PMA Labs Writeup: OllyDbg

by on under malware-analysis
13 minute read

Practical Malware Analysis is still a handbook for aspiring malware analysts, and while I’ve dabbled in the subject before, I’ve decided to work through the book for a better hands-on grasp of malware reverse engineering. Needless to say, this writeup will contain spoilers.

Chapter 9: OllyDbg

Lab 09-01

We’ll refer back to the basic dynamic analysis writeup from Chapter 3 rather than rehashing that for this executable. This was Lab03-04.exe. We saw at the time that the executable tried to delete itself when we ran it directly. It’s likely expecting a command-line parameter; let’s take a peek with IDA and find out.

Disassembly

Glancing briefly over the file, we see it checks for an initial parameter, and self-destructs if it doesn’t pass the complex test function. We’ll test this with OllyDbg to see what it does, exactly. Then it looks for more command-line arguments and tests a couple specifically:

-in [param1] password
-re [param1] password
-c param1 param2 param3 param4 password
-cc password

Then it checks for the number of additional arguments (up to 4!) and, if they do not match, it self-destructs. This makes the selfDestruct function fairly easy to pick out.

We’ll look at these command-line arguments, come up with some hypotheses about what they do, and then test them with OllyDbg.

The -in flag

This looks like the install argument. There’s a function call to get the current executable’s filename, and then a function call that has logic around creating a service. The service name appears to be partly comprised of the string “ Manager Service”, but it looks like there’s another component, probably the service name passed in as a function parameter. When invoked as Lab09-01.exe password -in, the service name comes from the executable; when invoked as Lab09-01.exe password -in param2, it comes from the optional second argument.

Once we dissect the password, we’ll test this dynamically and see if we can set the service name.

The -re flag

This looks like the uninstall argument. It’s very similar to the code under the -in flag, but it deletes the service instead of creating it.

Once we dissect the password, we’ll test this dynamically and see if we can remove the service previously created.

The -c flag

This creates a registry key under “SOFTWARE\Microsoft \XPS” (note the space) with a value pair named “Configuration”. It’s not immediately obvious what value is being stored, so we’ll analyze that later.

Once we dissect the password, we’ll test this dynamically and try adding a registry key with some junk data. Then we’ll watch what it does with OllyDbg.

The -cc flag

This seems to be the reverse of the -c flag, querying the registry key and performing some similarly obscure processing on the value. Then it runs the results through printf (not going to catch me again, rabbit hole!).

No flag

When we ran it before with no flag, nothing happened; but there is a path for it, if the registry key exists! In that case it does some processing based on certain commands which appear to be stored in the registry. Of note are “SLEEP”, “NOTHING”, “DOWNLOAD”, “CMD”, and “UPLOAD”. We’ll test these with the -c flag and see if we can decipher how they are to be used.

Debugging

From looking at the disassembly in IDA, there are references to 0x61 (a) and 0x63 (c), and it appears to be checking for a string of length 4, so we’ll start there. We’ll invoke OllyDbg, open the malware with parameters -in acac, and add a breakpoint on the password validation function.

On the first run through we note that the length is accepted (whew), it verifies the first character, and then fails the second. It’s doing some calculation that should result in a 1, but ends up as a 2. Hmm… well, let’s try a different password, -in attr. This time the difference is higher. Perhaps the calculation is the difference between the first and second characters? We try -in abcd and it passes!

We let the execution continue and then check our services. Sure enough, Lab09-01 Manager Service now appears in the list! Let’s patch the malware to remove the password - and maybe the self-destruct function while we’re at it - and then we’ll switch back to dynamic analysis to test our theories above.

To patch the malware, I am just replacing the body of those two functions with NOP instructions (0x90), being careful not to mess with the stack instructions at the beginning and end. I save the changes to Lab09-01_patched.exe. Now the moment of truth: if I run it without arguments, will it delete itself? It does not! And running it with a bogus password works too.

Back to the dynamic analysis. As predicted, the -in flag installs the service, by default using the name of the executable. The -re flag removes the service. The -c flag writes to the XPS registry key, and stores the four parameters we passed, encoded as binary. The -cc flag prints the string k:param1 h:param2 p:param3 per:param4. This may be a clue as to their purpose. Let’s open OllyDbg again and see what more we can find out.

