Nesting doll: unwrapping Vawtrak


Raul Alvarez

Fortinet, Canada
Editor: Martijn Grooten


Banking trojan Vawtrak wraps itself in layers, each of which gives rise to the next until a simple binary executable is exposed. Raul Alvarez likens Vawtrak to a nesting doll and in this article he unpacks the doll, looking at each layer in turn.

A lot of malware families perform their malicious activities within a single executable. Complete with body armour and shields, they load themselves into memory, decrypt just enough binaries, and then carry out their malicious actions under the protection of encryption and other stealth techniques. Thus it is hard to get a complete binary dump for simple analysis.

It is less common for a piece of malware to wrap itself in layers and at the heart of those layers, expose a simple complete binary executable – but that is exactly what a piece of malware known as Vawtrak does. In this article we will unravel Vawtrak’s layers, each of which gives rise to the next, just like a nesting doll.

Vawtrak, which is also known as Neverquest or Snifula, is a banking trojan that recently made the headlines because of its high level of sophistication. Initially it targeted only Japanese systems, but it has recently broadened its geographic scope [1]. Vawtrak spreads through drive-by downloads, attached to spam emails, and is also downloaded by other malware.

Other articles, such as [2], have looked at how the malware communicates with its C&C server and how it targets various banks. This paper looks specifically at how the malware installs itself persistently on a machine, frustrating both researchers and anti-virus products along the way.

Layer 1 (the outer doll)

In this section, we will look into the different algorithms performed by the outer layer (i.e. the malware itself) and at how it generates the executable binary that forms the next layer. The outer layer also contains the image file that is displayed once the malware is executed, to trick the victim into believing that is all the file does.

Simple anti-debugging trick

Anti-debugging tricks are common in modern malware. The variant of Vawtrak we looked at uses what appears to be a simple anti-debugging trick. There is nothing significant in its execution, and it doesn’t have a sneaky trick up its sleeve. However, when we look at how it is delivered, we see that it can easily stop heuristic scanning of the executable.

In its initial execution, Vawtrak calls the following APIs: GetModuleHandleA, InitCommonControlsEx and GetCommandLineA. None of these suggest that anything malicious is going to take place.

These instructions are followed by 1,600 (0x640) zeros (‘0’), where each pair of zeros is interpreted as ‘ADD BYTE PTR DS:[EAX],AL’, an irrelevant instruction that is repeated 800 times. This is a simple trick that could stop a simple anti-virus engine from emulating the malware.

Next, Vawtrak gets the address of the PEB (Process Environment Block). The third byte of the PEB is the ‘BeingDebugged’ flag, which has a value of TRUE if the process is run within the context of a debugger.

The malware doesn’t check specifically for the ‘BeingDebugged’ flag byte. Instead, it checks for the DWORD value from the third byte of the PEB, which contains the ‘BeingDebugged’ flag, the ‘SpareBool’ flag (reserved), and ‘Mutant’ DWORD value (reserved). Since the ‘Mutant’ value is a DWORD, only half of it is checked.

If the malware is being debugged, the DWORD value from the location PEB+2 will contain 0xFFFF0001. If this is not the case, the malware will continue execution.

After performing its anti-debugging check, the malware parses the stack memory to locate a return address that is located within kernel32.dll. Another check is then performed against the ‘BeingDebugged’ flag – in case the first anti-debug trick was unsuccessful. After performing a series of simple computations, the ImageBase of kernel32.dll is found.

String generator and API resolution

After finding kernel32’s ImageBase, the malware resolves the APIs it requires by performing the following routine:

  1. A simple string generator is used to produce the names of the APIs and libraries that it needs.

    The string generator starts by pushing DWORD values into the stack memory and adds each DWORD to another group of DWORD values. These DWORDs are constant in nature, pre-computed for specific strings such as ‘GetModuleFileNameA’.

  2. The malware parses the export table of the kernel32.dll library to look for an API name that matches the aforementioned string. When a match is found, the index of the API name is used to locate the address of the corresponding API.

Using the above routine, Vawtrak also resolves the following APIs: GetModuleFileNameA, CreateFileA, SetFilePointer, CloseHandle, ReadFile, WriteFile, GetTempPathA, LoadLibraryA and lstrcat.

