The Encrypted Logz - Some Simple Reverse Engineering
2020-04-18 09:30:22 Author: parsiya.net(查看原文) 阅读量:128 收藏

I was looking at an application (not related to my day job) and I decided to reverse engineer how it creates logs. I cannot name the app (yet) but hopefully, this is useful.

I executed the app and saw events in procmon creating logs in.

  • C:\Users\Parsia\AppData\Local\[brand]\[app-suite-name]\Logs.

Opened a log file and it looked encrypted. Why encrypted? Look at all the jumble of data with high entropy. These kinds of files usually either compressed files or encrypted data.

How do we know it's encrypted? Compressed files usually have large headers. So the start of the file does not have high entropy. This one has a tiny header and we can see the word LOGZ.

Encrypted log Encrypted log

How do we figure out who writes to it? Procmon.

  1. Run procmon.
  2. Set filter.Path - begins with - path/to/log/directory.
  3. Run the app.
  4. See who edits the log.

procmon filter procmon filter

We can see the file access events. It's the main app.exe.

Log write events in procmon Log write events in procmon

Double-click on an event and select the Stack tab. It's using Qt5Core.dll 5.9.2.0.

Log write stack Log write stack

Having symbols is nice. Qt5 is an open source project and we can download its symbols.

Qt5 Symbols

You can find the symbols for version 5.9.2.0 at:

Because we are running an x86 app (32-bit) and it's a desktop version (not UWP), we are looking for this file qt-opensource-windows-x86-pdb-files-desktop-5.9.2.zip on the page above.

Inside the archive, there are a bunch of Windows 8.1 and Windows 10 files. We do not need the Windows 8.1 files.

Windows 10 files come in four flavors:

  • x86 vs. x64: We will choose 32-bit.
  • MSVC 2015 vs. MSVC 2017: How do we figure this out?

Qt PDBs Qt PDBs

Looking inside the app directory we can see all the Qt5 DLLs.

Qt DLLs Qt DLLs

We already know it's a 32-bit DLL. Scroll back up and look at the Address column in the procmon event stack image. It has only 4 bytes or 32-bits.

To detect the MSVC version we must discover the linker info.

Detect It Easy does the job and tells us it's linker version 14 or MSVC 2015 1.

'Detect It Easy' results 'Detect It Easy' results

We can discard the unneeded symbols. Let's put the correct PDBs inside C:\Symbols\@MySymbols. The symbol lookup is alphabetical so you put all of your symbols further up in the chain to make it faster (is this still a thing?).

Ghidra PDB Issues

As an example, I dropped Qt5Core.dll in Ghidra and loaded the PDB via File (menu) > Load PDB File. I got a bunch of errors

PDB> Unable to create stack variable buffer at offset -4 in [email protected]@@[email protected]@@@Z
PDB> Unable to create stack variable dt at offset -4 in [email protected]@@[email protected]@[email protected]@[email protected]
PDB> Unable to create stack variable indices at offset -4 in [email protected]@@[email protected]@Z
PDB> Unable to create stack variable buf at offset -4 in [email protected]@@[email protected]@[email protected]
PDB> Unable to create stack variable tzi at offset -4 in getRegistryTzi

Unfortunately, Ghidra's PDB import is not that great. To troubleshoot we can also, generate the XML file via the pdb.exe command.

  • ghidra_9.1.2_PUBLIC\Ghidra\Features\PDB\os\win64\pdb.exe c:\path\to\pdb -fulloutput > somefile.pdb.xml

This resulting xml can be imported in Ghidra instead of the PDB file. This is not going to help us here but it can show us where the issue is. It's also useful for creating for Ghidra usage in non-Windows operating systems.

The tool creates a 100MB file for Qt5Core.dll. Navigating to the file we can see that it's cut-off somewhere.

PDB xml for Qt5Core.dll PDB xml for Qt5Core.dll

There were also issues with the 2000 character limit for symbols. This is a known problem.

Importing from IDA to Ghidra

I also tried importing the PDB in IDA Freeware 7.0 and it worked perfectly.