As noted before, if there is a registry key set, we can run the malware without any command-line parameters and it will go into a loop waiting for activity. We’ll skip past the part where it reads from the registry and put a breakpoint at the command handler at sub_402020.

After stepping through the execution, we find that it tries to convert “param3” to a number with atoi and fails. IDA automatically correlates this variable with hostshort, so this might be a port number. Let’s try setting this parameter in the registry to 80 and run again.

Lab09-01_patched.exe -c param1 param2 80 param4 abcd

Sure enough, this step now passes. The malware sets some parameters and calls a function with strings like “GET” and “HTTP/1.0”, suggesting an HTTP connection. One of the parameters is “param2”, one is 0x50 (which we’ve seen before as the hex for port 80), one is a mysterious string “QUAY/wVLs.3xG”, and the other two are memory addresses. Let’s see if we can figure out how these are used.

The first thing it does is try to establish a socket to “param2” on port 0x50. This must be where we specify the hostname. We have our mock network up and running, so let’s specify a domain and update our registry config:

Lab09-01_patched.exe -c param1 practicalmalwareanalysis.com 80 param4 abcd

On this run through the mysterious string changed to “7FDj/1cdS.bY4”. As we continue stepping through, it’s appended with a couple other strings to form “GET 7FDj/1cdS.by4 HTTP/1.0” (plus some carriage returns at the end) and then sent over the socket. Then it waits for a response. If it doesn’t get one, it quits. Otherwise, it keeps getting chunks until it gets one that contains \r\n\r\n, then returns the contents of the response.

Now that we’ve labeled this function in IDA, the following logic starts to make more sense. It checks for the string `'`'`, followed by the string '`'`', and makes sure they are less than 0x400 characters apart. Then, at a guess, it extracts the command from between those indicators and returns it.

Finally, jumping back up, there’s a series of if-checks to decide what to do for each command. UPLOAD and DOWNLOAD seem to expect a series of strings, like:

`'`'` UPLOAD [port] [filename] '`'`'
`'`'` DOWNLOAD [port] [filename] '`'`'

CMD opens a reverse shell, piping the output to a port, and wrapping the command in backticks to allow spaces:

`'`'` CMD [port] `[cmd]` '`'`'

SLEEP sleeps for the specified number of milliseconds:

`'`'` SLEEP [milliseconds] '`'`'

And NOTHING, as might be expected, does nothing.

The HTTP request the malware sends is malformed, so inetsim complains. And the malware is expecting a response after it sends its request, so a simple netcat listener won’t do. Instead, we can whip up a simple Python responder on our Ubuntu analysis machine:

import socket

sock = socket.create_server(("192.168.42.1", 80))
sock.listn(5)

while True:
  (client, address) = sock.accept()
  with client:
    print('Connected by ', address)
    while True:
      data = client.recv(1024)
      if not data:
        break
      print(data)
      print("Sending command...")
      client.sendall(b"`'`'`DOWNLOAD 15000 C:\\test.txt'`'`'\r\n\r\n")
      print("Command sent")

This may not be ideal Python socket code, but it works to test our hypotheses. We’ll set up a netcat listener on port 15000 with nc -l -p 15000 > output.txt and create a text file on the malware host as the data to be downloaded.

Then we run the server and test the malware. Sure enough, the command is sent, and when we inspect output.txt it contains the contents of our C:\test.txt data file!

Upload works similarly, but in reverse: we can pipe a file to netcat with nc -l -p 15000 < input.txt and the file will be created on the malware host.

One of our hypotheses was incorrect: when tested, SLEEP’s parameter turns out to be in seconds, not milliseconds. Everything else (even NOTHING) works as expected. It’s worth noting that the malware will exit after a successful CMD, UPLOAD, or DOWNLOAD, and otherwise loop indefinitely.

  • Host-based signatures
    • Services
      • /.* Management Service/
    • Registry Keys
      • SOFTWARE\Microsoft \XPS\Configuration
  • Network-based signatures
    • Traffic
      • /GET [a-zA-Z0-9]{4}/[a-zA-Z0-9]{4}\.[a-zA-Z0-9]{3} HTTP/1.0/
      • The lack of a leading slash may help eliminate false positives, but this is not a very reliable signature.

