Intro
Have you ever wondered how cool Ghidra is? I did for almost since the release. As someone who pretty much enjoy reading low-level code, I really dig what NSA did with this framework. To be honest, I'm still learning Ghidra potential but from what I can see, the possibilities are endless.
In this post, we will learn how to use Ghidra Scripting to patch a MacOS application called Image2Icon. This simple yet cool application allows you to change Folder Icons to look Sick! The application has been developed by the same company who created a famous Bear.App (a note taking ecosystem app). The application is rather cheap (10$), and you can buy it here which I highly recommend for the simplicity of use and the pricetag. Support these awesome developers.
In short, here is the example of the end results:
As you can see from above picture, basically all Sick modes in the application are disabled; which means there are just a couple of them possible (use-for-free). The Lock Icon represents the mode which are not available for free.
Also, there is no offline installation, nor key-based activation. This means we can't crack application logic and create keygen. What we can do is patch, as always.
btw. I use Sketch.App for all my image editings.
Well, onto Ghidra.
Patching
After loading the application in Ghidra, I imidietly went for Functions window. In here, I searched for few occurence of the words which would yield interesting results. Typing lock showed me there some interesting functions in the executable.
After retyping, 48 methods named isLocked showed up. I eventually realised that each Icon Mode corresponds to each of the method/function in here.
Above, I highlighted two things. With purple color, at the window title, there is total number of filtered function (equals to 43). The second one, highlighted in yellow is function size.
If you are wondering why the function size is different, it's pretty simple. The functions with size 8 are unlocked icons, while the size 11 are the locked icons.
The function size is different because of instructions. See the snippet below for explanation.
# For unlocked icons
100005815 48 89 e5 MOV RBP,RSP
100005818 31 c0 XOR EAX,EAX ; xoring EAX/EAX returns 0
# For locked icons
100004455 48 89 e5 MOV RBP,RSP
100004458 b8 01 00 MOV EAX,0x1 ; moving 1 to EAXIn the end, the pseudo-code of the function may look like this (plain C):
int isLocked()
{
return 1; // or 0 for unlocked icons
}Now, of course we could go through each function and patch MOV instruction to XOR via EAX, but doing that for all 43 icon modes seems pretty tedious. We can in fact patch instructions for those icons who actually seems interesting to us, since each isLocked method is called via Object Instance (ie. SFImageTemplateSD::isLocked, SFImageTemplateMiniDisc::isLocke) but that doesn't seem 1337, does it?
Therefore, Ghidra Scripts.
Ghidra Scripting
There are two options when we come to scripting. We can take the offset of each function named isLocked, and calculate return instruction (Bash, Ruby, Python). This direct patching is popular among Sublime Text tutorials. This time, we will use Ghidra.
Ghidra Scripts allows you to use underlying Ghidra API to automate some tasks in easier manner. It is quite powerful when it comes to malware researching and large input feedback. There are some extensions which allows you to write Ghidra Scripts in Python, but this time we will use official documented stuff, developed in plain Java.
Anytime in Ghidra, you can use Help menu button and then click Ghidra API Help. Ghidra documentation is awesome and very well documented even for begginers. Also, Ghidra comes with various default Scripts which you can view by clicking Window and then Script Manager.
You can run any script by clicking the Run Script button. Studying those will help you move forward, or give you clues anytime you came up to unknown problem.
Writing the script
The Ghidra Script Manager allows for rapid development of extended Ghidra functionality. Unlike conventional Ghidra plugins that require a full IDE for development, Ghidra scripts can be developed right inside of Ghidra while it is running. You can interactively change your script and immediately re-run it.
Lets create a new script. Select FunctionID from the Script Manager sidebar. Click on the New Script button (marked with purple), and select Java as preferred language.
We will give our script the shiny new name: FindPatchIsLocked.java. Ghidra spits our new editor window and autofill necessary imports and libraries.
Lets fix our run() method which will give us MD5 sum of the current program in use. Remember that run() function is mandatory for any Ghidra Script.
...
public void run() throws Exception {
println("Current Program MD5 Sum: " + currentProgram.getExecutableMD5());
}If we run the script, the Console window will yield correct output:
Successfully compiled: FindPatchIsLocked.java
FindPatchIsLocked.java> Running...
FindPatchIsLocked.java> Current Program MD5 Sum: f9a4cb81546dffa2f8d58c79d804e70f
FindPatchIsLocked.java> Finished!Now, we will write our new method for finding isLocked function. This method will accept Program which will be our currentProgram, and String as argument. The later will be used as to pass a function name.
...
private void findFunction(Program program, String name) {
// Define FunctionManager
FunctionManager functionManager = program.getFunctionManager();
FunctionIterator functions = functionManager.getFunctions(true);
// Iterate over all functions and find one which equals to `name`
for (Function function : functions) {
if (function.getName().equals(name)) {
println("Function " + name + " found!");
}
}
}Now fix the run() method to call this function.