Using the newly resolved LoadLibraryA API, Vawtrak loads the shell32.dll library into memory. This library name is also generated using the string generator. Using the technique described above, the ShellExecuteA API is also resolved from the shell32.dll library.

Simple API calling

When a regular application performs an API call, it does so by pushing all required parameters to the stack memory and then calling the API.

When, throughout its three layers, Vawtrak performs an API call, it first pushes all the required parameters to the stack, then it pushes the address of the API. Finally, it executes a RETN instruction, effectively jumping to the last value pushed, which is the address of the API.

At random locations between these instructions, NOP, DEC EDX and INC EDX instructions can be seen, which is another trick to make detection of the malware more difficult.

Main module

After a series of calls to the CreateFileA, SetFilePointer and ReadFile APIs that yield no meaningful results (which could either be a bug, or these calls could be intended as garbage code), Vawtrak gets the pathname of the current module using a call to the GetModuleFileNameA API and opens this path for reading using a call to the CreateFileA API.

This is followed by reading 260 (0x104) bytes from the physical file using a combination of the SetFilePointer and ReadFile APIs. These bytes hold the encrypted filename ‘mainOUT-crypted-5.exe’ , which will be used later for the dropped file. The filename is obfuscated using the single-byte XOR key 0xAA.

Using a combination of the GetTempPathA, lstrcat and CreateFileA APIs, Vawtrak creates the file ‘%temp%\ mainOUT-crypted-5.exe’ with GENERIC_WRITE access.

To fill this newly created file, the malware needs to decrypt a large block of data from the ‘overlay’ area of the original malware file. (The ‘overlay’ area of an executable file is not loaded into memory when the operating system executes the file.)

Vawtrak’s overlay area holds an encrypted copy of the executable binary that is used in the next layer. It is to be transferred and decrypted into the malware’s virtual memory space.

Decryption and the dropped executable file

Initially, the malware reads 1,024 (0x400) bytes of data from the overlay area into memory. Then it decrypts the bytes using a simple decryption algorithm with four byte-wise instructions: ‘SUB AH, 0x63’, ‘XOR AH, 0x42’, ‘XOR AH, 0x63’ and ‘NOT AH’. A variable number of NOP instructions are embedded in-between these instructions.

Upon decryption, the newly decrypted bytes are written to the previously created file, using the WriteFile API.

This routine is repeated until the full overlay area has been read and the decrypted file has been stored. It is then closed using the CloseHandle API.

Finally, Vawtrak executes ‘%temp%\ mainOUT-crypted-5.exe’ by calling the ShellExecuteA API.

Dropped image file

While ‘%temp%\ mainOUT-crypted-5.exe’ runs in the background, Vawtrak reads another 260 (0x104) bytes from the original malware into memory. Using the same simple XOR key, 0xAA, it generates the string ‘Diana-23.jpg’. The previously discussed set of instructions is then used to create an image file with this name in the temporary folder.

Finally, Vawtrak displays the image by calling the ShellExecuteA API; the operating system will automatically associate the file with the default image viewer. Figure 1 shows part of the image which is displayed once the main malware is executed. It also resembles the icon used by the original malware.

After displaying the image, the malware tries to generate another file but there seems to be a bug in generating the third filename, causing this part to fail.

Pixellated and cropped version of the image displayed

Figure 1. Pixellated and cropped version of the image displayed

Layer 2 (the second doll)

In this section, we will discuss the algorithms performed during the execution of the ‘%temp%\ mainOUT-crypted-5.exe’ file (the second ‘doll’ in our nesting doll analogy), which has been dropped by the outer layer.

Executing ‘mainOUT-crypted-5.exe’

First, Vawtrak resolves the GetProcessHeap and HeapAlloc APIs using the API resolution algorithm discussed below. This algorithm is also used to resolve the rest of the APIs needed by the malware.

API resolution algorithm

The API resolution algorithm starts with a series of function calls to locate the initial location of the PEB (Process Environment Block). Once the PEB has been located, Vawtrak parses every module name through the PEB’s linked list structure. Each module name is hashed using a simple mathematical calculation. Then, every computed hash value is compared against the hard-coded hash value (0xD56131B3) of ‘kernel32.dll’. Once the correct hash is found, the ImageBase of kernel32.dll is derived from the current structure of the PEB.

