The Story of a Trojan Dropper III

Introduction:

In previous posts (story of Trojan dropper part I, II) we performed both static and dynamic analysis on the threat and also developed a broad idea of what the malware was doing in it’s initial stages. We now set out to investigate the reason behind the crash of the dropped file (“Adobe.exe”) and at the same time we want to retrieve the payload from this particular malware sample.

Analysis:

Let’s go ahead and debug the file. This file (“Adobe.exe”) was found to be packed (custom variant) and the unpacking routine is similar to the approach detailed in part II of this post, so let’s skip all the gory details and dive right into it.

After unpacking the entire binary in memory, control is transferred to a newly unpacked region at the address "0x401634". Now the action begins from here and we get our first decryption routine, which is a simple 1-byte XOR with a static key as seen in the image below which reveals more code

Figure 1: Decrypt code

If we carefully look at the new code below, we can observe that it is riddled with JMP instructions and occasionally we find a few garbage values. The code is crafted in such a way that it resists disassemble attempts which makes reverse engineering the sample more difficult and somewhat painful.

Hidden in between these JMP’s in plain sight is an instruction that is familiar to us by now(i.e the instruction to fetch the address of PEB (Process Environment Block)).

Figure 2: Fetch address of PEB

At this point we can only speculate on why the malware needs the PEB address. Moving further in the code, we encounter another decryption point that is exactly same as the first one, which decrypts more code. Now the control is transferred to the newly decrypted code.

Figure 3: NtGlobalFlag (anti-debug)

In the image above, we have an interesting bit of code. Recall that the malware already collected the address of PEB. At this point, using the instruction CMP DWORD PTR DS:[EDI+68], ECX (here EDI holds the address of PEB which is "0x7FFD4000" and ECX holds a constant 0x70 ) a comparison is performed, after which a JNZ instruction decides the fate of the control flow.

Here we get our first glimpse of the anti-debugging technique employed by this malware. The instruction CMP DWORD PTR DS:[EDI+68], ECX compares the value of the ECX register (i.e. 0x70) to the location in the PEB structure known as “NtGlobalFlag”. The field is set to a value of “0x70”, if the process is spawned under a debugger.

In our case, this is true since we are debugging the file. Finally, the JNZ instruction at "0x401511" is not taken, which lands us in an invalid region in memory, thus triggering the anti-debugging. Let’s jump here and continue with our analysis. We now have another layer of decryption, which reveals more code after which, as usual, control is transferred to this region.

We then reach another piece of code, which is revealed only if we keep up with the control flow.

Figure 4: File-path, name identification (anti-debug)

The above code uses “strstr” to look for a string named “sample” anywhere in file path of our currently debugged file. If found, “strstr” returns a pointer to first occurrence of search string(i.e "sample") in file-path, or else it returns zero. The malware then checks the return value in EAX and takes a conditional jump in the form of JE instruction at "0x401135". If the jump is not taken, the code lands in “ExitProcess”, a call which terminates the process.

This is the second anti-debugging technique the malware employs, although not-an effective one in my opinion considering the odds of a file-path or malware name containing the name “sample”.

Figure 5: GetVolumeinformation, Volumeserial (anti-debug)

Again, here we have another anti-debugging technique where the malware retrieves the Volumeserial number using the API GetVolumeinformation and compares it with  “0CD1A40” and “70144646”. If either comparison matches, the code jumps to the ExitProcess call.

Another anti-debugging technique follows immediately thereafter in the form of “EnumSystemLocalesA”. The first argument that "EnumSystemLocalesA" accepts is a pointer to the callback function. Here the malware does a neat trick. If we look at the code below, at address “0x401179” a constant “0x2” is pushed onto the stack, which is followed immediately by a CALL. 

When this CALL is executed, it pushes the return address onto the top of the stack (which is "0x401180") and the EIP (Instruction pointer) now lands at the "EnumSystemLocalesA" call. Now if we observe the stack, the value on top of the stack is the return address, which naturally becomes the callback function address for "EnumSystemLocalesA" . When the "EnumSystemLocalesA" API is executed, control falls to the callback function, which continues the execution of the code. 

Figure 6: EnumSystemLocalesA (anti-debug)

Let's now continue our debugging from the address “0x401180”. Not far from here, yet another anti- debugging technique is uncovered. This time the malware retrieves the ‘Diskname’ from the registry. 

Figure 7: Diskname (anti-debug)

"RegOpenKeyExA" (HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\services\Disk\Enum)  and  "RegQueryValueExA" then uses “strstr” and searches for signs of a virtual environment such as “Xen”, “Vmware”, “Qemu” and also looks for string “virtual”.