PDB loaded in IDA Freeware 7.0 PDB loaded in IDA Freeware 7.0

In these situations, one option is to import the symbols in IDA and then use the Ghidra plugin to export the project. You can import it in Ghidra. This needs IDAPython which is not available in IDA Freeware.

At this point, I had enough to get started on decrypting the logs.

Now we can get into the main executable. Fortunately, (for me) the installer includes the PDB files for the main executables. This is great news.

PDBs included with the installer PDBs included with the installer

Dropping the main exe into Ghidra and the PDB resulted in a lot of errors. Similar to the errors above but it does not stop us. Then we can do Auto Analysis to continue. The result is not as great as IDA but it's great for paying 0 dollars.

How do we proceed from here? We can:

  1. Randomly search for things.
  2. Trace the writes to log files.

Just Guess Until You Find Something

When you encrypt a file, you have to be able to decrypt it too. Having symbols means we can search for decrypt (or encrypt). Decrypt is probably easier to trace because the log files are encrypted as the messages come but decryption usually happens in one go so everything might be in one place.

I searched for Decrypt and found the function that decrypts a log file named, well, DecryptLogFile.

DecryptLogFile function DecryptLogFile function

DecryptLogFile appears to be a wrapper for FUN_008f6100.

string __cdecl DecryptLogFile(path *param_1, path *param_2)
{
  FUN_008f6100(param_1,param_2);
  return param_1;
}

The parameters are of type boost::filesystem::path. We can guess one of them is the encrypted log file and the other is where the decrypted results will be stored. Let's look at the assembly for FUN_008f6100. I have cleaned up the decompiled code.

Notice the difference between the number of parameters above and in the function? I am not sure where param_1 comes from? Maybe it's a field and FUN_008f6100 is actually a method (i.e., thiscall)?

void FUN_008f6100(std::string *param_1, path inputFile, path outputFile)
{
    QIODevice::QIODevice qioDevice [8];
    inputFileString = QString::fromStdWString((std::basic_string *)&inputFile);
    QFile::QFile(qioDevice, inputFileString));
    
    isInputFileOpened = open(qioDevice,(QFlags<enum_QIODevice::OpenModeFlag>)0x1);
    if (isInputFileOpened == false) {
       cout << "Can\'t open input file";
    }
    else {
        inputSize = size(qioDevice);
        if ( inputSize > -1) {
            inputDataStream = QDataStream::QDataStream(qioDevice);
            fourCharString = "    ";
            fourCharStringPtr = *fourCharString;
            QDataStream::readRawData(inputDataStream, fourCharStringPtr, 4);
            if (fourCharStringPtr == c_header) // c_header = LOGZ {
                // Read four more bytes
                versionString = QDataStream(inputFileStream, 4);
                if (versionString == 0x2010ab01) {
                    outputFileString = QString::fromStdWString((std::basic_string *)&outputFile);
                    outputFile = QFile::QFile(outputFileString));
                    isOutputFileOpened = QFile::open(outputFileStream,(QFlags<enum_QIODevice::OpenModeFlag>)0xa);
                    if (isOutputFileOpened == false) {
                        cout << "Can\'t open output file";
                    } else {
                        // This is where decryption happens.
                        FUN_008f6bd0(param_1, inputDataStream, outputFileStream);
                    }
                else {
                } else {
                    cout << "Bad version";
                }
            } else {
                cout << "Bad header";
            }
    return;
}

This function does some sanity checks.

  1. Checks if the input file exists and can be opened.
  2. Reads the first four bytes and compares them with LOGZ. If not prints Bad Header.
  3. Reads the second four bytes and compares them with 0x2010ab01. If not prints Bad Version.
  4. Creates the output file.

We have already seen the header and version bytes in the encrypted log file above. They appear at the start of every encrypted log file, too.