Next, the malware decrypts the name of the required API using a rotating key string (0xFA A7 E9 F4 44 9C DF 43 5D C8 FD) similar to the one used during the decryption of the large data block (which will be discussed in the next section).

Then Vawtrak gets the ImageBase of the ntdll.dll library using the same routine as was used to find the ImageBase of kernel32.dll; in this case the hard-coded hash value is 0xA6196EA7. The same routine is also used to find the address of the LdrGetProcedureAddress API (hash value 0xB110618C).

Finally, the LdrGetProcedureAddress API is used to get the addresses of the APIs needed by Vawtrak (e.g. GetProcessHeap and HeapAlloc).

Decrypting the large data block

Vawtrak allocates 200,192 (0x30E00) bytes of heap memory using a combination of the GetProcessHeap and HeapAlloc (which is similar to RtlAllocateHeap) APIs. It decrypts and copies a large chunk of memory from the ‘.data’ section of the malware.

The malware decrypts the aforementioned large block of data using the key string ‘y>;=>u*SzvwnmWnj’ (0x79 3E 3B 3D 3E 75 2A 53 7A 76 77 6E 6D 57 6E 6A).

The decryption algorithm looks complicated as a result of there being garbage code embedded within the routine. However, on looking closely, one sees that the only relevant computation is the ‘SUB EAX, EDX’ instruction. It simply subtracts the key byte (taken from the rotating key) from the encrypted byte. The rest of the instructions (the majority) are irrelevant to the actual algorithm.

This decryption algorithm is similar to the one discussed earlier for decrypting the API name. Figure 2 shows the decryption algorithm with the relevant instructions highlighted.

Decryption algorithm, part of the encrypted block and part of the decrypted block in heap memory. Only the highlighted instructions are relevant; the rest are garbage instructions.

Figure 2. Decryption algorithm, part of the encrypted block and part of the decrypted block in heap memory. Only the highlighted instructions are relevant; the rest are garbage instructions.

(Click here to view a larger version of Figure 2.)

Decompressing the decrypted data

The decrypted data is now located in the heap memory, albeit in a compressed form. In order to decompress it to its normal state, the malware needs to call the RtlDecompressBuffer API. This API is resolved using the same technique as was used to resolve the LdrGetProcedureAddress API, as explained previously.

After getting the address of the RtlDecompressBuffer API, Vawtrak prepares the heap by calling the RtlAllocateHeap API. This is followed by decompressing the decrypted data using the RtlDecompressBuffer API with the COMPRESSION_FORMAT_LZNT1 format.

Decompressing the data produces an executable file image with the proper MZ/PE header, the sections, and the rest of the complete unencrypted, unpacked executable.

Moving from heap to virtual memory

After the decompression, Vawtrak resolves the following APIs using the same API resolution algorithm as used earlier: LoadLibraryA, GetProcAddress, VirtualAlloc, VirtualProtect, VirtualFree and UnmapViewOfFile.

Then, Vawtrak parses the PE header of the newly decompressed executable file image to acquire the SizeOfImage value. After that, it allocates a section of virtual memory of this size by calling the VirtualAlloc API.

This is followed by using SizeOfRawData to compute the size of each section of the executable file image. Then, the malware copies the contents of each section into the newly allocated virtual memory (at address 0x8a0000).

After carefully copying the decompressed executable file image, byte by byte, section by section, adjusted using SizeOfRawData to obtain the physical file size, a researcher can dump it to a file.

Fixing the IAT

When a given application runs, the operating system is responsible for filling the IAT (Import Address Table) with the corresponding API addresses. However, if the binary file is loaded without proper Import Address translation, the IAT will contain RVAs (Relative Virtual Addresses) instead of the actual API addresses, and any call that is executed to an RVA will not work properly.

The malware’s executable file image that was just copied to the virtual memory has an IAT that only contains RVAs. Vawtrak fills the IAT with the actual API addresses by performing the following routine:

Initially, the malware parses the PE header to locate the import table. This is followed by loading the first library that is found, kernel32.dll, using the LoadLibraryA API. Then it gets the address of each API using the GetProcAddress API and saves them to the corresponding locations in the IAT. Once all the kernel32-related APIs have been resolved, it goes back to load the rest of the libraries and thus resolves all the required APIs.