Figure 8: Search vm strings(anti-debug)

Once the malware detects a virtual environment, it changes the control flow and land us in an invalid memory region. If we play along, we can observe something interesting here.

Figure 9: Adobe.exe crash

There we go. We can finally reproduce the crash that happened during the first stage of our analysis.
Ok, so a through binary analysis is often not so straight forward after all !!!

Figure 10: sandbox identification(anti-debug)

Let’s move on and see what surprises the malware has this time. Below, we can see anotherpiece of code that checks for a DLL name called “sbiedll” . Here, the malware checks whether it is run in a popular sandbox called “sandboxie”. If found, as usual, the malware bails out.


Immediately afterward, the malware performs a decryption routine to reveal compressed data, which is again packed with aplib and below we can observe the aplib unpacking routine which decompresses the data to a PE file.

Figure 11: Aplib decompression routine

Soon after, control is passed to the newly decompressed file. Below, we can see the new file being executed.

Figure 12: New PE-file in memory

Debugging further, we can observe the malware revealing the final trick that it has up it’s sleeve ,which I will explain in the next post in this series.

The story of a Trojan Dropper II

Analysis:

Lets analyze the PE file in detail and see what it’s up to. Like most malware, this sample was packed and in order to properly analyze it, we must begin by unpacking the binary. Keeping this in mind, I began by debugging the file, hoping to find the reference to the data section in order to determine precisely where the encrypted portion of data was to be found.

Fortunately, I was not disappointed and was soon able to find the reference point.

Figure 1: Obfuscated data 

After further debugging, we are able to see the code decrypted in memory. The decryption occurs in multiple iterations, until the data is completely decrypted.

Figure 2: Decryption of obfuscated data

Now we have a full view of the decrypted code in memory. The portion that was decrypted contains position independent code (i.e shellcode).

Figure 3: Decrypted data in memory

Since the code is decrypted in memory, we can assume that at some point, control will be transferred to that region, which in this case happens immediately. We also can see that the VirtualProtectEx API is used to change the protection of the memory region and by doing so the malware will be able to execute and manipulate the memory.

Figure 4: Change memory protection

After this occurs, control is transferred to the region by an instruction of JMP EDI. Here, EDI will hold the address to which EIP (instruction pointer) lands and we can see that it is the same portion of the 

code that was decrypted earlier.

Figure 5: Control transferred to new code

There's an interesting bit of code here if we look at first couple of instructions on the landed region. We can see a NOP instruction, followed by SUB EAX,EAX and a CALL and POP EBX. If we carefully observe the address that is called, it is that of the POP EBX. This is a common technique found in shellcode and file infectors where one needs to get the address of the region that is currently being executed.

When this CALL is executed, it pushes the return address onto the stack (in this case it is the address of POP EBX). Now POP EBX is executed, as that instruction pops the value from the top of the stack to EBX. The address is then added to a constant of 0x33, to point to the region that is then decrypted by the decryption loop. This reveals more code, after which a JMP instruction transfers the control to the newly revealed code.

Further, I was able to identify another interesting piece of code here. The code below retrieves the address of the PEB (process environment block) and navigates to PEB_LDR_DATA->InLoadOrderModuleList, where it retrieves the names of the loaded modules (DLL’s) .

Figure 6: Fetch base address of kernel32.dll

There's another catch here. The malware looks for specific DLL’s (in this case kernel32.dll), but instead of using the string kernel32.dll to compare with retrieved module names from the PEB, it carries the hash of the DLL names and then calculates the hash value for the retrieved module names and compares them. This allows the malware to make minimum noise and avoid some antivirus rules.

Figure 7: Dll name hash 

Once the malware gets kernel32.dll, it then retrieves the base address of the kernel32.dll, which in this case is 0x7c800000. Now, using the PE file format, the malware moves to the export table of kernel32.dll, as illustrated in the code below,

Figure 8:  Finding exportaddresstable of kernel32.dll

Looking at code above the instruction MOV EBX, DWORD PTR DS:[EAX+78], lands us at the datadirectory-->exportaddresstable of the kernel32.dll. The malware then retrieves the value and adds it to the imagebase (ie 0x7c800000) in order to reach the export table, where it retrieves the address of the exported function. Here too, the malware never uses the names of the function, but instead it uses a stored hash.

After further analysis, we stumble onto another piece of code, which copies data again from the data section to a newly allocated memory region.

Figure 9: Copy more data 

Investigating further, we see that this data is decrypted to reveal what looks like some sort of an address table.

Figure 10: Address table 

The table has significance as it is used as an address calculator, To calculate the address of

the region from where it copies bulk data, Which is further decrypted to form what looks like

a compressed file.