Our next destination is FUN_008f6bd0(param_1, inputDataStream, outputFileStream) where the main decryption happens.

void FUN_008f6bd0(std::string param_1, QDataStream::QDataStream inputData, QDataStream::QDataStream outputData)
{
    // Initialize the PRNG.
    mersenne_twister_engine<unsigned_int> mtPRNG [2500];

    // See it with the default seed (0x1571). This is what the constructor does.
    seed(mtPRNG, &default_seed);
    // Removed
    appSeed = 0x25082011;
    // Seed the PRNG with the correct seed.
    seed(mtPRNG,&appSeed);
    // Get a buffer of 20000 bytes.
    // Removed code creates an output buffer named outputBuffer. 
    ByteArray((ByteArray *)&bufferByteArray20000,0x20000,'\0',0);
    while( true ) {
        // Try and read 20000 bytes.
        numberOfBytesRead = *(int *)(bufferByteArray20000 + 8);
        // Get a wrapper with the input buffer
        inputBuffer = (char **)Get((DataWrap *)&bufferByteArray20000);
        // Read 20000 bytes.
        numberOfBytesRead =
                QDataStream::ReadRawData(inputData,*inputBuffer,numberOfBytesRead);
        // Pass them to `FUN_008f6370`.
        FUN_008f6370(inputBuffer, outputBuffer, numberOfBytesRead);
        if (numberOfBytesRead < 1) break;
    }
    // removed

mersenne_twister_engine<unsigned_int, ... is a PRNG.

It's not cryptographically secure but we are encrypting some random local log files. It's seeded twice. First one is the default_seed followed by 0x25082011. The default seed is 0x1571 and the constructor does it automatically.

#include <random>
// Constructor automatically seeds it with 0x1571 and then 0x25082011.
mt19937 generator(0x25082011);

A 20000 byte buffer is read from input and passed to FUN_008f6370. This function is a wrapper for FUN_008f6370:

undefined4 FUN_008f6370(ByteArray* inputBuffer, byte* outputBuffer, uint numberOfBytesRead)
{
  FUN_008f5d90(inputBuffer, outputBuffer, numberOfBytesRead);
  return param_1;
}

FUN_008f5d90 is the most interesting function. We see our twister again.

Mersenne Twister Engine Mersenne Twister Engine

Further down we see references to ..\\..\\..\\src\\RcLogV2\\Output\\LogEncFileOutputV2.cpp and some classes that look interesting thanks to assert statements.

Assert statements Assert statements

If we search for local_80 in the decompiled code, we see this interesting function boost::random::mersenne_twister_engine::generate_uniform_int.

generate_uniform_int generate_uniform_int

Surprisingly, I could not find it in the docs but managed to find it in the hpp files.

emplate<class Engine, class T>
T generate_uniform_int(
    Engine& eng, T min_value, T max_value,
    boost::mpl::true_ /** is_integral<Engine::result_type> */)
{

It generates uint32 values (because we are using the int32 version of the Mersenne Twister) between min and max values. The function call in assembly is:


008f5fb3 8b 5d ec        MOV        EBX,dword ptr [EBP + local_84] ; param_1
        LAB_008f5fb6                                    XREF[1]:     008f5fe3(j)  
008f5fb6 8b 45 f0        MOV        EAX,dword ptr [EBP + local_80] ; mtPRNG
008f5fb9 53              PUSH       EBX
008f5fba ff b0 c8        PUSH       dword ptr [EAX + 0x9c8] ; *mtPRNG+2054
            09 00 00
008f5fc0 ff b0 c4        PUSH       dword ptr [EAX + 0x9c4] ; *mtPRNG+2058
            09 00 00
008f5fc6 50              PUSH       EAX ; mtPRNG
008f5fc7 e8 24 f5        CALL       boost::random::detail::generate_uniform_int
            ff ff

Which is generate_uniform_int(mtPRNG, *mtPRNG+2054, *mtPRNG+2058, param_1);. After the function call we see:

008f5fcc 8b 55 78        MOV        EDX,dword ptr [EBP + param_2]
008f5fcf 8d 7f 01        LEA        EDI,[EDI + 0x1]
008f5fd2 83 c4 10        ADD        ESP,0x10
008f5fd5 8a 0a           MOV        CL,byte ptr [EDX]
008f5fd7 42              INC        EDX
008f5fd8 32 c8           XOR        CL,AL
008f5fda 89 55 78        MOV        dword ptr [EBP + param_2],EDX
008f5fdd 88 4f ff        MOV        byte ptr [EDI + -0x1],CL
008f5fe0 83 ee 01        SUB        ESI,0x1
008f5fe3 75 d1           JNZ        LAB_008f5fb6 ; go back up to call `generate_uniform_int`

This is a typical XOR. Counter is esi (note SUB ESI,0x1 in the end). When esi is not zero, JNZ sends the execution back up to the function call to get another uint32. When esi reaches zero, the jump is not taken and we finish the loop. By tracing the app we see that edi is actually param_3. param_2 is probably the plaintext, param_3 is the size of ciphertext/plaintext.

for (int i=0; i<sizeOfPlaintext; i++) {
    // Get a uint32 from the mt19937 generator.
    key = generate_uniform_int(mtPRNG, *mtPRNG+2054, *mtPRNG+2058, param_1);
    ciphertext[i] = plaintext[i] XOR (AL)key;
    // Decompiled code here is `ciphertext[i] = plaintext[i] XOR (byte)key;` which is wrong
}

The Encryption Algorithm

The XOR is tricky here. We are XOR-ing AL with the plaintext. The decompiled code in Ghidra is ciphertext[i] = plaintext[i] XOR (byte)key; which is wrong.

If we go by decompiled code and cast the uint32 to byte we get the least significant byte. However, when numbers are stored in a register like eax AL will point to the most significant byte so (key & 0xFF000000) >> 24 does the job.

void FUN_008f5d90(ByteArray *param_1,byte *param_2,uint param_3)
    mersenne_twister_engine<unsigned_int> *mtPRNG;
    // removed
    for (int i=0; i<sizeOfPlaintext; i++) {
        // Get a uint32 from the mt19937 generator.
        key = generate_uniform_int(mtPRNG, *mtPRNG+2054, *mtPRNG+2058, param_1);
        ciphertext[i] = plaintext[i] XOR ((num & 0xFF000000) >> 24);
    }
    return ciphertext;
}

Min and Max Values

The generate_uniform_int function call has two parameters min and max. These point to the minimum and maximum values for the generated numbers. For std::mersenne_twister_engine these are 0 and 2^32 - 1 (for int32 version of the function).

These cannot be set for std::mersenne_twister_engine according to the docs. In the docs for std::mersenne_twister_engine::min we read:

Minimum value: Returns the minimum value potentially returned by member operator(), which for mersenne_twister_engine is always zero.

Similar for and max:

Returns the maximum value potentially returned by member operator(), which for mersenne_twister_engine is 2^w-1 (where w is the word size specified as the second class template parameter).

It seems like it's pointing to object fields (?) or methods (?). That means we do not need to define min and max values in our code.


The Bad Route

I initially started writing this route for this write-up. I was ashamed that I had found the original solution by searching for a function name. Then I realized why not?

This is the long route and not fun BUT I discovered some valuable info here.We will see how debug mode is detected and how console access is enabled.

For the long route, I started with the event I saw in procmon.

Qt5Core.dll > QFSFileEngine::write + 0x29. Searching for it in Ghidra or in the exports for Qt5Core.dll returns nothing. Searching online we can find references to it in the v4.8 docs.

Note: I do not know why it's not in the symbols. So I decided to do some dynamic analysis:

  1. Run the app via x64dbg.
  2. Go to the address for the procmon event.
    1. The new address in that run was 0x6f87b6b9.
  3. Go a bit up to see the call to [email protected]@@[email protected].
  4. Put a breakpoint and rerun the app.
  5. Go to the Call Stack tab in x64dbg after the breakpoint is reached.

Call to write in x64dbg Call to write in x64dbg Call stack in x64dbg Call stack in x64dbg

Now we can search for LogFileOutputBase in Ghidra.

LogFileOutputBase LogFileOutputBase

This is a method that writes the data passed in the parameter to the log file. I did not see any encryption in the source code. Meaning the parent class might handle it.

