Pevious: 2a: WinDbg
Next: 2c: List of common Windbg commands

2a: More Windbg

We already covered a lot of the basics last time: Breakpoints, how to look at memory, the stack and registers. How to <del>un</del>disassemble. We only scratched p, the step command. But you can get more on that from the MS documentation or my command list if you need it. There are of course step-in, step-out, but also step to next return and all sorts of ways to step around code.


With no better idea what to do you try the ?, .help and .extmatch * commands to list all possible windbg commands of the three types normal, meta and extensions. You can get additional help with .hh <command>. And .cls to clear the screen afterwards

Windbg Calculator

Bit and byte calculations like calculating offsets, shifting bits etc. are an important part of debugging the windows kernel. You can use these inside normal Windbg commands, for instance dq ` L1` will display the memory at the result of the formula. If you are not sure your formula is correct you can use `? `` to evaluate it. You can also use the `?` command as a simple calculator.

It comes in two flavors: ? will evaluate MASM ((Microsoft Assembler)) syntax and ?? will evaluate c++ syntax((you can also switch the default MASM syntax to C++ with a system variable)).

	? (ff2111 >> 8) + 8
	Evaluate expression: 65321 = 00000000`0000ff29
	kd> ?? (ff2111 >> 8) + 8
	#Couldn't resolve error at 'ff2111 >> 8) + 8'
	kd> ?? ( 0xff2111 >> 0x8 ) + 0x8
	int 0n65321
	kd> .formats 0n65321
	Evaluate expression:
	  Hex:     00000000`0000ff29

I would say the c++ parser is a bit dumb and requires more writing but if you have experience in c++ pointer shenanigans it may come more natural to you than MASM. I would suggest playing around with both options. You can always use c++ syntax inside MASM using @@(<expression>) or mix completely free with @@masm() and @@c++().

And while we are doing calculations we also need to look at pointers.



Pointers are variables that point (duh!) to memory locations. They are a big deal inside computers. For example an array in c is just a pointer to a place in memory. The array content at position x is accessed by accessing it's address+x*sizeof(whatever_the_array_holds). If you tell your c compiler you need an array of a 100 elements and then go and access position 200 -> no problem, it's just numbers. Of course this way stupid things can happen and that it why many programming languages don't give programmers direct access to pointers. But they are still everywhere inside and they are everywhere inside the windows kernel. And even deeper: in the last part I already talked about the Instruction Pointer (rip register on x64) that points to the memory location where the CPU will find the next instruction it should execute.

Just on semantics: 'Dereferencing a pointer' stands for 'retrieving the actual address it points add'.

Pointers can be dereferenced in MASM syntax by using the poi command. If you have some pointer inside the rax register you can use dp rax L1 to get the address it points at and this address with another d~ <address> L1 (~ depends on the data type: a, b, q, p.. ) to look at the actual data. Or you can use d~ poi(rax) to do the same in one command.


Let's look at an example. It combines a lot of concepts so don't be afraid to check other sources or ask me questions. I will also show all the data from my test run so you can trace what is happening. Sidenote: I will use 16 as Windbg number base (n 16), so 0n10 is a decimal 10 and 10 is a hex 10!

We will once again need a breakpoint at our friend nt!NtCreateUserProcess. Right at the very start before the prolog.

I forgot how to do it.
	bp nt!NtCreateUserProcess
	# if it doesn't break open debugee and start something

In the assembly primer of the last part I already wrote a few things about how functions are called and how the stack is used to do it. Now we get to another function calling convention: how parameters are transmitted. You should read some real documentation ((Like but the short version is this: The first four parameters are transmitted via registers (RCX, RDX, R8, and R9), the rest are put on the stack. But windows still reserves 4 spaces on the stack right before the return address as 'home space' for the parameters that are transmitted via registers (

So this is what it looks like:

	Stack Space nt!NtCreateUserProcess will use
	RETURN ADDRESS to calling function
	5th Parameter
	6th Parameter
	7th Parameter
	Stack of calling function

And NtCreateuserProcess has a lot of parameters. And as far as I can tell it isn't officially documented so beware that we are entering dangerous waters here. Luckily enough we are not the only ones looking at the windows kernel so google gave me a what is likely the correct function declaration:

	NTSTATUS NTAPI NtCreateUserProcess(
	    PHANDLE ProcessHandle,
	    PHANDLE ThreadHandle,
	    ACCESS_MASK ProcessDesiredAccess,
	    ACCESS_MASK ThreadDesiredAccess,
	    POBJECT_ATTRIBUTES ProcessObjectAttributes,
	    POBJECT_ATTRIBUTES ThreadObjectAttributes,
	    ULONG ProcessFlags,
	    ULONG ThreadFlags,

I don't want to go into handles right now, so let's ignore the first two parameters. The Access Mask parameters 3 and 4 are 0x20000000 for me. Checking the official ACCESS_MASK documentation at the MS website they could really be Access Masks with just the "MAXIMUM_ALLOWED" bit set. Take a look at yours with .formats r8.