Overwriting ‘mainOUT-crypted-5.exe’

Once the IAT has been filled with the proper API addresses, Vawtrak copies another 4,096 (0x1000) bytes of code to a different section of allocated virtual memory (0x890000) and transfers control to it.

Within the execution of the newly copied binaries, the malware changes the protection of the ‘mainOUT-crypted-5.exe’ module to PAGE_READWRITE using the VirtualProtect API. Then it clears the memory by overwriting the location at which ‘mainOUT-crypted-5.exe’ was stored with zeros, effectively removing it from memory.

This is followed by copying the contents of the virtual memory at 0x8a0000 to the original memory location of the ‘mainOUT-crypted-5.exe’ module. To recap: the content of the memory at 0x8a0000 is the decompressed executable image with properly filled IAT, that has just been generated.

After copying every byte from the virtual memory (0x8a0000), the section characteristics of the newly copied module are adjusted to take account of the new memory location.

Finally, the malware transfers control to the newly created module, which is the next layer of the malware.

Layer 3 (the third doll)

In this section, we will look into the third layer (the third ‘doll’ in our nesting doll analogy) – an executable binary produced from the second layer. The new executable binary contains no protection at all. The execution of this module is straightforward. Neither decryption nor hashing are used, and there is no digging around in a garbage bin.

In the malware’s current state, we can easily get a dumped copy of a complete executable binary file – but the malware is not done yet: it still has to reveal the last layer of the nesting doll.

Removing restrictions

Within the context of the third layer, Vawtrak’s first order of business is to remove software restrictions. These restrictions can be set through the software restriction policies of the operating system.

Some of these policies are accessed via the registry key ‘HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\Safer\CodeIdentifiers’.

Using the RegSetValueExA API, the following subkeys are set by Vawtrak:

  1. DefaultLevel = 0x40000 (unrestricted). This allows an administrator to define exceptions.

  2. TransparentEnabled = 1. This skips DLL checking.

  3. PolicyScope = 0. This makes policies applicable to all users.

After doing this, the malware deletes the key ‘HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\Safer\CodeIdentifiers\0\Paths\’, using the SHDeleteKeyA API. This basically removes any restrictions on any file under the ‘Path Rule’, which identifies whether an application is restricted or not.


After removing restrictions for any application, Vawtrak tries to restrict possible execution of some anti-malware applications using the routine described below.

Initially, it gets the %system% folder using the GetSystemDirectoryA API to extract the drive name, such as drive ‘C’. Then it adds the string ‘:\Program Files\’ to the drive name. Finally, it adds the anti-malware software name, resulting in the following string format: ‘[drivename]:\Program Files\[anti-malware name]’ (e.g. ‘C:\Program Files\[anti-malware name]’).

Vawtrak also looks into different folders by replacing the string ‘:\Program Files\’ with either ‘:\Program Files (x86)\’ or ‘:\Documents and Settings\All Users\Application Data\’.

Vawtrak calls the GetFileAttributesA API to check if the derived anti-malware’s pathname exists. If the pathname exists, the malware will generate a hash value for it. Then it will create the registry key ‘HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Windows\Safer\CodeIdentifiers\0\Paths\[hash value]’ with the subkeys ‘SaferFlags’ set to 0 and ‘ItemData’ set to the anti-malware’s pathname.

This newly created key will restrict the privileges and permissions of the given anti-malware application. It can also restrict execution of any applications under a given pathname.

Vawtrak will try to restrict the permissions granted to any anti-malware applications in its list.

Figure 3 shows a list of anti-malware names, the registry entries, and the warning message that appears when attempting to execute an application from a restricted folder.

A list of anti-malware names, the registry entries and the warning message displayed when attempting to execute an application from a restricted folder.

Figure 3. A list of anti-malware names, the registry entries and the warning message displayed when attempting to execute an application from a restricted folder.

(Click here to view a larger version of Figure 3.)

Generating the last doll

After restricting the execution of anti-malware applications, the malware is ready to unwrap its final component – the last ‘doll’.