Looking at the incoming references (bottom left in Ghidra), we can see:

Incoming References - LogFileOutputBase
    FUN_008f5720 <-- This references LogFileOutputBase
        CreateLogEncFileOutput <-- FUN_008f5720 calls this
    FUN_008f76c0  <-- This references LogFileOutputBase
        CreateLogFileOutput <-- FUN_008f5720 calls this

The parent class is created by using one of the two methods above. Just by looking at the names we can see that one creates an encrypted log file and the other does not.

Let's go to CreateLogEncFileOutput. It has one parameter (string) and returns a FileResult object. I could guess that it creates a log file and returns a handle to it. Then we can write to the handle and not worry about the underlying encryption.

FileResult __cdecl CreateLogEncFileOutput(char *param_1)
{
    // Removed
    local_30 = operator_new(0x9fc); // local_30 = new int(0x9fc);
    this = (LogFileOutputBase *)FUN_008f5720(); // What is this?
    FileName(this);
    shared_ptr<class_RcLogV2::IOutputDevice><class_`anonymous_namespace'::LogEncFileOutput>
            ((shared_ptr<class_RcLogV2::IOutputDevice> *)&stack0xffffffb0,(LogEncFileOutput *)this);
    FVar1 = FileResult((FileResult *)param_1,in_stack_ffffffb0,in_stack_ffffffb4);
    // Removed
    return FVar1;
}

It looks like FUN_008f5720 creates and returns the filehandle. We see some interesting stuff in it.

FUN_008f5720 FUN_008f5720

We have already seen mersenne_twister_engine<unsigned_int, ... so I will skip explaining it this time. An educated guess tells us that the PRNG is used to create a key to encrypt the file. But we do not know how, yet. A pointer to this object is stored in local_60.

After checking that it can create or open the file, we get to this code. The decompiled code in Ghidra gives us the general idea of what's happening but is not correct.

        LAB_008f5886                                    XREF[1]:     008f5821(j)  
008f5886 57              PUSH       EDI
008f5887 8d 4e 14        LEA        ECX,[ESI + 0x14]
008f588a ff 15 1c        CALL       dword ptr [->QT5CORE.DLL::?[email protected]
            c2 f1 01

This calls void QDataStream::setDevice(QIODevice *d) which sets the I/O device to most likely the log file we just created. This means we can now use the QDataStream object to write to it.

008f5890 6a 04           PUSH       0x4
008f5892 ff 35 e0        PUSH       dword ptr [->s_LOGZ]            = 030fa850
         3a 87 03                                                   = "LOGZ"
008f5898 8d 4e 14        LEA        ECX,[ESI + 0x14]
008f589b ff 15 7c        CALL       dword ptr [->QT5CORE.DLL::?[email protected]
            c0 f1 01

Calls int QDataStream::writeRawData("LOGZ", 4). This writes LOGZ to the start of the file. This is exactly what we saw above.

008f58a1 68 01 ab        PUSH       0x2010ab01
            10 20
008f58a6 8d 4e 14        LEA        ECX,[ESI + 0x14]
008f58a9 ff 15 78        CALL       dword ptr [->QT5CORE.DLL::??6[email protected]@QAEA
            c0 f1 01

This call writes 0x2010ab01 to the stream. This also appears in the log file but because it was not in printable hex I did not notice it.

IDA actually figures out what method it is.

QDataStream << 0x2010ab01 writes a the byte to the stream.

Next we have:

008f58af 8b 07           MOV        EAX,dword ptr [EDI]
008f58b1 8b cf           MOV        ECX,EDI

008f58b3 8b 40 38        MOV        EAX,dword ptr [EAX + 0x38]
008f58b6 ff d0           CALL       EAX
008f58b8 83 f8 08        CMP        EAX,0x8
008f58bb 0f 85 bb        JNZ        LAB_008f597c
            00 00 00

Honestly, I cannot figure out what it exactly does but we have written 8 bytes and there is a compare with 8. It's either calling a method or accessing a field. The most likely scenario is that it checks if it has written 8 bytes to the stream.