What we really want is the stack mine looks like this:

	kd> dp /c1 rsp
	fffff90a`3e1a3a88  fffff803`72990003 -> return address
	fffff90a`3e1a3a90  00000000`00000002 -> home
	fffff90a`3e1a3a98  00000000`00000000 -> home
	fffff90a`3e1a3aa0  00000000`00000000 -> home
	fffff90a`3e1a3aa8  00000000`00000001 -> home
	fffff90a`3e1a3ab0  00000000`00000000 -> 5th parameter
	fffff90a`3e1a3ab8  00000000`00000000 -> 6th parameter
	fffff90a`3e1a3ac0  000002b5`00000000 -> 7th parameter ulong ProcessFlags
	fffff90a`3e1a3ac8  000000fc`00000001 -> 8th parameter ulong ThreadFlags
	fffff90a`3e1a3ad0  000002b5`3e1b4d60 -> 9th paramter PRTL_USER_PROCESS_PARAMETERS ProcessParameters
	fffff90a`3e1a3ad8  000000fc`79c7d750 -> 10th parameter PPROCESS_CREATE_INFO CreateInfo
	fffff90a`3e1a3ae0  000000fc`79c7dc30 -> 11th parameter PPROCESS_ATTRIBUTE_LIST AttributeList
	fffff90a`3e1a3ae8  00000000`00000000 -> whatever
	fffff90a`3e1a3af0  00000000`00000001 -> something else

If I try to interpret parameter 7 as a pointer with dp 000002b5'00000000, I get garbage so it probably is a flag variable like it is supposed to. Parameter 9 is supposed to be a pointer to an RTL_USER_PROCESS_PARAMETERS structure. Once again using google I found the following declaration at the microsoft website:

	  BYTE           Reserved1[16];
	  PVOID          Reserved2[10];
	  UNICODE_STRING ImagePathName;
	  UNICODE_STRING CommandLine;

So it is easy as pie: first we have to print out the memory where parameter 9 points at (dq 000002b53e1b4d60 in my case). We count down 16 byte and 10 (8-byte) pointers and we will find a unicode string with the ImagePathName. In my example dq 000002b5'3e1b4d60 + 0n16 + 0n10*8` should give me the ImagePathName. Unfortunately it doesn't. For the simple reason that the structure declaration I found is not the correct one for Windows 10. And what I can gather from a quick look around the interweb it may very well be pre-vista. So be prepared for disappointment if you take random kernel code from the interweb, even (especially?) if it comes from the Microsoft website.

The correct one:

	     INT MaximumLength;
	     INT Length;
	     INT Flags;
	     INT DebugFlags;
	     PVOID ConsoleHandle;
	     INT ConsoleFlags;
	     PVOID StandardInput;
	     PVOID StandardOutput;
	     PVOID StandardError;
	     CURDIR CurrentDirectory;
	     UNICODE_STRING ImagePathName;
	     UNICODE_STRING CommandLine;

We still want the offset to the ImagePathName. The first four variables are put into memory as 4 byte integer, int ConsoleFlags can't be put into a 4 byte slot in between 8-byte pointer so it takes 8 bytes of space. CURDIR consists of a Pointer and a UNICODE_STRING(16 byte), so it takes 24 byte. That puts the DLLPath at offset 10*8 = 0x50 and ImagaPathName at 0x60.

Once again how that looks for me:

	kd> dq 000002b5`3e1b4d60 + 0x60 L2
	000002b5`3e1b4dc0  00000000`00440042 000002b5`3e1b5378

We are nearly there. UNICODE_STRING is once again a structure, that consists of two unsigned short variables - length and max length - and a pointer 8-) to a buffer that contains the actual unicode string. So in my example I can expect a unicode string at 000002b5'3e1b5378. Let's try it:

	kd> du /c42 000002b5`3e1b5378 
	000002b5`3e1b5378  "C:\Windows\system32\usoclient.exe"

Awesome. I found out that Windows sneakily tried to launch the UpdateOrchestrator. For you it will likely be something different, but the procedure is the same. Try it for yourself at least once. Play around a bit if you are unsure what is going on.

Bonus exercise:

Get the 'CurrentDirectory' string inside the CURDIR structure.

We are of course not done. We still need to combine all of this so we get the ImagePathName based on the rsp register so we don't have to do this every time. Try to do it yourself, just put it together one by one.

du /c40 poi(poi(rsp+0n9*8)+60+8)

kd> du /c40 poi(poi(rsp+0n9*8)+60+8)
000002b5`3e1b5378  "C:\Windows\system32\usoclient.exe"
kd> g
Breakpoint 0 hit
kd> du /c40 poi(poi(rsp+0n9*8)+60+8)
000002b5`3d2e1028  "c:\windows\system32\taskhostw.exe"

Since the RTL_USER_PROCESS_PARAMETERS structure is part of the public kernel symbols there is a much simpler way using the dt command:

	# -> find the structure name
	# -> view the structure definition
	# -> view the content of the structure that was used as parameter to nt!NtCreateUserProcess

Temporary registers

Windbg has t<0-9> registers as temporary variables of sort. Since there is no easy (at least I don't know any) way to use the output of one command as input for the next I will use these to show data agnostic examples. You will also see (and use) them a lot in Windbg scripts.

Final words

This concludes my Introduction to Windbg. There is more to learn: I didn't mention the thread commands and watches/traces for instance and we were rather short on registers. But I think I covered the basics and we are ready to dive deeper in tutorials:winkdbg:page3.