...
public void run() throws Exception {
println("Current Program MD5 Sum: " + currentProgram.getExecutableMD5());
findFunction(currentProgram, "isLocked");
}The running of this script will yield positive results in the Console window.
That seems quite easy doesn't it? And it works! You can always reference to Ghidra API and default scripts for further details. For example, we can find Function model on the API Docs and see what methods are supported.
Lets look at the simple isLocked function in Dissasemble view. This one returns (bool)True, meaning it's locked.
int __stdcall isLocked(void)
bool RAX:8 <RETURN>
SFImageTemplateMiniDisc::isLocked XREF[1]: 10025f540(*)
10004c500 55 PUSH RBP
10004c501 48 89 e5 MOV RBP,RSP
10004c504 b8 01 00 MOV EAX,0x1
00 00
10004c509 5d POP RBP
10004c50a c3 RETWhat we need from a *.function enumeration is EntryPoint & Return value. This way we can compare if function returns 0/1 depending if it's locked or not.
Also, based on EntryPoint, we can define the offset of request regression, either MOV or XOR. Since EntryPoint returns Address, we can call .next() (4 times), or call add() which accepts displacement as a paremeter. From the documentation:
Address add(long displacement) Creates a new address (possibly in a new space) by adding the displacement to this address.
Lets .add() offset of value 4 (number of instructions) to get correct instruction which either MOV or XOR the retval.
private void findFunction(Program program, String name) {
...
for (Function function : functions) {
if (function.getName().equals(name)) {
...
Address offset = function.getEntryPoint().add(4);
println(offset.toString()); Now that we got our offset address, we can patch it. To do so, we need two things.
One is getInstructionAt() method which spits Address instruction. We will use Assemblers (API) that offers methods for Patching and fixing instruction of correct offset address.
With a bit of refactoring, our current findFunction method looks like this:
private void findFunction(Program program, String name) {
FunctionManager functionManager = program.getFunctionManager();
FunctionIterator functions = functionManager.getFunctions(true);
int i = 0;
for (Function function : functions) {
if (function.getName().equals(name)) {
i++;
Address offset = function.getEntryPoint().add(4);
Instruction ins = getInstructionAt(offset);
String log = i + ": Function " + name +
" found at: " + function.getEntryPoint() + "! " +
"Offset: " + offset.toString() + " * " + ins.toString();
println(log);
}
}
}When we run our script from script manager, we will get correct output in the console which displays few things: isLocked() function Address, an offset of the instruction block which returns 0 or 1, and the instruction at the offset. As you can see, two types of instructions are possible, both XOR and MOV, depending on retval. Blue selection represent locked modes, while green one are available for free.
While we at it, lets work on our patchInstruction method. Write a new function named patchInstruction which will accept offset address as an input. Also import necessary modules into your script. This method needs to implement Exception as per documentation, as given Address may not be available in the currentProgram.
import ghidra.app.plugin.assembler.Assembler;
import ghidra.app.plugin.assembler.Assemblers;
...
private void patchInstruction(Address atOffset) throws Exception {
}In this method, first we will obtain the Assembler bound for current program. You can check above documentation for detailed class references and API. The documentation states:
static Assembler getAssembler(Program program) Get an assembler for the given program.
This method will return Assembler interface of a given Program. Checking the documentation again, we have assemble method used in SleighAssembler implementation which accepts Address and instruction block (either array of strings, or newline-separated instructions). We will use this method to patch our code.
InstructionBlock assemble(Address at, java.lang.String... assembly) Assemble a sequence of instructions and place them at the given address.
In the end, our patchInstruction method will do two things:
- Get Assembler of the Program
- Assemble fixed instruction on the offset
- this will patch instruction from
MOV EAX, 0x1toMOV EAX, 0x0
- this will patch instruction from
Lets first check our ASM code on difference of isLocked function, both which retval is 1 and 0.
# => isLocked, 0
int __stdcall isLocked(void)
bool RAX:8 <RETURN>
SF*::isLocked XREF[1]: 1002517b0(*)
100003b2f 55 PUSH RBP
100003b30 48 89 e5 MOV RBP,RSP
100003b33 31 c0 XOR EAX,EAX
100003b35 5d POP RBP
100003b36 c3 RET
# => isLocked, 1
int __stdcall isLocked(void)
bool RAX:8 <RETURN>
SF*::isLocked XREF[1]: 100264608(*)
10006583a 55 PUSH RBP
10006583b 48 89 e5 MOV RBP,RSP
10006583e b8 01 00 MOV EAX,0x1
00 00
100065843 5d POP RBP
100065844 c3 RETAs you can see, the function isLocked which returns value 1 (Locked) has a few padding bytes. The difference is 3 bytes (31 c0 >> * 00 00 00). The MOV EAX, 1 requires 5 bytes. XOR EAX, EAX requires just 2 bytes (xor opcode(1) + operand(1) = 2). This means we shall not edit XOR instructions as there would be overlapping in bytes padding.