// Create the datastream.
?[email protected]@@[email protected]@@Z(param_1 + 5,(QIODevice *)this);
// Write the header
?[email protected]@@[email protected](param_1 + 5,c_header,4);
// Write `0x2010ab01`. 
??6[email protected]@[email protected]@Z(param_1 + 5,0x2010ab01);
// Get the number of bytes written
lVar5 = (**(code **)(*this + 0x38))();
// Compare it with 8.
if (lVar5 != 8) {
    local_64 = (undefined4 *)0x0;
    local_8._0_1_ = 10;
    // If it was wrong, return an error message.
    local_5c = string("Bad header size");

The rest of the method is not interesting. So we have figured out the PRNG and the header. But we do not know how the encryption happens.

Returning into CreateLogEncFileOutput we see that a FileResult is created and returned.

FileResult __cdecl CreateLogEncFileOutput(char *param_1)
{
    // Removed
    local_30 = operator_new(0x9fc); // local_30 = new int(0x9fc);
    local_8 = 0;
    this = (LogFileOutputBase *)FUN_008f5720();
    // We are now here.
    FileName(this);
    // Removed
    FVar1 = FileResult(this);
    // Removed
    return FVar1;
}

Now we need to follow the returned FileResult and see how data is written to it.

Select the CreateLogEncFileOutput function, right-click and Find references to we get to FUN_013697c0. I have cleaned up the code and removed the unnecessary parts.

void __thiscall FUN_013697c0(void *param_1, undefined4 *param_1_00,
    LogDirections *param_2, undefined4 param_3, int param_4)
{
    // This checks if we are running in debug mode.
    isDebugMode = ?[email protected]@@YA_NXZ(param_1);
    if (isDebugMode) {
        // If running in debug mode create plaintext log.
        logFile = CreateLogFileOutput((char* local_28);
    }
    else {
        // If not debug mode create encrypted log.
        logFile = CreateLogEncFileOutput(char* local_68);
    }

    // removed
    if (isDebugMode) {
        ~FileResult(local_28);
    }
    // removed
    if (!isDebugMode) {
        ~FileResult(local_68);
    }

    // File permissions?
    param_1_00[4] = 0;
    param_1_00[5] = 0;
    param_1_00[5] = 7;
    param_1_00[4] = 0;
    puVar5 = param_1_00;
    if (7 < (uint)param_1_00[5]) {
        puVar5 = (undefined4 *)*param_1_00;
    }
    // removed
    return;
}

Seems like it creates a log file based on whether we have debug mode enabled or not. Let's rename it to CreateLogFile. There is not much in the rest of the function. As a detour let's check the debug mode.

Debug mode

Click on the [email protected] function.

bool __thiscall ?[email protected]@@YA_NXZ(void *this)
{
  if ((g_debugEnabled == (optional<bool>)0x0) &&
     (DAT_03b34bfd = (bool)FUN_0132a220(), g_debugEnabled == (optional<bool>)0x0)) {
    g_debugEnabled = (optional<bool>)0x1;
    return DAT_03b34bfd;
  }
  return DAT_03b34bfd;
}

It checks if the global variable g_debugEnabled is not zero. If it's zero, it returns with what we can assume is false.

If it's not zero, it calls FUN_0132a22. AL (the return value of this function is in EAX and AL is the first byte) is stored in DAT_03b34bfd and returned. It also sets g_debugEnabled to 1. Otherwise, DAT_03b34bfd is returned without being set.