Lab 09-02

Static Analysis

There are no strings of interest except one, cmd, suggesting any strings that are used are obfuscated somehow. The malware does not appear to be packed. There are imports for writing files, opening sockets, and creating processes.

Dynamic Analysis

The malware exits almost as soon as it starts. Initial dynamic analysis is fruitless. Let’s see if it’s expecting a command-line parameter.

Disassembly & Debugging

The first thing we notice is that it’s checking its own filename. We’ll fast-forward through the logic to strcmp and add a breakpoint to inspect the parameters. Sure enough: it’s checking if the filename equals ocl.exe. Stepping through the program and sniffing with Wireshark, we see it tries to connect to www[.]practicalmalwareanalysis.com on port 9999. If successful, it launches sub_401000.

We can set up a netcat listener on port 9999 of our Ubuntu analysis machine. When we do, and allow execution to proceed, it turns into a reverse shell! The malware uses CreateProcessA to spawn a cmd process, piping the stdin and stdout back through the socket. We now have an interactive command prompt.

Let’s step back for a moment and look at the string obfuscation. Starting at 0x00401133, the malware moves characters one byte at a time into the space at var_1B0. This is passed later, along with a copy of some scrambled data copied from 0x00405034, to sub_401089. This is a for loop that XORs the key (the string in var_1B0 from earlier) with the data (copied from 0x00405034).

Aside from the simple reverse shell, there does not appear to be any persistence or other interesting behavior.

Lab 09-03

Static Analysis

Lab09-03.exe shows no immediate signs of being packed. It imports a few functions from DLL1.dll and DLL2.dll, which we’ll get to in a moment. Other interesting imports include WriteFile from kernel32.dll and NetScheduleJobAdd from netapi32.dll.

DLL1.dll imports several functions, but none that are obviously of interest, and exports DLL1Print. Its image base is 0x10000000.

DLL2.dll imports CreateFileA from kernel32.dll, among others, and exports DLL2Print and DLL2ReturnJ. Its image base is 0x10000000.

DLL3.dll imports several functions, but none that are obviously of interest, and exports DLL3GetStructure and DLL3Print. Its image base is 0x10000000.

Interesting strings include:

Lab09-03.exe:
malwareanalysisbook.com

DLL1.dll:
DLL 1 mystery data %d

DLL2.dll:
temp.txt
DLL 2 mystery data %d

DLL3.dll:
ping www.malwareanalysisbook.com
DLL 3 mystery data %d

Debugging

When we initially load the malware into OllyDbg, the memory map reveals that DLL1 has been loaded into its preferred space of 0x10000000. DLL2 has been given the space 0x00330000, and DLL3 has not been loaded yet.

Stepping through the program, DLL1Print seems to just invoke printf with a global variable and the above “mystery data” template string. If we check in IDA, this global variable is being set in DLLMain to the current process ID.

Cheating a little bit (perhaps), we take a peek at DLL2.dll in IDA and see it’s following a similar pattern. Instead of saving the process ID to a global variable, however, it opens a file called temp.txt and saves the handle to a global variable. DLL2Print prints that handle, and DLL2ReturnJ returns the handle. To align our IDA analysis with what we see in OllyDbg, we can rebase the code at the actual load address rather than the DLL’s preferred load address.

DLL3 defines a data structure in DLLMain, which is returned by DLL3GetStructure. DLL3Print prints the pointer to the unicode command in the data structure.

Going back to the main executable, we see that the malware is getting the file handle from DLL2.dll and writing to that file. Then it gets the pointer to the data structure in DLL3, and passes that to NetScheduleJobAdd.

Wrap-up

I’ve enjoyed the labs thus far. I think they’ve given me a reasonable grounding in reverse engineering, especially for the Windows platform. Nevertheless, for a couple reasons, I’ve decided to skip the remaining labs and just read through the rest of the book. I expect this will still give me a base of knowledge about the field, but I don’t have any immediate plans to jump into malware analysis full time.

comments powered by Disqus