We can now create Assembler instance named asm and patch the opcodes to reflect above instruction. We will use asm.assemble method to change opcode on the given instruction. Our patchInstruction now looks like this:
private void patchInstruction(Address atOffset) throws Exception {
try {
Assembler asm = Assemblers.getAssembler(currentProgram);
asm.assemble(atOffset, "MOV EAX, 0x0");
} catch (Exception e) {
println("Unable to patch at offset: " + atOffset + " with err: " + e.toString());
}We need to fix our findFunction() method to skip XOR opcode in the iteration. This way, we will avoid overlapping XOR offsets with MOV instructions, as those Address already affects Icon modes (Unlocked).
private void findFunction(Program program, String name) {
...
for (Function function : functions) {
if (function.getName().equals(name)) {
// check if instruction is MOV
Boolean isMovOp = ins.toString().contains("MOV");
try {
// only patch instructions with MOV opcode
if (isMovOp == true) {
patchInstruction(offset);
}
...Finally, edit your logging info to affect the changes. The final code looks like this.
//This script finds isLocked function and patch the retval to 0.
//@author Halis Duraki
//@category FunctionID.Custom
//@keybinding
//@menupath
//@toolbar
import ghidra.app.script.GhidraScript;
import ghidra.program.model.util.*;
import ghidra.program.model.reloc.*;
import ghidra.program.model.data.*;
import ghidra.program.model.block.*;
import ghidra.program.model.symbol.*;
import ghidra.program.model.scalar.*;
import ghidra.program.model.mem.*;
import ghidra.program.model.listing.*;
import ghidra.program.model.lang.*;
import ghidra.program.model.pcode.*;
import ghidra.program.model.address.*;
import ghidra.app.plugin.assembler.Assembler;
import ghidra.app.plugin.assembler.Assemblers;
public class FindPatchIsLocked extends GhidraScript {
public void run() throws Exception {
println("Current Program MD5 Sum: " + currentProgram.getExecutableMD5());
findFunction(currentProgram, "isLocked");
}
private void findFunction(Program program, String name) {
FunctionManager functionManager = program.getFunctionManager();
FunctionIterator functions = functionManager.getFunctions(true);
int i = 0; // function index counter
for (Function function : functions) {
if (function.getName().equals(name)) {
i++;
Address offset = function.getEntryPoint().add(4);
Instruction ins = getInstructionAt(offset);
// check if instruction is MOV
Boolean isMovOp = ins.toString().contains("MOV");
try {
if (isMovOp == true) {
patchInstruction(offset);
}
} catch (Exception e) {
println("Error: Something happend while calling patchInstruction ...");
}
Instruction patchedIns = getInstructionAt(offset);
String log = i + ": Function " + name +
" found at: " + function.getEntryPoint() + "! " +
"Offset: " + offset.toString() + " * " +
ins.toString() + " >> " + patchedIns.toString() +
" patch: " + isMovOp.toString();
println(log);
}
}
}
private void patchInstruction(Address atOffset) throws Exception {
try {
Assembler asm = Assemblers.getAssembler(currentProgram);
asm.assemble(atOffset, "MOV EAX, 0x0");
} catch (Exception e) {
println("Unable to patch at offset: " + atOffset + " with err: " + e.toString());
}
}
}All you have to do is execute the script, and wait until completion. The script will automagically patch only those instructions for which isLocked function returns 1.
You have to admit Ghidra is truly a Reverse Engineering Framework. In the image below, I highlighted function address with Blue, offset of XOR/MOV Address with Yellow, and Green for information if instruction is patched or not.
Double-clicking on any Offset Address (in Console) will automatically send your cursor to given instruction; this way we can check if our script correctly patched the code. Lets have a look at Offset 0x10007c715 (last in Console). As you can see, the instruction was properly reversed.
...
10007c711 55 PUSH RBP
10007c712 48 89 e5 MOV RBP,RSP
10007c715 b8 00 00 MOV EAX,0x0
00 00
10007c71a 5d POP RBP
10007c71b c3 RETAll you have to do is build binary and codesign it. You may do so by checking end of my last post about patching Sketch for Unlimited Trial. Here is the end result:
For further resource, check out the references I've written at the end of the post which can help you understand more about Ghidra Scripting.
Hope you like it! Always available for help on my Twitter.
References:
- https://github.com/ghidraninja/ghidra_scripts
- https://github.com/ghidraninja/ghidra_scripts/wiki/Python-Scripting-Cheat-Sheet
- https://ghidra.re/courses/GhidraClass/Intermediate/Scripting_withNotes.html#Scripting.html