First, Vawtrak accesses its resource section by calling the FindResourceA API with an RT_RCDATA (raw data) type parameter. This is followed by getting the size and the handle of the only resource found for this module, using a combination of the SizeofResource and LoadResource APIs.

After a new heap memory has been allocated using the HeapCreate and RtlAllocateHeap APIs, the malware copies 184,258 (0x2cfc2) bytes of raw data resource to the heap memory, byte by byte. Upon finishing this, the resource is set free by calling the FreeResource API.

A marker (‘AP32’) at the beginning of the raw data is checked to make sure that the right binaries are copied to the heap memory. Another check is made by calculating the hash of the whole raw data and comparing it to the value 0x24D2EDEA.

If these checks are successful, another block of heap memory is allocated. Vawtrak decrypts the raw data and copies it to the newly allocated heap. Another hash computation is performed against the newly decrypted data and the resulting hash is compared against the value 0x52194545.

The new heap now contains a new executable binary – the last ‘doll’ – in the form of a dynamic link library (DLL) file.

A new file, with a random filename and an extension name of ‘dat’, is created in the %appdata% folder using a combination of the SHGetFolderPathA and CreateFileW APIs. Afterwards, Vawtrak copies the contents of the new heap memory to the newly created file.

This is followed by creating an autostart registry entry, ‘[HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Run]’ with value ‘[regsvr32.exe /s ‘C:\Documents and Settings\All Users\Application\{random filename}.dat’]’, using a combination of the RegCreateKeyA and RegSetValueExW APIs. This makes sure that the malware maintains persistence upon a restart.

Finally, to enable the current module to use the functions exported by the dropped DLL file, Vawtrak loads it into memory using a call to the LoadLibraryW API with the parameter ‘C:\Documents and Settings\All Users\Application\{random filename}.dat’.

The usage of the DLL file is not covered in this article. In principle, any DLL file could be included.


Like a nesting doll, the first executable binary (outer doll) generates the second executable binary from its overlay section; the second executable binary (second doll) decompresses a big chunk of data to generate the third executable binary; the third executable binary `(third doll) uses its resource section to generate the final executable binary (last doll).

Each ‘doll’ (executable binary) has its own set of algorithms and functions that leads to the unwrapping of the next one. Every binary (except for the last one) has an important role to perform in generating the next one.

The ingenuity and skills shown by Vawtrak are not simple, but concise. Do not be deceived by a nice picture. Be vigilant and stay safe.


[1] Leyden, J. Vawtrak challenges almighty ZeuS as king of the botnets. The Register. December 2014.

[2] Wyke, J. Vawtrak – International Crimeware as a Service. Sophos. December 2014.



Latest articles:

VB2017 paper: Browser attack points still abused by banking trojans

With the ever-increasing use of banking-related services on the web, browsers have naturally drawn the attention of malware authors. They are interested in adjusting the behaviour of the browsers for their purposes, namely intercepting the content of…

Does malware based on Spectre exist?

It is likely that, by now, everyone in computer science has at least heard of the Spectre attack. Since many excellent explanations of the attack already exist, this article focuses on the probability of finding Spectre being exploited on Android…

EternalBlue: a prominent threat actor of 2017–2018

At the centre of last year's infamous WannaCry ransomware attack was an NSA exploit leaked by the Shadow Brokers hacker group, known as ‘EternalBlue’. The worm-like functionality of the exploit made a deadly impact by propagating to interconnected…

VB99 paper: Giving the EICAR test file some teeth

There are situations that warrant the use of live viruses. There are also situations where the use of live viruses is unwarranted. Specifically, live viruses should not be used when safer and equally effective methods can be used to obtain the…

Powering the distribution of Tesla stealer with PowerShell and VBA macros

Since their return more than four years ago, Office macros have been one of the most common ways to spread malware. In this paper, Aditya K Sood and Rohit Bansal analyse a campaign in which VBA macros are used to execute PowerShell code, which in…

Bulletin Archive

We have placed cookies on your device in order to improve the functionality of this site, as outlined in our cookies policy. However, you may delete and block all cookies from this site and your use of the site will be unaffected. By continuing to browse this site, you are agreeing to Virus Bulletin's use of data as outlined in our privacy policy.