So FUN_0132a22 sets debug mode. Let's check it. Cleaned up decompiled code tells us what it does:

void FUN_0132a220(void)
{
    // Get the name of the executable.
    executableName = Utils::ExeName::Instance();
    // Get the executable directory.
    executableDirectory = Utils::ExeName::GetExeDirAsStr(executableName);
    // debugFilePath = "path/to/executable/directory/debug"
    debugFilePath = concat(executableDirectory, "debug");
    // Return the file_status of debugFilePath
    fileStatus = boost::filesystem::detail::status(debugFilePath, system::error_code& ec);
    // If the status is not `status_error` (0) and is not `file_not_found` (1)
    if ((fileStatus[0] != 0) && (fileStatus[0] != 1)) {
        // Show the console window.
        ShowConsolWindow();
    }
    return;
}

fileStatus is of type Boost::Filesystem::file_status and it can contain values of enum Boost::Filesystem::file_type.

enum file_type
{
    status_error, file_not_found, regular_file, directory_file,
    symlink_file, block_file, character_file, fifo_file, socket_file,
    type_unknown
};

0 and 1 are status_error and file_not_found respectively. The code explicitly checks if the result is 0 or 1.

In short, it checks if there is a file named debug in the executable directory. Enables debug mode and shows the console window.

To figure out what the console window is. Click on ShowConsolWindow. It prints the app output to the console. If you run the app from a command line and you have debug mode enabled you will see messages printed to the console:

void __cdecl ShowConsolWindow(void)
{
    Console *this;
    ulong uVar1;
    int iVar2;

    this = Win32::Console::Instance();
    if (*this == (Console)0x0) {
        uVar1 = Win32::ToolHelp::GetParentProcessId(0);
        iVar2 = Win32::Console::RedirectIOToConsole(uVar1);
        if (iVar2 == 0) {
            Win32::Console::Allocate(this);
            return;
        }
        Win32::Console::RedirectIOToConsole(this);
        *this = (Console)0x1;
    }
    return;
}

The debug mode is used in many places. Incoming references for the [email protected] are:

Functions that check for debug mode Functions that check for debug mode

Back to Hunting for Encryption

After our little side-quest, we are back here. We know CreateLogEncFileOutput creates an encrypted log file and returns a FileResult handle. It's referenced in FUN_013697c0/CreateLogFile which we saw it checked for debug mode and returned a log file accordingly (plaintext vs. encrypted).

Now we need to figure out what happens to the FileResult and how it's handled. Following CreateLogFile we can see it's called by FUN_01369160 which is a wrapper. We can rename it to CreateLogFileWrapper.

void CreateLogFileWrapper(undefined4 param_1,undefined4 param_2,undefined4 param_3,undefined4 param_4)
{
    // Removed
    return CreateLogFile(param_1,param_2,param_3,param_4);
}

This is in turn called by FUN_01368870.

Call stack Call stack

Rage Quit

I went down the rabbit hole and wasted a few hours trying to trace the calls with nothing to show. The only nice thing was finding how to enable the debug mode.

Then I realized there is an easier way of doing what I have been trying to do. We know we are using a mersenne_twister_engine and it's a somewhat unique string. Why not search for it in the binary?

I selected an instance that I saw above then did Right-click > References > find uses of.

The result has 135 locations but most of them are probably internal. We should start with the ones with FUN_00 labels because they are the most "fun" (dadjoke.png).

Mersenne uses Mersenne uses

And we find it in FUN_008f5d90. local_80 is the mt19937 variable. We have seen this function before above. But this time it's in the context of decrypting the logs. XOR is transitive so encryption and decryption do the same.

Now we can write code. I could not find implementations of mt19937 32-bit in Go. Python supposedly uses the same engine to generate random output but I did not get the result I wanted either. So I had to write C++ code.

Not having written C++ I spent a lot of time creating useless functions to warm up and get used to vectors. I have used the std::byte type in the code so you need to use the instructions in the code to use C++17.

The final code is at:

Decrypted log Decrypted log

  1. Installers might include symbols with the main executable.
  2. Symbols are fun.
  3. Ghidra's PDB processing has some issues.
  4. The Mersenne Twister Engine is part of the C++ standard library.
  5. Debug mode can be set with some silly methods (e.g., having a file named debug).
  6. Ghidra's decompiler is sometimes misleading so be sure to check the disassembly and not trust it blindly.

文章来源: https://parsiya.net/blog/2020-04-17-the-encrypted-logz-some-simple-reverse-engineering/
如有侵权请联系:admin#unsafe.sh