There are several ways how you can realize system wide API hooking. Due to the quite different system architecture of the two OS families "win9x" and "winNT" system wide hooking normally must be realized by using different solutions for each OS family. The win9x family has a much lower system security, so there we can hack around like crazy and can often do very effective system wide hooking without even having to write a DLL. The winNT family has a much more secure system base. Still we can do some hacks, but doing system wide API hooking in the winNT family without using DLLs is much more difficult than it is in the win9x family.
madCodeHook is meant to free you from having to care about in which OS your program is running. So for the "normal" programmer I strongly recommend to use madCodeHook's general OS independent solution for doing system wide API hooking. This solution is based on putting your hooking code into a little DLL and on then injecting this DLL into all current and in the future created processes. Actually using madCodeHook this is very easy to do. If you want to go this route, please skip the rest of this page and head directly to the How To Use madCodeHook guide.
What - you're still reading!?
That either means, you want to know it all. Or you don't care about writing OS independent code. Okay, so I'll explain to you know how you can hook most system APIs in win9x without using any DLLs. Sadly this solution only works in win9x, but hey, life isn't easy...
In win9x the memory/address area is splitted into 2 halfs. The area from $00000000 - $7FFFFFFF is private to each process. No process can directly access the private memory of another process, except by using the debug APIs "Read/WriteProcessMemory". The area from $80000000 - $FFFFFFFF is the so called "shared area", because the memory in this area is shared between all processes in win9x. In winNT/2000 there is no shared area. There the whole 32bit memory/address area is always private to each process.
In win9x that means, if one process changes memory in the shared area, all other processes will directly see this change. This is the reason why win9x is not a very stable operating system. One crashing application can accidently overwrite important stuff in the shared area and so make other processes or even the whole OS crash, too.
However, the shared area has also its pros. The most important for us now is that it gives us the possibility to hook "shared APIs" (I mean APIs that are located in the shared area) truely system wide. When "madCodeHook" overwrites the code of a shared API with a jump instruction, this ends up in that each and every process which calls this API thereafter, will execute that jump instruction and so jump to our callback function. Voilą - we have system wide API hooking.
But wait! There are some complications. Let us think about where your callback function is located. Normal applications and DLLs are not loaded into the shared area, so your callback function is most probably not located in the shared area, either. That means, it is private to your process, so not accessible in any other process. But what happens now if another process calls the hooked shared API and then executes that jump instruction which wants to jump to your callback function, which is unfortunately not accessible in that other process? This process will crash. You see the problem?
That means, when hooking shared APIs, your callback function MUST be located in the shared area. No way around that. So how we can move your callback function to the shared area? There are 2 possibilities:
|1.||You can put the callback function into a little DLL and make that DLL being loaded by win9x into the shared area. To do so the image base address must be greater than $80000000, furthermore the code section and all data sections must have the "shared" flag set. This way the whole DLL including your callback function will automatically be loaded into the shared area.|
|2.||We can copy your callback function to the shared area. That is not an easy task, because your callback function may contain relative jumps, which then have to be corrected before anyone can call your copied function successfully. Sounds difficult? Yes, it is. You even need a little disassembler to do something like that. But - we have a little disassembler - and it is also able to copy and relocate functions completely automatically (see CopyFunction)!|
Pratically spoken, when you hook a shared API, "madCodeHook" checks whether your callback function is already located in the shared area. If it is not, it is copied there, and all necessary corrections are also done automatically.
Unfortunately the location of the callback function (shared area or not) is not the only problem. There are several additional things you have to be careful about when writing a system wide callback function:
|1.||Because the hook is meant to be really system wide, and because Windows is a true multitasking system, your callback function has to be fully reentrant and thread safe.|
|2.||You must not access any functions, APIs or variables, that are not located in the shared area, because everything you access in the callback function must be accessible system wide.|
|3.||You can't know which DLLs are initialized in what processes, so you must only use APIs that are exported by "kernel32.dll". It's the only DLL that is guaranteed to be initialized in each and every win32 process. However, you can use "LoadLibrary" to initialize additional DLLs in your callback function, if you really need that. Furthermore when you hook e.g. a "user32.dll" API, you can also use other "user32.dll" APIs in your callback function, because your "user32.dll" API hook can't get called in the context of a process, which has not initialized "user32.dll". That would be paradox.|
|4.||You have to make sure that all CPU registers are restored correctly. Normally Delphi/C++ does that for us. But to be sure you should think about writing your system wide callback function in (inline) assembler.|
|5.||You must not use structured exception handling in the callback function.|
If your callback function fulfills all the 5 requirements noted above, you can safely hook any shared API in win9x without risking system stability. To check if a DLL is a shared one or not, please check whether the module handle of the DLL (see "GetModuleHandle") is > $80000000. The most important system DLLs are shared, e.g. "kernel32.dll" and "user32.dll".
Now let us talk about unhooking. If you don't unhook system wide APIs, they remain hooked even after your application has terminated. That's a very big problem, if your callback function was loaded into the shared area by using a little shared DLL, because this DLL gets unloaded automatically when your process terminates. But the API hook does NOT get unhooked automatically. So if someone now calls the still hooked API, it will try to jump to your callback function. But since the DLL was unloaded, your callback function is not there anymore, which will result in a crash. So when using callback functions which are located in a shared DLL, please make sure that you unhook all shared APIs, before you leave - even if your program crashes. Please install an exception handler that unhooks the shared APIs. Alternatively instead of loading your little shared DLL yourself, you can inject it into the "kernel32.dll" win9x system process. This way your DLL will stay loaded until the next reboot.
If your callback function was copied to the shared area by madCodeHook, things are a bit different, because such a copied function remains there in any case - even if your program terminates. That means, it's up to you in this case, whether you want to unhook the API before leaving your program or not. E.g. you can install a kind of patch, which remains installed until the next reboot, although your program has terminated already.
If you want to see a full example project about system wide API hooking, please look at the "SystemAPI" Example.