.
  3. Changed the license from 3-clause to 2-clause BSD License.
• 21 Jun, 2014: Updated the source to v1.2.2.
  1. Fixed compilation with Express Editions.
  2. Reduced the size of static libraries.
• 18 Jun, 2014: Updated the source to v1.2.1c.
  1. Translated the Japanese comments into English.
  2. Converted the source files from Shift-JIS to UTF-8.
• 16 Jun, 2014: Updated the source and binary files to v1.2.1.
  1. A lot of bug fixes and improvements. (thanks to jarredholman and RaMMicHaeL)
  2. Fixed compilation with VC2008, 2010, 2012 and 2013.
• 26 Nov, 2009: Updated the source, binary files and sample code
  1. Fixed a serious bug (thanks to xliqz)
  2. Changed the interface with the bug fix
• 23 Nov, 2009: Updated the source and binary files
  1. Fixed small bugs (internal type mismatch etc)
  2. Separated the .LIBs from the .DLLs
  3. Added the sample executables
• 22 Nov, 2009: Initial post