Figure 11: Compressed data

And there it is. Moving ahead, we land in the decompression routine, which quickly reveals that

the data is compressed using “aplib”.

Figure 12: Aplib decompression routine

Once the decompression is completed it does some familiar actions by flushing out the bytes of the original EXE file starting from the imagebase 0x400000 and copy the decompressed data to its new imagebase (i.e 0x400000)

Figure 13: Copy decompressed PE -file 

Finally using “LoadlibraryEx” and Getprocaddress the IAT is rebuild in the memory after which the control is transferred to the new code at the address 0x401021

Figure14: Rebuild IAT in memory

The job of this code is limited. It writes a PE-file which is embedded within itself into the temporary folder as “Adobe.exe” using the api “GetTempPathA”.

Figure 15: Transfer control to OEP

In the end, the file (Adobe.exe) is dropped in the temp folder and executed using the API “ShellExecuteA”.

Figure 16: Execute dropped "Adobe.exe"

A dummy PDF file is also written to the current directory named “Bestellung.pdf”. In a subsequent blog post, we will see why the malware dropped this PDF file.

That’s all for now. In the next post, we’ll continue the analysis of the dropped file “Adobe.exe”

Analyzing PDF exploits for finding payloads used

We have written a couple of previous blogs which focus on an in-depth analysis of PDF exploits as this is yet another techniques used by attackers to package malicious code and avoid antivirus detection. We have also written in the past about different decoding filters used to hide the malicious code inside PDF files. In this blog, we will examine yet another in the wild PDF exploit which has hidden it’s malicious code under different objects. We will also identify the final payload used to carry out the attack. The malicious PDF sample was retrieved from “hxxp://sj1s.co.cc/games/pdf.php?f=21”. Here is the PDF source code:

The above PDF file is small in size and contains clear text JavaScript. Let’s look at object 1, which contains malicious JavaScript code. This code is very simple to read and understand. The JavaScript accesses some property values declared elsewhere such as “this.producer”, “this.subject” etc. Now, where are those declared values coming from? If you look at the object 3, you will notice that all other variables are accessed from these object properties. The strings used like “eval” and “StringfromCharCode”, suggest this JavaScript is used for malicious purposes. Now, we have 3 different strings from object 3, which will be used in the malicious JavaScript code.

1) Property Producer contains a long array of values
2) Property Subject contains string “eval”
3) Property Title contains string “StringfromCharCode”

We will use Malzilla for decoding this malicious JavaScript. For that, we need to substitute the respective values into the variables. We will re-create the JavaScript code for Malzilla using those values to as it with the decoding process. We will need to look at the code carefully, because the array contains some substitution and arithmetic operations. Here is what we need to substitute:

Let’s take the first value from the Producer property array, which is “t9.5*w”. The malicious JavaScript contains one variable “w”, declared with a value of 4 and then there is another function (axp = axp.replace\(/t/g,'2'\);), which will replace character “t” with value 2. So the first integer of the array will become (29.5*4) = 118. When we substitute the whole array with values of “t” and “w” we can create the final simple JavaScript code:

We will now decode the above simplified code using Malzilla. The decoded content is shown below:

The decoded content contains malicious heap spray code, shellcode and code for attacking different Adobe vulnerabilities. However, we have to yet identify what this malicious code does once it exploits the vulnerability? What payloads does it use for the exploit? For this we need to identify the shellcode used. Here is what the shellcode looks like:

The shellcode used, is in %u Unicode-encoded format. We will convert this code into byte code or executable code for further reversing using IDA pro or OllyDbg. For this, we will use favorite online tool Shellcode 2 EXE. We will copy and paste the shellcode bytes from the variable, which will generate a sample “.exe” file to analyze. Here is the screenshot:

Now, we have executable file to analyze. So let’s open in IDA pro first to look at the strings used inside the payload. Here are the strings found,

The string shows that this payload is going to download additional files on the system. Now let’s open this file in OllyDbg for obtaining the malicious URL used inside the payload.

The shellcode starts with NOP instructions followed by another loop which will decode the malicious code. Look at the instructions above, inside the highlighted box. Those are the instructions which are used to decode everything. By stepping through the code, we come to know that there is an instruction that will compare the value with E9 to exit and another, which is XORing byte with a value of 31. We will put breakpoint at the RETN instruction. The code will successfully run and we will be presented with the decoded content, which contains more interesting strings.

Look at the highlighted string above in the dump area. This URL will be used to download another binary from the server. Now, we have identified this malicious payload and the URL used. For reference, here is the result from ThreatExpert for the same shellcode.

That’s it for now.

Umesh